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"
27 xlog_recover_inode_ra_pass2(
29 struct xlog_recover_item
*item
)
31 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
32 struct xfs_inode_log_format
*ilfp
= item
->ri_buf
[0].i_addr
;
34 xlog_buf_readahead(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
,
35 &xfs_inode_buf_ra_ops
);
37 struct xfs_inode_log_format_32
*ilfp
= item
->ri_buf
[0].i_addr
;
39 xlog_buf_readahead(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
,
40 &xfs_inode_buf_ra_ops
);
45 * Inode fork owner changes
47 * If we have been told that we have to reparent the inode fork, it's because an
48 * extent swap operation on a CRC enabled filesystem has been done and we are
49 * replaying it. We need to walk the BMBT of the appropriate fork and change the
52 * The complexity here is that we don't have an inode context to work with, so
53 * after we've replayed the inode we need to instantiate one. This is where the
56 * We are in the middle of log recovery, so we can't run transactions. That
57 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
58 * that will result in the corresponding iput() running the inode through
59 * xfs_inactive(). If we've just replayed an inode core that changes the link
60 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
61 * transactions (bad!).
63 * So, to avoid this, we instantiate an inode directly from the inode core we've
64 * just recovered. We have the buffer still locked, and all we really need to
65 * instantiate is the inode core and the forks being modified. We can do this
66 * manually, then run the inode btree owner change, and then tear down the
67 * xfs_inode without having to run any transactions at all.
69 * Also, because we don't have a transaction context available here but need to
70 * gather all the buffers we modify for writeback so we pass the buffer_list
71 * instead for the operation to use.
75 xfs_recover_inode_owner_change(
77 struct xfs_dinode
*dip
,
78 struct xfs_inode_log_format
*in_f
,
79 struct list_head
*buffer_list
)
84 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
86 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
90 /* instantiate the inode */
91 ASSERT(dip
->di_version
>= 3);
93 error
= xfs_inode_from_disk(ip
, dip
);
97 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
98 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
99 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
100 ip
->i_ino
, buffer_list
);
105 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
106 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
107 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
108 ip
->i_ino
, buffer_list
);
118 static inline bool xfs_log_dinode_has_bigtime(const struct xfs_log_dinode
*ld
)
120 return ld
->di_version
>= 3 &&
121 (ld
->di_flags2
& XFS_DIFLAG2_BIGTIME
);
124 /* Convert a log timestamp to an ondisk timestamp. */
125 static inline xfs_timestamp_t
126 xfs_log_dinode_to_disk_ts(
127 struct xfs_log_dinode
*from
,
128 const xfs_log_timestamp_t its
)
130 struct xfs_legacy_timestamp
*lts
;
131 struct xfs_log_legacy_timestamp
*lits
;
134 if (xfs_log_dinode_has_bigtime(from
))
135 return cpu_to_be64(its
);
137 lts
= (struct xfs_legacy_timestamp
*)&ts
;
138 lits
= (struct xfs_log_legacy_timestamp
*)&its
;
139 lts
->t_sec
= cpu_to_be32(lits
->t_sec
);
140 lts
->t_nsec
= cpu_to_be32(lits
->t_nsec
);
145 static inline bool xfs_log_dinode_has_large_extent_counts(
146 const struct xfs_log_dinode
*ld
)
148 return ld
->di_version
>= 3 &&
149 (ld
->di_flags2
& XFS_DIFLAG2_NREXT64
);
153 xfs_log_dinode_to_disk_iext_counters(
154 struct xfs_log_dinode
*from
,
155 struct xfs_dinode
*to
)
157 if (xfs_log_dinode_has_large_extent_counts(from
)) {
158 to
->di_big_nextents
= cpu_to_be64(from
->di_big_nextents
);
159 to
->di_big_anextents
= cpu_to_be32(from
->di_big_anextents
);
160 to
->di_nrext64_pad
= cpu_to_be16(from
->di_nrext64_pad
);
162 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
163 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
169 xfs_log_dinode_to_disk(
170 struct xfs_log_dinode
*from
,
171 struct xfs_dinode
*to
,
174 to
->di_magic
= cpu_to_be16(from
->di_magic
);
175 to
->di_mode
= cpu_to_be16(from
->di_mode
);
176 to
->di_version
= from
->di_version
;
177 to
->di_format
= from
->di_format
;
179 to
->di_uid
= cpu_to_be32(from
->di_uid
);
180 to
->di_gid
= cpu_to_be32(from
->di_gid
);
181 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
182 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
183 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
185 to
->di_atime
= xfs_log_dinode_to_disk_ts(from
, from
->di_atime
);
186 to
->di_mtime
= xfs_log_dinode_to_disk_ts(from
, from
->di_mtime
);
187 to
->di_ctime
= xfs_log_dinode_to_disk_ts(from
, from
->di_ctime
);
189 to
->di_size
= cpu_to_be64(from
->di_size
);
190 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
191 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
192 to
->di_forkoff
= from
->di_forkoff
;
193 to
->di_aformat
= from
->di_aformat
;
194 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
195 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
196 to
->di_flags
= cpu_to_be16(from
->di_flags
);
197 to
->di_gen
= cpu_to_be32(from
->di_gen
);
199 if (from
->di_version
== 3) {
200 to
->di_changecount
= cpu_to_be64(from
->di_changecount
);
201 to
->di_crtime
= xfs_log_dinode_to_disk_ts(from
,
203 to
->di_flags2
= cpu_to_be64(from
->di_flags2
);
204 to
->di_cowextsize
= cpu_to_be32(from
->di_cowextsize
);
205 to
->di_ino
= cpu_to_be64(from
->di_ino
);
206 to
->di_lsn
= cpu_to_be64(lsn
);
207 memset(to
->di_pad2
, 0, sizeof(to
->di_pad2
));
208 uuid_copy(&to
->di_uuid
, &from
->di_uuid
);
211 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
212 memset(to
->di_v2_pad
, 0, sizeof(to
->di_v2_pad
));
215 xfs_log_dinode_to_disk_iext_counters(from
, to
);
219 xlog_dinode_verify_extent_counts(
220 struct xfs_mount
*mp
,
221 struct xfs_log_dinode
*ldip
)
223 xfs_extnum_t nextents
;
224 xfs_aextnum_t anextents
;
226 if (xfs_log_dinode_has_large_extent_counts(ldip
)) {
227 if (!xfs_has_large_extent_counts(mp
) ||
228 (ldip
->di_nrext64_pad
!= 0)) {
229 XFS_CORRUPTION_ERROR(
230 "Bad log dinode large extent count format",
231 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
233 "Bad inode 0x%llx, large extent counts %d, padding 0x%x",
234 ldip
->di_ino
, xfs_has_large_extent_counts(mp
),
235 ldip
->di_nrext64_pad
);
236 return -EFSCORRUPTED
;
239 nextents
= ldip
->di_big_nextents
;
240 anextents
= ldip
->di_big_anextents
;
242 if (ldip
->di_version
== 3 && ldip
->di_v3_pad
!= 0) {
243 XFS_CORRUPTION_ERROR(
244 "Bad log dinode di_v3_pad",
245 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
247 "Bad inode 0x%llx, di_v3_pad 0x%llx",
248 ldip
->di_ino
, ldip
->di_v3_pad
);
249 return -EFSCORRUPTED
;
252 nextents
= ldip
->di_nextents
;
253 anextents
= ldip
->di_anextents
;
256 if (unlikely(nextents
+ anextents
> ldip
->di_nblocks
)) {
257 XFS_CORRUPTION_ERROR("Bad log dinode extent counts",
258 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
260 "Bad inode 0x%llx, large extent counts %d, nextents 0x%llx, anextents 0x%x, nblocks 0x%llx",
261 ldip
->di_ino
, xfs_has_large_extent_counts(mp
), nextents
,
262 anextents
, ldip
->di_nblocks
);
263 return -EFSCORRUPTED
;
270 xlog_recover_inode_commit_pass2(
272 struct list_head
*buffer_list
,
273 struct xlog_recover_item
*item
,
274 xfs_lsn_t current_lsn
)
276 struct xfs_inode_log_format
*in_f
;
277 struct xfs_mount
*mp
= log
->l_mp
;
279 struct xfs_dinode
*dip
;
286 struct xfs_log_dinode
*ldip
;
291 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
292 in_f
= item
->ri_buf
[0].i_addr
;
294 in_f
= kmalloc(sizeof(struct xfs_inode_log_format
),
295 GFP_KERNEL
| __GFP_NOFAIL
);
297 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
303 * Inode buffers can be freed, look out for it,
304 * and do not replay the inode.
306 if (xlog_is_buffer_cancelled(log
, in_f
->ilf_blkno
, in_f
->ilf_len
)) {
308 trace_xfs_log_recover_inode_cancel(log
, in_f
);
311 trace_xfs_log_recover_inode_recover(log
, in_f
);
313 error
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
,
314 0, &bp
, &xfs_inode_buf_ops
);
317 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
318 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
321 * Make sure the place we're flushing out to really looks
324 if (XFS_IS_CORRUPT(mp
, !xfs_verify_magic16(bp
, dip
->di_magic
))) {
326 "%s: Bad inode magic number, dip = "PTR_FMT
", dino bp = "PTR_FMT
", ino = %lld",
327 __func__
, dip
, bp
, in_f
->ilf_ino
);
328 error
= -EFSCORRUPTED
;
331 ldip
= item
->ri_buf
[1].i_addr
;
332 if (XFS_IS_CORRUPT(mp
, ldip
->di_magic
!= XFS_DINODE_MAGIC
)) {
334 "%s: Bad inode log record, rec ptr "PTR_FMT
", ino %lld",
335 __func__
, item
, in_f
->ilf_ino
);
336 error
= -EFSCORRUPTED
;
341 * If the inode has an LSN in it, recover the inode only if the on-disk
342 * inode's LSN is older than the lsn of the transaction we are
343 * replaying. We can have multiple checkpoints with the same start LSN,
344 * so the current LSN being equal to the on-disk LSN doesn't necessarily
345 * mean that the on-disk inode is more recent than the change being
348 * We must check the current_lsn against the on-disk inode
349 * here because the we can't trust the log dinode to contain a valid LSN
350 * (see comment below before replaying the log dinode for details).
352 * Note: we still need to replay an owner change even though the inode
353 * is more recent than the transaction as there is no guarantee that all
354 * the btree blocks are more recent than this transaction, too.
356 if (dip
->di_version
>= 3) {
357 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
359 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) > 0) {
360 trace_xfs_log_recover_inode_skip(log
, in_f
);
362 goto out_owner_change
;
367 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
368 * are transactional and if ordering is necessary we can determine that
369 * more accurately by the LSN field in the V3 inode core. Don't trust
370 * the inode versions we might be changing them here - use the
371 * superblock flag to determine whether we need to look at di_flushiter
372 * to skip replay when the on disk inode is newer than the log one
374 if (!xfs_has_v3inodes(mp
)) {
375 if (ldip
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
377 * Deal with the wrap case, DI_MAX_FLUSH is less
378 * than smaller numbers
380 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
381 ldip
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
384 trace_xfs_log_recover_inode_skip(log
, in_f
);
390 /* Take the opportunity to reset the flush iteration count */
391 ldip
->di_flushiter
= 0;
395 if (unlikely(S_ISREG(ldip
->di_mode
))) {
396 if ((ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
397 (ldip
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
398 XFS_CORRUPTION_ERROR(
399 "Bad log dinode data fork format for regular file",
400 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
402 "Bad inode 0x%llx, data fork format 0x%x",
403 in_f
->ilf_ino
, ldip
->di_format
);
404 error
= -EFSCORRUPTED
;
407 } else if (unlikely(S_ISDIR(ldip
->di_mode
))) {
408 if ((ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
409 (ldip
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
410 (ldip
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
411 XFS_CORRUPTION_ERROR(
412 "Bad log dinode data fork format for directory",
413 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
415 "Bad inode 0x%llx, data fork format 0x%x",
416 in_f
->ilf_ino
, ldip
->di_format
);
417 error
= -EFSCORRUPTED
;
422 error
= xlog_dinode_verify_extent_counts(mp
, ldip
);
426 if (unlikely(ldip
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
427 XFS_CORRUPTION_ERROR("Bad log dinode fork offset",
428 XFS_ERRLEVEL_LOW
, mp
, ldip
, sizeof(*ldip
));
430 "Bad inode 0x%llx, di_forkoff 0x%x",
431 in_f
->ilf_ino
, ldip
->di_forkoff
);
432 error
= -EFSCORRUPTED
;
435 isize
= xfs_log_dinode_size(mp
);
436 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
437 XFS_CORRUPTION_ERROR("Bad log dinode size", XFS_ERRLEVEL_LOW
,
438 mp
, ldip
, sizeof(*ldip
));
440 "Bad inode 0x%llx log dinode size 0x%x",
441 in_f
->ilf_ino
, item
->ri_buf
[1].i_len
);
442 error
= -EFSCORRUPTED
;
447 * Recover the log dinode inode into the on disk inode.
449 * The LSN in the log dinode is garbage - it can be zero or reflect
450 * stale in-memory runtime state that isn't coherent with the changes
451 * logged in this transaction or the changes written to the on-disk
452 * inode. Hence we write the current lSN into the inode because that
453 * matches what xfs_iflush() would write inode the inode when flushing
454 * the changes in this transaction.
456 xfs_log_dinode_to_disk(ldip
, dip
, current_lsn
);
458 fields
= in_f
->ilf_fields
;
459 if (fields
& XFS_ILOG_DEV
)
460 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
462 if (in_f
->ilf_size
== 2)
463 goto out_owner_change
;
464 len
= item
->ri_buf
[2].i_len
;
465 src
= item
->ri_buf
[2].i_addr
;
466 ASSERT(in_f
->ilf_size
<= 4);
467 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
468 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
469 (len
== xlog_calc_iovec_len(in_f
->ilf_dsize
)));
471 switch (fields
& XFS_ILOG_DFORK
) {
474 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
477 case XFS_ILOG_DBROOT
:
478 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
479 (struct xfs_bmdr_block
*)XFS_DFORK_DPTR(dip
),
480 XFS_DFORK_DSIZE(dip
, mp
));
485 * There are no data fork flags set.
487 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
492 * If we logged any attribute data, recover it. There may or
493 * may not have been any other non-core data logged in this
496 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
497 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
502 len
= item
->ri_buf
[attr_index
].i_len
;
503 src
= item
->ri_buf
[attr_index
].i_addr
;
504 ASSERT(len
== xlog_calc_iovec_len(in_f
->ilf_asize
));
506 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
509 dest
= XFS_DFORK_APTR(dip
);
510 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
511 memcpy(dest
, src
, len
);
514 case XFS_ILOG_ABROOT
:
515 dest
= XFS_DFORK_APTR(dip
);
516 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
517 len
, (struct xfs_bmdr_block
*)dest
,
518 XFS_DFORK_ASIZE(dip
, mp
));
522 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
524 error
= -EFSCORRUPTED
;
530 /* Recover the swapext owner change unless inode has been deleted */
531 if ((in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
)) &&
533 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
535 /* re-generate the checksum and validate the recovered inode. */
536 xfs_dinode_calc_crc(log
->l_mp
, dip
);
537 fa
= xfs_dinode_verify(log
->l_mp
, in_f
->ilf_ino
, dip
);
539 XFS_CORRUPTION_ERROR(
540 "Bad dinode after recovery",
541 XFS_ERRLEVEL_LOW
, mp
, dip
, sizeof(*dip
));
543 "Metadata corruption detected at %pS, inode 0x%llx",
545 error
= -EFSCORRUPTED
;
549 ASSERT(bp
->b_mount
== mp
);
550 bp
->b_flags
|= _XBF_LOGRECOVERY
;
551 xfs_buf_delwri_queue(bp
, buffer_list
);
561 const struct xlog_recover_item_ops xlog_inode_item_ops
= {
562 .item_type
= XFS_LI_INODE
,
563 .ra_pass2
= xlog_recover_inode_ra_pass2
,
564 .commit_pass2
= xlog_recover_inode_commit_pass2
,