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 * Generic key, ptr and record wrapper structures.
32 * These are disk format structures, and are converted where necessary
33 * by the btree specific code that needs to interpret them.
36 __be32 s; /* short form ptr */
37 __be64 l; /* long form ptr */
41 * The in-core btree key. Overlapping btrees actually store two keys
42 * per pointer, so we reserve enough memory to hold both. The __*bigkey
43 * items should never be accessed directly.
46 struct xfs_bmbt_key bmbt;
47 xfs_bmdr_key_t bmbr; /* bmbt root block */
48 xfs_alloc_key_t alloc;
49 struct xfs_inobt_key inobt;
50 struct xfs_rmap_key rmap;
51 struct xfs_rmap_key __rmap_bigkey[2];
55 struct xfs_bmbt_rec bmbt;
56 xfs_bmdr_rec_t bmbr; /* bmbt root block */
57 struct xfs_alloc_rec alloc;
58 struct xfs_inobt_rec inobt;
59 struct xfs_rmap_rec rmap;
63 * This nonsense is to make -wlint happy.
65 #define XFS_LOOKUP_EQ ((xfs_lookup_t)XFS_LOOKUP_EQi)
66 #define XFS_LOOKUP_LE ((xfs_lookup_t)XFS_LOOKUP_LEi)
67 #define XFS_LOOKUP_GE ((xfs_lookup_t)XFS_LOOKUP_GEi)
69 #define XFS_BTNUM_BNO ((xfs_btnum_t)XFS_BTNUM_BNOi)
70 #define XFS_BTNUM_CNT ((xfs_btnum_t)XFS_BTNUM_CNTi)
71 #define XFS_BTNUM_BMAP ((xfs_btnum_t)XFS_BTNUM_BMAPi)
72 #define XFS_BTNUM_INO ((xfs_btnum_t)XFS_BTNUM_INOi)
73 #define XFS_BTNUM_FINO ((xfs_btnum_t)XFS_BTNUM_FINOi)
74 #define XFS_BTNUM_RMAP ((xfs_btnum_t)XFS_BTNUM_RMAPi)
77 * For logging record fields.
79 #define XFS_BB_MAGIC (1 << 0)
80 #define XFS_BB_LEVEL (1 << 1)
81 #define XFS_BB_NUMRECS (1 << 2)
82 #define XFS_BB_LEFTSIB (1 << 3)
83 #define XFS_BB_RIGHTSIB (1 << 4)
84 #define XFS_BB_BLKNO (1 << 5)
85 #define XFS_BB_LSN (1 << 6)
86 #define XFS_BB_UUID (1 << 7)
87 #define XFS_BB_OWNER (1 << 8)
88 #define XFS_BB_NUM_BITS 5
89 #define XFS_BB_ALL_BITS ((1 << XFS_BB_NUM_BITS) - 1)
90 #define XFS_BB_NUM_BITS_CRC 9
91 #define XFS_BB_ALL_BITS_CRC ((1 << XFS_BB_NUM_BITS_CRC) - 1)
94 * Generic stats interface
96 #define __XFS_BTREE_STATS_INC(mp, type, stat) \
97 XFS_STATS_INC(mp, xs_ ## type ## _2_ ## stat)
98 #define XFS_BTREE_STATS_INC(cur, stat) \
100 struct xfs_mount *__mp = cur->bc_mp; \
101 switch (cur->bc_btnum) { \
102 case XFS_BTNUM_BNO: __XFS_BTREE_STATS_INC(__mp, abtb, stat); break; \
103 case XFS_BTNUM_CNT: __XFS_BTREE_STATS_INC(__mp, abtc, stat); break; \
104 case XFS_BTNUM_BMAP: __XFS_BTREE_STATS_INC(__mp, bmbt, stat); break; \
105 case XFS_BTNUM_INO: __XFS_BTREE_STATS_INC(__mp, ibt, stat); break; \
106 case XFS_BTNUM_FINO: __XFS_BTREE_STATS_INC(__mp, fibt, stat); break; \
107 case XFS_BTNUM_RMAP: __XFS_BTREE_STATS_INC(__mp, rmap, stat); break; \
108 case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break; \
112 #define __XFS_BTREE_STATS_ADD(mp, type, stat, val) \
113 XFS_STATS_ADD(mp, xs_ ## type ## _2_ ## stat, val)
114 #define XFS_BTREE_STATS_ADD(cur, stat, val) \
116 struct xfs_mount *__mp = cur->bc_mp; \
117 switch (cur->bc_btnum) { \
118 case XFS_BTNUM_BNO: \
119 __XFS_BTREE_STATS_ADD(__mp, abtb, stat, val); break; \
120 case XFS_BTNUM_CNT: \
121 __XFS_BTREE_STATS_ADD(__mp, abtc, stat, val); break; \
122 case XFS_BTNUM_BMAP: \
123 __XFS_BTREE_STATS_ADD(__mp, bmbt, stat, val); break; \
124 case XFS_BTNUM_INO: \
125 __XFS_BTREE_STATS_ADD(__mp, ibt, stat, val); break; \
126 case XFS_BTNUM_FINO: \
127 __XFS_BTREE_STATS_ADD(__mp, fibt, stat, val); break; \
128 case XFS_BTNUM_RMAP: \
129 __XFS_BTREE_STATS_ADD(__mp, rmap, stat, val); break; \
130 case XFS_BTNUM_MAX: ASSERT(0); /* fucking gcc */ ; break; \
134 #define XFS_BTREE_MAXLEVELS 9 /* max of all btrees */
136 struct xfs_btree_ops {
137 /* size of the key and record structures */
141 /* cursor operations */
142 struct xfs_btree_cur *(*dup_cursor)(struct xfs_btree_cur *);
143 void (*update_cursor)(struct xfs_btree_cur *src,
144 struct xfs_btree_cur *dst);
146 /* update btree root pointer */
147 void (*set_root)(struct xfs_btree_cur *cur,
148 union xfs_btree_ptr *nptr, int level_change);
150 /* block allocation / freeing */
151 int (*alloc_block)(struct xfs_btree_cur *cur,
152 union xfs_btree_ptr *start_bno,
153 union xfs_btree_ptr *new_bno,
155 int (*free_block)(struct xfs_btree_cur *cur, struct xfs_buf *bp);
157 /* update last record information */
158 void (*update_lastrec)(struct xfs_btree_cur *cur,
159 struct xfs_btree_block *block,
160 union xfs_btree_rec *rec,
161 int ptr, int reason);
163 /* records in block/level */
164 int (*get_minrecs)(struct xfs_btree_cur *cur, int level);
165 int (*get_maxrecs)(struct xfs_btree_cur *cur, int level);
167 /* records on disk. Matter for the root in inode case. */
168 int (*get_dmaxrecs)(struct xfs_btree_cur *cur, int level);
170 /* init values of btree structures */
171 void (*init_key_from_rec)(union xfs_btree_key *key,
172 union xfs_btree_rec *rec);
173 void (*init_rec_from_cur)(struct xfs_btree_cur *cur,
174 union xfs_btree_rec *rec);
175 void (*init_ptr_from_cur)(struct xfs_btree_cur *cur,
176 union xfs_btree_ptr *ptr);
177 void (*init_high_key_from_rec)(union xfs_btree_key *key,
178 union xfs_btree_rec *rec);
180 /* difference between key value and cursor value */
181 __int64_t (*key_diff)(struct xfs_btree_cur *cur,
182 union xfs_btree_key *key);
185 * Difference between key2 and key1 -- positive if key1 > key2,
186 * negative if key1 < key2, and zero if equal.
188 __int64_t (*diff_two_keys)(struct xfs_btree_cur *cur,
189 union xfs_btree_key *key1,
190 union xfs_btree_key *key2);
192 const struct xfs_buf_ops *buf_ops;
194 #if defined(DEBUG) || defined(XFS_WARN)
195 /* check that k1 is lower than k2 */
196 int (*keys_inorder)(struct xfs_btree_cur *cur,
197 union xfs_btree_key *k1,
198 union xfs_btree_key *k2);
200 /* check that r1 is lower than r2 */
201 int (*recs_inorder)(struct xfs_btree_cur *cur,
202 union xfs_btree_rec *r1,
203 union xfs_btree_rec *r2);
208 * Reasons for the update_lastrec method to be called.
210 #define LASTREC_UPDATE 0
211 #define LASTREC_INSREC 1
212 #define LASTREC_DELREC 2
215 union xfs_btree_irec {
216 struct xfs_alloc_rec_incore a;
217 struct xfs_bmbt_irec b;
218 struct xfs_inobt_rec_incore i;
219 struct xfs_rmap_irec r;
223 * Btree cursor structure.
224 * This collects all information needed by the btree code in one place.
226 typedef struct xfs_btree_cur
228 struct xfs_trans *bc_tp; /* transaction we're in, if any */
229 struct xfs_mount *bc_mp; /* file system mount struct */
230 const struct xfs_btree_ops *bc_ops;
231 uint bc_flags; /* btree features - below */
232 union xfs_btree_irec bc_rec; /* current insert/search record value */
233 struct xfs_buf *bc_bufs[XFS_BTREE_MAXLEVELS]; /* buf ptr per level */
234 int bc_ptrs[XFS_BTREE_MAXLEVELS]; /* key/record # */
235 __uint8_t bc_ra[XFS_BTREE_MAXLEVELS]; /* readahead bits */
236 #define XFS_BTCUR_LEFTRA 1 /* left sibling has been read-ahead */
237 #define XFS_BTCUR_RIGHTRA 2 /* right sibling has been read-ahead */
238 __uint8_t bc_nlevels; /* number of levels in the tree */
239 __uint8_t bc_blocklog; /* log2(blocksize) of btree blocks */
240 xfs_btnum_t bc_btnum; /* identifies which btree type */
242 struct { /* needed for BNO, CNT, INO */
243 struct xfs_buf *agbp; /* agf/agi buffer pointer */
244 struct xfs_defer_ops *dfops; /* deferred updates */
245 xfs_agnumber_t agno; /* ag number */
247 struct { /* needed for BMAP */
248 struct xfs_inode *ip; /* pointer to our inode */
249 struct xfs_defer_ops *dfops; /* deferred updates */
250 xfs_fsblock_t firstblock; /* 1st blk allocated */
251 int allocated; /* count of alloced */
252 short forksize; /* fork's inode space */
253 char whichfork; /* data or attr fork */
254 char flags; /* flags */
255 #define XFS_BTCUR_BPRV_WASDEL 1 /* was delayed */
257 } bc_private; /* per-btree type data */
261 #define XFS_BTREE_LONG_PTRS (1<<0) /* pointers are 64bits long */
262 #define XFS_BTREE_ROOT_IN_INODE (1<<1) /* root may be variable size */
263 #define XFS_BTREE_LASTREC_UPDATE (1<<2) /* track last rec externally */
264 #define XFS_BTREE_CRC_BLOCKS (1<<3) /* uses extended btree blocks */
265 #define XFS_BTREE_OVERLAPPING (1<<4) /* overlapping intervals */
268 #define XFS_BTREE_NOERROR 0
269 #define XFS_BTREE_ERROR 1
272 * Convert from buffer to btree block header.
274 #define XFS_BUF_TO_BLOCK(bp) ((struct xfs_btree_block *)((bp)->b_addr))
278 * Check that block header is ok.
281 xfs_btree_check_block(
282 struct xfs_btree_cur *cur, /* btree cursor */
283 struct xfs_btree_block *block, /* generic btree block pointer */
284 int level, /* level of the btree block */
285 struct xfs_buf *bp); /* buffer containing block, if any */
288 * Check that (long) pointer is ok.
290 int /* error (0 or EFSCORRUPTED) */
291 xfs_btree_check_lptr(
292 struct xfs_btree_cur *cur, /* btree cursor */
293 xfs_fsblock_t ptr, /* btree block disk address */
294 int level); /* btree block level */
297 * Delete the btree cursor.
300 xfs_btree_del_cursor(
301 xfs_btree_cur_t *cur, /* btree cursor */
302 int error); /* del because of error */
305 * Duplicate the btree cursor.
306 * Allocate a new one, copy the record, re-get the buffers.
309 xfs_btree_dup_cursor(
310 xfs_btree_cur_t *cur, /* input cursor */
311 xfs_btree_cur_t **ncur);/* output cursor */
314 * Get a buffer for the block, return it with no data read.
315 * Long-form addressing.
317 struct xfs_buf * /* buffer for fsbno */
319 struct xfs_mount *mp, /* file system mount point */
320 struct xfs_trans *tp, /* transaction pointer */
321 xfs_fsblock_t fsbno, /* file system block number */
322 uint lock); /* lock flags for get_buf */
325 * Get a buffer for the block, return it with no data read.
326 * Short-form addressing.
328 struct xfs_buf * /* buffer for agno/agbno */
330 struct xfs_mount *mp, /* file system mount point */
331 struct xfs_trans *tp, /* transaction pointer */
332 xfs_agnumber_t agno, /* allocation group number */
333 xfs_agblock_t agbno, /* allocation group block number */
334 uint lock); /* lock flags for get_buf */
337 * Check for the cursor referring to the last block at the given level.
339 int /* 1=is last block, 0=not last block */
340 xfs_btree_islastblock(
341 xfs_btree_cur_t *cur, /* btree cursor */
342 int level); /* level to check */
345 * Compute first and last byte offsets for the fields given.
346 * Interprets the offsets table, which contains struct field offsets.
350 __int64_t fields, /* bitmask of fields */
351 const short *offsets,/* table of field offsets */
352 int nbits, /* number of bits to inspect */
353 int *first, /* output: first byte offset */
354 int *last); /* output: last byte offset */
357 * Get a buffer for the block, return it read in.
358 * Long-form addressing.
362 struct xfs_mount *mp, /* file system mount point */
363 struct xfs_trans *tp, /* transaction pointer */
364 xfs_fsblock_t fsbno, /* file system block number */
365 uint lock, /* lock flags for read_buf */
366 struct xfs_buf **bpp, /* buffer for fsbno */
367 int refval, /* ref count value for buffer */
368 const struct xfs_buf_ops *ops);
371 * Read-ahead the block, don't wait for it, don't return a buffer.
372 * Long-form addressing.
375 xfs_btree_reada_bufl(
376 struct xfs_mount *mp, /* file system mount point */
377 xfs_fsblock_t fsbno, /* file system block number */
378 xfs_extlen_t count, /* count of filesystem blocks */
379 const struct xfs_buf_ops *ops);
382 * Read-ahead the block, don't wait for it, don't return a buffer.
383 * Short-form addressing.
386 xfs_btree_reada_bufs(
387 struct xfs_mount *mp, /* file system mount point */
388 xfs_agnumber_t agno, /* allocation group number */
389 xfs_agblock_t agbno, /* allocation group block number */
390 xfs_extlen_t count, /* count of filesystem blocks */
391 const struct xfs_buf_ops *ops);
394 * Initialise a new btree block header
397 xfs_btree_init_block(
398 struct xfs_mount *mp,
407 xfs_btree_init_block_int(
408 struct xfs_mount *mp,
409 struct xfs_btree_block *buf,
418 * Common btree core entry points.
420 int xfs_btree_increment(struct xfs_btree_cur *, int, int *);
421 int xfs_btree_decrement(struct xfs_btree_cur *, int, int *);
422 int xfs_btree_lookup(struct xfs_btree_cur *, xfs_lookup_t, int *);
423 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
424 int xfs_btree_new_iroot(struct xfs_btree_cur *, int *, int *);
425 int xfs_btree_insert(struct xfs_btree_cur *, int *);
426 int xfs_btree_delete(struct xfs_btree_cur *, int *);
427 int xfs_btree_get_rec(struct xfs_btree_cur *, union xfs_btree_rec **, int *);
428 int xfs_btree_change_owner(struct xfs_btree_cur *cur, __uint64_t new_owner,
429 struct list_head *buffer_list);
432 * btree block CRC helpers
434 void xfs_btree_lblock_calc_crc(struct xfs_buf *);
435 bool xfs_btree_lblock_verify_crc(struct xfs_buf *);
436 void xfs_btree_sblock_calc_crc(struct xfs_buf *);
437 bool xfs_btree_sblock_verify_crc(struct xfs_buf *);
440 * Internal btree helpers also used by xfs_bmap.c.
442 void xfs_btree_log_block(struct xfs_btree_cur *, struct xfs_buf *, int);
443 void xfs_btree_log_recs(struct xfs_btree_cur *, struct xfs_buf *, int, int);
448 static inline int xfs_btree_get_numrecs(struct xfs_btree_block *block)
450 return be16_to_cpu(block->bb_numrecs);
453 static inline void xfs_btree_set_numrecs(struct xfs_btree_block *block,
456 block->bb_numrecs = cpu_to_be16(numrecs);
459 static inline int xfs_btree_get_level(struct xfs_btree_block *block)
461 return be16_to_cpu(block->bb_level);
466 * Min and max functions for extlen, agblock, fileoff, and filblks types.
468 #define XFS_EXTLEN_MIN(a,b) min_t(xfs_extlen_t, (a), (b))
469 #define XFS_EXTLEN_MAX(a,b) max_t(xfs_extlen_t, (a), (b))
470 #define XFS_AGBLOCK_MIN(a,b) min_t(xfs_agblock_t, (a), (b))
471 #define XFS_AGBLOCK_MAX(a,b) max_t(xfs_agblock_t, (a), (b))
472 #define XFS_FILEOFF_MIN(a,b) min_t(xfs_fileoff_t, (a), (b))
473 #define XFS_FILEOFF_MAX(a,b) max_t(xfs_fileoff_t, (a), (b))
474 #define XFS_FILBLKS_MIN(a,b) min_t(xfs_filblks_t, (a), (b))
475 #define XFS_FILBLKS_MAX(a,b) max_t(xfs_filblks_t, (a), (b))
477 #define XFS_FSB_SANITY_CHECK(mp,fsb) \
478 (XFS_FSB_TO_AGNO(mp, fsb) < mp->m_sb.sb_agcount && \
479 XFS_FSB_TO_AGBNO(mp, fsb) < mp->m_sb.sb_agblocks)
482 * Trace hooks. Currently not implemented as they need to be ported
483 * over to the generic tracing functionality, which is some effort.
485 * i,j = integer (32 bit)
486 * b = btree block buffer (xfs_buf_t)
491 #define XFS_BTREE_TRACE_ARGBI(c, b, i)
492 #define XFS_BTREE_TRACE_ARGBII(c, b, i, j)
493 #define XFS_BTREE_TRACE_ARGI(c, i)
494 #define XFS_BTREE_TRACE_ARGIPK(c, i, p, s)
495 #define XFS_BTREE_TRACE_ARGIPR(c, i, p, r)
496 #define XFS_BTREE_TRACE_ARGIK(c, i, k)
497 #define XFS_BTREE_TRACE_ARGR(c, r)
498 #define XFS_BTREE_TRACE_CURSOR(c, t)
500 bool xfs_btree_sblock_v5hdr_verify(struct xfs_buf *bp);
501 bool xfs_btree_sblock_verify(struct xfs_buf *bp, unsigned int max_recs);
502 uint xfs_btree_compute_maxlevels(struct xfs_mount *mp, uint *limits,
504 xfs_extlen_t xfs_btree_calc_size(struct xfs_mount *mp, uint *limits,
505 unsigned long long len);
508 #define XFS_BTREE_QUERY_RANGE_CONTINUE 0 /* keep iterating */
509 #define XFS_BTREE_QUERY_RANGE_ABORT 1 /* stop iterating */
510 typedef int (*xfs_btree_query_range_fn)(struct xfs_btree_cur *cur,
511 union xfs_btree_rec *rec, void *priv);
513 int xfs_btree_query_range(struct xfs_btree_cur *cur,
514 union xfs_btree_irec *low_rec, union xfs_btree_irec *high_rec,
515 xfs_btree_query_range_fn fn, void *priv);
517 typedef int (*xfs_btree_visit_blocks_fn)(struct xfs_btree_cur *cur, int level,
519 int xfs_btree_visit_blocks(struct xfs_btree_cur *cur,
520 xfs_btree_visit_blocks_fn fn, void *data);
522 int xfs_btree_count_blocks(struct xfs_btree_cur *cur, xfs_extlen_t *blocks);
524 #endif /* __XFS_BTREE_H__ */