2 * Copyright (c) 2000-2006 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
18 #include <linux/log2.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
29 #include "xfs_mount.h"
30 #include "xfs_inode.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
34 #include "xfs_attr_sf.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_bmap_util.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_dinode.h"
46 #include "xfs_filestream.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
50 #include "xfs_symlink.h"
51 #include "xfs_trans_priv.h"
53 #include "xfs_bmap_btree.h"
55 kmem_zone_t
*xfs_inode_zone
;
58 * Used in xfs_itruncate_extents(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
66 * helper function to extract extent size hint from inode
72 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
73 return ip
->i_d
.di_extsize
;
74 if (XFS_IS_REALTIME_INODE(ip
))
75 return ip
->i_mount
->m_sb
.sb_rextsize
;
80 * These two are wrapper routines around the xfs_ilock() routine used to
81 * centralize some grungy code. They are used in places that wish to lock the
82 * inode solely for reading the extents. The reason these places can't just
83 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
84 * bringing in of the extents from disk for a file in b-tree format. If the
85 * inode is in b-tree format, then we need to lock the inode exclusively until
86 * the extents are read in. Locking it exclusively all the time would limit
87 * our parallelism unnecessarily, though. What we do instead is check to see
88 * if the extents have been read in yet, and only lock the inode exclusively
91 * The functions return a value which should be given to the corresponding
95 xfs_ilock_data_map_shared(
98 uint lock_mode
= XFS_ILOCK_SHARED
;
100 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
&&
101 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)
102 lock_mode
= XFS_ILOCK_EXCL
;
103 xfs_ilock(ip
, lock_mode
);
108 xfs_ilock_attr_map_shared(
109 struct xfs_inode
*ip
)
111 uint lock_mode
= XFS_ILOCK_SHARED
;
113 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_BTREE
&&
114 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
115 lock_mode
= XFS_ILOCK_EXCL
;
116 xfs_ilock(ip
, lock_mode
);
121 * The xfs inode contains 2 locks: a multi-reader lock called the
122 * i_iolock and a multi-reader lock called the i_lock. This routine
123 * allows either or both of the locks to be obtained.
125 * The 2 locks should always be ordered so that the IO lock is
126 * obtained first in order to prevent deadlock.
128 * ip -- the inode being locked
129 * lock_flags -- this parameter indicates the inode's locks
130 * to be locked. It can be:
135 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
136 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
137 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
138 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
145 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
148 * You can't set both SHARED and EXCL for the same lock,
149 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
150 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
152 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
153 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
154 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
155 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
156 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
158 if (lock_flags
& XFS_IOLOCK_EXCL
)
159 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
160 else if (lock_flags
& XFS_IOLOCK_SHARED
)
161 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
163 if (lock_flags
& XFS_ILOCK_EXCL
)
164 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
165 else if (lock_flags
& XFS_ILOCK_SHARED
)
166 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
170 * This is just like xfs_ilock(), except that the caller
171 * is guaranteed not to sleep. It returns 1 if it gets
172 * the requested locks and 0 otherwise. If the IO lock is
173 * obtained but the inode lock cannot be, then the IO lock
174 * is dropped before returning.
176 * ip -- the inode being locked
177 * lock_flags -- this parameter indicates the inode's locks to be
178 * to be locked. See the comment for xfs_ilock() for a list
186 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
189 * You can't set both SHARED and EXCL for the same lock,
190 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
191 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
193 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
194 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
195 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
196 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
197 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
199 if (lock_flags
& XFS_IOLOCK_EXCL
) {
200 if (!mrtryupdate(&ip
->i_iolock
))
202 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
203 if (!mrtryaccess(&ip
->i_iolock
))
206 if (lock_flags
& XFS_ILOCK_EXCL
) {
207 if (!mrtryupdate(&ip
->i_lock
))
208 goto out_undo_iolock
;
209 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
210 if (!mrtryaccess(&ip
->i_lock
))
211 goto out_undo_iolock
;
216 if (lock_flags
& XFS_IOLOCK_EXCL
)
217 mrunlock_excl(&ip
->i_iolock
);
218 else if (lock_flags
& XFS_IOLOCK_SHARED
)
219 mrunlock_shared(&ip
->i_iolock
);
225 * xfs_iunlock() is used to drop the inode locks acquired with
226 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
227 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
228 * that we know which locks to drop.
230 * ip -- the inode being unlocked
231 * lock_flags -- this parameter indicates the inode's locks to be
232 * to be unlocked. See the comment for xfs_ilock() for a list
233 * of valid values for this parameter.
242 * You can't set both SHARED and EXCL for the same lock,
243 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
244 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
246 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
247 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
248 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
249 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
250 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
251 ASSERT(lock_flags
!= 0);
253 if (lock_flags
& XFS_IOLOCK_EXCL
)
254 mrunlock_excl(&ip
->i_iolock
);
255 else if (lock_flags
& XFS_IOLOCK_SHARED
)
256 mrunlock_shared(&ip
->i_iolock
);
258 if (lock_flags
& XFS_ILOCK_EXCL
)
259 mrunlock_excl(&ip
->i_lock
);
260 else if (lock_flags
& XFS_ILOCK_SHARED
)
261 mrunlock_shared(&ip
->i_lock
);
263 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
267 * give up write locks. the i/o lock cannot be held nested
268 * if it is being demoted.
275 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
276 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
278 if (lock_flags
& XFS_ILOCK_EXCL
)
279 mrdemote(&ip
->i_lock
);
280 if (lock_flags
& XFS_IOLOCK_EXCL
)
281 mrdemote(&ip
->i_iolock
);
283 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
286 #if defined(DEBUG) || defined(XFS_WARN)
292 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
293 if (!(lock_flags
& XFS_ILOCK_SHARED
))
294 return !!ip
->i_lock
.mr_writer
;
295 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
298 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
299 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
300 return !!ip
->i_iolock
.mr_writer
;
301 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);
311 int xfs_small_retries
;
312 int xfs_middle_retries
;
313 int xfs_lots_retries
;
318 * Bump the subclass so xfs_lock_inodes() acquires each lock with
322 xfs_lock_inumorder(int lock_mode
, int subclass
)
324 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
325 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
326 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
327 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
333 * The following routine will lock n inodes in exclusive mode.
334 * We assume the caller calls us with the inodes in i_ino order.
336 * We need to detect deadlock where an inode that we lock
337 * is in the AIL and we start waiting for another inode that is locked
338 * by a thread in a long running transaction (such as truncate). This can
339 * result in deadlock since the long running trans might need to wait
340 * for the inode we just locked in order to push the tail and free space
349 int attempts
= 0, i
, j
, try_lock
;
352 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
358 for (; i
< inodes
; i
++) {
361 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
365 * If try_lock is not set yet, make sure all locked inodes
366 * are not in the AIL.
367 * If any are, set try_lock to be used later.
371 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
372 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
373 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
380 * If any of the previous locks we have locked is in the AIL,
381 * we must TRY to get the second and subsequent locks. If
382 * we can't get any, we must release all we have
387 /* try_lock must be 0 if i is 0. */
389 * try_lock means we have an inode locked
390 * that is in the AIL.
393 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
397 * Unlock all previous guys and try again.
398 * xfs_iunlock will try to push the tail
399 * if the inode is in the AIL.
402 for(j
= i
- 1; j
>= 0; j
--) {
405 * Check to see if we've already
407 * Not the first one going back,
408 * and the inode ptr is the same.
410 if ((j
!= (i
- 1)) && ips
[j
] ==
414 xfs_iunlock(ips
[j
], lock_mode
);
417 if ((attempts
% 5) == 0) {
418 delay(1); /* Don't just spin the CPU */
428 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
434 if (attempts
< 5) xfs_small_retries
++;
435 else if (attempts
< 100) xfs_middle_retries
++;
436 else xfs_lots_retries
++;
444 * xfs_lock_two_inodes() can only be used to lock one type of lock
445 * at a time - the iolock or the ilock, but not both at once. If
446 * we lock both at once, lockdep will report false positives saying
447 * we have violated locking orders.
459 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
460 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
461 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
463 if (ip0
->i_ino
> ip1
->i_ino
) {
470 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
473 * If the first lock we have locked is in the AIL, we must TRY to get
474 * the second lock. If we can't get it, we must release the first one
477 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
478 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
479 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
480 xfs_iunlock(ip0
, lock_mode
);
481 if ((++attempts
% 5) == 0)
482 delay(1); /* Don't just spin the CPU */
486 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
493 struct xfs_inode
*ip
)
495 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
496 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
499 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
500 if (xfs_isiflocked(ip
))
502 } while (!xfs_iflock_nowait(ip
));
504 finish_wait(wq
, &wait
.wait
);
513 if (di_flags
& XFS_DIFLAG_ANY
) {
514 if (di_flags
& XFS_DIFLAG_REALTIME
)
515 flags
|= XFS_XFLAG_REALTIME
;
516 if (di_flags
& XFS_DIFLAG_PREALLOC
)
517 flags
|= XFS_XFLAG_PREALLOC
;
518 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
519 flags
|= XFS_XFLAG_IMMUTABLE
;
520 if (di_flags
& XFS_DIFLAG_APPEND
)
521 flags
|= XFS_XFLAG_APPEND
;
522 if (di_flags
& XFS_DIFLAG_SYNC
)
523 flags
|= XFS_XFLAG_SYNC
;
524 if (di_flags
& XFS_DIFLAG_NOATIME
)
525 flags
|= XFS_XFLAG_NOATIME
;
526 if (di_flags
& XFS_DIFLAG_NODUMP
)
527 flags
|= XFS_XFLAG_NODUMP
;
528 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
529 flags
|= XFS_XFLAG_RTINHERIT
;
530 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
531 flags
|= XFS_XFLAG_PROJINHERIT
;
532 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
533 flags
|= XFS_XFLAG_NOSYMLINKS
;
534 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
535 flags
|= XFS_XFLAG_EXTSIZE
;
536 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
537 flags
|= XFS_XFLAG_EXTSZINHERIT
;
538 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
539 flags
|= XFS_XFLAG_NODEFRAG
;
540 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
541 flags
|= XFS_XFLAG_FILESTREAM
;
551 xfs_icdinode_t
*dic
= &ip
->i_d
;
553 return _xfs_dic2xflags(dic
->di_flags
) |
554 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
561 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
562 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
566 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
567 * is allowed, otherwise it has to be an exact match. If a CI match is found,
568 * ci_name->name will point to a the actual name (caller must free) or
569 * will be set to NULL if an exact match is found.
574 struct xfs_name
*name
,
576 struct xfs_name
*ci_name
)
582 trace_xfs_lookup(dp
, name
);
584 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
585 return XFS_ERROR(EIO
);
587 lock_mode
= xfs_ilock_data_map_shared(dp
);
588 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
589 xfs_iunlock(dp
, lock_mode
);
594 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
602 kmem_free(ci_name
->name
);
609 * Allocate an inode on disk and return a copy of its in-core version.
610 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
611 * appropriately within the inode. The uid and gid for the inode are
612 * set according to the contents of the given cred structure.
614 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
615 * has a free inode available, call xfs_iget() to obtain the in-core
616 * version of the allocated inode. Finally, fill in the inode and
617 * log its initial contents. In this case, ialloc_context would be
620 * If xfs_dialloc() does not have an available inode, it will replenish
621 * its supply by doing an allocation. Since we can only do one
622 * allocation within a transaction without deadlocks, we must commit
623 * the current transaction before returning the inode itself.
624 * In this case, therefore, we will set ialloc_context and return.
625 * The caller should then commit the current transaction, start a new
626 * transaction, and call xfs_ialloc() again to actually get the inode.
628 * To ensure that some other process does not grab the inode that
629 * was allocated during the first call to xfs_ialloc(), this routine
630 * also returns the [locked] bp pointing to the head of the freelist
631 * as ialloc_context. The caller should hold this buffer across
632 * the commit and pass it back into this routine on the second call.
634 * If we are allocating quota inodes, we do not have a parent inode
635 * to attach to or associate with (i.e. pip == NULL) because they
636 * are not linked into the directory structure - they are attached
637 * directly to the superblock - and so have no parent.
648 xfs_buf_t
**ialloc_context
,
651 struct xfs_mount
*mp
= tp
->t_mountp
;
660 * Call the space management code to pick
661 * the on-disk inode to be allocated.
663 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
664 ialloc_context
, &ino
);
667 if (*ialloc_context
|| ino
== NULLFSINO
) {
671 ASSERT(*ialloc_context
== NULL
);
674 * Get the in-core inode with the lock held exclusively.
675 * This is because we're setting fields here we need
676 * to prevent others from looking at until we're done.
678 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
679 XFS_ILOCK_EXCL
, &ip
);
684 ip
->i_d
.di_mode
= mode
;
685 ip
->i_d
.di_onlink
= 0;
686 ip
->i_d
.di_nlink
= nlink
;
687 ASSERT(ip
->i_d
.di_nlink
== nlink
);
688 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
689 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
690 xfs_set_projid(ip
, prid
);
691 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
694 * If the superblock version is up to where we support new format
695 * inodes and this is currently an old format inode, then change
696 * the inode version number now. This way we only do the conversion
697 * here rather than here and in the flush/logging code.
699 if (xfs_sb_version_hasnlink(&mp
->m_sb
) &&
700 ip
->i_d
.di_version
== 1) {
701 ip
->i_d
.di_version
= 2;
703 * We've already zeroed the old link count, the projid field,
709 * Project ids won't be stored on disk if we are using a version 1 inode.
711 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
712 xfs_bump_ino_vers2(tp
, ip
);
714 if (pip
&& XFS_INHERIT_GID(pip
)) {
715 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
716 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
717 ip
->i_d
.di_mode
|= S_ISGID
;
722 * If the group ID of the new file does not match the effective group
723 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
724 * (and only if the irix_sgid_inherit compatibility variable is set).
726 if ((irix_sgid_inherit
) &&
727 (ip
->i_d
.di_mode
& S_ISGID
) &&
728 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
)))) {
729 ip
->i_d
.di_mode
&= ~S_ISGID
;
733 ip
->i_d
.di_nextents
= 0;
734 ASSERT(ip
->i_d
.di_nblocks
== 0);
737 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
738 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
739 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
740 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
743 * di_gen will have been taken care of in xfs_iread.
745 ip
->i_d
.di_extsize
= 0;
746 ip
->i_d
.di_dmevmask
= 0;
747 ip
->i_d
.di_dmstate
= 0;
748 ip
->i_d
.di_flags
= 0;
750 if (ip
->i_d
.di_version
== 3) {
751 ASSERT(ip
->i_d
.di_ino
== ino
);
752 ASSERT(uuid_equal(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
));
754 ip
->i_d
.di_changecount
= 1;
756 ip
->i_d
.di_flags2
= 0;
757 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
758 ip
->i_d
.di_crtime
= ip
->i_d
.di_mtime
;
762 flags
= XFS_ILOG_CORE
;
763 switch (mode
& S_IFMT
) {
768 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
769 ip
->i_df
.if_u2
.if_rdev
= rdev
;
770 ip
->i_df
.if_flags
= 0;
771 flags
|= XFS_ILOG_DEV
;
775 * we can't set up filestreams until after the VFS inode
776 * is set up properly.
778 if (pip
&& xfs_inode_is_filestream(pip
))
782 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
786 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
787 di_flags
|= XFS_DIFLAG_RTINHERIT
;
788 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
789 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
790 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
792 } else if (S_ISREG(mode
)) {
793 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
794 di_flags
|= XFS_DIFLAG_REALTIME
;
795 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
796 di_flags
|= XFS_DIFLAG_EXTSIZE
;
797 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
800 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
802 di_flags
|= XFS_DIFLAG_NOATIME
;
803 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
805 di_flags
|= XFS_DIFLAG_NODUMP
;
806 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
808 di_flags
|= XFS_DIFLAG_SYNC
;
809 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
810 xfs_inherit_nosymlinks
)
811 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
812 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
813 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
814 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
815 xfs_inherit_nodefrag
)
816 di_flags
|= XFS_DIFLAG_NODEFRAG
;
817 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
818 di_flags
|= XFS_DIFLAG_FILESTREAM
;
819 ip
->i_d
.di_flags
|= di_flags
;
823 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
824 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
825 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
826 ip
->i_df
.if_u1
.if_extents
= NULL
;
832 * Attribute fork settings for new inode.
834 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
835 ip
->i_d
.di_anextents
= 0;
838 * Log the new values stuffed into the inode.
840 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
841 xfs_trans_log_inode(tp
, ip
, flags
);
843 /* now that we have an i_mode we can setup inode ops and unlock */
846 /* now we have set up the vfs inode we can associate the filestream */
848 error
= xfs_filestream_associate(pip
, ip
);
852 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
860 * Allocates a new inode from disk and return a pointer to the
861 * incore copy. This routine will internally commit the current
862 * transaction and allocate a new one if the Space Manager needed
863 * to do an allocation to replenish the inode free-list.
865 * This routine is designed to be called from xfs_create and
871 xfs_trans_t
**tpp
, /* input: current transaction;
872 output: may be a new transaction. */
873 xfs_inode_t
*dp
, /* directory within whose allocate
878 prid_t prid
, /* project id */
879 int okalloc
, /* ok to allocate new space */
880 xfs_inode_t
**ipp
, /* pointer to inode; it will be
888 xfs_buf_t
*ialloc_context
= NULL
;
894 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
897 * xfs_ialloc will return a pointer to an incore inode if
898 * the Space Manager has an available inode on the free
899 * list. Otherwise, it will do an allocation and replenish
900 * the freelist. Since we can only do one allocation per
901 * transaction without deadlocks, we will need to commit the
902 * current transaction and start a new one. We will then
903 * need to call xfs_ialloc again to get the inode.
905 * If xfs_ialloc did an allocation to replenish the freelist,
906 * it returns the bp containing the head of the freelist as
907 * ialloc_context. We will hold a lock on it across the
908 * transaction commit so that no other process can steal
909 * the inode(s) that we've just allocated.
911 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
912 &ialloc_context
, &ip
);
915 * Return an error if we were unable to allocate a new inode.
916 * This should only happen if we run out of space on disk or
917 * encounter a disk error.
923 if (!ialloc_context
&& !ip
) {
925 return XFS_ERROR(ENOSPC
);
929 * If the AGI buffer is non-NULL, then we were unable to get an
930 * inode in one operation. We need to commit the current
931 * transaction and call xfs_ialloc() again. It is guaranteed
932 * to succeed the second time.
934 if (ialloc_context
) {
935 struct xfs_trans_res tres
;
938 * Normally, xfs_trans_commit releases all the locks.
939 * We call bhold to hang on to the ialloc_context across
940 * the commit. Holding this buffer prevents any other
941 * processes from doing any allocations in this
944 xfs_trans_bhold(tp
, ialloc_context
);
946 * Save the log reservation so we can use
947 * them in the next transaction.
949 tres
.tr_logres
= xfs_trans_get_log_res(tp
);
950 tres
.tr_logcount
= xfs_trans_get_log_count(tp
);
953 * We want the quota changes to be associated with the next
954 * transaction, NOT this one. So, detach the dqinfo from this
955 * and attach it to the next transaction.
960 dqinfo
= (void *)tp
->t_dqinfo
;
962 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
963 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
966 ntp
= xfs_trans_dup(tp
);
967 code
= xfs_trans_commit(tp
, 0);
969 if (committed
!= NULL
) {
973 * If we get an error during the commit processing,
974 * release the buffer that is still held and return
978 xfs_buf_relse(ialloc_context
);
980 tp
->t_dqinfo
= dqinfo
;
981 xfs_trans_free_dqinfo(tp
);
989 * transaction commit worked ok so we can drop the extra ticket
990 * reference that we gained in xfs_trans_dup()
992 xfs_log_ticket_put(tp
->t_ticket
);
993 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
994 code
= xfs_trans_reserve(tp
, &tres
, 0, 0);
997 * Re-attach the quota info that we detached from prev trx.
1000 tp
->t_dqinfo
= dqinfo
;
1001 tp
->t_flags
|= tflags
;
1005 xfs_buf_relse(ialloc_context
);
1010 xfs_trans_bjoin(tp
, ialloc_context
);
1013 * Call ialloc again. Since we've locked out all
1014 * other allocations in this allocation group,
1015 * this call should always succeed.
1017 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1018 okalloc
, &ialloc_context
, &ip
);
1021 * If we get an error at this point, return to the caller
1022 * so that the current transaction can be aborted.
1029 ASSERT(!ialloc_context
&& ip
);
1032 if (committed
!= NULL
)
1043 * Decrement the link count on an inode & log the change.
1044 * If this causes the link count to go to zero, initiate the
1045 * logging activity required to truncate a file.
1054 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1056 ASSERT (ip
->i_d
.di_nlink
> 0);
1058 drop_nlink(VFS_I(ip
));
1059 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1062 if (ip
->i_d
.di_nlink
== 0) {
1064 * We're dropping the last link to this file.
1065 * Move the on-disk inode to the AGI unlinked list.
1066 * From xfs_inactive() we will pull the inode from
1067 * the list and free it.
1069 error
= xfs_iunlink(tp
, ip
);
1075 * This gets called when the inode's version needs to be changed from 1 to 2.
1076 * Currently this happens when the nlink field overflows the old 16-bit value
1077 * or when chproj is called to change the project for the first time.
1078 * As a side effect the superblock version will also get rev'd
1079 * to contain the NLINK bit.
1088 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1089 ASSERT(ip
->i_d
.di_version
== 1);
1091 ip
->i_d
.di_version
= 2;
1092 ip
->i_d
.di_onlink
= 0;
1093 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1095 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1096 spin_lock(&mp
->m_sb_lock
);
1097 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1098 xfs_sb_version_addnlink(&mp
->m_sb
);
1099 spin_unlock(&mp
->m_sb_lock
);
1100 xfs_mod_sb(tp
, XFS_SB_VERSIONNUM
);
1102 spin_unlock(&mp
->m_sb_lock
);
1105 /* Caller must log the inode */
1109 * Increment the link count on an inode & log the change.
1116 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1118 ASSERT(ip
->i_d
.di_nlink
> 0);
1120 inc_nlink(VFS_I(ip
));
1121 if ((ip
->i_d
.di_version
== 1) &&
1122 (ip
->i_d
.di_nlink
> XFS_MAXLINK_1
)) {
1124 * The inode has increased its number of links beyond
1125 * what can fit in an old format inode. It now needs
1126 * to be converted to a version 2 inode with a 32 bit
1127 * link count. If this is the first inode in the file
1128 * system to do this, then we need to bump the superblock
1129 * version number as well.
1131 xfs_bump_ino_vers2(tp
, ip
);
1134 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1141 struct xfs_name
*name
,
1146 int is_dir
= S_ISDIR(mode
);
1147 struct xfs_mount
*mp
= dp
->i_mount
;
1148 struct xfs_inode
*ip
= NULL
;
1149 struct xfs_trans
*tp
= NULL
;
1151 xfs_bmap_free_t free_list
;
1152 xfs_fsblock_t first_block
;
1153 bool unlock_dp_on_error
= false;
1157 struct xfs_dquot
*udqp
= NULL
;
1158 struct xfs_dquot
*gdqp
= NULL
;
1159 struct xfs_dquot
*pdqp
= NULL
;
1160 struct xfs_trans_res tres
;
1163 trace_xfs_create(dp
, name
);
1165 if (XFS_FORCED_SHUTDOWN(mp
))
1166 return XFS_ERROR(EIO
);
1168 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1169 prid
= xfs_get_projid(dp
);
1171 prid
= XFS_PROJID_DEFAULT
;
1174 * Make sure that we have allocated dquot(s) on disk.
1176 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1177 xfs_kgid_to_gid(current_fsgid()), prid
,
1178 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1179 &udqp
, &gdqp
, &pdqp
);
1185 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1186 tres
.tr_logres
= M_RES(mp
)->tr_mkdir
.tr_logres
;
1187 tres
.tr_logcount
= XFS_MKDIR_LOG_COUNT
;
1188 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1190 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1191 tres
.tr_logres
= M_RES(mp
)->tr_create
.tr_logres
;
1192 tres
.tr_logcount
= XFS_CREATE_LOG_COUNT
;
1193 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1196 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1199 * Initially assume that the file does not exist and
1200 * reserve the resources for that case. If that is not
1201 * the case we'll drop the one we have and get a more
1202 * appropriate transaction later.
1204 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1205 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1206 if (error
== ENOSPC
) {
1207 /* flush outstanding delalloc blocks and retry */
1208 xfs_flush_inodes(mp
);
1209 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1211 if (error
== ENOSPC
) {
1212 /* No space at all so try a "no-allocation" reservation */
1214 error
= xfs_trans_reserve(tp
, &tres
, 0, 0);
1218 goto out_trans_cancel
;
1221 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1222 unlock_dp_on_error
= true;
1224 xfs_bmap_init(&free_list
, &first_block
);
1227 * Reserve disk quota and the inode.
1229 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1230 pdqp
, resblks
, 1, 0);
1232 goto out_trans_cancel
;
1234 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1236 goto out_trans_cancel
;
1239 * A newly created regular or special file just has one directory
1240 * entry pointing to them, but a directory also the "." entry
1241 * pointing to itself.
1243 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1244 prid
, resblks
> 0, &ip
, &committed
);
1246 if (error
== ENOSPC
)
1247 goto out_trans_cancel
;
1248 goto out_trans_abort
;
1252 * Now we join the directory inode to the transaction. We do not do it
1253 * earlier because xfs_dir_ialloc might commit the previous transaction
1254 * (and release all the locks). An error from here on will result in
1255 * the transaction cancel unlocking dp so don't do it explicitly in the
1258 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1259 unlock_dp_on_error
= false;
1261 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1262 &first_block
, &free_list
, resblks
?
1263 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1265 ASSERT(error
!= ENOSPC
);
1266 goto out_trans_abort
;
1268 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1269 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1272 error
= xfs_dir_init(tp
, ip
, dp
);
1274 goto out_bmap_cancel
;
1276 error
= xfs_bumplink(tp
, dp
);
1278 goto out_bmap_cancel
;
1282 * If this is a synchronous mount, make sure that the
1283 * create transaction goes to disk before returning to
1286 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1287 xfs_trans_set_sync(tp
);
1290 * Attach the dquot(s) to the inodes and modify them incore.
1291 * These ids of the inode couldn't have changed since the new
1292 * inode has been locked ever since it was created.
1294 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1296 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1298 goto out_bmap_cancel
;
1300 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1302 goto out_release_inode
;
1304 xfs_qm_dqrele(udqp
);
1305 xfs_qm_dqrele(gdqp
);
1306 xfs_qm_dqrele(pdqp
);
1312 xfs_bmap_cancel(&free_list
);
1314 cancel_flags
|= XFS_TRANS_ABORT
;
1316 xfs_trans_cancel(tp
, cancel_flags
);
1319 * Wait until after the current transaction is aborted to
1320 * release the inode. This prevents recursive transactions
1321 * and deadlocks from xfs_inactive.
1326 xfs_qm_dqrele(udqp
);
1327 xfs_qm_dqrele(gdqp
);
1328 xfs_qm_dqrele(pdqp
);
1330 if (unlock_dp_on_error
)
1331 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1339 struct xfs_name
*target_name
)
1341 xfs_mount_t
*mp
= tdp
->i_mount
;
1344 xfs_bmap_free_t free_list
;
1345 xfs_fsblock_t first_block
;
1350 trace_xfs_link(tdp
, target_name
);
1352 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1354 if (XFS_FORCED_SHUTDOWN(mp
))
1355 return XFS_ERROR(EIO
);
1357 error
= xfs_qm_dqattach(sip
, 0);
1361 error
= xfs_qm_dqattach(tdp
, 0);
1365 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1366 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1367 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1368 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, resblks
, 0);
1369 if (error
== ENOSPC
) {
1371 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, 0, 0);
1378 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1380 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1381 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1384 * If we are using project inheritance, we only allow hard link
1385 * creation in our tree when the project IDs are the same; else
1386 * the tree quota mechanism could be circumvented.
1388 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1389 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1390 error
= XFS_ERROR(EXDEV
);
1394 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
1398 xfs_bmap_init(&free_list
, &first_block
);
1400 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1401 &first_block
, &free_list
, resblks
);
1404 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1405 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1407 error
= xfs_bumplink(tp
, sip
);
1412 * If this is a synchronous mount, make sure that the
1413 * link transaction goes to disk before returning to
1416 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1417 xfs_trans_set_sync(tp
);
1420 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1422 xfs_bmap_cancel(&free_list
);
1426 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1429 cancel_flags
|= XFS_TRANS_ABORT
;
1431 xfs_trans_cancel(tp
, cancel_flags
);
1437 * Free up the underlying blocks past new_size. The new size must be smaller
1438 * than the current size. This routine can be used both for the attribute and
1439 * data fork, and does not modify the inode size, which is left to the caller.
1441 * The transaction passed to this routine must have made a permanent log
1442 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1443 * given transaction and start new ones, so make sure everything involved in
1444 * the transaction is tidy before calling here. Some transaction will be
1445 * returned to the caller to be committed. The incoming transaction must
1446 * already include the inode, and both inode locks must be held exclusively.
1447 * The inode must also be "held" within the transaction. On return the inode
1448 * will be "held" within the returned transaction. This routine does NOT
1449 * require any disk space to be reserved for it within the transaction.
1451 * If we get an error, we must return with the inode locked and linked into the
1452 * current transaction. This keeps things simple for the higher level code,
1453 * because it always knows that the inode is locked and held in the transaction
1454 * that returns to it whether errors occur or not. We don't mark the inode
1455 * dirty on error so that transactions can be easily aborted if possible.
1458 xfs_itruncate_extents(
1459 struct xfs_trans
**tpp
,
1460 struct xfs_inode
*ip
,
1462 xfs_fsize_t new_size
)
1464 struct xfs_mount
*mp
= ip
->i_mount
;
1465 struct xfs_trans
*tp
= *tpp
;
1466 struct xfs_trans
*ntp
;
1467 xfs_bmap_free_t free_list
;
1468 xfs_fsblock_t first_block
;
1469 xfs_fileoff_t first_unmap_block
;
1470 xfs_fileoff_t last_block
;
1471 xfs_filblks_t unmap_len
;
1476 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1477 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1478 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1479 ASSERT(new_size
<= XFS_ISIZE(ip
));
1480 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1481 ASSERT(ip
->i_itemp
!= NULL
);
1482 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1483 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1485 trace_xfs_itruncate_extents_start(ip
, new_size
);
1488 * Since it is possible for space to become allocated beyond
1489 * the end of the file (in a crash where the space is allocated
1490 * but the inode size is not yet updated), simply remove any
1491 * blocks which show up between the new EOF and the maximum
1492 * possible file size. If the first block to be removed is
1493 * beyond the maximum file size (ie it is the same as last_block),
1494 * then there is nothing to do.
1496 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1497 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1498 if (first_unmap_block
== last_block
)
1501 ASSERT(first_unmap_block
< last_block
);
1502 unmap_len
= last_block
- first_unmap_block
+ 1;
1504 xfs_bmap_init(&free_list
, &first_block
);
1505 error
= xfs_bunmapi(tp
, ip
,
1506 first_unmap_block
, unmap_len
,
1507 xfs_bmapi_aflag(whichfork
),
1508 XFS_ITRUNC_MAX_EXTENTS
,
1509 &first_block
, &free_list
,
1512 goto out_bmap_cancel
;
1515 * Duplicate the transaction that has the permanent
1516 * reservation and commit the old transaction.
1518 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1520 xfs_trans_ijoin(tp
, ip
, 0);
1522 goto out_bmap_cancel
;
1526 * Mark the inode dirty so it will be logged and
1527 * moved forward in the log as part of every commit.
1529 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1532 ntp
= xfs_trans_dup(tp
);
1533 error
= xfs_trans_commit(tp
, 0);
1536 xfs_trans_ijoin(tp
, ip
, 0);
1542 * Transaction commit worked ok so we can drop the extra ticket
1543 * reference that we gained in xfs_trans_dup()
1545 xfs_log_ticket_put(tp
->t_ticket
);
1546 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1552 * Always re-log the inode so that our permanent transaction can keep
1553 * on rolling it forward in the log.
1555 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1557 trace_xfs_itruncate_extents_end(ip
, new_size
);
1564 * If the bunmapi call encounters an error, return to the caller where
1565 * the transaction can be properly aborted. We just need to make sure
1566 * we're not holding any resources that we were not when we came in.
1568 xfs_bmap_cancel(&free_list
);
1576 xfs_mount_t
*mp
= ip
->i_mount
;
1579 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1582 /* If this is a read-only mount, don't do this (would generate I/O) */
1583 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1586 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1590 * If we are using filestreams, and we have an unlinked
1591 * file that we are processing the last close on, then nothing
1592 * will be able to reopen and write to this file. Purge this
1593 * inode from the filestreams cache so that it doesn't delay
1594 * teardown of the inode.
1596 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1597 xfs_filestream_deassociate(ip
);
1600 * If we previously truncated this file and removed old data
1601 * in the process, we want to initiate "early" writeout on
1602 * the last close. This is an attempt to combat the notorious
1603 * NULL files problem which is particularly noticeable from a
1604 * truncate down, buffered (re-)write (delalloc), followed by
1605 * a crash. What we are effectively doing here is
1606 * significantly reducing the time window where we'd otherwise
1607 * be exposed to that problem.
1609 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1611 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1612 if (VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0) {
1613 error
= -filemap_flush(VFS_I(ip
)->i_mapping
);
1620 if (ip
->i_d
.di_nlink
== 0)
1623 if (xfs_can_free_eofblocks(ip
, false)) {
1626 * If we can't get the iolock just skip truncating the blocks
1627 * past EOF because we could deadlock with the mmap_sem
1628 * otherwise. We'll get another chance to drop them once the
1629 * last reference to the inode is dropped, so we'll never leak
1630 * blocks permanently.
1632 * Further, check if the inode is being opened, written and
1633 * closed frequently and we have delayed allocation blocks
1634 * outstanding (e.g. streaming writes from the NFS server),
1635 * truncating the blocks past EOF will cause fragmentation to
1638 * In this case don't do the truncation, either, but we have to
1639 * be careful how we detect this case. Blocks beyond EOF show
1640 * up as i_delayed_blks even when the inode is clean, so we
1641 * need to truncate them away first before checking for a dirty
1642 * release. Hence on the first dirty close we will still remove
1643 * the speculative allocation, but after that we will leave it
1646 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1649 error
= xfs_free_eofblocks(mp
, ip
, true);
1650 if (error
&& error
!= EAGAIN
)
1653 /* delalloc blocks after truncation means it really is dirty */
1654 if (ip
->i_delayed_blks
)
1655 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1661 * xfs_inactive_truncate
1663 * Called to perform a truncate when an inode becomes unlinked.
1666 xfs_inactive_truncate(
1667 struct xfs_inode
*ip
)
1669 struct xfs_mount
*mp
= ip
->i_mount
;
1670 struct xfs_trans
*tp
;
1673 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1674 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1676 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1677 xfs_trans_cancel(tp
, 0);
1681 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1682 xfs_trans_ijoin(tp
, ip
, 0);
1685 * Log the inode size first to prevent stale data exposure in the event
1686 * of a system crash before the truncate completes. See the related
1687 * comment in xfs_setattr_size() for details.
1689 ip
->i_d
.di_size
= 0;
1690 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1692 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1694 goto error_trans_cancel
;
1696 ASSERT(ip
->i_d
.di_nextents
== 0);
1698 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1702 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1706 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1708 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1713 * xfs_inactive_ifree()
1715 * Perform the inode free when an inode is unlinked.
1719 struct xfs_inode
*ip
)
1721 xfs_bmap_free_t free_list
;
1722 xfs_fsblock_t first_block
;
1724 struct xfs_mount
*mp
= ip
->i_mount
;
1725 struct xfs_trans
*tp
;
1728 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1729 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ifree
, 0, 0);
1731 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1732 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1736 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1737 xfs_trans_ijoin(tp
, ip
, 0);
1739 xfs_bmap_init(&free_list
, &first_block
);
1740 error
= xfs_ifree(tp
, ip
, &free_list
);
1743 * If we fail to free the inode, shut down. The cancel
1744 * might do that, we need to make sure. Otherwise the
1745 * inode might be lost for a long time or forever.
1747 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1748 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1750 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1752 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1753 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1758 * Credit the quota account(s). The inode is gone.
1760 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1763 * Just ignore errors at this point. There is nothing we can
1764 * do except to try to keep going. Make sure it's not a silent
1767 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1769 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1771 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1773 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1776 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1783 * This is called when the vnode reference count for the vnode
1784 * goes to zero. If the file has been unlinked, then it must
1785 * now be truncated. Also, we clear all of the read-ahead state
1786 * kept for the inode here since the file is now closed.
1792 struct xfs_mount
*mp
;
1797 * If the inode is already free, then there can be nothing
1800 if (ip
->i_d
.di_mode
== 0) {
1801 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1802 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1808 /* If this is a read-only mount, don't do this (would generate I/O) */
1809 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1812 if (ip
->i_d
.di_nlink
!= 0) {
1814 * force is true because we are evicting an inode from the
1815 * cache. Post-eof blocks must be freed, lest we end up with
1816 * broken free space accounting.
1818 if (xfs_can_free_eofblocks(ip
, true))
1819 xfs_free_eofblocks(mp
, ip
, false);
1824 if (S_ISREG(ip
->i_d
.di_mode
) &&
1825 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1826 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1829 error
= xfs_qm_dqattach(ip
, 0);
1833 if (S_ISLNK(ip
->i_d
.di_mode
))
1834 error
= xfs_inactive_symlink(ip
);
1836 error
= xfs_inactive_truncate(ip
);
1841 * If there are attributes associated with the file then blow them away
1842 * now. The code calls a routine that recursively deconstructs the
1843 * attribute fork. We need to just commit the current transaction
1844 * because we can't use it for xfs_attr_inactive().
1846 if (ip
->i_d
.di_anextents
> 0) {
1847 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1849 error
= xfs_attr_inactive(ip
);
1855 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1857 ASSERT(ip
->i_d
.di_anextents
== 0);
1862 error
= xfs_inactive_ifree(ip
);
1867 * Release the dquots held by inode, if any.
1869 xfs_qm_dqdetach(ip
);
1873 * This is called when the inode's link count goes to 0.
1874 * We place the on-disk inode on a list in the AGI. It
1875 * will be pulled from this list when the inode is freed.
1892 ASSERT(ip
->i_d
.di_nlink
== 0);
1893 ASSERT(ip
->i_d
.di_mode
!= 0);
1898 * Get the agi buffer first. It ensures lock ordering
1901 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1904 agi
= XFS_BUF_TO_AGI(agibp
);
1907 * Get the index into the agi hash table for the
1908 * list this inode will go on.
1910 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1912 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1913 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1914 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1916 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1918 * There is already another inode in the bucket we need
1919 * to add ourselves to. Add us at the front of the list.
1920 * Here we put the head pointer into our next pointer,
1921 * and then we fall through to point the head at us.
1923 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1928 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1929 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1930 offset
= ip
->i_imap
.im_boffset
+
1931 offsetof(xfs_dinode_t
, di_next_unlinked
);
1933 /* need to recalc the inode CRC if appropriate */
1934 xfs_dinode_calc_crc(mp
, dip
);
1936 xfs_trans_inode_buf(tp
, ibp
);
1937 xfs_trans_log_buf(tp
, ibp
, offset
,
1938 (offset
+ sizeof(xfs_agino_t
) - 1));
1939 xfs_inobp_check(mp
, ibp
);
1943 * Point the bucket head pointer at the inode being inserted.
1946 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1947 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1948 (sizeof(xfs_agino_t
) * bucket_index
);
1949 xfs_trans_log_buf(tp
, agibp
, offset
,
1950 (offset
+ sizeof(xfs_agino_t
) - 1));
1955 * Pull the on-disk inode from the AGI unlinked list.
1968 xfs_agnumber_t agno
;
1970 xfs_agino_t next_agino
;
1971 xfs_buf_t
*last_ibp
;
1972 xfs_dinode_t
*last_dip
= NULL
;
1974 int offset
, last_offset
= 0;
1978 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1981 * Get the agi buffer first. It ensures lock ordering
1984 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1988 agi
= XFS_BUF_TO_AGI(agibp
);
1991 * Get the index into the agi hash table for the
1992 * list this inode will go on.
1994 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1996 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1997 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
1998 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2000 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2002 * We're at the head of the list. Get the inode's on-disk
2003 * buffer to see if there is anyone after us on the list.
2004 * Only modify our next pointer if it is not already NULLAGINO.
2005 * This saves us the overhead of dealing with the buffer when
2006 * there is no need to change it.
2008 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2011 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2015 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2016 ASSERT(next_agino
!= 0);
2017 if (next_agino
!= NULLAGINO
) {
2018 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2019 offset
= ip
->i_imap
.im_boffset
+
2020 offsetof(xfs_dinode_t
, di_next_unlinked
);
2022 /* need to recalc the inode CRC if appropriate */
2023 xfs_dinode_calc_crc(mp
, dip
);
2025 xfs_trans_inode_buf(tp
, ibp
);
2026 xfs_trans_log_buf(tp
, ibp
, offset
,
2027 (offset
+ sizeof(xfs_agino_t
) - 1));
2028 xfs_inobp_check(mp
, ibp
);
2030 xfs_trans_brelse(tp
, ibp
);
2033 * Point the bucket head pointer at the next inode.
2035 ASSERT(next_agino
!= 0);
2036 ASSERT(next_agino
!= agino
);
2037 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2038 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2039 (sizeof(xfs_agino_t
) * bucket_index
);
2040 xfs_trans_log_buf(tp
, agibp
, offset
,
2041 (offset
+ sizeof(xfs_agino_t
) - 1));
2044 * We need to search the list for the inode being freed.
2046 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2048 while (next_agino
!= agino
) {
2049 struct xfs_imap imap
;
2052 xfs_trans_brelse(tp
, last_ibp
);
2055 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2057 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2060 "%s: xfs_imap returned error %d.",
2065 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2069 "%s: xfs_imap_to_bp returned error %d.",
2074 last_offset
= imap
.im_boffset
;
2075 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2076 ASSERT(next_agino
!= NULLAGINO
);
2077 ASSERT(next_agino
!= 0);
2081 * Now last_ibp points to the buffer previous to us on the
2082 * unlinked list. Pull us from the list.
2084 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2087 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2091 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2092 ASSERT(next_agino
!= 0);
2093 ASSERT(next_agino
!= agino
);
2094 if (next_agino
!= NULLAGINO
) {
2095 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2096 offset
= ip
->i_imap
.im_boffset
+
2097 offsetof(xfs_dinode_t
, di_next_unlinked
);
2099 /* need to recalc the inode CRC if appropriate */
2100 xfs_dinode_calc_crc(mp
, dip
);
2102 xfs_trans_inode_buf(tp
, ibp
);
2103 xfs_trans_log_buf(tp
, ibp
, offset
,
2104 (offset
+ sizeof(xfs_agino_t
) - 1));
2105 xfs_inobp_check(mp
, ibp
);
2107 xfs_trans_brelse(tp
, ibp
);
2110 * Point the previous inode on the list to the next inode.
2112 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2113 ASSERT(next_agino
!= 0);
2114 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2116 /* need to recalc the inode CRC if appropriate */
2117 xfs_dinode_calc_crc(mp
, last_dip
);
2119 xfs_trans_inode_buf(tp
, last_ibp
);
2120 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2121 (offset
+ sizeof(xfs_agino_t
) - 1));
2122 xfs_inobp_check(mp
, last_ibp
);
2128 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2129 * inodes that are in memory - they all must be marked stale and attached to
2130 * the cluster buffer.
2134 xfs_inode_t
*free_ip
,
2138 xfs_mount_t
*mp
= free_ip
->i_mount
;
2139 int blks_per_cluster
;
2146 xfs_inode_log_item_t
*iip
;
2147 xfs_log_item_t
*lip
;
2148 struct xfs_perag
*pag
;
2150 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2151 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2152 blks_per_cluster
= 1;
2153 ninodes
= mp
->m_sb
.sb_inopblock
;
2154 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2156 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2157 mp
->m_sb
.sb_blocksize
;
2158 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2159 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2162 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2163 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2164 XFS_INO_TO_AGBNO(mp
, inum
));
2167 * We obtain and lock the backing buffer first in the process
2168 * here, as we have to ensure that any dirty inode that we
2169 * can't get the flush lock on is attached to the buffer.
2170 * If we scan the in-memory inodes first, then buffer IO can
2171 * complete before we get a lock on it, and hence we may fail
2172 * to mark all the active inodes on the buffer stale.
2174 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2175 mp
->m_bsize
* blks_per_cluster
,
2182 * This buffer may not have been correctly initialised as we
2183 * didn't read it from disk. That's not important because we are
2184 * only using to mark the buffer as stale in the log, and to
2185 * attach stale cached inodes on it. That means it will never be
2186 * dispatched for IO. If it is, we want to know about it, and we
2187 * want it to fail. We can acheive this by adding a write
2188 * verifier to the buffer.
2190 bp
->b_ops
= &xfs_inode_buf_ops
;
2193 * Walk the inodes already attached to the buffer and mark them
2194 * stale. These will all have the flush locks held, so an
2195 * in-memory inode walk can't lock them. By marking them all
2196 * stale first, we will not attempt to lock them in the loop
2197 * below as the XFS_ISTALE flag will be set.
2201 if (lip
->li_type
== XFS_LI_INODE
) {
2202 iip
= (xfs_inode_log_item_t
*)lip
;
2203 ASSERT(iip
->ili_logged
== 1);
2204 lip
->li_cb
= xfs_istale_done
;
2205 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2206 &iip
->ili_flush_lsn
,
2207 &iip
->ili_item
.li_lsn
);
2208 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2210 lip
= lip
->li_bio_list
;
2215 * For each inode in memory attempt to add it to the inode
2216 * buffer and set it up for being staled on buffer IO
2217 * completion. This is safe as we've locked out tail pushing
2218 * and flushing by locking the buffer.
2220 * We have already marked every inode that was part of a
2221 * transaction stale above, which means there is no point in
2222 * even trying to lock them.
2224 for (i
= 0; i
< ninodes
; i
++) {
2227 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2228 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2230 /* Inode not in memory, nothing to do */
2237 * because this is an RCU protected lookup, we could
2238 * find a recently freed or even reallocated inode
2239 * during the lookup. We need to check under the
2240 * i_flags_lock for a valid inode here. Skip it if it
2241 * is not valid, the wrong inode or stale.
2243 spin_lock(&ip
->i_flags_lock
);
2244 if (ip
->i_ino
!= inum
+ i
||
2245 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2246 spin_unlock(&ip
->i_flags_lock
);
2250 spin_unlock(&ip
->i_flags_lock
);
2253 * Don't try to lock/unlock the current inode, but we
2254 * _cannot_ skip the other inodes that we did not find
2255 * in the list attached to the buffer and are not
2256 * already marked stale. If we can't lock it, back off
2259 if (ip
!= free_ip
&&
2260 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2268 xfs_iflags_set(ip
, XFS_ISTALE
);
2271 * we don't need to attach clean inodes or those only
2272 * with unlogged changes (which we throw away, anyway).
2275 if (!iip
|| xfs_inode_clean(ip
)) {
2276 ASSERT(ip
!= free_ip
);
2278 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2282 iip
->ili_last_fields
= iip
->ili_fields
;
2283 iip
->ili_fields
= 0;
2284 iip
->ili_logged
= 1;
2285 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2286 &iip
->ili_item
.li_lsn
);
2288 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2292 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2295 xfs_trans_stale_inode_buf(tp
, bp
);
2296 xfs_trans_binval(tp
, bp
);
2304 * This is called to return an inode to the inode free list.
2305 * The inode should already be truncated to 0 length and have
2306 * no pages associated with it. This routine also assumes that
2307 * the inode is already a part of the transaction.
2309 * The on-disk copy of the inode will have been added to the list
2310 * of unlinked inodes in the AGI. We need to remove the inode from
2311 * that list atomically with respect to freeing it here.
2317 xfs_bmap_free_t
*flist
)
2321 xfs_ino_t first_ino
;
2323 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2324 ASSERT(ip
->i_d
.di_nlink
== 0);
2325 ASSERT(ip
->i_d
.di_nextents
== 0);
2326 ASSERT(ip
->i_d
.di_anextents
== 0);
2327 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
2328 ASSERT(ip
->i_d
.di_nblocks
== 0);
2331 * Pull the on-disk inode from the AGI unlinked list.
2333 error
= xfs_iunlink_remove(tp
, ip
);
2337 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2341 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2342 ip
->i_d
.di_flags
= 0;
2343 ip
->i_d
.di_dmevmask
= 0;
2344 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2345 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2346 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2348 * Bump the generation count so no one will be confused
2349 * by reincarnations of this inode.
2352 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2355 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
2361 * This is called to unpin an inode. The caller must have the inode locked
2362 * in at least shared mode so that the buffer cannot be subsequently pinned
2363 * once someone is waiting for it to be unpinned.
2367 struct xfs_inode
*ip
)
2369 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2371 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2373 /* Give the log a push to start the unpinning I/O */
2374 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2380 struct xfs_inode
*ip
)
2382 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2383 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2388 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2389 if (xfs_ipincount(ip
))
2391 } while (xfs_ipincount(ip
));
2392 finish_wait(wq
, &wait
.wait
);
2397 struct xfs_inode
*ip
)
2399 if (xfs_ipincount(ip
))
2400 __xfs_iunpin_wait(ip
);
2404 * Removing an inode from the namespace involves removing the directory entry
2405 * and dropping the link count on the inode. Removing the directory entry can
2406 * result in locking an AGF (directory blocks were freed) and removing a link
2407 * count can result in placing the inode on an unlinked list which results in
2410 * The big problem here is that we have an ordering constraint on AGF and AGI
2411 * locking - inode allocation locks the AGI, then can allocate a new extent for
2412 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2413 * removes the inode from the unlinked list, requiring that we lock the AGI
2414 * first, and then freeing the inode can result in an inode chunk being freed
2415 * and hence freeing disk space requiring that we lock an AGF.
2417 * Hence the ordering that is imposed by other parts of the code is AGI before
2418 * AGF. This means we cannot remove the directory entry before we drop the inode
2419 * reference count and put it on the unlinked list as this results in a lock
2420 * order of AGF then AGI, and this can deadlock against inode allocation and
2421 * freeing. Therefore we must drop the link counts before we remove the
2424 * This is still safe from a transactional point of view - it is not until we
2425 * get to xfs_bmap_finish() that we have the possibility of multiple
2426 * transactions in this operation. Hence as long as we remove the directory
2427 * entry and drop the link count in the first transaction of the remove
2428 * operation, there are no transactional constraints on the ordering here.
2433 struct xfs_name
*name
,
2436 xfs_mount_t
*mp
= dp
->i_mount
;
2437 xfs_trans_t
*tp
= NULL
;
2438 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
2440 xfs_bmap_free_t free_list
;
2441 xfs_fsblock_t first_block
;
2448 trace_xfs_remove(dp
, name
);
2450 if (XFS_FORCED_SHUTDOWN(mp
))
2451 return XFS_ERROR(EIO
);
2453 error
= xfs_qm_dqattach(dp
, 0);
2457 error
= xfs_qm_dqattach(ip
, 0);
2462 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2463 log_count
= XFS_DEFAULT_LOG_COUNT
;
2465 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2466 log_count
= XFS_REMOVE_LOG_COUNT
;
2468 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2471 * We try to get the real space reservation first,
2472 * allowing for directory btree deletion(s) implying
2473 * possible bmap insert(s). If we can't get the space
2474 * reservation then we use 0 instead, and avoid the bmap
2475 * btree insert(s) in the directory code by, if the bmap
2476 * insert tries to happen, instead trimming the LAST
2477 * block from the directory.
2479 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2480 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, resblks
, 0);
2481 if (error
== ENOSPC
) {
2483 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
2486 ASSERT(error
!= ENOSPC
);
2488 goto out_trans_cancel
;
2491 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2493 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2494 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2497 * If we're removing a directory perform some additional validation.
2499 cancel_flags
|= XFS_TRANS_ABORT
;
2501 ASSERT(ip
->i_d
.di_nlink
>= 2);
2502 if (ip
->i_d
.di_nlink
!= 2) {
2503 error
= XFS_ERROR(ENOTEMPTY
);
2504 goto out_trans_cancel
;
2506 if (!xfs_dir_isempty(ip
)) {
2507 error
= XFS_ERROR(ENOTEMPTY
);
2508 goto out_trans_cancel
;
2511 /* Drop the link from ip's "..". */
2512 error
= xfs_droplink(tp
, dp
);
2514 goto out_trans_cancel
;
2516 /* Drop the "." link from ip to self. */
2517 error
= xfs_droplink(tp
, ip
);
2519 goto out_trans_cancel
;
2522 * When removing a non-directory we need to log the parent
2523 * inode here. For a directory this is done implicitly
2524 * by the xfs_droplink call for the ".." entry.
2526 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2528 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2530 /* Drop the link from dp to ip. */
2531 error
= xfs_droplink(tp
, ip
);
2533 goto out_trans_cancel
;
2535 /* Determine if this is the last link while the inode is locked */
2536 link_zero
= (ip
->i_d
.di_nlink
== 0);
2538 xfs_bmap_init(&free_list
, &first_block
);
2539 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2540 &first_block
, &free_list
, resblks
);
2542 ASSERT(error
!= ENOENT
);
2543 goto out_bmap_cancel
;
2547 * If this is a synchronous mount, make sure that the
2548 * remove transaction goes to disk before returning to
2551 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2552 xfs_trans_set_sync(tp
);
2554 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2556 goto out_bmap_cancel
;
2558 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2563 * If we are using filestreams, kill the stream association.
2564 * If the file is still open it may get a new one but that
2565 * will get killed on last close in xfs_close() so we don't
2566 * have to worry about that.
2568 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
2569 xfs_filestream_deassociate(ip
);
2574 xfs_bmap_cancel(&free_list
);
2576 xfs_trans_cancel(tp
, cancel_flags
);
2582 * Enter all inodes for a rename transaction into a sorted array.
2585 xfs_sort_for_rename(
2586 xfs_inode_t
*dp1
, /* in: old (source) directory inode */
2587 xfs_inode_t
*dp2
, /* in: new (target) directory inode */
2588 xfs_inode_t
*ip1
, /* in: inode of old entry */
2589 xfs_inode_t
*ip2
, /* in: inode of new entry, if it
2590 already exists, NULL otherwise. */
2591 xfs_inode_t
**i_tab
,/* out: array of inode returned, sorted */
2592 int *num_inodes
) /* out: number of inodes in array */
2598 * i_tab contains a list of pointers to inodes. We initialize
2599 * the table here & we'll sort it. We will then use it to
2600 * order the acquisition of the inode locks.
2602 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2616 * Sort the elements via bubble sort. (Remember, there are at
2617 * most 4 elements to sort, so this is adequate.)
2619 for (i
= 0; i
< *num_inodes
; i
++) {
2620 for (j
= 1; j
< *num_inodes
; j
++) {
2621 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2623 i_tab
[j
] = i_tab
[j
-1];
2635 xfs_inode_t
*src_dp
,
2636 struct xfs_name
*src_name
,
2637 xfs_inode_t
*src_ip
,
2638 xfs_inode_t
*target_dp
,
2639 struct xfs_name
*target_name
,
2640 xfs_inode_t
*target_ip
)
2642 xfs_trans_t
*tp
= NULL
;
2643 xfs_mount_t
*mp
= src_dp
->i_mount
;
2644 int new_parent
; /* moving to a new dir */
2645 int src_is_directory
; /* src_name is a directory */
2647 xfs_bmap_free_t free_list
;
2648 xfs_fsblock_t first_block
;
2651 xfs_inode_t
*inodes
[4];
2655 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2657 new_parent
= (src_dp
!= target_dp
);
2658 src_is_directory
= S_ISDIR(src_ip
->i_d
.di_mode
);
2660 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
,
2661 inodes
, &num_inodes
);
2663 xfs_bmap_init(&free_list
, &first_block
);
2664 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RENAME
);
2665 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2666 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2667 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, spaceres
, 0);
2668 if (error
== ENOSPC
) {
2670 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, 0, 0);
2673 xfs_trans_cancel(tp
, 0);
2678 * Attach the dquots to the inodes
2680 error
= xfs_qm_vop_rename_dqattach(inodes
);
2682 xfs_trans_cancel(tp
, cancel_flags
);
2687 * Lock all the participating inodes. Depending upon whether
2688 * the target_name exists in the target directory, and
2689 * whether the target directory is the same as the source
2690 * directory, we can lock from 2 to 4 inodes.
2692 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2695 * Join all the inodes to the transaction. From this point on,
2696 * we can rely on either trans_commit or trans_cancel to unlock
2699 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2701 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2702 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2704 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2707 * If we are using project inheritance, we only allow renames
2708 * into our tree when the project IDs are the same; else the
2709 * tree quota mechanism would be circumvented.
2711 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2712 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2713 error
= XFS_ERROR(EXDEV
);
2718 * Set up the target.
2720 if (target_ip
== NULL
) {
2722 * If there's no space reservation, check the entry will
2723 * fit before actually inserting it.
2725 error
= xfs_dir_canenter(tp
, target_dp
, target_name
, spaceres
);
2729 * If target does not exist and the rename crosses
2730 * directories, adjust the target directory link count
2731 * to account for the ".." reference from the new entry.
2733 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
2734 src_ip
->i_ino
, &first_block
,
2735 &free_list
, spaceres
);
2736 if (error
== ENOSPC
)
2741 xfs_trans_ichgtime(tp
, target_dp
,
2742 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2744 if (new_parent
&& src_is_directory
) {
2745 error
= xfs_bumplink(tp
, target_dp
);
2749 } else { /* target_ip != NULL */
2751 * If target exists and it's a directory, check that both
2752 * target and source are directories and that target can be
2753 * destroyed, or that neither is a directory.
2755 if (S_ISDIR(target_ip
->i_d
.di_mode
)) {
2757 * Make sure target dir is empty.
2759 if (!(xfs_dir_isempty(target_ip
)) ||
2760 (target_ip
->i_d
.di_nlink
> 2)) {
2761 error
= XFS_ERROR(EEXIST
);
2767 * Link the source inode under the target name.
2768 * If the source inode is a directory and we are moving
2769 * it across directories, its ".." entry will be
2770 * inconsistent until we replace that down below.
2772 * In case there is already an entry with the same
2773 * name at the destination directory, remove it first.
2775 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
2777 &first_block
, &free_list
, spaceres
);
2781 xfs_trans_ichgtime(tp
, target_dp
,
2782 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2785 * Decrement the link count on the target since the target
2786 * dir no longer points to it.
2788 error
= xfs_droplink(tp
, target_ip
);
2792 if (src_is_directory
) {
2794 * Drop the link from the old "." entry.
2796 error
= xfs_droplink(tp
, target_ip
);
2800 } /* target_ip != NULL */
2803 * Remove the source.
2805 if (new_parent
&& src_is_directory
) {
2807 * Rewrite the ".." entry to point to the new
2810 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
2812 &first_block
, &free_list
, spaceres
);
2813 ASSERT(error
!= EEXIST
);
2819 * We always want to hit the ctime on the source inode.
2821 * This isn't strictly required by the standards since the source
2822 * inode isn't really being changed, but old unix file systems did
2823 * it and some incremental backup programs won't work without it.
2825 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
2826 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
2829 * Adjust the link count on src_dp. This is necessary when
2830 * renaming a directory, either within one parent when
2831 * the target existed, or across two parent directories.
2833 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
2836 * Decrement link count on src_directory since the
2837 * entry that's moved no longer points to it.
2839 error
= xfs_droplink(tp
, src_dp
);
2844 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
2845 &first_block
, &free_list
, spaceres
);
2849 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2850 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
2852 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
2855 * If this is a synchronous mount, make sure that the
2856 * rename transaction goes to disk before returning to
2859 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2860 xfs_trans_set_sync(tp
);
2863 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2865 xfs_bmap_cancel(&free_list
);
2866 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
2872 * trans_commit will unlock src_ip, target_ip & decrement
2873 * the vnode references.
2875 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2878 cancel_flags
|= XFS_TRANS_ABORT
;
2880 xfs_bmap_cancel(&free_list
);
2881 xfs_trans_cancel(tp
, cancel_flags
);
2891 xfs_mount_t
*mp
= ip
->i_mount
;
2892 struct xfs_perag
*pag
;
2893 unsigned long first_index
, mask
;
2894 unsigned long inodes_per_cluster
;
2896 xfs_inode_t
**ilist
;
2903 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2905 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2906 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2907 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2911 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2912 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2914 /* really need a gang lookup range call here */
2915 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2916 first_index
, inodes_per_cluster
);
2920 for (i
= 0; i
< nr_found
; i
++) {
2926 * because this is an RCU protected lookup, we could find a
2927 * recently freed or even reallocated inode during the lookup.
2928 * We need to check under the i_flags_lock for a valid inode
2929 * here. Skip it if it is not valid or the wrong inode.
2931 spin_lock(&ip
->i_flags_lock
);
2933 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2934 spin_unlock(&ip
->i_flags_lock
);
2937 spin_unlock(&ip
->i_flags_lock
);
2940 * Do an un-protected check to see if the inode is dirty and
2941 * is a candidate for flushing. These checks will be repeated
2942 * later after the appropriate locks are acquired.
2944 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2948 * Try to get locks. If any are unavailable or it is pinned,
2949 * then this inode cannot be flushed and is skipped.
2952 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2954 if (!xfs_iflock_nowait(iq
)) {
2955 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2958 if (xfs_ipincount(iq
)) {
2960 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2965 * arriving here means that this inode can be flushed. First
2966 * re-check that it's dirty before flushing.
2968 if (!xfs_inode_clean(iq
)) {
2970 error
= xfs_iflush_int(iq
, bp
);
2972 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2973 goto cluster_corrupt_out
;
2979 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2983 XFS_STATS_INC(xs_icluster_flushcnt
);
2984 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2995 cluster_corrupt_out
:
2997 * Corruption detected in the clustering loop. Invalidate the
2998 * inode buffer and shut down the filesystem.
3002 * Clean up the buffer. If it was delwri, just release it --
3003 * brelse can handle it with no problems. If not, shut down the
3004 * filesystem before releasing the buffer.
3006 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3010 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3012 if (!bufwasdelwri
) {
3014 * Just like incore_relse: if we have b_iodone functions,
3015 * mark the buffer as an error and call them. Otherwise
3016 * mark it as stale and brelse.
3021 xfs_buf_ioerror(bp
, EIO
);
3022 xfs_buf_ioend(bp
, 0);
3030 * Unlocks the flush lock
3032 xfs_iflush_abort(iq
, false);
3035 return XFS_ERROR(EFSCORRUPTED
);
3039 * Flush dirty inode metadata into the backing buffer.
3041 * The caller must have the inode lock and the inode flush lock held. The
3042 * inode lock will still be held upon return to the caller, and the inode
3043 * flush lock will be released after the inode has reached the disk.
3045 * The caller must write out the buffer returned in *bpp and release it.
3049 struct xfs_inode
*ip
,
3050 struct xfs_buf
**bpp
)
3052 struct xfs_mount
*mp
= ip
->i_mount
;
3054 struct xfs_dinode
*dip
;
3057 XFS_STATS_INC(xs_iflush_count
);
3059 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3060 ASSERT(xfs_isiflocked(ip
));
3061 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3062 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3066 xfs_iunpin_wait(ip
);
3069 * For stale inodes we cannot rely on the backing buffer remaining
3070 * stale in cache for the remaining life of the stale inode and so
3071 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3072 * inodes below. We have to check this after ensuring the inode is
3073 * unpinned so that it is safe to reclaim the stale inode after the
3076 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3082 * This may have been unpinned because the filesystem is shutting
3083 * down forcibly. If that's the case we must not write this inode
3084 * to disk, because the log record didn't make it to disk.
3086 * We also have to remove the log item from the AIL in this case,
3087 * as we wait for an empty AIL as part of the unmount process.
3089 if (XFS_FORCED_SHUTDOWN(mp
)) {
3090 error
= XFS_ERROR(EIO
);
3095 * Get the buffer containing the on-disk inode.
3097 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3105 * First flush out the inode that xfs_iflush was called with.
3107 error
= xfs_iflush_int(ip
, bp
);
3112 * If the buffer is pinned then push on the log now so we won't
3113 * get stuck waiting in the write for too long.
3115 if (xfs_buf_ispinned(bp
))
3116 xfs_log_force(mp
, 0);
3120 * see if other inodes can be gathered into this write
3122 error
= xfs_iflush_cluster(ip
, bp
);
3124 goto cluster_corrupt_out
;
3131 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3132 cluster_corrupt_out
:
3133 error
= XFS_ERROR(EFSCORRUPTED
);
3136 * Unlocks the flush lock
3138 xfs_iflush_abort(ip
, false);
3144 struct xfs_inode
*ip
,
3147 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3148 struct xfs_dinode
*dip
;
3149 struct xfs_mount
*mp
= ip
->i_mount
;
3151 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3152 ASSERT(xfs_isiflocked(ip
));
3153 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3154 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3155 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3157 /* set *dip = inode's place in the buffer */
3158 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3160 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3161 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3162 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3163 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3164 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3167 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3168 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3169 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3170 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3171 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3174 if (S_ISREG(ip
->i_d
.di_mode
)) {
3176 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3177 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3178 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3179 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3180 "%s: Bad regular inode %Lu, ptr 0x%p",
3181 __func__
, ip
->i_ino
, ip
);
3184 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
3186 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3187 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3188 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3189 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3190 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3191 "%s: Bad directory inode %Lu, ptr 0x%p",
3192 __func__
, ip
->i_ino
, ip
);
3196 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3197 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3198 XFS_RANDOM_IFLUSH_5
)) {
3199 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3200 "%s: detected corrupt incore inode %Lu, "
3201 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3202 __func__
, ip
->i_ino
,
3203 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3204 ip
->i_d
.di_nblocks
, ip
);
3207 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3208 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3209 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3210 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3211 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3216 * Inode item log recovery for v1/v2 inodes are dependent on the
3217 * di_flushiter count for correct sequencing. We bump the flush
3218 * iteration count so we can detect flushes which postdate a log record
3219 * during recovery. This is redundant as we now log every change and
3220 * hence this can't happen but we need to still do it to ensure
3221 * backwards compatibility with old kernels that predate logging all
3224 if (ip
->i_d
.di_version
< 3)
3225 ip
->i_d
.di_flushiter
++;
3228 * Copy the dirty parts of the inode into the on-disk
3229 * inode. We always copy out the core of the inode,
3230 * because if the inode is dirty at all the core must
3233 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3235 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3236 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3237 ip
->i_d
.di_flushiter
= 0;
3240 * If this is really an old format inode and the superblock version
3241 * has not been updated to support only new format inodes, then
3242 * convert back to the old inode format. If the superblock version
3243 * has been updated, then make the conversion permanent.
3245 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
3246 if (ip
->i_d
.di_version
== 1) {
3247 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3251 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3252 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3255 * The superblock version has already been bumped,
3256 * so just make the conversion to the new inode
3259 ip
->i_d
.di_version
= 2;
3260 dip
->di_version
= 2;
3261 ip
->i_d
.di_onlink
= 0;
3263 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3264 memset(&(dip
->di_pad
[0]), 0,
3265 sizeof(dip
->di_pad
));
3266 ASSERT(xfs_get_projid(ip
) == 0);
3270 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3271 if (XFS_IFORK_Q(ip
))
3272 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3273 xfs_inobp_check(mp
, bp
);
3276 * We've recorded everything logged in the inode, so we'd like to clear
3277 * the ili_fields bits so we don't log and flush things unnecessarily.
3278 * However, we can't stop logging all this information until the data
3279 * we've copied into the disk buffer is written to disk. If we did we
3280 * might overwrite the copy of the inode in the log with all the data
3281 * after re-logging only part of it, and in the face of a crash we
3282 * wouldn't have all the data we need to recover.
3284 * What we do is move the bits to the ili_last_fields field. When
3285 * logging the inode, these bits are moved back to the ili_fields field.
3286 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3287 * know that the information those bits represent is permanently on
3288 * disk. As long as the flush completes before the inode is logged
3289 * again, then both ili_fields and ili_last_fields will be cleared.
3291 * We can play with the ili_fields bits here, because the inode lock
3292 * must be held exclusively in order to set bits there and the flush
3293 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3294 * done routine can tell whether or not to look in the AIL. Also, store
3295 * the current LSN of the inode so that we can tell whether the item has
3296 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3297 * need the AIL lock, because it is a 64 bit value that cannot be read
3300 iip
->ili_last_fields
= iip
->ili_fields
;
3301 iip
->ili_fields
= 0;
3302 iip
->ili_logged
= 1;
3304 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3305 &iip
->ili_item
.li_lsn
);
3308 * Attach the function xfs_iflush_done to the inode's
3309 * buffer. This will remove the inode from the AIL
3310 * and unlock the inode's flush lock when the inode is
3311 * completely written to disk.
3313 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3315 /* update the lsn in the on disk inode if required */
3316 if (ip
->i_d
.di_version
== 3)
3317 dip
->di_lsn
= cpu_to_be64(iip
->ili_item
.li_lsn
);
3319 /* generate the checksum. */
3320 xfs_dinode_calc_crc(mp
, dip
);
3322 ASSERT(bp
->b_fspriv
!= NULL
);
3323 ASSERT(bp
->b_iodone
!= NULL
);
3327 return XFS_ERROR(EFSCORRUPTED
);