2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
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.
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.
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
18 #ifndef __XFS_BTREE_H__
19 #define __XFS_BTREE_H__
27 extern kmem_zone_t *xfs_btree_cur_zone;
30 * This nonsense is to make -wlint happy.
32 #define XFS_LOOKUP_EQ ((xfs_lookup_t)XFS_LOOKUP_EQi)
33 #define XFS_LOOKUP_LE ((xfs_lookup_t)XFS_LOOKUP_LEi)
34 #define XFS_LOOKUP_GE ((xfs_lookup_t)XFS_LOOKUP_GEi)
36 #define XFS_BTNUM_BNO ((xfs_btnum_t)XFS_BTNUM_BNOi)
37 #define XFS_BTNUM_CNT ((xfs_btnum_t)XFS_BTNUM_CNTi)
38 #define XFS_BTNUM_BMAP ((xfs_btnum_t)XFS_BTNUM_BMAPi)
39 #define XFS_BTNUM_INO ((xfs_btnum_t)XFS_BTNUM_INOi)
42 * Short form header: space allocation btrees.
44 typedef struct xfs_btree_sblock {
45 __be32 bb_magic; /* magic number for block type */
46 __be16 bb_level; /* 0 is a leaf */
47 __be16 bb_numrecs; /* current # of data records */
48 __be32 bb_leftsib; /* left sibling block or NULLAGBLOCK */
49 __be32 bb_rightsib; /* right sibling block or NULLAGBLOCK */
53 * Long form header: bmap btrees.
55 typedef struct xfs_btree_lblock {
56 __be32 bb_magic; /* magic number for block type */
57 __be16 bb_level; /* 0 is a leaf */
58 __be16 bb_numrecs; /* current # of data records */
59 __be64 bb_leftsib; /* left sibling block or NULLDFSBNO */
60 __be64 bb_rightsib; /* right sibling block or NULLDFSBNO */
64 * Combined header and structure, used by common code.
66 typedef struct xfs_btree_block {
67 __be32 bb_magic; /* magic number for block type */
68 __be16 bb_level; /* 0 is a leaf */
69 __be16 bb_numrecs; /* current # of data records */
74 } s; /* short form pointers */
78 } l; /* long form pointers */
83 * Generic key, ptr and record wrapper structures.
85 * These are disk format structures, and are converted where necessary
86 * by the btree specific code that needs to interpret them.
89 __be32 s; /* short form ptr */
90 __be64 l; /* long form ptr */
95 xfs_bmdr_key_t bmbr; /* bmbt root block */
96 xfs_alloc_key_t alloc;
97 xfs_inobt_key_t inobt;
100 union xfs_btree_rec {
102 xfs_bmdr_rec_t bmbr; /* bmbt root block */
103 xfs_alloc_rec_t alloc;
104 xfs_inobt_rec_t inobt;
108 * For logging record fields.
110 #define XFS_BB_MAGIC 0x01
111 #define XFS_BB_LEVEL 0x02
112 #define XFS_BB_NUMRECS 0x04
113 #define XFS_BB_LEFTSIB 0x08
114 #define XFS_BB_RIGHTSIB 0x10
115 #define XFS_BB_NUM_BITS 5
116 #define XFS_BB_ALL_BITS ((1 << XFS_BB_NUM_BITS) - 1)
119 * Magic numbers for btree blocks.
121 extern const __uint32_t xfs_magics[];
124 * Generic stats interface
126 #define __XFS_BTREE_STATS_INC(type, stat) \
127 XFS_STATS_INC(xs_ ## type ## _2_ ## stat)
128 #define XFS_BTREE_STATS_INC(cur, stat) \
130 switch (cur->bc_btnum) { \
131 case XFS_BTNUM_BNO: __XFS_BTREE_STATS_INC(abtb, stat); break; \
132 case XFS_BTNUM_CNT: __XFS_BTREE_STATS_INC(abtc, stat); break; \
133 case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_INC(bmbt, stat); break; \
134 case XFS_BTNUM_INO: __XFS_BTREE_STATS_INC(ibt, stat); break; \
135 case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break; \
139 #define __XFS_BTREE_STATS_ADD(type, stat, val) \
140 XFS_STATS_ADD(xs_ ## type ## _2_ ## stat, val)
141 #define XFS_BTREE_STATS_ADD(cur, stat, val) \
143 switch (cur->bc_btnum) { \
144 case XFS_BTNUM_BNO: __XFS_BTREE_STATS_ADD(abtb, stat, val); break; \
145 case XFS_BTNUM_CNT: __XFS_BTREE_STATS_ADD(abtc, stat, val); break; \
146 case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_ADD(bmbt, stat, val); break; \
147 case XFS_BTNUM_INO: __XFS_BTREE_STATS_ADD(ibt, stat, val); break; \
148 case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break; \
152 * Maximum and minimum records in a btree block.
153 * Given block size, type prefix, and leaf flag (0 or 1).
154 * The divisor below is equivalent to lf ? (e1) : (e2) but that produces
157 #define XFS_BTREE_BLOCK_MAXRECS(bsz,t,lf) \
158 ((int)(((bsz) - (uint)sizeof(t ## _block_t)) / \
159 (((lf) * (uint)sizeof(t ## _rec_t)) + \
161 ((uint)sizeof(t ## _key_t) + (uint)sizeof(t ## _ptr_t))))))
162 #define XFS_BTREE_BLOCK_MINRECS(bsz,t,lf) \
163 (XFS_BTREE_BLOCK_MAXRECS(bsz,t,lf) / 2)
166 * Record, key, and pointer address calculation macros.
167 * Given block size, type prefix, block pointer, and index of requested entry
168 * (first entry numbered 1).
170 #define XFS_BTREE_REC_ADDR(t,bb,i) \
171 ((t ## _rec_t *)((char *)(bb) + sizeof(t ## _block_t) + \
172 ((i) - 1) * sizeof(t ## _rec_t)))
173 #define XFS_BTREE_KEY_ADDR(t,bb,i) \
174 ((t ## _key_t *)((char *)(bb) + sizeof(t ## _block_t) + \
175 ((i) - 1) * sizeof(t ## _key_t)))
176 #define XFS_BTREE_PTR_ADDR(t,bb,i,mxr) \
177 ((t ## _ptr_t *)((char *)(bb) + sizeof(t ## _block_t) + \
178 (mxr) * sizeof(t ## _key_t) + ((i) - 1) * sizeof(t ## _ptr_t)))
180 #define XFS_BTREE_MAXLEVELS 8 /* max of all btrees */
182 struct xfs_btree_ops {
183 /* size of the key and record structures */
187 /* cursor operations */
188 struct xfs_btree_cur *(*dup_cursor)(struct xfs_btree_cur *);
189 void (*update_cursor)(struct xfs_btree_cur *src,
190 struct xfs_btree_cur *dst);
192 /* update btree root pointer */
193 void (*set_root)(struct xfs_btree_cur *cur,
194 union xfs_btree_ptr *nptr, int level_change);
196 /* block allocation / freeing */
197 int (*alloc_block)(struct xfs_btree_cur *cur,
198 union xfs_btree_ptr *start_bno,
199 union xfs_btree_ptr *new_bno,
200 int length, int *stat);
201 int (*free_block)(struct xfs_btree_cur *cur, struct xfs_buf *bp);
203 /* update last record information */
204 void (*update_lastrec)(struct xfs_btree_cur *cur,
205 struct xfs_btree_block *block,
206 union xfs_btree_rec *rec,
207 int ptr, int reason);
209 /* records in block/level */
210 int (*get_maxrecs)(struct xfs_btree_cur *cur, int level);
212 /* records on disk. Matter for the root in inode case. */
213 int (*get_dmaxrecs)(struct xfs_btree_cur *cur, int level);
215 /* init values of btree structures */
216 void (*init_key_from_rec)(union xfs_btree_key *key,
217 union xfs_btree_rec *rec);
218 void (*init_rec_from_key)(union xfs_btree_key *key,
219 union xfs_btree_rec *rec);
220 void (*init_rec_from_cur)(struct xfs_btree_cur *cur,
221 union xfs_btree_rec *rec);
222 void (*init_ptr_from_cur)(struct xfs_btree_cur *cur,
223 union xfs_btree_ptr *ptr);
225 /* difference between key value and cursor value */
226 __int64_t (*key_diff)(struct xfs_btree_cur *cur,
227 union xfs_btree_key *key);
230 #ifdef XFS_BTREE_TRACE
231 void (*trace_enter)(struct xfs_btree_cur *, const char *,
232 char *, int, int, __psunsigned_t,
233 __psunsigned_t, __psunsigned_t,
234 __psunsigned_t, __psunsigned_t,
235 __psunsigned_t, __psunsigned_t,
236 __psunsigned_t, __psunsigned_t,
237 __psunsigned_t, __psunsigned_t);
238 void (*trace_cursor)(struct xfs_btree_cur *, __uint32_t *,
239 __uint64_t *, __uint64_t *);
240 void (*trace_key)(struct xfs_btree_cur *,
241 union xfs_btree_key *, __uint64_t *,
243 void (*trace_record)(struct xfs_btree_cur *,
244 union xfs_btree_rec *, __uint64_t *,
245 __uint64_t *, __uint64_t *);
250 * Reasons for the update_lastrec method to be called.
252 #define LASTREC_UPDATE 0
253 #define LASTREC_INSREC 1
257 * Btree cursor structure.
258 * This collects all information needed by the btree code in one place.
260 typedef struct xfs_btree_cur
262 struct xfs_trans *bc_tp; /* transaction we're in, if any */
263 struct xfs_mount *bc_mp; /* file system mount struct */
264 const struct xfs_btree_ops *bc_ops;
265 uint bc_flags; /* btree features - below */
267 xfs_alloc_rec_incore_t a;
269 xfs_inobt_rec_incore_t i;
270 } bc_rec; /* current insert/search record value */
271 struct xfs_buf *bc_bufs[XFS_BTREE_MAXLEVELS]; /* buf ptr per level */
272 int bc_ptrs[XFS_BTREE_MAXLEVELS]; /* key/record # */
273 __uint8_t bc_ra[XFS_BTREE_MAXLEVELS]; /* readahead bits */
274 #define XFS_BTCUR_LEFTRA 1 /* left sibling has been read-ahead */
275 #define XFS_BTCUR_RIGHTRA 2 /* right sibling has been read-ahead */
276 __uint8_t bc_nlevels; /* number of levels in the tree */
277 __uint8_t bc_blocklog; /* log2(blocksize) of btree blocks */
278 xfs_btnum_t bc_btnum; /* identifies which btree type */
280 struct { /* needed for BNO, CNT, INO */
281 struct xfs_buf *agbp; /* agf/agi buffer pointer */
282 xfs_agnumber_t agno; /* ag number */
284 struct { /* needed for BMAP */
285 struct xfs_inode *ip; /* pointer to our inode */
286 struct xfs_bmap_free *flist; /* list to free after */
287 xfs_fsblock_t firstblock; /* 1st blk allocated */
288 int allocated; /* count of alloced */
289 short forksize; /* fork's inode space */
290 char whichfork; /* data or attr fork */
291 char flags; /* flags */
292 #define XFS_BTCUR_BPRV_WASDEL 1 /* was delayed */
294 } bc_private; /* per-btree type data */
298 #define XFS_BTREE_LONG_PTRS (1<<0) /* pointers are 64bits long */
299 #define XFS_BTREE_ROOT_IN_INODE (1<<1) /* root may be variable size */
300 #define XFS_BTREE_LASTREC_UPDATE (1<<2) /* track last rec externally */
303 #define XFS_BTREE_NOERROR 0
304 #define XFS_BTREE_ERROR 1
307 * Convert from buffer to btree block header.
309 #define XFS_BUF_TO_BLOCK(bp) ((xfs_btree_block_t *)XFS_BUF_PTR(bp))
310 #define XFS_BUF_TO_LBLOCK(bp) ((xfs_btree_lblock_t *)XFS_BUF_PTR(bp))
311 #define XFS_BUF_TO_SBLOCK(bp) ((xfs_btree_sblock_t *)XFS_BUF_PTR(bp))
317 * Check that long form block header is ok.
319 int /* error (0 or EFSCORRUPTED) */
320 xfs_btree_check_lblock(
321 struct xfs_btree_cur *cur, /* btree cursor */
322 struct xfs_btree_lblock *block, /* btree long form block pointer */
323 int level, /* level of the btree block */
324 struct xfs_buf *bp); /* buffer containing block, if any */
327 * Check that short form block header is ok.
329 int /* error (0 or EFSCORRUPTED) */
330 xfs_btree_check_sblock(
331 struct xfs_btree_cur *cur, /* btree cursor */
332 struct xfs_btree_sblock *block, /* btree short form block pointer */
333 int level, /* level of the btree block */
334 struct xfs_buf *bp); /* buffer containing block */
337 * Check that block header is ok.
340 xfs_btree_check_block(
341 struct xfs_btree_cur *cur, /* btree cursor */
342 struct xfs_btree_block *block, /* generic btree block pointer */
343 int level, /* level of the btree block */
344 struct xfs_buf *bp); /* buffer containing block, if any */
347 * Check that (long) pointer is ok.
349 int /* error (0 or EFSCORRUPTED) */
350 xfs_btree_check_lptr(
351 struct xfs_btree_cur *cur, /* btree cursor */
352 xfs_dfsbno_t ptr, /* btree block disk address */
353 int level); /* btree block level */
355 #define xfs_btree_check_lptr_disk(cur, ptr, level) \
356 xfs_btree_check_lptr(cur, be64_to_cpu(ptr), level)
360 * Check that (short) pointer is ok.
362 int /* error (0 or EFSCORRUPTED) */
363 xfs_btree_check_sptr(
364 struct xfs_btree_cur *cur, /* btree cursor */
365 xfs_agblock_t ptr, /* btree block disk address */
366 int level); /* btree block level */
369 * Check that (short) pointer is ok.
371 int /* error (0 or EFSCORRUPTED) */
373 struct xfs_btree_cur *cur, /* btree cursor */
374 union xfs_btree_ptr *ptr, /* btree block disk address */
375 int index, /* offset from ptr to check */
376 int level); /* btree block level */
381 * Debug routine: check that keys are in the right order.
385 xfs_btnum_t btnum, /* btree identifier */
386 void *ak1, /* pointer to left (lower) key */
387 void *ak2); /* pointer to right (higher) key */
390 * Debug routine: check that records are in the right order.
394 xfs_btnum_t btnum, /* btree identifier */
395 void *ar1, /* pointer to left (lower) record */
396 void *ar2); /* pointer to right (higher) record */
398 #define xfs_btree_check_key(a, b, c)
399 #define xfs_btree_check_rec(a, b, c)
403 * Delete the btree cursor.
406 xfs_btree_del_cursor(
407 xfs_btree_cur_t *cur, /* btree cursor */
408 int error); /* del because of error */
411 * Duplicate the btree cursor.
412 * Allocate a new one, copy the record, re-get the buffers.
415 xfs_btree_dup_cursor(
416 xfs_btree_cur_t *cur, /* input cursor */
417 xfs_btree_cur_t **ncur);/* output cursor */
420 * Change the cursor to point to the first record in the current block
421 * at the given level. Other levels are unaffected.
423 int /* success=1, failure=0 */
425 xfs_btree_cur_t *cur, /* btree cursor */
426 int level); /* level to change */
429 * Get a buffer for the block, return it with no data read.
430 * Long-form addressing.
432 struct xfs_buf * /* buffer for fsbno */
434 struct xfs_mount *mp, /* file system mount point */
435 struct xfs_trans *tp, /* transaction pointer */
436 xfs_fsblock_t fsbno, /* file system block number */
437 uint lock); /* lock flags for get_buf */
440 * Get a buffer for the block, return it with no data read.
441 * Short-form addressing.
443 struct xfs_buf * /* buffer for agno/agbno */
445 struct xfs_mount *mp, /* file system mount point */
446 struct xfs_trans *tp, /* transaction pointer */
447 xfs_agnumber_t agno, /* allocation group number */
448 xfs_agblock_t agbno, /* allocation group block number */
449 uint lock); /* lock flags for get_buf */
452 * Check for the cursor referring to the last block at the given level.
454 int /* 1=is last block, 0=not last block */
455 xfs_btree_islastblock(
456 xfs_btree_cur_t *cur, /* btree cursor */
457 int level); /* level to check */
460 * Change the cursor to point to the last record in the current block
461 * at the given level. Other levels are unaffected.
463 int /* success=1, failure=0 */
465 xfs_btree_cur_t *cur, /* btree cursor */
466 int level); /* level to change */
469 * Compute first and last byte offsets for the fields given.
470 * Interprets the offsets table, which contains struct field offsets.
474 __int64_t fields, /* bitmask of fields */
475 const short *offsets,/* table of field offsets */
476 int nbits, /* number of bits to inspect */
477 int *first, /* output: first byte offset */
478 int *last); /* output: last byte offset */
481 * Get a buffer for the block, return it read in.
482 * Long-form addressing.
486 struct xfs_mount *mp, /* file system mount point */
487 struct xfs_trans *tp, /* transaction pointer */
488 xfs_fsblock_t fsbno, /* file system block number */
489 uint lock, /* lock flags for read_buf */
490 struct xfs_buf **bpp, /* buffer for fsbno */
491 int refval);/* ref count value for buffer */
494 * Get a buffer for the block, return it read in.
495 * Short-form addressing.
499 struct xfs_mount *mp, /* file system mount point */
500 struct xfs_trans *tp, /* transaction pointer */
501 xfs_agnumber_t agno, /* allocation group number */
502 xfs_agblock_t agbno, /* allocation group block number */
503 uint lock, /* lock flags for read_buf */
504 struct xfs_buf **bpp, /* buffer for agno/agbno */
505 int refval);/* ref count value for buffer */
508 * Read-ahead the block, don't wait for it, don't return a buffer.
509 * Long-form addressing.
512 xfs_btree_reada_bufl(
513 struct xfs_mount *mp, /* file system mount point */
514 xfs_fsblock_t fsbno, /* file system block number */
515 xfs_extlen_t count); /* count of filesystem blocks */
518 * Read-ahead the block, don't wait for it, don't return a buffer.
519 * Short-form addressing.
522 xfs_btree_reada_bufs(
523 struct xfs_mount *mp, /* file system mount point */
524 xfs_agnumber_t agno, /* allocation group number */
525 xfs_agblock_t agbno, /* allocation group block number */
526 xfs_extlen_t count); /* count of filesystem blocks */
529 * Read-ahead btree blocks, at the given level.
530 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
532 int /* readahead block count */
534 xfs_btree_cur_t *cur, /* btree cursor */
535 int lev, /* level in btree */
536 int lr); /* left/right bits */
539 * Set the buffer for level "lev" in the cursor to bp, releasing
540 * any previous buffer.
544 xfs_btree_cur_t *cur, /* btree cursor */
545 int lev, /* level in btree */
546 struct xfs_buf *bp); /* new buffer to set */
550 * Common btree core entry points.
552 int xfs_btree_increment(struct xfs_btree_cur *, int, int *);
553 int xfs_btree_decrement(struct xfs_btree_cur *, int, int *);
554 int xfs_btree_lookup(struct xfs_btree_cur *, xfs_lookup_t, int *);
555 int xfs_btree_updkey(struct xfs_btree_cur *, union xfs_btree_key *, int);
556 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
557 int xfs_btree_lshift(struct xfs_btree_cur *, int, int *);
558 int xfs_btree_rshift(struct xfs_btree_cur *, int, int *);
559 int xfs_btree_split(struct xfs_btree_cur *, int, union xfs_btree_ptr *,
560 union xfs_btree_key *, struct xfs_btree_cur **, int *);
561 int xfs_btree_new_root(struct xfs_btree_cur *, int *);
562 int xfs_btree_new_iroot(struct xfs_btree_cur *, int *, int *);
563 int xfs_btree_kill_iroot(struct xfs_btree_cur *);
564 int xfs_btree_insert(struct xfs_btree_cur *, int *);
569 static inline int xfs_btree_get_numrecs(struct xfs_btree_block *block)
571 return be16_to_cpu(block->bb_numrecs);
574 static inline void xfs_btree_set_numrecs(struct xfs_btree_block *block,
577 block->bb_numrecs = cpu_to_be16(numrecs);
580 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
582 return be16_to_cpu(block->bb_level);
585 #endif /* __KERNEL__ */
589 * Min and max functions for extlen, agblock, fileoff, and filblks types.
591 #define XFS_EXTLEN_MIN(a,b) min_t(xfs_extlen_t, (a), (b))
592 #define XFS_EXTLEN_MAX(a,b) max_t(xfs_extlen_t, (a), (b))
593 #define XFS_AGBLOCK_MIN(a,b) min_t(xfs_agblock_t, (a), (b))
594 #define XFS_AGBLOCK_MAX(a,b) max_t(xfs_agblock_t, (a), (b))
595 #define XFS_FILEOFF_MIN(a,b) min_t(xfs_fileoff_t, (a), (b))
596 #define XFS_FILEOFF_MAX(a,b) max_t(xfs_fileoff_t, (a), (b))
597 #define XFS_FILBLKS_MIN(a,b) min_t(xfs_filblks_t, (a), (b))
598 #define XFS_FILBLKS_MAX(a,b) max_t(xfs_filblks_t, (a), (b))
600 #define XFS_FSB_SANITY_CHECK(mp,fsb) \
601 (XFS_FSB_TO_AGNO(mp, fsb) < mp->m_sb.sb_agcount && \
602 XFS_FSB_TO_AGBNO(mp, fsb) < mp->m_sb.sb_agblocks)
604 #endif /* __XFS_BTREE_H__ */