1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 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"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_inode_item.h"
16 #include "xfs_trace.h"
17 #include "xfs_trans_priv.h"
18 #include "xfs_buf_item.h"
20 #include "xfs_error.h"
21 #include "xfs_log_priv.h"
22 #include "xfs_log_recover.h"
23 #include "xfs_icache.h"
24 #include "xfs_bmap_btree.h"
25 #include "xfs_rtrmap_btree.h"
26 #include "xfs_rtrefcount_btree.h"
29 xlog_recover_inode_ra_pass2(
31 struct xlog_recover_item
*item
)
33 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
34 struct xfs_inode_log_format
*ilfp
= item
->ri_buf
[0].i_addr
;
36 xlog_buf_readahead(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
,
37 &xfs_inode_buf_ra_ops
);
39 struct xfs_inode_log_format_32
*ilfp
= item
->ri_buf
[0].i_addr
;
41 xlog_buf_readahead(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
,
42 &xfs_inode_buf_ra_ops
);
47 * Inode fork owner changes
49 * If we have been told that we have to reparent the inode fork, it's because an
50 * extent swap operation on a CRC enabled filesystem has been done and we are
51 * replaying it. We need to walk the BMBT of the appropriate fork and change the
54 * The complexity here is that we don't have an inode context to work with, so
55 * after we've replayed the inode we need to instantiate one. This is where the
58 * We are in the middle of log recovery, so we can't run transactions. That
59 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
60 * that will result in the corresponding iput() running the inode through
61 * xfs_inactive(). If we've just replayed an inode core that changes the link
62 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
63 * transactions (bad!).
65 * So, to avoid this, we instantiate an inode directly from the inode core we've
66 * just recovered. We have the buffer still locked, and all we really need to
67 * instantiate is the inode core and the forks being modified. We can do this
68 * manually, then run the inode btree owner change, and then tear down the
69 * xfs_inode without having to run any transactions at all.
71 * Also, because we don't have a transaction context available here but need to
72 * gather all the buffers we modify for writeback so we pass the buffer_list
73 * instead for the operation to use.
77 xfs_recover_inode_owner_change(
79 struct xfs_dinode
*dip
,
80 struct xfs_inode_log_format
*in_f
,
81 struct list_head
*buffer_list
)
86 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
88 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
92 /* instantiate the inode */
93 ASSERT(dip
->di_version
>= 3);
95 error
= xfs_inode_from_disk(ip
, dip
);
99 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
100 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
101 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
102 ip
->i_ino
, buffer_list
);
107 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
108 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
109 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
110 ip
->i_ino
, buffer_list
);
120 static inline bool xfs_log_dinode_has_bigtime(const struct xfs_log_dinode
*ld
)
122 return ld
->di_version
>= 3 &&
123 (ld
->di_flags2
& XFS_DIFLAG2_BIGTIME
);
126 /* Convert a log timestamp to an ondisk timestamp. */
127 static inline xfs_timestamp_t
128 xfs_log_dinode_to_disk_ts(
129 struct xfs_log_dinode
*from
,
130 const xfs_log_timestamp_t its
)
132 struct xfs_legacy_timestamp
*lts
;
133 struct xfs_log_legacy_timestamp
*lits
;
136 if (xfs_log_dinode_has_bigtime(from
))
137 return cpu_to_be64(its
);
139 lts
= (struct xfs_legacy_timestamp
*)&ts
;
140 lits
= (struct xfs_log_legacy_timestamp
*)&its
;
141 lts
->t_sec
= cpu_to_be32(lits
->t_sec
);
142 lts
->t_nsec
= cpu_to_be32(lits
->t_nsec
);
147 static inline bool xfs_log_dinode_has_large_extent_counts(
148 const struct xfs_log_dinode
*ld
)
150 return ld
->di_version
>= 3 &&
151 (ld
->di_flags2
& XFS_DIFLAG2_NREXT64
);
155 xfs_log_dinode_to_disk_iext_counters(
156 struct xfs_log_dinode
*from
,
157 struct xfs_dinode
*to
)
159 if (xfs_log_dinode_has_large_extent_counts(from
)) {
160 to
->di_big_nextents
= cpu_to_be64(from
->di_big_nextents
);
161 to
->di_big_anextents
= cpu_to_be32(from
->di_big_anextents
);
162 to
->di_nrext64_pad
= cpu_to_be16(from
->di_nrext64_pad
);
164 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
165 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
171 xfs_log_dinode_to_disk(
172 struct xfs_log_dinode
*from
,
173 struct xfs_dinode
*to
,
176 to
->di_magic
= cpu_to_be16(from
->di_magic
);
177 to
->di_mode
= cpu_to_be16(from
->di_mode
);
178 to
->di_version
= from
->di_version
;
179 to
->di_format
= from
->di_format
;
180 to
->di_metatype
= cpu_to_be16(from
->di_metatype
);
181 to
->di_uid
= cpu_to_be32(from
->di_uid
);
182 to
->di_gid
= cpu_to_be32(from
->di_gid
);
183 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
184 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
185 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
187 to
->di_atime
= xfs_log_dinode_to_disk_ts(from
, from
->di_atime
);
188 to
->di_mtime
= xfs_log_dinode_to_disk_ts(from
, from
->di_mtime
);
189 to
->di_ctime
= xfs_log_dinode_to_disk_ts(from
, from
->di_ctime
);
191 to
->di_size
= cpu_to_be64(from
->di_size
);
192 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
193 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
194 to
->di_forkoff
= from
->di_forkoff
;
195 to
->di_aformat
= from
->di_aformat
;
196 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
197 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
198 to
->di_flags
= cpu_to_be16(from
->di_flags
);
199 to
->di_gen
= cpu_to_be32(from
->di_gen
);
201 if (from
->di_version
== 3) {
202 to
->di_changecount
= cpu_to_be64(from
->di_changecount
);
203 to
->di_crtime
= xfs_log_dinode_to_disk_ts(from
,
205 to
->di_flags2
= cpu_to_be64(from
->di_flags2
);
206 to
->di_cowextsize
= cpu_to_be32(from
->di_cowextsize
);
207 to
->di_ino
= cpu_to_be64(from
->di_ino
);
208 to
->di_lsn
= cpu_to_be64(lsn
);
209 memset(to
->di_pad2
, 0, sizeof(to
->di_pad2
));
210 uuid_copy(&to
->di_uuid
, &from
->di_uuid
);
213 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
214 memset(to
->di_v2_pad
, 0, sizeof(to
->di_v2_pad
));
217 xfs_log_dinode_to_disk_iext_counters(from
, to
);
221 xlog_dinode_verify_extent_counts(
222 struct xfs_mount
*mp
,
223 struct xfs_log_dinode
*ldip
)
225 xfs_extnum_t nextents
;
226 xfs_aextnum_t anextents
;
228 if (xfs_log_dinode_has_large_extent_counts(ldip
)) {
229 if (!xfs_has_large_extent_counts(mp
) ||
230 (ldip
->di_nrext64_pad
!= 0)) {
231 XFS_CORRUPTION_ERROR(
232 "Bad log dinode large extent count format",
233 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
235 "Bad inode 0x%llx, large extent counts %d, padding 0x%x",
236 ldip
->di_ino
, xfs_has_large_extent_counts(mp
),
237 ldip
->di_nrext64_pad
);
238 return -EFSCORRUPTED
;
241 nextents
= ldip
->di_big_nextents
;
242 anextents
= ldip
->di_big_anextents
;
244 if (ldip
->di_version
== 3 && ldip
->di_v3_pad
!= 0) {
245 XFS_CORRUPTION_ERROR(
246 "Bad log dinode di_v3_pad",
247 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
249 "Bad inode 0x%llx, di_v3_pad 0x%llx",
250 ldip
->di_ino
, ldip
->di_v3_pad
);
251 return -EFSCORRUPTED
;
254 nextents
= ldip
->di_nextents
;
255 anextents
= ldip
->di_anextents
;
258 if (unlikely(nextents
+ anextents
> ldip
->di_nblocks
)) {
259 XFS_CORRUPTION_ERROR("Bad log dinode extent counts",
260 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
262 "Bad inode 0x%llx, large extent counts %d, nextents 0x%llx, anextents 0x%x, nblocks 0x%llx",
263 ldip
->di_ino
, xfs_has_large_extent_counts(mp
), nextents
,
264 anextents
, ldip
->di_nblocks
);
265 return -EFSCORRUPTED
;
272 xlog_recover_inode_dbroot(
273 struct xfs_mount
*mp
,
276 struct xfs_dinode
*dip
)
278 void *dfork
= XFS_DFORK_DPTR(dip
);
279 unsigned int dsize
= XFS_DFORK_DSIZE(dip
, mp
);
281 switch (dip
->di_format
) {
282 case XFS_DINODE_FMT_BTREE
:
283 xfs_bmbt_to_bmdr(mp
, src
, len
, dfork
, dsize
);
285 case XFS_DINODE_FMT_META_BTREE
:
286 switch (be16_to_cpu(dip
->di_metatype
)) {
287 case XFS_METAFILE_RTRMAP
:
288 xfs_rtrmapbt_to_disk(mp
, src
, len
, dfork
, dsize
);
290 case XFS_METAFILE_RTREFCOUNT
:
291 xfs_rtrefcountbt_to_disk(mp
, src
, len
, dfork
, dsize
);
295 return -EFSCORRUPTED
;
300 return -EFSCORRUPTED
;
307 xlog_recover_inode_commit_pass2(
309 struct list_head
*buffer_list
,
310 struct xlog_recover_item
*item
,
311 xfs_lsn_t current_lsn
)
313 struct xfs_inode_log_format
*in_f
;
314 struct xfs_mount
*mp
= log
->l_mp
;
316 struct xfs_dinode
*dip
;
323 struct xfs_log_dinode
*ldip
;
328 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
329 in_f
= item
->ri_buf
[0].i_addr
;
331 in_f
= kmalloc(sizeof(struct xfs_inode_log_format
),
332 GFP_KERNEL
| __GFP_NOFAIL
);
334 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
340 * Inode buffers can be freed, look out for it,
341 * and do not replay the inode.
343 if (xlog_is_buffer_cancelled(log
, in_f
->ilf_blkno
, in_f
->ilf_len
)) {
345 trace_xfs_log_recover_inode_cancel(log
, in_f
);
348 trace_xfs_log_recover_inode_recover(log
, in_f
);
350 error
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
,
351 0, &bp
, &xfs_inode_buf_ops
);
354 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
355 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
358 * Make sure the place we're flushing out to really looks
361 if (XFS_IS_CORRUPT(mp
, !xfs_verify_magic16(bp
, dip
->di_magic
))) {
363 "%s: Bad inode magic number, dip = "PTR_FMT
", dino bp = "PTR_FMT
", ino = %lld",
364 __func__
, dip
, bp
, in_f
->ilf_ino
);
365 error
= -EFSCORRUPTED
;
368 ldip
= item
->ri_buf
[1].i_addr
;
369 if (XFS_IS_CORRUPT(mp
, ldip
->di_magic
!= XFS_DINODE_MAGIC
)) {
371 "%s: Bad inode log record, rec ptr "PTR_FMT
", ino %lld",
372 __func__
, item
, in_f
->ilf_ino
);
373 error
= -EFSCORRUPTED
;
378 * If the inode has an LSN in it, recover the inode only if the on-disk
379 * inode's LSN is older than the lsn of the transaction we are
380 * replaying. We can have multiple checkpoints with the same start LSN,
381 * so the current LSN being equal to the on-disk LSN doesn't necessarily
382 * mean that the on-disk inode is more recent than the change being
385 * We must check the current_lsn against the on-disk inode
386 * here because the we can't trust the log dinode to contain a valid LSN
387 * (see comment below before replaying the log dinode for details).
389 * Note: we still need to replay an owner change even though the inode
390 * is more recent than the transaction as there is no guarantee that all
391 * the btree blocks are more recent than this transaction, too.
393 if (dip
->di_version
>= 3) {
394 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
396 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) > 0) {
397 trace_xfs_log_recover_inode_skip(log
, in_f
);
399 goto out_owner_change
;
404 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
405 * are transactional and if ordering is necessary we can determine that
406 * more accurately by the LSN field in the V3 inode core. Don't trust
407 * the inode versions we might be changing them here - use the
408 * superblock flag to determine whether we need to look at di_flushiter
409 * to skip replay when the on disk inode is newer than the log one
411 if (!xfs_has_v3inodes(mp
)) {
412 if (ldip
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
414 * Deal with the wrap case, DI_MAX_FLUSH is less
415 * than smaller numbers
417 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
418 ldip
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
421 trace_xfs_log_recover_inode_skip(log
, in_f
);
427 /* Take the opportunity to reset the flush iteration count */
428 ldip
->di_flushiter
= 0;
432 if (unlikely(S_ISREG(ldip
->di_mode
))) {
433 if (ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
&&
434 ldip
->di_format
!= XFS_DINODE_FMT_BTREE
&&
435 ldip
->di_format
!= XFS_DINODE_FMT_META_BTREE
) {
436 XFS_CORRUPTION_ERROR(
437 "Bad log dinode data fork format for regular file",
438 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
440 "Bad inode 0x%llx, data fork format 0x%x",
441 in_f
->ilf_ino
, ldip
->di_format
);
442 error
= -EFSCORRUPTED
;
445 } else if (unlikely(S_ISDIR(ldip
->di_mode
))) {
446 if ((ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
447 (ldip
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
448 (ldip
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
449 XFS_CORRUPTION_ERROR(
450 "Bad log dinode data fork format for directory",
451 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
453 "Bad inode 0x%llx, data fork format 0x%x",
454 in_f
->ilf_ino
, ldip
->di_format
);
455 error
= -EFSCORRUPTED
;
460 error
= xlog_dinode_verify_extent_counts(mp
, ldip
);
464 if (unlikely(ldip
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
465 XFS_CORRUPTION_ERROR("Bad log dinode fork offset",
466 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
468 "Bad inode 0x%llx, di_forkoff 0x%x",
469 in_f
->ilf_ino
, ldip
->di_forkoff
);
470 error
= -EFSCORRUPTED
;
473 isize
= xfs_log_dinode_size(mp
);
474 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
475 XFS_CORRUPTION_ERROR("Bad log dinode size", XFS_ERRLEVEL_LOW
,
476 mp
, ldip
, sizeof(*ldip
));
478 "Bad inode 0x%llx log dinode size 0x%x",
479 in_f
->ilf_ino
, item
->ri_buf
[1].i_len
);
480 error
= -EFSCORRUPTED
;
485 * Recover the log dinode inode into the on disk inode.
487 * The LSN in the log dinode is garbage - it can be zero or reflect
488 * stale in-memory runtime state that isn't coherent with the changes
489 * logged in this transaction or the changes written to the on-disk
490 * inode. Hence we write the current lSN into the inode because that
491 * matches what xfs_iflush() would write inode the inode when flushing
492 * the changes in this transaction.
494 xfs_log_dinode_to_disk(ldip
, dip
, current_lsn
);
496 fields
= in_f
->ilf_fields
;
497 if (fields
& XFS_ILOG_DEV
)
498 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
500 if (in_f
->ilf_size
== 2)
501 goto out_owner_change
;
502 len
= item
->ri_buf
[2].i_len
;
503 src
= item
->ri_buf
[2].i_addr
;
504 ASSERT(in_f
->ilf_size
<= 4);
505 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
506 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
507 (len
== xlog_calc_iovec_len(in_f
->ilf_dsize
)));
509 switch (fields
& XFS_ILOG_DFORK
) {
512 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
515 case XFS_ILOG_DBROOT
:
516 error
= xlog_recover_inode_dbroot(mp
, src
, len
, dip
);
523 * There are no data fork flags set.
525 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
530 * If we logged any attribute data, recover it. There may or
531 * may not have been any other non-core data logged in this
534 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
535 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
540 len
= item
->ri_buf
[attr_index
].i_len
;
541 src
= item
->ri_buf
[attr_index
].i_addr
;
542 ASSERT(len
== xlog_calc_iovec_len(in_f
->ilf_asize
));
544 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
547 dest
= XFS_DFORK_APTR(dip
);
548 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
549 memcpy(dest
, src
, len
);
552 case XFS_ILOG_ABROOT
:
553 dest
= XFS_DFORK_APTR(dip
);
554 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
555 len
, (struct xfs_bmdr_block
*)dest
,
556 XFS_DFORK_ASIZE(dip
, mp
));
560 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
562 error
= -EFSCORRUPTED
;
568 /* Recover the swapext owner change unless inode has been deleted */
569 if ((in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
)) &&
571 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
573 /* re-generate the checksum and validate the recovered inode. */
574 xfs_dinode_calc_crc(log
->l_mp
, dip
);
575 fa
= xfs_dinode_verify(log
->l_mp
, in_f
->ilf_ino
, dip
);
577 XFS_CORRUPTION_ERROR(
578 "Bad dinode after recovery",
579 XFS_ERRLEVEL_LOW
, mp
, dip
, sizeof(*dip
));
581 "Metadata corruption detected at %pS, inode 0x%llx",
583 error
= -EFSCORRUPTED
;
587 ASSERT(bp
->b_mount
== mp
);
588 bp
->b_flags
|= _XBF_LOGRECOVERY
;
589 xfs_buf_delwri_queue(bp
, buffer_list
);
599 const struct xlog_recover_item_ops xlog_inode_item_ops
= {
600 .item_type
= XFS_LI_INODE
,
601 .ra_pass2
= xlog_recover_inode_ra_pass2
,
602 .commit_pass2
= xlog_recover_inode_commit_pass2
,