1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 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"
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_log_priv.h"
21 #include "xfs_error.h"
22 #include "xfs_rtbitmap.h"
24 #include <linux/iversion.h>
26 struct kmem_cache *xfs_ili_cache; /* inode log item */
28 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
30 return container_of(lip, struct xfs_inode_log_item, ili_item);
35 struct xfs_log_item *lip)
37 return INODE_ITEM(lip)->ili_inode->i_ino;
41 * Prior to finally logging the inode, we have to ensure that all the
42 * per-modification inode state changes are applied. This includes VFS inode
43 * state updates, format conversions, verifier state synchronisation and
44 * ensuring the inode buffer remains in memory whilst the inode is dirty.
46 * We have to be careful when we grab the inode cluster buffer due to lock
47 * ordering constraints. The unlinked inode modifications (xfs_iunlink_item)
48 * require AGI -> inode cluster buffer lock order. The inode cluster buffer is
49 * not locked until ->precommit, so it happens after everything else has been
52 * Further, we have AGI -> AGF lock ordering, and with O_TMPFILE handling we
53 * have AGI -> AGF -> iunlink item -> inode cluster buffer lock order. Hence we
54 * cannot safely lock the inode cluster buffer in xfs_trans_log_inode() because
55 * it can be called on a inode (e.g. via bumplink/droplink) before we take the
56 * AGF lock modifying directory blocks.
58 * Rather than force a complete rework of all the transactions to call
59 * xfs_trans_log_inode() once and once only at the end of every transaction, we
60 * move the pinning of the inode cluster buffer to a ->precommit operation. This
61 * matches how the xfs_iunlink_item locks the inode cluster buffer, and it
62 * ensures that the inode cluster buffer locking is always done last in a
63 * transaction. i.e. we ensure the lock order is always AGI -> AGF -> inode
66 * If we return the inode number as the precommit sort key then we'll also
67 * guarantee that the order all inode cluster buffer locking is the same all the
68 * inodes and unlink items in the transaction.
71 xfs_inode_item_precommit(
73 struct xfs_log_item *lip)
75 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
76 struct xfs_inode *ip = iip->ili_inode;
77 struct inode *inode = VFS_I(ip);
78 unsigned int flags = iip->ili_dirty_flags;
81 * Don't bother with i_lock for the I_DIRTY_TIME check here, as races
82 * don't matter - we either will need an extra transaction in 24 hours
83 * to log the timestamps, or will clear already cleared fields in the
86 if (inode->i_state & I_DIRTY_TIME) {
87 spin_lock(&inode->i_lock);
88 inode->i_state &= ~I_DIRTY_TIME;
89 spin_unlock(&inode->i_lock);
93 * If we're updating the inode core or the timestamps and it's possible
94 * to upgrade this inode to bigtime format, do so now.
96 if ((flags & (XFS_ILOG_CORE | XFS_ILOG_TIMESTAMP)) &&
97 xfs_has_bigtime(ip->i_mount) &&
98 !xfs_inode_has_bigtime(ip)) {
99 ip->i_diflags2 |= XFS_DIFLAG2_BIGTIME;
100 flags |= XFS_ILOG_CORE;
104 * Inode verifiers do not check that the extent size hint is an integer
105 * multiple of the rt extent size on a directory with both rtinherit
106 * and extszinherit flags set. If we're logging a directory that is
107 * misconfigured in this way, clear the hint.
109 if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
110 (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&
111 xfs_extlen_to_rtxmod(ip->i_mount, ip->i_extsize) > 0) {
112 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
113 XFS_DIFLAG_EXTSZINHERIT);
115 flags |= XFS_ILOG_CORE;
119 * Record the specific change for fdatasync optimisation. This allows
120 * fdatasync to skip log forces for inodes that are only timestamp
121 * dirty. Once we've processed the XFS_ILOG_IVERSION flag, convert it
122 * to XFS_ILOG_CORE so that the actual on-disk dirty tracking
123 * (ili_fields) correctly tracks that the version has changed.
125 spin_lock(&iip->ili_lock);
126 iip->ili_fsync_fields |= (flags & ~XFS_ILOG_IVERSION);
127 if (flags & XFS_ILOG_IVERSION)
128 flags = ((flags & ~XFS_ILOG_IVERSION) | XFS_ILOG_CORE);
130 if (!iip->ili_item.li_buf) {
135 * We hold the ILOCK here, so this inode is not going to be
136 * flushed while we are here. Further, because there is no
137 * buffer attached to the item, we know that there is no IO in
138 * progress, so nothing will clear the ili_fields while we read
139 * in the buffer. Hence we can safely drop the spin lock and
140 * read the buffer knowing that the state will not change from
143 spin_unlock(&iip->ili_lock);
144 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &bp);
149 * We need an explicit buffer reference for the log item but
150 * don't want the buffer to remain attached to the transaction.
151 * Hold the buffer but release the transaction reference once
152 * we've attached the inode log item to the buffer log item
156 spin_lock(&iip->ili_lock);
157 iip->ili_item.li_buf = bp;
158 bp->b_flags |= _XBF_INODES;
159 list_add_tail(&iip->ili_item.li_bio_list, &bp->b_li_list);
160 xfs_trans_brelse(tp, bp);
164 * Always OR in the bits from the ili_last_fields field. This is to
165 * coordinate with the xfs_iflush() and xfs_buf_inode_iodone() routines
166 * in the eventual clearing of the ili_fields bits. See the big comment
167 * in xfs_iflush() for an explanation of this coordination mechanism.
169 iip->ili_fields |= (flags | iip->ili_last_fields);
170 spin_unlock(&iip->ili_lock);
173 * We are done with the log item transaction dirty state, so clear it so
174 * that it doesn't pollute future transactions.
176 iip->ili_dirty_flags = 0;
181 * The logged size of an inode fork is always the current size of the inode
182 * fork. This means that when an inode fork is relogged, the size of the logged
183 * region is determined by the current state, not the combination of the
184 * previously logged state + the current state. This is different relogging
185 * behaviour to most other log items which will retain the size of the
186 * previously logged changes when smaller regions are relogged.
188 * Hence operations that remove data from the inode fork (e.g. shortform
189 * dir/attr remove, extent form extent removal, etc), the size of the relogged
190 * inode gets -smaller- rather than stays the same size as the previously logged
191 * size and this can result in the committing transaction reducing the amount of
192 * space being consumed by the CIL.
195 xfs_inode_item_data_fork_size(
196 struct xfs_inode_log_item *iip,
200 struct xfs_inode *ip = iip->ili_inode;
202 switch (ip->i_df.if_format) {
203 case XFS_DINODE_FMT_EXTENTS:
204 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
205 ip->i_df.if_nextents > 0 &&
206 ip->i_df.if_bytes > 0) {
207 /* worst case, doesn't subtract delalloc extents */
208 *nbytes += xfs_inode_data_fork_size(ip);
212 case XFS_DINODE_FMT_BTREE:
213 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
214 ip->i_df.if_broot_bytes > 0) {
215 *nbytes += ip->i_df.if_broot_bytes;
219 case XFS_DINODE_FMT_LOCAL:
220 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
221 ip->i_df.if_bytes > 0) {
222 *nbytes += xlog_calc_iovec_len(ip->i_df.if_bytes);
227 case XFS_DINODE_FMT_DEV:
236 xfs_inode_item_attr_fork_size(
237 struct xfs_inode_log_item *iip,
241 struct xfs_inode *ip = iip->ili_inode;
243 switch (ip->i_af.if_format) {
244 case XFS_DINODE_FMT_EXTENTS:
245 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
246 ip->i_af.if_nextents > 0 &&
247 ip->i_af.if_bytes > 0) {
248 /* worst case, doesn't subtract unused space */
249 *nbytes += xfs_inode_attr_fork_size(ip);
253 case XFS_DINODE_FMT_BTREE:
254 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
255 ip->i_af.if_broot_bytes > 0) {
256 *nbytes += ip->i_af.if_broot_bytes;
260 case XFS_DINODE_FMT_LOCAL:
261 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
262 ip->i_af.if_bytes > 0) {
263 *nbytes += xlog_calc_iovec_len(ip->i_af.if_bytes);
274 * This returns the number of iovecs needed to log the given inode item.
276 * We need one iovec for the inode log format structure, one for the
277 * inode core, and possibly one for the inode data/extents/b-tree root
278 * and one for the inode attribute data/extents/b-tree root.
282 struct xfs_log_item *lip,
286 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
287 struct xfs_inode *ip = iip->ili_inode;
290 *nbytes += sizeof(struct xfs_inode_log_format) +
291 xfs_log_dinode_size(ip->i_mount);
293 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
294 if (xfs_inode_has_attr_fork(ip))
295 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
299 xfs_inode_item_format_data_fork(
300 struct xfs_inode_log_item *iip,
301 struct xfs_inode_log_format *ilf,
302 struct xfs_log_vec *lv,
303 struct xfs_log_iovec **vecp)
305 struct xfs_inode *ip = iip->ili_inode;
308 switch (ip->i_df.if_format) {
309 case XFS_DINODE_FMT_EXTENTS:
311 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
313 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
314 ip->i_df.if_nextents > 0 &&
315 ip->i_df.if_bytes > 0) {
316 struct xfs_bmbt_rec *p;
318 ASSERT(xfs_iext_count(&ip->i_df) > 0);
320 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
321 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
322 xlog_finish_iovec(lv, *vecp, data_bytes);
324 ASSERT(data_bytes <= ip->i_df.if_bytes);
326 ilf->ilf_dsize = data_bytes;
329 iip->ili_fields &= ~XFS_ILOG_DEXT;
332 case XFS_DINODE_FMT_BTREE:
334 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
336 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
337 ip->i_df.if_broot_bytes > 0) {
338 ASSERT(ip->i_df.if_broot != NULL);
339 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
341 ip->i_df.if_broot_bytes);
342 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
345 ASSERT(!(iip->ili_fields &
347 iip->ili_fields &= ~XFS_ILOG_DBROOT;
350 case XFS_DINODE_FMT_LOCAL:
352 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
353 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
354 ip->i_df.if_bytes > 0) {
355 ASSERT(ip->i_df.if_data != NULL);
356 ASSERT(ip->i_disk_size > 0);
357 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
358 ip->i_df.if_data, ip->i_df.if_bytes);
359 ilf->ilf_dsize = (unsigned)ip->i_df.if_bytes;
362 iip->ili_fields &= ~XFS_ILOG_DDATA;
365 case XFS_DINODE_FMT_DEV:
367 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
368 if (iip->ili_fields & XFS_ILOG_DEV)
369 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
378 xfs_inode_item_format_attr_fork(
379 struct xfs_inode_log_item *iip,
380 struct xfs_inode_log_format *ilf,
381 struct xfs_log_vec *lv,
382 struct xfs_log_iovec **vecp)
384 struct xfs_inode *ip = iip->ili_inode;
387 switch (ip->i_af.if_format) {
388 case XFS_DINODE_FMT_EXTENTS:
390 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
392 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
393 ip->i_af.if_nextents > 0 &&
394 ip->i_af.if_bytes > 0) {
395 struct xfs_bmbt_rec *p;
397 ASSERT(xfs_iext_count(&ip->i_af) ==
398 ip->i_af.if_nextents);
400 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
401 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
402 xlog_finish_iovec(lv, *vecp, data_bytes);
404 ilf->ilf_asize = data_bytes;
407 iip->ili_fields &= ~XFS_ILOG_AEXT;
410 case XFS_DINODE_FMT_BTREE:
412 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
414 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
415 ip->i_af.if_broot_bytes > 0) {
416 ASSERT(ip->i_af.if_broot != NULL);
418 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
420 ip->i_af.if_broot_bytes);
421 ilf->ilf_asize = ip->i_af.if_broot_bytes;
424 iip->ili_fields &= ~XFS_ILOG_ABROOT;
427 case XFS_DINODE_FMT_LOCAL:
429 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
431 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
432 ip->i_af.if_bytes > 0) {
433 ASSERT(ip->i_af.if_data != NULL);
434 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
435 ip->i_af.if_data, ip->i_af.if_bytes);
436 ilf->ilf_asize = (unsigned)ip->i_af.if_bytes;
439 iip->ili_fields &= ~XFS_ILOG_ADATA;
449 * Convert an incore timestamp to a log timestamp. Note that the log format
450 * specifies host endian format!
452 static inline xfs_log_timestamp_t
453 xfs_inode_to_log_dinode_ts(
454 struct xfs_inode *ip,
455 const struct timespec64 tv)
457 struct xfs_log_legacy_timestamp *lits;
458 xfs_log_timestamp_t its;
460 if (xfs_inode_has_bigtime(ip))
461 return xfs_inode_encode_bigtime(tv);
463 lits = (struct xfs_log_legacy_timestamp *)&its;
464 lits->t_sec = tv.tv_sec;
465 lits->t_nsec = tv.tv_nsec;
471 * The legacy DMAPI fields are only present in the on-disk and in-log inodes,
472 * but not in the in-memory one. But we are guaranteed to have an inode buffer
473 * in memory when logging an inode, so we can just copy it from the on-disk
474 * inode to the in-log inode here so that recovery of file system with these
475 * fields set to non-zero values doesn't lose them. For all other cases we zero
479 xfs_copy_dm_fields_to_log_dinode(
480 struct xfs_inode *ip,
481 struct xfs_log_dinode *to)
483 struct xfs_dinode *dip;
485 dip = xfs_buf_offset(ip->i_itemp->ili_item.li_buf,
486 ip->i_imap.im_boffset);
488 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) {
489 to->di_dmevmask = be32_to_cpu(dip->di_dmevmask);
490 to->di_dmstate = be16_to_cpu(dip->di_dmstate);
498 xfs_inode_to_log_dinode_iext_counters(
499 struct xfs_inode *ip,
500 struct xfs_log_dinode *to)
502 if (xfs_inode_has_large_extent_counts(ip)) {
503 to->di_big_nextents = xfs_ifork_nextents(&ip->i_df);
504 to->di_big_anextents = xfs_ifork_nextents(&ip->i_af);
505 to->di_nrext64_pad = 0;
507 to->di_nextents = xfs_ifork_nextents(&ip->i_df);
508 to->di_anextents = xfs_ifork_nextents(&ip->i_af);
513 xfs_inode_to_log_dinode(
514 struct xfs_inode *ip,
515 struct xfs_log_dinode *to,
518 struct inode *inode = VFS_I(ip);
520 to->di_magic = XFS_DINODE_MAGIC;
521 to->di_format = xfs_ifork_format(&ip->i_df);
522 to->di_uid = i_uid_read(inode);
523 to->di_gid = i_gid_read(inode);
524 to->di_projid_lo = ip->i_projid & 0xffff;
525 to->di_projid_hi = ip->i_projid >> 16;
527 memset(to->di_pad3, 0, sizeof(to->di_pad3));
528 to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode_get_atime(inode));
529 to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode_get_mtime(inode));
530 to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode_get_ctime(inode));
531 to->di_nlink = inode->i_nlink;
532 to->di_gen = inode->i_generation;
533 to->di_mode = inode->i_mode;
535 to->di_size = ip->i_disk_size;
536 to->di_nblocks = ip->i_nblocks;
537 to->di_extsize = ip->i_extsize;
538 to->di_forkoff = ip->i_forkoff;
539 to->di_aformat = xfs_ifork_format(&ip->i_af);
540 to->di_flags = ip->i_diflags;
542 xfs_copy_dm_fields_to_log_dinode(ip, to);
544 /* log a dummy value to ensure log structure is fully initialised */
545 to->di_next_unlinked = NULLAGINO;
547 if (xfs_has_v3inodes(ip->i_mount)) {
549 to->di_changecount = inode_peek_iversion(inode);
550 to->di_crtime = xfs_inode_to_log_dinode_ts(ip, ip->i_crtime);
551 to->di_flags2 = ip->i_diflags2;
552 to->di_cowextsize = ip->i_cowextsize;
553 to->di_ino = ip->i_ino;
555 memset(to->di_pad2, 0, sizeof(to->di_pad2));
556 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
559 /* dummy value for initialisation */
563 to->di_flushiter = ip->i_flushiter;
564 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
567 xfs_inode_to_log_dinode_iext_counters(ip, to);
571 * Format the inode core. Current timestamp data is only in the VFS inode
572 * fields, so we need to grab them from there. Hence rather than just copying
573 * the XFS inode core structure, format the fields directly into the iovec.
576 xfs_inode_item_format_core(
577 struct xfs_inode *ip,
578 struct xfs_log_vec *lv,
579 struct xfs_log_iovec **vecp)
581 struct xfs_log_dinode *dic;
583 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
584 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
585 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
589 * This is called to fill in the vector of log iovecs for the given inode
590 * log item. It fills the first item with an inode log format structure,
591 * the second with the on-disk inode structure, and a possible third and/or
592 * fourth with the inode data/extents/b-tree root and inode attributes
593 * data/extents/b-tree root.
595 * Note: Always use the 64 bit inode log format structure so we don't
596 * leave an uninitialised hole in the format item on 64 bit systems. Log
597 * recovery on 32 bit systems handles this just fine, so there's no reason
598 * for not using an initialising the properly padded structure all the time.
601 xfs_inode_item_format(
602 struct xfs_log_item *lip,
603 struct xfs_log_vec *lv)
605 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
606 struct xfs_inode *ip = iip->ili_inode;
607 struct xfs_log_iovec *vecp = NULL;
608 struct xfs_inode_log_format *ilf;
610 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
611 ilf->ilf_type = XFS_LI_INODE;
612 ilf->ilf_ino = ip->i_ino;
613 ilf->ilf_blkno = ip->i_imap.im_blkno;
614 ilf->ilf_len = ip->i_imap.im_len;
615 ilf->ilf_boffset = ip->i_imap.im_boffset;
616 ilf->ilf_fields = XFS_ILOG_CORE;
617 ilf->ilf_size = 2; /* format + core */
620 * make sure we don't leak uninitialised data into the log in the case
621 * when we don't log every field in the inode.
626 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
628 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
630 xfs_inode_item_format_core(ip, lv, &vecp);
631 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
632 if (xfs_inode_has_attr_fork(ip)) {
633 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
636 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
639 /* update the format with the exact fields we actually logged */
640 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
644 * This is called to pin the inode associated with the inode log
645 * item in memory so it cannot be written out.
649 struct xfs_log_item *lip)
651 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
653 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
656 trace_xfs_inode_pin(ip, _RET_IP_);
657 atomic_inc(&ip->i_pincount);
662 * This is called to unpin the inode associated with the inode log
663 * item which was previously pinned with a call to xfs_inode_item_pin().
665 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
667 * Note that unpin can race with inode cluster buffer freeing marking the buffer
668 * stale. In that case, flush completions are run from the buffer unpin call,
669 * which may happen before the inode is unpinned. If we lose the race, there
670 * will be no buffer attached to the log item, but the inode will be marked
674 xfs_inode_item_unpin(
675 struct xfs_log_item *lip,
678 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
680 trace_xfs_inode_unpin(ip, _RET_IP_);
681 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
682 ASSERT(atomic_read(&ip->i_pincount) > 0);
683 if (atomic_dec_and_test(&ip->i_pincount))
684 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
689 struct xfs_log_item *lip,
690 struct list_head *buffer_list)
691 __releases(&lip->li_ailp->ail_lock)
692 __acquires(&lip->li_ailp->ail_lock)
694 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
695 struct xfs_inode *ip = iip->ili_inode;
696 struct xfs_buf *bp = lip->li_buf;
697 uint rval = XFS_ITEM_SUCCESS;
700 if (!bp || (ip->i_flags & XFS_ISTALE)) {
702 * Inode item/buffer is being aborted due to cluster
703 * buffer deletion. Trigger a log force to have that operation
704 * completed and items removed from the AIL before the next push
707 return XFS_ITEM_PINNED;
710 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp))
711 return XFS_ITEM_PINNED;
713 if (xfs_iflags_test(ip, XFS_IFLUSHING))
714 return XFS_ITEM_FLUSHING;
716 if (!xfs_buf_trylock(bp))
717 return XFS_ITEM_LOCKED;
719 spin_unlock(&lip->li_ailp->ail_lock);
722 * We need to hold a reference for flushing the cluster buffer as it may
723 * fail the buffer without IO submission. In which case, we better get a
724 * reference for that completion because otherwise we don't get a
725 * reference for IO until we queue the buffer for delwri submission.
728 error = xfs_iflush_cluster(bp);
730 if (!xfs_buf_delwri_queue(bp, buffer_list))
731 rval = XFS_ITEM_FLUSHING;
735 * Release the buffer if we were unable to flush anything. On
736 * any other error, the buffer has already been released.
738 if (error == -EAGAIN)
740 rval = XFS_ITEM_LOCKED;
743 spin_lock(&lip->li_ailp->ail_lock);
748 * Unlock the inode associated with the inode log item.
751 xfs_inode_item_release(
752 struct xfs_log_item *lip)
754 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
755 struct xfs_inode *ip = iip->ili_inode;
756 unsigned short lock_flags;
758 ASSERT(ip->i_itemp != NULL);
759 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
761 lock_flags = iip->ili_lock_flags;
762 iip->ili_lock_flags = 0;
764 xfs_iunlock(ip, lock_flags);
768 * This is called to find out where the oldest active copy of the inode log
769 * item in the on disk log resides now that the last log write of it completed
770 * at the given lsn. Since we always re-log all dirty data in an inode, the
771 * latest copy in the on disk log is the only one that matters. Therefore,
772 * simply return the given lsn.
774 * If the inode has been marked stale because the cluster is being freed, we
775 * don't want to (re-)insert this inode into the AIL. There is a race condition
776 * where the cluster buffer may be unpinned before the inode is inserted into
777 * the AIL during transaction committed processing. If the buffer is unpinned
778 * before the inode item has been committed and inserted, then it is possible
779 * for the buffer to be written and IO completes before the inode is inserted
780 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
781 * AIL which will never get removed. It will, however, get reclaimed which
782 * triggers an assert in xfs_inode_free() complaining about freein an inode
785 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
786 * transaction committed code knows that it does not need to do any further
787 * processing on the item.
790 xfs_inode_item_committed(
791 struct xfs_log_item *lip,
794 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
795 struct xfs_inode *ip = iip->ili_inode;
797 if (xfs_iflags_test(ip, XFS_ISTALE)) {
798 xfs_inode_item_unpin(lip, 0);
805 xfs_inode_item_committing(
806 struct xfs_log_item *lip,
809 INODE_ITEM(lip)->ili_commit_seq = seq;
810 return xfs_inode_item_release(lip);
813 static const struct xfs_item_ops xfs_inode_item_ops = {
814 .iop_sort = xfs_inode_item_sort,
815 .iop_precommit = xfs_inode_item_precommit,
816 .iop_size = xfs_inode_item_size,
817 .iop_format = xfs_inode_item_format,
818 .iop_pin = xfs_inode_item_pin,
819 .iop_unpin = xfs_inode_item_unpin,
820 .iop_release = xfs_inode_item_release,
821 .iop_committed = xfs_inode_item_committed,
822 .iop_push = xfs_inode_item_push,
823 .iop_committing = xfs_inode_item_committing,
828 * Initialize the inode log item for a newly allocated (in-core) inode.
832 struct xfs_inode *ip,
833 struct xfs_mount *mp)
835 struct xfs_inode_log_item *iip;
837 ASSERT(ip->i_itemp == NULL);
838 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_cache,
839 GFP_KERNEL | __GFP_NOFAIL);
842 spin_lock_init(&iip->ili_lock);
843 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
844 &xfs_inode_item_ops);
848 * Free the inode log item and any memory hanging off of it.
851 xfs_inode_item_destroy(
852 struct xfs_inode *ip)
854 struct xfs_inode_log_item *iip = ip->i_itemp;
856 ASSERT(iip->ili_item.li_buf == NULL);
859 kvfree(iip->ili_item.li_lv_shadow);
860 kmem_cache_free(xfs_ili_cache, iip);
865 * We only want to pull the item from the AIL if it is actually there
866 * and its location in the log has not changed since we started the
867 * flush. Thus, we only bother if the inode's lsn has not changed.
870 xfs_iflush_ail_updates(
871 struct xfs_ail *ailp,
872 struct list_head *list)
874 struct xfs_log_item *lip;
875 xfs_lsn_t tail_lsn = 0;
877 /* this is an opencoded batch version of xfs_trans_ail_delete */
878 spin_lock(&ailp->ail_lock);
879 list_for_each_entry(lip, list, li_bio_list) {
882 clear_bit(XFS_LI_FAILED, &lip->li_flags);
883 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
887 * dgc: Not sure how this happens, but it happens very
888 * occassionaly via generic/388. xfs_iflush_abort() also
889 * silently handles this same "under writeback but not in AIL at
890 * shutdown" condition via xfs_trans_ail_delete().
892 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
893 ASSERT(xlog_is_shutdown(lip->li_log));
897 lsn = xfs_ail_delete_one(ailp, lip);
898 if (!tail_lsn && lsn)
901 xfs_ail_update_finish(ailp, tail_lsn);
905 * Walk the list of inodes that have completed their IOs. If they are clean
906 * remove them from the list and dissociate them from the buffer. Buffers that
907 * are still dirty remain linked to the buffer and on the list. Caller must
908 * handle them appropriately.
913 struct list_head *list)
915 struct xfs_log_item *lip, *n;
917 list_for_each_entry_safe(lip, n, list, li_bio_list) {
918 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
919 bool drop_buffer = false;
921 spin_lock(&iip->ili_lock);
924 * Remove the reference to the cluster buffer if the inode is
925 * clean in memory and drop the buffer reference once we've
926 * dropped the locks we hold.
928 ASSERT(iip->ili_item.li_buf == bp);
929 if (!iip->ili_fields) {
930 iip->ili_item.li_buf = NULL;
931 list_del_init(&lip->li_bio_list);
934 iip->ili_last_fields = 0;
935 iip->ili_flush_lsn = 0;
936 spin_unlock(&iip->ili_lock);
937 xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);
944 * Inode buffer IO completion routine. It is responsible for removing inodes
945 * attached to the buffer from the AIL if they have not been re-logged and
946 * completing the inode flush.
949 xfs_buf_inode_iodone(
952 struct xfs_log_item *lip, *n;
953 LIST_HEAD(flushed_inodes);
954 LIST_HEAD(ail_updates);
957 * Pull the attached inodes from the buffer one at a time and take the
958 * appropriate action on them.
960 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
961 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
963 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
964 xfs_iflush_abort(iip->ili_inode);
967 if (!iip->ili_last_fields)
970 /* Do an unlocked check for needing the AIL lock. */
971 if (iip->ili_flush_lsn == lip->li_lsn ||
972 test_bit(XFS_LI_FAILED, &lip->li_flags))
973 list_move_tail(&lip->li_bio_list, &ail_updates);
975 list_move_tail(&lip->li_bio_list, &flushed_inodes);
978 if (!list_empty(&ail_updates)) {
979 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
980 list_splice_tail(&ail_updates, &flushed_inodes);
983 xfs_iflush_finish(bp, &flushed_inodes);
984 if (!list_empty(&flushed_inodes))
985 list_splice_tail(&flushed_inodes, &bp->b_li_list);
989 xfs_buf_inode_io_fail(
992 struct xfs_log_item *lip;
994 list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
995 set_bit(XFS_LI_FAILED, &lip->li_flags);
999 * Clear the inode logging fields so no more flushes are attempted. If we are
1000 * on a buffer list, it is now safe to remove it because the buffer is
1001 * guaranteed to be locked. The caller will drop the reference to the buffer
1002 * the log item held.
1005 xfs_iflush_abort_clean(
1006 struct xfs_inode_log_item *iip)
1008 iip->ili_last_fields = 0;
1009 iip->ili_fields = 0;
1010 iip->ili_fsync_fields = 0;
1011 iip->ili_flush_lsn = 0;
1012 iip->ili_item.li_buf = NULL;
1013 list_del_init(&iip->ili_item.li_bio_list);
1017 * Abort flushing the inode from a context holding the cluster buffer locked.
1019 * This is the normal runtime method of aborting writeback of an inode that is
1020 * attached to a cluster buffer. It occurs when the inode and the backing
1021 * cluster buffer have been freed (i.e. inode is XFS_ISTALE), or when cluster
1022 * flushing or buffer IO completion encounters a log shutdown situation.
1024 * If we need to abort inode writeback and we don't already hold the buffer
1025 * locked, call xfs_iflush_shutdown_abort() instead as this should only ever be
1026 * necessary in a shutdown situation.
1030 struct xfs_inode *ip)
1032 struct xfs_inode_log_item *iip = ip->i_itemp;
1036 /* clean inode, nothing to do */
1037 xfs_iflags_clear(ip, XFS_IFLUSHING);
1042 * Remove the inode item from the AIL before we clear its internal
1043 * state. Whilst the inode is in the AIL, it should have a valid buffer
1044 * pointer for push operations to access - it is only safe to remove the
1045 * inode from the buffer once it has been removed from the AIL.
1047 * We also clear the failed bit before removing the item from the AIL
1048 * as xfs_trans_ail_delete()->xfs_clear_li_failed() will release buffer
1049 * references the inode item owns and needs to hold until we've fully
1050 * aborted the inode log item and detached it from the buffer.
1052 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
1053 xfs_trans_ail_delete(&iip->ili_item, 0);
1056 * Grab the inode buffer so can we release the reference the inode log
1059 spin_lock(&iip->ili_lock);
1060 bp = iip->ili_item.li_buf;
1061 xfs_iflush_abort_clean(iip);
1062 spin_unlock(&iip->ili_lock);
1064 xfs_iflags_clear(ip, XFS_IFLUSHING);
1070 * Abort an inode flush in the case of a shutdown filesystem. This can be called
1071 * from anywhere with just an inode reference and does not require holding the
1072 * inode cluster buffer locked. If the inode is attached to a cluster buffer,
1073 * it will grab and lock it safely, then abort the inode flush.
1076 xfs_iflush_shutdown_abort(
1077 struct xfs_inode *ip)
1079 struct xfs_inode_log_item *iip = ip->i_itemp;
1083 /* clean inode, nothing to do */
1084 xfs_iflags_clear(ip, XFS_IFLUSHING);
1088 spin_lock(&iip->ili_lock);
1089 bp = iip->ili_item.li_buf;
1091 spin_unlock(&iip->ili_lock);
1092 xfs_iflush_abort(ip);
1097 * We have to take a reference to the buffer so that it doesn't get
1098 * freed when we drop the ili_lock and then wait to lock the buffer.
1099 * We'll clean up the extra reference after we pick up the ili_lock
1103 spin_unlock(&iip->ili_lock);
1106 spin_lock(&iip->ili_lock);
1107 if (!iip->ili_item.li_buf) {
1109 * Raced with another removal, hold the only reference
1110 * to bp now. Inode should not be in the AIL now, so just clean
1113 ASSERT(list_empty(&iip->ili_item.li_bio_list));
1114 ASSERT(!test_bit(XFS_LI_IN_AIL, &iip->ili_item.li_flags));
1115 xfs_iflush_abort_clean(iip);
1116 spin_unlock(&iip->ili_lock);
1117 xfs_iflags_clear(ip, XFS_IFLUSHING);
1123 * Got two references to bp. The first will get dropped by
1124 * xfs_iflush_abort() when the item is removed from the buffer list, but
1125 * we can't drop our reference until _abort() returns because we have to
1126 * unlock the buffer as well. Hence we abort and then unlock and release
1127 * our reference to the buffer.
1129 ASSERT(iip->ili_item.li_buf == bp);
1130 spin_unlock(&iip->ili_lock);
1131 xfs_iflush_abort(ip);
1137 * convert an xfs_inode_log_format struct from the old 32 bit version
1138 * (which can have different field alignments) to the native 64 bit version
1141 xfs_inode_item_format_convert(
1142 struct xfs_log_iovec *buf,
1143 struct xfs_inode_log_format *in_f)
1145 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
1147 if (buf->i_len != sizeof(*in_f32)) {
1148 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
1149 return -EFSCORRUPTED;
1152 in_f->ilf_type = in_f32->ilf_type;
1153 in_f->ilf_size = in_f32->ilf_size;
1154 in_f->ilf_fields = in_f32->ilf_fields;
1155 in_f->ilf_asize = in_f32->ilf_asize;
1156 in_f->ilf_dsize = in_f32->ilf_dsize;
1157 in_f->ilf_ino = in_f32->ilf_ino;
1158 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
1159 in_f->ilf_blkno = in_f32->ilf_blkno;
1160 in_f->ilf_len = in_f32->ilf_len;
1161 in_f->ilf_boffset = in_f32->ilf_boffset;