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_error.h"
21
22#include <linux/iversion.h>
23
24kmem_zone_t *xfs_ili_zone; /* inode log item zone */
25
26static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
27{
28 return container_of(lip, struct xfs_inode_log_item, ili_item);
29}
30
31STATIC void
32xfs_inode_item_data_fork_size(
33 struct xfs_inode_log_item *iip,
34 int *nvecs,
35 int *nbytes)
36{
37 struct xfs_inode *ip = iip->ili_inode;
38
39 switch (ip->i_df.if_format) {
40 case XFS_DINODE_FMT_EXTENTS:
41 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
42 ip->i_df.if_nextents > 0 &&
43 ip->i_df.if_bytes > 0) {
44 /* worst case, doesn't subtract delalloc extents */
45 *nbytes += XFS_IFORK_DSIZE(ip);
46 *nvecs += 1;
47 }
48 break;
49 case XFS_DINODE_FMT_BTREE:
50 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
51 ip->i_df.if_broot_bytes > 0) {
52 *nbytes += ip->i_df.if_broot_bytes;
53 *nvecs += 1;
54 }
55 break;
56 case XFS_DINODE_FMT_LOCAL:
57 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
58 ip->i_df.if_bytes > 0) {
59 *nbytes += roundup(ip->i_df.if_bytes, 4);
60 *nvecs += 1;
61 }
62 break;
63
64 case XFS_DINODE_FMT_DEV:
65 break;
66 default:
67 ASSERT(0);
68 break;
69 }
70}
71
72STATIC void
73xfs_inode_item_attr_fork_size(
74 struct xfs_inode_log_item *iip,
75 int *nvecs,
76 int *nbytes)
77{
78 struct xfs_inode *ip = iip->ili_inode;
79
80 switch (ip->i_afp->if_format) {
81 case XFS_DINODE_FMT_EXTENTS:
82 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
83 ip->i_afp->if_nextents > 0 &&
84 ip->i_afp->if_bytes > 0) {
85 /* worst case, doesn't subtract unused space */
86 *nbytes += XFS_IFORK_ASIZE(ip);
87 *nvecs += 1;
88 }
89 break;
90 case XFS_DINODE_FMT_BTREE:
91 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
92 ip->i_afp->if_broot_bytes > 0) {
93 *nbytes += ip->i_afp->if_broot_bytes;
94 *nvecs += 1;
95 }
96 break;
97 case XFS_DINODE_FMT_LOCAL:
98 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
99 ip->i_afp->if_bytes > 0) {
100 *nbytes += roundup(ip->i_afp->if_bytes, 4);
101 *nvecs += 1;
102 }
103 break;
104 default:
105 ASSERT(0);
106 break;
107 }
108}
109
110/*
111 * This returns the number of iovecs needed to log the given inode item.
112 *
113 * We need one iovec for the inode log format structure, one for the
114 * inode core, and possibly one for the inode data/extents/b-tree root
115 * and one for the inode attribute data/extents/b-tree root.
116 */
117STATIC void
118xfs_inode_item_size(
119 struct xfs_log_item *lip,
120 int *nvecs,
121 int *nbytes)
122{
123 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
124 struct xfs_inode *ip = iip->ili_inode;
125
126 *nvecs += 2;
127 *nbytes += sizeof(struct xfs_inode_log_format) +
128 xfs_log_dinode_size(ip->i_mount);
129
130 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
131 if (XFS_IFORK_Q(ip))
132 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
133}
134
135STATIC void
136xfs_inode_item_format_data_fork(
137 struct xfs_inode_log_item *iip,
138 struct xfs_inode_log_format *ilf,
139 struct xfs_log_vec *lv,
140 struct xfs_log_iovec **vecp)
141{
142 struct xfs_inode *ip = iip->ili_inode;
143 size_t data_bytes;
144
145 switch (ip->i_df.if_format) {
146 case XFS_DINODE_FMT_EXTENTS:
147 iip->ili_fields &=
148 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
149
150 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
151 ip->i_df.if_nextents > 0 &&
152 ip->i_df.if_bytes > 0) {
153 struct xfs_bmbt_rec *p;
154
155 ASSERT(xfs_iext_count(&ip->i_df) > 0);
156
157 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
158 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
159 xlog_finish_iovec(lv, *vecp, data_bytes);
160
161 ASSERT(data_bytes <= ip->i_df.if_bytes);
162
163 ilf->ilf_dsize = data_bytes;
164 ilf->ilf_size++;
165 } else {
166 iip->ili_fields &= ~XFS_ILOG_DEXT;
167 }
168 break;
169 case XFS_DINODE_FMT_BTREE:
170 iip->ili_fields &=
171 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
172
173 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
174 ip->i_df.if_broot_bytes > 0) {
175 ASSERT(ip->i_df.if_broot != NULL);
176 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
177 ip->i_df.if_broot,
178 ip->i_df.if_broot_bytes);
179 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
180 ilf->ilf_size++;
181 } else {
182 ASSERT(!(iip->ili_fields &
183 XFS_ILOG_DBROOT));
184 iip->ili_fields &= ~XFS_ILOG_DBROOT;
185 }
186 break;
187 case XFS_DINODE_FMT_LOCAL:
188 iip->ili_fields &=
189 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
190 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
191 ip->i_df.if_bytes > 0) {
192 /*
193 * Round i_bytes up to a word boundary.
194 * The underlying memory is guaranteed
195 * to be there by xfs_idata_realloc().
196 */
197 data_bytes = roundup(ip->i_df.if_bytes, 4);
198 ASSERT(ip->i_df.if_u1.if_data != NULL);
199 ASSERT(ip->i_d.di_size > 0);
200 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
201 ip->i_df.if_u1.if_data, data_bytes);
202 ilf->ilf_dsize = (unsigned)data_bytes;
203 ilf->ilf_size++;
204 } else {
205 iip->ili_fields &= ~XFS_ILOG_DDATA;
206 }
207 break;
208 case XFS_DINODE_FMT_DEV:
209 iip->ili_fields &=
210 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
211 if (iip->ili_fields & XFS_ILOG_DEV)
212 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
213 break;
214 default:
215 ASSERT(0);
216 break;
217 }
218}
219
220STATIC void
221xfs_inode_item_format_attr_fork(
222 struct xfs_inode_log_item *iip,
223 struct xfs_inode_log_format *ilf,
224 struct xfs_log_vec *lv,
225 struct xfs_log_iovec **vecp)
226{
227 struct xfs_inode *ip = iip->ili_inode;
228 size_t data_bytes;
229
230 switch (ip->i_afp->if_format) {
231 case XFS_DINODE_FMT_EXTENTS:
232 iip->ili_fields &=
233 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
234
235 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
236 ip->i_afp->if_nextents > 0 &&
237 ip->i_afp->if_bytes > 0) {
238 struct xfs_bmbt_rec *p;
239
240 ASSERT(xfs_iext_count(ip->i_afp) ==
241 ip->i_afp->if_nextents);
242
243 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
244 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
245 xlog_finish_iovec(lv, *vecp, data_bytes);
246
247 ilf->ilf_asize = data_bytes;
248 ilf->ilf_size++;
249 } else {
250 iip->ili_fields &= ~XFS_ILOG_AEXT;
251 }
252 break;
253 case XFS_DINODE_FMT_BTREE:
254 iip->ili_fields &=
255 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
256
257 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
258 ip->i_afp->if_broot_bytes > 0) {
259 ASSERT(ip->i_afp->if_broot != NULL);
260
261 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
262 ip->i_afp->if_broot,
263 ip->i_afp->if_broot_bytes);
264 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
265 ilf->ilf_size++;
266 } else {
267 iip->ili_fields &= ~XFS_ILOG_ABROOT;
268 }
269 break;
270 case XFS_DINODE_FMT_LOCAL:
271 iip->ili_fields &=
272 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
273
274 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
275 ip->i_afp->if_bytes > 0) {
276 /*
277 * Round i_bytes up to a word boundary.
278 * The underlying memory is guaranteed
279 * to be there by xfs_idata_realloc().
280 */
281 data_bytes = roundup(ip->i_afp->if_bytes, 4);
282 ASSERT(ip->i_afp->if_u1.if_data != NULL);
283 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
284 ip->i_afp->if_u1.if_data,
285 data_bytes);
286 ilf->ilf_asize = (unsigned)data_bytes;
287 ilf->ilf_size++;
288 } else {
289 iip->ili_fields &= ~XFS_ILOG_ADATA;
290 }
291 break;
292 default:
293 ASSERT(0);
294 break;
295 }
296}
297
298static void
299xfs_inode_to_log_dinode(
300 struct xfs_inode *ip,
301 struct xfs_log_dinode *to,
302 xfs_lsn_t lsn)
303{
304 struct xfs_icdinode *from = &ip->i_d;
305 struct inode *inode = VFS_I(ip);
306
307 to->di_magic = XFS_DINODE_MAGIC;
308 to->di_format = xfs_ifork_format(&ip->i_df);
309 to->di_uid = i_uid_read(inode);
310 to->di_gid = i_gid_read(inode);
311 to->di_projid_lo = from->di_projid & 0xffff;
312 to->di_projid_hi = from->di_projid >> 16;
313
314 memset(to->di_pad, 0, sizeof(to->di_pad));
315 memset(to->di_pad3, 0, sizeof(to->di_pad3));
316 to->di_atime.t_sec = inode->i_atime.tv_sec;
317 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
318 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
319 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
320 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
321 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
322 to->di_nlink = inode->i_nlink;
323 to->di_gen = inode->i_generation;
324 to->di_mode = inode->i_mode;
325
326 to->di_size = from->di_size;
327 to->di_nblocks = from->di_nblocks;
328 to->di_extsize = from->di_extsize;
329 to->di_nextents = xfs_ifork_nextents(&ip->i_df);
330 to->di_anextents = xfs_ifork_nextents(ip->i_afp);
331 to->di_forkoff = from->di_forkoff;
332 to->di_aformat = xfs_ifork_format(ip->i_afp);
333 to->di_dmevmask = from->di_dmevmask;
334 to->di_dmstate = from->di_dmstate;
335 to->di_flags = from->di_flags;
336
337 /* log a dummy value to ensure log structure is fully initialised */
338 to->di_next_unlinked = NULLAGINO;
339
340 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
341 to->di_version = 3;
342 to->di_changecount = inode_peek_iversion(inode);
343 to->di_crtime.t_sec = from->di_crtime.tv_sec;
344 to->di_crtime.t_nsec = from->di_crtime.tv_nsec;
345 to->di_flags2 = from->di_flags2;
346 to->di_cowextsize = from->di_cowextsize;
347 to->di_ino = ip->i_ino;
348 to->di_lsn = lsn;
349 memset(to->di_pad2, 0, sizeof(to->di_pad2));
350 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
351 to->di_flushiter = 0;
352 } else {
353 to->di_version = 2;
354 to->di_flushiter = from->di_flushiter;
355 }
356}
357
358/*
359 * Format the inode core. Current timestamp data is only in the VFS inode
360 * fields, so we need to grab them from there. Hence rather than just copying
361 * the XFS inode core structure, format the fields directly into the iovec.
362 */
363static void
364xfs_inode_item_format_core(
365 struct xfs_inode *ip,
366 struct xfs_log_vec *lv,
367 struct xfs_log_iovec **vecp)
368{
369 struct xfs_log_dinode *dic;
370
371 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
372 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
373 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
374}
375
376/*
377 * This is called to fill in the vector of log iovecs for the given inode
378 * log item. It fills the first item with an inode log format structure,
379 * the second with the on-disk inode structure, and a possible third and/or
380 * fourth with the inode data/extents/b-tree root and inode attributes
381 * data/extents/b-tree root.
382 *
383 * Note: Always use the 64 bit inode log format structure so we don't
384 * leave an uninitialised hole in the format item on 64 bit systems. Log
385 * recovery on 32 bit systems handles this just fine, so there's no reason
386 * for not using an initialising the properly padded structure all the time.
387 */
388STATIC void
389xfs_inode_item_format(
390 struct xfs_log_item *lip,
391 struct xfs_log_vec *lv)
392{
393 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
394 struct xfs_inode *ip = iip->ili_inode;
395 struct xfs_log_iovec *vecp = NULL;
396 struct xfs_inode_log_format *ilf;
397
398 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
399 ilf->ilf_type = XFS_LI_INODE;
400 ilf->ilf_ino = ip->i_ino;
401 ilf->ilf_blkno = ip->i_imap.im_blkno;
402 ilf->ilf_len = ip->i_imap.im_len;
403 ilf->ilf_boffset = ip->i_imap.im_boffset;
404 ilf->ilf_fields = XFS_ILOG_CORE;
405 ilf->ilf_size = 2; /* format + core */
406
407 /*
408 * make sure we don't leak uninitialised data into the log in the case
409 * when we don't log every field in the inode.
410 */
411 ilf->ilf_dsize = 0;
412 ilf->ilf_asize = 0;
413 ilf->ilf_pad = 0;
414 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
415
416 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
417
418 xfs_inode_item_format_core(ip, lv, &vecp);
419 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
420 if (XFS_IFORK_Q(ip)) {
421 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
422 } else {
423 iip->ili_fields &=
424 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
425 }
426
427 /* update the format with the exact fields we actually logged */
428 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
429}
430
431/*
432 * This is called to pin the inode associated with the inode log
433 * item in memory so it cannot be written out.
434 */
435STATIC void
436xfs_inode_item_pin(
437 struct xfs_log_item *lip)
438{
439 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
440
441 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
442 ASSERT(lip->li_buf);
443
444 trace_xfs_inode_pin(ip, _RET_IP_);
445 atomic_inc(&ip->i_pincount);
446}
447
448
449/*
450 * This is called to unpin the inode associated with the inode log
451 * item which was previously pinned with a call to xfs_inode_item_pin().
452 *
453 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
454 *
455 * Note that unpin can race with inode cluster buffer freeing marking the buffer
456 * stale. In that case, flush completions are run from the buffer unpin call,
457 * which may happen before the inode is unpinned. If we lose the race, there
458 * will be no buffer attached to the log item, but the inode will be marked
459 * XFS_ISTALE.
460 */
461STATIC void
462xfs_inode_item_unpin(
463 struct xfs_log_item *lip,
464 int remove)
465{
466 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
467
468 trace_xfs_inode_unpin(ip, _RET_IP_);
469 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
470 ASSERT(atomic_read(&ip->i_pincount) > 0);
471 if (atomic_dec_and_test(&ip->i_pincount))
472 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
473}
474
475STATIC uint
476xfs_inode_item_push(
477 struct xfs_log_item *lip,
478 struct list_head *buffer_list)
479 __releases(&lip->li_ailp->ail_lock)
480 __acquires(&lip->li_ailp->ail_lock)
481{
482 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
483 struct xfs_inode *ip = iip->ili_inode;
484 struct xfs_buf *bp = lip->li_buf;
485 uint rval = XFS_ITEM_SUCCESS;
486 int error;
487
488 ASSERT(iip->ili_item.li_buf);
489
490 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp) ||
491 (ip->i_flags & XFS_ISTALE))
492 return XFS_ITEM_PINNED;
493
494 /* If the inode is already flush locked, we're already flushing. */
495 if (xfs_isiflocked(ip))
496 return XFS_ITEM_FLUSHING;
497
498 if (!xfs_buf_trylock(bp))
499 return XFS_ITEM_LOCKED;
500
501 spin_unlock(&lip->li_ailp->ail_lock);
502
503 /*
504 * We need to hold a reference for flushing the cluster buffer as it may
505 * fail the buffer without IO submission. In which case, we better get a
506 * reference for that completion because otherwise we don't get a
507 * reference for IO until we queue the buffer for delwri submission.
508 */
509 xfs_buf_hold(bp);
510 error = xfs_iflush_cluster(bp);
511 if (!error) {
512 if (!xfs_buf_delwri_queue(bp, buffer_list))
513 rval = XFS_ITEM_FLUSHING;
514 xfs_buf_relse(bp);
515 } else {
516 /*
517 * Release the buffer if we were unable to flush anything. On
518 * any other error, the buffer has already been released.
519 */
520 if (error == -EAGAIN)
521 xfs_buf_relse(bp);
522 rval = XFS_ITEM_LOCKED;
523 }
524
525 spin_lock(&lip->li_ailp->ail_lock);
526 return rval;
527}
528
529/*
530 * Unlock the inode associated with the inode log item.
531 */
532STATIC void
533xfs_inode_item_release(
534 struct xfs_log_item *lip)
535{
536 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
537 struct xfs_inode *ip = iip->ili_inode;
538 unsigned short lock_flags;
539
540 ASSERT(ip->i_itemp != NULL);
541 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
542
543 lock_flags = iip->ili_lock_flags;
544 iip->ili_lock_flags = 0;
545 if (lock_flags)
546 xfs_iunlock(ip, lock_flags);
547}
548
549/*
550 * This is called to find out where the oldest active copy of the inode log
551 * item in the on disk log resides now that the last log write of it completed
552 * at the given lsn. Since we always re-log all dirty data in an inode, the
553 * latest copy in the on disk log is the only one that matters. Therefore,
554 * simply return the given lsn.
555 *
556 * If the inode has been marked stale because the cluster is being freed, we
557 * don't want to (re-)insert this inode into the AIL. There is a race condition
558 * where the cluster buffer may be unpinned before the inode is inserted into
559 * the AIL during transaction committed processing. If the buffer is unpinned
560 * before the inode item has been committed and inserted, then it is possible
561 * for the buffer to be written and IO completes before the inode is inserted
562 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
563 * AIL which will never get removed. It will, however, get reclaimed which
564 * triggers an assert in xfs_inode_free() complaining about freein an inode
565 * still in the AIL.
566 *
567 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
568 * transaction committed code knows that it does not need to do any further
569 * processing on the item.
570 */
571STATIC xfs_lsn_t
572xfs_inode_item_committed(
573 struct xfs_log_item *lip,
574 xfs_lsn_t lsn)
575{
576 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
577 struct xfs_inode *ip = iip->ili_inode;
578
579 if (xfs_iflags_test(ip, XFS_ISTALE)) {
580 xfs_inode_item_unpin(lip, 0);
581 return -1;
582 }
583 return lsn;
584}
585
586STATIC void
587xfs_inode_item_committing(
588 struct xfs_log_item *lip,
589 xfs_lsn_t commit_lsn)
590{
591 INODE_ITEM(lip)->ili_last_lsn = commit_lsn;
592 return xfs_inode_item_release(lip);
593}
594
595static const struct xfs_item_ops xfs_inode_item_ops = {
596 .iop_size = xfs_inode_item_size,
597 .iop_format = xfs_inode_item_format,
598 .iop_pin = xfs_inode_item_pin,
599 .iop_unpin = xfs_inode_item_unpin,
600 .iop_release = xfs_inode_item_release,
601 .iop_committed = xfs_inode_item_committed,
602 .iop_push = xfs_inode_item_push,
603 .iop_committing = xfs_inode_item_committing,
604};
605
606
607/*
608 * Initialize the inode log item for a newly allocated (in-core) inode.
609 */
610void
611xfs_inode_item_init(
612 struct xfs_inode *ip,
613 struct xfs_mount *mp)
614{
615 struct xfs_inode_log_item *iip;
616
617 ASSERT(ip->i_itemp == NULL);
618 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_zone,
619 GFP_KERNEL | __GFP_NOFAIL);
620
621 iip->ili_inode = ip;
622 spin_lock_init(&iip->ili_lock);
623 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
624 &xfs_inode_item_ops);
625}
626
627/*
628 * Free the inode log item and any memory hanging off of it.
629 */
630void
631xfs_inode_item_destroy(
632 struct xfs_inode *ip)
633{
634 struct xfs_inode_log_item *iip = ip->i_itemp;
635
636 ASSERT(iip->ili_item.li_buf == NULL);
637
638 ip->i_itemp = NULL;
639 kmem_free(iip->ili_item.li_lv_shadow);
640 kmem_cache_free(xfs_ili_zone, iip);
641}
642
643
644/*
645 * We only want to pull the item from the AIL if it is actually there
646 * and its location in the log has not changed since we started the
647 * flush. Thus, we only bother if the inode's lsn has not changed.
648 */
649static void
650xfs_iflush_ail_updates(
651 struct xfs_ail *ailp,
652 struct list_head *list)
653{
654 struct xfs_log_item *lip;
655 xfs_lsn_t tail_lsn = 0;
656
657 /* this is an opencoded batch version of xfs_trans_ail_delete */
658 spin_lock(&ailp->ail_lock);
659 list_for_each_entry(lip, list, li_bio_list) {
660 xfs_lsn_t lsn;
661
662 clear_bit(XFS_LI_FAILED, &lip->li_flags);
663 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
664 continue;
665
666 lsn = xfs_ail_delete_one(ailp, lip);
667 if (!tail_lsn && lsn)
668 tail_lsn = lsn;
669 }
670 xfs_ail_update_finish(ailp, tail_lsn);
671}
672
673/*
674 * Walk the list of inodes that have completed their IOs. If they are clean
675 * remove them from the list and dissociate them from the buffer. Buffers that
676 * are still dirty remain linked to the buffer and on the list. Caller must
677 * handle them appropriately.
678 */
679static void
680xfs_iflush_finish(
681 struct xfs_buf *bp,
682 struct list_head *list)
683{
684 struct xfs_log_item *lip, *n;
685
686 list_for_each_entry_safe(lip, n, list, li_bio_list) {
687 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
688 bool drop_buffer = false;
689
690 spin_lock(&iip->ili_lock);
691
692 /*
693 * Remove the reference to the cluster buffer if the inode is
694 * clean in memory and drop the buffer reference once we've
695 * dropped the locks we hold.
696 */
697 ASSERT(iip->ili_item.li_buf == bp);
698 if (!iip->ili_fields) {
699 iip->ili_item.li_buf = NULL;
700 list_del_init(&lip->li_bio_list);
701 drop_buffer = true;
702 }
703 iip->ili_last_fields = 0;
704 iip->ili_flush_lsn = 0;
705 spin_unlock(&iip->ili_lock);
706 xfs_ifunlock(iip->ili_inode);
707 if (drop_buffer)
708 xfs_buf_rele(bp);
709 }
710}
711
712/*
713 * Inode buffer IO completion routine. It is responsible for removing inodes
714 * attached to the buffer from the AIL if they have not been re-logged, as well
715 * as completing the flush and unlocking the inode.
716 */
717void
718xfs_iflush_done(
719 struct xfs_buf *bp)
720{
721 struct xfs_log_item *lip, *n;
722 LIST_HEAD(flushed_inodes);
723 LIST_HEAD(ail_updates);
724
725 /*
726 * Pull the attached inodes from the buffer one at a time and take the
727 * appropriate action on them.
728 */
729 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
730 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
731
732 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
733 xfs_iflush_abort(iip->ili_inode);
734 continue;
735 }
736 if (!iip->ili_last_fields)
737 continue;
738
739 /* Do an unlocked check for needing the AIL lock. */
740 if (iip->ili_flush_lsn == lip->li_lsn ||
741 test_bit(XFS_LI_FAILED, &lip->li_flags))
742 list_move_tail(&lip->li_bio_list, &ail_updates);
743 else
744 list_move_tail(&lip->li_bio_list, &flushed_inodes);
745 }
746
747 if (!list_empty(&ail_updates)) {
748 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
749 list_splice_tail(&ail_updates, &flushed_inodes);
750 }
751
752 xfs_iflush_finish(bp, &flushed_inodes);
753 if (!list_empty(&flushed_inodes))
754 list_splice_tail(&flushed_inodes, &bp->b_li_list);
755}
756
757/*
758 * This is the inode flushing abort routine. It is called from xfs_iflush when
759 * the filesystem is shutting down to clean up the inode state. It is
760 * responsible for removing the inode item from the AIL if it has not been
761 * re-logged, and unlocking the inode's flush lock.
762 */
763void
764xfs_iflush_abort(
765 struct xfs_inode *ip)
766{
767 struct xfs_inode_log_item *iip = ip->i_itemp;
768 struct xfs_buf *bp = NULL;
769
770 if (iip) {
771 /*
772 * Clear the failed bit before removing the item from the AIL so
773 * xfs_trans_ail_delete() doesn't try to clear and release the
774 * buffer attached to the log item before we are done with it.
775 */
776 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
777 xfs_trans_ail_delete(&iip->ili_item, 0);
778
779 /*
780 * Clear the inode logging fields so no more flushes are
781 * attempted.
782 */
783 spin_lock(&iip->ili_lock);
784 iip->ili_last_fields = 0;
785 iip->ili_fields = 0;
786 iip->ili_fsync_fields = 0;
787 iip->ili_flush_lsn = 0;
788 bp = iip->ili_item.li_buf;
789 iip->ili_item.li_buf = NULL;
790 list_del_init(&iip->ili_item.li_bio_list);
791 spin_unlock(&iip->ili_lock);
792 }
793 xfs_ifunlock(ip);
794 if (bp)
795 xfs_buf_rele(bp);
796}
797
798/*
799 * convert an xfs_inode_log_format struct from the old 32 bit version
800 * (which can have different field alignments) to the native 64 bit version
801 */
802int
803xfs_inode_item_format_convert(
804 struct xfs_log_iovec *buf,
805 struct xfs_inode_log_format *in_f)
806{
807 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
808
809 if (buf->i_len != sizeof(*in_f32)) {
810 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
811 return -EFSCORRUPTED;
812 }
813
814 in_f->ilf_type = in_f32->ilf_type;
815 in_f->ilf_size = in_f32->ilf_size;
816 in_f->ilf_fields = in_f32->ilf_fields;
817 in_f->ilf_asize = in_f32->ilf_asize;
818 in_f->ilf_dsize = in_f32->ilf_dsize;
819 in_f->ilf_ino = in_f32->ilf_ino;
820 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
821 in_f->ilf_blkno = in_f32->ilf_blkno;
822 in_f->ilf_len = in_f32->ilf_len;
823 in_f->ilf_boffset = in_f32->ilf_boffset;
824 return 0;
825}