2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
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
20 #include "xfs_types.h"
24 #include "xfs_trans.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.
54 struct xfs_log_item
*lip
)
56 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
57 struct xfs_inode
*ip
= iip
->ili_inode
;
61 * Only log the data/extents/b-tree root if there is something
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
);
77 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_DEXT
;
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
);
92 ASSERT(!(iip
->ili_format
.ilf_fields
&
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
,
100 iip
->ili_root_size
) == 0);
102 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
105 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_DBROOT
;
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);
119 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_DDATA
;
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
);
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
);
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
);
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
);
164 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_AEXT
;
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
);
176 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_ABROOT
;
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
);
188 iip
->ili_format
.ilf_fields
&= ~XFS_ILOG_ADATA
;
201 * xfs_inode_item_format_extents - convert in-core extents to on-disk form
203 * For either the data or attr fork in extent format, we need to endian convert
204 * the in-core extent as we place them into the on-disk inode. In this case, we
205 * need to do this conversion before we write the extents into the log. Because
206 * we don't have the disk inode to write into here, we allocate a buffer and
207 * format the extents into it via xfs_iextents_copy(). We free the buffer in
208 * the unlock routine after the copy for the log has been made.
210 * In the case of the data fork, the in-core and on-disk fork sizes can be
211 * different due to delayed allocation extents. We only log on-disk extents
212 * here, so always use the physical fork size to determine the size of the
213 * buffer we need to allocate.
216 xfs_inode_item_format_extents(
217 struct xfs_inode
*ip
,
218 struct xfs_log_iovec
*vecp
,
222 xfs_bmbt_rec_t
*ext_buffer
;
224 ext_buffer
= kmem_alloc(XFS_IFORK_SIZE(ip
, whichfork
), KM_SLEEP
);
225 if (whichfork
== XFS_DATA_FORK
)
226 ip
->i_itemp
->ili_extents_buf
= ext_buffer
;
228 ip
->i_itemp
->ili_aextents_buf
= ext_buffer
;
230 vecp
->i_addr
= ext_buffer
;
231 vecp
->i_len
= xfs_iextents_copy(ip
, ext_buffer
, whichfork
);
236 * This is called to fill in the vector of log iovecs for the
237 * given inode log item. It fills the first item with an inode
238 * log format structure, the second with the on-disk inode structure,
239 * and a possible third and/or fourth with the inode data/extents/b-tree
240 * root and inode attributes data/extents/b-tree root.
243 xfs_inode_item_format(
244 struct xfs_log_item
*lip
,
245 struct xfs_log_iovec
*vecp
)
247 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
248 struct xfs_inode
*ip
= iip
->ili_inode
;
253 vecp
->i_addr
= &iip
->ili_format
;
254 vecp
->i_len
= sizeof(xfs_inode_log_format_t
);
255 vecp
->i_type
= XLOG_REG_TYPE_IFORMAT
;
260 * Clear i_update_core if the timestamps (or any other
261 * non-transactional modification) need flushing/logging
262 * and we're about to log them with the rest of the core.
264 * This is the same logic as xfs_iflush() but this code can't
265 * run at the same time as xfs_iflush because we're in commit
266 * processing here and so we have the inode lock held in
267 * exclusive mode. Although it doesn't really matter
268 * for the timestamps if both routines were to grab the
269 * timestamps or not. That would be ok.
271 * We clear i_update_core before copying out the data.
272 * This is for coordination with our timestamp updates
273 * that don't hold the inode lock. They will always
274 * update the timestamps BEFORE setting i_update_core,
275 * so if we clear i_update_core after they set it we
276 * are guaranteed to see their updates to the timestamps
277 * either here. Likewise, if they set it after we clear it
278 * here, we'll see it either on the next commit of this
279 * inode or the next time the inode gets flushed via
280 * xfs_iflush(). This depends on strongly ordered memory
281 * semantics, but we have that. We use the SYNCHRONIZE
282 * macro to make sure that the compiler does not reorder
283 * the i_update_core access below the data copy below.
285 if (ip
->i_update_core
) {
286 ip
->i_update_core
= 0;
291 * Make sure to get the latest timestamps from the Linux inode.
293 xfs_synchronize_times(ip
);
295 vecp
->i_addr
= &ip
->i_d
;
296 vecp
->i_len
= sizeof(struct xfs_icdinode
);
297 vecp
->i_type
= XLOG_REG_TYPE_ICORE
;
300 iip
->ili_format
.ilf_fields
|= XFS_ILOG_CORE
;
303 * If this is really an old format inode, then we need to
304 * log it as such. This means that we have to copy the link
305 * count from the new field to the old. We don't have to worry
306 * about the new fields, because nothing trusts them as long as
307 * the old inode version number is there. If the superblock already
308 * has a new version number, then we don't bother converting back.
311 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
312 if (ip
->i_d
.di_version
== 1) {
313 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
317 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
318 ip
->i_d
.di_onlink
= ip
->i_d
.di_nlink
;
321 * The superblock version has already been bumped,
322 * so just make the conversion to the new inode
325 ip
->i_d
.di_version
= 2;
326 ip
->i_d
.di_onlink
= 0;
327 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
331 switch (ip
->i_d
.di_format
) {
332 case XFS_DINODE_FMT_EXTENTS
:
333 ASSERT(!(iip
->ili_format
.ilf_fields
&
334 (XFS_ILOG_DDATA
| XFS_ILOG_DBROOT
|
335 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
336 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
) {
337 ASSERT(ip
->i_df
.if_bytes
> 0);
338 ASSERT(ip
->i_df
.if_u1
.if_extents
!= NULL
);
339 ASSERT(ip
->i_d
.di_nextents
> 0);
340 ASSERT(iip
->ili_extents_buf
== NULL
);
341 ASSERT((ip
->i_df
.if_bytes
/
342 (uint
)sizeof(xfs_bmbt_rec_t
)) > 0);
343 #ifdef XFS_NATIVE_HOST
344 if (ip
->i_d
.di_nextents
== ip
->i_df
.if_bytes
/
345 (uint
)sizeof(xfs_bmbt_rec_t
)) {
347 * There are no delayed allocation
348 * extents, so just point to the
349 * real extents array.
351 vecp
->i_addr
= ip
->i_df
.if_u1
.if_extents
;
352 vecp
->i_len
= ip
->i_df
.if_bytes
;
353 vecp
->i_type
= XLOG_REG_TYPE_IEXT
;
357 xfs_inode_item_format_extents(ip
, vecp
,
358 XFS_DATA_FORK
, XLOG_REG_TYPE_IEXT
);
360 ASSERT(vecp
->i_len
<= ip
->i_df
.if_bytes
);
361 iip
->ili_format
.ilf_dsize
= vecp
->i_len
;
367 case XFS_DINODE_FMT_BTREE
:
368 ASSERT(!(iip
->ili_format
.ilf_fields
&
369 (XFS_ILOG_DDATA
| XFS_ILOG_DEXT
|
370 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
371 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DBROOT
) {
372 ASSERT(ip
->i_df
.if_broot_bytes
> 0);
373 ASSERT(ip
->i_df
.if_broot
!= NULL
);
374 vecp
->i_addr
= ip
->i_df
.if_broot
;
375 vecp
->i_len
= ip
->i_df
.if_broot_bytes
;
376 vecp
->i_type
= XLOG_REG_TYPE_IBROOT
;
379 iip
->ili_format
.ilf_dsize
= ip
->i_df
.if_broot_bytes
;
383 case XFS_DINODE_FMT_LOCAL
:
384 ASSERT(!(iip
->ili_format
.ilf_fields
&
385 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
386 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
387 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DDATA
) {
388 ASSERT(ip
->i_df
.if_bytes
> 0);
389 ASSERT(ip
->i_df
.if_u1
.if_data
!= NULL
);
390 ASSERT(ip
->i_d
.di_size
> 0);
392 vecp
->i_addr
= ip
->i_df
.if_u1
.if_data
;
394 * Round i_bytes up to a word boundary.
395 * The underlying memory is guaranteed to
396 * to be there by xfs_idata_realloc().
398 data_bytes
= roundup(ip
->i_df
.if_bytes
, 4);
399 ASSERT((ip
->i_df
.if_real_bytes
== 0) ||
400 (ip
->i_df
.if_real_bytes
== data_bytes
));
401 vecp
->i_len
= (int)data_bytes
;
402 vecp
->i_type
= XLOG_REG_TYPE_ILOCAL
;
405 iip
->ili_format
.ilf_dsize
= (unsigned)data_bytes
;
409 case XFS_DINODE_FMT_DEV
:
410 ASSERT(!(iip
->ili_format
.ilf_fields
&
411 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
412 XFS_ILOG_DDATA
| XFS_ILOG_UUID
)));
413 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
414 iip
->ili_format
.ilf_u
.ilfu_rdev
=
415 ip
->i_df
.if_u2
.if_rdev
;
419 case XFS_DINODE_FMT_UUID
:
420 ASSERT(!(iip
->ili_format
.ilf_fields
&
421 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
422 XFS_ILOG_DDATA
| XFS_ILOG_DEV
)));
423 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
424 iip
->ili_format
.ilf_u
.ilfu_uuid
=
425 ip
->i_df
.if_u2
.if_uuid
;
435 * If there are no attributes associated with the file,
437 * Assert that no attribute-related log flags are set.
439 if (!XFS_IFORK_Q(ip
)) {
440 iip
->ili_format
.ilf_size
= nvecs
;
441 ASSERT(!(iip
->ili_format
.ilf_fields
&
442 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
446 switch (ip
->i_d
.di_aformat
) {
447 case XFS_DINODE_FMT_EXTENTS
:
448 ASSERT(!(iip
->ili_format
.ilf_fields
&
449 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
)));
450 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
) {
452 int nrecs
= ip
->i_afp
->if_bytes
/
453 (uint
)sizeof(xfs_bmbt_rec_t
);
455 ASSERT(nrecs
== ip
->i_d
.di_anextents
);
456 ASSERT(ip
->i_afp
->if_bytes
> 0);
457 ASSERT(ip
->i_afp
->if_u1
.if_extents
!= NULL
);
458 ASSERT(ip
->i_d
.di_anextents
> 0);
460 #ifdef XFS_NATIVE_HOST
462 * There are not delayed allocation extents
463 * for attributes, so just point at the array.
465 vecp
->i_addr
= ip
->i_afp
->if_u1
.if_extents
;
466 vecp
->i_len
= ip
->i_afp
->if_bytes
;
467 vecp
->i_type
= XLOG_REG_TYPE_IATTR_EXT
;
469 ASSERT(iip
->ili_aextents_buf
== NULL
);
470 xfs_inode_item_format_extents(ip
, vecp
,
471 XFS_ATTR_FORK
, XLOG_REG_TYPE_IATTR_EXT
);
473 iip
->ili_format
.ilf_asize
= vecp
->i_len
;
479 case XFS_DINODE_FMT_BTREE
:
480 ASSERT(!(iip
->ili_format
.ilf_fields
&
481 (XFS_ILOG_ADATA
| XFS_ILOG_AEXT
)));
482 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ABROOT
) {
483 ASSERT(ip
->i_afp
->if_broot_bytes
> 0);
484 ASSERT(ip
->i_afp
->if_broot
!= NULL
);
485 vecp
->i_addr
= ip
->i_afp
->if_broot
;
486 vecp
->i_len
= ip
->i_afp
->if_broot_bytes
;
487 vecp
->i_type
= XLOG_REG_TYPE_IATTR_BROOT
;
490 iip
->ili_format
.ilf_asize
= ip
->i_afp
->if_broot_bytes
;
494 case XFS_DINODE_FMT_LOCAL
:
495 ASSERT(!(iip
->ili_format
.ilf_fields
&
496 (XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
497 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ADATA
) {
498 ASSERT(ip
->i_afp
->if_bytes
> 0);
499 ASSERT(ip
->i_afp
->if_u1
.if_data
!= NULL
);
501 vecp
->i_addr
= ip
->i_afp
->if_u1
.if_data
;
503 * Round i_bytes up to a word boundary.
504 * The underlying memory is guaranteed to
505 * to be there by xfs_idata_realloc().
507 data_bytes
= roundup(ip
->i_afp
->if_bytes
, 4);
508 ASSERT((ip
->i_afp
->if_real_bytes
== 0) ||
509 (ip
->i_afp
->if_real_bytes
== data_bytes
));
510 vecp
->i_len
= (int)data_bytes
;
511 vecp
->i_type
= XLOG_REG_TYPE_IATTR_LOCAL
;
514 iip
->ili_format
.ilf_asize
= (unsigned)data_bytes
;
523 iip
->ili_format
.ilf_size
= nvecs
;
528 * This is called to pin the inode associated with the inode log
529 * item in memory so it cannot be written out.
533 struct xfs_log_item
*lip
)
535 struct xfs_inode
*ip
= INODE_ITEM(lip
)->ili_inode
;
537 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
539 trace_xfs_inode_pin(ip
, _RET_IP_
);
540 atomic_inc(&ip
->i_pincount
);
545 * This is called to unpin the inode associated with the inode log
546 * item which was previously pinned with a call to xfs_inode_item_pin().
548 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
551 xfs_inode_item_unpin(
552 struct xfs_log_item
*lip
,
555 struct xfs_inode
*ip
= INODE_ITEM(lip
)->ili_inode
;
557 trace_xfs_inode_unpin(ip
, _RET_IP_
);
558 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
559 if (atomic_dec_and_test(&ip
->i_pincount
))
560 wake_up(&ip
->i_ipin_wait
);
564 * This is called to attempt to lock the inode associated with this
565 * inode log item, in preparation for the push routine which does the actual
566 * iflush. Don't sleep on the inode lock or the flush lock.
568 * If the flush lock is already held, indicating that the inode has
569 * been or is in the process of being flushed, then (ideally) we'd like to
570 * see if the inode's buffer is still incore, and if so give it a nudge.
571 * We delay doing so until the pushbuf routine, though, to avoid holding
572 * the AIL lock across a call to the blackhole which is the buffer cache.
573 * Also we don't want to sleep in any device strategy routines, which can happen
574 * if we do the subsequent bawrite in here.
577 xfs_inode_item_trylock(
578 struct xfs_log_item
*lip
)
580 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
581 struct xfs_inode
*ip
= iip
->ili_inode
;
583 if (xfs_ipincount(ip
) > 0)
584 return XFS_ITEM_PINNED
;
586 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
))
587 return XFS_ITEM_LOCKED
;
589 if (!xfs_iflock_nowait(ip
)) {
591 * inode has already been flushed to the backing buffer,
592 * leave it locked in shared mode, pushbuf routine will
595 return XFS_ITEM_PUSHBUF
;
598 /* Stale items should force out the iclog */
599 if (ip
->i_flags
& XFS_ISTALE
) {
602 * we hold the AIL lock - notify the unlock routine of this
603 * so it doesn't try to get the lock again.
605 xfs_iunlock(ip
, XFS_ILOCK_SHARED
|XFS_IUNLOCK_NONOTIFY
);
606 return XFS_ITEM_PINNED
;
610 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
611 ASSERT(iip
->ili_format
.ilf_fields
!= 0);
612 ASSERT(iip
->ili_logged
== 0);
613 ASSERT(lip
->li_flags
& XFS_LI_IN_AIL
);
616 return XFS_ITEM_SUCCESS
;
620 * Unlock the inode associated with the inode log item.
621 * Clear the fields of the inode and inode log item that
622 * are specific to the current transaction. If the
623 * hold flags is set, do not unlock the inode.
626 xfs_inode_item_unlock(
627 struct xfs_log_item
*lip
)
629 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
630 struct xfs_inode
*ip
= iip
->ili_inode
;
631 unsigned short lock_flags
;
633 ASSERT(ip
->i_itemp
!= NULL
);
634 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
637 * If the inode needed a separate buffer with which to log
638 * its extents, then free it now.
640 if (iip
->ili_extents_buf
!= NULL
) {
641 ASSERT(ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
);
642 ASSERT(ip
->i_d
.di_nextents
> 0);
643 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
);
644 ASSERT(ip
->i_df
.if_bytes
> 0);
645 kmem_free(iip
->ili_extents_buf
);
646 iip
->ili_extents_buf
= NULL
;
648 if (iip
->ili_aextents_buf
!= NULL
) {
649 ASSERT(ip
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
);
650 ASSERT(ip
->i_d
.di_anextents
> 0);
651 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
);
652 ASSERT(ip
->i_afp
->if_bytes
> 0);
653 kmem_free(iip
->ili_aextents_buf
);
654 iip
->ili_aextents_buf
= NULL
;
657 lock_flags
= iip
->ili_lock_flags
;
658 iip
->ili_lock_flags
= 0;
660 xfs_iunlock(ip
, lock_flags
);
664 * This is called to find out where the oldest active copy of the inode log
665 * item in the on disk log resides now that the last log write of it completed
666 * at the given lsn. Since we always re-log all dirty data in an inode, the
667 * latest copy in the on disk log is the only one that matters. Therefore,
668 * simply return the given lsn.
670 * If the inode has been marked stale because the cluster is being freed, we
671 * don't want to (re-)insert this inode into the AIL. There is a race condition
672 * where the cluster buffer may be unpinned before the inode is inserted into
673 * the AIL during transaction committed processing. If the buffer is unpinned
674 * before the inode item has been committed and inserted, then it is possible
675 * for the buffer to be written and IO completes before the inode is inserted
676 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
677 * AIL which will never get removed. It will, however, get reclaimed which
678 * triggers an assert in xfs_inode_free() complaining about freein an inode
681 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
682 * transaction committed code knows that it does not need to do any further
683 * processing on the item.
686 xfs_inode_item_committed(
687 struct xfs_log_item
*lip
,
690 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
691 struct xfs_inode
*ip
= iip
->ili_inode
;
693 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
694 xfs_inode_item_unpin(lip
, 0);
701 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
702 * failed to get the inode flush lock but did get the inode locked SHARED.
703 * Here we're trying to see if the inode buffer is incore, and if so whether it's
704 * marked delayed write. If that's the case, we'll promote it and that will
705 * allow the caller to write the buffer by triggering the xfsbufd to run.
708 xfs_inode_item_pushbuf(
709 struct xfs_log_item
*lip
)
711 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
712 struct xfs_inode
*ip
= iip
->ili_inode
;
716 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
719 * If a flush is not in progress anymore, chances are that the
720 * inode was taken off the AIL. So, just get out.
722 if (completion_done(&ip
->i_flush
) ||
723 !(lip
->li_flags
& XFS_LI_IN_AIL
)) {
724 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
728 bp
= xfs_incore(ip
->i_mount
->m_ddev_targp
, iip
->ili_format
.ilf_blkno
,
729 iip
->ili_format
.ilf_len
, XBF_TRYLOCK
);
731 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
734 if (XFS_BUF_ISDELAYWRITE(bp
))
735 xfs_buf_delwri_promote(bp
);
736 if (xfs_buf_ispinned(bp
))
743 * This is called to asynchronously write the inode associated with this
744 * inode log item out to disk. The inode will already have been locked by
745 * a successful call to xfs_inode_item_trylock().
749 struct xfs_log_item
*lip
)
751 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
752 struct xfs_inode
*ip
= iip
->ili_inode
;
754 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
755 ASSERT(!completion_done(&ip
->i_flush
));
758 * Since we were able to lock the inode's flush lock and
759 * we found it on the AIL, the inode must be dirty. This
760 * is because the inode is removed from the AIL while still
761 * holding the flush lock in xfs_iflush_done(). Thus, if
762 * we found it in the AIL and were able to obtain the flush
763 * lock without sleeping, then there must not have been
764 * anyone in the process of flushing the inode.
766 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) ||
767 iip
->ili_format
.ilf_fields
!= 0);
770 * Push the inode to it's backing buffer. This will not remove the
771 * inode from the AIL - a further push will be required to trigger a
772 * buffer push. However, this allows all the dirty inodes to be pushed
773 * to the buffer before it is pushed to disk. The buffer IO completion
774 * will pull the inode from the AIL, mark it clean and unlock the flush
777 (void) xfs_iflush(ip
, SYNC_TRYLOCK
);
778 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
782 * XXX rcc - this one really has to do something. Probably needs
783 * to stamp in a new field in the incore inode.
786 xfs_inode_item_committing(
787 struct xfs_log_item
*lip
,
790 INODE_ITEM(lip
)->ili_last_lsn
= lsn
;
794 * This is the ops vector shared by all buf log items.
796 static const struct xfs_item_ops xfs_inode_item_ops
= {
797 .iop_size
= xfs_inode_item_size
,
798 .iop_format
= xfs_inode_item_format
,
799 .iop_pin
= xfs_inode_item_pin
,
800 .iop_unpin
= xfs_inode_item_unpin
,
801 .iop_trylock
= xfs_inode_item_trylock
,
802 .iop_unlock
= xfs_inode_item_unlock
,
803 .iop_committed
= xfs_inode_item_committed
,
804 .iop_push
= xfs_inode_item_push
,
805 .iop_pushbuf
= xfs_inode_item_pushbuf
,
806 .iop_committing
= xfs_inode_item_committing
811 * Initialize the inode log item for a newly allocated (in-core) inode.
815 struct xfs_inode
*ip
,
816 struct xfs_mount
*mp
)
818 struct xfs_inode_log_item
*iip
;
820 ASSERT(ip
->i_itemp
== NULL
);
821 iip
= ip
->i_itemp
= kmem_zone_zalloc(xfs_ili_zone
, KM_SLEEP
);
824 xfs_log_item_init(mp
, &iip
->ili_item
, XFS_LI_INODE
,
825 &xfs_inode_item_ops
);
826 iip
->ili_format
.ilf_type
= XFS_LI_INODE
;
827 iip
->ili_format
.ilf_ino
= ip
->i_ino
;
828 iip
->ili_format
.ilf_blkno
= ip
->i_imap
.im_blkno
;
829 iip
->ili_format
.ilf_len
= ip
->i_imap
.im_len
;
830 iip
->ili_format
.ilf_boffset
= ip
->i_imap
.im_boffset
;
834 * Free the inode log item and any memory hanging off of it.
837 xfs_inode_item_destroy(
840 #ifdef XFS_TRANS_DEBUG
841 if (ip
->i_itemp
->ili_root_size
!= 0) {
842 kmem_free(ip
->i_itemp
->ili_orig_root
);
845 kmem_zone_free(xfs_ili_zone
, ip
->i_itemp
);
850 * This is the inode flushing I/O completion routine. It is called
851 * from interrupt level when the buffer containing the inode is
852 * flushed to disk. It is responsible for removing the inode item
853 * from the AIL if it has not been re-logged, and unlocking the inode's
856 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
857 * list for other inodes that will run this function. We remove them from the
858 * buffer list so we can process all the inode IO completions in one AIL lock
864 struct xfs_log_item
*lip
)
866 struct xfs_inode_log_item
*iip
;
867 struct xfs_log_item
*blip
;
868 struct xfs_log_item
*next
;
869 struct xfs_log_item
*prev
;
870 struct xfs_ail
*ailp
= lip
->li_ailp
;
874 * Scan the buffer IO completions for other inodes being completed and
875 * attach them to the current inode log item.
879 while (blip
!= NULL
) {
880 if (lip
->li_cb
!= xfs_iflush_done
) {
882 blip
= blip
->li_bio_list
;
886 /* remove from list */
887 next
= blip
->li_bio_list
;
891 prev
->li_bio_list
= next
;
894 /* add to current list */
895 blip
->li_bio_list
= lip
->li_bio_list
;
896 lip
->li_bio_list
= blip
;
899 * while we have the item, do the unlocked check for needing
902 iip
= INODE_ITEM(blip
);
903 if (iip
->ili_logged
&& blip
->li_lsn
== iip
->ili_flush_lsn
)
909 /* make sure we capture the state of the initial inode. */
910 iip
= INODE_ITEM(lip
);
911 if (iip
->ili_logged
&& lip
->li_lsn
== iip
->ili_flush_lsn
)
915 * We only want to pull the item from the AIL if it is
916 * actually there and its location in the log has not
917 * changed since we started the flush. Thus, we only bother
918 * if the ili_logged flag is set and the inode's lsn has not
919 * changed. First we check the lsn outside
920 * the lock since it's cheaper, and then we recheck while
921 * holding the lock before removing the inode from the AIL.
924 struct xfs_log_item
*log_items
[need_ail
];
926 spin_lock(&ailp
->xa_lock
);
927 for (blip
= lip
; blip
; blip
= blip
->li_bio_list
) {
928 iip
= INODE_ITEM(blip
);
929 if (iip
->ili_logged
&&
930 blip
->li_lsn
== iip
->ili_flush_lsn
) {
931 log_items
[i
++] = blip
;
933 ASSERT(i
<= need_ail
);
935 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
936 xfs_trans_ail_delete_bulk(ailp
, log_items
, i
);
941 * clean up and unlock the flush lock now we are done. We can clear the
942 * ili_last_fields bits now that we know that the data corresponding to
943 * them is safely on disk.
945 for (blip
= lip
; blip
; blip
= next
) {
946 next
= blip
->li_bio_list
;
947 blip
->li_bio_list
= NULL
;
949 iip
= INODE_ITEM(blip
);
951 iip
->ili_last_fields
= 0;
952 xfs_ifunlock(iip
->ili_inode
);
957 * This is the inode flushing abort routine. It is called
958 * from xfs_iflush when the filesystem is shutting down to clean
959 * up the inode state.
960 * It is responsible for removing the inode item
961 * from the AIL if it has not been re-logged, and unlocking the inode's
968 xfs_inode_log_item_t
*iip
= ip
->i_itemp
;
971 struct xfs_ail
*ailp
= iip
->ili_item
.li_ailp
;
972 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
973 spin_lock(&ailp
->xa_lock
);
974 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
975 /* xfs_trans_ail_delete() drops the AIL lock. */
976 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)iip
);
978 spin_unlock(&ailp
->xa_lock
);
982 * Clear the ili_last_fields bits now that we know that the
983 * data corresponding to them is safely on disk.
985 iip
->ili_last_fields
= 0;
987 * Clear the inode logging fields so no more flushes are
990 iip
->ili_format
.ilf_fields
= 0;
993 * Release the inode's flush lock since we're done with it.
1001 struct xfs_log_item
*lip
)
1003 xfs_iflush_abort(INODE_ITEM(lip
)->ili_inode
);
1007 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
1008 * (which can have different field alignments) to the native version
1011 xfs_inode_item_format_convert(
1012 xfs_log_iovec_t
*buf
,
1013 xfs_inode_log_format_t
*in_f
)
1015 if (buf
->i_len
== sizeof(xfs_inode_log_format_32_t
)) {
1016 xfs_inode_log_format_32_t
*in_f32
= buf
->i_addr
;
1018 in_f
->ilf_type
= in_f32
->ilf_type
;
1019 in_f
->ilf_size
= in_f32
->ilf_size
;
1020 in_f
->ilf_fields
= in_f32
->ilf_fields
;
1021 in_f
->ilf_asize
= in_f32
->ilf_asize
;
1022 in_f
->ilf_dsize
= in_f32
->ilf_dsize
;
1023 in_f
->ilf_ino
= in_f32
->ilf_ino
;
1024 /* copy biggest field of ilf_u */
1025 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
1026 in_f32
->ilf_u
.ilfu_uuid
.__u_bits
,
1028 in_f
->ilf_blkno
= in_f32
->ilf_blkno
;
1029 in_f
->ilf_len
= in_f32
->ilf_len
;
1030 in_f
->ilf_boffset
= in_f32
->ilf_boffset
;
1032 } else if (buf
->i_len
== sizeof(xfs_inode_log_format_64_t
)){
1033 xfs_inode_log_format_64_t
*in_f64
= buf
->i_addr
;
1035 in_f
->ilf_type
= in_f64
->ilf_type
;
1036 in_f
->ilf_size
= in_f64
->ilf_size
;
1037 in_f
->ilf_fields
= in_f64
->ilf_fields
;
1038 in_f
->ilf_asize
= in_f64
->ilf_asize
;
1039 in_f
->ilf_dsize
= in_f64
->ilf_dsize
;
1040 in_f
->ilf_ino
= in_f64
->ilf_ino
;
1041 /* copy biggest field of ilf_u */
1042 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
1043 in_f64
->ilf_u
.ilfu_uuid
.__u_bits
,
1045 in_f
->ilf_blkno
= in_f64
->ilf_blkno
;
1046 in_f
->ilf_len
= in_f64
->ilf_len
;
1047 in_f
->ilf_boffset
= in_f64
->ilf_boffset
;
1050 return EFSCORRUPTED
;