fs/xfs/xfs_iomap.c

   1 /*
   2  * Copyright (c) 2000-2005 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_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_alloc.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_quota.h"
  31 #include "xfs_mount.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_bmap.h"
  43 #include "xfs_bit.h"
  44 #include "xfs_rtalloc.h"
  45 #include "xfs_error.h"
  46 #include "xfs_itable.h"
  47 #include "xfs_rw.h"
  48 #include "xfs_acl.h"
  49 #include "xfs_cap.h"
  50 #include "xfs_mac.h"
  51 #include "xfs_attr.h"
  52 #include "xfs_buf_item.h"
  53 #include "xfs_trans_space.h"
  54 #include "xfs_utils.h"
  55 #include "xfs_iomap.h"
  56
  57 #if defined(XFS_RW_TRACE)
  58 void
  59 xfs_iomap_enter_trace(
  60         int             tag,
  61         xfs_iocore_t    *io,
  62         xfs_off_t       offset,
  63         ssize_t         count)
  64 {
  65         xfs_inode_t     *ip = XFS_IO_INODE(io);
  66
  67         if (!ip->i_rwtrace)
  68                 return;
  69
  70         ktrace_enter(ip->i_rwtrace,
  71                 (void *)((unsigned long)tag),
  72                 (void *)ip,
  73                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  74                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  75                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  76                 (void *)((unsigned long)(offset & 0xffffffff)),
  77                 (void *)((unsigned long)count),
  78                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  79                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  80                 (void *)NULL,
  81                 (void *)NULL,
  82                 (void *)NULL,
  83                 (void *)NULL,
  84                 (void *)NULL,
  85                 (void *)NULL,
  86                 (void *)NULL);
  87 }
  88
  89 void
  90 xfs_iomap_map_trace(
  91         int             tag,
  92         xfs_iocore_t    *io,
  93         xfs_off_t       offset,
  94         ssize_t         count,
  95         xfs_iomap_t     *iomapp,
  96         xfs_bmbt_irec_t *imapp,
  97         int             flags)
  98 {
  99         xfs_inode_t     *ip = XFS_IO_INODE(io);
 100
 101         if (!ip->i_rwtrace)
 102                 return;
 103
 104         ktrace_enter(ip->i_rwtrace,
 105                 (void *)((unsigned long)tag),
 106                 (void *)ip,
 107                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
 108                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
 109                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 110                 (void *)((unsigned long)(offset & 0xffffffff)),
 111                 (void *)((unsigned long)count),
 112                 (void *)((unsigned long)flags),
 113                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
 114                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
 115                 (void *)((unsigned long)(iomapp->iomap_delta)),
 116                 (void *)((unsigned long)(iomapp->iomap_bsize)),
 117                 (void *)((unsigned long)(iomapp->iomap_bn)),
 118                 (void *)(__psint_t)(imapp->br_startoff),
 119                 (void *)((unsigned long)(imapp->br_blockcount)),
 120                 (void *)(__psint_t)(imapp->br_startblock));
 121 }
 122 #else
 123 #define xfs_iomap_enter_trace(tag, io, offset, count)
 124 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
 125 #endif
 126
 127 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
 128                                                 << mp->m_writeio_log)
 129 #define XFS_STRAT_WRITE_IMAPS   2
 130 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
 131
 132 STATIC int
 133 xfs_imap_to_bmap(
 134         xfs_iocore_t    *io,
 135         xfs_off_t       offset,
 136         xfs_bmbt_irec_t *imap,
 137         xfs_iomap_t     *iomapp,
 138         int             imaps,                  /* Number of imap entries */
 139         int             iomaps,                 /* Number of iomap entries */
 140         int             flags)
 141 {
 142         xfs_mount_t     *mp;
 143         xfs_fsize_t     nisize;
 144         int             pbm;
 145         xfs_fsblock_t   start_block;
 146
 147         mp = io->io_mount;
 148         nisize = XFS_SIZE(mp, io);
 149         if (io->io_new_size > nisize)
 150                 nisize = io->io_new_size;
 151
 152         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
 153                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 154                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 155                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 156                 iomapp->iomap_flags = flags;
 157
 158                 if (io->io_flags & XFS_IOCORE_RT) {
 159                         iomapp->iomap_flags |= IOMAP_REALTIME;
 160                         iomapp->iomap_target = mp->m_rtdev_targp;
 161                 } else {
 162                         iomapp->iomap_target = mp->m_ddev_targp;
 163                 }
 164                 start_block = imap->br_startblock;
 165                 if (start_block == HOLESTARTBLOCK) {
 166                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 167                         iomapp->iomap_flags |= IOMAP_HOLE;
 168                 } else if (start_block == DELAYSTARTBLOCK) {
 169                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 170                         iomapp->iomap_flags |= IOMAP_DELAY;
 171                 } else {
 172                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 173                         if (ISUNWRITTEN(imap))
 174                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 175                 }
 176
 177                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 178                         iomapp->iomap_flags |= IOMAP_EOF;
 179                 }
 180
 181                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 182         }
 183         return pbm;     /* Return the number filled */
 184 }
 185
 186 int
 187 xfs_iomap(
 188         xfs_iocore_t    *io,
 189         xfs_off_t       offset,
 190         ssize_t         count,
 191         int             flags,
 192         xfs_iomap_t     *iomapp,
 193         int             *niomaps)
 194 {
 195         xfs_mount_t     *mp = io->io_mount;
 196         xfs_fileoff_t   offset_fsb, end_fsb;
 197         int             error = 0;
 198         int             lockmode = 0;
 199         xfs_bmbt_irec_t imap;
 200         int             nimaps = 1;
 201         int             bmapi_flags = 0;
 202         int             iomap_flags = 0;
 203
 204         if (XFS_FORCED_SHUTDOWN(mp))
 205                 return XFS_ERROR(EIO);
 206
 207         switch (flags &
 208                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 209                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 210         case BMAPI_READ:
 211                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 212                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 213                 bmapi_flags = XFS_BMAPI_ENTIRE;
 214                 break;
 215         case BMAPI_WRITE:
 216                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 217                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 218                 if (flags & BMAPI_IGNSTATE)
 219                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
 220                 XFS_ILOCK(mp, io, lockmode);
 221                 break;
 222         case BMAPI_ALLOCATE:
 223                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 224                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 225                 bmapi_flags = XFS_BMAPI_ENTIRE;
 226                 /* Attempt non-blocking lock */
 227                 if (flags & BMAPI_TRYLOCK) {
 228                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 229                                 return XFS_ERROR(EAGAIN);
 230                 } else {
 231                         XFS_ILOCK(mp, io, lockmode);
 232                 }
 233                 break;
 234         case BMAPI_UNWRITTEN:
 235                 goto phase2;
 236         case BMAPI_DEVICE:
 237                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 238                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 239                         mp->m_rtdev_targp : mp->m_ddev_targp;
 240                 error = 0;
 241                 *niomaps = 1;
 242                 goto out;
 243         default:
 244                 BUG();
 245         }
 246
 247         ASSERT(offset <= mp->m_maxioffset);
 248         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 249                 count = mp->m_maxioffset - offset;
 250         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 251         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 252
 253         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 254                         (xfs_filblks_t)(end_fsb - offset_fsb),
 255                         bmapi_flags,  NULL, 0, &imap,
 256                         &nimaps, NULL);
 257
 258         if (error)
 259                 goto out;
 260
 261 phase2:
 262         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 263         case BMAPI_WRITE:
 264                 /* If we found an extent, return it */
 265                 if (nimaps &&
 266                     (imap.br_startblock != HOLESTARTBLOCK) &&
 267                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 268                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 269                                         offset, count, iomapp, &imap, flags);
 270                         break;
 271                 }
 272
 273                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 274                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 275                                         count, flags, &imap, &nimaps, nimaps);
 276                 } else {
 277                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 278                                         flags, &imap, &nimaps);
 279                 }
 280                 if (!error) {
 281                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 282                                         offset, count, iomapp, &imap, flags);
 283                 }
 284                 iomap_flags = IOMAP_NEW;
 285                 break;
 286         case BMAPI_ALLOCATE:
 287                 /* If we found an extent, return it */
 288                 XFS_IUNLOCK(mp, io, lockmode);
 289                 lockmode = 0;
 290
 291                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 292                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 293                                         offset, count, iomapp, &imap, flags);
 294                         break;
 295                 }
 296
 297                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
 298                                                  &imap, &nimaps);
 299                 break;
 300         case BMAPI_UNWRITTEN:
 301                 lockmode = 0;
 302                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 303                 nimaps = 0;
 304                 break;
 305         }
 306
 307         if (nimaps) {
 308                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 309                                             iomapp, nimaps, *niomaps, iomap_flags);
 310         } else if (niomaps) {
 311                 *niomaps = 0;
 312         }
 313
 314 out:
 315         if (lockmode)
 316                 XFS_IUNLOCK(mp, io, lockmode);
 317         return XFS_ERROR(error);
 318 }
 319
 320 STATIC int
 321 xfs_flush_space(
 322         xfs_inode_t     *ip,
 323         int             *fsynced,
 324         int             *ioflags)
 325 {
 326         switch (*fsynced) {
 327         case 0:
 328                 if (ip->i_delayed_blks) {
 329                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 330                         xfs_flush_inode(ip);
 331                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 332                         *fsynced = 1;
 333                 } else {
 334                         *ioflags |= BMAPI_SYNC;
 335                         *fsynced = 2;
 336                 }
 337                 return 0;
 338         case 1:
 339                 *fsynced = 2;
 340                 *ioflags |= BMAPI_SYNC;
 341                 return 0;
 342         case 2:
 343                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 344                 xfs_flush_device(ip);
 345                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 346                 *fsynced = 3;
 347                 return 0;
 348         }
 349         return 1;
 350 }
 351
 352 int
 353 xfs_iomap_write_direct(
 354         xfs_inode_t     *ip,
 355         xfs_off_t       offset,
 356         size_t          count,
 357         int             flags,
 358         xfs_bmbt_irec_t *ret_imap,
 359         int             *nmaps,
 360         int             found)
 361 {
 362         xfs_mount_t     *mp = ip->i_mount;
 363         xfs_iocore_t    *io = &ip->i_iocore;
 364         xfs_fileoff_t   offset_fsb;
 365         xfs_fileoff_t   last_fsb;
 366         xfs_filblks_t   count_fsb;
 367         xfs_fsize_t     isize;
 368         xfs_fsblock_t   firstfsb;
 369         int             nimaps;
 370         int             error;
 371         int             bmapi_flag;
 372         int             quota_flag;
 373         int             rt;
 374         xfs_trans_t     *tp;
 375         xfs_bmbt_irec_t imap;
 376         xfs_bmap_free_t free_list;
 377         int             aeof;
 378         xfs_filblks_t   qblocks, resblks;
 379         int             committed;
 380         int             resrtextents;
 381
 382         /*
 383          * Make sure that the dquots are there. This doesn't hold
 384          * the ilock across a disk read.
 385          */
 386         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 387         if (error)
 388                 return XFS_ERROR(error);
 389
 390         isize = ip->i_d.di_size;
 391         aeof = (offset + count) > isize;
 392
 393         if (io->io_new_size > isize)
 394                 isize = io->io_new_size;
 395
 396         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 397         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 398         count_fsb = last_fsb - offset_fsb;
 399         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
 400                 xfs_fileoff_t   map_last_fsb;
 401
 402                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
 403                 if (map_last_fsb < last_fsb) {
 404                         last_fsb = map_last_fsb;
 405                         count_fsb = last_fsb - offset_fsb;
 406                 }
 407                 ASSERT(count_fsb > 0);
 408         }
 409
 410         /*
 411          * Determine if reserving space on the data or realtime partition.
 412          */
 413         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
 414                 xfs_extlen_t    extsz;
 415
 416                 if (!(extsz = ip->i_d.di_extsize))
 417                         extsz = mp->m_sb.sb_rextsize;
 418                 resrtextents = qblocks = (count_fsb + extsz - 1);
 419                 do_div(resrtextents, mp->m_sb.sb_rextsize);
 420                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 421                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 422         } else {
 423                 resrtextents = 0;
 424                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
 425                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 426         }
 427
 428         /*
 429          * Allocate and setup the transaction
 430          */
 431         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 432         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 433         error = xfs_trans_reserve(tp, resblks,
 434                         XFS_WRITE_LOG_RES(mp), resrtextents,
 435                         XFS_TRANS_PERM_LOG_RES,
 436                         XFS_WRITE_LOG_COUNT);
 437
 438         /*
 439          * Check for running out of space, note: need lock to return
 440          */
 441         if (error)
 442                 xfs_trans_cancel(tp, 0);
 443         xfs_ilock(ip, XFS_ILOCK_EXCL);
 444         if (error)
 445                 goto error_out;
 446
 447         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
 448                 error = (EDQUOT);
 449                 goto error1;
 450         }
 451
 452         bmapi_flag = XFS_BMAPI_WRITE;
 453         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 454         xfs_trans_ihold(tp, ip);
 455
 456         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
 457                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 458
 459         /*
 460          * Issue the bmapi() call to allocate the blocks
 461          */
 462         XFS_BMAP_INIT(&free_list, &firstfsb);
 463         nimaps = 1;
 464         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 465                 bmapi_flag, &firstfsb, 0, &imap, &nimaps, &free_list);
 466         if (error)
 467                 goto error0;
 468
 469         /*
 470          * Complete the transaction
 471          */
 472         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 473         if (error)
 474                 goto error0;
 475         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 476         if (error)
 477                 goto error_out;
 478
 479         /*
 480          * Copy any maps to caller's array and return any error.
 481          */
 482         if (nimaps == 0) {
 483                 error = (ENOSPC);
 484                 goto error_out;
 485         }
 486
 487         *ret_imap = imap;
 488         *nmaps = 1;
 489         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 490                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 491                         "start_block : %llx start_off : %llx blkcnt : %llx "
 492                         "extent-state : %x \n",
 493                         (ip->i_mount)->m_fsname,
 494                         (long long)ip->i_ino,
 495                         ret_imap->br_startblock, ret_imap->br_startoff,
 496                         ret_imap->br_blockcount,ret_imap->br_state);
 497         }
 498         return 0;
 499
 500 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 501         xfs_bmap_cancel(&free_list);
 502         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
 503
 504 error1: /* Just cancel transaction */
 505         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 506         *nmaps = 0;     /* nothing set-up here */
 507
 508 error_out:
 509         return XFS_ERROR(error);
 510 }
 511
 512 int
 513 xfs_iomap_write_delay(
 514         xfs_inode_t     *ip,
 515         xfs_off_t       offset,
 516         size_t          count,
 517         int             ioflag,
 518         xfs_bmbt_irec_t *ret_imap,
 519         int             *nmaps)
 520 {
 521         xfs_mount_t     *mp = ip->i_mount;
 522         xfs_iocore_t    *io = &ip->i_iocore;
 523         xfs_fileoff_t   offset_fsb;
 524         xfs_fileoff_t   last_fsb;
 525         xfs_fsize_t     isize;
 526         xfs_fsblock_t   firstblock;
 527         int             nimaps;
 528         int             error;
 529         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 530         int             aeof;
 531         int             fsynced = 0;
 532
 533         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 534
 535         /*
 536          * Make sure that the dquots are there. This doesn't hold
 537          * the ilock across a disk read.
 538          */
 539
 540         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 541         if (error)
 542                 return XFS_ERROR(error);
 543
 544 retry:
 545         isize = ip->i_d.di_size;
 546         if (io->io_new_size > isize) {
 547                 isize = io->io_new_size;
 548         }
 549
 550         aeof = 0;
 551         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 552         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 553         /*
 554          * If the caller is doing a write at the end of the file,
 555          * then extend the allocation (and the buffer used for the write)
 556          * out to the file system's write iosize.  We clean up any extra
 557          * space left over when the file is closed in xfs_inactive().
 558          *
 559          * For sync writes, we are flushing delayed allocate space to
 560          * try to make additional space available for allocation near
 561          * the filesystem full boundary - preallocation hurts in that
 562          * situation, of course.
 563          */
 564         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
 565                 xfs_off_t       aligned_offset;
 566                 xfs_filblks_t   count_fsb;
 567                 unsigned int    iosize;
 568                 xfs_fileoff_t   ioalign;
 569                 int             n;
 570                 xfs_fileoff_t   start_fsb;
 571
 572                 /*
 573                  * If there are any real blocks past eof, then don't
 574                  * do any speculative allocation.
 575                  */
 576                 start_fsb = XFS_B_TO_FSBT(mp,
 577                                         ((xfs_ufsize_t)(offset + count - 1)));
 578                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 579                 while (count_fsb > 0) {
 580                         nimaps = XFS_WRITE_IMAPS;
 581                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 582                                         0, &firstblock, 0, imap, &nimaps, NULL);
 583                         if (error) {
 584                                 return error;
 585                         }
 586                         for (n = 0; n < nimaps; n++) {
 587                                 if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 588                                         !imap[n].br_startblock) {
 589                                         cmn_err(CE_PANIC,"Access to block "
 590                                                 "zero:  fs <%s> inode: %lld "
 591                                                 "start_block : %llx start_off "
 592                                                 ": %llx blkcnt : %llx "
 593                                                 "extent-state : %x \n",
 594                                                 (ip->i_mount)->m_fsname,
 595                                                 (long long)ip->i_ino,
 596                                                 imap[n].br_startblock,
 597                                                 imap[n].br_startoff,
 598                                                 imap[n].br_blockcount,
 599                                                 imap[n].br_state);
 600                                 }
 601                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 602                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
 603                                         goto write_map;
 604                                 }
 605                                 start_fsb += imap[n].br_blockcount;
 606                                 count_fsb -= imap[n].br_blockcount;
 607                         }
 608                 }
 609                 iosize = mp->m_writeio_blocks;
 610                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 611                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 612                 last_fsb = ioalign + iosize;
 613                 aeof = 1;
 614         }
 615 write_map:
 616         nimaps = XFS_WRITE_IMAPS;
 617         firstblock = NULLFSBLOCK;
 618
 619         /*
 620          * If mounted with the "-o swalloc" option, roundup the allocation
 621          * request to a stripe width boundary if the file size is >=
 622          * stripe width and we are allocating past the allocation eof.
 623          */
 624         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
 625             && (mp->m_flags & XFS_MOUNT_SWALLOC)
 626             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
 627                 int eof;
 628                 xfs_fileoff_t new_last_fsb;
 629
 630                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
 631                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 632                 if (error) {
 633                         return error;
 634                 }
 635                 if (eof) {
 636                         last_fsb = new_last_fsb;
 637                 }
 638         /*
 639          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 640          * if the file size is >= stripe unit size, and we are allocating past
 641          * the allocation eof.
 642          */
 643         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
 644                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
 645                 int eof;
 646                 xfs_fileoff_t new_last_fsb;
 647                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
 648                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 649                 if (error) {
 650                         return error;
 651                 }
 652                 if (eof) {
 653                         last_fsb = new_last_fsb;
 654                 }
 655         /*
 656          * Round up the allocation request to a real-time extent boundary
 657          * if the file is on the real-time subvolume.
 658          */
 659         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
 660                 int eof;
 661                 xfs_fileoff_t new_last_fsb;
 662
 663                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
 664                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
 665                 if (error) {
 666                         return error;
 667                 }
 668                 if (eof)
 669                         last_fsb = new_last_fsb;
 670         }
 671         error = xfs_bmapi(NULL, ip, offset_fsb,
 672                           (xfs_filblks_t)(last_fsb - offset_fsb),
 673                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 674                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 675                           &nimaps, NULL);
 676         /*
 677          * This can be EDQUOT, if nimaps == 0
 678          */
 679         if (error && (error != ENOSPC)) {
 680                 return XFS_ERROR(error);
 681         }
 682         /*
 683          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 684          * then we must have run out of space.
 685          */
 686         if (nimaps == 0) {
 687                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 688                                         io, offset, count);
 689                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 690                         return XFS_ERROR(ENOSPC);
 691
 692                 error = 0;
 693                 goto retry;
 694         }
 695
 696         *ret_imap = imap[0];
 697         *nmaps = 1;
 698         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 699                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 700                         "start_block : %llx start_off : %llx blkcnt : %llx "
 701                         "extent-state : %x \n",
 702                         (ip->i_mount)->m_fsname,
 703                         (long long)ip->i_ino,
 704                         ret_imap->br_startblock, ret_imap->br_startoff,
 705                         ret_imap->br_blockcount,ret_imap->br_state);
 706         }
 707         return 0;
 708 }
 709
 710 /*
 711  * Pass in a delayed allocate extent, convert it to real extents;
 712  * return to the caller the extent we create which maps on top of
 713  * the originating callers request.
 714  *
 715  * Called without a lock on the inode.
 716  */
 717 int
 718 xfs_iomap_write_allocate(
 719         xfs_inode_t     *ip,
 720         xfs_off_t       offset,
 721         size_t          count,
 722         xfs_bmbt_irec_t *map,
 723         int             *retmap)
 724 {
 725         xfs_mount_t     *mp = ip->i_mount;
 726         xfs_iocore_t    *io = &ip->i_iocore;
 727         xfs_fileoff_t   offset_fsb, last_block;
 728         xfs_fileoff_t   end_fsb, map_start_fsb;
 729         xfs_fsblock_t   first_block;
 730         xfs_bmap_free_t free_list;
 731         xfs_filblks_t   count_fsb;
 732         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 733         xfs_trans_t     *tp;
 734         int             i, nimaps, committed;
 735         int             error = 0;
 736         int             nres;
 737
 738         *retmap = 0;
 739
 740         /*
 741          * Make sure that the dquots are there.
 742          */
 743         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 744                 return XFS_ERROR(error);
 745
 746         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 747         count_fsb = map->br_blockcount;
 748         map_start_fsb = map->br_startoff;
 749
 750         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 751
 752         while (count_fsb != 0) {
 753                 /*
 754                  * Set up a transaction with which to allocate the
 755                  * backing store for the file.  Do allocations in a
 756                  * loop until we get some space in the range we are
 757                  * interested in.  The other space that might be allocated
 758                  * is in the delayed allocation extent on which we sit
 759                  * but before our buffer starts.
 760                  */
 761
 762                 nimaps = 0;
 763                 while (nimaps == 0) {
 764                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 765                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 766                         error = xfs_trans_reserve(tp, nres,
 767                                         XFS_WRITE_LOG_RES(mp),
 768                                         0, XFS_TRANS_PERM_LOG_RES,
 769                                         XFS_WRITE_LOG_COUNT);
 770                         if (error == ENOSPC) {
 771                                 error = xfs_trans_reserve(tp, 0,
 772                                                 XFS_WRITE_LOG_RES(mp),
 773                                                 0,
 774                                                 XFS_TRANS_PERM_LOG_RES,
 775                                                 XFS_WRITE_LOG_COUNT);
 776                         }
 777                         if (error) {
 778                                 xfs_trans_cancel(tp, 0);
 779                                 return XFS_ERROR(error);
 780                         }
 781                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 782                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 783                         xfs_trans_ihold(tp, ip);
 784
 785                         XFS_BMAP_INIT(&free_list, &first_block);
 786
 787                         nimaps = XFS_STRAT_WRITE_IMAPS;
 788                         /*
 789                          * Ensure we don't go beyond eof - it is possible
 790                          * the extents changed since we did the read call,
 791                          * we dropped the ilock in the interim.
 792                          */
 793
 794                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 795                         xfs_bmap_last_offset(NULL, ip, &last_block,
 796                                 XFS_DATA_FORK);
 797                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 798                         if ((map_start_fsb + count_fsb) > last_block) {
 799                                 count_fsb = last_block - map_start_fsb;
 800                                 if (count_fsb == 0) {
 801                                         error = EAGAIN;
 802                                         goto trans_cancel;
 803                                 }
 804                         }
 805
 806                         /* Go get the actual blocks */
 807                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 808                                         XFS_BMAPI_WRITE, &first_block, 1,
 809                                         imap, &nimaps, &free_list);
 810                         if (error)
 811                                 goto trans_cancel;
 812
 813                         error = xfs_bmap_finish(&tp, &free_list,
 814                                         first_block, &committed);
 815                         if (error)
 816                                 goto trans_cancel;
 817
 818                         error = xfs_trans_commit(tp,
 819                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 820                         if (error)
 821                                 goto error0;
 822
 823                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 824                 }
 825
 826                 /*
 827                  * See if we were able to allocate an extent that
 828                  * covers at least part of the callers request
 829                  */
 830
 831                 for (i = 0; i < nimaps; i++) {
 832                         if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 833                                 !imap[i].br_startblock) {
 834                                 cmn_err(CE_PANIC,"Access to block zero:  "
 835                                         "fs <%s> inode: %lld "
 836                                         "start_block : %llx start_off : %llx "
 837                                         "blkcnt : %llx extent-state : %x \n",
 838                                         (ip->i_mount)->m_fsname,
 839                                         (long long)ip->i_ino,
 840                                         imap[i].br_startblock,
 841                                         imap[i].br_startoff,
 842                                         imap[i].br_blockcount,imap[i].br_state);
 843                         }
 844                         if ((offset_fsb >= imap[i].br_startoff) &&
 845                             (offset_fsb < (imap[i].br_startoff +
 846                                            imap[i].br_blockcount))) {
 847                                 *map = imap[i];
 848                                 *retmap = 1;
 849                                 XFS_STATS_INC(xs_xstrat_quick);
 850                                 return 0;
 851                         }
 852                         count_fsb -= imap[i].br_blockcount;
 853                 }
 854
 855                 /* So far we have not mapped the requested part of the
 856                  * file, just surrounding data, try again.
 857                  */
 858                 nimaps--;
 859                 map_start_fsb = imap[nimaps].br_startoff +
 860                                 imap[nimaps].br_blockcount;
 861         }
 862
 863 trans_cancel:
 864         xfs_bmap_cancel(&free_list);
 865         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 866 error0:
 867         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 868         return XFS_ERROR(error);
 869 }
 870
 871 int
 872 xfs_iomap_write_unwritten(
 873         xfs_inode_t     *ip,
 874         xfs_off_t       offset,
 875         size_t          count)
 876 {
 877         xfs_mount_t     *mp = ip->i_mount;
 878         xfs_iocore_t    *io = &ip->i_iocore;
 879         xfs_trans_t     *tp;
 880         xfs_fileoff_t   offset_fsb;
 881         xfs_filblks_t   count_fsb;
 882         xfs_filblks_t   numblks_fsb;
 883         xfs_bmbt_irec_t imap;
 884         int             committed;
 885         int             error;
 886         int             nres;
 887         int             nimaps;
 888         xfs_fsblock_t   firstfsb;
 889         xfs_bmap_free_t free_list;
 890
 891         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 892                                 &ip->i_iocore, offset, count);
 893
 894         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 895         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 896         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 897
 898         do {
 899                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 900
 901                 /*
 902                  * set up a transaction to convert the range of extents
 903                  * from unwritten to real. Do allocations in a loop until
 904                  * we have covered the range passed in.
 905                  */
 906
 907                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 908                 error = xfs_trans_reserve(tp, nres,
 909                                 XFS_WRITE_LOG_RES(mp), 0,
 910                                 XFS_TRANS_PERM_LOG_RES,
 911                                 XFS_WRITE_LOG_COUNT);
 912                 if (error) {
 913                         xfs_trans_cancel(tp, 0);
 914                         goto error0;
 915                 }
 916
 917                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 918                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 919                 xfs_trans_ihold(tp, ip);
 920
 921                 /*
 922                  * Modify the unwritten extent state of the buffer.
 923                  */
 924                 XFS_BMAP_INIT(&free_list, &firstfsb);
 925                 nimaps = 1;
 926                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 927                                   XFS_BMAPI_WRITE, &firstfsb,
 928                                   1, &imap, &nimaps, &free_list);
 929                 if (error)
 930                         goto error_on_bmapi_transaction;
 931
 932                 error = xfs_bmap_finish(&(tp), &(free_list),
 933                                 firstfsb, &committed);
 934                 if (error)
 935                         goto error_on_bmapi_transaction;
 936
 937                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 938                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 939                 if (error)
 940                         goto error0;
 941
 942                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 943                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 944                                 "inode: %lld start_block : %llx start_off : "
 945                                 "%llx blkcnt : %llx extent-state : %x \n",
 946                                 (ip->i_mount)->m_fsname,
 947                                 (long long)ip->i_ino,
 948                                 imap.br_startblock,imap.br_startoff,
 949                                 imap.br_blockcount,imap.br_state);
 950                 }
 951
 952                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 953                         /*
 954                          * The numblks_fsb value should always get
 955                          * smaller, otherwise the loop is stuck.
 956                          */
 957                         ASSERT(imap.br_blockcount);
 958                         break;
 959                 }
 960                 offset_fsb += numblks_fsb;
 961                 count_fsb -= numblks_fsb;
 962         } while (count_fsb > 0);
 963
 964         return 0;
 965
 966 error_on_bmapi_transaction:
 967         xfs_bmap_cancel(&free_list);
 968         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 969         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 970 error0:
 971         return XFS_ERROR(error);
 972 }