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
;
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
;
226 * Make sure the linux inode is dirty. We do this before
227 * clearing i_update_core as the VFS will call back into
228 * XFS here and set i_update_core, so we need to dirty the
229 * inode first so that the ordering of i_update_core and
230 * unlogged modifications still works as described below.
232 xfs_mark_inode_dirty_sync(ip
);
235 * Clear i_update_core if the timestamps (or any other
236 * non-transactional modification) need flushing/logging
237 * and we're about to log them with the rest of the core.
239 * This is the same logic as xfs_iflush() but this code can't
240 * run at the same time as xfs_iflush because we're in commit
241 * processing here and so we have the inode lock held in
242 * exclusive mode. Although it doesn't really matter
243 * for the timestamps if both routines were to grab the
244 * timestamps or not. That would be ok.
246 * We clear i_update_core before copying out the data.
247 * This is for coordination with our timestamp updates
248 * that don't hold the inode lock. They will always
249 * update the timestamps BEFORE setting i_update_core,
250 * so if we clear i_update_core after they set it we
251 * are guaranteed to see their updates to the timestamps
252 * either here. Likewise, if they set it after we clear it
253 * here, we'll see it either on the next commit of this
254 * inode or the next time the inode gets flushed via
255 * xfs_iflush(). This depends on strongly ordered memory
256 * semantics, but we have that. We use the SYNCHRONIZE
257 * macro to make sure that the compiler does not reorder
258 * the i_update_core access below the data copy below.
260 if (ip
->i_update_core
) {
261 ip
->i_update_core
= 0;
266 * Make sure to get the latest timestamps from the Linux inode.
268 xfs_synchronize_times(ip
);
270 vecp
->i_addr
= &ip
->i_d
;
271 vecp
->i_len
= sizeof(struct xfs_icdinode
);
272 vecp
->i_type
= XLOG_REG_TYPE_ICORE
;
275 iip
->ili_format
.ilf_fields
|= XFS_ILOG_CORE
;
278 * If this is really an old format inode, then we need to
279 * log it as such. This means that we have to copy the link
280 * count from the new field to the old. We don't have to worry
281 * about the new fields, because nothing trusts them as long as
282 * the old inode version number is there. If the superblock already
283 * has a new version number, then we don't bother converting back.
286 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
287 if (ip
->i_d
.di_version
== 1) {
288 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
292 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
293 ip
->i_d
.di_onlink
= ip
->i_d
.di_nlink
;
296 * The superblock version has already been bumped,
297 * so just make the conversion to the new inode
300 ip
->i_d
.di_version
= 2;
301 ip
->i_d
.di_onlink
= 0;
302 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
306 switch (ip
->i_d
.di_format
) {
307 case XFS_DINODE_FMT_EXTENTS
:
308 ASSERT(!(iip
->ili_format
.ilf_fields
&
309 (XFS_ILOG_DDATA
| XFS_ILOG_DBROOT
|
310 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
311 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
) {
312 ASSERT(ip
->i_df
.if_bytes
> 0);
313 ASSERT(ip
->i_df
.if_u1
.if_extents
!= NULL
);
314 ASSERT(ip
->i_d
.di_nextents
> 0);
315 ASSERT(iip
->ili_extents_buf
== NULL
);
316 ASSERT((ip
->i_df
.if_bytes
/
317 (uint
)sizeof(xfs_bmbt_rec_t
)) > 0);
318 #ifdef XFS_NATIVE_HOST
319 if (nrecs
== ip
->i_d
.di_nextents
) {
321 * There are no delayed allocation
322 * extents, so just point to the
323 * real extents array.
325 vecp
->i_addr
= ip
->i_df
.if_u1
.if_extents
;
326 vecp
->i_len
= ip
->i_df
.if_bytes
;
327 vecp
->i_type
= XLOG_REG_TYPE_IEXT
;
332 * There are delayed allocation extents
333 * in the inode, or we need to convert
334 * the extents to on disk format.
335 * Use xfs_iextents_copy()
336 * to copy only the real extents into
337 * a separate buffer. We'll free the
338 * buffer in the unlock routine.
340 ext_buffer
= kmem_alloc(ip
->i_df
.if_bytes
,
342 iip
->ili_extents_buf
= ext_buffer
;
343 vecp
->i_addr
= ext_buffer
;
344 vecp
->i_len
= xfs_iextents_copy(ip
, ext_buffer
,
346 vecp
->i_type
= XLOG_REG_TYPE_IEXT
;
348 ASSERT(vecp
->i_len
<= ip
->i_df
.if_bytes
);
349 iip
->ili_format
.ilf_dsize
= vecp
->i_len
;
355 case XFS_DINODE_FMT_BTREE
:
356 ASSERT(!(iip
->ili_format
.ilf_fields
&
357 (XFS_ILOG_DDATA
| XFS_ILOG_DEXT
|
358 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
359 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DBROOT
) {
360 ASSERT(ip
->i_df
.if_broot_bytes
> 0);
361 ASSERT(ip
->i_df
.if_broot
!= NULL
);
362 vecp
->i_addr
= ip
->i_df
.if_broot
;
363 vecp
->i_len
= ip
->i_df
.if_broot_bytes
;
364 vecp
->i_type
= XLOG_REG_TYPE_IBROOT
;
367 iip
->ili_format
.ilf_dsize
= ip
->i_df
.if_broot_bytes
;
371 case XFS_DINODE_FMT_LOCAL
:
372 ASSERT(!(iip
->ili_format
.ilf_fields
&
373 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
374 XFS_ILOG_DEV
| XFS_ILOG_UUID
)));
375 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DDATA
) {
376 ASSERT(ip
->i_df
.if_bytes
> 0);
377 ASSERT(ip
->i_df
.if_u1
.if_data
!= NULL
);
378 ASSERT(ip
->i_d
.di_size
> 0);
380 vecp
->i_addr
= ip
->i_df
.if_u1
.if_data
;
382 * Round i_bytes up to a word boundary.
383 * The underlying memory is guaranteed to
384 * to be there by xfs_idata_realloc().
386 data_bytes
= roundup(ip
->i_df
.if_bytes
, 4);
387 ASSERT((ip
->i_df
.if_real_bytes
== 0) ||
388 (ip
->i_df
.if_real_bytes
== data_bytes
));
389 vecp
->i_len
= (int)data_bytes
;
390 vecp
->i_type
= XLOG_REG_TYPE_ILOCAL
;
393 iip
->ili_format
.ilf_dsize
= (unsigned)data_bytes
;
397 case XFS_DINODE_FMT_DEV
:
398 ASSERT(!(iip
->ili_format
.ilf_fields
&
399 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
400 XFS_ILOG_DDATA
| XFS_ILOG_UUID
)));
401 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
402 iip
->ili_format
.ilf_u
.ilfu_rdev
=
403 ip
->i_df
.if_u2
.if_rdev
;
407 case XFS_DINODE_FMT_UUID
:
408 ASSERT(!(iip
->ili_format
.ilf_fields
&
409 (XFS_ILOG_DBROOT
| XFS_ILOG_DEXT
|
410 XFS_ILOG_DDATA
| XFS_ILOG_DEV
)));
411 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
412 iip
->ili_format
.ilf_u
.ilfu_uuid
=
413 ip
->i_df
.if_u2
.if_uuid
;
423 * If there are no attributes associated with the file,
425 * Assert that no attribute-related log flags are set.
427 if (!XFS_IFORK_Q(ip
)) {
428 ASSERT(nvecs
== lip
->li_desc
->lid_size
);
429 iip
->ili_format
.ilf_size
= nvecs
;
430 ASSERT(!(iip
->ili_format
.ilf_fields
&
431 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
435 switch (ip
->i_d
.di_aformat
) {
436 case XFS_DINODE_FMT_EXTENTS
:
437 ASSERT(!(iip
->ili_format
.ilf_fields
&
438 (XFS_ILOG_ADATA
| XFS_ILOG_ABROOT
)));
439 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
) {
441 int nrecs
= ip
->i_afp
->if_bytes
/
442 (uint
)sizeof(xfs_bmbt_rec_t
);
444 ASSERT(nrecs
== ip
->i_d
.di_anextents
);
445 ASSERT(ip
->i_afp
->if_bytes
> 0);
446 ASSERT(ip
->i_afp
->if_u1
.if_extents
!= NULL
);
447 ASSERT(ip
->i_d
.di_anextents
> 0);
449 #ifdef XFS_NATIVE_HOST
451 * There are not delayed allocation extents
452 * for attributes, so just point at the array.
454 vecp
->i_addr
= ip
->i_afp
->if_u1
.if_extents
;
455 vecp
->i_len
= ip
->i_afp
->if_bytes
;
457 ASSERT(iip
->ili_aextents_buf
== NULL
);
459 * Need to endian flip before logging
461 ext_buffer
= kmem_alloc(ip
->i_afp
->if_bytes
,
463 iip
->ili_aextents_buf
= ext_buffer
;
464 vecp
->i_addr
= ext_buffer
;
465 vecp
->i_len
= xfs_iextents_copy(ip
, ext_buffer
,
468 vecp
->i_type
= XLOG_REG_TYPE_IATTR_EXT
;
469 iip
->ili_format
.ilf_asize
= vecp
->i_len
;
475 case XFS_DINODE_FMT_BTREE
:
476 ASSERT(!(iip
->ili_format
.ilf_fields
&
477 (XFS_ILOG_ADATA
| XFS_ILOG_AEXT
)));
478 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ABROOT
) {
479 ASSERT(ip
->i_afp
->if_broot_bytes
> 0);
480 ASSERT(ip
->i_afp
->if_broot
!= NULL
);
481 vecp
->i_addr
= ip
->i_afp
->if_broot
;
482 vecp
->i_len
= ip
->i_afp
->if_broot_bytes
;
483 vecp
->i_type
= XLOG_REG_TYPE_IATTR_BROOT
;
486 iip
->ili_format
.ilf_asize
= ip
->i_afp
->if_broot_bytes
;
490 case XFS_DINODE_FMT_LOCAL
:
491 ASSERT(!(iip
->ili_format
.ilf_fields
&
492 (XFS_ILOG_ABROOT
| XFS_ILOG_AEXT
)));
493 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_ADATA
) {
494 ASSERT(ip
->i_afp
->if_bytes
> 0);
495 ASSERT(ip
->i_afp
->if_u1
.if_data
!= NULL
);
497 vecp
->i_addr
= ip
->i_afp
->if_u1
.if_data
;
499 * Round i_bytes up to a word boundary.
500 * The underlying memory is guaranteed to
501 * to be there by xfs_idata_realloc().
503 data_bytes
= roundup(ip
->i_afp
->if_bytes
, 4);
504 ASSERT((ip
->i_afp
->if_real_bytes
== 0) ||
505 (ip
->i_afp
->if_real_bytes
== data_bytes
));
506 vecp
->i_len
= (int)data_bytes
;
507 vecp
->i_type
= XLOG_REG_TYPE_IATTR_LOCAL
;
510 iip
->ili_format
.ilf_asize
= (unsigned)data_bytes
;
519 ASSERT(nvecs
== lip
->li_desc
->lid_size
);
520 iip
->ili_format
.ilf_size
= nvecs
;
525 * This is called to pin the inode associated with the inode log
526 * item in memory so it cannot be written out.
530 struct xfs_log_item
*lip
)
532 struct xfs_inode
*ip
= INODE_ITEM(lip
)->ili_inode
;
534 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
536 trace_xfs_inode_pin(ip
, _RET_IP_
);
537 atomic_inc(&ip
->i_pincount
);
542 * This is called to unpin the inode associated with the inode log
543 * item which was previously pinned with a call to xfs_inode_item_pin().
545 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
548 xfs_inode_item_unpin(
549 struct xfs_log_item
*lip
,
552 struct xfs_inode
*ip
= INODE_ITEM(lip
)->ili_inode
;
554 trace_xfs_inode_unpin(ip
, _RET_IP_
);
555 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
556 if (atomic_dec_and_test(&ip
->i_pincount
))
557 wake_up(&ip
->i_ipin_wait
);
561 * This is called to attempt to lock the inode associated with this
562 * inode log item, in preparation for the push routine which does the actual
563 * iflush. Don't sleep on the inode lock or the flush lock.
565 * If the flush lock is already held, indicating that the inode has
566 * been or is in the process of being flushed, then (ideally) we'd like to
567 * see if the inode's buffer is still incore, and if so give it a nudge.
568 * We delay doing so until the pushbuf routine, though, to avoid holding
569 * the AIL lock across a call to the blackhole which is the buffer cache.
570 * Also we don't want to sleep in any device strategy routines, which can happen
571 * if we do the subsequent bawrite in here.
574 xfs_inode_item_trylock(
575 struct xfs_log_item
*lip
)
577 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
578 struct xfs_inode
*ip
= iip
->ili_inode
;
580 if (xfs_ipincount(ip
) > 0)
581 return XFS_ITEM_PINNED
;
583 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
))
584 return XFS_ITEM_LOCKED
;
586 if (!xfs_iflock_nowait(ip
)) {
588 * inode has already been flushed to the backing buffer,
589 * leave it locked in shared mode, pushbuf routine will
592 return XFS_ITEM_PUSHBUF
;
595 /* Stale items should force out the iclog */
596 if (ip
->i_flags
& XFS_ISTALE
) {
599 * we hold the AIL lock - notify the unlock routine of this
600 * so it doesn't try to get the lock again.
602 xfs_iunlock(ip
, XFS_ILOCK_SHARED
|XFS_IUNLOCK_NONOTIFY
);
603 return XFS_ITEM_PINNED
;
607 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
608 ASSERT(iip
->ili_format
.ilf_fields
!= 0);
609 ASSERT(iip
->ili_logged
== 0);
610 ASSERT(lip
->li_flags
& XFS_LI_IN_AIL
);
613 return XFS_ITEM_SUCCESS
;
617 * Unlock the inode associated with the inode log item.
618 * Clear the fields of the inode and inode log item that
619 * are specific to the current transaction. If the
620 * hold flags is set, do not unlock the inode.
623 xfs_inode_item_unlock(
624 struct xfs_log_item
*lip
)
626 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
627 struct xfs_inode
*ip
= iip
->ili_inode
;
628 unsigned short lock_flags
;
630 ASSERT(iip
->ili_inode
->i_itemp
!= NULL
);
631 ASSERT(xfs_isilocked(iip
->ili_inode
, XFS_ILOCK_EXCL
));
634 * Clear the transaction pointer in the inode.
639 * If the inode needed a separate buffer with which to log
640 * its extents, then free it now.
642 if (iip
->ili_extents_buf
!= NULL
) {
643 ASSERT(ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
);
644 ASSERT(ip
->i_d
.di_nextents
> 0);
645 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_DEXT
);
646 ASSERT(ip
->i_df
.if_bytes
> 0);
647 kmem_free(iip
->ili_extents_buf
);
648 iip
->ili_extents_buf
= NULL
;
650 if (iip
->ili_aextents_buf
!= NULL
) {
651 ASSERT(ip
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
);
652 ASSERT(ip
->i_d
.di_anextents
> 0);
653 ASSERT(iip
->ili_format
.ilf_fields
& XFS_ILOG_AEXT
);
654 ASSERT(ip
->i_afp
->if_bytes
> 0);
655 kmem_free(iip
->ili_aextents_buf
);
656 iip
->ili_aextents_buf
= NULL
;
659 lock_flags
= iip
->ili_lock_flags
;
660 iip
->ili_lock_flags
= 0;
662 xfs_iunlock(iip
->ili_inode
, lock_flags
);
663 IRELE(iip
->ili_inode
);
668 * This is called to find out where the oldest active copy of the
669 * inode log item in the on disk log resides now that the last log
670 * write of it completed at the given lsn. Since we always re-log
671 * all dirty data in an inode, the latest copy in the on disk log
672 * is the only one that matters. Therefore, simply return the
676 xfs_inode_item_committed(
677 struct xfs_log_item
*lip
,
684 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
685 * failed to get the inode flush lock but did get the inode locked SHARED.
686 * Here we're trying to see if the inode buffer is incore, and if so whether it's
687 * marked delayed write. If that's the case, we'll promote it and that will
688 * allow the caller to write the buffer by triggering the xfsbufd to run.
691 xfs_inode_item_pushbuf(
692 struct xfs_log_item
*lip
)
694 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
695 struct xfs_inode
*ip
= iip
->ili_inode
;
698 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
701 * If a flush is not in progress anymore, chances are that the
702 * inode was taken off the AIL. So, just get out.
704 if (completion_done(&ip
->i_flush
) ||
705 !(lip
->li_flags
& XFS_LI_IN_AIL
)) {
706 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
710 bp
= xfs_incore(ip
->i_mount
->m_ddev_targp
, iip
->ili_format
.ilf_blkno
,
711 iip
->ili_format
.ilf_len
, XBF_TRYLOCK
);
713 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
716 if (XFS_BUF_ISDELAYWRITE(bp
))
717 xfs_buf_delwri_promote(bp
);
722 * This is called to asynchronously write the inode associated with this
723 * inode log item out to disk. The inode will already have been locked by
724 * a successful call to xfs_inode_item_trylock().
728 struct xfs_log_item
*lip
)
730 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
731 struct xfs_inode
*ip
= iip
->ili_inode
;
733 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
));
734 ASSERT(!completion_done(&ip
->i_flush
));
737 * Since we were able to lock the inode's flush lock and
738 * we found it on the AIL, the inode must be dirty. This
739 * is because the inode is removed from the AIL while still
740 * holding the flush lock in xfs_iflush_done(). Thus, if
741 * we found it in the AIL and were able to obtain the flush
742 * lock without sleeping, then there must not have been
743 * anyone in the process of flushing the inode.
745 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) ||
746 iip
->ili_format
.ilf_fields
!= 0);
749 * Push the inode to it's backing buffer. This will not remove the
750 * inode from the AIL - a further push will be required to trigger a
751 * buffer push. However, this allows all the dirty inodes to be pushed
752 * to the buffer before it is pushed to disk. THe buffer IO completion
753 * will pull th einode from the AIL, mark it clean and unlock the flush
756 (void) xfs_iflush(ip
, 0);
757 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
761 * XXX rcc - this one really has to do something. Probably needs
762 * to stamp in a new field in the incore inode.
765 xfs_inode_item_committing(
766 struct xfs_log_item
*lip
,
769 INODE_ITEM(lip
)->ili_last_lsn
= lsn
;
773 * This is the ops vector shared by all buf log items.
775 static struct xfs_item_ops xfs_inode_item_ops
= {
776 .iop_size
= xfs_inode_item_size
,
777 .iop_format
= xfs_inode_item_format
,
778 .iop_pin
= xfs_inode_item_pin
,
779 .iop_unpin
= xfs_inode_item_unpin
,
780 .iop_trylock
= xfs_inode_item_trylock
,
781 .iop_unlock
= xfs_inode_item_unlock
,
782 .iop_committed
= xfs_inode_item_committed
,
783 .iop_push
= xfs_inode_item_push
,
784 .iop_pushbuf
= xfs_inode_item_pushbuf
,
785 .iop_committing
= xfs_inode_item_committing
790 * Initialize the inode log item for a newly allocated (in-core) inode.
794 struct xfs_inode
*ip
,
795 struct xfs_mount
*mp
)
797 struct xfs_inode_log_item
*iip
;
799 ASSERT(ip
->i_itemp
== NULL
);
800 iip
= ip
->i_itemp
= kmem_zone_zalloc(xfs_ili_zone
, KM_SLEEP
);
803 xfs_log_item_init(mp
, &iip
->ili_item
, XFS_LI_INODE
,
804 &xfs_inode_item_ops
);
805 iip
->ili_format
.ilf_type
= XFS_LI_INODE
;
806 iip
->ili_format
.ilf_ino
= ip
->i_ino
;
807 iip
->ili_format
.ilf_blkno
= ip
->i_imap
.im_blkno
;
808 iip
->ili_format
.ilf_len
= ip
->i_imap
.im_len
;
809 iip
->ili_format
.ilf_boffset
= ip
->i_imap
.im_boffset
;
813 * Free the inode log item and any memory hanging off of it.
816 xfs_inode_item_destroy(
819 #ifdef XFS_TRANS_DEBUG
820 if (ip
->i_itemp
->ili_root_size
!= 0) {
821 kmem_free(ip
->i_itemp
->ili_orig_root
);
824 kmem_zone_free(xfs_ili_zone
, ip
->i_itemp
);
829 * This is the inode flushing I/O completion routine. It is called
830 * from interrupt level when the buffer containing the inode is
831 * flushed to disk. It is responsible for removing the inode item
832 * from the AIL if it has not been re-logged, and unlocking the inode's
838 struct xfs_log_item
*lip
)
840 struct xfs_inode_log_item
*iip
= INODE_ITEM(lip
);
841 xfs_inode_t
*ip
= iip
->ili_inode
;
842 struct xfs_ail
*ailp
= lip
->li_ailp
;
845 * We only want to pull the item from the AIL if it is
846 * actually there and its location in the log has not
847 * changed since we started the flush. Thus, we only bother
848 * if the ili_logged flag is set and the inode's lsn has not
849 * changed. First we check the lsn outside
850 * the lock since it's cheaper, and then we recheck while
851 * holding the lock before removing the inode from the AIL.
853 if (iip
->ili_logged
&& lip
->li_lsn
== iip
->ili_flush_lsn
) {
854 spin_lock(&ailp
->xa_lock
);
855 if (lip
->li_lsn
== iip
->ili_flush_lsn
) {
856 /* xfs_trans_ail_delete() drops the AIL lock. */
857 xfs_trans_ail_delete(ailp
, lip
);
859 spin_unlock(&ailp
->xa_lock
);
866 * Clear the ili_last_fields bits now that we know that the
867 * data corresponding to them is safely on disk.
869 iip
->ili_last_fields
= 0;
872 * Release the inode's flush lock since we're done with it.
878 * This is the inode flushing abort routine. It is called
879 * from xfs_iflush when the filesystem is shutting down to clean
880 * up the inode state.
881 * It is responsible for removing the inode item
882 * from the AIL if it has not been re-logged, and unlocking the inode's
889 xfs_inode_log_item_t
*iip
= ip
->i_itemp
;
893 struct xfs_ail
*ailp
= iip
->ili_item
.li_ailp
;
894 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
895 spin_lock(&ailp
->xa_lock
);
896 if (iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) {
897 /* xfs_trans_ail_delete() drops the AIL lock. */
898 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)iip
);
900 spin_unlock(&ailp
->xa_lock
);
904 * Clear the ili_last_fields bits now that we know that the
905 * data corresponding to them is safely on disk.
907 iip
->ili_last_fields
= 0;
909 * Clear the inode logging fields so no more flushes are
912 iip
->ili_format
.ilf_fields
= 0;
915 * Release the inode's flush lock since we're done with it.
923 struct xfs_log_item
*lip
)
925 xfs_iflush_abort(INODE_ITEM(lip
)->ili_inode
);
929 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
930 * (which can have different field alignments) to the native version
933 xfs_inode_item_format_convert(
934 xfs_log_iovec_t
*buf
,
935 xfs_inode_log_format_t
*in_f
)
937 if (buf
->i_len
== sizeof(xfs_inode_log_format_32_t
)) {
938 xfs_inode_log_format_32_t
*in_f32
= buf
->i_addr
;
940 in_f
->ilf_type
= in_f32
->ilf_type
;
941 in_f
->ilf_size
= in_f32
->ilf_size
;
942 in_f
->ilf_fields
= in_f32
->ilf_fields
;
943 in_f
->ilf_asize
= in_f32
->ilf_asize
;
944 in_f
->ilf_dsize
= in_f32
->ilf_dsize
;
945 in_f
->ilf_ino
= in_f32
->ilf_ino
;
946 /* copy biggest field of ilf_u */
947 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
948 in_f32
->ilf_u
.ilfu_uuid
.__u_bits
,
950 in_f
->ilf_blkno
= in_f32
->ilf_blkno
;
951 in_f
->ilf_len
= in_f32
->ilf_len
;
952 in_f
->ilf_boffset
= in_f32
->ilf_boffset
;
954 } else if (buf
->i_len
== sizeof(xfs_inode_log_format_64_t
)){
955 xfs_inode_log_format_64_t
*in_f64
= buf
->i_addr
;
957 in_f
->ilf_type
= in_f64
->ilf_type
;
958 in_f
->ilf_size
= in_f64
->ilf_size
;
959 in_f
->ilf_fields
= in_f64
->ilf_fields
;
960 in_f
->ilf_asize
= in_f64
->ilf_asize
;
961 in_f
->ilf_dsize
= in_f64
->ilf_dsize
;
962 in_f
->ilf_ino
= in_f64
->ilf_ino
;
963 /* copy biggest field of ilf_u */
964 memcpy(in_f
->ilf_u
.ilfu_uuid
.__u_bits
,
965 in_f64
->ilf_u
.ilfu_uuid
.__u_bits
,
967 in_f
->ilf_blkno
= in_f64
->ilf_blkno
;
968 in_f
->ilf_len
= in_f64
->ilf_len
;
969 in_f
->ilf_boffset
= in_f64
->ilf_boffset
;