1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
6 #include <linux/iversion.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
15 #include "xfs_mount.h"
16 #include "xfs_defer.h"
17 #include "xfs_inode.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_ialloc.h"
26 #include "xfs_bmap_util.h"
27 #include "xfs_errortag.h"
28 #include "xfs_error.h"
29 #include "xfs_quota.h"
30 #include "xfs_filestream.h"
31 #include "xfs_trace.h"
32 #include "xfs_icache.h"
33 #include "xfs_symlink.h"
34 #include "xfs_trans_priv.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_reflink.h"
39 kmem_zone_t
*xfs_inode_zone
;
42 * Used in xfs_itruncate_extents(). This is the maximum number of extents
43 * freed from a file in a single transaction.
45 #define XFS_ITRUNC_MAX_EXTENTS 2
47 STATIC
int xfs_iflush_int(struct xfs_inode
*, struct xfs_buf
*);
48 STATIC
int xfs_iunlink(struct xfs_trans
*, struct xfs_inode
*);
49 STATIC
int xfs_iunlink_remove(struct xfs_trans
*, struct xfs_inode
*);
52 * helper function to extract extent size hint from inode
59 * No point in aligning allocations if we need to COW to actually
62 if (xfs_is_always_cow_inode(ip
))
64 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
65 return ip
->i_d
.di_extsize
;
66 if (XFS_IS_REALTIME_INODE(ip
))
67 return ip
->i_mount
->m_sb
.sb_rextsize
;
72 * Helper function to extract CoW extent size hint from inode.
73 * Between the extent size hint and the CoW extent size hint, we
74 * return the greater of the two. If the value is zero (automatic),
75 * use the default size.
78 xfs_get_cowextsz_hint(
84 if (ip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
85 a
= ip
->i_d
.di_cowextsize
;
86 b
= xfs_get_extsz_hint(ip
);
90 return XFS_DEFAULT_COWEXTSZ_HINT
;
95 * These two are wrapper routines around the xfs_ilock() routine used to
96 * centralize some grungy code. They are used in places that wish to lock the
97 * inode solely for reading the extents. The reason these places can't just
98 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
99 * bringing in of the extents from disk for a file in b-tree format. If the
100 * inode is in b-tree format, then we need to lock the inode exclusively until
101 * the extents are read in. Locking it exclusively all the time would limit
102 * our parallelism unnecessarily, though. What we do instead is check to see
103 * if the extents have been read in yet, and only lock the inode exclusively
106 * The functions return a value which should be given to the corresponding
107 * xfs_iunlock() call.
110 xfs_ilock_data_map_shared(
111 struct xfs_inode
*ip
)
113 uint lock_mode
= XFS_ILOCK_SHARED
;
115 if (ip
->i_df
.if_format
== XFS_DINODE_FMT_BTREE
&&
116 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)
117 lock_mode
= XFS_ILOCK_EXCL
;
118 xfs_ilock(ip
, lock_mode
);
123 xfs_ilock_attr_map_shared(
124 struct xfs_inode
*ip
)
126 uint lock_mode
= XFS_ILOCK_SHARED
;
129 ip
->i_afp
->if_format
== XFS_DINODE_FMT_BTREE
&&
130 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
131 lock_mode
= XFS_ILOCK_EXCL
;
132 xfs_ilock(ip
, lock_mode
);
137 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
138 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
139 * various combinations of the locks to be obtained.
141 * The 3 locks should always be ordered so that the IO lock is obtained first,
142 * the mmap lock second and the ilock last in order to prevent deadlock.
144 * Basic locking order:
146 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
148 * mmap_lock locking order:
150 * i_rwsem -> page lock -> mmap_lock
151 * mmap_lock -> i_mmap_lock -> page_lock
153 * The difference in mmap_lock locking order mean that we cannot hold the
154 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
155 * fault in pages during copy in/out (for buffered IO) or require the mmap_lock
156 * in get_user_pages() to map the user pages into the kernel address space for
157 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
158 * page faults already hold the mmap_lock.
160 * Hence to serialise fully against both syscall and mmap based IO, we need to
161 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
162 * taken in places where we need to invalidate the page cache in a race
163 * free manner (e.g. truncate, hole punch and other extent manipulation
171 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
174 * You can't set both SHARED and EXCL for the same lock,
175 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
176 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
178 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
179 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
180 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
181 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
182 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
183 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
184 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
186 if (lock_flags
& XFS_IOLOCK_EXCL
) {
187 down_write_nested(&VFS_I(ip
)->i_rwsem
,
188 XFS_IOLOCK_DEP(lock_flags
));
189 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
190 down_read_nested(&VFS_I(ip
)->i_rwsem
,
191 XFS_IOLOCK_DEP(lock_flags
));
194 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
195 mrupdate_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
196 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
197 mraccess_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
199 if (lock_flags
& XFS_ILOCK_EXCL
)
200 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
201 else if (lock_flags
& XFS_ILOCK_SHARED
)
202 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
206 * This is just like xfs_ilock(), except that the caller
207 * is guaranteed not to sleep. It returns 1 if it gets
208 * the requested locks and 0 otherwise. If the IO lock is
209 * obtained but the inode lock cannot be, then the IO lock
210 * is dropped before returning.
212 * ip -- the inode being locked
213 * lock_flags -- this parameter indicates the inode's locks to be
214 * to be locked. See the comment for xfs_ilock() for a list
222 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
225 * You can't set both SHARED and EXCL for the same lock,
226 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
227 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
229 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
230 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
231 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
232 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
233 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
234 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
235 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
237 if (lock_flags
& XFS_IOLOCK_EXCL
) {
238 if (!down_write_trylock(&VFS_I(ip
)->i_rwsem
))
240 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
241 if (!down_read_trylock(&VFS_I(ip
)->i_rwsem
))
245 if (lock_flags
& XFS_MMAPLOCK_EXCL
) {
246 if (!mrtryupdate(&ip
->i_mmaplock
))
247 goto out_undo_iolock
;
248 } else if (lock_flags
& XFS_MMAPLOCK_SHARED
) {
249 if (!mrtryaccess(&ip
->i_mmaplock
))
250 goto out_undo_iolock
;
253 if (lock_flags
& XFS_ILOCK_EXCL
) {
254 if (!mrtryupdate(&ip
->i_lock
))
255 goto out_undo_mmaplock
;
256 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
257 if (!mrtryaccess(&ip
->i_lock
))
258 goto out_undo_mmaplock
;
263 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
264 mrunlock_excl(&ip
->i_mmaplock
);
265 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
266 mrunlock_shared(&ip
->i_mmaplock
);
268 if (lock_flags
& XFS_IOLOCK_EXCL
)
269 up_write(&VFS_I(ip
)->i_rwsem
);
270 else if (lock_flags
& XFS_IOLOCK_SHARED
)
271 up_read(&VFS_I(ip
)->i_rwsem
);
277 * xfs_iunlock() is used to drop the inode locks acquired with
278 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
280 * that we know which locks to drop.
282 * ip -- the inode being unlocked
283 * lock_flags -- this parameter indicates the inode's locks to be
284 * to be unlocked. See the comment for xfs_ilock() for a list
285 * of valid values for this parameter.
294 * You can't set both SHARED and EXCL for the same lock,
295 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
296 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
298 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
299 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
300 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
301 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
302 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
303 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
304 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
305 ASSERT(lock_flags
!= 0);
307 if (lock_flags
& XFS_IOLOCK_EXCL
)
308 up_write(&VFS_I(ip
)->i_rwsem
);
309 else if (lock_flags
& XFS_IOLOCK_SHARED
)
310 up_read(&VFS_I(ip
)->i_rwsem
);
312 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
313 mrunlock_excl(&ip
->i_mmaplock
);
314 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
315 mrunlock_shared(&ip
->i_mmaplock
);
317 if (lock_flags
& XFS_ILOCK_EXCL
)
318 mrunlock_excl(&ip
->i_lock
);
319 else if (lock_flags
& XFS_ILOCK_SHARED
)
320 mrunlock_shared(&ip
->i_lock
);
322 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
326 * give up write locks. the i/o lock cannot be held nested
327 * if it is being demoted.
334 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
));
336 ~(XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
338 if (lock_flags
& XFS_ILOCK_EXCL
)
339 mrdemote(&ip
->i_lock
);
340 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
341 mrdemote(&ip
->i_mmaplock
);
342 if (lock_flags
& XFS_IOLOCK_EXCL
)
343 downgrade_write(&VFS_I(ip
)->i_rwsem
);
345 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
348 #if defined(DEBUG) || defined(XFS_WARN)
354 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
355 if (!(lock_flags
& XFS_ILOCK_SHARED
))
356 return !!ip
->i_lock
.mr_writer
;
357 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
360 if (lock_flags
& (XFS_MMAPLOCK_EXCL
|XFS_MMAPLOCK_SHARED
)) {
361 if (!(lock_flags
& XFS_MMAPLOCK_SHARED
))
362 return !!ip
->i_mmaplock
.mr_writer
;
363 return rwsem_is_locked(&ip
->i_mmaplock
.mr_lock
);
366 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
367 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
368 return !debug_locks
||
369 lockdep_is_held_type(&VFS_I(ip
)->i_rwsem
, 0);
370 return rwsem_is_locked(&VFS_I(ip
)->i_rwsem
);
379 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
380 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
381 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
382 * errors and warnings.
384 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
386 xfs_lockdep_subclass_ok(
389 return subclass
< MAX_LOCKDEP_SUBCLASSES
;
392 #define xfs_lockdep_subclass_ok(subclass) (true)
396 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
397 * value. This can be called for any type of inode lock combination, including
398 * parent locking. Care must be taken to ensure we don't overrun the subclass
399 * storage fields in the class mask we build.
402 xfs_lock_inumorder(int lock_mode
, int subclass
)
406 ASSERT(!(lock_mode
& (XFS_ILOCK_PARENT
| XFS_ILOCK_RTBITMAP
|
408 ASSERT(xfs_lockdep_subclass_ok(subclass
));
410 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
411 ASSERT(subclass
<= XFS_IOLOCK_MAX_SUBCLASS
);
412 class += subclass
<< XFS_IOLOCK_SHIFT
;
415 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) {
416 ASSERT(subclass
<= XFS_MMAPLOCK_MAX_SUBCLASS
);
417 class += subclass
<< XFS_MMAPLOCK_SHIFT
;
420 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) {
421 ASSERT(subclass
<= XFS_ILOCK_MAX_SUBCLASS
);
422 class += subclass
<< XFS_ILOCK_SHIFT
;
425 return (lock_mode
& ~XFS_LOCK_SUBCLASS_MASK
) | class;
429 * The following routine will lock n inodes in exclusive mode. We assume the
430 * caller calls us with the inodes in i_ino order.
432 * We need to detect deadlock where an inode that we lock is in the AIL and we
433 * start waiting for another inode that is locked by a thread in a long running
434 * transaction (such as truncate). This can result in deadlock since the long
435 * running trans might need to wait for the inode we just locked in order to
436 * push the tail and free space in the log.
438 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
439 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
440 * lock more than one at a time, lockdep will report false positives saying we
441 * have violated locking orders.
445 struct xfs_inode
**ips
,
449 int attempts
= 0, i
, j
, try_lock
;
450 struct xfs_log_item
*lp
;
453 * Currently supports between 2 and 5 inodes with exclusive locking. We
454 * support an arbitrary depth of locking here, but absolute limits on
455 * inodes depend on the the type of locking and the limits placed by
456 * lockdep annotations in xfs_lock_inumorder. These are all checked by
459 ASSERT(ips
&& inodes
>= 2 && inodes
<= 5);
460 ASSERT(lock_mode
& (XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
|
462 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
| XFS_MMAPLOCK_SHARED
|
464 ASSERT(!(lock_mode
& XFS_MMAPLOCK_EXCL
) ||
465 inodes
<= XFS_MMAPLOCK_MAX_SUBCLASS
+ 1);
466 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
) ||
467 inodes
<= XFS_ILOCK_MAX_SUBCLASS
+ 1);
469 if (lock_mode
& XFS_IOLOCK_EXCL
) {
470 ASSERT(!(lock_mode
& (XFS_MMAPLOCK_EXCL
| XFS_ILOCK_EXCL
)));
471 } else if (lock_mode
& XFS_MMAPLOCK_EXCL
)
472 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
));
477 for (; i
< inodes
; i
++) {
480 if (i
&& (ips
[i
] == ips
[i
- 1])) /* Already locked */
484 * If try_lock is not set yet, make sure all locked inodes are
485 * not in the AIL. If any are, set try_lock to be used later.
488 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
489 lp
= &ips
[j
]->i_itemp
->ili_item
;
490 if (lp
&& test_bit(XFS_LI_IN_AIL
, &lp
->li_flags
))
496 * If any of the previous locks we have locked is in the AIL,
497 * we must TRY to get the second and subsequent locks. If
498 * we can't get any, we must release all we have
502 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
506 /* try_lock means we have an inode locked that is in the AIL. */
508 if (xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
)))
512 * Unlock all previous guys and try again. xfs_iunlock will try
513 * to push the tail if the inode is in the AIL.
516 for (j
= i
- 1; j
>= 0; j
--) {
518 * Check to see if we've already unlocked this one. Not
519 * the first one going back, and the inode ptr is the
522 if (j
!= (i
- 1) && ips
[j
] == ips
[j
+ 1])
525 xfs_iunlock(ips
[j
], lock_mode
);
528 if ((attempts
% 5) == 0) {
529 delay(1); /* Don't just spin the CPU */
538 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
539 * the mmaplock or the ilock, but not more than one type at a time. If we lock
540 * more than one at a time, lockdep will report false positives saying we have
541 * violated locking orders. The iolock must be double-locked separately since
542 * we use i_rwsem for that. We now support taking one lock EXCL and the other
547 struct xfs_inode
*ip0
,
549 struct xfs_inode
*ip1
,
552 struct xfs_inode
*temp
;
555 struct xfs_log_item
*lp
;
557 ASSERT(hweight32(ip0_mode
) == 1);
558 ASSERT(hweight32(ip1_mode
) == 1);
559 ASSERT(!(ip0_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
560 ASSERT(!(ip1_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
561 ASSERT(!(ip0_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
562 !(ip0_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
563 ASSERT(!(ip1_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
564 !(ip1_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
565 ASSERT(!(ip1_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
566 !(ip0_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
567 ASSERT(!(ip0_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
568 !(ip1_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
570 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
572 if (ip0
->i_ino
> ip1
->i_ino
) {
576 mode_temp
= ip0_mode
;
578 ip1_mode
= mode_temp
;
582 xfs_ilock(ip0
, xfs_lock_inumorder(ip0_mode
, 0));
585 * If the first lock we have locked is in the AIL, we must TRY to get
586 * the second lock. If we can't get it, we must release the first one
589 lp
= &ip0
->i_itemp
->ili_item
;
590 if (lp
&& test_bit(XFS_LI_IN_AIL
, &lp
->li_flags
)) {
591 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(ip1_mode
, 1))) {
592 xfs_iunlock(ip0
, ip0_mode
);
593 if ((++attempts
% 5) == 0)
594 delay(1); /* Don't just spin the CPU */
598 xfs_ilock(ip1
, xfs_lock_inumorder(ip1_mode
, 1));
604 struct xfs_inode
*ip
)
606 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
607 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
610 prepare_to_wait_exclusive(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
611 if (xfs_isiflocked(ip
))
613 } while (!xfs_iflock_nowait(ip
));
615 finish_wait(wq
, &wait
.wq_entry
);
626 if (di_flags
& XFS_DIFLAG_ANY
) {
627 if (di_flags
& XFS_DIFLAG_REALTIME
)
628 flags
|= FS_XFLAG_REALTIME
;
629 if (di_flags
& XFS_DIFLAG_PREALLOC
)
630 flags
|= FS_XFLAG_PREALLOC
;
631 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
632 flags
|= FS_XFLAG_IMMUTABLE
;
633 if (di_flags
& XFS_DIFLAG_APPEND
)
634 flags
|= FS_XFLAG_APPEND
;
635 if (di_flags
& XFS_DIFLAG_SYNC
)
636 flags
|= FS_XFLAG_SYNC
;
637 if (di_flags
& XFS_DIFLAG_NOATIME
)
638 flags
|= FS_XFLAG_NOATIME
;
639 if (di_flags
& XFS_DIFLAG_NODUMP
)
640 flags
|= FS_XFLAG_NODUMP
;
641 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
642 flags
|= FS_XFLAG_RTINHERIT
;
643 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
644 flags
|= FS_XFLAG_PROJINHERIT
;
645 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
646 flags
|= FS_XFLAG_NOSYMLINKS
;
647 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
648 flags
|= FS_XFLAG_EXTSIZE
;
649 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
650 flags
|= FS_XFLAG_EXTSZINHERIT
;
651 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
652 flags
|= FS_XFLAG_NODEFRAG
;
653 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
654 flags
|= FS_XFLAG_FILESTREAM
;
657 if (di_flags2
& XFS_DIFLAG2_ANY
) {
658 if (di_flags2
& XFS_DIFLAG2_DAX
)
659 flags
|= FS_XFLAG_DAX
;
660 if (di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
661 flags
|= FS_XFLAG_COWEXTSIZE
;
665 flags
|= FS_XFLAG_HASATTR
;
672 struct xfs_inode
*ip
)
674 struct xfs_icdinode
*dic
= &ip
->i_d
;
676 return _xfs_dic2xflags(dic
->di_flags
, dic
->di_flags2
, XFS_IFORK_Q(ip
));
680 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
681 * is allowed, otherwise it has to be an exact match. If a CI match is found,
682 * ci_name->name will point to a the actual name (caller must free) or
683 * will be set to NULL if an exact match is found.
688 struct xfs_name
*name
,
690 struct xfs_name
*ci_name
)
695 trace_xfs_lookup(dp
, name
);
697 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
700 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
704 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
712 kmem_free(ci_name
->name
);
719 * Allocate an inode on disk and return a copy of its in-core version.
720 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
721 * appropriately within the inode. The uid and gid for the inode are
722 * set according to the contents of the given cred structure.
724 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
725 * has a free inode available, call xfs_iget() to obtain the in-core
726 * version of the allocated inode. Finally, fill in the inode and
727 * log its initial contents. In this case, ialloc_context would be
730 * If xfs_dialloc() does not have an available inode, it will replenish
731 * its supply by doing an allocation. Since we can only do one
732 * allocation within a transaction without deadlocks, we must commit
733 * the current transaction before returning the inode itself.
734 * In this case, therefore, we will set ialloc_context and return.
735 * The caller should then commit the current transaction, start a new
736 * transaction, and call xfs_ialloc() again to actually get the inode.
738 * To ensure that some other process does not grab the inode that
739 * was allocated during the first call to xfs_ialloc(), this routine
740 * also returns the [locked] bp pointing to the head of the freelist
741 * as ialloc_context. The caller should hold this buffer across
742 * the commit and pass it back into this routine on the second call.
744 * If we are allocating quota inodes, we do not have a parent inode
745 * to attach to or associate with (i.e. pip == NULL) because they
746 * are not linked into the directory structure - they are attached
747 * directly to the superblock - and so have no parent.
757 xfs_buf_t
**ialloc_context
,
760 struct xfs_mount
*mp
= tp
->t_mountp
;
765 struct timespec64 tv
;
769 * Call the space management code to pick
770 * the on-disk inode to be allocated.
772 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
,
773 ialloc_context
, &ino
);
776 if (*ialloc_context
|| ino
== NULLFSINO
) {
780 ASSERT(*ialloc_context
== NULL
);
783 * Protect against obviously corrupt allocation btree records. Later
784 * xfs_iget checks will catch re-allocation of other active in-memory
785 * and on-disk inodes. If we don't catch reallocating the parent inode
786 * here we will deadlock in xfs_iget() so we have to do these checks
789 if ((pip
&& ino
== pip
->i_ino
) || !xfs_verify_dir_ino(mp
, ino
)) {
790 xfs_alert(mp
, "Allocated a known in-use inode 0x%llx!", ino
);
791 return -EFSCORRUPTED
;
795 * Get the in-core inode with the lock held exclusively.
796 * This is because we're setting fields here we need
797 * to prevent others from looking at until we're done.
799 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
800 XFS_ILOCK_EXCL
, &ip
);
805 inode
->i_mode
= mode
;
806 set_nlink(inode
, nlink
);
807 inode
->i_uid
= current_fsuid();
808 inode
->i_rdev
= rdev
;
809 ip
->i_d
.di_projid
= prid
;
811 if (pip
&& XFS_INHERIT_GID(pip
)) {
812 inode
->i_gid
= VFS_I(pip
)->i_gid
;
813 if ((VFS_I(pip
)->i_mode
& S_ISGID
) && S_ISDIR(mode
))
814 inode
->i_mode
|= S_ISGID
;
816 inode
->i_gid
= current_fsgid();
820 * If the group ID of the new file does not match the effective group
821 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
822 * (and only if the irix_sgid_inherit compatibility variable is set).
824 if (irix_sgid_inherit
&&
825 (inode
->i_mode
& S_ISGID
) && !in_group_p(inode
->i_gid
))
826 inode
->i_mode
&= ~S_ISGID
;
829 ip
->i_df
.if_nextents
= 0;
830 ASSERT(ip
->i_d
.di_nblocks
== 0);
832 tv
= current_time(inode
);
837 ip
->i_d
.di_extsize
= 0;
838 ip
->i_d
.di_dmevmask
= 0;
839 ip
->i_d
.di_dmstate
= 0;
840 ip
->i_d
.di_flags
= 0;
842 if (xfs_sb_version_has_v3inode(&mp
->m_sb
)) {
843 inode_set_iversion(inode
, 1);
844 ip
->i_d
.di_flags2
= 0;
845 ip
->i_d
.di_cowextsize
= 0;
846 ip
->i_d
.di_crtime
= tv
;
849 flags
= XFS_ILOG_CORE
;
850 switch (mode
& S_IFMT
) {
855 ip
->i_df
.if_format
= XFS_DINODE_FMT_DEV
;
856 ip
->i_df
.if_flags
= 0;
857 flags
|= XFS_ILOG_DEV
;
861 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
865 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
866 di_flags
|= XFS_DIFLAG_RTINHERIT
;
867 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
868 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
869 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
871 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
872 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
873 } else if (S_ISREG(mode
)) {
874 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
875 di_flags
|= XFS_DIFLAG_REALTIME
;
876 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
877 di_flags
|= XFS_DIFLAG_EXTSIZE
;
878 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
881 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
883 di_flags
|= XFS_DIFLAG_NOATIME
;
884 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
886 di_flags
|= XFS_DIFLAG_NODUMP
;
887 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
889 di_flags
|= XFS_DIFLAG_SYNC
;
890 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
891 xfs_inherit_nosymlinks
)
892 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
893 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
894 xfs_inherit_nodefrag
)
895 di_flags
|= XFS_DIFLAG_NODEFRAG
;
896 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
897 di_flags
|= XFS_DIFLAG_FILESTREAM
;
899 ip
->i_d
.di_flags
|= di_flags
;
901 if (pip
&& (pip
->i_d
.di_flags2
& XFS_DIFLAG2_ANY
)) {
902 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) {
903 ip
->i_d
.di_flags2
|= XFS_DIFLAG2_COWEXTSIZE
;
904 ip
->i_d
.di_cowextsize
= pip
->i_d
.di_cowextsize
;
906 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
)
907 ip
->i_d
.di_flags2
|= XFS_DIFLAG2_DAX
;
911 ip
->i_df
.if_format
= XFS_DINODE_FMT_EXTENTS
;
912 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
913 ip
->i_df
.if_bytes
= 0;
914 ip
->i_df
.if_u1
.if_root
= NULL
;
921 * Log the new values stuffed into the inode.
923 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
924 xfs_trans_log_inode(tp
, ip
, flags
);
926 /* now that we have an i_mode we can setup the inode structure */
934 * Allocates a new inode from disk and return a pointer to the
935 * incore copy. This routine will internally commit the current
936 * transaction and allocate a new one if the Space Manager needed
937 * to do an allocation to replenish the inode free-list.
939 * This routine is designed to be called from xfs_create and
945 xfs_trans_t
**tpp
, /* input: current transaction;
946 output: may be a new transaction. */
947 xfs_inode_t
*dp
, /* directory within whose allocate
952 prid_t prid
, /* project id */
953 xfs_inode_t
**ipp
) /* pointer to inode; it will be
958 xfs_buf_t
*ialloc_context
= NULL
;
964 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
967 * xfs_ialloc will return a pointer to an incore inode if
968 * the Space Manager has an available inode on the free
969 * list. Otherwise, it will do an allocation and replenish
970 * the freelist. Since we can only do one allocation per
971 * transaction without deadlocks, we will need to commit the
972 * current transaction and start a new one. We will then
973 * need to call xfs_ialloc again to get the inode.
975 * If xfs_ialloc did an allocation to replenish the freelist,
976 * it returns the bp containing the head of the freelist as
977 * ialloc_context. We will hold a lock on it across the
978 * transaction commit so that no other process can steal
979 * the inode(s) that we've just allocated.
981 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, &ialloc_context
,
985 * Return an error if we were unable to allocate a new inode.
986 * This should only happen if we run out of space on disk or
987 * encounter a disk error.
993 if (!ialloc_context
&& !ip
) {
999 * If the AGI buffer is non-NULL, then we were unable to get an
1000 * inode in one operation. We need to commit the current
1001 * transaction and call xfs_ialloc() again. It is guaranteed
1002 * to succeed the second time.
1004 if (ialloc_context
) {
1006 * Normally, xfs_trans_commit releases all the locks.
1007 * We call bhold to hang on to the ialloc_context across
1008 * the commit. Holding this buffer prevents any other
1009 * processes from doing any allocations in this
1012 xfs_trans_bhold(tp
, ialloc_context
);
1015 * We want the quota changes to be associated with the next
1016 * transaction, NOT this one. So, detach the dqinfo from this
1017 * and attach it to the next transaction.
1022 dqinfo
= (void *)tp
->t_dqinfo
;
1023 tp
->t_dqinfo
= NULL
;
1024 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
1025 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
1028 code
= xfs_trans_roll(&tp
);
1031 * Re-attach the quota info that we detached from prev trx.
1034 tp
->t_dqinfo
= dqinfo
;
1035 tp
->t_flags
|= tflags
;
1039 xfs_buf_relse(ialloc_context
);
1044 xfs_trans_bjoin(tp
, ialloc_context
);
1047 * Call ialloc again. Since we've locked out all
1048 * other allocations in this allocation group,
1049 * this call should always succeed.
1051 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1052 &ialloc_context
, &ip
);
1055 * If we get an error at this point, return to the caller
1056 * so that the current transaction can be aborted.
1063 ASSERT(!ialloc_context
&& ip
);
1074 * Decrement the link count on an inode & log the change. If this causes the
1075 * link count to go to zero, move the inode to AGI unlinked list so that it can
1076 * be freed when the last active reference goes away via xfs_inactive().
1078 static int /* error */
1083 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1085 drop_nlink(VFS_I(ip
));
1086 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1088 if (VFS_I(ip
)->i_nlink
)
1091 return xfs_iunlink(tp
, ip
);
1095 * Increment the link count on an inode & log the change.
1102 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1104 inc_nlink(VFS_I(ip
));
1105 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1111 struct xfs_name
*name
,
1116 int is_dir
= S_ISDIR(mode
);
1117 struct xfs_mount
*mp
= dp
->i_mount
;
1118 struct xfs_inode
*ip
= NULL
;
1119 struct xfs_trans
*tp
= NULL
;
1121 bool unlock_dp_on_error
= false;
1123 struct xfs_dquot
*udqp
= NULL
;
1124 struct xfs_dquot
*gdqp
= NULL
;
1125 struct xfs_dquot
*pdqp
= NULL
;
1126 struct xfs_trans_res
*tres
;
1129 trace_xfs_create(dp
, name
);
1131 if (XFS_FORCED_SHUTDOWN(mp
))
1134 prid
= xfs_get_initial_prid(dp
);
1137 * Make sure that we have allocated dquot(s) on disk.
1139 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1140 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1141 &udqp
, &gdqp
, &pdqp
);
1146 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1147 tres
= &M_RES(mp
)->tr_mkdir
;
1149 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1150 tres
= &M_RES(mp
)->tr_create
;
1154 * Initially assume that the file does not exist and
1155 * reserve the resources for that case. If that is not
1156 * the case we'll drop the one we have and get a more
1157 * appropriate transaction later.
1159 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1160 if (error
== -ENOSPC
) {
1161 /* flush outstanding delalloc blocks and retry */
1162 xfs_flush_inodes(mp
);
1163 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1166 goto out_release_inode
;
1168 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1169 unlock_dp_on_error
= true;
1172 * Reserve disk quota and the inode.
1174 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1175 pdqp
, resblks
, 1, 0);
1177 goto out_trans_cancel
;
1180 * A newly created regular or special file just has one directory
1181 * entry pointing to them, but a directory also the "." entry
1182 * pointing to itself.
1184 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
, prid
, &ip
);
1186 goto out_trans_cancel
;
1189 * Now we join the directory inode to the transaction. We do not do it
1190 * earlier because xfs_dir_ialloc might commit the previous transaction
1191 * (and release all the locks). An error from here on will result in
1192 * the transaction cancel unlocking dp so don't do it explicitly in the
1195 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1196 unlock_dp_on_error
= false;
1198 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1199 resblks
- XFS_IALLOC_SPACE_RES(mp
));
1201 ASSERT(error
!= -ENOSPC
);
1202 goto out_trans_cancel
;
1204 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1205 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1208 error
= xfs_dir_init(tp
, ip
, dp
);
1210 goto out_trans_cancel
;
1212 xfs_bumplink(tp
, dp
);
1216 * If this is a synchronous mount, make sure that the
1217 * create transaction goes to disk before returning to
1220 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1221 xfs_trans_set_sync(tp
);
1224 * Attach the dquot(s) to the inodes and modify them incore.
1225 * These ids of the inode couldn't have changed since the new
1226 * inode has been locked ever since it was created.
1228 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1230 error
= xfs_trans_commit(tp
);
1232 goto out_release_inode
;
1234 xfs_qm_dqrele(udqp
);
1235 xfs_qm_dqrele(gdqp
);
1236 xfs_qm_dqrele(pdqp
);
1242 xfs_trans_cancel(tp
);
1245 * Wait until after the current transaction is aborted to finish the
1246 * setup of the inode and release the inode. This prevents recursive
1247 * transactions and deadlocks from xfs_inactive.
1250 xfs_finish_inode_setup(ip
);
1254 xfs_qm_dqrele(udqp
);
1255 xfs_qm_dqrele(gdqp
);
1256 xfs_qm_dqrele(pdqp
);
1258 if (unlock_dp_on_error
)
1259 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1265 struct xfs_inode
*dp
,
1267 struct xfs_inode
**ipp
)
1269 struct xfs_mount
*mp
= dp
->i_mount
;
1270 struct xfs_inode
*ip
= NULL
;
1271 struct xfs_trans
*tp
= NULL
;
1274 struct xfs_dquot
*udqp
= NULL
;
1275 struct xfs_dquot
*gdqp
= NULL
;
1276 struct xfs_dquot
*pdqp
= NULL
;
1277 struct xfs_trans_res
*tres
;
1280 if (XFS_FORCED_SHUTDOWN(mp
))
1283 prid
= xfs_get_initial_prid(dp
);
1286 * Make sure that we have allocated dquot(s) on disk.
1288 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1289 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1290 &udqp
, &gdqp
, &pdqp
);
1294 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1295 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1297 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1299 goto out_release_inode
;
1301 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1302 pdqp
, resblks
, 1, 0);
1304 goto out_trans_cancel
;
1306 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 0, 0, prid
, &ip
);
1308 goto out_trans_cancel
;
1310 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1311 xfs_trans_set_sync(tp
);
1314 * Attach the dquot(s) to the inodes and modify them incore.
1315 * These ids of the inode couldn't have changed since the new
1316 * inode has been locked ever since it was created.
1318 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1320 error
= xfs_iunlink(tp
, ip
);
1322 goto out_trans_cancel
;
1324 error
= xfs_trans_commit(tp
);
1326 goto out_release_inode
;
1328 xfs_qm_dqrele(udqp
);
1329 xfs_qm_dqrele(gdqp
);
1330 xfs_qm_dqrele(pdqp
);
1336 xfs_trans_cancel(tp
);
1339 * Wait until after the current transaction is aborted to finish the
1340 * setup of the inode and release the inode. This prevents recursive
1341 * transactions and deadlocks from xfs_inactive.
1344 xfs_finish_inode_setup(ip
);
1348 xfs_qm_dqrele(udqp
);
1349 xfs_qm_dqrele(gdqp
);
1350 xfs_qm_dqrele(pdqp
);
1359 struct xfs_name
*target_name
)
1361 xfs_mount_t
*mp
= tdp
->i_mount
;
1366 trace_xfs_link(tdp
, target_name
);
1368 ASSERT(!S_ISDIR(VFS_I(sip
)->i_mode
));
1370 if (XFS_FORCED_SHUTDOWN(mp
))
1373 error
= xfs_qm_dqattach(sip
);
1377 error
= xfs_qm_dqattach(tdp
);
1381 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1382 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, resblks
, 0, 0, &tp
);
1383 if (error
== -ENOSPC
) {
1385 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, 0, 0, 0, &tp
);
1390 xfs_lock_two_inodes(sip
, XFS_ILOCK_EXCL
, tdp
, XFS_ILOCK_EXCL
);
1392 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1393 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1396 * If we are using project inheritance, we only allow hard link
1397 * creation in our tree when the project IDs are the same; else
1398 * the tree quota mechanism could be circumvented.
1400 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1401 tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
)) {
1407 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1413 * Handle initial link state of O_TMPFILE inode
1415 if (VFS_I(sip
)->i_nlink
== 0) {
1416 error
= xfs_iunlink_remove(tp
, sip
);
1421 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1425 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1426 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1428 xfs_bumplink(tp
, sip
);
1431 * If this is a synchronous mount, make sure that the
1432 * link transaction goes to disk before returning to
1435 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1436 xfs_trans_set_sync(tp
);
1438 return xfs_trans_commit(tp
);
1441 xfs_trans_cancel(tp
);
1446 /* Clear the reflink flag and the cowblocks tag if possible. */
1448 xfs_itruncate_clear_reflink_flags(
1449 struct xfs_inode
*ip
)
1451 struct xfs_ifork
*dfork
;
1452 struct xfs_ifork
*cfork
;
1454 if (!xfs_is_reflink_inode(ip
))
1456 dfork
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1457 cfork
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
1458 if (dfork
->if_bytes
== 0 && cfork
->if_bytes
== 0)
1459 ip
->i_d
.di_flags2
&= ~XFS_DIFLAG2_REFLINK
;
1460 if (cfork
->if_bytes
== 0)
1461 xfs_inode_clear_cowblocks_tag(ip
);
1465 * Free up the underlying blocks past new_size. The new size must be smaller
1466 * than the current size. This routine can be used both for the attribute and
1467 * data fork, and does not modify the inode size, which is left to the caller.
1469 * The transaction passed to this routine must have made a permanent log
1470 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1471 * given transaction and start new ones, so make sure everything involved in
1472 * the transaction is tidy before calling here. Some transaction will be
1473 * returned to the caller to be committed. The incoming transaction must
1474 * already include the inode, and both inode locks must be held exclusively.
1475 * The inode must also be "held" within the transaction. On return the inode
1476 * will be "held" within the returned transaction. This routine does NOT
1477 * require any disk space to be reserved for it within the transaction.
1479 * If we get an error, we must return with the inode locked and linked into the
1480 * current transaction. This keeps things simple for the higher level code,
1481 * because it always knows that the inode is locked and held in the transaction
1482 * that returns to it whether errors occur or not. We don't mark the inode
1483 * dirty on error so that transactions can be easily aborted if possible.
1486 xfs_itruncate_extents_flags(
1487 struct xfs_trans
**tpp
,
1488 struct xfs_inode
*ip
,
1490 xfs_fsize_t new_size
,
1493 struct xfs_mount
*mp
= ip
->i_mount
;
1494 struct xfs_trans
*tp
= *tpp
;
1495 xfs_fileoff_t first_unmap_block
;
1496 xfs_filblks_t unmap_len
;
1499 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1500 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1501 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1502 ASSERT(new_size
<= XFS_ISIZE(ip
));
1503 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1504 ASSERT(ip
->i_itemp
!= NULL
);
1505 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1506 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1508 trace_xfs_itruncate_extents_start(ip
, new_size
);
1510 flags
|= xfs_bmapi_aflag(whichfork
);
1513 * Since it is possible for space to become allocated beyond
1514 * the end of the file (in a crash where the space is allocated
1515 * but the inode size is not yet updated), simply remove any
1516 * blocks which show up between the new EOF and the maximum
1517 * possible file size.
1519 * We have to free all the blocks to the bmbt maximum offset, even if
1520 * the page cache can't scale that far.
1522 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1523 if (first_unmap_block
>= XFS_MAX_FILEOFF
) {
1524 WARN_ON_ONCE(first_unmap_block
> XFS_MAX_FILEOFF
);
1528 unmap_len
= XFS_MAX_FILEOFF
- first_unmap_block
+ 1;
1529 while (unmap_len
> 0) {
1530 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
);
1531 error
= __xfs_bunmapi(tp
, ip
, first_unmap_block
, &unmap_len
,
1532 flags
, XFS_ITRUNC_MAX_EXTENTS
);
1537 * Duplicate the transaction that has the permanent
1538 * reservation and commit the old transaction.
1540 error
= xfs_defer_finish(&tp
);
1544 error
= xfs_trans_roll_inode(&tp
, ip
);
1549 if (whichfork
== XFS_DATA_FORK
) {
1550 /* Remove all pending CoW reservations. */
1551 error
= xfs_reflink_cancel_cow_blocks(ip
, &tp
,
1552 first_unmap_block
, XFS_MAX_FILEOFF
, true);
1556 xfs_itruncate_clear_reflink_flags(ip
);
1560 * Always re-log the inode so that our permanent transaction can keep
1561 * on rolling it forward in the log.
1563 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1565 trace_xfs_itruncate_extents_end(ip
, new_size
);
1576 xfs_mount_t
*mp
= ip
->i_mount
;
1579 if (!S_ISREG(VFS_I(ip
)->i_mode
) || (VFS_I(ip
)->i_mode
== 0))
1582 /* If this is a read-only mount, don't do this (would generate I/O) */
1583 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1586 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1590 * If we previously truncated this file and removed old data
1591 * in the process, we want to initiate "early" writeout on
1592 * the last close. This is an attempt to combat the notorious
1593 * NULL files problem which is particularly noticeable from a
1594 * truncate down, buffered (re-)write (delalloc), followed by
1595 * a crash. What we are effectively doing here is
1596 * significantly reducing the time window where we'd otherwise
1597 * be exposed to that problem.
1599 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1601 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1602 if (ip
->i_delayed_blks
> 0) {
1603 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1610 if (VFS_I(ip
)->i_nlink
== 0)
1613 if (xfs_can_free_eofblocks(ip
, false)) {
1616 * Check if the inode is being opened, written and closed
1617 * frequently and we have delayed allocation blocks outstanding
1618 * (e.g. streaming writes from the NFS server), truncating the
1619 * blocks past EOF will cause fragmentation to occur.
1621 * In this case don't do the truncation, but we have to be
1622 * careful how we detect this case. Blocks beyond EOF show up as
1623 * i_delayed_blks even when the inode is clean, so we need to
1624 * truncate them away first before checking for a dirty release.
1625 * Hence on the first dirty close we will still remove the
1626 * speculative allocation, but after that we will leave it in
1629 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1632 * If we can't get the iolock just skip truncating the blocks
1633 * past EOF because we could deadlock with the mmap_lock
1634 * otherwise. We'll get another chance to drop them once the
1635 * last reference to the inode is dropped, so we'll never leak
1636 * blocks permanently.
1638 if (xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1639 error
= xfs_free_eofblocks(ip
);
1640 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1645 /* delalloc blocks after truncation means it really is dirty */
1646 if (ip
->i_delayed_blks
)
1647 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1653 * xfs_inactive_truncate
1655 * Called to perform a truncate when an inode becomes unlinked.
1658 xfs_inactive_truncate(
1659 struct xfs_inode
*ip
)
1661 struct xfs_mount
*mp
= ip
->i_mount
;
1662 struct xfs_trans
*tp
;
1665 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
1667 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1670 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1671 xfs_trans_ijoin(tp
, ip
, 0);
1674 * Log the inode size first to prevent stale data exposure in the event
1675 * of a system crash before the truncate completes. See the related
1676 * comment in xfs_vn_setattr_size() for details.
1678 ip
->i_d
.di_size
= 0;
1679 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1681 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1683 goto error_trans_cancel
;
1685 ASSERT(ip
->i_df
.if_nextents
== 0);
1687 error
= xfs_trans_commit(tp
);
1691 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1695 xfs_trans_cancel(tp
);
1697 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1702 * xfs_inactive_ifree()
1704 * Perform the inode free when an inode is unlinked.
1708 struct xfs_inode
*ip
)
1710 struct xfs_mount
*mp
= ip
->i_mount
;
1711 struct xfs_trans
*tp
;
1715 * We try to use a per-AG reservation for any block needed by the finobt
1716 * tree, but as the finobt feature predates the per-AG reservation
1717 * support a degraded file system might not have enough space for the
1718 * reservation at mount time. In that case try to dip into the reserved
1721 * Send a warning if the reservation does happen to fail, as the inode
1722 * now remains allocated and sits on the unlinked list until the fs is
1725 if (unlikely(mp
->m_finobt_nores
)) {
1726 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
,
1727 XFS_IFREE_SPACE_RES(mp
), 0, XFS_TRANS_RESERVE
,
1730 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
, 0, 0, 0, &tp
);
1733 if (error
== -ENOSPC
) {
1734 xfs_warn_ratelimited(mp
,
1735 "Failed to remove inode(s) from unlinked list. "
1736 "Please free space, unmount and run xfs_repair.");
1738 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1743 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1744 xfs_trans_ijoin(tp
, ip
, 0);
1746 error
= xfs_ifree(tp
, ip
);
1749 * If we fail to free the inode, shut down. The cancel
1750 * might do that, we need to make sure. Otherwise the
1751 * inode might be lost for a long time or forever.
1753 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1754 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1756 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1758 xfs_trans_cancel(tp
);
1759 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1764 * Credit the quota account(s). The inode is gone.
1766 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1769 * Just ignore errors at this point. There is nothing we can do except
1770 * to try to keep going. Make sure it's not a silent error.
1772 error
= xfs_trans_commit(tp
);
1774 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1777 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1784 * This is called when the vnode reference count for the vnode
1785 * goes to zero. If the file has been unlinked, then it must
1786 * now be truncated. Also, we clear all of the read-ahead state
1787 * kept for the inode here since the file is now closed.
1793 struct xfs_mount
*mp
;
1798 * If the inode is already free, then there can be nothing
1801 if (VFS_I(ip
)->i_mode
== 0) {
1802 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1807 ASSERT(!xfs_iflags_test(ip
, XFS_IRECOVERY
));
1809 /* If this is a read-only mount, don't do this (would generate I/O) */
1810 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1813 /* Try to clean out the cow blocks if there are any. */
1814 if (xfs_inode_has_cow_data(ip
))
1815 xfs_reflink_cancel_cow_range(ip
, 0, NULLFILEOFF
, true);
1817 if (VFS_I(ip
)->i_nlink
!= 0) {
1819 * force is true because we are evicting an inode from the
1820 * cache. Post-eof blocks must be freed, lest we end up with
1821 * broken free space accounting.
1823 * Note: don't bother with iolock here since lockdep complains
1824 * about acquiring it in reclaim context. We have the only
1825 * reference to the inode at this point anyways.
1827 if (xfs_can_free_eofblocks(ip
, true))
1828 xfs_free_eofblocks(ip
);
1833 if (S_ISREG(VFS_I(ip
)->i_mode
) &&
1834 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1835 ip
->i_df
.if_nextents
> 0 || ip
->i_delayed_blks
> 0))
1838 error
= xfs_qm_dqattach(ip
);
1842 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1843 error
= xfs_inactive_symlink(ip
);
1845 error
= xfs_inactive_truncate(ip
);
1850 * If there are attributes associated with the file then blow them away
1851 * now. The code calls a routine that recursively deconstructs the
1852 * attribute fork. If also blows away the in-core attribute fork.
1854 if (XFS_IFORK_Q(ip
)) {
1855 error
= xfs_attr_inactive(ip
);
1861 ASSERT(ip
->i_d
.di_forkoff
== 0);
1866 error
= xfs_inactive_ifree(ip
);
1871 * Release the dquots held by inode, if any.
1873 xfs_qm_dqdetach(ip
);
1877 * In-Core Unlinked List Lookups
1878 * =============================
1880 * Every inode is supposed to be reachable from some other piece of metadata
1881 * with the exception of the root directory. Inodes with a connection to a
1882 * file descriptor but not linked from anywhere in the on-disk directory tree
1883 * are collectively known as unlinked inodes, though the filesystem itself
1884 * maintains links to these inodes so that on-disk metadata are consistent.
1886 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1887 * header contains a number of buckets that point to an inode, and each inode
1888 * record has a pointer to the next inode in the hash chain. This
1889 * singly-linked list causes scaling problems in the iunlink remove function
1890 * because we must walk that list to find the inode that points to the inode
1891 * being removed from the unlinked hash bucket list.
1893 * What if we modelled the unlinked list as a collection of records capturing
1894 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1895 * have a fast way to look up unlinked list predecessors, which avoids the
1896 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1899 * Because this is a backref cache, we ignore operational failures since the
1900 * iunlink code can fall back to the slow bucket walk. The only errors that
1901 * should bubble out are for obviously incorrect situations.
1903 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1904 * access or have otherwise provided for concurrency control.
1907 /* Capture a "X.next_unlinked = Y" relationship. */
1908 struct xfs_iunlink
{
1909 struct rhash_head iu_rhash_head
;
1910 xfs_agino_t iu_agino
; /* X */
1911 xfs_agino_t iu_next_unlinked
; /* Y */
1914 /* Unlinked list predecessor lookup hashtable construction */
1916 xfs_iunlink_obj_cmpfn(
1917 struct rhashtable_compare_arg
*arg
,
1920 const xfs_agino_t
*key
= arg
->key
;
1921 const struct xfs_iunlink
*iu
= obj
;
1923 if (iu
->iu_next_unlinked
!= *key
)
1928 static const struct rhashtable_params xfs_iunlink_hash_params
= {
1929 .min_size
= XFS_AGI_UNLINKED_BUCKETS
,
1930 .key_len
= sizeof(xfs_agino_t
),
1931 .key_offset
= offsetof(struct xfs_iunlink
,
1933 .head_offset
= offsetof(struct xfs_iunlink
, iu_rhash_head
),
1934 .automatic_shrinking
= true,
1935 .obj_cmpfn
= xfs_iunlink_obj_cmpfn
,
1939 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1940 * relation is found.
1943 xfs_iunlink_lookup_backref(
1944 struct xfs_perag
*pag
,
1947 struct xfs_iunlink
*iu
;
1949 iu
= rhashtable_lookup_fast(&pag
->pagi_unlinked_hash
, &agino
,
1950 xfs_iunlink_hash_params
);
1951 return iu
? iu
->iu_agino
: NULLAGINO
;
1955 * Take ownership of an iunlink cache entry and insert it into the hash table.
1956 * If successful, the entry will be owned by the cache; if not, it is freed.
1957 * Either way, the caller does not own @iu after this call.
1960 xfs_iunlink_insert_backref(
1961 struct xfs_perag
*pag
,
1962 struct xfs_iunlink
*iu
)
1966 error
= rhashtable_insert_fast(&pag
->pagi_unlinked_hash
,
1967 &iu
->iu_rhash_head
, xfs_iunlink_hash_params
);
1969 * Fail loudly if there already was an entry because that's a sign of
1970 * corruption of in-memory data. Also fail loudly if we see an error
1971 * code we didn't anticipate from the rhashtable code. Currently we
1972 * only anticipate ENOMEM.
1975 WARN(error
!= -ENOMEM
, "iunlink cache insert error %d", error
);
1979 * Absorb any runtime errors that aren't a result of corruption because
1980 * this is a cache and we can always fall back to bucket list scanning.
1982 if (error
!= 0 && error
!= -EEXIST
)
1987 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1989 xfs_iunlink_add_backref(
1990 struct xfs_perag
*pag
,
1991 xfs_agino_t prev_agino
,
1992 xfs_agino_t this_agino
)
1994 struct xfs_iunlink
*iu
;
1996 if (XFS_TEST_ERROR(false, pag
->pag_mount
, XFS_ERRTAG_IUNLINK_FALLBACK
))
1999 iu
= kmem_zalloc(sizeof(*iu
), KM_NOFS
);
2000 iu
->iu_agino
= prev_agino
;
2001 iu
->iu_next_unlinked
= this_agino
;
2003 return xfs_iunlink_insert_backref(pag
, iu
);
2007 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2008 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
2009 * wasn't any such entry then we don't bother.
2012 xfs_iunlink_change_backref(
2013 struct xfs_perag
*pag
,
2015 xfs_agino_t next_unlinked
)
2017 struct xfs_iunlink
*iu
;
2020 /* Look up the old entry; if there wasn't one then exit. */
2021 iu
= rhashtable_lookup_fast(&pag
->pagi_unlinked_hash
, &agino
,
2022 xfs_iunlink_hash_params
);
2027 * Remove the entry. This shouldn't ever return an error, but if we
2028 * couldn't remove the old entry we don't want to add it again to the
2029 * hash table, and if the entry disappeared on us then someone's
2030 * violated the locking rules and we need to fail loudly. Either way
2031 * we cannot remove the inode because internal state is or would have
2034 error
= rhashtable_remove_fast(&pag
->pagi_unlinked_hash
,
2035 &iu
->iu_rhash_head
, xfs_iunlink_hash_params
);
2039 /* If there is no new next entry just free our item and return. */
2040 if (next_unlinked
== NULLAGINO
) {
2045 /* Update the entry and re-add it to the hash table. */
2046 iu
->iu_next_unlinked
= next_unlinked
;
2047 return xfs_iunlink_insert_backref(pag
, iu
);
2050 /* Set up the in-core predecessor structures. */
2053 struct xfs_perag
*pag
)
2055 return rhashtable_init(&pag
->pagi_unlinked_hash
,
2056 &xfs_iunlink_hash_params
);
2059 /* Free the in-core predecessor structures. */
2061 xfs_iunlink_free_item(
2065 struct xfs_iunlink
*iu
= ptr
;
2066 bool *freed_anything
= arg
;
2068 *freed_anything
= true;
2073 xfs_iunlink_destroy(
2074 struct xfs_perag
*pag
)
2076 bool freed_anything
= false;
2078 rhashtable_free_and_destroy(&pag
->pagi_unlinked_hash
,
2079 xfs_iunlink_free_item
, &freed_anything
);
2081 ASSERT(freed_anything
== false || XFS_FORCED_SHUTDOWN(pag
->pag_mount
));
2085 * Point the AGI unlinked bucket at an inode and log the results. The caller
2086 * is responsible for validating the old value.
2089 xfs_iunlink_update_bucket(
2090 struct xfs_trans
*tp
,
2091 xfs_agnumber_t agno
,
2092 struct xfs_buf
*agibp
,
2093 unsigned int bucket_index
,
2094 xfs_agino_t new_agino
)
2096 struct xfs_agi
*agi
= agibp
->b_addr
;
2097 xfs_agino_t old_value
;
2100 ASSERT(xfs_verify_agino_or_null(tp
->t_mountp
, agno
, new_agino
));
2102 old_value
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2103 trace_xfs_iunlink_update_bucket(tp
->t_mountp
, agno
, bucket_index
,
2104 old_value
, new_agino
);
2107 * We should never find the head of the list already set to the value
2108 * passed in because either we're adding or removing ourselves from the
2111 if (old_value
== new_agino
) {
2112 xfs_buf_mark_corrupt(agibp
);
2113 return -EFSCORRUPTED
;
2116 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(new_agino
);
2117 offset
= offsetof(struct xfs_agi
, agi_unlinked
) +
2118 (sizeof(xfs_agino_t
) * bucket_index
);
2119 xfs_trans_log_buf(tp
, agibp
, offset
, offset
+ sizeof(xfs_agino_t
) - 1);
2123 /* Set an on-disk inode's next_unlinked pointer. */
2125 xfs_iunlink_update_dinode(
2126 struct xfs_trans
*tp
,
2127 xfs_agnumber_t agno
,
2129 struct xfs_buf
*ibp
,
2130 struct xfs_dinode
*dip
,
2131 struct xfs_imap
*imap
,
2132 xfs_agino_t next_agino
)
2134 struct xfs_mount
*mp
= tp
->t_mountp
;
2137 ASSERT(xfs_verify_agino_or_null(mp
, agno
, next_agino
));
2139 trace_xfs_iunlink_update_dinode(mp
, agno
, agino
,
2140 be32_to_cpu(dip
->di_next_unlinked
), next_agino
);
2142 dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2143 offset
= imap
->im_boffset
+
2144 offsetof(struct xfs_dinode
, di_next_unlinked
);
2146 /* need to recalc the inode CRC if appropriate */
2147 xfs_dinode_calc_crc(mp
, dip
);
2148 xfs_trans_inode_buf(tp
, ibp
);
2149 xfs_trans_log_buf(tp
, ibp
, offset
, offset
+ sizeof(xfs_agino_t
) - 1);
2150 xfs_inobp_check(mp
, ibp
);
2153 /* Set an in-core inode's unlinked pointer and return the old value. */
2155 xfs_iunlink_update_inode(
2156 struct xfs_trans
*tp
,
2157 struct xfs_inode
*ip
,
2158 xfs_agnumber_t agno
,
2159 xfs_agino_t next_agino
,
2160 xfs_agino_t
*old_next_agino
)
2162 struct xfs_mount
*mp
= tp
->t_mountp
;
2163 struct xfs_dinode
*dip
;
2164 struct xfs_buf
*ibp
;
2165 xfs_agino_t old_value
;
2168 ASSERT(xfs_verify_agino_or_null(mp
, agno
, next_agino
));
2170 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
, 0);
2174 /* Make sure the old pointer isn't garbage. */
2175 old_value
= be32_to_cpu(dip
->di_next_unlinked
);
2176 if (!xfs_verify_agino_or_null(mp
, agno
, old_value
)) {
2177 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
, dip
,
2178 sizeof(*dip
), __this_address
);
2179 error
= -EFSCORRUPTED
;
2184 * Since we're updating a linked list, we should never find that the
2185 * current pointer is the same as the new value, unless we're
2186 * terminating the list.
2188 *old_next_agino
= old_value
;
2189 if (old_value
== next_agino
) {
2190 if (next_agino
!= NULLAGINO
) {
2191 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
,
2192 dip
, sizeof(*dip
), __this_address
);
2193 error
= -EFSCORRUPTED
;
2198 /* Ok, update the new pointer. */
2199 xfs_iunlink_update_dinode(tp
, agno
, XFS_INO_TO_AGINO(mp
, ip
->i_ino
),
2200 ibp
, dip
, &ip
->i_imap
, next_agino
);
2203 xfs_trans_brelse(tp
, ibp
);
2208 * This is called when the inode's link count has gone to 0 or we are creating
2209 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2211 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2212 * list when the inode is freed.
2216 struct xfs_trans
*tp
,
2217 struct xfs_inode
*ip
)
2219 struct xfs_mount
*mp
= tp
->t_mountp
;
2220 struct xfs_agi
*agi
;
2221 struct xfs_buf
*agibp
;
2222 xfs_agino_t next_agino
;
2223 xfs_agnumber_t agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2224 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2225 short bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2228 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2229 ASSERT(VFS_I(ip
)->i_mode
!= 0);
2230 trace_xfs_iunlink(ip
);
2232 /* Get the agi buffer first. It ensures lock ordering on the list. */
2233 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2236 agi
= agibp
->b_addr
;
2239 * Get the index into the agi hash table for the list this inode will
2240 * go on. Make sure the pointer isn't garbage and that this inode
2241 * isn't already on the list.
2243 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2244 if (next_agino
== agino
||
2245 !xfs_verify_agino_or_null(mp
, agno
, next_agino
)) {
2246 xfs_buf_mark_corrupt(agibp
);
2247 return -EFSCORRUPTED
;
2250 if (next_agino
!= NULLAGINO
) {
2251 struct xfs_perag
*pag
;
2252 xfs_agino_t old_agino
;
2255 * There is already another inode in the bucket, so point this
2256 * inode to the current head of the list.
2258 error
= xfs_iunlink_update_inode(tp
, ip
, agno
, next_agino
,
2262 ASSERT(old_agino
== NULLAGINO
);
2265 * agino has been unlinked, add a backref from the next inode
2268 pag
= xfs_perag_get(mp
, agno
);
2269 error
= xfs_iunlink_add_backref(pag
, agino
, next_agino
);
2275 /* Point the head of the list to point to this inode. */
2276 return xfs_iunlink_update_bucket(tp
, agno
, agibp
, bucket_index
, agino
);
2279 /* Return the imap, dinode pointer, and buffer for an inode. */
2281 xfs_iunlink_map_ino(
2282 struct xfs_trans
*tp
,
2283 xfs_agnumber_t agno
,
2285 struct xfs_imap
*imap
,
2286 struct xfs_dinode
**dipp
,
2287 struct xfs_buf
**bpp
)
2289 struct xfs_mount
*mp
= tp
->t_mountp
;
2293 error
= xfs_imap(mp
, tp
, XFS_AGINO_TO_INO(mp
, agno
, agino
), imap
, 0);
2295 xfs_warn(mp
, "%s: xfs_imap returned error %d.",
2300 error
= xfs_imap_to_bp(mp
, tp
, imap
, dipp
, bpp
, 0);
2302 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2311 * Walk the unlinked chain from @head_agino until we find the inode that
2312 * points to @target_agino. Return the inode number, map, dinode pointer,
2313 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2315 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2316 * @agino, @imap, @dipp, and @bpp are all output parameters.
2318 * Do not call this function if @target_agino is the head of the list.
2321 xfs_iunlink_map_prev(
2322 struct xfs_trans
*tp
,
2323 xfs_agnumber_t agno
,
2324 xfs_agino_t head_agino
,
2325 xfs_agino_t target_agino
,
2327 struct xfs_imap
*imap
,
2328 struct xfs_dinode
**dipp
,
2329 struct xfs_buf
**bpp
,
2330 struct xfs_perag
*pag
)
2332 struct xfs_mount
*mp
= tp
->t_mountp
;
2333 xfs_agino_t next_agino
;
2336 ASSERT(head_agino
!= target_agino
);
2339 /* See if our backref cache can find it faster. */
2340 *agino
= xfs_iunlink_lookup_backref(pag
, target_agino
);
2341 if (*agino
!= NULLAGINO
) {
2342 error
= xfs_iunlink_map_ino(tp
, agno
, *agino
, imap
, dipp
, bpp
);
2346 if (be32_to_cpu((*dipp
)->di_next_unlinked
) == target_agino
)
2350 * If we get here the cache contents were corrupt, so drop the
2351 * buffer and fall back to walking the bucket list.
2353 xfs_trans_brelse(tp
, *bpp
);
2358 trace_xfs_iunlink_map_prev_fallback(mp
, agno
);
2360 /* Otherwise, walk the entire bucket until we find it. */
2361 next_agino
= head_agino
;
2362 while (next_agino
!= target_agino
) {
2363 xfs_agino_t unlinked_agino
;
2366 xfs_trans_brelse(tp
, *bpp
);
2368 *agino
= next_agino
;
2369 error
= xfs_iunlink_map_ino(tp
, agno
, next_agino
, imap
, dipp
,
2374 unlinked_agino
= be32_to_cpu((*dipp
)->di_next_unlinked
);
2376 * Make sure this pointer is valid and isn't an obvious
2379 if (!xfs_verify_agino(mp
, agno
, unlinked_agino
) ||
2380 next_agino
== unlinked_agino
) {
2381 XFS_CORRUPTION_ERROR(__func__
,
2382 XFS_ERRLEVEL_LOW
, mp
,
2383 *dipp
, sizeof(**dipp
));
2384 error
= -EFSCORRUPTED
;
2387 next_agino
= unlinked_agino
;
2394 * Pull the on-disk inode from the AGI unlinked list.
2398 struct xfs_trans
*tp
,
2399 struct xfs_inode
*ip
)
2401 struct xfs_mount
*mp
= tp
->t_mountp
;
2402 struct xfs_agi
*agi
;
2403 struct xfs_buf
*agibp
;
2404 struct xfs_buf
*last_ibp
;
2405 struct xfs_dinode
*last_dip
= NULL
;
2406 struct xfs_perag
*pag
= NULL
;
2407 xfs_agnumber_t agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2408 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2409 xfs_agino_t next_agino
;
2410 xfs_agino_t head_agino
;
2411 short bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2414 trace_xfs_iunlink_remove(ip
);
2416 /* Get the agi buffer first. It ensures lock ordering on the list. */
2417 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2420 agi
= agibp
->b_addr
;
2423 * Get the index into the agi hash table for the list this inode will
2424 * go on. Make sure the head pointer isn't garbage.
2426 head_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2427 if (!xfs_verify_agino(mp
, agno
, head_agino
)) {
2428 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
,
2430 return -EFSCORRUPTED
;
2434 * Set our inode's next_unlinked pointer to NULL and then return
2435 * the old pointer value so that we can update whatever was previous
2436 * to us in the list to point to whatever was next in the list.
2438 error
= xfs_iunlink_update_inode(tp
, ip
, agno
, NULLAGINO
, &next_agino
);
2443 * If there was a backref pointing from the next inode back to this
2444 * one, remove it because we've removed this inode from the list.
2446 * Later, if this inode was in the middle of the list we'll update
2447 * this inode's backref to point from the next inode.
2449 if (next_agino
!= NULLAGINO
) {
2450 pag
= xfs_perag_get(mp
, agno
);
2451 error
= xfs_iunlink_change_backref(pag
, next_agino
,
2457 if (head_agino
== agino
) {
2458 /* Point the head of the list to the next unlinked inode. */
2459 error
= xfs_iunlink_update_bucket(tp
, agno
, agibp
, bucket_index
,
2464 struct xfs_imap imap
;
2465 xfs_agino_t prev_agino
;
2468 pag
= xfs_perag_get(mp
, agno
);
2470 /* We need to search the list for the inode being freed. */
2471 error
= xfs_iunlink_map_prev(tp
, agno
, head_agino
, agino
,
2472 &prev_agino
, &imap
, &last_dip
, &last_ibp
,
2477 /* Point the previous inode on the list to the next inode. */
2478 xfs_iunlink_update_dinode(tp
, agno
, prev_agino
, last_ibp
,
2479 last_dip
, &imap
, next_agino
);
2482 * Now we deal with the backref for this inode. If this inode
2483 * pointed at a real inode, change the backref that pointed to
2484 * us to point to our old next. If this inode was the end of
2485 * the list, delete the backref that pointed to us. Note that
2486 * change_backref takes care of deleting the backref if
2487 * next_agino is NULLAGINO.
2489 error
= xfs_iunlink_change_backref(pag
, agino
, next_agino
);
2501 * Look up the inode number specified and mark it stale if it is found. If it is
2502 * dirty, return the inode so it can be attached to the cluster buffer so it can
2503 * be processed appropriately when the cluster free transaction completes.
2505 static struct xfs_inode
*
2506 xfs_ifree_get_one_inode(
2507 struct xfs_perag
*pag
,
2508 struct xfs_inode
*free_ip
,
2511 struct xfs_mount
*mp
= pag
->pag_mount
;
2512 struct xfs_inode
*ip
;
2516 ip
= radix_tree_lookup(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, inum
));
2518 /* Inode not in memory, nothing to do */
2520 goto out_rcu_unlock
;
2523 * because this is an RCU protected lookup, we could find a recently
2524 * freed or even reallocated inode during the lookup. We need to check
2525 * under the i_flags_lock for a valid inode here. Skip it if it is not
2526 * valid, the wrong inode or stale.
2528 spin_lock(&ip
->i_flags_lock
);
2529 if (ip
->i_ino
!= inum
|| __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2530 spin_unlock(&ip
->i_flags_lock
);
2531 goto out_rcu_unlock
;
2533 spin_unlock(&ip
->i_flags_lock
);
2536 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2537 * other inodes that we did not find in the list attached to the buffer
2538 * and are not already marked stale. If we can't lock it, back off and
2541 if (ip
!= free_ip
) {
2542 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2549 * Check the inode number again in case we're racing with
2550 * freeing in xfs_reclaim_inode(). See the comments in that
2551 * function for more information as to why the initial check is
2554 if (ip
->i_ino
!= inum
) {
2555 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2556 goto out_rcu_unlock
;
2562 xfs_iflags_set(ip
, XFS_ISTALE
);
2565 * We don't need to attach clean inodes or those only with unlogged
2566 * changes (which we throw away, anyway).
2568 if (!ip
->i_itemp
|| xfs_inode_clean(ip
)) {
2569 ASSERT(ip
!= free_ip
);
2571 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2583 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2584 * inodes that are in memory - they all must be marked stale and attached to
2585 * the cluster buffer.
2589 xfs_inode_t
*free_ip
,
2591 struct xfs_icluster
*xic
)
2593 xfs_mount_t
*mp
= free_ip
->i_mount
;
2600 struct xfs_inode_log_item
*iip
;
2601 struct xfs_log_item
*lip
;
2602 struct xfs_perag
*pag
;
2603 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
2607 inum
= xic
->first_ino
;
2608 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2609 nbufs
= igeo
->ialloc_blks
/ igeo
->blocks_per_cluster
;
2611 for (j
= 0; j
< nbufs
; j
++, inum
+= igeo
->inodes_per_cluster
) {
2613 * The allocation bitmap tells us which inodes of the chunk were
2614 * physically allocated. Skip the cluster if an inode falls into
2617 ioffset
= inum
- xic
->first_ino
;
2618 if ((xic
->alloc
& XFS_INOBT_MASK(ioffset
)) == 0) {
2619 ASSERT(ioffset
% igeo
->inodes_per_cluster
== 0);
2623 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2624 XFS_INO_TO_AGBNO(mp
, inum
));
2627 * We obtain and lock the backing buffer first in the process
2628 * here, as we have to ensure that any dirty inode that we
2629 * can't get the flush lock on is attached to the buffer.
2630 * If we scan the in-memory inodes first, then buffer IO can
2631 * complete before we get a lock on it, and hence we may fail
2632 * to mark all the active inodes on the buffer stale.
2634 error
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2635 mp
->m_bsize
* igeo
->blocks_per_cluster
,
2643 * This buffer may not have been correctly initialised as we
2644 * didn't read it from disk. That's not important because we are
2645 * only using to mark the buffer as stale in the log, and to
2646 * attach stale cached inodes on it. That means it will never be
2647 * dispatched for IO. If it is, we want to know about it, and we
2648 * want it to fail. We can acheive this by adding a write
2649 * verifier to the buffer.
2651 bp
->b_ops
= &xfs_inode_buf_ops
;
2654 * Walk the inodes already attached to the buffer and mark them
2655 * stale. These will all have the flush locks held, so an
2656 * in-memory inode walk can't lock them. By marking them all
2657 * stale first, we will not attempt to lock them in the loop
2658 * below as the XFS_ISTALE flag will be set.
2660 list_for_each_entry(lip
, &bp
->b_li_list
, li_bio_list
) {
2661 if (lip
->li_type
== XFS_LI_INODE
) {
2662 iip
= (struct xfs_inode_log_item
*)lip
;
2663 ASSERT(iip
->ili_logged
== 1);
2664 lip
->li_cb
= xfs_istale_done
;
2665 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2666 &iip
->ili_flush_lsn
,
2667 &iip
->ili_item
.li_lsn
);
2668 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2674 * For each inode in memory attempt to add it to the inode
2675 * buffer and set it up for being staled on buffer IO
2676 * completion. This is safe as we've locked out tail pushing
2677 * and flushing by locking the buffer.
2679 * We have already marked every inode that was part of a
2680 * transaction stale above, which means there is no point in
2681 * even trying to lock them.
2683 for (i
= 0; i
< igeo
->inodes_per_cluster
; i
++) {
2684 ip
= xfs_ifree_get_one_inode(pag
, free_ip
, inum
+ i
);
2689 iip
->ili_last_fields
= iip
->ili_fields
;
2690 iip
->ili_fields
= 0;
2691 iip
->ili_fsync_fields
= 0;
2692 iip
->ili_logged
= 1;
2693 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2694 &iip
->ili_item
.li_lsn
);
2696 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2700 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2703 xfs_trans_stale_inode_buf(tp
, bp
);
2704 xfs_trans_binval(tp
, bp
);
2712 * This is called to return an inode to the inode free list.
2713 * The inode should already be truncated to 0 length and have
2714 * no pages associated with it. This routine also assumes that
2715 * the inode is already a part of the transaction.
2717 * The on-disk copy of the inode will have been added to the list
2718 * of unlinked inodes in the AGI. We need to remove the inode from
2719 * that list atomically with respect to freeing it here.
2723 struct xfs_trans
*tp
,
2724 struct xfs_inode
*ip
)
2727 struct xfs_icluster xic
= { 0 };
2729 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2730 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2731 ASSERT(ip
->i_df
.if_nextents
== 0);
2732 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(VFS_I(ip
)->i_mode
));
2733 ASSERT(ip
->i_d
.di_nblocks
== 0);
2736 * Pull the on-disk inode from the AGI unlinked list.
2738 error
= xfs_iunlink_remove(tp
, ip
);
2742 error
= xfs_difree(tp
, ip
->i_ino
, &xic
);
2747 * Free any local-format data sitting around before we reset the
2748 * data fork to extents format. Note that the attr fork data has
2749 * already been freed by xfs_attr_inactive.
2751 if (ip
->i_df
.if_format
== XFS_DINODE_FMT_LOCAL
) {
2752 kmem_free(ip
->i_df
.if_u1
.if_data
);
2753 ip
->i_df
.if_u1
.if_data
= NULL
;
2754 ip
->i_df
.if_bytes
= 0;
2757 VFS_I(ip
)->i_mode
= 0; /* mark incore inode as free */
2758 ip
->i_d
.di_flags
= 0;
2759 ip
->i_d
.di_flags2
= 0;
2760 ip
->i_d
.di_dmevmask
= 0;
2761 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2762 ip
->i_df
.if_format
= XFS_DINODE_FMT_EXTENTS
;
2764 /* Don't attempt to replay owner changes for a deleted inode */
2765 ip
->i_itemp
->ili_fields
&= ~(XFS_ILOG_AOWNER
|XFS_ILOG_DOWNER
);
2768 * Bump the generation count so no one will be confused
2769 * by reincarnations of this inode.
2771 VFS_I(ip
)->i_generation
++;
2772 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2775 error
= xfs_ifree_cluster(ip
, tp
, &xic
);
2781 * This is called to unpin an inode. The caller must have the inode locked
2782 * in at least shared mode so that the buffer cannot be subsequently pinned
2783 * once someone is waiting for it to be unpinned.
2787 struct xfs_inode
*ip
)
2789 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2791 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2793 /* Give the log a push to start the unpinning I/O */
2794 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0, NULL
);
2800 struct xfs_inode
*ip
)
2802 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2803 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2808 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2809 if (xfs_ipincount(ip
))
2811 } while (xfs_ipincount(ip
));
2812 finish_wait(wq
, &wait
.wq_entry
);
2817 struct xfs_inode
*ip
)
2819 if (xfs_ipincount(ip
))
2820 __xfs_iunpin_wait(ip
);
2824 * Removing an inode from the namespace involves removing the directory entry
2825 * and dropping the link count on the inode. Removing the directory entry can
2826 * result in locking an AGF (directory blocks were freed) and removing a link
2827 * count can result in placing the inode on an unlinked list which results in
2830 * The big problem here is that we have an ordering constraint on AGF and AGI
2831 * locking - inode allocation locks the AGI, then can allocate a new extent for
2832 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2833 * removes the inode from the unlinked list, requiring that we lock the AGI
2834 * first, and then freeing the inode can result in an inode chunk being freed
2835 * and hence freeing disk space requiring that we lock an AGF.
2837 * Hence the ordering that is imposed by other parts of the code is AGI before
2838 * AGF. This means we cannot remove the directory entry before we drop the inode
2839 * reference count and put it on the unlinked list as this results in a lock
2840 * order of AGF then AGI, and this can deadlock against inode allocation and
2841 * freeing. Therefore we must drop the link counts before we remove the
2844 * This is still safe from a transactional point of view - it is not until we
2845 * get to xfs_defer_finish() that we have the possibility of multiple
2846 * transactions in this operation. Hence as long as we remove the directory
2847 * entry and drop the link count in the first transaction of the remove
2848 * operation, there are no transactional constraints on the ordering here.
2853 struct xfs_name
*name
,
2856 xfs_mount_t
*mp
= dp
->i_mount
;
2857 xfs_trans_t
*tp
= NULL
;
2858 int is_dir
= S_ISDIR(VFS_I(ip
)->i_mode
);
2862 trace_xfs_remove(dp
, name
);
2864 if (XFS_FORCED_SHUTDOWN(mp
))
2867 error
= xfs_qm_dqattach(dp
);
2871 error
= xfs_qm_dqattach(ip
);
2876 * We try to get the real space reservation first,
2877 * allowing for directory btree deletion(s) implying
2878 * possible bmap insert(s). If we can't get the space
2879 * reservation then we use 0 instead, and avoid the bmap
2880 * btree insert(s) in the directory code by, if the bmap
2881 * insert tries to happen, instead trimming the LAST
2882 * block from the directory.
2884 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2885 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, resblks
, 0, 0, &tp
);
2886 if (error
== -ENOSPC
) {
2888 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, 0, 0, 0,
2892 ASSERT(error
!= -ENOSPC
);
2896 xfs_lock_two_inodes(dp
, XFS_ILOCK_EXCL
, ip
, XFS_ILOCK_EXCL
);
2898 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2899 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2902 * If we're removing a directory perform some additional validation.
2905 ASSERT(VFS_I(ip
)->i_nlink
>= 2);
2906 if (VFS_I(ip
)->i_nlink
!= 2) {
2908 goto out_trans_cancel
;
2910 if (!xfs_dir_isempty(ip
)) {
2912 goto out_trans_cancel
;
2915 /* Drop the link from ip's "..". */
2916 error
= xfs_droplink(tp
, dp
);
2918 goto out_trans_cancel
;
2920 /* Drop the "." link from ip to self. */
2921 error
= xfs_droplink(tp
, ip
);
2923 goto out_trans_cancel
;
2926 * When removing a non-directory we need to log the parent
2927 * inode here. For a directory this is done implicitly
2928 * by the xfs_droplink call for the ".." entry.
2930 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2932 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2934 /* Drop the link from dp to ip. */
2935 error
= xfs_droplink(tp
, ip
);
2937 goto out_trans_cancel
;
2939 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
, resblks
);
2941 ASSERT(error
!= -ENOENT
);
2942 goto out_trans_cancel
;
2946 * If this is a synchronous mount, make sure that the
2947 * remove transaction goes to disk before returning to
2950 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2951 xfs_trans_set_sync(tp
);
2953 error
= xfs_trans_commit(tp
);
2957 if (is_dir
&& xfs_inode_is_filestream(ip
))
2958 xfs_filestream_deassociate(ip
);
2963 xfs_trans_cancel(tp
);
2969 * Enter all inodes for a rename transaction into a sorted array.
2971 #define __XFS_SORT_INODES 5
2973 xfs_sort_for_rename(
2974 struct xfs_inode
*dp1
, /* in: old (source) directory inode */
2975 struct xfs_inode
*dp2
, /* in: new (target) directory inode */
2976 struct xfs_inode
*ip1
, /* in: inode of old entry */
2977 struct xfs_inode
*ip2
, /* in: inode of new entry */
2978 struct xfs_inode
*wip
, /* in: whiteout inode */
2979 struct xfs_inode
**i_tab
,/* out: sorted array of inodes */
2980 int *num_inodes
) /* in/out: inodes in array */
2984 ASSERT(*num_inodes
== __XFS_SORT_INODES
);
2985 memset(i_tab
, 0, *num_inodes
* sizeof(struct xfs_inode
*));
2988 * i_tab contains a list of pointers to inodes. We initialize
2989 * the table here & we'll sort it. We will then use it to
2990 * order the acquisition of the inode locks.
2992 * Note that the table may contain duplicates. e.g., dp1 == dp2.
3005 * Sort the elements via bubble sort. (Remember, there are at
3006 * most 5 elements to sort, so this is adequate.)
3008 for (i
= 0; i
< *num_inodes
; i
++) {
3009 for (j
= 1; j
< *num_inodes
; j
++) {
3010 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
3011 struct xfs_inode
*temp
= i_tab
[j
];
3012 i_tab
[j
] = i_tab
[j
-1];
3021 struct xfs_trans
*tp
)
3024 * If this is a synchronous mount, make sure that the rename transaction
3025 * goes to disk before returning to the user.
3027 if (tp
->t_mountp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
3028 xfs_trans_set_sync(tp
);
3030 return xfs_trans_commit(tp
);
3034 * xfs_cross_rename()
3036 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3040 struct xfs_trans
*tp
,
3041 struct xfs_inode
*dp1
,
3042 struct xfs_name
*name1
,
3043 struct xfs_inode
*ip1
,
3044 struct xfs_inode
*dp2
,
3045 struct xfs_name
*name2
,
3046 struct xfs_inode
*ip2
,
3054 /* Swap inode number for dirent in first parent */
3055 error
= xfs_dir_replace(tp
, dp1
, name1
, ip2
->i_ino
, spaceres
);
3057 goto out_trans_abort
;
3059 /* Swap inode number for dirent in second parent */
3060 error
= xfs_dir_replace(tp
, dp2
, name2
, ip1
->i_ino
, spaceres
);
3062 goto out_trans_abort
;
3065 * If we're renaming one or more directories across different parents,
3066 * update the respective ".." entries (and link counts) to match the new
3070 dp2_flags
= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
3072 if (S_ISDIR(VFS_I(ip2
)->i_mode
)) {
3073 error
= xfs_dir_replace(tp
, ip2
, &xfs_name_dotdot
,
3074 dp1
->i_ino
, spaceres
);
3076 goto out_trans_abort
;
3078 /* transfer ip2 ".." reference to dp1 */
3079 if (!S_ISDIR(VFS_I(ip1
)->i_mode
)) {
3080 error
= xfs_droplink(tp
, dp2
);
3082 goto out_trans_abort
;
3083 xfs_bumplink(tp
, dp1
);
3087 * Although ip1 isn't changed here, userspace needs
3088 * to be warned about the change, so that applications
3089 * relying on it (like backup ones), will properly
3092 ip1_flags
|= XFS_ICHGTIME_CHG
;
3093 ip2_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
3096 if (S_ISDIR(VFS_I(ip1
)->i_mode
)) {
3097 error
= xfs_dir_replace(tp
, ip1
, &xfs_name_dotdot
,
3098 dp2
->i_ino
, spaceres
);
3100 goto out_trans_abort
;
3102 /* transfer ip1 ".." reference to dp2 */
3103 if (!S_ISDIR(VFS_I(ip2
)->i_mode
)) {
3104 error
= xfs_droplink(tp
, dp1
);
3106 goto out_trans_abort
;
3107 xfs_bumplink(tp
, dp2
);
3111 * Although ip2 isn't changed here, userspace needs
3112 * to be warned about the change, so that applications
3113 * relying on it (like backup ones), will properly
3116 ip1_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
3117 ip2_flags
|= XFS_ICHGTIME_CHG
;
3122 xfs_trans_ichgtime(tp
, ip1
, ip1_flags
);
3123 xfs_trans_log_inode(tp
, ip1
, XFS_ILOG_CORE
);
3126 xfs_trans_ichgtime(tp
, ip2
, ip2_flags
);
3127 xfs_trans_log_inode(tp
, ip2
, XFS_ILOG_CORE
);
3130 xfs_trans_ichgtime(tp
, dp2
, dp2_flags
);
3131 xfs_trans_log_inode(tp
, dp2
, XFS_ILOG_CORE
);
3133 xfs_trans_ichgtime(tp
, dp1
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3134 xfs_trans_log_inode(tp
, dp1
, XFS_ILOG_CORE
);
3135 return xfs_finish_rename(tp
);
3138 xfs_trans_cancel(tp
);
3143 * xfs_rename_alloc_whiteout()
3145 * Return a referenced, unlinked, unlocked inode that that can be used as a
3146 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3147 * crash between allocating the inode and linking it into the rename transaction
3148 * recovery will free the inode and we won't leak it.
3151 xfs_rename_alloc_whiteout(
3152 struct xfs_inode
*dp
,
3153 struct xfs_inode
**wip
)
3155 struct xfs_inode
*tmpfile
;
3158 error
= xfs_create_tmpfile(dp
, S_IFCHR
| WHITEOUT_MODE
, &tmpfile
);
3163 * Prepare the tmpfile inode as if it were created through the VFS.
3164 * Complete the inode setup and flag it as linkable. nlink is already
3165 * zero, so we can skip the drop_nlink.
3167 xfs_setup_iops(tmpfile
);
3168 xfs_finish_inode_setup(tmpfile
);
3169 VFS_I(tmpfile
)->i_state
|= I_LINKABLE
;
3180 struct xfs_inode
*src_dp
,
3181 struct xfs_name
*src_name
,
3182 struct xfs_inode
*src_ip
,
3183 struct xfs_inode
*target_dp
,
3184 struct xfs_name
*target_name
,
3185 struct xfs_inode
*target_ip
,
3188 struct xfs_mount
*mp
= src_dp
->i_mount
;
3189 struct xfs_trans
*tp
;
3190 struct xfs_inode
*wip
= NULL
; /* whiteout inode */
3191 struct xfs_inode
*inodes
[__XFS_SORT_INODES
];
3192 struct xfs_buf
*agibp
;
3193 int num_inodes
= __XFS_SORT_INODES
;
3194 bool new_parent
= (src_dp
!= target_dp
);
3195 bool src_is_directory
= S_ISDIR(VFS_I(src_ip
)->i_mode
);
3199 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
3201 if ((flags
& RENAME_EXCHANGE
) && !target_ip
)
3205 * If we are doing a whiteout operation, allocate the whiteout inode
3206 * we will be placing at the target and ensure the type is set
3209 if (flags
& RENAME_WHITEOUT
) {
3210 ASSERT(!(flags
& (RENAME_NOREPLACE
| RENAME_EXCHANGE
)));
3211 error
= xfs_rename_alloc_whiteout(target_dp
, &wip
);
3215 /* setup target dirent info as whiteout */
3216 src_name
->type
= XFS_DIR3_FT_CHRDEV
;
3219 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
, wip
,
3220 inodes
, &num_inodes
);
3222 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
3223 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, spaceres
, 0, 0, &tp
);
3224 if (error
== -ENOSPC
) {
3226 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, 0, 0, 0,
3230 goto out_release_wip
;
3233 * Attach the dquots to the inodes
3235 error
= xfs_qm_vop_rename_dqattach(inodes
);
3237 goto out_trans_cancel
;
3240 * Lock all the participating inodes. Depending upon whether
3241 * the target_name exists in the target directory, and
3242 * whether the target directory is the same as the source
3243 * directory, we can lock from 2 to 4 inodes.
3245 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
3248 * Join all the inodes to the transaction. From this point on,
3249 * we can rely on either trans_commit or trans_cancel to unlock
3252 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
3254 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
3255 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
3257 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
3259 xfs_trans_ijoin(tp
, wip
, XFS_ILOCK_EXCL
);
3262 * If we are using project inheritance, we only allow renames
3263 * into our tree when the project IDs are the same; else the
3264 * tree quota mechanism would be circumvented.
3266 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
3267 target_dp
->i_d
.di_projid
!= src_ip
->i_d
.di_projid
)) {
3269 goto out_trans_cancel
;
3272 /* RENAME_EXCHANGE is unique from here on. */
3273 if (flags
& RENAME_EXCHANGE
)
3274 return xfs_cross_rename(tp
, src_dp
, src_name
, src_ip
,
3275 target_dp
, target_name
, target_ip
,
3279 * Check for expected errors before we dirty the transaction
3280 * so we can return an error without a transaction abort.
3282 if (target_ip
== NULL
) {
3284 * If there's no space reservation, check the entry will
3285 * fit before actually inserting it.
3288 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
3290 goto out_trans_cancel
;
3294 * If target exists and it's a directory, check that whether
3295 * it can be destroyed.
3297 if (S_ISDIR(VFS_I(target_ip
)->i_mode
) &&
3298 (!xfs_dir_isempty(target_ip
) ||
3299 (VFS_I(target_ip
)->i_nlink
> 2))) {
3301 goto out_trans_cancel
;
3306 * Directory entry creation below may acquire the AGF. Remove
3307 * the whiteout from the unlinked list first to preserve correct
3308 * AGI/AGF locking order. This dirties the transaction so failures
3309 * after this point will abort and log recovery will clean up the
3312 * For whiteouts, we need to bump the link count on the whiteout
3313 * inode. After this point, we have a real link, clear the tmpfile
3314 * state flag from the inode so it doesn't accidentally get misused
3318 ASSERT(VFS_I(wip
)->i_nlink
== 0);
3319 error
= xfs_iunlink_remove(tp
, wip
);
3321 goto out_trans_cancel
;
3323 xfs_bumplink(tp
, wip
);
3324 VFS_I(wip
)->i_state
&= ~I_LINKABLE
;
3328 * Set up the target.
3330 if (target_ip
== NULL
) {
3332 * If target does not exist and the rename crosses
3333 * directories, adjust the target directory link count
3334 * to account for the ".." reference from the new entry.
3336 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
3337 src_ip
->i_ino
, spaceres
);
3339 goto out_trans_cancel
;
3341 xfs_trans_ichgtime(tp
, target_dp
,
3342 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3344 if (new_parent
&& src_is_directory
) {
3345 xfs_bumplink(tp
, target_dp
);
3347 } else { /* target_ip != NULL */
3349 * Link the source inode under the target name.
3350 * If the source inode is a directory and we are moving
3351 * it across directories, its ".." entry will be
3352 * inconsistent until we replace that down below.
3354 * In case there is already an entry with the same
3355 * name at the destination directory, remove it first.
3359 * Check whether the replace operation will need to allocate
3360 * blocks. This happens when the shortform directory lacks
3361 * space and we have to convert it to a block format directory.
3362 * When more blocks are necessary, we must lock the AGI first
3363 * to preserve locking order (AGI -> AGF).
3365 if (xfs_dir2_sf_replace_needblock(target_dp
, src_ip
->i_ino
)) {
3366 error
= xfs_read_agi(mp
, tp
,
3367 XFS_INO_TO_AGNO(mp
, target_ip
->i_ino
),
3370 goto out_trans_cancel
;
3373 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
3374 src_ip
->i_ino
, spaceres
);
3376 goto out_trans_cancel
;
3378 xfs_trans_ichgtime(tp
, target_dp
,
3379 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3382 * Decrement the link count on the target since the target
3383 * dir no longer points to it.
3385 error
= xfs_droplink(tp
, target_ip
);
3387 goto out_trans_cancel
;
3389 if (src_is_directory
) {
3391 * Drop the link from the old "." entry.
3393 error
= xfs_droplink(tp
, target_ip
);
3395 goto out_trans_cancel
;
3397 } /* target_ip != NULL */
3400 * Remove the source.
3402 if (new_parent
&& src_is_directory
) {
3404 * Rewrite the ".." entry to point to the new
3407 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
3408 target_dp
->i_ino
, spaceres
);
3409 ASSERT(error
!= -EEXIST
);
3411 goto out_trans_cancel
;
3415 * We always want to hit the ctime on the source inode.
3417 * This isn't strictly required by the standards since the source
3418 * inode isn't really being changed, but old unix file systems did
3419 * it and some incremental backup programs won't work without it.
3421 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
3422 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
3425 * Adjust the link count on src_dp. This is necessary when
3426 * renaming a directory, either within one parent when
3427 * the target existed, or across two parent directories.
3429 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
3432 * Decrement link count on src_directory since the
3433 * entry that's moved no longer points to it.
3435 error
= xfs_droplink(tp
, src_dp
);
3437 goto out_trans_cancel
;
3441 * For whiteouts, we only need to update the source dirent with the
3442 * inode number of the whiteout inode rather than removing it
3446 error
= xfs_dir_replace(tp
, src_dp
, src_name
, wip
->i_ino
,
3449 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
3452 goto out_trans_cancel
;
3454 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3455 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
3457 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
3459 error
= xfs_finish_rename(tp
);
3465 xfs_trans_cancel(tp
);
3474 struct xfs_inode
*ip
,
3477 struct xfs_mount
*mp
= ip
->i_mount
;
3478 struct xfs_perag
*pag
;
3479 unsigned long first_index
, mask
;
3481 struct xfs_inode
**cilist
;
3482 struct xfs_inode
*cip
;
3483 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
3489 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
3491 cilist_size
= igeo
->inodes_per_cluster
* sizeof(struct xfs_inode
*);
3492 cilist
= kmem_alloc(cilist_size
, KM_MAYFAIL
|KM_NOFS
);
3496 mask
= ~(igeo
->inodes_per_cluster
- 1);
3497 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
3499 /* really need a gang lookup range call here */
3500 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)cilist
,
3501 first_index
, igeo
->inodes_per_cluster
);
3505 for (i
= 0; i
< nr_found
; i
++) {
3511 * because this is an RCU protected lookup, we could find a
3512 * recently freed or even reallocated inode during the lookup.
3513 * We need to check under the i_flags_lock for a valid inode
3514 * here. Skip it if it is not valid or the wrong inode.
3516 spin_lock(&cip
->i_flags_lock
);
3518 __xfs_iflags_test(cip
, XFS_ISTALE
)) {
3519 spin_unlock(&cip
->i_flags_lock
);
3524 * Once we fall off the end of the cluster, no point checking
3525 * any more inodes in the list because they will also all be
3526 * outside the cluster.
3528 if ((XFS_INO_TO_AGINO(mp
, cip
->i_ino
) & mask
) != first_index
) {
3529 spin_unlock(&cip
->i_flags_lock
);
3532 spin_unlock(&cip
->i_flags_lock
);
3535 * Do an un-protected check to see if the inode is dirty and
3536 * is a candidate for flushing. These checks will be repeated
3537 * later after the appropriate locks are acquired.
3539 if (xfs_inode_clean(cip
) && xfs_ipincount(cip
) == 0)
3543 * Try to get locks. If any are unavailable or it is pinned,
3544 * then this inode cannot be flushed and is skipped.
3547 if (!xfs_ilock_nowait(cip
, XFS_ILOCK_SHARED
))
3549 if (!xfs_iflock_nowait(cip
)) {
3550 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3553 if (xfs_ipincount(cip
)) {
3555 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3561 * Check the inode number again, just to be certain we are not
3562 * racing with freeing in xfs_reclaim_inode(). See the comments
3563 * in that function for more information as to why the initial
3564 * check is not sufficient.
3568 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3573 * arriving here means that this inode can be flushed. First
3574 * re-check that it's dirty before flushing.
3576 if (!xfs_inode_clean(cip
)) {
3577 error
= xfs_iflush_int(cip
, bp
);
3579 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3586 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3590 XFS_STATS_INC(mp
, xs_icluster_flushcnt
);
3591 XFS_STATS_ADD(mp
, xs_icluster_flushinode
, clcount
);
3603 * Flush dirty inode metadata into the backing buffer.
3605 * The caller must have the inode lock and the inode flush lock held. The
3606 * inode lock will still be held upon return to the caller, and the inode
3607 * flush lock will be released after the inode has reached the disk.
3609 * The caller must write out the buffer returned in *bpp and release it.
3613 struct xfs_inode
*ip
,
3614 struct xfs_buf
**bpp
)
3616 struct xfs_mount
*mp
= ip
->i_mount
;
3617 struct xfs_buf
*bp
= NULL
;
3618 struct xfs_dinode
*dip
;
3621 XFS_STATS_INC(mp
, xs_iflush_count
);
3623 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3624 ASSERT(xfs_isiflocked(ip
));
3625 ASSERT(ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
||
3626 ip
->i_df
.if_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3630 xfs_iunpin_wait(ip
);
3633 * For stale inodes we cannot rely on the backing buffer remaining
3634 * stale in cache for the remaining life of the stale inode and so
3635 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3636 * inodes below. We have to check this after ensuring the inode is
3637 * unpinned so that it is safe to reclaim the stale inode after the
3640 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3646 * Get the buffer containing the on-disk inode. We are doing a try-lock
3647 * operation here, so we may get an EAGAIN error. In that case, return
3648 * leaving the inode dirty.
3650 * If we get any other error, we effectively have a corruption situation
3651 * and we cannot flush the inode. Abort the flush and shut down.
3653 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
);
3654 if (error
== -EAGAIN
) {
3662 * If the buffer is pinned then push on the log now so we won't
3663 * get stuck waiting in the write for too long.
3665 if (xfs_buf_ispinned(bp
))
3666 xfs_log_force(mp
, 0);
3669 * Flush the provided inode then attempt to gather others from the
3670 * cluster into the write.
3672 * Note: Once we attempt to flush an inode, we must run buffer
3673 * completion callbacks on any failure. If this fails, simulate an I/O
3674 * failure on the buffer and shut down.
3676 error
= xfs_iflush_int(ip
, bp
);
3678 error
= xfs_iflush_cluster(ip
, bp
);
3680 bp
->b_flags
|= XBF_ASYNC
;
3681 xfs_buf_ioend_fail(bp
);
3689 xfs_iflush_abort(ip
);
3691 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3697 struct xfs_inode
*ip
,
3700 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3701 struct xfs_dinode
*dip
;
3702 struct xfs_mount
*mp
= ip
->i_mount
;
3705 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3706 ASSERT(xfs_isiflocked(ip
));
3707 ASSERT(ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
||
3708 ip
->i_df
.if_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3709 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3711 dip
= xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3714 * We don't flush the inode if any of the following checks fail, but we
3715 * do still update the log item and attach to the backing buffer as if
3716 * the flush happened. This is a formality to facilitate predictable
3717 * error handling as the caller will shutdown and fail the buffer.
3719 error
= -EFSCORRUPTED
;
3720 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3721 mp
, XFS_ERRTAG_IFLUSH_1
)) {
3722 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3723 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT
,
3724 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3727 if (S_ISREG(VFS_I(ip
)->i_mode
)) {
3729 ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
&&
3730 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
,
3731 mp
, XFS_ERRTAG_IFLUSH_3
)) {
3732 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3733 "%s: Bad regular inode %Lu, ptr "PTR_FMT
,
3734 __func__
, ip
->i_ino
, ip
);
3737 } else if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
3739 ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
&&
3740 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
&&
3741 ip
->i_df
.if_format
!= XFS_DINODE_FMT_LOCAL
,
3742 mp
, XFS_ERRTAG_IFLUSH_4
)) {
3743 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3744 "%s: Bad directory inode %Lu, ptr "PTR_FMT
,
3745 __func__
, ip
->i_ino
, ip
);
3749 if (XFS_TEST_ERROR(ip
->i_df
.if_nextents
+ xfs_ifork_nextents(ip
->i_afp
) >
3750 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
)) {
3751 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3752 "%s: detected corrupt incore inode %Lu, "
3753 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT
,
3754 __func__
, ip
->i_ino
,
3755 ip
->i_df
.if_nextents
+ xfs_ifork_nextents(ip
->i_afp
),
3756 ip
->i_d
.di_nblocks
, ip
);
3759 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3760 mp
, XFS_ERRTAG_IFLUSH_6
)) {
3761 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3762 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT
,
3763 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3768 * Inode item log recovery for v2 inodes are dependent on the
3769 * di_flushiter count for correct sequencing. We bump the flush
3770 * iteration count so we can detect flushes which postdate a log record
3771 * during recovery. This is redundant as we now log every change and
3772 * hence this can't happen but we need to still do it to ensure
3773 * backwards compatibility with old kernels that predate logging all
3776 if (!xfs_sb_version_has_v3inode(&mp
->m_sb
))
3777 ip
->i_d
.di_flushiter
++;
3780 * If there are inline format data / attr forks attached to this inode,
3781 * make sure they are not corrupt.
3783 if (ip
->i_df
.if_format
== XFS_DINODE_FMT_LOCAL
&&
3784 xfs_ifork_verify_local_data(ip
))
3786 if (ip
->i_afp
&& ip
->i_afp
->if_format
== XFS_DINODE_FMT_LOCAL
&&
3787 xfs_ifork_verify_local_attr(ip
))
3791 * Copy the dirty parts of the inode into the on-disk inode. We always
3792 * copy out the core of the inode, because if the inode is dirty at all
3795 xfs_inode_to_disk(ip
, dip
, iip
->ili_item
.li_lsn
);
3797 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3798 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3799 ip
->i_d
.di_flushiter
= 0;
3801 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3802 if (XFS_IFORK_Q(ip
))
3803 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3804 xfs_inobp_check(mp
, bp
);
3807 * We've recorded everything logged in the inode, so we'd like to clear
3808 * the ili_fields bits so we don't log and flush things unnecessarily.
3809 * However, we can't stop logging all this information until the data
3810 * we've copied into the disk buffer is written to disk. If we did we
3811 * might overwrite the copy of the inode in the log with all the data
3812 * after re-logging only part of it, and in the face of a crash we
3813 * wouldn't have all the data we need to recover.
3815 * What we do is move the bits to the ili_last_fields field. When
3816 * logging the inode, these bits are moved back to the ili_fields field.
3817 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3818 * know that the information those bits represent is permanently on
3819 * disk. As long as the flush completes before the inode is logged
3820 * again, then both ili_fields and ili_last_fields will be cleared.
3822 * We can play with the ili_fields bits here, because the inode lock
3823 * must be held exclusively in order to set bits there and the flush
3824 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3825 * done routine can tell whether or not to look in the AIL. Also, store
3826 * the current LSN of the inode so that we can tell whether the item has
3827 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3828 * need the AIL lock, because it is a 64 bit value that cannot be read
3833 iip
->ili_last_fields
= iip
->ili_fields
;
3834 iip
->ili_fields
= 0;
3835 iip
->ili_fsync_fields
= 0;
3836 iip
->ili_logged
= 1;
3838 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3839 &iip
->ili_item
.li_lsn
);
3842 * Attach the inode item callback to the buffer whether the flush
3843 * succeeded or not. If not, the caller will shut down and fail I/O
3844 * completion on the buffer to remove the inode from the AIL and release
3847 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3849 /* generate the checksum. */
3850 xfs_dinode_calc_crc(mp
, dip
);
3852 ASSERT(!list_empty(&bp
->b_li_list
));
3853 ASSERT(bp
->b_iodone
!= NULL
);
3857 /* Release an inode. */
3860 struct xfs_inode
*ip
)
3862 trace_xfs_irele(ip
, _RET_IP_
);
3867 * Ensure all commited transactions touching the inode are written to the log.
3870 xfs_log_force_inode(
3871 struct xfs_inode
*ip
)
3875 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3876 if (xfs_ipincount(ip
))
3877 lsn
= ip
->i_itemp
->ili_last_lsn
;
3878 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3882 return xfs_log_force_lsn(ip
->i_mount
, lsn
, XFS_LOG_SYNC
, NULL
);