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"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_attr_sf.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
53 #include "xfs_dir2_priv.h"
55 kmem_zone_t
*xfs_inode_zone
;
58 * Used in xfs_itruncate_extents(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC
int xfs_iflush_int(struct xfs_inode
*, struct xfs_buf
*);
64 STATIC
int xfs_iunlink(struct xfs_trans
*, struct xfs_inode
*);
65 STATIC
int xfs_iunlink_remove(struct xfs_trans
*, struct xfs_inode
*);
68 * helper function to extract extent size hint from inode
74 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
75 return ip
->i_d
.di_extsize
;
76 if (XFS_IS_REALTIME_INODE(ip
))
77 return ip
->i_mount
->m_sb
.sb_rextsize
;
82 * Helper function to extract CoW extent size hint from inode.
83 * Between the extent size hint and the CoW extent size hint, we
84 * return the greater of the two. If the value is zero (automatic),
85 * use the default size.
88 xfs_get_cowextsz_hint(
94 if (ip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
95 a
= ip
->i_d
.di_cowextsize
;
96 b
= xfs_get_extsz_hint(ip
);
100 return XFS_DEFAULT_COWEXTSZ_HINT
;
105 * These two are wrapper routines around the xfs_ilock() routine used to
106 * centralize some grungy code. They are used in places that wish to lock the
107 * inode solely for reading the extents. The reason these places can't just
108 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
109 * bringing in of the extents from disk for a file in b-tree format. If the
110 * inode is in b-tree format, then we need to lock the inode exclusively until
111 * the extents are read in. Locking it exclusively all the time would limit
112 * our parallelism unnecessarily, though. What we do instead is check to see
113 * if the extents have been read in yet, and only lock the inode exclusively
116 * The functions return a value which should be given to the corresponding
117 * xfs_iunlock() call.
120 xfs_ilock_data_map_shared(
121 struct xfs_inode
*ip
)
123 uint lock_mode
= XFS_ILOCK_SHARED
;
125 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
&&
126 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)
127 lock_mode
= XFS_ILOCK_EXCL
;
128 xfs_ilock(ip
, lock_mode
);
133 xfs_ilock_attr_map_shared(
134 struct xfs_inode
*ip
)
136 uint lock_mode
= XFS_ILOCK_SHARED
;
138 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_BTREE
&&
139 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
140 lock_mode
= XFS_ILOCK_EXCL
;
141 xfs_ilock(ip
, lock_mode
);
146 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
147 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
148 * various combinations of the locks to be obtained.
150 * The 3 locks should always be ordered so that the IO lock is obtained first,
151 * the mmap lock second and the ilock last in order to prevent deadlock.
153 * Basic locking order:
155 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
157 * mmap_sem locking order:
159 * i_rwsem -> page lock -> mmap_sem
160 * mmap_sem -> i_mmap_lock -> page_lock
162 * The difference in mmap_sem locking order mean that we cannot hold the
163 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
164 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
165 * in get_user_pages() to map the user pages into the kernel address space for
166 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
167 * page faults already hold the mmap_sem.
169 * Hence to serialise fully against both syscall and mmap based IO, we need to
170 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
171 * taken in places where we need to invalidate the page cache in a race
172 * free manner (e.g. truncate, hole punch and other extent manipulation
180 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
183 * You can't set both SHARED and EXCL for the same lock,
184 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
185 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
187 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
188 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
189 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
190 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
191 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
192 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
193 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
195 if (lock_flags
& XFS_IOLOCK_EXCL
) {
196 down_write_nested(&VFS_I(ip
)->i_rwsem
,
197 XFS_IOLOCK_DEP(lock_flags
));
198 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
199 down_read_nested(&VFS_I(ip
)->i_rwsem
,
200 XFS_IOLOCK_DEP(lock_flags
));
203 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
204 mrupdate_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
205 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
206 mraccess_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
208 if (lock_flags
& XFS_ILOCK_EXCL
)
209 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
210 else if (lock_flags
& XFS_ILOCK_SHARED
)
211 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
215 * This is just like xfs_ilock(), except that the caller
216 * is guaranteed not to sleep. It returns 1 if it gets
217 * the requested locks and 0 otherwise. If the IO lock is
218 * obtained but the inode lock cannot be, then the IO lock
219 * is dropped before returning.
221 * ip -- the inode being locked
222 * lock_flags -- this parameter indicates the inode's locks to be
223 * to be locked. See the comment for xfs_ilock() for a list
231 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
234 * You can't set both SHARED and EXCL for the same lock,
235 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
236 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
238 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
239 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
240 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
241 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
242 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
243 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
244 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
246 if (lock_flags
& XFS_IOLOCK_EXCL
) {
247 if (!down_write_trylock(&VFS_I(ip
)->i_rwsem
))
249 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
250 if (!down_read_trylock(&VFS_I(ip
)->i_rwsem
))
254 if (lock_flags
& XFS_MMAPLOCK_EXCL
) {
255 if (!mrtryupdate(&ip
->i_mmaplock
))
256 goto out_undo_iolock
;
257 } else if (lock_flags
& XFS_MMAPLOCK_SHARED
) {
258 if (!mrtryaccess(&ip
->i_mmaplock
))
259 goto out_undo_iolock
;
262 if (lock_flags
& XFS_ILOCK_EXCL
) {
263 if (!mrtryupdate(&ip
->i_lock
))
264 goto out_undo_mmaplock
;
265 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
266 if (!mrtryaccess(&ip
->i_lock
))
267 goto out_undo_mmaplock
;
272 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
273 mrunlock_excl(&ip
->i_mmaplock
);
274 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
275 mrunlock_shared(&ip
->i_mmaplock
);
277 if (lock_flags
& XFS_IOLOCK_EXCL
)
278 up_write(&VFS_I(ip
)->i_rwsem
);
279 else if (lock_flags
& XFS_IOLOCK_SHARED
)
280 up_read(&VFS_I(ip
)->i_rwsem
);
286 * xfs_iunlock() is used to drop the inode locks acquired with
287 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
288 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
289 * that we know which locks to drop.
291 * ip -- the inode being unlocked
292 * lock_flags -- this parameter indicates the inode's locks to be
293 * to be unlocked. See the comment for xfs_ilock() for a list
294 * of valid values for this parameter.
303 * You can't set both SHARED and EXCL for the same lock,
304 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
305 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
307 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
308 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
309 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
310 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
311 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
312 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
313 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
314 ASSERT(lock_flags
!= 0);
316 if (lock_flags
& XFS_IOLOCK_EXCL
)
317 up_write(&VFS_I(ip
)->i_rwsem
);
318 else if (lock_flags
& XFS_IOLOCK_SHARED
)
319 up_read(&VFS_I(ip
)->i_rwsem
);
321 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
322 mrunlock_excl(&ip
->i_mmaplock
);
323 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
324 mrunlock_shared(&ip
->i_mmaplock
);
326 if (lock_flags
& XFS_ILOCK_EXCL
)
327 mrunlock_excl(&ip
->i_lock
);
328 else if (lock_flags
& XFS_ILOCK_SHARED
)
329 mrunlock_shared(&ip
->i_lock
);
331 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
335 * give up write locks. the i/o lock cannot be held nested
336 * if it is being demoted.
343 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
));
345 ~(XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
347 if (lock_flags
& XFS_ILOCK_EXCL
)
348 mrdemote(&ip
->i_lock
);
349 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
350 mrdemote(&ip
->i_mmaplock
);
351 if (lock_flags
& XFS_IOLOCK_EXCL
)
352 downgrade_write(&VFS_I(ip
)->i_rwsem
);
354 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
357 #if defined(DEBUG) || defined(XFS_WARN)
363 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
364 if (!(lock_flags
& XFS_ILOCK_SHARED
))
365 return !!ip
->i_lock
.mr_writer
;
366 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
369 if (lock_flags
& (XFS_MMAPLOCK_EXCL
|XFS_MMAPLOCK_SHARED
)) {
370 if (!(lock_flags
& XFS_MMAPLOCK_SHARED
))
371 return !!ip
->i_mmaplock
.mr_writer
;
372 return rwsem_is_locked(&ip
->i_mmaplock
.mr_lock
);
375 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
376 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
377 return !debug_locks
||
378 lockdep_is_held_type(&VFS_I(ip
)->i_rwsem
, 0);
379 return rwsem_is_locked(&VFS_I(ip
)->i_rwsem
);
389 int xfs_small_retries
;
390 int xfs_middle_retries
;
391 int xfs_lots_retries
;
396 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
397 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
398 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
399 * errors and warnings.
401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
403 xfs_lockdep_subclass_ok(
406 return subclass
< MAX_LOCKDEP_SUBCLASSES
;
409 #define xfs_lockdep_subclass_ok(subclass) (true)
413 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
414 * value. This can be called for any type of inode lock combination, including
415 * parent locking. Care must be taken to ensure we don't overrun the subclass
416 * storage fields in the class mask we build.
419 xfs_lock_inumorder(int lock_mode
, int subclass
)
423 ASSERT(!(lock_mode
& (XFS_ILOCK_PARENT
| XFS_ILOCK_RTBITMAP
|
425 ASSERT(xfs_lockdep_subclass_ok(subclass
));
427 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
428 ASSERT(subclass
<= XFS_IOLOCK_MAX_SUBCLASS
);
429 class += subclass
<< XFS_IOLOCK_SHIFT
;
432 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) {
433 ASSERT(subclass
<= XFS_MMAPLOCK_MAX_SUBCLASS
);
434 class += subclass
<< XFS_MMAPLOCK_SHIFT
;
437 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) {
438 ASSERT(subclass
<= XFS_ILOCK_MAX_SUBCLASS
);
439 class += subclass
<< XFS_ILOCK_SHIFT
;
442 return (lock_mode
& ~XFS_LOCK_SUBCLASS_MASK
) | class;
446 * The following routine will lock n inodes in exclusive mode. We assume the
447 * caller calls us with the inodes in i_ino order.
449 * We need to detect deadlock where an inode that we lock is in the AIL and we
450 * start waiting for another inode that is locked by a thread in a long running
451 * transaction (such as truncate). This can result in deadlock since the long
452 * running trans might need to wait for the inode we just locked in order to
453 * push the tail and free space in the log.
455 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457 * lock more than one at a time, lockdep will report false positives saying we
458 * have violated locking orders.
466 int attempts
= 0, i
, j
, try_lock
;
470 * Currently supports between 2 and 5 inodes with exclusive locking. We
471 * support an arbitrary depth of locking here, but absolute limits on
472 * inodes depend on the the type of locking and the limits placed by
473 * lockdep annotations in xfs_lock_inumorder. These are all checked by
476 ASSERT(ips
&& inodes
>= 2 && inodes
<= 5);
477 ASSERT(lock_mode
& (XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
|
479 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
| XFS_MMAPLOCK_SHARED
|
481 ASSERT(!(lock_mode
& XFS_MMAPLOCK_EXCL
) ||
482 inodes
<= XFS_MMAPLOCK_MAX_SUBCLASS
+ 1);
483 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
) ||
484 inodes
<= XFS_ILOCK_MAX_SUBCLASS
+ 1);
486 if (lock_mode
& XFS_IOLOCK_EXCL
) {
487 ASSERT(!(lock_mode
& (XFS_MMAPLOCK_EXCL
| XFS_ILOCK_EXCL
)));
488 } else if (lock_mode
& XFS_MMAPLOCK_EXCL
)
489 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
));
494 for (; i
< inodes
; i
++) {
497 if (i
&& (ips
[i
] == ips
[i
- 1])) /* Already locked */
501 * If try_lock is not set yet, make sure all locked inodes are
502 * not in the AIL. If any are, set try_lock to be used later.
505 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
506 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
507 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
))
513 * If any of the previous locks we have locked is in the AIL,
514 * we must TRY to get the second and subsequent locks. If
515 * we can't get any, we must release all we have
519 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
523 /* try_lock means we have an inode locked that is in the AIL. */
525 if (xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
)))
529 * Unlock all previous guys and try again. xfs_iunlock will try
530 * to push the tail if the inode is in the AIL.
533 for (j
= i
- 1; j
>= 0; j
--) {
535 * Check to see if we've already unlocked this one. Not
536 * the first one going back, and the inode ptr is the
539 if (j
!= (i
- 1) && ips
[j
] == ips
[j
+ 1])
542 xfs_iunlock(ips
[j
], lock_mode
);
545 if ((attempts
% 5) == 0) {
546 delay(1); /* Don't just spin the CPU */
558 if (attempts
< 5) xfs_small_retries
++;
559 else if (attempts
< 100) xfs_middle_retries
++;
560 else xfs_lots_retries
++;
568 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
569 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
570 * lock more than one at a time, lockdep will report false positives saying we
571 * have violated locking orders.
583 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
584 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
))
585 ASSERT(!(lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
587 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
589 if (ip0
->i_ino
> ip1
->i_ino
) {
596 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
599 * If the first lock we have locked is in the AIL, we must TRY to get
600 * the second lock. If we can't get it, we must release the first one
603 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
604 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
605 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
606 xfs_iunlock(ip0
, lock_mode
);
607 if ((++attempts
% 5) == 0)
608 delay(1); /* Don't just spin the CPU */
612 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
619 struct xfs_inode
*ip
)
621 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
622 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
625 prepare_to_wait_exclusive(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
626 if (xfs_isiflocked(ip
))
628 } while (!xfs_iflock_nowait(ip
));
630 finish_wait(wq
, &wait
.wq_entry
);
641 if (di_flags
& XFS_DIFLAG_ANY
) {
642 if (di_flags
& XFS_DIFLAG_REALTIME
)
643 flags
|= FS_XFLAG_REALTIME
;
644 if (di_flags
& XFS_DIFLAG_PREALLOC
)
645 flags
|= FS_XFLAG_PREALLOC
;
646 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
647 flags
|= FS_XFLAG_IMMUTABLE
;
648 if (di_flags
& XFS_DIFLAG_APPEND
)
649 flags
|= FS_XFLAG_APPEND
;
650 if (di_flags
& XFS_DIFLAG_SYNC
)
651 flags
|= FS_XFLAG_SYNC
;
652 if (di_flags
& XFS_DIFLAG_NOATIME
)
653 flags
|= FS_XFLAG_NOATIME
;
654 if (di_flags
& XFS_DIFLAG_NODUMP
)
655 flags
|= FS_XFLAG_NODUMP
;
656 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
657 flags
|= FS_XFLAG_RTINHERIT
;
658 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
659 flags
|= FS_XFLAG_PROJINHERIT
;
660 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
661 flags
|= FS_XFLAG_NOSYMLINKS
;
662 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
663 flags
|= FS_XFLAG_EXTSIZE
;
664 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
665 flags
|= FS_XFLAG_EXTSZINHERIT
;
666 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
667 flags
|= FS_XFLAG_NODEFRAG
;
668 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
669 flags
|= FS_XFLAG_FILESTREAM
;
672 if (di_flags2
& XFS_DIFLAG2_ANY
) {
673 if (di_flags2
& XFS_DIFLAG2_DAX
)
674 flags
|= FS_XFLAG_DAX
;
675 if (di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
676 flags
|= FS_XFLAG_COWEXTSIZE
;
680 flags
|= FS_XFLAG_HASATTR
;
687 struct xfs_inode
*ip
)
689 struct xfs_icdinode
*dic
= &ip
->i_d
;
691 return _xfs_dic2xflags(dic
->di_flags
, dic
->di_flags2
, XFS_IFORK_Q(ip
));
695 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
696 * is allowed, otherwise it has to be an exact match. If a CI match is found,
697 * ci_name->name will point to a the actual name (caller must free) or
698 * will be set to NULL if an exact match is found.
703 struct xfs_name
*name
,
705 struct xfs_name
*ci_name
)
710 trace_xfs_lookup(dp
, name
);
712 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
715 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
719 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
727 kmem_free(ci_name
->name
);
734 * Allocate an inode on disk and return a copy of its in-core version.
735 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
736 * appropriately within the inode. The uid and gid for the inode are
737 * set according to the contents of the given cred structure.
739 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
740 * has a free inode available, call xfs_iget() to obtain the in-core
741 * version of the allocated inode. Finally, fill in the inode and
742 * log its initial contents. In this case, ialloc_context would be
745 * If xfs_dialloc() does not have an available inode, it will replenish
746 * its supply by doing an allocation. Since we can only do one
747 * allocation within a transaction without deadlocks, we must commit
748 * the current transaction before returning the inode itself.
749 * In this case, therefore, we will set ialloc_context and return.
750 * The caller should then commit the current transaction, start a new
751 * transaction, and call xfs_ialloc() again to actually get the inode.
753 * To ensure that some other process does not grab the inode that
754 * was allocated during the first call to xfs_ialloc(), this routine
755 * also returns the [locked] bp pointing to the head of the freelist
756 * as ialloc_context. The caller should hold this buffer across
757 * the commit and pass it back into this routine on the second call.
759 * If we are allocating quota inodes, we do not have a parent inode
760 * to attach to or associate with (i.e. pip == NULL) because they
761 * are not linked into the directory structure - they are attached
762 * directly to the superblock - and so have no parent.
773 xfs_buf_t
**ialloc_context
,
776 struct xfs_mount
*mp
= tp
->t_mountp
;
785 * Call the space management code to pick
786 * the on-disk inode to be allocated.
788 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
789 ialloc_context
, &ino
);
792 if (*ialloc_context
|| ino
== NULLFSINO
) {
796 ASSERT(*ialloc_context
== NULL
);
799 * Get the in-core inode with the lock held exclusively.
800 * This is because we're setting fields here we need
801 * to prevent others from looking at until we're done.
803 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
804 XFS_ILOCK_EXCL
, &ip
);
811 * We always convert v1 inodes to v2 now - we only support filesystems
812 * with >= v2 inode capability, so there is no reason for ever leaving
813 * an inode in v1 format.
815 if (ip
->i_d
.di_version
== 1)
816 ip
->i_d
.di_version
= 2;
818 inode
->i_mode
= mode
;
819 set_nlink(inode
, nlink
);
820 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
821 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
822 xfs_set_projid(ip
, prid
);
824 if (pip
&& XFS_INHERIT_GID(pip
)) {
825 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
826 if ((VFS_I(pip
)->i_mode
& S_ISGID
) && S_ISDIR(mode
))
827 inode
->i_mode
|= S_ISGID
;
831 * If the group ID of the new file does not match the effective group
832 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
833 * (and only if the irix_sgid_inherit compatibility variable is set).
835 if ((irix_sgid_inherit
) &&
836 (inode
->i_mode
& S_ISGID
) &&
837 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
))))
838 inode
->i_mode
&= ~S_ISGID
;
841 ip
->i_d
.di_nextents
= 0;
842 ASSERT(ip
->i_d
.di_nblocks
== 0);
844 tv
= current_time(inode
);
849 ip
->i_d
.di_extsize
= 0;
850 ip
->i_d
.di_dmevmask
= 0;
851 ip
->i_d
.di_dmstate
= 0;
852 ip
->i_d
.di_flags
= 0;
854 if (ip
->i_d
.di_version
== 3) {
855 inode
->i_version
= 1;
856 ip
->i_d
.di_flags2
= 0;
857 ip
->i_d
.di_cowextsize
= 0;
858 ip
->i_d
.di_crtime
.t_sec
= (int32_t)tv
.tv_sec
;
859 ip
->i_d
.di_crtime
.t_nsec
= (int32_t)tv
.tv_nsec
;
863 flags
= XFS_ILOG_CORE
;
864 switch (mode
& S_IFMT
) {
869 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
870 ip
->i_df
.if_u2
.if_rdev
= rdev
;
871 ip
->i_df
.if_flags
= 0;
872 flags
|= XFS_ILOG_DEV
;
876 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
880 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
881 di_flags
|= XFS_DIFLAG_RTINHERIT
;
882 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
883 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
884 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
886 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
887 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
888 } else if (S_ISREG(mode
)) {
889 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
890 di_flags
|= XFS_DIFLAG_REALTIME
;
891 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
892 di_flags
|= XFS_DIFLAG_EXTSIZE
;
893 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
896 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
898 di_flags
|= XFS_DIFLAG_NOATIME
;
899 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
901 di_flags
|= XFS_DIFLAG_NODUMP
;
902 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
904 di_flags
|= XFS_DIFLAG_SYNC
;
905 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
906 xfs_inherit_nosymlinks
)
907 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
908 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
909 xfs_inherit_nodefrag
)
910 di_flags
|= XFS_DIFLAG_NODEFRAG
;
911 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
912 di_flags
|= XFS_DIFLAG_FILESTREAM
;
914 ip
->i_d
.di_flags
|= di_flags
;
917 (pip
->i_d
.di_flags2
& XFS_DIFLAG2_ANY
) &&
918 pip
->i_d
.di_version
== 3 &&
919 ip
->i_d
.di_version
== 3) {
920 uint64_t di_flags2
= 0;
922 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) {
923 di_flags2
|= XFS_DIFLAG2_COWEXTSIZE
;
924 ip
->i_d
.di_cowextsize
= pip
->i_d
.di_cowextsize
;
926 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
)
927 di_flags2
|= XFS_DIFLAG2_DAX
;
929 ip
->i_d
.di_flags2
|= di_flags2
;
933 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
934 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
935 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
936 ip
->i_df
.if_u1
.if_extents
= NULL
;
942 * Attribute fork settings for new inode.
944 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
945 ip
->i_d
.di_anextents
= 0;
948 * Log the new values stuffed into the inode.
950 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
951 xfs_trans_log_inode(tp
, ip
, flags
);
953 /* now that we have an i_mode we can setup the inode structure */
961 * Allocates a new inode from disk and return a pointer to the
962 * incore copy. This routine will internally commit the current
963 * transaction and allocate a new one if the Space Manager needed
964 * to do an allocation to replenish the inode free-list.
966 * This routine is designed to be called from xfs_create and
972 xfs_trans_t
**tpp
, /* input: current transaction;
973 output: may be a new transaction. */
974 xfs_inode_t
*dp
, /* directory within whose allocate
979 prid_t prid
, /* project id */
980 int okalloc
, /* ok to allocate new space */
981 xfs_inode_t
**ipp
, /* pointer to inode; it will be
988 xfs_buf_t
*ialloc_context
= NULL
;
994 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
997 * xfs_ialloc will return a pointer to an incore inode if
998 * the Space Manager has an available inode on the free
999 * list. Otherwise, it will do an allocation and replenish
1000 * the freelist. Since we can only do one allocation per
1001 * transaction without deadlocks, we will need to commit the
1002 * current transaction and start a new one. We will then
1003 * need to call xfs_ialloc again to get the inode.
1005 * If xfs_ialloc did an allocation to replenish the freelist,
1006 * it returns the bp containing the head of the freelist as
1007 * ialloc_context. We will hold a lock on it across the
1008 * transaction commit so that no other process can steal
1009 * the inode(s) that we've just allocated.
1011 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
1012 &ialloc_context
, &ip
);
1015 * Return an error if we were unable to allocate a new inode.
1016 * This should only happen if we run out of space on disk or
1017 * encounter a disk error.
1023 if (!ialloc_context
&& !ip
) {
1029 * If the AGI buffer is non-NULL, then we were unable to get an
1030 * inode in one operation. We need to commit the current
1031 * transaction and call xfs_ialloc() again. It is guaranteed
1032 * to succeed the second time.
1034 if (ialloc_context
) {
1036 * Normally, xfs_trans_commit releases all the locks.
1037 * We call bhold to hang on to the ialloc_context across
1038 * the commit. Holding this buffer prevents any other
1039 * processes from doing any allocations in this
1042 xfs_trans_bhold(tp
, ialloc_context
);
1045 * We want the quota changes to be associated with the next
1046 * transaction, NOT this one. So, detach the dqinfo from this
1047 * and attach it to the next transaction.
1052 dqinfo
= (void *)tp
->t_dqinfo
;
1053 tp
->t_dqinfo
= NULL
;
1054 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
1055 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
1058 code
= xfs_trans_roll(&tp
, NULL
);
1059 if (committed
!= NULL
)
1063 * Re-attach the quota info that we detached from prev trx.
1066 tp
->t_dqinfo
= dqinfo
;
1067 tp
->t_flags
|= tflags
;
1071 xfs_buf_relse(ialloc_context
);
1076 xfs_trans_bjoin(tp
, ialloc_context
);
1079 * Call ialloc again. Since we've locked out all
1080 * other allocations in this allocation group,
1081 * this call should always succeed.
1083 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1084 okalloc
, &ialloc_context
, &ip
);
1087 * If we get an error at this point, return to the caller
1088 * so that the current transaction can be aborted.
1095 ASSERT(!ialloc_context
&& ip
);
1098 if (committed
!= NULL
)
1109 * Decrement the link count on an inode & log the change. If this causes the
1110 * link count to go to zero, move the inode to AGI unlinked list so that it can
1111 * be freed when the last active reference goes away via xfs_inactive().
1113 static int /* error */
1118 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1120 drop_nlink(VFS_I(ip
));
1121 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1123 if (VFS_I(ip
)->i_nlink
)
1126 return xfs_iunlink(tp
, ip
);
1130 * Increment the link count on an inode & log the change.
1137 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1139 ASSERT(ip
->i_d
.di_version
> 1);
1140 inc_nlink(VFS_I(ip
));
1141 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1148 struct xfs_name
*name
,
1153 int is_dir
= S_ISDIR(mode
);
1154 struct xfs_mount
*mp
= dp
->i_mount
;
1155 struct xfs_inode
*ip
= NULL
;
1156 struct xfs_trans
*tp
= NULL
;
1158 struct xfs_defer_ops dfops
;
1159 xfs_fsblock_t first_block
;
1160 bool unlock_dp_on_error
= false;
1162 struct xfs_dquot
*udqp
= NULL
;
1163 struct xfs_dquot
*gdqp
= NULL
;
1164 struct xfs_dquot
*pdqp
= NULL
;
1165 struct xfs_trans_res
*tres
;
1168 trace_xfs_create(dp
, name
);
1170 if (XFS_FORCED_SHUTDOWN(mp
))
1173 prid
= xfs_get_initial_prid(dp
);
1176 * Make sure that we have allocated dquot(s) on disk.
1178 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1179 xfs_kgid_to_gid(current_fsgid()), prid
,
1180 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1181 &udqp
, &gdqp
, &pdqp
);
1187 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1188 tres
= &M_RES(mp
)->tr_mkdir
;
1190 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1191 tres
= &M_RES(mp
)->tr_create
;
1195 * Initially assume that the file does not exist and
1196 * reserve the resources for that case. If that is not
1197 * the case we'll drop the one we have and get a more
1198 * appropriate transaction later.
1200 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1201 if (error
== -ENOSPC
) {
1202 /* flush outstanding delalloc blocks and retry */
1203 xfs_flush_inodes(mp
);
1204 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1206 if (error
== -ENOSPC
) {
1207 /* No space at all so try a "no-allocation" reservation */
1209 error
= xfs_trans_alloc(mp
, tres
, 0, 0, 0, &tp
);
1212 goto out_release_inode
;
1214 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1215 unlock_dp_on_error
= true;
1217 xfs_defer_init(&dfops
, &first_block
);
1220 * Reserve disk quota and the inode.
1222 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1223 pdqp
, resblks
, 1, 0);
1225 goto out_trans_cancel
;
1228 error
= xfs_dir_canenter(tp
, dp
, name
);
1230 goto out_trans_cancel
;
1234 * A newly created regular or special file just has one directory
1235 * entry pointing to them, but a directory also the "." entry
1236 * pointing to itself.
1238 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1239 prid
, resblks
> 0, &ip
, NULL
);
1241 goto out_trans_cancel
;
1244 * Now we join the directory inode to the transaction. We do not do it
1245 * earlier because xfs_dir_ialloc might commit the previous transaction
1246 * (and release all the locks). An error from here on will result in
1247 * the transaction cancel unlocking dp so don't do it explicitly in the
1250 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1251 unlock_dp_on_error
= false;
1253 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1254 &first_block
, &dfops
, resblks
?
1255 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1257 ASSERT(error
!= -ENOSPC
);
1258 goto out_trans_cancel
;
1260 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1261 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1264 error
= xfs_dir_init(tp
, ip
, dp
);
1266 goto out_bmap_cancel
;
1268 error
= xfs_bumplink(tp
, dp
);
1270 goto out_bmap_cancel
;
1274 * If this is a synchronous mount, make sure that the
1275 * create transaction goes to disk before returning to
1278 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1279 xfs_trans_set_sync(tp
);
1282 * Attach the dquot(s) to the inodes and modify them incore.
1283 * These ids of the inode couldn't have changed since the new
1284 * inode has been locked ever since it was created.
1286 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1288 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1290 goto out_bmap_cancel
;
1292 error
= xfs_trans_commit(tp
);
1294 goto out_release_inode
;
1296 xfs_qm_dqrele(udqp
);
1297 xfs_qm_dqrele(gdqp
);
1298 xfs_qm_dqrele(pdqp
);
1304 xfs_defer_cancel(&dfops
);
1306 xfs_trans_cancel(tp
);
1309 * Wait until after the current transaction is aborted to finish the
1310 * setup of the inode and release the inode. This prevents recursive
1311 * transactions and deadlocks from xfs_inactive.
1314 xfs_finish_inode_setup(ip
);
1318 xfs_qm_dqrele(udqp
);
1319 xfs_qm_dqrele(gdqp
);
1320 xfs_qm_dqrele(pdqp
);
1322 if (unlock_dp_on_error
)
1323 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1329 struct xfs_inode
*dp
,
1330 struct dentry
*dentry
,
1332 struct xfs_inode
**ipp
)
1334 struct xfs_mount
*mp
= dp
->i_mount
;
1335 struct xfs_inode
*ip
= NULL
;
1336 struct xfs_trans
*tp
= NULL
;
1339 struct xfs_dquot
*udqp
= NULL
;
1340 struct xfs_dquot
*gdqp
= NULL
;
1341 struct xfs_dquot
*pdqp
= NULL
;
1342 struct xfs_trans_res
*tres
;
1345 if (XFS_FORCED_SHUTDOWN(mp
))
1348 prid
= xfs_get_initial_prid(dp
);
1351 * Make sure that we have allocated dquot(s) on disk.
1353 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1354 xfs_kgid_to_gid(current_fsgid()), prid
,
1355 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1356 &udqp
, &gdqp
, &pdqp
);
1360 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1361 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1363 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1364 if (error
== -ENOSPC
) {
1365 /* No space at all so try a "no-allocation" reservation */
1367 error
= xfs_trans_alloc(mp
, tres
, 0, 0, 0, &tp
);
1370 goto out_release_inode
;
1372 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1373 pdqp
, resblks
, 1, 0);
1375 goto out_trans_cancel
;
1377 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1, 0,
1378 prid
, resblks
> 0, &ip
, NULL
);
1380 goto out_trans_cancel
;
1382 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1383 xfs_trans_set_sync(tp
);
1386 * Attach the dquot(s) to the inodes and modify them incore.
1387 * These ids of the inode couldn't have changed since the new
1388 * inode has been locked ever since it was created.
1390 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1392 error
= xfs_iunlink(tp
, ip
);
1394 goto out_trans_cancel
;
1396 error
= xfs_trans_commit(tp
);
1398 goto out_release_inode
;
1400 xfs_qm_dqrele(udqp
);
1401 xfs_qm_dqrele(gdqp
);
1402 xfs_qm_dqrele(pdqp
);
1408 xfs_trans_cancel(tp
);
1411 * Wait until after the current transaction is aborted to finish the
1412 * setup of the inode and release the inode. This prevents recursive
1413 * transactions and deadlocks from xfs_inactive.
1416 xfs_finish_inode_setup(ip
);
1420 xfs_qm_dqrele(udqp
);
1421 xfs_qm_dqrele(gdqp
);
1422 xfs_qm_dqrele(pdqp
);
1431 struct xfs_name
*target_name
)
1433 xfs_mount_t
*mp
= tdp
->i_mount
;
1436 struct xfs_defer_ops dfops
;
1437 xfs_fsblock_t first_block
;
1440 trace_xfs_link(tdp
, target_name
);
1442 ASSERT(!S_ISDIR(VFS_I(sip
)->i_mode
));
1444 if (XFS_FORCED_SHUTDOWN(mp
))
1447 error
= xfs_qm_dqattach(sip
, 0);
1451 error
= xfs_qm_dqattach(tdp
, 0);
1455 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1456 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, resblks
, 0, 0, &tp
);
1457 if (error
== -ENOSPC
) {
1459 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, 0, 0, 0, &tp
);
1464 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1466 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1467 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1470 * If we are using project inheritance, we only allow hard link
1471 * creation in our tree when the project IDs are the same; else
1472 * the tree quota mechanism could be circumvented.
1474 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1475 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1481 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1486 xfs_defer_init(&dfops
, &first_block
);
1489 * Handle initial link state of O_TMPFILE inode
1491 if (VFS_I(sip
)->i_nlink
== 0) {
1492 error
= xfs_iunlink_remove(tp
, sip
);
1497 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1498 &first_block
, &dfops
, resblks
);
1501 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1502 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1504 error
= xfs_bumplink(tp
, sip
);
1509 * If this is a synchronous mount, make sure that the
1510 * link transaction goes to disk before returning to
1513 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1514 xfs_trans_set_sync(tp
);
1516 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1518 xfs_defer_cancel(&dfops
);
1522 return xfs_trans_commit(tp
);
1525 xfs_trans_cancel(tp
);
1531 * Free up the underlying blocks past new_size. The new size must be smaller
1532 * than the current size. This routine can be used both for the attribute and
1533 * data fork, and does not modify the inode size, which is left to the caller.
1535 * The transaction passed to this routine must have made a permanent log
1536 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1537 * given transaction and start new ones, so make sure everything involved in
1538 * the transaction is tidy before calling here. Some transaction will be
1539 * returned to the caller to be committed. The incoming transaction must
1540 * already include the inode, and both inode locks must be held exclusively.
1541 * The inode must also be "held" within the transaction. On return the inode
1542 * will be "held" within the returned transaction. This routine does NOT
1543 * require any disk space to be reserved for it within the transaction.
1545 * If we get an error, we must return with the inode locked and linked into the
1546 * current transaction. This keeps things simple for the higher level code,
1547 * because it always knows that the inode is locked and held in the transaction
1548 * that returns to it whether errors occur or not. We don't mark the inode
1549 * dirty on error so that transactions can be easily aborted if possible.
1552 xfs_itruncate_extents(
1553 struct xfs_trans
**tpp
,
1554 struct xfs_inode
*ip
,
1556 xfs_fsize_t new_size
)
1558 struct xfs_mount
*mp
= ip
->i_mount
;
1559 struct xfs_trans
*tp
= *tpp
;
1560 struct xfs_defer_ops dfops
;
1561 xfs_fsblock_t first_block
;
1562 xfs_fileoff_t first_unmap_block
;
1563 xfs_fileoff_t last_block
;
1564 xfs_filblks_t unmap_len
;
1568 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1569 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1570 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1571 ASSERT(new_size
<= XFS_ISIZE(ip
));
1572 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1573 ASSERT(ip
->i_itemp
!= NULL
);
1574 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1575 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1577 trace_xfs_itruncate_extents_start(ip
, new_size
);
1580 * Since it is possible for space to become allocated beyond
1581 * the end of the file (in a crash where the space is allocated
1582 * but the inode size is not yet updated), simply remove any
1583 * blocks which show up between the new EOF and the maximum
1584 * possible file size. If the first block to be removed is
1585 * beyond the maximum file size (ie it is the same as last_block),
1586 * then there is nothing to do.
1588 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1589 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1590 if (first_unmap_block
== last_block
)
1593 ASSERT(first_unmap_block
< last_block
);
1594 unmap_len
= last_block
- first_unmap_block
+ 1;
1596 xfs_defer_init(&dfops
, &first_block
);
1597 error
= xfs_bunmapi(tp
, ip
,
1598 first_unmap_block
, unmap_len
,
1599 xfs_bmapi_aflag(whichfork
),
1600 XFS_ITRUNC_MAX_EXTENTS
,
1601 &first_block
, &dfops
,
1604 goto out_bmap_cancel
;
1607 * Duplicate the transaction that has the permanent
1608 * reservation and commit the old transaction.
1610 error
= xfs_defer_finish(&tp
, &dfops
, ip
);
1612 goto out_bmap_cancel
;
1614 error
= xfs_trans_roll(&tp
, ip
);
1619 /* Remove all pending CoW reservations. */
1620 error
= xfs_reflink_cancel_cow_blocks(ip
, &tp
, first_unmap_block
,
1626 * Clear the reflink flag if we truncated everything.
1628 if (ip
->i_d
.di_nblocks
== 0 && xfs_is_reflink_inode(ip
)) {
1629 ip
->i_d
.di_flags2
&= ~XFS_DIFLAG2_REFLINK
;
1630 xfs_inode_clear_cowblocks_tag(ip
);
1634 * Always re-log the inode so that our permanent transaction can keep
1635 * on rolling it forward in the log.
1637 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1639 trace_xfs_itruncate_extents_end(ip
, new_size
);
1646 * If the bunmapi call encounters an error, return to the caller where
1647 * the transaction can be properly aborted. We just need to make sure
1648 * we're not holding any resources that we were not when we came in.
1650 xfs_defer_cancel(&dfops
);
1658 xfs_mount_t
*mp
= ip
->i_mount
;
1661 if (!S_ISREG(VFS_I(ip
)->i_mode
) || (VFS_I(ip
)->i_mode
== 0))
1664 /* If this is a read-only mount, don't do this (would generate I/O) */
1665 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1668 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1672 * If we previously truncated this file and removed old data
1673 * in the process, we want to initiate "early" writeout on
1674 * the last close. This is an attempt to combat the notorious
1675 * NULL files problem which is particularly noticeable from a
1676 * truncate down, buffered (re-)write (delalloc), followed by
1677 * a crash. What we are effectively doing here is
1678 * significantly reducing the time window where we'd otherwise
1679 * be exposed to that problem.
1681 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1683 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1684 if (ip
->i_delayed_blks
> 0) {
1685 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1692 if (VFS_I(ip
)->i_nlink
== 0)
1695 if (xfs_can_free_eofblocks(ip
, false)) {
1698 * Check if the inode is being opened, written and closed
1699 * frequently and we have delayed allocation blocks outstanding
1700 * (e.g. streaming writes from the NFS server), truncating the
1701 * blocks past EOF will cause fragmentation to occur.
1703 * In this case don't do the truncation, but we have to be
1704 * careful how we detect this case. Blocks beyond EOF show up as
1705 * i_delayed_blks even when the inode is clean, so we need to
1706 * truncate them away first before checking for a dirty release.
1707 * Hence on the first dirty close we will still remove the
1708 * speculative allocation, but after that we will leave it in
1711 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1714 * If we can't get the iolock just skip truncating the blocks
1715 * past EOF because we could deadlock with the mmap_sem
1716 * otherwise. We'll get another chance to drop them once the
1717 * last reference to the inode is dropped, so we'll never leak
1718 * blocks permanently.
1720 if (xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1721 error
= xfs_free_eofblocks(ip
);
1722 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1727 /* delalloc blocks after truncation means it really is dirty */
1728 if (ip
->i_delayed_blks
)
1729 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1735 * xfs_inactive_truncate
1737 * Called to perform a truncate when an inode becomes unlinked.
1740 xfs_inactive_truncate(
1741 struct xfs_inode
*ip
)
1743 struct xfs_mount
*mp
= ip
->i_mount
;
1744 struct xfs_trans
*tp
;
1747 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
1749 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1753 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1754 xfs_trans_ijoin(tp
, ip
, 0);
1757 * Log the inode size first to prevent stale data exposure in the event
1758 * of a system crash before the truncate completes. See the related
1759 * comment in xfs_vn_setattr_size() for details.
1761 ip
->i_d
.di_size
= 0;
1762 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1764 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1766 goto error_trans_cancel
;
1768 ASSERT(ip
->i_d
.di_nextents
== 0);
1770 error
= xfs_trans_commit(tp
);
1774 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1778 xfs_trans_cancel(tp
);
1780 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1785 * xfs_inactive_ifree()
1787 * Perform the inode free when an inode is unlinked.
1791 struct xfs_inode
*ip
)
1793 struct xfs_defer_ops dfops
;
1794 xfs_fsblock_t first_block
;
1795 struct xfs_mount
*mp
= ip
->i_mount
;
1796 struct xfs_trans
*tp
;
1800 * We try to use a per-AG reservation for any block needed by the finobt
1801 * tree, but as the finobt feature predates the per-AG reservation
1802 * support a degraded file system might not have enough space for the
1803 * reservation at mount time. In that case try to dip into the reserved
1806 * Send a warning if the reservation does happen to fail, as the inode
1807 * now remains allocated and sits on the unlinked list until the fs is
1810 if (unlikely(mp
->m_inotbt_nores
)) {
1811 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
,
1812 XFS_IFREE_SPACE_RES(mp
), 0, XFS_TRANS_RESERVE
,
1815 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
, 0, 0, 0, &tp
);
1818 if (error
== -ENOSPC
) {
1819 xfs_warn_ratelimited(mp
,
1820 "Failed to remove inode(s) from unlinked list. "
1821 "Please free space, unmount and run xfs_repair.");
1823 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1828 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1829 xfs_trans_ijoin(tp
, ip
, 0);
1831 xfs_defer_init(&dfops
, &first_block
);
1832 error
= xfs_ifree(tp
, ip
, &dfops
);
1835 * If we fail to free the inode, shut down. The cancel
1836 * might do that, we need to make sure. Otherwise the
1837 * inode might be lost for a long time or forever.
1839 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1840 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1842 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1844 xfs_trans_cancel(tp
);
1845 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1850 * Credit the quota account(s). The inode is gone.
1852 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1855 * Just ignore errors at this point. There is nothing we can do except
1856 * to try to keep going. Make sure it's not a silent error.
1858 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1860 xfs_notice(mp
, "%s: xfs_defer_finish returned error %d",
1862 xfs_defer_cancel(&dfops
);
1864 error
= xfs_trans_commit(tp
);
1866 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1869 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1876 * This is called when the vnode reference count for the vnode
1877 * goes to zero. If the file has been unlinked, then it must
1878 * now be truncated. Also, we clear all of the read-ahead state
1879 * kept for the inode here since the file is now closed.
1885 struct xfs_mount
*mp
;
1890 * If the inode is already free, then there can be nothing
1893 if (VFS_I(ip
)->i_mode
== 0) {
1894 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1895 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1900 ASSERT(!xfs_iflags_test(ip
, XFS_IRECOVERY
));
1902 /* If this is a read-only mount, don't do this (would generate I/O) */
1903 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1906 if (VFS_I(ip
)->i_nlink
!= 0) {
1908 * force is true because we are evicting an inode from the
1909 * cache. Post-eof blocks must be freed, lest we end up with
1910 * broken free space accounting.
1912 * Note: don't bother with iolock here since lockdep complains
1913 * about acquiring it in reclaim context. We have the only
1914 * reference to the inode at this point anyways.
1916 if (xfs_can_free_eofblocks(ip
, true))
1917 xfs_free_eofblocks(ip
);
1922 if (S_ISREG(VFS_I(ip
)->i_mode
) &&
1923 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1924 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1927 error
= xfs_qm_dqattach(ip
, 0);
1931 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1932 error
= xfs_inactive_symlink(ip
);
1934 error
= xfs_inactive_truncate(ip
);
1939 * If there are attributes associated with the file then blow them away
1940 * now. The code calls a routine that recursively deconstructs the
1941 * attribute fork. If also blows away the in-core attribute fork.
1943 if (XFS_IFORK_Q(ip
)) {
1944 error
= xfs_attr_inactive(ip
);
1950 ASSERT(ip
->i_d
.di_anextents
== 0);
1951 ASSERT(ip
->i_d
.di_forkoff
== 0);
1956 error
= xfs_inactive_ifree(ip
);
1961 * Release the dquots held by inode, if any.
1963 xfs_qm_dqdetach(ip
);
1967 * This is called when the inode's link count goes to 0 or we are creating a
1968 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1969 * set to true as the link count is dropped to zero by the VFS after we've
1970 * created the file successfully, so we have to add it to the unlinked list
1971 * while the link count is non-zero.
1973 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1974 * list when the inode is freed.
1978 struct xfs_trans
*tp
,
1979 struct xfs_inode
*ip
)
1981 xfs_mount_t
*mp
= tp
->t_mountp
;
1991 ASSERT(VFS_I(ip
)->i_mode
!= 0);
1994 * Get the agi buffer first. It ensures lock ordering
1997 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
2000 agi
= XFS_BUF_TO_AGI(agibp
);
2003 * Get the index into the agi hash table for the
2004 * list this inode will go on.
2006 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2008 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2009 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2010 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
2012 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
2014 * There is already another inode in the bucket we need
2015 * to add ourselves to. Add us at the front of the list.
2016 * Here we put the head pointer into our next pointer,
2017 * and then we fall through to point the head at us.
2019 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2024 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
2025 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
2026 offset
= ip
->i_imap
.im_boffset
+
2027 offsetof(xfs_dinode_t
, di_next_unlinked
);
2029 /* need to recalc the inode CRC if appropriate */
2030 xfs_dinode_calc_crc(mp
, dip
);
2032 xfs_trans_inode_buf(tp
, ibp
);
2033 xfs_trans_log_buf(tp
, ibp
, offset
,
2034 (offset
+ sizeof(xfs_agino_t
) - 1));
2035 xfs_inobp_check(mp
, ibp
);
2039 * Point the bucket head pointer at the inode being inserted.
2042 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
2043 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2044 (sizeof(xfs_agino_t
) * bucket_index
);
2045 xfs_trans_log_buf(tp
, agibp
, offset
,
2046 (offset
+ sizeof(xfs_agino_t
) - 1));
2051 * Pull the on-disk inode from the AGI unlinked list.
2064 xfs_agnumber_t agno
;
2066 xfs_agino_t next_agino
;
2067 xfs_buf_t
*last_ibp
;
2068 xfs_dinode_t
*last_dip
= NULL
;
2070 int offset
, last_offset
= 0;
2074 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2077 * Get the agi buffer first. It ensures lock ordering
2080 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2084 agi
= XFS_BUF_TO_AGI(agibp
);
2087 * Get the index into the agi hash table for the
2088 * list this inode will go on.
2090 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2092 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2093 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
2094 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2096 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2098 * We're at the head of the list. Get the inode's on-disk
2099 * buffer to see if there is anyone after us on the list.
2100 * Only modify our next pointer if it is not already NULLAGINO.
2101 * This saves us the overhead of dealing with the buffer when
2102 * there is no need to change it.
2104 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2107 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2111 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2112 ASSERT(next_agino
!= 0);
2113 if (next_agino
!= NULLAGINO
) {
2114 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2115 offset
= ip
->i_imap
.im_boffset
+
2116 offsetof(xfs_dinode_t
, di_next_unlinked
);
2118 /* need to recalc the inode CRC if appropriate */
2119 xfs_dinode_calc_crc(mp
, dip
);
2121 xfs_trans_inode_buf(tp
, ibp
);
2122 xfs_trans_log_buf(tp
, ibp
, offset
,
2123 (offset
+ sizeof(xfs_agino_t
) - 1));
2124 xfs_inobp_check(mp
, ibp
);
2126 xfs_trans_brelse(tp
, ibp
);
2129 * Point the bucket head pointer at the next inode.
2131 ASSERT(next_agino
!= 0);
2132 ASSERT(next_agino
!= agino
);
2133 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2134 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2135 (sizeof(xfs_agino_t
) * bucket_index
);
2136 xfs_trans_log_buf(tp
, agibp
, offset
,
2137 (offset
+ sizeof(xfs_agino_t
) - 1));
2140 * We need to search the list for the inode being freed.
2142 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2144 while (next_agino
!= agino
) {
2145 struct xfs_imap imap
;
2148 xfs_trans_brelse(tp
, last_ibp
);
2151 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2153 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2156 "%s: xfs_imap returned error %d.",
2161 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2165 "%s: xfs_imap_to_bp returned error %d.",
2170 last_offset
= imap
.im_boffset
;
2171 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2172 ASSERT(next_agino
!= NULLAGINO
);
2173 ASSERT(next_agino
!= 0);
2177 * Now last_ibp points to the buffer previous to us on the
2178 * unlinked list. Pull us from the list.
2180 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2183 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2187 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2188 ASSERT(next_agino
!= 0);
2189 ASSERT(next_agino
!= agino
);
2190 if (next_agino
!= NULLAGINO
) {
2191 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2192 offset
= ip
->i_imap
.im_boffset
+
2193 offsetof(xfs_dinode_t
, di_next_unlinked
);
2195 /* need to recalc the inode CRC if appropriate */
2196 xfs_dinode_calc_crc(mp
, dip
);
2198 xfs_trans_inode_buf(tp
, ibp
);
2199 xfs_trans_log_buf(tp
, ibp
, offset
,
2200 (offset
+ sizeof(xfs_agino_t
) - 1));
2201 xfs_inobp_check(mp
, ibp
);
2203 xfs_trans_brelse(tp
, ibp
);
2206 * Point the previous inode on the list to the next inode.
2208 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2209 ASSERT(next_agino
!= 0);
2210 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2212 /* need to recalc the inode CRC if appropriate */
2213 xfs_dinode_calc_crc(mp
, last_dip
);
2215 xfs_trans_inode_buf(tp
, last_ibp
);
2216 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2217 (offset
+ sizeof(xfs_agino_t
) - 1));
2218 xfs_inobp_check(mp
, last_ibp
);
2224 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2225 * inodes that are in memory - they all must be marked stale and attached to
2226 * the cluster buffer.
2230 xfs_inode_t
*free_ip
,
2232 struct xfs_icluster
*xic
)
2234 xfs_mount_t
*mp
= free_ip
->i_mount
;
2235 int blks_per_cluster
;
2236 int inodes_per_cluster
;
2243 xfs_inode_log_item_t
*iip
;
2244 xfs_log_item_t
*lip
;
2245 struct xfs_perag
*pag
;
2248 inum
= xic
->first_ino
;
2249 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2250 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2251 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
2252 nbufs
= mp
->m_ialloc_blks
/ blks_per_cluster
;
2254 for (j
= 0; j
< nbufs
; j
++, inum
+= inodes_per_cluster
) {
2256 * The allocation bitmap tells us which inodes of the chunk were
2257 * physically allocated. Skip the cluster if an inode falls into
2260 ioffset
= inum
- xic
->first_ino
;
2261 if ((xic
->alloc
& XFS_INOBT_MASK(ioffset
)) == 0) {
2262 ASSERT(do_mod(ioffset
, inodes_per_cluster
) == 0);
2266 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2267 XFS_INO_TO_AGBNO(mp
, inum
));
2270 * We obtain and lock the backing buffer first in the process
2271 * here, as we have to ensure that any dirty inode that we
2272 * can't get the flush lock on is attached to the buffer.
2273 * If we scan the in-memory inodes first, then buffer IO can
2274 * complete before we get a lock on it, and hence we may fail
2275 * to mark all the active inodes on the buffer stale.
2277 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2278 mp
->m_bsize
* blks_per_cluster
,
2285 * This buffer may not have been correctly initialised as we
2286 * didn't read it from disk. That's not important because we are
2287 * only using to mark the buffer as stale in the log, and to
2288 * attach stale cached inodes on it. That means it will never be
2289 * dispatched for IO. If it is, we want to know about it, and we
2290 * want it to fail. We can acheive this by adding a write
2291 * verifier to the buffer.
2293 bp
->b_ops
= &xfs_inode_buf_ops
;
2296 * Walk the inodes already attached to the buffer and mark them
2297 * stale. These will all have the flush locks held, so an
2298 * in-memory inode walk can't lock them. By marking them all
2299 * stale first, we will not attempt to lock them in the loop
2300 * below as the XFS_ISTALE flag will be set.
2304 if (lip
->li_type
== XFS_LI_INODE
) {
2305 iip
= (xfs_inode_log_item_t
*)lip
;
2306 ASSERT(iip
->ili_logged
== 1);
2307 lip
->li_cb
= xfs_istale_done
;
2308 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2309 &iip
->ili_flush_lsn
,
2310 &iip
->ili_item
.li_lsn
);
2311 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2313 lip
= lip
->li_bio_list
;
2318 * For each inode in memory attempt to add it to the inode
2319 * buffer and set it up for being staled on buffer IO
2320 * completion. This is safe as we've locked out tail pushing
2321 * and flushing by locking the buffer.
2323 * We have already marked every inode that was part of a
2324 * transaction stale above, which means there is no point in
2325 * even trying to lock them.
2327 for (i
= 0; i
< inodes_per_cluster
; i
++) {
2330 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2331 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2333 /* Inode not in memory, nothing to do */
2340 * because this is an RCU protected lookup, we could
2341 * find a recently freed or even reallocated inode
2342 * during the lookup. We need to check under the
2343 * i_flags_lock for a valid inode here. Skip it if it
2344 * is not valid, the wrong inode or stale.
2346 spin_lock(&ip
->i_flags_lock
);
2347 if (ip
->i_ino
!= inum
+ i
||
2348 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2349 spin_unlock(&ip
->i_flags_lock
);
2353 spin_unlock(&ip
->i_flags_lock
);
2356 * Don't try to lock/unlock the current inode, but we
2357 * _cannot_ skip the other inodes that we did not find
2358 * in the list attached to the buffer and are not
2359 * already marked stale. If we can't lock it, back off
2362 if (ip
!= free_ip
&&
2363 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2371 xfs_iflags_set(ip
, XFS_ISTALE
);
2374 * we don't need to attach clean inodes or those only
2375 * with unlogged changes (which we throw away, anyway).
2378 if (!iip
|| xfs_inode_clean(ip
)) {
2379 ASSERT(ip
!= free_ip
);
2381 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2385 iip
->ili_last_fields
= iip
->ili_fields
;
2386 iip
->ili_fields
= 0;
2387 iip
->ili_fsync_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 struct xfs_defer_ops
*dfops
)
2424 struct xfs_icluster xic
= { 0 };
2426 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2427 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2428 ASSERT(ip
->i_d
.di_nextents
== 0);
2429 ASSERT(ip
->i_d
.di_anextents
== 0);
2430 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(VFS_I(ip
)->i_mode
));
2431 ASSERT(ip
->i_d
.di_nblocks
== 0);
2434 * Pull the on-disk inode from the AGI unlinked list.
2436 error
= xfs_iunlink_remove(tp
, ip
);
2440 error
= xfs_difree(tp
, ip
->i_ino
, dfops
, &xic
);
2444 VFS_I(ip
)->i_mode
= 0; /* mark incore inode as free */
2445 ip
->i_d
.di_flags
= 0;
2446 ip
->i_d
.di_dmevmask
= 0;
2447 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2448 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2449 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2451 * Bump the generation count so no one will be confused
2452 * by reincarnations of this inode.
2454 VFS_I(ip
)->i_generation
++;
2455 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2458 error
= xfs_ifree_cluster(ip
, tp
, &xic
);
2464 * This is called to unpin an inode. The caller must have the inode locked
2465 * in at least shared mode so that the buffer cannot be subsequently pinned
2466 * once someone is waiting for it to be unpinned.
2470 struct xfs_inode
*ip
)
2472 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2474 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2476 /* Give the log a push to start the unpinning I/O */
2477 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2483 struct xfs_inode
*ip
)
2485 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2486 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2491 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2492 if (xfs_ipincount(ip
))
2494 } while (xfs_ipincount(ip
));
2495 finish_wait(wq
, &wait
.wq_entry
);
2500 struct xfs_inode
*ip
)
2502 if (xfs_ipincount(ip
))
2503 __xfs_iunpin_wait(ip
);
2507 * Removing an inode from the namespace involves removing the directory entry
2508 * and dropping the link count on the inode. Removing the directory entry can
2509 * result in locking an AGF (directory blocks were freed) and removing a link
2510 * count can result in placing the inode on an unlinked list which results in
2513 * The big problem here is that we have an ordering constraint on AGF and AGI
2514 * locking - inode allocation locks the AGI, then can allocate a new extent for
2515 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2516 * removes the inode from the unlinked list, requiring that we lock the AGI
2517 * first, and then freeing the inode can result in an inode chunk being freed
2518 * and hence freeing disk space requiring that we lock an AGF.
2520 * Hence the ordering that is imposed by other parts of the code is AGI before
2521 * AGF. This means we cannot remove the directory entry before we drop the inode
2522 * reference count and put it on the unlinked list as this results in a lock
2523 * order of AGF then AGI, and this can deadlock against inode allocation and
2524 * freeing. Therefore we must drop the link counts before we remove the
2527 * This is still safe from a transactional point of view - it is not until we
2528 * get to xfs_defer_finish() that we have the possibility of multiple
2529 * transactions in this operation. Hence as long as we remove the directory
2530 * entry and drop the link count in the first transaction of the remove
2531 * operation, there are no transactional constraints on the ordering here.
2536 struct xfs_name
*name
,
2539 xfs_mount_t
*mp
= dp
->i_mount
;
2540 xfs_trans_t
*tp
= NULL
;
2541 int is_dir
= S_ISDIR(VFS_I(ip
)->i_mode
);
2543 struct xfs_defer_ops dfops
;
2544 xfs_fsblock_t first_block
;
2547 trace_xfs_remove(dp
, name
);
2549 if (XFS_FORCED_SHUTDOWN(mp
))
2552 error
= xfs_qm_dqattach(dp
, 0);
2556 error
= xfs_qm_dqattach(ip
, 0);
2561 * We try to get the real space reservation first,
2562 * allowing for directory btree deletion(s) implying
2563 * possible bmap insert(s). If we can't get the space
2564 * reservation then we use 0 instead, and avoid the bmap
2565 * btree insert(s) in the directory code by, if the bmap
2566 * insert tries to happen, instead trimming the LAST
2567 * block from the directory.
2569 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2570 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, resblks
, 0, 0, &tp
);
2571 if (error
== -ENOSPC
) {
2573 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, 0, 0, 0,
2577 ASSERT(error
!= -ENOSPC
);
2581 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2583 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2584 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2587 * If we're removing a directory perform some additional validation.
2590 ASSERT(VFS_I(ip
)->i_nlink
>= 2);
2591 if (VFS_I(ip
)->i_nlink
!= 2) {
2593 goto out_trans_cancel
;
2595 if (!xfs_dir_isempty(ip
)) {
2597 goto out_trans_cancel
;
2600 /* Drop the link from ip's "..". */
2601 error
= xfs_droplink(tp
, dp
);
2603 goto out_trans_cancel
;
2605 /* Drop the "." link from ip to self. */
2606 error
= xfs_droplink(tp
, ip
);
2608 goto out_trans_cancel
;
2611 * When removing a non-directory we need to log the parent
2612 * inode here. For a directory this is done implicitly
2613 * by the xfs_droplink call for the ".." entry.
2615 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2617 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2619 /* Drop the link from dp to ip. */
2620 error
= xfs_droplink(tp
, ip
);
2622 goto out_trans_cancel
;
2624 xfs_defer_init(&dfops
, &first_block
);
2625 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2626 &first_block
, &dfops
, resblks
);
2628 ASSERT(error
!= -ENOENT
);
2629 goto out_bmap_cancel
;
2633 * If this is a synchronous mount, make sure that the
2634 * remove transaction goes to disk before returning to
2637 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2638 xfs_trans_set_sync(tp
);
2640 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
2642 goto out_bmap_cancel
;
2644 error
= xfs_trans_commit(tp
);
2648 if (is_dir
&& xfs_inode_is_filestream(ip
))
2649 xfs_filestream_deassociate(ip
);
2654 xfs_defer_cancel(&dfops
);
2656 xfs_trans_cancel(tp
);
2662 * Enter all inodes for a rename transaction into a sorted array.
2664 #define __XFS_SORT_INODES 5
2666 xfs_sort_for_rename(
2667 struct xfs_inode
*dp1
, /* in: old (source) directory inode */
2668 struct xfs_inode
*dp2
, /* in: new (target) directory inode */
2669 struct xfs_inode
*ip1
, /* in: inode of old entry */
2670 struct xfs_inode
*ip2
, /* in: inode of new entry */
2671 struct xfs_inode
*wip
, /* in: whiteout inode */
2672 struct xfs_inode
**i_tab
,/* out: sorted array of inodes */
2673 int *num_inodes
) /* in/out: inodes in array */
2677 ASSERT(*num_inodes
== __XFS_SORT_INODES
);
2678 memset(i_tab
, 0, *num_inodes
* sizeof(struct xfs_inode
*));
2681 * i_tab contains a list of pointers to inodes. We initialize
2682 * the table here & we'll sort it. We will then use it to
2683 * order the acquisition of the inode locks.
2685 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2698 * Sort the elements via bubble sort. (Remember, there are at
2699 * most 5 elements to sort, so this is adequate.)
2701 for (i
= 0; i
< *num_inodes
; i
++) {
2702 for (j
= 1; j
< *num_inodes
; j
++) {
2703 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2704 struct xfs_inode
*temp
= i_tab
[j
];
2705 i_tab
[j
] = i_tab
[j
-1];
2714 struct xfs_trans
*tp
,
2715 struct xfs_defer_ops
*dfops
)
2720 * If this is a synchronous mount, make sure that the rename transaction
2721 * goes to disk before returning to the user.
2723 if (tp
->t_mountp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2724 xfs_trans_set_sync(tp
);
2726 error
= xfs_defer_finish(&tp
, dfops
, NULL
);
2728 xfs_defer_cancel(dfops
);
2729 xfs_trans_cancel(tp
);
2733 return xfs_trans_commit(tp
);
2737 * xfs_cross_rename()
2739 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2743 struct xfs_trans
*tp
,
2744 struct xfs_inode
*dp1
,
2745 struct xfs_name
*name1
,
2746 struct xfs_inode
*ip1
,
2747 struct xfs_inode
*dp2
,
2748 struct xfs_name
*name2
,
2749 struct xfs_inode
*ip2
,
2750 struct xfs_defer_ops
*dfops
,
2751 xfs_fsblock_t
*first_block
,
2759 /* Swap inode number for dirent in first parent */
2760 error
= xfs_dir_replace(tp
, dp1
, name1
,
2762 first_block
, dfops
, spaceres
);
2764 goto out_trans_abort
;
2766 /* Swap inode number for dirent in second parent */
2767 error
= xfs_dir_replace(tp
, dp2
, name2
,
2769 first_block
, dfops
, spaceres
);
2771 goto out_trans_abort
;
2774 * If we're renaming one or more directories across different parents,
2775 * update the respective ".." entries (and link counts) to match the new
2779 dp2_flags
= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2781 if (S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2782 error
= xfs_dir_replace(tp
, ip2
, &xfs_name_dotdot
,
2783 dp1
->i_ino
, first_block
,
2786 goto out_trans_abort
;
2788 /* transfer ip2 ".." reference to dp1 */
2789 if (!S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2790 error
= xfs_droplink(tp
, dp2
);
2792 goto out_trans_abort
;
2793 error
= xfs_bumplink(tp
, dp1
);
2795 goto out_trans_abort
;
2799 * Although ip1 isn't changed here, userspace needs
2800 * to be warned about the change, so that applications
2801 * relying on it (like backup ones), will properly
2804 ip1_flags
|= XFS_ICHGTIME_CHG
;
2805 ip2_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2808 if (S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2809 error
= xfs_dir_replace(tp
, ip1
, &xfs_name_dotdot
,
2810 dp2
->i_ino
, first_block
,
2813 goto out_trans_abort
;
2815 /* transfer ip1 ".." reference to dp2 */
2816 if (!S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2817 error
= xfs_droplink(tp
, dp1
);
2819 goto out_trans_abort
;
2820 error
= xfs_bumplink(tp
, dp2
);
2822 goto out_trans_abort
;
2826 * Although ip2 isn't changed here, userspace needs
2827 * to be warned about the change, so that applications
2828 * relying on it (like backup ones), will properly
2831 ip1_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2832 ip2_flags
|= XFS_ICHGTIME_CHG
;
2837 xfs_trans_ichgtime(tp
, ip1
, ip1_flags
);
2838 xfs_trans_log_inode(tp
, ip1
, XFS_ILOG_CORE
);
2841 xfs_trans_ichgtime(tp
, ip2
, ip2_flags
);
2842 xfs_trans_log_inode(tp
, ip2
, XFS_ILOG_CORE
);
2845 xfs_trans_ichgtime(tp
, dp2
, dp2_flags
);
2846 xfs_trans_log_inode(tp
, dp2
, XFS_ILOG_CORE
);
2848 xfs_trans_ichgtime(tp
, dp1
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2849 xfs_trans_log_inode(tp
, dp1
, XFS_ILOG_CORE
);
2850 return xfs_finish_rename(tp
, dfops
);
2853 xfs_defer_cancel(dfops
);
2854 xfs_trans_cancel(tp
);
2859 * xfs_rename_alloc_whiteout()
2861 * Return a referenced, unlinked, unlocked inode that that can be used as a
2862 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2863 * crash between allocating the inode and linking it into the rename transaction
2864 * recovery will free the inode and we won't leak it.
2867 xfs_rename_alloc_whiteout(
2868 struct xfs_inode
*dp
,
2869 struct xfs_inode
**wip
)
2871 struct xfs_inode
*tmpfile
;
2874 error
= xfs_create_tmpfile(dp
, NULL
, S_IFCHR
| WHITEOUT_MODE
, &tmpfile
);
2879 * Prepare the tmpfile inode as if it were created through the VFS.
2880 * Otherwise, the link increment paths will complain about nlink 0->1.
2881 * Drop the link count as done by d_tmpfile(), complete the inode setup
2882 * and flag it as linkable.
2884 drop_nlink(VFS_I(tmpfile
));
2885 xfs_setup_iops(tmpfile
);
2886 xfs_finish_inode_setup(tmpfile
);
2887 VFS_I(tmpfile
)->i_state
|= I_LINKABLE
;
2898 struct xfs_inode
*src_dp
,
2899 struct xfs_name
*src_name
,
2900 struct xfs_inode
*src_ip
,
2901 struct xfs_inode
*target_dp
,
2902 struct xfs_name
*target_name
,
2903 struct xfs_inode
*target_ip
,
2906 struct xfs_mount
*mp
= src_dp
->i_mount
;
2907 struct xfs_trans
*tp
;
2908 struct xfs_defer_ops dfops
;
2909 xfs_fsblock_t first_block
;
2910 struct xfs_inode
*wip
= NULL
; /* whiteout inode */
2911 struct xfs_inode
*inodes
[__XFS_SORT_INODES
];
2912 int num_inodes
= __XFS_SORT_INODES
;
2913 bool new_parent
= (src_dp
!= target_dp
);
2914 bool src_is_directory
= S_ISDIR(VFS_I(src_ip
)->i_mode
);
2918 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2920 if ((flags
& RENAME_EXCHANGE
) && !target_ip
)
2924 * If we are doing a whiteout operation, allocate the whiteout inode
2925 * we will be placing at the target and ensure the type is set
2928 if (flags
& RENAME_WHITEOUT
) {
2929 ASSERT(!(flags
& (RENAME_NOREPLACE
| RENAME_EXCHANGE
)));
2930 error
= xfs_rename_alloc_whiteout(target_dp
, &wip
);
2934 /* setup target dirent info as whiteout */
2935 src_name
->type
= XFS_DIR3_FT_CHRDEV
;
2938 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
, wip
,
2939 inodes
, &num_inodes
);
2941 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2942 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, spaceres
, 0, 0, &tp
);
2943 if (error
== -ENOSPC
) {
2945 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, 0, 0, 0,
2949 goto out_release_wip
;
2952 * Attach the dquots to the inodes
2954 error
= xfs_qm_vop_rename_dqattach(inodes
);
2956 goto out_trans_cancel
;
2959 * Lock all the participating inodes. Depending upon whether
2960 * the target_name exists in the target directory, and
2961 * whether the target directory is the same as the source
2962 * directory, we can lock from 2 to 4 inodes.
2964 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2967 * Join all the inodes to the transaction. From this point on,
2968 * we can rely on either trans_commit or trans_cancel to unlock
2971 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2973 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2974 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2976 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2978 xfs_trans_ijoin(tp
, wip
, XFS_ILOCK_EXCL
);
2981 * If we are using project inheritance, we only allow renames
2982 * into our tree when the project IDs are the same; else the
2983 * tree quota mechanism would be circumvented.
2985 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2986 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2988 goto out_trans_cancel
;
2991 xfs_defer_init(&dfops
, &first_block
);
2993 /* RENAME_EXCHANGE is unique from here on. */
2994 if (flags
& RENAME_EXCHANGE
)
2995 return xfs_cross_rename(tp
, src_dp
, src_name
, src_ip
,
2996 target_dp
, target_name
, target_ip
,
2997 &dfops
, &first_block
, spaceres
);
3000 * Set up the target.
3002 if (target_ip
== NULL
) {
3004 * If there's no space reservation, check the entry will
3005 * fit before actually inserting it.
3008 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
3010 goto out_trans_cancel
;
3013 * If target does not exist and the rename crosses
3014 * directories, adjust the target directory link count
3015 * to account for the ".." reference from the new entry.
3017 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
3018 src_ip
->i_ino
, &first_block
,
3021 goto out_bmap_cancel
;
3023 xfs_trans_ichgtime(tp
, target_dp
,
3024 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3026 if (new_parent
&& src_is_directory
) {
3027 error
= xfs_bumplink(tp
, target_dp
);
3029 goto out_bmap_cancel
;
3031 } else { /* target_ip != NULL */
3033 * If target exists and it's a directory, check that both
3034 * target and source are directories and that target can be
3035 * destroyed, or that neither is a directory.
3037 if (S_ISDIR(VFS_I(target_ip
)->i_mode
)) {
3039 * Make sure target dir is empty.
3041 if (!(xfs_dir_isempty(target_ip
)) ||
3042 (VFS_I(target_ip
)->i_nlink
> 2)) {
3044 goto out_trans_cancel
;
3049 * Link the source inode under the target name.
3050 * If the source inode is a directory and we are moving
3051 * it across directories, its ".." entry will be
3052 * inconsistent until we replace that down below.
3054 * In case there is already an entry with the same
3055 * name at the destination directory, remove it first.
3057 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
3059 &first_block
, &dfops
, spaceres
);
3061 goto out_bmap_cancel
;
3063 xfs_trans_ichgtime(tp
, target_dp
,
3064 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3067 * Decrement the link count on the target since the target
3068 * dir no longer points to it.
3070 error
= xfs_droplink(tp
, target_ip
);
3072 goto out_bmap_cancel
;
3074 if (src_is_directory
) {
3076 * Drop the link from the old "." entry.
3078 error
= xfs_droplink(tp
, target_ip
);
3080 goto out_bmap_cancel
;
3082 } /* target_ip != NULL */
3085 * Remove the source.
3087 if (new_parent
&& src_is_directory
) {
3089 * Rewrite the ".." entry to point to the new
3092 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
3094 &first_block
, &dfops
, spaceres
);
3095 ASSERT(error
!= -EEXIST
);
3097 goto out_bmap_cancel
;
3101 * We always want to hit the ctime on the source inode.
3103 * This isn't strictly required by the standards since the source
3104 * inode isn't really being changed, but old unix file systems did
3105 * it and some incremental backup programs won't work without it.
3107 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
3108 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
3111 * Adjust the link count on src_dp. This is necessary when
3112 * renaming a directory, either within one parent when
3113 * the target existed, or across two parent directories.
3115 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
3118 * Decrement link count on src_directory since the
3119 * entry that's moved no longer points to it.
3121 error
= xfs_droplink(tp
, src_dp
);
3123 goto out_bmap_cancel
;
3127 * For whiteouts, we only need to update the source dirent with the
3128 * inode number of the whiteout inode rather than removing it
3132 error
= xfs_dir_replace(tp
, src_dp
, src_name
, wip
->i_ino
,
3133 &first_block
, &dfops
, spaceres
);
3135 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
3136 &first_block
, &dfops
, spaceres
);
3138 goto out_bmap_cancel
;
3141 * For whiteouts, we need to bump the link count on the whiteout inode.
3142 * This means that failures all the way up to this point leave the inode
3143 * on the unlinked list and so cleanup is a simple matter of dropping
3144 * the remaining reference to it. If we fail here after bumping the link
3145 * count, we're shutting down the filesystem so we'll never see the
3146 * intermediate state on disk.
3149 ASSERT(VFS_I(wip
)->i_nlink
== 0);
3150 error
= xfs_bumplink(tp
, wip
);
3152 goto out_bmap_cancel
;
3153 error
= xfs_iunlink_remove(tp
, wip
);
3155 goto out_bmap_cancel
;
3156 xfs_trans_log_inode(tp
, wip
, XFS_ILOG_CORE
);
3159 * Now we have a real link, clear the "I'm a tmpfile" state
3160 * flag from the inode so it doesn't accidentally get misused in
3163 VFS_I(wip
)->i_state
&= ~I_LINKABLE
;
3166 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3167 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
3169 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
3171 error
= xfs_finish_rename(tp
, &dfops
);
3177 xfs_defer_cancel(&dfops
);
3179 xfs_trans_cancel(tp
);
3188 struct xfs_inode
*ip
,
3191 struct xfs_mount
*mp
= ip
->i_mount
;
3192 struct xfs_perag
*pag
;
3193 unsigned long first_index
, mask
;
3194 unsigned long inodes_per_cluster
;
3196 struct xfs_inode
**cilist
;
3197 struct xfs_inode
*cip
;
3203 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
3205 inodes_per_cluster
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
3206 cilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
3207 cilist
= kmem_alloc(cilist_size
, KM_MAYFAIL
|KM_NOFS
);
3211 mask
= ~(((mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
)) - 1);
3212 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
3214 /* really need a gang lookup range call here */
3215 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)cilist
,
3216 first_index
, inodes_per_cluster
);
3220 for (i
= 0; i
< nr_found
; i
++) {
3226 * because this is an RCU protected lookup, we could find a
3227 * recently freed or even reallocated inode during the lookup.
3228 * We need to check under the i_flags_lock for a valid inode
3229 * here. Skip it if it is not valid or the wrong inode.
3231 spin_lock(&cip
->i_flags_lock
);
3233 __xfs_iflags_test(cip
, XFS_ISTALE
)) {
3234 spin_unlock(&cip
->i_flags_lock
);
3239 * Once we fall off the end of the cluster, no point checking
3240 * any more inodes in the list because they will also all be
3241 * outside the cluster.
3243 if ((XFS_INO_TO_AGINO(mp
, cip
->i_ino
) & mask
) != first_index
) {
3244 spin_unlock(&cip
->i_flags_lock
);
3247 spin_unlock(&cip
->i_flags_lock
);
3250 * Do an un-protected check to see if the inode is dirty and
3251 * is a candidate for flushing. These checks will be repeated
3252 * later after the appropriate locks are acquired.
3254 if (xfs_inode_clean(cip
) && xfs_ipincount(cip
) == 0)
3258 * Try to get locks. If any are unavailable or it is pinned,
3259 * then this inode cannot be flushed and is skipped.
3262 if (!xfs_ilock_nowait(cip
, XFS_ILOCK_SHARED
))
3264 if (!xfs_iflock_nowait(cip
)) {
3265 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3268 if (xfs_ipincount(cip
)) {
3270 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3276 * Check the inode number again, just to be certain we are not
3277 * racing with freeing in xfs_reclaim_inode(). See the comments
3278 * in that function for more information as to why the initial
3279 * check is not sufficient.
3283 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3288 * arriving here means that this inode can be flushed. First
3289 * re-check that it's dirty before flushing.
3291 if (!xfs_inode_clean(cip
)) {
3293 error
= xfs_iflush_int(cip
, bp
);
3295 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3296 goto cluster_corrupt_out
;
3302 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3306 XFS_STATS_INC(mp
, xs_icluster_flushcnt
);
3307 XFS_STATS_ADD(mp
, xs_icluster_flushinode
, clcount
);
3318 cluster_corrupt_out
:
3320 * Corruption detected in the clustering loop. Invalidate the
3321 * inode buffer and shut down the filesystem.
3325 * Clean up the buffer. If it was delwri, just release it --
3326 * brelse can handle it with no problems. If not, shut down the
3327 * filesystem before releasing the buffer.
3329 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3333 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3335 if (!bufwasdelwri
) {
3337 * Just like incore_relse: if we have b_iodone functions,
3338 * mark the buffer as an error and call them. Otherwise
3339 * mark it as stale and brelse.
3342 bp
->b_flags
&= ~XBF_DONE
;
3344 xfs_buf_ioerror(bp
, -EIO
);
3353 * Unlocks the flush lock
3355 xfs_iflush_abort(cip
, false);
3358 return -EFSCORRUPTED
;
3362 * Flush dirty inode metadata into the backing buffer.
3364 * The caller must have the inode lock and the inode flush lock held. The
3365 * inode lock will still be held upon return to the caller, and the inode
3366 * flush lock will be released after the inode has reached the disk.
3368 * The caller must write out the buffer returned in *bpp and release it.
3372 struct xfs_inode
*ip
,
3373 struct xfs_buf
**bpp
)
3375 struct xfs_mount
*mp
= ip
->i_mount
;
3376 struct xfs_buf
*bp
= NULL
;
3377 struct xfs_dinode
*dip
;
3380 XFS_STATS_INC(mp
, xs_iflush_count
);
3382 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3383 ASSERT(xfs_isiflocked(ip
));
3384 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3385 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3389 xfs_iunpin_wait(ip
);
3392 * For stale inodes we cannot rely on the backing buffer remaining
3393 * stale in cache for the remaining life of the stale inode and so
3394 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3395 * inodes below. We have to check this after ensuring the inode is
3396 * unpinned so that it is safe to reclaim the stale inode after the
3399 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3405 * This may have been unpinned because the filesystem is shutting
3406 * down forcibly. If that's the case we must not write this inode
3407 * to disk, because the log record didn't make it to disk.
3409 * We also have to remove the log item from the AIL in this case,
3410 * as we wait for an empty AIL as part of the unmount process.
3412 if (XFS_FORCED_SHUTDOWN(mp
)) {
3418 * Get the buffer containing the on-disk inode. We are doing a try-lock
3419 * operation here, so we may get an EAGAIN error. In that case, we
3420 * simply want to return with the inode still dirty.
3422 * If we get any other error, we effectively have a corruption situation
3423 * and we cannot flush the inode, so we treat it the same as failing
3426 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3428 if (error
== -EAGAIN
) {
3436 * First flush out the inode that xfs_iflush was called with.
3438 error
= xfs_iflush_int(ip
, bp
);
3443 * If the buffer is pinned then push on the log now so we won't
3444 * get stuck waiting in the write for too long.
3446 if (xfs_buf_ispinned(bp
))
3447 xfs_log_force(mp
, 0);
3451 * see if other inodes can be gathered into this write
3453 error
= xfs_iflush_cluster(ip
, bp
);
3455 goto cluster_corrupt_out
;
3463 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3464 cluster_corrupt_out
:
3465 error
= -EFSCORRUPTED
;
3468 * Unlocks the flush lock
3470 xfs_iflush_abort(ip
, false);
3476 struct xfs_inode
*ip
,
3479 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3480 struct xfs_dinode
*dip
;
3481 struct xfs_mount
*mp
= ip
->i_mount
;
3483 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3484 ASSERT(xfs_isiflocked(ip
));
3485 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3486 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3487 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3488 ASSERT(ip
->i_d
.di_version
> 1);
3490 /* set *dip = inode's place in the buffer */
3491 dip
= xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3493 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3494 mp
, XFS_ERRTAG_IFLUSH_1
)) {
3495 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3496 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3497 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3500 if (S_ISREG(VFS_I(ip
)->i_mode
)) {
3502 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3503 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3504 mp
, XFS_ERRTAG_IFLUSH_3
)) {
3505 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3506 "%s: Bad regular inode %Lu, ptr 0x%p",
3507 __func__
, ip
->i_ino
, ip
);
3510 } else if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
3512 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3513 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3514 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3515 mp
, XFS_ERRTAG_IFLUSH_4
)) {
3516 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3517 "%s: Bad directory inode %Lu, ptr 0x%p",
3518 __func__
, ip
->i_ino
, ip
);
3522 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3523 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
)) {
3524 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3525 "%s: detected corrupt incore inode %Lu, "
3526 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3527 __func__
, ip
->i_ino
,
3528 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3529 ip
->i_d
.di_nblocks
, ip
);
3532 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3533 mp
, XFS_ERRTAG_IFLUSH_6
)) {
3534 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3535 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3536 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3541 * Inode item log recovery for v2 inodes are dependent on the
3542 * di_flushiter count for correct sequencing. We bump the flush
3543 * iteration count so we can detect flushes which postdate a log record
3544 * during recovery. This is redundant as we now log every change and
3545 * hence this can't happen but we need to still do it to ensure
3546 * backwards compatibility with old kernels that predate logging all
3549 if (ip
->i_d
.di_version
< 3)
3550 ip
->i_d
.di_flushiter
++;
3552 /* Check the inline directory data. */
3553 if (S_ISDIR(VFS_I(ip
)->i_mode
) &&
3554 ip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
&&
3555 xfs_dir2_sf_verify(ip
))
3559 * Copy the dirty parts of the inode into the on-disk inode. We always
3560 * copy out the core of the inode, because if the inode is dirty at all
3563 xfs_inode_to_disk(ip
, dip
, iip
->ili_item
.li_lsn
);
3565 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3566 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3567 ip
->i_d
.di_flushiter
= 0;
3569 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3570 if (XFS_IFORK_Q(ip
))
3571 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3572 xfs_inobp_check(mp
, bp
);
3575 * We've recorded everything logged in the inode, so we'd like to clear
3576 * the ili_fields bits so we don't log and flush things unnecessarily.
3577 * However, we can't stop logging all this information until the data
3578 * we've copied into the disk buffer is written to disk. If we did we
3579 * might overwrite the copy of the inode in the log with all the data
3580 * after re-logging only part of it, and in the face of a crash we
3581 * wouldn't have all the data we need to recover.
3583 * What we do is move the bits to the ili_last_fields field. When
3584 * logging the inode, these bits are moved back to the ili_fields field.
3585 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3586 * know that the information those bits represent is permanently on
3587 * disk. As long as the flush completes before the inode is logged
3588 * again, then both ili_fields and ili_last_fields will be cleared.
3590 * We can play with the ili_fields bits here, because the inode lock
3591 * must be held exclusively in order to set bits there and the flush
3592 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3593 * done routine can tell whether or not to look in the AIL. Also, store
3594 * the current LSN of the inode so that we can tell whether the item has
3595 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3596 * need the AIL lock, because it is a 64 bit value that cannot be read
3599 iip
->ili_last_fields
= iip
->ili_fields
;
3600 iip
->ili_fields
= 0;
3601 iip
->ili_fsync_fields
= 0;
3602 iip
->ili_logged
= 1;
3604 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3605 &iip
->ili_item
.li_lsn
);
3608 * Attach the function xfs_iflush_done to the inode's
3609 * buffer. This will remove the inode from the AIL
3610 * and unlock the inode's flush lock when the inode is
3611 * completely written to disk.
3613 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3615 /* generate the checksum. */
3616 xfs_dinode_calc_crc(mp
, dip
);
3618 ASSERT(bp
->b_fspriv
!= NULL
);
3619 ASSERT(bp
->b_iodone
!= NULL
);
3623 return -EFSCORRUPTED
;