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/*
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_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_inode.h"
25#include "xfs_trans.h"
26#include "xfs_inode_item.h"
27#include "xfs_error.h"
28#include "xfs_trace.h"
29#include "xfs_trans_priv.h"
30#include "xfs_log.h"
31
32
33kmem_zone_t *xfs_ili_zone; /* inode log item zone */
34
35static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
36{
37 return container_of(lip, struct xfs_inode_log_item, ili_item);
38}
39
40STATIC void
41xfs_inode_item_data_fork_size(
42 struct xfs_inode_log_item *iip,
43 int *nvecs,
44 int *nbytes)
45{
46 struct xfs_inode *ip = iip->ili_inode;
47
48 switch (ip->i_d.di_format) {
49 case XFS_DINODE_FMT_EXTENTS:
50 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
51 ip->i_d.di_nextents > 0 &&
52 ip->i_df.if_bytes > 0) {
53 /* worst case, doesn't subtract delalloc extents */
54 *nbytes += XFS_IFORK_DSIZE(ip);
55 *nvecs += 1;
56 }
57 break;
58 case XFS_DINODE_FMT_BTREE:
59 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
60 ip->i_df.if_broot_bytes > 0) {
61 *nbytes += ip->i_df.if_broot_bytes;
62 *nvecs += 1;
63 }
64 break;
65 case XFS_DINODE_FMT_LOCAL:
66 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
67 ip->i_df.if_bytes > 0) {
68 *nbytes += roundup(ip->i_df.if_bytes, 4);
69 *nvecs += 1;
70 }
71 break;
72
73 case XFS_DINODE_FMT_DEV:
74 case XFS_DINODE_FMT_UUID:
75 break;
76 default:
77 ASSERT(0);
78 break;
79 }
80}
81
82STATIC void
83xfs_inode_item_attr_fork_size(
84 struct xfs_inode_log_item *iip,
85 int *nvecs,
86 int *nbytes)
87{
88 struct xfs_inode *ip = iip->ili_inode;
89
90 switch (ip->i_d.di_aformat) {
91 case XFS_DINODE_FMT_EXTENTS:
92 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
93 ip->i_d.di_anextents > 0 &&
94 ip->i_afp->if_bytes > 0) {
95 /* worst case, doesn't subtract unused space */
96 *nbytes += XFS_IFORK_ASIZE(ip);
97 *nvecs += 1;
98 }
99 break;
100 case XFS_DINODE_FMT_BTREE:
101 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
102 ip->i_afp->if_broot_bytes > 0) {
103 *nbytes += ip->i_afp->if_broot_bytes;
104 *nvecs += 1;
105 }
106 break;
107 case XFS_DINODE_FMT_LOCAL:
108 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
109 ip->i_afp->if_bytes > 0) {
110 *nbytes += roundup(ip->i_afp->if_bytes, 4);
111 *nvecs += 1;
112 }
113 break;
114 default:
115 ASSERT(0);
116 break;
117 }
118}
119
120/*
121 * This returns the number of iovecs needed to log the given inode item.
122 *
123 * We need one iovec for the inode log format structure, one for the
124 * inode core, and possibly one for the inode data/extents/b-tree root
125 * and one for the inode attribute data/extents/b-tree root.
126 */
127STATIC void
128xfs_inode_item_size(
129 struct xfs_log_item *lip,
130 int *nvecs,
131 int *nbytes)
132{
133 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
134 struct xfs_inode *ip = iip->ili_inode;
135
136 *nvecs += 2;
137 *nbytes += sizeof(struct xfs_inode_log_format) +
138 xfs_log_dinode_size(ip->i_d.di_version);
139
140 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
141 if (XFS_IFORK_Q(ip))
142 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
143}
144
145STATIC void
146xfs_inode_item_format_data_fork(
147 struct xfs_inode_log_item *iip,
148 struct xfs_inode_log_format *ilf,
149 struct xfs_log_vec *lv,
150 struct xfs_log_iovec **vecp)
151{
152 struct xfs_inode *ip = iip->ili_inode;
153 size_t data_bytes;
154
155 switch (ip->i_d.di_format) {
156 case XFS_DINODE_FMT_EXTENTS:
157 iip->ili_fields &=
158 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
159 XFS_ILOG_DEV | XFS_ILOG_UUID);
160
161 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
162 ip->i_d.di_nextents > 0 &&
163 ip->i_df.if_bytes > 0) {
164 struct xfs_bmbt_rec *p;
165
166 ASSERT(ip->i_df.if_u1.if_extents != NULL);
167 ASSERT(xfs_iext_count(&ip->i_df) > 0);
168
169 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
170 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
171 xlog_finish_iovec(lv, *vecp, data_bytes);
172
173 ASSERT(data_bytes <= ip->i_df.if_bytes);
174
175 ilf->ilf_dsize = data_bytes;
176 ilf->ilf_size++;
177 } else {
178 iip->ili_fields &= ~XFS_ILOG_DEXT;
179 }
180 break;
181 case XFS_DINODE_FMT_BTREE:
182 iip->ili_fields &=
183 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
184 XFS_ILOG_DEV | XFS_ILOG_UUID);
185
186 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
187 ip->i_df.if_broot_bytes > 0) {
188 ASSERT(ip->i_df.if_broot != NULL);
189 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
190 ip->i_df.if_broot,
191 ip->i_df.if_broot_bytes);
192 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
193 ilf->ilf_size++;
194 } else {
195 ASSERT(!(iip->ili_fields &
196 XFS_ILOG_DBROOT));
197 iip->ili_fields &= ~XFS_ILOG_DBROOT;
198 }
199 break;
200 case XFS_DINODE_FMT_LOCAL:
201 iip->ili_fields &=
202 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
203 XFS_ILOG_DEV | XFS_ILOG_UUID);
204 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
205 ip->i_df.if_bytes > 0) {
206 /*
207 * Round i_bytes up to a word boundary.
208 * The underlying memory is guaranteed to
209 * to be there by xfs_idata_realloc().
210 */
211 data_bytes = roundup(ip->i_df.if_bytes, 4);
212 ASSERT(ip->i_df.if_real_bytes == 0 ||
213 ip->i_df.if_real_bytes >= data_bytes);
214 ASSERT(ip->i_df.if_u1.if_data != NULL);
215 ASSERT(ip->i_d.di_size > 0);
216 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
217 ip->i_df.if_u1.if_data, data_bytes);
218 ilf->ilf_dsize = (unsigned)data_bytes;
219 ilf->ilf_size++;
220 } else {
221 iip->ili_fields &= ~XFS_ILOG_DDATA;
222 }
223 break;
224 case XFS_DINODE_FMT_DEV:
225 iip->ili_fields &=
226 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
227 XFS_ILOG_DEXT | XFS_ILOG_UUID);
228 if (iip->ili_fields & XFS_ILOG_DEV)
229 ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
230 break;
231 case XFS_DINODE_FMT_UUID:
232 iip->ili_fields &=
233 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
234 XFS_ILOG_DEXT | XFS_ILOG_DEV);
235 if (iip->ili_fields & XFS_ILOG_UUID)
236 ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
237 break;
238 default:
239 ASSERT(0);
240 break;
241 }
242}
243
244STATIC void
245xfs_inode_item_format_attr_fork(
246 struct xfs_inode_log_item *iip,
247 struct xfs_inode_log_format *ilf,
248 struct xfs_log_vec *lv,
249 struct xfs_log_iovec **vecp)
250{
251 struct xfs_inode *ip = iip->ili_inode;
252 size_t data_bytes;
253
254 switch (ip->i_d.di_aformat) {
255 case XFS_DINODE_FMT_EXTENTS:
256 iip->ili_fields &=
257 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
258
259 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
260 ip->i_d.di_anextents > 0 &&
261 ip->i_afp->if_bytes > 0) {
262 struct xfs_bmbt_rec *p;
263
264 ASSERT(xfs_iext_count(ip->i_afp) ==
265 ip->i_d.di_anextents);
266 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
267
268 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
269 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
270 xlog_finish_iovec(lv, *vecp, data_bytes);
271
272 ilf->ilf_asize = data_bytes;
273 ilf->ilf_size++;
274 } else {
275 iip->ili_fields &= ~XFS_ILOG_AEXT;
276 }
277 break;
278 case XFS_DINODE_FMT_BTREE:
279 iip->ili_fields &=
280 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
281
282 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
283 ip->i_afp->if_broot_bytes > 0) {
284 ASSERT(ip->i_afp->if_broot != NULL);
285
286 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
287 ip->i_afp->if_broot,
288 ip->i_afp->if_broot_bytes);
289 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
290 ilf->ilf_size++;
291 } else {
292 iip->ili_fields &= ~XFS_ILOG_ABROOT;
293 }
294 break;
295 case XFS_DINODE_FMT_LOCAL:
296 iip->ili_fields &=
297 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
298
299 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
300 ip->i_afp->if_bytes > 0) {
301 /*
302 * Round i_bytes up to a word boundary.
303 * The underlying memory is guaranteed to
304 * to be there by xfs_idata_realloc().
305 */
306 data_bytes = roundup(ip->i_afp->if_bytes, 4);
307 ASSERT(ip->i_afp->if_real_bytes == 0 ||
308 ip->i_afp->if_real_bytes >= data_bytes);
309 ASSERT(ip->i_afp->if_u1.if_data != NULL);
310 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
311 ip->i_afp->if_u1.if_data,
312 data_bytes);
313 ilf->ilf_asize = (unsigned)data_bytes;
314 ilf->ilf_size++;
315 } else {
316 iip->ili_fields &= ~XFS_ILOG_ADATA;
317 }
318 break;
319 default:
320 ASSERT(0);
321 break;
322 }
323}
324
325static void
326xfs_inode_to_log_dinode(
327 struct xfs_inode *ip,
328 struct xfs_log_dinode *to,
329 xfs_lsn_t lsn)
330{
331 struct xfs_icdinode *from = &ip->i_d;
332 struct inode *inode = VFS_I(ip);
333
334 to->di_magic = XFS_DINODE_MAGIC;
335
336 to->di_version = from->di_version;
337 to->di_format = from->di_format;
338 to->di_uid = from->di_uid;
339 to->di_gid = from->di_gid;
340 to->di_projid_lo = from->di_projid_lo;
341 to->di_projid_hi = from->di_projid_hi;
342
343 memset(to->di_pad, 0, sizeof(to->di_pad));
344 memset(to->di_pad3, 0, sizeof(to->di_pad3));
345 to->di_atime.t_sec = inode->i_atime.tv_sec;
346 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
347 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
348 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
349 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
350 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
351 to->di_nlink = inode->i_nlink;
352 to->di_gen = inode->i_generation;
353 to->di_mode = inode->i_mode;
354
355 to->di_size = from->di_size;
356 to->di_nblocks = from->di_nblocks;
357 to->di_extsize = from->di_extsize;
358 to->di_nextents = from->di_nextents;
359 to->di_anextents = from->di_anextents;
360 to->di_forkoff = from->di_forkoff;
361 to->di_aformat = from->di_aformat;
362 to->di_dmevmask = from->di_dmevmask;
363 to->di_dmstate = from->di_dmstate;
364 to->di_flags = from->di_flags;
365
366 if (from->di_version == 3) {
367 to->di_changecount = inode->i_version;
368 to->di_crtime.t_sec = from->di_crtime.t_sec;
369 to->di_crtime.t_nsec = from->di_crtime.t_nsec;
370 to->di_flags2 = from->di_flags2;
371 to->di_cowextsize = from->di_cowextsize;
372 to->di_ino = ip->i_ino;
373 to->di_lsn = lsn;
374 memset(to->di_pad2, 0, sizeof(to->di_pad2));
375 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
376 to->di_flushiter = 0;
377 } else {
378 to->di_flushiter = from->di_flushiter;
379 }
380}
381
382/*
383 * Format the inode core. Current timestamp data is only in the VFS inode
384 * fields, so we need to grab them from there. Hence rather than just copying
385 * the XFS inode core structure, format the fields directly into the iovec.
386 */
387static void
388xfs_inode_item_format_core(
389 struct xfs_inode *ip,
390 struct xfs_log_vec *lv,
391 struct xfs_log_iovec **vecp)
392{
393 struct xfs_log_dinode *dic;
394
395 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
396 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
397 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
398}
399
400/*
401 * This is called to fill in the vector of log iovecs for the given inode
402 * log item. It fills the first item with an inode log format structure,
403 * the second with the on-disk inode structure, and a possible third and/or
404 * fourth with the inode data/extents/b-tree root and inode attributes
405 * data/extents/b-tree root.
406 */
407STATIC void
408xfs_inode_item_format(
409 struct xfs_log_item *lip,
410 struct xfs_log_vec *lv)
411{
412 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
413 struct xfs_inode *ip = iip->ili_inode;
414 struct xfs_inode_log_format *ilf;
415 struct xfs_log_iovec *vecp = NULL;
416
417 ASSERT(ip->i_d.di_version > 1);
418
419 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
420 ilf->ilf_type = XFS_LI_INODE;
421 ilf->ilf_ino = ip->i_ino;
422 ilf->ilf_blkno = ip->i_imap.im_blkno;
423 ilf->ilf_len = ip->i_imap.im_len;
424 ilf->ilf_boffset = ip->i_imap.im_boffset;
425 ilf->ilf_fields = XFS_ILOG_CORE;
426 ilf->ilf_size = 2; /* format + core */
427 xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
428
429 xfs_inode_item_format_core(ip, lv, &vecp);
430 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
431 if (XFS_IFORK_Q(ip)) {
432 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
433 } else {
434 iip->ili_fields &=
435 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
436 }
437
438 /* update the format with the exact fields we actually logged */
439 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
440}
441
442/*
443 * This is called to pin the inode associated with the inode log
444 * item in memory so it cannot be written out.
445 */
446STATIC void
447xfs_inode_item_pin(
448 struct xfs_log_item *lip)
449{
450 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
451
452 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
453
454 trace_xfs_inode_pin(ip, _RET_IP_);
455 atomic_inc(&ip->i_pincount);
456}
457
458
459/*
460 * This is called to unpin the inode associated with the inode log
461 * item which was previously pinned with a call to xfs_inode_item_pin().
462 *
463 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
464 */
465STATIC void
466xfs_inode_item_unpin(
467 struct xfs_log_item *lip,
468 int remove)
469{
470 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
471
472 trace_xfs_inode_unpin(ip, _RET_IP_);
473 ASSERT(atomic_read(&ip->i_pincount) > 0);
474 if (atomic_dec_and_test(&ip->i_pincount))
475 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
476}
477
478STATIC uint
479xfs_inode_item_push(
480 struct xfs_log_item *lip,
481 struct list_head *buffer_list)
482 __releases(&lip->li_ailp->xa_lock)
483 __acquires(&lip->li_ailp->xa_lock)
484{
485 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
486 struct xfs_inode *ip = iip->ili_inode;
487 struct xfs_buf *bp = NULL;
488 uint rval = XFS_ITEM_SUCCESS;
489 int error;
490
491 if (xfs_ipincount(ip) > 0)
492 return XFS_ITEM_PINNED;
493
494 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
495 return XFS_ITEM_LOCKED;
496
497 /*
498 * Re-check the pincount now that we stabilized the value by
499 * taking the ilock.
500 */
501 if (xfs_ipincount(ip) > 0) {
502 rval = XFS_ITEM_PINNED;
503 goto out_unlock;
504 }
505
506 /*
507 * Stale inode items should force out the iclog.
508 */
509 if (ip->i_flags & XFS_ISTALE) {
510 rval = XFS_ITEM_PINNED;
511 goto out_unlock;
512 }
513
514 /*
515 * Someone else is already flushing the inode. Nothing we can do
516 * here but wait for the flush to finish and remove the item from
517 * the AIL.
518 */
519 if (!xfs_iflock_nowait(ip)) {
520 rval = XFS_ITEM_FLUSHING;
521 goto out_unlock;
522 }
523
524 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
525 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
526
527 spin_unlock(&lip->li_ailp->xa_lock);
528
529 error = xfs_iflush(ip, &bp);
530 if (!error) {
531 if (!xfs_buf_delwri_queue(bp, buffer_list))
532 rval = XFS_ITEM_FLUSHING;
533 xfs_buf_relse(bp);
534 }
535
536 spin_lock(&lip->li_ailp->xa_lock);
537out_unlock:
538 xfs_iunlock(ip, XFS_ILOCK_SHARED);
539 return rval;
540}
541
542/*
543 * Unlock the inode associated with the inode log item.
544 * Clear the fields of the inode and inode log item that
545 * are specific to the current transaction. If the
546 * hold flags is set, do not unlock the inode.
547 */
548STATIC void
549xfs_inode_item_unlock(
550 struct xfs_log_item *lip)
551{
552 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
553 struct xfs_inode *ip = iip->ili_inode;
554 unsigned short lock_flags;
555
556 ASSERT(ip->i_itemp != NULL);
557 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
558
559 lock_flags = iip->ili_lock_flags;
560 iip->ili_lock_flags = 0;
561 if (lock_flags)
562 xfs_iunlock(ip, lock_flags);
563}
564
565/*
566 * This is called to find out where the oldest active copy of the inode log
567 * item in the on disk log resides now that the last log write of it completed
568 * at the given lsn. Since we always re-log all dirty data in an inode, the
569 * latest copy in the on disk log is the only one that matters. Therefore,
570 * simply return the given lsn.
571 *
572 * If the inode has been marked stale because the cluster is being freed, we
573 * don't want to (re-)insert this inode into the AIL. There is a race condition
574 * where the cluster buffer may be unpinned before the inode is inserted into
575 * the AIL during transaction committed processing. If the buffer is unpinned
576 * before the inode item has been committed and inserted, then it is possible
577 * for the buffer to be written and IO completes before the inode is inserted
578 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
579 * AIL which will never get removed. It will, however, get reclaimed which
580 * triggers an assert in xfs_inode_free() complaining about freein an inode
581 * still in the AIL.
582 *
583 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
584 * transaction committed code knows that it does not need to do any further
585 * processing on the item.
586 */
587STATIC xfs_lsn_t
588xfs_inode_item_committed(
589 struct xfs_log_item *lip,
590 xfs_lsn_t lsn)
591{
592 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
593 struct xfs_inode *ip = iip->ili_inode;
594
595 if (xfs_iflags_test(ip, XFS_ISTALE)) {
596 xfs_inode_item_unpin(lip, 0);
597 return -1;
598 }
599 return lsn;
600}
601
602/*
603 * XXX rcc - this one really has to do something. Probably needs
604 * to stamp in a new field in the incore inode.
605 */
606STATIC void
607xfs_inode_item_committing(
608 struct xfs_log_item *lip,
609 xfs_lsn_t lsn)
610{
611 INODE_ITEM(lip)->ili_last_lsn = lsn;
612}
613
614/*
615 * This is the ops vector shared by all buf log items.
616 */
617static const struct xfs_item_ops xfs_inode_item_ops = {
618 .iop_size = xfs_inode_item_size,
619 .iop_format = xfs_inode_item_format,
620 .iop_pin = xfs_inode_item_pin,
621 .iop_unpin = xfs_inode_item_unpin,
622 .iop_unlock = xfs_inode_item_unlock,
623 .iop_committed = xfs_inode_item_committed,
624 .iop_push = xfs_inode_item_push,
625 .iop_committing = xfs_inode_item_committing
626};
627
628
629/*
630 * Initialize the inode log item for a newly allocated (in-core) inode.
631 */
632void
633xfs_inode_item_init(
634 struct xfs_inode *ip,
635 struct xfs_mount *mp)
636{
637 struct xfs_inode_log_item *iip;
638
639 ASSERT(ip->i_itemp == NULL);
640 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
641
642 iip->ili_inode = ip;
643 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
644 &xfs_inode_item_ops);
645}
646
647/*
648 * Free the inode log item and any memory hanging off of it.
649 */
650void
651xfs_inode_item_destroy(
652 xfs_inode_t *ip)
653{
654 kmem_free(ip->i_itemp->ili_item.li_lv_shadow);
655 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
656}
657
658
659/*
660 * This is the inode flushing I/O completion routine. It is called
661 * from interrupt level when the buffer containing the inode is
662 * flushed to disk. It is responsible for removing the inode item
663 * from the AIL if it has not been re-logged, and unlocking the inode's
664 * flush lock.
665 *
666 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
667 * list for other inodes that will run this function. We remove them from the
668 * buffer list so we can process all the inode IO completions in one AIL lock
669 * traversal.
670 */
671void
672xfs_iflush_done(
673 struct xfs_buf *bp,
674 struct xfs_log_item *lip)
675{
676 struct xfs_inode_log_item *iip;
677 struct xfs_log_item *blip;
678 struct xfs_log_item *next;
679 struct xfs_log_item *prev;
680 struct xfs_ail *ailp = lip->li_ailp;
681 int need_ail = 0;
682
683 /*
684 * Scan the buffer IO completions for other inodes being completed and
685 * attach them to the current inode log item.
686 */
687 blip = bp->b_fspriv;
688 prev = NULL;
689 while (blip != NULL) {
690 if (blip->li_cb != xfs_iflush_done) {
691 prev = blip;
692 blip = blip->li_bio_list;
693 continue;
694 }
695
696 /* remove from list */
697 next = blip->li_bio_list;
698 if (!prev) {
699 bp->b_fspriv = next;
700 } else {
701 prev->li_bio_list = next;
702 }
703
704 /* add to current list */
705 blip->li_bio_list = lip->li_bio_list;
706 lip->li_bio_list = blip;
707
708 /*
709 * while we have the item, do the unlocked check for needing
710 * the AIL lock.
711 */
712 iip = INODE_ITEM(blip);
713 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
714 need_ail++;
715
716 blip = next;
717 }
718
719 /* make sure we capture the state of the initial inode. */
720 iip = INODE_ITEM(lip);
721 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
722 need_ail++;
723
724 /*
725 * We only want to pull the item from the AIL if it is
726 * actually there and its location in the log has not
727 * changed since we started the flush. Thus, we only bother
728 * if the ili_logged flag is set and the inode's lsn has not
729 * changed. First we check the lsn outside
730 * the lock since it's cheaper, and then we recheck while
731 * holding the lock before removing the inode from the AIL.
732 */
733 if (need_ail) {
734 struct xfs_log_item *log_items[need_ail];
735 int i = 0;
736 spin_lock(&ailp->xa_lock);
737 for (blip = lip; blip; blip = blip->li_bio_list) {
738 iip = INODE_ITEM(blip);
739 if (iip->ili_logged &&
740 blip->li_lsn == iip->ili_flush_lsn) {
741 log_items[i++] = blip;
742 }
743 ASSERT(i <= need_ail);
744 }
745 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
746 xfs_trans_ail_delete_bulk(ailp, log_items, i,
747 SHUTDOWN_CORRUPT_INCORE);
748 }
749
750
751 /*
752 * clean up and unlock the flush lock now we are done. We can clear the
753 * ili_last_fields bits now that we know that the data corresponding to
754 * them is safely on disk.
755 */
756 for (blip = lip; blip; blip = next) {
757 next = blip->li_bio_list;
758 blip->li_bio_list = NULL;
759
760 iip = INODE_ITEM(blip);
761 iip->ili_logged = 0;
762 iip->ili_last_fields = 0;
763 xfs_ifunlock(iip->ili_inode);
764 }
765}
766
767/*
768 * This is the inode flushing abort routine. It is called from xfs_iflush when
769 * the filesystem is shutting down to clean up the inode state. It is
770 * responsible for removing the inode item from the AIL if it has not been
771 * re-logged, and unlocking the inode's flush lock.
772 */
773void
774xfs_iflush_abort(
775 xfs_inode_t *ip,
776 bool stale)
777{
778 xfs_inode_log_item_t *iip = ip->i_itemp;
779
780 if (iip) {
781 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
782 xfs_trans_ail_remove(&iip->ili_item,
783 stale ? SHUTDOWN_LOG_IO_ERROR :
784 SHUTDOWN_CORRUPT_INCORE);
785 }
786 iip->ili_logged = 0;
787 /*
788 * Clear the ili_last_fields bits now that we know that the
789 * data corresponding to them is safely on disk.
790 */
791 iip->ili_last_fields = 0;
792 /*
793 * Clear the inode logging fields so no more flushes are
794 * attempted.
795 */
796 iip->ili_fields = 0;
797 iip->ili_fsync_fields = 0;
798 }
799 /*
800 * Release the inode's flush lock since we're done with it.
801 */
802 xfs_ifunlock(ip);
803}
804
805void
806xfs_istale_done(
807 struct xfs_buf *bp,
808 struct xfs_log_item *lip)
809{
810 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
811}
812
813/*
814 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
815 * (which can have different field alignments) to the native version
816 */
817int
818xfs_inode_item_format_convert(
819 xfs_log_iovec_t *buf,
820 xfs_inode_log_format_t *in_f)
821{
822 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
823 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
824
825 in_f->ilf_type = in_f32->ilf_type;
826 in_f->ilf_size = in_f32->ilf_size;
827 in_f->ilf_fields = in_f32->ilf_fields;
828 in_f->ilf_asize = in_f32->ilf_asize;
829 in_f->ilf_dsize = in_f32->ilf_dsize;
830 in_f->ilf_ino = in_f32->ilf_ino;
831 /* copy biggest field of ilf_u */
832 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
833 in_f32->ilf_u.ilfu_uuid.__u_bits,
834 sizeof(uuid_t));
835 in_f->ilf_blkno = in_f32->ilf_blkno;
836 in_f->ilf_len = in_f32->ilf_len;
837 in_f->ilf_boffset = in_f32->ilf_boffset;
838 return 0;
839 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
840 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
841
842 in_f->ilf_type = in_f64->ilf_type;
843 in_f->ilf_size = in_f64->ilf_size;
844 in_f->ilf_fields = in_f64->ilf_fields;
845 in_f->ilf_asize = in_f64->ilf_asize;
846 in_f->ilf_dsize = in_f64->ilf_dsize;
847 in_f->ilf_ino = in_f64->ilf_ino;
848 /* copy biggest field of ilf_u */
849 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
850 in_f64->ilf_u.ilfu_uuid.__u_bits,
851 sizeof(uuid_t));
852 in_f->ilf_blkno = in_f64->ilf_blkno;
853 in_f->ilf_len = in_f64->ilf_len;
854 in_f->ilf_boffset = in_f64->ilf_boffset;
855 return 0;
856 }
857 return -EFSCORRUPTED;
858}