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_format.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trans_priv.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_dir2_format.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
47 #include "xfs_bmap_util.h"
48 #include "xfs_error.h"
49 #include "xfs_quota.h"
50 #include "xfs_filestream.h"
51 #include "xfs_cksum.h"
52 #include "xfs_trace.h"
53 #include "xfs_icache.h"
54 #include "xfs_symlink.h"
56 kmem_zone_t
*xfs_inode_zone
;
59 * Used in xfs_itruncate_extents(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
67 * helper function to extract extent size hint from inode
73 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
74 return ip
->i_d
.di_extsize
;
75 if (XFS_IS_REALTIME_INODE(ip
))
76 return ip
->i_mount
->m_sb
.sb_rextsize
;
81 * This is a wrapper routine around the xfs_ilock() routine used to centralize
82 * some grungy code. It is used in places that wish to lock the inode solely
83 * for reading the extents. The reason these places can't just call
84 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
85 * extents from disk for a file in b-tree format. If the inode is in b-tree
86 * format, then we need to lock the inode exclusively until the extents are read
87 * in. Locking it exclusively all the time would limit our parallelism
88 * unnecessarily, though. What we do instead is check to see if the extents
89 * have been read in yet, and only lock the inode exclusively if they have not.
91 * The function returns a value which should be given to the corresponding
92 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
101 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
102 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
103 lock_mode
= XFS_ILOCK_EXCL
;
105 lock_mode
= XFS_ILOCK_SHARED
;
108 xfs_ilock(ip
, lock_mode
);
114 * This is simply the unlock routine to go with xfs_ilock_map_shared().
115 * All it does is call xfs_iunlock() with the given lock_mode.
118 xfs_iunlock_map_shared(
120 unsigned int lock_mode
)
122 xfs_iunlock(ip
, lock_mode
);
126 * The xfs inode contains 2 locks: a multi-reader lock called the
127 * i_iolock and a multi-reader lock called the i_lock. This routine
128 * allows either or both of the locks to be obtained.
130 * The 2 locks should always be ordered so that the IO lock is
131 * obtained first in order to prevent deadlock.
133 * ip -- the inode being locked
134 * lock_flags -- this parameter indicates the inode's locks
135 * to be locked. It can be:
140 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
141 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
142 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
143 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
150 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
153 * You can't set both SHARED and EXCL for the same lock,
154 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
155 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
157 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
158 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
159 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
160 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
161 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
163 if (lock_flags
& XFS_IOLOCK_EXCL
)
164 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
165 else if (lock_flags
& XFS_IOLOCK_SHARED
)
166 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
168 if (lock_flags
& XFS_ILOCK_EXCL
)
169 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
170 else if (lock_flags
& XFS_ILOCK_SHARED
)
171 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
175 * This is just like xfs_ilock(), except that the caller
176 * is guaranteed not to sleep. It returns 1 if it gets
177 * the requested locks and 0 otherwise. If the IO lock is
178 * obtained but the inode lock cannot be, then the IO lock
179 * is dropped before returning.
181 * ip -- the inode being locked
182 * lock_flags -- this parameter indicates the inode's locks to be
183 * to be locked. See the comment for xfs_ilock() for a list
191 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
194 * You can't set both SHARED and EXCL for the same lock,
195 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
196 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
198 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
199 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
200 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
201 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
202 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
204 if (lock_flags
& XFS_IOLOCK_EXCL
) {
205 if (!mrtryupdate(&ip
->i_iolock
))
207 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
208 if (!mrtryaccess(&ip
->i_iolock
))
211 if (lock_flags
& XFS_ILOCK_EXCL
) {
212 if (!mrtryupdate(&ip
->i_lock
))
213 goto out_undo_iolock
;
214 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
215 if (!mrtryaccess(&ip
->i_lock
))
216 goto out_undo_iolock
;
221 if (lock_flags
& XFS_IOLOCK_EXCL
)
222 mrunlock_excl(&ip
->i_iolock
);
223 else if (lock_flags
& XFS_IOLOCK_SHARED
)
224 mrunlock_shared(&ip
->i_iolock
);
230 * xfs_iunlock() is used to drop the inode locks acquired with
231 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
232 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
233 * that we know which locks to drop.
235 * ip -- the inode being unlocked
236 * lock_flags -- this parameter indicates the inode's locks to be
237 * to be unlocked. See the comment for xfs_ilock() for a list
238 * of valid values for this parameter.
247 * You can't set both SHARED and EXCL for the same lock,
248 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
249 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
251 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
252 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
253 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
254 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
255 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
256 ASSERT(lock_flags
!= 0);
258 if (lock_flags
& XFS_IOLOCK_EXCL
)
259 mrunlock_excl(&ip
->i_iolock
);
260 else if (lock_flags
& XFS_IOLOCK_SHARED
)
261 mrunlock_shared(&ip
->i_iolock
);
263 if (lock_flags
& XFS_ILOCK_EXCL
)
264 mrunlock_excl(&ip
->i_lock
);
265 else if (lock_flags
& XFS_ILOCK_SHARED
)
266 mrunlock_shared(&ip
->i_lock
);
268 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
272 * give up write locks. the i/o lock cannot be held nested
273 * if it is being demoted.
280 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
281 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
283 if (lock_flags
& XFS_ILOCK_EXCL
)
284 mrdemote(&ip
->i_lock
);
285 if (lock_flags
& XFS_IOLOCK_EXCL
)
286 mrdemote(&ip
->i_iolock
);
288 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
291 #if defined(DEBUG) || defined(XFS_WARN)
297 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
298 if (!(lock_flags
& XFS_ILOCK_SHARED
))
299 return !!ip
->i_lock
.mr_writer
;
300 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
303 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
304 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
305 return !!ip
->i_iolock
.mr_writer
;
306 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);
316 int xfs_small_retries
;
317 int xfs_middle_retries
;
318 int xfs_lots_retries
;
323 * Bump the subclass so xfs_lock_inodes() acquires each lock with
327 xfs_lock_inumorder(int lock_mode
, int subclass
)
329 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
330 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
331 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
332 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
338 * The following routine will lock n inodes in exclusive mode.
339 * We assume the caller calls us with the inodes in i_ino order.
341 * We need to detect deadlock where an inode that we lock
342 * is in the AIL and we start waiting for another inode that is locked
343 * by a thread in a long running transaction (such as truncate). This can
344 * result in deadlock since the long running trans might need to wait
345 * for the inode we just locked in order to push the tail and free space
354 int attempts
= 0, i
, j
, try_lock
;
357 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
363 for (; i
< inodes
; i
++) {
366 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
370 * If try_lock is not set yet, make sure all locked inodes
371 * are not in the AIL.
372 * If any are, set try_lock to be used later.
376 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
377 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
378 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
385 * If any of the previous locks we have locked is in the AIL,
386 * we must TRY to get the second and subsequent locks. If
387 * we can't get any, we must release all we have
392 /* try_lock must be 0 if i is 0. */
394 * try_lock means we have an inode locked
395 * that is in the AIL.
398 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
402 * Unlock all previous guys and try again.
403 * xfs_iunlock will try to push the tail
404 * if the inode is in the AIL.
407 for(j
= i
- 1; j
>= 0; j
--) {
410 * Check to see if we've already
412 * Not the first one going back,
413 * and the inode ptr is the same.
415 if ((j
!= (i
- 1)) && ips
[j
] ==
419 xfs_iunlock(ips
[j
], lock_mode
);
422 if ((attempts
% 5) == 0) {
423 delay(1); /* Don't just spin the CPU */
433 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
439 if (attempts
< 5) xfs_small_retries
++;
440 else if (attempts
< 100) xfs_middle_retries
++;
441 else xfs_lots_retries
++;
449 * xfs_lock_two_inodes() can only be used to lock one type of lock
450 * at a time - the iolock or the ilock, but not both at once. If
451 * we lock both at once, lockdep will report false positives saying
452 * we have violated locking orders.
464 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
465 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
466 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
468 if (ip0
->i_ino
> ip1
->i_ino
) {
475 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
478 * If the first lock we have locked is in the AIL, we must TRY to get
479 * the second lock. If we can't get it, we must release the first one
482 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
483 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
484 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
485 xfs_iunlock(ip0
, lock_mode
);
486 if ((++attempts
% 5) == 0)
487 delay(1); /* Don't just spin the CPU */
491 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
498 struct xfs_inode
*ip
)
500 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
501 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
504 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
505 if (xfs_isiflocked(ip
))
507 } while (!xfs_iflock_nowait(ip
));
509 finish_wait(wq
, &wait
.wait
);
518 if (di_flags
& XFS_DIFLAG_ANY
) {
519 if (di_flags
& XFS_DIFLAG_REALTIME
)
520 flags
|= XFS_XFLAG_REALTIME
;
521 if (di_flags
& XFS_DIFLAG_PREALLOC
)
522 flags
|= XFS_XFLAG_PREALLOC
;
523 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
524 flags
|= XFS_XFLAG_IMMUTABLE
;
525 if (di_flags
& XFS_DIFLAG_APPEND
)
526 flags
|= XFS_XFLAG_APPEND
;
527 if (di_flags
& XFS_DIFLAG_SYNC
)
528 flags
|= XFS_XFLAG_SYNC
;
529 if (di_flags
& XFS_DIFLAG_NOATIME
)
530 flags
|= XFS_XFLAG_NOATIME
;
531 if (di_flags
& XFS_DIFLAG_NODUMP
)
532 flags
|= XFS_XFLAG_NODUMP
;
533 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
534 flags
|= XFS_XFLAG_RTINHERIT
;
535 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
536 flags
|= XFS_XFLAG_PROJINHERIT
;
537 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
538 flags
|= XFS_XFLAG_NOSYMLINKS
;
539 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
540 flags
|= XFS_XFLAG_EXTSIZE
;
541 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
542 flags
|= XFS_XFLAG_EXTSZINHERIT
;
543 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
544 flags
|= XFS_XFLAG_NODEFRAG
;
545 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
546 flags
|= XFS_XFLAG_FILESTREAM
;
556 xfs_icdinode_t
*dic
= &ip
->i_d
;
558 return _xfs_dic2xflags(dic
->di_flags
) |
559 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
566 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
567 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
571 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
572 * is allowed, otherwise it has to be an exact match. If a CI match is found,
573 * ci_name->name will point to a the actual name (caller must free) or
574 * will be set to NULL if an exact match is found.
579 struct xfs_name
*name
,
581 struct xfs_name
*ci_name
)
587 trace_xfs_lookup(dp
, name
);
589 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
590 return XFS_ERROR(EIO
);
592 lock_mode
= xfs_ilock_map_shared(dp
);
593 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
594 xfs_iunlock_map_shared(dp
, lock_mode
);
599 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
607 kmem_free(ci_name
->name
);
614 * Allocate an inode on disk and return a copy of its in-core version.
615 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
616 * appropriately within the inode. The uid and gid for the inode are
617 * set according to the contents of the given cred structure.
619 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
620 * has a free inode available, call xfs_iget() to obtain the in-core
621 * version of the allocated inode. Finally, fill in the inode and
622 * log its initial contents. In this case, ialloc_context would be
625 * If xfs_dialloc() does not have an available inode, it will replenish
626 * its supply by doing an allocation. Since we can only do one
627 * allocation within a transaction without deadlocks, we must commit
628 * the current transaction before returning the inode itself.
629 * In this case, therefore, we will set ialloc_context and return.
630 * The caller should then commit the current transaction, start a new
631 * transaction, and call xfs_ialloc() again to actually get the inode.
633 * To ensure that some other process does not grab the inode that
634 * was allocated during the first call to xfs_ialloc(), this routine
635 * also returns the [locked] bp pointing to the head of the freelist
636 * as ialloc_context. The caller should hold this buffer across
637 * the commit and pass it back into this routine on the second call.
639 * If we are allocating quota inodes, we do not have a parent inode
640 * to attach to or associate with (i.e. pip == NULL) because they
641 * are not linked into the directory structure - they are attached
642 * directly to the superblock - and so have no parent.
653 xfs_buf_t
**ialloc_context
,
656 struct xfs_mount
*mp
= tp
->t_mountp
;
665 * Call the space management code to pick
666 * the on-disk inode to be allocated.
668 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
669 ialloc_context
, &ino
);
672 if (*ialloc_context
|| ino
== NULLFSINO
) {
676 ASSERT(*ialloc_context
== NULL
);
679 * Get the in-core inode with the lock held exclusively.
680 * This is because we're setting fields here we need
681 * to prevent others from looking at until we're done.
683 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
684 XFS_ILOCK_EXCL
, &ip
);
689 ip
->i_d
.di_mode
= mode
;
690 ip
->i_d
.di_onlink
= 0;
691 ip
->i_d
.di_nlink
= nlink
;
692 ASSERT(ip
->i_d
.di_nlink
== nlink
);
693 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
694 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
695 xfs_set_projid(ip
, prid
);
696 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
699 * If the superblock version is up to where we support new format
700 * inodes and this is currently an old format inode, then change
701 * the inode version number now. This way we only do the conversion
702 * here rather than here and in the flush/logging code.
704 if (xfs_sb_version_hasnlink(&mp
->m_sb
) &&
705 ip
->i_d
.di_version
== 1) {
706 ip
->i_d
.di_version
= 2;
708 * We've already zeroed the old link count, the projid field,
714 * Project ids won't be stored on disk if we are using a version 1 inode.
716 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
717 xfs_bump_ino_vers2(tp
, ip
);
719 if (pip
&& XFS_INHERIT_GID(pip
)) {
720 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
721 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
722 ip
->i_d
.di_mode
|= S_ISGID
;
727 * If the group ID of the new file does not match the effective group
728 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
729 * (and only if the irix_sgid_inherit compatibility variable is set).
731 if ((irix_sgid_inherit
) &&
732 (ip
->i_d
.di_mode
& S_ISGID
) &&
733 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
)))) {
734 ip
->i_d
.di_mode
&= ~S_ISGID
;
738 ip
->i_d
.di_nextents
= 0;
739 ASSERT(ip
->i_d
.di_nblocks
== 0);
742 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
743 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
744 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
745 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
748 * di_gen will have been taken care of in xfs_iread.
750 ip
->i_d
.di_extsize
= 0;
751 ip
->i_d
.di_dmevmask
= 0;
752 ip
->i_d
.di_dmstate
= 0;
753 ip
->i_d
.di_flags
= 0;
755 if (ip
->i_d
.di_version
== 3) {
756 ASSERT(ip
->i_d
.di_ino
== ino
);
757 ASSERT(uuid_equal(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
));
759 ip
->i_d
.di_changecount
= 1;
761 ip
->i_d
.di_flags2
= 0;
762 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
763 ip
->i_d
.di_crtime
= ip
->i_d
.di_mtime
;
767 flags
= XFS_ILOG_CORE
;
768 switch (mode
& S_IFMT
) {
773 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
774 ip
->i_df
.if_u2
.if_rdev
= rdev
;
775 ip
->i_df
.if_flags
= 0;
776 flags
|= XFS_ILOG_DEV
;
780 * we can't set up filestreams until after the VFS inode
781 * is set up properly.
783 if (pip
&& xfs_inode_is_filestream(pip
))
787 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
791 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
792 di_flags
|= XFS_DIFLAG_RTINHERIT
;
793 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
794 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
795 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
797 } else if (S_ISREG(mode
)) {
798 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
799 di_flags
|= XFS_DIFLAG_REALTIME
;
800 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
801 di_flags
|= XFS_DIFLAG_EXTSIZE
;
802 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
805 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
807 di_flags
|= XFS_DIFLAG_NOATIME
;
808 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
810 di_flags
|= XFS_DIFLAG_NODUMP
;
811 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
813 di_flags
|= XFS_DIFLAG_SYNC
;
814 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
815 xfs_inherit_nosymlinks
)
816 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
817 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
818 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
819 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
820 xfs_inherit_nodefrag
)
821 di_flags
|= XFS_DIFLAG_NODEFRAG
;
822 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
823 di_flags
|= XFS_DIFLAG_FILESTREAM
;
824 ip
->i_d
.di_flags
|= di_flags
;
828 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
829 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
830 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
831 ip
->i_df
.if_u1
.if_extents
= NULL
;
837 * Attribute fork settings for new inode.
839 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
840 ip
->i_d
.di_anextents
= 0;
843 * Log the new values stuffed into the inode.
845 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
846 xfs_trans_log_inode(tp
, ip
, flags
);
848 /* now that we have an i_mode we can setup inode ops and unlock */
851 /* now we have set up the vfs inode we can associate the filestream */
853 error
= xfs_filestream_associate(pip
, ip
);
857 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
865 * Allocates a new inode from disk and return a pointer to the
866 * incore copy. This routine will internally commit the current
867 * transaction and allocate a new one if the Space Manager needed
868 * to do an allocation to replenish the inode free-list.
870 * This routine is designed to be called from xfs_create and
876 xfs_trans_t
**tpp
, /* input: current transaction;
877 output: may be a new transaction. */
878 xfs_inode_t
*dp
, /* directory within whose allocate
883 prid_t prid
, /* project id */
884 int okalloc
, /* ok to allocate new space */
885 xfs_inode_t
**ipp
, /* pointer to inode; it will be
893 xfs_buf_t
*ialloc_context
= NULL
;
899 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
902 * xfs_ialloc will return a pointer to an incore inode if
903 * the Space Manager has an available inode on the free
904 * list. Otherwise, it will do an allocation and replenish
905 * the freelist. Since we can only do one allocation per
906 * transaction without deadlocks, we will need to commit the
907 * current transaction and start a new one. We will then
908 * need to call xfs_ialloc again to get the inode.
910 * If xfs_ialloc did an allocation to replenish the freelist,
911 * it returns the bp containing the head of the freelist as
912 * ialloc_context. We will hold a lock on it across the
913 * transaction commit so that no other process can steal
914 * the inode(s) that we've just allocated.
916 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
917 &ialloc_context
, &ip
);
920 * Return an error if we were unable to allocate a new inode.
921 * This should only happen if we run out of space on disk or
922 * encounter a disk error.
928 if (!ialloc_context
&& !ip
) {
930 return XFS_ERROR(ENOSPC
);
934 * If the AGI buffer is non-NULL, then we were unable to get an
935 * inode in one operation. We need to commit the current
936 * transaction and call xfs_ialloc() again. It is guaranteed
937 * to succeed the second time.
939 if (ialloc_context
) {
940 struct xfs_trans_res tres
;
943 * Normally, xfs_trans_commit releases all the locks.
944 * We call bhold to hang on to the ialloc_context across
945 * the commit. Holding this buffer prevents any other
946 * processes from doing any allocations in this
949 xfs_trans_bhold(tp
, ialloc_context
);
951 * Save the log reservation so we can use
952 * them in the next transaction.
954 tres
.tr_logres
= xfs_trans_get_log_res(tp
);
955 tres
.tr_logcount
= xfs_trans_get_log_count(tp
);
958 * We want the quota changes to be associated with the next
959 * transaction, NOT this one. So, detach the dqinfo from this
960 * and attach it to the next transaction.
965 dqinfo
= (void *)tp
->t_dqinfo
;
967 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
968 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
971 ntp
= xfs_trans_dup(tp
);
972 code
= xfs_trans_commit(tp
, 0);
974 if (committed
!= NULL
) {
978 * If we get an error during the commit processing,
979 * release the buffer that is still held and return
983 xfs_buf_relse(ialloc_context
);
985 tp
->t_dqinfo
= dqinfo
;
986 xfs_trans_free_dqinfo(tp
);
994 * transaction commit worked ok so we can drop the extra ticket
995 * reference that we gained in xfs_trans_dup()
997 xfs_log_ticket_put(tp
->t_ticket
);
998 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
999 code
= xfs_trans_reserve(tp
, &tres
, 0, 0);
1002 * Re-attach the quota info that we detached from prev trx.
1005 tp
->t_dqinfo
= dqinfo
;
1006 tp
->t_flags
|= tflags
;
1010 xfs_buf_relse(ialloc_context
);
1015 xfs_trans_bjoin(tp
, ialloc_context
);
1018 * Call ialloc again. Since we've locked out all
1019 * other allocations in this allocation group,
1020 * this call should always succeed.
1022 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1023 okalloc
, &ialloc_context
, &ip
);
1026 * If we get an error at this point, return to the caller
1027 * so that the current transaction can be aborted.
1034 ASSERT(!ialloc_context
&& ip
);
1037 if (committed
!= NULL
)
1048 * Decrement the link count on an inode & log the change.
1049 * If this causes the link count to go to zero, initiate the
1050 * logging activity required to truncate a file.
1059 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1061 ASSERT (ip
->i_d
.di_nlink
> 0);
1063 drop_nlink(VFS_I(ip
));
1064 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1067 if (ip
->i_d
.di_nlink
== 0) {
1069 * We're dropping the last link to this file.
1070 * Move the on-disk inode to the AGI unlinked list.
1071 * From xfs_inactive() we will pull the inode from
1072 * the list and free it.
1074 error
= xfs_iunlink(tp
, ip
);
1080 * This gets called when the inode's version needs to be changed from 1 to 2.
1081 * Currently this happens when the nlink field overflows the old 16-bit value
1082 * or when chproj is called to change the project for the first time.
1083 * As a side effect the superblock version will also get rev'd
1084 * to contain the NLINK bit.
1093 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1094 ASSERT(ip
->i_d
.di_version
== 1);
1096 ip
->i_d
.di_version
= 2;
1097 ip
->i_d
.di_onlink
= 0;
1098 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1100 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1101 spin_lock(&mp
->m_sb_lock
);
1102 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1103 xfs_sb_version_addnlink(&mp
->m_sb
);
1104 spin_unlock(&mp
->m_sb_lock
);
1105 xfs_mod_sb(tp
, XFS_SB_VERSIONNUM
);
1107 spin_unlock(&mp
->m_sb_lock
);
1110 /* Caller must log the inode */
1114 * Increment the link count on an inode & log the change.
1121 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1123 ASSERT(ip
->i_d
.di_nlink
> 0);
1125 inc_nlink(VFS_I(ip
));
1126 if ((ip
->i_d
.di_version
== 1) &&
1127 (ip
->i_d
.di_nlink
> XFS_MAXLINK_1
)) {
1129 * The inode has increased its number of links beyond
1130 * what can fit in an old format inode. It now needs
1131 * to be converted to a version 2 inode with a 32 bit
1132 * link count. If this is the first inode in the file
1133 * system to do this, then we need to bump the superblock
1134 * version number as well.
1136 xfs_bump_ino_vers2(tp
, ip
);
1139 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1146 struct xfs_name
*name
,
1151 int is_dir
= S_ISDIR(mode
);
1152 struct xfs_mount
*mp
= dp
->i_mount
;
1153 struct xfs_inode
*ip
= NULL
;
1154 struct xfs_trans
*tp
= NULL
;
1156 xfs_bmap_free_t free_list
;
1157 xfs_fsblock_t first_block
;
1158 bool unlock_dp_on_error
= false;
1162 struct xfs_dquot
*udqp
= NULL
;
1163 struct xfs_dquot
*gdqp
= NULL
;
1164 struct xfs_dquot
*pdqp
= NULL
;
1165 struct xfs_trans_res tres
;
1168 trace_xfs_create(dp
, name
);
1170 if (XFS_FORCED_SHUTDOWN(mp
))
1171 return XFS_ERROR(EIO
);
1173 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1174 prid
= xfs_get_projid(dp
);
1176 prid
= XFS_PROJID_DEFAULT
;
1179 * Make sure that we have allocated dquot(s) on disk.
1181 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1182 xfs_kgid_to_gid(current_fsgid()), prid
,
1183 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1184 &udqp
, &gdqp
, &pdqp
);
1190 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1191 tres
.tr_logres
= M_RES(mp
)->tr_mkdir
.tr_logres
;
1192 tres
.tr_logcount
= XFS_MKDIR_LOG_COUNT
;
1193 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1195 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1196 tres
.tr_logres
= M_RES(mp
)->tr_create
.tr_logres
;
1197 tres
.tr_logcount
= XFS_CREATE_LOG_COUNT
;
1198 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1201 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1204 * Initially assume that the file does not exist and
1205 * reserve the resources for that case. If that is not
1206 * the case we'll drop the one we have and get a more
1207 * appropriate transaction later.
1209 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1210 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1211 if (error
== ENOSPC
) {
1212 /* flush outstanding delalloc blocks and retry */
1213 xfs_flush_inodes(mp
);
1214 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1216 if (error
== ENOSPC
) {
1217 /* No space at all so try a "no-allocation" reservation */
1219 error
= xfs_trans_reserve(tp
, &tres
, 0, 0);
1223 goto out_trans_cancel
;
1226 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1227 unlock_dp_on_error
= true;
1229 xfs_bmap_init(&free_list
, &first_block
);
1232 * Reserve disk quota and the inode.
1234 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1235 pdqp
, resblks
, 1, 0);
1237 goto out_trans_cancel
;
1239 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1241 goto out_trans_cancel
;
1244 * A newly created regular or special file just has one directory
1245 * entry pointing to them, but a directory also the "." entry
1246 * pointing to itself.
1248 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1249 prid
, resblks
> 0, &ip
, &committed
);
1251 if (error
== ENOSPC
)
1252 goto out_trans_cancel
;
1253 goto out_trans_abort
;
1257 * Now we join the directory inode to the transaction. We do not do it
1258 * earlier because xfs_dir_ialloc might commit the previous transaction
1259 * (and release all the locks). An error from here on will result in
1260 * the transaction cancel unlocking dp so don't do it explicitly in the
1263 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1264 unlock_dp_on_error
= false;
1266 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1267 &first_block
, &free_list
, resblks
?
1268 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1270 ASSERT(error
!= ENOSPC
);
1271 goto out_trans_abort
;
1273 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1274 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1277 error
= xfs_dir_init(tp
, ip
, dp
);
1279 goto out_bmap_cancel
;
1281 error
= xfs_bumplink(tp
, dp
);
1283 goto out_bmap_cancel
;
1287 * If this is a synchronous mount, make sure that the
1288 * create transaction goes to disk before returning to
1291 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1292 xfs_trans_set_sync(tp
);
1295 * Attach the dquot(s) to the inodes and modify them incore.
1296 * These ids of the inode couldn't have changed since the new
1297 * inode has been locked ever since it was created.
1299 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1301 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1303 goto out_bmap_cancel
;
1305 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1307 goto out_release_inode
;
1309 xfs_qm_dqrele(udqp
);
1310 xfs_qm_dqrele(gdqp
);
1311 xfs_qm_dqrele(pdqp
);
1317 xfs_bmap_cancel(&free_list
);
1319 cancel_flags
|= XFS_TRANS_ABORT
;
1321 xfs_trans_cancel(tp
, cancel_flags
);
1324 * Wait until after the current transaction is aborted to
1325 * release the inode. This prevents recursive transactions
1326 * and deadlocks from xfs_inactive.
1331 xfs_qm_dqrele(udqp
);
1332 xfs_qm_dqrele(gdqp
);
1333 xfs_qm_dqrele(pdqp
);
1335 if (unlock_dp_on_error
)
1336 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1344 struct xfs_name
*target_name
)
1346 xfs_mount_t
*mp
= tdp
->i_mount
;
1349 xfs_bmap_free_t free_list
;
1350 xfs_fsblock_t first_block
;
1355 trace_xfs_link(tdp
, target_name
);
1357 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1359 if (XFS_FORCED_SHUTDOWN(mp
))
1360 return XFS_ERROR(EIO
);
1362 error
= xfs_qm_dqattach(sip
, 0);
1366 error
= xfs_qm_dqattach(tdp
, 0);
1370 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1371 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1372 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1373 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, resblks
, 0);
1374 if (error
== ENOSPC
) {
1376 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, 0, 0);
1383 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1385 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1386 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1389 * If we are using project inheritance, we only allow hard link
1390 * creation in our tree when the project IDs are the same; else
1391 * the tree quota mechanism could be circumvented.
1393 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1394 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1395 error
= XFS_ERROR(EXDEV
);
1399 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
1403 xfs_bmap_init(&free_list
, &first_block
);
1405 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1406 &first_block
, &free_list
, resblks
);
1409 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1410 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1412 error
= xfs_bumplink(tp
, sip
);
1417 * If this is a synchronous mount, make sure that the
1418 * link transaction goes to disk before returning to
1421 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1422 xfs_trans_set_sync(tp
);
1425 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1427 xfs_bmap_cancel(&free_list
);
1431 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1434 cancel_flags
|= XFS_TRANS_ABORT
;
1436 xfs_trans_cancel(tp
, cancel_flags
);
1442 * Free up the underlying blocks past new_size. The new size must be smaller
1443 * than the current size. This routine can be used both for the attribute and
1444 * data fork, and does not modify the inode size, which is left to the caller.
1446 * The transaction passed to this routine must have made a permanent log
1447 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1448 * given transaction and start new ones, so make sure everything involved in
1449 * the transaction is tidy before calling here. Some transaction will be
1450 * returned to the caller to be committed. The incoming transaction must
1451 * already include the inode, and both inode locks must be held exclusively.
1452 * The inode must also be "held" within the transaction. On return the inode
1453 * will be "held" within the returned transaction. This routine does NOT
1454 * require any disk space to be reserved for it within the transaction.
1456 * If we get an error, we must return with the inode locked and linked into the
1457 * current transaction. This keeps things simple for the higher level code,
1458 * because it always knows that the inode is locked and held in the transaction
1459 * that returns to it whether errors occur or not. We don't mark the inode
1460 * dirty on error so that transactions can be easily aborted if possible.
1463 xfs_itruncate_extents(
1464 struct xfs_trans
**tpp
,
1465 struct xfs_inode
*ip
,
1467 xfs_fsize_t new_size
)
1469 struct xfs_mount
*mp
= ip
->i_mount
;
1470 struct xfs_trans
*tp
= *tpp
;
1471 struct xfs_trans
*ntp
;
1472 xfs_bmap_free_t free_list
;
1473 xfs_fsblock_t first_block
;
1474 xfs_fileoff_t first_unmap_block
;
1475 xfs_fileoff_t last_block
;
1476 xfs_filblks_t unmap_len
;
1481 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1482 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1483 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1484 ASSERT(new_size
<= XFS_ISIZE(ip
));
1485 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1486 ASSERT(ip
->i_itemp
!= NULL
);
1487 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1488 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1490 trace_xfs_itruncate_extents_start(ip
, new_size
);
1493 * Since it is possible for space to become allocated beyond
1494 * the end of the file (in a crash where the space is allocated
1495 * but the inode size is not yet updated), simply remove any
1496 * blocks which show up between the new EOF and the maximum
1497 * possible file size. If the first block to be removed is
1498 * beyond the maximum file size (ie it is the same as last_block),
1499 * then there is nothing to do.
1501 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1502 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1503 if (first_unmap_block
== last_block
)
1506 ASSERT(first_unmap_block
< last_block
);
1507 unmap_len
= last_block
- first_unmap_block
+ 1;
1509 xfs_bmap_init(&free_list
, &first_block
);
1510 error
= xfs_bunmapi(tp
, ip
,
1511 first_unmap_block
, unmap_len
,
1512 xfs_bmapi_aflag(whichfork
),
1513 XFS_ITRUNC_MAX_EXTENTS
,
1514 &first_block
, &free_list
,
1517 goto out_bmap_cancel
;
1520 * Duplicate the transaction that has the permanent
1521 * reservation and commit the old transaction.
1523 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1525 xfs_trans_ijoin(tp
, ip
, 0);
1527 goto out_bmap_cancel
;
1531 * Mark the inode dirty so it will be logged and
1532 * moved forward in the log as part of every commit.
1534 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1537 ntp
= xfs_trans_dup(tp
);
1538 error
= xfs_trans_commit(tp
, 0);
1541 xfs_trans_ijoin(tp
, ip
, 0);
1547 * Transaction commit worked ok so we can drop the extra ticket
1548 * reference that we gained in xfs_trans_dup()
1550 xfs_log_ticket_put(tp
->t_ticket
);
1551 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1557 * Always re-log the inode so that our permanent transaction can keep
1558 * on rolling it forward in the log.
1560 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1562 trace_xfs_itruncate_extents_end(ip
, new_size
);
1569 * If the bunmapi call encounters an error, return to the caller where
1570 * the transaction can be properly aborted. We just need to make sure
1571 * we're not holding any resources that we were not when we came in.
1573 xfs_bmap_cancel(&free_list
);
1581 xfs_mount_t
*mp
= ip
->i_mount
;
1584 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1587 /* If this is a read-only mount, don't do this (would generate I/O) */
1588 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1591 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1595 * If we are using filestreams, and we have an unlinked
1596 * file that we are processing the last close on, then nothing
1597 * will be able to reopen and write to this file. Purge this
1598 * inode from the filestreams cache so that it doesn't delay
1599 * teardown of the inode.
1601 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1602 xfs_filestream_deassociate(ip
);
1605 * If we previously truncated this file and removed old data
1606 * in the process, we want to initiate "early" writeout on
1607 * the last close. This is an attempt to combat the notorious
1608 * NULL files problem which is particularly noticeable from a
1609 * truncate down, buffered (re-)write (delalloc), followed by
1610 * a crash. What we are effectively doing here is
1611 * significantly reducing the time window where we'd otherwise
1612 * be exposed to that problem.
1614 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1616 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1617 if (VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0) {
1618 error
= -filemap_flush(VFS_I(ip
)->i_mapping
);
1625 if (ip
->i_d
.di_nlink
== 0)
1628 if (xfs_can_free_eofblocks(ip
, false)) {
1631 * If we can't get the iolock just skip truncating the blocks
1632 * past EOF because we could deadlock with the mmap_sem
1633 * otherwise. We'll get another chance to drop them once the
1634 * last reference to the inode is dropped, so we'll never leak
1635 * blocks permanently.
1637 * Further, check if the inode is being opened, written and
1638 * closed frequently and we have delayed allocation blocks
1639 * outstanding (e.g. streaming writes from the NFS server),
1640 * truncating the blocks past EOF will cause fragmentation to
1643 * In this case don't do the truncation, either, but we have to
1644 * be careful how we detect this case. Blocks beyond EOF show
1645 * up as i_delayed_blks even when the inode is clean, so we
1646 * need to truncate them away first before checking for a dirty
1647 * release. Hence on the first dirty close we will still remove
1648 * the speculative allocation, but after that we will leave it
1651 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1654 error
= xfs_free_eofblocks(mp
, ip
, true);
1655 if (error
&& error
!= EAGAIN
)
1658 /* delalloc blocks after truncation means it really is dirty */
1659 if (ip
->i_delayed_blks
)
1660 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1668 * This is called when the vnode reference count for the vnode
1669 * goes to zero. If the file has been unlinked, then it must
1670 * now be truncated. Also, we clear all of the read-ahead state
1671 * kept for the inode here since the file is now closed.
1677 xfs_bmap_free_t free_list
;
1678 xfs_fsblock_t first_block
;
1680 struct xfs_trans
*tp
;
1681 struct xfs_mount
*mp
;
1682 struct xfs_trans_res
*resp
;
1687 * If the inode is already free, then there can be nothing
1690 if (ip
->i_d
.di_mode
== 0 || is_bad_inode(VFS_I(ip
))) {
1691 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1692 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1693 return VN_INACTIVE_CACHE
;
1700 /* If this is a read-only mount, don't do this (would generate I/O) */
1701 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1704 if (ip
->i_d
.di_nlink
!= 0) {
1706 * force is true because we are evicting an inode from the
1707 * cache. Post-eof blocks must be freed, lest we end up with
1708 * broken free space accounting.
1710 if (xfs_can_free_eofblocks(ip
, true)) {
1711 error
= xfs_free_eofblocks(mp
, ip
, false);
1713 return VN_INACTIVE_CACHE
;
1718 if (S_ISREG(ip
->i_d
.di_mode
) &&
1719 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1720 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1723 error
= xfs_qm_dqattach(ip
, 0);
1725 return VN_INACTIVE_CACHE
;
1727 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1728 resp
= (truncate
|| S_ISLNK(ip
->i_d
.di_mode
)) ?
1729 &M_RES(mp
)->tr_itruncate
: &M_RES(mp
)->tr_ifree
;
1731 error
= xfs_trans_reserve(tp
, resp
, 0, 0);
1733 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1734 xfs_trans_cancel(tp
, 0);
1735 return VN_INACTIVE_CACHE
;
1738 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1739 xfs_trans_ijoin(tp
, ip
, 0);
1741 if (S_ISLNK(ip
->i_d
.di_mode
)) {
1742 error
= xfs_inactive_symlink(ip
, &tp
);
1745 } else if (truncate
) {
1746 ip
->i_d
.di_size
= 0;
1747 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1749 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1753 ASSERT(ip
->i_d
.di_nextents
== 0);
1757 * If there are attributes associated with the file then blow them away
1758 * now. The code calls a routine that recursively deconstructs the
1759 * attribute fork. We need to just commit the current transaction
1760 * because we can't use it for xfs_attr_inactive().
1762 if (ip
->i_d
.di_anextents
> 0) {
1763 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1765 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1769 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1771 error
= xfs_attr_inactive(ip
);
1775 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1776 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ifree
, 0, 0);
1778 xfs_trans_cancel(tp
, 0);
1782 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1783 xfs_trans_ijoin(tp
, ip
, 0);
1787 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1789 ASSERT(ip
->i_d
.di_anextents
== 0);
1794 xfs_bmap_init(&free_list
, &first_block
);
1795 error
= xfs_ifree(tp
, ip
, &free_list
);
1798 * If we fail to free the inode, shut down. The cancel
1799 * might do that, we need to make sure. Otherwise the
1800 * inode might be lost for a long time or forever.
1802 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1803 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1805 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1807 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1810 * Credit the quota account(s). The inode is gone.
1812 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1815 * Just ignore errors at this point. There is nothing we can
1816 * do except to try to keep going. Make sure it's not a silent
1819 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1821 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1823 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1825 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1830 * Release the dquots held by inode, if any.
1832 xfs_qm_dqdetach(ip
);
1834 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1836 return VN_INACTIVE_CACHE
;
1838 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1843 * This is called when the inode's link count goes to 0.
1844 * We place the on-disk inode on a list in the AGI. It
1845 * will be pulled from this list when the inode is freed.
1862 ASSERT(ip
->i_d
.di_nlink
== 0);
1863 ASSERT(ip
->i_d
.di_mode
!= 0);
1868 * Get the agi buffer first. It ensures lock ordering
1871 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1874 agi
= XFS_BUF_TO_AGI(agibp
);
1877 * Get the index into the agi hash table for the
1878 * list this inode will go on.
1880 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1882 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1883 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1884 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1886 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1888 * There is already another inode in the bucket we need
1889 * to add ourselves to. Add us at the front of the list.
1890 * Here we put the head pointer into our next pointer,
1891 * and then we fall through to point the head at us.
1893 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1898 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1899 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1900 offset
= ip
->i_imap
.im_boffset
+
1901 offsetof(xfs_dinode_t
, di_next_unlinked
);
1903 /* need to recalc the inode CRC if appropriate */
1904 xfs_dinode_calc_crc(mp
, dip
);
1906 xfs_trans_inode_buf(tp
, ibp
);
1907 xfs_trans_log_buf(tp
, ibp
, offset
,
1908 (offset
+ sizeof(xfs_agino_t
) - 1));
1909 xfs_inobp_check(mp
, ibp
);
1913 * Point the bucket head pointer at the inode being inserted.
1916 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1917 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1918 (sizeof(xfs_agino_t
) * bucket_index
);
1919 xfs_trans_log_buf(tp
, agibp
, offset
,
1920 (offset
+ sizeof(xfs_agino_t
) - 1));
1925 * Pull the on-disk inode from the AGI unlinked list.
1938 xfs_agnumber_t agno
;
1940 xfs_agino_t next_agino
;
1941 xfs_buf_t
*last_ibp
;
1942 xfs_dinode_t
*last_dip
= NULL
;
1944 int offset
, last_offset
= 0;
1948 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1951 * Get the agi buffer first. It ensures lock ordering
1954 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1958 agi
= XFS_BUF_TO_AGI(agibp
);
1961 * Get the index into the agi hash table for the
1962 * list this inode will go on.
1964 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1966 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1967 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
1968 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1970 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1972 * We're at the head of the list. Get the inode's on-disk
1973 * buffer to see if there is anyone after us on the list.
1974 * Only modify our next pointer if it is not already NULLAGINO.
1975 * This saves us the overhead of dealing with the buffer when
1976 * there is no need to change it.
1978 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1981 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
1985 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1986 ASSERT(next_agino
!= 0);
1987 if (next_agino
!= NULLAGINO
) {
1988 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1989 offset
= ip
->i_imap
.im_boffset
+
1990 offsetof(xfs_dinode_t
, di_next_unlinked
);
1992 /* need to recalc the inode CRC if appropriate */
1993 xfs_dinode_calc_crc(mp
, dip
);
1995 xfs_trans_inode_buf(tp
, ibp
);
1996 xfs_trans_log_buf(tp
, ibp
, offset
,
1997 (offset
+ sizeof(xfs_agino_t
) - 1));
1998 xfs_inobp_check(mp
, ibp
);
2000 xfs_trans_brelse(tp
, ibp
);
2003 * Point the bucket head pointer at the next inode.
2005 ASSERT(next_agino
!= 0);
2006 ASSERT(next_agino
!= agino
);
2007 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2008 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2009 (sizeof(xfs_agino_t
) * bucket_index
);
2010 xfs_trans_log_buf(tp
, agibp
, offset
,
2011 (offset
+ sizeof(xfs_agino_t
) - 1));
2014 * We need to search the list for the inode being freed.
2016 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2018 while (next_agino
!= agino
) {
2019 struct xfs_imap imap
;
2022 xfs_trans_brelse(tp
, last_ibp
);
2025 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2027 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2030 "%s: xfs_imap returned error %d.",
2035 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2039 "%s: xfs_imap_to_bp returned error %d.",
2044 last_offset
= imap
.im_boffset
;
2045 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2046 ASSERT(next_agino
!= NULLAGINO
);
2047 ASSERT(next_agino
!= 0);
2051 * Now last_ibp points to the buffer previous to us on the
2052 * unlinked list. Pull us from the list.
2054 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2057 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2061 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2062 ASSERT(next_agino
!= 0);
2063 ASSERT(next_agino
!= agino
);
2064 if (next_agino
!= NULLAGINO
) {
2065 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2066 offset
= ip
->i_imap
.im_boffset
+
2067 offsetof(xfs_dinode_t
, di_next_unlinked
);
2069 /* need to recalc the inode CRC if appropriate */
2070 xfs_dinode_calc_crc(mp
, dip
);
2072 xfs_trans_inode_buf(tp
, ibp
);
2073 xfs_trans_log_buf(tp
, ibp
, offset
,
2074 (offset
+ sizeof(xfs_agino_t
) - 1));
2075 xfs_inobp_check(mp
, ibp
);
2077 xfs_trans_brelse(tp
, ibp
);
2080 * Point the previous inode on the list to the next inode.
2082 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2083 ASSERT(next_agino
!= 0);
2084 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2086 /* need to recalc the inode CRC if appropriate */
2087 xfs_dinode_calc_crc(mp
, last_dip
);
2089 xfs_trans_inode_buf(tp
, last_ibp
);
2090 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2091 (offset
+ sizeof(xfs_agino_t
) - 1));
2092 xfs_inobp_check(mp
, last_ibp
);
2098 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2099 * inodes that are in memory - they all must be marked stale and attached to
2100 * the cluster buffer.
2104 xfs_inode_t
*free_ip
,
2108 xfs_mount_t
*mp
= free_ip
->i_mount
;
2109 int blks_per_cluster
;
2116 xfs_inode_log_item_t
*iip
;
2117 xfs_log_item_t
*lip
;
2118 struct xfs_perag
*pag
;
2120 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2121 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2122 blks_per_cluster
= 1;
2123 ninodes
= mp
->m_sb
.sb_inopblock
;
2124 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2126 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2127 mp
->m_sb
.sb_blocksize
;
2128 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2129 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2132 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2133 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2134 XFS_INO_TO_AGBNO(mp
, inum
));
2137 * We obtain and lock the backing buffer first in the process
2138 * here, as we have to ensure that any dirty inode that we
2139 * can't get the flush lock on is attached to the buffer.
2140 * If we scan the in-memory inodes first, then buffer IO can
2141 * complete before we get a lock on it, and hence we may fail
2142 * to mark all the active inodes on the buffer stale.
2144 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2145 mp
->m_bsize
* blks_per_cluster
,
2152 * This buffer may not have been correctly initialised as we
2153 * didn't read it from disk. That's not important because we are
2154 * only using to mark the buffer as stale in the log, and to
2155 * attach stale cached inodes on it. That means it will never be
2156 * dispatched for IO. If it is, we want to know about it, and we
2157 * want it to fail. We can acheive this by adding a write
2158 * verifier to the buffer.
2160 bp
->b_ops
= &xfs_inode_buf_ops
;
2163 * Walk the inodes already attached to the buffer and mark them
2164 * stale. These will all have the flush locks held, so an
2165 * in-memory inode walk can't lock them. By marking them all
2166 * stale first, we will not attempt to lock them in the loop
2167 * below as the XFS_ISTALE flag will be set.
2171 if (lip
->li_type
== XFS_LI_INODE
) {
2172 iip
= (xfs_inode_log_item_t
*)lip
;
2173 ASSERT(iip
->ili_logged
== 1);
2174 lip
->li_cb
= xfs_istale_done
;
2175 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2176 &iip
->ili_flush_lsn
,
2177 &iip
->ili_item
.li_lsn
);
2178 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2180 lip
= lip
->li_bio_list
;
2185 * For each inode in memory attempt to add it to the inode
2186 * buffer and set it up for being staled on buffer IO
2187 * completion. This is safe as we've locked out tail pushing
2188 * and flushing by locking the buffer.
2190 * We have already marked every inode that was part of a
2191 * transaction stale above, which means there is no point in
2192 * even trying to lock them.
2194 for (i
= 0; i
< ninodes
; i
++) {
2197 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2198 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2200 /* Inode not in memory, nothing to do */
2207 * because this is an RCU protected lookup, we could
2208 * find a recently freed or even reallocated inode
2209 * during the lookup. We need to check under the
2210 * i_flags_lock for a valid inode here. Skip it if it
2211 * is not valid, the wrong inode or stale.
2213 spin_lock(&ip
->i_flags_lock
);
2214 if (ip
->i_ino
!= inum
+ i
||
2215 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2216 spin_unlock(&ip
->i_flags_lock
);
2220 spin_unlock(&ip
->i_flags_lock
);
2223 * Don't try to lock/unlock the current inode, but we
2224 * _cannot_ skip the other inodes that we did not find
2225 * in the list attached to the buffer and are not
2226 * already marked stale. If we can't lock it, back off
2229 if (ip
!= free_ip
&&
2230 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2238 xfs_iflags_set(ip
, XFS_ISTALE
);
2241 * we don't need to attach clean inodes or those only
2242 * with unlogged changes (which we throw away, anyway).
2245 if (!iip
|| xfs_inode_clean(ip
)) {
2246 ASSERT(ip
!= free_ip
);
2248 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2252 iip
->ili_last_fields
= iip
->ili_fields
;
2253 iip
->ili_fields
= 0;
2254 iip
->ili_logged
= 1;
2255 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2256 &iip
->ili_item
.li_lsn
);
2258 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2262 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2265 xfs_trans_stale_inode_buf(tp
, bp
);
2266 xfs_trans_binval(tp
, bp
);
2274 * This is called to return an inode to the inode free list.
2275 * The inode should already be truncated to 0 length and have
2276 * no pages associated with it. This routine also assumes that
2277 * the inode is already a part of the transaction.
2279 * The on-disk copy of the inode will have been added to the list
2280 * of unlinked inodes in the AGI. We need to remove the inode from
2281 * that list atomically with respect to freeing it here.
2287 xfs_bmap_free_t
*flist
)
2291 xfs_ino_t first_ino
;
2293 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2294 ASSERT(ip
->i_d
.di_nlink
== 0);
2295 ASSERT(ip
->i_d
.di_nextents
== 0);
2296 ASSERT(ip
->i_d
.di_anextents
== 0);
2297 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
2298 ASSERT(ip
->i_d
.di_nblocks
== 0);
2301 * Pull the on-disk inode from the AGI unlinked list.
2303 error
= xfs_iunlink_remove(tp
, ip
);
2307 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2311 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2312 ip
->i_d
.di_flags
= 0;
2313 ip
->i_d
.di_dmevmask
= 0;
2314 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2315 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2316 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2318 * Bump the generation count so no one will be confused
2319 * by reincarnations of this inode.
2322 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2325 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
2331 * This is called to unpin an inode. The caller must have the inode locked
2332 * in at least shared mode so that the buffer cannot be subsequently pinned
2333 * once someone is waiting for it to be unpinned.
2337 struct xfs_inode
*ip
)
2339 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2341 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2343 /* Give the log a push to start the unpinning I/O */
2344 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2350 struct xfs_inode
*ip
)
2352 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2353 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2358 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2359 if (xfs_ipincount(ip
))
2361 } while (xfs_ipincount(ip
));
2362 finish_wait(wq
, &wait
.wait
);
2367 struct xfs_inode
*ip
)
2369 if (xfs_ipincount(ip
))
2370 __xfs_iunpin_wait(ip
);
2374 * Removing an inode from the namespace involves removing the directory entry
2375 * and dropping the link count on the inode. Removing the directory entry can
2376 * result in locking an AGF (directory blocks were freed) and removing a link
2377 * count can result in placing the inode on an unlinked list which results in
2380 * The big problem here is that we have an ordering constraint on AGF and AGI
2381 * locking - inode allocation locks the AGI, then can allocate a new extent for
2382 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2383 * removes the inode from the unlinked list, requiring that we lock the AGI
2384 * first, and then freeing the inode can result in an inode chunk being freed
2385 * and hence freeing disk space requiring that we lock an AGF.
2387 * Hence the ordering that is imposed by other parts of the code is AGI before
2388 * AGF. This means we cannot remove the directory entry before we drop the inode
2389 * reference count and put it on the unlinked list as this results in a lock
2390 * order of AGF then AGI, and this can deadlock against inode allocation and
2391 * freeing. Therefore we must drop the link counts before we remove the
2394 * This is still safe from a transactional point of view - it is not until we
2395 * get to xfs_bmap_finish() that we have the possibility of multiple
2396 * transactions in this operation. Hence as long as we remove the directory
2397 * entry and drop the link count in the first transaction of the remove
2398 * operation, there are no transactional constraints on the ordering here.
2403 struct xfs_name
*name
,
2406 xfs_mount_t
*mp
= dp
->i_mount
;
2407 xfs_trans_t
*tp
= NULL
;
2408 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
2410 xfs_bmap_free_t free_list
;
2411 xfs_fsblock_t first_block
;
2418 trace_xfs_remove(dp
, name
);
2420 if (XFS_FORCED_SHUTDOWN(mp
))
2421 return XFS_ERROR(EIO
);
2423 error
= xfs_qm_dqattach(dp
, 0);
2427 error
= xfs_qm_dqattach(ip
, 0);
2432 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2433 log_count
= XFS_DEFAULT_LOG_COUNT
;
2435 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2436 log_count
= XFS_REMOVE_LOG_COUNT
;
2438 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2441 * We try to get the real space reservation first,
2442 * allowing for directory btree deletion(s) implying
2443 * possible bmap insert(s). If we can't get the space
2444 * reservation then we use 0 instead, and avoid the bmap
2445 * btree insert(s) in the directory code by, if the bmap
2446 * insert tries to happen, instead trimming the LAST
2447 * block from the directory.
2449 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2450 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, resblks
, 0);
2451 if (error
== ENOSPC
) {
2453 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
2456 ASSERT(error
!= ENOSPC
);
2458 goto out_trans_cancel
;
2461 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2463 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2464 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2467 * If we're removing a directory perform some additional validation.
2469 cancel_flags
|= XFS_TRANS_ABORT
;
2471 ASSERT(ip
->i_d
.di_nlink
>= 2);
2472 if (ip
->i_d
.di_nlink
!= 2) {
2473 error
= XFS_ERROR(ENOTEMPTY
);
2474 goto out_trans_cancel
;
2476 if (!xfs_dir_isempty(ip
)) {
2477 error
= XFS_ERROR(ENOTEMPTY
);
2478 goto out_trans_cancel
;
2481 /* Drop the link from ip's "..". */
2482 error
= xfs_droplink(tp
, dp
);
2484 goto out_trans_cancel
;
2486 /* Drop the "." link from ip to self. */
2487 error
= xfs_droplink(tp
, ip
);
2489 goto out_trans_cancel
;
2492 * When removing a non-directory we need to log the parent
2493 * inode here. For a directory this is done implicitly
2494 * by the xfs_droplink call for the ".." entry.
2496 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2498 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2500 /* Drop the link from dp to ip. */
2501 error
= xfs_droplink(tp
, ip
);
2503 goto out_trans_cancel
;
2505 /* Determine if this is the last link while the inode is locked */
2506 link_zero
= (ip
->i_d
.di_nlink
== 0);
2508 xfs_bmap_init(&free_list
, &first_block
);
2509 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2510 &first_block
, &free_list
, resblks
);
2512 ASSERT(error
!= ENOENT
);
2513 goto out_bmap_cancel
;
2517 * If this is a synchronous mount, make sure that the
2518 * remove transaction goes to disk before returning to
2521 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2522 xfs_trans_set_sync(tp
);
2524 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2526 goto out_bmap_cancel
;
2528 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2533 * If we are using filestreams, kill the stream association.
2534 * If the file is still open it may get a new one but that
2535 * will get killed on last close in xfs_close() so we don't
2536 * have to worry about that.
2538 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
2539 xfs_filestream_deassociate(ip
);
2544 xfs_bmap_cancel(&free_list
);
2546 xfs_trans_cancel(tp
, cancel_flags
);
2552 * Enter all inodes for a rename transaction into a sorted array.
2555 xfs_sort_for_rename(
2556 xfs_inode_t
*dp1
, /* in: old (source) directory inode */
2557 xfs_inode_t
*dp2
, /* in: new (target) directory inode */
2558 xfs_inode_t
*ip1
, /* in: inode of old entry */
2559 xfs_inode_t
*ip2
, /* in: inode of new entry, if it
2560 already exists, NULL otherwise. */
2561 xfs_inode_t
**i_tab
,/* out: array of inode returned, sorted */
2562 int *num_inodes
) /* out: number of inodes in array */
2568 * i_tab contains a list of pointers to inodes. We initialize
2569 * the table here & we'll sort it. We will then use it to
2570 * order the acquisition of the inode locks.
2572 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2586 * Sort the elements via bubble sort. (Remember, there are at
2587 * most 4 elements to sort, so this is adequate.)
2589 for (i
= 0; i
< *num_inodes
; i
++) {
2590 for (j
= 1; j
< *num_inodes
; j
++) {
2591 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2593 i_tab
[j
] = i_tab
[j
-1];
2605 xfs_inode_t
*src_dp
,
2606 struct xfs_name
*src_name
,
2607 xfs_inode_t
*src_ip
,
2608 xfs_inode_t
*target_dp
,
2609 struct xfs_name
*target_name
,
2610 xfs_inode_t
*target_ip
)
2612 xfs_trans_t
*tp
= NULL
;
2613 xfs_mount_t
*mp
= src_dp
->i_mount
;
2614 int new_parent
; /* moving to a new dir */
2615 int src_is_directory
; /* src_name is a directory */
2617 xfs_bmap_free_t free_list
;
2618 xfs_fsblock_t first_block
;
2621 xfs_inode_t
*inodes
[4];
2625 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2627 new_parent
= (src_dp
!= target_dp
);
2628 src_is_directory
= S_ISDIR(src_ip
->i_d
.di_mode
);
2630 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
,
2631 inodes
, &num_inodes
);
2633 xfs_bmap_init(&free_list
, &first_block
);
2634 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RENAME
);
2635 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2636 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2637 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, spaceres
, 0);
2638 if (error
== ENOSPC
) {
2640 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, 0, 0);
2643 xfs_trans_cancel(tp
, 0);
2648 * Attach the dquots to the inodes
2650 error
= xfs_qm_vop_rename_dqattach(inodes
);
2652 xfs_trans_cancel(tp
, cancel_flags
);
2657 * Lock all the participating inodes. Depending upon whether
2658 * the target_name exists in the target directory, and
2659 * whether the target directory is the same as the source
2660 * directory, we can lock from 2 to 4 inodes.
2662 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2665 * Join all the inodes to the transaction. From this point on,
2666 * we can rely on either trans_commit or trans_cancel to unlock
2669 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2671 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2672 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2674 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2677 * If we are using project inheritance, we only allow renames
2678 * into our tree when the project IDs are the same; else the
2679 * tree quota mechanism would be circumvented.
2681 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2682 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2683 error
= XFS_ERROR(EXDEV
);
2688 * Set up the target.
2690 if (target_ip
== NULL
) {
2692 * If there's no space reservation, check the entry will
2693 * fit before actually inserting it.
2695 error
= xfs_dir_canenter(tp
, target_dp
, target_name
, spaceres
);
2699 * If target does not exist and the rename crosses
2700 * directories, adjust the target directory link count
2701 * to account for the ".." reference from the new entry.
2703 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
2704 src_ip
->i_ino
, &first_block
,
2705 &free_list
, spaceres
);
2706 if (error
== ENOSPC
)
2711 xfs_trans_ichgtime(tp
, target_dp
,
2712 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2714 if (new_parent
&& src_is_directory
) {
2715 error
= xfs_bumplink(tp
, target_dp
);
2719 } else { /* target_ip != NULL */
2721 * If target exists and it's a directory, check that both
2722 * target and source are directories and that target can be
2723 * destroyed, or that neither is a directory.
2725 if (S_ISDIR(target_ip
->i_d
.di_mode
)) {
2727 * Make sure target dir is empty.
2729 if (!(xfs_dir_isempty(target_ip
)) ||
2730 (target_ip
->i_d
.di_nlink
> 2)) {
2731 error
= XFS_ERROR(EEXIST
);
2737 * Link the source inode under the target name.
2738 * If the source inode is a directory and we are moving
2739 * it across directories, its ".." entry will be
2740 * inconsistent until we replace that down below.
2742 * In case there is already an entry with the same
2743 * name at the destination directory, remove it first.
2745 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
2747 &first_block
, &free_list
, spaceres
);
2751 xfs_trans_ichgtime(tp
, target_dp
,
2752 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2755 * Decrement the link count on the target since the target
2756 * dir no longer points to it.
2758 error
= xfs_droplink(tp
, target_ip
);
2762 if (src_is_directory
) {
2764 * Drop the link from the old "." entry.
2766 error
= xfs_droplink(tp
, target_ip
);
2770 } /* target_ip != NULL */
2773 * Remove the source.
2775 if (new_parent
&& src_is_directory
) {
2777 * Rewrite the ".." entry to point to the new
2780 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
2782 &first_block
, &free_list
, spaceres
);
2783 ASSERT(error
!= EEXIST
);
2789 * We always want to hit the ctime on the source inode.
2791 * This isn't strictly required by the standards since the source
2792 * inode isn't really being changed, but old unix file systems did
2793 * it and some incremental backup programs won't work without it.
2795 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
2796 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
2799 * Adjust the link count on src_dp. This is necessary when
2800 * renaming a directory, either within one parent when
2801 * the target existed, or across two parent directories.
2803 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
2806 * Decrement link count on src_directory since the
2807 * entry that's moved no longer points to it.
2809 error
= xfs_droplink(tp
, src_dp
);
2814 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
2815 &first_block
, &free_list
, spaceres
);
2819 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2820 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
2822 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
2825 * If this is a synchronous mount, make sure that the
2826 * rename transaction goes to disk before returning to
2829 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2830 xfs_trans_set_sync(tp
);
2833 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2835 xfs_bmap_cancel(&free_list
);
2836 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
2842 * trans_commit will unlock src_ip, target_ip & decrement
2843 * the vnode references.
2845 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2848 cancel_flags
|= XFS_TRANS_ABORT
;
2850 xfs_bmap_cancel(&free_list
);
2851 xfs_trans_cancel(tp
, cancel_flags
);
2861 xfs_mount_t
*mp
= ip
->i_mount
;
2862 struct xfs_perag
*pag
;
2863 unsigned long first_index
, mask
;
2864 unsigned long inodes_per_cluster
;
2866 xfs_inode_t
**ilist
;
2873 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2875 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2876 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2877 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2881 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2882 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2884 /* really need a gang lookup range call here */
2885 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2886 first_index
, inodes_per_cluster
);
2890 for (i
= 0; i
< nr_found
; i
++) {
2896 * because this is an RCU protected lookup, we could find a
2897 * recently freed or even reallocated inode during the lookup.
2898 * We need to check under the i_flags_lock for a valid inode
2899 * here. Skip it if it is not valid or the wrong inode.
2901 spin_lock(&ip
->i_flags_lock
);
2903 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2904 spin_unlock(&ip
->i_flags_lock
);
2907 spin_unlock(&ip
->i_flags_lock
);
2910 * Do an un-protected check to see if the inode is dirty and
2911 * is a candidate for flushing. These checks will be repeated
2912 * later after the appropriate locks are acquired.
2914 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2918 * Try to get locks. If any are unavailable or it is pinned,
2919 * then this inode cannot be flushed and is skipped.
2922 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2924 if (!xfs_iflock_nowait(iq
)) {
2925 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2928 if (xfs_ipincount(iq
)) {
2930 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2935 * arriving here means that this inode can be flushed. First
2936 * re-check that it's dirty before flushing.
2938 if (!xfs_inode_clean(iq
)) {
2940 error
= xfs_iflush_int(iq
, bp
);
2942 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2943 goto cluster_corrupt_out
;
2949 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2953 XFS_STATS_INC(xs_icluster_flushcnt
);
2954 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2965 cluster_corrupt_out
:
2967 * Corruption detected in the clustering loop. Invalidate the
2968 * inode buffer and shut down the filesystem.
2972 * Clean up the buffer. If it was delwri, just release it --
2973 * brelse can handle it with no problems. If not, shut down the
2974 * filesystem before releasing the buffer.
2976 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
2980 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2982 if (!bufwasdelwri
) {
2984 * Just like incore_relse: if we have b_iodone functions,
2985 * mark the buffer as an error and call them. Otherwise
2986 * mark it as stale and brelse.
2991 xfs_buf_ioerror(bp
, EIO
);
2992 xfs_buf_ioend(bp
, 0);
3000 * Unlocks the flush lock
3002 xfs_iflush_abort(iq
, false);
3005 return XFS_ERROR(EFSCORRUPTED
);
3009 * Flush dirty inode metadata into the backing buffer.
3011 * The caller must have the inode lock and the inode flush lock held. The
3012 * inode lock will still be held upon return to the caller, and the inode
3013 * flush lock will be released after the inode has reached the disk.
3015 * The caller must write out the buffer returned in *bpp and release it.
3019 struct xfs_inode
*ip
,
3020 struct xfs_buf
**bpp
)
3022 struct xfs_mount
*mp
= ip
->i_mount
;
3024 struct xfs_dinode
*dip
;
3027 XFS_STATS_INC(xs_iflush_count
);
3029 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3030 ASSERT(xfs_isiflocked(ip
));
3031 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3032 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3036 xfs_iunpin_wait(ip
);
3039 * For stale inodes we cannot rely on the backing buffer remaining
3040 * stale in cache for the remaining life of the stale inode and so
3041 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3042 * inodes below. We have to check this after ensuring the inode is
3043 * unpinned so that it is safe to reclaim the stale inode after the
3046 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3052 * This may have been unpinned because the filesystem is shutting
3053 * down forcibly. If that's the case we must not write this inode
3054 * to disk, because the log record didn't make it to disk.
3056 * We also have to remove the log item from the AIL in this case,
3057 * as we wait for an empty AIL as part of the unmount process.
3059 if (XFS_FORCED_SHUTDOWN(mp
)) {
3060 error
= XFS_ERROR(EIO
);
3065 * Get the buffer containing the on-disk inode.
3067 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3075 * First flush out the inode that xfs_iflush was called with.
3077 error
= xfs_iflush_int(ip
, bp
);
3082 * If the buffer is pinned then push on the log now so we won't
3083 * get stuck waiting in the write for too long.
3085 if (xfs_buf_ispinned(bp
))
3086 xfs_log_force(mp
, 0);
3090 * see if other inodes can be gathered into this write
3092 error
= xfs_iflush_cluster(ip
, bp
);
3094 goto cluster_corrupt_out
;
3101 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3102 cluster_corrupt_out
:
3103 error
= XFS_ERROR(EFSCORRUPTED
);
3106 * Unlocks the flush lock
3108 xfs_iflush_abort(ip
, false);
3114 struct xfs_inode
*ip
,
3117 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3118 struct xfs_dinode
*dip
;
3119 struct xfs_mount
*mp
= ip
->i_mount
;
3121 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3122 ASSERT(xfs_isiflocked(ip
));
3123 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3124 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3125 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3127 /* set *dip = inode's place in the buffer */
3128 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3130 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3131 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3132 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3133 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3134 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3137 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3138 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3139 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3140 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3141 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3144 if (S_ISREG(ip
->i_d
.di_mode
)) {
3146 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3147 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3148 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3149 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3150 "%s: Bad regular inode %Lu, ptr 0x%p",
3151 __func__
, ip
->i_ino
, ip
);
3154 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
3156 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3157 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3158 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3159 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3160 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3161 "%s: Bad directory inode %Lu, ptr 0x%p",
3162 __func__
, ip
->i_ino
, ip
);
3166 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3167 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3168 XFS_RANDOM_IFLUSH_5
)) {
3169 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3170 "%s: detected corrupt incore inode %Lu, "
3171 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3172 __func__
, ip
->i_ino
,
3173 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3174 ip
->i_d
.di_nblocks
, ip
);
3177 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3178 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3179 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3180 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3181 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3186 * Inode item log recovery for v1/v2 inodes are dependent on the
3187 * di_flushiter count for correct sequencing. We bump the flush
3188 * iteration count so we can detect flushes which postdate a log record
3189 * during recovery. This is redundant as we now log every change and
3190 * hence this can't happen but we need to still do it to ensure
3191 * backwards compatibility with old kernels that predate logging all
3194 if (ip
->i_d
.di_version
< 3)
3195 ip
->i_d
.di_flushiter
++;
3198 * Copy the dirty parts of the inode into the on-disk
3199 * inode. We always copy out the core of the inode,
3200 * because if the inode is dirty at all the core must
3203 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3205 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3206 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3207 ip
->i_d
.di_flushiter
= 0;
3210 * If this is really an old format inode and the superblock version
3211 * has not been updated to support only new format inodes, then
3212 * convert back to the old inode format. If the superblock version
3213 * has been updated, then make the conversion permanent.
3215 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
3216 if (ip
->i_d
.di_version
== 1) {
3217 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3221 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3222 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3225 * The superblock version has already been bumped,
3226 * so just make the conversion to the new inode
3229 ip
->i_d
.di_version
= 2;
3230 dip
->di_version
= 2;
3231 ip
->i_d
.di_onlink
= 0;
3233 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3234 memset(&(dip
->di_pad
[0]), 0,
3235 sizeof(dip
->di_pad
));
3236 ASSERT(xfs_get_projid(ip
) == 0);
3240 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3241 if (XFS_IFORK_Q(ip
))
3242 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3243 xfs_inobp_check(mp
, bp
);
3246 * We've recorded everything logged in the inode, so we'd like to clear
3247 * the ili_fields bits so we don't log and flush things unnecessarily.
3248 * However, we can't stop logging all this information until the data
3249 * we've copied into the disk buffer is written to disk. If we did we
3250 * might overwrite the copy of the inode in the log with all the data
3251 * after re-logging only part of it, and in the face of a crash we
3252 * wouldn't have all the data we need to recover.
3254 * What we do is move the bits to the ili_last_fields field. When
3255 * logging the inode, these bits are moved back to the ili_fields field.
3256 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3257 * know that the information those bits represent is permanently on
3258 * disk. As long as the flush completes before the inode is logged
3259 * again, then both ili_fields and ili_last_fields will be cleared.
3261 * We can play with the ili_fields bits here, because the inode lock
3262 * must be held exclusively in order to set bits there and the flush
3263 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3264 * done routine can tell whether or not to look in the AIL. Also, store
3265 * the current LSN of the inode so that we can tell whether the item has
3266 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3267 * need the AIL lock, because it is a 64 bit value that cannot be read
3270 iip
->ili_last_fields
= iip
->ili_fields
;
3271 iip
->ili_fields
= 0;
3272 iip
->ili_logged
= 1;
3274 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3275 &iip
->ili_item
.li_lsn
);
3278 * Attach the function xfs_iflush_done to the inode's
3279 * buffer. This will remove the inode from the AIL
3280 * and unlock the inode's flush lock when the inode is
3281 * completely written to disk.
3283 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3285 /* update the lsn in the on disk inode if required */
3286 if (ip
->i_d
.di_version
== 3)
3287 dip
->di_lsn
= cpu_to_be64(iip
->ili_item
.li_lsn
);
3289 /* generate the checksum. */
3290 xfs_dinode_calc_crc(mp
, dip
);
3292 ASSERT(bp
->b_fspriv
!= NULL
);
3293 ASSERT(bp
->b_iodone
!= NULL
);
3297 return XFS_ERROR(EFSCORRUPTED
);