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_defer.h"
15 #include "xfs_inode.h"
16 #include "xfs_trans.h"
17 #include "xfs_inode_item.h"
18 #include "xfs_alloc.h"
19 #include "xfs_btree.h"
20 #include "xfs_bmap_btree.h"
22 #include "xfs_error.h"
23 #include "xfs_quota.h"
24 #include "xfs_trace.h"
25 #include "xfs_cksum.h"
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 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(
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(
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_sb_version_hascrc(&mp->m_sb)) {
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_private.b.ip, cur->bc_private.b.whichfork);
175 * Copy the firstblock, dfops, and flags values,
176 * since init cursor doesn't get them.
178 new->bc_private.b.firstblock = cur->bc_private.b.firstblock;
179 new->bc_private.b.dfops = cur->bc_private.b.dfops;
180 new->bc_private.b.flags = cur->bc_private.b.flags;
186 xfs_bmbt_update_cursor(
187 struct xfs_btree_cur *src,
188 struct xfs_btree_cur *dst)
190 ASSERT((dst->bc_private.b.firstblock != NULLFSBLOCK) ||
191 (dst->bc_private.b.ip->i_d.di_flags & XFS_DIFLAG_REALTIME));
192 ASSERT(dst->bc_private.b.dfops == src->bc_private.b.dfops);
194 dst->bc_private.b.allocated += src->bc_private.b.allocated;
195 dst->bc_private.b.firstblock = src->bc_private.b.firstblock;
197 src->bc_private.b.allocated = 0;
201 xfs_bmbt_alloc_block(
202 struct xfs_btree_cur *cur,
203 union xfs_btree_ptr *start,
204 union xfs_btree_ptr *new,
207 xfs_alloc_arg_t args; /* block allocation args */
208 int error; /* error return value */
210 memset(&args, 0, sizeof(args));
211 args.tp = cur->bc_tp;
212 args.mp = cur->bc_mp;
213 args.fsbno = cur->bc_private.b.firstblock;
214 args.firstblock = args.fsbno;
215 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_private.b.ip->i_ino,
216 cur->bc_private.b.whichfork);
218 if (args.fsbno == NULLFSBLOCK) {
219 args.fsbno = be64_to_cpu(start->l);
220 args.type = XFS_ALLOCTYPE_START_BNO;
222 * Make sure there is sufficient room left in the AG to
223 * complete a full tree split for an extent insert. If
224 * we are converting the middle part of an extent then
225 * we may need space for two tree splits.
227 * We are relying on the caller to make the correct block
228 * reservation for this operation to succeed. If the
229 * reservation amount is insufficient then we may fail a
230 * block allocation here and corrupt the filesystem.
232 args.minleft = args.tp->t_blk_res;
233 } else if (cur->bc_private.b.dfops->dop_low) {
234 args.type = XFS_ALLOCTYPE_START_BNO;
236 args.type = XFS_ALLOCTYPE_NEAR_BNO;
239 args.minlen = args.maxlen = args.prod = 1;
240 args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
241 if (!args.wasdel && args.tp->t_blk_res == 0) {
245 error = xfs_alloc_vextent(&args);
249 if (args.fsbno == NULLFSBLOCK && args.minleft) {
251 * Could not find an AG with enough free space to satisfy
252 * a full btree split. Try again and if
253 * successful activate the lowspace algorithm.
256 args.type = XFS_ALLOCTYPE_FIRST_AG;
257 error = xfs_alloc_vextent(&args);
260 cur->bc_private.b.dfops->dop_low = true;
262 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
267 ASSERT(args.len == 1);
268 cur->bc_private.b.firstblock = args.fsbno;
269 cur->bc_private.b.allocated++;
270 cur->bc_private.b.ip->i_d.di_nblocks++;
271 xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
272 xfs_trans_mod_dquot_byino(args.tp, cur->bc_private.b.ip,
273 XFS_TRANS_DQ_BCOUNT, 1L);
275 new->l = cpu_to_be64(args.fsbno);
286 struct xfs_btree_cur *cur,
289 struct xfs_mount *mp = cur->bc_mp;
290 struct xfs_inode *ip = cur->bc_private.b.ip;
291 struct xfs_trans *tp = cur->bc_tp;
292 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
293 struct xfs_owner_info oinfo;
295 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_private.b.whichfork);
296 xfs_bmap_add_free(mp, cur->bc_private.b.dfops, fsbno, 1, &oinfo);
297 ip->i_d.di_nblocks--;
299 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
300 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
305 xfs_bmbt_get_minrecs(
306 struct xfs_btree_cur *cur,
309 if (level == cur->bc_nlevels - 1) {
310 struct xfs_ifork *ifp;
312 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
313 cur->bc_private.b.whichfork);
315 return xfs_bmbt_maxrecs(cur->bc_mp,
316 ifp->if_broot_bytes, level == 0) / 2;
319 return cur->bc_mp->m_bmap_dmnr[level != 0];
323 xfs_bmbt_get_maxrecs(
324 struct xfs_btree_cur *cur,
327 if (level == cur->bc_nlevels - 1) {
328 struct xfs_ifork *ifp;
330 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
331 cur->bc_private.b.whichfork);
333 return xfs_bmbt_maxrecs(cur->bc_mp,
334 ifp->if_broot_bytes, level == 0);
337 return cur->bc_mp->m_bmap_dmxr[level != 0];
342 * Get the maximum records we could store in the on-disk format.
344 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
345 * for the root node this checks the available space in the dinode fork
346 * so that we can resize the in-memory buffer to match it. After a
347 * resize to the maximum size this function returns the same value
348 * as xfs_bmbt_get_maxrecs for the root node, too.
351 xfs_bmbt_get_dmaxrecs(
352 struct xfs_btree_cur *cur,
355 if (level != cur->bc_nlevels - 1)
356 return cur->bc_mp->m_bmap_dmxr[level != 0];
357 return xfs_bmdr_maxrecs(cur->bc_private.b.forksize, level == 0);
361 xfs_bmbt_init_key_from_rec(
362 union xfs_btree_key *key,
363 union xfs_btree_rec *rec)
365 key->bmbt.br_startoff =
366 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
370 xfs_bmbt_init_high_key_from_rec(
371 union xfs_btree_key *key,
372 union xfs_btree_rec *rec)
374 key->bmbt.br_startoff = cpu_to_be64(
375 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
376 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
380 xfs_bmbt_init_rec_from_cur(
381 struct xfs_btree_cur *cur,
382 union xfs_btree_rec *rec)
384 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
388 xfs_bmbt_init_ptr_from_cur(
389 struct xfs_btree_cur *cur,
390 union xfs_btree_ptr *ptr)
397 struct xfs_btree_cur *cur,
398 union xfs_btree_key *key)
400 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
401 cur->bc_rec.b.br_startoff;
405 xfs_bmbt_diff_two_keys(
406 struct xfs_btree_cur *cur,
407 union xfs_btree_key *k1,
408 union xfs_btree_key *k2)
410 return (int64_t)be64_to_cpu(k1->bmbt.br_startoff) -
411 be64_to_cpu(k2->bmbt.br_startoff);
414 static xfs_failaddr_t
418 struct xfs_mount *mp = bp->b_target->bt_mount;
419 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
423 switch (block->bb_magic) {
424 case cpu_to_be32(XFS_BMAP_CRC_MAGIC):
426 * XXX: need a better way of verifying the owner here. Right now
427 * just make sure there has been one set.
429 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
433 case cpu_to_be32(XFS_BMAP_MAGIC):
436 return __this_address;
440 * numrecs and level verification.
442 * We don't know what fork we belong to, so just verify that the level
443 * is less than the maximum of the two. Later checks will be more
446 level = be16_to_cpu(block->bb_level);
447 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
448 return __this_address;
450 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
454 xfs_bmbt_read_verify(
459 if (!xfs_btree_lblock_verify_crc(bp))
460 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
462 fa = xfs_bmbt_verify(bp);
464 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
468 trace_xfs_btree_corrupt(bp, _RET_IP_);
472 xfs_bmbt_write_verify(
477 fa = xfs_bmbt_verify(bp);
479 trace_xfs_btree_corrupt(bp, _RET_IP_);
480 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
483 xfs_btree_lblock_calc_crc(bp);
486 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
488 .verify_read = xfs_bmbt_read_verify,
489 .verify_write = xfs_bmbt_write_verify,
490 .verify_struct = xfs_bmbt_verify,
495 xfs_bmbt_keys_inorder(
496 struct xfs_btree_cur *cur,
497 union xfs_btree_key *k1,
498 union xfs_btree_key *k2)
500 return be64_to_cpu(k1->bmbt.br_startoff) <
501 be64_to_cpu(k2->bmbt.br_startoff);
505 xfs_bmbt_recs_inorder(
506 struct xfs_btree_cur *cur,
507 union xfs_btree_rec *r1,
508 union xfs_btree_rec *r2)
510 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
511 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
512 xfs_bmbt_disk_get_startoff(&r2->bmbt);
515 static const struct xfs_btree_ops xfs_bmbt_ops = {
516 .rec_len = sizeof(xfs_bmbt_rec_t),
517 .key_len = sizeof(xfs_bmbt_key_t),
519 .dup_cursor = xfs_bmbt_dup_cursor,
520 .update_cursor = xfs_bmbt_update_cursor,
521 .alloc_block = xfs_bmbt_alloc_block,
522 .free_block = xfs_bmbt_free_block,
523 .get_maxrecs = xfs_bmbt_get_maxrecs,
524 .get_minrecs = xfs_bmbt_get_minrecs,
525 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
526 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
527 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
528 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
529 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
530 .key_diff = xfs_bmbt_key_diff,
531 .diff_two_keys = xfs_bmbt_diff_two_keys,
532 .buf_ops = &xfs_bmbt_buf_ops,
533 .keys_inorder = xfs_bmbt_keys_inorder,
534 .recs_inorder = xfs_bmbt_recs_inorder,
538 * Allocate a new bmap btree cursor.
540 struct xfs_btree_cur * /* new bmap btree cursor */
541 xfs_bmbt_init_cursor(
542 struct xfs_mount *mp, /* file system mount point */
543 struct xfs_trans *tp, /* transaction pointer */
544 struct xfs_inode *ip, /* inode owning the btree */
545 int whichfork) /* data or attr fork */
547 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
548 struct xfs_btree_cur *cur;
549 ASSERT(whichfork != XFS_COW_FORK);
551 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
555 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
556 cur->bc_btnum = XFS_BTNUM_BMAP;
557 cur->bc_blocklog = mp->m_sb.sb_blocklog;
558 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
560 cur->bc_ops = &xfs_bmbt_ops;
561 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
562 if (xfs_sb_version_hascrc(&mp->m_sb))
563 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
565 cur->bc_private.b.forksize = XFS_IFORK_SIZE(ip, whichfork);
566 cur->bc_private.b.ip = ip;
567 cur->bc_private.b.firstblock = NULLFSBLOCK;
568 cur->bc_private.b.dfops = NULL;
569 cur->bc_private.b.allocated = 0;
570 cur->bc_private.b.flags = 0;
571 cur->bc_private.b.whichfork = whichfork;
577 * Calculate number of records in a bmap btree block.
581 struct xfs_mount *mp,
585 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
588 return blocklen / sizeof(xfs_bmbt_rec_t);
589 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
593 * Calculate number of records in a bmap btree inode root.
600 blocklen -= sizeof(xfs_bmdr_block_t);
603 return blocklen / sizeof(xfs_bmdr_rec_t);
604 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
608 * Change the owner of a btree format fork fo the inode passed in. Change it to
609 * the owner of that is passed in so that we can change owners before or after
610 * we switch forks between inodes. The operation that the caller is doing will
611 * determine whether is needs to change owner before or after the switch.
613 * For demand paged transactional modification, the fork switch should be done
614 * after reading in all the blocks, modifying them and pinning them in the
615 * transaction. For modification when the buffers are already pinned in memory,
616 * the fork switch can be done before changing the owner as we won't need to
617 * validate the owner until the btree buffers are unpinned and writes can occur
620 * For recovery based ownership change, there is no transactional context and
621 * so a buffer list must be supplied so that we can record the buffers that we
622 * modified for the caller to issue IO on.
625 xfs_bmbt_change_owner(
626 struct xfs_trans *tp,
627 struct xfs_inode *ip,
630 struct list_head *buffer_list)
632 struct xfs_btree_cur *cur;
635 ASSERT(tp || buffer_list);
636 ASSERT(!(tp && buffer_list));
637 if (whichfork == XFS_DATA_FORK)
638 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_BTREE);
640 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE);
642 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
645 cur->bc_private.b.flags |= XFS_BTCUR_BPRV_INVALID_OWNER;
647 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
648 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
652 /* Calculate the bmap btree size for some records. */
655 struct xfs_mount *mp,
656 unsigned long long len)
658 return xfs_btree_calc_size(mp->m_bmap_dmnr, len);