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 *);
190 /* records in block/level */
191 int (*get_maxrecs)(struct xfs_btree_cur *cur, int level);
194 #ifdef XFS_BTREE_TRACE
195 void (*trace_enter)(struct xfs_btree_cur *, const char *,
196 char *, int, int, __psunsigned_t,
197 __psunsigned_t, __psunsigned_t,
198 __psunsigned_t, __psunsigned_t,
199 __psunsigned_t, __psunsigned_t,
200 __psunsigned_t, __psunsigned_t,
201 __psunsigned_t, __psunsigned_t);
202 void (*trace_cursor)(struct xfs_btree_cur *, __uint32_t *,
203 __uint64_t *, __uint64_t *);
204 void (*trace_key)(struct xfs_btree_cur *,
205 union xfs_btree_key *, __uint64_t *,
207 void (*trace_record)(struct xfs_btree_cur *,
208 union xfs_btree_rec *, __uint64_t *,
209 __uint64_t *, __uint64_t *);
214 * Btree cursor structure.
215 * This collects all information needed by the btree code in one place.
217 typedef struct xfs_btree_cur
219 struct xfs_trans *bc_tp; /* transaction we're in, if any */
220 struct xfs_mount *bc_mp; /* file system mount struct */
221 const struct xfs_btree_ops *bc_ops;
222 uint bc_flags; /* btree features - below */
224 xfs_alloc_rec_incore_t a;
226 xfs_inobt_rec_incore_t i;
227 } bc_rec; /* current insert/search record value */
228 struct xfs_buf *bc_bufs[XFS_BTREE_MAXLEVELS]; /* buf ptr per level */
229 int bc_ptrs[XFS_BTREE_MAXLEVELS]; /* key/record # */
230 __uint8_t bc_ra[XFS_BTREE_MAXLEVELS]; /* readahead bits */
231 #define XFS_BTCUR_LEFTRA 1 /* left sibling has been read-ahead */
232 #define XFS_BTCUR_RIGHTRA 2 /* right sibling has been read-ahead */
233 __uint8_t bc_nlevels; /* number of levels in the tree */
234 __uint8_t bc_blocklog; /* log2(blocksize) of btree blocks */
235 xfs_btnum_t bc_btnum; /* identifies which btree type */
237 struct { /* needed for BNO, CNT, INO */
238 struct xfs_buf *agbp; /* agf/agi buffer pointer */
239 xfs_agnumber_t agno; /* ag number */
241 struct { /* needed for BMAP */
242 struct xfs_inode *ip; /* pointer to our inode */
243 struct xfs_bmap_free *flist; /* list to free after */
244 xfs_fsblock_t firstblock; /* 1st blk allocated */
245 int allocated; /* count of alloced */
246 short forksize; /* fork's inode space */
247 char whichfork; /* data or attr fork */
248 char flags; /* flags */
249 #define XFS_BTCUR_BPRV_WASDEL 1 /* was delayed */
251 } bc_private; /* per-btree type data */
255 #define XFS_BTREE_LONG_PTRS (1<<0) /* pointers are 64bits long */
256 #define XFS_BTREE_ROOT_IN_INODE (1<<1) /* root may be variable size */
259 #define XFS_BTREE_NOERROR 0
260 #define XFS_BTREE_ERROR 1
263 * Convert from buffer to btree block header.
265 #define XFS_BUF_TO_BLOCK(bp) ((xfs_btree_block_t *)XFS_BUF_PTR(bp))
266 #define XFS_BUF_TO_LBLOCK(bp) ((xfs_btree_lblock_t *)XFS_BUF_PTR(bp))
267 #define XFS_BUF_TO_SBLOCK(bp) ((xfs_btree_sblock_t *)XFS_BUF_PTR(bp))
273 * Check that long form block header is ok.
275 int /* error (0 or EFSCORRUPTED) */
276 xfs_btree_check_lblock(
277 struct xfs_btree_cur *cur, /* btree cursor */
278 struct xfs_btree_lblock *block, /* btree long form block pointer */
279 int level, /* level of the btree block */
280 struct xfs_buf *bp); /* buffer containing block, if any */
283 * Check that short form block header is ok.
285 int /* error (0 or EFSCORRUPTED) */
286 xfs_btree_check_sblock(
287 struct xfs_btree_cur *cur, /* btree cursor */
288 struct xfs_btree_sblock *block, /* btree short form block pointer */
289 int level, /* level of the btree block */
290 struct xfs_buf *bp); /* buffer containing block */
293 * Check that block header is ok.
296 xfs_btree_check_block(
297 struct xfs_btree_cur *cur, /* btree cursor */
298 struct xfs_btree_block *block, /* generic btree block pointer */
299 int level, /* level of the btree block */
300 struct xfs_buf *bp); /* buffer containing block, if any */
303 * Check that (long) pointer is ok.
305 int /* error (0 or EFSCORRUPTED) */
306 xfs_btree_check_lptr(
307 struct xfs_btree_cur *cur, /* btree cursor */
308 xfs_dfsbno_t ptr, /* btree block disk address */
309 int level); /* btree block level */
311 #define xfs_btree_check_lptr_disk(cur, ptr, level) \
312 xfs_btree_check_lptr(cur, be64_to_cpu(ptr), level)
316 * Check that (short) pointer is ok.
318 int /* error (0 or EFSCORRUPTED) */
319 xfs_btree_check_sptr(
320 struct xfs_btree_cur *cur, /* btree cursor */
321 xfs_agblock_t ptr, /* btree block disk address */
322 int level); /* btree block level */
325 * Check that (short) pointer is ok.
327 int /* error (0 or EFSCORRUPTED) */
329 struct xfs_btree_cur *cur, /* btree cursor */
330 union xfs_btree_ptr *ptr, /* btree block disk address */
331 int index, /* offset from ptr to check */
332 int level); /* btree block level */
337 * Debug routine: check that keys are in the right order.
341 xfs_btnum_t btnum, /* btree identifier */
342 void *ak1, /* pointer to left (lower) key */
343 void *ak2); /* pointer to right (higher) key */
346 * Debug routine: check that records are in the right order.
350 xfs_btnum_t btnum, /* btree identifier */
351 void *ar1, /* pointer to left (lower) record */
352 void *ar2); /* pointer to right (higher) record */
354 #define xfs_btree_check_key(a, b, c)
355 #define xfs_btree_check_rec(a, b, c)
359 * Delete the btree cursor.
362 xfs_btree_del_cursor(
363 xfs_btree_cur_t *cur, /* btree cursor */
364 int error); /* del because of error */
367 * Duplicate the btree cursor.
368 * Allocate a new one, copy the record, re-get the buffers.
371 xfs_btree_dup_cursor(
372 xfs_btree_cur_t *cur, /* input cursor */
373 xfs_btree_cur_t **ncur);/* output cursor */
376 * Change the cursor to point to the first record in the current block
377 * at the given level. Other levels are unaffected.
379 int /* success=1, failure=0 */
381 xfs_btree_cur_t *cur, /* btree cursor */
382 int level); /* level to change */
385 * Get a buffer for the block, return it with no data read.
386 * Long-form addressing.
388 struct xfs_buf * /* buffer for fsbno */
390 struct xfs_mount *mp, /* file system mount point */
391 struct xfs_trans *tp, /* transaction pointer */
392 xfs_fsblock_t fsbno, /* file system block number */
393 uint lock); /* lock flags for get_buf */
396 * Get a buffer for the block, return it with no data read.
397 * Short-form addressing.
399 struct xfs_buf * /* buffer for agno/agbno */
401 struct xfs_mount *mp, /* file system mount point */
402 struct xfs_trans *tp, /* transaction pointer */
403 xfs_agnumber_t agno, /* allocation group number */
404 xfs_agblock_t agbno, /* allocation group block number */
405 uint lock); /* lock flags for get_buf */
408 * Check for the cursor referring to the last block at the given level.
410 int /* 1=is last block, 0=not last block */
411 xfs_btree_islastblock(
412 xfs_btree_cur_t *cur, /* btree cursor */
413 int level); /* level to check */
416 * Change the cursor to point to the last record in the current block
417 * at the given level. Other levels are unaffected.
419 int /* success=1, failure=0 */
421 xfs_btree_cur_t *cur, /* btree cursor */
422 int level); /* level to change */
425 * Compute first and last byte offsets for the fields given.
426 * Interprets the offsets table, which contains struct field offsets.
430 __int64_t fields, /* bitmask of fields */
431 const short *offsets,/* table of field offsets */
432 int nbits, /* number of bits to inspect */
433 int *first, /* output: first byte offset */
434 int *last); /* output: last byte offset */
437 * Get a buffer for the block, return it read in.
438 * Long-form addressing.
442 struct xfs_mount *mp, /* file system mount point */
443 struct xfs_trans *tp, /* transaction pointer */
444 xfs_fsblock_t fsbno, /* file system block number */
445 uint lock, /* lock flags for read_buf */
446 struct xfs_buf **bpp, /* buffer for fsbno */
447 int refval);/* ref count value for buffer */
450 * Get a buffer for the block, return it read in.
451 * Short-form addressing.
455 struct xfs_mount *mp, /* file system mount point */
456 struct xfs_trans *tp, /* transaction pointer */
457 xfs_agnumber_t agno, /* allocation group number */
458 xfs_agblock_t agbno, /* allocation group block number */
459 uint lock, /* lock flags for read_buf */
460 struct xfs_buf **bpp, /* buffer for agno/agbno */
461 int refval);/* ref count value for buffer */
464 * Read-ahead the block, don't wait for it, don't return a buffer.
465 * Long-form addressing.
468 xfs_btree_reada_bufl(
469 struct xfs_mount *mp, /* file system mount point */
470 xfs_fsblock_t fsbno, /* file system block number */
471 xfs_extlen_t count); /* count of filesystem blocks */
474 * Read-ahead the block, don't wait for it, don't return a buffer.
475 * Short-form addressing.
478 xfs_btree_reada_bufs(
479 struct xfs_mount *mp, /* file system mount point */
480 xfs_agnumber_t agno, /* allocation group number */
481 xfs_agblock_t agbno, /* allocation group block number */
482 xfs_extlen_t count); /* count of filesystem blocks */
485 * Read-ahead btree blocks, at the given level.
486 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
488 int /* readahead block count */
490 xfs_btree_cur_t *cur, /* btree cursor */
491 int lev, /* level in btree */
492 int lr); /* left/right bits */
495 * Set the buffer for level "lev" in the cursor to bp, releasing
496 * any previous buffer.
500 xfs_btree_cur_t *cur, /* btree cursor */
501 int lev, /* level in btree */
502 struct xfs_buf *bp); /* new buffer to set */
508 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
510 return be16_to_cpu(block->bb_level);
513 #endif /* __KERNEL__ */
517 * Min and max functions for extlen, agblock, fileoff, and filblks types.
519 #define XFS_EXTLEN_MIN(a,b) min_t(xfs_extlen_t, (a), (b))
520 #define XFS_EXTLEN_MAX(a,b) max_t(xfs_extlen_t, (a), (b))
521 #define XFS_AGBLOCK_MIN(a,b) min_t(xfs_agblock_t, (a), (b))
522 #define XFS_AGBLOCK_MAX(a,b) max_t(xfs_agblock_t, (a), (b))
523 #define XFS_FILEOFF_MIN(a,b) min_t(xfs_fileoff_t, (a), (b))
524 #define XFS_FILEOFF_MAX(a,b) max_t(xfs_fileoff_t, (a), (b))
525 #define XFS_FILBLKS_MIN(a,b) min_t(xfs_filblks_t, (a), (b))
526 #define XFS_FILBLKS_MAX(a,b) max_t(xfs_filblks_t, (a), (b))
528 #define XFS_FSB_SANITY_CHECK(mp,fsb) \
529 (XFS_FSB_TO_AGNO(mp, fsb) < mp->m_sb.sb_agcount && \
530 XFS_FSB_TO_AGBNO(mp, fsb) < mp->m_sb.sb_agblocks)
532 #endif /* __XFS_BTREE_H__ */