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 * 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.
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
;
216 xfs_bmbt_rec_t
*ext_buffer
;
220 vecp
->i_addr
= (xfs_caddr_t
)&iip
->ili_format
;
221 vecp
->i_len
= sizeof(xfs_inode_log_format_t
);
222 vecp
->i_type
= XLOG_REG_TYPE_IFORMAT
;
227 * Make sure the linux inode is dirty. We do this before
228 * clearing i_update_core as the VFS will call back into
229 * XFS here and set i_update_core, so we need to dirty the
230 * inode first so that the ordering of i_update_core and
231 * unlogged modifications still works as described below.
233 xfs_mark_inode_dirty_sync(ip
);
236 * Clear i_update_core if the timestamps (or any other
237 * non-transactional modification) need flushing/logging
238 * and we're about to log them with the rest of the core.
240 * This is the same logic as xfs_iflush() but this code can't
241 * run at the same time as xfs_iflush because we're in commit
242 * processing here and so we have the inode lock held in
243 * exclusive mode. Although it doesn't really matter
244 * for the timestamps if both routines were to grab the
245 * timestamps or not. That would be ok.
247 * We clear i_update_core before copying out the data.
248 * This is for coordination with our timestamp updates
249 * that don't hold the inode lock. They will always
250 * update the timestamps BEFORE setting i_update_core,
251 * so if we clear i_update_core after they set it we
252 * are guaranteed to see their updates to the timestamps
253 * either here. Likewise, if they set it after we clear it
254 * here, we'll see it either on the next commit of this
255 * inode or the next time the inode gets flushed via
256 * xfs_iflush(). This depends on strongly ordered memory
257 * semantics, but we have that. We use the SYNCHRONIZE
258 * macro to make sure that the compiler does not reorder
259 * the i_update_core access below the data copy below.
261 if (ip
->i_update_core
) {
262 ip
->i_update_core
= 0;
267 * Make sure to get the latest timestamps from the Linux inode.
269 xfs_synchronize_times(ip
);
271 vecp
->i_addr
= (xfs_caddr_t
)&ip
->i_d
;
272 vecp
->i_len
= sizeof(struct xfs_icdinode
);
273 vecp
->i_type
= XLOG_REG_TYPE_ICORE
;
276 iip
->ili_format
.ilf_fields
|= XFS_ILOG_CORE
;
279 * If this is really an old format inode, then we need to
280 * log it as such. This means that we have to copy the link
281 * count from the new field to the old. We don't have to worry
282 * about the new fields, because nothing trusts them as long as
283 * the old inode version number is there. If the superblock already
284 * has a new version number, then we don't bother converting back.
287 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
288 if (ip
->i_d
.di_version
== 1) {
289 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
293 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
294 ip
->i_d
.di_onlink
= ip
->i_d
.di_nlink
;
297 * The superblock version has already been bumped,
298 * so just make the conversion to the new inode
301 ip
->i_d
.di_version
= 2;
302 ip
->i_d
.di_onlink
= 0;
303 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
307 switch (ip
->i_d
.di_format
) {
308 case XFS_DINODE_FMT_EXTENTS
:
309 ASSERT(!(iip
->ili_format
.ilf_fields
&
310 (XFS_ILOG_DDATA
| XFS_ILOG_DBROOT
|
311 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
312 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
) {
313 ASSERT(ip
->i_df
.if_bytes
> 0);
314 ASSERT(ip
->i_df
.if_u1
.if_extents
!= NULL
);
315 ASSERT(ip
->i_d
.di_nextents
> 0);
316 ASSERT(iip
->ili_extents_buf
== NULL
);
317 nrecs
= ip
->i_df
.if_bytes
/
318 (uint
)sizeof(xfs_bmbt_rec_t
);
320 #ifdef XFS_NATIVE_HOST
321 if (nrecs
== ip
->i_d
.di_nextents
) {
323 * There are no delayed allocation
324 * extents, so just point to the
325 * real extents array.
328 (char *)(ip
->i_df
.if_u1
.if_extents
);
329 vecp
->i_len
= ip
->i_df
.if_bytes
;
330 vecp
->i_type
= XLOG_REG_TYPE_IEXT
;
335 * There are delayed allocation extents
336 * in the inode, or we need to convert
337 * the extents to on disk format.
338 * Use xfs_iextents_copy()
339 * to copy only the real extents into
340 * a separate buffer. We'll free the
341 * buffer in the unlock routine.
343 ext_buffer
= kmem_alloc(ip
->i_df
.if_bytes
,
345 iip
->ili_extents_buf
= ext_buffer
;
346 vecp
->i_addr
= (xfs_caddr_t
)ext_buffer
;
347 vecp
->i_len
= xfs_iextents_copy(ip
, ext_buffer
,
349 vecp
->i_type
= XLOG_REG_TYPE_IEXT
;
351 ASSERT(vecp
->i_len
<= ip
->i_df
.if_bytes
);
352 iip
->ili_format
.ilf_dsize
= vecp
->i_len
;
358 case XFS_DINODE_FMT_BTREE
:
359 ASSERT(!(iip
->ili_format
.ilf_fields
&
360 (XFS_ILOG_DDATA
| XFS_ILOG_DEXT
|
361 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
362 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DBROOT
) {
363 ASSERT(ip
->i_df
.if_broot_bytes
> 0);
364 ASSERT(ip
->i_df
.if_broot
!= NULL
);
365 vecp
->i_addr
= (xfs_caddr_t
)ip
->i_df
.if_broot
;
366 vecp
->i_len
= ip
->i_df
.if_broot_bytes
;
367 vecp
->i_type
= XLOG_REG_TYPE_IBROOT
;
370 iip
->ili_format
.ilf_dsize
= ip
->i_df
.if_broot_bytes
;
374 case XFS_DINODE_FMT_LOCAL
:
375 ASSERT(!(iip
->ili_format
.ilf_fields
&
376 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
377 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
378 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DDATA
) {
379 ASSERT(ip
->i_df
.if_bytes
> 0);
380 ASSERT(ip
->i_df
.if_u1
.if_data
!= NULL
);
381 ASSERT(ip
->i_d
.di_size
> 0);
383 vecp
->i_addr
= (xfs_caddr_t
)ip
->i_df
.if_u1
.if_data
;
385 * Round i_bytes up to a word boundary.
386 * The underlying memory is guaranteed to
387 * to be there by xfs_idata_realloc().
389 data_bytes
= roundup(ip
->i_df
.if_bytes
, 4);
390 ASSERT((ip
->i_df
.if_real_bytes
== 0) ||
391 (ip
->i_df
.if_real_bytes
== data_bytes
));
392 vecp
->i_len
= (int)data_bytes
;
393 vecp
->i_type
= XLOG_REG_TYPE_ILOCAL
;
396 iip
->ili_format
.ilf_dsize
= (unsigned)data_bytes
;
400 case XFS_DINODE_FMT_DEV
:
401 ASSERT(!(iip
->ili_format
.ilf_fields
&
402 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
403 XFS_ILOG_DDATA
| XFS_ILOG_UUID
)));
404 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
405 iip
->ili_format
.ilf_u
.ilfu_rdev
=
406 ip
->i_df
.if_u2
.if_rdev
;
410 case XFS_DINODE_FMT_UUID
:
411 ASSERT(!(iip
->ili_format
.ilf_fields
&
412 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
413 XFS_ILOG_DDATA
| XFS_ILOG_DEV
)));
414 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
415 iip
->ili_format
.ilf_u
.ilfu_uuid
=
416 ip
->i_df
.if_u2
.if_uuid
;
426 * If there are no attributes associated with the file,
428 * Assert that no attribute-related log flags are set.
430 if (!XFS_IFORK_Q(ip
)) {
431 ASSERT(nvecs
== lip
->li_desc
->lid_size
);
432 iip
->ili_format
.ilf_size
= nvecs
;
433 ASSERT(!(iip
->ili_format
.ilf_fields
&
434 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
438 switch (ip
->i_d
.di_aformat
) {
439 case XFS_DINODE_FMT_EXTENTS
:
440 ASSERT(!(iip
->ili_format
.ilf_fields
&
441 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
)));
442 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
) {
443 ASSERT(ip
->i_afp
->if_bytes
> 0);
444 ASSERT(ip
->i_afp
->if_u1
.if_extents
!= NULL
);
445 ASSERT(ip
->i_d
.di_anextents
> 0);
447 nrecs
= ip
->i_afp
->if_bytes
/
448 (uint
)sizeof(xfs_bmbt_rec_t
);
451 ASSERT(nrecs
== ip
->i_d
.di_anextents
);
452 #ifdef XFS_NATIVE_HOST
454 * There are not delayed allocation extents
455 * for attributes, so just point at the array.
457 vecp
->i_addr
= (char *)(ip
->i_afp
->if_u1
.if_extents
);
458 vecp
->i_len
= ip
->i_afp
->if_bytes
;
460 ASSERT(iip
->ili_aextents_buf
== NULL
);
462 * Need to endian flip before logging
464 ext_buffer
= kmem_alloc(ip
->i_afp
->if_bytes
,
466 iip
->ili_aextents_buf
= ext_buffer
;
467 vecp
->i_addr
= (xfs_caddr_t
)ext_buffer
;
468 vecp
->i_len
= xfs_iextents_copy(ip
, ext_buffer
,
471 vecp
->i_type
= XLOG_REG_TYPE_IATTR_EXT
;
472 iip
->ili_format
.ilf_asize
= vecp
->i_len
;
478 case XFS_DINODE_FMT_BTREE
:
479 ASSERT(!(iip
->ili_format
.ilf_fields
&
480 (XFS_ILOG_ADATA
| XFS_ILOG_AEXT
)));
481 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ABROOT
) {
482 ASSERT(ip
->i_afp
->if_broot_bytes
> 0);
483 ASSERT(ip
->i_afp
->if_broot
!= NULL
);
484 vecp
->i_addr
= (xfs_caddr_t
)ip
->i_afp
->if_broot
;
485 vecp
->i_len
= ip
->i_afp
->if_broot_bytes
;
486 vecp
->i_type
= XLOG_REG_TYPE_IATTR_BROOT
;
489 iip
->ili_format
.ilf_asize
= ip
->i_afp
->if_broot_bytes
;
493 case XFS_DINODE_FMT_LOCAL
:
494 ASSERT(!(iip
->ili_format
.ilf_fields
&
495 (XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
496 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ADATA
) {
497 ASSERT(ip
->i_afp
->if_bytes
> 0);
498 ASSERT(ip
->i_afp
->if_u1
.if_data
!= NULL
);
500 vecp
->i_addr
= (xfs_caddr_t
)ip
->i_afp
->if_u1
.if_data
;
502 * Round i_bytes up to a word boundary.
503 * The underlying memory is guaranteed to
504 * to be there by xfs_idata_realloc().
506 data_bytes
= roundup(ip
->i_afp
->if_bytes
, 4);
507 ASSERT((ip
->i_afp
->if_real_bytes
== 0) ||
508 (ip
->i_afp
->if_real_bytes
== data_bytes
));
509 vecp
->i_len
= (int)data_bytes
;
510 vecp
->i_type
= XLOG_REG_TYPE_IATTR_LOCAL
;
513 iip
->ili_format
.ilf_asize
= (unsigned)data_bytes
;
522 ASSERT(nvecs
== lip
->li_desc
->lid_size
);
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
;
636 ASSERT(iip
->ili_inode
->i_itemp
!= NULL
);
637 ASSERT(xfs_isilocked(iip
->ili_inode
, XFS_ILOCK_EXCL
));
638 ASSERT((!(iip
->ili_inode
->i_itemp
->ili_flags
&
639 XFS_ILI_IOLOCKED_EXCL
)) ||
640 xfs_isilocked(iip
->ili_inode
, XFS_IOLOCK_EXCL
));
641 ASSERT((!(iip
->ili_inode
->i_itemp
->ili_flags
&
642 XFS_ILI_IOLOCKED_SHARED
)) ||
643 xfs_isilocked(iip
->ili_inode
, XFS_IOLOCK_SHARED
));
646 * Clear the transaction pointer in the inode.
651 * If the inode needed a separate buffer with which to log
652 * its extents, then free it now.
654 if (iip
->ili_extents_buf
!= NULL
) {
655 ASSERT(ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
);
656 ASSERT(ip
->i_d
.di_nextents
> 0);
657 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
);
658 ASSERT(ip
->i_df
.if_bytes
> 0);
659 kmem_free(iip
->ili_extents_buf
);
660 iip
->ili_extents_buf
= NULL
;
662 if (iip
->ili_aextents_buf
!= NULL
) {
663 ASSERT(ip
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
);
664 ASSERT(ip
->i_d
.di_anextents
> 0);
665 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
);
666 ASSERT(ip
->i_afp
->if_bytes
> 0);
667 kmem_free(iip
->ili_aextents_buf
);
668 iip
->ili_aextents_buf
= NULL
;
672 * Figure out if we should unlock the inode or not.
674 hold
= iip
->ili_flags
& XFS_ILI_HOLD
;
677 * Before clearing out the flags, remember whether we
678 * are holding the inode's IO lock.
680 iolocked
= iip
->ili_flags
& XFS_ILI_IOLOCKED_ANY
;
683 * Clear out the fields of the inode log item particular
684 * to the current transaction.
689 * Unlock the inode if XFS_ILI_HOLD was not set.
692 lock_flags
= XFS_ILOCK_EXCL
;
693 if (iolocked
& XFS_ILI_IOLOCKED_EXCL
) {
694 lock_flags
|= XFS_IOLOCK_EXCL
;
695 } else if (iolocked
& XFS_ILI_IOLOCKED_SHARED
) {
696 lock_flags
|= XFS_IOLOCK_SHARED
;
698 xfs_iput(iip
->ili_inode
, lock_flags
);
703 * This is called to find out where the oldest active copy of the
704 * inode log item in the on disk log resides now that the last log
705 * write of it completed at the given lsn. Since we always re-log
706 * all dirty data in an inode, the latest copy in the on disk log
707 * is the only one that matters. Therefore, simply return the
711 xfs_inode_item_committed(
712 struct xfs_log_item
*lip
,
719 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
720 * failed to get the inode flush lock but did get the inode locked SHARED.
721 * Here we're trying to see if the inode buffer is incore, and if so whether it's
722 * marked delayed write. If that's the case, we'll promote it and that will
723 * allow the caller to write the buffer by triggering the xfsbufd to run.
726 xfs_inode_item_pushbuf(
727 struct xfs_log_item
*lip
)
729 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
730 struct xfs_inode
*ip
= iip
->ili_inode
;
733 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
736 * If a flush is not in progress anymore, chances are that the
737 * inode was taken off the AIL. So, just get out.
739 if (completion_done(&ip
->i_flush
) ||
740 !(lip
->li_flags
& XFS_LI_IN_AIL
)) {
741 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
745 bp
= xfs_incore(ip
->i_mount
->m_ddev_targp
, iip
->ili_format
.ilf_blkno
,
746 iip
->ili_format
.ilf_len
, XBF_TRYLOCK
);
748 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
751 if (XFS_BUF_ISDELAYWRITE(bp
))
752 xfs_buf_delwri_promote(bp
);
757 * This is called to asynchronously write the inode associated with this
758 * inode log item out to disk. The inode will already have been locked by
759 * a successful call to xfs_inode_item_trylock().
763 struct xfs_log_item
*lip
)
765 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
766 struct xfs_inode
*ip
= iip
->ili_inode
;
768 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
769 ASSERT(!completion_done(&ip
->i_flush
));
772 * Since we were able to lock the inode's flush lock and
773 * we found it on the AIL, the inode must be dirty. This
774 * is because the inode is removed from the AIL while still
775 * holding the flush lock in xfs_iflush_done(). Thus, if
776 * we found it in the AIL and were able to obtain the flush
777 * lock without sleeping, then there must not have been
778 * anyone in the process of flushing the inode.
780 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) ||
781 iip
->ili_format
.ilf_fields
!= 0);
784 * Push the inode to it's backing buffer. This will not remove the
785 * inode from the AIL - a further push will be required to trigger a
786 * buffer push. However, this allows all the dirty inodes to be pushed
787 * to the buffer before it is pushed to disk. THe buffer IO completion
788 * will pull th einode from the AIL, mark it clean and unlock the flush
791 (void) xfs_iflush(ip
, 0);
792 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
796 * XXX rcc - this one really has to do something. Probably needs
797 * to stamp in a new field in the incore inode.
800 xfs_inode_item_committing(
801 struct xfs_log_item
*lip
,
804 INODE_ITEM(lip
)->ili_last_lsn
= lsn
;
808 * This is the ops vector shared by all buf log items.
810 static struct xfs_item_ops xfs_inode_item_ops
= {
811 .iop_size
= xfs_inode_item_size
,
812 .iop_format
= xfs_inode_item_format
,
813 .iop_pin
= xfs_inode_item_pin
,
814 .iop_unpin
= xfs_inode_item_unpin
,
815 .iop_trylock
= xfs_inode_item_trylock
,
816 .iop_unlock
= xfs_inode_item_unlock
,
817 .iop_committed
= xfs_inode_item_committed
,
818 .iop_push
= xfs_inode_item_push
,
819 .iop_pushbuf
= xfs_inode_item_pushbuf
,
820 .iop_committing
= xfs_inode_item_committing
825 * Initialize the inode log item for a newly allocated (in-core) inode.
829 struct xfs_inode
*ip
,
830 struct xfs_mount
*mp
)
832 struct xfs_inode_log_item
*iip
;
834 ASSERT(ip
->i_itemp
== NULL
);
835 iip
= ip
->i_itemp
= kmem_zone_zalloc(xfs_ili_zone
, KM_SLEEP
);
838 xfs_log_item_init(mp
, &iip
->ili_item
, XFS_LI_INODE
,
839 &xfs_inode_item_ops
);
840 iip
->ili_format
.ilf_type
= XFS_LI_INODE
;
841 iip
->ili_format
.ilf_ino
= ip
->i_ino
;
842 iip
->ili_format
.ilf_blkno
= ip
->i_imap
.im_blkno
;
843 iip
->ili_format
.ilf_len
= ip
->i_imap
.im_len
;
844 iip
->ili_format
.ilf_boffset
= ip
->i_imap
.im_boffset
;
848 * Free the inode log item and any memory hanging off of it.
851 xfs_inode_item_destroy(
854 #ifdef XFS_TRANS_DEBUG
855 if (ip
->i_itemp
->ili_root_size
!= 0) {
856 kmem_free(ip
->i_itemp
->ili_orig_root
);
859 kmem_zone_free(xfs_ili_zone
, ip
->i_itemp
);
864 * This is the inode flushing I/O completion routine. It is called
865 * from interrupt level when the buffer containing the inode is
866 * flushed to disk. It is responsible for removing the inode item
867 * from the AIL if it has not been re-logged, and unlocking the inode's
874 xfs_inode_log_item_t
*iip
)
876 xfs_inode_t
*ip
= iip
->ili_inode
;
877 struct xfs_ail
*ailp
= iip
->ili_item
.li_ailp
;
880 * We only want to pull the item from the AIL if it is
881 * actually there and its location in the log has not
882 * changed since we started the flush. Thus, we only bother
883 * if the ili_logged flag is set and the inode's lsn has not
884 * changed. First we check the lsn outside
885 * the lock since it's cheaper, and then we recheck while
886 * holding the lock before removing the inode from the AIL.
888 if (iip
->ili_logged
&&
889 (iip
->ili_item
.li_lsn
== iip
->ili_flush_lsn
)) {
890 spin_lock(&ailp
->xa_lock
);
891 if (iip
->ili_item
.li_lsn
== iip
->ili_flush_lsn
) {
892 /* xfs_trans_ail_delete() drops the AIL lock. */
893 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)iip
);
895 spin_unlock(&ailp
->xa_lock
);
902 * Clear the ili_last_fields bits now that we know that the
903 * data corresponding to them is safely on disk.
905 iip
->ili_last_fields
= 0;
908 * Release the inode's flush lock since we're done with it.
916 * This is the inode flushing abort routine. It is called
917 * from xfs_iflush when the filesystem is shutting down to clean
918 * up the inode state.
919 * It is responsible for removing the inode item
920 * from the AIL if it has not been re-logged, and unlocking the inode's
927 xfs_inode_log_item_t
*iip
= ip
->i_itemp
;
933 struct xfs_ail
*ailp
= iip
->ili_item
.li_ailp
;
934 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
935 spin_lock(&ailp
->xa_lock
);
936 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
937 /* xfs_trans_ail_delete() drops the AIL lock. */
938 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)iip
);
940 spin_unlock(&ailp
->xa_lock
);
944 * Clear the ili_last_fields bits now that we know that the
945 * data corresponding to them is safely on disk.
947 iip
->ili_last_fields
= 0;
949 * Clear the inode logging fields so no more flushes are
952 iip
->ili_format
.ilf_fields
= 0;
955 * Release the inode's flush lock since we're done with it.
963 xfs_inode_log_item_t
*iip
)
965 xfs_iflush_abort(iip
->ili_inode
);
969 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
970 * (which can have different field alignments) to the native version
973 xfs_inode_item_format_convert(
974 xfs_log_iovec_t
*buf
,
975 xfs_inode_log_format_t
*in_f
)
977 if (buf
->i_len
== sizeof(xfs_inode_log_format_32_t
)) {
978 xfs_inode_log_format_32_t
*in_f32
;
980 in_f32
= (xfs_inode_log_format_32_t
*)buf
->i_addr
;
981 in_f
->ilf_type
= in_f32
->ilf_type
;
982 in_f
->ilf_size
= in_f32
->ilf_size
;
983 in_f
->ilf_fields
= in_f32
->ilf_fields
;
984 in_f
->ilf_asize
= in_f32
->ilf_asize
;
985 in_f
->ilf_dsize
= in_f32
->ilf_dsize
;
986 in_f
->ilf_ino
= in_f32
->ilf_ino
;
987 /* copy biggest field of ilf_u */
988 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
989 in_f32
->ilf_u
.ilfu_uuid
.__u_bits
,
991 in_f
->ilf_blkno
= in_f32
->ilf_blkno
;
992 in_f
->ilf_len
= in_f32
->ilf_len
;
993 in_f
->ilf_boffset
= in_f32
->ilf_boffset
;
995 } else if (buf
->i_len
== sizeof(xfs_inode_log_format_64_t
)){
996 xfs_inode_log_format_64_t
*in_f64
;
998 in_f64
= (xfs_inode_log_format_64_t
*)buf
->i_addr
;
999 in_f
->ilf_type
= in_f64
->ilf_type
;
1000 in_f
->ilf_size
= in_f64
->ilf_size
;
1001 in_f
->ilf_fields
= in_f64
->ilf_fields
;
1002 in_f
->ilf_asize
= in_f64
->ilf_asize
;
1003 in_f
->ilf_dsize
= in_f64
->ilf_dsize
;
1004 in_f
->ilf_ino
= in_f64
->ilf_ino
;
1005 /* copy biggest field of ilf_u */
1006 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
1007 in_f64
->ilf_u
.ilfu_uuid
.__u_bits
,
1009 in_f
->ilf_blkno
= in_f64
->ilf_blkno
;
1010 in_f
->ilf_len
= in_f64
->ilf_len
;
1011 in_f
->ilf_boffset
= in_f64
->ilf_boffset
;
1014 return EFSCORRUPTED
;