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_inode.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
32 #include "xfs_attr_sf.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_ialloc.h"
40 #include "xfs_bmap_util.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_filestream.h"
44 #include "xfs_cksum.h"
45 #include "xfs_trace.h"
46 #include "xfs_icache.h"
47 #include "xfs_symlink.h"
48 #include "xfs_trans_priv.h"
50 #include "xfs_bmap_btree.h"
52 kmem_zone_t
*xfs_inode_zone
;
55 * Used in xfs_itruncate_extents(). This is the maximum number of extents
56 * freed from a file in a single transaction.
58 #define XFS_ITRUNC_MAX_EXTENTS 2
60 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
62 STATIC
int xfs_iunlink_remove(xfs_trans_t
*, xfs_inode_t
*);
65 * helper function to extract extent size hint from inode
71 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
72 return ip
->i_d
.di_extsize
;
73 if (XFS_IS_REALTIME_INODE(ip
))
74 return ip
->i_mount
->m_sb
.sb_rextsize
;
79 * These two are wrapper routines around the xfs_ilock() routine used to
80 * centralize some grungy code. They are used in places that wish to lock the
81 * inode solely for reading the extents. The reason these places can't just
82 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83 * bringing in of the extents from disk for a file in b-tree format. If the
84 * inode is in b-tree format, then we need to lock the inode exclusively until
85 * the extents are read in. Locking it exclusively all the time would limit
86 * our parallelism unnecessarily, though. What we do instead is check to see
87 * if the extents have been read in yet, and only lock the inode exclusively
90 * The functions return a value which should be given to the corresponding
94 xfs_ilock_data_map_shared(
97 uint lock_mode
= XFS_ILOCK_SHARED
;
99 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
&&
100 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)
101 lock_mode
= XFS_ILOCK_EXCL
;
102 xfs_ilock(ip
, lock_mode
);
107 xfs_ilock_attr_map_shared(
108 struct xfs_inode
*ip
)
110 uint lock_mode
= XFS_ILOCK_SHARED
;
112 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_BTREE
&&
113 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) == 0)
114 lock_mode
= XFS_ILOCK_EXCL
;
115 xfs_ilock(ip
, lock_mode
);
120 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
121 * the i_lock. This routine allows various combinations of the locks to be
124 * The 3 locks should always be ordered so that the IO lock is obtained first,
125 * the mmap lock second and the ilock last in order to prevent deadlock.
127 * Basic locking order:
129 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
131 * mmap_sem locking order:
133 * i_iolock -> page lock -> mmap_sem
134 * mmap_sem -> i_mmap_lock -> page_lock
136 * The difference in mmap_sem locking order mean that we cannot hold the
137 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
138 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
139 * in get_user_pages() to map the user pages into the kernel address space for
140 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
141 * page faults already hold the mmap_sem.
143 * Hence to serialise fully against both syscall and mmap based IO, we need to
144 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
145 * taken in places where we need to invalidate the page cache in a race
146 * free manner (e.g. truncate, hole punch and other extent manipulation
154 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
157 * You can't set both SHARED and EXCL for the same lock,
158 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
159 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
161 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
162 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
163 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
164 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
165 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
166 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
167 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
169 if (lock_flags
& XFS_IOLOCK_EXCL
)
170 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
171 else if (lock_flags
& XFS_IOLOCK_SHARED
)
172 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
174 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
175 mrupdate_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
176 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
177 mraccess_nested(&ip
->i_mmaplock
, XFS_MMAPLOCK_DEP(lock_flags
));
179 if (lock_flags
& XFS_ILOCK_EXCL
)
180 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
181 else if (lock_flags
& XFS_ILOCK_SHARED
)
182 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
186 * This is just like xfs_ilock(), except that the caller
187 * is guaranteed not to sleep. It returns 1 if it gets
188 * the requested locks and 0 otherwise. If the IO lock is
189 * obtained but the inode lock cannot be, then the IO lock
190 * is dropped before returning.
192 * ip -- the inode being locked
193 * lock_flags -- this parameter indicates the inode's locks to be
194 * to be locked. See the comment for xfs_ilock() for a list
202 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
205 * You can't set both SHARED and EXCL for the same lock,
206 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
207 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
209 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
210 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
211 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
212 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
213 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
214 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
215 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
217 if (lock_flags
& XFS_IOLOCK_EXCL
) {
218 if (!mrtryupdate(&ip
->i_iolock
))
220 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
221 if (!mrtryaccess(&ip
->i_iolock
))
225 if (lock_flags
& XFS_MMAPLOCK_EXCL
) {
226 if (!mrtryupdate(&ip
->i_mmaplock
))
227 goto out_undo_iolock
;
228 } else if (lock_flags
& XFS_MMAPLOCK_SHARED
) {
229 if (!mrtryaccess(&ip
->i_mmaplock
))
230 goto out_undo_iolock
;
233 if (lock_flags
& XFS_ILOCK_EXCL
) {
234 if (!mrtryupdate(&ip
->i_lock
))
235 goto out_undo_mmaplock
;
236 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
237 if (!mrtryaccess(&ip
->i_lock
))
238 goto out_undo_mmaplock
;
243 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
244 mrunlock_excl(&ip
->i_mmaplock
);
245 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
246 mrunlock_shared(&ip
->i_mmaplock
);
248 if (lock_flags
& XFS_IOLOCK_EXCL
)
249 mrunlock_excl(&ip
->i_iolock
);
250 else if (lock_flags
& XFS_IOLOCK_SHARED
)
251 mrunlock_shared(&ip
->i_iolock
);
257 * xfs_iunlock() is used to drop the inode locks acquired with
258 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
259 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
260 * that we know which locks to drop.
262 * ip -- the inode being unlocked
263 * lock_flags -- this parameter indicates the inode's locks to be
264 * to be unlocked. See the comment for xfs_ilock() for a list
265 * of valid values for this parameter.
274 * You can't set both SHARED and EXCL for the same lock,
275 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
276 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
278 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
279 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
280 ASSERT((lock_flags
& (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
)) !=
281 (XFS_MMAPLOCK_SHARED
| XFS_MMAPLOCK_EXCL
));
282 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
283 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
284 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
285 ASSERT(lock_flags
!= 0);
287 if (lock_flags
& XFS_IOLOCK_EXCL
)
288 mrunlock_excl(&ip
->i_iolock
);
289 else if (lock_flags
& XFS_IOLOCK_SHARED
)
290 mrunlock_shared(&ip
->i_iolock
);
292 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
293 mrunlock_excl(&ip
->i_mmaplock
);
294 else if (lock_flags
& XFS_MMAPLOCK_SHARED
)
295 mrunlock_shared(&ip
->i_mmaplock
);
297 if (lock_flags
& XFS_ILOCK_EXCL
)
298 mrunlock_excl(&ip
->i_lock
);
299 else if (lock_flags
& XFS_ILOCK_SHARED
)
300 mrunlock_shared(&ip
->i_lock
);
302 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
306 * give up write locks. the i/o lock cannot be held nested
307 * if it is being demoted.
314 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
));
316 ~(XFS_IOLOCK_EXCL
|XFS_MMAPLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
318 if (lock_flags
& XFS_ILOCK_EXCL
)
319 mrdemote(&ip
->i_lock
);
320 if (lock_flags
& XFS_MMAPLOCK_EXCL
)
321 mrdemote(&ip
->i_mmaplock
);
322 if (lock_flags
& XFS_IOLOCK_EXCL
)
323 mrdemote(&ip
->i_iolock
);
325 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
328 #if defined(DEBUG) || defined(XFS_WARN)
334 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
335 if (!(lock_flags
& XFS_ILOCK_SHARED
))
336 return !!ip
->i_lock
.mr_writer
;
337 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
340 if (lock_flags
& (XFS_MMAPLOCK_EXCL
|XFS_MMAPLOCK_SHARED
)) {
341 if (!(lock_flags
& XFS_MMAPLOCK_SHARED
))
342 return !!ip
->i_mmaplock
.mr_writer
;
343 return rwsem_is_locked(&ip
->i_mmaplock
.mr_lock
);
346 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
347 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
348 return !!ip
->i_iolock
.mr_writer
;
349 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);
359 int xfs_small_retries
;
360 int xfs_middle_retries
;
361 int xfs_lots_retries
;
366 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
367 * value. This shouldn't be called for page fault locking, but we also need to
368 * ensure we don't overrun the number of lockdep subclasses for the iolock or
369 * mmaplock as that is limited to 12 by the mmap lock lockdep annotations.
372 xfs_lock_inumorder(int lock_mode
, int subclass
)
374 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
375 ASSERT(subclass
+ XFS_LOCK_INUMORDER
<
376 (1 << (XFS_MMAPLOCK_SHIFT
- XFS_IOLOCK_SHIFT
)));
377 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
380 if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)) {
381 ASSERT(subclass
+ XFS_LOCK_INUMORDER
<
382 (1 << (XFS_ILOCK_SHIFT
- XFS_MMAPLOCK_SHIFT
)));
383 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) <<
387 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
388 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
394 * The following routine will lock n inodes in exclusive mode. We assume the
395 * caller calls us with the inodes in i_ino order.
397 * We need to detect deadlock where an inode that we lock is in the AIL and we
398 * start waiting for another inode that is locked by a thread in a long running
399 * transaction (such as truncate). This can result in deadlock since the long
400 * running trans might need to wait for the inode we just locked in order to
401 * push the tail and free space in the log.
409 int attempts
= 0, i
, j
, try_lock
;
412 /* currently supports between 2 and 5 inodes */
413 ASSERT(ips
&& inodes
>= 2 && inodes
<= 5);
418 for (; i
< inodes
; i
++) {
421 if (i
&& (ips
[i
] == ips
[i
- 1])) /* Already locked */
425 * If try_lock is not set yet, make sure all locked inodes are
426 * not in the AIL. If any are, set try_lock to be used later.
429 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
430 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
431 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
))
437 * If any of the previous locks we have locked is in the AIL,
438 * we must TRY to get the second and subsequent locks. If
439 * we can't get any, we must release all we have
443 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
447 /* try_lock means we have an inode locked that is in the AIL. */
449 if (xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
)))
453 * Unlock all previous guys and try again. xfs_iunlock will try
454 * to push the tail if the inode is in the AIL.
457 for (j
= i
- 1; j
>= 0; j
--) {
459 * Check to see if we've already unlocked this one. Not
460 * the first one going back, and the inode ptr is the
463 if (j
!= (i
- 1) && ips
[j
] == ips
[j
+ 1])
466 xfs_iunlock(ips
[j
], lock_mode
);
469 if ((attempts
% 5) == 0) {
470 delay(1); /* Don't just spin the CPU */
482 if (attempts
< 5) xfs_small_retries
++;
483 else if (attempts
< 100) xfs_middle_retries
++;
484 else xfs_lots_retries
++;
492 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
493 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
494 * lock more than one at a time, lockdep will report false positives saying we
495 * have violated locking orders.
507 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
)) {
508 ASSERT(!(lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
)));
509 ASSERT(!(lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
510 } else if (lock_mode
& (XFS_MMAPLOCK_SHARED
|XFS_MMAPLOCK_EXCL
))
511 ASSERT(!(lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)));
513 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
515 if (ip0
->i_ino
> ip1
->i_ino
) {
522 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
525 * If the first lock we have locked is in the AIL, we must TRY to get
526 * the second lock. If we can't get it, we must release the first one
529 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
530 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
531 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
532 xfs_iunlock(ip0
, lock_mode
);
533 if ((++attempts
% 5) == 0)
534 delay(1); /* Don't just spin the CPU */
538 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
545 struct xfs_inode
*ip
)
547 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
548 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
551 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
552 if (xfs_isiflocked(ip
))
554 } while (!xfs_iflock_nowait(ip
));
556 finish_wait(wq
, &wait
.wait
);
565 if (di_flags
& XFS_DIFLAG_ANY
) {
566 if (di_flags
& XFS_DIFLAG_REALTIME
)
567 flags
|= XFS_XFLAG_REALTIME
;
568 if (di_flags
& XFS_DIFLAG_PREALLOC
)
569 flags
|= XFS_XFLAG_PREALLOC
;
570 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
571 flags
|= XFS_XFLAG_IMMUTABLE
;
572 if (di_flags
& XFS_DIFLAG_APPEND
)
573 flags
|= XFS_XFLAG_APPEND
;
574 if (di_flags
& XFS_DIFLAG_SYNC
)
575 flags
|= XFS_XFLAG_SYNC
;
576 if (di_flags
& XFS_DIFLAG_NOATIME
)
577 flags
|= XFS_XFLAG_NOATIME
;
578 if (di_flags
& XFS_DIFLAG_NODUMP
)
579 flags
|= XFS_XFLAG_NODUMP
;
580 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
581 flags
|= XFS_XFLAG_RTINHERIT
;
582 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
583 flags
|= XFS_XFLAG_PROJINHERIT
;
584 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
585 flags
|= XFS_XFLAG_NOSYMLINKS
;
586 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
587 flags
|= XFS_XFLAG_EXTSIZE
;
588 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
589 flags
|= XFS_XFLAG_EXTSZINHERIT
;
590 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
591 flags
|= XFS_XFLAG_NODEFRAG
;
592 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
593 flags
|= XFS_XFLAG_FILESTREAM
;
603 xfs_icdinode_t
*dic
= &ip
->i_d
;
605 return _xfs_dic2xflags(dic
->di_flags
) |
606 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
613 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
614 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
618 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
619 * is allowed, otherwise it has to be an exact match. If a CI match is found,
620 * ci_name->name will point to a the actual name (caller must free) or
621 * will be set to NULL if an exact match is found.
626 struct xfs_name
*name
,
628 struct xfs_name
*ci_name
)
634 trace_xfs_lookup(dp
, name
);
636 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
639 lock_mode
= xfs_ilock_data_map_shared(dp
);
640 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
641 xfs_iunlock(dp
, lock_mode
);
646 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
654 kmem_free(ci_name
->name
);
661 * Allocate an inode on disk and return a copy of its in-core version.
662 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
663 * appropriately within the inode. The uid and gid for the inode are
664 * set according to the contents of the given cred structure.
666 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
667 * has a free inode available, call xfs_iget() to obtain the in-core
668 * version of the allocated inode. Finally, fill in the inode and
669 * log its initial contents. In this case, ialloc_context would be
672 * If xfs_dialloc() does not have an available inode, it will replenish
673 * its supply by doing an allocation. Since we can only do one
674 * allocation within a transaction without deadlocks, we must commit
675 * the current transaction before returning the inode itself.
676 * In this case, therefore, we will set ialloc_context and return.
677 * The caller should then commit the current transaction, start a new
678 * transaction, and call xfs_ialloc() again to actually get the inode.
680 * To ensure that some other process does not grab the inode that
681 * was allocated during the first call to xfs_ialloc(), this routine
682 * also returns the [locked] bp pointing to the head of the freelist
683 * as ialloc_context. The caller should hold this buffer across
684 * the commit and pass it back into this routine on the second call.
686 * If we are allocating quota inodes, we do not have a parent inode
687 * to attach to or associate with (i.e. pip == NULL) because they
688 * are not linked into the directory structure - they are attached
689 * directly to the superblock - and so have no parent.
700 xfs_buf_t
**ialloc_context
,
703 struct xfs_mount
*mp
= tp
->t_mountp
;
711 * Call the space management code to pick
712 * the on-disk inode to be allocated.
714 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
715 ialloc_context
, &ino
);
718 if (*ialloc_context
|| ino
== NULLFSINO
) {
722 ASSERT(*ialloc_context
== NULL
);
725 * Get the in-core inode with the lock held exclusively.
726 * This is because we're setting fields here we need
727 * to prevent others from looking at until we're done.
729 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_CREATE
,
730 XFS_ILOCK_EXCL
, &ip
);
736 * We always convert v1 inodes to v2 now - we only support filesystems
737 * with >= v2 inode capability, so there is no reason for ever leaving
738 * an inode in v1 format.
740 if (ip
->i_d
.di_version
== 1)
741 ip
->i_d
.di_version
= 2;
743 ip
->i_d
.di_mode
= mode
;
744 ip
->i_d
.di_onlink
= 0;
745 ip
->i_d
.di_nlink
= nlink
;
746 ASSERT(ip
->i_d
.di_nlink
== nlink
);
747 ip
->i_d
.di_uid
= xfs_kuid_to_uid(current_fsuid());
748 ip
->i_d
.di_gid
= xfs_kgid_to_gid(current_fsgid());
749 xfs_set_projid(ip
, prid
);
750 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
752 if (pip
&& XFS_INHERIT_GID(pip
)) {
753 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
754 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
755 ip
->i_d
.di_mode
|= S_ISGID
;
760 * If the group ID of the new file does not match the effective group
761 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
762 * (and only if the irix_sgid_inherit compatibility variable is set).
764 if ((irix_sgid_inherit
) &&
765 (ip
->i_d
.di_mode
& S_ISGID
) &&
766 (!in_group_p(xfs_gid_to_kgid(ip
->i_d
.di_gid
)))) {
767 ip
->i_d
.di_mode
&= ~S_ISGID
;
771 ip
->i_d
.di_nextents
= 0;
772 ASSERT(ip
->i_d
.di_nblocks
== 0);
774 tv
= current_fs_time(mp
->m_super
);
775 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
776 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
777 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
778 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
781 * di_gen will have been taken care of in xfs_iread.
783 ip
->i_d
.di_extsize
= 0;
784 ip
->i_d
.di_dmevmask
= 0;
785 ip
->i_d
.di_dmstate
= 0;
786 ip
->i_d
.di_flags
= 0;
788 if (ip
->i_d
.di_version
== 3) {
789 ASSERT(ip
->i_d
.di_ino
== ino
);
790 ASSERT(uuid_equal(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
));
792 ip
->i_d
.di_changecount
= 1;
794 ip
->i_d
.di_flags2
= 0;
795 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
796 ip
->i_d
.di_crtime
= ip
->i_d
.di_mtime
;
800 flags
= XFS_ILOG_CORE
;
801 switch (mode
& S_IFMT
) {
806 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
807 ip
->i_df
.if_u2
.if_rdev
= rdev
;
808 ip
->i_df
.if_flags
= 0;
809 flags
|= XFS_ILOG_DEV
;
813 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
817 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
818 di_flags
|= XFS_DIFLAG_RTINHERIT
;
819 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
820 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
821 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
823 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
824 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
825 } else if (S_ISREG(mode
)) {
826 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
827 di_flags
|= XFS_DIFLAG_REALTIME
;
828 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
829 di_flags
|= XFS_DIFLAG_EXTSIZE
;
830 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
833 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
835 di_flags
|= XFS_DIFLAG_NOATIME
;
836 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
838 di_flags
|= XFS_DIFLAG_NODUMP
;
839 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
841 di_flags
|= XFS_DIFLAG_SYNC
;
842 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
843 xfs_inherit_nosymlinks
)
844 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
845 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
846 xfs_inherit_nodefrag
)
847 di_flags
|= XFS_DIFLAG_NODEFRAG
;
848 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
849 di_flags
|= XFS_DIFLAG_FILESTREAM
;
850 ip
->i_d
.di_flags
|= di_flags
;
854 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
855 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
856 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
857 ip
->i_df
.if_u1
.if_extents
= NULL
;
863 * Attribute fork settings for new inode.
865 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
866 ip
->i_d
.di_anextents
= 0;
869 * Log the new values stuffed into the inode.
871 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
872 xfs_trans_log_inode(tp
, ip
, flags
);
874 /* now that we have an i_mode we can setup the inode structure */
882 * Allocates a new inode from disk and return a pointer to the
883 * incore copy. This routine will internally commit the current
884 * transaction and allocate a new one if the Space Manager needed
885 * to do an allocation to replenish the inode free-list.
887 * This routine is designed to be called from xfs_create and
893 xfs_trans_t
**tpp
, /* input: current transaction;
894 output: may be a new transaction. */
895 xfs_inode_t
*dp
, /* directory within whose allocate
900 prid_t prid
, /* project id */
901 int okalloc
, /* ok to allocate new space */
902 xfs_inode_t
**ipp
, /* pointer to inode; it will be
909 xfs_buf_t
*ialloc_context
= NULL
;
915 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
918 * xfs_ialloc will return a pointer to an incore inode if
919 * the Space Manager has an available inode on the free
920 * list. Otherwise, it will do an allocation and replenish
921 * the freelist. Since we can only do one allocation per
922 * transaction without deadlocks, we will need to commit the
923 * current transaction and start a new one. We will then
924 * need to call xfs_ialloc again to get the inode.
926 * If xfs_ialloc did an allocation to replenish the freelist,
927 * it returns the bp containing the head of the freelist as
928 * ialloc_context. We will hold a lock on it across the
929 * transaction commit so that no other process can steal
930 * the inode(s) that we've just allocated.
932 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
, okalloc
,
933 &ialloc_context
, &ip
);
936 * Return an error if we were unable to allocate a new inode.
937 * This should only happen if we run out of space on disk or
938 * encounter a disk error.
944 if (!ialloc_context
&& !ip
) {
950 * If the AGI buffer is non-NULL, then we were unable to get an
951 * inode in one operation. We need to commit the current
952 * transaction and call xfs_ialloc() again. It is guaranteed
953 * to succeed the second time.
955 if (ialloc_context
) {
957 * Normally, xfs_trans_commit releases all the locks.
958 * We call bhold to hang on to the ialloc_context across
959 * the commit. Holding this buffer prevents any other
960 * processes from doing any allocations in this
963 xfs_trans_bhold(tp
, ialloc_context
);
966 * We want the quota changes to be associated with the next
967 * transaction, NOT this one. So, detach the dqinfo from this
968 * and attach it to the next transaction.
973 dqinfo
= (void *)tp
->t_dqinfo
;
975 tflags
= tp
->t_flags
& XFS_TRANS_DQ_DIRTY
;
976 tp
->t_flags
&= ~(XFS_TRANS_DQ_DIRTY
);
979 code
= xfs_trans_roll(&tp
, 0);
980 if (committed
!= NULL
)
984 * Re-attach the quota info that we detached from prev trx.
987 tp
->t_dqinfo
= dqinfo
;
988 tp
->t_flags
|= tflags
;
992 xfs_buf_relse(ialloc_context
);
997 xfs_trans_bjoin(tp
, ialloc_context
);
1000 * Call ialloc again. Since we've locked out all
1001 * other allocations in this allocation group,
1002 * this call should always succeed.
1004 code
= xfs_ialloc(tp
, dp
, mode
, nlink
, rdev
, prid
,
1005 okalloc
, &ialloc_context
, &ip
);
1008 * If we get an error at this point, return to the caller
1009 * so that the current transaction can be aborted.
1016 ASSERT(!ialloc_context
&& ip
);
1019 if (committed
!= NULL
)
1030 * Decrement the link count on an inode & log the change.
1031 * If this causes the link count to go to zero, initiate the
1032 * logging activity required to truncate a file.
1041 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1043 ASSERT (ip
->i_d
.di_nlink
> 0);
1045 drop_nlink(VFS_I(ip
));
1046 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1049 if (ip
->i_d
.di_nlink
== 0) {
1051 * We're dropping the last link to this file.
1052 * Move the on-disk inode to the AGI unlinked list.
1053 * From xfs_inactive() we will pull the inode from
1054 * the list and free it.
1056 error
= xfs_iunlink(tp
, ip
);
1062 * Increment the link count on an inode & log the change.
1069 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_CHG
);
1071 ASSERT(ip
->i_d
.di_version
> 1);
1072 ASSERT(ip
->i_d
.di_nlink
> 0 || (VFS_I(ip
)->i_state
& I_LINKABLE
));
1074 inc_nlink(VFS_I(ip
));
1075 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1082 struct xfs_name
*name
,
1087 int is_dir
= S_ISDIR(mode
);
1088 struct xfs_mount
*mp
= dp
->i_mount
;
1089 struct xfs_inode
*ip
= NULL
;
1090 struct xfs_trans
*tp
= NULL
;
1092 xfs_bmap_free_t free_list
;
1093 xfs_fsblock_t first_block
;
1094 bool unlock_dp_on_error
= false;
1097 struct xfs_dquot
*udqp
= NULL
;
1098 struct xfs_dquot
*gdqp
= NULL
;
1099 struct xfs_dquot
*pdqp
= NULL
;
1100 struct xfs_trans_res
*tres
;
1103 trace_xfs_create(dp
, name
);
1105 if (XFS_FORCED_SHUTDOWN(mp
))
1108 prid
= xfs_get_initial_prid(dp
);
1111 * Make sure that we have allocated dquot(s) on disk.
1113 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1114 xfs_kgid_to_gid(current_fsgid()), prid
,
1115 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1116 &udqp
, &gdqp
, &pdqp
);
1122 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1123 tres
= &M_RES(mp
)->tr_mkdir
;
1124 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1126 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1127 tres
= &M_RES(mp
)->tr_create
;
1128 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1132 * Initially assume that the file does not exist and
1133 * reserve the resources for that case. If that is not
1134 * the case we'll drop the one we have and get a more
1135 * appropriate transaction later.
1137 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1138 if (error
== -ENOSPC
) {
1139 /* flush outstanding delalloc blocks and retry */
1140 xfs_flush_inodes(mp
);
1141 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1143 if (error
== -ENOSPC
) {
1144 /* No space at all so try a "no-allocation" reservation */
1146 error
= xfs_trans_reserve(tp
, tres
, 0, 0);
1149 goto out_trans_cancel
;
1152 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1153 unlock_dp_on_error
= true;
1155 xfs_bmap_init(&free_list
, &first_block
);
1158 * Reserve disk quota and the inode.
1160 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1161 pdqp
, resblks
, 1, 0);
1163 goto out_trans_cancel
;
1166 error
= xfs_dir_canenter(tp
, dp
, name
);
1168 goto out_trans_cancel
;
1172 * A newly created regular or special file just has one directory
1173 * entry pointing to them, but a directory also the "." entry
1174 * pointing to itself.
1176 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1177 prid
, resblks
> 0, &ip
, &committed
);
1179 goto out_trans_cancel
;
1182 * Now we join the directory inode to the transaction. We do not do it
1183 * earlier because xfs_dir_ialloc might commit the previous transaction
1184 * (and release all the locks). An error from here on will result in
1185 * the transaction cancel unlocking dp so don't do it explicitly in the
1188 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1189 unlock_dp_on_error
= false;
1191 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1192 &first_block
, &free_list
, resblks
?
1193 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1195 ASSERT(error
!= -ENOSPC
);
1196 goto out_trans_cancel
;
1198 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1199 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1202 error
= xfs_dir_init(tp
, ip
, dp
);
1204 goto out_bmap_cancel
;
1206 error
= xfs_bumplink(tp
, dp
);
1208 goto out_bmap_cancel
;
1212 * If this is a synchronous mount, make sure that the
1213 * create transaction goes to disk before returning to
1216 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1217 xfs_trans_set_sync(tp
);
1220 * Attach the dquot(s) to the inodes and modify them incore.
1221 * These ids of the inode couldn't have changed since the new
1222 * inode has been locked ever since it was created.
1224 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1226 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1228 goto out_bmap_cancel
;
1230 error
= xfs_trans_commit(tp
);
1232 goto out_release_inode
;
1234 xfs_qm_dqrele(udqp
);
1235 xfs_qm_dqrele(gdqp
);
1236 xfs_qm_dqrele(pdqp
);
1242 xfs_bmap_cancel(&free_list
);
1244 xfs_trans_cancel(tp
);
1247 * Wait until after the current transaction is aborted to finish the
1248 * setup of the inode and release the inode. This prevents recursive
1249 * transactions and deadlocks from xfs_inactive.
1252 xfs_finish_inode_setup(ip
);
1256 xfs_qm_dqrele(udqp
);
1257 xfs_qm_dqrele(gdqp
);
1258 xfs_qm_dqrele(pdqp
);
1260 if (unlock_dp_on_error
)
1261 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1267 struct xfs_inode
*dp
,
1268 struct dentry
*dentry
,
1270 struct xfs_inode
**ipp
)
1272 struct xfs_mount
*mp
= dp
->i_mount
;
1273 struct xfs_inode
*ip
= NULL
;
1274 struct xfs_trans
*tp
= NULL
;
1277 struct xfs_dquot
*udqp
= NULL
;
1278 struct xfs_dquot
*gdqp
= NULL
;
1279 struct xfs_dquot
*pdqp
= NULL
;
1280 struct xfs_trans_res
*tres
;
1283 if (XFS_FORCED_SHUTDOWN(mp
))
1286 prid
= xfs_get_initial_prid(dp
);
1289 * Make sure that we have allocated dquot(s) on disk.
1291 error
= xfs_qm_vop_dqalloc(dp
, xfs_kuid_to_uid(current_fsuid()),
1292 xfs_kgid_to_gid(current_fsgid()), prid
,
1293 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
,
1294 &udqp
, &gdqp
, &pdqp
);
1298 resblks
= XFS_IALLOC_SPACE_RES(mp
);
1299 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE_TMPFILE
);
1301 tres
= &M_RES(mp
)->tr_create_tmpfile
;
1302 error
= xfs_trans_reserve(tp
, tres
, resblks
, 0);
1303 if (error
== -ENOSPC
) {
1304 /* No space at all so try a "no-allocation" reservation */
1306 error
= xfs_trans_reserve(tp
, tres
, 0, 0);
1309 goto out_trans_cancel
;
1311 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
,
1312 pdqp
, resblks
, 1, 0);
1314 goto out_trans_cancel
;
1316 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1, 0,
1317 prid
, resblks
> 0, &ip
, NULL
);
1319 goto out_trans_cancel
;
1321 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1322 xfs_trans_set_sync(tp
);
1325 * Attach the dquot(s) to the inodes and modify them incore.
1326 * These ids of the inode couldn't have changed since the new
1327 * inode has been locked ever since it was created.
1329 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
, pdqp
);
1332 error
= xfs_iunlink(tp
, ip
);
1334 goto out_trans_cancel
;
1336 error
= xfs_trans_commit(tp
);
1338 goto out_release_inode
;
1340 xfs_qm_dqrele(udqp
);
1341 xfs_qm_dqrele(gdqp
);
1342 xfs_qm_dqrele(pdqp
);
1348 xfs_trans_cancel(tp
);
1351 * Wait until after the current transaction is aborted to finish the
1352 * setup of the inode and release the inode. This prevents recursive
1353 * transactions and deadlocks from xfs_inactive.
1356 xfs_finish_inode_setup(ip
);
1360 xfs_qm_dqrele(udqp
);
1361 xfs_qm_dqrele(gdqp
);
1362 xfs_qm_dqrele(pdqp
);
1371 struct xfs_name
*target_name
)
1373 xfs_mount_t
*mp
= tdp
->i_mount
;
1376 xfs_bmap_free_t free_list
;
1377 xfs_fsblock_t first_block
;
1381 trace_xfs_link(tdp
, target_name
);
1383 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1385 if (XFS_FORCED_SHUTDOWN(mp
))
1388 error
= xfs_qm_dqattach(sip
, 0);
1392 error
= xfs_qm_dqattach(tdp
, 0);
1396 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1397 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1398 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, resblks
, 0);
1399 if (error
== -ENOSPC
) {
1401 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_link
, 0, 0);
1406 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1408 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1409 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1412 * If we are using project inheritance, we only allow hard link
1413 * creation in our tree when the project IDs are the same; else
1414 * the tree quota mechanism could be circumvented.
1416 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1417 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1423 error
= xfs_dir_canenter(tp
, tdp
, target_name
);
1428 xfs_bmap_init(&free_list
, &first_block
);
1430 if (sip
->i_d
.di_nlink
== 0) {
1431 error
= xfs_iunlink_remove(tp
, sip
);
1436 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1437 &first_block
, &free_list
, resblks
);
1440 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1441 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1443 error
= xfs_bumplink(tp
, sip
);
1448 * If this is a synchronous mount, make sure that the
1449 * link transaction goes to disk before returning to
1452 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1453 xfs_trans_set_sync(tp
);
1456 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1458 xfs_bmap_cancel(&free_list
);
1462 return xfs_trans_commit(tp
);
1465 xfs_trans_cancel(tp
);
1471 * Free up the underlying blocks past new_size. The new size must be smaller
1472 * than the current size. This routine can be used both for the attribute and
1473 * data fork, and does not modify the inode size, which is left to the caller.
1475 * The transaction passed to this routine must have made a permanent log
1476 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1477 * given transaction and start new ones, so make sure everything involved in
1478 * the transaction is tidy before calling here. Some transaction will be
1479 * returned to the caller to be committed. The incoming transaction must
1480 * already include the inode, and both inode locks must be held exclusively.
1481 * The inode must also be "held" within the transaction. On return the inode
1482 * will be "held" within the returned transaction. This routine does NOT
1483 * require any disk space to be reserved for it within the transaction.
1485 * If we get an error, we must return with the inode locked and linked into the
1486 * current transaction. This keeps things simple for the higher level code,
1487 * because it always knows that the inode is locked and held in the transaction
1488 * that returns to it whether errors occur or not. We don't mark the inode
1489 * dirty on error so that transactions can be easily aborted if possible.
1492 xfs_itruncate_extents(
1493 struct xfs_trans
**tpp
,
1494 struct xfs_inode
*ip
,
1496 xfs_fsize_t new_size
)
1498 struct xfs_mount
*mp
= ip
->i_mount
;
1499 struct xfs_trans
*tp
= *tpp
;
1500 xfs_bmap_free_t free_list
;
1501 xfs_fsblock_t first_block
;
1502 xfs_fileoff_t first_unmap_block
;
1503 xfs_fileoff_t last_block
;
1504 xfs_filblks_t unmap_len
;
1509 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1510 ASSERT(!atomic_read(&VFS_I(ip
)->i_count
) ||
1511 xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1512 ASSERT(new_size
<= XFS_ISIZE(ip
));
1513 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1514 ASSERT(ip
->i_itemp
!= NULL
);
1515 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1516 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1518 trace_xfs_itruncate_extents_start(ip
, new_size
);
1521 * Since it is possible for space to become allocated beyond
1522 * the end of the file (in a crash where the space is allocated
1523 * but the inode size is not yet updated), simply remove any
1524 * blocks which show up between the new EOF and the maximum
1525 * possible file size. If the first block to be removed is
1526 * beyond the maximum file size (ie it is the same as last_block),
1527 * then there is nothing to do.
1529 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1530 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1531 if (first_unmap_block
== last_block
)
1534 ASSERT(first_unmap_block
< last_block
);
1535 unmap_len
= last_block
- first_unmap_block
+ 1;
1537 xfs_bmap_init(&free_list
, &first_block
);
1538 error
= xfs_bunmapi(tp
, ip
,
1539 first_unmap_block
, unmap_len
,
1540 xfs_bmapi_aflag(whichfork
),
1541 XFS_ITRUNC_MAX_EXTENTS
,
1542 &first_block
, &free_list
,
1545 goto out_bmap_cancel
;
1548 * Duplicate the transaction that has the permanent
1549 * reservation and commit the old transaction.
1551 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1553 xfs_trans_ijoin(tp
, ip
, 0);
1555 goto out_bmap_cancel
;
1557 error
= xfs_trans_roll(&tp
, ip
);
1563 * Always re-log the inode so that our permanent transaction can keep
1564 * on rolling it forward in the log.
1566 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1568 trace_xfs_itruncate_extents_end(ip
, new_size
);
1575 * If the bunmapi call encounters an error, return to the caller where
1576 * the transaction can be properly aborted. We just need to make sure
1577 * we're not holding any resources that we were not when we came in.
1579 xfs_bmap_cancel(&free_list
);
1587 xfs_mount_t
*mp
= ip
->i_mount
;
1590 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1593 /* If this is a read-only mount, don't do this (would generate I/O) */
1594 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1597 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1601 * If we previously truncated this file and removed old data
1602 * in the process, we want to initiate "early" writeout on
1603 * the last close. This is an attempt to combat the notorious
1604 * NULL files problem which is particularly noticeable from a
1605 * truncate down, buffered (re-)write (delalloc), followed by
1606 * a crash. What we are effectively doing here is
1607 * significantly reducing the time window where we'd otherwise
1608 * be exposed to that problem.
1610 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1612 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
1613 if (ip
->i_delayed_blks
> 0) {
1614 error
= filemap_flush(VFS_I(ip
)->i_mapping
);
1621 if (ip
->i_d
.di_nlink
== 0)
1624 if (xfs_can_free_eofblocks(ip
, false)) {
1627 * If we can't get the iolock just skip truncating the blocks
1628 * past EOF because we could deadlock with the mmap_sem
1629 * otherwise. We'll get another chance to drop them once the
1630 * last reference to the inode is dropped, so we'll never leak
1631 * blocks permanently.
1633 * Further, check if the inode is being opened, written and
1634 * closed frequently and we have delayed allocation blocks
1635 * outstanding (e.g. streaming writes from the NFS server),
1636 * truncating the blocks past EOF will cause fragmentation to
1639 * In this case don't do the truncation, either, but we have to
1640 * be careful how we detect this case. Blocks beyond EOF show
1641 * up as i_delayed_blks even when the inode is clean, so we
1642 * need to truncate them away first before checking for a dirty
1643 * release. Hence on the first dirty close we will still remove
1644 * the speculative allocation, but after that we will leave it
1647 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1650 error
= xfs_free_eofblocks(mp
, ip
, true);
1651 if (error
&& error
!= -EAGAIN
)
1654 /* delalloc blocks after truncation means it really is dirty */
1655 if (ip
->i_delayed_blks
)
1656 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1662 * xfs_inactive_truncate
1664 * Called to perform a truncate when an inode becomes unlinked.
1667 xfs_inactive_truncate(
1668 struct xfs_inode
*ip
)
1670 struct xfs_mount
*mp
= ip
->i_mount
;
1671 struct xfs_trans
*tp
;
1674 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1675 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
1677 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1678 xfs_trans_cancel(tp
);
1682 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1683 xfs_trans_ijoin(tp
, ip
, 0);
1686 * Log the inode size first to prevent stale data exposure in the event
1687 * of a system crash before the truncate completes. See the related
1688 * comment in xfs_setattr_size() for details.
1690 ip
->i_d
.di_size
= 0;
1691 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1693 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
1695 goto error_trans_cancel
;
1697 ASSERT(ip
->i_d
.di_nextents
== 0);
1699 error
= xfs_trans_commit(tp
);
1703 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1707 xfs_trans_cancel(tp
);
1709 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1714 * xfs_inactive_ifree()
1716 * Perform the inode free when an inode is unlinked.
1720 struct xfs_inode
*ip
)
1722 xfs_bmap_free_t free_list
;
1723 xfs_fsblock_t first_block
;
1725 struct xfs_mount
*mp
= ip
->i_mount
;
1726 struct xfs_trans
*tp
;
1729 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1732 * The ifree transaction might need to allocate blocks for record
1733 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1734 * allow ifree to dip into the reserved block pool if necessary.
1736 * Freeing large sets of inodes generally means freeing inode chunks,
1737 * directory and file data blocks, so this should be relatively safe.
1738 * Only under severe circumstances should it be possible to free enough
1739 * inodes to exhaust the reserve block pool via finobt expansion while
1740 * at the same time not creating free space in the filesystem.
1742 * Send a warning if the reservation does happen to fail, as the inode
1743 * now remains allocated and sits on the unlinked list until the fs is
1746 tp
->t_flags
|= XFS_TRANS_RESERVE
;
1747 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ifree
,
1748 XFS_IFREE_SPACE_RES(mp
), 0);
1750 if (error
== -ENOSPC
) {
1751 xfs_warn_ratelimited(mp
,
1752 "Failed to remove inode(s) from unlinked list. "
1753 "Please free space, unmount and run xfs_repair.");
1755 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1757 xfs_trans_cancel(tp
);
1761 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1762 xfs_trans_ijoin(tp
, ip
, 0);
1764 xfs_bmap_init(&free_list
, &first_block
);
1765 error
= xfs_ifree(tp
, ip
, &free_list
);
1768 * If we fail to free the inode, shut down. The cancel
1769 * might do that, we need to make sure. Otherwise the
1770 * inode might be lost for a long time or forever.
1772 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1773 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1775 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1777 xfs_trans_cancel(tp
);
1778 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1783 * Credit the quota account(s). The inode is gone.
1785 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1788 * Just ignore errors at this point. There is nothing we can do except
1789 * to try to keep going. Make sure it's not a silent error.
1791 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1793 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1795 xfs_bmap_cancel(&free_list
);
1797 error
= xfs_trans_commit(tp
);
1799 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1802 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1809 * This is called when the vnode reference count for the vnode
1810 * goes to zero. If the file has been unlinked, then it must
1811 * now be truncated. Also, we clear all of the read-ahead state
1812 * kept for the inode here since the file is now closed.
1818 struct xfs_mount
*mp
;
1823 * If the inode is already free, then there can be nothing
1826 if (ip
->i_d
.di_mode
== 0) {
1827 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1828 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1834 /* If this is a read-only mount, don't do this (would generate I/O) */
1835 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1838 if (ip
->i_d
.di_nlink
!= 0) {
1840 * force is true because we are evicting an inode from the
1841 * cache. Post-eof blocks must be freed, lest we end up with
1842 * broken free space accounting.
1844 if (xfs_can_free_eofblocks(ip
, true))
1845 xfs_free_eofblocks(mp
, ip
, false);
1850 if (S_ISREG(ip
->i_d
.di_mode
) &&
1851 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
1852 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
1855 error
= xfs_qm_dqattach(ip
, 0);
1859 if (S_ISLNK(ip
->i_d
.di_mode
))
1860 error
= xfs_inactive_symlink(ip
);
1862 error
= xfs_inactive_truncate(ip
);
1867 * If there are attributes associated with the file then blow them away
1868 * now. The code calls a routine that recursively deconstructs the
1869 * attribute fork. If also blows away the in-core attribute fork.
1871 if (XFS_IFORK_Q(ip
)) {
1872 error
= xfs_attr_inactive(ip
);
1878 ASSERT(ip
->i_d
.di_anextents
== 0);
1879 ASSERT(ip
->i_d
.di_forkoff
== 0);
1884 error
= xfs_inactive_ifree(ip
);
1889 * Release the dquots held by inode, if any.
1891 xfs_qm_dqdetach(ip
);
1895 * This is called when the inode's link count goes to 0.
1896 * We place the on-disk inode on a list in the AGI. It
1897 * will be pulled from this list when the inode is freed.
1914 ASSERT(ip
->i_d
.di_nlink
== 0);
1915 ASSERT(ip
->i_d
.di_mode
!= 0);
1920 * Get the agi buffer first. It ensures lock ordering
1923 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1926 agi
= XFS_BUF_TO_AGI(agibp
);
1929 * Get the index into the agi hash table for the
1930 * list this inode will go on.
1932 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1934 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1935 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1936 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1938 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1940 * There is already another inode in the bucket we need
1941 * to add ourselves to. Add us at the front of the list.
1942 * Here we put the head pointer into our next pointer,
1943 * and then we fall through to point the head at us.
1945 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1950 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1951 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1952 offset
= ip
->i_imap
.im_boffset
+
1953 offsetof(xfs_dinode_t
, di_next_unlinked
);
1955 /* need to recalc the inode CRC if appropriate */
1956 xfs_dinode_calc_crc(mp
, dip
);
1958 xfs_trans_inode_buf(tp
, ibp
);
1959 xfs_trans_log_buf(tp
, ibp
, offset
,
1960 (offset
+ sizeof(xfs_agino_t
) - 1));
1961 xfs_inobp_check(mp
, ibp
);
1965 * Point the bucket head pointer at the inode being inserted.
1968 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1969 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1970 (sizeof(xfs_agino_t
) * bucket_index
);
1971 xfs_trans_buf_set_type(tp
, agibp
, XFS_BLFT_AGI_BUF
);
1972 xfs_trans_log_buf(tp
, agibp
, offset
,
1973 (offset
+ sizeof(xfs_agino_t
) - 1));
1978 * Pull the on-disk inode from the AGI unlinked list.
1991 xfs_agnumber_t agno
;
1993 xfs_agino_t next_agino
;
1994 xfs_buf_t
*last_ibp
;
1995 xfs_dinode_t
*last_dip
= NULL
;
1997 int offset
, last_offset
= 0;
2001 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2004 * Get the agi buffer first. It ensures lock ordering
2007 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
2011 agi
= XFS_BUF_TO_AGI(agibp
);
2014 * Get the index into the agi hash table for the
2015 * list this inode will go on.
2017 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2019 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2020 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
2021 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2023 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2025 * We're at the head of the list. Get the inode's on-disk
2026 * buffer to see if there is anyone after us on the list.
2027 * Only modify our next pointer if it is not already NULLAGINO.
2028 * This saves us the overhead of dealing with the buffer when
2029 * there is no need to change it.
2031 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2034 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
2038 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2039 ASSERT(next_agino
!= 0);
2040 if (next_agino
!= NULLAGINO
) {
2041 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2042 offset
= ip
->i_imap
.im_boffset
+
2043 offsetof(xfs_dinode_t
, di_next_unlinked
);
2045 /* need to recalc the inode CRC if appropriate */
2046 xfs_dinode_calc_crc(mp
, dip
);
2048 xfs_trans_inode_buf(tp
, ibp
);
2049 xfs_trans_log_buf(tp
, ibp
, offset
,
2050 (offset
+ sizeof(xfs_agino_t
) - 1));
2051 xfs_inobp_check(mp
, ibp
);
2053 xfs_trans_brelse(tp
, ibp
);
2056 * Point the bucket head pointer at the next inode.
2058 ASSERT(next_agino
!= 0);
2059 ASSERT(next_agino
!= agino
);
2060 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2061 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2062 (sizeof(xfs_agino_t
) * bucket_index
);
2063 xfs_trans_buf_set_type(tp
, agibp
, XFS_BLFT_AGI_BUF
);
2064 xfs_trans_log_buf(tp
, agibp
, offset
,
2065 (offset
+ sizeof(xfs_agino_t
) - 1));
2068 * We need to search the list for the inode being freed.
2070 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2072 while (next_agino
!= agino
) {
2073 struct xfs_imap imap
;
2076 xfs_trans_brelse(tp
, last_ibp
);
2079 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2081 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
2084 "%s: xfs_imap returned error %d.",
2089 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
2093 "%s: xfs_imap_to_bp returned error %d.",
2098 last_offset
= imap
.im_boffset
;
2099 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2100 ASSERT(next_agino
!= NULLAGINO
);
2101 ASSERT(next_agino
!= 0);
2105 * Now last_ibp points to the buffer previous to us on the
2106 * unlinked list. Pull us from the list.
2108 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
2111 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
2115 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2116 ASSERT(next_agino
!= 0);
2117 ASSERT(next_agino
!= agino
);
2118 if (next_agino
!= NULLAGINO
) {
2119 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2120 offset
= ip
->i_imap
.im_boffset
+
2121 offsetof(xfs_dinode_t
, di_next_unlinked
);
2123 /* need to recalc the inode CRC if appropriate */
2124 xfs_dinode_calc_crc(mp
, dip
);
2126 xfs_trans_inode_buf(tp
, ibp
);
2127 xfs_trans_log_buf(tp
, ibp
, offset
,
2128 (offset
+ sizeof(xfs_agino_t
) - 1));
2129 xfs_inobp_check(mp
, ibp
);
2131 xfs_trans_brelse(tp
, ibp
);
2134 * Point the previous inode on the list to the next inode.
2136 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2137 ASSERT(next_agino
!= 0);
2138 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2140 /* need to recalc the inode CRC if appropriate */
2141 xfs_dinode_calc_crc(mp
, last_dip
);
2143 xfs_trans_inode_buf(tp
, last_ibp
);
2144 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2145 (offset
+ sizeof(xfs_agino_t
) - 1));
2146 xfs_inobp_check(mp
, last_ibp
);
2152 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2153 * inodes that are in memory - they all must be marked stale and attached to
2154 * the cluster buffer.
2158 xfs_inode_t
*free_ip
,
2160 struct xfs_icluster
*xic
)
2162 xfs_mount_t
*mp
= free_ip
->i_mount
;
2163 int blks_per_cluster
;
2164 int inodes_per_cluster
;
2171 xfs_inode_log_item_t
*iip
;
2172 xfs_log_item_t
*lip
;
2173 struct xfs_perag
*pag
;
2176 inum
= xic
->first_ino
;
2177 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
2178 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2179 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
2180 nbufs
= mp
->m_ialloc_blks
/ blks_per_cluster
;
2182 for (j
= 0; j
< nbufs
; j
++, inum
+= inodes_per_cluster
) {
2184 * The allocation bitmap tells us which inodes of the chunk were
2185 * physically allocated. Skip the cluster if an inode falls into
2188 ioffset
= inum
- xic
->first_ino
;
2189 if ((xic
->alloc
& XFS_INOBT_MASK(ioffset
)) == 0) {
2190 ASSERT(do_mod(ioffset
, inodes_per_cluster
) == 0);
2194 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2195 XFS_INO_TO_AGBNO(mp
, inum
));
2198 * We obtain and lock the backing buffer first in the process
2199 * here, as we have to ensure that any dirty inode that we
2200 * can't get the flush lock on is attached to the buffer.
2201 * If we scan the in-memory inodes first, then buffer IO can
2202 * complete before we get a lock on it, and hence we may fail
2203 * to mark all the active inodes on the buffer stale.
2205 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2206 mp
->m_bsize
* blks_per_cluster
,
2213 * This buffer may not have been correctly initialised as we
2214 * didn't read it from disk. That's not important because we are
2215 * only using to mark the buffer as stale in the log, and to
2216 * attach stale cached inodes on it. That means it will never be
2217 * dispatched for IO. If it is, we want to know about it, and we
2218 * want it to fail. We can acheive this by adding a write
2219 * verifier to the buffer.
2221 bp
->b_ops
= &xfs_inode_buf_ops
;
2224 * Walk the inodes already attached to the buffer and mark them
2225 * stale. These will all have the flush locks held, so an
2226 * in-memory inode walk can't lock them. By marking them all
2227 * stale first, we will not attempt to lock them in the loop
2228 * below as the XFS_ISTALE flag will be set.
2232 if (lip
->li_type
== XFS_LI_INODE
) {
2233 iip
= (xfs_inode_log_item_t
*)lip
;
2234 ASSERT(iip
->ili_logged
== 1);
2235 lip
->li_cb
= xfs_istale_done
;
2236 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2237 &iip
->ili_flush_lsn
,
2238 &iip
->ili_item
.li_lsn
);
2239 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2241 lip
= lip
->li_bio_list
;
2246 * For each inode in memory attempt to add it to the inode
2247 * buffer and set it up for being staled on buffer IO
2248 * completion. This is safe as we've locked out tail pushing
2249 * and flushing by locking the buffer.
2251 * We have already marked every inode that was part of a
2252 * transaction stale above, which means there is no point in
2253 * even trying to lock them.
2255 for (i
= 0; i
< inodes_per_cluster
; i
++) {
2258 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2259 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2261 /* Inode not in memory, nothing to do */
2268 * because this is an RCU protected lookup, we could
2269 * find a recently freed or even reallocated inode
2270 * during the lookup. We need to check under the
2271 * i_flags_lock for a valid inode here. Skip it if it
2272 * is not valid, the wrong inode or stale.
2274 spin_lock(&ip
->i_flags_lock
);
2275 if (ip
->i_ino
!= inum
+ i
||
2276 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2277 spin_unlock(&ip
->i_flags_lock
);
2281 spin_unlock(&ip
->i_flags_lock
);
2284 * Don't try to lock/unlock the current inode, but we
2285 * _cannot_ skip the other inodes that we did not find
2286 * in the list attached to the buffer and are not
2287 * already marked stale. If we can't lock it, back off
2290 if (ip
!= free_ip
&&
2291 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2299 xfs_iflags_set(ip
, XFS_ISTALE
);
2302 * we don't need to attach clean inodes or those only
2303 * with unlogged changes (which we throw away, anyway).
2306 if (!iip
|| xfs_inode_clean(ip
)) {
2307 ASSERT(ip
!= free_ip
);
2309 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2313 iip
->ili_last_fields
= iip
->ili_fields
;
2314 iip
->ili_fields
= 0;
2315 iip
->ili_logged
= 1;
2316 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2317 &iip
->ili_item
.li_lsn
);
2319 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2323 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2326 xfs_trans_stale_inode_buf(tp
, bp
);
2327 xfs_trans_binval(tp
, bp
);
2335 * This is called to return an inode to the inode free list.
2336 * The inode should already be truncated to 0 length and have
2337 * no pages associated with it. This routine also assumes that
2338 * the inode is already a part of the transaction.
2340 * The on-disk copy of the inode will have been added to the list
2341 * of unlinked inodes in the AGI. We need to remove the inode from
2342 * that list atomically with respect to freeing it here.
2348 xfs_bmap_free_t
*flist
)
2351 struct xfs_icluster xic
= { 0 };
2353 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2354 ASSERT(ip
->i_d
.di_nlink
== 0);
2355 ASSERT(ip
->i_d
.di_nextents
== 0);
2356 ASSERT(ip
->i_d
.di_anextents
== 0);
2357 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
2358 ASSERT(ip
->i_d
.di_nblocks
== 0);
2361 * Pull the on-disk inode from the AGI unlinked list.
2363 error
= xfs_iunlink_remove(tp
, ip
);
2367 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &xic
);
2371 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2372 ip
->i_d
.di_flags
= 0;
2373 ip
->i_d
.di_dmevmask
= 0;
2374 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2375 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2376 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2378 * Bump the generation count so no one will be confused
2379 * by reincarnations of this inode.
2382 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2385 error
= xfs_ifree_cluster(ip
, tp
, &xic
);
2391 * This is called to unpin an inode. The caller must have the inode locked
2392 * in at least shared mode so that the buffer cannot be subsequently pinned
2393 * once someone is waiting for it to be unpinned.
2397 struct xfs_inode
*ip
)
2399 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2401 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2403 /* Give the log a push to start the unpinning I/O */
2404 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2410 struct xfs_inode
*ip
)
2412 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2413 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2418 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2419 if (xfs_ipincount(ip
))
2421 } while (xfs_ipincount(ip
));
2422 finish_wait(wq
, &wait
.wait
);
2427 struct xfs_inode
*ip
)
2429 if (xfs_ipincount(ip
))
2430 __xfs_iunpin_wait(ip
);
2434 * Removing an inode from the namespace involves removing the directory entry
2435 * and dropping the link count on the inode. Removing the directory entry can
2436 * result in locking an AGF (directory blocks were freed) and removing a link
2437 * count can result in placing the inode on an unlinked list which results in
2440 * The big problem here is that we have an ordering constraint on AGF and AGI
2441 * locking - inode allocation locks the AGI, then can allocate a new extent for
2442 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2443 * removes the inode from the unlinked list, requiring that we lock the AGI
2444 * first, and then freeing the inode can result in an inode chunk being freed
2445 * and hence freeing disk space requiring that we lock an AGF.
2447 * Hence the ordering that is imposed by other parts of the code is AGI before
2448 * AGF. This means we cannot remove the directory entry before we drop the inode
2449 * reference count and put it on the unlinked list as this results in a lock
2450 * order of AGF then AGI, and this can deadlock against inode allocation and
2451 * freeing. Therefore we must drop the link counts before we remove the
2454 * This is still safe from a transactional point of view - it is not until we
2455 * get to xfs_bmap_finish() that we have the possibility of multiple
2456 * transactions in this operation. Hence as long as we remove the directory
2457 * entry and drop the link count in the first transaction of the remove
2458 * operation, there are no transactional constraints on the ordering here.
2463 struct xfs_name
*name
,
2466 xfs_mount_t
*mp
= dp
->i_mount
;
2467 xfs_trans_t
*tp
= NULL
;
2468 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
2470 xfs_bmap_free_t free_list
;
2471 xfs_fsblock_t first_block
;
2475 trace_xfs_remove(dp
, name
);
2477 if (XFS_FORCED_SHUTDOWN(mp
))
2480 error
= xfs_qm_dqattach(dp
, 0);
2484 error
= xfs_qm_dqattach(ip
, 0);
2489 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2491 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2494 * We try to get the real space reservation first,
2495 * allowing for directory btree deletion(s) implying
2496 * possible bmap insert(s). If we can't get the space
2497 * reservation then we use 0 instead, and avoid the bmap
2498 * btree insert(s) in the directory code by, if the bmap
2499 * insert tries to happen, instead trimming the LAST
2500 * block from the directory.
2502 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2503 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, resblks
, 0);
2504 if (error
== -ENOSPC
) {
2506 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
2509 ASSERT(error
!= -ENOSPC
);
2510 goto out_trans_cancel
;
2513 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
2515 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2516 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2519 * If we're removing a directory perform some additional validation.
2522 ASSERT(ip
->i_d
.di_nlink
>= 2);
2523 if (ip
->i_d
.di_nlink
!= 2) {
2525 goto out_trans_cancel
;
2527 if (!xfs_dir_isempty(ip
)) {
2529 goto out_trans_cancel
;
2532 /* Drop the link from ip's "..". */
2533 error
= xfs_droplink(tp
, dp
);
2535 goto out_trans_cancel
;
2537 /* Drop the "." link from ip to self. */
2538 error
= xfs_droplink(tp
, ip
);
2540 goto out_trans_cancel
;
2543 * When removing a non-directory we need to log the parent
2544 * inode here. For a directory this is done implicitly
2545 * by the xfs_droplink call for the ".." entry.
2547 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2549 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2551 /* Drop the link from dp to ip. */
2552 error
= xfs_droplink(tp
, ip
);
2554 goto out_trans_cancel
;
2556 xfs_bmap_init(&free_list
, &first_block
);
2557 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
2558 &first_block
, &free_list
, resblks
);
2560 ASSERT(error
!= -ENOENT
);
2561 goto out_bmap_cancel
;
2565 * If this is a synchronous mount, make sure that the
2566 * remove transaction goes to disk before returning to
2569 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2570 xfs_trans_set_sync(tp
);
2572 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2574 goto out_bmap_cancel
;
2576 error
= xfs_trans_commit(tp
);
2580 if (is_dir
&& xfs_inode_is_filestream(ip
))
2581 xfs_filestream_deassociate(ip
);
2586 xfs_bmap_cancel(&free_list
);
2588 xfs_trans_cancel(tp
);
2594 * Enter all inodes for a rename transaction into a sorted array.
2596 #define __XFS_SORT_INODES 5
2598 xfs_sort_for_rename(
2599 struct xfs_inode
*dp1
, /* in: old (source) directory inode */
2600 struct xfs_inode
*dp2
, /* in: new (target) directory inode */
2601 struct xfs_inode
*ip1
, /* in: inode of old entry */
2602 struct xfs_inode
*ip2
, /* in: inode of new entry */
2603 struct xfs_inode
*wip
, /* in: whiteout inode */
2604 struct xfs_inode
**i_tab
,/* out: sorted array of inodes */
2605 int *num_inodes
) /* in/out: inodes in array */
2609 ASSERT(*num_inodes
== __XFS_SORT_INODES
);
2610 memset(i_tab
, 0, *num_inodes
* sizeof(struct xfs_inode
*));
2613 * i_tab contains a list of pointers to inodes. We initialize
2614 * the table here & we'll sort it. We will then use it to
2615 * order the acquisition of the inode locks.
2617 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2630 * Sort the elements via bubble sort. (Remember, there are at
2631 * most 5 elements to sort, so this is adequate.)
2633 for (i
= 0; i
< *num_inodes
; i
++) {
2634 for (j
= 1; j
< *num_inodes
; j
++) {
2635 if (i_tab
[j
]->i_ino
< i_tab
[j
-1]->i_ino
) {
2636 struct xfs_inode
*temp
= i_tab
[j
];
2637 i_tab
[j
] = i_tab
[j
-1];
2646 struct xfs_trans
*tp
,
2647 struct xfs_bmap_free
*free_list
)
2653 * If this is a synchronous mount, make sure that the rename transaction
2654 * goes to disk before returning to the user.
2656 if (tp
->t_mountp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
2657 xfs_trans_set_sync(tp
);
2659 error
= xfs_bmap_finish(&tp
, free_list
, &committed
);
2661 xfs_bmap_cancel(free_list
);
2662 xfs_trans_cancel(tp
);
2666 return xfs_trans_commit(tp
);
2670 * xfs_cross_rename()
2672 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2676 struct xfs_trans
*tp
,
2677 struct xfs_inode
*dp1
,
2678 struct xfs_name
*name1
,
2679 struct xfs_inode
*ip1
,
2680 struct xfs_inode
*dp2
,
2681 struct xfs_name
*name2
,
2682 struct xfs_inode
*ip2
,
2683 struct xfs_bmap_free
*free_list
,
2684 xfs_fsblock_t
*first_block
,
2692 /* Swap inode number for dirent in first parent */
2693 error
= xfs_dir_replace(tp
, dp1
, name1
,
2695 first_block
, free_list
, spaceres
);
2697 goto out_trans_abort
;
2699 /* Swap inode number for dirent in second parent */
2700 error
= xfs_dir_replace(tp
, dp2
, name2
,
2702 first_block
, free_list
, spaceres
);
2704 goto out_trans_abort
;
2707 * If we're renaming one or more directories across different parents,
2708 * update the respective ".." entries (and link counts) to match the new
2712 dp2_flags
= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2714 if (S_ISDIR(ip2
->i_d
.di_mode
)) {
2715 error
= xfs_dir_replace(tp
, ip2
, &xfs_name_dotdot
,
2716 dp1
->i_ino
, first_block
,
2717 free_list
, spaceres
);
2719 goto out_trans_abort
;
2721 /* transfer ip2 ".." reference to dp1 */
2722 if (!S_ISDIR(ip1
->i_d
.di_mode
)) {
2723 error
= xfs_droplink(tp
, dp2
);
2725 goto out_trans_abort
;
2726 error
= xfs_bumplink(tp
, dp1
);
2728 goto out_trans_abort
;
2732 * Although ip1 isn't changed here, userspace needs
2733 * to be warned about the change, so that applications
2734 * relying on it (like backup ones), will properly
2737 ip1_flags
|= XFS_ICHGTIME_CHG
;
2738 ip2_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2741 if (S_ISDIR(ip1
->i_d
.di_mode
)) {
2742 error
= xfs_dir_replace(tp
, ip1
, &xfs_name_dotdot
,
2743 dp2
->i_ino
, first_block
,
2744 free_list
, spaceres
);
2746 goto out_trans_abort
;
2748 /* transfer ip1 ".." reference to dp2 */
2749 if (!S_ISDIR(ip2
->i_d
.di_mode
)) {
2750 error
= xfs_droplink(tp
, dp1
);
2752 goto out_trans_abort
;
2753 error
= xfs_bumplink(tp
, dp2
);
2755 goto out_trans_abort
;
2759 * Although ip2 isn't changed here, userspace needs
2760 * to be warned about the change, so that applications
2761 * relying on it (like backup ones), will properly
2764 ip1_flags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
2765 ip2_flags
|= XFS_ICHGTIME_CHG
;
2770 xfs_trans_ichgtime(tp
, ip1
, ip1_flags
);
2771 xfs_trans_log_inode(tp
, ip1
, XFS_ILOG_CORE
);
2774 xfs_trans_ichgtime(tp
, ip2
, ip2_flags
);
2775 xfs_trans_log_inode(tp
, ip2
, XFS_ILOG_CORE
);
2778 xfs_trans_ichgtime(tp
, dp2
, dp2_flags
);
2779 xfs_trans_log_inode(tp
, dp2
, XFS_ILOG_CORE
);
2781 xfs_trans_ichgtime(tp
, dp1
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2782 xfs_trans_log_inode(tp
, dp1
, XFS_ILOG_CORE
);
2783 return xfs_finish_rename(tp
, free_list
);
2786 xfs_bmap_cancel(free_list
);
2787 xfs_trans_cancel(tp
);
2792 * xfs_rename_alloc_whiteout()
2794 * Return a referenced, unlinked, unlocked inode that that can be used as a
2795 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2796 * crash between allocating the inode and linking it into the rename transaction
2797 * recovery will free the inode and we won't leak it.
2800 xfs_rename_alloc_whiteout(
2801 struct xfs_inode
*dp
,
2802 struct xfs_inode
**wip
)
2804 struct xfs_inode
*tmpfile
;
2807 error
= xfs_create_tmpfile(dp
, NULL
, S_IFCHR
| WHITEOUT_MODE
, &tmpfile
);
2812 * Prepare the tmpfile inode as if it were created through the VFS.
2813 * Otherwise, the link increment paths will complain about nlink 0->1.
2814 * Drop the link count as done by d_tmpfile(), complete the inode setup
2815 * and flag it as linkable.
2817 drop_nlink(VFS_I(tmpfile
));
2818 xfs_finish_inode_setup(tmpfile
);
2819 VFS_I(tmpfile
)->i_state
|= I_LINKABLE
;
2830 struct xfs_inode
*src_dp
,
2831 struct xfs_name
*src_name
,
2832 struct xfs_inode
*src_ip
,
2833 struct xfs_inode
*target_dp
,
2834 struct xfs_name
*target_name
,
2835 struct xfs_inode
*target_ip
,
2838 struct xfs_mount
*mp
= src_dp
->i_mount
;
2839 struct xfs_trans
*tp
;
2840 struct xfs_bmap_free free_list
;
2841 xfs_fsblock_t first_block
;
2842 struct xfs_inode
*wip
= NULL
; /* whiteout inode */
2843 struct xfs_inode
*inodes
[__XFS_SORT_INODES
];
2844 int num_inodes
= __XFS_SORT_INODES
;
2845 bool new_parent
= (src_dp
!= target_dp
);
2846 bool src_is_directory
= S_ISDIR(src_ip
->i_d
.di_mode
);
2850 trace_xfs_rename(src_dp
, target_dp
, src_name
, target_name
);
2852 if ((flags
& RENAME_EXCHANGE
) && !target_ip
)
2856 * If we are doing a whiteout operation, allocate the whiteout inode
2857 * we will be placing at the target and ensure the type is set
2860 if (flags
& RENAME_WHITEOUT
) {
2861 ASSERT(!(flags
& (RENAME_NOREPLACE
| RENAME_EXCHANGE
)));
2862 error
= xfs_rename_alloc_whiteout(target_dp
, &wip
);
2866 /* setup target dirent info as whiteout */
2867 src_name
->type
= XFS_DIR3_FT_CHRDEV
;
2870 xfs_sort_for_rename(src_dp
, target_dp
, src_ip
, target_ip
, wip
,
2871 inodes
, &num_inodes
);
2873 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RENAME
);
2874 spaceres
= XFS_RENAME_SPACE_RES(mp
, target_name
->len
);
2875 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, spaceres
, 0);
2876 if (error
== -ENOSPC
) {
2878 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
, 0, 0);
2881 goto out_trans_cancel
;
2884 * Attach the dquots to the inodes
2886 error
= xfs_qm_vop_rename_dqattach(inodes
);
2888 goto out_trans_cancel
;
2891 * Lock all the participating inodes. Depending upon whether
2892 * the target_name exists in the target directory, and
2893 * whether the target directory is the same as the source
2894 * directory, we can lock from 2 to 4 inodes.
2896 xfs_lock_inodes(inodes
, num_inodes
, XFS_ILOCK_EXCL
);
2899 * Join all the inodes to the transaction. From this point on,
2900 * we can rely on either trans_commit or trans_cancel to unlock
2903 xfs_trans_ijoin(tp
, src_dp
, XFS_ILOCK_EXCL
);
2905 xfs_trans_ijoin(tp
, target_dp
, XFS_ILOCK_EXCL
);
2906 xfs_trans_ijoin(tp
, src_ip
, XFS_ILOCK_EXCL
);
2908 xfs_trans_ijoin(tp
, target_ip
, XFS_ILOCK_EXCL
);
2910 xfs_trans_ijoin(tp
, wip
, XFS_ILOCK_EXCL
);
2913 * If we are using project inheritance, we only allow renames
2914 * into our tree when the project IDs are the same; else the
2915 * tree quota mechanism would be circumvented.
2917 if (unlikely((target_dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2918 (xfs_get_projid(target_dp
) != xfs_get_projid(src_ip
)))) {
2920 goto out_trans_cancel
;
2923 xfs_bmap_init(&free_list
, &first_block
);
2925 /* RENAME_EXCHANGE is unique from here on. */
2926 if (flags
& RENAME_EXCHANGE
)
2927 return xfs_cross_rename(tp
, src_dp
, src_name
, src_ip
,
2928 target_dp
, target_name
, target_ip
,
2929 &free_list
, &first_block
, spaceres
);
2932 * Set up the target.
2934 if (target_ip
== NULL
) {
2936 * If there's no space reservation, check the entry will
2937 * fit before actually inserting it.
2940 error
= xfs_dir_canenter(tp
, target_dp
, target_name
);
2942 goto out_trans_cancel
;
2945 * If target does not exist and the rename crosses
2946 * directories, adjust the target directory link count
2947 * to account for the ".." reference from the new entry.
2949 error
= xfs_dir_createname(tp
, target_dp
, target_name
,
2950 src_ip
->i_ino
, &first_block
,
2951 &free_list
, spaceres
);
2953 goto out_bmap_cancel
;
2955 xfs_trans_ichgtime(tp
, target_dp
,
2956 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2958 if (new_parent
&& src_is_directory
) {
2959 error
= xfs_bumplink(tp
, target_dp
);
2961 goto out_bmap_cancel
;
2963 } else { /* target_ip != NULL */
2965 * If target exists and it's a directory, check that both
2966 * target and source are directories and that target can be
2967 * destroyed, or that neither is a directory.
2969 if (S_ISDIR(target_ip
->i_d
.di_mode
)) {
2971 * Make sure target dir is empty.
2973 if (!(xfs_dir_isempty(target_ip
)) ||
2974 (target_ip
->i_d
.di_nlink
> 2)) {
2976 goto out_trans_cancel
;
2981 * Link the source inode under the target name.
2982 * If the source inode is a directory and we are moving
2983 * it across directories, its ".." entry will be
2984 * inconsistent until we replace that down below.
2986 * In case there is already an entry with the same
2987 * name at the destination directory, remove it first.
2989 error
= xfs_dir_replace(tp
, target_dp
, target_name
,
2991 &first_block
, &free_list
, spaceres
);
2993 goto out_bmap_cancel
;
2995 xfs_trans_ichgtime(tp
, target_dp
,
2996 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2999 * Decrement the link count on the target since the target
3000 * dir no longer points to it.
3002 error
= xfs_droplink(tp
, target_ip
);
3004 goto out_bmap_cancel
;
3006 if (src_is_directory
) {
3008 * Drop the link from the old "." entry.
3010 error
= xfs_droplink(tp
, target_ip
);
3012 goto out_bmap_cancel
;
3014 } /* target_ip != NULL */
3017 * Remove the source.
3019 if (new_parent
&& src_is_directory
) {
3021 * Rewrite the ".." entry to point to the new
3024 error
= xfs_dir_replace(tp
, src_ip
, &xfs_name_dotdot
,
3026 &first_block
, &free_list
, spaceres
);
3027 ASSERT(error
!= -EEXIST
);
3029 goto out_bmap_cancel
;
3033 * We always want to hit the ctime on the source inode.
3035 * This isn't strictly required by the standards since the source
3036 * inode isn't really being changed, but old unix file systems did
3037 * it and some incremental backup programs won't work without it.
3039 xfs_trans_ichgtime(tp
, src_ip
, XFS_ICHGTIME_CHG
);
3040 xfs_trans_log_inode(tp
, src_ip
, XFS_ILOG_CORE
);
3043 * Adjust the link count on src_dp. This is necessary when
3044 * renaming a directory, either within one parent when
3045 * the target existed, or across two parent directories.
3047 if (src_is_directory
&& (new_parent
|| target_ip
!= NULL
)) {
3050 * Decrement link count on src_directory since the
3051 * entry that's moved no longer points to it.
3053 error
= xfs_droplink(tp
, src_dp
);
3055 goto out_bmap_cancel
;
3059 * For whiteouts, we only need to update the source dirent with the
3060 * inode number of the whiteout inode rather than removing it
3064 error
= xfs_dir_replace(tp
, src_dp
, src_name
, wip
->i_ino
,
3065 &first_block
, &free_list
, spaceres
);
3067 error
= xfs_dir_removename(tp
, src_dp
, src_name
, src_ip
->i_ino
,
3068 &first_block
, &free_list
, spaceres
);
3070 goto out_bmap_cancel
;
3073 * For whiteouts, we need to bump the link count on the whiteout inode.
3074 * This means that failures all the way up to this point leave the inode
3075 * on the unlinked list and so cleanup is a simple matter of dropping
3076 * the remaining reference to it. If we fail here after bumping the link
3077 * count, we're shutting down the filesystem so we'll never see the
3078 * intermediate state on disk.
3081 ASSERT(VFS_I(wip
)->i_nlink
== 0 && wip
->i_d
.di_nlink
== 0);
3082 error
= xfs_bumplink(tp
, wip
);
3084 goto out_bmap_cancel
;
3085 error
= xfs_iunlink_remove(tp
, wip
);
3087 goto out_bmap_cancel
;
3088 xfs_trans_log_inode(tp
, wip
, XFS_ILOG_CORE
);
3091 * Now we have a real link, clear the "I'm a tmpfile" state
3092 * flag from the inode so it doesn't accidentally get misused in
3095 VFS_I(wip
)->i_state
&= ~I_LINKABLE
;
3098 xfs_trans_ichgtime(tp
, src_dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3099 xfs_trans_log_inode(tp
, src_dp
, XFS_ILOG_CORE
);
3101 xfs_trans_log_inode(tp
, target_dp
, XFS_ILOG_CORE
);
3103 error
= xfs_finish_rename(tp
, &free_list
);
3109 xfs_bmap_cancel(&free_list
);
3111 xfs_trans_cancel(tp
);
3122 xfs_mount_t
*mp
= ip
->i_mount
;
3123 struct xfs_perag
*pag
;
3124 unsigned long first_index
, mask
;
3125 unsigned long inodes_per_cluster
;
3127 xfs_inode_t
**ilist
;
3134 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
3136 inodes_per_cluster
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
3137 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
3138 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
3142 mask
= ~(((mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
)) - 1);
3143 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
3145 /* really need a gang lookup range call here */
3146 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
3147 first_index
, inodes_per_cluster
);
3151 for (i
= 0; i
< nr_found
; i
++) {
3157 * because this is an RCU protected lookup, we could find a
3158 * recently freed or even reallocated inode during the lookup.
3159 * We need to check under the i_flags_lock for a valid inode
3160 * here. Skip it if it is not valid or the wrong inode.
3162 spin_lock(&ip
->i_flags_lock
);
3164 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
3165 spin_unlock(&ip
->i_flags_lock
);
3168 spin_unlock(&ip
->i_flags_lock
);
3171 * Do an un-protected check to see if the inode is dirty and
3172 * is a candidate for flushing. These checks will be repeated
3173 * later after the appropriate locks are acquired.
3175 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
3179 * Try to get locks. If any are unavailable or it is pinned,
3180 * then this inode cannot be flushed and is skipped.
3183 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
3185 if (!xfs_iflock_nowait(iq
)) {
3186 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3189 if (xfs_ipincount(iq
)) {
3191 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3196 * arriving here means that this inode can be flushed. First
3197 * re-check that it's dirty before flushing.
3199 if (!xfs_inode_clean(iq
)) {
3201 error
= xfs_iflush_int(iq
, bp
);
3203 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3204 goto cluster_corrupt_out
;
3210 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3214 XFS_STATS_INC(xs_icluster_flushcnt
);
3215 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3226 cluster_corrupt_out
:
3228 * Corruption detected in the clustering loop. Invalidate the
3229 * inode buffer and shut down the filesystem.
3233 * Clean up the buffer. If it was delwri, just release it --
3234 * brelse can handle it with no problems. If not, shut down the
3235 * filesystem before releasing the buffer.
3237 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
3241 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3243 if (!bufwasdelwri
) {
3245 * Just like incore_relse: if we have b_iodone functions,
3246 * mark the buffer as an error and call them. Otherwise
3247 * mark it as stale and brelse.
3252 xfs_buf_ioerror(bp
, -EIO
);
3261 * Unlocks the flush lock
3263 xfs_iflush_abort(iq
, false);
3266 return -EFSCORRUPTED
;
3270 * Flush dirty inode metadata into the backing buffer.
3272 * The caller must have the inode lock and the inode flush lock held. The
3273 * inode lock will still be held upon return to the caller, and the inode
3274 * flush lock will be released after the inode has reached the disk.
3276 * The caller must write out the buffer returned in *bpp and release it.
3280 struct xfs_inode
*ip
,
3281 struct xfs_buf
**bpp
)
3283 struct xfs_mount
*mp
= ip
->i_mount
;
3285 struct xfs_dinode
*dip
;
3288 XFS_STATS_INC(xs_iflush_count
);
3290 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3291 ASSERT(xfs_isiflocked(ip
));
3292 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3293 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3297 xfs_iunpin_wait(ip
);
3300 * For stale inodes we cannot rely on the backing buffer remaining
3301 * stale in cache for the remaining life of the stale inode and so
3302 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3303 * inodes below. We have to check this after ensuring the inode is
3304 * unpinned so that it is safe to reclaim the stale inode after the
3307 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
3313 * This may have been unpinned because the filesystem is shutting
3314 * down forcibly. If that's the case we must not write this inode
3315 * to disk, because the log record didn't make it to disk.
3317 * We also have to remove the log item from the AIL in this case,
3318 * as we wait for an empty AIL as part of the unmount process.
3320 if (XFS_FORCED_SHUTDOWN(mp
)) {
3326 * Get the buffer containing the on-disk inode.
3328 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
3336 * First flush out the inode that xfs_iflush was called with.
3338 error
= xfs_iflush_int(ip
, bp
);
3343 * If the buffer is pinned then push on the log now so we won't
3344 * get stuck waiting in the write for too long.
3346 if (xfs_buf_ispinned(bp
))
3347 xfs_log_force(mp
, 0);
3351 * see if other inodes can be gathered into this write
3353 error
= xfs_iflush_cluster(ip
, bp
);
3355 goto cluster_corrupt_out
;
3362 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3363 cluster_corrupt_out
:
3364 error
= -EFSCORRUPTED
;
3367 * Unlocks the flush lock
3369 xfs_iflush_abort(ip
, false);
3375 struct xfs_inode
*ip
,
3378 struct xfs_inode_log_item
*iip
= ip
->i_itemp
;
3379 struct xfs_dinode
*dip
;
3380 struct xfs_mount
*mp
= ip
->i_mount
;
3382 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3383 ASSERT(xfs_isiflocked(ip
));
3384 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3385 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
3386 ASSERT(iip
!= NULL
&& iip
->ili_fields
!= 0);
3387 ASSERT(ip
->i_d
.di_version
> 1);
3389 /* set *dip = inode's place in the buffer */
3390 dip
= xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3392 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
3393 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3394 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3395 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3396 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3399 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3400 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3401 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3402 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3403 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3406 if (S_ISREG(ip
->i_d
.di_mode
)) {
3408 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3409 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3410 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3411 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3412 "%s: Bad regular inode %Lu, ptr 0x%p",
3413 __func__
, ip
->i_ino
, ip
);
3416 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
3418 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3419 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3420 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3421 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3422 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3423 "%s: Bad directory inode %Lu, ptr 0x%p",
3424 __func__
, ip
->i_ino
, ip
);
3428 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3429 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3430 XFS_RANDOM_IFLUSH_5
)) {
3431 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3432 "%s: detected corrupt incore inode %Lu, "
3433 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3434 __func__
, ip
->i_ino
,
3435 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3436 ip
->i_d
.di_nblocks
, ip
);
3439 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3440 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3441 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
3442 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3443 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3448 * Inode item log recovery for v2 inodes are dependent on the
3449 * di_flushiter count for correct sequencing. We bump the flush
3450 * iteration count so we can detect flushes which postdate a log record
3451 * during recovery. This is redundant as we now log every change and
3452 * hence this can't happen but we need to still do it to ensure
3453 * backwards compatibility with old kernels that predate logging all
3456 if (ip
->i_d
.di_version
< 3)
3457 ip
->i_d
.di_flushiter
++;
3460 * Copy the dirty parts of the inode into the on-disk
3461 * inode. We always copy out the core of the inode,
3462 * because if the inode is dirty at all the core must
3465 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3467 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3468 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3469 ip
->i_d
.di_flushiter
= 0;
3471 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
);
3472 if (XFS_IFORK_Q(ip
))
3473 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
);
3474 xfs_inobp_check(mp
, bp
);
3477 * We've recorded everything logged in the inode, so we'd like to clear
3478 * the ili_fields bits so we don't log and flush things unnecessarily.
3479 * However, we can't stop logging all this information until the data
3480 * we've copied into the disk buffer is written to disk. If we did we
3481 * might overwrite the copy of the inode in the log with all the data
3482 * after re-logging only part of it, and in the face of a crash we
3483 * wouldn't have all the data we need to recover.
3485 * What we do is move the bits to the ili_last_fields field. When
3486 * logging the inode, these bits are moved back to the ili_fields field.
3487 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3488 * know that the information those bits represent is permanently on
3489 * disk. As long as the flush completes before the inode is logged
3490 * again, then both ili_fields and ili_last_fields will be cleared.
3492 * We can play with the ili_fields bits here, because the inode lock
3493 * must be held exclusively in order to set bits there and the flush
3494 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3495 * done routine can tell whether or not to look in the AIL. Also, store
3496 * the current LSN of the inode so that we can tell whether the item has
3497 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3498 * need the AIL lock, because it is a 64 bit value that cannot be read
3501 iip
->ili_last_fields
= iip
->ili_fields
;
3502 iip
->ili_fields
= 0;
3503 iip
->ili_logged
= 1;
3505 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3506 &iip
->ili_item
.li_lsn
);
3509 * Attach the function xfs_iflush_done to the inode's
3510 * buffer. This will remove the inode from the AIL
3511 * and unlock the inode's flush lock when the inode is
3512 * completely written to disk.
3514 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3516 /* update the lsn in the on disk inode if required */
3517 if (ip
->i_d
.di_version
== 3)
3518 dip
->di_lsn
= cpu_to_be64(iip
->ili_item
.li_lsn
);
3520 /* generate the checksum. */
3521 xfs_dinode_calc_crc(mp
, dip
);
3523 ASSERT(bp
->b_fspriv
!= NULL
);
3524 ASSERT(bp
->b_iodone
!= NULL
);
3528 return -EFSCORRUPTED
;