fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dir2_sf.h"
  33 #include "xfs_attr_sf.h"
  34 #include "xfs_dinode.h"
  35 #include "xfs_inode.h"
  36 #include "xfs_btree.h"
  37 #include "xfs_dmapi.h"
  38 #include "xfs_mount.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_itable.h"
  41 #include "xfs_dir2_data.h"
  42 #include "xfs_dir2_leaf.h"
  43 #include "xfs_dir2_block.h"
  44 #include "xfs_inode_item.h"
  45 #include "xfs_extfree_item.h"
  46 #include "xfs_alloc.h"
  47 #include "xfs_bmap.h"
  48 #include "xfs_rtalloc.h"
  49 #include "xfs_error.h"
  50 #include "xfs_attr_leaf.h"
  51 #include "xfs_rw.h"
  52 #include "xfs_quota.h"
  53 #include "xfs_trans_space.h"
  54 #include "xfs_buf_item.h"
  55 #include "xfs_filestream.h"
  56 #include "xfs_vnodeops.h"
  57
  58
  59 #ifdef DEBUG
  60 STATIC void
  61 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  62 #endif
  63
  64 kmem_zone_t             *xfs_bmap_free_item_zone;
  65
  66 /*
  67  * Prototypes for internal bmap routines.
  68  */
  69
  70
  71 /*
  72  * Called from xfs_bmap_add_attrfork to handle extents format files.
  73  */
  74 STATIC int                                      /* error */
  75 xfs_bmap_add_attrfork_extents(
  76         xfs_trans_t             *tp,            /* transaction pointer */
  77         xfs_inode_t             *ip,            /* incore inode pointer */
  78         xfs_fsblock_t           *firstblock,    /* first block allocated */
  79         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  80         int                     *flags);        /* inode logging flags */
  81
  82 /*
  83  * Called from xfs_bmap_add_attrfork to handle local format files.
  84  */
  85 STATIC int                                      /* error */
  86 xfs_bmap_add_attrfork_local(
  87         xfs_trans_t             *tp,            /* transaction pointer */
  88         xfs_inode_t             *ip,            /* incore inode pointer */
  89         xfs_fsblock_t           *firstblock,    /* first block allocated */
  90         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  91         int                     *flags);        /* inode logging flags */
  92
  93 /*
  94  * Called by xfs_bmapi to update file extent records and the btree
  95  * after allocating space (or doing a delayed allocation).
  96  */
  97 STATIC int                              /* error */
  98 xfs_bmap_add_extent(
  99         xfs_inode_t             *ip,    /* incore inode pointer */
 100         xfs_extnum_t            idx,    /* extent number to update/insert */
 101         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 102         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 103         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 104         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 105         int                     *logflagsp, /* inode logging flags */
 106         xfs_extdelta_t          *delta, /* Change made to incore extents */
 107         int                     whichfork, /* data or attr fork */
 108         int                     rsvd);  /* OK to allocate reserved blocks */
 109
 110 /*
 111  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 112  * allocation to a real allocation.
 113  */
 114 STATIC int                              /* error */
 115 xfs_bmap_add_extent_delay_real(
 116         xfs_inode_t             *ip,    /* incore inode pointer */
 117         xfs_extnum_t            idx,    /* extent number to update/insert */
 118         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 119         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 120         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 121         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 122         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 123         int                     *logflagsp, /* inode logging flags */
 124         xfs_extdelta_t          *delta, /* Change made to incore extents */
 125         int                     rsvd);  /* OK to allocate reserved blocks */
 126
 127 /*
 128  * Called by xfs_bmap_add_extent to handle cases converting a hole
 129  * to a delayed allocation.
 130  */
 131 STATIC int                              /* error */
 132 xfs_bmap_add_extent_hole_delay(
 133         xfs_inode_t             *ip,    /* incore inode pointer */
 134         xfs_extnum_t            idx,    /* extent number to update/insert */
 135         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 136         int                     *logflagsp,/* inode logging flags */
 137         xfs_extdelta_t          *delta, /* Change made to incore extents */
 138         int                     rsvd);  /* OK to allocate reserved blocks */
 139
 140 /*
 141  * Called by xfs_bmap_add_extent to handle cases converting a hole
 142  * to a real allocation.
 143  */
 144 STATIC int                              /* error */
 145 xfs_bmap_add_extent_hole_real(
 146         xfs_inode_t             *ip,    /* incore inode pointer */
 147         xfs_extnum_t            idx,    /* extent number to update/insert */
 148         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 149         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 150         int                     *logflagsp, /* inode logging flags */
 151         xfs_extdelta_t          *delta, /* Change made to incore extents */
 152         int                     whichfork); /* data or attr fork */
 153
 154 /*
 155  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 156  * allocation to a real allocation or vice versa.
 157  */
 158 STATIC int                              /* error */
 159 xfs_bmap_add_extent_unwritten_real(
 160         xfs_inode_t             *ip,    /* incore inode pointer */
 161         xfs_extnum_t            idx,    /* extent number to update/insert */
 162         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 163         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 164         int                     *logflagsp, /* inode logging flags */
 165         xfs_extdelta_t          *delta); /* Change made to incore extents */
 166
 167 /*
 168  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 169  * It figures out where to ask the underlying allocator to put the new extent.
 170  */
 171 STATIC int                              /* error */
 172 xfs_bmap_alloc(
 173         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 174
 175 /*
 176  * Transform a btree format file with only one leaf node, where the
 177  * extents list will fit in the inode, into an extents format file.
 178  * Since the file extents are already in-core, all we have to do is
 179  * give up the space for the btree root and pitch the leaf block.
 180  */
 181 STATIC int                              /* error */
 182 xfs_bmap_btree_to_extents(
 183         xfs_trans_t             *tp,    /* transaction pointer */
 184         xfs_inode_t             *ip,    /* incore inode pointer */
 185         xfs_btree_cur_t         *cur,   /* btree cursor */
 186         int                     *logflagsp, /* inode logging flags */
 187         int                     whichfork); /* data or attr fork */
 188
 189 /*
 190  * Called by xfs_bmapi to update file extent records and the btree
 191  * after removing space (or undoing a delayed allocation).
 192  */
 193 STATIC int                              /* error */
 194 xfs_bmap_del_extent(
 195         xfs_inode_t             *ip,    /* incore inode pointer */
 196         xfs_trans_t             *tp,    /* current trans pointer */
 197         xfs_extnum_t            idx,    /* extent number to update/insert */
 198         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 199         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 200         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 201         int                     *logflagsp,/* inode logging flags */
 202         xfs_extdelta_t          *delta, /* Change made to incore extents */
 203         int                     whichfork, /* data or attr fork */
 204         int                     rsvd);   /* OK to allocate reserved blocks */
 205
 206 /*
 207  * Remove the entry "free" from the free item list.  Prev points to the
 208  * previous entry, unless "free" is the head of the list.
 209  */
 210 STATIC void
 211 xfs_bmap_del_free(
 212         xfs_bmap_free_t         *flist, /* free item list header */
 213         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 214         xfs_bmap_free_item_t    *free); /* list item to be freed */
 215
 216 /*
 217  * Convert an extents-format file into a btree-format file.
 218  * The new file will have a root block (in the inode) and a single child block.
 219  */
 220 STATIC int                                      /* error */
 221 xfs_bmap_extents_to_btree(
 222         xfs_trans_t             *tp,            /* transaction pointer */
 223         xfs_inode_t             *ip,            /* incore inode pointer */
 224         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 225         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 226         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 227         int                     wasdel,         /* converting a delayed alloc */
 228         int                     *logflagsp,     /* inode logging flags */
 229         int                     whichfork);     /* data or attr fork */
 230
 231 /*
 232  * Convert a local file to an extents file.
 233  * This code is sort of bogus, since the file data needs to get
 234  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 235  */
 236 STATIC int                              /* error */
 237 xfs_bmap_local_to_extents(
 238         xfs_trans_t     *tp,            /* transaction pointer */
 239         xfs_inode_t     *ip,            /* incore inode pointer */
 240         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 241         xfs_extlen_t    total,          /* total blocks needed by transaction */
 242         int             *logflagsp,     /* inode logging flags */
 243         int             whichfork);     /* data or attr fork */
 244
 245 /*
 246  * Search the extents list for the inode, for the extent containing bno.
 247  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 248  * *eofp will be set, and *prevp will contain the last entry (null if none).
 249  * Else, *lastxp will be set to the index of the found
 250  * entry; *gotp will contain the entry.
 251  */
 252 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 253 xfs_bmap_search_extents(
 254         xfs_inode_t     *ip,            /* incore inode pointer */
 255         xfs_fileoff_t   bno,            /* block number searched for */
 256         int             whichfork,      /* data or attr fork */
 257         int             *eofp,          /* out: end of file found */
 258         xfs_extnum_t    *lastxp,        /* out: last extent index */
 259         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 260         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 261
 262 /*
 263  * Check the last inode extent to determine whether this allocation will result
 264  * in blocks being allocated at the end of the file. When we allocate new data
 265  * blocks at the end of the file which do not start at the previous data block,
 266  * we will try to align the new blocks at stripe unit boundaries.
 267  */
 268 STATIC int                              /* error */
 269 xfs_bmap_isaeof(
 270         xfs_inode_t     *ip,            /* incore inode pointer */
 271         xfs_fileoff_t   off,            /* file offset in fsblocks */
 272         int             whichfork,      /* data or attribute fork */
 273         char            *aeof);         /* return value */
 274
 275 #ifdef XFS_BMAP_TRACE
 276 /*
 277  * Add bmap trace entry prior to a call to xfs_iext_remove.
 278  */
 279 STATIC void
 280 xfs_bmap_trace_delete(
 281         const char      *fname,         /* function name */
 282         char            *desc,          /* operation description */
 283         xfs_inode_t     *ip,            /* incore inode pointer */
 284         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
 285         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
 286         int             whichfork);     /* data or attr fork */
 287
 288 /*
 289  * Add bmap trace entry prior to a call to xfs_iext_insert, or
 290  * reading in the extents list from the disk (in the btree).
 291  */
 292 STATIC void
 293 xfs_bmap_trace_insert(
 294         const char      *fname,         /* function name */
 295         char            *desc,          /* operation description */
 296         xfs_inode_t     *ip,            /* incore inode pointer */
 297         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
 298         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
 299         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
 300         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
 301         int             whichfork);     /* data or attr fork */
 302
 303 /*
 304  * Add bmap trace entry after updating an extent record in place.
 305  */
 306 STATIC void
 307 xfs_bmap_trace_post_update(
 308         const char      *fname,         /* function name */
 309         char            *desc,          /* operation description */
 310         xfs_inode_t     *ip,            /* incore inode pointer */
 311         xfs_extnum_t    idx,            /* index of entry updated */
 312         int             whichfork);     /* data or attr fork */
 313
 314 /*
 315  * Add bmap trace entry prior to updating an extent record in place.
 316  */
 317 STATIC void
 318 xfs_bmap_trace_pre_update(
 319         const char      *fname,         /* function name */
 320         char            *desc,          /* operation description */
 321         xfs_inode_t     *ip,            /* incore inode pointer */
 322         xfs_extnum_t    idx,            /* index of entry to be updated */
 323         int             whichfork);     /* data or attr fork */
 324
 325 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)       \
 326         xfs_bmap_trace_delete(__func__,d,ip,i,c,w)
 327 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w) \
 328         xfs_bmap_trace_insert(__func__,d,ip,i,c,r1,r2,w)
 329 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)    \
 330         xfs_bmap_trace_post_update(__func__,d,ip,i,w)
 331 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)     \
 332         xfs_bmap_trace_pre_update(__func__,d,ip,i,w)
 333 #else
 334 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)
 335 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w)
 336 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)
 337 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)
 338 #endif  /* XFS_BMAP_TRACE */
 339
 340 /*
 341  * Compute the worst-case number of indirect blocks that will be used
 342  * for ip's delayed extent of length "len".
 343  */
 344 STATIC xfs_filblks_t
 345 xfs_bmap_worst_indlen(
 346         xfs_inode_t             *ip,    /* incore inode pointer */
 347         xfs_filblks_t           len);   /* delayed extent length */
 348
 349 #ifdef DEBUG
 350 /*
 351  * Perform various validation checks on the values being returned
 352  * from xfs_bmapi().
 353  */
 354 STATIC void
 355 xfs_bmap_validate_ret(
 356         xfs_fileoff_t           bno,
 357         xfs_filblks_t           len,
 358         int                     flags,
 359         xfs_bmbt_irec_t         *mval,
 360         int                     nmap,
 361         int                     ret_nmap);
 362 #else
 363 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 364 #endif /* DEBUG */
 365
 366 #if defined(XFS_RW_TRACE)
 367 STATIC void
 368 xfs_bunmap_trace(
 369         xfs_inode_t             *ip,
 370         xfs_fileoff_t           bno,
 371         xfs_filblks_t           len,
 372         int                     flags,
 373         inst_t                  *ra);
 374 #else
 375 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
 376 #endif  /* XFS_RW_TRACE */
 377
 378 STATIC int
 379 xfs_bmap_count_tree(
 380         xfs_mount_t     *mp,
 381         xfs_trans_t     *tp,
 382         xfs_ifork_t     *ifp,
 383         xfs_fsblock_t   blockno,
 384         int             levelin,
 385         int             *count);
 386
 387 STATIC int
 388 xfs_bmap_count_leaves(
 389         xfs_ifork_t             *ifp,
 390         xfs_extnum_t            idx,
 391         int                     numrecs,
 392         int                     *count);
 393
 394 STATIC int
 395 xfs_bmap_disk_count_leaves(
 396         xfs_extnum_t            idx,
 397         xfs_bmbt_block_t        *block,
 398         int                     numrecs,
 399         int                     *count);
 400
 401 /*
 402  * Bmap internal routines.
 403  */
 404
 405 /*
 406  * Called from xfs_bmap_add_attrfork to handle btree format files.
 407  */
 408 STATIC int                                      /* error */
 409 xfs_bmap_add_attrfork_btree(
 410         xfs_trans_t             *tp,            /* transaction pointer */
 411         xfs_inode_t             *ip,            /* incore inode pointer */
 412         xfs_fsblock_t           *firstblock,    /* first block allocated */
 413         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 414         int                     *flags)         /* inode logging flags */
 415 {
 416         xfs_btree_cur_t         *cur;           /* btree cursor */
 417         int                     error;          /* error return value */
 418         xfs_mount_t             *mp;            /* file system mount struct */
 419         int                     stat;           /* newroot status */
 420
 421         mp = ip->i_mount;
 422         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 423                 *flags |= XFS_ILOG_DBROOT;
 424         else {
 425                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
 426                         XFS_DATA_FORK);
 427                 cur->bc_private.b.flist = flist;
 428                 cur->bc_private.b.firstblock = *firstblock;
 429                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 430                         goto error0;
 431                 /* must be at least one entry */
 432                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 433                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
 434                         goto error0;
 435                 if (stat == 0) {
 436                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 437                         return XFS_ERROR(ENOSPC);
 438                 }
 439                 *firstblock = cur->bc_private.b.firstblock;
 440                 cur->bc_private.b.allocated = 0;
 441                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 442         }
 443         return 0;
 444 error0:
 445         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 446         return error;
 447 }
 448
 449 /*
 450  * Called from xfs_bmap_add_attrfork to handle extents format files.
 451  */
 452 STATIC int                                      /* error */
 453 xfs_bmap_add_attrfork_extents(
 454         xfs_trans_t             *tp,            /* transaction pointer */
 455         xfs_inode_t             *ip,            /* incore inode pointer */
 456         xfs_fsblock_t           *firstblock,    /* first block allocated */
 457         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 458         int                     *flags)         /* inode logging flags */
 459 {
 460         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 461         int                     error;          /* error return value */
 462
 463         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 464                 return 0;
 465         cur = NULL;
 466         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 467                 flags, XFS_DATA_FORK);
 468         if (cur) {
 469                 cur->bc_private.b.allocated = 0;
 470                 xfs_btree_del_cursor(cur,
 471                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 472         }
 473         return error;
 474 }
 475
 476 /*
 477  * Called from xfs_bmap_add_attrfork to handle local format files.
 478  */
 479 STATIC int                                      /* error */
 480 xfs_bmap_add_attrfork_local(
 481         xfs_trans_t             *tp,            /* transaction pointer */
 482         xfs_inode_t             *ip,            /* incore inode pointer */
 483         xfs_fsblock_t           *firstblock,    /* first block allocated */
 484         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 485         int                     *flags)         /* inode logging flags */
 486 {
 487         xfs_da_args_t           dargs;          /* args for dir/attr code */
 488         int                     error;          /* error return value */
 489         xfs_mount_t             *mp;            /* mount structure pointer */
 490
 491         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 492                 return 0;
 493         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 494                 mp = ip->i_mount;
 495                 memset(&dargs, 0, sizeof(dargs));
 496                 dargs.dp = ip;
 497                 dargs.firstblock = firstblock;
 498                 dargs.flist = flist;
 499                 dargs.total = mp->m_dirblkfsbs;
 500                 dargs.whichfork = XFS_DATA_FORK;
 501                 dargs.trans = tp;
 502                 error = xfs_dir2_sf_to_block(&dargs);
 503         } else
 504                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 505                         XFS_DATA_FORK);
 506         return error;
 507 }
 508
 509 /*
 510  * Called by xfs_bmapi to update file extent records and the btree
 511  * after allocating space (or doing a delayed allocation).
 512  */
 513 STATIC int                              /* error */
 514 xfs_bmap_add_extent(
 515         xfs_inode_t             *ip,    /* incore inode pointer */
 516         xfs_extnum_t            idx,    /* extent number to update/insert */
 517         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 518         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 519         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 520         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 521         int                     *logflagsp, /* inode logging flags */
 522         xfs_extdelta_t          *delta, /* Change made to incore extents */
 523         int                     whichfork, /* data or attr fork */
 524         int                     rsvd)   /* OK to use reserved data blocks */
 525 {
 526         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 527         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 528         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 529         int                     error;  /* error return value */
 530         xfs_ifork_t             *ifp;   /* inode fork ptr */
 531         int                     logflags; /* returned value */
 532         xfs_extnum_t            nextents; /* number of extents in file now */
 533
 534         XFS_STATS_INC(xs_add_exlist);
 535         cur = *curp;
 536         ifp = XFS_IFORK_PTR(ip, whichfork);
 537         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 538         ASSERT(idx <= nextents);
 539         da_old = da_new = 0;
 540         error = 0;
 541         /*
 542          * This is the first extent added to a new/empty file.
 543          * Special case this one, so other routines get to assume there are
 544          * already extents in the list.
 545          */
 546         if (nextents == 0) {
 547                 XFS_BMAP_TRACE_INSERT("insert empty", ip, 0, 1, new, NULL,
 548                         whichfork);
 549                 xfs_iext_insert(ifp, 0, 1, new);
 550                 ASSERT(cur == NULL);
 551                 ifp->if_lastex = 0;
 552                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
 553                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 554                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
 555                 } else
 556                         logflags = 0;
 557                 /* DELTA: single new extent */
 558                 if (delta) {
 559                         if (delta->xed_startoff > new->br_startoff)
 560                                 delta->xed_startoff = new->br_startoff;
 561                         if (delta->xed_blockcount <
 562                                         new->br_startoff + new->br_blockcount)
 563                                 delta->xed_blockcount = new->br_startoff +
 564                                                 new->br_blockcount;
 565                 }
 566         }
 567         /*
 568          * Any kind of new delayed allocation goes here.
 569          */
 570         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
 571                 if (cur)
 572                         ASSERT((cur->bc_private.b.flags &
 573                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 574                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 575                                 &logflags, delta, rsvd)))
 576                         goto done;
 577         }
 578         /*
 579          * Real allocation off the end of the file.
 580          */
 581         else if (idx == nextents) {
 582                 if (cur)
 583                         ASSERT((cur->bc_private.b.flags &
 584                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 585                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 586                                 &logflags, delta, whichfork)))
 587                         goto done;
 588         } else {
 589                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 590
 591                 /*
 592                  * Get the record referred to by idx.
 593                  */
 594                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
 595                 /*
 596                  * If it's a real allocation record, and the new allocation ends
 597                  * after the start of the referred to record, then we're filling
 598                  * in a delayed or unwritten allocation with a real one, or
 599                  * converting real back to unwritten.
 600                  */
 601                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
 602                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 603                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 604                             ISNULLSTARTBLOCK(prev.br_startblock)) {
 605                                 da_old = STARTBLOCKVAL(prev.br_startblock);
 606                                 if (cur)
 607                                         ASSERT(cur->bc_private.b.flags &
 608                                                 XFS_BTCUR_BPRV_WASDEL);
 609                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 610                                         idx, &cur, new, &da_new, first, flist,
 611                                         &logflags, delta, rsvd)))
 612                                         goto done;
 613                         } else if (new->br_state == XFS_EXT_NORM) {
 614                                 ASSERT(new->br_state == XFS_EXT_NORM);
 615                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 616                                         ip, idx, &cur, new, &logflags, delta)))
 617                                         goto done;
 618                         } else {
 619                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 620                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 621                                         ip, idx, &cur, new, &logflags, delta)))
 622                                         goto done;
 623                         }
 624                         ASSERT(*curp == cur || *curp == NULL);
 625                 }
 626                 /*
 627                  * Otherwise we're filling in a hole with an allocation.
 628                  */
 629                 else {
 630                         if (cur)
 631                                 ASSERT((cur->bc_private.b.flags &
 632                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 633                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 634                                         new, &logflags, delta, whichfork)))
 635                                 goto done;
 636                 }
 637         }
 638
 639         ASSERT(*curp == cur || *curp == NULL);
 640         /*
 641          * Convert to a btree if necessary.
 642          */
 643         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 644             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 645                 int     tmp_logflags;   /* partial log flag return val */
 646
 647                 ASSERT(cur == NULL);
 648                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 649                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 650                 logflags |= tmp_logflags;
 651                 if (error)
 652                         goto done;
 653         }
 654         /*
 655          * Adjust for changes in reserved delayed indirect blocks.
 656          * Nothing to do for disk quotas here.
 657          */
 658         if (da_old || da_new) {
 659                 xfs_filblks_t   nblks;
 660
 661                 nblks = da_new;
 662                 if (cur)
 663                         nblks += cur->bc_private.b.allocated;
 664                 ASSERT(nblks <= da_old);
 665                 if (nblks < da_old)
 666                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 667                                 (int64_t)(da_old - nblks), rsvd);
 668         }
 669         /*
 670          * Clear out the allocated field, done with it now in any case.
 671          */
 672         if (cur) {
 673                 cur->bc_private.b.allocated = 0;
 674                 *curp = cur;
 675         }
 676 done:
 677 #ifdef DEBUG
 678         if (!error)
 679                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 680 #endif
 681         *logflagsp = logflags;
 682         return error;
 683 }
 684
 685 /*
 686  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 687  * allocation to a real allocation.
 688  */
 689 STATIC int                              /* error */
 690 xfs_bmap_add_extent_delay_real(
 691         xfs_inode_t             *ip,    /* incore inode pointer */
 692         xfs_extnum_t            idx,    /* extent number to update/insert */
 693         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 694         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 695         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 696         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 697         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 698         int                     *logflagsp, /* inode logging flags */
 699         xfs_extdelta_t          *delta, /* Change made to incore extents */
 700         int                     rsvd)   /* OK to use reserved data block allocation */
 701 {
 702         xfs_btree_cur_t         *cur;   /* btree cursor */
 703         int                     diff;   /* temp value */
 704         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 705         int                     error;  /* error return value */
 706         int                     i;      /* temp state */
 707         xfs_ifork_t             *ifp;   /* inode fork pointer */
 708         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 709         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 710                                         /* left is 0, right is 1, prev is 2 */
 711         int                     rval=0; /* return value (logging flags) */
 712         int                     state = 0;/* state bits, accessed thru macros */
 713         xfs_filblks_t           temp=0; /* value for dnew calculations */
 714         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 715         int                     tmp_rval;       /* partial logging flags */
 716         enum {                          /* bit number definitions for state */
 717                 LEFT_CONTIG,    RIGHT_CONTIG,
 718                 LEFT_FILLING,   RIGHT_FILLING,
 719                 LEFT_DELAY,     RIGHT_DELAY,
 720                 LEFT_VALID,     RIGHT_VALID
 721         };
 722
 723 #define LEFT            r[0]
 724 #define RIGHT           r[1]
 725 #define PREV            r[2]
 726 #define MASK(b)         (1 << (b))
 727 #define MASK2(a,b)      (MASK(a) | MASK(b))
 728 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
 729 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
 730 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
 731 #define STATE_TEST(b)   (state & MASK(b))
 732 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
 733                                        ((state &= ~MASK(b)), 0))
 734 #define SWITCH_STATE            \
 735         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
 736
 737         /*
 738          * Set up a bunch of variables to make the tests simpler.
 739          */
 740         cur = *curp;
 741         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 742         ep = xfs_iext_get_ext(ifp, idx);
 743         xfs_bmbt_get_all(ep, &PREV);
 744         new_endoff = new->br_startoff + new->br_blockcount;
 745         ASSERT(PREV.br_startoff <= new->br_startoff);
 746         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 747         /*
 748          * Set flags determining what part of the previous delayed allocation
 749          * extent is being replaced by a real allocation.
 750          */
 751         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
 752         STATE_SET(RIGHT_FILLING,
 753                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
 754         /*
 755          * Check and set flags if this segment has a left neighbor.
 756          * Don't set contiguous if the combined extent would be too large.
 757          */
 758         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
 759                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
 760                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
 761         }
 762         STATE_SET(LEFT_CONTIG,
 763                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
 764                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 765                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 766                 LEFT.br_state == new->br_state &&
 767                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
 768         /*
 769          * Check and set flags if this segment has a right neighbor.
 770          * Don't set contiguous if the combined extent would be too large.
 771          * Also check for all-three-contiguous being too large.
 772          */
 773         if (STATE_SET_TEST(RIGHT_VALID,
 774                         idx <
 775                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
 776                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
 777                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
 778         }
 779         STATE_SET(RIGHT_CONTIG,
 780                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
 781                 new_endoff == RIGHT.br_startoff &&
 782                 new->br_startblock + new->br_blockcount ==
 783                     RIGHT.br_startblock &&
 784                 new->br_state == RIGHT.br_state &&
 785                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 786                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
 787                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
 788                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 789                      <= MAXEXTLEN));
 790         error = 0;
 791         /*
 792          * Switch out based on the FILLING and CONTIG state bits.
 793          */
 794         switch (SWITCH_STATE) {
 795
 796         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
 797                 /*
 798                  * Filling in all of a previously delayed allocation extent.
 799                  * The left and right neighbors are both contiguous with new.
 800                  */
 801                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 802                         XFS_DATA_FORK);
 803                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 804                         LEFT.br_blockcount + PREV.br_blockcount +
 805                         RIGHT.br_blockcount);
 806                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 807                         XFS_DATA_FORK);
 808                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
 809                 xfs_iext_remove(ifp, idx, 2);
 810                 ip->i_df.if_lastex = idx - 1;
 811                 ip->i_d.di_nextents--;
 812                 if (cur == NULL)
 813                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 814                 else {
 815                         rval = XFS_ILOG_CORE;
 816                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 817                                         RIGHT.br_startblock,
 818                                         RIGHT.br_blockcount, &i)))
 819                                 goto done;
 820                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 821                         if ((error = xfs_bmbt_delete(cur, &i)))
 822                                 goto done;
 823                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 824                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
 825                                 goto done;
 826                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 827                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 828                                         LEFT.br_startblock,
 829                                         LEFT.br_blockcount +
 830                                         PREV.br_blockcount +
 831                                         RIGHT.br_blockcount, LEFT.br_state)))
 832                                 goto done;
 833                 }
 834                 *dnew = 0;
 835                 /* DELTA: Three in-core extents are replaced by one. */
 836                 temp = LEFT.br_startoff;
 837                 temp2 = LEFT.br_blockcount +
 838                         PREV.br_blockcount +
 839                         RIGHT.br_blockcount;
 840                 break;
 841
 842         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
 843                 /*
 844                  * Filling in all of a previously delayed allocation extent.
 845                  * The left neighbor is contiguous, the right is not.
 846                  */
 847                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
 848                         XFS_DATA_FORK);
 849                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 850                         LEFT.br_blockcount + PREV.br_blockcount);
 851                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
 852                         XFS_DATA_FORK);
 853                 ip->i_df.if_lastex = idx - 1;
 854                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
 855                 xfs_iext_remove(ifp, idx, 1);
 856                 if (cur == NULL)
 857                         rval = XFS_ILOG_DEXT;
 858                 else {
 859                         rval = 0;
 860                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 861                                         LEFT.br_startblock, LEFT.br_blockcount,
 862                                         &i)))
 863                                 goto done;
 864                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 865                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 866                                         LEFT.br_startblock,
 867                                         LEFT.br_blockcount +
 868                                         PREV.br_blockcount, LEFT.br_state)))
 869                                 goto done;
 870                 }
 871                 *dnew = 0;
 872                 /* DELTA: Two in-core extents are replaced by one. */
 873                 temp = LEFT.br_startoff;
 874                 temp2 = LEFT.br_blockcount +
 875                         PREV.br_blockcount;
 876                 break;
 877
 878         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
 879                 /*
 880                  * Filling in all of a previously delayed allocation extent.
 881                  * The right neighbor is contiguous, the left is not.
 882                  */
 883                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 884                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 885                 xfs_bmbt_set_blockcount(ep,
 886                         PREV.br_blockcount + RIGHT.br_blockcount);
 887                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 888                 ip->i_df.if_lastex = idx;
 889                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
 890                 xfs_iext_remove(ifp, idx + 1, 1);
 891                 if (cur == NULL)
 892                         rval = XFS_ILOG_DEXT;
 893                 else {
 894                         rval = 0;
 895                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 896                                         RIGHT.br_startblock,
 897                                         RIGHT.br_blockcount, &i)))
 898                                 goto done;
 899                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 900                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 901                                         new->br_startblock,
 902                                         PREV.br_blockcount +
 903                                         RIGHT.br_blockcount, PREV.br_state)))
 904                                 goto done;
 905                 }
 906                 *dnew = 0;
 907                 /* DELTA: Two in-core extents are replaced by one. */
 908                 temp = PREV.br_startoff;
 909                 temp2 = PREV.br_blockcount +
 910                         RIGHT.br_blockcount;
 911                 break;
 912
 913         case MASK2(LEFT_FILLING, RIGHT_FILLING):
 914                 /*
 915                  * Filling in all of a previously delayed allocation extent.
 916                  * Neither the left nor right neighbors are contiguous with
 917                  * the new one.
 918                  */
 919                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 920                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 921                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 922                 ip->i_df.if_lastex = idx;
 923                 ip->i_d.di_nextents++;
 924                 if (cur == NULL)
 925                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 926                 else {
 927                         rval = XFS_ILOG_CORE;
 928                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 929                                         new->br_startblock, new->br_blockcount,
 930                                         &i)))
 931                                 goto done;
 932                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 933                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 934                         if ((error = xfs_bmbt_insert(cur, &i)))
 935                                 goto done;
 936                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 937                 }
 938                 *dnew = 0;
 939                 /* DELTA: The in-core extent described by new changed type. */
 940                 temp = new->br_startoff;
 941                 temp2 = new->br_blockcount;
 942                 break;
 943
 944         case MASK2(LEFT_FILLING, LEFT_CONTIG):
 945                 /*
 946                  * Filling in the first part of a previous delayed allocation.
 947                  * The left neighbor is contiguous.
 948                  */
 949                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 950                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 951                         LEFT.br_blockcount + new->br_blockcount);
 952                 xfs_bmbt_set_startoff(ep,
 953                         PREV.br_startoff + new->br_blockcount);
 954                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 955                 temp = PREV.br_blockcount - new->br_blockcount;
 956                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 957                 xfs_bmbt_set_blockcount(ep, temp);
 958                 ip->i_df.if_lastex = idx - 1;
 959                 if (cur == NULL)
 960                         rval = XFS_ILOG_DEXT;
 961                 else {
 962                         rval = 0;
 963                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 964                                         LEFT.br_startblock, LEFT.br_blockcount,
 965                                         &i)))
 966                                 goto done;
 967                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 968                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 969                                         LEFT.br_startblock,
 970                                         LEFT.br_blockcount +
 971                                         new->br_blockcount,
 972                                         LEFT.br_state)))
 973                                 goto done;
 974                 }
 975                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 976                         STARTBLOCKVAL(PREV.br_startblock));
 977                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
 978                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 979                 *dnew = temp;
 980                 /* DELTA: The boundary between two in-core extents moved. */
 981                 temp = LEFT.br_startoff;
 982                 temp2 = LEFT.br_blockcount +
 983                         PREV.br_blockcount;
 984                 break;
 985
 986         case MASK(LEFT_FILLING):
 987                 /*
 988                  * Filling in the first part of a previous delayed allocation.
 989                  * The left neighbor is not contiguous.
 990                  */
 991                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
 992                 xfs_bmbt_set_startoff(ep, new_endoff);
 993                 temp = PREV.br_blockcount - new->br_blockcount;
 994                 xfs_bmbt_set_blockcount(ep, temp);
 995                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
 996                         XFS_DATA_FORK);
 997                 xfs_iext_insert(ifp, idx, 1, new);
 998                 ip->i_df.if_lastex = idx;
 999                 ip->i_d.di_nextents++;
1000                 if (cur == NULL)
1001                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1002                 else {
1003                         rval = XFS_ILOG_CORE;
1004                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1005                                         new->br_startblock, new->br_blockcount,
1006                                         &i)))
1007                                 goto done;
1008                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1009                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1010                         if ((error = xfs_bmbt_insert(cur, &i)))
1011                                 goto done;
1012                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1013                 }
1014                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1015                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1016                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1017                                         first, flist, &cur, 1, &tmp_rval,
1018                                         XFS_DATA_FORK);
1019                         rval |= tmp_rval;
1020                         if (error)
1021                                 goto done;
1022                 }
1023                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1024                         STARTBLOCKVAL(PREV.br_startblock) -
1025                         (cur ? cur->bc_private.b.allocated : 0));
1026                 ep = xfs_iext_get_ext(ifp, idx + 1);
1027                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1028                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx + 1, XFS_DATA_FORK);
1029                 *dnew = temp;
1030                 /* DELTA: One in-core extent is split in two. */
1031                 temp = PREV.br_startoff;
1032                 temp2 = PREV.br_blockcount;
1033                 break;
1034
1035         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1036                 /*
1037                  * Filling in the last part of a previous delayed allocation.
1038                  * The right neighbor is contiguous with the new allocation.
1039                  */
1040                 temp = PREV.br_blockcount - new->br_blockcount;
1041                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1042                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1043                 xfs_bmbt_set_blockcount(ep, temp);
1044                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1045                         new->br_startoff, new->br_startblock,
1046                         new->br_blockcount + RIGHT.br_blockcount,
1047                         RIGHT.br_state);
1048                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1049                 ip->i_df.if_lastex = idx + 1;
1050                 if (cur == NULL)
1051                         rval = XFS_ILOG_DEXT;
1052                 else {
1053                         rval = 0;
1054                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1055                                         RIGHT.br_startblock,
1056                                         RIGHT.br_blockcount, &i)))
1057                                 goto done;
1058                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1059                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1060                                         new->br_startblock,
1061                                         new->br_blockcount +
1062                                         RIGHT.br_blockcount,
1063                                         RIGHT.br_state)))
1064                                 goto done;
1065                 }
1066                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1067                         STARTBLOCKVAL(PREV.br_startblock));
1068                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1069                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1070                 *dnew = temp;
1071                 /* DELTA: The boundary between two in-core extents moved. */
1072                 temp = PREV.br_startoff;
1073                 temp2 = PREV.br_blockcount +
1074                         RIGHT.br_blockcount;
1075                 break;
1076
1077         case MASK(RIGHT_FILLING):
1078                 /*
1079                  * Filling in the last part of a previous delayed allocation.
1080                  * The right neighbor is not contiguous.
1081                  */
1082                 temp = PREV.br_blockcount - new->br_blockcount;
1083                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1084                 xfs_bmbt_set_blockcount(ep, temp);
1085                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1086                         XFS_DATA_FORK);
1087                 xfs_iext_insert(ifp, idx + 1, 1, new);
1088                 ip->i_df.if_lastex = idx + 1;
1089                 ip->i_d.di_nextents++;
1090                 if (cur == NULL)
1091                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1092                 else {
1093                         rval = XFS_ILOG_CORE;
1094                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1095                                         new->br_startblock, new->br_blockcount,
1096                                         &i)))
1097                                 goto done;
1098                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1099                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1100                         if ((error = xfs_bmbt_insert(cur, &i)))
1101                                 goto done;
1102                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1103                 }
1104                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1105                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1106                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1107                                 first, flist, &cur, 1, &tmp_rval,
1108                                 XFS_DATA_FORK);
1109                         rval |= tmp_rval;
1110                         if (error)
1111                                 goto done;
1112                 }
1113                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1114                         STARTBLOCKVAL(PREV.br_startblock) -
1115                         (cur ? cur->bc_private.b.allocated : 0));
1116                 ep = xfs_iext_get_ext(ifp, idx);
1117                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1118                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1119                 *dnew = temp;
1120                 /* DELTA: One in-core extent is split in two. */
1121                 temp = PREV.br_startoff;
1122                 temp2 = PREV.br_blockcount;
1123                 break;
1124
1125         case 0:
1126                 /*
1127                  * Filling in the middle part of a previous delayed allocation.
1128                  * Contiguity is impossible here.
1129                  * This case is avoided almost all the time.
1130                  */
1131                 temp = new->br_startoff - PREV.br_startoff;
1132                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1133                 xfs_bmbt_set_blockcount(ep, temp);
1134                 r[0] = *new;
1135                 r[1].br_state = PREV.br_state;
1136                 r[1].br_startblock = 0;
1137                 r[1].br_startoff = new_endoff;
1138                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1139                 r[1].br_blockcount = temp2;
1140                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1141                         XFS_DATA_FORK);
1142                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1143                 ip->i_df.if_lastex = idx + 1;
1144                 ip->i_d.di_nextents++;
1145                 if (cur == NULL)
1146                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1147                 else {
1148                         rval = XFS_ILOG_CORE;
1149                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1150                                         new->br_startblock, new->br_blockcount,
1151                                         &i)))
1152                                 goto done;
1153                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1154                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1155                         if ((error = xfs_bmbt_insert(cur, &i)))
1156                                 goto done;
1157                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1158                 }
1159                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1160                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1161                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1162                                         first, flist, &cur, 1, &tmp_rval,
1163                                         XFS_DATA_FORK);
1164                         rval |= tmp_rval;
1165                         if (error)
1166                                 goto done;
1167                 }
1168                 temp = xfs_bmap_worst_indlen(ip, temp);
1169                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1170                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1171                         (cur ? cur->bc_private.b.allocated : 0));
1172                 if (diff > 0 &&
1173                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1174                         /*
1175                          * Ick gross gag me with a spoon.
1176                          */
1177                         ASSERT(0);      /* want to see if this ever happens! */
1178                         while (diff > 0) {
1179                                 if (temp) {
1180                                         temp--;
1181                                         diff--;
1182                                         if (!diff ||
1183                                             !xfs_mod_incore_sb(ip->i_mount,
1184                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1185                                                 break;
1186                                 }
1187                                 if (temp2) {
1188                                         temp2--;
1189                                         diff--;
1190                                         if (!diff ||
1191                                             !xfs_mod_incore_sb(ip->i_mount,
1192                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1193                                                 break;
1194                                 }
1195                         }
1196                 }
1197                 ep = xfs_iext_get_ext(ifp, idx);
1198                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1199                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1200                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1201                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1202                         NULLSTARTBLOCK((int)temp2));
1203                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1204                 *dnew = temp + temp2;
1205                 /* DELTA: One in-core extent is split in three. */
1206                 temp = PREV.br_startoff;
1207                 temp2 = PREV.br_blockcount;
1208                 break;
1209
1210         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1211         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1212         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1213         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1214         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1215         case MASK(LEFT_CONTIG):
1216         case MASK(RIGHT_CONTIG):
1217                 /*
1218                  * These cases are all impossible.
1219                  */
1220                 ASSERT(0);
1221         }
1222         *curp = cur;
1223         if (delta) {
1224                 temp2 += temp;
1225                 if (delta->xed_startoff > temp)
1226                         delta->xed_startoff = temp;
1227                 if (delta->xed_blockcount < temp2)
1228                         delta->xed_blockcount = temp2;
1229         }
1230 done:
1231         *logflagsp = rval;
1232         return error;
1233 #undef  LEFT
1234 #undef  RIGHT
1235 #undef  PREV
1236 #undef  MASK
1237 #undef  MASK2
1238 #undef  MASK3
1239 #undef  MASK4
1240 #undef  STATE_SET
1241 #undef  STATE_TEST
1242 #undef  STATE_SET_TEST
1243 #undef  SWITCH_STATE
1244 }
1245
1246 /*
1247  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1248  * allocation to a real allocation or vice versa.
1249  */
1250 STATIC int                              /* error */
1251 xfs_bmap_add_extent_unwritten_real(
1252         xfs_inode_t             *ip,    /* incore inode pointer */
1253         xfs_extnum_t            idx,    /* extent number to update/insert */
1254         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1255         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1256         int                     *logflagsp, /* inode logging flags */
1257         xfs_extdelta_t          *delta) /* Change made to incore extents */
1258 {
1259         xfs_btree_cur_t         *cur;   /* btree cursor */
1260         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1261         int                     error;  /* error return value */
1262         int                     i;      /* temp state */
1263         xfs_ifork_t             *ifp;   /* inode fork pointer */
1264         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1265         xfs_exntst_t            newext; /* new extent state */
1266         xfs_exntst_t            oldext; /* old extent state */
1267         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1268                                         /* left is 0, right is 1, prev is 2 */
1269         int                     rval=0; /* return value (logging flags) */
1270         int                     state = 0;/* state bits, accessed thru macros */
1271         xfs_filblks_t           temp=0;
1272         xfs_filblks_t           temp2=0;
1273         enum {                          /* bit number definitions for state */
1274                 LEFT_CONTIG,    RIGHT_CONTIG,
1275                 LEFT_FILLING,   RIGHT_FILLING,
1276                 LEFT_DELAY,     RIGHT_DELAY,
1277                 LEFT_VALID,     RIGHT_VALID
1278         };
1279
1280 #define LEFT            r[0]
1281 #define RIGHT           r[1]
1282 #define PREV            r[2]
1283 #define MASK(b)         (1 << (b))
1284 #define MASK2(a,b)      (MASK(a) | MASK(b))
1285 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1286 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1287 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1288 #define STATE_TEST(b)   (state & MASK(b))
1289 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1290                                        ((state &= ~MASK(b)), 0))
1291 #define SWITCH_STATE            \
1292         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1293
1294         /*
1295          * Set up a bunch of variables to make the tests simpler.
1296          */
1297         error = 0;
1298         cur = *curp;
1299         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1300         ep = xfs_iext_get_ext(ifp, idx);
1301         xfs_bmbt_get_all(ep, &PREV);
1302         newext = new->br_state;
1303         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1304                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1305         ASSERT(PREV.br_state == oldext);
1306         new_endoff = new->br_startoff + new->br_blockcount;
1307         ASSERT(PREV.br_startoff <= new->br_startoff);
1308         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1309         /*
1310          * Set flags determining what part of the previous oldext allocation
1311          * extent is being replaced by a newext allocation.
1312          */
1313         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1314         STATE_SET(RIGHT_FILLING,
1315                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1316         /*
1317          * Check and set flags if this segment has a left neighbor.
1318          * Don't set contiguous if the combined extent would be too large.
1319          */
1320         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1321                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1322                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1323         }
1324         STATE_SET(LEFT_CONTIG,
1325                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1326                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1327                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1328                 LEFT.br_state == newext &&
1329                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1330         /*
1331          * Check and set flags if this segment has a right neighbor.
1332          * Don't set contiguous if the combined extent would be too large.
1333          * Also check for all-three-contiguous being too large.
1334          */
1335         if (STATE_SET_TEST(RIGHT_VALID,
1336                         idx <
1337                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1338                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1339                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1340         }
1341         STATE_SET(RIGHT_CONTIG,
1342                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1343                 new_endoff == RIGHT.br_startoff &&
1344                 new->br_startblock + new->br_blockcount ==
1345                     RIGHT.br_startblock &&
1346                 newext == RIGHT.br_state &&
1347                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1348                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1349                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1350                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1351                      <= MAXEXTLEN));
1352         /*
1353          * Switch out based on the FILLING and CONTIG state bits.
1354          */
1355         switch (SWITCH_STATE) {
1356
1357         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1358                 /*
1359                  * Setting all of a previous oldext extent to newext.
1360                  * The left and right neighbors are both contiguous with new.
1361                  */
1362                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1363                         XFS_DATA_FORK);
1364                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1365                         LEFT.br_blockcount + PREV.br_blockcount +
1366                         RIGHT.br_blockcount);
1367                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1368                         XFS_DATA_FORK);
1369                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
1370                 xfs_iext_remove(ifp, idx, 2);
1371                 ip->i_df.if_lastex = idx - 1;
1372                 ip->i_d.di_nextents -= 2;
1373                 if (cur == NULL)
1374                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1375                 else {
1376                         rval = XFS_ILOG_CORE;
1377                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1378                                         RIGHT.br_startblock,
1379                                         RIGHT.br_blockcount, &i)))
1380                                 goto done;
1381                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1382                         if ((error = xfs_bmbt_delete(cur, &i)))
1383                                 goto done;
1384                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1385                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1386                                 goto done;
1387                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1388                         if ((error = xfs_bmbt_delete(cur, &i)))
1389                                 goto done;
1390                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1391                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1392                                 goto done;
1393                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1394                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1395                                 LEFT.br_startblock,
1396                                 LEFT.br_blockcount + PREV.br_blockcount +
1397                                 RIGHT.br_blockcount, LEFT.br_state)))
1398                                 goto done;
1399                 }
1400                 /* DELTA: Three in-core extents are replaced by one. */
1401                 temp = LEFT.br_startoff;
1402                 temp2 = LEFT.br_blockcount +
1403                         PREV.br_blockcount +
1404                         RIGHT.br_blockcount;
1405                 break;
1406
1407         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1408                 /*
1409                  * Setting all of a previous oldext extent to newext.
1410                  * The left neighbor is contiguous, the right is not.
1411                  */
1412                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
1413                         XFS_DATA_FORK);
1414                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1415                         LEFT.br_blockcount + PREV.br_blockcount);
1416                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
1417                         XFS_DATA_FORK);
1418                 ip->i_df.if_lastex = idx - 1;
1419                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
1420                 xfs_iext_remove(ifp, idx, 1);
1421                 ip->i_d.di_nextents--;
1422                 if (cur == NULL)
1423                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1424                 else {
1425                         rval = XFS_ILOG_CORE;
1426                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1427                                         PREV.br_startblock, PREV.br_blockcount,
1428                                         &i)))
1429                                 goto done;
1430                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1431                         if ((error = xfs_bmbt_delete(cur, &i)))
1432                                 goto done;
1433                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1434                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1435                                 goto done;
1436                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1437                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1438                                 LEFT.br_startblock,
1439                                 LEFT.br_blockcount + PREV.br_blockcount,
1440                                 LEFT.br_state)))
1441                                 goto done;
1442                 }
1443                 /* DELTA: Two in-core extents are replaced by one. */
1444                 temp = LEFT.br_startoff;
1445                 temp2 = LEFT.br_blockcount +
1446                         PREV.br_blockcount;
1447                 break;
1448
1449         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1450                 /*
1451                  * Setting all of a previous oldext extent to newext.
1452                  * The right neighbor is contiguous, the left is not.
1453                  */
1454                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx,
1455                         XFS_DATA_FORK);
1456                 xfs_bmbt_set_blockcount(ep,
1457                         PREV.br_blockcount + RIGHT.br_blockcount);
1458                 xfs_bmbt_set_state(ep, newext);
1459                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx,
1460                         XFS_DATA_FORK);
1461                 ip->i_df.if_lastex = idx;
1462                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
1463                 xfs_iext_remove(ifp, idx + 1, 1);
1464                 ip->i_d.di_nextents--;
1465                 if (cur == NULL)
1466                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1467                 else {
1468                         rval = XFS_ILOG_CORE;
1469                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1470                                         RIGHT.br_startblock,
1471                                         RIGHT.br_blockcount, &i)))
1472                                 goto done;
1473                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1474                         if ((error = xfs_bmbt_delete(cur, &i)))
1475                                 goto done;
1476                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1477                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1478                                 goto done;
1479                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1480                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1481                                 new->br_startblock,
1482                                 new->br_blockcount + RIGHT.br_blockcount,
1483                                 newext)))
1484                                 goto done;
1485                 }
1486                 /* DELTA: Two in-core extents are replaced by one. */
1487                 temp = PREV.br_startoff;
1488                 temp2 = PREV.br_blockcount +
1489                         RIGHT.br_blockcount;
1490                 break;
1491
1492         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1493                 /*
1494                  * Setting all of a previous oldext extent to newext.
1495                  * Neither the left nor right neighbors are contiguous with
1496                  * the new one.
1497                  */
1498                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx,
1499                         XFS_DATA_FORK);
1500                 xfs_bmbt_set_state(ep, newext);
1501                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx,
1502                         XFS_DATA_FORK);
1503                 ip->i_df.if_lastex = idx;
1504                 if (cur == NULL)
1505                         rval = XFS_ILOG_DEXT;
1506                 else {
1507                         rval = 0;
1508                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1509                                         new->br_startblock, new->br_blockcount,
1510                                         &i)))
1511                                 goto done;
1512                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1513                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1514                                 new->br_startblock, new->br_blockcount,
1515                                 newext)))
1516                                 goto done;
1517                 }
1518                 /* DELTA: The in-core extent described by new changed type. */
1519                 temp = new->br_startoff;
1520                 temp2 = new->br_blockcount;
1521                 break;
1522
1523         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1524                 /*
1525                  * Setting the first part of a previous oldext extent to newext.
1526                  * The left neighbor is contiguous.
1527                  */
1528                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1,
1529                         XFS_DATA_FORK);
1530                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1531                         LEFT.br_blockcount + new->br_blockcount);
1532                 xfs_bmbt_set_startoff(ep,
1533                         PREV.br_startoff + new->br_blockcount);
1534                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1,
1535                         XFS_DATA_FORK);
1536                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx,
1537                         XFS_DATA_FORK);
1538                 xfs_bmbt_set_startblock(ep,
1539                         new->br_startblock + new->br_blockcount);
1540                 xfs_bmbt_set_blockcount(ep,
1541                         PREV.br_blockcount - new->br_blockcount);
1542                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx,
1543                         XFS_DATA_FORK);
1544                 ip->i_df.if_lastex = idx - 1;
1545                 if (cur == NULL)
1546                         rval = XFS_ILOG_DEXT;
1547                 else {
1548                         rval = 0;
1549                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1550                                         PREV.br_startblock, PREV.br_blockcount,
1551                                         &i)))
1552                                 goto done;
1553                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1554                         if ((error = xfs_bmbt_update(cur,
1555                                 PREV.br_startoff + new->br_blockcount,
1556                                 PREV.br_startblock + new->br_blockcount,
1557                                 PREV.br_blockcount - new->br_blockcount,
1558                                 oldext)))
1559                                 goto done;
1560                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1561                                 goto done;
1562                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1563                                 LEFT.br_startblock,
1564                                 LEFT.br_blockcount + new->br_blockcount,
1565                                 LEFT.br_state))
1566                                 goto done;
1567                 }
1568                 /* DELTA: The boundary between two in-core extents moved. */
1569                 temp = LEFT.br_startoff;
1570                 temp2 = LEFT.br_blockcount +
1571                         PREV.br_blockcount;
1572                 break;
1573
1574         case MASK(LEFT_FILLING):
1575                 /*
1576                  * Setting the first part of a previous oldext extent to newext.
1577                  * The left neighbor is not contiguous.
1578                  */
1579                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1580                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1581                 xfs_bmbt_set_startoff(ep, new_endoff);
1582                 xfs_bmbt_set_blockcount(ep,
1583                         PREV.br_blockcount - new->br_blockcount);
1584                 xfs_bmbt_set_startblock(ep,
1585                         new->br_startblock + new->br_blockcount);
1586                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1587                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1588                         XFS_DATA_FORK);
1589                 xfs_iext_insert(ifp, idx, 1, new);
1590                 ip->i_df.if_lastex = idx;
1591                 ip->i_d.di_nextents++;
1592                 if (cur == NULL)
1593                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1594                 else {
1595                         rval = XFS_ILOG_CORE;
1596                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1597                                         PREV.br_startblock, PREV.br_blockcount,
1598                                         &i)))
1599                                 goto done;
1600                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1601                         if ((error = xfs_bmbt_update(cur,
1602                                 PREV.br_startoff + new->br_blockcount,
1603                                 PREV.br_startblock + new->br_blockcount,
1604                                 PREV.br_blockcount - new->br_blockcount,
1605                                 oldext)))
1606                                 goto done;
1607                         cur->bc_rec.b = *new;
1608                         if ((error = xfs_bmbt_insert(cur, &i)))
1609                                 goto done;
1610                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1611                 }
1612                 /* DELTA: One in-core extent is split in two. */
1613                 temp = PREV.br_startoff;
1614                 temp2 = PREV.br_blockcount;
1615                 break;
1616
1617         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1618                 /*
1619                  * Setting the last part of a previous oldext extent to newext.
1620                  * The right neighbor is contiguous with the new allocation.
1621                  */
1622                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx,
1623                         XFS_DATA_FORK);
1624                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1,
1625                         XFS_DATA_FORK);
1626                 xfs_bmbt_set_blockcount(ep,
1627                         PREV.br_blockcount - new->br_blockcount);
1628                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx,
1629                         XFS_DATA_FORK);
1630                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1631                         new->br_startoff, new->br_startblock,
1632                         new->br_blockcount + RIGHT.br_blockcount, newext);
1633                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1,
1634                         XFS_DATA_FORK);
1635                 ip->i_df.if_lastex = idx + 1;
1636                 if (cur == NULL)
1637                         rval = XFS_ILOG_DEXT;
1638                 else {
1639                         rval = 0;
1640                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1641                                         PREV.br_startblock,
1642                                         PREV.br_blockcount, &i)))
1643                                 goto done;
1644                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1645                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1646                                 PREV.br_startblock,
1647                                 PREV.br_blockcount - new->br_blockcount,
1648                                 oldext)))
1649                                 goto done;
1650                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1651                                 goto done;
1652                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1653                                 new->br_startblock,
1654                                 new->br_blockcount + RIGHT.br_blockcount,
1655                                 newext)))
1656                                 goto done;
1657                 }
1658                 /* DELTA: The boundary between two in-core extents moved. */
1659                 temp = PREV.br_startoff;
1660                 temp2 = PREV.br_blockcount +
1661                         RIGHT.br_blockcount;
1662                 break;
1663
1664         case MASK(RIGHT_FILLING):
1665                 /*
1666                  * Setting the last part of a previous oldext extent to newext.
1667                  * The right neighbor is not contiguous.
1668                  */
1669                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1670                 xfs_bmbt_set_blockcount(ep,
1671                         PREV.br_blockcount - new->br_blockcount);
1672                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1673                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1674                         XFS_DATA_FORK);
1675                 xfs_iext_insert(ifp, idx + 1, 1, new);
1676                 ip->i_df.if_lastex = idx + 1;
1677                 ip->i_d.di_nextents++;
1678                 if (cur == NULL)
1679                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1680                 else {
1681                         rval = XFS_ILOG_CORE;
1682                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1683                                         PREV.br_startblock, PREV.br_blockcount,
1684                                         &i)))
1685                                 goto done;
1686                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1687                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1688                                 PREV.br_startblock,
1689                                 PREV.br_blockcount - new->br_blockcount,
1690                                 oldext)))
1691                                 goto done;
1692                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1693                                         new->br_startblock, new->br_blockcount,
1694                                         &i)))
1695                                 goto done;
1696                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1697                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1698                         if ((error = xfs_bmbt_insert(cur, &i)))
1699                                 goto done;
1700                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1701                 }
1702                 /* DELTA: One in-core extent is split in two. */
1703                 temp = PREV.br_startoff;
1704                 temp2 = PREV.br_blockcount;
1705                 break;
1706
1707         case 0:
1708                 /*
1709                  * Setting the middle part of a previous oldext extent to
1710                  * newext.  Contiguity is impossible here.
1711                  * One extent becomes three extents.
1712                  */
1713                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1714                 xfs_bmbt_set_blockcount(ep,
1715                         new->br_startoff - PREV.br_startoff);
1716                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1717                 r[0] = *new;
1718                 r[1].br_startoff = new_endoff;
1719                 r[1].br_blockcount =
1720                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1721                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1722                 r[1].br_state = oldext;
1723                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1724                         XFS_DATA_FORK);
1725                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1726                 ip->i_df.if_lastex = idx + 1;
1727                 ip->i_d.di_nextents += 2;
1728                 if (cur == NULL)
1729                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1730                 else {
1731                         rval = XFS_ILOG_CORE;
1732                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1733                                         PREV.br_startblock, PREV.br_blockcount,
1734                                         &i)))
1735                                 goto done;
1736                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1737                         /* new right extent - oldext */
1738                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1739                                 r[1].br_startblock, r[1].br_blockcount,
1740                                 r[1].br_state)))
1741                                 goto done;
1742                         /* new left extent - oldext */
1743                         PREV.br_blockcount =
1744                                 new->br_startoff - PREV.br_startoff;
1745                         cur->bc_rec.b = PREV;
1746                         if ((error = xfs_bmbt_insert(cur, &i)))
1747                                 goto done;
1748                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1749                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1750                                 goto done;
1751                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1752                         /* new middle extent - newext */
1753                         cur->bc_rec.b = *new;
1754                         if ((error = xfs_bmbt_insert(cur, &i)))
1755                                 goto done;
1756                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1757                 }
1758                 /* DELTA: One in-core extent is split in three. */
1759                 temp = PREV.br_startoff;
1760                 temp2 = PREV.br_blockcount;
1761                 break;
1762
1763         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1764         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1765         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1766         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1767         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1768         case MASK(LEFT_CONTIG):
1769         case MASK(RIGHT_CONTIG):
1770                 /*
1771                  * These cases are all impossible.
1772                  */
1773                 ASSERT(0);
1774         }
1775         *curp = cur;
1776         if (delta) {
1777                 temp2 += temp;
1778                 if (delta->xed_startoff > temp)
1779                         delta->xed_startoff = temp;
1780                 if (delta->xed_blockcount < temp2)
1781                         delta->xed_blockcount = temp2;
1782         }
1783 done:
1784         *logflagsp = rval;
1785         return error;
1786 #undef  LEFT
1787 #undef  RIGHT
1788 #undef  PREV
1789 #undef  MASK
1790 #undef  MASK2
1791 #undef  MASK3
1792 #undef  MASK4
1793 #undef  STATE_SET
1794 #undef  STATE_TEST
1795 #undef  STATE_SET_TEST
1796 #undef  SWITCH_STATE
1797 }
1798
1799 /*
1800  * Called by xfs_bmap_add_extent to handle cases converting a hole
1801  * to a delayed allocation.
1802  */
1803 /*ARGSUSED*/
1804 STATIC int                              /* error */
1805 xfs_bmap_add_extent_hole_delay(
1806         xfs_inode_t             *ip,    /* incore inode pointer */
1807         xfs_extnum_t            idx,    /* extent number to update/insert */
1808         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1809         int                     *logflagsp, /* inode logging flags */
1810         xfs_extdelta_t          *delta, /* Change made to incore extents */
1811         int                     rsvd)           /* OK to allocate reserved blocks */
1812 {
1813         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1814         xfs_ifork_t             *ifp;   /* inode fork pointer */
1815         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1816         xfs_filblks_t           newlen=0;       /* new indirect size */
1817         xfs_filblks_t           oldlen=0;       /* old indirect size */
1818         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1819         int                     state;  /* state bits, accessed thru macros */
1820         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1821         xfs_filblks_t           temp2=0;
1822         enum {                          /* bit number definitions for state */
1823                 LEFT_CONTIG,    RIGHT_CONTIG,
1824                 LEFT_DELAY,     RIGHT_DELAY,
1825                 LEFT_VALID,     RIGHT_VALID
1826         };
1827
1828 #define MASK(b)                 (1 << (b))
1829 #define MASK2(a,b)              (MASK(a) | MASK(b))
1830 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1831 #define STATE_TEST(b)           (state & MASK(b))
1832 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1833                                        ((state &= ~MASK(b)), 0))
1834 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1835
1836         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1837         ep = xfs_iext_get_ext(ifp, idx);
1838         state = 0;
1839         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1840         /*
1841          * Check and set flags if this segment has a left neighbor
1842          */
1843         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1844                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1845                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1846         }
1847         /*
1848          * Check and set flags if the current (right) segment exists.
1849          * If it doesn't exist, we're converting the hole at end-of-file.
1850          */
1851         if (STATE_SET_TEST(RIGHT_VALID,
1852                            idx <
1853                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1854                 xfs_bmbt_get_all(ep, &right);
1855                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1856         }
1857         /*
1858          * Set contiguity flags on the left and right neighbors.
1859          * Don't let extents get too large, even if the pieces are contiguous.
1860          */
1861         STATE_SET(LEFT_CONTIG,
1862                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1863                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1864                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1865         STATE_SET(RIGHT_CONTIG,
1866                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1867                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1868                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1869                 (!STATE_TEST(LEFT_CONTIG) ||
1870                  (left.br_blockcount + new->br_blockcount +
1871                      right.br_blockcount <= MAXEXTLEN)));
1872         /*
1873          * Switch out based on the contiguity flags.
1874          */
1875         switch (SWITCH_STATE) {
1876
1877         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1878                 /*
1879                  * New allocation is contiguous with delayed allocations
1880                  * on the left and on the right.
1881                  * Merge all three into a single extent record.
1882                  */
1883                 temp = left.br_blockcount + new->br_blockcount +
1884                         right.br_blockcount;
1885                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
1886                         XFS_DATA_FORK);
1887                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1888                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1889                         STARTBLOCKVAL(new->br_startblock) +
1890                         STARTBLOCKVAL(right.br_startblock);
1891                 newlen = xfs_bmap_worst_indlen(ip, temp);
1892                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1893                         NULLSTARTBLOCK((int)newlen));
1894                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
1895                         XFS_DATA_FORK);
1896                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, XFS_DATA_FORK);
1897                 xfs_iext_remove(ifp, idx, 1);
1898                 ip->i_df.if_lastex = idx - 1;
1899                 /* DELTA: Two in-core extents were replaced by one. */
1900                 temp2 = temp;
1901                 temp = left.br_startoff;
1902                 break;
1903
1904         case MASK(LEFT_CONTIG):
1905                 /*
1906                  * New allocation is contiguous with a delayed allocation
1907                  * on the left.
1908                  * Merge the new allocation with the left neighbor.
1909                  */
1910                 temp = left.br_blockcount + new->br_blockcount;
1911                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1,
1912                         XFS_DATA_FORK);
1913                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1914                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1915                         STARTBLOCKVAL(new->br_startblock);
1916                 newlen = xfs_bmap_worst_indlen(ip, temp);
1917                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1918                         NULLSTARTBLOCK((int)newlen));
1919                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1,
1920                         XFS_DATA_FORK);
1921                 ip->i_df.if_lastex = idx - 1;
1922                 /* DELTA: One in-core extent grew into a hole. */
1923                 temp2 = temp;
1924                 temp = left.br_startoff;
1925                 break;
1926
1927         case MASK(RIGHT_CONTIG):
1928                 /*
1929                  * New allocation is contiguous with a delayed allocation
1930                  * on the right.
1931                  * Merge the new allocation with the right neighbor.
1932                  */
1933                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1934                 temp = new->br_blockcount + right.br_blockcount;
1935                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1936                         STARTBLOCKVAL(right.br_startblock);
1937                 newlen = xfs_bmap_worst_indlen(ip, temp);
1938                 xfs_bmbt_set_allf(ep, new->br_startoff,
1939                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1940                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1941                 ip->i_df.if_lastex = idx;
1942                 /* DELTA: One in-core extent grew into a hole. */
1943                 temp2 = temp;
1944                 temp = new->br_startoff;
1945                 break;
1946
1947         case 0:
1948                 /*
1949                  * New allocation is not contiguous with another
1950                  * delayed allocation.
1951                  * Insert a new entry.
1952                  */
1953                 oldlen = newlen = 0;
1954                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL,
1955                         XFS_DATA_FORK);
1956                 xfs_iext_insert(ifp, idx, 1, new);
1957                 ip->i_df.if_lastex = idx;
1958                 /* DELTA: A new in-core extent was added in a hole. */
1959                 temp2 = new->br_blockcount;
1960                 temp = new->br_startoff;
1961                 break;
1962         }
1963         if (oldlen != newlen) {
1964                 ASSERT(oldlen > newlen);
1965                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1966                         (int64_t)(oldlen - newlen), rsvd);
1967                 /*
1968                  * Nothing to do for disk quota accounting here.
1969                  */
1970         }
1971         if (delta) {
1972                 temp2 += temp;
1973                 if (delta->xed_startoff > temp)
1974                         delta->xed_startoff = temp;
1975                 if (delta->xed_blockcount < temp2)
1976                         delta->xed_blockcount = temp2;
1977         }
1978         *logflagsp = 0;
1979         return 0;
1980 #undef  MASK
1981 #undef  MASK2
1982 #undef  STATE_SET
1983 #undef  STATE_TEST
1984 #undef  STATE_SET_TEST
1985 #undef  SWITCH_STATE
1986 }
1987
1988 /*
1989  * Called by xfs_bmap_add_extent to handle cases converting a hole
1990  * to a real allocation.
1991  */
1992 STATIC int                              /* error */
1993 xfs_bmap_add_extent_hole_real(
1994         xfs_inode_t             *ip,    /* incore inode pointer */
1995         xfs_extnum_t            idx,    /* extent number to update/insert */
1996         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1997         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1998         int                     *logflagsp, /* inode logging flags */
1999         xfs_extdelta_t          *delta, /* Change made to incore extents */
2000         int                     whichfork) /* data or attr fork */
2001 {
2002         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
2003         int                     error;  /* error return value */
2004         int                     i;      /* temp state */
2005         xfs_ifork_t             *ifp;   /* inode fork pointer */
2006         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2007         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2008         int                     rval=0; /* return value (logging flags) */
2009         int                     state;  /* state bits, accessed thru macros */
2010         xfs_filblks_t           temp=0;
2011         xfs_filblks_t           temp2=0;
2012         enum {                          /* bit number definitions for state */
2013                 LEFT_CONTIG,    RIGHT_CONTIG,
2014                 LEFT_DELAY,     RIGHT_DELAY,
2015                 LEFT_VALID,     RIGHT_VALID
2016         };
2017
2018 #define MASK(b)                 (1 << (b))
2019 #define MASK2(a,b)              (MASK(a) | MASK(b))
2020 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2021 #define STATE_TEST(b)           (state & MASK(b))
2022 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2023                                        ((state &= ~MASK(b)), 0))
2024 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2025
2026         ifp = XFS_IFORK_PTR(ip, whichfork);
2027         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2028         ep = xfs_iext_get_ext(ifp, idx);
2029         state = 0;
2030         /*
2031          * Check and set flags if this segment has a left neighbor.
2032          */
2033         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2034                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2035                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2036         }
2037         /*
2038          * Check and set flags if this segment has a current value.
2039          * Not true if we're inserting into the "hole" at eof.
2040          */
2041         if (STATE_SET_TEST(RIGHT_VALID,
2042                            idx <
2043                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2044                 xfs_bmbt_get_all(ep, &right);
2045                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2046         }
2047         /*
2048          * We're inserting a real allocation between "left" and "right".
2049          * Set the contiguity flags.  Don't let extents get too large.
2050          */
2051         STATE_SET(LEFT_CONTIG,
2052                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2053                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2054                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2055                 left.br_state == new->br_state &&
2056                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2057         STATE_SET(RIGHT_CONTIG,
2058                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2059                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2060                 new->br_startblock + new->br_blockcount ==
2061                     right.br_startblock &&
2062                 new->br_state == right.br_state &&
2063                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2064                 (!STATE_TEST(LEFT_CONTIG) ||
2065                  left.br_blockcount + new->br_blockcount +
2066                      right.br_blockcount <= MAXEXTLEN));
2067
2068         error = 0;
2069         /*
2070          * Select which case we're in here, and implement it.
2071          */
2072         switch (SWITCH_STATE) {
2073
2074         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2075                 /*
2076                  * New allocation is contiguous with real allocations on the
2077                  * left and on the right.
2078                  * Merge all three into a single extent record.
2079                  */
2080                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
2081                         whichfork);
2082                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2083                         left.br_blockcount + new->br_blockcount +
2084                         right.br_blockcount);
2085                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
2086                         whichfork);
2087                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, whichfork);
2088                 xfs_iext_remove(ifp, idx, 1);
2089                 ifp->if_lastex = idx - 1;
2090                 XFS_IFORK_NEXT_SET(ip, whichfork,
2091                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2092                 if (cur == NULL) {
2093                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2094                 } else {
2095                         rval = XFS_ILOG_CORE;
2096                         if ((error = xfs_bmbt_lookup_eq(cur,
2097                                         right.br_startoff,
2098                                         right.br_startblock,
2099                                         right.br_blockcount, &i)))
2100                                 goto done;
2101                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2102                         if ((error = xfs_bmbt_delete(cur, &i)))
2103                                 goto done;
2104                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2105                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2106                                 goto done;
2107                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2108                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2109                                         left.br_startblock,
2110                                         left.br_blockcount +
2111                                                 new->br_blockcount +
2112                                                 right.br_blockcount,
2113                                         left.br_state)))
2114                                 goto done;
2115                 }
2116                 /* DELTA: Two in-core extents were replaced by one. */
2117                 temp = left.br_startoff;
2118                 temp2 = left.br_blockcount +
2119                         new->br_blockcount +
2120                         right.br_blockcount;
2121                 break;
2122
2123         case MASK(LEFT_CONTIG):
2124                 /*
2125                  * New allocation is contiguous with a real allocation
2126                  * on the left.
2127                  * Merge the new allocation with the left neighbor.
2128                  */
2129                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1, whichfork);
2130                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2131                         left.br_blockcount + new->br_blockcount);
2132                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1, whichfork);
2133                 ifp->if_lastex = idx - 1;
2134                 if (cur == NULL) {
2135                         rval = XFS_ILOG_FEXT(whichfork);
2136                 } else {
2137                         rval = 0;
2138                         if ((error = xfs_bmbt_lookup_eq(cur,
2139                                         left.br_startoff,
2140                                         left.br_startblock,
2141                                         left.br_blockcount, &i)))
2142                                 goto done;
2143                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2144                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2145                                         left.br_startblock,
2146                                         left.br_blockcount +
2147                                                 new->br_blockcount,
2148                                         left.br_state)))
2149                                 goto done;
2150                 }
2151                 /* DELTA: One in-core extent grew. */
2152                 temp = left.br_startoff;
2153                 temp2 = left.br_blockcount +
2154                         new->br_blockcount;
2155                 break;
2156
2157         case MASK(RIGHT_CONTIG):
2158                 /*
2159                  * New allocation is contiguous with a real allocation
2160                  * on the right.
2161                  * Merge the new allocation with the right neighbor.
2162                  */
2163                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, whichfork);
2164                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2165                         new->br_blockcount + right.br_blockcount,
2166                         right.br_state);
2167                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, whichfork);
2168                 ifp->if_lastex = idx;
2169                 if (cur == NULL) {
2170                         rval = XFS_ILOG_FEXT(whichfork);
2171                 } else {
2172                         rval = 0;
2173                         if ((error = xfs_bmbt_lookup_eq(cur,
2174                                         right.br_startoff,
2175                                         right.br_startblock,
2176                                         right.br_blockcount, &i)))
2177                                 goto done;
2178                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2179                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2180                                         new->br_startblock,
2181                                         new->br_blockcount +
2182                                                 right.br_blockcount,
2183                                         right.br_state)))
2184                                 goto done;
2185                 }
2186                 /* DELTA: One in-core extent grew. */
2187                 temp = new->br_startoff;
2188                 temp2 = new->br_blockcount +
2189                         right.br_blockcount;
2190                 break;
2191
2192         case 0:
2193                 /*
2194                  * New allocation is not contiguous with another
2195                  * real allocation.
2196                  * Insert a new entry.
2197                  */
2198                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL, whichfork);
2199                 xfs_iext_insert(ifp, idx, 1, new);
2200                 ifp->if_lastex = idx;
2201                 XFS_IFORK_NEXT_SET(ip, whichfork,
2202                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2203                 if (cur == NULL) {
2204                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2205                 } else {
2206                         rval = XFS_ILOG_CORE;
2207                         if ((error = xfs_bmbt_lookup_eq(cur,
2208                                         new->br_startoff,
2209                                         new->br_startblock,
2210                                         new->br_blockcount, &i)))
2211                                 goto done;
2212                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2213                         cur->bc_rec.b.br_state = new->br_state;
2214                         if ((error = xfs_bmbt_insert(cur, &i)))
2215                                 goto done;
2216                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2217                 }
2218                 /* DELTA: A new extent was added in a hole. */
2219                 temp = new->br_startoff;
2220                 temp2 = new->br_blockcount;
2221                 break;
2222         }
2223         if (delta) {
2224                 temp2 += temp;
2225                 if (delta->xed_startoff > temp)
2226                         delta->xed_startoff = temp;
2227                 if (delta->xed_blockcount < temp2)
2228                         delta->xed_blockcount = temp2;
2229         }
2230 done:
2231         *logflagsp = rval;
2232         return error;
2233 #undef  MASK
2234 #undef  MASK2
2235 #undef  STATE_SET
2236 #undef  STATE_TEST
2237 #undef  STATE_SET_TEST
2238 #undef  SWITCH_STATE
2239 }
2240
2241 /*
2242  * Adjust the size of the new extent based on di_extsize and rt extsize.
2243  */
2244 STATIC int
2245 xfs_bmap_extsize_align(
2246         xfs_mount_t     *mp,
2247         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2248         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2249         xfs_extlen_t    extsz,          /* align to this extent size */
2250         int             rt,             /* is this a realtime inode? */
2251         int             eof,            /* is extent at end-of-file? */
2252         int             delay,          /* creating delalloc extent? */
2253         int             convert,        /* overwriting unwritten extent? */
2254         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2255         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2256 {
2257         xfs_fileoff_t   orig_off;       /* original offset */
2258         xfs_extlen_t    orig_alen;      /* original length */
2259         xfs_fileoff_t   orig_end;       /* original off+len */
2260         xfs_fileoff_t   nexto;          /* next file offset */
2261         xfs_fileoff_t   prevo;          /* previous file offset */
2262         xfs_fileoff_t   align_off;      /* temp for offset */
2263         xfs_extlen_t    align_alen;     /* temp for length */
2264         xfs_extlen_t    temp;           /* temp for calculations */
2265
2266         if (convert)
2267                 return 0;
2268
2269         orig_off = align_off = *offp;
2270         orig_alen = align_alen = *lenp;
2271         orig_end = orig_off + orig_alen;
2272
2273         /*
2274          * If this request overlaps an existing extent, then don't
2275          * attempt to perform any additional alignment.
2276          */
2277         if (!delay && !eof &&
2278             (orig_off >= gotp->br_startoff) &&
2279             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2280                 return 0;
2281         }
2282
2283         /*
2284          * If the file offset is unaligned vs. the extent size
2285          * we need to align it.  This will be possible unless
2286          * the file was previously written with a kernel that didn't
2287          * perform this alignment, or if a truncate shot us in the
2288          * foot.
2289          */
2290         temp = do_mod(orig_off, extsz);
2291         if (temp) {
2292                 align_alen += temp;
2293                 align_off -= temp;
2294         }
2295         /*
2296          * Same adjustment for the end of the requested area.
2297          */
2298         if ((temp = (align_alen % extsz))) {
2299                 align_alen += extsz - temp;
2300         }
2301         /*
2302          * If the previous block overlaps with this proposed allocation
2303          * then move the start forward without adjusting the length.
2304          */
2305         if (prevp->br_startoff != NULLFILEOFF) {
2306                 if (prevp->br_startblock == HOLESTARTBLOCK)
2307                         prevo = prevp->br_startoff;
2308                 else
2309                         prevo = prevp->br_startoff + prevp->br_blockcount;
2310         } else
2311                 prevo = 0;
2312         if (align_off != orig_off && align_off < prevo)
2313                 align_off = prevo;
2314         /*
2315          * If the next block overlaps with this proposed allocation
2316          * then move the start back without adjusting the length,
2317          * but not before offset 0.
2318          * This may of course make the start overlap previous block,
2319          * and if we hit the offset 0 limit then the next block
2320          * can still overlap too.
2321          */
2322         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2323                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2324                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2325                         nexto = gotp->br_startoff + gotp->br_blockcount;
2326                 else
2327                         nexto = gotp->br_startoff;
2328         } else
2329                 nexto = NULLFILEOFF;
2330         if (!eof &&
2331             align_off + align_alen != orig_end &&
2332             align_off + align_alen > nexto)
2333                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2334         /*
2335          * If we're now overlapping the next or previous extent that
2336          * means we can't fit an extsz piece in this hole.  Just move
2337          * the start forward to the first valid spot and set
2338          * the length so we hit the end.
2339          */
2340         if (align_off != orig_off && align_off < prevo)
2341                 align_off = prevo;
2342         if (align_off + align_alen != orig_end &&
2343             align_off + align_alen > nexto &&
2344             nexto != NULLFILEOFF) {
2345                 ASSERT(nexto > prevo);
2346                 align_alen = nexto - align_off;
2347         }
2348
2349         /*
2350          * If realtime, and the result isn't a multiple of the realtime
2351          * extent size we need to remove blocks until it is.
2352          */
2353         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2354                 /*
2355                  * We're not covering the original request, or
2356                  * we won't be able to once we fix the length.
2357                  */
2358                 if (orig_off < align_off ||
2359                     orig_end > align_off + align_alen ||
2360                     align_alen - temp < orig_alen)
2361                         return XFS_ERROR(EINVAL);
2362                 /*
2363                  * Try to fix it by moving the start up.
2364                  */
2365                 if (align_off + temp <= orig_off) {
2366                         align_alen -= temp;
2367                         align_off += temp;
2368                 }
2369                 /*
2370                  * Try to fix it by moving the end in.
2371                  */
2372                 else if (align_off + align_alen - temp >= orig_end)
2373                         align_alen -= temp;
2374                 /*
2375                  * Set the start to the minimum then trim the length.
2376                  */
2377                 else {
2378                         align_alen -= orig_off - align_off;
2379                         align_off = orig_off;
2380                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2381                 }
2382                 /*
2383                  * Result doesn't cover the request, fail it.
2384                  */
2385                 if (orig_off < align_off || orig_end > align_off + align_alen)
2386                         return XFS_ERROR(EINVAL);
2387         } else {
2388                 ASSERT(orig_off >= align_off);
2389                 ASSERT(orig_end <= align_off + align_alen);
2390         }
2391
2392 #ifdef DEBUG
2393         if (!eof && gotp->br_startoff != NULLFILEOFF)
2394                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2395         if (prevp->br_startoff != NULLFILEOFF)
2396                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2397 #endif
2398
2399         *lenp = align_alen;
2400         *offp = align_off;
2401         return 0;
2402 }
2403
2404 #define XFS_ALLOC_GAP_UNITS     4
2405
2406 STATIC void
2407 xfs_bmap_adjacent(
2408         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2409 {
2410         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2411         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2412         xfs_mount_t     *mp;            /* mount point structure */
2413         int             nullfb;         /* true if ap->firstblock isn't set */
2414         int             rt;             /* true if inode is realtime */
2415
2416 #define ISVALID(x,y)    \
2417         (rt ? \
2418                 (x) < mp->m_sb.sb_rblocks : \
2419                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2420                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2421                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2422
2423         mp = ap->ip->i_mount;
2424         nullfb = ap->firstblock == NULLFSBLOCK;
2425         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2426         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2427         /*
2428          * If allocating at eof, and there's a previous real block,
2429          * try to use it's last block as our starting point.
2430          */
2431         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2432             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2433             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2434                     ap->prevp->br_startblock)) {
2435                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2436                 /*
2437                  * Adjust for the gap between prevp and us.
2438                  */
2439                 adjust = ap->off -
2440                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2441                 if (adjust &&
2442                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2443                         ap->rval += adjust;
2444         }
2445         /*
2446          * If not at eof, then compare the two neighbor blocks.
2447          * Figure out whether either one gives us a good starting point,
2448          * and pick the better one.
2449          */
2450         else if (!ap->eof) {
2451                 xfs_fsblock_t   gotbno;         /* right side block number */
2452                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2453                 xfs_fsblock_t   prevbno;        /* left side block number */
2454                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2455
2456                 /*
2457                  * If there's a previous (left) block, select a requested
2458                  * start block based on it.
2459                  */
2460                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2461                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2462                     (prevbno = ap->prevp->br_startblock +
2463                                ap->prevp->br_blockcount) &&
2464                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2465                         /*
2466                          * Calculate gap to end of previous block.
2467                          */
2468                         adjust = prevdiff = ap->off -
2469                                 (ap->prevp->br_startoff +
2470                                  ap->prevp->br_blockcount);
2471                         /*
2472                          * Figure the startblock based on the previous block's
2473                          * end and the gap size.
2474                          * Heuristic!
2475                          * If the gap is large relative to the piece we're
2476                          * allocating, or using it gives us an invalid block
2477                          * number, then just use the end of the previous block.
2478                          */
2479                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2480                             ISVALID(prevbno + prevdiff,
2481                                     ap->prevp->br_startblock))
2482                                 prevbno += adjust;
2483                         else
2484                                 prevdiff += adjust;
2485                         /*
2486                          * If the firstblock forbids it, can't use it,
2487                          * must use default.
2488                          */
2489                         if (!rt && !nullfb &&
2490                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2491                                 prevbno = NULLFSBLOCK;
2492                 }
2493                 /*
2494                  * No previous block or can't follow it, just default.
2495                  */
2496                 else
2497                         prevbno = NULLFSBLOCK;
2498                 /*
2499                  * If there's a following (right) block, select a requested
2500                  * start block based on it.
2501                  */
2502                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2503                         /*
2504                          * Calculate gap to start of next block.
2505                          */
2506                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2507                         /*
2508                          * Figure the startblock based on the next block's
2509                          * start and the gap size.
2510                          */
2511                         gotbno = ap->gotp->br_startblock;
2512                         /*
2513                          * Heuristic!
2514                          * If the gap is large relative to the piece we're
2515                          * allocating, or using it gives us an invalid block
2516                          * number, then just use the start of the next block
2517                          * offset by our length.
2518                          */
2519                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2520                             ISVALID(gotbno - gotdiff, gotbno))
2521                                 gotbno -= adjust;
2522                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2523                                 gotbno -= ap->alen;
2524                                 gotdiff += adjust - ap->alen;
2525                         } else
2526                                 gotdiff += adjust;
2527                         /*
2528                          * If the firstblock forbids it, can't use it,
2529                          * must use default.
2530                          */
2531                         if (!rt && !nullfb &&
2532                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2533                                 gotbno = NULLFSBLOCK;
2534                 }
2535                 /*
2536                  * No next block, just default.
2537                  */
2538                 else
2539                         gotbno = NULLFSBLOCK;
2540                 /*
2541                  * If both valid, pick the better one, else the only good
2542                  * one, else ap->rval is already set (to 0 or the inode block).
2543                  */
2544                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2545                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2546                 else if (prevbno != NULLFSBLOCK)
2547                         ap->rval = prevbno;
2548                 else if (gotbno != NULLFSBLOCK)
2549                         ap->rval = gotbno;
2550         }
2551 #undef ISVALID
2552 }
2553
2554 STATIC int
2555 xfs_bmap_rtalloc(
2556         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2557 {
2558         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2559         int             error;          /* error return value */
2560         xfs_mount_t     *mp;            /* mount point structure */
2561         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2562         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2563         xfs_extlen_t    align;          /* minimum allocation alignment */
2564         xfs_rtblock_t   rtb;
2565
2566         mp = ap->ip->i_mount;
2567         align = xfs_get_extsz_hint(ap->ip);
2568         prod = align / mp->m_sb.sb_rextsize;
2569         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2570                                         align, 1, ap->eof, 0,
2571                                         ap->conv, &ap->off, &ap->alen);
2572         if (error)
2573                 return error;
2574         ASSERT(ap->alen);
2575         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2576
2577         /*
2578          * If the offset & length are not perfectly aligned
2579          * then kill prod, it will just get us in trouble.
2580          */
2581         if (do_mod(ap->off, align) || ap->alen % align)
2582                 prod = 1;
2583         /*
2584          * Set ralen to be the actual requested length in rtextents.
2585          */
2586         ralen = ap->alen / mp->m_sb.sb_rextsize;
2587         /*
2588          * If the old value was close enough to MAXEXTLEN that
2589          * we rounded up to it, cut it back so it's valid again.
2590          * Note that if it's a really large request (bigger than
2591          * MAXEXTLEN), we don't hear about that number, and can't
2592          * adjust the starting point to match it.
2593          */
2594         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2595                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2596         /*
2597          * If it's an allocation to an empty file at offset 0,
2598          * pick an extent that will space things out in the rt area.
2599          */
2600         if (ap->eof && ap->off == 0) {
2601                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2602
2603                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2604                 if (error)
2605                         return error;
2606                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2607         } else {
2608                 ap->rval = 0;
2609         }
2610
2611         xfs_bmap_adjacent(ap);
2612
2613         /*
2614          * Realtime allocation, done through xfs_rtallocate_extent.
2615          */
2616         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2617         do_div(ap->rval, mp->m_sb.sb_rextsize);
2618         rtb = ap->rval;
2619         ap->alen = ralen;
2620         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2621                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2622                 return error;
2623         if (rtb == NULLFSBLOCK && prod > 1 &&
2624             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2625                                            ap->alen, &ralen, atype,
2626                                            ap->wasdel, 1, &rtb)))
2627                 return error;
2628         ap->rval = rtb;
2629         if (ap->rval != NULLFSBLOCK) {
2630                 ap->rval *= mp->m_sb.sb_rextsize;
2631                 ralen *= mp->m_sb.sb_rextsize;
2632                 ap->alen = ralen;
2633                 ap->ip->i_d.di_nblocks += ralen;
2634                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2635                 if (ap->wasdel)
2636                         ap->ip->i_delayed_blks -= ralen;
2637                 /*
2638                  * Adjust the disk quota also. This was reserved
2639                  * earlier.
2640                  */
2641                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2642                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2643                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2644         } else {
2645                 ap->alen = 0;
2646         }
2647         return 0;
2648 }
2649
2650 STATIC int
2651 xfs_bmap_btalloc(
2652         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2653 {
2654         xfs_mount_t     *mp;            /* mount point structure */
2655         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2656         xfs_extlen_t    align;          /* minimum allocation alignment */
2657         xfs_agnumber_t  ag;
2658         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2659         xfs_agnumber_t  startag;
2660         xfs_alloc_arg_t args;
2661         xfs_extlen_t    blen;
2662         xfs_extlen_t    delta;
2663         xfs_extlen_t    longest;
2664         xfs_extlen_t    need;
2665         xfs_extlen_t    nextminlen = 0;
2666         xfs_perag_t     *pag;
2667         int             nullfb;         /* true if ap->firstblock isn't set */
2668         int             isaligned;
2669         int             notinit;
2670         int             tryagain;
2671         int             error;
2672
2673         mp = ap->ip->i_mount;
2674         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2675         if (unlikely(align)) {
2676                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2677                                                 align, 0, ap->eof, 0, ap->conv,
2678                                                 &ap->off, &ap->alen);
2679                 ASSERT(!error);
2680                 ASSERT(ap->alen);
2681         }
2682         nullfb = ap->firstblock == NULLFSBLOCK;
2683         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2684         if (nullfb) {
2685                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2686                         ag = xfs_filestream_lookup_ag(ap->ip);
2687                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2688                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2689                 } else {
2690                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2691                 }
2692         } else
2693                 ap->rval = ap->firstblock;
2694
2695         xfs_bmap_adjacent(ap);
2696
2697         /*
2698          * If allowed, use ap->rval; otherwise must use firstblock since
2699          * it's in the right allocation group.
2700          */
2701         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2702                 ;
2703         else
2704                 ap->rval = ap->firstblock;
2705         /*
2706          * Normal allocation, done through xfs_alloc_vextent.
2707          */
2708         tryagain = isaligned = 0;
2709         args.tp = ap->tp;
2710         args.mp = mp;
2711         args.fsbno = ap->rval;
2712         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2713         args.firstblock = ap->firstblock;
2714         blen = 0;
2715         if (nullfb) {
2716                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2717                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2718                 else
2719                         args.type = XFS_ALLOCTYPE_START_BNO;
2720                 args.total = ap->total;
2721
2722                 /*
2723                  * Search for an allocation group with a single extent
2724                  * large enough for the request.
2725                  *
2726                  * If one isn't found, then adjust the minimum allocation
2727                  * size to the largest space found.
2728                  */
2729                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2730                 if (startag == NULLAGNUMBER)
2731                         startag = ag = 0;
2732                 notinit = 0;
2733                 down_read(&mp->m_peraglock);
2734                 while (blen < ap->alen) {
2735                         pag = &mp->m_perag[ag];
2736                         if (!pag->pagf_init &&
2737                             (error = xfs_alloc_pagf_init(mp, args.tp,
2738                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2739                                 up_read(&mp->m_peraglock);
2740                                 return error;
2741                         }
2742                         /*
2743                          * See xfs_alloc_fix_freelist...
2744                          */
2745                         if (pag->pagf_init) {
2746                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2747                                 delta = need > pag->pagf_flcount ?
2748                                         need - pag->pagf_flcount : 0;
2749                                 longest = (pag->pagf_longest > delta) ?
2750                                         (pag->pagf_longest - delta) :
2751                                         (pag->pagf_flcount > 0 ||
2752                                          pag->pagf_longest > 0);
2753                                 if (blen < longest)
2754                                         blen = longest;
2755                         } else
2756                                 notinit = 1;
2757
2758                         if (xfs_inode_is_filestream(ap->ip)) {
2759                                 if (blen >= ap->alen)
2760                                         break;
2761
2762                                 if (ap->userdata) {
2763                                         /*
2764                                          * If startag is an invalid AG, we've
2765                                          * come here once before and
2766                                          * xfs_filestream_new_ag picked the
2767                                          * best currently available.
2768                                          *
2769                                          * Don't continue looping, since we
2770                                          * could loop forever.
2771                                          */
2772                                         if (startag == NULLAGNUMBER)
2773                                                 break;
2774
2775                                         error = xfs_filestream_new_ag(ap, &ag);
2776                                         if (error) {
2777                                                 up_read(&mp->m_peraglock);
2778                                                 return error;
2779                                         }
2780
2781                                         /* loop again to set 'blen'*/
2782                                         startag = NULLAGNUMBER;
2783                                         continue;
2784                                 }
2785                         }
2786                         if (++ag == mp->m_sb.sb_agcount)
2787                                 ag = 0;
2788                         if (ag == startag)
2789                                 break;
2790                 }
2791                 up_read(&mp->m_peraglock);
2792                 /*
2793                  * Since the above loop did a BUF_TRYLOCK, it is
2794                  * possible that there is space for this request.
2795                  */
2796                 if (notinit || blen < ap->minlen)
2797                         args.minlen = ap->minlen;
2798                 /*
2799                  * If the best seen length is less than the request
2800                  * length, use the best as the minimum.
2801                  */
2802                 else if (blen < ap->alen)
2803                         args.minlen = blen;
2804                 /*
2805                  * Otherwise we've seen an extent as big as alen,
2806                  * use that as the minimum.
2807                  */
2808                 else
2809                         args.minlen = ap->alen;
2810
2811                 /*
2812                  * set the failure fallback case to look in the selected
2813                  * AG as the stream may have moved.
2814                  */
2815                 if (xfs_inode_is_filestream(ap->ip))
2816                         ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2817         } else if (ap->low) {
2818                 if (xfs_inode_is_filestream(ap->ip))
2819                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2820                 else
2821                         args.type = XFS_ALLOCTYPE_START_BNO;
2822                 args.total = args.minlen = ap->minlen;
2823         } else {
2824                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2825                 args.total = ap->total;
2826                 args.minlen = ap->minlen;
2827         }
2828         /* apply extent size hints if obtained earlier */
2829         if (unlikely(align)) {
2830                 args.prod = align;
2831                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2832                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2833         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2834                 args.prod = 1;
2835                 args.mod = 0;
2836         } else {
2837                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2838                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2839                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2840         }
2841         /*
2842          * If we are not low on available data blocks, and the
2843          * underlying logical volume manager is a stripe, and
2844          * the file offset is zero then try to allocate data
2845          * blocks on stripe unit boundary.
2846          * NOTE: ap->aeof is only set if the allocation length
2847          * is >= the stripe unit and the allocation offset is
2848          * at the end of file.
2849          */
2850         if (!ap->low && ap->aeof) {
2851                 if (!ap->off) {
2852                         args.alignment = mp->m_dalign;
2853                         atype = args.type;
2854                         isaligned = 1;
2855                         /*
2856                          * Adjust for alignment
2857                          */
2858                         if (blen > args.alignment && blen <= ap->alen)
2859                                 args.minlen = blen - args.alignment;
2860                         args.minalignslop = 0;
2861                 } else {
2862                         /*
2863                          * First try an exact bno allocation.
2864                          * If it fails then do a near or start bno
2865                          * allocation with alignment turned on.
2866                          */
2867                         atype = args.type;
2868                         tryagain = 1;
2869                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2870                         args.alignment = 1;
2871                         /*
2872                          * Compute the minlen+alignment for the
2873                          * next case.  Set slop so that the value
2874                          * of minlen+alignment+slop doesn't go up
2875                          * between the calls.
2876                          */
2877                         if (blen > mp->m_dalign && blen <= ap->alen)
2878                                 nextminlen = blen - mp->m_dalign;
2879                         else
2880                                 nextminlen = args.minlen;
2881                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2882                                 args.minalignslop =
2883                                         nextminlen + mp->m_dalign -
2884                                         args.minlen - 1;
2885                         else
2886                                 args.minalignslop = 0;
2887                 }
2888         } else {
2889                 args.alignment = 1;
2890                 args.minalignslop = 0;
2891         }
2892         args.minleft = ap->minleft;
2893         args.wasdel = ap->wasdel;
2894         args.isfl = 0;
2895         args.userdata = ap->userdata;
2896         if ((error = xfs_alloc_vextent(&args)))
2897                 return error;
2898         if (tryagain && args.fsbno == NULLFSBLOCK) {
2899                 /*
2900                  * Exact allocation failed. Now try with alignment
2901                  * turned on.
2902                  */
2903                 args.type = atype;
2904                 args.fsbno = ap->rval;
2905                 args.alignment = mp->m_dalign;
2906                 args.minlen = nextminlen;
2907                 args.minalignslop = 0;
2908                 isaligned = 1;
2909                 if ((error = xfs_alloc_vextent(&args)))
2910                         return error;
2911         }
2912         if (isaligned && args.fsbno == NULLFSBLOCK) {
2913                 /*
2914                  * allocation failed, so turn off alignment and
2915                  * try again.
2916                  */
2917                 args.type = atype;
2918                 args.fsbno = ap->rval;
2919                 args.alignment = 0;
2920                 if ((error = xfs_alloc_vextent(&args)))
2921                         return error;
2922         }
2923         if (args.fsbno == NULLFSBLOCK && nullfb &&
2924             args.minlen > ap->minlen) {
2925                 args.minlen = ap->minlen;
2926                 args.type = XFS_ALLOCTYPE_START_BNO;
2927                 args.fsbno = ap->rval;
2928                 if ((error = xfs_alloc_vextent(&args)))
2929                         return error;
2930         }
2931         if (args.fsbno == NULLFSBLOCK && nullfb) {
2932                 args.fsbno = 0;
2933                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2934                 args.total = ap->minlen;
2935                 args.minleft = 0;
2936                 if ((error = xfs_alloc_vextent(&args)))
2937                         return error;
2938                 ap->low = 1;
2939         }
2940         if (args.fsbno != NULLFSBLOCK) {
2941                 ap->firstblock = ap->rval = args.fsbno;
2942                 ASSERT(nullfb || fb_agno == args.agno ||
2943                        (ap->low && fb_agno < args.agno));
2944                 ap->alen = args.len;
2945                 ap->ip->i_d.di_nblocks += args.len;
2946                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2947                 if (ap->wasdel)
2948                         ap->ip->i_delayed_blks -= args.len;
2949                 /*
2950                  * Adjust the disk quota also. This was reserved
2951                  * earlier.
2952                  */
2953                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2954                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2955                                         XFS_TRANS_DQ_BCOUNT,
2956                         (long) args.len);
2957         } else {
2958                 ap->rval = NULLFSBLOCK;
2959                 ap->alen = 0;
2960         }
2961         return 0;
2962 }
2963
2964 /*
2965  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2966  * It figures out where to ask the underlying allocator to put the new extent.
2967  */
2968 STATIC int
2969 xfs_bmap_alloc(
2970         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2971 {
2972         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2973                 return xfs_bmap_rtalloc(ap);
2974         return xfs_bmap_btalloc(ap);
2975 }
2976
2977 /*
2978  * Transform a btree format file with only one leaf node, where the
2979  * extents list will fit in the inode, into an extents format file.
2980  * Since the file extents are already in-core, all we have to do is
2981  * give up the space for the btree root and pitch the leaf block.
2982  */
2983 STATIC int                              /* error */
2984 xfs_bmap_btree_to_extents(
2985         xfs_trans_t             *tp,    /* transaction pointer */
2986         xfs_inode_t             *ip,    /* incore inode pointer */
2987         xfs_btree_cur_t         *cur,   /* btree cursor */
2988         int                     *logflagsp, /* inode logging flags */
2989         int                     whichfork)  /* data or attr fork */
2990 {
2991         /* REFERENCED */
2992         xfs_bmbt_block_t        *cblock;/* child btree block */
2993         xfs_fsblock_t           cbno;   /* child block number */
2994         xfs_buf_t               *cbp;   /* child block's buffer */
2995         int                     error;  /* error return value */
2996         xfs_ifork_t             *ifp;   /* inode fork data */
2997         xfs_mount_t             *mp;    /* mount point structure */
2998         __be64                  *pp;    /* ptr to block address */
2999         xfs_bmbt_block_t        *rblock;/* root btree block */
3000
3001         ifp = XFS_IFORK_PTR(ip, whichfork);
3002         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3003         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3004         rblock = ifp->if_broot;
3005         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3006         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3007         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3008         mp = ip->i_mount;
3009         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3010         cbno = be64_to_cpu(*pp);
3011         *logflagsp = 0;
3012 #ifdef DEBUG
3013         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3014                 return error;
3015 #endif
3016         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3017                         XFS_BMAP_BTREE_REF)))
3018                 return error;
3019         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3020         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3021                 return error;
3022         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3023         ip->i_d.di_nblocks--;
3024         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3025         xfs_trans_binval(tp, cbp);
3026         if (cur->bc_bufs[0] == cbp)
3027                 cur->bc_bufs[0] = NULL;
3028         xfs_iroot_realloc(ip, -1, whichfork);
3029         ASSERT(ifp->if_broot == NULL);
3030         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3031         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3032         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3033         return 0;
3034 }
3035
3036 /*
3037  * Called by xfs_bmapi to update file extent records and the btree
3038  * after removing space (or undoing a delayed allocation).
3039  */
3040 STATIC int                              /* error */
3041 xfs_bmap_del_extent(
3042         xfs_inode_t             *ip,    /* incore inode pointer */
3043         xfs_trans_t             *tp,    /* current transaction pointer */
3044         xfs_extnum_t            idx,    /* extent number to update/delete */
3045         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3046         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3047         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3048         int                     *logflagsp, /* inode logging flags */
3049         xfs_extdelta_t          *delta, /* Change made to incore extents */
3050         int                     whichfork, /* data or attr fork */
3051         int                     rsvd)   /* OK to allocate reserved blocks */
3052 {
3053         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3054         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3055         xfs_fsblock_t           del_endblock=0; /* first block past del */
3056         xfs_fileoff_t           del_endoff;     /* first offset past del */
3057         int                     delay;  /* current block is delayed allocated */
3058         int                     do_fx;  /* free extent at end of routine */
3059         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
3060         int                     error;  /* error return value */
3061         int                     flags;  /* inode logging flags */
3062         xfs_bmbt_irec_t         got;    /* current extent entry */
3063         xfs_fileoff_t           got_endoff;     /* first offset past got */
3064         int                     i;      /* temp state */
3065         xfs_ifork_t             *ifp;   /* inode fork pointer */
3066         xfs_mount_t             *mp;    /* mount structure */
3067         xfs_filblks_t           nblks;  /* quota/sb block count */
3068         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3069         /* REFERENCED */
3070         uint                    qfield; /* quota field to update */
3071         xfs_filblks_t           temp;   /* for indirect length calculations */
3072         xfs_filblks_t           temp2;  /* for indirect length calculations */
3073
3074         XFS_STATS_INC(xs_del_exlist);
3075         mp = ip->i_mount;
3076         ifp = XFS_IFORK_PTR(ip, whichfork);
3077         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3078                 (uint)sizeof(xfs_bmbt_rec_t)));
3079         ASSERT(del->br_blockcount > 0);
3080         ep = xfs_iext_get_ext(ifp, idx);
3081         xfs_bmbt_get_all(ep, &got);
3082         ASSERT(got.br_startoff <= del->br_startoff);
3083         del_endoff = del->br_startoff + del->br_blockcount;
3084         got_endoff = got.br_startoff + got.br_blockcount;
3085         ASSERT(got_endoff >= del_endoff);
3086         delay = ISNULLSTARTBLOCK(got.br_startblock);
3087         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3088         flags = 0;
3089         qfield = 0;
3090         error = 0;
3091         /*
3092          * If deleting a real allocation, must free up the disk space.
3093          */
3094         if (!delay) {
3095                 flags = XFS_ILOG_CORE;
3096                 /*
3097                  * Realtime allocation.  Free it and record di_nblocks update.
3098                  */
3099                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
3100                         xfs_fsblock_t   bno;
3101                         xfs_filblks_t   len;
3102
3103                         ASSERT(do_mod(del->br_blockcount,
3104                                       mp->m_sb.sb_rextsize) == 0);
3105                         ASSERT(do_mod(del->br_startblock,
3106                                       mp->m_sb.sb_rextsize) == 0);
3107                         bno = del->br_startblock;
3108                         len = del->br_blockcount;
3109                         do_div(bno, mp->m_sb.sb_rextsize);
3110                         do_div(len, mp->m_sb.sb_rextsize);
3111                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3112                                         (xfs_extlen_t)len)))
3113                                 goto done;
3114                         do_fx = 0;
3115                         nblks = len * mp->m_sb.sb_rextsize;
3116                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3117                 }
3118                 /*
3119                  * Ordinary allocation.
3120                  */
3121                 else {
3122                         do_fx = 1;
3123                         nblks = del->br_blockcount;
3124                         qfield = XFS_TRANS_DQ_BCOUNT;
3125                 }
3126                 /*
3127                  * Set up del_endblock and cur for later.
3128                  */
3129                 del_endblock = del->br_startblock + del->br_blockcount;
3130                 if (cur) {
3131                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3132                                         got.br_startblock, got.br_blockcount,
3133                                         &i)))
3134                                 goto done;
3135                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3136                 }
3137                 da_old = da_new = 0;
3138         } else {
3139                 da_old = STARTBLOCKVAL(got.br_startblock);
3140                 da_new = 0;
3141                 nblks = 0;
3142                 do_fx = 0;
3143         }
3144         /*
3145          * Set flag value to use in switch statement.
3146          * Left-contig is 2, right-contig is 1.
3147          */
3148         switch (((got.br_startoff == del->br_startoff) << 1) |
3149                 (got_endoff == del_endoff)) {
3150         case 3:
3151                 /*
3152                  * Matches the whole extent.  Delete the entry.
3153                  */
3154                 XFS_BMAP_TRACE_DELETE("3", ip, idx, 1, whichfork);
3155                 xfs_iext_remove(ifp, idx, 1);
3156                 ifp->if_lastex = idx;
3157                 if (delay)
3158                         break;
3159                 XFS_IFORK_NEXT_SET(ip, whichfork,
3160                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3161                 flags |= XFS_ILOG_CORE;
3162                 if (!cur) {
3163                         flags |= XFS_ILOG_FEXT(whichfork);
3164                         break;
3165                 }
3166                 if ((error = xfs_bmbt_delete(cur, &i)))
3167                         goto done;
3168                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3169                 break;
3170
3171         case 2:
3172                 /*
3173                  * Deleting the first part of the extent.
3174                  */
3175                 XFS_BMAP_TRACE_PRE_UPDATE("2", ip, idx, whichfork);
3176                 xfs_bmbt_set_startoff(ep, del_endoff);
3177                 temp = got.br_blockcount - del->br_blockcount;
3178                 xfs_bmbt_set_blockcount(ep, temp);
3179                 ifp->if_lastex = idx;
3180                 if (delay) {
3181                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3182                                 da_old);
3183                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3184                         XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx,
3185                                 whichfork);
3186                         da_new = temp;
3187                         break;
3188                 }
3189                 xfs_bmbt_set_startblock(ep, del_endblock);
3190                 XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx, whichfork);
3191                 if (!cur) {
3192                         flags |= XFS_ILOG_FEXT(whichfork);
3193                         break;
3194                 }
3195                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3196                                 got.br_blockcount - del->br_blockcount,
3197                                 got.br_state)))
3198                         goto done;
3199                 break;
3200
3201         case 1:
3202                 /*
3203                  * Deleting the last part of the extent.
3204                  */
3205                 temp = got.br_blockcount - del->br_blockcount;
3206                 XFS_BMAP_TRACE_PRE_UPDATE("1", ip, idx, whichfork);
3207                 xfs_bmbt_set_blockcount(ep, temp);
3208                 ifp->if_lastex = idx;
3209                 if (delay) {
3210                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3211                                 da_old);
3212                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3213                         XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx,
3214                                 whichfork);
3215                         da_new = temp;
3216                         break;
3217                 }
3218                 XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx, whichfork);
3219                 if (!cur) {
3220                         flags |= XFS_ILOG_FEXT(whichfork);
3221                         break;
3222                 }
3223                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3224                                 got.br_startblock,
3225                                 got.br_blockcount - del->br_blockcount,
3226                                 got.br_state)))
3227                         goto done;
3228                 break;
3229
3230         case 0:
3231                 /*
3232                  * Deleting the middle of the extent.
3233                  */
3234                 temp = del->br_startoff - got.br_startoff;
3235                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, whichfork);
3236                 xfs_bmbt_set_blockcount(ep, temp);
3237                 new.br_startoff = del_endoff;
3238                 temp2 = got_endoff - del_endoff;
3239                 new.br_blockcount = temp2;
3240                 new.br_state = got.br_state;
3241                 if (!delay) {
3242                         new.br_startblock = del_endblock;
3243                         flags |= XFS_ILOG_CORE;
3244                         if (cur) {
3245                                 if ((error = xfs_bmbt_update(cur,
3246                                                 got.br_startoff,
3247                                                 got.br_startblock, temp,
3248                                                 got.br_state)))
3249                                         goto done;
3250                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3251                                         goto done;
3252                                 cur->bc_rec.b = new;
3253                                 error = xfs_bmbt_insert(cur, &i);
3254                                 if (error && error != ENOSPC)
3255                                         goto done;
3256                                 /*
3257                                  * If get no-space back from btree insert,
3258                                  * it tried a split, and we have a zero
3259                                  * block reservation.
3260                                  * Fix up our state and return the error.
3261                                  */
3262                                 if (error == ENOSPC) {
3263                                         /*
3264                                          * Reset the cursor, don't trust
3265                                          * it after any insert operation.
3266                                          */
3267                                         if ((error = xfs_bmbt_lookup_eq(cur,
3268                                                         got.br_startoff,
3269                                                         got.br_startblock,
3270                                                         temp, &i)))
3271                                                 goto done;
3272                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3273                                         /*
3274                                          * Update the btree record back
3275                                          * to the original value.
3276                                          */
3277                                         if ((error = xfs_bmbt_update(cur,
3278                                                         got.br_startoff,
3279                                                         got.br_startblock,
3280                                                         got.br_blockcount,
3281                                                         got.br_state)))
3282                                                 goto done;
3283                                         /*
3284                                          * Reset the extent record back
3285                                          * to the original value.
3286                                          */
3287                                         xfs_bmbt_set_blockcount(ep,
3288                                                 got.br_blockcount);
3289                                         flags = 0;
3290                                         error = XFS_ERROR(ENOSPC);
3291                                         goto done;
3292                                 }
3293                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3294                         } else
3295                                 flags |= XFS_ILOG_FEXT(whichfork);
3296                         XFS_IFORK_NEXT_SET(ip, whichfork,
3297                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3298                 } else {
3299                         ASSERT(whichfork == XFS_DATA_FORK);
3300                         temp = xfs_bmap_worst_indlen(ip, temp);
3301                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3302                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3303                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3304                         da_new = temp + temp2;
3305                         while (da_new > da_old) {
3306                                 if (temp) {
3307                                         temp--;
3308                                         da_new--;
3309                                         xfs_bmbt_set_startblock(ep,
3310                                                 NULLSTARTBLOCK((int)temp));
3311                                 }
3312                                 if (da_new == da_old)
3313                                         break;
3314                                 if (temp2) {
3315                                         temp2--;
3316                                         da_new--;
3317                                         new.br_startblock =
3318                                                 NULLSTARTBLOCK((int)temp2);
3319                                 }
3320                         }
3321                 }
3322                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, whichfork);
3323                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 1, &new, NULL,
3324                         whichfork);
3325                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3326                 ifp->if_lastex = idx + 1;
3327                 break;
3328         }
3329         /*
3330          * If we need to, add to list of extents to delete.
3331          */
3332         if (do_fx)
3333                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3334                         mp);
3335         /*
3336          * Adjust inode # blocks in the file.
3337          */
3338         if (nblks)
3339                 ip->i_d.di_nblocks -= nblks;
3340         /*
3341          * Adjust quota data.
3342          */
3343         if (qfield)
3344                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3345
3346         /*
3347          * Account for change in delayed indirect blocks.
3348          * Nothing to do for disk quota accounting here.
3349          */
3350         ASSERT(da_old >= da_new);
3351         if (da_old > da_new)
3352                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3353                         rsvd);
3354         if (delta) {
3355                 /* DELTA: report the original extent. */
3356                 if (delta->xed_startoff > got.br_startoff)
3357                         delta->xed_startoff = got.br_startoff;
3358                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3359                         delta->xed_blockcount = got.br_startoff +
3360                                                         got.br_blockcount;
3361         }
3362 done:
3363         *logflagsp = flags;
3364         return error;
3365 }
3366
3367 /*
3368  * Remove the entry "free" from the free item list.  Prev points to the
3369  * previous entry, unless "free" is the head of the list.
3370  */
3371 STATIC void
3372 xfs_bmap_del_free(
3373         xfs_bmap_free_t         *flist, /* free item list header */
3374         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3375         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3376 {
3377         if (prev)
3378                 prev->xbfi_next = free->xbfi_next;
3379         else
3380                 flist->xbf_first = free->xbfi_next;
3381         flist->xbf_count--;
3382         kmem_zone_free(xfs_bmap_free_item_zone, free);
3383 }
3384
3385 /*
3386  * Convert an extents-format file into a btree-format file.
3387  * The new file will have a root block (in the inode) and a single child block.
3388  */
3389 STATIC int                                      /* error */
3390 xfs_bmap_extents_to_btree(
3391         xfs_trans_t             *tp,            /* transaction pointer */
3392         xfs_inode_t             *ip,            /* incore inode pointer */
3393         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3394         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3395         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3396         int                     wasdel,         /* converting a delayed alloc */
3397         int                     *logflagsp,     /* inode logging flags */
3398         int                     whichfork)      /* data or attr fork */
3399 {
3400         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3401         xfs_buf_t               *abp;           /* buffer for ablock */
3402         xfs_alloc_arg_t         args;           /* allocation arguments */
3403         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3404         xfs_bmbt_block_t        *block;         /* btree root block */
3405         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3406         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3407         int                     error;          /* error return value */
3408         xfs_extnum_t            i, cnt;         /* extent record index */
3409         xfs_ifork_t             *ifp;           /* inode fork pointer */
3410         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3411         xfs_mount_t             *mp;            /* mount structure */
3412         xfs_extnum_t            nextents;       /* number of file extents */
3413         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3414
3415         ifp = XFS_IFORK_PTR(ip, whichfork);
3416         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3417         ASSERT(ifp->if_ext_max ==
3418                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3419         /*
3420          * Make space in the inode incore.
3421          */
3422         xfs_iroot_realloc(ip, 1, whichfork);
3423         ifp->if_flags |= XFS_IFBROOT;
3424         /*
3425          * Fill in the root.
3426          */
3427         block = ifp->if_broot;
3428         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3429         block->bb_level = cpu_to_be16(1);
3430         block->bb_numrecs = cpu_to_be16(1);
3431         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3432         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3433         /*
3434          * Need a cursor.  Can't allocate until bb_level is filled in.
3435          */
3436         mp = ip->i_mount;
3437         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3438                 whichfork);
3439         cur->bc_private.b.firstblock = *firstblock;
3440         cur->bc_private.b.flist = flist;
3441         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3442         /*
3443          * Convert to a btree with two levels, one record in root.
3444          */
3445         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3446         args.tp = tp;
3447         args.mp = mp;
3448         args.firstblock = *firstblock;
3449         if (*firstblock == NULLFSBLOCK) {
3450                 args.type = XFS_ALLOCTYPE_START_BNO;
3451                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3452         } else if (flist->xbf_low) {
3453                 args.type = XFS_ALLOCTYPE_START_BNO;
3454                 args.fsbno = *firstblock;
3455         } else {
3456                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3457                 args.fsbno = *firstblock;
3458         }
3459         args.minlen = args.maxlen = args.prod = 1;
3460         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3461                 args.minalignslop = 0;
3462         args.wasdel = wasdel;
3463         *logflagsp = 0;
3464         if ((error = xfs_alloc_vextent(&args))) {
3465                 xfs_iroot_realloc(ip, -1, whichfork);
3466                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3467                 return error;
3468         }
3469         /*
3470          * Allocation can't fail, the space was reserved.
3471          */
3472         ASSERT(args.fsbno != NULLFSBLOCK);
3473         ASSERT(*firstblock == NULLFSBLOCK ||
3474                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3475                (flist->xbf_low &&
3476                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3477         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3478         cur->bc_private.b.allocated++;
3479         ip->i_d.di_nblocks++;
3480         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3481         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3482         /*
3483          * Fill in the child block.
3484          */
3485         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3486         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3487         ablock->bb_level = 0;
3488         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3489         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3490         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3491         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3492         for (cnt = i = 0; i < nextents; i++) {
3493                 ep = xfs_iext_get_ext(ifp, i);
3494                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3495                         arp->l0 = cpu_to_be64(ep->l0);
3496                         arp->l1 = cpu_to_be64(ep->l1);
3497                         arp++; cnt++;
3498                 }
3499         }
3500         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3501         ablock->bb_numrecs = cpu_to_be16(cnt);
3502         /*
3503          * Fill in the root key and pointer.
3504          */
3505         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3506         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3507         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3508         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3509         *pp = cpu_to_be64(args.fsbno);
3510         /*
3511          * Do all this logging at the end so that
3512          * the root is at the right level.
3513          */
3514         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3515         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3516         ASSERT(*curp == NULL);
3517         *curp = cur;
3518         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3519         return 0;
3520 }
3521
3522 /*
3523  * Helper routine to reset inode di_forkoff field when switching
3524  * attribute fork from local to extent format - we reset it where
3525  * possible to make space available for inline data fork extents.
3526  */
3527 STATIC void
3528 xfs_bmap_forkoff_reset(
3529         xfs_mount_t     *mp,
3530         xfs_inode_t     *ip,
3531         int             whichfork)
3532 {
3533         if (whichfork == XFS_ATTR_FORK &&
3534             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3535             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3536             (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3537             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3538                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3539                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3540                                         (uint)sizeof(xfs_bmbt_rec_t);
3541                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3542                                         (uint)sizeof(xfs_bmbt_rec_t);
3543         }
3544 }
3545
3546 /*
3547  * Convert a local file to an extents file.
3548  * This code is out of bounds for data forks of regular files,
3549  * since the file data needs to get logged so things will stay consistent.
3550  * (The bmap-level manipulations are ok, though).
3551  */
3552 STATIC int                              /* error */
3553 xfs_bmap_local_to_extents(
3554         xfs_trans_t     *tp,            /* transaction pointer */
3555         xfs_inode_t     *ip,            /* incore inode pointer */
3556         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3557         xfs_extlen_t    total,          /* total blocks needed by transaction */
3558         int             *logflagsp,     /* inode logging flags */
3559         int             whichfork)      /* data or attr fork */
3560 {
3561         int             error;          /* error return value */
3562         int             flags;          /* logging flags returned */
3563         xfs_ifork_t     *ifp;           /* inode fork pointer */
3564
3565         /*
3566          * We don't want to deal with the case of keeping inode data inline yet.
3567          * So sending the data fork of a regular inode is invalid.
3568          */
3569         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3570                  whichfork == XFS_DATA_FORK));
3571         ifp = XFS_IFORK_PTR(ip, whichfork);
3572         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3573         flags = 0;
3574         error = 0;
3575         if (ifp->if_bytes) {
3576                 xfs_alloc_arg_t args;   /* allocation arguments */
3577                 xfs_buf_t       *bp;    /* buffer for extent block */
3578                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3579
3580                 args.tp = tp;
3581                 args.mp = ip->i_mount;
3582                 args.firstblock = *firstblock;
3583                 ASSERT((ifp->if_flags &
3584                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3585                 /*
3586                  * Allocate a block.  We know we need only one, since the
3587                  * file currently fits in an inode.
3588                  */
3589                 if (*firstblock == NULLFSBLOCK) {
3590                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3591                         args.type = XFS_ALLOCTYPE_START_BNO;
3592                 } else {
3593                         args.fsbno = *firstblock;
3594                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3595                 }
3596                 args.total = total;
3597                 args.mod = args.minleft = args.alignment = args.wasdel =
3598                         args.isfl = args.minalignslop = 0;
3599                 args.minlen = args.maxlen = args.prod = 1;
3600                 if ((error = xfs_alloc_vextent(&args)))
3601                         goto done;
3602                 /*
3603                  * Can't fail, the space was reserved.
3604                  */
3605                 ASSERT(args.fsbno != NULLFSBLOCK);
3606                 ASSERT(args.len == 1);
3607                 *firstblock = args.fsbno;
3608                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3609                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3610                         ifp->if_bytes);
3611                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3612                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3613                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3614                 xfs_iext_add(ifp, 0, 1);
3615                 ep = xfs_iext_get_ext(ifp, 0);
3616                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3617                 XFS_BMAP_TRACE_POST_UPDATE("new", ip, 0, whichfork);
3618                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3619                 ip->i_d.di_nblocks = 1;
3620                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3621                         XFS_TRANS_DQ_BCOUNT, 1L);
3622                 flags |= XFS_ILOG_FEXT(whichfork);
3623         } else {
3624                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3625                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3626         }
3627         ifp->if_flags &= ~XFS_IFINLINE;
3628         ifp->if_flags |= XFS_IFEXTENTS;
3629         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3630         flags |= XFS_ILOG_CORE;
3631 done:
3632         *logflagsp = flags;
3633         return error;
3634 }
3635
3636 /*
3637  * Search the extent records for the entry containing block bno.
3638  * If bno lies in a hole, point to the next entry.  If bno lies
3639  * past eof, *eofp will be set, and *prevp will contain the last
3640  * entry (null if none).  Else, *lastxp will be set to the index
3641  * of the found entry; *gotp will contain the entry.
3642  */
3643 xfs_bmbt_rec_host_t *                   /* pointer to found extent entry */
3644 xfs_bmap_search_multi_extents(
3645         xfs_ifork_t     *ifp,           /* inode fork pointer */
3646         xfs_fileoff_t   bno,            /* block number searched for */
3647         int             *eofp,          /* out: end of file found */
3648         xfs_extnum_t    *lastxp,        /* out: last extent index */
3649         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3650         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3651 {
3652         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3653         xfs_extnum_t    lastx;          /* last extent index */
3654
3655         /*
3656          * Initialize the extent entry structure to catch access to
3657          * uninitialized br_startblock field.
3658          */
3659         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3660         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3661         gotp->br_state = XFS_EXT_INVALID;
3662 #if XFS_BIG_BLKNOS
3663         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3664 #else
3665         gotp->br_startblock = 0xffffa5a5;
3666 #endif
3667         prevp->br_startoff = NULLFILEOFF;
3668
3669         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3670         if (lastx > 0) {
3671                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3672         }
3673         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3674                 xfs_bmbt_get_all(ep, gotp);
3675                 *eofp = 0;
3676         } else {
3677                 if (lastx > 0) {
3678                         *gotp = *prevp;
3679                 }
3680                 *eofp = 1;
3681                 ep = NULL;
3682         }
3683         *lastxp = lastx;
3684         return ep;
3685 }
3686
3687 /*
3688  * Search the extents list for the inode, for the extent containing bno.
3689  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3690  * *eofp will be set, and *prevp will contain the last entry (null if none).
3691  * Else, *lastxp will be set to the index of the found
3692  * entry; *gotp will contain the entry.
3693  */
3694 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3695 xfs_bmap_search_extents(
3696         xfs_inode_t     *ip,            /* incore inode pointer */
3697         xfs_fileoff_t   bno,            /* block number searched for */
3698         int             fork,           /* data or attr fork */
3699         int             *eofp,          /* out: end of file found */
3700         xfs_extnum_t    *lastxp,        /* out: last extent index */
3701         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3702         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3703 {
3704         xfs_ifork_t     *ifp;           /* inode fork pointer */
3705         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3706
3707         XFS_STATS_INC(xs_look_exlist);
3708         ifp = XFS_IFORK_PTR(ip, fork);
3709
3710         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3711
3712         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3713                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3714                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3715                                 "Access to block zero in inode %llu "
3716                                 "start_block: %llx start_off: %llx "
3717                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3718                         (unsigned long long)ip->i_ino,
3719                         (unsigned long long)gotp->br_startblock,
3720                         (unsigned long long)gotp->br_startoff,
3721                         (unsigned long long)gotp->br_blockcount,
3722                         gotp->br_state, *lastxp);
3723                 *lastxp = NULLEXTNUM;
3724                 *eofp = 1;
3725                 return NULL;
3726         }
3727         return ep;
3728 }
3729
3730
3731 #ifdef XFS_BMAP_TRACE
3732 ktrace_t        *xfs_bmap_trace_buf;
3733
3734 /*
3735  * Add a bmap trace buffer entry.  Base routine for the others.
3736  */
3737 STATIC void
3738 xfs_bmap_trace_addentry(
3739         int             opcode,         /* operation */
3740         const char      *fname,         /* function name */
3741         char            *desc,          /* operation description */
3742         xfs_inode_t     *ip,            /* incore inode pointer */
3743         xfs_extnum_t    idx,            /* index of entry(ies) */
3744         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3745         xfs_bmbt_rec_host_t *r1,        /* first record */
3746         xfs_bmbt_rec_host_t *r2,        /* second record or null */
3747         int             whichfork)      /* data or attr fork */
3748 {
3749         xfs_bmbt_rec_host_t tr2;
3750
3751         ASSERT(cnt == 1 || cnt == 2);
3752         ASSERT(r1 != NULL);
3753         if (cnt == 1) {
3754                 ASSERT(r2 == NULL);
3755                 r2 = &tr2;
3756                 memset(&tr2, 0, sizeof(tr2));
3757         } else
3758                 ASSERT(r2 != NULL);
3759         ktrace_enter(xfs_bmap_trace_buf,
3760                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3761                 (void *)fname, (void *)desc, (void *)ip,
3762                 (void *)(__psint_t)idx,
3763                 (void *)(__psint_t)cnt,
3764                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3765                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3766                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3767                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3768                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3769                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3770                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3771                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3772                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3773                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3774                 );
3775         ASSERT(ip->i_xtrace);
3776         ktrace_enter(ip->i_xtrace,
3777                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3778                 (void *)fname, (void *)desc, (void *)ip,
3779                 (void *)(__psint_t)idx,
3780                 (void *)(__psint_t)cnt,
3781                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3782                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3783                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3784                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3785                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3786                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3787                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3788                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3789                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3790                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3791                 );
3792 }
3793
3794 /*
3795  * Add bmap trace entry prior to a call to xfs_iext_remove.
3796  */
3797 STATIC void
3798 xfs_bmap_trace_delete(
3799         const char      *fname,         /* function name */
3800         char            *desc,          /* operation description */
3801         xfs_inode_t     *ip,            /* incore inode pointer */
3802         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3803         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3804         int             whichfork)      /* data or attr fork */
3805 {
3806         xfs_ifork_t     *ifp;           /* inode fork pointer */
3807
3808         ifp = XFS_IFORK_PTR(ip, whichfork);
3809         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3810                 cnt, xfs_iext_get_ext(ifp, idx),
3811                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3812                 whichfork);
3813 }
3814
3815 /*
3816  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3817  * reading in the extents list from the disk (in the btree).
3818  */
3819 STATIC void
3820 xfs_bmap_trace_insert(
3821         const char      *fname,         /* function name */
3822         char            *desc,          /* operation description */
3823         xfs_inode_t     *ip,            /* incore inode pointer */
3824         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3825         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3826         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3827         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3828         int             whichfork)      /* data or attr fork */
3829 {
3830         xfs_bmbt_rec_host_t tr1;        /* compressed record 1 */
3831         xfs_bmbt_rec_host_t tr2;        /* compressed record 2 if needed */
3832
3833         xfs_bmbt_set_all(&tr1, r1);
3834         if (cnt == 2) {
3835                 ASSERT(r2 != NULL);
3836                 xfs_bmbt_set_all(&tr2, r2);
3837         } else {
3838                 ASSERT(cnt == 1);
3839                 ASSERT(r2 == NULL);
3840         }
3841         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3842                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3843 }
3844
3845 /*
3846  * Add bmap trace entry after updating an extent record in place.
3847  */
3848 STATIC void
3849 xfs_bmap_trace_post_update(
3850         const char      *fname,         /* function name */
3851         char            *desc,          /* operation description */
3852         xfs_inode_t     *ip,            /* incore inode pointer */
3853         xfs_extnum_t    idx,            /* index of entry updated */
3854         int             whichfork)      /* data or attr fork */
3855 {
3856         xfs_ifork_t     *ifp;           /* inode fork pointer */
3857
3858         ifp = XFS_IFORK_PTR(ip, whichfork);
3859         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3860                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3861 }
3862
3863 /*
3864  * Add bmap trace entry prior to updating an extent record in place.
3865  */
3866 STATIC void
3867 xfs_bmap_trace_pre_update(
3868         const char      *fname,         /* function name */
3869         char            *desc,          /* operation description */
3870         xfs_inode_t     *ip,            /* incore inode pointer */
3871         xfs_extnum_t    idx,            /* index of entry to be updated */
3872         int             whichfork)      /* data or attr fork */
3873 {
3874         xfs_ifork_t     *ifp;           /* inode fork pointer */
3875
3876         ifp = XFS_IFORK_PTR(ip, whichfork);
3877         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3878                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3879 }
3880 #endif  /* XFS_BMAP_TRACE */
3881
3882 /*
3883  * Compute the worst-case number of indirect blocks that will be used
3884  * for ip's delayed extent of length "len".
3885  */
3886 STATIC xfs_filblks_t
3887 xfs_bmap_worst_indlen(
3888         xfs_inode_t     *ip,            /* incore inode pointer */
3889         xfs_filblks_t   len)            /* delayed extent length */
3890 {
3891         int             level;          /* btree level number */
3892         int             maxrecs;        /* maximum record count at this level */
3893         xfs_mount_t     *mp;            /* mount structure */
3894         xfs_filblks_t   rval;           /* return value */
3895
3896         mp = ip->i_mount;
3897         maxrecs = mp->m_bmap_dmxr[0];
3898         for (level = 0, rval = 0;
3899              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3900              level++) {
3901                 len += maxrecs - 1;
3902                 do_div(len, maxrecs);
3903                 rval += len;
3904                 if (len == 1)
3905                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3906                                 level - 1;
3907                 if (level == 0)
3908                         maxrecs = mp->m_bmap_dmxr[1];
3909         }
3910         return rval;
3911 }
3912
3913 #if defined(XFS_RW_TRACE)
3914 STATIC void
3915 xfs_bunmap_trace(
3916         xfs_inode_t             *ip,
3917         xfs_fileoff_t           bno,
3918         xfs_filblks_t           len,
3919         int                     flags,
3920         inst_t                  *ra)
3921 {
3922         if (ip->i_rwtrace == NULL)
3923                 return;
3924         ktrace_enter(ip->i_rwtrace,
3925                 (void *)(__psint_t)XFS_BUNMAP,
3926                 (void *)ip,
3927                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3928                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3929                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3930                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3931                 (void *)(__psint_t)len,
3932                 (void *)(__psint_t)flags,
3933                 (void *)(unsigned long)current_cpu(),
3934                 (void *)ra,
3935                 (void *)0,
3936                 (void *)0,
3937                 (void *)0,
3938                 (void *)0,
3939                 (void *)0,
3940                 (void *)0);
3941 }
3942 #endif
3943
3944 /*
3945  * Convert inode from non-attributed to attributed.
3946  * Must not be in a transaction, ip must not be locked.
3947  */
3948 int                                             /* error code */
3949 xfs_bmap_add_attrfork(
3950         xfs_inode_t             *ip,            /* incore inode pointer */
3951         int                     size,           /* space new attribute needs */
3952         int                     rsvd)           /* xact may use reserved blks */
3953 {
3954         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3955         xfs_bmap_free_t         flist;          /* freed extent records */
3956         xfs_mount_t             *mp;            /* mount structure */
3957         xfs_trans_t             *tp;            /* transaction pointer */
3958         int                     blks;           /* space reservation */
3959         int                     version = 1;    /* superblock attr version */
3960         int                     committed;      /* xaction was committed */
3961         int                     logflags;       /* logging flags */
3962         int                     error;          /* error return value */
3963
3964         ASSERT(XFS_IFORK_Q(ip) == 0);
3965         ASSERT(ip->i_df.if_ext_max ==
3966                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3967
3968         mp = ip->i_mount;
3969         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3970         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3971         blks = XFS_ADDAFORK_SPACE_RES(mp);
3972         if (rsvd)
3973                 tp->t_flags |= XFS_TRANS_RESERVE;
3974         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3975                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3976                 goto error0;
3977         xfs_ilock(ip, XFS_ILOCK_EXCL);
3978         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3979                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3980                         XFS_QMOPT_RES_REGBLKS);
3981         if (error) {
3982                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3983                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3984                 return error;
3985         }
3986         if (XFS_IFORK_Q(ip))
3987                 goto error1;
3988         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3989                 /*
3990                  * For inodes coming from pre-6.2 filesystems.
3991                  */
3992                 ASSERT(ip->i_d.di_aformat == 0);
3993                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3994         }
3995         ASSERT(ip->i_d.di_anextents == 0);
3996         VN_HOLD(XFS_ITOV(ip));
3997         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3998         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3999         switch (ip->i_d.di_format) {
4000         case XFS_DINODE_FMT_DEV:
4001                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4002                 break;
4003         case XFS_DINODE_FMT_UUID:
4004                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4005                 break;
4006         case XFS_DINODE_FMT_LOCAL:
4007         case XFS_DINODE_FMT_EXTENTS:
4008         case XFS_DINODE_FMT_BTREE:
4009                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4010                 if (!ip->i_d.di_forkoff)
4011                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4012                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4013                         version = 2;
4014                 break;
4015         default:
4016                 ASSERT(0);
4017                 error = XFS_ERROR(EINVAL);
4018                 goto error1;
4019         }
4020         ip->i_df.if_ext_max =
4021                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4022         ASSERT(ip->i_afp == NULL);
4023         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4024         ip->i_afp->if_ext_max =
4025                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4026         ip->i_afp->if_flags = XFS_IFEXTENTS;
4027         logflags = 0;
4028         XFS_BMAP_INIT(&flist, &firstblock);
4029         switch (ip->i_d.di_format) {
4030         case XFS_DINODE_FMT_LOCAL:
4031                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4032                         &logflags);
4033                 break;
4034         case XFS_DINODE_FMT_EXTENTS:
4035                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4036                         &flist, &logflags);
4037                 break;
4038         case XFS_DINODE_FMT_BTREE:
4039                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4040                         &logflags);
4041                 break;
4042         default:
4043                 error = 0;
4044                 break;
4045         }
4046         if (logflags)
4047                 xfs_trans_log_inode(tp, ip, logflags);
4048         if (error)
4049                 goto error2;
4050         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
4051            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
4052                 __int64_t sbfields = 0;
4053
4054                 spin_lock(&mp->m_sb_lock);
4055                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
4056                         xfs_sb_version_addattr(&mp->m_sb);
4057                         sbfields |= XFS_SB_VERSIONNUM;
4058                 }
4059                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
4060                         xfs_sb_version_addattr2(&mp->m_sb);
4061                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4062                 }
4063                 if (sbfields) {
4064                         spin_unlock(&mp->m_sb_lock);
4065                         xfs_mod_sb(tp, sbfields);
4066                 } else
4067                         spin_unlock(&mp->m_sb_lock);
4068         }
4069         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4070                 goto error2;
4071         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4072         ASSERT(ip->i_df.if_ext_max ==
4073                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4074         return error;
4075 error2:
4076         xfs_bmap_cancel(&flist);
4077 error1:
4078         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4079 error0:
4080         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4081         ASSERT(ip->i_df.if_ext_max ==
4082                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4083         return error;
4084 }
4085
4086 /*
4087  * Add the extent to the list of extents to be free at transaction end.
4088  * The list is maintained sorted (by block number).
4089  */
4090 /* ARGSUSED */
4091 void
4092 xfs_bmap_add_free(
4093         xfs_fsblock_t           bno,            /* fs block number of extent */
4094         xfs_filblks_t           len,            /* length of extent */
4095         xfs_bmap_free_t         *flist,         /* list of extents */
4096         xfs_mount_t             *mp)            /* mount point structure */
4097 {
4098         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4099         xfs_bmap_free_item_t    *new;           /* new element */
4100         xfs_bmap_free_item_t    *prev;          /* previous element */
4101 #ifdef DEBUG
4102         xfs_agnumber_t          agno;
4103         xfs_agblock_t           agbno;
4104
4105         ASSERT(bno != NULLFSBLOCK);
4106         ASSERT(len > 0);
4107         ASSERT(len <= MAXEXTLEN);
4108         ASSERT(!ISNULLSTARTBLOCK(bno));
4109         agno = XFS_FSB_TO_AGNO(mp, bno);
4110         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4111         ASSERT(agno < mp->m_sb.sb_agcount);
4112         ASSERT(agbno < mp->m_sb.sb_agblocks);
4113         ASSERT(len < mp->m_sb.sb_agblocks);
4114         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4115 #endif
4116         ASSERT(xfs_bmap_free_item_zone != NULL);
4117         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4118         new->xbfi_startblock = bno;
4119         new->xbfi_blockcount = (xfs_extlen_t)len;
4120         for (prev = NULL, cur = flist->xbf_first;
4121              cur != NULL;
4122              prev = cur, cur = cur->xbfi_next) {
4123                 if (cur->xbfi_startblock >= bno)
4124                         break;
4125         }
4126         if (prev)
4127                 prev->xbfi_next = new;
4128         else
4129                 flist->xbf_first = new;
4130         new->xbfi_next = cur;
4131         flist->xbf_count++;
4132 }
4133
4134 /*
4135  * Compute and fill in the value of the maximum depth of a bmap btree
4136  * in this filesystem.  Done once, during mount.
4137  */
4138 void
4139 xfs_bmap_compute_maxlevels(
4140         xfs_mount_t     *mp,            /* file system mount structure */
4141         int             whichfork)      /* data or attr fork */
4142 {
4143         int             level;          /* btree level */
4144         uint            maxblocks;      /* max blocks at this level */
4145         uint            maxleafents;    /* max leaf entries possible */
4146         int             maxrootrecs;    /* max records in root block */
4147         int             minleafrecs;    /* min records in leaf block */
4148         int             minnoderecs;    /* min records in node block */
4149         int             sz;             /* root block size */
4150
4151         /*
4152          * The maximum number of extents in a file, hence the maximum
4153          * number of leaf entries, is controlled by the type of di_nextents
4154          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4155          * (a signed 16-bit number, xfs_aextnum_t).
4156          *
4157          * Note that we can no longer assume that if we are in ATTR1 that
4158          * the fork offset of all the inodes will be (m_attroffset >> 3)
4159          * because we could have mounted with ATTR2 and then mounted back
4160          * with ATTR1, keeping the di_forkoff's fixed but probably at
4161          * various positions. Therefore, for both ATTR1 and ATTR2
4162          * we have to assume the worst case scenario of a minimum size
4163          * available.
4164          */
4165         if (whichfork == XFS_DATA_FORK) {
4166                 maxleafents = MAXEXTNUM;
4167                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4168         } else {
4169                 maxleafents = MAXAEXTNUM;
4170                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
4171         }
4172         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4173         minleafrecs = mp->m_bmap_dmnr[0];
4174         minnoderecs = mp->m_bmap_dmnr[1];
4175         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4176         for (level = 1; maxblocks > 1; level++) {
4177                 if (maxblocks <= maxrootrecs)
4178                         maxblocks = 1;
4179                 else
4180                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4181         }
4182         mp->m_bm_maxlevels[whichfork] = level;
4183 }
4184
4185 /*
4186  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4187  * caller.  Frees all the extents that need freeing, which must be done
4188  * last due to locking considerations.  We never free any extents in
4189  * the first transaction.  This is to allow the caller to make the first
4190  * transaction a synchronous one so that the pointers to the data being
4191  * broken in this transaction will be permanent before the data is actually
4192  * freed.  This is necessary to prevent blocks from being reallocated
4193  * and written to before the free and reallocation are actually permanent.
4194  * We do not just make the first transaction synchronous here, because
4195  * there are more efficient ways to gain the same protection in some cases
4196  * (see the file truncation code).
4197  *
4198  * Return 1 if the given transaction was committed and a new one
4199  * started, and 0 otherwise in the committed parameter.
4200  */
4201 /*ARGSUSED*/
4202 int                                             /* error */
4203 xfs_bmap_finish(
4204         xfs_trans_t             **tp,           /* transaction pointer addr */
4205         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4206         int                     *committed)     /* xact committed or not */
4207 {
4208         xfs_efd_log_item_t      *efd;           /* extent free data */
4209         xfs_efi_log_item_t      *efi;           /* extent free intention */
4210         int                     error;          /* error return value */
4211         xfs_bmap_free_item_t    *free;          /* free extent item */
4212         unsigned int            logres;         /* new log reservation */
4213         unsigned int            logcount;       /* new log count */
4214         xfs_mount_t             *mp;            /* filesystem mount structure */
4215         xfs_bmap_free_item_t    *next;          /* next item on free list */
4216         xfs_trans_t             *ntp;           /* new transaction pointer */
4217
4218         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4219         if (flist->xbf_count == 0) {
4220                 *committed = 0;
4221                 return 0;
4222         }
4223         ntp = *tp;
4224         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4225         for (free = flist->xbf_first; free; free = free->xbfi_next)
4226                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4227                         free->xbfi_blockcount);
4228         logres = ntp->t_log_res;
4229         logcount = ntp->t_log_count;
4230         ntp = xfs_trans_dup(*tp);
4231         error = xfs_trans_commit(*tp, 0);
4232         *tp = ntp;
4233         *committed = 1;
4234         /*
4235          * We have a new transaction, so we should return committed=1,
4236          * even though we're returning an error.
4237          */
4238         if (error) {
4239                 return error;
4240         }
4241         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4242                         logcount)))
4243                 return error;
4244         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4245         for (free = flist->xbf_first; free != NULL; free = next) {
4246                 next = free->xbfi_next;
4247                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4248                                 free->xbfi_blockcount))) {
4249                         /*
4250                          * The bmap free list will be cleaned up at a
4251                          * higher level.  The EFI will be canceled when
4252                          * this transaction is aborted.
4253                          * Need to force shutdown here to make sure it
4254                          * happens, since this transaction may not be
4255                          * dirty yet.
4256                          */
4257                         mp = ntp->t_mountp;
4258                         if (!XFS_FORCED_SHUTDOWN(mp))
4259                                 xfs_force_shutdown(mp,
4260                                                    (error == EFSCORRUPTED) ?
4261                                                    SHUTDOWN_CORRUPT_INCORE :
4262                                                    SHUTDOWN_META_IO_ERROR);
4263                         return error;
4264                 }
4265                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4266                         free->xbfi_blockcount);
4267                 xfs_bmap_del_free(flist, NULL, free);
4268         }
4269         return 0;
4270 }
4271
4272 /*
4273  * Free up any items left in the list.
4274  */
4275 void
4276 xfs_bmap_cancel(
4277         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4278 {
4279         xfs_bmap_free_item_t    *free;  /* free list item */
4280         xfs_bmap_free_item_t    *next;
4281
4282         if (flist->xbf_count == 0)
4283                 return;
4284         ASSERT(flist->xbf_first != NULL);
4285         for (free = flist->xbf_first; free; free = next) {
4286                 next = free->xbfi_next;
4287                 xfs_bmap_del_free(flist, NULL, free);
4288         }
4289         ASSERT(flist->xbf_count == 0);
4290 }
4291
4292 /*
4293  * Returns the file-relative block number of the first unused block(s)
4294  * in the file with at least "len" logically contiguous blocks free.
4295  * This is the lowest-address hole if the file has holes, else the first block
4296  * past the end of file.
4297  * Return 0 if the file is currently local (in-inode).
4298  */
4299 int                                             /* error */
4300 xfs_bmap_first_unused(
4301         xfs_trans_t     *tp,                    /* transaction pointer */
4302         xfs_inode_t     *ip,                    /* incore inode */
4303         xfs_extlen_t    len,                    /* size of hole to find */
4304         xfs_fileoff_t   *first_unused,          /* unused block */
4305         int             whichfork)              /* data or attr fork */
4306 {
4307         int             error;                  /* error return value */
4308         int             idx;                    /* extent record index */
4309         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4310         xfs_fileoff_t   lastaddr;               /* last block number seen */
4311         xfs_fileoff_t   lowest;                 /* lowest useful block */
4312         xfs_fileoff_t   max;                    /* starting useful block */
4313         xfs_fileoff_t   off;                    /* offset for this block */
4314         xfs_extnum_t    nextents;               /* number of extent entries */
4315
4316         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4317                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4318                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4319         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4320                 *first_unused = 0;
4321                 return 0;
4322         }
4323         ifp = XFS_IFORK_PTR(ip, whichfork);
4324         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4325             (error = xfs_iread_extents(tp, ip, whichfork)))
4326                 return error;
4327         lowest = *first_unused;
4328         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4329         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4330                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
4331                 off = xfs_bmbt_get_startoff(ep);
4332                 /*
4333                  * See if the hole before this extent will work.
4334                  */
4335                 if (off >= lowest + len && off - max >= len) {
4336                         *first_unused = max;
4337                         return 0;
4338                 }
4339                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4340                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4341         }
4342         *first_unused = max;
4343         return 0;
4344 }
4345
4346 /*
4347  * Returns the file-relative block number of the last block + 1 before
4348  * last_block (input value) in the file.
4349  * This is not based on i_size, it is based on the extent records.
4350  * Returns 0 for local files, as they do not have extent records.
4351  */
4352 int                                             /* error */
4353 xfs_bmap_last_before(
4354         xfs_trans_t     *tp,                    /* transaction pointer */
4355         xfs_inode_t     *ip,                    /* incore inode */
4356         xfs_fileoff_t   *last_block,            /* last block */
4357         int             whichfork)              /* data or attr fork */
4358 {
4359         xfs_fileoff_t   bno;                    /* input file offset */
4360         int             eof;                    /* hit end of file */
4361         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4362         int             error;                  /* error return value */
4363         xfs_bmbt_irec_t got;                    /* current extent value */
4364         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4365         xfs_extnum_t    lastx;                  /* last extent used */
4366         xfs_bmbt_irec_t prev;                   /* previous extent value */
4367
4368         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4369             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4370             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4371                return XFS_ERROR(EIO);
4372         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4373                 *last_block = 0;
4374                 return 0;
4375         }
4376         ifp = XFS_IFORK_PTR(ip, whichfork);
4377         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4378             (error = xfs_iread_extents(tp, ip, whichfork)))
4379                 return error;
4380         bno = *last_block - 1;
4381         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4382                 &prev);
4383         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4384                 if (prev.br_startoff == NULLFILEOFF)
4385                         *last_block = 0;
4386                 else
4387                         *last_block = prev.br_startoff + prev.br_blockcount;
4388         }
4389         /*
4390          * Otherwise *last_block is already the right answer.
4391          */
4392         return 0;
4393 }
4394
4395 /*
4396  * Returns the file-relative block number of the first block past eof in
4397  * the file.  This is not based on i_size, it is based on the extent records.
4398  * Returns 0 for local files, as they do not have extent records.
4399  */
4400 int                                             /* error */
4401 xfs_bmap_last_offset(
4402         xfs_trans_t     *tp,                    /* transaction pointer */
4403         xfs_inode_t     *ip,                    /* incore inode */
4404         xfs_fileoff_t   *last_block,            /* last block */
4405         int             whichfork)              /* data or attr fork */
4406 {
4407         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4408         int             error;                  /* error return value */
4409         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4410         xfs_extnum_t    nextents;               /* number of extent entries */
4411
4412         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4413             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4414             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4415                return XFS_ERROR(EIO);
4416         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4417                 *last_block = 0;
4418                 return 0;
4419         }
4420         ifp = XFS_IFORK_PTR(ip, whichfork);
4421         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4422             (error = xfs_iread_extents(tp, ip, whichfork)))
4423                 return error;
4424         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4425         if (!nextents) {
4426                 *last_block = 0;
4427                 return 0;
4428         }
4429         ep = xfs_iext_get_ext(ifp, nextents - 1);
4430         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4431         return 0;
4432 }
4433
4434 /*
4435  * Returns whether the selected fork of the inode has exactly one
4436  * block or not.  For the data fork we check this matches di_size,
4437  * implying the file's range is 0..bsize-1.
4438  */
4439 int                                     /* 1=>1 block, 0=>otherwise */
4440 xfs_bmap_one_block(
4441         xfs_inode_t     *ip,            /* incore inode */
4442         int             whichfork)      /* data or attr fork */
4443 {
4444         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4445         xfs_ifork_t     *ifp;           /* inode fork pointer */
4446         int             rval;           /* return value */
4447         xfs_bmbt_irec_t s;              /* internal version of extent */
4448
4449 #ifndef DEBUG
4450         if (whichfork == XFS_DATA_FORK) {
4451                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4452                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4453                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4454         }
4455 #endif  /* !DEBUG */
4456         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4457                 return 0;
4458         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4459                 return 0;
4460         ifp = XFS_IFORK_PTR(ip, whichfork);
4461         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4462         ep = xfs_iext_get_ext(ifp, 0);
4463         xfs_bmbt_get_all(ep, &s);
4464         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4465         if (rval && whichfork == XFS_DATA_FORK)
4466                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4467         return rval;
4468 }
4469
4470 /*
4471  * Read in the extents to if_extents.
4472  * All inode fields are set up by caller, we just traverse the btree
4473  * and copy the records in. If the file system cannot contain unwritten
4474  * extents, the records are checked for no "state" flags.
4475  */
4476 int                                     /* error */
4477 xfs_bmap_read_extents(
4478         xfs_trans_t             *tp,    /* transaction pointer */
4479         xfs_inode_t             *ip,    /* incore inode */
4480         int                     whichfork) /* data or attr fork */
4481 {
4482         xfs_bmbt_block_t        *block; /* current btree block */
4483         xfs_fsblock_t           bno;    /* block # of "block" */
4484         xfs_buf_t               *bp;    /* buffer for "block" */
4485         int                     error;  /* error return value */
4486         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4487         xfs_extnum_t            i, j;   /* index into the extents list */
4488         xfs_ifork_t             *ifp;   /* fork structure */
4489         int                     level;  /* btree level, for checking */
4490         xfs_mount_t             *mp;    /* file system mount structure */
4491         __be64                  *pp;    /* pointer to block address */
4492         /* REFERENCED */
4493         xfs_extnum_t            room;   /* number of entries there's room for */
4494
4495         bno = NULLFSBLOCK;
4496         mp = ip->i_mount;
4497         ifp = XFS_IFORK_PTR(ip, whichfork);
4498         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4499                                         XFS_EXTFMT_INODE(ip);
4500         block = ifp->if_broot;
4501         /*
4502          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4503          */
4504         level = be16_to_cpu(block->bb_level);
4505         ASSERT(level > 0);
4506         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4507         bno = be64_to_cpu(*pp);
4508         ASSERT(bno != NULLDFSBNO);
4509         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4510         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4511         /*
4512          * Go down the tree until leaf level is reached, following the first
4513          * pointer (leftmost) at each level.
4514          */
4515         while (level-- > 0) {
4516                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4517                                 XFS_BMAP_BTREE_REF)))
4518                         return error;
4519                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4520                 XFS_WANT_CORRUPTED_GOTO(
4521                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4522                         error0);
4523                 if (level == 0)
4524                         break;
4525                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4526                 bno = be64_to_cpu(*pp);
4527                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4528                 xfs_trans_brelse(tp, bp);
4529         }
4530         /*
4531          * Here with bp and block set to the leftmost leaf node in the tree.
4532          */
4533         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4534         i = 0;
4535         /*
4536          * Loop over all leaf nodes.  Copy information to the extent records.
4537          */
4538         for (;;) {
4539                 xfs_bmbt_rec_t  *frp;
4540                 xfs_fsblock_t   nextbno;
4541                 xfs_extnum_t    num_recs;
4542                 xfs_extnum_t    start;
4543
4544
4545                 num_recs = be16_to_cpu(block->bb_numrecs);
4546                 if (unlikely(i + num_recs > room)) {
4547                         ASSERT(i + num_recs <= room);
4548                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4549                                 "corrupt dinode %Lu, (btree extents).",
4550                                 (unsigned long long) ip->i_ino);
4551                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4552                                          XFS_ERRLEVEL_LOW,
4553                                         ip->i_mount);
4554                         goto error0;
4555                 }
4556                 XFS_WANT_CORRUPTED_GOTO(
4557                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4558                         error0);
4559                 /*
4560                  * Read-ahead the next leaf block, if any.
4561                  */
4562                 nextbno = be64_to_cpu(block->bb_rightsib);
4563                 if (nextbno != NULLFSBLOCK)
4564                         xfs_btree_reada_bufl(mp, nextbno, 1);
4565                 /*
4566                  * Copy records into the extent records.
4567                  */
4568                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4569                 start = i;
4570                 for (j = 0; j < num_recs; j++, i++, frp++) {
4571                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4572                         trp->l0 = be64_to_cpu(frp->l0);
4573                         trp->l1 = be64_to_cpu(frp->l1);
4574                 }
4575                 if (exntf == XFS_EXTFMT_NOSTATE) {
4576                         /*
4577                          * Check all attribute bmap btree records and
4578                          * any "older" data bmap btree records for a
4579                          * set bit in the "extent flag" position.
4580                          */
4581                         if (unlikely(xfs_check_nostate_extents(ifp,
4582                                         start, num_recs))) {
4583                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4584                                                  XFS_ERRLEVEL_LOW,
4585                                                  ip->i_mount);
4586                                 goto error0;
4587                         }
4588                 }
4589                 xfs_trans_brelse(tp, bp);
4590                 bno = nextbno;
4591                 /*
4592                  * If we've reached the end, stop.
4593                  */
4594                 if (bno == NULLFSBLOCK)
4595                         break;
4596                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4597                                 XFS_BMAP_BTREE_REF)))
4598                         return error;
4599                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4600         }
4601         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4602         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4603         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4604         return 0;
4605 error0:
4606         xfs_trans_brelse(tp, bp);
4607         return XFS_ERROR(EFSCORRUPTED);
4608 }
4609
4610 #ifdef XFS_BMAP_TRACE
4611 /*
4612  * Add bmap trace insert entries for all the contents of the extent records.
4613  */
4614 void
4615 xfs_bmap_trace_exlist(
4616         const char      *fname,         /* function name */
4617         xfs_inode_t     *ip,            /* incore inode pointer */
4618         xfs_extnum_t    cnt,            /* count of entries in the list */
4619         int             whichfork)      /* data or attr fork */
4620 {
4621         xfs_bmbt_rec_host_t *ep;        /* current extent record */
4622         xfs_extnum_t    idx;            /* extent record index */
4623         xfs_ifork_t     *ifp;           /* inode fork pointer */
4624         xfs_bmbt_irec_t s;              /* file extent record */
4625
4626         ifp = XFS_IFORK_PTR(ip, whichfork);
4627         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4628         for (idx = 0; idx < cnt; idx++) {
4629                 ep = xfs_iext_get_ext(ifp, idx);
4630                 xfs_bmbt_get_all(ep, &s);
4631                 XFS_BMAP_TRACE_INSERT("exlist", ip, idx, 1, &s, NULL,
4632                         whichfork);
4633         }
4634 }
4635 #endif
4636
4637 #ifdef DEBUG
4638 /*
4639  * Validate that the bmbt_irecs being returned from bmapi are valid
4640  * given the callers original parameters.  Specifically check the
4641  * ranges of the returned irecs to ensure that they only extent beyond
4642  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4643  */
4644 STATIC void
4645 xfs_bmap_validate_ret(
4646         xfs_fileoff_t           bno,
4647         xfs_filblks_t           len,
4648         int                     flags,
4649         xfs_bmbt_irec_t         *mval,
4650         int                     nmap,
4651         int                     ret_nmap)
4652 {
4653         int                     i;              /* index to map values */
4654
4655         ASSERT(ret_nmap <= nmap);
4656
4657         for (i = 0; i < ret_nmap; i++) {
4658                 ASSERT(mval[i].br_blockcount > 0);
4659                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4660                         ASSERT(mval[i].br_startoff >= bno);
4661                         ASSERT(mval[i].br_blockcount <= len);
4662                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4663                                bno + len);
4664                 } else {
4665                         ASSERT(mval[i].br_startoff < bno + len);
4666                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4667                                bno);
4668                 }
4669                 ASSERT(i == 0 ||
4670                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4671                        mval[i].br_startoff);
4672                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4673                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4674                                mval[i].br_startblock != HOLESTARTBLOCK);
4675                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4676                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4677         }
4678 }
4679 #endif /* DEBUG */
4680
4681
4682 /*
4683  * Map file blocks to filesystem blocks.
4684  * File range is given by the bno/len pair.
4685  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4686  * into a hole or past eof.
4687  * Only allocates blocks from a single allocation group,
4688  * to avoid locking problems.
4689  * The returned value in "firstblock" from the first call in a transaction
4690  * must be remembered and presented to subsequent calls in "firstblock".
4691  * An upper bound for the number of blocks to be allocated is supplied to
4692  * the first call in "total"; if no allocation group has that many free
4693  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4694  */
4695 int                                     /* error */
4696 xfs_bmapi(
4697         xfs_trans_t     *tp,            /* transaction pointer */
4698         xfs_inode_t     *ip,            /* incore inode */
4699         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4700         xfs_filblks_t   len,            /* length to map in file */
4701         int             flags,          /* XFS_BMAPI_... */
4702         xfs_fsblock_t   *firstblock,    /* first allocated block
4703                                            controls a.g. for allocs */
4704         xfs_extlen_t    total,          /* total blocks needed */
4705         xfs_bmbt_irec_t *mval,          /* output: map values */
4706         int             *nmap,          /* i/o: mval size/count */
4707         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4708         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4709 {
4710         xfs_fsblock_t   abno;           /* allocated block number */
4711         xfs_extlen_t    alen;           /* allocated extent length */
4712         xfs_fileoff_t   aoff;           /* allocated file offset */
4713         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4714         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4715         xfs_fileoff_t   end;            /* end of mapped file region */
4716         int             eof;            /* we've hit the end of extents */
4717         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4718         int             error;          /* error return */
4719         xfs_bmbt_irec_t got;            /* current file extent record */
4720         xfs_ifork_t     *ifp;           /* inode fork pointer */
4721         xfs_extlen_t    indlen;         /* indirect blocks length */
4722         xfs_extnum_t    lastx;          /* last useful extent number */
4723         int             logflags;       /* flags for transaction logging */
4724         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4725         xfs_extlen_t    minlen;         /* min allocation size */
4726         xfs_mount_t     *mp;            /* xfs mount structure */
4727         int             n;              /* current extent index */
4728         int             nallocs;        /* number of extents alloc\'d */
4729         xfs_extnum_t    nextents;       /* number of extents in file */
4730         xfs_fileoff_t   obno;           /* old block number (offset) */
4731         xfs_bmbt_irec_t prev;           /* previous file extent record */
4732         int             tmp_logflags;   /* temp flags holder */
4733         int             whichfork;      /* data or attr fork */
4734         char            inhole;         /* current location is hole in file */
4735         char            wasdelay;       /* old extent was delayed */
4736         char            wr;             /* this is a write request */
4737         char            rt;             /* this is a realtime file */
4738 #ifdef DEBUG
4739         xfs_fileoff_t   orig_bno;       /* original block number value */
4740         int             orig_flags;     /* original flags arg value */
4741         xfs_filblks_t   orig_len;       /* original value of len arg */
4742         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4743         int             orig_nmap;      /* original value of *nmap */
4744
4745         orig_bno = bno;
4746         orig_len = len;
4747         orig_flags = flags;
4748         orig_mval = mval;
4749         orig_nmap = *nmap;
4750 #endif
4751         ASSERT(*nmap >= 1);
4752         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4753         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4754                 XFS_ATTR_FORK : XFS_DATA_FORK;
4755         mp = ip->i_mount;
4756         if (unlikely(XFS_TEST_ERROR(
4757             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4758              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4759              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4760              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4761                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4762                 return XFS_ERROR(EFSCORRUPTED);
4763         }
4764         if (XFS_FORCED_SHUTDOWN(mp))
4765                 return XFS_ERROR(EIO);
4766         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4767         ifp = XFS_IFORK_PTR(ip, whichfork);
4768         ASSERT(ifp->if_ext_max ==
4769                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4770         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4771                 XFS_STATS_INC(xs_blk_mapw);
4772         else
4773                 XFS_STATS_INC(xs_blk_mapr);
4774         /*
4775          * IGSTATE flag is used to combine extents which
4776          * differ only due to the state of the extents.
4777          * This technique is used from xfs_getbmap()
4778          * when the caller does not wish to see the
4779          * separation (which is the default).
4780          *
4781          * This technique is also used when writing a
4782          * buffer which has been partially written,
4783          * (usually by being flushed during a chunkread),
4784          * to ensure one write takes place. This also
4785          * prevents a change in the xfs inode extents at
4786          * this time, intentionally. This change occurs
4787          * on completion of the write operation, in
4788          * xfs_strat_comp(), where the xfs_bmapi() call
4789          * is transactioned, and the extents combined.
4790          */
4791         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4792                 wr = 0;                         /* no allocations are allowed */
4793         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4794         logflags = 0;
4795         nallocs = 0;
4796         cur = NULL;
4797         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4798                 ASSERT(wr && tp);
4799                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4800                                 firstblock, total, &logflags, whichfork)))
4801                         goto error0;
4802         }
4803         if (wr && *firstblock == NULLFSBLOCK) {
4804                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4805                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4806                 else
4807                         minleft = 1;
4808         } else
4809                 minleft = 0;
4810         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4811             (error = xfs_iread_extents(tp, ip, whichfork)))
4812                 goto error0;
4813         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4814                 &prev);
4815         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4816         n = 0;
4817         end = bno + len;
4818         obno = bno;
4819         bma.ip = NULL;
4820         if (delta) {
4821                 delta->xed_startoff = NULLFILEOFF;
4822                 delta->xed_blockcount = 0;
4823         }
4824         while (bno < end && n < *nmap) {
4825                 /*
4826                  * Reading past eof, act as though there's a hole
4827                  * up to end.
4828                  */
4829                 if (eof && !wr)
4830                         got.br_startoff = end;
4831                 inhole = eof || got.br_startoff > bno;
4832                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4833                         ISNULLSTARTBLOCK(got.br_startblock);
4834                 /*
4835                  * First, deal with the hole before the allocated space
4836                  * that we found, if any.
4837                  */
4838                 if (wr && (inhole || wasdelay)) {
4839                         /*
4840                          * For the wasdelay case, we could also just
4841                          * allocate the stuff asked for in this bmap call
4842                          * but that wouldn't be as good.
4843                          */
4844                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4845                                 alen = (xfs_extlen_t)got.br_blockcount;
4846                                 aoff = got.br_startoff;
4847                                 if (lastx != NULLEXTNUM && lastx) {
4848                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4849                                         xfs_bmbt_get_all(ep, &prev);
4850                                 }
4851                         } else if (wasdelay) {
4852                                 alen = (xfs_extlen_t)
4853                                         XFS_FILBLKS_MIN(len,
4854                                                 (got.br_startoff +
4855                                                  got.br_blockcount) - bno);
4856                                 aoff = bno;
4857                         } else {
4858                                 alen = (xfs_extlen_t)
4859                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4860                                 if (!eof)
4861                                         alen = (xfs_extlen_t)
4862                                                 XFS_FILBLKS_MIN(alen,
4863                                                         got.br_startoff - bno);
4864                                 aoff = bno;
4865                         }
4866                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4867                         if (flags & XFS_BMAPI_DELAY) {
4868                                 xfs_extlen_t    extsz;
4869
4870                                 /* Figure out the extent size, adjust alen */
4871                                 extsz = xfs_get_extsz_hint(ip);
4872                                 if (extsz) {
4873                                         error = xfs_bmap_extsize_align(mp,
4874                                                         &got, &prev, extsz,
4875                                                         rt, eof,
4876                                                         flags&XFS_BMAPI_DELAY,
4877                                                         flags&XFS_BMAPI_CONVERT,
4878                                                         &aoff, &alen);
4879                                         ASSERT(!error);
4880                                 }
4881
4882                                 if (rt)
4883                                         extsz = alen / mp->m_sb.sb_rextsize;
4884
4885                                 /*
4886                                  * Make a transaction-less quota reservation for
4887                                  * delayed allocation blocks. This number gets
4888                                  * adjusted later.  We return if we haven't
4889                                  * allocated blocks already inside this loop.
4890                                  */
4891                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4892                                                 mp, NULL, ip, (long)alen, 0,
4893                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4894                                                      XFS_QMOPT_RES_REGBLKS))) {
4895                                         if (n == 0) {
4896                                                 *nmap = 0;
4897                                                 ASSERT(cur == NULL);
4898                                                 return error;
4899                                         }
4900                                         break;
4901                                 }
4902
4903                                 /*
4904                                  * Split changing sb for alen and indlen since
4905                                  * they could be coming from different places.
4906                                  */
4907                                 indlen = (xfs_extlen_t)
4908                                         xfs_bmap_worst_indlen(ip, alen);
4909                                 ASSERT(indlen > 0);
4910
4911                                 if (rt) {
4912                                         error = xfs_mod_incore_sb(mp,
4913                                                         XFS_SBS_FREXTENTS,
4914                                                         -((int64_t)extsz), (flags &
4915                                                         XFS_BMAPI_RSVBLOCKS));
4916                                 } else {
4917                                         error = xfs_mod_incore_sb(mp,
4918                                                         XFS_SBS_FDBLOCKS,
4919                                                         -((int64_t)alen), (flags &
4920                                                         XFS_BMAPI_RSVBLOCKS));
4921                                 }
4922                                 if (!error) {
4923                                         error = xfs_mod_incore_sb(mp,
4924                                                         XFS_SBS_FDBLOCKS,
4925                                                         -((int64_t)indlen), (flags &
4926                                                         XFS_BMAPI_RSVBLOCKS));
4927                                         if (error && rt)
4928                                                 xfs_mod_incore_sb(mp,
4929                                                         XFS_SBS_FREXTENTS,
4930                                                         (int64_t)extsz, (flags &
4931                                                         XFS_BMAPI_RSVBLOCKS));
4932                                         else if (error)
4933                                                 xfs_mod_incore_sb(mp,
4934                                                         XFS_SBS_FDBLOCKS,
4935                                                         (int64_t)alen, (flags &
4936                                                         XFS_BMAPI_RSVBLOCKS));
4937                                 }
4938
4939                                 if (error) {
4940                                         if (XFS_IS_QUOTA_ON(mp))
4941                                                 /* unreserve the blocks now */
4942                                                 (void)
4943                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4944                                                         mp, NULL, ip,
4945                                                         (long)alen, 0, rt ?
4946                                                         XFS_QMOPT_RES_RTBLKS :
4947                                                         XFS_QMOPT_RES_REGBLKS);
4948                                         break;
4949                                 }
4950
4951                                 ip->i_delayed_blks += alen;
4952                                 abno = NULLSTARTBLOCK(indlen);
4953                         } else {
4954                                 /*
4955                                  * If first time, allocate and fill in
4956                                  * once-only bma fields.
4957                                  */
4958                                 if (bma.ip == NULL) {
4959                                         bma.tp = tp;
4960                                         bma.ip = ip;
4961                                         bma.prevp = &prev;
4962                                         bma.gotp = &got;
4963                                         bma.total = total;
4964                                         bma.userdata = 0;
4965                                 }
4966                                 /* Indicate if this is the first user data
4967                                  * in the file, or just any user data.
4968                                  */
4969                                 if (!(flags & XFS_BMAPI_METADATA)) {
4970                                         bma.userdata = (aoff == 0) ?
4971                                                 XFS_ALLOC_INITIAL_USER_DATA :
4972                                                 XFS_ALLOC_USERDATA;
4973                                 }
4974                                 /*
4975                                  * Fill in changeable bma fields.
4976                                  */
4977                                 bma.eof = eof;
4978                                 bma.firstblock = *firstblock;
4979                                 bma.alen = alen;
4980                                 bma.off = aoff;
4981                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4982                                 bma.wasdel = wasdelay;
4983                                 bma.minlen = minlen;
4984                                 bma.low = flist->xbf_low;
4985                                 bma.minleft = minleft;
4986                                 /*
4987                                  * Only want to do the alignment at the
4988                                  * eof if it is userdata and allocation length
4989                                  * is larger than a stripe unit.
4990                                  */
4991                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4992                                     (!(flags & XFS_BMAPI_METADATA)) &&
4993                                     (whichfork == XFS_DATA_FORK)) {
4994                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4995                                                         whichfork, &bma.aeof)))
4996                                                 goto error0;
4997                                 } else
4998                                         bma.aeof = 0;
4999                                 /*
5000                                  * Call allocator.
5001                                  */
5002                                 if ((error = xfs_bmap_alloc(&bma)))
5003                                         goto error0;
5004                                 /*
5005                                  * Copy out result fields.
5006                                  */
5007                                 abno = bma.rval;
5008                                 if ((flist->xbf_low = bma.low))
5009                                         minleft = 0;
5010                                 alen = bma.alen;
5011                                 aoff = bma.off;
5012                                 ASSERT(*firstblock == NULLFSBLOCK ||
5013                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5014                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5015                                        (flist->xbf_low &&
5016                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5017                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5018                                 *firstblock = bma.firstblock;
5019                                 if (cur)
5020                                         cur->bc_private.b.firstblock =
5021                                                 *firstblock;
5022                                 if (abno == NULLFSBLOCK)
5023                                         break;
5024                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5025                                         cur = xfs_btree_init_cursor(mp,
5026                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
5027                                                 ip, whichfork);
5028                                         cur->bc_private.b.firstblock =
5029                                                 *firstblock;
5030                                         cur->bc_private.b.flist = flist;
5031                                 }
5032                                 /*
5033                                  * Bump the number of extents we've allocated
5034                                  * in this call.
5035                                  */
5036                                 nallocs++;
5037                         }
5038                         if (cur)
5039                                 cur->bc_private.b.flags =
5040                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5041                         got.br_startoff = aoff;
5042                         got.br_startblock = abno;
5043                         got.br_blockcount = alen;
5044                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5045                         /*
5046                          * Determine state of extent, and the filesystem.
5047                          * A wasdelay extent has been initialized, so
5048                          * shouldn't be flagged as unwritten.
5049                          */
5050                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5051                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5052                                         got.br_state = XFS_EXT_UNWRITTEN;
5053                         }
5054                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5055                                 firstblock, flist, &tmp_logflags, delta,
5056                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5057                         logflags |= tmp_logflags;
5058                         if (error)
5059                                 goto error0;
5060                         lastx = ifp->if_lastex;
5061                         ep = xfs_iext_get_ext(ifp, lastx);
5062                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5063                         xfs_bmbt_get_all(ep, &got);
5064                         ASSERT(got.br_startoff <= aoff);
5065                         ASSERT(got.br_startoff + got.br_blockcount >=
5066                                 aoff + alen);
5067 #ifdef DEBUG
5068                         if (flags & XFS_BMAPI_DELAY) {
5069                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5070                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5071                         }
5072                         ASSERT(got.br_state == XFS_EXT_NORM ||
5073                                got.br_state == XFS_EXT_UNWRITTEN);
5074 #endif
5075                         /*
5076                          * Fall down into the found allocated space case.
5077                          */
5078                 } else if (inhole) {
5079                         /*
5080                          * Reading in a hole.
5081                          */
5082                         mval->br_startoff = bno;
5083                         mval->br_startblock = HOLESTARTBLOCK;
5084                         mval->br_blockcount =
5085                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5086                         mval->br_state = XFS_EXT_NORM;
5087                         bno += mval->br_blockcount;
5088                         len -= mval->br_blockcount;
5089                         mval++;
5090                         n++;
5091                         continue;
5092                 }
5093                 /*
5094                  * Then deal with the allocated space we found.
5095                  */
5096                 ASSERT(ep != NULL);
5097                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5098                     (got.br_startoff + got.br_blockcount > obno)) {
5099                         if (obno > bno)
5100                                 bno = obno;
5101                         ASSERT((bno >= obno) || (n == 0));
5102                         ASSERT(bno < end);
5103                         mval->br_startoff = bno;
5104                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5105                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5106                                 mval->br_startblock = DELAYSTARTBLOCK;
5107                         } else
5108                                 mval->br_startblock =
5109                                         got.br_startblock +
5110                                         (bno - got.br_startoff);
5111                         /*
5112                          * Return the minimum of what we got and what we
5113                          * asked for for the length.  We can use the len
5114                          * variable here because it is modified below
5115                          * and we could have been there before coming
5116                          * here if the first part of the allocation
5117                          * didn't overlap what was asked for.
5118                          */
5119                         mval->br_blockcount =
5120                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5121                                         (bno - got.br_startoff));
5122                         mval->br_state = got.br_state;
5123                         ASSERT(mval->br_blockcount <= len);
5124                 } else {
5125                         *mval = got;
5126                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5127                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5128                                 mval->br_startblock = DELAYSTARTBLOCK;
5129                         }
5130                 }
5131
5132                 /*
5133                  * Check if writing previously allocated but
5134                  * unwritten extents.
5135                  */
5136                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5137                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5138                         /*
5139                          * Modify (by adding) the state flag, if writing.
5140                          */
5141                         ASSERT(mval->br_blockcount <= len);
5142                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5143                                 cur = xfs_btree_init_cursor(mp,
5144                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5145                                         ip, whichfork);
5146                                 cur->bc_private.b.firstblock =
5147                                         *firstblock;
5148                                 cur->bc_private.b.flist = flist;
5149                         }
5150                         mval->br_state = XFS_EXT_NORM;
5151                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5152                                 firstblock, flist, &tmp_logflags, delta,
5153                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5154                         logflags |= tmp_logflags;
5155                         if (error)
5156                                 goto error0;
5157                         lastx = ifp->if_lastex;
5158                         ep = xfs_iext_get_ext(ifp, lastx);
5159                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5160                         xfs_bmbt_get_all(ep, &got);
5161                         /*
5162                          * We may have combined previously unwritten
5163                          * space with written space, so generate
5164                          * another request.
5165                          */
5166                         if (mval->br_blockcount < len)
5167                                 continue;
5168                 }
5169
5170                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5171                        ((mval->br_startoff + mval->br_blockcount) <= end));
5172                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5173                        (mval->br_blockcount <= len) ||
5174                        (mval->br_startoff < obno));
5175                 bno = mval->br_startoff + mval->br_blockcount;
5176                 len = end - bno;
5177                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5178                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5179                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5180                         ASSERT(mval->br_state == mval[-1].br_state);
5181                         mval[-1].br_blockcount = mval->br_blockcount;
5182                         mval[-1].br_state = mval->br_state;
5183                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5184                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5185                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5186                            mval->br_startblock ==
5187                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5188                            ((flags & XFS_BMAPI_IGSTATE) ||
5189                                 mval[-1].br_state == mval->br_state)) {
5190                         ASSERT(mval->br_startoff ==
5191                                mval[-1].br_startoff + mval[-1].br_blockcount);
5192                         mval[-1].br_blockcount += mval->br_blockcount;
5193                 } else if (n > 0 &&
5194                            mval->br_startblock == DELAYSTARTBLOCK &&
5195                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5196                            mval->br_startoff ==
5197                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5198                         mval[-1].br_blockcount += mval->br_blockcount;
5199                         mval[-1].br_state = mval->br_state;
5200                 } else if (!((n == 0) &&
5201                              ((mval->br_startoff + mval->br_blockcount) <=
5202                               obno))) {
5203                         mval++;
5204                         n++;
5205                 }
5206                 /*
5207                  * If we're done, stop now.  Stop when we've allocated
5208                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5209                  * the transaction may get too big.
5210                  */
5211                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5212                         break;
5213                 /*
5214                  * Else go on to the next record.
5215                  */
5216                 ep = xfs_iext_get_ext(ifp, ++lastx);
5217                 prev = got;
5218                 if (lastx >= nextents)
5219                         eof = 1;
5220                 else
5221                         xfs_bmbt_get_all(ep, &got);
5222         }
5223         ifp->if_lastex = lastx;
5224         *nmap = n;
5225         /*
5226          * Transform from btree to extents, give it cur.
5227          */
5228         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5229             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5230                 ASSERT(wr && cur);
5231                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5232                         &tmp_logflags, whichfork);
5233                 logflags |= tmp_logflags;
5234                 if (error)
5235                         goto error0;
5236         }
5237         ASSERT(ifp->if_ext_max ==
5238                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5239         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5240                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5241         error = 0;
5242         if (delta && delta->xed_startoff != NULLFILEOFF) {
5243                 /* A change was actually made.
5244                  * Note that delta->xed_blockount is an offset at this
5245                  * point and needs to be converted to a block count.
5246                  */
5247                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5248                 delta->xed_blockcount -= delta->xed_startoff;
5249         }
5250 error0:
5251         /*
5252          * Log everything.  Do this after conversion, there's no point in
5253          * logging the extent records if we've converted to btree format.
5254          */
5255         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5256             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5257                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5258         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5259                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5260                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5261         /*
5262          * Log whatever the flags say, even if error.  Otherwise we might miss
5263          * detecting a case where the data is changed, there's an error,
5264          * and it's not logged so we don't shutdown when we should.
5265          */
5266         if (logflags) {
5267                 ASSERT(tp && wr);
5268                 xfs_trans_log_inode(tp, ip, logflags);
5269         }
5270         if (cur) {
5271                 if (!error) {
5272                         ASSERT(*firstblock == NULLFSBLOCK ||
5273                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5274                                XFS_FSB_TO_AGNO(mp,
5275                                        cur->bc_private.b.firstblock) ||
5276                                (flist->xbf_low &&
5277                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5278                                 XFS_FSB_TO_AGNO(mp,
5279                                         cur->bc_private.b.firstblock)));
5280                         *firstblock = cur->bc_private.b.firstblock;
5281                 }
5282                 xfs_btree_del_cursor(cur,
5283                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5284         }
5285         if (!error)
5286                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5287                         orig_nmap, *nmap);
5288         return error;
5289 }
5290
5291 /*
5292  * Map file blocks to filesystem blocks, simple version.
5293  * One block (extent) only, read-only.
5294  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5295  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5296  * was set and all the others were clear.
5297  */
5298 int                                             /* error */
5299 xfs_bmapi_single(
5300         xfs_trans_t     *tp,            /* transaction pointer */
5301         xfs_inode_t     *ip,            /* incore inode */
5302         int             whichfork,      /* data or attr fork */
5303         xfs_fsblock_t   *fsb,           /* output: mapped block */
5304         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5305 {
5306         int             eof;            /* we've hit the end of extents */
5307         int             error;          /* error return */
5308         xfs_bmbt_irec_t got;            /* current file extent record */
5309         xfs_ifork_t     *ifp;           /* inode fork pointer */
5310         xfs_extnum_t    lastx;          /* last useful extent number */
5311         xfs_bmbt_irec_t prev;           /* previous file extent record */
5312
5313         ifp = XFS_IFORK_PTR(ip, whichfork);
5314         if (unlikely(
5315             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5316             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5317                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5318                                 ip->i_mount);
5319                return XFS_ERROR(EFSCORRUPTED);
5320         }
5321         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5322                 return XFS_ERROR(EIO);
5323         XFS_STATS_INC(xs_blk_mapr);
5324         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5325             (error = xfs_iread_extents(tp, ip, whichfork)))
5326                 return error;
5327         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5328                 &prev);
5329         /*
5330          * Reading past eof, act as though there's a hole
5331          * up to end.
5332          */
5333         if (eof || got.br_startoff > bno) {
5334                 *fsb = NULLFSBLOCK;
5335                 return 0;
5336         }
5337         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5338         ASSERT(bno < got.br_startoff + got.br_blockcount);
5339         *fsb = got.br_startblock + (bno - got.br_startoff);
5340         ifp->if_lastex = lastx;
5341         return 0;
5342 }
5343
5344 /*
5345  * Unmap (remove) blocks from a file.
5346  * If nexts is nonzero then the number of extents to remove is limited to
5347  * that value.  If not all extents in the block range can be removed then
5348  * *done is set.
5349  */
5350 int                                             /* error */
5351 xfs_bunmapi(
5352         xfs_trans_t             *tp,            /* transaction pointer */
5353         struct xfs_inode        *ip,            /* incore inode */
5354         xfs_fileoff_t           bno,            /* starting offset to unmap */
5355         xfs_filblks_t           len,            /* length to unmap in file */
5356         int                     flags,          /* misc flags */
5357         xfs_extnum_t            nexts,          /* number of extents max */
5358         xfs_fsblock_t           *firstblock,    /* first allocated block
5359                                                    controls a.g. for allocs */
5360         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5361         xfs_extdelta_t          *delta,         /* o: change made to incore
5362                                                    extents */
5363         int                     *done)          /* set if not done yet */
5364 {
5365         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5366         xfs_bmbt_irec_t         del;            /* extent being deleted */
5367         int                     eof;            /* is deleting at eof */
5368         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5369         int                     error;          /* error return value */
5370         xfs_extnum_t            extno;          /* extent number in list */
5371         xfs_bmbt_irec_t         got;            /* current extent record */
5372         xfs_ifork_t             *ifp;           /* inode fork pointer */
5373         int                     isrt;           /* freeing in rt area */
5374         xfs_extnum_t            lastx;          /* last extent index used */
5375         int                     logflags;       /* transaction logging flags */
5376         xfs_extlen_t            mod;            /* rt extent offset */
5377         xfs_mount_t             *mp;            /* mount structure */
5378         xfs_extnum_t            nextents;       /* number of file extents */
5379         xfs_bmbt_irec_t         prev;           /* previous extent record */
5380         xfs_fileoff_t           start;          /* first file offset deleted */
5381         int                     tmp_logflags;   /* partial logging flags */
5382         int                     wasdel;         /* was a delayed alloc extent */
5383         int                     whichfork;      /* data or attribute fork */
5384         int                     rsvd;           /* OK to allocate reserved blocks */
5385         xfs_fsblock_t           sum;
5386
5387         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5388         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5389                 XFS_ATTR_FORK : XFS_DATA_FORK;
5390         ifp = XFS_IFORK_PTR(ip, whichfork);
5391         if (unlikely(
5392             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5393             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5394                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5395                                  ip->i_mount);
5396                 return XFS_ERROR(EFSCORRUPTED);
5397         }
5398         mp = ip->i_mount;
5399         if (XFS_FORCED_SHUTDOWN(mp))
5400                 return XFS_ERROR(EIO);
5401         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5402         ASSERT(len > 0);
5403         ASSERT(nexts >= 0);
5404         ASSERT(ifp->if_ext_max ==
5405                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5406         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5407             (error = xfs_iread_extents(tp, ip, whichfork)))
5408                 return error;
5409         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5410         if (nextents == 0) {
5411                 *done = 1;
5412                 return 0;
5413         }
5414         XFS_STATS_INC(xs_blk_unmap);
5415         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5416         start = bno;
5417         bno = start + len - 1;
5418         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5419                 &prev);
5420         if (delta) {
5421                 delta->xed_startoff = NULLFILEOFF;
5422                 delta->xed_blockcount = 0;
5423         }
5424         /*
5425          * Check to see if the given block number is past the end of the
5426          * file, back up to the last block if so...
5427          */
5428         if (eof) {
5429                 ep = xfs_iext_get_ext(ifp, --lastx);
5430                 xfs_bmbt_get_all(ep, &got);
5431                 bno = got.br_startoff + got.br_blockcount - 1;
5432         }
5433         logflags = 0;
5434         if (ifp->if_flags & XFS_IFBROOT) {
5435                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5436                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5437                         whichfork);
5438                 cur->bc_private.b.firstblock = *firstblock;
5439                 cur->bc_private.b.flist = flist;
5440                 cur->bc_private.b.flags = 0;
5441         } else
5442                 cur = NULL;
5443         extno = 0;
5444         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5445                (nexts == 0 || extno < nexts)) {
5446                 /*
5447                  * Is the found extent after a hole in which bno lives?
5448                  * Just back up to the previous extent, if so.
5449                  */
5450                 if (got.br_startoff > bno) {
5451                         if (--lastx < 0)
5452                                 break;
5453                         ep = xfs_iext_get_ext(ifp, lastx);
5454                         xfs_bmbt_get_all(ep, &got);
5455                 }
5456                 /*
5457                  * Is the last block of this extent before the range
5458                  * we're supposed to delete?  If so, we're done.
5459                  */
5460                 bno = XFS_FILEOFF_MIN(bno,
5461                         got.br_startoff + got.br_blockcount - 1);
5462                 if (bno < start)
5463                         break;
5464                 /*
5465                  * Then deal with the (possibly delayed) allocated space
5466                  * we found.
5467                  */
5468                 ASSERT(ep != NULL);
5469                 del = got;
5470                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5471                 if (got.br_startoff < start) {
5472                         del.br_startoff = start;
5473                         del.br_blockcount -= start - got.br_startoff;
5474                         if (!wasdel)
5475                                 del.br_startblock += start - got.br_startoff;
5476                 }
5477                 if (del.br_startoff + del.br_blockcount > bno + 1)
5478                         del.br_blockcount = bno + 1 - del.br_startoff;
5479                 sum = del.br_startblock + del.br_blockcount;
5480                 if (isrt &&
5481                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5482                         /*
5483                          * Realtime extent not lined up at the end.
5484                          * The extent could have been split into written
5485                          * and unwritten pieces, or we could just be
5486                          * unmapping part of it.  But we can't really
5487                          * get rid of part of a realtime extent.
5488                          */
5489                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5490                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5491                                 /*
5492                                  * This piece is unwritten, or we're not
5493                                  * using unwritten extents.  Skip over it.
5494                                  */
5495                                 ASSERT(bno >= mod);
5496                                 bno -= mod > del.br_blockcount ?
5497                                         del.br_blockcount : mod;
5498                                 if (bno < got.br_startoff) {
5499                                         if (--lastx >= 0)
5500                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5501                                                         ifp, lastx), &got);
5502                                 }
5503                                 continue;
5504                         }
5505                         /*
5506                          * It's written, turn it unwritten.
5507                          * This is better than zeroing it.
5508                          */
5509                         ASSERT(del.br_state == XFS_EXT_NORM);
5510                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5511                         /*
5512                          * If this spans a realtime extent boundary,
5513                          * chop it back to the start of the one we end at.
5514                          */
5515                         if (del.br_blockcount > mod) {
5516                                 del.br_startoff += del.br_blockcount - mod;
5517                                 del.br_startblock += del.br_blockcount - mod;
5518                                 del.br_blockcount = mod;
5519                         }
5520                         del.br_state = XFS_EXT_UNWRITTEN;
5521                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5522                                 firstblock, flist, &logflags, delta,
5523                                 XFS_DATA_FORK, 0);
5524                         if (error)
5525                                 goto error0;
5526                         goto nodelete;
5527                 }
5528                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5529                         /*
5530                          * Realtime extent is lined up at the end but not
5531                          * at the front.  We'll get rid of full extents if
5532                          * we can.
5533                          */
5534                         mod = mp->m_sb.sb_rextsize - mod;
5535                         if (del.br_blockcount > mod) {
5536                                 del.br_blockcount -= mod;
5537                                 del.br_startoff += mod;
5538                                 del.br_startblock += mod;
5539                         } else if ((del.br_startoff == start &&
5540                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5541                                      xfs_trans_get_block_res(tp) == 0)) ||
5542                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5543                                 /*
5544                                  * Can't make it unwritten.  There isn't
5545                                  * a full extent here so just skip it.
5546                                  */
5547                                 ASSERT(bno >= del.br_blockcount);
5548                                 bno -= del.br_blockcount;
5549                                 if (bno < got.br_startoff) {
5550                                         if (--lastx >= 0)
5551                                                 xfs_bmbt_get_all(--ep, &got);
5552                                 }
5553                                 continue;
5554                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5555                                 /*
5556                                  * This one is already unwritten.
5557                                  * It must have a written left neighbor.
5558                                  * Unwrite the killed part of that one and
5559                                  * try again.
5560                                  */
5561                                 ASSERT(lastx > 0);
5562                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5563                                                 lastx - 1), &prev);
5564                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5565                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5566                                 ASSERT(del.br_startblock ==
5567                                        prev.br_startblock + prev.br_blockcount);
5568                                 if (prev.br_startoff < start) {
5569                                         mod = start - prev.br_startoff;
5570                                         prev.br_blockcount -= mod;
5571                                         prev.br_startblock += mod;
5572                                         prev.br_startoff = start;
5573                                 }
5574                                 prev.br_state = XFS_EXT_UNWRITTEN;
5575                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5576                                         &prev, firstblock, flist, &logflags,
5577                                         delta, XFS_DATA_FORK, 0);
5578                                 if (error)
5579                                         goto error0;
5580                                 goto nodelete;
5581                         } else {
5582                                 ASSERT(del.br_state == XFS_EXT_NORM);
5583                                 del.br_state = XFS_EXT_UNWRITTEN;
5584                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5585                                         &del, firstblock, flist, &logflags,
5586                                         delta, XFS_DATA_FORK, 0);
5587                                 if (error)
5588                                         goto error0;
5589                                 goto nodelete;
5590                         }
5591                 }
5592                 if (wasdel) {
5593                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5594                         /* Update realtime/data freespace, unreserve quota */
5595                         if (isrt) {
5596                                 xfs_filblks_t rtexts;
5597
5598                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5599                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5600                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5601                                                 (int64_t)rtexts, rsvd);
5602                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5603                                         NULL, ip, -((long)del.br_blockcount), 0,
5604                                         XFS_QMOPT_RES_RTBLKS);
5605                         } else {
5606                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5607                                                 (int64_t)del.br_blockcount, rsvd);
5608                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5609                                         NULL, ip, -((long)del.br_blockcount), 0,
5610                                         XFS_QMOPT_RES_REGBLKS);
5611                         }
5612                         ip->i_delayed_blks -= del.br_blockcount;
5613                         if (cur)
5614                                 cur->bc_private.b.flags |=
5615                                         XFS_BTCUR_BPRV_WASDEL;
5616                 } else if (cur)
5617                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5618                 /*
5619                  * If it's the case where the directory code is running
5620                  * with no block reservation, and the deleted block is in
5621                  * the middle of its extent, and the resulting insert
5622                  * of an extent would cause transformation to btree format,
5623                  * then reject it.  The calling code will then swap
5624                  * blocks around instead.
5625                  * We have to do this now, rather than waiting for the
5626                  * conversion to btree format, since the transaction
5627                  * will be dirty.
5628                  */
5629                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5630                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5631                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5632                     del.br_startoff > got.br_startoff &&
5633                     del.br_startoff + del.br_blockcount <
5634                     got.br_startoff + got.br_blockcount) {
5635                         error = XFS_ERROR(ENOSPC);
5636                         goto error0;
5637                 }
5638                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5639                                 &tmp_logflags, delta, whichfork, rsvd);
5640                 logflags |= tmp_logflags;
5641                 if (error)
5642                         goto error0;
5643                 bno = del.br_startoff - 1;
5644 nodelete:
5645                 lastx = ifp->if_lastex;
5646                 /*
5647                  * If not done go on to the next (previous) record.
5648                  * Reset ep in case the extents array was re-alloced.
5649                  */
5650                 ep = xfs_iext_get_ext(ifp, lastx);
5651                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5652                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5653                             xfs_bmbt_get_startoff(ep) > bno) {
5654                                 if (--lastx >= 0)
5655                                         ep = xfs_iext_get_ext(ifp, lastx);
5656                         }
5657                         if (lastx >= 0)
5658                                 xfs_bmbt_get_all(ep, &got);
5659                         extno++;
5660                 }
5661         }
5662         ifp->if_lastex = lastx;
5663         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5664         ASSERT(ifp->if_ext_max ==
5665                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5666         /*
5667          * Convert to a btree if necessary.
5668          */
5669         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5670             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5671                 ASSERT(cur == NULL);
5672                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5673                         &cur, 0, &tmp_logflags, whichfork);
5674                 logflags |= tmp_logflags;
5675                 if (error)
5676                         goto error0;
5677         }
5678         /*
5679          * transform from btree to extents, give it cur
5680          */
5681         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5682                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5683                 ASSERT(cur != NULL);
5684                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5685                         whichfork);
5686                 logflags |= tmp_logflags;
5687                 if (error)
5688                         goto error0;
5689         }
5690         /*
5691          * transform from extents to local?
5692          */
5693         ASSERT(ifp->if_ext_max ==
5694                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5695         error = 0;
5696         if (delta && delta->xed_startoff != NULLFILEOFF) {
5697                 /* A change was actually made.
5698                  * Note that delta->xed_blockount is an offset at this
5699                  * point and needs to be converted to a block count.
5700                  */
5701                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5702                 delta->xed_blockcount -= delta->xed_startoff;
5703         }
5704 error0:
5705         /*
5706          * Log everything.  Do this after conversion, there's no point in
5707          * logging the extent records if we've converted to btree format.
5708          */
5709         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5710             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5711                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5712         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5713                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5714                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5715         /*
5716          * Log inode even in the error case, if the transaction
5717          * is dirty we'll need to shut down the filesystem.
5718          */
5719         if (logflags)
5720                 xfs_trans_log_inode(tp, ip, logflags);
5721         if (cur) {
5722                 if (!error) {
5723                         *firstblock = cur->bc_private.b.firstblock;
5724                         cur->bc_private.b.allocated = 0;
5725                 }
5726                 xfs_btree_del_cursor(cur,
5727                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5728         }
5729         return error;
5730 }
5731
5732 /*
5733  * returns 1 for success, 0 if we failed to map the extent.
5734  */
5735 STATIC int
5736 xfs_getbmapx_fix_eof_hole(
5737         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5738         struct getbmap          *out,           /* output structure */
5739         int                     prealloced,     /* this is a file with
5740                                                 * preallocated data space */
5741         __int64_t               end,            /* last block requested */
5742         xfs_fsblock_t           startblock)
5743 {
5744         __int64_t               fixlen;
5745         xfs_mount_t             *mp;            /* file system mount point */
5746
5747         if (startblock == HOLESTARTBLOCK) {
5748                 mp = ip->i_mount;
5749                 out->bmv_block = -1;
5750                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5751                 fixlen -= out->bmv_offset;
5752                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5753                         /* Came to hole at EOF. Trim it. */
5754                         if (fixlen <= 0)
5755                                 return 0;
5756                         out->bmv_length = fixlen;
5757                 }
5758         } else {
5759                 out->bmv_block = XFS_FSB_TO_DB(ip, startblock);
5760         }
5761
5762         return 1;
5763 }
5764
5765 /*
5766  * Fcntl interface to xfs_bmapi.
5767  */
5768 int                                             /* error code */
5769 xfs_getbmap(
5770         xfs_inode_t             *ip,
5771         struct getbmap          *bmv,           /* user bmap structure */
5772         void                    __user *ap,     /* pointer to user's array */
5773         int                     interface)      /* interface flags */
5774 {
5775         __int64_t               bmvend;         /* last block requested */
5776         int                     error;          /* return value */
5777         __int64_t               fixlen;         /* length for -1 case */
5778         int                     i;              /* extent number */
5779         int                     lock;           /* lock state */
5780         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5781         xfs_mount_t             *mp;            /* file system mount point */
5782         int                     nex;            /* # of user extents can do */
5783         int                     nexleft;        /* # of user extents left */
5784         int                     subnex;         /* # of bmapi's can do */
5785         int                     nmap;           /* number of map entries */
5786         struct getbmap          out;            /* output structure */
5787         int                     whichfork;      /* data or attr fork */
5788         int                     prealloced;     /* this is a file with
5789                                                  * preallocated data space */
5790         int                     sh_unwritten;   /* true, if unwritten */
5791                                                 /* extents listed separately */
5792         int                     bmapi_flags;    /* flags for xfs_bmapi */
5793         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5794
5795         mp = ip->i_mount;
5796
5797         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5798         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5799
5800         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5801          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5802          *      bit is set for the file, generate a read event in order
5803          *      that the DMAPI application may do its thing before we return
5804          *      the extents.  Usually this means restoring user file data to
5805          *      regions of the file that look like holes.
5806          *
5807          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5808          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5809          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5810          *      could misinterpret holes in a DMAPI file as true holes,
5811          *      when in fact they may represent offline user data.
5812          */
5813         if ((interface & BMV_IF_NO_DMAPI_READ) == 0 &&
5814             DM_EVENT_ENABLED(ip, DM_EVENT_READ) &&
5815             whichfork == XFS_DATA_FORK) {
5816                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, 0, 0, 0, NULL);
5817                 if (error)
5818                         return XFS_ERROR(error);
5819         }
5820
5821         if (whichfork == XFS_ATTR_FORK) {
5822                 if (XFS_IFORK_Q(ip)) {
5823                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5824                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5825                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5826                                 return XFS_ERROR(EINVAL);
5827                 } else if (unlikely(
5828                            ip->i_d.di_aformat != 0 &&
5829                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5830                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5831                                          ip->i_mount);
5832                         return XFS_ERROR(EFSCORRUPTED);
5833                 }
5834         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5835                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5836                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5837                 return XFS_ERROR(EINVAL);
5838         if (whichfork == XFS_DATA_FORK) {
5839                 if (xfs_get_extsz_hint(ip) ||
5840                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5841                         prealloced = 1;
5842                         fixlen = XFS_MAXIOFFSET(mp);
5843                 } else {
5844                         prealloced = 0;
5845                         fixlen = ip->i_size;
5846                 }
5847         } else {
5848                 prealloced = 0;
5849                 fixlen = 1LL << 32;
5850         }
5851
5852         if (bmv->bmv_length == -1) {
5853                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5854                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5855                                         (__int64_t)0);
5856         } else if (bmv->bmv_length < 0)
5857                 return XFS_ERROR(EINVAL);
5858         if (bmv->bmv_length == 0) {
5859                 bmv->bmv_entries = 0;
5860                 return 0;
5861         }
5862         nex = bmv->bmv_count - 1;
5863         if (nex <= 0)
5864                 return XFS_ERROR(EINVAL);
5865         bmvend = bmv->bmv_offset + bmv->bmv_length;
5866
5867         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5868
5869         if (whichfork == XFS_DATA_FORK &&
5870                 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5871                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5872                 error = xfs_flush_pages(ip, (xfs_off_t)0,
5873                                                -1, 0, FI_REMAPF);
5874                 if (error) {
5875                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5876                 return error;
5877                 }
5878         }
5879
5880         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5881
5882         lock = xfs_ilock_map_shared(ip);
5883
5884         /*
5885          * Don't let nex be bigger than the number of extents
5886          * we can have assuming alternating holes and real extents.
5887          */
5888         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5889                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5890
5891         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5892                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5893
5894         /*
5895          * Allocate enough space to handle "subnex" maps at a time.
5896          */
5897         subnex = 16;
5898         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5899
5900         bmv->bmv_entries = 0;
5901
5902         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5903                 error = 0;
5904                 goto unlock_and_return;
5905         }
5906
5907         nexleft = nex;
5908
5909         do {
5910                 nmap = (nexleft > subnex) ? subnex : nexleft;
5911                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5912                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5913                                   bmapi_flags, NULL, 0, map, &nmap,
5914                                   NULL, NULL);
5915                 if (error)
5916                         goto unlock_and_return;
5917                 ASSERT(nmap <= subnex);
5918
5919                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5920                         nexleft--;
5921                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5922                                         BMV_OF_PREALLOC : 0;
5923                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5924                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5925                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5926                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5927                             whichfork == XFS_ATTR_FORK) {
5928                                 /* came to the end of attribute fork */
5929                                 goto unlock_and_return;
5930                         } else {
5931                                 if (!xfs_getbmapx_fix_eof_hole(ip, &out,
5932                                                         prealloced, bmvend,
5933                                                         map[i].br_startblock)) {
5934                                         goto unlock_and_return;
5935                                 }
5936
5937                                 /* return either getbmap/getbmapx structure. */
5938                                 if (interface & BMV_IF_EXTENDED) {
5939                                         struct  getbmapx        outx;
5940
5941                                         GETBMAP_CONVERT(out,outx);
5942                                         outx.bmv_oflags = oflags;
5943                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5944                                         if (copy_to_user(ap, &outx,
5945                                                         sizeof(outx))) {
5946                                                 error = XFS_ERROR(EFAULT);
5947                                                 goto unlock_and_return;
5948                                         }
5949                                 } else {
5950                                         if (copy_to_user(ap, &out,
5951                                                         sizeof(out))) {
5952                                                 error = XFS_ERROR(EFAULT);
5953                                                 goto unlock_and_return;
5954                                         }
5955                                 }
5956                                 bmv->bmv_offset =
5957                                         out.bmv_offset + out.bmv_length;
5958                                 bmv->bmv_length = MAX((__int64_t)0,
5959                                         (__int64_t)(bmvend - bmv->bmv_offset));
5960                                 bmv->bmv_entries++;
5961                                 ap = (interface & BMV_IF_EXTENDED) ?
5962                                                 (void __user *)
5963                                         ((struct getbmapx __user *)ap + 1) :
5964                                                 (void __user *)
5965                                         ((struct getbmap __user *)ap + 1);
5966                         }
5967                 }
5968         } while (nmap && nexleft && bmv->bmv_length);
5969
5970 unlock_and_return:
5971         xfs_iunlock_map_shared(ip, lock);
5972         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5973
5974         kmem_free(map);
5975
5976         return error;
5977 }
5978
5979 /*
5980  * Check the last inode extent to determine whether this allocation will result
5981  * in blocks being allocated at the end of the file. When we allocate new data
5982  * blocks at the end of the file which do not start at the previous data block,
5983  * we will try to align the new blocks at stripe unit boundaries.
5984  */
5985 STATIC int                              /* error */
5986 xfs_bmap_isaeof(
5987         xfs_inode_t     *ip,            /* incore inode pointer */
5988         xfs_fileoff_t   off,            /* file offset in fsblocks */
5989         int             whichfork,      /* data or attribute fork */
5990         char            *aeof)          /* return value */
5991 {
5992         int             error;          /* error return value */
5993         xfs_ifork_t     *ifp;           /* inode fork pointer */
5994         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5995         xfs_extnum_t    nextents;       /* number of file extents */
5996         xfs_bmbt_irec_t s;              /* expanded extent record */
5997
5998         ASSERT(whichfork == XFS_DATA_FORK);
5999         ifp = XFS_IFORK_PTR(ip, whichfork);
6000         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6001             (error = xfs_iread_extents(NULL, ip, whichfork)))
6002                 return error;
6003         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6004         if (nextents == 0) {
6005                 *aeof = 1;
6006                 return 0;
6007         }
6008         /*
6009          * Go to the last extent
6010          */
6011         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6012         xfs_bmbt_get_all(lastrec, &s);
6013         /*
6014          * Check we are allocating in the last extent (for delayed allocations)
6015          * or past the last extent for non-delayed allocations.
6016          */
6017         *aeof = (off >= s.br_startoff &&
6018                  off < s.br_startoff + s.br_blockcount &&
6019                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6020                 off >= s.br_startoff + s.br_blockcount;
6021         return 0;
6022 }
6023
6024 /*
6025  * Check if the endoff is outside the last extent. If so the caller will grow
6026  * the allocation to a stripe unit boundary.
6027  */
6028 int                                     /* error */
6029 xfs_bmap_eof(
6030         xfs_inode_t     *ip,            /* incore inode pointer */
6031         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6032         int             whichfork,      /* data or attribute fork */
6033         int             *eof)           /* result value */
6034 {
6035         xfs_fsblock_t   blockcount;     /* extent block count */
6036         int             error;          /* error return value */
6037         xfs_ifork_t     *ifp;           /* inode fork pointer */
6038         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6039         xfs_extnum_t    nextents;       /* number of file extents */
6040         xfs_fileoff_t   startoff;       /* extent starting file offset */
6041
6042         ASSERT(whichfork == XFS_DATA_FORK);
6043         ifp = XFS_IFORK_PTR(ip, whichfork);
6044         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6045             (error = xfs_iread_extents(NULL, ip, whichfork)))
6046                 return error;
6047         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6048         if (nextents == 0) {
6049                 *eof = 1;
6050                 return 0;
6051         }
6052         /*
6053          * Go to the last extent
6054          */
6055         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6056         startoff = xfs_bmbt_get_startoff(lastrec);
6057         blockcount = xfs_bmbt_get_blockcount(lastrec);
6058         *eof = endoff >= startoff + blockcount;
6059         return 0;
6060 }
6061
6062 #ifdef DEBUG
6063 STATIC
6064 xfs_buf_t *
6065 xfs_bmap_get_bp(
6066         xfs_btree_cur_t         *cur,
6067         xfs_fsblock_t           bno)
6068 {
6069         int i;
6070         xfs_buf_t *bp;
6071
6072         if (!cur)
6073                 return(NULL);
6074
6075         bp = NULL;
6076         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6077                 bp = cur->bc_bufs[i];
6078                 if (!bp) break;
6079                 if (XFS_BUF_ADDR(bp) == bno)
6080                         break;  /* Found it */
6081         }
6082         if (i == XFS_BTREE_MAXLEVELS)
6083                 bp = NULL;
6084
6085         if (!bp) { /* Chase down all the log items to see if the bp is there */
6086                 xfs_log_item_chunk_t    *licp;
6087                 xfs_trans_t             *tp;
6088
6089                 tp = cur->bc_tp;
6090                 licp = &tp->t_items;
6091                 while (!bp && licp != NULL) {
6092                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
6093                                 licp = licp->lic_next;
6094                                 continue;
6095                         }
6096                         for (i = 0; i < licp->lic_unused; i++) {
6097                                 xfs_log_item_desc_t     *lidp;
6098                                 xfs_log_item_t          *lip;
6099                                 xfs_buf_log_item_t      *bip;
6100                                 xfs_buf_t               *lbp;
6101
6102                                 if (XFS_LIC_ISFREE(licp, i)) {
6103                                         continue;
6104                                 }
6105
6106                                 lidp = XFS_LIC_SLOT(licp, i);
6107                                 lip = lidp->lid_item;
6108                                 if (lip->li_type != XFS_LI_BUF)
6109                                         continue;
6110
6111                                 bip = (xfs_buf_log_item_t *)lip;
6112                                 lbp = bip->bli_buf;
6113
6114                                 if (XFS_BUF_ADDR(lbp) == bno) {
6115                                         bp = lbp;
6116                                         break; /* Found it */
6117                                 }
6118                         }
6119                         licp = licp->lic_next;
6120                 }
6121         }
6122         return(bp);
6123 }
6124
6125 void
6126 xfs_check_block(
6127         xfs_bmbt_block_t        *block,
6128         xfs_mount_t             *mp,
6129         int                     root,
6130         short                   sz)
6131 {
6132         int                     i, j, dmxr;
6133         __be64                  *pp, *thispa;   /* pointer to block address */
6134         xfs_bmbt_key_t          *prevp, *keyp;
6135
6136         ASSERT(be16_to_cpu(block->bb_level) > 0);
6137
6138         prevp = NULL;
6139         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6140                 dmxr = mp->m_bmap_dmxr[0];
6141
6142                 if (root) {
6143                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6144                 } else {
6145                         keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6146                 }
6147
6148                 if (prevp) {
6149                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6150                 }
6151                 prevp = keyp;
6152
6153                 /*
6154                  * Compare the block numbers to see if there are dups.
6155                  */
6156
6157                 if (root) {
6158                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6159                 } else {
6160                         pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6161                 }
6162                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6163                         if (root) {
6164                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6165                         } else {
6166                                 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6167                                                             dmxr);
6168                         }
6169                         if (*thispa == *pp) {
6170                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6171                                         __func__, j, i,
6172                                         (unsigned long long)be64_to_cpu(*thispa));
6173                                 panic("%s: ptrs are equal in node\n",
6174                                         __func__);
6175                         }
6176                 }
6177         }
6178 }
6179
6180 /*
6181  * Check that the extents for the inode ip are in the right order in all
6182  * btree leaves.
6183  */
6184
6185 STATIC void
6186 xfs_bmap_check_leaf_extents(
6187         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6188         xfs_inode_t             *ip,            /* incore inode pointer */
6189         int                     whichfork)      /* data or attr fork */
6190 {
6191         xfs_bmbt_block_t        *block; /* current btree block */
6192         xfs_fsblock_t           bno;    /* block # of "block" */
6193         xfs_buf_t               *bp;    /* buffer for "block" */
6194         int                     error;  /* error return value */
6195         xfs_extnum_t            i=0, j; /* index into the extents list */
6196         xfs_ifork_t             *ifp;   /* fork structure */
6197         int                     level;  /* btree level, for checking */
6198         xfs_mount_t             *mp;    /* file system mount structure */
6199         __be64                  *pp;    /* pointer to block address */
6200         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6201         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
6202         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6203         int                     bp_release = 0;
6204
6205         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6206                 return;
6207         }
6208
6209         bno = NULLFSBLOCK;
6210         mp = ip->i_mount;
6211         ifp = XFS_IFORK_PTR(ip, whichfork);
6212         block = ifp->if_broot;
6213         /*
6214          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6215          */
6216         level = be16_to_cpu(block->bb_level);
6217         ASSERT(level > 0);
6218         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6219         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6220         bno = be64_to_cpu(*pp);
6221
6222         ASSERT(bno != NULLDFSBNO);
6223         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6224         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6225
6226         /*
6227          * Go down the tree until leaf level is reached, following the first
6228          * pointer (leftmost) at each level.
6229          */
6230         while (level-- > 0) {
6231                 /* See if buf is in cur first */
6232                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6233                 if (bp) {
6234                         bp_release = 0;
6235                 } else {
6236                         bp_release = 1;
6237                 }
6238                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6239                                 XFS_BMAP_BTREE_REF)))
6240                         goto error_norelse;
6241                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6242                 XFS_WANT_CORRUPTED_GOTO(
6243                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6244                         error0);
6245                 if (level == 0)
6246                         break;
6247
6248                 /*
6249                  * Check this block for basic sanity (increasing keys and
6250                  * no duplicate blocks).
6251                  */
6252
6253                 xfs_check_block(block, mp, 0, 0);
6254                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6255                 bno = be64_to_cpu(*pp);
6256                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6257                 if (bp_release) {
6258                         bp_release = 0;
6259                         xfs_trans_brelse(NULL, bp);
6260                 }
6261         }
6262
6263         /*
6264          * Here with bp and block set to the leftmost leaf node in the tree.
6265          */
6266         i = 0;
6267
6268         /*
6269          * Loop over all leaf nodes checking that all extents are in the right order.
6270          */
6271         for (;;) {
6272                 xfs_fsblock_t   nextbno;
6273                 xfs_extnum_t    num_recs;
6274
6275
6276                 num_recs = be16_to_cpu(block->bb_numrecs);
6277
6278                 /*
6279                  * Read-ahead the next leaf block, if any.
6280                  */
6281
6282                 nextbno = be64_to_cpu(block->bb_rightsib);
6283
6284                 /*
6285                  * Check all the extents to make sure they are OK.
6286                  * If we had a previous block, the last entry should
6287                  * conform with the first entry in this one.
6288                  */
6289
6290                 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6291                 if (i) {
6292                         xfs_btree_check_rec(XFS_BTNUM_BMAP, &last, ep);
6293                 }
6294                 for (j = 1; j < num_recs; j++) {
6295                         nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6296                         xfs_btree_check_rec(XFS_BTNUM_BMAP, ep, nextp);
6297                         ep = nextp;
6298                 }
6299
6300                 last = *ep;
6301                 i += num_recs;
6302                 if (bp_release) {
6303                         bp_release = 0;
6304                         xfs_trans_brelse(NULL, bp);
6305                 }
6306                 bno = nextbno;
6307                 /*
6308                  * If we've reached the end, stop.
6309                  */
6310                 if (bno == NULLFSBLOCK)
6311                         break;
6312
6313                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6314                 if (bp) {
6315                         bp_release = 0;
6316                 } else {
6317                         bp_release = 1;
6318                 }
6319                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6320                                 XFS_BMAP_BTREE_REF)))
6321                         goto error_norelse;
6322                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6323         }
6324         if (bp_release) {
6325                 bp_release = 0;
6326                 xfs_trans_brelse(NULL, bp);
6327         }
6328         return;
6329
6330 error0:
6331         cmn_err(CE_WARN, "%s: at error0", __func__);
6332         if (bp_release)
6333                 xfs_trans_brelse(NULL, bp);
6334 error_norelse:
6335         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6336                 __func__, i);
6337         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6338         return;
6339 }
6340 #endif
6341
6342 /*
6343  * Count fsblocks of the given fork.
6344  */
6345 int                                             /* error */
6346 xfs_bmap_count_blocks(
6347         xfs_trans_t             *tp,            /* transaction pointer */
6348         xfs_inode_t             *ip,            /* incore inode */
6349         int                     whichfork,      /* data or attr fork */
6350         int                     *count)         /* out: count of blocks */
6351 {
6352         xfs_bmbt_block_t        *block; /* current btree block */
6353         xfs_fsblock_t           bno;    /* block # of "block" */
6354         xfs_ifork_t             *ifp;   /* fork structure */
6355         int                     level;  /* btree level, for checking */
6356         xfs_mount_t             *mp;    /* file system mount structure */
6357         __be64                  *pp;    /* pointer to block address */
6358
6359         bno = NULLFSBLOCK;
6360         mp = ip->i_mount;
6361         ifp = XFS_IFORK_PTR(ip, whichfork);
6362         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6363                 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6364                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6365                         count) < 0)) {
6366                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6367                                          XFS_ERRLEVEL_LOW, mp);
6368                         return XFS_ERROR(EFSCORRUPTED);
6369                 }
6370                 return 0;
6371         }
6372
6373         /*
6374          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6375          */
6376         block = ifp->if_broot;
6377         level = be16_to_cpu(block->bb_level);
6378         ASSERT(level > 0);
6379         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6380         bno = be64_to_cpu(*pp);
6381         ASSERT(bno != NULLDFSBNO);
6382         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6383         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6384
6385         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6386                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6387                                  mp);
6388                 return XFS_ERROR(EFSCORRUPTED);
6389         }
6390
6391         return 0;
6392 }
6393
6394 /*
6395  * Recursively walks each level of a btree
6396  * to count total fsblocks is use.
6397  */
6398 STATIC int                                     /* error */
6399 xfs_bmap_count_tree(
6400         xfs_mount_t     *mp,            /* file system mount point */
6401         xfs_trans_t     *tp,            /* transaction pointer */
6402         xfs_ifork_t     *ifp,           /* inode fork pointer */
6403         xfs_fsblock_t   blockno,        /* file system block number */
6404         int             levelin,        /* level in btree */
6405         int             *count)         /* Count of blocks */
6406 {
6407         int                     error;
6408         xfs_buf_t               *bp, *nbp;
6409         int                     level = levelin;
6410         __be64                  *pp;
6411         xfs_fsblock_t           bno = blockno;
6412         xfs_fsblock_t           nextbno;
6413         xfs_bmbt_block_t        *block, *nextblock;
6414         int                     numrecs;
6415
6416         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6417                 return error;
6418         *count += 1;
6419         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6420
6421         if (--level) {
6422                 /* Not at node above leafs, count this level of nodes */
6423                 nextbno = be64_to_cpu(block->bb_rightsib);
6424                 while (nextbno != NULLFSBLOCK) {
6425                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6426                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6427                                 return error;
6428                         *count += 1;
6429                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6430                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6431                         xfs_trans_brelse(tp, nbp);
6432                 }
6433
6434                 /* Dive to the next level */
6435                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6436                 bno = be64_to_cpu(*pp);
6437                 if (unlikely((error =
6438                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6439                         xfs_trans_brelse(tp, bp);
6440                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6441                                          XFS_ERRLEVEL_LOW, mp);
6442                         return XFS_ERROR(EFSCORRUPTED);
6443                 }
6444                 xfs_trans_brelse(tp, bp);
6445         } else {
6446                 /* count all level 1 nodes and their leaves */
6447                 for (;;) {
6448                         nextbno = be64_to_cpu(block->bb_rightsib);
6449                         numrecs = be16_to_cpu(block->bb_numrecs);
6450                         if (unlikely(xfs_bmap_disk_count_leaves(0,
6451                                         block, numrecs, count) < 0)) {
6452                                 xfs_trans_brelse(tp, bp);
6453                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6454                                                  XFS_ERRLEVEL_LOW, mp);
6455                                 return XFS_ERROR(EFSCORRUPTED);
6456                         }
6457                         xfs_trans_brelse(tp, bp);
6458                         if (nextbno == NULLFSBLOCK)
6459                                 break;
6460                         bno = nextbno;
6461                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6462                                 XFS_BMAP_BTREE_REF)))
6463                                 return error;
6464                         *count += 1;
6465                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6466                 }
6467         }
6468         return 0;
6469 }
6470
6471 /*
6472  * Count leaf blocks given a range of extent records.
6473  */
6474 STATIC int
6475 xfs_bmap_count_leaves(
6476         xfs_ifork_t             *ifp,
6477         xfs_extnum_t            idx,
6478         int                     numrecs,
6479         int                     *count)
6480 {
6481         int             b;
6482
6483         for (b = 0; b < numrecs; b++) {
6484                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6485                 *count += xfs_bmbt_get_blockcount(frp);
6486         }
6487         return 0;
6488 }
6489
6490 /*
6491  * Count leaf blocks given a range of extent records originally
6492  * in btree format.
6493  */
6494 STATIC int
6495 xfs_bmap_disk_count_leaves(
6496         xfs_extnum_t            idx,
6497         xfs_bmbt_block_t        *block,
6498         int                     numrecs,
6499         int                     *count)
6500 {
6501         int             b;
6502         xfs_bmbt_rec_t  *frp;
6503
6504         for (b = 1; b <= numrecs; b++) {
6505                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6506                 *count += xfs_bmbt_disk_get_blockcount(frp);
6507         }
6508         return 0;
6509 }