FRV: Use generic show_interrupts()
[cris-mirror.git] / fs / xfs / xfs_inode_item.c
blobfd4f398bd6f1d8d01cca49046a65991d4d696d77
1 /*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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.
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.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
37 kmem_zone_t *xfs_ili_zone; /* inode log item zone */
39 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
41 return container_of(lip, struct xfs_inode_log_item, ili_item);
46 * This returns the number of iovecs needed to log the given inode item.
48 * We need one iovec for the inode log format structure, one for the
49 * inode core, and possibly one for the inode data/extents/b-tree root
50 * and one for the inode attribute data/extents/b-tree root.
52 STATIC uint
53 xfs_inode_item_size(
54 struct xfs_log_item *lip)
56 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
57 struct xfs_inode *ip = iip->ili_inode;
58 uint nvecs = 2;
61 * Only log the data/extents/b-tree root if there is something
62 * left to log.
64 iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
66 switch (ip->i_d.di_format) {
67 case XFS_DINODE_FMT_EXTENTS:
68 iip->ili_format.ilf_fields &=
69 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
70 XFS_ILOG_DEV | XFS_ILOG_UUID);
71 if ((iip->ili_format.ilf_fields & XFS_ILOG_DEXT) &&
72 (ip->i_d.di_nextents > 0) &&
73 (ip->i_df.if_bytes > 0)) {
74 ASSERT(ip->i_df.if_u1.if_extents != NULL);
75 nvecs++;
76 } else {
77 iip->ili_format.ilf_fields &= ~XFS_ILOG_DEXT;
79 break;
81 case XFS_DINODE_FMT_BTREE:
82 ASSERT(ip->i_df.if_ext_max ==
83 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
84 iip->ili_format.ilf_fields &=
85 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
86 XFS_ILOG_DEV | XFS_ILOG_UUID);
87 if ((iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) &&
88 (ip->i_df.if_broot_bytes > 0)) {
89 ASSERT(ip->i_df.if_broot != NULL);
90 nvecs++;
91 } else {
92 ASSERT(!(iip->ili_format.ilf_fields &
93 XFS_ILOG_DBROOT));
94 #ifdef XFS_TRANS_DEBUG
95 if (iip->ili_root_size > 0) {
96 ASSERT(iip->ili_root_size ==
97 ip->i_df.if_broot_bytes);
98 ASSERT(memcmp(iip->ili_orig_root,
99 ip->i_df.if_broot,
100 iip->ili_root_size) == 0);
101 } else {
102 ASSERT(ip->i_df.if_broot_bytes == 0);
104 #endif
105 iip->ili_format.ilf_fields &= ~XFS_ILOG_DBROOT;
107 break;
109 case XFS_DINODE_FMT_LOCAL:
110 iip->ili_format.ilf_fields &=
111 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
112 XFS_ILOG_DEV | XFS_ILOG_UUID);
113 if ((iip->ili_format.ilf_fields & XFS_ILOG_DDATA) &&
114 (ip->i_df.if_bytes > 0)) {
115 ASSERT(ip->i_df.if_u1.if_data != NULL);
116 ASSERT(ip->i_d.di_size > 0);
117 nvecs++;
118 } else {
119 iip->ili_format.ilf_fields &= ~XFS_ILOG_DDATA;
121 break;
123 case XFS_DINODE_FMT_DEV:
124 iip->ili_format.ilf_fields &=
125 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
126 XFS_ILOG_DEXT | XFS_ILOG_UUID);
127 break;
129 case XFS_DINODE_FMT_UUID:
130 iip->ili_format.ilf_fields &=
131 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
132 XFS_ILOG_DEXT | XFS_ILOG_DEV);
133 break;
135 default:
136 ASSERT(0);
137 break;
141 * If there are no attributes associated with this file,
142 * then there cannot be anything more to log.
143 * Clear all attribute-related log flags.
145 if (!XFS_IFORK_Q(ip)) {
146 iip->ili_format.ilf_fields &=
147 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
148 return nvecs;
152 * Log any necessary attribute data.
154 switch (ip->i_d.di_aformat) {
155 case XFS_DINODE_FMT_EXTENTS:
156 iip->ili_format.ilf_fields &=
157 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
158 if ((iip->ili_format.ilf_fields & XFS_ILOG_AEXT) &&
159 (ip->i_d.di_anextents > 0) &&
160 (ip->i_afp->if_bytes > 0)) {
161 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
162 nvecs++;
163 } else {
164 iip->ili_format.ilf_fields &= ~XFS_ILOG_AEXT;
166 break;
168 case XFS_DINODE_FMT_BTREE:
169 iip->ili_format.ilf_fields &=
170 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
171 if ((iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) &&
172 (ip->i_afp->if_broot_bytes > 0)) {
173 ASSERT(ip->i_afp->if_broot != NULL);
174 nvecs++;
175 } else {
176 iip->ili_format.ilf_fields &= ~XFS_ILOG_ABROOT;
178 break;
180 case XFS_DINODE_FMT_LOCAL:
181 iip->ili_format.ilf_fields &=
182 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
183 if ((iip->ili_format.ilf_fields & XFS_ILOG_ADATA) &&
184 (ip->i_afp->if_bytes > 0)) {
185 ASSERT(ip->i_afp->if_u1.if_data != NULL);
186 nvecs++;
187 } else {
188 iip->ili_format.ilf_fields &= ~XFS_ILOG_ADATA;
190 break;
192 default:
193 ASSERT(0);
194 break;
197 return nvecs;
201 * This is called to fill in the vector of log iovecs for the
202 * given inode log item. It fills the first item with an inode
203 * log format structure, the second with the on-disk inode structure,
204 * and a possible third and/or fourth with the inode data/extents/b-tree
205 * root and inode attributes data/extents/b-tree root.
207 STATIC void
208 xfs_inode_item_format(
209 struct xfs_log_item *lip,
210 struct xfs_log_iovec *vecp)
212 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
213 struct xfs_inode *ip = iip->ili_inode;
214 uint nvecs;
215 size_t data_bytes;
216 xfs_bmbt_rec_t *ext_buffer;
217 xfs_mount_t *mp;
219 vecp->i_addr = &iip->ili_format;
220 vecp->i_len = sizeof(xfs_inode_log_format_t);
221 vecp->i_type = XLOG_REG_TYPE_IFORMAT;
222 vecp++;
223 nvecs = 1;
226 * Clear i_update_core if the timestamps (or any other
227 * non-transactional modification) need flushing/logging
228 * and we're about to log them with the rest of the core.
230 * This is the same logic as xfs_iflush() but this code can't
231 * run at the same time as xfs_iflush because we're in commit
232 * processing here and so we have the inode lock held in
233 * exclusive mode. Although it doesn't really matter
234 * for the timestamps if both routines were to grab the
235 * timestamps or not. That would be ok.
237 * We clear i_update_core before copying out the data.
238 * This is for coordination with our timestamp updates
239 * that don't hold the inode lock. They will always
240 * update the timestamps BEFORE setting i_update_core,
241 * so if we clear i_update_core after they set it we
242 * are guaranteed to see their updates to the timestamps
243 * either here. Likewise, if they set it after we clear it
244 * here, we'll see it either on the next commit of this
245 * inode or the next time the inode gets flushed via
246 * xfs_iflush(). This depends on strongly ordered memory
247 * semantics, but we have that. We use the SYNCHRONIZE
248 * macro to make sure that the compiler does not reorder
249 * the i_update_core access below the data copy below.
251 if (ip->i_update_core) {
252 ip->i_update_core = 0;
253 SYNCHRONIZE();
257 * Make sure to get the latest timestamps from the Linux inode.
259 xfs_synchronize_times(ip);
261 vecp->i_addr = &ip->i_d;
262 vecp->i_len = sizeof(struct xfs_icdinode);
263 vecp->i_type = XLOG_REG_TYPE_ICORE;
264 vecp++;
265 nvecs++;
266 iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
269 * If this is really an old format inode, then we need to
270 * log it as such. This means that we have to copy the link
271 * count from the new field to the old. We don't have to worry
272 * about the new fields, because nothing trusts them as long as
273 * the old inode version number is there. If the superblock already
274 * has a new version number, then we don't bother converting back.
276 mp = ip->i_mount;
277 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
278 if (ip->i_d.di_version == 1) {
279 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
281 * Convert it back.
283 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
284 ip->i_d.di_onlink = ip->i_d.di_nlink;
285 } else {
287 * The superblock version has already been bumped,
288 * so just make the conversion to the new inode
289 * format permanent.
291 ip->i_d.di_version = 2;
292 ip->i_d.di_onlink = 0;
293 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
297 switch (ip->i_d.di_format) {
298 case XFS_DINODE_FMT_EXTENTS:
299 ASSERT(!(iip->ili_format.ilf_fields &
300 (XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
301 XFS_ILOG_DEV | XFS_ILOG_UUID)));
302 if (iip->ili_format.ilf_fields & XFS_ILOG_DEXT) {
303 ASSERT(ip->i_df.if_bytes > 0);
304 ASSERT(ip->i_df.if_u1.if_extents != NULL);
305 ASSERT(ip->i_d.di_nextents > 0);
306 ASSERT(iip->ili_extents_buf == NULL);
307 ASSERT((ip->i_df.if_bytes /
308 (uint)sizeof(xfs_bmbt_rec_t)) > 0);
309 #ifdef XFS_NATIVE_HOST
310 if (ip->i_d.di_nextents == ip->i_df.if_bytes /
311 (uint)sizeof(xfs_bmbt_rec_t)) {
313 * There are no delayed allocation
314 * extents, so just point to the
315 * real extents array.
317 vecp->i_addr = ip->i_df.if_u1.if_extents;
318 vecp->i_len = ip->i_df.if_bytes;
319 vecp->i_type = XLOG_REG_TYPE_IEXT;
320 } else
321 #endif
324 * There are delayed allocation extents
325 * in the inode, or we need to convert
326 * the extents to on disk format.
327 * Use xfs_iextents_copy()
328 * to copy only the real extents into
329 * a separate buffer. We'll free the
330 * buffer in the unlock routine.
332 ext_buffer = kmem_alloc(ip->i_df.if_bytes,
333 KM_SLEEP);
334 iip->ili_extents_buf = ext_buffer;
335 vecp->i_addr = ext_buffer;
336 vecp->i_len = xfs_iextents_copy(ip, ext_buffer,
337 XFS_DATA_FORK);
338 vecp->i_type = XLOG_REG_TYPE_IEXT;
340 ASSERT(vecp->i_len <= ip->i_df.if_bytes);
341 iip->ili_format.ilf_dsize = vecp->i_len;
342 vecp++;
343 nvecs++;
345 break;
347 case XFS_DINODE_FMT_BTREE:
348 ASSERT(!(iip->ili_format.ilf_fields &
349 (XFS_ILOG_DDATA | XFS_ILOG_DEXT |
350 XFS_ILOG_DEV | XFS_ILOG_UUID)));
351 if (iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) {
352 ASSERT(ip->i_df.if_broot_bytes > 0);
353 ASSERT(ip->i_df.if_broot != NULL);
354 vecp->i_addr = ip->i_df.if_broot;
355 vecp->i_len = ip->i_df.if_broot_bytes;
356 vecp->i_type = XLOG_REG_TYPE_IBROOT;
357 vecp++;
358 nvecs++;
359 iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
361 break;
363 case XFS_DINODE_FMT_LOCAL:
364 ASSERT(!(iip->ili_format.ilf_fields &
365 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
366 XFS_ILOG_DEV | XFS_ILOG_UUID)));
367 if (iip->ili_format.ilf_fields & XFS_ILOG_DDATA) {
368 ASSERT(ip->i_df.if_bytes > 0);
369 ASSERT(ip->i_df.if_u1.if_data != NULL);
370 ASSERT(ip->i_d.di_size > 0);
372 vecp->i_addr = ip->i_df.if_u1.if_data;
374 * Round i_bytes up to a word boundary.
375 * The underlying memory is guaranteed to
376 * to be there by xfs_idata_realloc().
378 data_bytes = roundup(ip->i_df.if_bytes, 4);
379 ASSERT((ip->i_df.if_real_bytes == 0) ||
380 (ip->i_df.if_real_bytes == data_bytes));
381 vecp->i_len = (int)data_bytes;
382 vecp->i_type = XLOG_REG_TYPE_ILOCAL;
383 vecp++;
384 nvecs++;
385 iip->ili_format.ilf_dsize = (unsigned)data_bytes;
387 break;
389 case XFS_DINODE_FMT_DEV:
390 ASSERT(!(iip->ili_format.ilf_fields &
391 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
392 XFS_ILOG_DDATA | XFS_ILOG_UUID)));
393 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
394 iip->ili_format.ilf_u.ilfu_rdev =
395 ip->i_df.if_u2.if_rdev;
397 break;
399 case XFS_DINODE_FMT_UUID:
400 ASSERT(!(iip->ili_format.ilf_fields &
401 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
402 XFS_ILOG_DDATA | XFS_ILOG_DEV)));
403 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
404 iip->ili_format.ilf_u.ilfu_uuid =
405 ip->i_df.if_u2.if_uuid;
407 break;
409 default:
410 ASSERT(0);
411 break;
415 * If there are no attributes associated with the file,
416 * then we're done.
417 * Assert that no attribute-related log flags are set.
419 if (!XFS_IFORK_Q(ip)) {
420 ASSERT(nvecs == lip->li_desc->lid_size);
421 iip->ili_format.ilf_size = nvecs;
422 ASSERT(!(iip->ili_format.ilf_fields &
423 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
424 return;
427 switch (ip->i_d.di_aformat) {
428 case XFS_DINODE_FMT_EXTENTS:
429 ASSERT(!(iip->ili_format.ilf_fields &
430 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT)));
431 if (iip->ili_format.ilf_fields & XFS_ILOG_AEXT) {
432 #ifdef DEBUG
433 int nrecs = ip->i_afp->if_bytes /
434 (uint)sizeof(xfs_bmbt_rec_t);
435 ASSERT(nrecs > 0);
436 ASSERT(nrecs == ip->i_d.di_anextents);
437 ASSERT(ip->i_afp->if_bytes > 0);
438 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
439 ASSERT(ip->i_d.di_anextents > 0);
440 #endif
441 #ifdef XFS_NATIVE_HOST
443 * There are not delayed allocation extents
444 * for attributes, so just point at the array.
446 vecp->i_addr = ip->i_afp->if_u1.if_extents;
447 vecp->i_len = ip->i_afp->if_bytes;
448 #else
449 ASSERT(iip->ili_aextents_buf == NULL);
451 * Need to endian flip before logging
453 ext_buffer = kmem_alloc(ip->i_afp->if_bytes,
454 KM_SLEEP);
455 iip->ili_aextents_buf = ext_buffer;
456 vecp->i_addr = ext_buffer;
457 vecp->i_len = xfs_iextents_copy(ip, ext_buffer,
458 XFS_ATTR_FORK);
459 #endif
460 vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
461 iip->ili_format.ilf_asize = vecp->i_len;
462 vecp++;
463 nvecs++;
465 break;
467 case XFS_DINODE_FMT_BTREE:
468 ASSERT(!(iip->ili_format.ilf_fields &
469 (XFS_ILOG_ADATA | XFS_ILOG_AEXT)));
470 if (iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) {
471 ASSERT(ip->i_afp->if_broot_bytes > 0);
472 ASSERT(ip->i_afp->if_broot != NULL);
473 vecp->i_addr = ip->i_afp->if_broot;
474 vecp->i_len = ip->i_afp->if_broot_bytes;
475 vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
476 vecp++;
477 nvecs++;
478 iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
480 break;
482 case XFS_DINODE_FMT_LOCAL:
483 ASSERT(!(iip->ili_format.ilf_fields &
484 (XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
485 if (iip->ili_format.ilf_fields & XFS_ILOG_ADATA) {
486 ASSERT(ip->i_afp->if_bytes > 0);
487 ASSERT(ip->i_afp->if_u1.if_data != NULL);
489 vecp->i_addr = ip->i_afp->if_u1.if_data;
491 * Round i_bytes up to a word boundary.
492 * The underlying memory is guaranteed to
493 * to be there by xfs_idata_realloc().
495 data_bytes = roundup(ip->i_afp->if_bytes, 4);
496 ASSERT((ip->i_afp->if_real_bytes == 0) ||
497 (ip->i_afp->if_real_bytes == data_bytes));
498 vecp->i_len = (int)data_bytes;
499 vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
500 vecp++;
501 nvecs++;
502 iip->ili_format.ilf_asize = (unsigned)data_bytes;
504 break;
506 default:
507 ASSERT(0);
508 break;
511 ASSERT(nvecs == lip->li_desc->lid_size);
512 iip->ili_format.ilf_size = nvecs;
517 * This is called to pin the inode associated with the inode log
518 * item in memory so it cannot be written out.
520 STATIC void
521 xfs_inode_item_pin(
522 struct xfs_log_item *lip)
524 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
526 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
528 trace_xfs_inode_pin(ip, _RET_IP_);
529 atomic_inc(&ip->i_pincount);
534 * This is called to unpin the inode associated with the inode log
535 * item which was previously pinned with a call to xfs_inode_item_pin().
537 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
539 STATIC void
540 xfs_inode_item_unpin(
541 struct xfs_log_item *lip,
542 int remove)
544 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
546 trace_xfs_inode_unpin(ip, _RET_IP_);
547 ASSERT(atomic_read(&ip->i_pincount) > 0);
548 if (atomic_dec_and_test(&ip->i_pincount))
549 wake_up(&ip->i_ipin_wait);
553 * This is called to attempt to lock the inode associated with this
554 * inode log item, in preparation for the push routine which does the actual
555 * iflush. Don't sleep on the inode lock or the flush lock.
557 * If the flush lock is already held, indicating that the inode has
558 * been or is in the process of being flushed, then (ideally) we'd like to
559 * see if the inode's buffer is still incore, and if so give it a nudge.
560 * We delay doing so until the pushbuf routine, though, to avoid holding
561 * the AIL lock across a call to the blackhole which is the buffer cache.
562 * Also we don't want to sleep in any device strategy routines, which can happen
563 * if we do the subsequent bawrite in here.
565 STATIC uint
566 xfs_inode_item_trylock(
567 struct xfs_log_item *lip)
569 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
570 struct xfs_inode *ip = iip->ili_inode;
572 if (xfs_ipincount(ip) > 0)
573 return XFS_ITEM_PINNED;
575 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
576 return XFS_ITEM_LOCKED;
578 if (!xfs_iflock_nowait(ip)) {
580 * inode has already been flushed to the backing buffer,
581 * leave it locked in shared mode, pushbuf routine will
582 * unlock it.
584 return XFS_ITEM_PUSHBUF;
587 /* Stale items should force out the iclog */
588 if (ip->i_flags & XFS_ISTALE) {
589 xfs_ifunlock(ip);
591 * we hold the AIL lock - notify the unlock routine of this
592 * so it doesn't try to get the lock again.
594 xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
595 return XFS_ITEM_PINNED;
598 #ifdef DEBUG
599 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
600 ASSERT(iip->ili_format.ilf_fields != 0);
601 ASSERT(iip->ili_logged == 0);
602 ASSERT(lip->li_flags & XFS_LI_IN_AIL);
604 #endif
605 return XFS_ITEM_SUCCESS;
609 * Unlock the inode associated with the inode log item.
610 * Clear the fields of the inode and inode log item that
611 * are specific to the current transaction. If the
612 * hold flags is set, do not unlock the inode.
614 STATIC void
615 xfs_inode_item_unlock(
616 struct xfs_log_item *lip)
618 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
619 struct xfs_inode *ip = iip->ili_inode;
620 unsigned short lock_flags;
622 ASSERT(iip->ili_inode->i_itemp != NULL);
623 ASSERT(xfs_isilocked(iip->ili_inode, XFS_ILOCK_EXCL));
626 * Clear the transaction pointer in the inode.
628 ip->i_transp = NULL;
631 * If the inode needed a separate buffer with which to log
632 * its extents, then free it now.
634 if (iip->ili_extents_buf != NULL) {
635 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
636 ASSERT(ip->i_d.di_nextents > 0);
637 ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_DEXT);
638 ASSERT(ip->i_df.if_bytes > 0);
639 kmem_free(iip->ili_extents_buf);
640 iip->ili_extents_buf = NULL;
642 if (iip->ili_aextents_buf != NULL) {
643 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
644 ASSERT(ip->i_d.di_anextents > 0);
645 ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_AEXT);
646 ASSERT(ip->i_afp->if_bytes > 0);
647 kmem_free(iip->ili_aextents_buf);
648 iip->ili_aextents_buf = NULL;
651 lock_flags = iip->ili_lock_flags;
652 iip->ili_lock_flags = 0;
653 if (lock_flags) {
654 xfs_iunlock(iip->ili_inode, lock_flags);
655 IRELE(iip->ili_inode);
660 * This is called to find out where the oldest active copy of the inode log
661 * item in the on disk log resides now that the last log write of it completed
662 * at the given lsn. Since we always re-log all dirty data in an inode, the
663 * latest copy in the on disk log is the only one that matters. Therefore,
664 * simply return the given lsn.
666 * If the inode has been marked stale because the cluster is being freed, we
667 * don't want to (re-)insert this inode into the AIL. There is a race condition
668 * where the cluster buffer may be unpinned before the inode is inserted into
669 * the AIL during transaction committed processing. If the buffer is unpinned
670 * before the inode item has been committed and inserted, then it is possible
671 * for the buffer to be written and IO completions before the inode is inserted
672 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
673 * AIL which will never get removed. It will, however, get reclaimed which
674 * triggers an assert in xfs_inode_free() complaining about freein an inode
675 * still in the AIL.
677 * To avoid this, return a lower LSN than the one passed in so that the
678 * transaction committed code will not move the inode forward in the AIL but
679 * will still unpin it properly.
681 STATIC xfs_lsn_t
682 xfs_inode_item_committed(
683 struct xfs_log_item *lip,
684 xfs_lsn_t lsn)
686 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
687 struct xfs_inode *ip = iip->ili_inode;
689 if (xfs_iflags_test(ip, XFS_ISTALE))
690 return lsn - 1;
691 return lsn;
695 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
696 * failed to get the inode flush lock but did get the inode locked SHARED.
697 * Here we're trying to see if the inode buffer is incore, and if so whether it's
698 * marked delayed write. If that's the case, we'll promote it and that will
699 * allow the caller to write the buffer by triggering the xfsbufd to run.
701 STATIC void
702 xfs_inode_item_pushbuf(
703 struct xfs_log_item *lip)
705 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
706 struct xfs_inode *ip = iip->ili_inode;
707 struct xfs_buf *bp;
709 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
712 * If a flush is not in progress anymore, chances are that the
713 * inode was taken off the AIL. So, just get out.
715 if (completion_done(&ip->i_flush) ||
716 !(lip->li_flags & XFS_LI_IN_AIL)) {
717 xfs_iunlock(ip, XFS_ILOCK_SHARED);
718 return;
721 bp = xfs_incore(ip->i_mount->m_ddev_targp, iip->ili_format.ilf_blkno,
722 iip->ili_format.ilf_len, XBF_TRYLOCK);
724 xfs_iunlock(ip, XFS_ILOCK_SHARED);
725 if (!bp)
726 return;
727 if (XFS_BUF_ISDELAYWRITE(bp))
728 xfs_buf_delwri_promote(bp);
729 xfs_buf_relse(bp);
733 * This is called to asynchronously write the inode associated with this
734 * inode log item out to disk. The inode will already have been locked by
735 * a successful call to xfs_inode_item_trylock().
737 STATIC void
738 xfs_inode_item_push(
739 struct xfs_log_item *lip)
741 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
742 struct xfs_inode *ip = iip->ili_inode;
744 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
745 ASSERT(!completion_done(&ip->i_flush));
748 * Since we were able to lock the inode's flush lock and
749 * we found it on the AIL, the inode must be dirty. This
750 * is because the inode is removed from the AIL while still
751 * holding the flush lock in xfs_iflush_done(). Thus, if
752 * we found it in the AIL and were able to obtain the flush
753 * lock without sleeping, then there must not have been
754 * anyone in the process of flushing the inode.
756 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
757 iip->ili_format.ilf_fields != 0);
760 * Push the inode to it's backing buffer. This will not remove the
761 * inode from the AIL - a further push will be required to trigger a
762 * buffer push. However, this allows all the dirty inodes to be pushed
763 * to the buffer before it is pushed to disk. THe buffer IO completion
764 * will pull th einode from the AIL, mark it clean and unlock the flush
765 * lock.
767 (void) xfs_iflush(ip, 0);
768 xfs_iunlock(ip, XFS_ILOCK_SHARED);
772 * XXX rcc - this one really has to do something. Probably needs
773 * to stamp in a new field in the incore inode.
775 STATIC void
776 xfs_inode_item_committing(
777 struct xfs_log_item *lip,
778 xfs_lsn_t lsn)
780 INODE_ITEM(lip)->ili_last_lsn = lsn;
784 * This is the ops vector shared by all buf log items.
786 static struct xfs_item_ops xfs_inode_item_ops = {
787 .iop_size = xfs_inode_item_size,
788 .iop_format = xfs_inode_item_format,
789 .iop_pin = xfs_inode_item_pin,
790 .iop_unpin = xfs_inode_item_unpin,
791 .iop_trylock = xfs_inode_item_trylock,
792 .iop_unlock = xfs_inode_item_unlock,
793 .iop_committed = xfs_inode_item_committed,
794 .iop_push = xfs_inode_item_push,
795 .iop_pushbuf = xfs_inode_item_pushbuf,
796 .iop_committing = xfs_inode_item_committing
801 * Initialize the inode log item for a newly allocated (in-core) inode.
803 void
804 xfs_inode_item_init(
805 struct xfs_inode *ip,
806 struct xfs_mount *mp)
808 struct xfs_inode_log_item *iip;
810 ASSERT(ip->i_itemp == NULL);
811 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
813 iip->ili_inode = ip;
814 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
815 &xfs_inode_item_ops);
816 iip->ili_format.ilf_type = XFS_LI_INODE;
817 iip->ili_format.ilf_ino = ip->i_ino;
818 iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
819 iip->ili_format.ilf_len = ip->i_imap.im_len;
820 iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
824 * Free the inode log item and any memory hanging off of it.
826 void
827 xfs_inode_item_destroy(
828 xfs_inode_t *ip)
830 #ifdef XFS_TRANS_DEBUG
831 if (ip->i_itemp->ili_root_size != 0) {
832 kmem_free(ip->i_itemp->ili_orig_root);
834 #endif
835 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
840 * This is the inode flushing I/O completion routine. It is called
841 * from interrupt level when the buffer containing the inode is
842 * flushed to disk. It is responsible for removing the inode item
843 * from the AIL if it has not been re-logged, and unlocking the inode's
844 * flush lock.
846 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
847 * list for other inodes that will run this function. We remove them from the
848 * buffer list so we can process all the inode IO completions in one AIL lock
849 * traversal.
851 void
852 xfs_iflush_done(
853 struct xfs_buf *bp,
854 struct xfs_log_item *lip)
856 struct xfs_inode_log_item *iip;
857 struct xfs_log_item *blip;
858 struct xfs_log_item *next;
859 struct xfs_log_item *prev;
860 struct xfs_ail *ailp = lip->li_ailp;
861 int need_ail = 0;
864 * Scan the buffer IO completions for other inodes being completed and
865 * attach them to the current inode log item.
867 blip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
868 prev = NULL;
869 while (blip != NULL) {
870 if (lip->li_cb != xfs_iflush_done) {
871 prev = blip;
872 blip = blip->li_bio_list;
873 continue;
876 /* remove from list */
877 next = blip->li_bio_list;
878 if (!prev) {
879 XFS_BUF_SET_FSPRIVATE(bp, next);
880 } else {
881 prev->li_bio_list = next;
884 /* add to current list */
885 blip->li_bio_list = lip->li_bio_list;
886 lip->li_bio_list = blip;
889 * while we have the item, do the unlocked check for needing
890 * the AIL lock.
892 iip = INODE_ITEM(blip);
893 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
894 need_ail++;
896 blip = next;
899 /* make sure we capture the state of the initial inode. */
900 iip = INODE_ITEM(lip);
901 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
902 need_ail++;
905 * We only want to pull the item from the AIL if it is
906 * actually there and its location in the log has not
907 * changed since we started the flush. Thus, we only bother
908 * if the ili_logged flag is set and the inode's lsn has not
909 * changed. First we check the lsn outside
910 * the lock since it's cheaper, and then we recheck while
911 * holding the lock before removing the inode from the AIL.
913 if (need_ail) {
914 struct xfs_log_item *log_items[need_ail];
915 int i = 0;
916 spin_lock(&ailp->xa_lock);
917 for (blip = lip; blip; blip = blip->li_bio_list) {
918 iip = INODE_ITEM(blip);
919 if (iip->ili_logged &&
920 blip->li_lsn == iip->ili_flush_lsn) {
921 log_items[i++] = blip;
923 ASSERT(i <= need_ail);
925 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
926 xfs_trans_ail_delete_bulk(ailp, log_items, i);
931 * clean up and unlock the flush lock now we are done. We can clear the
932 * ili_last_fields bits now that we know that the data corresponding to
933 * them is safely on disk.
935 for (blip = lip; blip; blip = next) {
936 next = blip->li_bio_list;
937 blip->li_bio_list = NULL;
939 iip = INODE_ITEM(blip);
940 iip->ili_logged = 0;
941 iip->ili_last_fields = 0;
942 xfs_ifunlock(iip->ili_inode);
947 * This is the inode flushing abort routine. It is called
948 * from xfs_iflush when the filesystem is shutting down to clean
949 * up the inode state.
950 * It is responsible for removing the inode item
951 * from the AIL if it has not been re-logged, and unlocking the inode's
952 * flush lock.
954 void
955 xfs_iflush_abort(
956 xfs_inode_t *ip)
958 xfs_inode_log_item_t *iip = ip->i_itemp;
960 iip = ip->i_itemp;
961 if (iip) {
962 struct xfs_ail *ailp = iip->ili_item.li_ailp;
963 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
964 spin_lock(&ailp->xa_lock);
965 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
966 /* xfs_trans_ail_delete() drops the AIL lock. */
967 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip);
968 } else
969 spin_unlock(&ailp->xa_lock);
971 iip->ili_logged = 0;
973 * Clear the ili_last_fields bits now that we know that the
974 * data corresponding to them is safely on disk.
976 iip->ili_last_fields = 0;
978 * Clear the inode logging fields so no more flushes are
979 * attempted.
981 iip->ili_format.ilf_fields = 0;
984 * Release the inode's flush lock since we're done with it.
986 xfs_ifunlock(ip);
989 void
990 xfs_istale_done(
991 struct xfs_buf *bp,
992 struct xfs_log_item *lip)
994 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode);
998 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
999 * (which can have different field alignments) to the native version
1002 xfs_inode_item_format_convert(
1003 xfs_log_iovec_t *buf,
1004 xfs_inode_log_format_t *in_f)
1006 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
1007 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
1009 in_f->ilf_type = in_f32->ilf_type;
1010 in_f->ilf_size = in_f32->ilf_size;
1011 in_f->ilf_fields = in_f32->ilf_fields;
1012 in_f->ilf_asize = in_f32->ilf_asize;
1013 in_f->ilf_dsize = in_f32->ilf_dsize;
1014 in_f->ilf_ino = in_f32->ilf_ino;
1015 /* copy biggest field of ilf_u */
1016 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1017 in_f32->ilf_u.ilfu_uuid.__u_bits,
1018 sizeof(uuid_t));
1019 in_f->ilf_blkno = in_f32->ilf_blkno;
1020 in_f->ilf_len = in_f32->ilf_len;
1021 in_f->ilf_boffset = in_f32->ilf_boffset;
1022 return 0;
1023 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
1024 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
1026 in_f->ilf_type = in_f64->ilf_type;
1027 in_f->ilf_size = in_f64->ilf_size;
1028 in_f->ilf_fields = in_f64->ilf_fields;
1029 in_f->ilf_asize = in_f64->ilf_asize;
1030 in_f->ilf_dsize = in_f64->ilf_dsize;
1031 in_f->ilf_ino = in_f64->ilf_ino;
1032 /* copy biggest field of ilf_u */
1033 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1034 in_f64->ilf_u.ilfu_uuid.__u_bits,
1035 sizeof(uuid_t));
1036 in_f->ilf_blkno = in_f64->ilf_blkno;
1037 in_f->ilf_len = in_f64->ilf_len;
1038 in_f->ilf_boffset = in_f64->ilf_boffset;
1039 return 0;
1041 return EFSCORRUPTED;