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);
146 xfs_log_dinode_to_disk(
147 struct xfs_log_dinode *from,
148 struct xfs_dinode *to,
151 to->di_magic = cpu_to_be16(from->di_magic);
152 to->di_mode = cpu_to_be16(from->di_mode);
153 to->di_version = from->di_version;
154 to->di_format = from->di_format;
156 to->di_uid = cpu_to_be32(from->di_uid);
157 to->di_gid = cpu_to_be32(from->di_gid);
158 to->di_nlink = cpu_to_be32(from->di_nlink);
159 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
160 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
161 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
163 to->di_atime = xfs_log_dinode_to_disk_ts(from, from->di_atime);
164 to->di_mtime = xfs_log_dinode_to_disk_ts(from, from->di_mtime);
165 to->di_ctime = xfs_log_dinode_to_disk_ts(from, from->di_ctime);
167 to->di_size = cpu_to_be64(from->di_size);
168 to->di_nblocks = cpu_to_be64(from->di_nblocks);
169 to->di_extsize = cpu_to_be32(from->di_extsize);
170 to->di_nextents = cpu_to_be32(from->di_nextents);
171 to->di_anextents = cpu_to_be16(from->di_anextents);
172 to->di_forkoff = from->di_forkoff;
173 to->di_aformat = from->di_aformat;
174 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
175 to->di_dmstate = cpu_to_be16(from->di_dmstate);
176 to->di_flags = cpu_to_be16(from->di_flags);
177 to->di_gen = cpu_to_be32(from->di_gen);
179 if (from->di_version == 3) {
180 to->di_changecount = cpu_to_be64(from->di_changecount);
181 to->di_crtime = xfs_log_dinode_to_disk_ts(from,
183 to->di_flags2 = cpu_to_be64(from->di_flags2);
184 to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
185 to->di_ino = cpu_to_be64(from->di_ino);
186 to->di_lsn = cpu_to_be64(lsn);
187 memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
188 uuid_copy(&to->di_uuid, &from->di_uuid);
189 to->di_flushiter = 0;
191 to->di_flushiter = cpu_to_be16(from->di_flushiter);
196 xlog_recover_inode_commit_pass2(
198 struct list_head *buffer_list,
199 struct xlog_recover_item *item,
200 xfs_lsn_t current_lsn)
202 struct xfs_inode_log_format *in_f;
203 struct xfs_mount *mp = log->l_mp;
205 struct xfs_dinode *dip;
212 struct xfs_log_dinode *ldip;
216 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
217 in_f = item->ri_buf[0].i_addr;
219 in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), 0);
221 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
227 * Inode buffers can be freed, look out for it,
228 * and do not replay the inode.
230 if (xlog_is_buffer_cancelled(log, in_f->ilf_blkno, in_f->ilf_len)) {
232 trace_xfs_log_recover_inode_cancel(log, in_f);
235 trace_xfs_log_recover_inode_recover(log, in_f);
237 error = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
238 0, &bp, &xfs_inode_buf_ops);
241 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
242 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
245 * Make sure the place we're flushing out to really looks
248 if (XFS_IS_CORRUPT(mp, !xfs_verify_magic16(bp, dip->di_magic))) {
250 "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld",
251 __func__, dip, bp, in_f->ilf_ino);
252 error = -EFSCORRUPTED;
255 ldip = item->ri_buf[1].i_addr;
256 if (XFS_IS_CORRUPT(mp, ldip->di_magic != XFS_DINODE_MAGIC)) {
258 "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld",
259 __func__, item, in_f->ilf_ino);
260 error = -EFSCORRUPTED;
265 * If the inode has an LSN in it, recover the inode only if the on-disk
266 * inode's LSN is older than the lsn of the transaction we are
267 * replaying. We can have multiple checkpoints with the same start LSN,
268 * so the current LSN being equal to the on-disk LSN doesn't necessarily
269 * mean that the on-disk inode is more recent than the change being
272 * We must check the current_lsn against the on-disk inode
273 * here because the we can't trust the log dinode to contain a valid LSN
274 * (see comment below before replaying the log dinode for details).
276 * Note: we still need to replay an owner change even though the inode
277 * is more recent than the transaction as there is no guarantee that all
278 * the btree blocks are more recent than this transaction, too.
280 if (dip->di_version >= 3) {
281 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
283 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) > 0) {
284 trace_xfs_log_recover_inode_skip(log, in_f);
286 goto out_owner_change;
291 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
292 * are transactional and if ordering is necessary we can determine that
293 * more accurately by the LSN field in the V3 inode core. Don't trust
294 * the inode versions we might be changing them here - use the
295 * superblock flag to determine whether we need to look at di_flushiter
296 * to skip replay when the on disk inode is newer than the log one
298 if (!xfs_sb_version_has_v3inode(&mp->m_sb) &&
299 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
301 * Deal with the wrap case, DI_MAX_FLUSH is less
302 * than smaller numbers
304 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
305 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
308 trace_xfs_log_recover_inode_skip(log, in_f);
314 /* Take the opportunity to reset the flush iteration count */
315 ldip->di_flushiter = 0;
317 if (unlikely(S_ISREG(ldip->di_mode))) {
318 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
319 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
320 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
321 XFS_ERRLEVEL_LOW, mp, ldip,
324 "%s: Bad regular inode log record, rec ptr "PTR_FMT", "
325 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
326 __func__, item, dip, bp, in_f->ilf_ino);
327 error = -EFSCORRUPTED;
330 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
331 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
332 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
333 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
334 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
335 XFS_ERRLEVEL_LOW, mp, ldip,
338 "%s: Bad dir inode log record, rec ptr "PTR_FMT", "
339 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
340 __func__, item, dip, bp, in_f->ilf_ino);
341 error = -EFSCORRUPTED;
345 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
346 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
347 XFS_ERRLEVEL_LOW, mp, ldip,
350 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
351 "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld",
352 __func__, item, dip, bp, in_f->ilf_ino,
353 ldip->di_nextents + ldip->di_anextents,
355 error = -EFSCORRUPTED;
358 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
359 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
360 XFS_ERRLEVEL_LOW, mp, ldip,
363 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
364 "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__,
365 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
366 error = -EFSCORRUPTED;
369 isize = xfs_log_dinode_size(mp);
370 if (unlikely(item->ri_buf[1].i_len > isize)) {
371 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
372 XFS_ERRLEVEL_LOW, mp, ldip,
375 "%s: Bad inode log record length %d, rec ptr "PTR_FMT,
376 __func__, item->ri_buf[1].i_len, item);
377 error = -EFSCORRUPTED;
382 * Recover the log dinode inode into the on disk inode.
384 * The LSN in the log dinode is garbage - it can be zero or reflect
385 * stale in-memory runtime state that isn't coherent with the changes
386 * logged in this transaction or the changes written to the on-disk
387 * inode. Hence we write the current lSN into the inode because that
388 * matches what xfs_iflush() would write inode the inode when flushing
389 * the changes in this transaction.
391 xfs_log_dinode_to_disk(ldip, dip, current_lsn);
393 fields = in_f->ilf_fields;
394 if (fields & XFS_ILOG_DEV)
395 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
397 if (in_f->ilf_size == 2)
398 goto out_owner_change;
399 len = item->ri_buf[2].i_len;
400 src = item->ri_buf[2].i_addr;
401 ASSERT(in_f->ilf_size <= 4);
402 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
403 ASSERT(!(fields & XFS_ILOG_DFORK) ||
404 (len == in_f->ilf_dsize));
406 switch (fields & XFS_ILOG_DFORK) {
409 memcpy(XFS_DFORK_DPTR(dip), src, len);
412 case XFS_ILOG_DBROOT:
413 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
414 (struct xfs_bmdr_block *)XFS_DFORK_DPTR(dip),
415 XFS_DFORK_DSIZE(dip, mp));
420 * There are no data fork flags set.
422 ASSERT((fields & XFS_ILOG_DFORK) == 0);
427 * If we logged any attribute data, recover it. There may or
428 * may not have been any other non-core data logged in this
431 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
432 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
437 len = item->ri_buf[attr_index].i_len;
438 src = item->ri_buf[attr_index].i_addr;
439 ASSERT(len == in_f->ilf_asize);
441 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
444 dest = XFS_DFORK_APTR(dip);
445 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
446 memcpy(dest, src, len);
449 case XFS_ILOG_ABROOT:
450 dest = XFS_DFORK_APTR(dip);
451 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
452 len, (struct xfs_bmdr_block *)dest,
453 XFS_DFORK_ASIZE(dip, mp));
457 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
459 error = -EFSCORRUPTED;
465 /* Recover the swapext owner change unless inode has been deleted */
466 if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) &&
468 error = xfs_recover_inode_owner_change(mp, dip, in_f,
470 /* re-generate the checksum. */
471 xfs_dinode_calc_crc(log->l_mp, dip);
473 ASSERT(bp->b_mount == mp);
474 bp->b_flags |= _XBF_LOGRECOVERY;
475 xfs_buf_delwri_queue(bp, buffer_list);
485 const struct xlog_recover_item_ops xlog_inode_item_ops = {
486 .item_type = XFS_LI_INODE,
487 .ra_pass2 = xlog_recover_inode_ra_pass2,
488 .commit_pass2 = xlog_recover_inode_commit_pass2,