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"
25 static struct kmem_cache *xfs_bmbt_cur_cache;
28 * Convert on-disk form of btree root to in-memory form.
33 xfs_bmdr_block_t *dblock,
35 struct xfs_btree_block *rblock,
38 struct xfs_mount *mp = ip->i_mount;
45 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
46 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
48 rblock->bb_level = dblock->bb_level;
49 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
50 rblock->bb_numrecs = dblock->bb_numrecs;
51 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
52 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
53 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
54 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
55 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
56 dmxr = be16_to_cpu(dblock->bb_numrecs);
57 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
58 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
62 xfs_bmbt_disk_get_all(
63 const struct xfs_bmbt_rec *rec,
64 struct xfs_bmbt_irec *irec)
66 uint64_t l0 = get_unaligned_be64(&rec->l0);
67 uint64_t l1 = get_unaligned_be64(&rec->l1);
69 irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
70 irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
71 irec->br_blockcount = l1 & xfs_mask64lo(21);
72 if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
73 irec->br_state = XFS_EXT_UNWRITTEN;
75 irec->br_state = XFS_EXT_NORM;
79 * Extract the blockcount field from an on disk bmap extent record.
82 xfs_bmbt_disk_get_blockcount(
83 const struct xfs_bmbt_rec *r)
85 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
89 * Extract the startoff field from a disk format bmap extent record.
92 xfs_bmbt_disk_get_startoff(
93 const struct xfs_bmbt_rec *r)
95 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
96 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
100 * Set all the fields in a bmap extent record from the uncompressed form.
103 xfs_bmbt_disk_set_all(
104 struct xfs_bmbt_rec *r,
105 struct xfs_bmbt_irec *s)
107 int extent_flag = (s->br_state != XFS_EXT_NORM);
109 ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
110 ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
111 ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
112 ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
115 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
116 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
117 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
119 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
120 ((xfs_bmbt_rec_base_t)s->br_blockcount &
121 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
125 * Convert in-memory form of btree root to on-disk form.
129 struct xfs_mount *mp,
130 struct xfs_btree_block *rblock,
132 xfs_bmdr_block_t *dblock,
141 if (xfs_has_crc(mp)) {
142 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
143 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
144 &mp->m_sb.sb_meta_uuid));
145 ASSERT(rblock->bb_u.l.bb_blkno ==
146 cpu_to_be64(XFS_BUF_DADDR_NULL));
148 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
149 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
150 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
151 ASSERT(rblock->bb_level != 0);
152 dblock->bb_level = rblock->bb_level;
153 dblock->bb_numrecs = rblock->bb_numrecs;
154 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
155 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
156 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
157 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
158 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
159 dmxr = be16_to_cpu(dblock->bb_numrecs);
160 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
161 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
164 STATIC struct xfs_btree_cur *
166 struct xfs_btree_cur *cur)
168 struct xfs_btree_cur *new;
170 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
171 cur->bc_ino.ip, cur->bc_ino.whichfork);
174 * Copy the firstblock, dfops, and flags values,
175 * since init cursor doesn't get them.
177 new->bc_ino.flags = cur->bc_ino.flags;
183 xfs_bmbt_update_cursor(
184 struct xfs_btree_cur *src,
185 struct xfs_btree_cur *dst)
187 ASSERT((dst->bc_tp->t_firstblock != NULLFSBLOCK) ||
188 (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
190 dst->bc_ino.allocated += src->bc_ino.allocated;
191 dst->bc_tp->t_firstblock = src->bc_tp->t_firstblock;
193 src->bc_ino.allocated = 0;
197 xfs_bmbt_alloc_block(
198 struct xfs_btree_cur *cur,
199 const union xfs_btree_ptr *start,
200 union xfs_btree_ptr *new,
203 xfs_alloc_arg_t args; /* block allocation args */
204 int error; /* error return value */
206 memset(&args, 0, sizeof(args));
207 args.tp = cur->bc_tp;
208 args.mp = cur->bc_mp;
209 args.fsbno = cur->bc_tp->t_firstblock;
210 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
211 cur->bc_ino.whichfork);
213 if (args.fsbno == NULLFSBLOCK) {
214 args.fsbno = be64_to_cpu(start->l);
215 args.type = XFS_ALLOCTYPE_START_BNO;
217 * Make sure there is sufficient room left in the AG to
218 * complete a full tree split for an extent insert. If
219 * we are converting the middle part of an extent then
220 * we may need space for two tree splits.
222 * We are relying on the caller to make the correct block
223 * reservation for this operation to succeed. If the
224 * reservation amount is insufficient then we may fail a
225 * block allocation here and corrupt the filesystem.
227 args.minleft = args.tp->t_blk_res;
228 } else if (cur->bc_tp->t_flags & XFS_TRANS_LOWMODE) {
229 args.type = XFS_ALLOCTYPE_START_BNO;
231 args.type = XFS_ALLOCTYPE_NEAR_BNO;
234 args.minlen = args.maxlen = args.prod = 1;
235 args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
236 if (!args.wasdel && args.tp->t_blk_res == 0) {
240 error = xfs_alloc_vextent(&args);
244 if (args.fsbno == NULLFSBLOCK && args.minleft) {
246 * Could not find an AG with enough free space to satisfy
247 * a full btree split. Try again and if
248 * successful activate the lowspace algorithm.
251 args.type = XFS_ALLOCTYPE_FIRST_AG;
252 error = xfs_alloc_vextent(&args);
255 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
257 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
262 ASSERT(args.len == 1);
263 cur->bc_tp->t_firstblock = args.fsbno;
264 cur->bc_ino.allocated++;
265 cur->bc_ino.ip->i_nblocks++;
266 xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
267 xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
268 XFS_TRANS_DQ_BCOUNT, 1L);
270 new->l = cpu_to_be64(args.fsbno);
281 struct xfs_btree_cur *cur,
284 struct xfs_mount *mp = cur->bc_mp;
285 struct xfs_inode *ip = cur->bc_ino.ip;
286 struct xfs_trans *tp = cur->bc_tp;
287 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
288 struct xfs_owner_info oinfo;
290 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
291 xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
294 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
295 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
300 xfs_bmbt_get_minrecs(
301 struct xfs_btree_cur *cur,
304 if (level == cur->bc_nlevels - 1) {
305 struct xfs_ifork *ifp;
307 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
308 cur->bc_ino.whichfork);
310 return xfs_bmbt_maxrecs(cur->bc_mp,
311 ifp->if_broot_bytes, level == 0) / 2;
314 return cur->bc_mp->m_bmap_dmnr[level != 0];
318 xfs_bmbt_get_maxrecs(
319 struct xfs_btree_cur *cur,
322 if (level == cur->bc_nlevels - 1) {
323 struct xfs_ifork *ifp;
325 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
326 cur->bc_ino.whichfork);
328 return xfs_bmbt_maxrecs(cur->bc_mp,
329 ifp->if_broot_bytes, level == 0);
332 return cur->bc_mp->m_bmap_dmxr[level != 0];
337 * Get the maximum records we could store in the on-disk format.
339 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
340 * for the root node this checks the available space in the dinode fork
341 * so that we can resize the in-memory buffer to match it. After a
342 * resize to the maximum size this function returns the same value
343 * as xfs_bmbt_get_maxrecs for the root node, too.
346 xfs_bmbt_get_dmaxrecs(
347 struct xfs_btree_cur *cur,
350 if (level != cur->bc_nlevels - 1)
351 return cur->bc_mp->m_bmap_dmxr[level != 0];
352 return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
356 xfs_bmbt_init_key_from_rec(
357 union xfs_btree_key *key,
358 const union xfs_btree_rec *rec)
360 key->bmbt.br_startoff =
361 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
365 xfs_bmbt_init_high_key_from_rec(
366 union xfs_btree_key *key,
367 const union xfs_btree_rec *rec)
369 key->bmbt.br_startoff = cpu_to_be64(
370 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
371 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
375 xfs_bmbt_init_rec_from_cur(
376 struct xfs_btree_cur *cur,
377 union xfs_btree_rec *rec)
379 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
383 xfs_bmbt_init_ptr_from_cur(
384 struct xfs_btree_cur *cur,
385 union xfs_btree_ptr *ptr)
392 struct xfs_btree_cur *cur,
393 const union xfs_btree_key *key)
395 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
396 cur->bc_rec.b.br_startoff;
400 xfs_bmbt_diff_two_keys(
401 struct xfs_btree_cur *cur,
402 const union xfs_btree_key *k1,
403 const union xfs_btree_key *k2)
405 uint64_t a = be64_to_cpu(k1->bmbt.br_startoff);
406 uint64_t b = be64_to_cpu(k2->bmbt.br_startoff);
409 * Note: This routine previously casted a and b to int64 and subtracted
410 * them to generate a result. This lead to problems if b was the
411 * "maximum" key value (all ones) being signed incorrectly, hence this
412 * somewhat less efficient version.
421 static xfs_failaddr_t
425 struct xfs_mount *mp = bp->b_mount;
426 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
430 if (!xfs_verify_magic(bp, block->bb_magic))
431 return __this_address;
433 if (xfs_has_crc(mp)) {
435 * XXX: need a better way of verifying the owner here. Right now
436 * just make sure there has been one set.
438 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
444 * numrecs and level verification.
446 * We don't know what fork we belong to, so just verify that the level
447 * is less than the maximum of the two. Later checks will be more
450 level = be16_to_cpu(block->bb_level);
451 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
452 return __this_address;
454 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
458 xfs_bmbt_read_verify(
463 if (!xfs_btree_lblock_verify_crc(bp))
464 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
466 fa = xfs_bmbt_verify(bp);
468 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
472 trace_xfs_btree_corrupt(bp, _RET_IP_);
476 xfs_bmbt_write_verify(
481 fa = xfs_bmbt_verify(bp);
483 trace_xfs_btree_corrupt(bp, _RET_IP_);
484 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
487 xfs_btree_lblock_calc_crc(bp);
490 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
492 .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
493 cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
494 .verify_read = xfs_bmbt_read_verify,
495 .verify_write = xfs_bmbt_write_verify,
496 .verify_struct = xfs_bmbt_verify,
501 xfs_bmbt_keys_inorder(
502 struct xfs_btree_cur *cur,
503 const union xfs_btree_key *k1,
504 const union xfs_btree_key *k2)
506 return be64_to_cpu(k1->bmbt.br_startoff) <
507 be64_to_cpu(k2->bmbt.br_startoff);
511 xfs_bmbt_recs_inorder(
512 struct xfs_btree_cur *cur,
513 const union xfs_btree_rec *r1,
514 const union xfs_btree_rec *r2)
516 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
517 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
518 xfs_bmbt_disk_get_startoff(&r2->bmbt);
521 static const struct xfs_btree_ops xfs_bmbt_ops = {
522 .rec_len = sizeof(xfs_bmbt_rec_t),
523 .key_len = sizeof(xfs_bmbt_key_t),
525 .dup_cursor = xfs_bmbt_dup_cursor,
526 .update_cursor = xfs_bmbt_update_cursor,
527 .alloc_block = xfs_bmbt_alloc_block,
528 .free_block = xfs_bmbt_free_block,
529 .get_maxrecs = xfs_bmbt_get_maxrecs,
530 .get_minrecs = xfs_bmbt_get_minrecs,
531 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
532 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
533 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
534 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
535 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
536 .key_diff = xfs_bmbt_key_diff,
537 .diff_two_keys = xfs_bmbt_diff_two_keys,
538 .buf_ops = &xfs_bmbt_buf_ops,
539 .keys_inorder = xfs_bmbt_keys_inorder,
540 .recs_inorder = xfs_bmbt_recs_inorder,
544 * Allocate a new bmap btree cursor.
546 struct xfs_btree_cur * /* new bmap btree cursor */
547 xfs_bmbt_init_cursor(
548 struct xfs_mount *mp, /* file system mount point */
549 struct xfs_trans *tp, /* transaction pointer */
550 struct xfs_inode *ip, /* inode owning the btree */
551 int whichfork) /* data or attr fork */
553 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
554 struct xfs_btree_cur *cur;
555 ASSERT(whichfork != XFS_COW_FORK);
557 cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
558 mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
559 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
560 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
562 cur->bc_ops = &xfs_bmbt_ops;
563 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
565 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
567 cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
569 cur->bc_ino.allocated = 0;
570 cur->bc_ino.flags = 0;
571 cur->bc_ino.whichfork = whichfork;
576 /* Calculate number of records in a block mapping btree block. */
577 static inline unsigned int
578 xfs_bmbt_block_maxrecs(
579 unsigned int blocklen,
583 return blocklen / sizeof(xfs_bmbt_rec_t);
584 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
588 * Calculate number of records in a bmap btree block.
592 struct xfs_mount *mp,
596 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
597 return xfs_bmbt_block_maxrecs(blocklen, leaf);
601 * Calculate the maximum possible height of the btree that the on-disk format
602 * supports. This is used for sizing structures large enough to support every
603 * possible configuration of a filesystem that might get mounted.
606 xfs_bmbt_maxlevels_ondisk(void)
608 unsigned int minrecs[2];
609 unsigned int blocklen;
611 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
612 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
614 minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
615 minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
617 /* One extra level for the inode root. */
618 return xfs_btree_compute_maxlevels(minrecs,
619 XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
623 * Calculate number of records in a bmap btree inode root.
630 blocklen -= sizeof(xfs_bmdr_block_t);
633 return blocklen / sizeof(xfs_bmdr_rec_t);
634 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
638 * Change the owner of a btree format fork fo the inode passed in. Change it to
639 * the owner of that is passed in so that we can change owners before or after
640 * we switch forks between inodes. The operation that the caller is doing will
641 * determine whether is needs to change owner before or after the switch.
643 * For demand paged transactional modification, the fork switch should be done
644 * after reading in all the blocks, modifying them and pinning them in the
645 * transaction. For modification when the buffers are already pinned in memory,
646 * the fork switch can be done before changing the owner as we won't need to
647 * validate the owner until the btree buffers are unpinned and writes can occur
650 * For recovery based ownership change, there is no transactional context and
651 * so a buffer list must be supplied so that we can record the buffers that we
652 * modified for the caller to issue IO on.
655 xfs_bmbt_change_owner(
656 struct xfs_trans *tp,
657 struct xfs_inode *ip,
660 struct list_head *buffer_list)
662 struct xfs_btree_cur *cur;
665 ASSERT(tp || buffer_list);
666 ASSERT(!(tp && buffer_list));
667 ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
669 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
670 cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
672 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
673 xfs_btree_del_cursor(cur, error);
677 /* Calculate the bmap btree size for some records. */
680 struct xfs_mount *mp,
681 unsigned long long len)
683 return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
687 xfs_bmbt_init_cur_cache(void)
689 xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
690 xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
693 if (!xfs_bmbt_cur_cache)
699 xfs_bmbt_destroy_cur_cache(void)
701 kmem_cache_destroy(xfs_bmbt_cur_cache);
702 xfs_bmbt_cur_cache = NULL;