2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_attr_sf.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
54 kmem_zone_t
*xfs_inode_zone
;
57 * Used in xfs_itruncate_extents(). This is the maximum number of extents
58 * freed from a file in a single transaction.
60 #define XFS_ITRUNC_MAX_EXTENTS 2
62 STATIC
int xfs_iflush_int(struct xfs_inode
*, struct xfs_buf
*);
63 STATIC
int xfs_iunlink(struct xfs_trans
*, struct xfs_inode
*);
64 STATIC
int xfs_iunlink_remove(struct xfs_trans
*, struct xfs_inode
*);
67 * helper function to extract extent size hint from inode
73 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
74 return ip
->i_d
.di_extsize
;
75 if (XFS_IS_REALTIME_INODE(ip
))
76 return ip
->i_mount
->m_sb
.sb_rextsize
;
81 * Helper function to extract CoW extent size hint from inode.
82 * Between the extent size hint and the CoW extent size hint, we
83 * return the greater of the two. If the value is zero (automatic),
84 * use the default size.
87 xfs_get_cowextsz_hint(
93 if (ip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
94 a
= ip
->i_d
.di_cowextsize
;
95 b
= xfs_get_extsz_hint(ip
);
99 return XFS_DEFAULT_COWEXTSZ_HINT
;
104 * These two are wrapper routines around the xfs_ilock() routine used to
105 * centralize some grungy code. They are used in places that wish to lock the
106 * inode solely for reading the extents. The reason these places can't just
107 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
108 * bringing in of the extents from disk for a file in b-tree format. If the
109 * inode is in b-tree format, then we need to lock the inode exclusively until
110 * the extents are read in. Locking it exclusively all the time would limit
111 * our parallelism unnecessarily, though. What we do instead is check to see
112 * if the extents have been read in yet, and only lock the inode exclusively
115 * The functions return a value which should be given to the corresponding
116 * xfs_iunlock() call.
119 xfs_ilock_data_map_shared(
120 struct xfs_inode
*ip
)
122 uint lock_mode
= XFS_ILOCK_SHARED
;
124 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
&&
125 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)
126 lock_mode
= XFS_ILOCK_EXCL
;
127 xfs_ilock(ip
, lock_mode
);
132 xfs_ilock_attr_map_shared(
133 struct xfs_inode
*ip
)
135 uint lock_mode
= XFS_ILOCK_SHARED
;
137 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_BTREE
&&
138 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
139 lock_mode
= XFS_ILOCK_EXCL
;
140 xfs_ilock(ip
, lock_mode
);
145 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
146 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
147 * various combinations of the locks to be obtained.
149 * The 3 locks should always be ordered so that the IO lock is obtained first,
150 * the mmap lock second and the ilock last in order to prevent deadlock.
152 * Basic locking order:
154 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
156 * mmap_sem locking order:
158 * i_rwsem -> page lock -> mmap_sem
159 * mmap_sem -> i_mmap_lock -> page_lock
161 * The difference in mmap_sem locking order mean that we cannot hold the
162 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
163 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
164 * in get_user_pages() to map the user pages into the kernel address space for
165 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
166 * page faults already hold the mmap_sem.
168 * Hence to serialise fully against both syscall and mmap based IO, we need to
169 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
170 * taken in places where we need to invalidate the page cache in a race
171 * free manner (e.g. truncate, hole punch and other extent manipulation
179 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
182 * You can't set both SHARED and EXCL for the same lock,
183 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
184 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
186 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
187 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
188 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
189 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
190 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
191 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
192 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
194 if (lock_flags
& XFS_IOLOCK_EXCL
) {
195 down_write_nested(&VFS_I(ip
)->i_rwsem
,
196 XFS_IOLOCK_DEP(lock_flags
));
197 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
198 down_read_nested(&VFS_I(ip
)->i_rwsem
,
199 XFS_IOLOCK_DEP(lock_flags
));
202 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
203 mrupdate_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
204 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
205 mraccess_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
207 if (lock_flags
& XFS_ILOCK_EXCL
)
208 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
209 else if (lock_flags
& XFS_ILOCK_SHARED
)
210 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
214 * This is just like xfs_ilock(), except that the caller
215 * is guaranteed not to sleep. It returns 1 if it gets
216 * the requested locks and 0 otherwise. If the IO lock is
217 * obtained but the inode lock cannot be, then the IO lock
218 * is dropped before returning.
220 * ip -- the inode being locked
221 * lock_flags -- this parameter indicates the inode's locks to be
222 * to be locked. See the comment for xfs_ilock() for a list
230 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
233 * You can't set both SHARED and EXCL for the same lock,
234 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
235 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
237 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
238 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
239 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
240 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
241 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
242 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
243 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
245 if (lock_flags
& XFS_IOLOCK_EXCL
) {
246 if (!down_write_trylock(&VFS_I(ip
)->i_rwsem
))
248 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
249 if (!down_read_trylock(&VFS_I(ip
)->i_rwsem
))
253 if (lock_flags
& XFS_MMAPLOCK_EXCL
) {
254 if (!mrtryupdate(&ip
->i_mmaplock
))
255 goto out_undo_iolock
;
256 } else if (lock_flags
& XFS_MMAPLOCK_SHARED
) {
257 if (!mrtryaccess(&ip
->i_mmaplock
))
258 goto out_undo_iolock
;
261 if (lock_flags
& XFS_ILOCK_EXCL
) {
262 if (!mrtryupdate(&ip
->i_lock
))
263 goto out_undo_mmaplock
;
264 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
265 if (!mrtryaccess(&ip
->i_lock
))
266 goto out_undo_mmaplock
;
271 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
272 mrunlock_excl(&ip
->i_mmaplock
);
273 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
274 mrunlock_shared(&ip
->i_mmaplock
);
276 if (lock_flags
& XFS_IOLOCK_EXCL
)
277 up_write(&VFS_I(ip
)->i_rwsem
);
278 else if (lock_flags
& XFS_IOLOCK_SHARED
)
279 up_read(&VFS_I(ip
)->i_rwsem
);
285 * xfs_iunlock() is used to drop the inode locks acquired with
286 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
287 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
288 * that we know which locks to drop.
290 * ip -- the inode being unlocked
291 * lock_flags -- this parameter indicates the inode's locks to be
292 * to be unlocked. See the comment for xfs_ilock() for a list
293 * of valid values for this parameter.
302 * You can't set both SHARED and EXCL for the same lock,
303 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
304 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
306 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
307 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
308 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
309 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
310 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
311 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
312 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_SUBCLASS_MASK
)) == 0);
313 ASSERT(lock_flags
!= 0);
315 if (lock_flags
& XFS_IOLOCK_EXCL
)
316 up_write(&VFS_I(ip
)->i_rwsem
);
317 else if (lock_flags
& XFS_IOLOCK_SHARED
)
318 up_read(&VFS_I(ip
)->i_rwsem
);
320 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
321 mrunlock_excl(&ip
->i_mmaplock
);
322 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
323 mrunlock_shared(&ip
->i_mmaplock
);
325 if (lock_flags
& XFS_ILOCK_EXCL
)
326 mrunlock_excl(&ip
->i_lock
);
327 else if (lock_flags
& XFS_ILOCK_SHARED
)
328 mrunlock_shared(&ip
->i_lock
);
330 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
334 * give up write locks. the i/o lock cannot be held nested
335 * if it is being demoted.
342 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
));
344 ~(XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
346 if (lock_flags
& XFS_ILOCK_EXCL
)
347 mrdemote(&ip
->i_lock
);
348 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
349 mrdemote(&ip
->i_mmaplock
);
350 if (lock_flags
& XFS_IOLOCK_EXCL
)
351 downgrade_write(&VFS_I(ip
)->i_rwsem
);
353 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
356 #if defined(DEBUG) || defined(XFS_WARN)
362 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
363 if (!(lock_flags
& XFS_ILOCK_SHARED
))
364 return !!ip
->i_lock
.mr_writer
;
365 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
368 if (lock_flags
& (XFS_MMAPLOCK_EXCL
|XFS_MMAPLOCK_SHARED
)) {
369 if (!(lock_flags
& XFS_MMAPLOCK_SHARED
))
370 return !!ip
->i_mmaplock
.mr_writer
;
371 return rwsem_is_locked(&ip
->i_mmaplock
.mr_lock
);
374 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
375 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
376 return !debug_locks
||
377 lockdep_is_held_type(&VFS_I(ip
)->i_rwsem
, 0);
378 return rwsem_is_locked(&VFS_I(ip
)->i_rwsem
);
388 int xfs_small_retries
;
389 int xfs_middle_retries
;
390 int xfs_lots_retries
;
395 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
396 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
397 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
398 * errors and warnings.
400 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
402 xfs_lockdep_subclass_ok(
405 return subclass
< MAX_LOCKDEP_SUBCLASSES
;
408 #define xfs_lockdep_subclass_ok(subclass) (true)
412 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
413 * value. This can be called for any type of inode lock combination, including
414 * parent locking. Care must be taken to ensure we don't overrun the subclass
415 * storage fields in the class mask we build.
418 xfs_lock_inumorder(int lock_mode
, int subclass
)
422 ASSERT(!(lock_mode
& (XFS_ILOCK_PARENT
| XFS_ILOCK_RTBITMAP
|
424 ASSERT(xfs_lockdep_subclass_ok(subclass
));
426 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
427 ASSERT(subclass
<= XFS_IOLOCK_MAX_SUBCLASS
);
428 class += subclass
<< XFS_IOLOCK_SHIFT
;
431 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) {
432 ASSERT(subclass
<= XFS_MMAPLOCK_MAX_SUBCLASS
);
433 class += subclass
<< XFS_MMAPLOCK_SHIFT
;
436 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) {
437 ASSERT(subclass
<= XFS_ILOCK_MAX_SUBCLASS
);
438 class += subclass
<< XFS_ILOCK_SHIFT
;
441 return (lock_mode
& ~XFS_LOCK_SUBCLASS_MASK
) | class;
445 * The following routine will lock n inodes in exclusive mode. We assume the
446 * caller calls us with the inodes in i_ino order.
448 * We need to detect deadlock where an inode that we lock is in the AIL and we
449 * start waiting for another inode that is locked by a thread in a long running
450 * transaction (such as truncate). This can result in deadlock since the long
451 * running trans might need to wait for the inode we just locked in order to
452 * push the tail and free space in the log.
454 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
455 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
456 * lock more than one at a time, lockdep will report false positives saying we
457 * have violated locking orders.
465 int attempts
= 0, i
, j
, try_lock
;
469 * Currently supports between 2 and 5 inodes with exclusive locking. We
470 * support an arbitrary depth of locking here, but absolute limits on
471 * inodes depend on the the type of locking and the limits placed by
472 * lockdep annotations in xfs_lock_inumorder. These are all checked by
475 ASSERT(ips
&& inodes
>= 2 && inodes
<= 5);
476 ASSERT(lock_mode
& (XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
|
478 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
| XFS_MMAPLOCK_SHARED
|
480 ASSERT(!(lock_mode
& XFS_MMAPLOCK_EXCL
) ||
481 inodes
<= XFS_MMAPLOCK_MAX_SUBCLASS
+ 1);
482 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
) ||
483 inodes
<= XFS_ILOCK_MAX_SUBCLASS
+ 1);
485 if (lock_mode
& XFS_IOLOCK_EXCL
) {
486 ASSERT(!(lock_mode
& (XFS_MMAPLOCK_EXCL
| XFS_ILOCK_EXCL
)));
487 } else if (lock_mode
& XFS_MMAPLOCK_EXCL
)
488 ASSERT(!(lock_mode
& XFS_ILOCK_EXCL
));
493 for (; i
< inodes
; i
++) {
496 if (i
&& (ips
[i
] == ips
[i
- 1])) /* Already locked */
500 * If try_lock is not set yet, make sure all locked inodes are
501 * not in the AIL. If any are, set try_lock to be used later.
504 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
505 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
506 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
))
512 * If any of the previous locks we have locked is in the AIL,
513 * we must TRY to get the second and subsequent locks. If
514 * we can't get any, we must release all we have
518 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
522 /* try_lock means we have an inode locked that is in the AIL. */
524 if (xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
)))
528 * Unlock all previous guys and try again. xfs_iunlock will try
529 * to push the tail if the inode is in the AIL.
532 for (j
= i
- 1; j
>= 0; j
--) {
534 * Check to see if we've already unlocked this one. Not
535 * the first one going back, and the inode ptr is the
538 if (j
!= (i
- 1) && ips
[j
] == ips
[j
+ 1])
541 xfs_iunlock(ips
[j
], lock_mode
);
544 if ((attempts
% 5) == 0) {
545 delay(1); /* Don't just spin the CPU */
557 if (attempts
< 5) xfs_small_retries
++;
558 else if (attempts
< 100) xfs_middle_retries
++;
559 else xfs_lots_retries
++;
567 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
568 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
569 * lock more than one at a time, lockdep will report false positives saying we
570 * have violated locking orders.
582 ASSERT(!(lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)));
583 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
))
584 ASSERT(!(lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
586 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
588 if (ip0
->i_ino
> ip1
->i_ino
) {
595 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
598 * If the first lock we have locked is in the AIL, we must TRY to get
599 * the second lock. If we can't get it, we must release the first one
602 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
603 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
604 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
605 xfs_iunlock(ip0
, lock_mode
);
606 if ((++attempts
% 5) == 0)
607 delay(1); /* Don't just spin the CPU */
611 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
618 struct xfs_inode
*ip
)
620 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
621 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
624 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
625 if (xfs_isiflocked(ip
))
627 } while (!xfs_iflock_nowait(ip
));
629 finish_wait(wq
, &wait
.wait
);
640 if (di_flags
& XFS_DIFLAG_ANY
) {
641 if (di_flags
& XFS_DIFLAG_REALTIME
)
642 flags
|= FS_XFLAG_REALTIME
;
643 if (di_flags
& XFS_DIFLAG_PREALLOC
)
644 flags
|= FS_XFLAG_PREALLOC
;
645 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
646 flags
|= FS_XFLAG_IMMUTABLE
;
647 if (di_flags
& XFS_DIFLAG_APPEND
)
648 flags
|= FS_XFLAG_APPEND
;
649 if (di_flags
& XFS_DIFLAG_SYNC
)
650 flags
|= FS_XFLAG_SYNC
;
651 if (di_flags
& XFS_DIFLAG_NOATIME
)
652 flags
|= FS_XFLAG_NOATIME
;
653 if (di_flags
& XFS_DIFLAG_NODUMP
)
654 flags
|= FS_XFLAG_NODUMP
;
655 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
656 flags
|= FS_XFLAG_RTINHERIT
;
657 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
658 flags
|= FS_XFLAG_PROJINHERIT
;
659 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
660 flags
|= FS_XFLAG_NOSYMLINKS
;
661 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
662 flags
|= FS_XFLAG_EXTSIZE
;
663 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
664 flags
|= FS_XFLAG_EXTSZINHERIT
;
665 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
666 flags
|= FS_XFLAG_NODEFRAG
;
667 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
668 flags
|= FS_XFLAG_FILESTREAM
;
671 if (di_flags2
& XFS_DIFLAG2_ANY
) {
672 if (di_flags2
& XFS_DIFLAG2_DAX
)
673 flags
|= FS_XFLAG_DAX
;
674 if (di_flags2
& XFS_DIFLAG2_COWEXTSIZE
)
675 flags
|= FS_XFLAG_COWEXTSIZE
;
679 flags
|= FS_XFLAG_HASATTR
;
686 struct xfs_inode
*ip
)
688 struct xfs_icdinode
*dic
= &ip
->i_d
;
690 return _xfs_dic2xflags(dic
->di_flags
, dic
->di_flags2
, XFS_IFORK_Q(ip
));
694 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
695 * is allowed, otherwise it has to be an exact match. If a CI match is found,
696 * ci_name->name will point to a the actual name (caller must free) or
697 * will be set to NULL if an exact match is found.
702 struct xfs_name
*name
,
704 struct xfs_name
*ci_name
)
709 trace_xfs_lookup(dp
, name
);
711 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
714 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
718 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
726 kmem_free(ci_name
->name
);
733 * Allocate an inode on disk and return a copy of its in-core version.
734 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
735 * appropriately within the inode. The uid and gid for the inode are
736 * set according to the contents of the given cred structure.
738 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
739 * has a free inode available, call xfs_iget() to obtain the in-core
740 * version of the allocated inode. Finally, fill in the inode and
741 * log its initial contents. In this case, ialloc_context would be
744 * If xfs_dialloc() does not have an available inode, it will replenish
745 * its supply by doing an allocation. Since we can only do one
746 * allocation within a transaction without deadlocks, we must commit
747 * the current transaction before returning the inode itself.
748 * In this case, therefore, we will set ialloc_context and return.
749 * The caller should then commit the current transaction, start a new
750 * transaction, and call xfs_ialloc() again to actually get the inode.
752 * To ensure that some other process does not grab the inode that
753 * was allocated during the first call to xfs_ialloc(), this routine
754 * also returns the [locked] bp pointing to the head of the freelist
755 * as ialloc_context. The caller should hold this buffer across
756 * the commit and pass it back into this routine on the second call.
758 * If we are allocating quota inodes, we do not have a parent inode
759 * to attach to or associate with (i.e. pip == NULL) because they
760 * are not linked into the directory structure - they are attached
761 * directly to the superblock - and so have no parent.
772 xfs_buf_t
**ialloc_context
,
775 struct xfs_mount
*mp
= tp
->t_mountp
;
784 * Call the space management code to pick
785 * the on-disk inode to be allocated.
787 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
788 ialloc_context
, &ino
);
791 if (*ialloc_context
|| ino
== NULLFSINO
) {
795 ASSERT(*ialloc_context
== NULL
);
798 * Get the in-core inode with the lock held exclusively.
799 * This is because we're setting fields here we need
800 * to prevent others from looking at until we're done.
802 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
803 XFS_ILOCK_EXCL
, &ip
);
810 * We always convert v1 inodes to v2 now - we only support filesystems
811 * with >= v2 inode capability, so there is no reason for ever leaving
812 * an inode in v1 format.
814 if (ip
->i_d
.di_version
== 1)
815 ip
->i_d
.di_version
= 2;
817 inode
->i_mode
= mode
;
818 set_nlink(inode
, nlink
);
819 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
820 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
821 xfs_set_projid(ip
, prid
);
823 if (pip
&& XFS_INHERIT_GID(pip
)) {
824 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
825 if ((VFS_I(pip
)->i_mode
& S_ISGID
) && S_ISDIR(mode
))
826 inode
->i_mode
|= S_ISGID
;
830 * If the group ID of the new file does not match the effective group
831 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
832 * (and only if the irix_sgid_inherit compatibility variable is set).
834 if ((irix_sgid_inherit
) &&
835 (inode
->i_mode
& S_ISGID
) &&
836 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
))))
837 inode
->i_mode
&= ~S_ISGID
;
840 ip
->i_d
.di_nextents
= 0;
841 ASSERT(ip
->i_d
.di_nblocks
== 0);
843 tv
= current_time(inode
);
848 ip
->i_d
.di_extsize
= 0;
849 ip
->i_d
.di_dmevmask
= 0;
850 ip
->i_d
.di_dmstate
= 0;
851 ip
->i_d
.di_flags
= 0;
853 if (ip
->i_d
.di_version
== 3) {
854 inode
->i_version
= 1;
855 ip
->i_d
.di_flags2
= 0;
856 ip
->i_d
.di_cowextsize
= 0;
857 ip
->i_d
.di_crtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
858 ip
->i_d
.di_crtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
862 flags
= XFS_ILOG_CORE
;
863 switch (mode
& S_IFMT
) {
868 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
869 ip
->i_df
.if_u2
.if_rdev
= rdev
;
870 ip
->i_df
.if_flags
= 0;
871 flags
|= XFS_ILOG_DEV
;
875 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
876 uint64_t di_flags2
= 0;
880 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
881 di_flags
|= XFS_DIFLAG_RTINHERIT
;
882 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
883 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
884 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
886 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
887 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
888 } else if (S_ISREG(mode
)) {
889 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
890 di_flags
|= XFS_DIFLAG_REALTIME
;
891 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
892 di_flags
|= XFS_DIFLAG_EXTSIZE
;
893 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
896 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
898 di_flags
|= XFS_DIFLAG_NOATIME
;
899 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
901 di_flags
|= XFS_DIFLAG_NODUMP
;
902 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
904 di_flags
|= XFS_DIFLAG_SYNC
;
905 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
906 xfs_inherit_nosymlinks
)
907 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
908 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
909 xfs_inherit_nodefrag
)
910 di_flags
|= XFS_DIFLAG_NODEFRAG
;
911 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
912 di_flags
|= XFS_DIFLAG_FILESTREAM
;
913 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
)
914 di_flags2
|= XFS_DIFLAG2_DAX
;
916 ip
->i_d
.di_flags
|= di_flags
;
917 ip
->i_d
.di_flags2
|= di_flags2
;
920 (pip
->i_d
.di_flags2
& XFS_DIFLAG2_ANY
) &&
921 pip
->i_d
.di_version
== 3 &&
922 ip
->i_d
.di_version
== 3) {
923 if (pip
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) {
924 ip
->i_d
.di_flags2
|= XFS_DIFLAG2_COWEXTSIZE
;
925 ip
->i_d
.di_cowextsize
= pip
->i_d
.di_cowextsize
;
930 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
931 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
932 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
933 ip
->i_df
.if_u1
.if_extents
= NULL
;
939 * Attribute fork settings for new inode.
941 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
942 ip
->i_d
.di_anextents
= 0;
945 * Log the new values stuffed into the inode.
947 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
948 xfs_trans_log_inode(tp
, ip
, flags
);
950 /* now that we have an i_mode we can setup the inode structure */
958 * Allocates a new inode from disk and return a pointer to the
959 * incore copy. This routine will internally commit the current
960 * transaction and allocate a new one if the Space Manager needed
961 * to do an allocation to replenish the inode free-list.
963 * This routine is designed to be called from xfs_create and
969 xfs_trans_t
**tpp
, /* input: current transaction;
970 output: may be a new transaction. */
971 xfs_inode_t
*dp
, /* directory within whose allocate
976 prid_t prid
, /* project id */
977 int okalloc
, /* ok to allocate new space */
978 xfs_inode_t
**ipp
, /* pointer to inode; it will be
985 xfs_buf_t
*ialloc_context
= NULL
;
991 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
994 * xfs_ialloc will return a pointer to an incore inode if
995 * the Space Manager has an available inode on the free
996 * list. Otherwise, it will do an allocation and replenish
997 * the freelist. Since we can only do one allocation per
998 * transaction without deadlocks, we will need to commit the
999 * current transaction and start a new one. We will then
1000 * need to call xfs_ialloc again to get the inode.
1002 * If xfs_ialloc did an allocation to replenish the freelist,
1003 * it returns the bp containing the head of the freelist as
1004 * ialloc_context. We will hold a lock on it across the
1005 * transaction commit so that no other process can steal
1006 * the inode(s) that we've just allocated.
1008 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
1009 &ialloc_context
, &ip
);
1012 * Return an error if we were unable to allocate a new inode.
1013 * This should only happen if we run out of space on disk or
1014 * encounter a disk error.
1020 if (!ialloc_context
&& !ip
) {
1026 * If the AGI buffer is non-NULL, then we were unable to get an
1027 * inode in one operation. We need to commit the current
1028 * transaction and call xfs_ialloc() again. It is guaranteed
1029 * to succeed the second time.
1031 if (ialloc_context
) {
1033 * Normally, xfs_trans_commit releases all the locks.
1034 * We call bhold to hang on to the ialloc_context across
1035 * the commit. Holding this buffer prevents any other
1036 * processes from doing any allocations in this
1039 xfs_trans_bhold(tp
, ialloc_context
);
1042 * We want the quota changes to be associated with the next
1043 * transaction, NOT this one. So, detach the dqinfo from this
1044 * and attach it to the next transaction.
1049 dqinfo
= (void *)tp
->t_dqinfo
;
1050 tp
->t_dqinfo
= NULL
;
1051 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
1052 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
1055 code
= xfs_trans_roll(&tp
, NULL
);
1056 if (committed
!= NULL
)
1060 * Re-attach the quota info that we detached from prev trx.
1063 tp
->t_dqinfo
= dqinfo
;
1064 tp
->t_flags
|= tflags
;
1068 xfs_buf_relse(ialloc_context
);
1073 xfs_trans_bjoin(tp
, ialloc_context
);
1076 * Call ialloc again. Since we've locked out all
1077 * other allocations in this allocation group,
1078 * this call should always succeed.
1080 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1081 okalloc
, &ialloc_context
, &ip
);
1084 * If we get an error at this point, return to the caller
1085 * so that the current transaction can be aborted.
1092 ASSERT(!ialloc_context
&& ip
);
1095 if (committed
!= NULL
)
1106 * Decrement the link count on an inode & log the change. If this causes the
1107 * link count to go to zero, move the inode to AGI unlinked list so that it can
1108 * be freed when the last active reference goes away via xfs_inactive().
1110 static int /* error */
1115 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1117 drop_nlink(VFS_I(ip
));
1118 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1120 if (VFS_I(ip
)->i_nlink
)
1123 return xfs_iunlink(tp
, ip
);
1127 * Increment the link count on an inode & log the change.
1134 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1136 ASSERT(ip
->i_d
.di_version
> 1);
1137 inc_nlink(VFS_I(ip
));
1138 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1145 struct xfs_name
*name
,
1150 int is_dir
= S_ISDIR(mode
);
1151 struct xfs_mount
*mp
= dp
->i_mount
;
1152 struct xfs_inode
*ip
= NULL
;
1153 struct xfs_trans
*tp
= NULL
;
1155 struct xfs_defer_ops dfops
;
1156 xfs_fsblock_t first_block
;
1157 bool unlock_dp_on_error
= false;
1159 struct xfs_dquot
*udqp
= NULL
;
1160 struct xfs_dquot
*gdqp
= NULL
;
1161 struct xfs_dquot
*pdqp
= NULL
;
1162 struct xfs_trans_res
*tres
;
1165 trace_xfs_create(dp
, name
);
1167 if (XFS_FORCED_SHUTDOWN(mp
))
1170 prid
= xfs_get_initial_prid(dp
);
1173 * Make sure that we have allocated dquot(s) on disk.
1175 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1176 xfs_kgid_to_gid(current_fsgid()), prid
,
1177 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1178 &udqp
, &gdqp
, &pdqp
);
1184 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1185 tres
= &M_RES(mp
)->tr_mkdir
;
1187 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1188 tres
= &M_RES(mp
)->tr_create
;
1192 * Initially assume that the file does not exist and
1193 * reserve the resources for that case. If that is not
1194 * the case we'll drop the one we have and get a more
1195 * appropriate transaction later.
1197 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1198 if (error
== -ENOSPC
) {
1199 /* flush outstanding delalloc blocks and retry */
1200 xfs_flush_inodes(mp
);
1201 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1203 if (error
== -ENOSPC
) {
1204 /* No space at all so try a "no-allocation" reservation */
1206 error
= xfs_trans_alloc(mp
, tres
, 0, 0, 0, &tp
);
1209 goto out_release_inode
;
1211 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1212 unlock_dp_on_error
= true;
1214 xfs_defer_init(&dfops
, &first_block
);
1217 * Reserve disk quota and the inode.
1219 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1220 pdqp
, resblks
, 1, 0);
1222 goto out_trans_cancel
;
1225 error
= xfs_dir_canenter(tp
, dp
, name
);
1227 goto out_trans_cancel
;
1231 * A newly created regular or special file just has one directory
1232 * entry pointing to them, but a directory also the "." entry
1233 * pointing to itself.
1235 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1236 prid
, resblks
> 0, &ip
, NULL
);
1238 goto out_trans_cancel
;
1241 * Now we join the directory inode to the transaction. We do not do it
1242 * earlier because xfs_dir_ialloc might commit the previous transaction
1243 * (and release all the locks). An error from here on will result in
1244 * the transaction cancel unlocking dp so don't do it explicitly in the
1247 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1248 unlock_dp_on_error
= false;
1250 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1251 &first_block
, &dfops
, resblks
?
1252 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1254 ASSERT(error
!= -ENOSPC
);
1255 goto out_trans_cancel
;
1257 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1258 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1261 error
= xfs_dir_init(tp
, ip
, dp
);
1263 goto out_bmap_cancel
;
1265 error
= xfs_bumplink(tp
, dp
);
1267 goto out_bmap_cancel
;
1271 * If this is a synchronous mount, make sure that the
1272 * create transaction goes to disk before returning to
1275 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1276 xfs_trans_set_sync(tp
);
1279 * Attach the dquot(s) to the inodes and modify them incore.
1280 * These ids of the inode couldn't have changed since the new
1281 * inode has been locked ever since it was created.
1283 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1285 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1287 goto out_bmap_cancel
;
1289 error
= xfs_trans_commit(tp
);
1291 goto out_release_inode
;
1293 xfs_qm_dqrele(udqp
);
1294 xfs_qm_dqrele(gdqp
);
1295 xfs_qm_dqrele(pdqp
);
1301 xfs_defer_cancel(&dfops
);
1303 xfs_trans_cancel(tp
);
1306 * Wait until after the current transaction is aborted to finish the
1307 * setup of the inode and release the inode. This prevents recursive
1308 * transactions and deadlocks from xfs_inactive.
1311 xfs_finish_inode_setup(ip
);
1315 xfs_qm_dqrele(udqp
);
1316 xfs_qm_dqrele(gdqp
);
1317 xfs_qm_dqrele(pdqp
);
1319 if (unlock_dp_on_error
)
1320 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1326 struct xfs_inode
*dp
,
1327 struct dentry
*dentry
,
1329 struct xfs_inode
**ipp
)
1331 struct xfs_mount
*mp
= dp
->i_mount
;
1332 struct xfs_inode
*ip
= NULL
;
1333 struct xfs_trans
*tp
= NULL
;
1336 struct xfs_dquot
*udqp
= NULL
;
1337 struct xfs_dquot
*gdqp
= NULL
;
1338 struct xfs_dquot
*pdqp
= NULL
;
1339 struct xfs_trans_res
*tres
;
1342 if (XFS_FORCED_SHUTDOWN(mp
))
1345 prid
= xfs_get_initial_prid(dp
);
1348 * Make sure that we have allocated dquot(s) on disk.
1350 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1351 xfs_kgid_to_gid(current_fsgid()), prid
,
1352 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1353 &udqp
, &gdqp
, &pdqp
);
1357 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1358 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1360 error
= xfs_trans_alloc(mp
, tres
, resblks
, 0, 0, &tp
);
1361 if (error
== -ENOSPC
) {
1362 /* No space at all so try a "no-allocation" reservation */
1364 error
= xfs_trans_alloc(mp
, tres
, 0, 0, 0, &tp
);
1367 goto out_release_inode
;
1369 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1370 pdqp
, resblks
, 1, 0);
1372 goto out_trans_cancel
;
1374 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1, 0,
1375 prid
, resblks
> 0, &ip
, NULL
);
1377 goto out_trans_cancel
;
1379 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1380 xfs_trans_set_sync(tp
);
1383 * Attach the dquot(s) to the inodes and modify them incore.
1384 * These ids of the inode couldn't have changed since the new
1385 * inode has been locked ever since it was created.
1387 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1389 error
= xfs_iunlink(tp
, ip
);
1391 goto out_trans_cancel
;
1393 error
= xfs_trans_commit(tp
);
1395 goto out_release_inode
;
1397 xfs_qm_dqrele(udqp
);
1398 xfs_qm_dqrele(gdqp
);
1399 xfs_qm_dqrele(pdqp
);
1405 xfs_trans_cancel(tp
);
1408 * Wait until after the current transaction is aborted to finish the
1409 * setup of the inode and release the inode. This prevents recursive
1410 * transactions and deadlocks from xfs_inactive.
1413 xfs_finish_inode_setup(ip
);
1417 xfs_qm_dqrele(udqp
);
1418 xfs_qm_dqrele(gdqp
);
1419 xfs_qm_dqrele(pdqp
);
1428 struct xfs_name
*target_name
)
1430 xfs_mount_t
*mp
= tdp
->i_mount
;
1433 struct xfs_defer_ops dfops
;
1434 xfs_fsblock_t first_block
;
1437 trace_xfs_link(tdp
, target_name
);
1439 ASSERT(!S_ISDIR(VFS_I(sip
)->i_mode
));
1441 if (XFS_FORCED_SHUTDOWN(mp
))
1444 error
= xfs_qm_dqattach(sip
, 0);
1448 error
= xfs_qm_dqattach(tdp
, 0);
1452 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1453 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, resblks
, 0, 0, &tp
);
1454 if (error
== -ENOSPC
) {
1456 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_link
, 0, 0, 0, &tp
);
1461 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1463 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1464 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1467 * If we are using project inheritance, we only allow hard link
1468 * creation in our tree when the project IDs are the same; else
1469 * the tree quota mechanism could be circumvented.
1471 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1472 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1478 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1483 xfs_defer_init(&dfops
, &first_block
);
1486 * Handle initial link state of O_TMPFILE inode
1488 if (VFS_I(sip
)->i_nlink
== 0) {
1489 error
= xfs_iunlink_remove(tp
, sip
);
1494 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1495 &first_block
, &dfops
, resblks
);
1498 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1499 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1501 error
= xfs_bumplink(tp
, sip
);
1506 * If this is a synchronous mount, make sure that the
1507 * link transaction goes to disk before returning to
1510 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1511 xfs_trans_set_sync(tp
);
1513 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1515 xfs_defer_cancel(&dfops
);
1519 return xfs_trans_commit(tp
);
1522 xfs_trans_cancel(tp
);
1528 * Free up the underlying blocks past new_size. The new size must be smaller
1529 * than the current size. This routine can be used both for the attribute and
1530 * data fork, and does not modify the inode size, which is left to the caller.
1532 * The transaction passed to this routine must have made a permanent log
1533 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1534 * given transaction and start new ones, so make sure everything involved in
1535 * the transaction is tidy before calling here. Some transaction will be
1536 * returned to the caller to be committed. The incoming transaction must
1537 * already include the inode, and both inode locks must be held exclusively.
1538 * The inode must also be "held" within the transaction. On return the inode
1539 * will be "held" within the returned transaction. This routine does NOT
1540 * require any disk space to be reserved for it within the transaction.
1542 * If we get an error, we must return with the inode locked and linked into the
1543 * current transaction. This keeps things simple for the higher level code,
1544 * because it always knows that the inode is locked and held in the transaction
1545 * that returns to it whether errors occur or not. We don't mark the inode
1546 * dirty on error so that transactions can be easily aborted if possible.
1549 xfs_itruncate_extents(
1550 struct xfs_trans
**tpp
,
1551 struct xfs_inode
*ip
,
1553 xfs_fsize_t new_size
)
1555 struct xfs_mount
*mp
= ip
->i_mount
;
1556 struct xfs_trans
*tp
= *tpp
;
1557 struct xfs_defer_ops dfops
;
1558 xfs_fsblock_t first_block
;
1559 xfs_fileoff_t first_unmap_block
;
1560 xfs_fileoff_t last_block
;
1561 xfs_filblks_t unmap_len
;
1565 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1566 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1567 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1568 ASSERT(new_size
<= XFS_ISIZE(ip
));
1569 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1570 ASSERT(ip
->i_itemp
!= NULL
);
1571 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1572 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1574 trace_xfs_itruncate_extents_start(ip
, new_size
);
1577 * Since it is possible for space to become allocated beyond
1578 * the end of the file (in a crash where the space is allocated
1579 * but the inode size is not yet updated), simply remove any
1580 * blocks which show up between the new EOF and the maximum
1581 * possible file size. If the first block to be removed is
1582 * beyond the maximum file size (ie it is the same as last_block),
1583 * then there is nothing to do.
1585 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1586 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1587 if (first_unmap_block
== last_block
)
1590 ASSERT(first_unmap_block
< last_block
);
1591 unmap_len
= last_block
- first_unmap_block
+ 1;
1593 xfs_defer_init(&dfops
, &first_block
);
1594 error
= xfs_bunmapi(tp
, ip
,
1595 first_unmap_block
, unmap_len
,
1596 xfs_bmapi_aflag(whichfork
),
1597 XFS_ITRUNC_MAX_EXTENTS
,
1598 &first_block
, &dfops
,
1601 goto out_bmap_cancel
;
1604 * Duplicate the transaction that has the permanent
1605 * reservation and commit the old transaction.
1607 error
= xfs_defer_finish(&tp
, &dfops
, ip
);
1609 goto out_bmap_cancel
;
1611 error
= xfs_trans_roll(&tp
, ip
);
1616 /* Remove all pending CoW reservations. */
1617 error
= xfs_reflink_cancel_cow_blocks(ip
, &tp
, first_unmap_block
,
1623 * Clear the reflink flag if we truncated everything.
1625 if (ip
->i_d
.di_nblocks
== 0 && xfs_is_reflink_inode(ip
)) {
1626 ip
->i_d
.di_flags2
&= ~XFS_DIFLAG2_REFLINK
;
1627 xfs_inode_clear_cowblocks_tag(ip
);
1631 * Always re-log the inode so that our permanent transaction can keep
1632 * on rolling it forward in the log.
1634 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1636 trace_xfs_itruncate_extents_end(ip
, new_size
);
1643 * If the bunmapi call encounters an error, return to the caller where
1644 * the transaction can be properly aborted. We just need to make sure
1645 * we're not holding any resources that we were not when we came in.
1647 xfs_defer_cancel(&dfops
);
1655 xfs_mount_t
*mp
= ip
->i_mount
;
1658 if (!S_ISREG(VFS_I(ip
)->i_mode
) || (VFS_I(ip
)->i_mode
== 0))
1661 /* If this is a read-only mount, don't do this (would generate I/O) */
1662 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1665 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1669 * If we previously truncated this file and removed old data
1670 * in the process, we want to initiate "early" writeout on
1671 * the last close. This is an attempt to combat the notorious
1672 * NULL files problem which is particularly noticeable from a
1673 * truncate down, buffered (re-)write (delalloc), followed by
1674 * a crash. What we are effectively doing here is
1675 * significantly reducing the time window where we'd otherwise
1676 * be exposed to that problem.
1678 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1680 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1681 if (ip
->i_delayed_blks
> 0) {
1682 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1689 if (VFS_I(ip
)->i_nlink
== 0)
1692 if (xfs_can_free_eofblocks(ip
, false)) {
1695 * Check if the inode is being opened, written and closed
1696 * frequently and we have delayed allocation blocks outstanding
1697 * (e.g. streaming writes from the NFS server), truncating the
1698 * blocks past EOF will cause fragmentation to occur.
1700 * In this case don't do the truncation, but we have to be
1701 * careful how we detect this case. Blocks beyond EOF show up as
1702 * i_delayed_blks even when the inode is clean, so we need to
1703 * truncate them away first before checking for a dirty release.
1704 * Hence on the first dirty close we will still remove the
1705 * speculative allocation, but after that we will leave it in
1708 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1711 * If we can't get the iolock just skip truncating the blocks
1712 * past EOF because we could deadlock with the mmap_sem
1713 * otherwise. We'll get another chance to drop them once the
1714 * last reference to the inode is dropped, so we'll never leak
1715 * blocks permanently.
1717 if (xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1718 error
= xfs_free_eofblocks(ip
);
1719 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1724 /* delalloc blocks after truncation means it really is dirty */
1725 if (ip
->i_delayed_blks
)
1726 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1732 * xfs_inactive_truncate
1734 * Called to perform a truncate when an inode becomes unlinked.
1737 xfs_inactive_truncate(
1738 struct xfs_inode
*ip
)
1740 struct xfs_mount
*mp
= ip
->i_mount
;
1741 struct xfs_trans
*tp
;
1744 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
1746 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1750 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1751 xfs_trans_ijoin(tp
, ip
, 0);
1754 * Log the inode size first to prevent stale data exposure in the event
1755 * of a system crash before the truncate completes. See the related
1756 * comment in xfs_vn_setattr_size() for details.
1758 ip
->i_d
.di_size
= 0;
1759 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1761 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1763 goto error_trans_cancel
;
1765 ASSERT(ip
->i_d
.di_nextents
== 0);
1767 error
= xfs_trans_commit(tp
);
1771 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1775 xfs_trans_cancel(tp
);
1777 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1782 * xfs_inactive_ifree()
1784 * Perform the inode free when an inode is unlinked.
1788 struct xfs_inode
*ip
)
1790 struct xfs_defer_ops dfops
;
1791 xfs_fsblock_t first_block
;
1792 struct xfs_mount
*mp
= ip
->i_mount
;
1793 struct xfs_trans
*tp
;
1797 * We try to use a per-AG reservation for any block needed by the finobt
1798 * tree, but as the finobt feature predates the per-AG reservation
1799 * support a degraded file system might not have enough space for the
1800 * reservation at mount time. In that case try to dip into the reserved
1803 * Send a warning if the reservation does happen to fail, as the inode
1804 * now remains allocated and sits on the unlinked list until the fs is
1807 if (unlikely(mp
->m_inotbt_nores
)) {
1808 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
,
1809 XFS_IFREE_SPACE_RES(mp
), 0, XFS_TRANS_RESERVE
,
1812 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ifree
, 0, 0, 0, &tp
);
1815 if (error
== -ENOSPC
) {
1816 xfs_warn_ratelimited(mp
,
1817 "Failed to remove inode(s) from unlinked list. "
1818 "Please free space, unmount and run xfs_repair.");
1820 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1825 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1826 xfs_trans_ijoin(tp
, ip
, 0);
1828 xfs_defer_init(&dfops
, &first_block
);
1829 error
= xfs_ifree(tp
, ip
, &dfops
);
1832 * If we fail to free the inode, shut down. The cancel
1833 * might do that, we need to make sure. Otherwise the
1834 * inode might be lost for a long time or forever.
1836 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1837 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1839 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1841 xfs_trans_cancel(tp
);
1842 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1847 * Credit the quota account(s). The inode is gone.
1849 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1852 * Just ignore errors at this point. There is nothing we can do except
1853 * to try to keep going. Make sure it's not a silent error.
1855 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1857 xfs_notice(mp
, "%s: xfs_defer_finish returned error %d",
1859 xfs_defer_cancel(&dfops
);
1861 error
= xfs_trans_commit(tp
);
1863 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1866 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1873 * This is called when the vnode reference count for the vnode
1874 * goes to zero. If the file has been unlinked, then it must
1875 * now be truncated. Also, we clear all of the read-ahead state
1876 * kept for the inode here since the file is now closed.
1882 struct xfs_mount
*mp
;
1887 * If the inode is already free, then there can be nothing
1890 if (VFS_I(ip
)->i_mode
== 0) {
1891 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1892 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1897 ASSERT(!xfs_iflags_test(ip
, XFS_IRECOVERY
));
1899 /* If this is a read-only mount, don't do this (would generate I/O) */
1900 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1903 if (VFS_I(ip
)->i_nlink
!= 0) {
1905 * force is true because we are evicting an inode from the
1906 * cache. Post-eof blocks must be freed, lest we end up with
1907 * broken free space accounting.
1909 if (xfs_can_free_eofblocks(ip
, true)) {
1910 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1911 xfs_free_eofblocks(ip
);
1912 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1918 if (S_ISREG(VFS_I(ip
)->i_mode
) &&
1919 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1920 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1923 error
= xfs_qm_dqattach(ip
, 0);
1927 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1928 error
= xfs_inactive_symlink(ip
);
1930 error
= xfs_inactive_truncate(ip
);
1935 * If there are attributes associated with the file then blow them away
1936 * now. The code calls a routine that recursively deconstructs the
1937 * attribute fork. If also blows away the in-core attribute fork.
1939 if (XFS_IFORK_Q(ip
)) {
1940 error
= xfs_attr_inactive(ip
);
1946 ASSERT(ip
->i_d
.di_anextents
== 0);
1947 ASSERT(ip
->i_d
.di_forkoff
== 0);
1952 error
= xfs_inactive_ifree(ip
);
1957 * Release the dquots held by inode, if any.
1959 xfs_qm_dqdetach(ip
);
1963 * This is called when the inode's link count goes to 0 or we are creating a
1964 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1965 * set to true as the link count is dropped to zero by the VFS after we've
1966 * created the file successfully, so we have to add it to the unlinked list
1967 * while the link count is non-zero.
1969 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1970 * list when the inode is freed.
1974 struct xfs_trans
*tp
,
1975 struct xfs_inode
*ip
)
1977 xfs_mount_t
*mp
= tp
->t_mountp
;
1987 ASSERT(VFS_I(ip
)->i_mode
!= 0);
1990 * Get the agi buffer first. It ensures lock ordering
1993 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1996 agi
= XFS_BUF_TO_AGI(agibp
);
1999 * Get the index into the agi hash table for the
2000 * list this inode will go on.
2002 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2004 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2005 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2006 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
2008 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
2010 * There is already another inode in the bucket we need
2011 * to add ourselves to. Add us at the front of the list.
2012 * Here we put the head pointer into our next pointer,
2013 * and then we fall through to point the head at us.
2015 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2020 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
2021 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
2022 offset
= ip
->i_imap
.im_boffset
+
2023 offsetof(xfs_dinode_t
, di_next_unlinked
);
2025 /* need to recalc the inode CRC if appropriate */
2026 xfs_dinode_calc_crc(mp
, dip
);
2028 xfs_trans_inode_buf(tp
, ibp
);
2029 xfs_trans_log_buf(tp
, ibp
, offset
,
2030 (offset
+ sizeof(xfs_agino_t
) - 1));
2031 xfs_inobp_check(mp
, ibp
);
2035 * Point the bucket head pointer at the inode being inserted.
2038 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
2039 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2040 (sizeof(xfs_agino_t
) * bucket_index
);
2041 xfs_trans_log_buf(tp
, agibp
, offset
,
2042 (offset
+ sizeof(xfs_agino_t
) - 1));
2047 * Pull the on-disk inode from the AGI unlinked list.
2060 xfs_agnumber_t agno
;
2062 xfs_agino_t next_agino
;
2063 xfs_buf_t
*last_ibp
;
2064 xfs_dinode_t
*last_dip
= NULL
;
2066 int offset
, last_offset
= 0;
2070 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2073 * Get the agi buffer first. It ensures lock ordering
2076 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2080 agi
= XFS_BUF_TO_AGI(agibp
);
2083 * Get the index into the agi hash table for the
2084 * list this inode will go on.
2086 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2088 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2089 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
2090 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2092 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2094 * We're at the head of the list. Get the inode's on-disk
2095 * buffer to see if there is anyone after us on the list.
2096 * Only modify our next pointer if it is not already NULLAGINO.
2097 * This saves us the overhead of dealing with the buffer when
2098 * there is no need to change it.
2100 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2103 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2107 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2108 ASSERT(next_agino
!= 0);
2109 if (next_agino
!= NULLAGINO
) {
2110 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2111 offset
= ip
->i_imap
.im_boffset
+
2112 offsetof(xfs_dinode_t
, di_next_unlinked
);
2114 /* need to recalc the inode CRC if appropriate */
2115 xfs_dinode_calc_crc(mp
, dip
);
2117 xfs_trans_inode_buf(tp
, ibp
);
2118 xfs_trans_log_buf(tp
, ibp
, offset
,
2119 (offset
+ sizeof(xfs_agino_t
) - 1));
2120 xfs_inobp_check(mp
, ibp
);
2122 xfs_trans_brelse(tp
, ibp
);
2125 * Point the bucket head pointer at the next inode.
2127 ASSERT(next_agino
!= 0);
2128 ASSERT(next_agino
!= agino
);
2129 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2130 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2131 (sizeof(xfs_agino_t
) * bucket_index
);
2132 xfs_trans_log_buf(tp
, agibp
, offset
,
2133 (offset
+ sizeof(xfs_agino_t
) - 1));
2136 * We need to search the list for the inode being freed.
2138 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2140 while (next_agino
!= agino
) {
2141 struct xfs_imap imap
;
2144 xfs_trans_brelse(tp
, last_ibp
);
2147 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2149 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2152 "%s: xfs_imap returned error %d.",
2157 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2161 "%s: xfs_imap_to_bp returned error %d.",
2166 last_offset
= imap
.im_boffset
;
2167 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2168 ASSERT(next_agino
!= NULLAGINO
);
2169 ASSERT(next_agino
!= 0);
2173 * Now last_ibp points to the buffer previous to us on the
2174 * unlinked list. Pull us from the list.
2176 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2179 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2183 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2184 ASSERT(next_agino
!= 0);
2185 ASSERT(next_agino
!= agino
);
2186 if (next_agino
!= NULLAGINO
) {
2187 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2188 offset
= ip
->i_imap
.im_boffset
+
2189 offsetof(xfs_dinode_t
, di_next_unlinked
);
2191 /* need to recalc the inode CRC if appropriate */
2192 xfs_dinode_calc_crc(mp
, dip
);
2194 xfs_trans_inode_buf(tp
, ibp
);
2195 xfs_trans_log_buf(tp
, ibp
, offset
,
2196 (offset
+ sizeof(xfs_agino_t
) - 1));
2197 xfs_inobp_check(mp
, ibp
);
2199 xfs_trans_brelse(tp
, ibp
);
2202 * Point the previous inode on the list to the next inode.
2204 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2205 ASSERT(next_agino
!= 0);
2206 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2208 /* need to recalc the inode CRC if appropriate */
2209 xfs_dinode_calc_crc(mp
, last_dip
);
2211 xfs_trans_inode_buf(tp
, last_ibp
);
2212 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2213 (offset
+ sizeof(xfs_agino_t
) - 1));
2214 xfs_inobp_check(mp
, last_ibp
);
2220 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2221 * inodes that are in memory - they all must be marked stale and attached to
2222 * the cluster buffer.
2226 xfs_inode_t
*free_ip
,
2228 struct xfs_icluster
*xic
)
2230 xfs_mount_t
*mp
= free_ip
->i_mount
;
2231 int blks_per_cluster
;
2232 int inodes_per_cluster
;
2239 xfs_inode_log_item_t
*iip
;
2240 xfs_log_item_t
*lip
;
2241 struct xfs_perag
*pag
;
2244 inum
= xic
->first_ino
;
2245 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2246 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2247 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
2248 nbufs
= mp
->m_ialloc_blks
/ blks_per_cluster
;
2250 for (j
= 0; j
< nbufs
; j
++, inum
+= inodes_per_cluster
) {
2252 * The allocation bitmap tells us which inodes of the chunk were
2253 * physically allocated. Skip the cluster if an inode falls into
2256 ioffset
= inum
- xic
->first_ino
;
2257 if ((xic
->alloc
& XFS_INOBT_MASK(ioffset
)) == 0) {
2258 ASSERT(do_mod(ioffset
, inodes_per_cluster
) == 0);
2262 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2263 XFS_INO_TO_AGBNO(mp
, inum
));
2266 * We obtain and lock the backing buffer first in the process
2267 * here, as we have to ensure that any dirty inode that we
2268 * can't get the flush lock on is attached to the buffer.
2269 * If we scan the in-memory inodes first, then buffer IO can
2270 * complete before we get a lock on it, and hence we may fail
2271 * to mark all the active inodes on the buffer stale.
2273 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2274 mp
->m_bsize
* blks_per_cluster
,
2281 * This buffer may not have been correctly initialised as we
2282 * didn't read it from disk. That's not important because we are
2283 * only using to mark the buffer as stale in the log, and to
2284 * attach stale cached inodes on it. That means it will never be
2285 * dispatched for IO. If it is, we want to know about it, and we
2286 * want it to fail. We can acheive this by adding a write
2287 * verifier to the buffer.
2289 bp
->b_ops
= &xfs_inode_buf_ops
;
2292 * Walk the inodes already attached to the buffer and mark them
2293 * stale. These will all have the flush locks held, so an
2294 * in-memory inode walk can't lock them. By marking them all
2295 * stale first, we will not attempt to lock them in the loop
2296 * below as the XFS_ISTALE flag will be set.
2300 if (lip
->li_type
== XFS_LI_INODE
) {
2301 iip
= (xfs_inode_log_item_t
*)lip
;
2302 ASSERT(iip
->ili_logged
== 1);
2303 lip
->li_cb
= xfs_istale_done
;
2304 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2305 &iip
->ili_flush_lsn
,
2306 &iip
->ili_item
.li_lsn
);
2307 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2309 lip
= lip
->li_bio_list
;
2314 * For each inode in memory attempt to add it to the inode
2315 * buffer and set it up for being staled on buffer IO
2316 * completion. This is safe as we've locked out tail pushing
2317 * and flushing by locking the buffer.
2319 * We have already marked every inode that was part of a
2320 * transaction stale above, which means there is no point in
2321 * even trying to lock them.
2323 for (i
= 0; i
< inodes_per_cluster
; i
++) {
2326 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2327 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2329 /* Inode not in memory, nothing to do */
2336 * because this is an RCU protected lookup, we could
2337 * find a recently freed or even reallocated inode
2338 * during the lookup. We need to check under the
2339 * i_flags_lock for a valid inode here. Skip it if it
2340 * is not valid, the wrong inode or stale.
2342 spin_lock(&ip
->i_flags_lock
);
2343 if (ip
->i_ino
!= inum
+ i
||
2344 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2345 spin_unlock(&ip
->i_flags_lock
);
2349 spin_unlock(&ip
->i_flags_lock
);
2352 * Don't try to lock/unlock the current inode, but we
2353 * _cannot_ skip the other inodes that we did not find
2354 * in the list attached to the buffer and are not
2355 * already marked stale. If we can't lock it, back off
2358 if (ip
!= free_ip
&&
2359 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2367 xfs_iflags_set(ip
, XFS_ISTALE
);
2370 * we don't need to attach clean inodes or those only
2371 * with unlogged changes (which we throw away, anyway).
2374 if (!iip
|| xfs_inode_clean(ip
)) {
2375 ASSERT(ip
!= free_ip
);
2377 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2381 iip
->ili_last_fields
= iip
->ili_fields
;
2382 iip
->ili_fields
= 0;
2383 iip
->ili_fsync_fields
= 0;
2384 iip
->ili_logged
= 1;
2385 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2386 &iip
->ili_item
.li_lsn
);
2388 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2392 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2395 xfs_trans_stale_inode_buf(tp
, bp
);
2396 xfs_trans_binval(tp
, bp
);
2404 * This is called to return an inode to the inode free list.
2405 * The inode should already be truncated to 0 length and have
2406 * no pages associated with it. This routine also assumes that
2407 * the inode is already a part of the transaction.
2409 * The on-disk copy of the inode will have been added to the list
2410 * of unlinked inodes in the AGI. We need to remove the inode from
2411 * that list atomically with respect to freeing it here.
2417 struct xfs_defer_ops
*dfops
)
2420 struct xfs_icluster xic
= { 0 };
2422 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2423 ASSERT(VFS_I(ip
)->i_nlink
== 0);
2424 ASSERT(ip
->i_d
.di_nextents
== 0);
2425 ASSERT(ip
->i_d
.di_anextents
== 0);
2426 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(VFS_I(ip
)->i_mode
));
2427 ASSERT(ip
->i_d
.di_nblocks
== 0);
2430 * Pull the on-disk inode from the AGI unlinked list.
2432 error
= xfs_iunlink_remove(tp
, ip
);
2436 error
= xfs_difree(tp
, ip
->i_ino
, dfops
, &xic
);
2440 VFS_I(ip
)->i_mode
= 0; /* mark incore inode as free */
2441 ip
->i_d
.di_flags
= 0;
2442 ip
->i_d
.di_dmevmask
= 0;
2443 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2444 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2445 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2447 * Bump the generation count so no one will be confused
2448 * by reincarnations of this inode.
2450 VFS_I(ip
)->i_generation
++;
2451 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2454 error
= xfs_ifree_cluster(ip
, tp
, &xic
);
2460 * This is called to unpin an inode. The caller must have the inode locked
2461 * in at least shared mode so that the buffer cannot be subsequently pinned
2462 * once someone is waiting for it to be unpinned.
2466 struct xfs_inode
*ip
)
2468 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2470 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2472 /* Give the log a push to start the unpinning I/O */
2473 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2479 struct xfs_inode
*ip
)
2481 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2482 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2487 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2488 if (xfs_ipincount(ip
))
2490 } while (xfs_ipincount(ip
));
2491 finish_wait(wq
, &wait
.wait
);
2496 struct xfs_inode
*ip
)
2498 if (xfs_ipincount(ip
))
2499 __xfs_iunpin_wait(ip
);
2503 * Removing an inode from the namespace involves removing the directory entry
2504 * and dropping the link count on the inode. Removing the directory entry can
2505 * result in locking an AGF (directory blocks were freed) and removing a link
2506 * count can result in placing the inode on an unlinked list which results in
2509 * The big problem here is that we have an ordering constraint on AGF and AGI
2510 * locking - inode allocation locks the AGI, then can allocate a new extent for
2511 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2512 * removes the inode from the unlinked list, requiring that we lock the AGI
2513 * first, and then freeing the inode can result in an inode chunk being freed
2514 * and hence freeing disk space requiring that we lock an AGF.
2516 * Hence the ordering that is imposed by other parts of the code is AGI before
2517 * AGF. This means we cannot remove the directory entry before we drop the inode
2518 * reference count and put it on the unlinked list as this results in a lock
2519 * order of AGF then AGI, and this can deadlock against inode allocation and
2520 * freeing. Therefore we must drop the link counts before we remove the
2523 * This is still safe from a transactional point of view - it is not until we
2524 * get to xfs_defer_finish() that we have the possibility of multiple
2525 * transactions in this operation. Hence as long as we remove the directory
2526 * entry and drop the link count in the first transaction of the remove
2527 * operation, there are no transactional constraints on the ordering here.
2532 struct xfs_name
*name
,
2535 xfs_mount_t
*mp
= dp
->i_mount
;
2536 xfs_trans_t
*tp
= NULL
;
2537 int is_dir
= S_ISDIR(VFS_I(ip
)->i_mode
);
2539 struct xfs_defer_ops dfops
;
2540 xfs_fsblock_t first_block
;
2543 trace_xfs_remove(dp
, name
);
2545 if (XFS_FORCED_SHUTDOWN(mp
))
2548 error
= xfs_qm_dqattach(dp
, 0);
2552 error
= xfs_qm_dqattach(ip
, 0);
2557 * We try to get the real space reservation first,
2558 * allowing for directory btree deletion(s) implying
2559 * possible bmap insert(s). If we can't get the space
2560 * reservation then we use 0 instead, and avoid the bmap
2561 * btree insert(s) in the directory code by, if the bmap
2562 * insert tries to happen, instead trimming the LAST
2563 * block from the directory.
2565 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2566 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, resblks
, 0, 0, &tp
);
2567 if (error
== -ENOSPC
) {
2569 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_remove
, 0, 0, 0,
2573 ASSERT(error
!= -ENOSPC
);
2577 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2579 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2580 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2583 * If we're removing a directory perform some additional validation.
2586 ASSERT(VFS_I(ip
)->i_nlink
>= 2);
2587 if (VFS_I(ip
)->i_nlink
!= 2) {
2589 goto out_trans_cancel
;
2591 if (!xfs_dir_isempty(ip
)) {
2593 goto out_trans_cancel
;
2596 /* Drop the link from ip's "..". */
2597 error
= xfs_droplink(tp
, dp
);
2599 goto out_trans_cancel
;
2601 /* Drop the "." link from ip to self. */
2602 error
= xfs_droplink(tp
, ip
);
2604 goto out_trans_cancel
;
2607 * When removing a non-directory we need to log the parent
2608 * inode here. For a directory this is done implicitly
2609 * by the xfs_droplink call for the ".." entry.
2611 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2613 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2615 /* Drop the link from dp to ip. */
2616 error
= xfs_droplink(tp
, ip
);
2618 goto out_trans_cancel
;
2620 xfs_defer_init(&dfops
, &first_block
);
2621 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2622 &first_block
, &dfops
, resblks
);
2624 ASSERT(error
!= -ENOENT
);
2625 goto out_bmap_cancel
;
2629 * If this is a synchronous mount, make sure that the
2630 * remove transaction goes to disk before returning to
2633 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2634 xfs_trans_set_sync(tp
);
2636 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
2638 goto out_bmap_cancel
;
2640 error
= xfs_trans_commit(tp
);
2644 if (is_dir
&& xfs_inode_is_filestream(ip
))
2645 xfs_filestream_deassociate(ip
);
2650 xfs_defer_cancel(&dfops
);
2652 xfs_trans_cancel(tp
);
2658 * Enter all inodes for a rename transaction into a sorted array.
2660 #define __XFS_SORT_INODES 5
2662 xfs_sort_for_rename(
2663 struct xfs_inode
*dp1
, /* in: old (source) directory inode */
2664 struct xfs_inode
*dp2
, /* in: new (target) directory inode */
2665 struct xfs_inode
*ip1
, /* in: inode of old entry */
2666 struct xfs_inode
*ip2
, /* in: inode of new entry */
2667 struct xfs_inode
*wip
, /* in: whiteout inode */
2668 struct xfs_inode
**i_tab
,/* out: sorted array of inodes */
2669 int *num_inodes
) /* in/out: inodes in array */
2673 ASSERT(*num_inodes
== __XFS_SORT_INODES
);
2674 memset(i_tab
, 0, *num_inodes
* sizeof(struct xfs_inode
*));
2677 * i_tab contains a list of pointers to inodes. We initialize
2678 * the table here & we'll sort it. We will then use it to
2679 * order the acquisition of the inode locks.
2681 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2694 * Sort the elements via bubble sort. (Remember, there are at
2695 * most 5 elements to sort, so this is adequate.)
2697 for (i
= 0; i
< *num_inodes
; i
++) {
2698 for (j
= 1; j
< *num_inodes
; j
++) {
2699 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2700 struct xfs_inode
*temp
= i_tab
[j
];
2701 i_tab
[j
] = i_tab
[j
-1];
2710 struct xfs_trans
*tp
,
2711 struct xfs_defer_ops
*dfops
)
2716 * If this is a synchronous mount, make sure that the rename transaction
2717 * goes to disk before returning to the user.
2719 if (tp
->t_mountp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2720 xfs_trans_set_sync(tp
);
2722 error
= xfs_defer_finish(&tp
, dfops
, NULL
);
2724 xfs_defer_cancel(dfops
);
2725 xfs_trans_cancel(tp
);
2729 return xfs_trans_commit(tp
);
2733 * xfs_cross_rename()
2735 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2739 struct xfs_trans
*tp
,
2740 struct xfs_inode
*dp1
,
2741 struct xfs_name
*name1
,
2742 struct xfs_inode
*ip1
,
2743 struct xfs_inode
*dp2
,
2744 struct xfs_name
*name2
,
2745 struct xfs_inode
*ip2
,
2746 struct xfs_defer_ops
*dfops
,
2747 xfs_fsblock_t
*first_block
,
2755 /* Swap inode number for dirent in first parent */
2756 error
= xfs_dir_replace(tp
, dp1
, name1
,
2758 first_block
, dfops
, spaceres
);
2760 goto out_trans_abort
;
2762 /* Swap inode number for dirent in second parent */
2763 error
= xfs_dir_replace(tp
, dp2
, name2
,
2765 first_block
, dfops
, spaceres
);
2767 goto out_trans_abort
;
2770 * If we're renaming one or more directories across different parents,
2771 * update the respective ".." entries (and link counts) to match the new
2775 dp2_flags
= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2777 if (S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2778 error
= xfs_dir_replace(tp
, ip2
, &xfs_name_dotdot
,
2779 dp1
->i_ino
, first_block
,
2782 goto out_trans_abort
;
2784 /* transfer ip2 ".." reference to dp1 */
2785 if (!S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2786 error
= xfs_droplink(tp
, dp2
);
2788 goto out_trans_abort
;
2789 error
= xfs_bumplink(tp
, dp1
);
2791 goto out_trans_abort
;
2795 * Although ip1 isn't changed here, userspace needs
2796 * to be warned about the change, so that applications
2797 * relying on it (like backup ones), will properly
2800 ip1_flags
|= XFS_ICHGTIME_CHG
;
2801 ip2_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2804 if (S_ISDIR(VFS_I(ip1
)->i_mode
)) {
2805 error
= xfs_dir_replace(tp
, ip1
, &xfs_name_dotdot
,
2806 dp2
->i_ino
, first_block
,
2809 goto out_trans_abort
;
2811 /* transfer ip1 ".." reference to dp2 */
2812 if (!S_ISDIR(VFS_I(ip2
)->i_mode
)) {
2813 error
= xfs_droplink(tp
, dp1
);
2815 goto out_trans_abort
;
2816 error
= xfs_bumplink(tp
, dp2
);
2818 goto out_trans_abort
;
2822 * Although ip2 isn't changed here, userspace needs
2823 * to be warned about the change, so that applications
2824 * relying on it (like backup ones), will properly
2827 ip1_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2828 ip2_flags
|= XFS_ICHGTIME_CHG
;
2833 xfs_trans_ichgtime(tp
, ip1
, ip1_flags
);
2834 xfs_trans_log_inode(tp
, ip1
, XFS_ILOG_CORE
);
2837 xfs_trans_ichgtime(tp
, ip2
, ip2_flags
);
2838 xfs_trans_log_inode(tp
, ip2
, XFS_ILOG_CORE
);
2841 xfs_trans_ichgtime(tp
, dp2
, dp2_flags
);
2842 xfs_trans_log_inode(tp
, dp2
, XFS_ILOG_CORE
);
2844 xfs_trans_ichgtime(tp
, dp1
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2845 xfs_trans_log_inode(tp
, dp1
, XFS_ILOG_CORE
);
2846 return xfs_finish_rename(tp
, dfops
);
2849 xfs_defer_cancel(dfops
);
2850 xfs_trans_cancel(tp
);
2855 * xfs_rename_alloc_whiteout()
2857 * Return a referenced, unlinked, unlocked inode that that can be used as a
2858 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2859 * crash between allocating the inode and linking it into the rename transaction
2860 * recovery will free the inode and we won't leak it.
2863 xfs_rename_alloc_whiteout(
2864 struct xfs_inode
*dp
,
2865 struct xfs_inode
**wip
)
2867 struct xfs_inode
*tmpfile
;
2870 error
= xfs_create_tmpfile(dp
, NULL
, S_IFCHR
| WHITEOUT_MODE
, &tmpfile
);
2875 * Prepare the tmpfile inode as if it were created through the VFS.
2876 * Otherwise, the link increment paths will complain about nlink 0->1.
2877 * Drop the link count as done by d_tmpfile(), complete the inode setup
2878 * and flag it as linkable.
2880 drop_nlink(VFS_I(tmpfile
));
2881 xfs_setup_iops(tmpfile
);
2882 xfs_finish_inode_setup(tmpfile
);
2883 VFS_I(tmpfile
)->i_state
|= I_LINKABLE
;
2894 struct xfs_inode
*src_dp
,
2895 struct xfs_name
*src_name
,
2896 struct xfs_inode
*src_ip
,
2897 struct xfs_inode
*target_dp
,
2898 struct xfs_name
*target_name
,
2899 struct xfs_inode
*target_ip
,
2902 struct xfs_mount
*mp
= src_dp
->i_mount
;
2903 struct xfs_trans
*tp
;
2904 struct xfs_defer_ops dfops
;
2905 xfs_fsblock_t first_block
;
2906 struct xfs_inode
*wip
= NULL
; /* whiteout inode */
2907 struct xfs_inode
*inodes
[__XFS_SORT_INODES
];
2908 int num_inodes
= __XFS_SORT_INODES
;
2909 bool new_parent
= (src_dp
!= target_dp
);
2910 bool src_is_directory
= S_ISDIR(VFS_I(src_ip
)->i_mode
);
2914 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2916 if ((flags
& RENAME_EXCHANGE
) && !target_ip
)
2920 * If we are doing a whiteout operation, allocate the whiteout inode
2921 * we will be placing at the target and ensure the type is set
2924 if (flags
& RENAME_WHITEOUT
) {
2925 ASSERT(!(flags
& (RENAME_NOREPLACE
| RENAME_EXCHANGE
)));
2926 error
= xfs_rename_alloc_whiteout(target_dp
, &wip
);
2930 /* setup target dirent info as whiteout */
2931 src_name
->type
= XFS_DIR3_FT_CHRDEV
;
2934 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
, wip
,
2935 inodes
, &num_inodes
);
2937 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2938 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, spaceres
, 0, 0, &tp
);
2939 if (error
== -ENOSPC
) {
2941 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_rename
, 0, 0, 0,
2945 goto out_release_wip
;
2948 * Attach the dquots to the inodes
2950 error
= xfs_qm_vop_rename_dqattach(inodes
);
2952 goto out_trans_cancel
;
2955 * Lock all the participating inodes. Depending upon whether
2956 * the target_name exists in the target directory, and
2957 * whether the target directory is the same as the source
2958 * directory, we can lock from 2 to 4 inodes.
2960 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2963 * Join all the inodes to the transaction. From this point on,
2964 * we can rely on either trans_commit or trans_cancel to unlock
2967 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2969 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2970 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2972 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2974 xfs_trans_ijoin(tp
, wip
, XFS_ILOCK_EXCL
);
2977 * If we are using project inheritance, we only allow renames
2978 * into our tree when the project IDs are the same; else the
2979 * tree quota mechanism would be circumvented.
2981 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2982 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2984 goto out_trans_cancel
;
2987 xfs_defer_init(&dfops
, &first_block
);
2989 /* RENAME_EXCHANGE is unique from here on. */
2990 if (flags
& RENAME_EXCHANGE
)
2991 return xfs_cross_rename(tp
, src_dp
, src_name
, src_ip
,
2992 target_dp
, target_name
, target_ip
,
2993 &dfops
, &first_block
, spaceres
);
2996 * Set up the target.
2998 if (target_ip
== NULL
) {
3000 * If there's no space reservation, check the entry will
3001 * fit before actually inserting it.
3004 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
3006 goto out_trans_cancel
;
3009 * If target does not exist and the rename crosses
3010 * directories, adjust the target directory link count
3011 * to account for the ".." reference from the new entry.
3013 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
3014 src_ip
->i_ino
, &first_block
,
3017 goto out_bmap_cancel
;
3019 xfs_trans_ichgtime(tp
, target_dp
,
3020 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3022 if (new_parent
&& src_is_directory
) {
3023 error
= xfs_bumplink(tp
, target_dp
);
3025 goto out_bmap_cancel
;
3027 } else { /* target_ip != NULL */
3029 * If target exists and it's a directory, check that both
3030 * target and source are directories and that target can be
3031 * destroyed, or that neither is a directory.
3033 if (S_ISDIR(VFS_I(target_ip
)->i_mode
)) {
3035 * Make sure target dir is empty.
3037 if (!(xfs_dir_isempty(target_ip
)) ||
3038 (VFS_I(target_ip
)->i_nlink
> 2)) {
3040 goto out_trans_cancel
;
3045 * Link the source inode under the target name.
3046 * If the source inode is a directory and we are moving
3047 * it across directories, its ".." entry will be
3048 * inconsistent until we replace that down below.
3050 * In case there is already an entry with the same
3051 * name at the destination directory, remove it first.
3053 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
3055 &first_block
, &dfops
, spaceres
);
3057 goto out_bmap_cancel
;
3059 xfs_trans_ichgtime(tp
, target_dp
,
3060 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3063 * Decrement the link count on the target since the target
3064 * dir no longer points to it.
3066 error
= xfs_droplink(tp
, target_ip
);
3068 goto out_bmap_cancel
;
3070 if (src_is_directory
) {
3072 * Drop the link from the old "." entry.
3074 error
= xfs_droplink(tp
, target_ip
);
3076 goto out_bmap_cancel
;
3078 } /* target_ip != NULL */
3081 * Remove the source.
3083 if (new_parent
&& src_is_directory
) {
3085 * Rewrite the ".." entry to point to the new
3088 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
3090 &first_block
, &dfops
, spaceres
);
3091 ASSERT(error
!= -EEXIST
);
3093 goto out_bmap_cancel
;
3097 * We always want to hit the ctime on the source inode.
3099 * This isn't strictly required by the standards since the source
3100 * inode isn't really being changed, but old unix file systems did
3101 * it and some incremental backup programs won't work without it.
3103 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
3104 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
3107 * Adjust the link count on src_dp. This is necessary when
3108 * renaming a directory, either within one parent when
3109 * the target existed, or across two parent directories.
3111 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
3114 * Decrement link count on src_directory since the
3115 * entry that's moved no longer points to it.
3117 error
= xfs_droplink(tp
, src_dp
);
3119 goto out_bmap_cancel
;
3123 * For whiteouts, we only need to update the source dirent with the
3124 * inode number of the whiteout inode rather than removing it
3128 error
= xfs_dir_replace(tp
, src_dp
, src_name
, wip
->i_ino
,
3129 &first_block
, &dfops
, spaceres
);
3131 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
3132 &first_block
, &dfops
, spaceres
);
3134 goto out_bmap_cancel
;
3137 * For whiteouts, we need to bump the link count on the whiteout inode.
3138 * This means that failures all the way up to this point leave the inode
3139 * on the unlinked list and so cleanup is a simple matter of dropping
3140 * the remaining reference to it. If we fail here after bumping the link
3141 * count, we're shutting down the filesystem so we'll never see the
3142 * intermediate state on disk.
3145 ASSERT(VFS_I(wip
)->i_nlink
== 0);
3146 error
= xfs_bumplink(tp
, wip
);
3148 goto out_bmap_cancel
;
3149 error
= xfs_iunlink_remove(tp
, wip
);
3151 goto out_bmap_cancel
;
3152 xfs_trans_log_inode(tp
, wip
, XFS_ILOG_CORE
);
3155 * Now we have a real link, clear the "I'm a tmpfile" state
3156 * flag from the inode so it doesn't accidentally get misused in
3159 VFS_I(wip
)->i_state
&= ~I_LINKABLE
;
3162 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3163 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
3165 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
3167 error
= xfs_finish_rename(tp
, &dfops
);
3173 xfs_defer_cancel(&dfops
);
3175 xfs_trans_cancel(tp
);
3184 struct xfs_inode
*ip
,
3187 struct xfs_mount
*mp
= ip
->i_mount
;
3188 struct xfs_perag
*pag
;
3189 unsigned long first_index
, mask
;
3190 unsigned long inodes_per_cluster
;
3192 struct xfs_inode
**cilist
;
3193 struct xfs_inode
*cip
;
3199 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
3201 inodes_per_cluster
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
3202 cilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
3203 cilist
= kmem_alloc(cilist_size
, KM_MAYFAIL
|KM_NOFS
);
3207 mask
= ~(((mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
)) - 1);
3208 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
3210 /* really need a gang lookup range call here */
3211 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)cilist
,
3212 first_index
, inodes_per_cluster
);
3216 for (i
= 0; i
< nr_found
; i
++) {
3222 * because this is an RCU protected lookup, we could find a
3223 * recently freed or even reallocated inode during the lookup.
3224 * We need to check under the i_flags_lock for a valid inode
3225 * here. Skip it if it is not valid or the wrong inode.
3227 spin_lock(&cip
->i_flags_lock
);
3229 __xfs_iflags_test(cip
, XFS_ISTALE
)) {
3230 spin_unlock(&cip
->i_flags_lock
);
3235 * Once we fall off the end of the cluster, no point checking
3236 * any more inodes in the list because they will also all be
3237 * outside the cluster.
3239 if ((XFS_INO_TO_AGINO(mp
, cip
->i_ino
) & mask
) != first_index
) {
3240 spin_unlock(&cip
->i_flags_lock
);
3243 spin_unlock(&cip
->i_flags_lock
);
3246 * Do an un-protected check to see if the inode is dirty and
3247 * is a candidate for flushing. These checks will be repeated
3248 * later after the appropriate locks are acquired.
3250 if (xfs_inode_clean(cip
) && xfs_ipincount(cip
) == 0)
3254 * Try to get locks. If any are unavailable or it is pinned,
3255 * then this inode cannot be flushed and is skipped.
3258 if (!xfs_ilock_nowait(cip
, XFS_ILOCK_SHARED
))
3260 if (!xfs_iflock_nowait(cip
)) {
3261 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3264 if (xfs_ipincount(cip
)) {
3266 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3272 * Check the inode number again, just to be certain we are not
3273 * racing with freeing in xfs_reclaim_inode(). See the comments
3274 * in that function for more information as to why the initial
3275 * check is not sufficient.
3279 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3284 * arriving here means that this inode can be flushed. First
3285 * re-check that it's dirty before flushing.
3287 if (!xfs_inode_clean(cip
)) {
3289 error
= xfs_iflush_int(cip
, bp
);
3291 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3292 goto cluster_corrupt_out
;
3298 xfs_iunlock(cip
, XFS_ILOCK_SHARED
);
3302 XFS_STATS_INC(mp
, xs_icluster_flushcnt
);
3303 XFS_STATS_ADD(mp
, xs_icluster_flushinode
, clcount
);
3314 cluster_corrupt_out
:
3316 * Corruption detected in the clustering loop. Invalidate the
3317 * inode buffer and shut down the filesystem.
3321 * Clean up the buffer. If it was delwri, just release it --
3322 * brelse can handle it with no problems. If not, shut down the
3323 * filesystem before releasing the buffer.
3325 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3329 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3331 if (!bufwasdelwri
) {
3333 * Just like incore_relse: if we have b_iodone functions,
3334 * mark the buffer as an error and call them. Otherwise
3335 * mark it as stale and brelse.
3338 bp
->b_flags
&= ~XBF_DONE
;
3340 xfs_buf_ioerror(bp
, -EIO
);
3349 * Unlocks the flush lock
3351 xfs_iflush_abort(cip
, false);
3354 return -EFSCORRUPTED
;
3358 * Flush dirty inode metadata into the backing buffer.
3360 * The caller must have the inode lock and the inode flush lock held. The
3361 * inode lock will still be held upon return to the caller, and the inode
3362 * flush lock will be released after the inode has reached the disk.
3364 * The caller must write out the buffer returned in *bpp and release it.
3368 struct xfs_inode
*ip
,
3369 struct xfs_buf
**bpp
)
3371 struct xfs_mount
*mp
= ip
->i_mount
;
3372 struct xfs_buf
*bp
= NULL
;
3373 struct xfs_dinode
*dip
;
3376 XFS_STATS_INC(mp
, xs_iflush_count
);
3378 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3379 ASSERT(xfs_isiflocked(ip
));
3380 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3381 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3385 xfs_iunpin_wait(ip
);
3388 * For stale inodes we cannot rely on the backing buffer remaining
3389 * stale in cache for the remaining life of the stale inode and so
3390 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3391 * inodes below. We have to check this after ensuring the inode is
3392 * unpinned so that it is safe to reclaim the stale inode after the
3395 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3401 * This may have been unpinned because the filesystem is shutting
3402 * down forcibly. If that's the case we must not write this inode
3403 * to disk, because the log record didn't make it to disk.
3405 * We also have to remove the log item from the AIL in this case,
3406 * as we wait for an empty AIL as part of the unmount process.
3408 if (XFS_FORCED_SHUTDOWN(mp
)) {
3414 * Get the buffer containing the on-disk inode. We are doing a try-lock
3415 * operation here, so we may get an EAGAIN error. In that case, we
3416 * simply want to return with the inode still dirty.
3418 * If we get any other error, we effectively have a corruption situation
3419 * and we cannot flush the inode, so we treat it the same as failing
3422 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3424 if (error
== -EAGAIN
) {
3432 * First flush out the inode that xfs_iflush was called with.
3434 error
= xfs_iflush_int(ip
, bp
);
3439 * If the buffer is pinned then push on the log now so we won't
3440 * get stuck waiting in the write for too long.
3442 if (xfs_buf_ispinned(bp
))
3443 xfs_log_force(mp
, 0);
3447 * see if other inodes can be gathered into this write
3449 error
= xfs_iflush_cluster(ip
, bp
);
3451 goto cluster_corrupt_out
;
3459 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3460 cluster_corrupt_out
:
3461 error
= -EFSCORRUPTED
;
3464 * Unlocks the flush lock
3466 xfs_iflush_abort(ip
, false);
3472 struct xfs_inode
*ip
,
3475 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3476 struct xfs_dinode
*dip
;
3477 struct xfs_mount
*mp
= ip
->i_mount
;
3479 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3480 ASSERT(xfs_isiflocked(ip
));
3481 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3482 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3483 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3484 ASSERT(ip
->i_d
.di_version
> 1);
3486 /* set *dip = inode's place in the buffer */
3487 dip
= xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3489 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3490 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3491 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3492 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3493 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3496 if (S_ISREG(VFS_I(ip
)->i_mode
)) {
3498 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3499 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3500 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3501 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3502 "%s: Bad regular inode %Lu, ptr 0x%p",
3503 __func__
, ip
->i_ino
, ip
);
3506 } else if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
3508 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3509 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3510 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3511 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3512 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3513 "%s: Bad directory inode %Lu, ptr 0x%p",
3514 __func__
, ip
->i_ino
, ip
);
3518 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3519 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3520 XFS_RANDOM_IFLUSH_5
)) {
3521 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3522 "%s: detected corrupt incore inode %Lu, "
3523 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3524 __func__
, ip
->i_ino
,
3525 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3526 ip
->i_d
.di_nblocks
, ip
);
3529 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3530 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3531 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3532 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3533 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3538 * Inode item log recovery for v2 inodes are dependent on the
3539 * di_flushiter count for correct sequencing. We bump the flush
3540 * iteration count so we can detect flushes which postdate a log record
3541 * during recovery. This is redundant as we now log every change and
3542 * hence this can't happen but we need to still do it to ensure
3543 * backwards compatibility with old kernels that predate logging all
3546 if (ip
->i_d
.di_version
< 3)
3547 ip
->i_d
.di_flushiter
++;
3550 * Copy the dirty parts of the inode into the on-disk inode. We always
3551 * copy out the core of the inode, because if the inode is dirty at all
3554 xfs_inode_to_disk(ip
, dip
, iip
->ili_item
.li_lsn
);
3556 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3557 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3558 ip
->i_d
.di_flushiter
= 0;
3560 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3561 if (XFS_IFORK_Q(ip
))
3562 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3563 xfs_inobp_check(mp
, bp
);
3566 * We've recorded everything logged in the inode, so we'd like to clear
3567 * the ili_fields bits so we don't log and flush things unnecessarily.
3568 * However, we can't stop logging all this information until the data
3569 * we've copied into the disk buffer is written to disk. If we did we
3570 * might overwrite the copy of the inode in the log with all the data
3571 * after re-logging only part of it, and in the face of a crash we
3572 * wouldn't have all the data we need to recover.
3574 * What we do is move the bits to the ili_last_fields field. When
3575 * logging the inode, these bits are moved back to the ili_fields field.
3576 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3577 * know that the information those bits represent is permanently on
3578 * disk. As long as the flush completes before the inode is logged
3579 * again, then both ili_fields and ili_last_fields will be cleared.
3581 * We can play with the ili_fields bits here, because the inode lock
3582 * must be held exclusively in order to set bits there and the flush
3583 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3584 * done routine can tell whether or not to look in the AIL. Also, store
3585 * the current LSN of the inode so that we can tell whether the item has
3586 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3587 * need the AIL lock, because it is a 64 bit value that cannot be read
3590 iip
->ili_last_fields
= iip
->ili_fields
;
3591 iip
->ili_fields
= 0;
3592 iip
->ili_fsync_fields
= 0;
3593 iip
->ili_logged
= 1;
3595 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3596 &iip
->ili_item
.li_lsn
);
3599 * Attach the function xfs_iflush_done to the inode's
3600 * buffer. This will remove the inode from the AIL
3601 * and unlock the inode's flush lock when the inode is
3602 * completely written to disk.
3604 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3606 /* generate the checksum. */
3607 xfs_dinode_calc_crc(mp
, dip
);
3609 ASSERT(bp
->b_fspriv
!= NULL
);
3610 ASSERT(bp
->b_iodone
!= NULL
);
3614 return -EFSCORRUPTED
;