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_iunlink(struct xfs_trans
*, struct xfs_inode
*);
48 STATIC
int xfs_iunlink_remove(struct xfs_trans
*, struct xfs_inode
*);
51 * helper function to extract extent size hint from inode
58 * No point in aligning allocations if we need to COW to actually
61 if (xfs_is_always_cow_inode(ip
))
63 if ((ip
->i_diflags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_extsize
)
65 if (XFS_IS_REALTIME_INODE(ip
))
66 return ip
->i_mount
->m_sb
.sb_rextsize
;
71 * Helper function to extract CoW extent size hint from inode.
72 * Between the extent size hint and the CoW extent size hint, we
73 * return the greater of the two. If the value is zero (automatic),
74 * use the default size.
77 xfs_get_cowextsz_hint(
83 if (ip
->i_diflags2
& XFS_DIFLAG2_COWEXTSIZE
)
85 b
= xfs_get_extsz_hint(ip
);
89 return XFS_DEFAULT_COWEXTSZ_HINT
;
94 * These two are wrapper routines around the xfs_ilock() routine used to
95 * centralize some grungy code. They are used in places that wish to lock the
96 * inode solely for reading the extents. The reason these places can't just
97 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
98 * bringing in of the extents from disk for a file in b-tree format. If the
99 * inode is in b-tree format, then we need to lock the inode exclusively until
100 * the extents are read in. Locking it exclusively all the time would limit
101 * our parallelism unnecessarily, though. What we do instead is check to see
102 * if the extents have been read in yet, and only lock the inode exclusively
105 * The functions return a value which should be given to the corresponding
106 * xfs_iunlock() call.
109 xfs_ilock_data_map_shared(
110 struct xfs_inode
*ip
)
112 uint lock_mode
= XFS_ILOCK_SHARED
;
114 if (xfs_need_iread_extents(&ip
->i_df
))
115 lock_mode
= XFS_ILOCK_EXCL
;
116 xfs_ilock(ip
, lock_mode
);
121 xfs_ilock_attr_map_shared(
122 struct xfs_inode
*ip
)
124 uint lock_mode
= XFS_ILOCK_SHARED
;
126 if (ip
->i_afp
&& xfs_need_iread_extents(ip
->i_afp
))
127 lock_mode
= XFS_ILOCK_EXCL
;
128 xfs_ilock(ip
, lock_mode
);
133 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
134 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
135 * various combinations of the locks to be obtained.
137 * The 3 locks should always be ordered so that the IO lock is obtained first,
138 * the mmap lock second and the ilock last in order to prevent deadlock.
140 * Basic locking order:
142 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
144 * mmap_lock locking order:
146 * i_rwsem -> page lock -> mmap_lock
147 * mmap_lock -> i_mmap_lock -> page_lock
149 * The difference in mmap_lock locking order mean that we cannot hold the
150 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
151 * fault in pages during copy in/out (for buffered IO) or require the mmap_lock
152 * in get_user_pages() to map the user pages into the kernel address space for
153 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
154 * page faults already hold the mmap_lock.
156 * Hence to serialise fully against both syscall and mmap based IO, we need to
157 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
158 * taken in places where we need to invalidate the page cache in a race
159 * free manner (e.g. truncate, hole punch and other extent manipulation
167 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
170 * You can't set both SHARED and EXCL for the same lock,
171 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
172 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
174 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
175 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
176 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
177 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
178 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
179 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
180 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
182 if (lock_flags
& XFS_IOLOCK_EXCL
) {
183 down_write_nested(&VFS_I(ip
)->i_rwsem
,
184 XFS_IOLOCK_DEP(lock_flags
));
185 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
186 down_read_nested(&VFS_I(ip
)->i_rwsem
,
187 XFS_IOLOCK_DEP(lock_flags
));
190 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
191 mrupdate_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
192 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
193 mraccess_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
195 if (lock_flags
& XFS_ILOCK_EXCL
)
196 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
197 else if (lock_flags
& XFS_ILOCK_SHARED
)
198 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
202 * This is just like xfs_ilock(), except that the caller
203 * is guaranteed not to sleep. It returns 1 if it gets
204 * the requested locks and 0 otherwise. If the IO lock is
205 * obtained but the inode lock cannot be, then the IO lock
206 * is dropped before returning.
208 * ip -- the inode being locked
209 * lock_flags -- this parameter indicates the inode's locks to be
210 * to be locked. See the comment for xfs_ilock() for a list
218 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
221 * You can't set both SHARED and EXCL for the same lock,
222 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
223 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
225 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
226 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
227 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
228 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
229 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
230 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
231 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
233 if (lock_flags
& XFS_IOLOCK_EXCL
) {
234 if (!down_write_trylock(&VFS_I(ip
)->i_rwsem
))
236 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
237 if (!down_read_trylock(&VFS_I(ip
)->i_rwsem
))
241 if (lock_flags
& XFS_MMAPLOCK_EXCL
) {
242 if (!mrtryupdate(&ip
->i_mmaplock
))
243 goto out_undo_iolock
;
244 } else if (lock_flags
& XFS_MMAPLOCK_SHARED
) {
245 if (!mrtryaccess(&ip
->i_mmaplock
))
246 goto out_undo_iolock
;
249 if (lock_flags
& XFS_ILOCK_EXCL
) {
250 if (!mrtryupdate(&ip
->i_lock
))
251 goto out_undo_mmaplock
;
252 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
253 if (!mrtryaccess(&ip
->i_lock
))
254 goto out_undo_mmaplock
;
259 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
260 mrunlock_excl(&ip
->i_mmaplock
);
261 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
262 mrunlock_shared(&ip
->i_mmaplock
);
264 if (lock_flags
& XFS_IOLOCK_EXCL
)
265 up_write(&VFS_I(ip
)->i_rwsem
);
266 else if (lock_flags
& XFS_IOLOCK_SHARED
)
267 up_read(&VFS_I(ip
)->i_rwsem
);
273 * xfs_iunlock() is used to drop the inode locks acquired with
274 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
275 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
276 * that we know which locks to drop.
278 * ip -- the inode being unlocked
279 * lock_flags -- this parameter indicates the inode's locks to be
280 * to be unlocked. See the comment for xfs_ilock() for a list
281 * of valid values for this parameter.
290 * You can't set both SHARED and EXCL for the same lock,
291 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
292 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
294 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
295 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
296 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
297 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
298 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
299 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
300 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
301 ASSERT(lock_flags
!= 0);
303 if (lock_flags
& XFS_IOLOCK_EXCL
)
304 up_write(&VFS_I(ip
)->i_rwsem
);
305 else if (lock_flags
& XFS_IOLOCK_SHARED
)
306 up_read(&VFS_I(ip
)->i_rwsem
);
308 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
309 mrunlock_excl(&ip
->i_mmaplock
);
310 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
311 mrunlock_shared(&ip
->i_mmaplock
);
313 if (lock_flags
& XFS_ILOCK_EXCL
)
314 mrunlock_excl(&ip
->i_lock
);
315 else if (lock_flags
& XFS_ILOCK_SHARED
)
316 mrunlock_shared(&ip
->i_lock
);
318 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
322 * give up write locks. the i/o lock cannot be held nested
323 * if it is being demoted.
330 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
));
332 ~(XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
334 if (lock_flags
& XFS_ILOCK_EXCL
)
335 mrdemote(&ip
->i_lock
);
336 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
337 mrdemote(&ip
->i_mmaplock
);
338 if (lock_flags
& XFS_IOLOCK_EXCL
)
339 downgrade_write(&VFS_I(ip
)->i_rwsem
);
341 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
344 #if defined(DEBUG) || defined(XFS_WARN)
350 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
351 if (!(lock_flags
& XFS_ILOCK_SHARED
))
352 return !!ip
->i_lock
.mr_writer
;
353 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
356 if (lock_flags
& (XFS_MMAPLOCK_EXCL
|XFS_MMAPLOCK_SHARED
)) {
357 if (!(lock_flags
& XFS_MMAPLOCK_SHARED
))
358 return !!ip
->i_mmaplock
.mr_writer
;
359 return rwsem_is_locked(&ip
->i_mmaplock
.mr_lock
);
362 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
363 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
364 return !debug_locks
||
365 lockdep_is_held_type(&VFS_I(ip
)->i_rwsem
, 0);
366 return rwsem_is_locked(&VFS_I(ip
)->i_rwsem
);
375 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
376 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
377 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
378 * errors and warnings.
380 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
382 xfs_lockdep_subclass_ok(
385 return subclass
< MAX_LOCKDEP_SUBCLASSES
;
388 #define xfs_lockdep_subclass_ok(subclass) (true)
392 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
393 * value. This can be called for any type of inode lock combination, including
394 * parent locking. Care must be taken to ensure we don't overrun the subclass
395 * storage fields in the class mask we build.
398 xfs_lock_inumorder(int lock_mode
, int subclass
)
402 ASSERT(!(lock_mode
& (XFS_ILOCK_PARENT
| XFS_ILOCK_RTBITMAP
|
404 ASSERT(xfs_lockdep_subclass_ok(subclass
));
406 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
407 ASSERT(subclass
<= XFS_IOLOCK_MAX_SUBCLASS
);
408 class += subclass
<< XFS_IOLOCK_SHIFT
;
411 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) {
412 ASSERT(subclass
<= XFS_MMAPLOCK_MAX_SUBCLASS
);
413 class += subclass
<< XFS_MMAPLOCK_SHIFT
;
416 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) {
417 ASSERT(subclass
<= XFS_ILOCK_MAX_SUBCLASS
);
418 class += subclass
<< XFS_ILOCK_SHIFT
;
421 return (lock_mode
& ~XFS_LOCK_SUBCLASS_MASK
) | class;
425 * The following routine will lock n inodes in exclusive mode. We assume the
426 * caller calls us with the inodes in i_ino order.
428 * We need to detect deadlock where an inode that we lock is in the AIL and we
429 * start waiting for another inode that is locked by a thread in a long running
430 * transaction (such as truncate). This can result in deadlock since the long
431 * running trans might need to wait for the inode we just locked in order to
432 * push the tail and free space in the log.
434 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
435 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
436 * lock more than one at a time, lockdep will report false positives saying we
437 * have violated locking orders.
441 struct xfs_inode
**ips
,
445 int attempts
= 0, i
, j
, try_lock
;
446 struct xfs_log_item
*lp
;
449 * Currently supports between 2 and 5 inodes with exclusive locking. We
450 * support an arbitrary depth of locking here, but absolute limits on
451 * inodes depend on the type of locking and the limits placed by
452 * lockdep annotations in xfs_lock_inumorder. These are all checked by
455 ASSERT(ips
&& inodes
>= 2 && inodes
<= 5);
456 ASSERT(lock_mode
& (XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
|
458 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
| XFS_MMAPLOCK_SHARED
|
460 ASSERT(!(lock_mode
& XFS_MMAPLOCK_EXCL
) ||
461 inodes
<= XFS_MMAPLOCK_MAX_SUBCLASS
+ 1);
462 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
) ||
463 inodes
<= XFS_ILOCK_MAX_SUBCLASS
+ 1);
465 if (lock_mode
& XFS_IOLOCK_EXCL
) {
466 ASSERT(!(lock_mode
& (XFS_MMAPLOCK_EXCL
| XFS_ILOCK_EXCL
)));
467 } else if (lock_mode
& XFS_MMAPLOCK_EXCL
)
468 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
));
473 for (; i
< inodes
; i
++) {
476 if (i
&& (ips
[i
] == ips
[i
- 1])) /* Already locked */
480 * If try_lock is not set yet, make sure all locked inodes are
481 * not in the AIL. If any are, set try_lock to be used later.
484 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
485 lp
= &ips
[j
]->i_itemp
->ili_item
;
486 if (lp
&& test_bit(XFS_LI_IN_AIL
, &lp
->li_flags
))
492 * If any of the previous locks we have locked is in the AIL,
493 * we must TRY to get the second and subsequent locks. If
494 * we can't get any, we must release all we have
498 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
502 /* try_lock means we have an inode locked that is in the AIL. */
504 if (xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
)))
508 * Unlock all previous guys and try again. xfs_iunlock will try
509 * to push the tail if the inode is in the AIL.
512 for (j
= i
- 1; j
>= 0; j
--) {
514 * Check to see if we've already unlocked this one. Not
515 * the first one going back, and the inode ptr is the
518 if (j
!= (i
- 1) && ips
[j
] == ips
[j
+ 1])
521 xfs_iunlock(ips
[j
], lock_mode
);
524 if ((attempts
% 5) == 0) {
525 delay(1); /* Don't just spin the CPU */
534 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
535 * the mmaplock or the ilock, but not more than one type at a time. If we lock
536 * more than one at a time, lockdep will report false positives saying we have
537 * violated locking orders. The iolock must be double-locked separately since
538 * we use i_rwsem for that. We now support taking one lock EXCL and the other
543 struct xfs_inode
*ip0
,
545 struct xfs_inode
*ip1
,
548 struct xfs_inode
*temp
;
551 struct xfs_log_item
*lp
;
553 ASSERT(hweight32(ip0_mode
) == 1);
554 ASSERT(hweight32(ip1_mode
) == 1);
555 ASSERT(!(ip0_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
556 ASSERT(!(ip1_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
557 ASSERT(!(ip0_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
558 !(ip0_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
559 ASSERT(!(ip1_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
560 !(ip1_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
561 ASSERT(!(ip1_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
562 !(ip0_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
563 ASSERT(!(ip0_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) ||
564 !(ip1_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
566 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
568 if (ip0
->i_ino
> ip1
->i_ino
) {
572 mode_temp
= ip0_mode
;
574 ip1_mode
= mode_temp
;
578 xfs_ilock(ip0
, xfs_lock_inumorder(ip0_mode
, 0));
581 * If the first lock we have locked is in the AIL, we must TRY to get
582 * the second lock. If we can't get it, we must release the first one
585 lp
= &ip0
->i_itemp
->ili_item
;
586 if (lp
&& test_bit(XFS_LI_IN_AIL
, &lp
->li_flags
)) {
587 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(ip1_mode
, 1))) {
588 xfs_iunlock(ip0
, ip0_mode
);
589 if ((++attempts
% 5) == 0)
590 delay(1); /* Don't just spin the CPU */
594 xfs_ilock(ip1
, xfs_lock_inumorder(ip1_mode
, 1));
600 struct xfs_inode
*ip
)
604 if (ip
->i_diflags
& XFS_DIFLAG_ANY
) {
605 if (ip
->i_diflags
& XFS_DIFLAG_REALTIME
)
606 flags
|= FS_XFLAG_REALTIME
;
607 if (ip
->i_diflags
& XFS_DIFLAG_PREALLOC
)
608 flags
|= FS_XFLAG_PREALLOC
;
609 if (ip
->i_diflags
& XFS_DIFLAG_IMMUTABLE
)
610 flags
|= FS_XFLAG_IMMUTABLE
;
611 if (ip
->i_diflags
& XFS_DIFLAG_APPEND
)
612 flags
|= FS_XFLAG_APPEND
;
613 if (ip
->i_diflags
& XFS_DIFLAG_SYNC
)
614 flags
|= FS_XFLAG_SYNC
;
615 if (ip
->i_diflags
& XFS_DIFLAG_NOATIME
)
616 flags
|= FS_XFLAG_NOATIME
;
617 if (ip
->i_diflags
& XFS_DIFLAG_NODUMP
)
618 flags
|= FS_XFLAG_NODUMP
;
619 if (ip
->i_diflags
& XFS_DIFLAG_RTINHERIT
)
620 flags
|= FS_XFLAG_RTINHERIT
;
621 if (ip
->i_diflags
& XFS_DIFLAG_PROJINHERIT
)
622 flags
|= FS_XFLAG_PROJINHERIT
;
623 if (ip
->i_diflags
& XFS_DIFLAG_NOSYMLINKS
)
624 flags
|= FS_XFLAG_NOSYMLINKS
;
625 if (ip
->i_diflags
& XFS_DIFLAG_EXTSIZE
)
626 flags
|= FS_XFLAG_EXTSIZE
;
627 if (ip
->i_diflags
& XFS_DIFLAG_EXTSZINHERIT
)
628 flags
|= FS_XFLAG_EXTSZINHERIT
;
629 if (ip
->i_diflags
& XFS_DIFLAG_NODEFRAG
)
630 flags
|= FS_XFLAG_NODEFRAG
;
631 if (ip
->i_diflags
& XFS_DIFLAG_FILESTREAM
)
632 flags
|= FS_XFLAG_FILESTREAM
;
635 if (ip
->i_diflags2
& XFS_DIFLAG2_ANY
) {
636 if (ip
->i_diflags2
& XFS_DIFLAG2_DAX
)
637 flags
|= FS_XFLAG_DAX
;
638 if (ip
->i_diflags2
& XFS_DIFLAG2_COWEXTSIZE
)
639 flags
|= FS_XFLAG_COWEXTSIZE
;
643 flags
|= FS_XFLAG_HASATTR
;
648 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
649 * is allowed, otherwise it has to be an exact match. If a CI match is found,
650 * ci_name->name will point to a the actual name (caller must free) or
651 * will be set to NULL if an exact match is found.
656 struct xfs_name
*name
,
658 struct xfs_name
*ci_name
)
663 trace_xfs_lookup(dp
, name
);
665 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
668 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
672 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
680 kmem_free(ci_name
->name
);
686 /* Propagate di_flags from a parent inode to a child inode. */
688 xfs_inode_inherit_flags(
689 struct xfs_inode
*ip
,
690 const struct xfs_inode
*pip
)
692 unsigned int di_flags
= 0;
693 umode_t mode
= VFS_I(ip
)->i_mode
;
696 if (pip
->i_diflags
& XFS_DIFLAG_RTINHERIT
)
697 di_flags
|= XFS_DIFLAG_RTINHERIT
;
698 if (pip
->i_diflags
& XFS_DIFLAG_EXTSZINHERIT
) {
699 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
700 ip
->i_extsize
= pip
->i_extsize
;
702 if (pip
->i_diflags
& XFS_DIFLAG_PROJINHERIT
)
703 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
704 } else if (S_ISREG(mode
)) {
705 if ((pip
->i_diflags
& XFS_DIFLAG_RTINHERIT
) &&
706 xfs_sb_version_hasrealtime(&ip
->i_mount
->m_sb
))
707 di_flags
|= XFS_DIFLAG_REALTIME
;
708 if (pip
->i_diflags
& XFS_DIFLAG_EXTSZINHERIT
) {
709 di_flags
|= XFS_DIFLAG_EXTSIZE
;
710 ip
->i_extsize
= pip
->i_extsize
;
713 if ((pip
->i_diflags
& XFS_DIFLAG_NOATIME
) &&
715 di_flags
|= XFS_DIFLAG_NOATIME
;
716 if ((pip
->i_diflags
& XFS_DIFLAG_NODUMP
) &&
718 di_flags
|= XFS_DIFLAG_NODUMP
;
719 if ((pip
->i_diflags
& XFS_DIFLAG_SYNC
) &&
721 di_flags
|= XFS_DIFLAG_SYNC
;
722 if ((pip
->i_diflags
& XFS_DIFLAG_NOSYMLINKS
) &&
723 xfs_inherit_nosymlinks
)
724 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
725 if ((pip
->i_diflags
& XFS_DIFLAG_NODEFRAG
) &&
726 xfs_inherit_nodefrag
)
727 di_flags
|= XFS_DIFLAG_NODEFRAG
;
728 if (pip
->i_diflags
& XFS_DIFLAG_FILESTREAM
)
729 di_flags
|= XFS_DIFLAG_FILESTREAM
;
731 ip
->i_diflags
|= di_flags
;
734 /* Propagate di_flags2 from a parent inode to a child inode. */
736 xfs_inode_inherit_flags2(
737 struct xfs_inode
*ip
,
738 const struct xfs_inode
*pip
)
740 if (pip
->i_diflags2
& XFS_DIFLAG2_COWEXTSIZE
) {
741 ip
->i_diflags2
|= XFS_DIFLAG2_COWEXTSIZE
;
742 ip
->i_cowextsize
= pip
->i_cowextsize
;
744 if (pip
->i_diflags2
& XFS_DIFLAG2_DAX
)
745 ip
->i_diflags2
|= XFS_DIFLAG2_DAX
;
749 * Initialise a newly allocated inode and return the in-core inode to the
750 * caller locked exclusively.
754 struct user_namespace
*mnt_userns
,
755 struct xfs_trans
*tp
,
756 struct xfs_inode
*pip
,
763 struct xfs_inode
**ipp
)
765 struct inode
*dir
= pip
? VFS_I(pip
) : NULL
;
766 struct xfs_mount
*mp
= tp
->t_mountp
;
767 struct xfs_inode
*ip
;
770 struct timespec64 tv
;
774 * Protect against obviously corrupt allocation btree records. Later
775 * xfs_iget checks will catch re-allocation of other active in-memory
776 * and on-disk inodes. If we don't catch reallocating the parent inode
777 * here we will deadlock in xfs_iget() so we have to do these checks
780 if ((pip
&& ino
== pip
->i_ino
) || !xfs_verify_dir_ino(mp
, ino
)) {
781 xfs_alert(mp
, "Allocated a known in-use inode 0x%llx!", ino
);
782 return -EFSCORRUPTED
;
786 * Get the in-core inode with the lock held exclusively to prevent
787 * others from looking at until we're done.
789 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
795 set_nlink(inode
, nlink
);
796 inode
->i_rdev
= rdev
;
799 if (dir
&& !(dir
->i_mode
& S_ISGID
) &&
800 (mp
->m_flags
& XFS_MOUNT_GRPID
)) {
801 inode_fsuid_set(inode
, mnt_userns
);
802 inode
->i_gid
= dir
->i_gid
;
803 inode
->i_mode
= mode
;
805 inode_init_owner(mnt_userns
, inode
, dir
, mode
);
809 * If the group ID of the new file does not match the effective group
810 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
811 * (and only if the irix_sgid_inherit compatibility variable is set).
813 if (irix_sgid_inherit
&&
814 (inode
->i_mode
& S_ISGID
) &&
815 !in_group_p(i_gid_into_mnt(mnt_userns
, inode
)))
816 inode
->i_mode
&= ~S_ISGID
;
819 ip
->i_df
.if_nextents
= 0;
820 ASSERT(ip
->i_nblocks
== 0);
822 tv
= current_time(inode
);
830 if (xfs_sb_version_has_v3inode(&mp
->m_sb
)) {
831 inode_set_iversion(inode
, 1);
832 ip
->i_cowextsize
= 0;
836 flags
= XFS_ILOG_CORE
;
837 switch (mode
& S_IFMT
) {
842 ip
->i_df
.if_format
= XFS_DINODE_FMT_DEV
;
843 flags
|= XFS_ILOG_DEV
;
847 if (pip
&& (pip
->i_diflags
& XFS_DIFLAG_ANY
))
848 xfs_inode_inherit_flags(ip
, pip
);
849 if (pip
&& (pip
->i_diflags2
& XFS_DIFLAG2_ANY
))
850 xfs_inode_inherit_flags2(ip
, pip
);
853 ip
->i_df
.if_format
= XFS_DINODE_FMT_EXTENTS
;
854 ip
->i_df
.if_bytes
= 0;
855 ip
->i_df
.if_u1
.if_root
= NULL
;
862 * If we need to create attributes immediately after allocating the
863 * inode, initialise an empty attribute fork right now. We use the
864 * default fork offset for attributes here as we don't know exactly what
865 * size or how many attributes we might be adding. We can do this
866 * safely here because we know the data fork is completely empty and
867 * this saves us from needing to run a separate transaction to set the
868 * fork offset in the immediate future.
870 if (init_xattrs
&& xfs_sb_version_hasattr(&mp
->m_sb
)) {
871 ip
->i_forkoff
= xfs_default_attroffset(ip
) >> 3;
872 ip
->i_afp
= xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS
, 0);
876 * Log the new values stuffed into the inode.
878 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
879 xfs_trans_log_inode(tp
, ip
, flags
);
881 /* now that we have an i_mode we can setup the inode structure */
889 * Allocates a new inode from disk and return a pointer to the incore copy. This
890 * routine will internally commit the current transaction and allocate a new one
891 * if we needed to allocate more on-disk free inodes to perform the requested
894 * If we are allocating quota inodes, we do not have a parent inode to attach to
895 * or associate with (i.e. dp == NULL) because they are not linked into the
896 * directory structure - they are attached directly to the superblock - and so
901 struct user_namespace
*mnt_userns
,
902 struct xfs_trans
**tpp
,
903 struct xfs_inode
*dp
,
909 struct xfs_inode
**ipp
)
911 struct xfs_buf
*agibp
;
912 xfs_ino_t parent_ino
= dp
? dp
->i_ino
: 0;
916 ASSERT((*tpp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
919 * Call the space management code to pick the on-disk inode to be
922 error
= xfs_dialloc_select_ag(tpp
, parent_ino
, mode
, &agibp
);
929 /* Allocate an inode from the selected AG */
930 error
= xfs_dialloc_ag(*tpp
, agibp
, parent_ino
, &ino
);
933 ASSERT(ino
!= NULLFSINO
);
935 return xfs_init_new_inode(mnt_userns
, *tpp
, dp
, ino
, mode
, nlink
, rdev
,
936 prid
, init_xattrs
, ipp
);
940 * Decrement the link count on an inode & log the change. If this causes the
941 * link count to go to zero, move the inode to AGI unlinked list so that it can
942 * be freed when the last active reference goes away via xfs_inactive().
944 static int /* error */
949 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
951 drop_nlink(VFS_I(ip
));
952 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
954 if (VFS_I(ip
)->i_nlink
)
957 return xfs_iunlink(tp
, ip
);
961 * Increment the link count on an inode & log the change.
968 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
970 inc_nlink(VFS_I(ip
));
971 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
976 struct user_namespace
*mnt_userns
,
978 struct xfs_name
*name
,
984 int is_dir
= S_ISDIR(mode
);
985 struct xfs_mount
*mp
= dp
->i_mount
;
986 struct xfs_inode
*ip
= NULL
;
987 struct xfs_trans
*tp
= NULL
;
989 bool unlock_dp_on_error
= false;
991 struct xfs_dquot
*udqp
= NULL
;
992 struct xfs_dquot
*gdqp
= NULL
;
993 struct xfs_dquot
*pdqp
= NULL
;
994 struct xfs_trans_res
*tres
;
997 trace_xfs_create(dp
, name
);
999 if (XFS_FORCED_SHUTDOWN(mp
))
1002 prid
= xfs_get_initial_prid(dp
);
1005 * Make sure that we have allocated dquot(s) on disk.
1007 error
= xfs_qm_vop_dqalloc(dp
, mapped_fsuid(mnt_userns
),
1008 mapped_fsgid(mnt_userns
), prid
,
1009 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1010 &udqp
, &gdqp
, &pdqp
);
1015 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1016 tres
= &M_RES(mp
)->tr_mkdir
;
1018 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1019 tres
= &M_RES(mp
)->tr_create
;
1023 * Initially assume that the file does not exist and
1024 * reserve the resources for that case. If that is not
1025 * the case we'll drop the one we have and get a more
1026 * appropriate transaction later.
1028 error
= xfs_trans_alloc_icreate(mp
, tres
, udqp
, gdqp
, pdqp
, resblks
,
1030 if (error
== -ENOSPC
) {
1031 /* flush outstanding delalloc blocks and retry */
1032 xfs_flush_inodes(mp
);
1033 error
= xfs_trans_alloc_icreate(mp
, tres
, udqp
, gdqp
, pdqp
,
1037 goto out_release_dquots
;
1039 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1040 unlock_dp_on_error
= true;
1042 error
= xfs_iext_count_may_overflow(dp
, XFS_DATA_FORK
,
1043 XFS_IEXT_DIR_MANIP_CNT(mp
));
1045 goto out_trans_cancel
;
1048 * A newly created regular or special file just has one directory
1049 * entry pointing to them, but a directory also the "." entry
1050 * pointing to itself.
1052 error
= xfs_dir_ialloc(mnt_userns
, &tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1053 prid
, init_xattrs
, &ip
);
1055 goto out_trans_cancel
;
1058 * Now we join the directory inode to the transaction. We do not do it
1059 * earlier because xfs_dir_ialloc might commit the previous transaction
1060 * (and release all the locks). An error from here on will result in
1061 * the transaction cancel unlocking dp so don't do it explicitly in the
1064 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1065 unlock_dp_on_error
= false;
1067 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1068 resblks
- XFS_IALLOC_SPACE_RES(mp
));
1070 ASSERT(error
!= -ENOSPC
);
1071 goto out_trans_cancel
;
1073 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1074 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1077 error
= xfs_dir_init(tp
, ip
, dp
);
1079 goto out_trans_cancel
;
1081 xfs_bumplink(tp
, dp
);
1085 * If this is a synchronous mount, make sure that the
1086 * create transaction goes to disk before returning to
1089 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1090 xfs_trans_set_sync(tp
);
1093 * Attach the dquot(s) to the inodes and modify them incore.
1094 * These ids of the inode couldn't have changed since the new
1095 * inode has been locked ever since it was created.
1097 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1099 error
= xfs_trans_commit(tp
);
1101 goto out_release_inode
;
1103 xfs_qm_dqrele(udqp
);
1104 xfs_qm_dqrele(gdqp
);
1105 xfs_qm_dqrele(pdqp
);
1111 xfs_trans_cancel(tp
);
1114 * Wait until after the current transaction is aborted to finish the
1115 * setup of the inode and release the inode. This prevents recursive
1116 * transactions and deadlocks from xfs_inactive.
1119 xfs_finish_inode_setup(ip
);
1123 xfs_qm_dqrele(udqp
);
1124 xfs_qm_dqrele(gdqp
);
1125 xfs_qm_dqrele(pdqp
);
1127 if (unlock_dp_on_error
)
1128 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1134 struct user_namespace
*mnt_userns
,
1135 struct xfs_inode
*dp
,
1137 struct xfs_inode
**ipp
)
1139 struct xfs_mount
*mp
= dp
->i_mount
;
1140 struct xfs_inode
*ip
= NULL
;
1141 struct xfs_trans
*tp
= NULL
;
1144 struct xfs_dquot
*udqp
= NULL
;
1145 struct xfs_dquot
*gdqp
= NULL
;
1146 struct xfs_dquot
*pdqp
= NULL
;
1147 struct xfs_trans_res
*tres
;
1150 if (XFS_FORCED_SHUTDOWN(mp
))
1153 prid
= xfs_get_initial_prid(dp
);
1156 * Make sure that we have allocated dquot(s) on disk.
1158 error
= xfs_qm_vop_dqalloc(dp
, mapped_fsuid(mnt_userns
),
1159 mapped_fsgid(mnt_userns
), prid
,
1160 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1161 &udqp
, &gdqp
, &pdqp
);
1165 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1166 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1168 error
= xfs_trans_alloc_icreate(mp
, tres
, udqp
, gdqp
, pdqp
, resblks
,
1171 goto out_release_dquots
;
1173 error
= xfs_dir_ialloc(mnt_userns
, &tp
, dp
, mode
, 0, 0, prid
,
1176 goto out_trans_cancel
;
1178 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1179 xfs_trans_set_sync(tp
);
1182 * Attach the dquot(s) to the inodes and modify them incore.
1183 * These ids of the inode couldn't have changed since the new
1184 * inode has been locked ever since it was created.
1186 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1188 error
= xfs_iunlink(tp
, ip
);
1190 goto out_trans_cancel
;
1192 error
= xfs_trans_commit(tp
);
1194 goto out_release_inode
;
1196 xfs_qm_dqrele(udqp
);
1197 xfs_qm_dqrele(gdqp
);
1198 xfs_qm_dqrele(pdqp
);
1204 xfs_trans_cancel(tp
);
1207 * Wait until after the current transaction is aborted to finish the
1208 * setup of the inode and release the inode. This prevents recursive
1209 * transactions and deadlocks from xfs_inactive.
1212 xfs_finish_inode_setup(ip
);
1216 xfs_qm_dqrele(udqp
);
1217 xfs_qm_dqrele(gdqp
);
1218 xfs_qm_dqrele(pdqp
);
1227 struct xfs_name
*target_name
)
1229 xfs_mount_t
*mp
= tdp
->i_mount
;
1234 trace_xfs_link(tdp
, target_name
);
1236 ASSERT(!S_ISDIR(VFS_I(sip
)->i_mode
));
1238 if (XFS_FORCED_SHUTDOWN(mp
))
1241 error
= xfs_qm_dqattach(sip
);
1245 error
= xfs_qm_dqattach(tdp
);
1249 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1250 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, resblks
, 0, 0, &tp
);
1251 if (error
== -ENOSPC
) {
1253 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, 0, 0, 0, &tp
);
1258 xfs_lock_two_inodes(sip
, XFS_ILOCK_EXCL
, tdp
, XFS_ILOCK_EXCL
);
1260 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1261 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1263 error
= xfs_iext_count_may_overflow(tdp
, XFS_DATA_FORK
,
1264 XFS_IEXT_DIR_MANIP_CNT(mp
));
1269 * If we are using project inheritance, we only allow hard link
1270 * creation in our tree when the project IDs are the same; else
1271 * the tree quota mechanism could be circumvented.
1273 if (unlikely((tdp
->i_diflags
& XFS_DIFLAG_PROJINHERIT
) &&
1274 tdp
->i_projid
!= sip
->i_projid
)) {
1280 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1286 * Handle initial link state of O_TMPFILE inode
1288 if (VFS_I(sip
)->i_nlink
== 0) {
1289 error
= xfs_iunlink_remove(tp
, sip
);
1294 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1298 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1299 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1301 xfs_bumplink(tp
, sip
);
1304 * If this is a synchronous mount, make sure that the
1305 * link transaction goes to disk before returning to
1308 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1309 xfs_trans_set_sync(tp
);
1311 return xfs_trans_commit(tp
);
1314 xfs_trans_cancel(tp
);
1319 /* Clear the reflink flag and the cowblocks tag if possible. */
1321 xfs_itruncate_clear_reflink_flags(
1322 struct xfs_inode
*ip
)
1324 struct xfs_ifork
*dfork
;
1325 struct xfs_ifork
*cfork
;
1327 if (!xfs_is_reflink_inode(ip
))
1329 dfork
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1330 cfork
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
1331 if (dfork
->if_bytes
== 0 && cfork
->if_bytes
== 0)
1332 ip
->i_diflags2
&= ~XFS_DIFLAG2_REFLINK
;
1333 if (cfork
->if_bytes
== 0)
1334 xfs_inode_clear_cowblocks_tag(ip
);
1338 * Free up the underlying blocks past new_size. The new size must be smaller
1339 * than the current size. This routine can be used both for the attribute and
1340 * data fork, and does not modify the inode size, which is left to the caller.
1342 * The transaction passed to this routine must have made a permanent log
1343 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1344 * given transaction and start new ones, so make sure everything involved in
1345 * the transaction is tidy before calling here. Some transaction will be
1346 * returned to the caller to be committed. The incoming transaction must
1347 * already include the inode, and both inode locks must be held exclusively.
1348 * The inode must also be "held" within the transaction. On return the inode
1349 * will be "held" within the returned transaction. This routine does NOT
1350 * require any disk space to be reserved for it within the transaction.
1352 * If we get an error, we must return with the inode locked and linked into the
1353 * current transaction. This keeps things simple for the higher level code,
1354 * because it always knows that the inode is locked and held in the transaction
1355 * that returns to it whether errors occur or not. We don't mark the inode
1356 * dirty on error so that transactions can be easily aborted if possible.
1359 xfs_itruncate_extents_flags(
1360 struct xfs_trans
**tpp
,
1361 struct xfs_inode
*ip
,
1363 xfs_fsize_t new_size
,
1366 struct xfs_mount
*mp
= ip
->i_mount
;
1367 struct xfs_trans
*tp
= *tpp
;
1368 xfs_fileoff_t first_unmap_block
;
1369 xfs_filblks_t unmap_len
;
1372 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1373 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1374 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1375 ASSERT(new_size
<= XFS_ISIZE(ip
));
1376 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1377 ASSERT(ip
->i_itemp
!= NULL
);
1378 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1379 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1381 trace_xfs_itruncate_extents_start(ip
, new_size
);
1383 flags
|= xfs_bmapi_aflag(whichfork
);
1386 * Since it is possible for space to become allocated beyond
1387 * the end of the file (in a crash where the space is allocated
1388 * but the inode size is not yet updated), simply remove any
1389 * blocks which show up between the new EOF and the maximum
1390 * possible file size.
1392 * We have to free all the blocks to the bmbt maximum offset, even if
1393 * the page cache can't scale that far.
1395 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1396 if (!xfs_verify_fileoff(mp
, first_unmap_block
)) {
1397 WARN_ON_ONCE(first_unmap_block
> XFS_MAX_FILEOFF
);
1401 unmap_len
= XFS_MAX_FILEOFF
- first_unmap_block
+ 1;
1402 while (unmap_len
> 0) {
1403 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
);
1404 error
= __xfs_bunmapi(tp
, ip
, first_unmap_block
, &unmap_len
,
1405 flags
, XFS_ITRUNC_MAX_EXTENTS
);
1409 /* free the just unmapped extents */
1410 error
= xfs_defer_finish(&tp
);
1415 if (whichfork
== XFS_DATA_FORK
) {
1416 /* Remove all pending CoW reservations. */
1417 error
= xfs_reflink_cancel_cow_blocks(ip
, &tp
,
1418 first_unmap_block
, XFS_MAX_FILEOFF
, true);
1422 xfs_itruncate_clear_reflink_flags(ip
);
1426 * Always re-log the inode so that our permanent transaction can keep
1427 * on rolling it forward in the log.
1429 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1431 trace_xfs_itruncate_extents_end(ip
, new_size
);
1442 xfs_mount_t
*mp
= ip
->i_mount
;
1445 if (!S_ISREG(VFS_I(ip
)->i_mode
) || (VFS_I(ip
)->i_mode
== 0))
1448 /* If this is a read-only mount, don't do this (would generate I/O) */
1449 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1452 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1456 * If we previously truncated this file and removed old data
1457 * in the process, we want to initiate "early" writeout on
1458 * the last close. This is an attempt to combat the notorious
1459 * NULL files problem which is particularly noticeable from a
1460 * truncate down, buffered (re-)write (delalloc), followed by
1461 * a crash. What we are effectively doing here is
1462 * significantly reducing the time window where we'd otherwise
1463 * be exposed to that problem.
1465 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1467 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1468 if (ip
->i_delayed_blks
> 0) {
1469 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1476 if (VFS_I(ip
)->i_nlink
== 0)
1480 * If we can't get the iolock just skip truncating the blocks past EOF
1481 * because we could deadlock with the mmap_lock otherwise. We'll get
1482 * another chance to drop them once the last reference to the inode is
1483 * dropped, so we'll never leak blocks permanently.
1485 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
))
1488 if (xfs_can_free_eofblocks(ip
, false)) {
1490 * Check if the inode is being opened, written and closed
1491 * frequently and we have delayed allocation blocks outstanding
1492 * (e.g. streaming writes from the NFS server), truncating the
1493 * blocks past EOF will cause fragmentation to occur.
1495 * In this case don't do the truncation, but we have to be
1496 * careful how we detect this case. Blocks beyond EOF show up as
1497 * i_delayed_blks even when the inode is clean, so we need to
1498 * truncate them away first before checking for a dirty release.
1499 * Hence on the first dirty close we will still remove the
1500 * speculative allocation, but after that we will leave it in
1503 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1506 error
= xfs_free_eofblocks(ip
);
1510 /* delalloc blocks after truncation means it really is dirty */
1511 if (ip
->i_delayed_blks
)
1512 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1516 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1521 * xfs_inactive_truncate
1523 * Called to perform a truncate when an inode becomes unlinked.
1526 xfs_inactive_truncate(
1527 struct xfs_inode
*ip
)
1529 struct xfs_mount
*mp
= ip
->i_mount
;
1530 struct xfs_trans
*tp
;
1533 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
1535 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1538 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1539 xfs_trans_ijoin(tp
, ip
, 0);
1542 * Log the inode size first to prevent stale data exposure in the event
1543 * of a system crash before the truncate completes. See the related
1544 * comment in xfs_vn_setattr_size() for details.
1546 ip
->i_disk_size
= 0;
1547 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1549 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1551 goto error_trans_cancel
;
1553 ASSERT(ip
->i_df
.if_nextents
== 0);
1555 error
= xfs_trans_commit(tp
);
1559 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1563 xfs_trans_cancel(tp
);
1565 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1570 * xfs_inactive_ifree()
1572 * Perform the inode free when an inode is unlinked.
1576 struct xfs_inode
*ip
)
1578 struct xfs_mount
*mp
= ip
->i_mount
;
1579 struct xfs_trans
*tp
;
1583 * We try to use a per-AG reservation for any block needed by the finobt
1584 * tree, but as the finobt feature predates the per-AG reservation
1585 * support a degraded file system might not have enough space for the
1586 * reservation at mount time. In that case try to dip into the reserved
1589 * Send a warning if the reservation does happen to fail, as the inode
1590 * now remains allocated and sits on the unlinked list until the fs is
1593 if (unlikely(mp
->m_finobt_nores
)) {
1594 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
,
1595 XFS_IFREE_SPACE_RES(mp
), 0, XFS_TRANS_RESERVE
,
1598 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
, 0, 0, 0, &tp
);
1601 if (error
== -ENOSPC
) {
1602 xfs_warn_ratelimited(mp
,
1603 "Failed to remove inode(s) from unlinked list. "
1604 "Please free space, unmount and run xfs_repair.");
1606 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1612 * We do not hold the inode locked across the entire rolling transaction
1613 * here. We only need to hold it for the first transaction that
1614 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1615 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1616 * here breaks the relationship between cluster buffer invalidation and
1617 * stale inode invalidation on cluster buffer item journal commit
1618 * completion, and can result in leaving dirty stale inodes hanging
1621 * We have no need for serialising this inode operation against other
1622 * operations - we freed the inode and hence reallocation is required
1623 * and that will serialise on reallocating the space the deferops need
1624 * to free. Hence we can unlock the inode on the first commit of
1625 * the transaction rather than roll it right through the deferops. This
1626 * avoids relogging the XFS_ISTALE inode.
1628 * We check that xfs_ifree() hasn't grown an internal transaction roll
1629 * by asserting that the inode is still locked when it returns.
1631 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1632 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1634 error
= xfs_ifree(tp
, ip
);
1635 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1638 * If we fail to free the inode, shut down. The cancel
1639 * might do that, we need to make sure. Otherwise the
1640 * inode might be lost for a long time or forever.
1642 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1643 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1645 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1647 xfs_trans_cancel(tp
);
1652 * Credit the quota account(s). The inode is gone.
1654 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1657 * Just ignore errors at this point. There is nothing we can do except
1658 * to try to keep going. Make sure it's not a silent error.
1660 error
= xfs_trans_commit(tp
);
1662 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1671 * This is called when the vnode reference count for the vnode
1672 * goes to zero. If the file has been unlinked, then it must
1673 * now be truncated. Also, we clear all of the read-ahead state
1674 * kept for the inode here since the file is now closed.
1680 struct xfs_mount
*mp
;
1685 * If the inode is already free, then there can be nothing
1688 if (VFS_I(ip
)->i_mode
== 0) {
1689 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1694 ASSERT(!xfs_iflags_test(ip
, XFS_IRECOVERY
));
1696 /* If this is a read-only mount, don't do this (would generate I/O) */
1697 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1700 /* Metadata inodes require explicit resource cleanup. */
1701 if (xfs_is_metadata_inode(ip
))
1704 /* Try to clean out the cow blocks if there are any. */
1705 if (xfs_inode_has_cow_data(ip
))
1706 xfs_reflink_cancel_cow_range(ip
, 0, NULLFILEOFF
, true);
1708 if (VFS_I(ip
)->i_nlink
!= 0) {
1710 * force is true because we are evicting an inode from the
1711 * cache. Post-eof blocks must be freed, lest we end up with
1712 * broken free space accounting.
1714 * Note: don't bother with iolock here since lockdep complains
1715 * about acquiring it in reclaim context. We have the only
1716 * reference to the inode at this point anyways.
1718 if (xfs_can_free_eofblocks(ip
, true))
1719 xfs_free_eofblocks(ip
);
1724 if (S_ISREG(VFS_I(ip
)->i_mode
) &&
1725 (ip
->i_disk_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1726 ip
->i_df
.if_nextents
> 0 || ip
->i_delayed_blks
> 0))
1729 error
= xfs_qm_dqattach(ip
);
1733 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1734 error
= xfs_inactive_symlink(ip
);
1736 error
= xfs_inactive_truncate(ip
);
1741 * If there are attributes associated with the file then blow them away
1742 * now. The code calls a routine that recursively deconstructs the
1743 * attribute fork. If also blows away the in-core attribute fork.
1745 if (XFS_IFORK_Q(ip
)) {
1746 error
= xfs_attr_inactive(ip
);
1752 ASSERT(ip
->i_forkoff
== 0);
1757 error
= xfs_inactive_ifree(ip
);
1762 * Release the dquots held by inode, if any.
1764 xfs_qm_dqdetach(ip
);
1768 * In-Core Unlinked List Lookups
1769 * =============================
1771 * Every inode is supposed to be reachable from some other piece of metadata
1772 * with the exception of the root directory. Inodes with a connection to a
1773 * file descriptor but not linked from anywhere in the on-disk directory tree
1774 * are collectively known as unlinked inodes, though the filesystem itself
1775 * maintains links to these inodes so that on-disk metadata are consistent.
1777 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1778 * header contains a number of buckets that point to an inode, and each inode
1779 * record has a pointer to the next inode in the hash chain. This
1780 * singly-linked list causes scaling problems in the iunlink remove function
1781 * because we must walk that list to find the inode that points to the inode
1782 * being removed from the unlinked hash bucket list.
1784 * What if we modelled the unlinked list as a collection of records capturing
1785 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1786 * have a fast way to look up unlinked list predecessors, which avoids the
1787 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1790 * Because this is a backref cache, we ignore operational failures since the
1791 * iunlink code can fall back to the slow bucket walk. The only errors that
1792 * should bubble out are for obviously incorrect situations.
1794 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1795 * access or have otherwise provided for concurrency control.
1798 /* Capture a "X.next_unlinked = Y" relationship. */
1799 struct xfs_iunlink
{
1800 struct rhash_head iu_rhash_head
;
1801 xfs_agino_t iu_agino
; /* X */
1802 xfs_agino_t iu_next_unlinked
; /* Y */
1805 /* Unlinked list predecessor lookup hashtable construction */
1807 xfs_iunlink_obj_cmpfn(
1808 struct rhashtable_compare_arg
*arg
,
1811 const xfs_agino_t
*key
= arg
->key
;
1812 const struct xfs_iunlink
*iu
= obj
;
1814 if (iu
->iu_next_unlinked
!= *key
)
1819 static const struct rhashtable_params xfs_iunlink_hash_params
= {
1820 .min_size
= XFS_AGI_UNLINKED_BUCKETS
,
1821 .key_len
= sizeof(xfs_agino_t
),
1822 .key_offset
= offsetof(struct xfs_iunlink
,
1824 .head_offset
= offsetof(struct xfs_iunlink
, iu_rhash_head
),
1825 .automatic_shrinking
= true,
1826 .obj_cmpfn
= xfs_iunlink_obj_cmpfn
,
1830 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1831 * relation is found.
1834 xfs_iunlink_lookup_backref(
1835 struct xfs_perag
*pag
,
1838 struct xfs_iunlink
*iu
;
1840 iu
= rhashtable_lookup_fast(&pag
->pagi_unlinked_hash
, &agino
,
1841 xfs_iunlink_hash_params
);
1842 return iu
? iu
->iu_agino
: NULLAGINO
;
1846 * Take ownership of an iunlink cache entry and insert it into the hash table.
1847 * If successful, the entry will be owned by the cache; if not, it is freed.
1848 * Either way, the caller does not own @iu after this call.
1851 xfs_iunlink_insert_backref(
1852 struct xfs_perag
*pag
,
1853 struct xfs_iunlink
*iu
)
1857 error
= rhashtable_insert_fast(&pag
->pagi_unlinked_hash
,
1858 &iu
->iu_rhash_head
, xfs_iunlink_hash_params
);
1860 * Fail loudly if there already was an entry because that's a sign of
1861 * corruption of in-memory data. Also fail loudly if we see an error
1862 * code we didn't anticipate from the rhashtable code. Currently we
1863 * only anticipate ENOMEM.
1866 WARN(error
!= -ENOMEM
, "iunlink cache insert error %d", error
);
1870 * Absorb any runtime errors that aren't a result of corruption because
1871 * this is a cache and we can always fall back to bucket list scanning.
1873 if (error
!= 0 && error
!= -EEXIST
)
1878 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1880 xfs_iunlink_add_backref(
1881 struct xfs_perag
*pag
,
1882 xfs_agino_t prev_agino
,
1883 xfs_agino_t this_agino
)
1885 struct xfs_iunlink
*iu
;
1887 if (XFS_TEST_ERROR(false, pag
->pag_mount
, XFS_ERRTAG_IUNLINK_FALLBACK
))
1890 iu
= kmem_zalloc(sizeof(*iu
), KM_NOFS
);
1891 iu
->iu_agino
= prev_agino
;
1892 iu
->iu_next_unlinked
= this_agino
;
1894 return xfs_iunlink_insert_backref(pag
, iu
);
1898 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
1899 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
1900 * wasn't any such entry then we don't bother.
1903 xfs_iunlink_change_backref(
1904 struct xfs_perag
*pag
,
1906 xfs_agino_t next_unlinked
)
1908 struct xfs_iunlink
*iu
;
1911 /* Look up the old entry; if there wasn't one then exit. */
1912 iu
= rhashtable_lookup_fast(&pag
->pagi_unlinked_hash
, &agino
,
1913 xfs_iunlink_hash_params
);
1918 * Remove the entry. This shouldn't ever return an error, but if we
1919 * couldn't remove the old entry we don't want to add it again to the
1920 * hash table, and if the entry disappeared on us then someone's
1921 * violated the locking rules and we need to fail loudly. Either way
1922 * we cannot remove the inode because internal state is or would have
1925 error
= rhashtable_remove_fast(&pag
->pagi_unlinked_hash
,
1926 &iu
->iu_rhash_head
, xfs_iunlink_hash_params
);
1930 /* If there is no new next entry just free our item and return. */
1931 if (next_unlinked
== NULLAGINO
) {
1936 /* Update the entry and re-add it to the hash table. */
1937 iu
->iu_next_unlinked
= next_unlinked
;
1938 return xfs_iunlink_insert_backref(pag
, iu
);
1941 /* Set up the in-core predecessor structures. */
1944 struct xfs_perag
*pag
)
1946 return rhashtable_init(&pag
->pagi_unlinked_hash
,
1947 &xfs_iunlink_hash_params
);
1950 /* Free the in-core predecessor structures. */
1952 xfs_iunlink_free_item(
1956 struct xfs_iunlink
*iu
= ptr
;
1957 bool *freed_anything
= arg
;
1959 *freed_anything
= true;
1964 xfs_iunlink_destroy(
1965 struct xfs_perag
*pag
)
1967 bool freed_anything
= false;
1969 rhashtable_free_and_destroy(&pag
->pagi_unlinked_hash
,
1970 xfs_iunlink_free_item
, &freed_anything
);
1972 ASSERT(freed_anything
== false || XFS_FORCED_SHUTDOWN(pag
->pag_mount
));
1976 * Point the AGI unlinked bucket at an inode and log the results. The caller
1977 * is responsible for validating the old value.
1980 xfs_iunlink_update_bucket(
1981 struct xfs_trans
*tp
,
1982 xfs_agnumber_t agno
,
1983 struct xfs_buf
*agibp
,
1984 unsigned int bucket_index
,
1985 xfs_agino_t new_agino
)
1987 struct xfs_agi
*agi
= agibp
->b_addr
;
1988 xfs_agino_t old_value
;
1991 ASSERT(xfs_verify_agino_or_null(tp
->t_mountp
, agno
, new_agino
));
1993 old_value
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1994 trace_xfs_iunlink_update_bucket(tp
->t_mountp
, agno
, bucket_index
,
1995 old_value
, new_agino
);
1998 * We should never find the head of the list already set to the value
1999 * passed in because either we're adding or removing ourselves from the
2002 if (old_value
== new_agino
) {
2003 xfs_buf_mark_corrupt(agibp
);
2004 return -EFSCORRUPTED
;
2007 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(new_agino
);
2008 offset
= offsetof(struct xfs_agi
, agi_unlinked
) +
2009 (sizeof(xfs_agino_t
) * bucket_index
);
2010 xfs_trans_log_buf(tp
, agibp
, offset
, offset
+ sizeof(xfs_agino_t
) - 1);
2014 /* Set an on-disk inode's next_unlinked pointer. */
2016 xfs_iunlink_update_dinode(
2017 struct xfs_trans
*tp
,
2018 xfs_agnumber_t agno
,
2020 struct xfs_buf
*ibp
,
2021 struct xfs_dinode
*dip
,
2022 struct xfs_imap
*imap
,
2023 xfs_agino_t next_agino
)
2025 struct xfs_mount
*mp
= tp
->t_mountp
;
2028 ASSERT(xfs_verify_agino_or_null(mp
, agno
, next_agino
));
2030 trace_xfs_iunlink_update_dinode(mp
, agno
, agino
,
2031 be32_to_cpu(dip
->di_next_unlinked
), next_agino
);
2033 dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2034 offset
= imap
->im_boffset
+
2035 offsetof(struct xfs_dinode
, di_next_unlinked
);
2037 /* need to recalc the inode CRC if appropriate */
2038 xfs_dinode_calc_crc(mp
, dip
);
2039 xfs_trans_inode_buf(tp
, ibp
);
2040 xfs_trans_log_buf(tp
, ibp
, offset
, offset
+ sizeof(xfs_agino_t
) - 1);
2043 /* Set an in-core inode's unlinked pointer and return the old value. */
2045 xfs_iunlink_update_inode(
2046 struct xfs_trans
*tp
,
2047 struct xfs_inode
*ip
,
2048 xfs_agnumber_t agno
,
2049 xfs_agino_t next_agino
,
2050 xfs_agino_t
*old_next_agino
)
2052 struct xfs_mount
*mp
= tp
->t_mountp
;
2053 struct xfs_dinode
*dip
;
2054 struct xfs_buf
*ibp
;
2055 xfs_agino_t old_value
;
2058 ASSERT(xfs_verify_agino_or_null(mp
, agno
, next_agino
));
2060 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &ibp
);
2063 dip
= xfs_buf_offset(ibp
, ip
->i_imap
.im_boffset
);
2065 /* Make sure the old pointer isn't garbage. */
2066 old_value
= be32_to_cpu(dip
->di_next_unlinked
);
2067 if (!xfs_verify_agino_or_null(mp
, agno
, old_value
)) {
2068 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
, dip
,
2069 sizeof(*dip
), __this_address
);
2070 error
= -EFSCORRUPTED
;
2075 * Since we're updating a linked list, we should never find that the
2076 * current pointer is the same as the new value, unless we're
2077 * terminating the list.
2079 *old_next_agino
= old_value
;
2080 if (old_value
== next_agino
) {
2081 if (next_agino
!= NULLAGINO
) {
2082 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
,
2083 dip
, sizeof(*dip
), __this_address
);
2084 error
= -EFSCORRUPTED
;
2089 /* Ok, update the new pointer. */
2090 xfs_iunlink_update_dinode(tp
, agno
, XFS_INO_TO_AGINO(mp
, ip
->i_ino
),
2091 ibp
, dip
, &ip
->i_imap
, next_agino
);
2094 xfs_trans_brelse(tp
, ibp
);
2099 * This is called when the inode's link count has gone to 0 or we are creating
2100 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2102 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2103 * list when the inode is freed.
2107 struct xfs_trans
*tp
,
2108 struct xfs_inode
*ip
)
2110 struct xfs_mount
*mp
= tp
->t_mountp
;
2111 struct xfs_agi
*agi
;
2112 struct xfs_buf
*agibp
;
2113 xfs_agino_t next_agino
;
2114 xfs_agnumber_t agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2115 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2116 short bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2119 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2120 ASSERT(VFS_I(ip
)->i_mode
!= 0);
2121 trace_xfs_iunlink(ip
);
2123 /* Get the agi buffer first. It ensures lock ordering on the list. */
2124 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2127 agi
= agibp
->b_addr
;
2130 * Get the index into the agi hash table for the list this inode will
2131 * go on. Make sure the pointer isn't garbage and that this inode
2132 * isn't already on the list.
2134 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2135 if (next_agino
== agino
||
2136 !xfs_verify_agino_or_null(mp
, agno
, next_agino
)) {
2137 xfs_buf_mark_corrupt(agibp
);
2138 return -EFSCORRUPTED
;
2141 if (next_agino
!= NULLAGINO
) {
2142 xfs_agino_t old_agino
;
2145 * There is already another inode in the bucket, so point this
2146 * inode to the current head of the list.
2148 error
= xfs_iunlink_update_inode(tp
, ip
, agno
, next_agino
,
2152 ASSERT(old_agino
== NULLAGINO
);
2155 * agino has been unlinked, add a backref from the next inode
2158 error
= xfs_iunlink_add_backref(agibp
->b_pag
, agino
, next_agino
);
2163 /* Point the head of the list to point to this inode. */
2164 return xfs_iunlink_update_bucket(tp
, agno
, agibp
, bucket_index
, agino
);
2167 /* Return the imap, dinode pointer, and buffer for an inode. */
2169 xfs_iunlink_map_ino(
2170 struct xfs_trans
*tp
,
2171 xfs_agnumber_t agno
,
2173 struct xfs_imap
*imap
,
2174 struct xfs_dinode
**dipp
,
2175 struct xfs_buf
**bpp
)
2177 struct xfs_mount
*mp
= tp
->t_mountp
;
2181 error
= xfs_imap(mp
, tp
, XFS_AGINO_TO_INO(mp
, agno
, agino
), imap
, 0);
2183 xfs_warn(mp
, "%s: xfs_imap returned error %d.",
2188 error
= xfs_imap_to_bp(mp
, tp
, imap
, bpp
);
2190 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2195 *dipp
= xfs_buf_offset(*bpp
, imap
->im_boffset
);
2200 * Walk the unlinked chain from @head_agino until we find the inode that
2201 * points to @target_agino. Return the inode number, map, dinode pointer,
2202 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2204 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2205 * @agino, @imap, @dipp, and @bpp are all output parameters.
2207 * Do not call this function if @target_agino is the head of the list.
2210 xfs_iunlink_map_prev(
2211 struct xfs_trans
*tp
,
2212 xfs_agnumber_t agno
,
2213 xfs_agino_t head_agino
,
2214 xfs_agino_t target_agino
,
2216 struct xfs_imap
*imap
,
2217 struct xfs_dinode
**dipp
,
2218 struct xfs_buf
**bpp
,
2219 struct xfs_perag
*pag
)
2221 struct xfs_mount
*mp
= tp
->t_mountp
;
2222 xfs_agino_t next_agino
;
2225 ASSERT(head_agino
!= target_agino
);
2228 /* See if our backref cache can find it faster. */
2229 *agino
= xfs_iunlink_lookup_backref(pag
, target_agino
);
2230 if (*agino
!= NULLAGINO
) {
2231 error
= xfs_iunlink_map_ino(tp
, agno
, *agino
, imap
, dipp
, bpp
);
2235 if (be32_to_cpu((*dipp
)->di_next_unlinked
) == target_agino
)
2239 * If we get here the cache contents were corrupt, so drop the
2240 * buffer and fall back to walking the bucket list.
2242 xfs_trans_brelse(tp
, *bpp
);
2247 trace_xfs_iunlink_map_prev_fallback(mp
, agno
);
2249 /* Otherwise, walk the entire bucket until we find it. */
2250 next_agino
= head_agino
;
2251 while (next_agino
!= target_agino
) {
2252 xfs_agino_t unlinked_agino
;
2255 xfs_trans_brelse(tp
, *bpp
);
2257 *agino
= next_agino
;
2258 error
= xfs_iunlink_map_ino(tp
, agno
, next_agino
, imap
, dipp
,
2263 unlinked_agino
= be32_to_cpu((*dipp
)->di_next_unlinked
);
2265 * Make sure this pointer is valid and isn't an obvious
2268 if (!xfs_verify_agino(mp
, agno
, unlinked_agino
) ||
2269 next_agino
== unlinked_agino
) {
2270 XFS_CORRUPTION_ERROR(__func__
,
2271 XFS_ERRLEVEL_LOW
, mp
,
2272 *dipp
, sizeof(**dipp
));
2273 error
= -EFSCORRUPTED
;
2276 next_agino
= unlinked_agino
;
2283 * Pull the on-disk inode from the AGI unlinked list.
2287 struct xfs_trans
*tp
,
2288 struct xfs_inode
*ip
)
2290 struct xfs_mount
*mp
= tp
->t_mountp
;
2291 struct xfs_agi
*agi
;
2292 struct xfs_buf
*agibp
;
2293 struct xfs_buf
*last_ibp
;
2294 struct xfs_dinode
*last_dip
= NULL
;
2295 xfs_agnumber_t agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2296 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2297 xfs_agino_t next_agino
;
2298 xfs_agino_t head_agino
;
2299 short bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2302 trace_xfs_iunlink_remove(ip
);
2304 /* Get the agi buffer first. It ensures lock ordering on the list. */
2305 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2308 agi
= agibp
->b_addr
;
2311 * Get the index into the agi hash table for the list this inode will
2312 * go on. Make sure the head pointer isn't garbage.
2314 head_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2315 if (!xfs_verify_agino(mp
, agno
, head_agino
)) {
2316 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
,
2318 return -EFSCORRUPTED
;
2322 * Set our inode's next_unlinked pointer to NULL and then return
2323 * the old pointer value so that we can update whatever was previous
2324 * to us in the list to point to whatever was next in the list.
2326 error
= xfs_iunlink_update_inode(tp
, ip
, agno
, NULLAGINO
, &next_agino
);
2331 * If there was a backref pointing from the next inode back to this
2332 * one, remove it because we've removed this inode from the list.
2334 * Later, if this inode was in the middle of the list we'll update
2335 * this inode's backref to point from the next inode.
2337 if (next_agino
!= NULLAGINO
) {
2338 error
= xfs_iunlink_change_backref(agibp
->b_pag
, next_agino
,
2344 if (head_agino
!= agino
) {
2345 struct xfs_imap imap
;
2346 xfs_agino_t prev_agino
;
2348 /* We need to search the list for the inode being freed. */
2349 error
= xfs_iunlink_map_prev(tp
, agno
, head_agino
, agino
,
2350 &prev_agino
, &imap
, &last_dip
, &last_ibp
,
2355 /* Point the previous inode on the list to the next inode. */
2356 xfs_iunlink_update_dinode(tp
, agno
, prev_agino
, last_ibp
,
2357 last_dip
, &imap
, next_agino
);
2360 * Now we deal with the backref for this inode. If this inode
2361 * pointed at a real inode, change the backref that pointed to
2362 * us to point to our old next. If this inode was the end of
2363 * the list, delete the backref that pointed to us. Note that
2364 * change_backref takes care of deleting the backref if
2365 * next_agino is NULLAGINO.
2367 return xfs_iunlink_change_backref(agibp
->b_pag
, agino
,
2371 /* Point the head of the list to the next unlinked inode. */
2372 return xfs_iunlink_update_bucket(tp
, agno
, agibp
, bucket_index
,
2377 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2378 * mark it stale. We should only find clean inodes in this lookup that aren't
2382 xfs_ifree_mark_inode_stale(
2384 struct xfs_inode
*free_ip
,
2387 struct xfs_mount
*mp
= bp
->b_mount
;
2388 struct xfs_perag
*pag
= bp
->b_pag
;
2389 struct xfs_inode_log_item
*iip
;
2390 struct xfs_inode
*ip
;
2394 ip
= radix_tree_lookup(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, inum
));
2396 /* Inode not in memory, nothing to do */
2403 * because this is an RCU protected lookup, we could find a recently
2404 * freed or even reallocated inode during the lookup. We need to check
2405 * under the i_flags_lock for a valid inode here. Skip it if it is not
2406 * valid, the wrong inode or stale.
2408 spin_lock(&ip
->i_flags_lock
);
2409 if (ip
->i_ino
!= inum
|| __xfs_iflags_test(ip
, XFS_ISTALE
))
2410 goto out_iflags_unlock
;
2413 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2414 * other inodes that we did not find in the list attached to the buffer
2415 * and are not already marked stale. If we can't lock it, back off and
2418 if (ip
!= free_ip
) {
2419 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2420 spin_unlock(&ip
->i_flags_lock
);
2426 ip
->i_flags
|= XFS_ISTALE
;
2429 * If the inode is flushing, it is already attached to the buffer. All
2430 * we needed to do here is mark the inode stale so buffer IO completion
2431 * will remove it from the AIL.
2434 if (__xfs_iflags_test(ip
, XFS_IFLUSHING
)) {
2435 ASSERT(!list_empty(&iip
->ili_item
.li_bio_list
));
2436 ASSERT(iip
->ili_last_fields
);
2441 * Inodes not attached to the buffer can be released immediately.
2442 * Everything else has to go through xfs_iflush_abort() on journal
2443 * commit as the flock synchronises removal of the inode from the
2444 * cluster buffer against inode reclaim.
2446 if (!iip
|| list_empty(&iip
->ili_item
.li_bio_list
))
2449 __xfs_iflags_set(ip
, XFS_IFLUSHING
);
2450 spin_unlock(&ip
->i_flags_lock
);
2453 /* we have a dirty inode in memory that has not yet been flushed. */
2454 spin_lock(&iip
->ili_lock
);
2455 iip
->ili_last_fields
= iip
->ili_fields
;
2456 iip
->ili_fields
= 0;
2457 iip
->ili_fsync_fields
= 0;
2458 spin_unlock(&iip
->ili_lock
);
2459 ASSERT(iip
->ili_last_fields
);
2462 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2467 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2469 spin_unlock(&ip
->i_flags_lock
);
2474 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2475 * inodes that are in memory - they all must be marked stale and attached to
2476 * the cluster buffer.
2480 struct xfs_inode
*free_ip
,
2481 struct xfs_trans
*tp
,
2482 struct xfs_icluster
*xic
)
2484 struct xfs_mount
*mp
= free_ip
->i_mount
;
2485 struct xfs_ino_geometry
*igeo
= M_IGEO(mp
);
2488 xfs_ino_t inum
= xic
->first_ino
;
2494 nbufs
= igeo
->ialloc_blks
/ igeo
->blocks_per_cluster
;
2496 for (j
= 0; j
< nbufs
; j
++, inum
+= igeo
->inodes_per_cluster
) {
2498 * The allocation bitmap tells us which inodes of the chunk were
2499 * physically allocated. Skip the cluster if an inode falls into
2502 ioffset
= inum
- xic
->first_ino
;
2503 if ((xic
->alloc
& XFS_INOBT_MASK(ioffset
)) == 0) {
2504 ASSERT(ioffset
% igeo
->inodes_per_cluster
== 0);
2508 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2509 XFS_INO_TO_AGBNO(mp
, inum
));
2512 * We obtain and lock the backing buffer first in the process
2513 * here to ensure dirty inodes attached to the buffer remain in
2514 * the flushing state while we mark them stale.
2516 * If we scan the in-memory inodes first, then buffer IO can
2517 * complete before we get a lock on it, and hence we may fail
2518 * to mark all the active inodes on the buffer stale.
2520 error
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2521 mp
->m_bsize
* igeo
->blocks_per_cluster
,
2527 * This buffer may not have been correctly initialised as we
2528 * didn't read it from disk. That's not important because we are
2529 * only using to mark the buffer as stale in the log, and to
2530 * attach stale cached inodes on it. That means it will never be
2531 * dispatched for IO. If it is, we want to know about it, and we
2532 * want it to fail. We can acheive this by adding a write
2533 * verifier to the buffer.
2535 bp
->b_ops
= &xfs_inode_buf_ops
;
2538 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2539 * too. This requires lookups, and will skip inodes that we've
2540 * already marked XFS_ISTALE.
2542 for (i
= 0; i
< igeo
->inodes_per_cluster
; i
++)
2543 xfs_ifree_mark_inode_stale(bp
, free_ip
, inum
+ i
);
2545 xfs_trans_stale_inode_buf(tp
, bp
);
2546 xfs_trans_binval(tp
, bp
);
2552 * This is called to return an inode to the inode free list.
2553 * The inode should already be truncated to 0 length and have
2554 * no pages associated with it. This routine also assumes that
2555 * the inode is already a part of the transaction.
2557 * The on-disk copy of the inode will have been added to the list
2558 * of unlinked inodes in the AGI. We need to remove the inode from
2559 * that list atomically with respect to freeing it here.
2563 struct xfs_trans
*tp
,
2564 struct xfs_inode
*ip
)
2567 struct xfs_icluster xic
= { 0 };
2568 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
2570 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2571 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2572 ASSERT(ip
->i_df
.if_nextents
== 0);
2573 ASSERT(ip
->i_disk_size
== 0 || !S_ISREG(VFS_I(ip
)->i_mode
));
2574 ASSERT(ip
->i_nblocks
== 0);
2577 * Pull the on-disk inode from the AGI unlinked list.
2579 error
= xfs_iunlink_remove(tp
, ip
);
2583 error
= xfs_difree(tp
, ip
->i_ino
, &xic
);
2588 * Free any local-format data sitting around before we reset the
2589 * data fork to extents format. Note that the attr fork data has
2590 * already been freed by xfs_attr_inactive.
2592 if (ip
->i_df
.if_format
== XFS_DINODE_FMT_LOCAL
) {
2593 kmem_free(ip
->i_df
.if_u1
.if_data
);
2594 ip
->i_df
.if_u1
.if_data
= NULL
;
2595 ip
->i_df
.if_bytes
= 0;
2598 VFS_I(ip
)->i_mode
= 0; /* mark incore inode as free */
2600 ip
->i_diflags2
= ip
->i_mount
->m_ino_geo
.new_diflags2
;
2601 ip
->i_forkoff
= 0; /* mark the attr fork not in use */
2602 ip
->i_df
.if_format
= XFS_DINODE_FMT_EXTENTS
;
2603 if (xfs_iflags_test(ip
, XFS_IPRESERVE_DM_FIELDS
))
2604 xfs_iflags_clear(ip
, XFS_IPRESERVE_DM_FIELDS
);
2606 /* Don't attempt to replay owner changes for a deleted inode */
2607 spin_lock(&iip
->ili_lock
);
2608 iip
->ili_fields
&= ~(XFS_ILOG_AOWNER
| XFS_ILOG_DOWNER
);
2609 spin_unlock(&iip
->ili_lock
);
2612 * Bump the generation count so no one will be confused
2613 * by reincarnations of this inode.
2615 VFS_I(ip
)->i_generation
++;
2616 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2619 error
= xfs_ifree_cluster(ip
, tp
, &xic
);
2625 * This is called to unpin an inode. The caller must have the inode locked
2626 * in at least shared mode so that the buffer cannot be subsequently pinned
2627 * once someone is waiting for it to be unpinned.
2631 struct xfs_inode
*ip
)
2633 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2635 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2637 /* Give the log a push to start the unpinning I/O */
2638 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0, NULL
);
2644 struct xfs_inode
*ip
)
2646 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2647 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2652 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2653 if (xfs_ipincount(ip
))
2655 } while (xfs_ipincount(ip
));
2656 finish_wait(wq
, &wait
.wq_entry
);
2661 struct xfs_inode
*ip
)
2663 if (xfs_ipincount(ip
))
2664 __xfs_iunpin_wait(ip
);
2668 * Removing an inode from the namespace involves removing the directory entry
2669 * and dropping the link count on the inode. Removing the directory entry can
2670 * result in locking an AGF (directory blocks were freed) and removing a link
2671 * count can result in placing the inode on an unlinked list which results in
2674 * The big problem here is that we have an ordering constraint on AGF and AGI
2675 * locking - inode allocation locks the AGI, then can allocate a new extent for
2676 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2677 * removes the inode from the unlinked list, requiring that we lock the AGI
2678 * first, and then freeing the inode can result in an inode chunk being freed
2679 * and hence freeing disk space requiring that we lock an AGF.
2681 * Hence the ordering that is imposed by other parts of the code is AGI before
2682 * AGF. This means we cannot remove the directory entry before we drop the inode
2683 * reference count and put it on the unlinked list as this results in a lock
2684 * order of AGF then AGI, and this can deadlock against inode allocation and
2685 * freeing. Therefore we must drop the link counts before we remove the
2688 * This is still safe from a transactional point of view - it is not until we
2689 * get to xfs_defer_finish() that we have the possibility of multiple
2690 * transactions in this operation. Hence as long as we remove the directory
2691 * entry and drop the link count in the first transaction of the remove
2692 * operation, there are no transactional constraints on the ordering here.
2697 struct xfs_name
*name
,
2700 xfs_mount_t
*mp
= dp
->i_mount
;
2701 xfs_trans_t
*tp
= NULL
;
2702 int is_dir
= S_ISDIR(VFS_I(ip
)->i_mode
);
2706 trace_xfs_remove(dp
, name
);
2708 if (XFS_FORCED_SHUTDOWN(mp
))
2711 error
= xfs_qm_dqattach(dp
);
2715 error
= xfs_qm_dqattach(ip
);
2720 * We try to get the real space reservation first,
2721 * allowing for directory btree deletion(s) implying
2722 * possible bmap insert(s). If we can't get the space
2723 * reservation then we use 0 instead, and avoid the bmap
2724 * btree insert(s) in the directory code by, if the bmap
2725 * insert tries to happen, instead trimming the LAST
2726 * block from the directory.
2728 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2729 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, resblks
, 0, 0, &tp
);
2730 if (error
== -ENOSPC
) {
2732 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, 0, 0, 0,
2736 ASSERT(error
!= -ENOSPC
);
2740 xfs_lock_two_inodes(dp
, XFS_ILOCK_EXCL
, ip
, XFS_ILOCK_EXCL
);
2742 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2743 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2746 * If we're removing a directory perform some additional validation.
2749 ASSERT(VFS_I(ip
)->i_nlink
>= 2);
2750 if (VFS_I(ip
)->i_nlink
!= 2) {
2752 goto out_trans_cancel
;
2754 if (!xfs_dir_isempty(ip
)) {
2756 goto out_trans_cancel
;
2759 /* Drop the link from ip's "..". */
2760 error
= xfs_droplink(tp
, dp
);
2762 goto out_trans_cancel
;
2764 /* Drop the "." link from ip to self. */
2765 error
= xfs_droplink(tp
, ip
);
2767 goto out_trans_cancel
;
2770 * When removing a non-directory we need to log the parent
2771 * inode here. For a directory this is done implicitly
2772 * by the xfs_droplink call for the ".." entry.
2774 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2776 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2778 /* Drop the link from dp to ip. */
2779 error
= xfs_droplink(tp
, ip
);
2781 goto out_trans_cancel
;
2783 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
, resblks
);
2785 ASSERT(error
!= -ENOENT
);
2786 goto out_trans_cancel
;
2790 * If this is a synchronous mount, make sure that the
2791 * remove transaction goes to disk before returning to
2794 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2795 xfs_trans_set_sync(tp
);
2797 error
= xfs_trans_commit(tp
);
2801 if (is_dir
&& xfs_inode_is_filestream(ip
))
2802 xfs_filestream_deassociate(ip
);
2807 xfs_trans_cancel(tp
);
2813 * Enter all inodes for a rename transaction into a sorted array.
2815 #define __XFS_SORT_INODES 5
2817 xfs_sort_for_rename(
2818 struct xfs_inode
*dp1
, /* in: old (source) directory inode */
2819 struct xfs_inode
*dp2
, /* in: new (target) directory inode */
2820 struct xfs_inode
*ip1
, /* in: inode of old entry */
2821 struct xfs_inode
*ip2
, /* in: inode of new entry */
2822 struct xfs_inode
*wip
, /* in: whiteout inode */
2823 struct xfs_inode
**i_tab
,/* out: sorted array of inodes */
2824 int *num_inodes
) /* in/out: inodes in array */
2828 ASSERT(*num_inodes
== __XFS_SORT_INODES
);
2829 memset(i_tab
, 0, *num_inodes
* sizeof(struct xfs_inode
*));
2832 * i_tab contains a list of pointers to inodes. We initialize
2833 * the table here & we'll sort it. We will then use it to
2834 * order the acquisition of the inode locks.
2836 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2849 * Sort the elements via bubble sort. (Remember, there are at
2850 * most 5 elements to sort, so this is adequate.)
2852 for (i
= 0; i
< *num_inodes
; i
++) {
2853 for (j
= 1; j
< *num_inodes
; j
++) {
2854 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2855 struct xfs_inode
*temp
= i_tab
[j
];
2856 i_tab
[j
] = i_tab
[j
-1];
2865 struct xfs_trans
*tp
)
2868 * If this is a synchronous mount, make sure that the rename transaction
2869 * goes to disk before returning to the user.
2871 if (tp
->t_mountp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2872 xfs_trans_set_sync(tp
);
2874 return xfs_trans_commit(tp
);
2878 * xfs_cross_rename()
2880 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2884 struct xfs_trans
*tp
,
2885 struct xfs_inode
*dp1
,
2886 struct xfs_name
*name1
,
2887 struct xfs_inode
*ip1
,
2888 struct xfs_inode
*dp2
,
2889 struct xfs_name
*name2
,
2890 struct xfs_inode
*ip2
,
2898 /* Swap inode number for dirent in first parent */
2899 error
= xfs_dir_replace(tp
, dp1
, name1
, ip2
->i_ino
, spaceres
);
2901 goto out_trans_abort
;
2903 /* Swap inode number for dirent in second parent */
2904 error
= xfs_dir_replace(tp
, dp2
, name2
, ip1
->i_ino
, spaceres
);
2906 goto out_trans_abort
;
2909 * If we're renaming one or more directories across different parents,
2910 * update the respective ".." entries (and link counts) to match the new
2914 dp2_flags
= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2916 if (S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2917 error
= xfs_dir_replace(tp
, ip2
, &xfs_name_dotdot
,
2918 dp1
->i_ino
, spaceres
);
2920 goto out_trans_abort
;
2922 /* transfer ip2 ".." reference to dp1 */
2923 if (!S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2924 error
= xfs_droplink(tp
, dp2
);
2926 goto out_trans_abort
;
2927 xfs_bumplink(tp
, dp1
);
2931 * Although ip1 isn't changed here, userspace needs
2932 * to be warned about the change, so that applications
2933 * relying on it (like backup ones), will properly
2936 ip1_flags
|= XFS_ICHGTIME_CHG
;
2937 ip2_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2940 if (S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2941 error
= xfs_dir_replace(tp
, ip1
, &xfs_name_dotdot
,
2942 dp2
->i_ino
, spaceres
);
2944 goto out_trans_abort
;
2946 /* transfer ip1 ".." reference to dp2 */
2947 if (!S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2948 error
= xfs_droplink(tp
, dp1
);
2950 goto out_trans_abort
;
2951 xfs_bumplink(tp
, dp2
);
2955 * Although ip2 isn't changed here, userspace needs
2956 * to be warned about the change, so that applications
2957 * relying on it (like backup ones), will properly
2960 ip1_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2961 ip2_flags
|= XFS_ICHGTIME_CHG
;
2966 xfs_trans_ichgtime(tp
, ip1
, ip1_flags
);
2967 xfs_trans_log_inode(tp
, ip1
, XFS_ILOG_CORE
);
2970 xfs_trans_ichgtime(tp
, ip2
, ip2_flags
);
2971 xfs_trans_log_inode(tp
, ip2
, XFS_ILOG_CORE
);
2974 xfs_trans_ichgtime(tp
, dp2
, dp2_flags
);
2975 xfs_trans_log_inode(tp
, dp2
, XFS_ILOG_CORE
);
2977 xfs_trans_ichgtime(tp
, dp1
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2978 xfs_trans_log_inode(tp
, dp1
, XFS_ILOG_CORE
);
2979 return xfs_finish_rename(tp
);
2982 xfs_trans_cancel(tp
);
2987 * xfs_rename_alloc_whiteout()
2989 * Return a referenced, unlinked, unlocked inode that can be used as a
2990 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2991 * crash between allocating the inode and linking it into the rename transaction
2992 * recovery will free the inode and we won't leak it.
2995 xfs_rename_alloc_whiteout(
2996 struct user_namespace
*mnt_userns
,
2997 struct xfs_inode
*dp
,
2998 struct xfs_inode
**wip
)
3000 struct xfs_inode
*tmpfile
;
3003 error
= xfs_create_tmpfile(mnt_userns
, dp
, S_IFCHR
| WHITEOUT_MODE
,
3009 * Prepare the tmpfile inode as if it were created through the VFS.
3010 * Complete the inode setup and flag it as linkable. nlink is already
3011 * zero, so we can skip the drop_nlink.
3013 xfs_setup_iops(tmpfile
);
3014 xfs_finish_inode_setup(tmpfile
);
3015 VFS_I(tmpfile
)->i_state
|= I_LINKABLE
;
3026 struct user_namespace
*mnt_userns
,
3027 struct xfs_inode
*src_dp
,
3028 struct xfs_name
*src_name
,
3029 struct xfs_inode
*src_ip
,
3030 struct xfs_inode
*target_dp
,
3031 struct xfs_name
*target_name
,
3032 struct xfs_inode
*target_ip
,
3035 struct xfs_mount
*mp
= src_dp
->i_mount
;
3036 struct xfs_trans
*tp
;
3037 struct xfs_inode
*wip
= NULL
; /* whiteout inode */
3038 struct xfs_inode
*inodes
[__XFS_SORT_INODES
];
3040 int num_inodes
= __XFS_SORT_INODES
;
3041 bool new_parent
= (src_dp
!= target_dp
);
3042 bool src_is_directory
= S_ISDIR(VFS_I(src_ip
)->i_mode
);
3046 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
3048 if ((flags
& RENAME_EXCHANGE
) && !target_ip
)
3052 * If we are doing a whiteout operation, allocate the whiteout inode
3053 * we will be placing at the target and ensure the type is set
3056 if (flags
& RENAME_WHITEOUT
) {
3057 ASSERT(!(flags
& (RENAME_NOREPLACE
| RENAME_EXCHANGE
)));
3058 error
= xfs_rename_alloc_whiteout(mnt_userns
, target_dp
, &wip
);
3062 /* setup target dirent info as whiteout */
3063 src_name
->type
= XFS_DIR3_FT_CHRDEV
;
3066 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
, wip
,
3067 inodes
, &num_inodes
);
3069 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
3070 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, spaceres
, 0, 0, &tp
);
3071 if (error
== -ENOSPC
) {
3073 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, 0, 0, 0,
3077 goto out_release_wip
;
3080 * Attach the dquots to the inodes
3082 error
= xfs_qm_vop_rename_dqattach(inodes
);
3084 goto out_trans_cancel
;
3087 * Lock all the participating inodes. Depending upon whether
3088 * the target_name exists in the target directory, and
3089 * whether the target directory is the same as the source
3090 * directory, we can lock from 2 to 4 inodes.
3092 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
3095 * Join all the inodes to the transaction. From this point on,
3096 * we can rely on either trans_commit or trans_cancel to unlock
3099 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
3101 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
3102 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
3104 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
3106 xfs_trans_ijoin(tp
, wip
, XFS_ILOCK_EXCL
);
3109 * If we are using project inheritance, we only allow renames
3110 * into our tree when the project IDs are the same; else the
3111 * tree quota mechanism would be circumvented.
3113 if (unlikely((target_dp
->i_diflags
& XFS_DIFLAG_PROJINHERIT
) &&
3114 target_dp
->i_projid
!= src_ip
->i_projid
)) {
3116 goto out_trans_cancel
;
3119 /* RENAME_EXCHANGE is unique from here on. */
3120 if (flags
& RENAME_EXCHANGE
)
3121 return xfs_cross_rename(tp
, src_dp
, src_name
, src_ip
,
3122 target_dp
, target_name
, target_ip
,
3126 * Check for expected errors before we dirty the transaction
3127 * so we can return an error without a transaction abort.
3129 * Extent count overflow check:
3131 * From the perspective of src_dp, a rename operation is essentially a
3132 * directory entry remove operation. Hence the only place where we check
3133 * for extent count overflow for src_dp is in
3134 * xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
3135 * -ENOSPC when it detects a possible extent count overflow and in
3136 * response, the higher layers of directory handling code do the
3138 * 1. Data/Free blocks: XFS lets these blocks linger until a
3139 * future remove operation removes them.
3140 * 2. Dabtree blocks: XFS swaps the blocks with the last block in the
3141 * Leaf space and unmaps the last block.
3143 * For target_dp, there are two cases depending on whether the
3144 * destination directory entry exists or not.
3146 * When destination directory entry does not exist (i.e. target_ip ==
3147 * NULL), extent count overflow check is performed only when transaction
3148 * has a non-zero sized space reservation associated with it. With a
3149 * zero-sized space reservation, XFS allows a rename operation to
3150 * continue only when the directory has sufficient free space in its
3151 * data/leaf/free space blocks to hold the new entry.
3153 * When destination directory entry exists (i.e. target_ip != NULL), all
3154 * we need to do is change the inode number associated with the already
3155 * existing entry. Hence there is no need to perform an extent count
3158 if (target_ip
== NULL
) {
3160 * If there's no space reservation, check the entry will
3161 * fit before actually inserting it.
3164 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
3166 goto out_trans_cancel
;
3168 error
= xfs_iext_count_may_overflow(target_dp
,
3170 XFS_IEXT_DIR_MANIP_CNT(mp
));
3172 goto out_trans_cancel
;
3176 * If target exists and it's a directory, check that whether
3177 * it can be destroyed.
3179 if (S_ISDIR(VFS_I(target_ip
)->i_mode
) &&
3180 (!xfs_dir_isempty(target_ip
) ||
3181 (VFS_I(target_ip
)->i_nlink
> 2))) {
3183 goto out_trans_cancel
;
3188 * Lock the AGI buffers we need to handle bumping the nlink of the
3189 * whiteout inode off the unlinked list and to handle dropping the
3190 * nlink of the target inode. Per locking order rules, do this in
3191 * increasing AG order and before directory block allocation tries to
3192 * grab AGFs because we grab AGIs before AGFs.
3194 * The (vfs) caller must ensure that if src is a directory then
3195 * target_ip is either null or an empty directory.
3197 for (i
= 0; i
< num_inodes
&& inodes
[i
] != NULL
; i
++) {
3198 if (inodes
[i
] == wip
||
3199 (inodes
[i
] == target_ip
&&
3200 (VFS_I(target_ip
)->i_nlink
== 1 || src_is_directory
))) {
3202 xfs_agnumber_t agno
;
3204 agno
= XFS_INO_TO_AGNO(mp
, inodes
[i
]->i_ino
);
3205 error
= xfs_read_agi(mp
, tp
, agno
, &bp
);
3207 goto out_trans_cancel
;
3212 * Directory entry creation below may acquire the AGF. Remove
3213 * the whiteout from the unlinked list first to preserve correct
3214 * AGI/AGF locking order. This dirties the transaction so failures
3215 * after this point will abort and log recovery will clean up the
3218 * For whiteouts, we need to bump the link count on the whiteout
3219 * inode. After this point, we have a real link, clear the tmpfile
3220 * state flag from the inode so it doesn't accidentally get misused
3224 ASSERT(VFS_I(wip
)->i_nlink
== 0);
3225 error
= xfs_iunlink_remove(tp
, wip
);
3227 goto out_trans_cancel
;
3229 xfs_bumplink(tp
, wip
);
3230 VFS_I(wip
)->i_state
&= ~I_LINKABLE
;
3234 * Set up the target.
3236 if (target_ip
== NULL
) {
3238 * If target does not exist and the rename crosses
3239 * directories, adjust the target directory link count
3240 * to account for the ".." reference from the new entry.
3242 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
3243 src_ip
->i_ino
, spaceres
);
3245 goto out_trans_cancel
;
3247 xfs_trans_ichgtime(tp
, target_dp
,
3248 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3250 if (new_parent
&& src_is_directory
) {
3251 xfs_bumplink(tp
, target_dp
);
3253 } else { /* target_ip != NULL */
3255 * Link the source inode under the target name.
3256 * If the source inode is a directory and we are moving
3257 * it across directories, its ".." entry will be
3258 * inconsistent until we replace that down below.
3260 * In case there is already an entry with the same
3261 * name at the destination directory, remove it first.
3263 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
3264 src_ip
->i_ino
, spaceres
);
3266 goto out_trans_cancel
;
3268 xfs_trans_ichgtime(tp
, target_dp
,
3269 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3272 * Decrement the link count on the target since the target
3273 * dir no longer points to it.
3275 error
= xfs_droplink(tp
, target_ip
);
3277 goto out_trans_cancel
;
3279 if (src_is_directory
) {
3281 * Drop the link from the old "." entry.
3283 error
= xfs_droplink(tp
, target_ip
);
3285 goto out_trans_cancel
;
3287 } /* target_ip != NULL */
3290 * Remove the source.
3292 if (new_parent
&& src_is_directory
) {
3294 * Rewrite the ".." entry to point to the new
3297 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
3298 target_dp
->i_ino
, spaceres
);
3299 ASSERT(error
!= -EEXIST
);
3301 goto out_trans_cancel
;
3305 * We always want to hit the ctime on the source inode.
3307 * This isn't strictly required by the standards since the source
3308 * inode isn't really being changed, but old unix file systems did
3309 * it and some incremental backup programs won't work without it.
3311 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
3312 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
3315 * Adjust the link count on src_dp. This is necessary when
3316 * renaming a directory, either within one parent when
3317 * the target existed, or across two parent directories.
3319 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
3322 * Decrement link count on src_directory since the
3323 * entry that's moved no longer points to it.
3325 error
= xfs_droplink(tp
, src_dp
);
3327 goto out_trans_cancel
;
3331 * For whiteouts, we only need to update the source dirent with the
3332 * inode number of the whiteout inode rather than removing it
3336 error
= xfs_dir_replace(tp
, src_dp
, src_name
, wip
->i_ino
,
3340 * NOTE: We don't need to check for extent count overflow here
3341 * because the dir remove name code will leave the dir block in
3342 * place if the extent count would overflow.
3344 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
3349 goto out_trans_cancel
;
3351 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3352 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
3354 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
3356 error
= xfs_finish_rename(tp
);
3362 xfs_trans_cancel(tp
);
3371 struct xfs_inode
*ip
,
3374 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3375 struct xfs_dinode
*dip
;
3376 struct xfs_mount
*mp
= ip
->i_mount
;
3379 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3380 ASSERT(xfs_iflags_test(ip
, XFS_IFLUSHING
));
3381 ASSERT(ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
||
3382 ip
->i_df
.if_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3383 ASSERT(iip
->ili_item
.li_buf
== bp
);
3385 dip
= xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3388 * We don't flush the inode if any of the following checks fail, but we
3389 * do still update the log item and attach to the backing buffer as if
3390 * the flush happened. This is a formality to facilitate predictable
3391 * error handling as the caller will shutdown and fail the buffer.
3393 error
= -EFSCORRUPTED
;
3394 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3395 mp
, XFS_ERRTAG_IFLUSH_1
)) {
3396 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3397 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT
,
3398 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3401 if (S_ISREG(VFS_I(ip
)->i_mode
)) {
3403 ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
&&
3404 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
,
3405 mp
, XFS_ERRTAG_IFLUSH_3
)) {
3406 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3407 "%s: Bad regular inode %Lu, ptr "PTR_FMT
,
3408 __func__
, ip
->i_ino
, ip
);
3411 } else if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
3413 ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
&&
3414 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
&&
3415 ip
->i_df
.if_format
!= XFS_DINODE_FMT_LOCAL
,
3416 mp
, XFS_ERRTAG_IFLUSH_4
)) {
3417 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3418 "%s: Bad directory inode %Lu, ptr "PTR_FMT
,
3419 __func__
, ip
->i_ino
, ip
);
3423 if (XFS_TEST_ERROR(ip
->i_df
.if_nextents
+ xfs_ifork_nextents(ip
->i_afp
) >
3424 ip
->i_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
)) {
3425 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3426 "%s: detected corrupt incore inode %Lu, "
3427 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT
,
3428 __func__
, ip
->i_ino
,
3429 ip
->i_df
.if_nextents
+ xfs_ifork_nextents(ip
->i_afp
),
3433 if (XFS_TEST_ERROR(ip
->i_forkoff
> mp
->m_sb
.sb_inodesize
,
3434 mp
, XFS_ERRTAG_IFLUSH_6
)) {
3435 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3436 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT
,
3437 __func__
, ip
->i_ino
, ip
->i_forkoff
, ip
);
3442 * Inode item log recovery for v2 inodes are dependent on the flushiter
3443 * count for correct sequencing. We bump the flush iteration count so
3444 * we can detect flushes which postdate a log record during recovery.
3445 * This is redundant as we now log every change and hence this can't
3446 * happen but we need to still do it to ensure backwards compatibility
3447 * with old kernels that predate logging all inode changes.
3449 if (!xfs_sb_version_has_v3inode(&mp
->m_sb
))
3453 * If there are inline format data / attr forks attached to this inode,
3454 * make sure they are not corrupt.
3456 if (ip
->i_df
.if_format
== XFS_DINODE_FMT_LOCAL
&&
3457 xfs_ifork_verify_local_data(ip
))
3459 if (ip
->i_afp
&& ip
->i_afp
->if_format
== XFS_DINODE_FMT_LOCAL
&&
3460 xfs_ifork_verify_local_attr(ip
))
3464 * Copy the dirty parts of the inode into the on-disk inode. We always
3465 * copy out the core of the inode, because if the inode is dirty at all
3468 xfs_inode_to_disk(ip
, dip
, iip
->ili_item
.li_lsn
);
3470 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3471 if (!xfs_sb_version_has_v3inode(&mp
->m_sb
)) {
3472 if (ip
->i_flushiter
== DI_MAX_FLUSH
)
3473 ip
->i_flushiter
= 0;
3476 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3477 if (XFS_IFORK_Q(ip
))
3478 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3481 * We've recorded everything logged in the inode, so we'd like to clear
3482 * the ili_fields bits so we don't log and flush things unnecessarily.
3483 * However, we can't stop logging all this information until the data
3484 * we've copied into the disk buffer is written to disk. If we did we
3485 * might overwrite the copy of the inode in the log with all the data
3486 * after re-logging only part of it, and in the face of a crash we
3487 * wouldn't have all the data we need to recover.
3489 * What we do is move the bits to the ili_last_fields field. When
3490 * logging the inode, these bits are moved back to the ili_fields field.
3491 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3492 * we know that the information those bits represent is permanently on
3493 * disk. As long as the flush completes before the inode is logged
3494 * again, then both ili_fields and ili_last_fields will be cleared.
3498 spin_lock(&iip
->ili_lock
);
3499 iip
->ili_last_fields
= iip
->ili_fields
;
3500 iip
->ili_fields
= 0;
3501 iip
->ili_fsync_fields
= 0;
3502 spin_unlock(&iip
->ili_lock
);
3505 * Store the current LSN of the inode so that we can tell whether the
3506 * item has moved in the AIL from xfs_buf_inode_iodone().
3508 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3509 &iip
->ili_item
.li_lsn
);
3511 /* generate the checksum. */
3512 xfs_dinode_calc_crc(mp
, dip
);
3517 * Non-blocking flush of dirty inode metadata into the backing buffer.
3519 * The caller must have a reference to the inode and hold the cluster buffer
3520 * locked. The function will walk across all the inodes on the cluster buffer it
3521 * can find and lock without blocking, and flush them to the cluster buffer.
3523 * On successful flushing of at least one inode, the caller must write out the
3524 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3525 * the caller needs to release the buffer. On failure, the filesystem will be
3526 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3533 struct xfs_mount
*mp
= bp
->b_mount
;
3534 struct xfs_log_item
*lip
, *n
;
3535 struct xfs_inode
*ip
;
3536 struct xfs_inode_log_item
*iip
;
3541 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3542 * can remove itself from the list.
3544 list_for_each_entry_safe(lip
, n
, &bp
->b_li_list
, li_bio_list
) {
3545 iip
= (struct xfs_inode_log_item
*)lip
;
3546 ip
= iip
->ili_inode
;
3549 * Quick and dirty check to avoid locks if possible.
3551 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
| XFS_IFLUSHING
))
3553 if (xfs_ipincount(ip
))
3557 * The inode is still attached to the buffer, which means it is
3558 * dirty but reclaim might try to grab it. Check carefully for
3559 * that, and grab the ilock while still holding the i_flags_lock
3560 * to guarantee reclaim will not be able to reclaim this inode
3561 * once we drop the i_flags_lock.
3563 spin_lock(&ip
->i_flags_lock
);
3564 ASSERT(!__xfs_iflags_test(ip
, XFS_ISTALE
));
3565 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
| XFS_IFLUSHING
)) {
3566 spin_unlock(&ip
->i_flags_lock
);
3571 * ILOCK will pin the inode against reclaim and prevent
3572 * concurrent transactions modifying the inode while we are
3573 * flushing the inode. If we get the lock, set the flushing
3574 * state before we drop the i_flags_lock.
3576 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
3577 spin_unlock(&ip
->i_flags_lock
);
3580 __xfs_iflags_set(ip
, XFS_IFLUSHING
);
3581 spin_unlock(&ip
->i_flags_lock
);
3584 * Abort flushing this inode if we are shut down because the
3585 * inode may not currently be in the AIL. This can occur when
3586 * log I/O failure unpins the inode without inserting into the
3587 * AIL, leaving a dirty/unpinned inode attached to the buffer
3588 * that otherwise looks like it should be flushed.
3590 if (XFS_FORCED_SHUTDOWN(mp
)) {
3591 xfs_iunpin_wait(ip
);
3592 xfs_iflush_abort(ip
);
3593 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3598 /* don't block waiting on a log force to unpin dirty inodes */
3599 if (xfs_ipincount(ip
)) {
3600 xfs_iflags_clear(ip
, XFS_IFLUSHING
);
3601 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3605 if (!xfs_inode_clean(ip
))
3606 error
= xfs_iflush(ip
, bp
);
3608 xfs_iflags_clear(ip
, XFS_IFLUSHING
);
3609 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3616 bp
->b_flags
|= XBF_ASYNC
;
3617 xfs_buf_ioend_fail(bp
);
3618 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3625 XFS_STATS_INC(mp
, xs_icluster_flushcnt
);
3626 XFS_STATS_ADD(mp
, xs_icluster_flushinode
, clcount
);
3631 /* Release an inode. */
3634 struct xfs_inode
*ip
)
3636 trace_xfs_irele(ip
, _RET_IP_
);
3641 * Ensure all commited transactions touching the inode are written to the log.
3644 xfs_log_force_inode(
3645 struct xfs_inode
*ip
)
3649 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3650 if (xfs_ipincount(ip
))
3651 lsn
= ip
->i_itemp
->ili_last_lsn
;
3652 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3656 return xfs_log_force_lsn(ip
->i_mount
, lsn
, XFS_LOG_SYNC
, NULL
);
3660 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3661 * abide vfs locking order (lowest pointer value goes first) and breaking the
3662 * layout leases before proceeding. The loop is needed because we cannot call
3663 * the blocking break_layout() with the iolocks held, and therefore have to
3664 * back out both locks.
3667 xfs_iolock_two_inodes_and_break_layout(
3677 /* Wait to break both inodes' layouts before we start locking. */
3678 error
= break_layout(src
, true);
3682 error
= break_layout(dest
, true);
3687 /* Lock one inode and make sure nobody got in and leased it. */
3689 error
= break_layout(src
, false);
3692 if (error
== -EWOULDBLOCK
)
3700 /* Lock the other inode and make sure nobody got in and leased it. */
3701 inode_lock_nested(dest
, I_MUTEX_NONDIR2
);
3702 error
= break_layout(dest
, false);
3706 if (error
== -EWOULDBLOCK
)
3715 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3720 struct xfs_inode
*ip1
,
3721 struct xfs_inode
*ip2
)
3725 ret
= xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1
), VFS_I(ip2
));
3729 xfs_ilock(ip1
, XFS_MMAPLOCK_EXCL
);
3731 xfs_lock_two_inodes(ip1
, XFS_MMAPLOCK_EXCL
,
3732 ip2
, XFS_MMAPLOCK_EXCL
);
3736 /* Unlock both inodes to allow IO and mmap activity. */
3738 xfs_iunlock2_io_mmap(
3739 struct xfs_inode
*ip1
,
3740 struct xfs_inode
*ip2
)
3742 bool same_inode
= (ip1
== ip2
);
3744 xfs_iunlock(ip2
, XFS_MMAPLOCK_EXCL
);
3746 xfs_iunlock(ip1
, XFS_MMAPLOCK_EXCL
);
3747 inode_unlock(VFS_I(ip2
));
3749 inode_unlock(VFS_I(ip1
));