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
))
586 return XFS_ERROR(EIO
);
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
;
661 * Call the space management code to pick
662 * the on-disk inode to be allocated.
664 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
665 ialloc_context
, &ino
);
668 if (*ialloc_context
|| ino
== NULLFSINO
) {
672 ASSERT(*ialloc_context
== NULL
);
675 * Get the in-core inode with the lock held exclusively.
676 * This is because we're setting fields here we need
677 * to prevent others from looking at until we're done.
679 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
680 XFS_ILOCK_EXCL
, &ip
);
685 ip
->i_d
.di_mode
= mode
;
686 ip
->i_d
.di_onlink
= 0;
687 ip
->i_d
.di_nlink
= nlink
;
688 ASSERT(ip
->i_d
.di_nlink
== nlink
);
689 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
690 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
691 xfs_set_projid(ip
, prid
);
692 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
695 * If the superblock version is up to where we support new format
696 * inodes and this is currently an old format inode, then change
697 * the inode version number now. This way we only do the conversion
698 * here rather than here and in the flush/logging code.
700 if (xfs_sb_version_hasnlink(&mp
->m_sb
) &&
701 ip
->i_d
.di_version
== 1) {
702 ip
->i_d
.di_version
= 2;
704 * We've already zeroed the old link count, the projid field,
710 * Project ids won't be stored on disk if we are using a version 1 inode.
712 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
713 xfs_bump_ino_vers2(tp
, ip
);
715 if (pip
&& XFS_INHERIT_GID(pip
)) {
716 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
717 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
718 ip
->i_d
.di_mode
|= S_ISGID
;
723 * If the group ID of the new file does not match the effective group
724 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
725 * (and only if the irix_sgid_inherit compatibility variable is set).
727 if ((irix_sgid_inherit
) &&
728 (ip
->i_d
.di_mode
& S_ISGID
) &&
729 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
)))) {
730 ip
->i_d
.di_mode
&= ~S_ISGID
;
734 ip
->i_d
.di_nextents
= 0;
735 ASSERT(ip
->i_d
.di_nblocks
== 0);
738 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
739 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
740 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
741 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
744 * di_gen will have been taken care of in xfs_iread.
746 ip
->i_d
.di_extsize
= 0;
747 ip
->i_d
.di_dmevmask
= 0;
748 ip
->i_d
.di_dmstate
= 0;
749 ip
->i_d
.di_flags
= 0;
751 if (ip
->i_d
.di_version
== 3) {
752 ASSERT(ip
->i_d
.di_ino
== ino
);
753 ASSERT(uuid_equal(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
));
755 ip
->i_d
.di_changecount
= 1;
757 ip
->i_d
.di_flags2
= 0;
758 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
759 ip
->i_d
.di_crtime
= ip
->i_d
.di_mtime
;
763 flags
= XFS_ILOG_CORE
;
764 switch (mode
& S_IFMT
) {
769 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
770 ip
->i_df
.if_u2
.if_rdev
= rdev
;
771 ip
->i_df
.if_flags
= 0;
772 flags
|= XFS_ILOG_DEV
;
776 * we can't set up filestreams until after the VFS inode
777 * is set up properly.
779 if (pip
&& xfs_inode_is_filestream(pip
))
783 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
787 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
788 di_flags
|= XFS_DIFLAG_RTINHERIT
;
789 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
790 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
791 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
793 } else if (S_ISREG(mode
)) {
794 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
795 di_flags
|= XFS_DIFLAG_REALTIME
;
796 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
797 di_flags
|= XFS_DIFLAG_EXTSIZE
;
798 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
801 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
803 di_flags
|= XFS_DIFLAG_NOATIME
;
804 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
806 di_flags
|= XFS_DIFLAG_NODUMP
;
807 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
809 di_flags
|= XFS_DIFLAG_SYNC
;
810 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
811 xfs_inherit_nosymlinks
)
812 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
813 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
814 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
815 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
816 xfs_inherit_nodefrag
)
817 di_flags
|= XFS_DIFLAG_NODEFRAG
;
818 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
819 di_flags
|= XFS_DIFLAG_FILESTREAM
;
820 ip
->i_d
.di_flags
|= di_flags
;
824 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
825 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
826 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
827 ip
->i_df
.if_u1
.if_extents
= NULL
;
833 * Attribute fork settings for new inode.
835 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
836 ip
->i_d
.di_anextents
= 0;
839 * Log the new values stuffed into the inode.
841 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
842 xfs_trans_log_inode(tp
, ip
, flags
);
844 /* now that we have an i_mode we can setup inode ops and unlock */
847 /* now we have set up the vfs inode we can associate the filestream */
849 error
= xfs_filestream_associate(pip
, ip
);
853 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
861 * Allocates a new inode from disk and return a pointer to the
862 * incore copy. This routine will internally commit the current
863 * transaction and allocate a new one if the Space Manager needed
864 * to do an allocation to replenish the inode free-list.
866 * This routine is designed to be called from xfs_create and
872 xfs_trans_t
**tpp
, /* input: current transaction;
873 output: may be a new transaction. */
874 xfs_inode_t
*dp
, /* directory within whose allocate
879 prid_t prid
, /* project id */
880 int okalloc
, /* ok to allocate new space */
881 xfs_inode_t
**ipp
, /* pointer to inode; it will be
889 xfs_buf_t
*ialloc_context
= NULL
;
895 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
898 * xfs_ialloc will return a pointer to an incore inode if
899 * the Space Manager has an available inode on the free
900 * list. Otherwise, it will do an allocation and replenish
901 * the freelist. Since we can only do one allocation per
902 * transaction without deadlocks, we will need to commit the
903 * current transaction and start a new one. We will then
904 * need to call xfs_ialloc again to get the inode.
906 * If xfs_ialloc did an allocation to replenish the freelist,
907 * it returns the bp containing the head of the freelist as
908 * ialloc_context. We will hold a lock on it across the
909 * transaction commit so that no other process can steal
910 * the inode(s) that we've just allocated.
912 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
913 &ialloc_context
, &ip
);
916 * Return an error if we were unable to allocate a new inode.
917 * This should only happen if we run out of space on disk or
918 * encounter a disk error.
924 if (!ialloc_context
&& !ip
) {
926 return XFS_ERROR(ENOSPC
);
930 * If the AGI buffer is non-NULL, then we were unable to get an
931 * inode in one operation. We need to commit the current
932 * transaction and call xfs_ialloc() again. It is guaranteed
933 * to succeed the second time.
935 if (ialloc_context
) {
936 struct xfs_trans_res tres
;
939 * Normally, xfs_trans_commit releases all the locks.
940 * We call bhold to hang on to the ialloc_context across
941 * the commit. Holding this buffer prevents any other
942 * processes from doing any allocations in this
945 xfs_trans_bhold(tp
, ialloc_context
);
947 * Save the log reservation so we can use
948 * them in the next transaction.
950 tres
.tr_logres
= xfs_trans_get_log_res(tp
);
951 tres
.tr_logcount
= xfs_trans_get_log_count(tp
);
954 * We want the quota changes to be associated with the next
955 * transaction, NOT this one. So, detach the dqinfo from this
956 * and attach it to the next transaction.
961 dqinfo
= (void *)tp
->t_dqinfo
;
963 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
964 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
967 ntp
= xfs_trans_dup(tp
);
968 code
= xfs_trans_commit(tp
, 0);
970 if (committed
!= NULL
) {
974 * If we get an error during the commit processing,
975 * release the buffer that is still held and return
979 xfs_buf_relse(ialloc_context
);
981 tp
->t_dqinfo
= dqinfo
;
982 xfs_trans_free_dqinfo(tp
);
990 * transaction commit worked ok so we can drop the extra ticket
991 * reference that we gained in xfs_trans_dup()
993 xfs_log_ticket_put(tp
->t_ticket
);
994 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
995 code
= xfs_trans_reserve(tp
, &tres
, 0, 0);
998 * Re-attach the quota info that we detached from prev trx.
1001 tp
->t_dqinfo
= dqinfo
;
1002 tp
->t_flags
|= tflags
;
1006 xfs_buf_relse(ialloc_context
);
1011 xfs_trans_bjoin(tp
, ialloc_context
);
1014 * Call ialloc again. Since we've locked out all
1015 * other allocations in this allocation group,
1016 * this call should always succeed.
1018 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1019 okalloc
, &ialloc_context
, &ip
);
1022 * If we get an error at this point, return to the caller
1023 * so that the current transaction can be aborted.
1030 ASSERT(!ialloc_context
&& ip
);
1033 if (committed
!= NULL
)
1044 * Decrement the link count on an inode & log the change.
1045 * If this causes the link count to go to zero, initiate the
1046 * logging activity required to truncate a file.
1055 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1057 ASSERT (ip
->i_d
.di_nlink
> 0);
1059 drop_nlink(VFS_I(ip
));
1060 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1063 if (ip
->i_d
.di_nlink
== 0) {
1065 * We're dropping the last link to this file.
1066 * Move the on-disk inode to the AGI unlinked list.
1067 * From xfs_inactive() we will pull the inode from
1068 * the list and free it.
1070 error
= xfs_iunlink(tp
, ip
);
1076 * This gets called when the inode's version needs to be changed from 1 to 2.
1077 * Currently this happens when the nlink field overflows the old 16-bit value
1078 * or when chproj is called to change the project for the first time.
1079 * As a side effect the superblock version will also get rev'd
1080 * to contain the NLINK bit.
1089 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1090 ASSERT(ip
->i_d
.di_version
== 1);
1092 ip
->i_d
.di_version
= 2;
1093 ip
->i_d
.di_onlink
= 0;
1094 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1096 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1097 spin_lock(&mp
->m_sb_lock
);
1098 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
1099 xfs_sb_version_addnlink(&mp
->m_sb
);
1100 spin_unlock(&mp
->m_sb_lock
);
1101 xfs_mod_sb(tp
, XFS_SB_VERSIONNUM
);
1103 spin_unlock(&mp
->m_sb_lock
);
1106 /* Caller must log the inode */
1110 * Increment the link count on an inode & log the change.
1117 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1119 ASSERT(ip
->i_d
.di_nlink
> 0 || (VFS_I(ip
)->i_state
& I_LINKABLE
));
1121 inc_nlink(VFS_I(ip
));
1122 if ((ip
->i_d
.di_version
== 1) &&
1123 (ip
->i_d
.di_nlink
> XFS_MAXLINK_1
)) {
1125 * The inode has increased its number of links beyond
1126 * what can fit in an old format inode. It now needs
1127 * to be converted to a version 2 inode with a 32 bit
1128 * link count. If this is the first inode in the file
1129 * system to do this, then we need to bump the superblock
1130 * version number as well.
1132 xfs_bump_ino_vers2(tp
, ip
);
1135 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1142 struct xfs_name
*name
,
1147 int is_dir
= S_ISDIR(mode
);
1148 struct xfs_mount
*mp
= dp
->i_mount
;
1149 struct xfs_inode
*ip
= NULL
;
1150 struct xfs_trans
*tp
= NULL
;
1152 xfs_bmap_free_t free_list
;
1153 xfs_fsblock_t first_block
;
1154 bool unlock_dp_on_error
= false;
1158 struct xfs_dquot
*udqp
= NULL
;
1159 struct xfs_dquot
*gdqp
= NULL
;
1160 struct xfs_dquot
*pdqp
= NULL
;
1161 struct xfs_trans_res tres
;
1164 trace_xfs_create(dp
, name
);
1166 if (XFS_FORCED_SHUTDOWN(mp
))
1167 return XFS_ERROR(EIO
);
1169 prid
= xfs_get_initial_prid(dp
);
1172 * Make sure that we have allocated dquot(s) on disk.
1174 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1175 xfs_kgid_to_gid(current_fsgid()), prid
,
1176 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1177 &udqp
, &gdqp
, &pdqp
);
1183 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1184 tres
.tr_logres
= M_RES(mp
)->tr_mkdir
.tr_logres
;
1185 tres
.tr_logcount
= XFS_MKDIR_LOG_COUNT
;
1186 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1188 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1189 tres
.tr_logres
= M_RES(mp
)->tr_create
.tr_logres
;
1190 tres
.tr_logcount
= XFS_CREATE_LOG_COUNT
;
1191 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1194 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1197 * Initially assume that the file does not exist and
1198 * reserve the resources for that case. If that is not
1199 * the case we'll drop the one we have and get a more
1200 * appropriate transaction later.
1202 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1203 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1204 if (error
== ENOSPC
) {
1205 /* flush outstanding delalloc blocks and retry */
1206 xfs_flush_inodes(mp
);
1207 error
= xfs_trans_reserve(tp
, &tres
, resblks
, 0);
1209 if (error
== ENOSPC
) {
1210 /* No space at all so try a "no-allocation" reservation */
1212 error
= xfs_trans_reserve(tp
, &tres
, 0, 0);
1216 goto out_trans_cancel
;
1219 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1220 unlock_dp_on_error
= true;
1222 xfs_bmap_init(&free_list
, &first_block
);
1225 * Reserve disk quota and the inode.
1227 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1228 pdqp
, resblks
, 1, 0);
1230 goto out_trans_cancel
;
1232 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1234 goto out_trans_cancel
;
1237 * A newly created regular or special file just has one directory
1238 * entry pointing to them, but a directory also the "." entry
1239 * pointing to itself.
1241 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1242 prid
, resblks
> 0, &ip
, &committed
);
1244 if (error
== ENOSPC
)
1245 goto out_trans_cancel
;
1246 goto out_trans_abort
;
1250 * Now we join the directory inode to the transaction. We do not do it
1251 * earlier because xfs_dir_ialloc might commit the previous transaction
1252 * (and release all the locks). An error from here on will result in
1253 * the transaction cancel unlocking dp so don't do it explicitly in the
1256 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1257 unlock_dp_on_error
= false;
1259 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1260 &first_block
, &free_list
, resblks
?
1261 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1263 ASSERT(error
!= ENOSPC
);
1264 goto out_trans_abort
;
1266 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1267 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1270 error
= xfs_dir_init(tp
, ip
, dp
);
1272 goto out_bmap_cancel
;
1274 error
= xfs_bumplink(tp
, dp
);
1276 goto out_bmap_cancel
;
1280 * If this is a synchronous mount, make sure that the
1281 * create transaction goes to disk before returning to
1284 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1285 xfs_trans_set_sync(tp
);
1288 * Attach the dquot(s) to the inodes and modify them incore.
1289 * These ids of the inode couldn't have changed since the new
1290 * inode has been locked ever since it was created.
1292 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1294 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1296 goto out_bmap_cancel
;
1298 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1300 goto out_release_inode
;
1302 xfs_qm_dqrele(udqp
);
1303 xfs_qm_dqrele(gdqp
);
1304 xfs_qm_dqrele(pdqp
);
1310 xfs_bmap_cancel(&free_list
);
1312 cancel_flags
|= XFS_TRANS_ABORT
;
1314 xfs_trans_cancel(tp
, cancel_flags
);
1317 * Wait until after the current transaction is aborted to
1318 * release the inode. This prevents recursive transactions
1319 * and deadlocks from xfs_inactive.
1324 xfs_qm_dqrele(udqp
);
1325 xfs_qm_dqrele(gdqp
);
1326 xfs_qm_dqrele(pdqp
);
1328 if (unlock_dp_on_error
)
1329 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1335 struct xfs_inode
*dp
,
1336 struct dentry
*dentry
,
1339 struct xfs_mount
*mp
= dp
->i_mount
;
1340 struct xfs_inode
*ip
= NULL
;
1341 struct xfs_trans
*tp
= NULL
;
1343 uint cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1345 struct xfs_dquot
*udqp
= NULL
;
1346 struct xfs_dquot
*gdqp
= NULL
;
1347 struct xfs_dquot
*pdqp
= NULL
;
1348 struct xfs_trans_res
*tres
;
1351 if (XFS_FORCED_SHUTDOWN(mp
))
1352 return XFS_ERROR(EIO
);
1354 prid
= xfs_get_initial_prid(dp
);
1357 * Make sure that we have allocated dquot(s) on disk.
1359 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1360 xfs_kgid_to_gid(current_fsgid()), prid
,
1361 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1362 &udqp
, &gdqp
, &pdqp
);
1366 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1367 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE_TMPFILE
);
1369 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1370 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1371 if (error
== ENOSPC
) {
1372 /* No space at all so try a "no-allocation" reservation */
1374 error
= xfs_trans_reserve(tp
, tres
, 0, 0);
1378 goto out_trans_cancel
;
1381 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1382 pdqp
, resblks
, 1, 0);
1384 goto out_trans_cancel
;
1386 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1, 0,
1387 prid
, resblks
> 0, &ip
, NULL
);
1389 if (error
== ENOSPC
)
1390 goto out_trans_cancel
;
1391 goto out_trans_abort
;
1394 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1395 xfs_trans_set_sync(tp
);
1398 * Attach the dquot(s) to the inodes and modify them incore.
1399 * These ids of the inode couldn't have changed since the new
1400 * inode has been locked ever since it was created.
1402 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1405 d_tmpfile(dentry
, VFS_I(ip
));
1406 error
= xfs_iunlink(tp
, ip
);
1408 goto out_trans_abort
;
1410 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1412 goto out_release_inode
;
1414 xfs_qm_dqrele(udqp
);
1415 xfs_qm_dqrele(gdqp
);
1416 xfs_qm_dqrele(pdqp
);
1421 cancel_flags
|= XFS_TRANS_ABORT
;
1423 xfs_trans_cancel(tp
, cancel_flags
);
1426 * Wait until after the current transaction is aborted to
1427 * release the inode. This prevents recursive transactions
1428 * and deadlocks from xfs_inactive.
1433 xfs_qm_dqrele(udqp
);
1434 xfs_qm_dqrele(gdqp
);
1435 xfs_qm_dqrele(pdqp
);
1444 struct xfs_name
*target_name
)
1446 xfs_mount_t
*mp
= tdp
->i_mount
;
1449 xfs_bmap_free_t free_list
;
1450 xfs_fsblock_t first_block
;
1455 trace_xfs_link(tdp
, target_name
);
1457 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1459 if (XFS_FORCED_SHUTDOWN(mp
))
1460 return XFS_ERROR(EIO
);
1462 error
= xfs_qm_dqattach(sip
, 0);
1466 error
= xfs_qm_dqattach(tdp
, 0);
1470 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1471 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1472 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1473 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, resblks
, 0);
1474 if (error
== ENOSPC
) {
1476 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, 0, 0);
1483 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1485 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1486 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1489 * If we are using project inheritance, we only allow hard link
1490 * creation in our tree when the project IDs are the same; else
1491 * the tree quota mechanism could be circumvented.
1493 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1494 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1495 error
= XFS_ERROR(EXDEV
);
1499 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
1503 xfs_bmap_init(&free_list
, &first_block
);
1505 if (sip
->i_d
.di_nlink
== 0) {
1506 error
= xfs_iunlink_remove(tp
, sip
);
1511 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1512 &first_block
, &free_list
, resblks
);
1515 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1516 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1518 error
= xfs_bumplink(tp
, sip
);
1523 * If this is a synchronous mount, make sure that the
1524 * link transaction goes to disk before returning to
1527 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1528 xfs_trans_set_sync(tp
);
1531 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1533 xfs_bmap_cancel(&free_list
);
1537 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1540 cancel_flags
|= XFS_TRANS_ABORT
;
1542 xfs_trans_cancel(tp
, cancel_flags
);
1548 * Free up the underlying blocks past new_size. The new size must be smaller
1549 * than the current size. This routine can be used both for the attribute and
1550 * data fork, and does not modify the inode size, which is left to the caller.
1552 * The transaction passed to this routine must have made a permanent log
1553 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1554 * given transaction and start new ones, so make sure everything involved in
1555 * the transaction is tidy before calling here. Some transaction will be
1556 * returned to the caller to be committed. The incoming transaction must
1557 * already include the inode, and both inode locks must be held exclusively.
1558 * The inode must also be "held" within the transaction. On return the inode
1559 * will be "held" within the returned transaction. This routine does NOT
1560 * require any disk space to be reserved for it within the transaction.
1562 * If we get an error, we must return with the inode locked and linked into the
1563 * current transaction. This keeps things simple for the higher level code,
1564 * because it always knows that the inode is locked and held in the transaction
1565 * that returns to it whether errors occur or not. We don't mark the inode
1566 * dirty on error so that transactions can be easily aborted if possible.
1569 xfs_itruncate_extents(
1570 struct xfs_trans
**tpp
,
1571 struct xfs_inode
*ip
,
1573 xfs_fsize_t new_size
)
1575 struct xfs_mount
*mp
= ip
->i_mount
;
1576 struct xfs_trans
*tp
= *tpp
;
1577 struct xfs_trans
*ntp
;
1578 xfs_bmap_free_t free_list
;
1579 xfs_fsblock_t first_block
;
1580 xfs_fileoff_t first_unmap_block
;
1581 xfs_fileoff_t last_block
;
1582 xfs_filblks_t unmap_len
;
1587 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1588 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1589 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1590 ASSERT(new_size
<= XFS_ISIZE(ip
));
1591 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1592 ASSERT(ip
->i_itemp
!= NULL
);
1593 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1594 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1596 trace_xfs_itruncate_extents_start(ip
, new_size
);
1599 * Since it is possible for space to become allocated beyond
1600 * the end of the file (in a crash where the space is allocated
1601 * but the inode size is not yet updated), simply remove any
1602 * blocks which show up between the new EOF and the maximum
1603 * possible file size. If the first block to be removed is
1604 * beyond the maximum file size (ie it is the same as last_block),
1605 * then there is nothing to do.
1607 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1608 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1609 if (first_unmap_block
== last_block
)
1612 ASSERT(first_unmap_block
< last_block
);
1613 unmap_len
= last_block
- first_unmap_block
+ 1;
1615 xfs_bmap_init(&free_list
, &first_block
);
1616 error
= xfs_bunmapi(tp
, ip
,
1617 first_unmap_block
, unmap_len
,
1618 xfs_bmapi_aflag(whichfork
),
1619 XFS_ITRUNC_MAX_EXTENTS
,
1620 &first_block
, &free_list
,
1623 goto out_bmap_cancel
;
1626 * Duplicate the transaction that has the permanent
1627 * reservation and commit the old transaction.
1629 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1631 xfs_trans_ijoin(tp
, ip
, 0);
1633 goto out_bmap_cancel
;
1637 * Mark the inode dirty so it will be logged and
1638 * moved forward in the log as part of every commit.
1640 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1643 ntp
= xfs_trans_dup(tp
);
1644 error
= xfs_trans_commit(tp
, 0);
1647 xfs_trans_ijoin(tp
, ip
, 0);
1653 * Transaction commit worked ok so we can drop the extra ticket
1654 * reference that we gained in xfs_trans_dup()
1656 xfs_log_ticket_put(tp
->t_ticket
);
1657 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1663 * Always re-log the inode so that our permanent transaction can keep
1664 * on rolling it forward in the log.
1666 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1668 trace_xfs_itruncate_extents_end(ip
, new_size
);
1675 * If the bunmapi call encounters an error, return to the caller where
1676 * the transaction can be properly aborted. We just need to make sure
1677 * we're not holding any resources that we were not when we came in.
1679 xfs_bmap_cancel(&free_list
);
1687 xfs_mount_t
*mp
= ip
->i_mount
;
1690 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1693 /* If this is a read-only mount, don't do this (would generate I/O) */
1694 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1697 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1701 * If we are using filestreams, and we have an unlinked
1702 * file that we are processing the last close on, then nothing
1703 * will be able to reopen and write to this file. Purge this
1704 * inode from the filestreams cache so that it doesn't delay
1705 * teardown of the inode.
1707 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1708 xfs_filestream_deassociate(ip
);
1711 * If we previously truncated this file and removed old data
1712 * in the process, we want to initiate "early" writeout on
1713 * the last close. This is an attempt to combat the notorious
1714 * NULL files problem which is particularly noticeable from a
1715 * truncate down, buffered (re-)write (delalloc), followed by
1716 * a crash. What we are effectively doing here is
1717 * significantly reducing the time window where we'd otherwise
1718 * be exposed to that problem.
1720 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1722 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1723 if (VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0) {
1724 error
= -filemap_flush(VFS_I(ip
)->i_mapping
);
1731 if (ip
->i_d
.di_nlink
== 0)
1734 if (xfs_can_free_eofblocks(ip
, false)) {
1737 * If we can't get the iolock just skip truncating the blocks
1738 * past EOF because we could deadlock with the mmap_sem
1739 * otherwise. We'll get another chance to drop them once the
1740 * last reference to the inode is dropped, so we'll never leak
1741 * blocks permanently.
1743 * Further, check if the inode is being opened, written and
1744 * closed frequently and we have delayed allocation blocks
1745 * outstanding (e.g. streaming writes from the NFS server),
1746 * truncating the blocks past EOF will cause fragmentation to
1749 * In this case don't do the truncation, either, but we have to
1750 * be careful how we detect this case. Blocks beyond EOF show
1751 * up as i_delayed_blks even when the inode is clean, so we
1752 * need to truncate them away first before checking for a dirty
1753 * release. Hence on the first dirty close we will still remove
1754 * the speculative allocation, but after that we will leave it
1757 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1760 error
= xfs_free_eofblocks(mp
, ip
, true);
1761 if (error
&& error
!= EAGAIN
)
1764 /* delalloc blocks after truncation means it really is dirty */
1765 if (ip
->i_delayed_blks
)
1766 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1772 * xfs_inactive_truncate
1774 * Called to perform a truncate when an inode becomes unlinked.
1777 xfs_inactive_truncate(
1778 struct xfs_inode
*ip
)
1780 struct xfs_mount
*mp
= ip
->i_mount
;
1781 struct xfs_trans
*tp
;
1784 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1785 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1787 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1788 xfs_trans_cancel(tp
, 0);
1792 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1793 xfs_trans_ijoin(tp
, ip
, 0);
1796 * Log the inode size first to prevent stale data exposure in the event
1797 * of a system crash before the truncate completes. See the related
1798 * comment in xfs_setattr_size() for details.
1800 ip
->i_d
.di_size
= 0;
1801 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1803 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1805 goto error_trans_cancel
;
1807 ASSERT(ip
->i_d
.di_nextents
== 0);
1809 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1813 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1817 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1819 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1824 * xfs_inactive_ifree()
1826 * Perform the inode free when an inode is unlinked.
1830 struct xfs_inode
*ip
)
1832 xfs_bmap_free_t free_list
;
1833 xfs_fsblock_t first_block
;
1835 struct xfs_mount
*mp
= ip
->i_mount
;
1836 struct xfs_trans
*tp
;
1839 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1840 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ifree
, 0, 0);
1842 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1843 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1847 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1848 xfs_trans_ijoin(tp
, ip
, 0);
1850 xfs_bmap_init(&free_list
, &first_block
);
1851 error
= xfs_ifree(tp
, ip
, &free_list
);
1854 * If we fail to free the inode, shut down. The cancel
1855 * might do that, we need to make sure. Otherwise the
1856 * inode might be lost for a long time or forever.
1858 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1859 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1861 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1863 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1864 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1869 * Credit the quota account(s). The inode is gone.
1871 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1874 * Just ignore errors at this point. There is nothing we can
1875 * do except to try to keep going. Make sure it's not a silent
1878 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1880 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1882 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1884 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1887 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1894 * This is called when the vnode reference count for the vnode
1895 * goes to zero. If the file has been unlinked, then it must
1896 * now be truncated. Also, we clear all of the read-ahead state
1897 * kept for the inode here since the file is now closed.
1903 struct xfs_mount
*mp
;
1908 * If the inode is already free, then there can be nothing
1911 if (ip
->i_d
.di_mode
== 0) {
1912 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1913 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1919 /* If this is a read-only mount, don't do this (would generate I/O) */
1920 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1923 if (ip
->i_d
.di_nlink
!= 0) {
1925 * force is true because we are evicting an inode from the
1926 * cache. Post-eof blocks must be freed, lest we end up with
1927 * broken free space accounting.
1929 if (xfs_can_free_eofblocks(ip
, true))
1930 xfs_free_eofblocks(mp
, ip
, false);
1935 if (S_ISREG(ip
->i_d
.di_mode
) &&
1936 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1937 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1940 error
= xfs_qm_dqattach(ip
, 0);
1944 if (S_ISLNK(ip
->i_d
.di_mode
))
1945 error
= xfs_inactive_symlink(ip
);
1947 error
= xfs_inactive_truncate(ip
);
1952 * If there are attributes associated with the file then blow them away
1953 * now. The code calls a routine that recursively deconstructs the
1954 * attribute fork. We need to just commit the current transaction
1955 * because we can't use it for xfs_attr_inactive().
1957 if (ip
->i_d
.di_anextents
> 0) {
1958 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1960 error
= xfs_attr_inactive(ip
);
1966 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1968 ASSERT(ip
->i_d
.di_anextents
== 0);
1973 error
= xfs_inactive_ifree(ip
);
1978 * Release the dquots held by inode, if any.
1980 xfs_qm_dqdetach(ip
);
1984 * This is called when the inode's link count goes to 0.
1985 * We place the on-disk inode on a list in the AGI. It
1986 * will be pulled from this list when the inode is freed.
2003 ASSERT(ip
->i_d
.di_nlink
== 0);
2004 ASSERT(ip
->i_d
.di_mode
!= 0);
2009 * Get the agi buffer first. It ensures lock ordering
2012 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
2015 agi
= XFS_BUF_TO_AGI(agibp
);
2018 * Get the index into the agi hash table for the
2019 * list this inode will go on.
2021 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2023 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2024 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2025 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
2027 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
2029 * There is already another inode in the bucket we need
2030 * to add ourselves to. Add us at the front of the list.
2031 * Here we put the head pointer into our next pointer,
2032 * and then we fall through to point the head at us.
2034 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2039 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
2040 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
2041 offset
= ip
->i_imap
.im_boffset
+
2042 offsetof(xfs_dinode_t
, di_next_unlinked
);
2044 /* need to recalc the inode CRC if appropriate */
2045 xfs_dinode_calc_crc(mp
, dip
);
2047 xfs_trans_inode_buf(tp
, ibp
);
2048 xfs_trans_log_buf(tp
, ibp
, offset
,
2049 (offset
+ sizeof(xfs_agino_t
) - 1));
2050 xfs_inobp_check(mp
, ibp
);
2054 * Point the bucket head pointer at the inode being inserted.
2057 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
2058 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2059 (sizeof(xfs_agino_t
) * bucket_index
);
2060 xfs_trans_log_buf(tp
, agibp
, offset
,
2061 (offset
+ sizeof(xfs_agino_t
) - 1));
2066 * Pull the on-disk inode from the AGI unlinked list.
2079 xfs_agnumber_t agno
;
2081 xfs_agino_t next_agino
;
2082 xfs_buf_t
*last_ibp
;
2083 xfs_dinode_t
*last_dip
= NULL
;
2085 int offset
, last_offset
= 0;
2089 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2092 * Get the agi buffer first. It ensures lock ordering
2095 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2099 agi
= XFS_BUF_TO_AGI(agibp
);
2102 * Get the index into the agi hash table for the
2103 * list this inode will go on.
2105 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2107 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2108 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
2109 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2111 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2113 * We're at the head of the list. Get the inode's on-disk
2114 * buffer to see if there is anyone after us on the list.
2115 * Only modify our next pointer if it is not already NULLAGINO.
2116 * This saves us the overhead of dealing with the buffer when
2117 * there is no need to change it.
2119 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2122 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2126 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2127 ASSERT(next_agino
!= 0);
2128 if (next_agino
!= NULLAGINO
) {
2129 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2130 offset
= ip
->i_imap
.im_boffset
+
2131 offsetof(xfs_dinode_t
, di_next_unlinked
);
2133 /* need to recalc the inode CRC if appropriate */
2134 xfs_dinode_calc_crc(mp
, dip
);
2136 xfs_trans_inode_buf(tp
, ibp
);
2137 xfs_trans_log_buf(tp
, ibp
, offset
,
2138 (offset
+ sizeof(xfs_agino_t
) - 1));
2139 xfs_inobp_check(mp
, ibp
);
2141 xfs_trans_brelse(tp
, ibp
);
2144 * Point the bucket head pointer at the next inode.
2146 ASSERT(next_agino
!= 0);
2147 ASSERT(next_agino
!= agino
);
2148 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2149 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2150 (sizeof(xfs_agino_t
) * bucket_index
);
2151 xfs_trans_log_buf(tp
, agibp
, offset
,
2152 (offset
+ sizeof(xfs_agino_t
) - 1));
2155 * We need to search the list for the inode being freed.
2157 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2159 while (next_agino
!= agino
) {
2160 struct xfs_imap imap
;
2163 xfs_trans_brelse(tp
, last_ibp
);
2166 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2168 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2171 "%s: xfs_imap returned error %d.",
2176 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2180 "%s: xfs_imap_to_bp returned error %d.",
2185 last_offset
= imap
.im_boffset
;
2186 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2187 ASSERT(next_agino
!= NULLAGINO
);
2188 ASSERT(next_agino
!= 0);
2192 * Now last_ibp points to the buffer previous to us on the
2193 * unlinked list. Pull us from the list.
2195 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2198 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2202 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2203 ASSERT(next_agino
!= 0);
2204 ASSERT(next_agino
!= agino
);
2205 if (next_agino
!= NULLAGINO
) {
2206 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2207 offset
= ip
->i_imap
.im_boffset
+
2208 offsetof(xfs_dinode_t
, di_next_unlinked
);
2210 /* need to recalc the inode CRC if appropriate */
2211 xfs_dinode_calc_crc(mp
, dip
);
2213 xfs_trans_inode_buf(tp
, ibp
);
2214 xfs_trans_log_buf(tp
, ibp
, offset
,
2215 (offset
+ sizeof(xfs_agino_t
) - 1));
2216 xfs_inobp_check(mp
, ibp
);
2218 xfs_trans_brelse(tp
, ibp
);
2221 * Point the previous inode on the list to the next inode.
2223 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2224 ASSERT(next_agino
!= 0);
2225 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2227 /* need to recalc the inode CRC if appropriate */
2228 xfs_dinode_calc_crc(mp
, last_dip
);
2230 xfs_trans_inode_buf(tp
, last_ibp
);
2231 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2232 (offset
+ sizeof(xfs_agino_t
) - 1));
2233 xfs_inobp_check(mp
, last_ibp
);
2239 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2240 * inodes that are in memory - they all must be marked stale and attached to
2241 * the cluster buffer.
2245 xfs_inode_t
*free_ip
,
2249 xfs_mount_t
*mp
= free_ip
->i_mount
;
2250 int blks_per_cluster
;
2251 int inodes_per_cluster
;
2257 xfs_inode_log_item_t
*iip
;
2258 xfs_log_item_t
*lip
;
2259 struct xfs_perag
*pag
;
2261 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2262 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2263 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
2264 nbufs
= mp
->m_ialloc_blks
/ blks_per_cluster
;
2266 for (j
= 0; j
< nbufs
; j
++, inum
+= inodes_per_cluster
) {
2267 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2268 XFS_INO_TO_AGBNO(mp
, inum
));
2271 * We obtain and lock the backing buffer first in the process
2272 * here, as we have to ensure that any dirty inode that we
2273 * can't get the flush lock on is attached to the buffer.
2274 * If we scan the in-memory inodes first, then buffer IO can
2275 * complete before we get a lock on it, and hence we may fail
2276 * to mark all the active inodes on the buffer stale.
2278 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2279 mp
->m_bsize
* blks_per_cluster
,
2286 * This buffer may not have been correctly initialised as we
2287 * didn't read it from disk. That's not important because we are
2288 * only using to mark the buffer as stale in the log, and to
2289 * attach stale cached inodes on it. That means it will never be
2290 * dispatched for IO. If it is, we want to know about it, and we
2291 * want it to fail. We can acheive this by adding a write
2292 * verifier to the buffer.
2294 bp
->b_ops
= &xfs_inode_buf_ops
;
2297 * Walk the inodes already attached to the buffer and mark them
2298 * stale. These will all have the flush locks held, so an
2299 * in-memory inode walk can't lock them. By marking them all
2300 * stale first, we will not attempt to lock them in the loop
2301 * below as the XFS_ISTALE flag will be set.
2305 if (lip
->li_type
== XFS_LI_INODE
) {
2306 iip
= (xfs_inode_log_item_t
*)lip
;
2307 ASSERT(iip
->ili_logged
== 1);
2308 lip
->li_cb
= xfs_istale_done
;
2309 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2310 &iip
->ili_flush_lsn
,
2311 &iip
->ili_item
.li_lsn
);
2312 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2314 lip
= lip
->li_bio_list
;
2319 * For each inode in memory attempt to add it to the inode
2320 * buffer and set it up for being staled on buffer IO
2321 * completion. This is safe as we've locked out tail pushing
2322 * and flushing by locking the buffer.
2324 * We have already marked every inode that was part of a
2325 * transaction stale above, which means there is no point in
2326 * even trying to lock them.
2328 for (i
= 0; i
< inodes_per_cluster
; i
++) {
2331 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2332 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2334 /* Inode not in memory, nothing to do */
2341 * because this is an RCU protected lookup, we could
2342 * find a recently freed or even reallocated inode
2343 * during the lookup. We need to check under the
2344 * i_flags_lock for a valid inode here. Skip it if it
2345 * is not valid, the wrong inode or stale.
2347 spin_lock(&ip
->i_flags_lock
);
2348 if (ip
->i_ino
!= inum
+ i
||
2349 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2350 spin_unlock(&ip
->i_flags_lock
);
2354 spin_unlock(&ip
->i_flags_lock
);
2357 * Don't try to lock/unlock the current inode, but we
2358 * _cannot_ skip the other inodes that we did not find
2359 * in the list attached to the buffer and are not
2360 * already marked stale. If we can't lock it, back off
2363 if (ip
!= free_ip
&&
2364 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2372 xfs_iflags_set(ip
, XFS_ISTALE
);
2375 * we don't need to attach clean inodes or those only
2376 * with unlogged changes (which we throw away, anyway).
2379 if (!iip
|| xfs_inode_clean(ip
)) {
2380 ASSERT(ip
!= free_ip
);
2382 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2386 iip
->ili_last_fields
= iip
->ili_fields
;
2387 iip
->ili_fields
= 0;
2388 iip
->ili_logged
= 1;
2389 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2390 &iip
->ili_item
.li_lsn
);
2392 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2396 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2399 xfs_trans_stale_inode_buf(tp
, bp
);
2400 xfs_trans_binval(tp
, bp
);
2408 * This is called to return an inode to the inode free list.
2409 * The inode should already be truncated to 0 length and have
2410 * no pages associated with it. This routine also assumes that
2411 * the inode is already a part of the transaction.
2413 * The on-disk copy of the inode will have been added to the list
2414 * of unlinked inodes in the AGI. We need to remove the inode from
2415 * that list atomically with respect to freeing it here.
2421 xfs_bmap_free_t
*flist
)
2425 xfs_ino_t first_ino
;
2427 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2428 ASSERT(ip
->i_d
.di_nlink
== 0);
2429 ASSERT(ip
->i_d
.di_nextents
== 0);
2430 ASSERT(ip
->i_d
.di_anextents
== 0);
2431 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
2432 ASSERT(ip
->i_d
.di_nblocks
== 0);
2435 * Pull the on-disk inode from the AGI unlinked list.
2437 error
= xfs_iunlink_remove(tp
, ip
);
2441 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2445 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2446 ip
->i_d
.di_flags
= 0;
2447 ip
->i_d
.di_dmevmask
= 0;
2448 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2449 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2450 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2452 * Bump the generation count so no one will be confused
2453 * by reincarnations of this inode.
2456 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2459 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
2465 * This is called to unpin an inode. The caller must have the inode locked
2466 * in at least shared mode so that the buffer cannot be subsequently pinned
2467 * once someone is waiting for it to be unpinned.
2471 struct xfs_inode
*ip
)
2473 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2475 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2477 /* Give the log a push to start the unpinning I/O */
2478 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2484 struct xfs_inode
*ip
)
2486 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2487 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2492 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2493 if (xfs_ipincount(ip
))
2495 } while (xfs_ipincount(ip
));
2496 finish_wait(wq
, &wait
.wait
);
2501 struct xfs_inode
*ip
)
2503 if (xfs_ipincount(ip
))
2504 __xfs_iunpin_wait(ip
);
2508 * Removing an inode from the namespace involves removing the directory entry
2509 * and dropping the link count on the inode. Removing the directory entry can
2510 * result in locking an AGF (directory blocks were freed) and removing a link
2511 * count can result in placing the inode on an unlinked list which results in
2514 * The big problem here is that we have an ordering constraint on AGF and AGI
2515 * locking - inode allocation locks the AGI, then can allocate a new extent for
2516 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2517 * removes the inode from the unlinked list, requiring that we lock the AGI
2518 * first, and then freeing the inode can result in an inode chunk being freed
2519 * and hence freeing disk space requiring that we lock an AGF.
2521 * Hence the ordering that is imposed by other parts of the code is AGI before
2522 * AGF. This means we cannot remove the directory entry before we drop the inode
2523 * reference count and put it on the unlinked list as this results in a lock
2524 * order of AGF then AGI, and this can deadlock against inode allocation and
2525 * freeing. Therefore we must drop the link counts before we remove the
2528 * This is still safe from a transactional point of view - it is not until we
2529 * get to xfs_bmap_finish() that we have the possibility of multiple
2530 * transactions in this operation. Hence as long as we remove the directory
2531 * entry and drop the link count in the first transaction of the remove
2532 * operation, there are no transactional constraints on the ordering here.
2537 struct xfs_name
*name
,
2540 xfs_mount_t
*mp
= dp
->i_mount
;
2541 xfs_trans_t
*tp
= NULL
;
2542 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
2544 xfs_bmap_free_t free_list
;
2545 xfs_fsblock_t first_block
;
2552 trace_xfs_remove(dp
, name
);
2554 if (XFS_FORCED_SHUTDOWN(mp
))
2555 return XFS_ERROR(EIO
);
2557 error
= xfs_qm_dqattach(dp
, 0);
2561 error
= xfs_qm_dqattach(ip
, 0);
2566 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2567 log_count
= XFS_DEFAULT_LOG_COUNT
;
2569 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2570 log_count
= XFS_REMOVE_LOG_COUNT
;
2572 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2575 * We try to get the real space reservation first,
2576 * allowing for directory btree deletion(s) implying
2577 * possible bmap insert(s). If we can't get the space
2578 * reservation then we use 0 instead, and avoid the bmap
2579 * btree insert(s) in the directory code by, if the bmap
2580 * insert tries to happen, instead trimming the LAST
2581 * block from the directory.
2583 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2584 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, resblks
, 0);
2585 if (error
== ENOSPC
) {
2587 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
2590 ASSERT(error
!= ENOSPC
);
2592 goto out_trans_cancel
;
2595 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2597 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2598 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2601 * If we're removing a directory perform some additional validation.
2603 cancel_flags
|= XFS_TRANS_ABORT
;
2605 ASSERT(ip
->i_d
.di_nlink
>= 2);
2606 if (ip
->i_d
.di_nlink
!= 2) {
2607 error
= XFS_ERROR(ENOTEMPTY
);
2608 goto out_trans_cancel
;
2610 if (!xfs_dir_isempty(ip
)) {
2611 error
= XFS_ERROR(ENOTEMPTY
);
2612 goto out_trans_cancel
;
2615 /* Drop the link from ip's "..". */
2616 error
= xfs_droplink(tp
, dp
);
2618 goto out_trans_cancel
;
2620 /* Drop the "." link from ip to self. */
2621 error
= xfs_droplink(tp
, ip
);
2623 goto out_trans_cancel
;
2626 * When removing a non-directory we need to log the parent
2627 * inode here. For a directory this is done implicitly
2628 * by the xfs_droplink call for the ".." entry.
2630 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2632 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2634 /* Drop the link from dp to ip. */
2635 error
= xfs_droplink(tp
, ip
);
2637 goto out_trans_cancel
;
2639 /* Determine if this is the last link while the inode is locked */
2640 link_zero
= (ip
->i_d
.di_nlink
== 0);
2642 xfs_bmap_init(&free_list
, &first_block
);
2643 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2644 &first_block
, &free_list
, resblks
);
2646 ASSERT(error
!= ENOENT
);
2647 goto out_bmap_cancel
;
2651 * If this is a synchronous mount, make sure that the
2652 * remove transaction goes to disk before returning to
2655 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2656 xfs_trans_set_sync(tp
);
2658 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2660 goto out_bmap_cancel
;
2662 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2667 * If we are using filestreams, kill the stream association.
2668 * If the file is still open it may get a new one but that
2669 * will get killed on last close in xfs_close() so we don't
2670 * have to worry about that.
2672 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
2673 xfs_filestream_deassociate(ip
);
2678 xfs_bmap_cancel(&free_list
);
2680 xfs_trans_cancel(tp
, cancel_flags
);
2686 * Enter all inodes for a rename transaction into a sorted array.
2689 xfs_sort_for_rename(
2690 xfs_inode_t
*dp1
, /* in: old (source) directory inode */
2691 xfs_inode_t
*dp2
, /* in: new (target) directory inode */
2692 xfs_inode_t
*ip1
, /* in: inode of old entry */
2693 xfs_inode_t
*ip2
, /* in: inode of new entry, if it
2694 already exists, NULL otherwise. */
2695 xfs_inode_t
**i_tab
,/* out: array of inode returned, sorted */
2696 int *num_inodes
) /* out: number of inodes in array */
2702 * i_tab contains a list of pointers to inodes. We initialize
2703 * the table here & we'll sort it. We will then use it to
2704 * order the acquisition of the inode locks.
2706 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2720 * Sort the elements via bubble sort. (Remember, there are at
2721 * most 4 elements to sort, so this is adequate.)
2723 for (i
= 0; i
< *num_inodes
; i
++) {
2724 for (j
= 1; j
< *num_inodes
; j
++) {
2725 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2727 i_tab
[j
] = i_tab
[j
-1];
2739 xfs_inode_t
*src_dp
,
2740 struct xfs_name
*src_name
,
2741 xfs_inode_t
*src_ip
,
2742 xfs_inode_t
*target_dp
,
2743 struct xfs_name
*target_name
,
2744 xfs_inode_t
*target_ip
)
2746 xfs_trans_t
*tp
= NULL
;
2747 xfs_mount_t
*mp
= src_dp
->i_mount
;
2748 int new_parent
; /* moving to a new dir */
2749 int src_is_directory
; /* src_name is a directory */
2751 xfs_bmap_free_t free_list
;
2752 xfs_fsblock_t first_block
;
2755 xfs_inode_t
*inodes
[4];
2759 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2761 new_parent
= (src_dp
!= target_dp
);
2762 src_is_directory
= S_ISDIR(src_ip
->i_d
.di_mode
);
2764 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
,
2765 inodes
, &num_inodes
);
2767 xfs_bmap_init(&free_list
, &first_block
);
2768 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RENAME
);
2769 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2770 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2771 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, spaceres
, 0);
2772 if (error
== ENOSPC
) {
2774 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, 0, 0);
2777 xfs_trans_cancel(tp
, 0);
2782 * Attach the dquots to the inodes
2784 error
= xfs_qm_vop_rename_dqattach(inodes
);
2786 xfs_trans_cancel(tp
, cancel_flags
);
2791 * Lock all the participating inodes. Depending upon whether
2792 * the target_name exists in the target directory, and
2793 * whether the target directory is the same as the source
2794 * directory, we can lock from 2 to 4 inodes.
2796 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2799 * Join all the inodes to the transaction. From this point on,
2800 * we can rely on either trans_commit or trans_cancel to unlock
2803 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2805 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2806 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2808 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2811 * If we are using project inheritance, we only allow renames
2812 * into our tree when the project IDs are the same; else the
2813 * tree quota mechanism would be circumvented.
2815 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2816 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2817 error
= XFS_ERROR(EXDEV
);
2822 * Set up the target.
2824 if (target_ip
== NULL
) {
2826 * If there's no space reservation, check the entry will
2827 * fit before actually inserting it.
2829 error
= xfs_dir_canenter(tp
, target_dp
, target_name
, spaceres
);
2833 * If target does not exist and the rename crosses
2834 * directories, adjust the target directory link count
2835 * to account for the ".." reference from the new entry.
2837 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
2838 src_ip
->i_ino
, &first_block
,
2839 &free_list
, spaceres
);
2840 if (error
== ENOSPC
)
2845 xfs_trans_ichgtime(tp
, target_dp
,
2846 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2848 if (new_parent
&& src_is_directory
) {
2849 error
= xfs_bumplink(tp
, target_dp
);
2853 } else { /* target_ip != NULL */
2855 * If target exists and it's a directory, check that both
2856 * target and source are directories and that target can be
2857 * destroyed, or that neither is a directory.
2859 if (S_ISDIR(target_ip
->i_d
.di_mode
)) {
2861 * Make sure target dir is empty.
2863 if (!(xfs_dir_isempty(target_ip
)) ||
2864 (target_ip
->i_d
.di_nlink
> 2)) {
2865 error
= XFS_ERROR(EEXIST
);
2871 * Link the source inode under the target name.
2872 * If the source inode is a directory and we are moving
2873 * it across directories, its ".." entry will be
2874 * inconsistent until we replace that down below.
2876 * In case there is already an entry with the same
2877 * name at the destination directory, remove it first.
2879 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
2881 &first_block
, &free_list
, spaceres
);
2885 xfs_trans_ichgtime(tp
, target_dp
,
2886 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2889 * Decrement the link count on the target since the target
2890 * dir no longer points to it.
2892 error
= xfs_droplink(tp
, target_ip
);
2896 if (src_is_directory
) {
2898 * Drop the link from the old "." entry.
2900 error
= xfs_droplink(tp
, target_ip
);
2904 } /* target_ip != NULL */
2907 * Remove the source.
2909 if (new_parent
&& src_is_directory
) {
2911 * Rewrite the ".." entry to point to the new
2914 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
2916 &first_block
, &free_list
, spaceres
);
2917 ASSERT(error
!= EEXIST
);
2923 * We always want to hit the ctime on the source inode.
2925 * This isn't strictly required by the standards since the source
2926 * inode isn't really being changed, but old unix file systems did
2927 * it and some incremental backup programs won't work without it.
2929 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
2930 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
2933 * Adjust the link count on src_dp. This is necessary when
2934 * renaming a directory, either within one parent when
2935 * the target existed, or across two parent directories.
2937 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
2940 * Decrement link count on src_directory since the
2941 * entry that's moved no longer points to it.
2943 error
= xfs_droplink(tp
, src_dp
);
2948 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
2949 &first_block
, &free_list
, spaceres
);
2953 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2954 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
2956 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
2959 * If this is a synchronous mount, make sure that the
2960 * rename transaction goes to disk before returning to
2963 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2964 xfs_trans_set_sync(tp
);
2967 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2969 xfs_bmap_cancel(&free_list
);
2970 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
2976 * trans_commit will unlock src_ip, target_ip & decrement
2977 * the vnode references.
2979 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2982 cancel_flags
|= XFS_TRANS_ABORT
;
2984 xfs_bmap_cancel(&free_list
);
2985 xfs_trans_cancel(tp
, cancel_flags
);
2995 xfs_mount_t
*mp
= ip
->i_mount
;
2996 struct xfs_perag
*pag
;
2997 unsigned long first_index
, mask
;
2998 unsigned long inodes_per_cluster
;
3000 xfs_inode_t
**ilist
;
3007 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
3009 inodes_per_cluster
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
3010 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
3011 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
3015 mask
= ~(((mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
)) - 1);
3016 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
3018 /* really need a gang lookup range call here */
3019 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
3020 first_index
, inodes_per_cluster
);
3024 for (i
= 0; i
< nr_found
; i
++) {
3030 * because this is an RCU protected lookup, we could find a
3031 * recently freed or even reallocated inode during the lookup.
3032 * We need to check under the i_flags_lock for a valid inode
3033 * here. Skip it if it is not valid or the wrong inode.
3035 spin_lock(&ip
->i_flags_lock
);
3037 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
3038 spin_unlock(&ip
->i_flags_lock
);
3041 spin_unlock(&ip
->i_flags_lock
);
3044 * Do an un-protected check to see if the inode is dirty and
3045 * is a candidate for flushing. These checks will be repeated
3046 * later after the appropriate locks are acquired.
3048 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
3052 * Try to get locks. If any are unavailable or it is pinned,
3053 * then this inode cannot be flushed and is skipped.
3056 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
3058 if (!xfs_iflock_nowait(iq
)) {
3059 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3062 if (xfs_ipincount(iq
)) {
3064 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3069 * arriving here means that this inode can be flushed. First
3070 * re-check that it's dirty before flushing.
3072 if (!xfs_inode_clean(iq
)) {
3074 error
= xfs_iflush_int(iq
, bp
);
3076 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3077 goto cluster_corrupt_out
;
3083 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3087 XFS_STATS_INC(xs_icluster_flushcnt
);
3088 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3099 cluster_corrupt_out
:
3101 * Corruption detected in the clustering loop. Invalidate the
3102 * inode buffer and shut down the filesystem.
3106 * Clean up the buffer. If it was delwri, just release it --
3107 * brelse can handle it with no problems. If not, shut down the
3108 * filesystem before releasing the buffer.
3110 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3114 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3116 if (!bufwasdelwri
) {
3118 * Just like incore_relse: if we have b_iodone functions,
3119 * mark the buffer as an error and call them. Otherwise
3120 * mark it as stale and brelse.
3125 xfs_buf_ioerror(bp
, EIO
);
3126 xfs_buf_ioend(bp
, 0);
3134 * Unlocks the flush lock
3136 xfs_iflush_abort(iq
, false);
3139 return XFS_ERROR(EFSCORRUPTED
);
3143 * Flush dirty inode metadata into the backing buffer.
3145 * The caller must have the inode lock and the inode flush lock held. The
3146 * inode lock will still be held upon return to the caller, and the inode
3147 * flush lock will be released after the inode has reached the disk.
3149 * The caller must write out the buffer returned in *bpp and release it.
3153 struct xfs_inode
*ip
,
3154 struct xfs_buf
**bpp
)
3156 struct xfs_mount
*mp
= ip
->i_mount
;
3158 struct xfs_dinode
*dip
;
3161 XFS_STATS_INC(xs_iflush_count
);
3163 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3164 ASSERT(xfs_isiflocked(ip
));
3165 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3166 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3170 xfs_iunpin_wait(ip
);
3173 * For stale inodes we cannot rely on the backing buffer remaining
3174 * stale in cache for the remaining life of the stale inode and so
3175 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3176 * inodes below. We have to check this after ensuring the inode is
3177 * unpinned so that it is safe to reclaim the stale inode after the
3180 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3186 * This may have been unpinned because the filesystem is shutting
3187 * down forcibly. If that's the case we must not write this inode
3188 * to disk, because the log record didn't make it to disk.
3190 * We also have to remove the log item from the AIL in this case,
3191 * as we wait for an empty AIL as part of the unmount process.
3193 if (XFS_FORCED_SHUTDOWN(mp
)) {
3194 error
= XFS_ERROR(EIO
);
3199 * Get the buffer containing the on-disk inode.
3201 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3209 * First flush out the inode that xfs_iflush was called with.
3211 error
= xfs_iflush_int(ip
, bp
);
3216 * If the buffer is pinned then push on the log now so we won't
3217 * get stuck waiting in the write for too long.
3219 if (xfs_buf_ispinned(bp
))
3220 xfs_log_force(mp
, 0);
3224 * see if other inodes can be gathered into this write
3226 error
= xfs_iflush_cluster(ip
, bp
);
3228 goto cluster_corrupt_out
;
3235 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3236 cluster_corrupt_out
:
3237 error
= XFS_ERROR(EFSCORRUPTED
);
3240 * Unlocks the flush lock
3242 xfs_iflush_abort(ip
, false);
3248 struct xfs_inode
*ip
,
3251 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3252 struct xfs_dinode
*dip
;
3253 struct xfs_mount
*mp
= ip
->i_mount
;
3255 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3256 ASSERT(xfs_isiflocked(ip
));
3257 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3258 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3259 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3261 /* set *dip = inode's place in the buffer */
3262 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3264 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3265 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3266 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3267 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3268 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3271 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3272 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3273 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3274 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3275 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3278 if (S_ISREG(ip
->i_d
.di_mode
)) {
3280 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3281 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3282 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3283 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3284 "%s: Bad regular inode %Lu, ptr 0x%p",
3285 __func__
, ip
->i_ino
, ip
);
3288 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
3290 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3291 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3292 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3293 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3294 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3295 "%s: Bad directory inode %Lu, ptr 0x%p",
3296 __func__
, ip
->i_ino
, ip
);
3300 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3301 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3302 XFS_RANDOM_IFLUSH_5
)) {
3303 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3304 "%s: detected corrupt incore inode %Lu, "
3305 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3306 __func__
, ip
->i_ino
,
3307 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3308 ip
->i_d
.di_nblocks
, ip
);
3311 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3312 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3313 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3314 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3315 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3320 * Inode item log recovery for v1/v2 inodes are dependent on the
3321 * di_flushiter count for correct sequencing. We bump the flush
3322 * iteration count so we can detect flushes which postdate a log record
3323 * during recovery. This is redundant as we now log every change and
3324 * hence this can't happen but we need to still do it to ensure
3325 * backwards compatibility with old kernels that predate logging all
3328 if (ip
->i_d
.di_version
< 3)
3329 ip
->i_d
.di_flushiter
++;
3332 * Copy the dirty parts of the inode into the on-disk
3333 * inode. We always copy out the core of the inode,
3334 * because if the inode is dirty at all the core must
3337 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3339 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3340 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3341 ip
->i_d
.di_flushiter
= 0;
3344 * If this is really an old format inode and the superblock version
3345 * has not been updated to support only new format inodes, then
3346 * convert back to the old inode format. If the superblock version
3347 * has been updated, then make the conversion permanent.
3349 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
3350 if (ip
->i_d
.di_version
== 1) {
3351 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3355 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3356 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3359 * The superblock version has already been bumped,
3360 * so just make the conversion to the new inode
3363 ip
->i_d
.di_version
= 2;
3364 dip
->di_version
= 2;
3365 ip
->i_d
.di_onlink
= 0;
3367 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3368 memset(&(dip
->di_pad
[0]), 0,
3369 sizeof(dip
->di_pad
));
3370 ASSERT(xfs_get_projid(ip
) == 0);
3374 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3375 if (XFS_IFORK_Q(ip
))
3376 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3377 xfs_inobp_check(mp
, bp
);
3380 * We've recorded everything logged in the inode, so we'd like to clear
3381 * the ili_fields bits so we don't log and flush things unnecessarily.
3382 * However, we can't stop logging all this information until the data
3383 * we've copied into the disk buffer is written to disk. If we did we
3384 * might overwrite the copy of the inode in the log with all the data
3385 * after re-logging only part of it, and in the face of a crash we
3386 * wouldn't have all the data we need to recover.
3388 * What we do is move the bits to the ili_last_fields field. When
3389 * logging the inode, these bits are moved back to the ili_fields field.
3390 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3391 * know that the information those bits represent is permanently on
3392 * disk. As long as the flush completes before the inode is logged
3393 * again, then both ili_fields and ili_last_fields will be cleared.
3395 * We can play with the ili_fields bits here, because the inode lock
3396 * must be held exclusively in order to set bits there and the flush
3397 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3398 * done routine can tell whether or not to look in the AIL. Also, store
3399 * the current LSN of the inode so that we can tell whether the item has
3400 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3401 * need the AIL lock, because it is a 64 bit value that cannot be read
3404 iip
->ili_last_fields
= iip
->ili_fields
;
3405 iip
->ili_fields
= 0;
3406 iip
->ili_logged
= 1;
3408 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3409 &iip
->ili_item
.li_lsn
);
3412 * Attach the function xfs_iflush_done to the inode's
3413 * buffer. This will remove the inode from the AIL
3414 * and unlock the inode's flush lock when the inode is
3415 * completely written to disk.
3417 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3419 /* update the lsn in the on disk inode if required */
3420 if (ip
->i_d
.di_version
== 3)
3421 dip
->di_lsn
= cpu_to_be64(iip
->ili_item
.li_lsn
);
3423 /* generate the checksum. */
3424 xfs_dinode_calc_crc(mp
, dip
);
3426 ASSERT(bp
->b_fspriv
!= NULL
);
3427 ASSERT(bp
->b_iodone
!= NULL
);
3431 return XFS_ERROR(EFSCORRUPTED
);