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 * Convert on-disk form of btree root to in-memory form.
31 xfs_bmdr_block_t *dblock,
33 struct xfs_btree_block *rblock,
36 struct xfs_mount *mp = ip->i_mount;
43 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
44 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
46 rblock->bb_level = dblock->bb_level;
47 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
48 rblock->bb_numrecs = dblock->bb_numrecs;
49 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
50 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
51 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
52 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
53 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
54 dmxr = be16_to_cpu(dblock->bb_numrecs);
55 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
56 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
60 xfs_bmbt_disk_get_all(
61 struct xfs_bmbt_rec *rec,
62 struct xfs_bmbt_irec *irec)
64 uint64_t l0 = get_unaligned_be64(&rec->l0);
65 uint64_t l1 = get_unaligned_be64(&rec->l1);
67 irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
68 irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
69 irec->br_blockcount = l1 & xfs_mask64lo(21);
70 if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
71 irec->br_state = XFS_EXT_UNWRITTEN;
73 irec->br_state = XFS_EXT_NORM;
77 * Extract the blockcount field from an on disk bmap extent record.
80 xfs_bmbt_disk_get_blockcount(
83 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
87 * Extract the startoff field from a disk format bmap extent record.
90 xfs_bmbt_disk_get_startoff(
93 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
94 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
98 * Set all the fields in a bmap extent record from the uncompressed form.
101 xfs_bmbt_disk_set_all(
102 struct xfs_bmbt_rec *r,
103 struct xfs_bmbt_irec *s)
105 int extent_flag = (s->br_state != XFS_EXT_NORM);
107 ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
108 ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
109 ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
110 ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
113 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
114 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
115 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
117 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
118 ((xfs_bmbt_rec_base_t)s->br_blockcount &
119 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
123 * Convert in-memory form of btree root to on-disk form.
127 struct xfs_mount *mp,
128 struct xfs_btree_block *rblock,
130 xfs_bmdr_block_t *dblock,
139 if (xfs_sb_version_hascrc(&mp->m_sb)) {
140 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
141 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
142 &mp->m_sb.sb_meta_uuid));
143 ASSERT(rblock->bb_u.l.bb_blkno ==
144 cpu_to_be64(XFS_BUF_DADDR_NULL));
146 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
147 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
148 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
149 ASSERT(rblock->bb_level != 0);
150 dblock->bb_level = rblock->bb_level;
151 dblock->bb_numrecs = rblock->bb_numrecs;
152 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
153 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
154 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
155 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
156 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
157 dmxr = be16_to_cpu(dblock->bb_numrecs);
158 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
159 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
162 STATIC struct xfs_btree_cur *
164 struct xfs_btree_cur *cur)
166 struct xfs_btree_cur *new;
168 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
169 cur->bc_private.b.ip, cur->bc_private.b.whichfork);
172 * Copy the firstblock, dfops, and flags values,
173 * since init cursor doesn't get them.
175 new->bc_private.b.flags = cur->bc_private.b.flags;
181 xfs_bmbt_update_cursor(
182 struct xfs_btree_cur *src,
183 struct xfs_btree_cur *dst)
185 ASSERT((dst->bc_tp->t_firstblock != NULLFSBLOCK) ||
186 (dst->bc_private.b.ip->i_d.di_flags & XFS_DIFLAG_REALTIME));
188 dst->bc_private.b.allocated += src->bc_private.b.allocated;
189 dst->bc_tp->t_firstblock = src->bc_tp->t_firstblock;
191 src->bc_private.b.allocated = 0;
195 xfs_bmbt_alloc_block(
196 struct xfs_btree_cur *cur,
197 union xfs_btree_ptr *start,
198 union xfs_btree_ptr *new,
201 xfs_alloc_arg_t args; /* block allocation args */
202 int error; /* error return value */
204 memset(&args, 0, sizeof(args));
205 args.tp = cur->bc_tp;
206 args.mp = cur->bc_mp;
207 args.fsbno = cur->bc_tp->t_firstblock;
208 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_private.b.ip->i_ino,
209 cur->bc_private.b.whichfork);
211 if (args.fsbno == NULLFSBLOCK) {
212 args.fsbno = be64_to_cpu(start->l);
213 args.type = XFS_ALLOCTYPE_START_BNO;
215 * Make sure there is sufficient room left in the AG to
216 * complete a full tree split for an extent insert. If
217 * we are converting the middle part of an extent then
218 * we may need space for two tree splits.
220 * We are relying on the caller to make the correct block
221 * reservation for this operation to succeed. If the
222 * reservation amount is insufficient then we may fail a
223 * block allocation here and corrupt the filesystem.
225 args.minleft = args.tp->t_blk_res;
226 } else if (cur->bc_tp->t_flags & XFS_TRANS_LOWMODE) {
227 args.type = XFS_ALLOCTYPE_START_BNO;
229 args.type = XFS_ALLOCTYPE_NEAR_BNO;
232 args.minlen = args.maxlen = args.prod = 1;
233 args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
234 if (!args.wasdel && args.tp->t_blk_res == 0) {
238 error = xfs_alloc_vextent(&args);
242 if (args.fsbno == NULLFSBLOCK && args.minleft) {
244 * Could not find an AG with enough free space to satisfy
245 * a full btree split. Try again and if
246 * successful activate the lowspace algorithm.
249 args.type = XFS_ALLOCTYPE_FIRST_AG;
250 error = xfs_alloc_vextent(&args);
253 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
255 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
260 ASSERT(args.len == 1);
261 cur->bc_tp->t_firstblock = args.fsbno;
262 cur->bc_private.b.allocated++;
263 cur->bc_private.b.ip->i_d.di_nblocks++;
264 xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
265 xfs_trans_mod_dquot_byino(args.tp, cur->bc_private.b.ip,
266 XFS_TRANS_DQ_BCOUNT, 1L);
268 new->l = cpu_to_be64(args.fsbno);
279 struct xfs_btree_cur *cur,
282 struct xfs_mount *mp = cur->bc_mp;
283 struct xfs_inode *ip = cur->bc_private.b.ip;
284 struct xfs_trans *tp = cur->bc_tp;
285 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
286 struct xfs_owner_info oinfo;
288 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_private.b.whichfork);
289 xfs_bmap_add_free(cur->bc_tp, fsbno, 1, &oinfo);
290 ip->i_d.di_nblocks--;
292 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
293 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
298 xfs_bmbt_get_minrecs(
299 struct xfs_btree_cur *cur,
302 if (level == cur->bc_nlevels - 1) {
303 struct xfs_ifork *ifp;
305 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
306 cur->bc_private.b.whichfork);
308 return xfs_bmbt_maxrecs(cur->bc_mp,
309 ifp->if_broot_bytes, level == 0) / 2;
312 return cur->bc_mp->m_bmap_dmnr[level != 0];
316 xfs_bmbt_get_maxrecs(
317 struct xfs_btree_cur *cur,
320 if (level == cur->bc_nlevels - 1) {
321 struct xfs_ifork *ifp;
323 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
324 cur->bc_private.b.whichfork);
326 return xfs_bmbt_maxrecs(cur->bc_mp,
327 ifp->if_broot_bytes, level == 0);
330 return cur->bc_mp->m_bmap_dmxr[level != 0];
335 * Get the maximum records we could store in the on-disk format.
337 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
338 * for the root node this checks the available space in the dinode fork
339 * so that we can resize the in-memory buffer to match it. After a
340 * resize to the maximum size this function returns the same value
341 * as xfs_bmbt_get_maxrecs for the root node, too.
344 xfs_bmbt_get_dmaxrecs(
345 struct xfs_btree_cur *cur,
348 if (level != cur->bc_nlevels - 1)
349 return cur->bc_mp->m_bmap_dmxr[level != 0];
350 return xfs_bmdr_maxrecs(cur->bc_private.b.forksize, level == 0);
354 xfs_bmbt_init_key_from_rec(
355 union xfs_btree_key *key,
356 union xfs_btree_rec *rec)
358 key->bmbt.br_startoff =
359 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
363 xfs_bmbt_init_high_key_from_rec(
364 union xfs_btree_key *key,
365 union xfs_btree_rec *rec)
367 key->bmbt.br_startoff = cpu_to_be64(
368 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
369 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
373 xfs_bmbt_init_rec_from_cur(
374 struct xfs_btree_cur *cur,
375 union xfs_btree_rec *rec)
377 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
381 xfs_bmbt_init_ptr_from_cur(
382 struct xfs_btree_cur *cur,
383 union xfs_btree_ptr *ptr)
390 struct xfs_btree_cur *cur,
391 union xfs_btree_key *key)
393 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
394 cur->bc_rec.b.br_startoff;
398 xfs_bmbt_diff_two_keys(
399 struct xfs_btree_cur *cur,
400 union xfs_btree_key *k1,
401 union xfs_btree_key *k2)
403 return (int64_t)be64_to_cpu(k1->bmbt.br_startoff) -
404 be64_to_cpu(k2->bmbt.br_startoff);
407 static xfs_failaddr_t
411 struct xfs_mount *mp = bp->b_mount;
412 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
416 if (!xfs_verify_magic(bp, block->bb_magic))
417 return __this_address;
419 if (xfs_sb_version_hascrc(&mp->m_sb)) {
421 * XXX: need a better way of verifying the owner here. Right now
422 * just make sure there has been one set.
424 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
430 * numrecs and level verification.
432 * We don't know what fork we belong to, so just verify that the level
433 * is less than the maximum of the two. Later checks will be more
436 level = be16_to_cpu(block->bb_level);
437 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
438 return __this_address;
440 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
444 xfs_bmbt_read_verify(
449 if (!xfs_btree_lblock_verify_crc(bp))
450 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
452 fa = xfs_bmbt_verify(bp);
454 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
458 trace_xfs_btree_corrupt(bp, _RET_IP_);
462 xfs_bmbt_write_verify(
467 fa = xfs_bmbt_verify(bp);
469 trace_xfs_btree_corrupt(bp, _RET_IP_);
470 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
473 xfs_btree_lblock_calc_crc(bp);
476 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
478 .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
479 cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
480 .verify_read = xfs_bmbt_read_verify,
481 .verify_write = xfs_bmbt_write_verify,
482 .verify_struct = xfs_bmbt_verify,
487 xfs_bmbt_keys_inorder(
488 struct xfs_btree_cur *cur,
489 union xfs_btree_key *k1,
490 union xfs_btree_key *k2)
492 return be64_to_cpu(k1->bmbt.br_startoff) <
493 be64_to_cpu(k2->bmbt.br_startoff);
497 xfs_bmbt_recs_inorder(
498 struct xfs_btree_cur *cur,
499 union xfs_btree_rec *r1,
500 union xfs_btree_rec *r2)
502 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
503 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
504 xfs_bmbt_disk_get_startoff(&r2->bmbt);
507 static const struct xfs_btree_ops xfs_bmbt_ops = {
508 .rec_len = sizeof(xfs_bmbt_rec_t),
509 .key_len = sizeof(xfs_bmbt_key_t),
511 .dup_cursor = xfs_bmbt_dup_cursor,
512 .update_cursor = xfs_bmbt_update_cursor,
513 .alloc_block = xfs_bmbt_alloc_block,
514 .free_block = xfs_bmbt_free_block,
515 .get_maxrecs = xfs_bmbt_get_maxrecs,
516 .get_minrecs = xfs_bmbt_get_minrecs,
517 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
518 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
519 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
520 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
521 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
522 .key_diff = xfs_bmbt_key_diff,
523 .diff_two_keys = xfs_bmbt_diff_two_keys,
524 .buf_ops = &xfs_bmbt_buf_ops,
525 .keys_inorder = xfs_bmbt_keys_inorder,
526 .recs_inorder = xfs_bmbt_recs_inorder,
530 * Allocate a new bmap btree cursor.
532 struct xfs_btree_cur * /* new bmap btree cursor */
533 xfs_bmbt_init_cursor(
534 struct xfs_mount *mp, /* file system mount point */
535 struct xfs_trans *tp, /* transaction pointer */
536 struct xfs_inode *ip, /* inode owning the btree */
537 int whichfork) /* data or attr fork */
539 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
540 struct xfs_btree_cur *cur;
541 ASSERT(whichfork != XFS_COW_FORK);
543 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
547 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
548 cur->bc_btnum = XFS_BTNUM_BMAP;
549 cur->bc_blocklog = mp->m_sb.sb_blocklog;
550 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
552 cur->bc_ops = &xfs_bmbt_ops;
553 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
554 if (xfs_sb_version_hascrc(&mp->m_sb))
555 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
557 cur->bc_private.b.forksize = XFS_IFORK_SIZE(ip, whichfork);
558 cur->bc_private.b.ip = ip;
559 cur->bc_private.b.allocated = 0;
560 cur->bc_private.b.flags = 0;
561 cur->bc_private.b.whichfork = whichfork;
567 * Calculate number of records in a bmap btree block.
571 struct xfs_mount *mp,
575 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
578 return blocklen / sizeof(xfs_bmbt_rec_t);
579 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
583 * Calculate number of records in a bmap btree inode root.
590 blocklen -= sizeof(xfs_bmdr_block_t);
593 return blocklen / sizeof(xfs_bmdr_rec_t);
594 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
598 * Change the owner of a btree format fork fo the inode passed in. Change it to
599 * the owner of that is passed in so that we can change owners before or after
600 * we switch forks between inodes. The operation that the caller is doing will
601 * determine whether is needs to change owner before or after the switch.
603 * For demand paged transactional modification, the fork switch should be done
604 * after reading in all the blocks, modifying them and pinning them in the
605 * transaction. For modification when the buffers are already pinned in memory,
606 * the fork switch can be done before changing the owner as we won't need to
607 * validate the owner until the btree buffers are unpinned and writes can occur
610 * For recovery based ownership change, there is no transactional context and
611 * so a buffer list must be supplied so that we can record the buffers that we
612 * modified for the caller to issue IO on.
615 xfs_bmbt_change_owner(
616 struct xfs_trans *tp,
617 struct xfs_inode *ip,
620 struct list_head *buffer_list)
622 struct xfs_btree_cur *cur;
625 ASSERT(tp || buffer_list);
626 ASSERT(!(tp && buffer_list));
627 if (whichfork == XFS_DATA_FORK)
628 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_BTREE);
630 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE);
632 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
635 cur->bc_private.b.flags |= XFS_BTCUR_BPRV_INVALID_OWNER;
637 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
638 xfs_btree_del_cursor(cur, error);
642 /* Calculate the bmap btree size for some records. */
645 struct xfs_mount *mp,
646 unsigned long long len)
648 return xfs_btree_calc_size(mp->m_bmap_dmnr, len);