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_filestream.h"
46 #include "xfs_cksum.h"
47 #include "xfs_trace.h"
48 #include "xfs_icache.h"
49 #include "xfs_symlink.h"
50 #include "xfs_trans_priv.h"
52 #include "xfs_bmap_btree.h"
54 kmem_zone_t
*xfs_inode_zone
;
57 * Used in xfs_itruncate_extents(). This is the maximum number of extents
58 * freed from a file in a single transaction.
60 #define XFS_ITRUNC_MAX_EXTENTS 2
62 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
64 STATIC
int xfs_iunlink_remove(xfs_trans_t
*, xfs_inode_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 * These two are wrapper routines around the xfs_ilock() routine used to
82 * centralize some grungy code. They are used in places that wish to lock the
83 * inode solely for reading the extents. The reason these places can't just
84 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
85 * bringing in of the extents from disk for a file in b-tree format. If the
86 * inode is in b-tree format, then we need to lock the inode exclusively until
87 * the extents are read in. Locking it exclusively all the time would limit
88 * our parallelism unnecessarily, though. What we do instead is check to see
89 * if the extents have been read in yet, and only lock the inode exclusively
92 * The functions return a value which should be given to the corresponding
96 xfs_ilock_data_map_shared(
99 uint lock_mode
= XFS_ILOCK_SHARED
;
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
;
104 xfs_ilock(ip
, lock_mode
);
109 xfs_ilock_attr_map_shared(
110 struct xfs_inode
*ip
)
112 uint lock_mode
= XFS_ILOCK_SHARED
;
114 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_BTREE
&&
115 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
116 lock_mode
= XFS_ILOCK_EXCL
;
117 xfs_ilock(ip
, lock_mode
);
122 * The xfs inode contains 2 locks: a multi-reader lock called the
123 * i_iolock and a multi-reader lock called the i_lock. This routine
124 * allows either or both of the locks to be obtained.
126 * The 2 locks should always be ordered so that the IO lock is
127 * obtained first in order to prevent deadlock.
129 * ip -- the inode being locked
130 * lock_flags -- this parameter indicates the inode's locks
131 * to be locked. It can be:
136 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
137 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
138 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
139 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
146 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
149 * You can't set both SHARED and EXCL for the same lock,
150 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
151 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
153 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
154 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
155 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
156 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
157 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
159 if (lock_flags
& XFS_IOLOCK_EXCL
)
160 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
161 else if (lock_flags
& XFS_IOLOCK_SHARED
)
162 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
164 if (lock_flags
& XFS_ILOCK_EXCL
)
165 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
166 else if (lock_flags
& XFS_ILOCK_SHARED
)
167 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
171 * This is just like xfs_ilock(), except that the caller
172 * is guaranteed not to sleep. It returns 1 if it gets
173 * the requested locks and 0 otherwise. If the IO lock is
174 * obtained but the inode lock cannot be, then the IO lock
175 * is dropped before returning.
177 * ip -- the inode being locked
178 * lock_flags -- this parameter indicates the inode's locks to be
179 * to be locked. See the comment for xfs_ilock() for a list
187 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
190 * You can't set both SHARED and EXCL for the same lock,
191 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
192 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
194 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
195 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
196 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
197 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
198 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
200 if (lock_flags
& XFS_IOLOCK_EXCL
) {
201 if (!mrtryupdate(&ip
->i_iolock
))
203 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
204 if (!mrtryaccess(&ip
->i_iolock
))
207 if (lock_flags
& XFS_ILOCK_EXCL
) {
208 if (!mrtryupdate(&ip
->i_lock
))
209 goto out_undo_iolock
;
210 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
211 if (!mrtryaccess(&ip
->i_lock
))
212 goto out_undo_iolock
;
217 if (lock_flags
& XFS_IOLOCK_EXCL
)
218 mrunlock_excl(&ip
->i_iolock
);
219 else if (lock_flags
& XFS_IOLOCK_SHARED
)
220 mrunlock_shared(&ip
->i_iolock
);
226 * xfs_iunlock() is used to drop the inode locks acquired with
227 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
228 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
229 * that we know which locks to drop.
231 * ip -- the inode being unlocked
232 * lock_flags -- this parameter indicates the inode's locks to be
233 * to be unlocked. See the comment for xfs_ilock() for a list
234 * of valid values for this parameter.
243 * You can't set both SHARED and EXCL for the same lock,
244 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
245 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
247 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
248 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
249 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
250 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
251 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
252 ASSERT(lock_flags
!= 0);
254 if (lock_flags
& XFS_IOLOCK_EXCL
)
255 mrunlock_excl(&ip
->i_iolock
);
256 else if (lock_flags
& XFS_IOLOCK_SHARED
)
257 mrunlock_shared(&ip
->i_iolock
);
259 if (lock_flags
& XFS_ILOCK_EXCL
)
260 mrunlock_excl(&ip
->i_lock
);
261 else if (lock_flags
& XFS_ILOCK_SHARED
)
262 mrunlock_shared(&ip
->i_lock
);
264 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
268 * give up write locks. the i/o lock cannot be held nested
269 * if it is being demoted.
276 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
277 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
279 if (lock_flags
& XFS_ILOCK_EXCL
)
280 mrdemote(&ip
->i_lock
);
281 if (lock_flags
& XFS_IOLOCK_EXCL
)
282 mrdemote(&ip
->i_iolock
);
284 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
287 #if defined(DEBUG) || defined(XFS_WARN)
293 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
294 if (!(lock_flags
& XFS_ILOCK_SHARED
))
295 return !!ip
->i_lock
.mr_writer
;
296 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
299 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
300 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
301 return !!ip
->i_iolock
.mr_writer
;
302 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);
312 int xfs_small_retries
;
313 int xfs_middle_retries
;
314 int xfs_lots_retries
;
319 * Bump the subclass so xfs_lock_inodes() acquires each lock with
323 xfs_lock_inumorder(int lock_mode
, int subclass
)
325 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
326 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
327 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
328 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
334 * The following routine will lock n inodes in exclusive mode.
335 * We assume the caller calls us with the inodes in i_ino order.
337 * We need to detect deadlock where an inode that we lock
338 * is in the AIL and we start waiting for another inode that is locked
339 * by a thread in a long running transaction (such as truncate). This can
340 * result in deadlock since the long running trans might need to wait
341 * for the inode we just locked in order to push the tail and free space
350 int attempts
= 0, i
, j
, try_lock
;
353 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
359 for (; i
< inodes
; i
++) {
362 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
366 * If try_lock is not set yet, make sure all locked inodes
367 * are not in the AIL.
368 * If any are, set try_lock to be used later.
372 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
373 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
374 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
381 * If any of the previous locks we have locked is in the AIL,
382 * we must TRY to get the second and subsequent locks. If
383 * we can't get any, we must release all we have
388 /* try_lock must be 0 if i is 0. */
390 * try_lock means we have an inode locked
391 * that is in the AIL.
394 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
398 * Unlock all previous guys and try again.
399 * xfs_iunlock will try to push the tail
400 * if the inode is in the AIL.
403 for(j
= i
- 1; j
>= 0; j
--) {
406 * Check to see if we've already
408 * Not the first one going back,
409 * and the inode ptr is the same.
411 if ((j
!= (i
- 1)) && ips
[j
] ==
415 xfs_iunlock(ips
[j
], lock_mode
);
418 if ((attempts
% 5) == 0) {
419 delay(1); /* Don't just spin the CPU */
429 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
435 if (attempts
< 5) xfs_small_retries
++;
436 else if (attempts
< 100) xfs_middle_retries
++;
437 else xfs_lots_retries
++;
445 * xfs_lock_two_inodes() can only be used to lock one type of lock
446 * at a time - the iolock or the ilock, but not both at once. If
447 * we lock both at once, lockdep will report false positives saying
448 * we have violated locking orders.
460 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
461 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
462 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
464 if (ip0
->i_ino
> ip1
->i_ino
) {
471 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
474 * If the first lock we have locked is in the AIL, we must TRY to get
475 * the second lock. If we can't get it, we must release the first one
478 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
479 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
480 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
481 xfs_iunlock(ip0
, lock_mode
);
482 if ((++attempts
% 5) == 0)
483 delay(1); /* Don't just spin the CPU */
487 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
494 struct xfs_inode
*ip
)
496 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
497 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
500 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
501 if (xfs_isiflocked(ip
))
503 } while (!xfs_iflock_nowait(ip
));
505 finish_wait(wq
, &wait
.wait
);
514 if (di_flags
& XFS_DIFLAG_ANY
) {
515 if (di_flags
& XFS_DIFLAG_REALTIME
)
516 flags
|= XFS_XFLAG_REALTIME
;
517 if (di_flags
& XFS_DIFLAG_PREALLOC
)
518 flags
|= XFS_XFLAG_PREALLOC
;
519 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
520 flags
|= XFS_XFLAG_IMMUTABLE
;
521 if (di_flags
& XFS_DIFLAG_APPEND
)
522 flags
|= XFS_XFLAG_APPEND
;
523 if (di_flags
& XFS_DIFLAG_SYNC
)
524 flags
|= XFS_XFLAG_SYNC
;
525 if (di_flags
& XFS_DIFLAG_NOATIME
)
526 flags
|= XFS_XFLAG_NOATIME
;
527 if (di_flags
& XFS_DIFLAG_NODUMP
)
528 flags
|= XFS_XFLAG_NODUMP
;
529 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
530 flags
|= XFS_XFLAG_RTINHERIT
;
531 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
532 flags
|= XFS_XFLAG_PROJINHERIT
;
533 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
534 flags
|= XFS_XFLAG_NOSYMLINKS
;
535 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
536 flags
|= XFS_XFLAG_EXTSIZE
;
537 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
538 flags
|= XFS_XFLAG_EXTSZINHERIT
;
539 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
540 flags
|= XFS_XFLAG_NODEFRAG
;
541 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
542 flags
|= XFS_XFLAG_FILESTREAM
;
552 xfs_icdinode_t
*dic
= &ip
->i_d
;
554 return _xfs_dic2xflags(dic
->di_flags
) |
555 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
562 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
563 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
567 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
568 * is allowed, otherwise it has to be an exact match. If a CI match is found,
569 * ci_name->name will point to a the actual name (caller must free) or
570 * will be set to NULL if an exact match is found.
575 struct xfs_name
*name
,
577 struct xfs_name
*ci_name
)
583 trace_xfs_lookup(dp
, name
);
585 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
588 lock_mode
= xfs_ilock_data_map_shared(dp
);
589 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
590 xfs_iunlock(dp
, lock_mode
);
595 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
603 kmem_free(ci_name
->name
);
610 * Allocate an inode on disk and return a copy of its in-core version.
611 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
612 * appropriately within the inode. The uid and gid for the inode are
613 * set according to the contents of the given cred structure.
615 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
616 * has a free inode available, call xfs_iget() to obtain the in-core
617 * version of the allocated inode. Finally, fill in the inode and
618 * log its initial contents. In this case, ialloc_context would be
621 * If xfs_dialloc() does not have an available inode, it will replenish
622 * its supply by doing an allocation. Since we can only do one
623 * allocation within a transaction without deadlocks, we must commit
624 * the current transaction before returning the inode itself.
625 * In this case, therefore, we will set ialloc_context and return.
626 * The caller should then commit the current transaction, start a new
627 * transaction, and call xfs_ialloc() again to actually get the inode.
629 * To ensure that some other process does not grab the inode that
630 * was allocated during the first call to xfs_ialloc(), this routine
631 * also returns the [locked] bp pointing to the head of the freelist
632 * as ialloc_context. The caller should hold this buffer across
633 * the commit and pass it back into this routine on the second call.
635 * If we are allocating quota inodes, we do not have a parent inode
636 * to attach to or associate with (i.e. pip == NULL) because they
637 * are not linked into the directory structure - they are attached
638 * directly to the superblock - and so have no parent.
649 xfs_buf_t
**ialloc_context
,
652 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
);
685 * We always convert v1 inodes to v2 now - we only support filesystems
686 * with >= v2 inode capability, so there is no reason for ever leaving
687 * an inode in v1 format.
689 if (ip
->i_d
.di_version
== 1)
690 ip
->i_d
.di_version
= 2;
692 ip
->i_d
.di_mode
= mode
;
693 ip
->i_d
.di_onlink
= 0;
694 ip
->i_d
.di_nlink
= nlink
;
695 ASSERT(ip
->i_d
.di_nlink
== nlink
);
696 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
697 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
698 xfs_set_projid(ip
, prid
);
699 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
701 if (pip
&& XFS_INHERIT_GID(pip
)) {
702 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
703 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
704 ip
->i_d
.di_mode
|= S_ISGID
;
709 * If the group ID of the new file does not match the effective group
710 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
711 * (and only if the irix_sgid_inherit compatibility variable is set).
713 if ((irix_sgid_inherit
) &&
714 (ip
->i_d
.di_mode
& S_ISGID
) &&
715 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
)))) {
716 ip
->i_d
.di_mode
&= ~S_ISGID
;
720 ip
->i_d
.di_nextents
= 0;
721 ASSERT(ip
->i_d
.di_nblocks
== 0);
723 tv
= current_fs_time(mp
->m_super
);
724 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
725 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
726 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
727 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
730 * di_gen will have been taken care of in xfs_iread.
732 ip
->i_d
.di_extsize
= 0;
733 ip
->i_d
.di_dmevmask
= 0;
734 ip
->i_d
.di_dmstate
= 0;
735 ip
->i_d
.di_flags
= 0;
737 if (ip
->i_d
.di_version
== 3) {
738 ASSERT(ip
->i_d
.di_ino
== ino
);
739 ASSERT(uuid_equal(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
));
741 ip
->i_d
.di_changecount
= 1;
743 ip
->i_d
.di_flags2
= 0;
744 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
745 ip
->i_d
.di_crtime
= ip
->i_d
.di_mtime
;
749 flags
= XFS_ILOG_CORE
;
750 switch (mode
& S_IFMT
) {
755 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
756 ip
->i_df
.if_u2
.if_rdev
= rdev
;
757 ip
->i_df
.if_flags
= 0;
758 flags
|= XFS_ILOG_DEV
;
762 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
766 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
767 di_flags
|= XFS_DIFLAG_RTINHERIT
;
768 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
769 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
770 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
772 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
773 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
774 } else if (S_ISREG(mode
)) {
775 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
776 di_flags
|= XFS_DIFLAG_REALTIME
;
777 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
778 di_flags
|= XFS_DIFLAG_EXTSIZE
;
779 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
782 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
784 di_flags
|= XFS_DIFLAG_NOATIME
;
785 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
787 di_flags
|= XFS_DIFLAG_NODUMP
;
788 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
790 di_flags
|= XFS_DIFLAG_SYNC
;
791 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
792 xfs_inherit_nosymlinks
)
793 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
794 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
795 xfs_inherit_nodefrag
)
796 di_flags
|= XFS_DIFLAG_NODEFRAG
;
797 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
798 di_flags
|= XFS_DIFLAG_FILESTREAM
;
799 ip
->i_d
.di_flags
|= di_flags
;
803 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
804 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
805 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
806 ip
->i_df
.if_u1
.if_extents
= NULL
;
812 * Attribute fork settings for new inode.
814 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
815 ip
->i_d
.di_anextents
= 0;
818 * Log the new values stuffed into the inode.
820 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
821 xfs_trans_log_inode(tp
, ip
, flags
);
823 /* now that we have an i_mode we can setup inode ops and unlock */
831 * Allocates a new inode from disk and return a pointer to the
832 * incore copy. This routine will internally commit the current
833 * transaction and allocate a new one if the Space Manager needed
834 * to do an allocation to replenish the inode free-list.
836 * This routine is designed to be called from xfs_create and
842 xfs_trans_t
**tpp
, /* input: current transaction;
843 output: may be a new transaction. */
844 xfs_inode_t
*dp
, /* directory within whose allocate
849 prid_t prid
, /* project id */
850 int okalloc
, /* ok to allocate new space */
851 xfs_inode_t
**ipp
, /* pointer to inode; it will be
859 xfs_buf_t
*ialloc_context
= NULL
;
865 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
868 * xfs_ialloc will return a pointer to an incore inode if
869 * the Space Manager has an available inode on the free
870 * list. Otherwise, it will do an allocation and replenish
871 * the freelist. Since we can only do one allocation per
872 * transaction without deadlocks, we will need to commit the
873 * current transaction and start a new one. We will then
874 * need to call xfs_ialloc again to get the inode.
876 * If xfs_ialloc did an allocation to replenish the freelist,
877 * it returns the bp containing the head of the freelist as
878 * ialloc_context. We will hold a lock on it across the
879 * transaction commit so that no other process can steal
880 * the inode(s) that we've just allocated.
882 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
883 &ialloc_context
, &ip
);
886 * Return an error if we were unable to allocate a new inode.
887 * This should only happen if we run out of space on disk or
888 * encounter a disk error.
894 if (!ialloc_context
&& !ip
) {
900 * If the AGI buffer is non-NULL, then we were unable to get an
901 * inode in one operation. We need to commit the current
902 * transaction and call xfs_ialloc() again. It is guaranteed
903 * to succeed the second time.
905 if (ialloc_context
) {
906 struct xfs_trans_res tres
;
909 * Normally, xfs_trans_commit releases all the locks.
910 * We call bhold to hang on to the ialloc_context across
911 * the commit. Holding this buffer prevents any other
912 * processes from doing any allocations in this
915 xfs_trans_bhold(tp
, ialloc_context
);
917 * Save the log reservation so we can use
918 * them in the next transaction.
920 tres
.tr_logres
= xfs_trans_get_log_res(tp
);
921 tres
.tr_logcount
= xfs_trans_get_log_count(tp
);
924 * We want the quota changes to be associated with the next
925 * transaction, NOT this one. So, detach the dqinfo from this
926 * and attach it to the next transaction.
931 dqinfo
= (void *)tp
->t_dqinfo
;
933 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
934 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
937 ntp
= xfs_trans_dup(tp
);
938 code
= xfs_trans_commit(tp
, 0);
940 if (committed
!= NULL
) {
944 * If we get an error during the commit processing,
945 * release the buffer that is still held and return
949 xfs_buf_relse(ialloc_context
);
951 tp
->t_dqinfo
= dqinfo
;
952 xfs_trans_free_dqinfo(tp
);
960 * transaction commit worked ok so we can drop the extra ticket
961 * reference that we gained in xfs_trans_dup()
963 xfs_log_ticket_put(tp
->t_ticket
);
964 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
965 code
= xfs_trans_reserve(tp
, &tres
, 0, 0);
968 * Re-attach the quota info that we detached from prev trx.
971 tp
->t_dqinfo
= dqinfo
;
972 tp
->t_flags
|= tflags
;
976 xfs_buf_relse(ialloc_context
);
981 xfs_trans_bjoin(tp
, ialloc_context
);
984 * Call ialloc again. Since we've locked out all
985 * other allocations in this allocation group,
986 * this call should always succeed.
988 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
989 okalloc
, &ialloc_context
, &ip
);
992 * If we get an error at this point, return to the caller
993 * so that the current transaction can be aborted.
1000 ASSERT(!ialloc_context
&& ip
);
1003 if (committed
!= NULL
)
1014 * Decrement the link count on an inode & log the change.
1015 * If this causes the link count to go to zero, initiate the
1016 * logging activity required to truncate a file.
1025 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1027 ASSERT (ip
->i_d
.di_nlink
> 0);
1029 drop_nlink(VFS_I(ip
));
1030 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1033 if (ip
->i_d
.di_nlink
== 0) {
1035 * We're dropping the last link to this file.
1036 * Move the on-disk inode to the AGI unlinked list.
1037 * From xfs_inactive() we will pull the inode from
1038 * the list and free it.
1040 error
= xfs_iunlink(tp
, ip
);
1046 * Increment the link count on an inode & log the change.
1053 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1055 ASSERT(ip
->i_d
.di_version
> 1);
1056 ASSERT(ip
->i_d
.di_nlink
> 0 || (VFS_I(ip
)->i_state
& I_LINKABLE
));
1058 inc_nlink(VFS_I(ip
));
1059 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1066 struct xfs_name
*name
,
1071 int is_dir
= S_ISDIR(mode
);
1072 struct xfs_mount
*mp
= dp
->i_mount
;
1073 struct xfs_inode
*ip
= NULL
;
1074 struct xfs_trans
*tp
= NULL
;
1076 xfs_bmap_free_t free_list
;
1077 xfs_fsblock_t first_block
;
1078 bool unlock_dp_on_error
= false;
1082 struct xfs_dquot
*udqp
= NULL
;
1083 struct xfs_dquot
*gdqp
= NULL
;
1084 struct xfs_dquot
*pdqp
= NULL
;
1085 struct xfs_trans_res
*tres
;
1088 trace_xfs_create(dp
, name
);
1090 if (XFS_FORCED_SHUTDOWN(mp
))
1093 prid
= xfs_get_initial_prid(dp
);
1096 * Make sure that we have allocated dquot(s) on disk.
1098 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1099 xfs_kgid_to_gid(current_fsgid()), prid
,
1100 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1101 &udqp
, &gdqp
, &pdqp
);
1107 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1108 tres
= &M_RES(mp
)->tr_mkdir
;
1109 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1111 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1112 tres
= &M_RES(mp
)->tr_create
;
1113 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1116 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1119 * Initially assume that the file does not exist and
1120 * reserve the resources for that case. If that is not
1121 * the case we'll drop the one we have and get a more
1122 * appropriate transaction later.
1124 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1125 if (error
== -ENOSPC
) {
1126 /* flush outstanding delalloc blocks and retry */
1127 xfs_flush_inodes(mp
);
1128 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1130 if (error
== -ENOSPC
) {
1131 /* No space at all so try a "no-allocation" reservation */
1133 error
= xfs_trans_reserve(tp
, tres
, 0, 0);
1137 goto out_trans_cancel
;
1140 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1141 unlock_dp_on_error
= true;
1143 xfs_bmap_init(&free_list
, &first_block
);
1146 * Reserve disk quota and the inode.
1148 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1149 pdqp
, resblks
, 1, 0);
1151 goto out_trans_cancel
;
1154 error
= xfs_dir_canenter(tp
, dp
, name
);
1156 goto out_trans_cancel
;
1160 * A newly created regular or special file just has one directory
1161 * entry pointing to them, but a directory also the "." entry
1162 * pointing to itself.
1164 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1165 prid
, resblks
> 0, &ip
, &committed
);
1167 if (error
== -ENOSPC
)
1168 goto out_trans_cancel
;
1169 goto out_trans_abort
;
1173 * Now we join the directory inode to the transaction. We do not do it
1174 * earlier because xfs_dir_ialloc might commit the previous transaction
1175 * (and release all the locks). An error from here on will result in
1176 * the transaction cancel unlocking dp so don't do it explicitly in the
1179 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1180 unlock_dp_on_error
= false;
1182 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1183 &first_block
, &free_list
, resblks
?
1184 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1186 ASSERT(error
!= -ENOSPC
);
1187 goto out_trans_abort
;
1189 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1190 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1193 error
= xfs_dir_init(tp
, ip
, dp
);
1195 goto out_bmap_cancel
;
1197 error
= xfs_bumplink(tp
, dp
);
1199 goto out_bmap_cancel
;
1203 * If this is a synchronous mount, make sure that the
1204 * create transaction goes to disk before returning to
1207 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1208 xfs_trans_set_sync(tp
);
1211 * Attach the dquot(s) to the inodes and modify them incore.
1212 * These ids of the inode couldn't have changed since the new
1213 * inode has been locked ever since it was created.
1215 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1217 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1219 goto out_bmap_cancel
;
1221 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1223 goto out_release_inode
;
1225 xfs_qm_dqrele(udqp
);
1226 xfs_qm_dqrele(gdqp
);
1227 xfs_qm_dqrele(pdqp
);
1233 xfs_bmap_cancel(&free_list
);
1235 cancel_flags
|= XFS_TRANS_ABORT
;
1237 xfs_trans_cancel(tp
, cancel_flags
);
1240 * Wait until after the current transaction is aborted to
1241 * release the inode. This prevents recursive transactions
1242 * and deadlocks from xfs_inactive.
1247 xfs_qm_dqrele(udqp
);
1248 xfs_qm_dqrele(gdqp
);
1249 xfs_qm_dqrele(pdqp
);
1251 if (unlock_dp_on_error
)
1252 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1258 struct xfs_inode
*dp
,
1259 struct dentry
*dentry
,
1261 struct xfs_inode
**ipp
)
1263 struct xfs_mount
*mp
= dp
->i_mount
;
1264 struct xfs_inode
*ip
= NULL
;
1265 struct xfs_trans
*tp
= NULL
;
1267 uint cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1269 struct xfs_dquot
*udqp
= NULL
;
1270 struct xfs_dquot
*gdqp
= NULL
;
1271 struct xfs_dquot
*pdqp
= NULL
;
1272 struct xfs_trans_res
*tres
;
1275 if (XFS_FORCED_SHUTDOWN(mp
))
1278 prid
= xfs_get_initial_prid(dp
);
1281 * Make sure that we have allocated dquot(s) on disk.
1283 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1284 xfs_kgid_to_gid(current_fsgid()), prid
,
1285 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1286 &udqp
, &gdqp
, &pdqp
);
1290 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1291 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE_TMPFILE
);
1293 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1294 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1295 if (error
== -ENOSPC
) {
1296 /* No space at all so try a "no-allocation" reservation */
1298 error
= xfs_trans_reserve(tp
, tres
, 0, 0);
1302 goto out_trans_cancel
;
1305 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1306 pdqp
, resblks
, 1, 0);
1308 goto out_trans_cancel
;
1310 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1, 0,
1311 prid
, resblks
> 0, &ip
, NULL
);
1313 if (error
== -ENOSPC
)
1314 goto out_trans_cancel
;
1315 goto out_trans_abort
;
1318 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1319 xfs_trans_set_sync(tp
);
1322 * Attach the dquot(s) to the inodes and modify them incore.
1323 * These ids of the inode couldn't have changed since the new
1324 * inode has been locked ever since it was created.
1326 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1329 error
= xfs_iunlink(tp
, ip
);
1331 goto out_trans_abort
;
1333 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1335 goto out_release_inode
;
1337 xfs_qm_dqrele(udqp
);
1338 xfs_qm_dqrele(gdqp
);
1339 xfs_qm_dqrele(pdqp
);
1345 cancel_flags
|= XFS_TRANS_ABORT
;
1347 xfs_trans_cancel(tp
, cancel_flags
);
1350 * Wait until after the current transaction is aborted to
1351 * release the inode. This prevents recursive transactions
1352 * and deadlocks from xfs_inactive.
1357 xfs_qm_dqrele(udqp
);
1358 xfs_qm_dqrele(gdqp
);
1359 xfs_qm_dqrele(pdqp
);
1368 struct xfs_name
*target_name
)
1370 xfs_mount_t
*mp
= tdp
->i_mount
;
1373 xfs_bmap_free_t free_list
;
1374 xfs_fsblock_t first_block
;
1379 trace_xfs_link(tdp
, target_name
);
1381 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1383 if (XFS_FORCED_SHUTDOWN(mp
))
1386 error
= xfs_qm_dqattach(sip
, 0);
1390 error
= xfs_qm_dqattach(tdp
, 0);
1394 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1395 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1396 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1397 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, resblks
, 0);
1398 if (error
== -ENOSPC
) {
1400 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, 0, 0);
1407 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1409 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1410 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1413 * If we are using project inheritance, we only allow hard link
1414 * creation in our tree when the project IDs are the same; else
1415 * the tree quota mechanism could be circumvented.
1417 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1418 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1424 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1429 xfs_bmap_init(&free_list
, &first_block
);
1431 if (sip
->i_d
.di_nlink
== 0) {
1432 error
= xfs_iunlink_remove(tp
, sip
);
1437 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1438 &first_block
, &free_list
, resblks
);
1441 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1442 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1444 error
= xfs_bumplink(tp
, sip
);
1449 * If this is a synchronous mount, make sure that the
1450 * link transaction goes to disk before returning to
1453 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1454 xfs_trans_set_sync(tp
);
1457 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1459 xfs_bmap_cancel(&free_list
);
1463 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1466 cancel_flags
|= XFS_TRANS_ABORT
;
1468 xfs_trans_cancel(tp
, cancel_flags
);
1474 * Free up the underlying blocks past new_size. The new size must be smaller
1475 * than the current size. This routine can be used both for the attribute and
1476 * data fork, and does not modify the inode size, which is left to the caller.
1478 * The transaction passed to this routine must have made a permanent log
1479 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1480 * given transaction and start new ones, so make sure everything involved in
1481 * the transaction is tidy before calling here. Some transaction will be
1482 * returned to the caller to be committed. The incoming transaction must
1483 * already include the inode, and both inode locks must be held exclusively.
1484 * The inode must also be "held" within the transaction. On return the inode
1485 * will be "held" within the returned transaction. This routine does NOT
1486 * require any disk space to be reserved for it within the transaction.
1488 * If we get an error, we must return with the inode locked and linked into the
1489 * current transaction. This keeps things simple for the higher level code,
1490 * because it always knows that the inode is locked and held in the transaction
1491 * that returns to it whether errors occur or not. We don't mark the inode
1492 * dirty on error so that transactions can be easily aborted if possible.
1495 xfs_itruncate_extents(
1496 struct xfs_trans
**tpp
,
1497 struct xfs_inode
*ip
,
1499 xfs_fsize_t new_size
)
1501 struct xfs_mount
*mp
= ip
->i_mount
;
1502 struct xfs_trans
*tp
= *tpp
;
1503 struct xfs_trans
*ntp
;
1504 xfs_bmap_free_t free_list
;
1505 xfs_fsblock_t first_block
;
1506 xfs_fileoff_t first_unmap_block
;
1507 xfs_fileoff_t last_block
;
1508 xfs_filblks_t unmap_len
;
1513 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1514 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1515 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1516 ASSERT(new_size
<= XFS_ISIZE(ip
));
1517 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1518 ASSERT(ip
->i_itemp
!= NULL
);
1519 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1520 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1522 trace_xfs_itruncate_extents_start(ip
, new_size
);
1525 * Since it is possible for space to become allocated beyond
1526 * the end of the file (in a crash where the space is allocated
1527 * but the inode size is not yet updated), simply remove any
1528 * blocks which show up between the new EOF and the maximum
1529 * possible file size. If the first block to be removed is
1530 * beyond the maximum file size (ie it is the same as last_block),
1531 * then there is nothing to do.
1533 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1534 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1535 if (first_unmap_block
== last_block
)
1538 ASSERT(first_unmap_block
< last_block
);
1539 unmap_len
= last_block
- first_unmap_block
+ 1;
1541 xfs_bmap_init(&free_list
, &first_block
);
1542 error
= xfs_bunmapi(tp
, ip
,
1543 first_unmap_block
, unmap_len
,
1544 xfs_bmapi_aflag(whichfork
),
1545 XFS_ITRUNC_MAX_EXTENTS
,
1546 &first_block
, &free_list
,
1549 goto out_bmap_cancel
;
1552 * Duplicate the transaction that has the permanent
1553 * reservation and commit the old transaction.
1555 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1557 xfs_trans_ijoin(tp
, ip
, 0);
1559 goto out_bmap_cancel
;
1563 * Mark the inode dirty so it will be logged and
1564 * moved forward in the log as part of every commit.
1566 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1569 ntp
= xfs_trans_dup(tp
);
1570 error
= xfs_trans_commit(tp
, 0);
1573 xfs_trans_ijoin(tp
, ip
, 0);
1579 * Transaction commit worked ok so we can drop the extra ticket
1580 * reference that we gained in xfs_trans_dup()
1582 xfs_log_ticket_put(tp
->t_ticket
);
1583 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1589 * Always re-log the inode so that our permanent transaction can keep
1590 * on rolling it forward in the log.
1592 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1594 trace_xfs_itruncate_extents_end(ip
, new_size
);
1601 * If the bunmapi call encounters an error, return to the caller where
1602 * the transaction can be properly aborted. We just need to make sure
1603 * we're not holding any resources that we were not when we came in.
1605 xfs_bmap_cancel(&free_list
);
1613 xfs_mount_t
*mp
= ip
->i_mount
;
1616 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1619 /* If this is a read-only mount, don't do this (would generate I/O) */
1620 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1623 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1627 * If we previously truncated this file and removed old data
1628 * in the process, we want to initiate "early" writeout on
1629 * the last close. This is an attempt to combat the notorious
1630 * NULL files problem which is particularly noticeable from a
1631 * truncate down, buffered (re-)write (delalloc), followed by
1632 * a crash. What we are effectively doing here is
1633 * significantly reducing the time window where we'd otherwise
1634 * be exposed to that problem.
1636 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1638 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1639 if (ip
->i_delayed_blks
> 0) {
1640 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1647 if (ip
->i_d
.di_nlink
== 0)
1650 if (xfs_can_free_eofblocks(ip
, false)) {
1653 * If we can't get the iolock just skip truncating the blocks
1654 * past EOF because we could deadlock with the mmap_sem
1655 * otherwise. We'll get another chance to drop them once the
1656 * last reference to the inode is dropped, so we'll never leak
1657 * blocks permanently.
1659 * Further, check if the inode is being opened, written and
1660 * closed frequently and we have delayed allocation blocks
1661 * outstanding (e.g. streaming writes from the NFS server),
1662 * truncating the blocks past EOF will cause fragmentation to
1665 * In this case don't do the truncation, either, but we have to
1666 * be careful how we detect this case. Blocks beyond EOF show
1667 * up as i_delayed_blks even when the inode is clean, so we
1668 * need to truncate them away first before checking for a dirty
1669 * release. Hence on the first dirty close we will still remove
1670 * the speculative allocation, but after that we will leave it
1673 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1676 error
= xfs_free_eofblocks(mp
, ip
, true);
1677 if (error
&& error
!= -EAGAIN
)
1680 /* delalloc blocks after truncation means it really is dirty */
1681 if (ip
->i_delayed_blks
)
1682 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1688 * xfs_inactive_truncate
1690 * Called to perform a truncate when an inode becomes unlinked.
1693 xfs_inactive_truncate(
1694 struct xfs_inode
*ip
)
1696 struct xfs_mount
*mp
= ip
->i_mount
;
1697 struct xfs_trans
*tp
;
1700 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1701 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1703 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1704 xfs_trans_cancel(tp
, 0);
1708 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1709 xfs_trans_ijoin(tp
, ip
, 0);
1712 * Log the inode size first to prevent stale data exposure in the event
1713 * of a system crash before the truncate completes. See the related
1714 * comment in xfs_setattr_size() for details.
1716 ip
->i_d
.di_size
= 0;
1717 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1719 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1721 goto error_trans_cancel
;
1723 ASSERT(ip
->i_d
.di_nextents
== 0);
1725 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1729 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1733 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1735 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1740 * xfs_inactive_ifree()
1742 * Perform the inode free when an inode is unlinked.
1746 struct xfs_inode
*ip
)
1748 xfs_bmap_free_t free_list
;
1749 xfs_fsblock_t first_block
;
1751 struct xfs_mount
*mp
= ip
->i_mount
;
1752 struct xfs_trans
*tp
;
1755 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1758 * The ifree transaction might need to allocate blocks for record
1759 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1760 * allow ifree to dip into the reserved block pool if necessary.
1762 * Freeing large sets of inodes generally means freeing inode chunks,
1763 * directory and file data blocks, so this should be relatively safe.
1764 * Only under severe circumstances should it be possible to free enough
1765 * inodes to exhaust the reserve block pool via finobt expansion while
1766 * at the same time not creating free space in the filesystem.
1768 * Send a warning if the reservation does happen to fail, as the inode
1769 * now remains allocated and sits on the unlinked list until the fs is
1772 tp
->t_flags
|= XFS_TRANS_RESERVE
;
1773 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ifree
,
1774 XFS_IFREE_SPACE_RES(mp
), 0);
1776 if (error
== -ENOSPC
) {
1777 xfs_warn_ratelimited(mp
,
1778 "Failed to remove inode(s) from unlinked list. "
1779 "Please free space, unmount and run xfs_repair.");
1781 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1783 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1787 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1788 xfs_trans_ijoin(tp
, ip
, 0);
1790 xfs_bmap_init(&free_list
, &first_block
);
1791 error
= xfs_ifree(tp
, ip
, &free_list
);
1794 * If we fail to free the inode, shut down. The cancel
1795 * might do that, we need to make sure. Otherwise the
1796 * inode might be lost for a long time or forever.
1798 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1799 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1801 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1803 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1804 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1809 * Credit the quota account(s). The inode is gone.
1811 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1814 * Just ignore errors at this point. There is nothing we can
1815 * do except to try to keep going. Make sure it's not a silent
1818 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1820 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1822 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1824 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1827 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1834 * This is called when the vnode reference count for the vnode
1835 * goes to zero. If the file has been unlinked, then it must
1836 * now be truncated. Also, we clear all of the read-ahead state
1837 * kept for the inode here since the file is now closed.
1843 struct xfs_mount
*mp
;
1848 * If the inode is already free, then there can be nothing
1851 if (ip
->i_d
.di_mode
== 0) {
1852 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1853 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1859 /* If this is a read-only mount, don't do this (would generate I/O) */
1860 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1863 if (ip
->i_d
.di_nlink
!= 0) {
1865 * force is true because we are evicting an inode from the
1866 * cache. Post-eof blocks must be freed, lest we end up with
1867 * broken free space accounting.
1869 if (xfs_can_free_eofblocks(ip
, true))
1870 xfs_free_eofblocks(mp
, ip
, false);
1875 if (S_ISREG(ip
->i_d
.di_mode
) &&
1876 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1877 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1880 error
= xfs_qm_dqattach(ip
, 0);
1884 if (S_ISLNK(ip
->i_d
.di_mode
))
1885 error
= xfs_inactive_symlink(ip
);
1887 error
= xfs_inactive_truncate(ip
);
1892 * If there are attributes associated with the file then blow them away
1893 * now. The code calls a routine that recursively deconstructs the
1894 * attribute fork. We need to just commit the current transaction
1895 * because we can't use it for xfs_attr_inactive().
1897 if (ip
->i_d
.di_anextents
> 0) {
1898 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1900 error
= xfs_attr_inactive(ip
);
1906 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1908 ASSERT(ip
->i_d
.di_anextents
== 0);
1913 error
= xfs_inactive_ifree(ip
);
1918 * Release the dquots held by inode, if any.
1920 xfs_qm_dqdetach(ip
);
1924 * This is called when the inode's link count goes to 0.
1925 * We place the on-disk inode on a list in the AGI. It
1926 * will be pulled from this list when the inode is freed.
1943 ASSERT(ip
->i_d
.di_nlink
== 0);
1944 ASSERT(ip
->i_d
.di_mode
!= 0);
1949 * Get the agi buffer first. It ensures lock ordering
1952 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1955 agi
= XFS_BUF_TO_AGI(agibp
);
1958 * Get the index into the agi hash table for the
1959 * list this inode will go on.
1961 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1963 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1964 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1965 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1967 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1969 * There is already another inode in the bucket we need
1970 * to add ourselves to. Add us at the front of the list.
1971 * Here we put the head pointer into our next pointer,
1972 * and then we fall through to point the head at us.
1974 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1979 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1980 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1981 offset
= ip
->i_imap
.im_boffset
+
1982 offsetof(xfs_dinode_t
, di_next_unlinked
);
1984 /* need to recalc the inode CRC if appropriate */
1985 xfs_dinode_calc_crc(mp
, dip
);
1987 xfs_trans_inode_buf(tp
, ibp
);
1988 xfs_trans_log_buf(tp
, ibp
, offset
,
1989 (offset
+ sizeof(xfs_agino_t
) - 1));
1990 xfs_inobp_check(mp
, ibp
);
1994 * Point the bucket head pointer at the inode being inserted.
1997 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1998 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1999 (sizeof(xfs_agino_t
) * bucket_index
);
2000 xfs_trans_log_buf(tp
, agibp
, offset
,
2001 (offset
+ sizeof(xfs_agino_t
) - 1));
2006 * Pull the on-disk inode from the AGI unlinked list.
2019 xfs_agnumber_t agno
;
2021 xfs_agino_t next_agino
;
2022 xfs_buf_t
*last_ibp
;
2023 xfs_dinode_t
*last_dip
= NULL
;
2025 int offset
, last_offset
= 0;
2029 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2032 * Get the agi buffer first. It ensures lock ordering
2035 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2039 agi
= XFS_BUF_TO_AGI(agibp
);
2042 * Get the index into the agi hash table for the
2043 * list this inode will go on.
2045 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2047 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2048 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
2049 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2051 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2053 * We're at the head of the list. Get the inode's on-disk
2054 * buffer to see if there is anyone after us on the list.
2055 * Only modify our next pointer if it is not already NULLAGINO.
2056 * This saves us the overhead of dealing with the buffer when
2057 * there is no need to change it.
2059 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2062 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2066 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2067 ASSERT(next_agino
!= 0);
2068 if (next_agino
!= NULLAGINO
) {
2069 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2070 offset
= ip
->i_imap
.im_boffset
+
2071 offsetof(xfs_dinode_t
, di_next_unlinked
);
2073 /* need to recalc the inode CRC if appropriate */
2074 xfs_dinode_calc_crc(mp
, dip
);
2076 xfs_trans_inode_buf(tp
, ibp
);
2077 xfs_trans_log_buf(tp
, ibp
, offset
,
2078 (offset
+ sizeof(xfs_agino_t
) - 1));
2079 xfs_inobp_check(mp
, ibp
);
2081 xfs_trans_brelse(tp
, ibp
);
2084 * Point the bucket head pointer at the next inode.
2086 ASSERT(next_agino
!= 0);
2087 ASSERT(next_agino
!= agino
);
2088 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2089 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2090 (sizeof(xfs_agino_t
) * bucket_index
);
2091 xfs_trans_log_buf(tp
, agibp
, offset
,
2092 (offset
+ sizeof(xfs_agino_t
) - 1));
2095 * We need to search the list for the inode being freed.
2097 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2099 while (next_agino
!= agino
) {
2100 struct xfs_imap imap
;
2103 xfs_trans_brelse(tp
, last_ibp
);
2106 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2108 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2111 "%s: xfs_imap returned error %d.",
2116 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2120 "%s: xfs_imap_to_bp returned error %d.",
2125 last_offset
= imap
.im_boffset
;
2126 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2127 ASSERT(next_agino
!= NULLAGINO
);
2128 ASSERT(next_agino
!= 0);
2132 * Now last_ibp points to the buffer previous to us on the
2133 * unlinked list. Pull us from the list.
2135 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2138 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2142 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2143 ASSERT(next_agino
!= 0);
2144 ASSERT(next_agino
!= agino
);
2145 if (next_agino
!= NULLAGINO
) {
2146 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2147 offset
= ip
->i_imap
.im_boffset
+
2148 offsetof(xfs_dinode_t
, di_next_unlinked
);
2150 /* need to recalc the inode CRC if appropriate */
2151 xfs_dinode_calc_crc(mp
, dip
);
2153 xfs_trans_inode_buf(tp
, ibp
);
2154 xfs_trans_log_buf(tp
, ibp
, offset
,
2155 (offset
+ sizeof(xfs_agino_t
) - 1));
2156 xfs_inobp_check(mp
, ibp
);
2158 xfs_trans_brelse(tp
, ibp
);
2161 * Point the previous inode on the list to the next inode.
2163 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2164 ASSERT(next_agino
!= 0);
2165 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2167 /* need to recalc the inode CRC if appropriate */
2168 xfs_dinode_calc_crc(mp
, last_dip
);
2170 xfs_trans_inode_buf(tp
, last_ibp
);
2171 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2172 (offset
+ sizeof(xfs_agino_t
) - 1));
2173 xfs_inobp_check(mp
, last_ibp
);
2179 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2180 * inodes that are in memory - they all must be marked stale and attached to
2181 * the cluster buffer.
2185 xfs_inode_t
*free_ip
,
2189 xfs_mount_t
*mp
= free_ip
->i_mount
;
2190 int blks_per_cluster
;
2191 int inodes_per_cluster
;
2197 xfs_inode_log_item_t
*iip
;
2198 xfs_log_item_t
*lip
;
2199 struct xfs_perag
*pag
;
2201 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2202 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2203 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
2204 nbufs
= mp
->m_ialloc_blks
/ blks_per_cluster
;
2206 for (j
= 0; j
< nbufs
; j
++, inum
+= inodes_per_cluster
) {
2207 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2208 XFS_INO_TO_AGBNO(mp
, inum
));
2211 * We obtain and lock the backing buffer first in the process
2212 * here, as we have to ensure that any dirty inode that we
2213 * can't get the flush lock on is attached to the buffer.
2214 * If we scan the in-memory inodes first, then buffer IO can
2215 * complete before we get a lock on it, and hence we may fail
2216 * to mark all the active inodes on the buffer stale.
2218 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2219 mp
->m_bsize
* blks_per_cluster
,
2226 * This buffer may not have been correctly initialised as we
2227 * didn't read it from disk. That's not important because we are
2228 * only using to mark the buffer as stale in the log, and to
2229 * attach stale cached inodes on it. That means it will never be
2230 * dispatched for IO. If it is, we want to know about it, and we
2231 * want it to fail. We can acheive this by adding a write
2232 * verifier to the buffer.
2234 bp
->b_ops
= &xfs_inode_buf_ops
;
2237 * Walk the inodes already attached to the buffer and mark them
2238 * stale. These will all have the flush locks held, so an
2239 * in-memory inode walk can't lock them. By marking them all
2240 * stale first, we will not attempt to lock them in the loop
2241 * below as the XFS_ISTALE flag will be set.
2245 if (lip
->li_type
== XFS_LI_INODE
) {
2246 iip
= (xfs_inode_log_item_t
*)lip
;
2247 ASSERT(iip
->ili_logged
== 1);
2248 lip
->li_cb
= xfs_istale_done
;
2249 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2250 &iip
->ili_flush_lsn
,
2251 &iip
->ili_item
.li_lsn
);
2252 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2254 lip
= lip
->li_bio_list
;
2259 * For each inode in memory attempt to add it to the inode
2260 * buffer and set it up for being staled on buffer IO
2261 * completion. This is safe as we've locked out tail pushing
2262 * and flushing by locking the buffer.
2264 * We have already marked every inode that was part of a
2265 * transaction stale above, which means there is no point in
2266 * even trying to lock them.
2268 for (i
= 0; i
< inodes_per_cluster
; i
++) {
2271 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2272 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2274 /* Inode not in memory, nothing to do */
2281 * because this is an RCU protected lookup, we could
2282 * find a recently freed or even reallocated inode
2283 * during the lookup. We need to check under the
2284 * i_flags_lock for a valid inode here. Skip it if it
2285 * is not valid, the wrong inode or stale.
2287 spin_lock(&ip
->i_flags_lock
);
2288 if (ip
->i_ino
!= inum
+ i
||
2289 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2290 spin_unlock(&ip
->i_flags_lock
);
2294 spin_unlock(&ip
->i_flags_lock
);
2297 * Don't try to lock/unlock the current inode, but we
2298 * _cannot_ skip the other inodes that we did not find
2299 * in the list attached to the buffer and are not
2300 * already marked stale. If we can't lock it, back off
2303 if (ip
!= free_ip
&&
2304 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2312 xfs_iflags_set(ip
, XFS_ISTALE
);
2315 * we don't need to attach clean inodes or those only
2316 * with unlogged changes (which we throw away, anyway).
2319 if (!iip
|| xfs_inode_clean(ip
)) {
2320 ASSERT(ip
!= free_ip
);
2322 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2326 iip
->ili_last_fields
= iip
->ili_fields
;
2327 iip
->ili_fields
= 0;
2328 iip
->ili_logged
= 1;
2329 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2330 &iip
->ili_item
.li_lsn
);
2332 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2336 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2339 xfs_trans_stale_inode_buf(tp
, bp
);
2340 xfs_trans_binval(tp
, bp
);
2348 * This is called to return an inode to the inode free list.
2349 * The inode should already be truncated to 0 length and have
2350 * no pages associated with it. This routine also assumes that
2351 * the inode is already a part of the transaction.
2353 * The on-disk copy of the inode will have been added to the list
2354 * of unlinked inodes in the AGI. We need to remove the inode from
2355 * that list atomically with respect to freeing it here.
2361 xfs_bmap_free_t
*flist
)
2365 xfs_ino_t first_ino
;
2367 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2368 ASSERT(ip
->i_d
.di_nlink
== 0);
2369 ASSERT(ip
->i_d
.di_nextents
== 0);
2370 ASSERT(ip
->i_d
.di_anextents
== 0);
2371 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
2372 ASSERT(ip
->i_d
.di_nblocks
== 0);
2375 * Pull the on-disk inode from the AGI unlinked list.
2377 error
= xfs_iunlink_remove(tp
, ip
);
2381 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2385 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2386 ip
->i_d
.di_flags
= 0;
2387 ip
->i_d
.di_dmevmask
= 0;
2388 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2389 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2390 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2392 * Bump the generation count so no one will be confused
2393 * by reincarnations of this inode.
2396 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2399 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
2405 * This is called to unpin an inode. The caller must have the inode locked
2406 * in at least shared mode so that the buffer cannot be subsequently pinned
2407 * once someone is waiting for it to be unpinned.
2411 struct xfs_inode
*ip
)
2413 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2415 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2417 /* Give the log a push to start the unpinning I/O */
2418 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2424 struct xfs_inode
*ip
)
2426 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2427 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2432 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2433 if (xfs_ipincount(ip
))
2435 } while (xfs_ipincount(ip
));
2436 finish_wait(wq
, &wait
.wait
);
2441 struct xfs_inode
*ip
)
2443 if (xfs_ipincount(ip
))
2444 __xfs_iunpin_wait(ip
);
2448 * Removing an inode from the namespace involves removing the directory entry
2449 * and dropping the link count on the inode. Removing the directory entry can
2450 * result in locking an AGF (directory blocks were freed) and removing a link
2451 * count can result in placing the inode on an unlinked list which results in
2454 * The big problem here is that we have an ordering constraint on AGF and AGI
2455 * locking - inode allocation locks the AGI, then can allocate a new extent for
2456 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2457 * removes the inode from the unlinked list, requiring that we lock the AGI
2458 * first, and then freeing the inode can result in an inode chunk being freed
2459 * and hence freeing disk space requiring that we lock an AGF.
2461 * Hence the ordering that is imposed by other parts of the code is AGI before
2462 * AGF. This means we cannot remove the directory entry before we drop the inode
2463 * reference count and put it on the unlinked list as this results in a lock
2464 * order of AGF then AGI, and this can deadlock against inode allocation and
2465 * freeing. Therefore we must drop the link counts before we remove the
2468 * This is still safe from a transactional point of view - it is not until we
2469 * get to xfs_bmap_finish() that we have the possibility of multiple
2470 * transactions in this operation. Hence as long as we remove the directory
2471 * entry and drop the link count in the first transaction of the remove
2472 * operation, there are no transactional constraints on the ordering here.
2477 struct xfs_name
*name
,
2480 xfs_mount_t
*mp
= dp
->i_mount
;
2481 xfs_trans_t
*tp
= NULL
;
2482 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
2484 xfs_bmap_free_t free_list
;
2485 xfs_fsblock_t first_block
;
2490 trace_xfs_remove(dp
, name
);
2492 if (XFS_FORCED_SHUTDOWN(mp
))
2495 error
= xfs_qm_dqattach(dp
, 0);
2499 error
= xfs_qm_dqattach(ip
, 0);
2504 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2506 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2507 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2510 * We try to get the real space reservation first,
2511 * allowing for directory btree deletion(s) implying
2512 * possible bmap insert(s). If we can't get the space
2513 * reservation then we use 0 instead, and avoid the bmap
2514 * btree insert(s) in the directory code by, if the bmap
2515 * insert tries to happen, instead trimming the LAST
2516 * block from the directory.
2518 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2519 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, resblks
, 0);
2520 if (error
== -ENOSPC
) {
2522 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
2525 ASSERT(error
!= -ENOSPC
);
2527 goto out_trans_cancel
;
2530 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2532 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2533 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2536 * If we're removing a directory perform some additional validation.
2538 cancel_flags
|= XFS_TRANS_ABORT
;
2540 ASSERT(ip
->i_d
.di_nlink
>= 2);
2541 if (ip
->i_d
.di_nlink
!= 2) {
2543 goto out_trans_cancel
;
2545 if (!xfs_dir_isempty(ip
)) {
2547 goto out_trans_cancel
;
2550 /* Drop the link from ip's "..". */
2551 error
= xfs_droplink(tp
, dp
);
2553 goto out_trans_cancel
;
2555 /* Drop the "." link from ip to self. */
2556 error
= xfs_droplink(tp
, ip
);
2558 goto out_trans_cancel
;
2561 * When removing a non-directory we need to log the parent
2562 * inode here. For a directory this is done implicitly
2563 * by the xfs_droplink call for the ".." entry.
2565 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2567 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2569 /* Drop the link from dp to ip. */
2570 error
= xfs_droplink(tp
, ip
);
2572 goto out_trans_cancel
;
2574 xfs_bmap_init(&free_list
, &first_block
);
2575 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2576 &first_block
, &free_list
, resblks
);
2578 ASSERT(error
!= -ENOENT
);
2579 goto out_bmap_cancel
;
2583 * If this is a synchronous mount, make sure that the
2584 * remove transaction goes to disk before returning to
2587 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2588 xfs_trans_set_sync(tp
);
2590 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2592 goto out_bmap_cancel
;
2594 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2598 if (is_dir
&& xfs_inode_is_filestream(ip
))
2599 xfs_filestream_deassociate(ip
);
2604 xfs_bmap_cancel(&free_list
);
2606 xfs_trans_cancel(tp
, cancel_flags
);
2612 * Enter all inodes for a rename transaction into a sorted array.
2615 xfs_sort_for_rename(
2616 xfs_inode_t
*dp1
, /* in: old (source) directory inode */
2617 xfs_inode_t
*dp2
, /* in: new (target) directory inode */
2618 xfs_inode_t
*ip1
, /* in: inode of old entry */
2619 xfs_inode_t
*ip2
, /* in: inode of new entry, if it
2620 already exists, NULL otherwise. */
2621 xfs_inode_t
**i_tab
,/* out: array of inode returned, sorted */
2622 int *num_inodes
) /* out: number of inodes in array */
2628 * i_tab contains a list of pointers to inodes. We initialize
2629 * the table here & we'll sort it. We will then use it to
2630 * order the acquisition of the inode locks.
2632 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2646 * Sort the elements via bubble sort. (Remember, there are at
2647 * most 4 elements to sort, so this is adequate.)
2649 for (i
= 0; i
< *num_inodes
; i
++) {
2650 for (j
= 1; j
< *num_inodes
; j
++) {
2651 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2653 i_tab
[j
] = i_tab
[j
-1];
2665 xfs_inode_t
*src_dp
,
2666 struct xfs_name
*src_name
,
2667 xfs_inode_t
*src_ip
,
2668 xfs_inode_t
*target_dp
,
2669 struct xfs_name
*target_name
,
2670 xfs_inode_t
*target_ip
)
2672 xfs_trans_t
*tp
= NULL
;
2673 xfs_mount_t
*mp
= src_dp
->i_mount
;
2674 int new_parent
; /* moving to a new dir */
2675 int src_is_directory
; /* src_name is a directory */
2677 xfs_bmap_free_t free_list
;
2678 xfs_fsblock_t first_block
;
2681 xfs_inode_t
*inodes
[4];
2685 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2687 new_parent
= (src_dp
!= target_dp
);
2688 src_is_directory
= S_ISDIR(src_ip
->i_d
.di_mode
);
2690 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
,
2691 inodes
, &num_inodes
);
2693 xfs_bmap_init(&free_list
, &first_block
);
2694 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RENAME
);
2695 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2696 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2697 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, spaceres
, 0);
2698 if (error
== -ENOSPC
) {
2700 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, 0, 0);
2703 xfs_trans_cancel(tp
, 0);
2708 * Attach the dquots to the inodes
2710 error
= xfs_qm_vop_rename_dqattach(inodes
);
2712 xfs_trans_cancel(tp
, cancel_flags
);
2717 * Lock all the participating inodes. Depending upon whether
2718 * the target_name exists in the target directory, and
2719 * whether the target directory is the same as the source
2720 * directory, we can lock from 2 to 4 inodes.
2722 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2725 * Join all the inodes to the transaction. From this point on,
2726 * we can rely on either trans_commit or trans_cancel to unlock
2729 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2731 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2732 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2734 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2737 * If we are using project inheritance, we only allow renames
2738 * into our tree when the project IDs are the same; else the
2739 * tree quota mechanism would be circumvented.
2741 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2742 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2748 * Set up the target.
2750 if (target_ip
== NULL
) {
2752 * If there's no space reservation, check the entry will
2753 * fit before actually inserting it.
2756 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
2761 * If target does not exist and the rename crosses
2762 * directories, adjust the target directory link count
2763 * to account for the ".." reference from the new entry.
2765 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
2766 src_ip
->i_ino
, &first_block
,
2767 &free_list
, spaceres
);
2768 if (error
== -ENOSPC
)
2773 xfs_trans_ichgtime(tp
, target_dp
,
2774 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2776 if (new_parent
&& src_is_directory
) {
2777 error
= xfs_bumplink(tp
, target_dp
);
2781 } else { /* target_ip != NULL */
2783 * If target exists and it's a directory, check that both
2784 * target and source are directories and that target can be
2785 * destroyed, or that neither is a directory.
2787 if (S_ISDIR(target_ip
->i_d
.di_mode
)) {
2789 * Make sure target dir is empty.
2791 if (!(xfs_dir_isempty(target_ip
)) ||
2792 (target_ip
->i_d
.di_nlink
> 2)) {
2799 * Link the source inode under the target name.
2800 * If the source inode is a directory and we are moving
2801 * it across directories, its ".." entry will be
2802 * inconsistent until we replace that down below.
2804 * In case there is already an entry with the same
2805 * name at the destination directory, remove it first.
2807 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
2809 &first_block
, &free_list
, spaceres
);
2813 xfs_trans_ichgtime(tp
, target_dp
,
2814 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2817 * Decrement the link count on the target since the target
2818 * dir no longer points to it.
2820 error
= xfs_droplink(tp
, target_ip
);
2824 if (src_is_directory
) {
2826 * Drop the link from the old "." entry.
2828 error
= xfs_droplink(tp
, target_ip
);
2832 } /* target_ip != NULL */
2835 * Remove the source.
2837 if (new_parent
&& src_is_directory
) {
2839 * Rewrite the ".." entry to point to the new
2842 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
2844 &first_block
, &free_list
, spaceres
);
2845 ASSERT(error
!= -EEXIST
);
2851 * We always want to hit the ctime on the source inode.
2853 * This isn't strictly required by the standards since the source
2854 * inode isn't really being changed, but old unix file systems did
2855 * it and some incremental backup programs won't work without it.
2857 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
2858 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
2861 * Adjust the link count on src_dp. This is necessary when
2862 * renaming a directory, either within one parent when
2863 * the target existed, or across two parent directories.
2865 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
2868 * Decrement link count on src_directory since the
2869 * entry that's moved no longer points to it.
2871 error
= xfs_droplink(tp
, src_dp
);
2876 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
2877 &first_block
, &free_list
, spaceres
);
2881 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2882 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
2884 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
2887 * If this is a synchronous mount, make sure that the
2888 * rename transaction goes to disk before returning to
2891 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2892 xfs_trans_set_sync(tp
);
2895 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2897 xfs_bmap_cancel(&free_list
);
2898 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
2904 * trans_commit will unlock src_ip, target_ip & decrement
2905 * the vnode references.
2907 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2910 cancel_flags
|= XFS_TRANS_ABORT
;
2912 xfs_bmap_cancel(&free_list
);
2913 xfs_trans_cancel(tp
, cancel_flags
);
2923 xfs_mount_t
*mp
= ip
->i_mount
;
2924 struct xfs_perag
*pag
;
2925 unsigned long first_index
, mask
;
2926 unsigned long inodes_per_cluster
;
2928 xfs_inode_t
**ilist
;
2935 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2937 inodes_per_cluster
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
2938 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2939 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2943 mask
= ~(((mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
)) - 1);
2944 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2946 /* really need a gang lookup range call here */
2947 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2948 first_index
, inodes_per_cluster
);
2952 for (i
= 0; i
< nr_found
; i
++) {
2958 * because this is an RCU protected lookup, we could find a
2959 * recently freed or even reallocated inode during the lookup.
2960 * We need to check under the i_flags_lock for a valid inode
2961 * here. Skip it if it is not valid or the wrong inode.
2963 spin_lock(&ip
->i_flags_lock
);
2965 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2966 spin_unlock(&ip
->i_flags_lock
);
2969 spin_unlock(&ip
->i_flags_lock
);
2972 * Do an un-protected check to see if the inode is dirty and
2973 * is a candidate for flushing. These checks will be repeated
2974 * later after the appropriate locks are acquired.
2976 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2980 * Try to get locks. If any are unavailable or it is pinned,
2981 * then this inode cannot be flushed and is skipped.
2984 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2986 if (!xfs_iflock_nowait(iq
)) {
2987 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2990 if (xfs_ipincount(iq
)) {
2992 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2997 * arriving here means that this inode can be flushed. First
2998 * re-check that it's dirty before flushing.
3000 if (!xfs_inode_clean(iq
)) {
3002 error
= xfs_iflush_int(iq
, bp
);
3004 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3005 goto cluster_corrupt_out
;
3011 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3015 XFS_STATS_INC(xs_icluster_flushcnt
);
3016 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3027 cluster_corrupt_out
:
3029 * Corruption detected in the clustering loop. Invalidate the
3030 * inode buffer and shut down the filesystem.
3034 * Clean up the buffer. If it was delwri, just release it --
3035 * brelse can handle it with no problems. If not, shut down the
3036 * filesystem before releasing the buffer.
3038 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3042 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3044 if (!bufwasdelwri
) {
3046 * Just like incore_relse: if we have b_iodone functions,
3047 * mark the buffer as an error and call them. Otherwise
3048 * mark it as stale and brelse.
3053 xfs_buf_ioerror(bp
, -EIO
);
3062 * Unlocks the flush lock
3064 xfs_iflush_abort(iq
, false);
3067 return -EFSCORRUPTED
;
3071 * Flush dirty inode metadata into the backing buffer.
3073 * The caller must have the inode lock and the inode flush lock held. The
3074 * inode lock will still be held upon return to the caller, and the inode
3075 * flush lock will be released after the inode has reached the disk.
3077 * The caller must write out the buffer returned in *bpp and release it.
3081 struct xfs_inode
*ip
,
3082 struct xfs_buf
**bpp
)
3084 struct xfs_mount
*mp
= ip
->i_mount
;
3086 struct xfs_dinode
*dip
;
3089 XFS_STATS_INC(xs_iflush_count
);
3091 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3092 ASSERT(xfs_isiflocked(ip
));
3093 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3094 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3098 xfs_iunpin_wait(ip
);
3101 * For stale inodes we cannot rely on the backing buffer remaining
3102 * stale in cache for the remaining life of the stale inode and so
3103 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3104 * inodes below. We have to check this after ensuring the inode is
3105 * unpinned so that it is safe to reclaim the stale inode after the
3108 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3114 * This may have been unpinned because the filesystem is shutting
3115 * down forcibly. If that's the case we must not write this inode
3116 * to disk, because the log record didn't make it to disk.
3118 * We also have to remove the log item from the AIL in this case,
3119 * as we wait for an empty AIL as part of the unmount process.
3121 if (XFS_FORCED_SHUTDOWN(mp
)) {
3127 * Get the buffer containing the on-disk inode.
3129 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3137 * First flush out the inode that xfs_iflush was called with.
3139 error
= xfs_iflush_int(ip
, bp
);
3144 * If the buffer is pinned then push on the log now so we won't
3145 * get stuck waiting in the write for too long.
3147 if (xfs_buf_ispinned(bp
))
3148 xfs_log_force(mp
, 0);
3152 * see if other inodes can be gathered into this write
3154 error
= xfs_iflush_cluster(ip
, bp
);
3156 goto cluster_corrupt_out
;
3163 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3164 cluster_corrupt_out
:
3165 error
= -EFSCORRUPTED
;
3168 * Unlocks the flush lock
3170 xfs_iflush_abort(ip
, false);
3176 struct xfs_inode
*ip
,
3179 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3180 struct xfs_dinode
*dip
;
3181 struct xfs_mount
*mp
= ip
->i_mount
;
3183 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3184 ASSERT(xfs_isiflocked(ip
));
3185 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3186 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3187 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3188 ASSERT(ip
->i_d
.di_version
> 1);
3190 /* set *dip = inode's place in the buffer */
3191 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3193 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3194 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3195 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3196 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3197 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3200 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3201 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3202 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3203 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3204 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3207 if (S_ISREG(ip
->i_d
.di_mode
)) {
3209 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3210 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3211 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3212 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3213 "%s: Bad regular inode %Lu, ptr 0x%p",
3214 __func__
, ip
->i_ino
, ip
);
3217 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
3219 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3220 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3221 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3222 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3223 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3224 "%s: Bad directory inode %Lu, ptr 0x%p",
3225 __func__
, ip
->i_ino
, ip
);
3229 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3230 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3231 XFS_RANDOM_IFLUSH_5
)) {
3232 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3233 "%s: detected corrupt incore inode %Lu, "
3234 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3235 __func__
, ip
->i_ino
,
3236 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3237 ip
->i_d
.di_nblocks
, ip
);
3240 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3241 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3242 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3243 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3244 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3249 * Inode item log recovery for v2 inodes are dependent on the
3250 * di_flushiter count for correct sequencing. We bump the flush
3251 * iteration count so we can detect flushes which postdate a log record
3252 * during recovery. This is redundant as we now log every change and
3253 * hence this can't happen but we need to still do it to ensure
3254 * backwards compatibility with old kernels that predate logging all
3257 if (ip
->i_d
.di_version
< 3)
3258 ip
->i_d
.di_flushiter
++;
3261 * Copy the dirty parts of the inode into the on-disk
3262 * inode. We always copy out the core of the inode,
3263 * because if the inode is dirty at all the core must
3266 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3268 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3269 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3270 ip
->i_d
.di_flushiter
= 0;
3272 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3273 if (XFS_IFORK_Q(ip
))
3274 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3275 xfs_inobp_check(mp
, bp
);
3278 * We've recorded everything logged in the inode, so we'd like to clear
3279 * the ili_fields bits so we don't log and flush things unnecessarily.
3280 * However, we can't stop logging all this information until the data
3281 * we've copied into the disk buffer is written to disk. If we did we
3282 * might overwrite the copy of the inode in the log with all the data
3283 * after re-logging only part of it, and in the face of a crash we
3284 * wouldn't have all the data we need to recover.
3286 * What we do is move the bits to the ili_last_fields field. When
3287 * logging the inode, these bits are moved back to the ili_fields field.
3288 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3289 * know that the information those bits represent is permanently on
3290 * disk. As long as the flush completes before the inode is logged
3291 * again, then both ili_fields and ili_last_fields will be cleared.
3293 * We can play with the ili_fields bits here, because the inode lock
3294 * must be held exclusively in order to set bits there and the flush
3295 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3296 * done routine can tell whether or not to look in the AIL. Also, store
3297 * the current LSN of the inode so that we can tell whether the item has
3298 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3299 * need the AIL lock, because it is a 64 bit value that cannot be read
3302 iip
->ili_last_fields
= iip
->ili_fields
;
3303 iip
->ili_fields
= 0;
3304 iip
->ili_logged
= 1;
3306 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3307 &iip
->ili_item
.li_lsn
);
3310 * Attach the function xfs_iflush_done to the inode's
3311 * buffer. This will remove the inode from the AIL
3312 * and unlock the inode's flush lock when the inode is
3313 * completely written to disk.
3315 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3317 /* update the lsn in the on disk inode if required */
3318 if (ip
->i_d
.di_version
== 3)
3319 dip
->di_lsn
= cpu_to_be64(iip
->ili_item
.li_lsn
);
3321 /* generate the checksum. */
3322 xfs_dinode_calc_crc(mp
, dip
);
3324 ASSERT(bp
->b_fspriv
!= NULL
);
3325 ASSERT(bp
->b_iodone
!= NULL
);
3329 return -EFSCORRUPTED
;