Loading...
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_types.h"
21#include "xfs_log.h"
22#include "xfs_trans.h"
23#include "xfs_sb.h"
24#include "xfs_ag.h"
25#include "xfs_mount.h"
26#include "xfs_trans_priv.h"
27#include "xfs_bmap_btree.h"
28#include "xfs_dinode.h"
29#include "xfs_inode.h"
30#include "xfs_inode_item.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33
34
35kmem_zone_t *xfs_ili_zone; /* inode log item zone */
36
37static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
38{
39 return container_of(lip, struct xfs_inode_log_item, ili_item);
40}
41
42
43/*
44 * This returns the number of iovecs needed to log the given inode item.
45 *
46 * We need one iovec for the inode log format structure, one for the
47 * inode core, and possibly one for the inode data/extents/b-tree root
48 * and one for the inode attribute data/extents/b-tree root.
49 */
50STATIC uint
51xfs_inode_item_size(
52 struct xfs_log_item *lip)
53{
54 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
55 struct xfs_inode *ip = iip->ili_inode;
56 uint nvecs = 2;
57
58 switch (ip->i_d.di_format) {
59 case XFS_DINODE_FMT_EXTENTS:
60 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
61 ip->i_d.di_nextents > 0 &&
62 ip->i_df.if_bytes > 0)
63 nvecs++;
64 break;
65
66 case XFS_DINODE_FMT_BTREE:
67 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
68 ip->i_df.if_broot_bytes > 0)
69 nvecs++;
70 break;
71
72 case XFS_DINODE_FMT_LOCAL:
73 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
74 ip->i_df.if_bytes > 0)
75 nvecs++;
76 break;
77
78 case XFS_DINODE_FMT_DEV:
79 case XFS_DINODE_FMT_UUID:
80 break;
81
82 default:
83 ASSERT(0);
84 break;
85 }
86
87 if (!XFS_IFORK_Q(ip))
88 return nvecs;
89
90
91 /*
92 * Log any necessary attribute data.
93 */
94 switch (ip->i_d.di_aformat) {
95 case XFS_DINODE_FMT_EXTENTS:
96 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
97 ip->i_d.di_anextents > 0 &&
98 ip->i_afp->if_bytes > 0)
99 nvecs++;
100 break;
101
102 case XFS_DINODE_FMT_BTREE:
103 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
104 ip->i_afp->if_broot_bytes > 0)
105 nvecs++;
106 break;
107
108 case XFS_DINODE_FMT_LOCAL:
109 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
110 ip->i_afp->if_bytes > 0)
111 nvecs++;
112 break;
113
114 default:
115 ASSERT(0);
116 break;
117 }
118
119 return nvecs;
120}
121
122/*
123 * xfs_inode_item_format_extents - convert in-core extents to on-disk form
124 *
125 * For either the data or attr fork in extent format, we need to endian convert
126 * the in-core extent as we place them into the on-disk inode. In this case, we
127 * need to do this conversion before we write the extents into the log. Because
128 * we don't have the disk inode to write into here, we allocate a buffer and
129 * format the extents into it via xfs_iextents_copy(). We free the buffer in
130 * the unlock routine after the copy for the log has been made.
131 *
132 * In the case of the data fork, the in-core and on-disk fork sizes can be
133 * different due to delayed allocation extents. We only log on-disk extents
134 * here, so always use the physical fork size to determine the size of the
135 * buffer we need to allocate.
136 */
137STATIC void
138xfs_inode_item_format_extents(
139 struct xfs_inode *ip,
140 struct xfs_log_iovec *vecp,
141 int whichfork,
142 int type)
143{
144 xfs_bmbt_rec_t *ext_buffer;
145
146 ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
147 if (whichfork == XFS_DATA_FORK)
148 ip->i_itemp->ili_extents_buf = ext_buffer;
149 else
150 ip->i_itemp->ili_aextents_buf = ext_buffer;
151
152 vecp->i_addr = ext_buffer;
153 vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
154 vecp->i_type = type;
155}
156
157/*
158 * This is called to fill in the vector of log iovecs for the
159 * given inode log item. It fills the first item with an inode
160 * log format structure, the second with the on-disk inode structure,
161 * and a possible third and/or fourth with the inode data/extents/b-tree
162 * root and inode attributes data/extents/b-tree root.
163 */
164STATIC void
165xfs_inode_item_format(
166 struct xfs_log_item *lip,
167 struct xfs_log_iovec *vecp)
168{
169 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
170 struct xfs_inode *ip = iip->ili_inode;
171 uint nvecs;
172 size_t data_bytes;
173 xfs_mount_t *mp;
174
175 vecp->i_addr = &iip->ili_format;
176 vecp->i_len = sizeof(xfs_inode_log_format_t);
177 vecp->i_type = XLOG_REG_TYPE_IFORMAT;
178 vecp++;
179 nvecs = 1;
180
181 vecp->i_addr = &ip->i_d;
182 vecp->i_len = sizeof(struct xfs_icdinode);
183 vecp->i_type = XLOG_REG_TYPE_ICORE;
184 vecp++;
185 nvecs++;
186
187 /*
188 * If this is really an old format inode, then we need to
189 * log it as such. This means that we have to copy the link
190 * count from the new field to the old. We don't have to worry
191 * about the new fields, because nothing trusts them as long as
192 * the old inode version number is there. If the superblock already
193 * has a new version number, then we don't bother converting back.
194 */
195 mp = ip->i_mount;
196 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
197 if (ip->i_d.di_version == 1) {
198 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
199 /*
200 * Convert it back.
201 */
202 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
203 ip->i_d.di_onlink = ip->i_d.di_nlink;
204 } else {
205 /*
206 * The superblock version has already been bumped,
207 * so just make the conversion to the new inode
208 * format permanent.
209 */
210 ip->i_d.di_version = 2;
211 ip->i_d.di_onlink = 0;
212 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
213 }
214 }
215
216 switch (ip->i_d.di_format) {
217 case XFS_DINODE_FMT_EXTENTS:
218 iip->ili_fields &=
219 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
220 XFS_ILOG_DEV | XFS_ILOG_UUID);
221
222 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
223 ip->i_d.di_nextents > 0 &&
224 ip->i_df.if_bytes > 0) {
225 ASSERT(ip->i_df.if_u1.if_extents != NULL);
226 ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
227 ASSERT(iip->ili_extents_buf == NULL);
228
229#ifdef XFS_NATIVE_HOST
230 if (ip->i_d.di_nextents == ip->i_df.if_bytes /
231 (uint)sizeof(xfs_bmbt_rec_t)) {
232 /*
233 * There are no delayed allocation
234 * extents, so just point to the
235 * real extents array.
236 */
237 vecp->i_addr = ip->i_df.if_u1.if_extents;
238 vecp->i_len = ip->i_df.if_bytes;
239 vecp->i_type = XLOG_REG_TYPE_IEXT;
240 } else
241#endif
242 {
243 xfs_inode_item_format_extents(ip, vecp,
244 XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
245 }
246 ASSERT(vecp->i_len <= ip->i_df.if_bytes);
247 iip->ili_format.ilf_dsize = vecp->i_len;
248 vecp++;
249 nvecs++;
250 } else {
251 iip->ili_fields &= ~XFS_ILOG_DEXT;
252 }
253 break;
254
255 case XFS_DINODE_FMT_BTREE:
256 iip->ili_fields &=
257 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
258 XFS_ILOG_DEV | XFS_ILOG_UUID);
259
260 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
261 ip->i_df.if_broot_bytes > 0) {
262 ASSERT(ip->i_df.if_broot != NULL);
263 vecp->i_addr = ip->i_df.if_broot;
264 vecp->i_len = ip->i_df.if_broot_bytes;
265 vecp->i_type = XLOG_REG_TYPE_IBROOT;
266 vecp++;
267 nvecs++;
268 iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
269 } else {
270 ASSERT(!(iip->ili_fields &
271 XFS_ILOG_DBROOT));
272#ifdef XFS_TRANS_DEBUG
273 if (iip->ili_root_size > 0) {
274 ASSERT(iip->ili_root_size ==
275 ip->i_df.if_broot_bytes);
276 ASSERT(memcmp(iip->ili_orig_root,
277 ip->i_df.if_broot,
278 iip->ili_root_size) == 0);
279 } else {
280 ASSERT(ip->i_df.if_broot_bytes == 0);
281 }
282#endif
283 iip->ili_fields &= ~XFS_ILOG_DBROOT;
284 }
285 break;
286
287 case XFS_DINODE_FMT_LOCAL:
288 iip->ili_fields &=
289 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
290 XFS_ILOG_DEV | XFS_ILOG_UUID);
291 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
292 ip->i_df.if_bytes > 0) {
293 ASSERT(ip->i_df.if_u1.if_data != NULL);
294 ASSERT(ip->i_d.di_size > 0);
295
296 vecp->i_addr = ip->i_df.if_u1.if_data;
297 /*
298 * Round i_bytes up to a word boundary.
299 * The underlying memory is guaranteed to
300 * to be there by xfs_idata_realloc().
301 */
302 data_bytes = roundup(ip->i_df.if_bytes, 4);
303 ASSERT((ip->i_df.if_real_bytes == 0) ||
304 (ip->i_df.if_real_bytes == data_bytes));
305 vecp->i_len = (int)data_bytes;
306 vecp->i_type = XLOG_REG_TYPE_ILOCAL;
307 vecp++;
308 nvecs++;
309 iip->ili_format.ilf_dsize = (unsigned)data_bytes;
310 } else {
311 iip->ili_fields &= ~XFS_ILOG_DDATA;
312 }
313 break;
314
315 case XFS_DINODE_FMT_DEV:
316 iip->ili_fields &=
317 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
318 XFS_ILOG_DEXT | XFS_ILOG_UUID);
319 if (iip->ili_fields & XFS_ILOG_DEV) {
320 iip->ili_format.ilf_u.ilfu_rdev =
321 ip->i_df.if_u2.if_rdev;
322 }
323 break;
324
325 case XFS_DINODE_FMT_UUID:
326 iip->ili_fields &=
327 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
328 XFS_ILOG_DEXT | XFS_ILOG_DEV);
329 if (iip->ili_fields & XFS_ILOG_UUID) {
330 iip->ili_format.ilf_u.ilfu_uuid =
331 ip->i_df.if_u2.if_uuid;
332 }
333 break;
334
335 default:
336 ASSERT(0);
337 break;
338 }
339
340 /*
341 * If there are no attributes associated with the file, then we're done.
342 */
343 if (!XFS_IFORK_Q(ip)) {
344 iip->ili_fields &=
345 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
346 goto out;
347 }
348
349 switch (ip->i_d.di_aformat) {
350 case XFS_DINODE_FMT_EXTENTS:
351 iip->ili_fields &=
352 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
353
354 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
355 ip->i_d.di_anextents > 0 &&
356 ip->i_afp->if_bytes > 0) {
357 ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
358 ip->i_d.di_anextents);
359 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
360#ifdef XFS_NATIVE_HOST
361 /*
362 * There are not delayed allocation extents
363 * for attributes, so just point at the array.
364 */
365 vecp->i_addr = ip->i_afp->if_u1.if_extents;
366 vecp->i_len = ip->i_afp->if_bytes;
367 vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
368#else
369 ASSERT(iip->ili_aextents_buf == NULL);
370 xfs_inode_item_format_extents(ip, vecp,
371 XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
372#endif
373 iip->ili_format.ilf_asize = vecp->i_len;
374 vecp++;
375 nvecs++;
376 } else {
377 iip->ili_fields &= ~XFS_ILOG_AEXT;
378 }
379 break;
380
381 case XFS_DINODE_FMT_BTREE:
382 iip->ili_fields &=
383 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
384
385 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
386 ip->i_afp->if_broot_bytes > 0) {
387 ASSERT(ip->i_afp->if_broot != NULL);
388
389 vecp->i_addr = ip->i_afp->if_broot;
390 vecp->i_len = ip->i_afp->if_broot_bytes;
391 vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
392 vecp++;
393 nvecs++;
394 iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
395 } else {
396 iip->ili_fields &= ~XFS_ILOG_ABROOT;
397 }
398 break;
399
400 case XFS_DINODE_FMT_LOCAL:
401 iip->ili_fields &=
402 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
403
404 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
405 ip->i_afp->if_bytes > 0) {
406 ASSERT(ip->i_afp->if_u1.if_data != NULL);
407
408 vecp->i_addr = ip->i_afp->if_u1.if_data;
409 /*
410 * Round i_bytes up to a word boundary.
411 * The underlying memory is guaranteed to
412 * to be there by xfs_idata_realloc().
413 */
414 data_bytes = roundup(ip->i_afp->if_bytes, 4);
415 ASSERT((ip->i_afp->if_real_bytes == 0) ||
416 (ip->i_afp->if_real_bytes == data_bytes));
417 vecp->i_len = (int)data_bytes;
418 vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
419 vecp++;
420 nvecs++;
421 iip->ili_format.ilf_asize = (unsigned)data_bytes;
422 } else {
423 iip->ili_fields &= ~XFS_ILOG_ADATA;
424 }
425 break;
426
427 default:
428 ASSERT(0);
429 break;
430 }
431
432out:
433 /*
434 * Now update the log format that goes out to disk from the in-core
435 * values. We always write the inode core to make the arithmetic
436 * games in recovery easier, which isn't a big deal as just about any
437 * transaction would dirty it anyway.
438 */
439 iip->ili_format.ilf_fields = XFS_ILOG_CORE |
440 (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
441 iip->ili_format.ilf_size = nvecs;
442}
443
444
445/*
446 * This is called to pin the inode associated with the inode log
447 * item in memory so it cannot be written out.
448 */
449STATIC void
450xfs_inode_item_pin(
451 struct xfs_log_item *lip)
452{
453 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
454
455 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
456
457 trace_xfs_inode_pin(ip, _RET_IP_);
458 atomic_inc(&ip->i_pincount);
459}
460
461
462/*
463 * This is called to unpin the inode associated with the inode log
464 * item which was previously pinned with a call to xfs_inode_item_pin().
465 *
466 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
467 */
468STATIC void
469xfs_inode_item_unpin(
470 struct xfs_log_item *lip,
471 int remove)
472{
473 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
474
475 trace_xfs_inode_unpin(ip, _RET_IP_);
476 ASSERT(atomic_read(&ip->i_pincount) > 0);
477 if (atomic_dec_and_test(&ip->i_pincount))
478 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
479}
480
481STATIC uint
482xfs_inode_item_push(
483 struct xfs_log_item *lip,
484 struct list_head *buffer_list)
485{
486 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
487 struct xfs_inode *ip = iip->ili_inode;
488 struct xfs_buf *bp = NULL;
489 uint rval = XFS_ITEM_SUCCESS;
490 int error;
491
492 if (xfs_ipincount(ip) > 0)
493 return XFS_ITEM_PINNED;
494
495 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
496 return XFS_ITEM_LOCKED;
497
498 /*
499 * Re-check the pincount now that we stabilized the value by
500 * taking the ilock.
501 */
502 if (xfs_ipincount(ip) > 0) {
503 rval = XFS_ITEM_PINNED;
504 goto out_unlock;
505 }
506
507 /*
508 * Stale inode items should force out the iclog.
509 */
510 if (ip->i_flags & XFS_ISTALE) {
511 rval = XFS_ITEM_PINNED;
512 goto out_unlock;
513 }
514
515 /*
516 * Someone else is already flushing the inode. Nothing we can do
517 * here but wait for the flush to finish and remove the item from
518 * the AIL.
519 */
520 if (!xfs_iflock_nowait(ip)) {
521 rval = XFS_ITEM_FLUSHING;
522 goto out_unlock;
523 }
524
525 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
526 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
527
528 spin_unlock(&lip->li_ailp->xa_lock);
529
530 error = xfs_iflush(ip, &bp);
531 if (!error) {
532 if (!xfs_buf_delwri_queue(bp, buffer_list))
533 rval = XFS_ITEM_FLUSHING;
534 xfs_buf_relse(bp);
535 }
536
537 spin_lock(&lip->li_ailp->xa_lock);
538out_unlock:
539 xfs_iunlock(ip, XFS_ILOCK_SHARED);
540 return rval;
541}
542
543/*
544 * Unlock the inode associated with the inode log item.
545 * Clear the fields of the inode and inode log item that
546 * are specific to the current transaction. If the
547 * hold flags is set, do not unlock the inode.
548 */
549STATIC void
550xfs_inode_item_unlock(
551 struct xfs_log_item *lip)
552{
553 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
554 struct xfs_inode *ip = iip->ili_inode;
555 unsigned short lock_flags;
556
557 ASSERT(ip->i_itemp != NULL);
558 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
559
560 /*
561 * If the inode needed a separate buffer with which to log
562 * its extents, then free it now.
563 */
564 if (iip->ili_extents_buf != NULL) {
565 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
566 ASSERT(ip->i_d.di_nextents > 0);
567 ASSERT(iip->ili_fields & XFS_ILOG_DEXT);
568 ASSERT(ip->i_df.if_bytes > 0);
569 kmem_free(iip->ili_extents_buf);
570 iip->ili_extents_buf = NULL;
571 }
572 if (iip->ili_aextents_buf != NULL) {
573 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
574 ASSERT(ip->i_d.di_anextents > 0);
575 ASSERT(iip->ili_fields & XFS_ILOG_AEXT);
576 ASSERT(ip->i_afp->if_bytes > 0);
577 kmem_free(iip->ili_aextents_buf);
578 iip->ili_aextents_buf = NULL;
579 }
580
581 lock_flags = iip->ili_lock_flags;
582 iip->ili_lock_flags = 0;
583 if (lock_flags)
584 xfs_iunlock(ip, lock_flags);
585}
586
587/*
588 * This is called to find out where the oldest active copy of the inode log
589 * item in the on disk log resides now that the last log write of it completed
590 * at the given lsn. Since we always re-log all dirty data in an inode, the
591 * latest copy in the on disk log is the only one that matters. Therefore,
592 * simply return the given lsn.
593 *
594 * If the inode has been marked stale because the cluster is being freed, we
595 * don't want to (re-)insert this inode into the AIL. There is a race condition
596 * where the cluster buffer may be unpinned before the inode is inserted into
597 * the AIL during transaction committed processing. If the buffer is unpinned
598 * before the inode item has been committed and inserted, then it is possible
599 * for the buffer to be written and IO completes before the inode is inserted
600 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
601 * AIL which will never get removed. It will, however, get reclaimed which
602 * triggers an assert in xfs_inode_free() complaining about freein an inode
603 * still in the AIL.
604 *
605 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
606 * transaction committed code knows that it does not need to do any further
607 * processing on the item.
608 */
609STATIC xfs_lsn_t
610xfs_inode_item_committed(
611 struct xfs_log_item *lip,
612 xfs_lsn_t lsn)
613{
614 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
615 struct xfs_inode *ip = iip->ili_inode;
616
617 if (xfs_iflags_test(ip, XFS_ISTALE)) {
618 xfs_inode_item_unpin(lip, 0);
619 return -1;
620 }
621 return lsn;
622}
623
624/*
625 * XXX rcc - this one really has to do something. Probably needs
626 * to stamp in a new field in the incore inode.
627 */
628STATIC void
629xfs_inode_item_committing(
630 struct xfs_log_item *lip,
631 xfs_lsn_t lsn)
632{
633 INODE_ITEM(lip)->ili_last_lsn = lsn;
634}
635
636/*
637 * This is the ops vector shared by all buf log items.
638 */
639static const struct xfs_item_ops xfs_inode_item_ops = {
640 .iop_size = xfs_inode_item_size,
641 .iop_format = xfs_inode_item_format,
642 .iop_pin = xfs_inode_item_pin,
643 .iop_unpin = xfs_inode_item_unpin,
644 .iop_unlock = xfs_inode_item_unlock,
645 .iop_committed = xfs_inode_item_committed,
646 .iop_push = xfs_inode_item_push,
647 .iop_committing = xfs_inode_item_committing
648};
649
650
651/*
652 * Initialize the inode log item for a newly allocated (in-core) inode.
653 */
654void
655xfs_inode_item_init(
656 struct xfs_inode *ip,
657 struct xfs_mount *mp)
658{
659 struct xfs_inode_log_item *iip;
660
661 ASSERT(ip->i_itemp == NULL);
662 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
663
664 iip->ili_inode = ip;
665 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
666 &xfs_inode_item_ops);
667 iip->ili_format.ilf_type = XFS_LI_INODE;
668 iip->ili_format.ilf_ino = ip->i_ino;
669 iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
670 iip->ili_format.ilf_len = ip->i_imap.im_len;
671 iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
672}
673
674/*
675 * Free the inode log item and any memory hanging off of it.
676 */
677void
678xfs_inode_item_destroy(
679 xfs_inode_t *ip)
680{
681#ifdef XFS_TRANS_DEBUG
682 if (ip->i_itemp->ili_root_size != 0) {
683 kmem_free(ip->i_itemp->ili_orig_root);
684 }
685#endif
686 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
687}
688
689
690/*
691 * This is the inode flushing I/O completion routine. It is called
692 * from interrupt level when the buffer containing the inode is
693 * flushed to disk. It is responsible for removing the inode item
694 * from the AIL if it has not been re-logged, and unlocking the inode's
695 * flush lock.
696 *
697 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
698 * list for other inodes that will run this function. We remove them from the
699 * buffer list so we can process all the inode IO completions in one AIL lock
700 * traversal.
701 */
702void
703xfs_iflush_done(
704 struct xfs_buf *bp,
705 struct xfs_log_item *lip)
706{
707 struct xfs_inode_log_item *iip;
708 struct xfs_log_item *blip;
709 struct xfs_log_item *next;
710 struct xfs_log_item *prev;
711 struct xfs_ail *ailp = lip->li_ailp;
712 int need_ail = 0;
713
714 /*
715 * Scan the buffer IO completions for other inodes being completed and
716 * attach them to the current inode log item.
717 */
718 blip = bp->b_fspriv;
719 prev = NULL;
720 while (blip != NULL) {
721 if (lip->li_cb != xfs_iflush_done) {
722 prev = blip;
723 blip = blip->li_bio_list;
724 continue;
725 }
726
727 /* remove from list */
728 next = blip->li_bio_list;
729 if (!prev) {
730 bp->b_fspriv = next;
731 } else {
732 prev->li_bio_list = next;
733 }
734
735 /* add to current list */
736 blip->li_bio_list = lip->li_bio_list;
737 lip->li_bio_list = blip;
738
739 /*
740 * while we have the item, do the unlocked check for needing
741 * the AIL lock.
742 */
743 iip = INODE_ITEM(blip);
744 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
745 need_ail++;
746
747 blip = next;
748 }
749
750 /* make sure we capture the state of the initial inode. */
751 iip = INODE_ITEM(lip);
752 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
753 need_ail++;
754
755 /*
756 * We only want to pull the item from the AIL if it is
757 * actually there and its location in the log has not
758 * changed since we started the flush. Thus, we only bother
759 * if the ili_logged flag is set and the inode's lsn has not
760 * changed. First we check the lsn outside
761 * the lock since it's cheaper, and then we recheck while
762 * holding the lock before removing the inode from the AIL.
763 */
764 if (need_ail) {
765 struct xfs_log_item *log_items[need_ail];
766 int i = 0;
767 spin_lock(&ailp->xa_lock);
768 for (blip = lip; blip; blip = blip->li_bio_list) {
769 iip = INODE_ITEM(blip);
770 if (iip->ili_logged &&
771 blip->li_lsn == iip->ili_flush_lsn) {
772 log_items[i++] = blip;
773 }
774 ASSERT(i <= need_ail);
775 }
776 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
777 xfs_trans_ail_delete_bulk(ailp, log_items, i,
778 SHUTDOWN_CORRUPT_INCORE);
779 }
780
781
782 /*
783 * clean up and unlock the flush lock now we are done. We can clear the
784 * ili_last_fields bits now that we know that the data corresponding to
785 * them is safely on disk.
786 */
787 for (blip = lip; blip; blip = next) {
788 next = blip->li_bio_list;
789 blip->li_bio_list = NULL;
790
791 iip = INODE_ITEM(blip);
792 iip->ili_logged = 0;
793 iip->ili_last_fields = 0;
794 xfs_ifunlock(iip->ili_inode);
795 }
796}
797
798/*
799 * This is the inode flushing abort routine. It is called from xfs_iflush when
800 * the filesystem is shutting down to clean up the inode state. It is
801 * responsible for removing the inode item from the AIL if it has not been
802 * re-logged, and unlocking the inode's flush lock.
803 */
804void
805xfs_iflush_abort(
806 xfs_inode_t *ip,
807 bool stale)
808{
809 xfs_inode_log_item_t *iip = ip->i_itemp;
810
811 if (iip) {
812 struct xfs_ail *ailp = iip->ili_item.li_ailp;
813 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
814 spin_lock(&ailp->xa_lock);
815 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
816 /* xfs_trans_ail_delete() drops the AIL lock. */
817 xfs_trans_ail_delete(ailp, &iip->ili_item,
818 stale ?
819 SHUTDOWN_LOG_IO_ERROR :
820 SHUTDOWN_CORRUPT_INCORE);
821 } else
822 spin_unlock(&ailp->xa_lock);
823 }
824 iip->ili_logged = 0;
825 /*
826 * Clear the ili_last_fields bits now that we know that the
827 * data corresponding to them is safely on disk.
828 */
829 iip->ili_last_fields = 0;
830 /*
831 * Clear the inode logging fields so no more flushes are
832 * attempted.
833 */
834 iip->ili_fields = 0;
835 }
836 /*
837 * Release the inode's flush lock since we're done with it.
838 */
839 xfs_ifunlock(ip);
840}
841
842void
843xfs_istale_done(
844 struct xfs_buf *bp,
845 struct xfs_log_item *lip)
846{
847 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
848}
849
850/*
851 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
852 * (which can have different field alignments) to the native version
853 */
854int
855xfs_inode_item_format_convert(
856 xfs_log_iovec_t *buf,
857 xfs_inode_log_format_t *in_f)
858{
859 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
860 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
861
862 in_f->ilf_type = in_f32->ilf_type;
863 in_f->ilf_size = in_f32->ilf_size;
864 in_f->ilf_fields = in_f32->ilf_fields;
865 in_f->ilf_asize = in_f32->ilf_asize;
866 in_f->ilf_dsize = in_f32->ilf_dsize;
867 in_f->ilf_ino = in_f32->ilf_ino;
868 /* copy biggest field of ilf_u */
869 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
870 in_f32->ilf_u.ilfu_uuid.__u_bits,
871 sizeof(uuid_t));
872 in_f->ilf_blkno = in_f32->ilf_blkno;
873 in_f->ilf_len = in_f32->ilf_len;
874 in_f->ilf_boffset = in_f32->ilf_boffset;
875 return 0;
876 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
877 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
878
879 in_f->ilf_type = in_f64->ilf_type;
880 in_f->ilf_size = in_f64->ilf_size;
881 in_f->ilf_fields = in_f64->ilf_fields;
882 in_f->ilf_asize = in_f64->ilf_asize;
883 in_f->ilf_dsize = in_f64->ilf_dsize;
884 in_f->ilf_ino = in_f64->ilf_ino;
885 /* copy biggest field of ilf_u */
886 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
887 in_f64->ilf_u.ilfu_uuid.__u_bits,
888 sizeof(uuid_t));
889 in_f->ilf_blkno = in_f64->ilf_blkno;
890 in_f->ilf_len = in_f64->ilf_len;
891 in_f->ilf_boffset = in_f64->ilf_boffset;
892 return 0;
893 }
894 return EFSCORRUPTED;
895}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
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"
19#include "xfs_log.h"
20#include "xfs_log_priv.h"
21#include "xfs_error.h"
22#include "xfs_rtbitmap.h"
23
24#include <linux/iversion.h>
25
26struct kmem_cache *xfs_ili_cache; /* inode log item */
27
28static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
29{
30 return container_of(lip, struct xfs_inode_log_item, ili_item);
31}
32
33static uint64_t
34xfs_inode_item_sort(
35 struct xfs_log_item *lip)
36{
37 return INODE_ITEM(lip)->ili_inode->i_ino;
38}
39
40#ifdef DEBUG_EXPENSIVE
41static void
42xfs_inode_item_precommit_check(
43 struct xfs_inode *ip)
44{
45 struct xfs_mount *mp = ip->i_mount;
46 struct xfs_dinode *dip;
47 xfs_failaddr_t fa;
48
49 dip = kzalloc(mp->m_sb.sb_inodesize, GFP_KERNEL | GFP_NOFS);
50 if (!dip) {
51 ASSERT(dip != NULL);
52 return;
53 }
54
55 xfs_inode_to_disk(ip, dip, 0);
56 xfs_dinode_calc_crc(mp, dip);
57 fa = xfs_dinode_verify(mp, ip->i_ino, dip);
58 if (fa) {
59 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
60 sizeof(*dip), fa);
61 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
62 ASSERT(fa == NULL);
63 }
64 kfree(dip);
65}
66#else
67# define xfs_inode_item_precommit_check(ip) ((void)0)
68#endif
69
70/*
71 * Prior to finally logging the inode, we have to ensure that all the
72 * per-modification inode state changes are applied. This includes VFS inode
73 * state updates, format conversions, verifier state synchronisation and
74 * ensuring the inode buffer remains in memory whilst the inode is dirty.
75 *
76 * We have to be careful when we grab the inode cluster buffer due to lock
77 * ordering constraints. The unlinked inode modifications (xfs_iunlink_item)
78 * require AGI -> inode cluster buffer lock order. The inode cluster buffer is
79 * not locked until ->precommit, so it happens after everything else has been
80 * modified.
81 *
82 * Further, we have AGI -> AGF lock ordering, and with O_TMPFILE handling we
83 * have AGI -> AGF -> iunlink item -> inode cluster buffer lock order. Hence we
84 * cannot safely lock the inode cluster buffer in xfs_trans_log_inode() because
85 * it can be called on a inode (e.g. via bumplink/droplink) before we take the
86 * AGF lock modifying directory blocks.
87 *
88 * Rather than force a complete rework of all the transactions to call
89 * xfs_trans_log_inode() once and once only at the end of every transaction, we
90 * move the pinning of the inode cluster buffer to a ->precommit operation. This
91 * matches how the xfs_iunlink_item locks the inode cluster buffer, and it
92 * ensures that the inode cluster buffer locking is always done last in a
93 * transaction. i.e. we ensure the lock order is always AGI -> AGF -> inode
94 * cluster buffer.
95 *
96 * If we return the inode number as the precommit sort key then we'll also
97 * guarantee that the order all inode cluster buffer locking is the same all the
98 * inodes and unlink items in the transaction.
99 */
100static int
101xfs_inode_item_precommit(
102 struct xfs_trans *tp,
103 struct xfs_log_item *lip)
104{
105 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
106 struct xfs_inode *ip = iip->ili_inode;
107 struct inode *inode = VFS_I(ip);
108 unsigned int flags = iip->ili_dirty_flags;
109
110 /*
111 * Don't bother with i_lock for the I_DIRTY_TIME check here, as races
112 * don't matter - we either will need an extra transaction in 24 hours
113 * to log the timestamps, or will clear already cleared fields in the
114 * worst case.
115 */
116 if (inode->i_state & I_DIRTY_TIME) {
117 spin_lock(&inode->i_lock);
118 inode->i_state &= ~I_DIRTY_TIME;
119 spin_unlock(&inode->i_lock);
120 }
121
122 /*
123 * If we're updating the inode core or the timestamps and it's possible
124 * to upgrade this inode to bigtime format, do so now.
125 */
126 if ((flags & (XFS_ILOG_CORE | XFS_ILOG_TIMESTAMP)) &&
127 xfs_has_bigtime(ip->i_mount) &&
128 !xfs_inode_has_bigtime(ip)) {
129 ip->i_diflags2 |= XFS_DIFLAG2_BIGTIME;
130 flags |= XFS_ILOG_CORE;
131 }
132
133 /*
134 * Inode verifiers do not check that the extent size hint is an integer
135 * multiple of the rt extent size on a directory with both rtinherit
136 * and extszinherit flags set. If we're logging a directory that is
137 * misconfigured in this way, clear the hint.
138 */
139 if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
140 (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&
141 xfs_extlen_to_rtxmod(ip->i_mount, ip->i_extsize) > 0) {
142 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
143 XFS_DIFLAG_EXTSZINHERIT);
144 ip->i_extsize = 0;
145 flags |= XFS_ILOG_CORE;
146 }
147
148 /*
149 * Record the specific change for fdatasync optimisation. This allows
150 * fdatasync to skip log forces for inodes that are only timestamp
151 * dirty. Once we've processed the XFS_ILOG_IVERSION flag, convert it
152 * to XFS_ILOG_CORE so that the actual on-disk dirty tracking
153 * (ili_fields) correctly tracks that the version has changed.
154 */
155 spin_lock(&iip->ili_lock);
156 iip->ili_fsync_fields |= (flags & ~XFS_ILOG_IVERSION);
157 if (flags & XFS_ILOG_IVERSION)
158 flags = ((flags & ~XFS_ILOG_IVERSION) | XFS_ILOG_CORE);
159
160 if (!iip->ili_item.li_buf) {
161 struct xfs_buf *bp;
162 int error;
163
164 /*
165 * We hold the ILOCK here, so this inode is not going to be
166 * flushed while we are here. Further, because there is no
167 * buffer attached to the item, we know that there is no IO in
168 * progress, so nothing will clear the ili_fields while we read
169 * in the buffer. Hence we can safely drop the spin lock and
170 * read the buffer knowing that the state will not change from
171 * here.
172 */
173 spin_unlock(&iip->ili_lock);
174 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &bp);
175 if (error)
176 return error;
177
178 /*
179 * We need an explicit buffer reference for the log item but
180 * don't want the buffer to remain attached to the transaction.
181 * Hold the buffer but release the transaction reference once
182 * we've attached the inode log item to the buffer log item
183 * list.
184 */
185 xfs_buf_hold(bp);
186 spin_lock(&iip->ili_lock);
187 iip->ili_item.li_buf = bp;
188 bp->b_flags |= _XBF_INODES;
189 list_add_tail(&iip->ili_item.li_bio_list, &bp->b_li_list);
190 xfs_trans_brelse(tp, bp);
191 }
192
193 /*
194 * Always OR in the bits from the ili_last_fields field. This is to
195 * coordinate with the xfs_iflush() and xfs_buf_inode_iodone() routines
196 * in the eventual clearing of the ili_fields bits. See the big comment
197 * in xfs_iflush() for an explanation of this coordination mechanism.
198 */
199 iip->ili_fields |= (flags | iip->ili_last_fields);
200 spin_unlock(&iip->ili_lock);
201
202 xfs_inode_item_precommit_check(ip);
203
204 /*
205 * We are done with the log item transaction dirty state, so clear it so
206 * that it doesn't pollute future transactions.
207 */
208 iip->ili_dirty_flags = 0;
209 return 0;
210}
211
212/*
213 * The logged size of an inode fork is always the current size of the inode
214 * fork. This means that when an inode fork is relogged, the size of the logged
215 * region is determined by the current state, not the combination of the
216 * previously logged state + the current state. This is different relogging
217 * behaviour to most other log items which will retain the size of the
218 * previously logged changes when smaller regions are relogged.
219 *
220 * Hence operations that remove data from the inode fork (e.g. shortform
221 * dir/attr remove, extent form extent removal, etc), the size of the relogged
222 * inode gets -smaller- rather than stays the same size as the previously logged
223 * size and this can result in the committing transaction reducing the amount of
224 * space being consumed by the CIL.
225 */
226STATIC void
227xfs_inode_item_data_fork_size(
228 struct xfs_inode_log_item *iip,
229 int *nvecs,
230 int *nbytes)
231{
232 struct xfs_inode *ip = iip->ili_inode;
233
234 switch (ip->i_df.if_format) {
235 case XFS_DINODE_FMT_EXTENTS:
236 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
237 ip->i_df.if_nextents > 0 &&
238 ip->i_df.if_bytes > 0) {
239 /* worst case, doesn't subtract delalloc extents */
240 *nbytes += xfs_inode_data_fork_size(ip);
241 *nvecs += 1;
242 }
243 break;
244 case XFS_DINODE_FMT_BTREE:
245 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
246 ip->i_df.if_broot_bytes > 0) {
247 *nbytes += ip->i_df.if_broot_bytes;
248 *nvecs += 1;
249 }
250 break;
251 case XFS_DINODE_FMT_LOCAL:
252 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
253 ip->i_df.if_bytes > 0) {
254 *nbytes += xlog_calc_iovec_len(ip->i_df.if_bytes);
255 *nvecs += 1;
256 }
257 break;
258
259 case XFS_DINODE_FMT_DEV:
260 break;
261 default:
262 ASSERT(0);
263 break;
264 }
265}
266
267STATIC void
268xfs_inode_item_attr_fork_size(
269 struct xfs_inode_log_item *iip,
270 int *nvecs,
271 int *nbytes)
272{
273 struct xfs_inode *ip = iip->ili_inode;
274
275 switch (ip->i_af.if_format) {
276 case XFS_DINODE_FMT_EXTENTS:
277 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
278 ip->i_af.if_nextents > 0 &&
279 ip->i_af.if_bytes > 0) {
280 /* worst case, doesn't subtract unused space */
281 *nbytes += xfs_inode_attr_fork_size(ip);
282 *nvecs += 1;
283 }
284 break;
285 case XFS_DINODE_FMT_BTREE:
286 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
287 ip->i_af.if_broot_bytes > 0) {
288 *nbytes += ip->i_af.if_broot_bytes;
289 *nvecs += 1;
290 }
291 break;
292 case XFS_DINODE_FMT_LOCAL:
293 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
294 ip->i_af.if_bytes > 0) {
295 *nbytes += xlog_calc_iovec_len(ip->i_af.if_bytes);
296 *nvecs += 1;
297 }
298 break;
299 default:
300 ASSERT(0);
301 break;
302 }
303}
304
305/*
306 * This returns the number of iovecs needed to log the given inode item.
307 *
308 * We need one iovec for the inode log format structure, one for the
309 * inode core, and possibly one for the inode data/extents/b-tree root
310 * and one for the inode attribute data/extents/b-tree root.
311 */
312STATIC void
313xfs_inode_item_size(
314 struct xfs_log_item *lip,
315 int *nvecs,
316 int *nbytes)
317{
318 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
319 struct xfs_inode *ip = iip->ili_inode;
320
321 *nvecs += 2;
322 *nbytes += sizeof(struct xfs_inode_log_format) +
323 xfs_log_dinode_size(ip->i_mount);
324
325 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
326 if (xfs_inode_has_attr_fork(ip))
327 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
328}
329
330STATIC void
331xfs_inode_item_format_data_fork(
332 struct xfs_inode_log_item *iip,
333 struct xfs_inode_log_format *ilf,
334 struct xfs_log_vec *lv,
335 struct xfs_log_iovec **vecp)
336{
337 struct xfs_inode *ip = iip->ili_inode;
338 size_t data_bytes;
339
340 switch (ip->i_df.if_format) {
341 case XFS_DINODE_FMT_EXTENTS:
342 iip->ili_fields &=
343 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
344
345 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
346 ip->i_df.if_nextents > 0 &&
347 ip->i_df.if_bytes > 0) {
348 struct xfs_bmbt_rec *p;
349
350 ASSERT(xfs_iext_count(&ip->i_df) > 0);
351
352 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
353 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
354 xlog_finish_iovec(lv, *vecp, data_bytes);
355
356 ASSERT(data_bytes <= ip->i_df.if_bytes);
357
358 ilf->ilf_dsize = data_bytes;
359 ilf->ilf_size++;
360 } else {
361 iip->ili_fields &= ~XFS_ILOG_DEXT;
362 }
363 break;
364 case XFS_DINODE_FMT_BTREE:
365 iip->ili_fields &=
366 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
367
368 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
369 ip->i_df.if_broot_bytes > 0) {
370 ASSERT(ip->i_df.if_broot != NULL);
371 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
372 ip->i_df.if_broot,
373 ip->i_df.if_broot_bytes);
374 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
375 ilf->ilf_size++;
376 } else {
377 ASSERT(!(iip->ili_fields &
378 XFS_ILOG_DBROOT));
379 iip->ili_fields &= ~XFS_ILOG_DBROOT;
380 }
381 break;
382 case XFS_DINODE_FMT_LOCAL:
383 iip->ili_fields &=
384 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
385 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
386 ip->i_df.if_bytes > 0) {
387 ASSERT(ip->i_df.if_data != NULL);
388 ASSERT(ip->i_disk_size > 0);
389 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
390 ip->i_df.if_data, ip->i_df.if_bytes);
391 ilf->ilf_dsize = (unsigned)ip->i_df.if_bytes;
392 ilf->ilf_size++;
393 } else {
394 iip->ili_fields &= ~XFS_ILOG_DDATA;
395 }
396 break;
397 case XFS_DINODE_FMT_DEV:
398 iip->ili_fields &=
399 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
400 if (iip->ili_fields & XFS_ILOG_DEV)
401 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
402 break;
403 default:
404 ASSERT(0);
405 break;
406 }
407}
408
409STATIC void
410xfs_inode_item_format_attr_fork(
411 struct xfs_inode_log_item *iip,
412 struct xfs_inode_log_format *ilf,
413 struct xfs_log_vec *lv,
414 struct xfs_log_iovec **vecp)
415{
416 struct xfs_inode *ip = iip->ili_inode;
417 size_t data_bytes;
418
419 switch (ip->i_af.if_format) {
420 case XFS_DINODE_FMT_EXTENTS:
421 iip->ili_fields &=
422 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
423
424 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
425 ip->i_af.if_nextents > 0 &&
426 ip->i_af.if_bytes > 0) {
427 struct xfs_bmbt_rec *p;
428
429 ASSERT(xfs_iext_count(&ip->i_af) ==
430 ip->i_af.if_nextents);
431
432 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
433 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
434 xlog_finish_iovec(lv, *vecp, data_bytes);
435
436 ilf->ilf_asize = data_bytes;
437 ilf->ilf_size++;
438 } else {
439 iip->ili_fields &= ~XFS_ILOG_AEXT;
440 }
441 break;
442 case XFS_DINODE_FMT_BTREE:
443 iip->ili_fields &=
444 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
445
446 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
447 ip->i_af.if_broot_bytes > 0) {
448 ASSERT(ip->i_af.if_broot != NULL);
449
450 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
451 ip->i_af.if_broot,
452 ip->i_af.if_broot_bytes);
453 ilf->ilf_asize = ip->i_af.if_broot_bytes;
454 ilf->ilf_size++;
455 } else {
456 iip->ili_fields &= ~XFS_ILOG_ABROOT;
457 }
458 break;
459 case XFS_DINODE_FMT_LOCAL:
460 iip->ili_fields &=
461 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
462
463 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
464 ip->i_af.if_bytes > 0) {
465 ASSERT(ip->i_af.if_data != NULL);
466 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
467 ip->i_af.if_data, ip->i_af.if_bytes);
468 ilf->ilf_asize = (unsigned)ip->i_af.if_bytes;
469 ilf->ilf_size++;
470 } else {
471 iip->ili_fields &= ~XFS_ILOG_ADATA;
472 }
473 break;
474 default:
475 ASSERT(0);
476 break;
477 }
478}
479
480/*
481 * Convert an incore timestamp to a log timestamp. Note that the log format
482 * specifies host endian format!
483 */
484static inline xfs_log_timestamp_t
485xfs_inode_to_log_dinode_ts(
486 struct xfs_inode *ip,
487 const struct timespec64 tv)
488{
489 struct xfs_log_legacy_timestamp *lits;
490 xfs_log_timestamp_t its;
491
492 if (xfs_inode_has_bigtime(ip))
493 return xfs_inode_encode_bigtime(tv);
494
495 lits = (struct xfs_log_legacy_timestamp *)&its;
496 lits->t_sec = tv.tv_sec;
497 lits->t_nsec = tv.tv_nsec;
498
499 return its;
500}
501
502/*
503 * The legacy DMAPI fields are only present in the on-disk and in-log inodes,
504 * but not in the in-memory one. But we are guaranteed to have an inode buffer
505 * in memory when logging an inode, so we can just copy it from the on-disk
506 * inode to the in-log inode here so that recovery of file system with these
507 * fields set to non-zero values doesn't lose them. For all other cases we zero
508 * the fields.
509 */
510static void
511xfs_copy_dm_fields_to_log_dinode(
512 struct xfs_inode *ip,
513 struct xfs_log_dinode *to)
514{
515 struct xfs_dinode *dip;
516
517 dip = xfs_buf_offset(ip->i_itemp->ili_item.li_buf,
518 ip->i_imap.im_boffset);
519
520 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) {
521 to->di_dmevmask = be32_to_cpu(dip->di_dmevmask);
522 to->di_dmstate = be16_to_cpu(dip->di_dmstate);
523 } else {
524 to->di_dmevmask = 0;
525 to->di_dmstate = 0;
526 }
527}
528
529static inline void
530xfs_inode_to_log_dinode_iext_counters(
531 struct xfs_inode *ip,
532 struct xfs_log_dinode *to)
533{
534 if (xfs_inode_has_large_extent_counts(ip)) {
535 to->di_big_nextents = xfs_ifork_nextents(&ip->i_df);
536 to->di_big_anextents = xfs_ifork_nextents(&ip->i_af);
537 to->di_nrext64_pad = 0;
538 } else {
539 to->di_nextents = xfs_ifork_nextents(&ip->i_df);
540 to->di_anextents = xfs_ifork_nextents(&ip->i_af);
541 }
542}
543
544static void
545xfs_inode_to_log_dinode(
546 struct xfs_inode *ip,
547 struct xfs_log_dinode *to,
548 xfs_lsn_t lsn)
549{
550 struct inode *inode = VFS_I(ip);
551
552 to->di_magic = XFS_DINODE_MAGIC;
553 to->di_format = xfs_ifork_format(&ip->i_df);
554 to->di_uid = i_uid_read(inode);
555 to->di_gid = i_gid_read(inode);
556 to->di_projid_lo = ip->i_projid & 0xffff;
557 to->di_projid_hi = ip->i_projid >> 16;
558
559 to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode_get_atime(inode));
560 to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode_get_mtime(inode));
561 to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode_get_ctime(inode));
562 to->di_nlink = inode->i_nlink;
563 to->di_gen = inode->i_generation;
564 to->di_mode = inode->i_mode;
565
566 to->di_size = ip->i_disk_size;
567 to->di_nblocks = ip->i_nblocks;
568 to->di_extsize = ip->i_extsize;
569 to->di_forkoff = ip->i_forkoff;
570 to->di_aformat = xfs_ifork_format(&ip->i_af);
571 to->di_flags = ip->i_diflags;
572
573 xfs_copy_dm_fields_to_log_dinode(ip, to);
574
575 /* log a dummy value to ensure log structure is fully initialised */
576 to->di_next_unlinked = NULLAGINO;
577
578 if (xfs_has_v3inodes(ip->i_mount)) {
579 to->di_version = 3;
580 to->di_changecount = inode_peek_iversion(inode);
581 to->di_crtime = xfs_inode_to_log_dinode_ts(ip, ip->i_crtime);
582 to->di_flags2 = ip->i_diflags2;
583 to->di_cowextsize = ip->i_cowextsize;
584 to->di_ino = ip->i_ino;
585 to->di_lsn = lsn;
586 memset(to->di_pad2, 0, sizeof(to->di_pad2));
587 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
588 to->di_v3_pad = 0;
589
590 /* dummy value for initialisation */
591 to->di_crc = 0;
592
593 if (xfs_is_metadir_inode(ip))
594 to->di_metatype = ip->i_metatype;
595 else
596 to->di_metatype = 0;
597 } else {
598 to->di_version = 2;
599 to->di_flushiter = ip->i_flushiter;
600 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
601 to->di_metatype = 0;
602 }
603
604 xfs_inode_to_log_dinode_iext_counters(ip, to);
605}
606
607/*
608 * Format the inode core. Current timestamp data is only in the VFS inode
609 * fields, so we need to grab them from there. Hence rather than just copying
610 * the XFS inode core structure, format the fields directly into the iovec.
611 */
612static void
613xfs_inode_item_format_core(
614 struct xfs_inode *ip,
615 struct xfs_log_vec *lv,
616 struct xfs_log_iovec **vecp)
617{
618 struct xfs_log_dinode *dic;
619
620 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
621 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
622 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
623}
624
625/*
626 * This is called to fill in the vector of log iovecs for the given inode
627 * log item. It fills the first item with an inode log format structure,
628 * the second with the on-disk inode structure, and a possible third and/or
629 * fourth with the inode data/extents/b-tree root and inode attributes
630 * data/extents/b-tree root.
631 *
632 * Note: Always use the 64 bit inode log format structure so we don't
633 * leave an uninitialised hole in the format item on 64 bit systems. Log
634 * recovery on 32 bit systems handles this just fine, so there's no reason
635 * for not using an initialising the properly padded structure all the time.
636 */
637STATIC void
638xfs_inode_item_format(
639 struct xfs_log_item *lip,
640 struct xfs_log_vec *lv)
641{
642 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
643 struct xfs_inode *ip = iip->ili_inode;
644 struct xfs_log_iovec *vecp = NULL;
645 struct xfs_inode_log_format *ilf;
646
647 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
648 ilf->ilf_type = XFS_LI_INODE;
649 ilf->ilf_ino = ip->i_ino;
650 ilf->ilf_blkno = ip->i_imap.im_blkno;
651 ilf->ilf_len = ip->i_imap.im_len;
652 ilf->ilf_boffset = ip->i_imap.im_boffset;
653 ilf->ilf_fields = XFS_ILOG_CORE;
654 ilf->ilf_size = 2; /* format + core */
655
656 /*
657 * make sure we don't leak uninitialised data into the log in the case
658 * when we don't log every field in the inode.
659 */
660 ilf->ilf_dsize = 0;
661 ilf->ilf_asize = 0;
662 ilf->ilf_pad = 0;
663 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
664
665 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
666
667 xfs_inode_item_format_core(ip, lv, &vecp);
668 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
669 if (xfs_inode_has_attr_fork(ip)) {
670 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
671 } else {
672 iip->ili_fields &=
673 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
674 }
675
676 /* update the format with the exact fields we actually logged */
677 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
678}
679
680/*
681 * This is called to pin the inode associated with the inode log
682 * item in memory so it cannot be written out.
683 */
684STATIC void
685xfs_inode_item_pin(
686 struct xfs_log_item *lip)
687{
688 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
689
690 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
691 ASSERT(lip->li_buf);
692
693 trace_xfs_inode_pin(ip, _RET_IP_);
694 atomic_inc(&ip->i_pincount);
695}
696
697
698/*
699 * This is called to unpin the inode associated with the inode log
700 * item which was previously pinned with a call to xfs_inode_item_pin().
701 *
702 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
703 *
704 * Note that unpin can race with inode cluster buffer freeing marking the buffer
705 * stale. In that case, flush completions are run from the buffer unpin call,
706 * which may happen before the inode is unpinned. If we lose the race, there
707 * will be no buffer attached to the log item, but the inode will be marked
708 * XFS_ISTALE.
709 */
710STATIC void
711xfs_inode_item_unpin(
712 struct xfs_log_item *lip,
713 int remove)
714{
715 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
716
717 trace_xfs_inode_unpin(ip, _RET_IP_);
718 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
719 ASSERT(atomic_read(&ip->i_pincount) > 0);
720 if (atomic_dec_and_test(&ip->i_pincount))
721 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
722}
723
724STATIC uint
725xfs_inode_item_push(
726 struct xfs_log_item *lip,
727 struct list_head *buffer_list)
728 __releases(&lip->li_ailp->ail_lock)
729 __acquires(&lip->li_ailp->ail_lock)
730{
731 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
732 struct xfs_inode *ip = iip->ili_inode;
733 struct xfs_buf *bp = lip->li_buf;
734 uint rval = XFS_ITEM_SUCCESS;
735 int error;
736
737 if (!bp || (ip->i_flags & XFS_ISTALE)) {
738 /*
739 * Inode item/buffer is being aborted due to cluster
740 * buffer deletion. Trigger a log force to have that operation
741 * completed and items removed from the AIL before the next push
742 * attempt.
743 */
744 return XFS_ITEM_PINNED;
745 }
746
747 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp))
748 return XFS_ITEM_PINNED;
749
750 if (xfs_iflags_test(ip, XFS_IFLUSHING))
751 return XFS_ITEM_FLUSHING;
752
753 if (!xfs_buf_trylock(bp))
754 return XFS_ITEM_LOCKED;
755
756 spin_unlock(&lip->li_ailp->ail_lock);
757
758 /*
759 * We need to hold a reference for flushing the cluster buffer as it may
760 * fail the buffer without IO submission. In which case, we better get a
761 * reference for that completion because otherwise we don't get a
762 * reference for IO until we queue the buffer for delwri submission.
763 */
764 xfs_buf_hold(bp);
765 error = xfs_iflush_cluster(bp);
766 if (!error) {
767 if (!xfs_buf_delwri_queue(bp, buffer_list))
768 rval = XFS_ITEM_FLUSHING;
769 xfs_buf_relse(bp);
770 } else {
771 /*
772 * Release the buffer if we were unable to flush anything. On
773 * any other error, the buffer has already been released.
774 */
775 if (error == -EAGAIN)
776 xfs_buf_relse(bp);
777 rval = XFS_ITEM_LOCKED;
778 }
779
780 spin_lock(&lip->li_ailp->ail_lock);
781 return rval;
782}
783
784/*
785 * Unlock the inode associated with the inode log item.
786 */
787STATIC void
788xfs_inode_item_release(
789 struct xfs_log_item *lip)
790{
791 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
792 struct xfs_inode *ip = iip->ili_inode;
793 unsigned short lock_flags;
794
795 ASSERT(ip->i_itemp != NULL);
796 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
797
798 lock_flags = iip->ili_lock_flags;
799 iip->ili_lock_flags = 0;
800 if (lock_flags)
801 xfs_iunlock(ip, lock_flags);
802}
803
804/*
805 * This is called to find out where the oldest active copy of the inode log
806 * item in the on disk log resides now that the last log write of it completed
807 * at the given lsn. Since we always re-log all dirty data in an inode, the
808 * latest copy in the on disk log is the only one that matters. Therefore,
809 * simply return the given lsn.
810 *
811 * If the inode has been marked stale because the cluster is being freed, we
812 * don't want to (re-)insert this inode into the AIL. There is a race condition
813 * where the cluster buffer may be unpinned before the inode is inserted into
814 * the AIL during transaction committed processing. If the buffer is unpinned
815 * before the inode item has been committed and inserted, then it is possible
816 * for the buffer to be written and IO completes before the inode is inserted
817 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
818 * AIL which will never get removed. It will, however, get reclaimed which
819 * triggers an assert in xfs_inode_free() complaining about freein an inode
820 * still in the AIL.
821 *
822 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
823 * transaction committed code knows that it does not need to do any further
824 * processing on the item.
825 */
826STATIC xfs_lsn_t
827xfs_inode_item_committed(
828 struct xfs_log_item *lip,
829 xfs_lsn_t lsn)
830{
831 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
832 struct xfs_inode *ip = iip->ili_inode;
833
834 if (xfs_iflags_test(ip, XFS_ISTALE)) {
835 xfs_inode_item_unpin(lip, 0);
836 return -1;
837 }
838 return lsn;
839}
840
841STATIC void
842xfs_inode_item_committing(
843 struct xfs_log_item *lip,
844 xfs_csn_t seq)
845{
846 INODE_ITEM(lip)->ili_commit_seq = seq;
847 return xfs_inode_item_release(lip);
848}
849
850static const struct xfs_item_ops xfs_inode_item_ops = {
851 .iop_sort = xfs_inode_item_sort,
852 .iop_precommit = xfs_inode_item_precommit,
853 .iop_size = xfs_inode_item_size,
854 .iop_format = xfs_inode_item_format,
855 .iop_pin = xfs_inode_item_pin,
856 .iop_unpin = xfs_inode_item_unpin,
857 .iop_release = xfs_inode_item_release,
858 .iop_committed = xfs_inode_item_committed,
859 .iop_push = xfs_inode_item_push,
860 .iop_committing = xfs_inode_item_committing,
861};
862
863
864/*
865 * Initialize the inode log item for a newly allocated (in-core) inode.
866 */
867void
868xfs_inode_item_init(
869 struct xfs_inode *ip,
870 struct xfs_mount *mp)
871{
872 struct xfs_inode_log_item *iip;
873
874 ASSERT(ip->i_itemp == NULL);
875 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_cache,
876 GFP_KERNEL | __GFP_NOFAIL);
877
878 iip->ili_inode = ip;
879 spin_lock_init(&iip->ili_lock);
880 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
881 &xfs_inode_item_ops);
882}
883
884/*
885 * Free the inode log item and any memory hanging off of it.
886 */
887void
888xfs_inode_item_destroy(
889 struct xfs_inode *ip)
890{
891 struct xfs_inode_log_item *iip = ip->i_itemp;
892
893 ASSERT(iip->ili_item.li_buf == NULL);
894
895 ip->i_itemp = NULL;
896 kvfree(iip->ili_item.li_lv_shadow);
897 kmem_cache_free(xfs_ili_cache, iip);
898}
899
900
901/*
902 * We only want to pull the item from the AIL if it is actually there
903 * and its location in the log has not changed since we started the
904 * flush. Thus, we only bother if the inode's lsn has not changed.
905 */
906static void
907xfs_iflush_ail_updates(
908 struct xfs_ail *ailp,
909 struct list_head *list)
910{
911 struct xfs_log_item *lip;
912 xfs_lsn_t tail_lsn = 0;
913
914 /* this is an opencoded batch version of xfs_trans_ail_delete */
915 spin_lock(&ailp->ail_lock);
916 list_for_each_entry(lip, list, li_bio_list) {
917 xfs_lsn_t lsn;
918
919 clear_bit(XFS_LI_FAILED, &lip->li_flags);
920 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
921 continue;
922
923 /*
924 * dgc: Not sure how this happens, but it happens very
925 * occassionaly via generic/388. xfs_iflush_abort() also
926 * silently handles this same "under writeback but not in AIL at
927 * shutdown" condition via xfs_trans_ail_delete().
928 */
929 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
930 ASSERT(xlog_is_shutdown(lip->li_log));
931 continue;
932 }
933
934 lsn = xfs_ail_delete_one(ailp, lip);
935 if (!tail_lsn && lsn)
936 tail_lsn = lsn;
937 }
938 xfs_ail_update_finish(ailp, tail_lsn);
939}
940
941/*
942 * Walk the list of inodes that have completed their IOs. If they are clean
943 * remove them from the list and dissociate them from the buffer. Buffers that
944 * are still dirty remain linked to the buffer and on the list. Caller must
945 * handle them appropriately.
946 */
947static void
948xfs_iflush_finish(
949 struct xfs_buf *bp,
950 struct list_head *list)
951{
952 struct xfs_log_item *lip, *n;
953
954 list_for_each_entry_safe(lip, n, list, li_bio_list) {
955 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
956 bool drop_buffer = false;
957
958 spin_lock(&iip->ili_lock);
959
960 /*
961 * Remove the reference to the cluster buffer if the inode is
962 * clean in memory and drop the buffer reference once we've
963 * dropped the locks we hold.
964 */
965 ASSERT(iip->ili_item.li_buf == bp);
966 if (!iip->ili_fields) {
967 iip->ili_item.li_buf = NULL;
968 list_del_init(&lip->li_bio_list);
969 drop_buffer = true;
970 }
971 iip->ili_last_fields = 0;
972 iip->ili_flush_lsn = 0;
973 clear_bit(XFS_LI_FLUSHING, &lip->li_flags);
974 spin_unlock(&iip->ili_lock);
975 xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);
976 if (drop_buffer)
977 xfs_buf_rele(bp);
978 }
979}
980
981/*
982 * Inode buffer IO completion routine. It is responsible for removing inodes
983 * attached to the buffer from the AIL if they have not been re-logged and
984 * completing the inode flush.
985 */
986void
987xfs_buf_inode_iodone(
988 struct xfs_buf *bp)
989{
990 struct xfs_log_item *lip, *n;
991 LIST_HEAD(flushed_inodes);
992 LIST_HEAD(ail_updates);
993
994 /*
995 * Pull the attached inodes from the buffer one at a time and take the
996 * appropriate action on them.
997 */
998 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
999 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
1000
1001 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
1002 xfs_iflush_abort(iip->ili_inode);
1003 continue;
1004 }
1005 if (!iip->ili_last_fields)
1006 continue;
1007
1008 /* Do an unlocked check for needing the AIL lock. */
1009 if (iip->ili_flush_lsn == lip->li_lsn ||
1010 test_bit(XFS_LI_FAILED, &lip->li_flags))
1011 list_move_tail(&lip->li_bio_list, &ail_updates);
1012 else
1013 list_move_tail(&lip->li_bio_list, &flushed_inodes);
1014 }
1015
1016 if (!list_empty(&ail_updates)) {
1017 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
1018 list_splice_tail(&ail_updates, &flushed_inodes);
1019 }
1020
1021 xfs_iflush_finish(bp, &flushed_inodes);
1022 if (!list_empty(&flushed_inodes))
1023 list_splice_tail(&flushed_inodes, &bp->b_li_list);
1024}
1025
1026void
1027xfs_buf_inode_io_fail(
1028 struct xfs_buf *bp)
1029{
1030 struct xfs_log_item *lip;
1031
1032 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
1033 set_bit(XFS_LI_FAILED, &lip->li_flags);
1034 clear_bit(XFS_LI_FLUSHING, &lip->li_flags);
1035 }
1036}
1037
1038/*
1039 * Clear the inode logging fields so no more flushes are attempted. If we are
1040 * on a buffer list, it is now safe to remove it because the buffer is
1041 * guaranteed to be locked. The caller will drop the reference to the buffer
1042 * the log item held.
1043 */
1044static void
1045xfs_iflush_abort_clean(
1046 struct xfs_inode_log_item *iip)
1047{
1048 iip->ili_last_fields = 0;
1049 iip->ili_fields = 0;
1050 iip->ili_fsync_fields = 0;
1051 iip->ili_flush_lsn = 0;
1052 iip->ili_item.li_buf = NULL;
1053 list_del_init(&iip->ili_item.li_bio_list);
1054 clear_bit(XFS_LI_FLUSHING, &iip->ili_item.li_flags);
1055}
1056
1057/*
1058 * Abort flushing the inode from a context holding the cluster buffer locked.
1059 *
1060 * This is the normal runtime method of aborting writeback of an inode that is
1061 * attached to a cluster buffer. It occurs when the inode and the backing
1062 * cluster buffer have been freed (i.e. inode is XFS_ISTALE), or when cluster
1063 * flushing or buffer IO completion encounters a log shutdown situation.
1064 *
1065 * If we need to abort inode writeback and we don't already hold the buffer
1066 * locked, call xfs_iflush_shutdown_abort() instead as this should only ever be
1067 * necessary in a shutdown situation.
1068 */
1069void
1070xfs_iflush_abort(
1071 struct xfs_inode *ip)
1072{
1073 struct xfs_inode_log_item *iip = ip->i_itemp;
1074 struct xfs_buf *bp;
1075
1076 if (!iip) {
1077 /* clean inode, nothing to do */
1078 xfs_iflags_clear(ip, XFS_IFLUSHING);
1079 return;
1080 }
1081
1082 /*
1083 * Remove the inode item from the AIL before we clear its internal
1084 * state. Whilst the inode is in the AIL, it should have a valid buffer
1085 * pointer for push operations to access - it is only safe to remove the
1086 * inode from the buffer once it has been removed from the AIL.
1087 *
1088 * We also clear the failed bit before removing the item from the AIL
1089 * as xfs_trans_ail_delete()->xfs_clear_li_failed() will release buffer
1090 * references the inode item owns and needs to hold until we've fully
1091 * aborted the inode log item and detached it from the buffer.
1092 */
1093 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
1094 xfs_trans_ail_delete(&iip->ili_item, 0);
1095
1096 /*
1097 * Grab the inode buffer so can we release the reference the inode log
1098 * item holds on it.
1099 */
1100 spin_lock(&iip->ili_lock);
1101 bp = iip->ili_item.li_buf;
1102 xfs_iflush_abort_clean(iip);
1103 spin_unlock(&iip->ili_lock);
1104
1105 xfs_iflags_clear(ip, XFS_IFLUSHING);
1106 if (bp)
1107 xfs_buf_rele(bp);
1108}
1109
1110/*
1111 * Abort an inode flush in the case of a shutdown filesystem. This can be called
1112 * from anywhere with just an inode reference and does not require holding the
1113 * inode cluster buffer locked. If the inode is attached to a cluster buffer,
1114 * it will grab and lock it safely, then abort the inode flush.
1115 */
1116void
1117xfs_iflush_shutdown_abort(
1118 struct xfs_inode *ip)
1119{
1120 struct xfs_inode_log_item *iip = ip->i_itemp;
1121 struct xfs_buf *bp;
1122
1123 if (!iip) {
1124 /* clean inode, nothing to do */
1125 xfs_iflags_clear(ip, XFS_IFLUSHING);
1126 return;
1127 }
1128
1129 spin_lock(&iip->ili_lock);
1130 bp = iip->ili_item.li_buf;
1131 if (!bp) {
1132 spin_unlock(&iip->ili_lock);
1133 xfs_iflush_abort(ip);
1134 return;
1135 }
1136
1137 /*
1138 * We have to take a reference to the buffer so that it doesn't get
1139 * freed when we drop the ili_lock and then wait to lock the buffer.
1140 * We'll clean up the extra reference after we pick up the ili_lock
1141 * again.
1142 */
1143 xfs_buf_hold(bp);
1144 spin_unlock(&iip->ili_lock);
1145 xfs_buf_lock(bp);
1146
1147 spin_lock(&iip->ili_lock);
1148 if (!iip->ili_item.li_buf) {
1149 /*
1150 * Raced with another removal, hold the only reference
1151 * to bp now. Inode should not be in the AIL now, so just clean
1152 * up and return;
1153 */
1154 ASSERT(list_empty(&iip->ili_item.li_bio_list));
1155 ASSERT(!test_bit(XFS_LI_IN_AIL, &iip->ili_item.li_flags));
1156 xfs_iflush_abort_clean(iip);
1157 spin_unlock(&iip->ili_lock);
1158 xfs_iflags_clear(ip, XFS_IFLUSHING);
1159 xfs_buf_relse(bp);
1160 return;
1161 }
1162
1163 /*
1164 * Got two references to bp. The first will get dropped by
1165 * xfs_iflush_abort() when the item is removed from the buffer list, but
1166 * we can't drop our reference until _abort() returns because we have to
1167 * unlock the buffer as well. Hence we abort and then unlock and release
1168 * our reference to the buffer.
1169 */
1170 ASSERT(iip->ili_item.li_buf == bp);
1171 spin_unlock(&iip->ili_lock);
1172 xfs_iflush_abort(ip);
1173 xfs_buf_relse(bp);
1174}
1175
1176
1177/*
1178 * convert an xfs_inode_log_format struct from the old 32 bit version
1179 * (which can have different field alignments) to the native 64 bit version
1180 */
1181int
1182xfs_inode_item_format_convert(
1183 struct xfs_log_iovec *buf,
1184 struct xfs_inode_log_format *in_f)
1185{
1186 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
1187
1188 if (buf->i_len != sizeof(*in_f32)) {
1189 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
1190 return -EFSCORRUPTED;
1191 }
1192
1193 in_f->ilf_type = in_f32->ilf_type;
1194 in_f->ilf_size = in_f32->ilf_size;
1195 in_f->ilf_fields = in_f32->ilf_fields;
1196 in_f->ilf_asize = in_f32->ilf_asize;
1197 in_f->ilf_dsize = in_f32->ilf_dsize;
1198 in_f->ilf_ino = in_f32->ilf_ino;
1199 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
1200 in_f->ilf_blkno = in_f32->ilf_blkno;
1201 in_f->ilf_len = in_f32->ilf_len;
1202 in_f->ilf_boffset = in_f32->ilf_boffset;
1203 return 0;
1204}