1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_alloc.h"
17 #include "xfs_btree.h"
18 #include "xfs_bmap_btree.h"
20 #include "xfs_error.h"
21 #include "xfs_quota.h"
22 #include "xfs_trace.h"
26 static struct kmem_cache *xfs_bmbt_cur_cache;
29 * Convert on-disk form of btree root to in-memory form.
34 xfs_bmdr_block_t *dblock,
36 struct xfs_btree_block *rblock,
39 struct xfs_mount *mp = ip->i_mount;
46 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
47 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
49 rblock->bb_level = dblock->bb_level;
50 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
51 rblock->bb_numrecs = dblock->bb_numrecs;
52 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
53 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
54 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
55 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
56 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
57 dmxr = be16_to_cpu(dblock->bb_numrecs);
58 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
59 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
63 xfs_bmbt_disk_get_all(
64 const struct xfs_bmbt_rec *rec,
65 struct xfs_bmbt_irec *irec)
67 uint64_t l0 = get_unaligned_be64(&rec->l0);
68 uint64_t l1 = get_unaligned_be64(&rec->l1);
70 irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
71 irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
72 irec->br_blockcount = l1 & xfs_mask64lo(21);
73 if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
74 irec->br_state = XFS_EXT_UNWRITTEN;
76 irec->br_state = XFS_EXT_NORM;
80 * Extract the blockcount field from an on disk bmap extent record.
83 xfs_bmbt_disk_get_blockcount(
84 const struct xfs_bmbt_rec *r)
86 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
90 * Extract the startoff field from a disk format bmap extent record.
93 xfs_bmbt_disk_get_startoff(
94 const struct xfs_bmbt_rec *r)
96 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
97 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
101 * Set all the fields in a bmap extent record from the uncompressed form.
104 xfs_bmbt_disk_set_all(
105 struct xfs_bmbt_rec *r,
106 struct xfs_bmbt_irec *s)
108 int extent_flag = (s->br_state != XFS_EXT_NORM);
110 ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
111 ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
112 ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
113 ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
116 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
117 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
118 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
120 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
121 ((xfs_bmbt_rec_base_t)s->br_blockcount &
122 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
126 * Convert in-memory form of btree root to on-disk form.
130 struct xfs_mount *mp,
131 struct xfs_btree_block *rblock,
133 xfs_bmdr_block_t *dblock,
142 if (xfs_has_crc(mp)) {
143 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
144 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
145 &mp->m_sb.sb_meta_uuid));
146 ASSERT(rblock->bb_u.l.bb_blkno ==
147 cpu_to_be64(XFS_BUF_DADDR_NULL));
149 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
150 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
151 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
152 ASSERT(rblock->bb_level != 0);
153 dblock->bb_level = rblock->bb_level;
154 dblock->bb_numrecs = rblock->bb_numrecs;
155 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
156 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
157 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
158 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
159 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
160 dmxr = be16_to_cpu(dblock->bb_numrecs);
161 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
162 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
165 STATIC struct xfs_btree_cur *
167 struct xfs_btree_cur *cur)
169 struct xfs_btree_cur *new;
171 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
172 cur->bc_ino.ip, cur->bc_ino.whichfork);
175 * Copy the firstblock, dfops, and flags values,
176 * since init cursor doesn't get them.
178 new->bc_ino.flags = cur->bc_ino.flags;
184 xfs_bmbt_update_cursor(
185 struct xfs_btree_cur *src,
186 struct xfs_btree_cur *dst)
188 ASSERT((dst->bc_tp->t_highest_agno != NULLAGNUMBER) ||
189 (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
191 dst->bc_ino.allocated += src->bc_ino.allocated;
192 dst->bc_tp->t_highest_agno = src->bc_tp->t_highest_agno;
194 src->bc_ino.allocated = 0;
198 xfs_bmbt_alloc_block(
199 struct xfs_btree_cur *cur,
200 const union xfs_btree_ptr *start,
201 union xfs_btree_ptr *new,
204 struct xfs_alloc_arg args;
207 memset(&args, 0, sizeof(args));
208 args.tp = cur->bc_tp;
209 args.mp = cur->bc_mp;
210 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
211 cur->bc_ino.whichfork);
212 args.minlen = args.maxlen = args.prod = 1;
213 args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
214 if (!args.wasdel && args.tp->t_blk_res == 0)
218 * If we are coming here from something like unwritten extent
219 * conversion, there has been no data extent allocation already done, so
220 * we have to ensure that we attempt to locate the entire set of bmbt
221 * allocations in the same AG, as xfs_bmapi_write() would have reserved.
223 if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
224 args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
225 cur->bc_ino.whichfork);
227 error = xfs_alloc_vextent_start_ag(&args, be64_to_cpu(start->l));
231 if (args.fsbno == NULLFSBLOCK && args.minleft) {
233 * Could not find an AG with enough free space to satisfy
234 * a full btree split. Try again and if
235 * successful activate the lowspace algorithm.
238 error = xfs_alloc_vextent_start_ag(&args, 0);
241 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
243 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
248 ASSERT(args.len == 1);
249 cur->bc_ino.allocated++;
250 cur->bc_ino.ip->i_nblocks++;
251 xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
252 xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
253 XFS_TRANS_DQ_BCOUNT, 1L);
255 new->l = cpu_to_be64(args.fsbno);
263 struct xfs_btree_cur *cur,
266 struct xfs_mount *mp = cur->bc_mp;
267 struct xfs_inode *ip = cur->bc_ino.ip;
268 struct xfs_trans *tp = cur->bc_tp;
269 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
270 struct xfs_owner_info oinfo;
273 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
274 error = xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo,
280 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
281 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
286 xfs_bmbt_get_minrecs(
287 struct xfs_btree_cur *cur,
290 if (level == cur->bc_nlevels - 1) {
291 struct xfs_ifork *ifp;
293 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
294 cur->bc_ino.whichfork);
296 return xfs_bmbt_maxrecs(cur->bc_mp,
297 ifp->if_broot_bytes, level == 0) / 2;
300 return cur->bc_mp->m_bmap_dmnr[level != 0];
304 xfs_bmbt_get_maxrecs(
305 struct xfs_btree_cur *cur,
308 if (level == cur->bc_nlevels - 1) {
309 struct xfs_ifork *ifp;
311 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
312 cur->bc_ino.whichfork);
314 return xfs_bmbt_maxrecs(cur->bc_mp,
315 ifp->if_broot_bytes, level == 0);
318 return cur->bc_mp->m_bmap_dmxr[level != 0];
323 * Get the maximum records we could store in the on-disk format.
325 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
326 * for the root node this checks the available space in the dinode fork
327 * so that we can resize the in-memory buffer to match it. After a
328 * resize to the maximum size this function returns the same value
329 * as xfs_bmbt_get_maxrecs for the root node, too.
332 xfs_bmbt_get_dmaxrecs(
333 struct xfs_btree_cur *cur,
336 if (level != cur->bc_nlevels - 1)
337 return cur->bc_mp->m_bmap_dmxr[level != 0];
338 return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
342 xfs_bmbt_init_key_from_rec(
343 union xfs_btree_key *key,
344 const union xfs_btree_rec *rec)
346 key->bmbt.br_startoff =
347 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
351 xfs_bmbt_init_high_key_from_rec(
352 union xfs_btree_key *key,
353 const union xfs_btree_rec *rec)
355 key->bmbt.br_startoff = cpu_to_be64(
356 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
357 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
361 xfs_bmbt_init_rec_from_cur(
362 struct xfs_btree_cur *cur,
363 union xfs_btree_rec *rec)
365 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
369 xfs_bmbt_init_ptr_from_cur(
370 struct xfs_btree_cur *cur,
371 union xfs_btree_ptr *ptr)
378 struct xfs_btree_cur *cur,
379 const union xfs_btree_key *key)
381 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
382 cur->bc_rec.b.br_startoff;
386 xfs_bmbt_diff_two_keys(
387 struct xfs_btree_cur *cur,
388 const union xfs_btree_key *k1,
389 const union xfs_btree_key *k2,
390 const union xfs_btree_key *mask)
392 uint64_t a = be64_to_cpu(k1->bmbt.br_startoff);
393 uint64_t b = be64_to_cpu(k2->bmbt.br_startoff);
395 ASSERT(!mask || mask->bmbt.br_startoff);
398 * Note: This routine previously casted a and b to int64 and subtracted
399 * them to generate a result. This lead to problems if b was the
400 * "maximum" key value (all ones) being signed incorrectly, hence this
401 * somewhat less efficient version.
410 static xfs_failaddr_t
414 struct xfs_mount *mp = bp->b_mount;
415 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
419 if (!xfs_verify_magic(bp, block->bb_magic))
420 return __this_address;
422 if (xfs_has_crc(mp)) {
424 * XXX: need a better way of verifying the owner here. Right now
425 * just make sure there has been one set.
427 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
433 * numrecs and level verification.
435 * We don't know what fork we belong to, so just verify that the level
436 * is less than the maximum of the two. Later checks will be more
439 level = be16_to_cpu(block->bb_level);
440 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
441 return __this_address;
443 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
447 xfs_bmbt_read_verify(
452 if (!xfs_btree_lblock_verify_crc(bp))
453 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
455 fa = xfs_bmbt_verify(bp);
457 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
461 trace_xfs_btree_corrupt(bp, _RET_IP_);
465 xfs_bmbt_write_verify(
470 fa = xfs_bmbt_verify(bp);
472 trace_xfs_btree_corrupt(bp, _RET_IP_);
473 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
476 xfs_btree_lblock_calc_crc(bp);
479 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
481 .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
482 cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
483 .verify_read = xfs_bmbt_read_verify,
484 .verify_write = xfs_bmbt_write_verify,
485 .verify_struct = xfs_bmbt_verify,
490 xfs_bmbt_keys_inorder(
491 struct xfs_btree_cur *cur,
492 const union xfs_btree_key *k1,
493 const union xfs_btree_key *k2)
495 return be64_to_cpu(k1->bmbt.br_startoff) <
496 be64_to_cpu(k2->bmbt.br_startoff);
500 xfs_bmbt_recs_inorder(
501 struct xfs_btree_cur *cur,
502 const union xfs_btree_rec *r1,
503 const union xfs_btree_rec *r2)
505 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
506 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
507 xfs_bmbt_disk_get_startoff(&r2->bmbt);
510 STATIC enum xbtree_key_contig
511 xfs_bmbt_keys_contiguous(
512 struct xfs_btree_cur *cur,
513 const union xfs_btree_key *key1,
514 const union xfs_btree_key *key2,
515 const union xfs_btree_key *mask)
517 ASSERT(!mask || mask->bmbt.br_startoff);
519 return xbtree_key_contig(be64_to_cpu(key1->bmbt.br_startoff),
520 be64_to_cpu(key2->bmbt.br_startoff));
523 static const struct xfs_btree_ops xfs_bmbt_ops = {
524 .rec_len = sizeof(xfs_bmbt_rec_t),
525 .key_len = sizeof(xfs_bmbt_key_t),
527 .dup_cursor = xfs_bmbt_dup_cursor,
528 .update_cursor = xfs_bmbt_update_cursor,
529 .alloc_block = xfs_bmbt_alloc_block,
530 .free_block = xfs_bmbt_free_block,
531 .get_maxrecs = xfs_bmbt_get_maxrecs,
532 .get_minrecs = xfs_bmbt_get_minrecs,
533 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
534 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
535 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
536 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
537 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
538 .key_diff = xfs_bmbt_key_diff,
539 .diff_two_keys = xfs_bmbt_diff_two_keys,
540 .buf_ops = &xfs_bmbt_buf_ops,
541 .keys_inorder = xfs_bmbt_keys_inorder,
542 .recs_inorder = xfs_bmbt_recs_inorder,
543 .keys_contiguous = xfs_bmbt_keys_contiguous,
547 * Allocate a new bmap btree cursor.
549 struct xfs_btree_cur * /* new bmap btree cursor */
550 xfs_bmbt_init_cursor(
551 struct xfs_mount *mp, /* file system mount point */
552 struct xfs_trans *tp, /* transaction pointer */
553 struct xfs_inode *ip, /* inode owning the btree */
554 int whichfork) /* data or attr fork */
556 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
557 struct xfs_btree_cur *cur;
558 ASSERT(whichfork != XFS_COW_FORK);
560 cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
561 mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
562 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
563 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
565 cur->bc_ops = &xfs_bmbt_ops;
566 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
568 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
570 cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
572 cur->bc_ino.allocated = 0;
573 cur->bc_ino.flags = 0;
574 cur->bc_ino.whichfork = whichfork;
579 /* Calculate number of records in a block mapping btree block. */
580 static inline unsigned int
581 xfs_bmbt_block_maxrecs(
582 unsigned int blocklen,
586 return blocklen / sizeof(xfs_bmbt_rec_t);
587 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
591 * Calculate number of records in a bmap btree block.
595 struct xfs_mount *mp,
599 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
600 return xfs_bmbt_block_maxrecs(blocklen, leaf);
604 * Calculate the maximum possible height of the btree that the on-disk format
605 * supports. This is used for sizing structures large enough to support every
606 * possible configuration of a filesystem that might get mounted.
609 xfs_bmbt_maxlevels_ondisk(void)
611 unsigned int minrecs[2];
612 unsigned int blocklen;
614 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
615 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
617 minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
618 minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
620 /* One extra level for the inode root. */
621 return xfs_btree_compute_maxlevels(minrecs,
622 XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
626 * Calculate number of records in a bmap btree inode root.
633 blocklen -= sizeof(xfs_bmdr_block_t);
636 return blocklen / sizeof(xfs_bmdr_rec_t);
637 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
641 * Change the owner of a btree format fork fo the inode passed in. Change it to
642 * the owner of that is passed in so that we can change owners before or after
643 * we switch forks between inodes. The operation that the caller is doing will
644 * determine whether is needs to change owner before or after the switch.
646 * For demand paged transactional modification, the fork switch should be done
647 * after reading in all the blocks, modifying them and pinning them in the
648 * transaction. For modification when the buffers are already pinned in memory,
649 * the fork switch can be done before changing the owner as we won't need to
650 * validate the owner until the btree buffers are unpinned and writes can occur
653 * For recovery based ownership change, there is no transactional context and
654 * so a buffer list must be supplied so that we can record the buffers that we
655 * modified for the caller to issue IO on.
658 xfs_bmbt_change_owner(
659 struct xfs_trans *tp,
660 struct xfs_inode *ip,
663 struct list_head *buffer_list)
665 struct xfs_btree_cur *cur;
668 ASSERT(tp || buffer_list);
669 ASSERT(!(tp && buffer_list));
670 ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
672 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
673 cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
675 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
676 xfs_btree_del_cursor(cur, error);
680 /* Calculate the bmap btree size for some records. */
683 struct xfs_mount *mp,
684 unsigned long long len)
686 return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
690 xfs_bmbt_init_cur_cache(void)
692 xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
693 xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
696 if (!xfs_bmbt_cur_cache)
702 xfs_bmbt_destroy_cur_cache(void)
704 kmem_cache_destroy(xfs_bmbt_cur_cache);
705 xfs_bmbt_cur_cache = NULL;