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.
86 xfs_get_cowextsz_hint(
92 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
93 a = ip->i_d.di_cowextsize;
94 b = xfs_get_extsz_hint(ip);
102 * These two are wrapper routines around the xfs_ilock() routine used to
103 * centralize some grungy code. They are used in places that wish to lock the
104 * inode solely for reading the extents. The reason these places can't just
105 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
106 * bringing in of the extents from disk for a file in b-tree format. If the
107 * inode is in b-tree format, then we need to lock the inode exclusively until
108 * the extents are read in. Locking it exclusively all the time would limit
109 * our parallelism unnecessarily, though. What we do instead is check to see
110 * if the extents have been read in yet, and only lock the inode exclusively
113 * The functions return a value which should be given to the corresponding
114 * xfs_iunlock() call.
117 xfs_ilock_data_map_shared(
118 struct xfs_inode *ip)
120 uint lock_mode = XFS_ILOCK_SHARED;
122 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
123 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
124 lock_mode = XFS_ILOCK_EXCL;
125 xfs_ilock(ip, lock_mode);
130 xfs_ilock_attr_map_shared(
131 struct xfs_inode *ip)
133 uint lock_mode = XFS_ILOCK_SHARED;
135 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
136 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
137 lock_mode = XFS_ILOCK_EXCL;
138 xfs_ilock(ip, lock_mode);
143 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
144 * the i_lock. This routine allows various combinations of the locks to be
147 * The 3 locks should always be ordered so that the IO lock is obtained first,
148 * the mmap lock second and the ilock last in order to prevent deadlock.
150 * Basic locking order:
152 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
154 * mmap_sem locking order:
156 * i_iolock -> page lock -> mmap_sem
157 * mmap_sem -> i_mmap_lock -> page_lock
159 * The difference in mmap_sem locking order mean that we cannot hold the
160 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
161 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
162 * in get_user_pages() to map the user pages into the kernel address space for
163 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
164 * page faults already hold the mmap_sem.
166 * Hence to serialise fully against both syscall and mmap based IO, we need to
167 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
168 * taken in places where we need to invalidate the page cache in a race
169 * free manner (e.g. truncate, hole punch and other extent manipulation
177 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
180 * You can't set both SHARED and EXCL for the same lock,
181 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
182 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
184 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
185 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
186 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
187 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
188 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
189 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
190 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
192 if (lock_flags & XFS_IOLOCK_EXCL)
193 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
194 else if (lock_flags & XFS_IOLOCK_SHARED)
195 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
197 if (lock_flags & XFS_MMAPLOCK_EXCL)
198 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
199 else if (lock_flags & XFS_MMAPLOCK_SHARED)
200 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
202 if (lock_flags & XFS_ILOCK_EXCL)
203 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
204 else if (lock_flags & XFS_ILOCK_SHARED)
205 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
209 * This is just like xfs_ilock(), except that the caller
210 * is guaranteed not to sleep. It returns 1 if it gets
211 * the requested locks and 0 otherwise. If the IO lock is
212 * obtained but the inode lock cannot be, then the IO lock
213 * is dropped before returning.
215 * ip -- the inode being locked
216 * lock_flags -- this parameter indicates the inode's locks to be
217 * to be locked. See the comment for xfs_ilock() for a list
225 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
228 * You can't set both SHARED and EXCL for the same lock,
229 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
230 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
232 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
233 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
234 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
235 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
236 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
237 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
238 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
240 if (lock_flags & XFS_IOLOCK_EXCL) {
241 if (!mrtryupdate(&ip->i_iolock))
243 } else if (lock_flags & XFS_IOLOCK_SHARED) {
244 if (!mrtryaccess(&ip->i_iolock))
248 if (lock_flags & XFS_MMAPLOCK_EXCL) {
249 if (!mrtryupdate(&ip->i_mmaplock))
250 goto out_undo_iolock;
251 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
252 if (!mrtryaccess(&ip->i_mmaplock))
253 goto out_undo_iolock;
256 if (lock_flags & XFS_ILOCK_EXCL) {
257 if (!mrtryupdate(&ip->i_lock))
258 goto out_undo_mmaplock;
259 } else if (lock_flags & XFS_ILOCK_SHARED) {
260 if (!mrtryaccess(&ip->i_lock))
261 goto out_undo_mmaplock;
266 if (lock_flags & XFS_MMAPLOCK_EXCL)
267 mrunlock_excl(&ip->i_mmaplock);
268 else if (lock_flags & XFS_MMAPLOCK_SHARED)
269 mrunlock_shared(&ip->i_mmaplock);
271 if (lock_flags & XFS_IOLOCK_EXCL)
272 mrunlock_excl(&ip->i_iolock);
273 else if (lock_flags & XFS_IOLOCK_SHARED)
274 mrunlock_shared(&ip->i_iolock);
280 * xfs_iunlock() is used to drop the inode locks acquired with
281 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
282 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
283 * that we know which locks to drop.
285 * ip -- the inode being unlocked
286 * lock_flags -- this parameter indicates the inode's locks to be
287 * to be unlocked. See the comment for xfs_ilock() for a list
288 * of valid values for this parameter.
297 * You can't set both SHARED and EXCL for the same lock,
298 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
299 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
301 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
302 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
303 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
304 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
305 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
306 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
307 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
308 ASSERT(lock_flags != 0);
310 if (lock_flags & XFS_IOLOCK_EXCL)
311 mrunlock_excl(&ip->i_iolock);
312 else if (lock_flags & XFS_IOLOCK_SHARED)
313 mrunlock_shared(&ip->i_iolock);
315 if (lock_flags & XFS_MMAPLOCK_EXCL)
316 mrunlock_excl(&ip->i_mmaplock);
317 else if (lock_flags & XFS_MMAPLOCK_SHARED)
318 mrunlock_shared(&ip->i_mmaplock);
320 if (lock_flags & XFS_ILOCK_EXCL)
321 mrunlock_excl(&ip->i_lock);
322 else if (lock_flags & XFS_ILOCK_SHARED)
323 mrunlock_shared(&ip->i_lock);
325 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
329 * give up write locks. the i/o lock cannot be held nested
330 * if it is being demoted.
337 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
339 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
341 if (lock_flags & XFS_ILOCK_EXCL)
342 mrdemote(&ip->i_lock);
343 if (lock_flags & XFS_MMAPLOCK_EXCL)
344 mrdemote(&ip->i_mmaplock);
345 if (lock_flags & XFS_IOLOCK_EXCL)
346 mrdemote(&ip->i_iolock);
348 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
351 #if defined(DEBUG) || defined(XFS_WARN)
357 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
358 if (!(lock_flags & XFS_ILOCK_SHARED))
359 return !!ip->i_lock.mr_writer;
360 return rwsem_is_locked(&ip->i_lock.mr_lock);
363 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
364 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
365 return !!ip->i_mmaplock.mr_writer;
366 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
369 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
370 if (!(lock_flags & XFS_IOLOCK_SHARED))
371 return !!ip->i_iolock.mr_writer;
372 return rwsem_is_locked(&ip->i_iolock.mr_lock);
382 int xfs_small_retries;
383 int xfs_middle_retries;
384 int xfs_lots_retries;
389 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
390 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
391 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
392 * errors and warnings.
394 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
396 xfs_lockdep_subclass_ok(
399 return subclass < MAX_LOCKDEP_SUBCLASSES;
402 #define xfs_lockdep_subclass_ok(subclass) (true)
406 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
407 * value. This can be called for any type of inode lock combination, including
408 * parent locking. Care must be taken to ensure we don't overrun the subclass
409 * storage fields in the class mask we build.
412 xfs_lock_inumorder(int lock_mode, int subclass)
416 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
418 ASSERT(xfs_lockdep_subclass_ok(subclass));
420 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
421 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
422 ASSERT(xfs_lockdep_subclass_ok(subclass +
423 XFS_IOLOCK_PARENT_VAL));
424 class += subclass << XFS_IOLOCK_SHIFT;
425 if (lock_mode & XFS_IOLOCK_PARENT)
426 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
429 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
430 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
431 class += subclass << XFS_MMAPLOCK_SHIFT;
434 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
435 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
436 class += subclass << XFS_ILOCK_SHIFT;
439 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
443 * The following routine will lock n inodes in exclusive mode. We assume the
444 * caller calls us with the inodes in i_ino order.
446 * We need to detect deadlock where an inode that we lock is in the AIL and we
447 * start waiting for another inode that is locked by a thread in a long running
448 * transaction (such as truncate). This can result in deadlock since the long
449 * running trans might need to wait for the inode we just locked in order to
450 * push the tail and free space in the log.
452 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
453 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
454 * lock more than one at a time, lockdep will report false positives saying we
455 * have violated locking orders.
463 int attempts = 0, i, j, try_lock;
467 * Currently supports between 2 and 5 inodes with exclusive locking. We
468 * support an arbitrary depth of locking here, but absolute limits on
469 * inodes depend on the the type of locking and the limits placed by
470 * lockdep annotations in xfs_lock_inumorder. These are all checked by
473 ASSERT(ips && inodes >= 2 && inodes <= 5);
474 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
476 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
478 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
479 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
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 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
583 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
584 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
585 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
586 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
588 ASSERT(ip0->i_ino != ip1->i_ino);
590 if (ip0->i_ino > ip1->i_ino) {
597 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
600 * If the first lock we have locked is in the AIL, we must TRY to get
601 * the second lock. If we can't get it, we must release the first one
604 lp = (xfs_log_item_t *)ip0->i_itemp;
605 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
606 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
607 xfs_iunlock(ip0, lock_mode);
608 if ((++attempts % 5) == 0)
609 delay(1); /* Don't just spin the CPU */
613 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
620 struct xfs_inode *ip)
622 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
623 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
626 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
627 if (xfs_isiflocked(ip))
629 } while (!xfs_iflock_nowait(ip));
631 finish_wait(wq, &wait.wait);
642 if (di_flags & XFS_DIFLAG_ANY) {
643 if (di_flags & XFS_DIFLAG_REALTIME)
644 flags |= FS_XFLAG_REALTIME;
645 if (di_flags & XFS_DIFLAG_PREALLOC)
646 flags |= FS_XFLAG_PREALLOC;
647 if (di_flags & XFS_DIFLAG_IMMUTABLE)
648 flags |= FS_XFLAG_IMMUTABLE;
649 if (di_flags & XFS_DIFLAG_APPEND)
650 flags |= FS_XFLAG_APPEND;
651 if (di_flags & XFS_DIFLAG_SYNC)
652 flags |= FS_XFLAG_SYNC;
653 if (di_flags & XFS_DIFLAG_NOATIME)
654 flags |= FS_XFLAG_NOATIME;
655 if (di_flags & XFS_DIFLAG_NODUMP)
656 flags |= FS_XFLAG_NODUMP;
657 if (di_flags & XFS_DIFLAG_RTINHERIT)
658 flags |= FS_XFLAG_RTINHERIT;
659 if (di_flags & XFS_DIFLAG_PROJINHERIT)
660 flags |= FS_XFLAG_PROJINHERIT;
661 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
662 flags |= FS_XFLAG_NOSYMLINKS;
663 if (di_flags & XFS_DIFLAG_EXTSIZE)
664 flags |= FS_XFLAG_EXTSIZE;
665 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
666 flags |= FS_XFLAG_EXTSZINHERIT;
667 if (di_flags & XFS_DIFLAG_NODEFRAG)
668 flags |= FS_XFLAG_NODEFRAG;
669 if (di_flags & XFS_DIFLAG_FILESTREAM)
670 flags |= FS_XFLAG_FILESTREAM;
673 if (di_flags2 & XFS_DIFLAG2_ANY) {
674 if (di_flags2 & XFS_DIFLAG2_DAX)
675 flags |= FS_XFLAG_DAX;
676 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
677 flags |= FS_XFLAG_COWEXTSIZE;
681 flags |= FS_XFLAG_HASATTR;
688 struct xfs_inode *ip)
690 struct xfs_icdinode *dic = &ip->i_d;
692 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
696 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
697 * is allowed, otherwise it has to be an exact match. If a CI match is found,
698 * ci_name->name will point to a the actual name (caller must free) or
699 * will be set to NULL if an exact match is found.
704 struct xfs_name *name,
706 struct xfs_name *ci_name)
711 trace_xfs_lookup(dp, name);
713 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
716 xfs_ilock(dp, XFS_IOLOCK_SHARED);
717 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
721 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
725 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
730 kmem_free(ci_name->name);
732 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
738 * Allocate an inode on disk and return a copy of its in-core version.
739 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
740 * appropriately within the inode. The uid and gid for the inode are
741 * set according to the contents of the given cred structure.
743 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
744 * has a free inode available, call xfs_iget() to obtain the in-core
745 * version of the allocated inode. Finally, fill in the inode and
746 * log its initial contents. In this case, ialloc_context would be
749 * If xfs_dialloc() does not have an available inode, it will replenish
750 * its supply by doing an allocation. Since we can only do one
751 * allocation within a transaction without deadlocks, we must commit
752 * the current transaction before returning the inode itself.
753 * In this case, therefore, we will set ialloc_context and return.
754 * The caller should then commit the current transaction, start a new
755 * transaction, and call xfs_ialloc() again to actually get the inode.
757 * To ensure that some other process does not grab the inode that
758 * was allocated during the first call to xfs_ialloc(), this routine
759 * also returns the [locked] bp pointing to the head of the freelist
760 * as ialloc_context. The caller should hold this buffer across
761 * the commit and pass it back into this routine on the second call.
763 * If we are allocating quota inodes, we do not have a parent inode
764 * to attach to or associate with (i.e. pip == NULL) because they
765 * are not linked into the directory structure - they are attached
766 * directly to the superblock - and so have no parent.
777 xfs_buf_t **ialloc_context,
780 struct xfs_mount *mp = tp->t_mountp;
789 * Call the space management code to pick
790 * the on-disk inode to be allocated.
792 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
793 ialloc_context, &ino);
796 if (*ialloc_context || ino == NULLFSINO) {
800 ASSERT(*ialloc_context == NULL);
803 * Get the in-core inode with the lock held exclusively.
804 * This is because we're setting fields here we need
805 * to prevent others from looking at until we're done.
807 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
808 XFS_ILOCK_EXCL, &ip);
815 * We always convert v1 inodes to v2 now - we only support filesystems
816 * with >= v2 inode capability, so there is no reason for ever leaving
817 * an inode in v1 format.
819 if (ip->i_d.di_version == 1)
820 ip->i_d.di_version = 2;
822 inode->i_mode = mode;
823 set_nlink(inode, nlink);
824 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
825 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
826 xfs_set_projid(ip, prid);
828 if (pip && XFS_INHERIT_GID(pip)) {
829 ip->i_d.di_gid = pip->i_d.di_gid;
830 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
831 inode->i_mode |= S_ISGID;
835 * If the group ID of the new file does not match the effective group
836 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
837 * (and only if the irix_sgid_inherit compatibility variable is set).
839 if ((irix_sgid_inherit) &&
840 (inode->i_mode & S_ISGID) &&
841 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
842 inode->i_mode &= ~S_ISGID;
845 ip->i_d.di_nextents = 0;
846 ASSERT(ip->i_d.di_nblocks == 0);
848 tv = current_fs_time(mp->m_super);
853 ip->i_d.di_extsize = 0;
854 ip->i_d.di_dmevmask = 0;
855 ip->i_d.di_dmstate = 0;
856 ip->i_d.di_flags = 0;
858 if (ip->i_d.di_version == 3) {
859 inode->i_version = 1;
860 ip->i_d.di_flags2 = 0;
861 ip->i_d.di_cowextsize = 0;
862 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
863 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
867 flags = XFS_ILOG_CORE;
868 switch (mode & S_IFMT) {
873 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
874 ip->i_df.if_u2.if_rdev = rdev;
875 ip->i_df.if_flags = 0;
876 flags |= XFS_ILOG_DEV;
880 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
881 uint64_t di_flags2 = 0;
885 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
886 di_flags |= XFS_DIFLAG_RTINHERIT;
887 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
888 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
889 ip->i_d.di_extsize = pip->i_d.di_extsize;
891 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
892 di_flags |= XFS_DIFLAG_PROJINHERIT;
893 } else if (S_ISREG(mode)) {
894 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
895 di_flags |= XFS_DIFLAG_REALTIME;
896 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
897 di_flags |= XFS_DIFLAG_EXTSIZE;
898 ip->i_d.di_extsize = pip->i_d.di_extsize;
901 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
903 di_flags |= XFS_DIFLAG_NOATIME;
904 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
906 di_flags |= XFS_DIFLAG_NODUMP;
907 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
909 di_flags |= XFS_DIFLAG_SYNC;
910 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
911 xfs_inherit_nosymlinks)
912 di_flags |= XFS_DIFLAG_NOSYMLINKS;
913 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
914 xfs_inherit_nodefrag)
915 di_flags |= XFS_DIFLAG_NODEFRAG;
916 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
917 di_flags |= XFS_DIFLAG_FILESTREAM;
918 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
919 di_flags2 |= XFS_DIFLAG2_DAX;
921 ip->i_d.di_flags |= di_flags;
922 ip->i_d.di_flags2 |= di_flags2;
925 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
926 pip->i_d.di_version == 3 &&
927 ip->i_d.di_version == 3) {
928 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
929 ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
930 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
935 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
936 ip->i_df.if_flags = XFS_IFEXTENTS;
937 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
938 ip->i_df.if_u1.if_extents = NULL;
944 * Attribute fork settings for new inode.
946 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
947 ip->i_d.di_anextents = 0;
950 * Log the new values stuffed into the inode.
952 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
953 xfs_trans_log_inode(tp, ip, flags);
955 /* now that we have an i_mode we can setup the inode structure */
963 * Allocates a new inode from disk and return a pointer to the
964 * incore copy. This routine will internally commit the current
965 * transaction and allocate a new one if the Space Manager needed
966 * to do an allocation to replenish the inode free-list.
968 * This routine is designed to be called from xfs_create and
974 xfs_trans_t **tpp, /* input: current transaction;
975 output: may be a new transaction. */
976 xfs_inode_t *dp, /* directory within whose allocate
981 prid_t prid, /* project id */
982 int okalloc, /* ok to allocate new space */
983 xfs_inode_t **ipp, /* pointer to inode; it will be
990 xfs_buf_t *ialloc_context = NULL;
996 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
999 * xfs_ialloc will return a pointer to an incore inode if
1000 * the Space Manager has an available inode on the free
1001 * list. Otherwise, it will do an allocation and replenish
1002 * the freelist. Since we can only do one allocation per
1003 * transaction without deadlocks, we will need to commit the
1004 * current transaction and start a new one. We will then
1005 * need to call xfs_ialloc again to get the inode.
1007 * If xfs_ialloc did an allocation to replenish the freelist,
1008 * it returns the bp containing the head of the freelist as
1009 * ialloc_context. We will hold a lock on it across the
1010 * transaction commit so that no other process can steal
1011 * the inode(s) that we've just allocated.
1013 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1014 &ialloc_context, &ip);
1017 * Return an error if we were unable to allocate a new inode.
1018 * This should only happen if we run out of space on disk or
1019 * encounter a disk error.
1025 if (!ialloc_context && !ip) {
1031 * If the AGI buffer is non-NULL, then we were unable to get an
1032 * inode in one operation. We need to commit the current
1033 * transaction and call xfs_ialloc() again. It is guaranteed
1034 * to succeed the second time.
1036 if (ialloc_context) {
1038 * Normally, xfs_trans_commit releases all the locks.
1039 * We call bhold to hang on to the ialloc_context across
1040 * the commit. Holding this buffer prevents any other
1041 * processes from doing any allocations in this
1044 xfs_trans_bhold(tp, ialloc_context);
1047 * We want the quota changes to be associated with the next
1048 * transaction, NOT this one. So, detach the dqinfo from this
1049 * and attach it to the next transaction.
1054 dqinfo = (void *)tp->t_dqinfo;
1055 tp->t_dqinfo = NULL;
1056 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1057 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1060 code = xfs_trans_roll(&tp, NULL);
1061 if (committed != NULL)
1065 * Re-attach the quota info that we detached from prev trx.
1068 tp->t_dqinfo = dqinfo;
1069 tp->t_flags |= tflags;
1073 xfs_buf_relse(ialloc_context);
1078 xfs_trans_bjoin(tp, ialloc_context);
1081 * Call ialloc again. Since we've locked out all
1082 * other allocations in this allocation group,
1083 * this call should always succeed.
1085 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1086 okalloc, &ialloc_context, &ip);
1089 * If we get an error at this point, return to the caller
1090 * so that the current transaction can be aborted.
1097 ASSERT(!ialloc_context && ip);
1100 if (committed != NULL)
1111 * Decrement the link count on an inode & log the change. If this causes the
1112 * link count to go to zero, move the inode to AGI unlinked list so that it can
1113 * be freed when the last active reference goes away via xfs_inactive().
1115 static int /* error */
1120 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1122 drop_nlink(VFS_I(ip));
1123 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1125 if (VFS_I(ip)->i_nlink)
1128 return xfs_iunlink(tp, ip);
1132 * Increment the link count on an inode & log the change.
1139 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1141 ASSERT(ip->i_d.di_version > 1);
1142 inc_nlink(VFS_I(ip));
1143 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1150 struct xfs_name *name,
1155 int is_dir = S_ISDIR(mode);
1156 struct xfs_mount *mp = dp->i_mount;
1157 struct xfs_inode *ip = NULL;
1158 struct xfs_trans *tp = NULL;
1160 struct xfs_defer_ops dfops;
1161 xfs_fsblock_t first_block;
1162 bool unlock_dp_on_error = false;
1164 struct xfs_dquot *udqp = NULL;
1165 struct xfs_dquot *gdqp = NULL;
1166 struct xfs_dquot *pdqp = NULL;
1167 struct xfs_trans_res *tres;
1170 trace_xfs_create(dp, name);
1172 if (XFS_FORCED_SHUTDOWN(mp))
1175 prid = xfs_get_initial_prid(dp);
1178 * Make sure that we have allocated dquot(s) on disk.
1180 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1181 xfs_kgid_to_gid(current_fsgid()), prid,
1182 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1183 &udqp, &gdqp, &pdqp);
1189 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1190 tres = &M_RES(mp)->tr_mkdir;
1192 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1193 tres = &M_RES(mp)->tr_create;
1197 * Initially assume that the file does not exist and
1198 * reserve the resources for that case. If that is not
1199 * the case we'll drop the one we have and get a more
1200 * appropriate transaction later.
1202 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1203 if (error == -ENOSPC) {
1204 /* flush outstanding delalloc blocks and retry */
1205 xfs_flush_inodes(mp);
1206 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1208 if (error == -ENOSPC) {
1209 /* No space at all so try a "no-allocation" reservation */
1211 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1214 goto out_release_inode;
1216 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1217 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1218 unlock_dp_on_error = true;
1220 xfs_defer_init(&dfops, &first_block);
1223 * Reserve disk quota and the inode.
1225 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1226 pdqp, resblks, 1, 0);
1228 goto out_trans_cancel;
1231 error = xfs_dir_canenter(tp, dp, name);
1233 goto out_trans_cancel;
1237 * A newly created regular or special file just has one directory
1238 * entry pointing to them, but a directory also the "." entry
1239 * pointing to itself.
1241 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1242 prid, resblks > 0, &ip, NULL);
1244 goto out_trans_cancel;
1247 * Now we join the directory inode to the transaction. We do not do it
1248 * earlier because xfs_dir_ialloc might commit the previous transaction
1249 * (and release all the locks). An error from here on will result in
1250 * the transaction cancel unlocking dp so don't do it explicitly in the
1253 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1254 unlock_dp_on_error = false;
1256 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1257 &first_block, &dfops, resblks ?
1258 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1260 ASSERT(error != -ENOSPC);
1261 goto out_trans_cancel;
1263 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1264 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1267 error = xfs_dir_init(tp, ip, dp);
1269 goto out_bmap_cancel;
1271 error = xfs_bumplink(tp, dp);
1273 goto out_bmap_cancel;
1277 * If this is a synchronous mount, make sure that the
1278 * create transaction goes to disk before returning to
1281 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1282 xfs_trans_set_sync(tp);
1285 * Attach the dquot(s) to the inodes and modify them incore.
1286 * These ids of the inode couldn't have changed since the new
1287 * inode has been locked ever since it was created.
1289 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1291 error = xfs_defer_finish(&tp, &dfops, NULL);
1293 goto out_bmap_cancel;
1295 error = xfs_trans_commit(tp);
1297 goto out_release_inode;
1299 xfs_qm_dqrele(udqp);
1300 xfs_qm_dqrele(gdqp);
1301 xfs_qm_dqrele(pdqp);
1307 xfs_defer_cancel(&dfops);
1309 xfs_trans_cancel(tp);
1312 * Wait until after the current transaction is aborted to finish the
1313 * setup of the inode and release the inode. This prevents recursive
1314 * transactions and deadlocks from xfs_inactive.
1317 xfs_finish_inode_setup(ip);
1321 xfs_qm_dqrele(udqp);
1322 xfs_qm_dqrele(gdqp);
1323 xfs_qm_dqrele(pdqp);
1325 if (unlock_dp_on_error)
1326 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1332 struct xfs_inode *dp,
1333 struct dentry *dentry,
1335 struct xfs_inode **ipp)
1337 struct xfs_mount *mp = dp->i_mount;
1338 struct xfs_inode *ip = NULL;
1339 struct xfs_trans *tp = NULL;
1342 struct xfs_dquot *udqp = NULL;
1343 struct xfs_dquot *gdqp = NULL;
1344 struct xfs_dquot *pdqp = NULL;
1345 struct xfs_trans_res *tres;
1348 if (XFS_FORCED_SHUTDOWN(mp))
1351 prid = xfs_get_initial_prid(dp);
1354 * Make sure that we have allocated dquot(s) on disk.
1356 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1357 xfs_kgid_to_gid(current_fsgid()), prid,
1358 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1359 &udqp, &gdqp, &pdqp);
1363 resblks = XFS_IALLOC_SPACE_RES(mp);
1364 tres = &M_RES(mp)->tr_create_tmpfile;
1366 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1367 if (error == -ENOSPC) {
1368 /* No space at all so try a "no-allocation" reservation */
1370 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1373 goto out_release_inode;
1375 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1376 pdqp, resblks, 1, 0);
1378 goto out_trans_cancel;
1380 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1381 prid, resblks > 0, &ip, NULL);
1383 goto out_trans_cancel;
1385 if (mp->m_flags & XFS_MOUNT_WSYNC)
1386 xfs_trans_set_sync(tp);
1389 * Attach the dquot(s) to the inodes and modify them incore.
1390 * These ids of the inode couldn't have changed since the new
1391 * inode has been locked ever since it was created.
1393 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1395 error = xfs_iunlink(tp, ip);
1397 goto out_trans_cancel;
1399 error = xfs_trans_commit(tp);
1401 goto out_release_inode;
1403 xfs_qm_dqrele(udqp);
1404 xfs_qm_dqrele(gdqp);
1405 xfs_qm_dqrele(pdqp);
1411 xfs_trans_cancel(tp);
1414 * Wait until after the current transaction is aborted to finish the
1415 * setup of the inode and release the inode. This prevents recursive
1416 * transactions and deadlocks from xfs_inactive.
1419 xfs_finish_inode_setup(ip);
1423 xfs_qm_dqrele(udqp);
1424 xfs_qm_dqrele(gdqp);
1425 xfs_qm_dqrele(pdqp);
1434 struct xfs_name *target_name)
1436 xfs_mount_t *mp = tdp->i_mount;
1439 struct xfs_defer_ops dfops;
1440 xfs_fsblock_t first_block;
1443 trace_xfs_link(tdp, target_name);
1445 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1447 if (XFS_FORCED_SHUTDOWN(mp))
1450 error = xfs_qm_dqattach(sip, 0);
1454 error = xfs_qm_dqattach(tdp, 0);
1458 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1459 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1460 if (error == -ENOSPC) {
1462 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1467 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1468 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1470 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1471 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1474 * If we are using project inheritance, we only allow hard link
1475 * creation in our tree when the project IDs are the same; else
1476 * the tree quota mechanism could be circumvented.
1478 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1479 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1485 error = xfs_dir_canenter(tp, tdp, target_name);
1490 xfs_defer_init(&dfops, &first_block);
1493 * Handle initial link state of O_TMPFILE inode
1495 if (VFS_I(sip)->i_nlink == 0) {
1496 error = xfs_iunlink_remove(tp, sip);
1501 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1502 &first_block, &dfops, resblks);
1505 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1506 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1508 error = xfs_bumplink(tp, sip);
1513 * If this is a synchronous mount, make sure that the
1514 * link transaction goes to disk before returning to
1517 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1518 xfs_trans_set_sync(tp);
1520 error = xfs_defer_finish(&tp, &dfops, NULL);
1522 xfs_defer_cancel(&dfops);
1526 return xfs_trans_commit(tp);
1529 xfs_trans_cancel(tp);
1535 * Free up the underlying blocks past new_size. The new size must be smaller
1536 * than the current size. This routine can be used both for the attribute and
1537 * data fork, and does not modify the inode size, which is left to the caller.
1539 * The transaction passed to this routine must have made a permanent log
1540 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1541 * given transaction and start new ones, so make sure everything involved in
1542 * the transaction is tidy before calling here. Some transaction will be
1543 * returned to the caller to be committed. The incoming transaction must
1544 * already include the inode, and both inode locks must be held exclusively.
1545 * The inode must also be "held" within the transaction. On return the inode
1546 * will be "held" within the returned transaction. This routine does NOT
1547 * require any disk space to be reserved for it within the transaction.
1549 * If we get an error, we must return with the inode locked and linked into the
1550 * current transaction. This keeps things simple for the higher level code,
1551 * because it always knows that the inode is locked and held in the transaction
1552 * that returns to it whether errors occur or not. We don't mark the inode
1553 * dirty on error so that transactions can be easily aborted if possible.
1556 xfs_itruncate_extents(
1557 struct xfs_trans **tpp,
1558 struct xfs_inode *ip,
1560 xfs_fsize_t new_size)
1562 struct xfs_mount *mp = ip->i_mount;
1563 struct xfs_trans *tp = *tpp;
1564 struct xfs_defer_ops dfops;
1565 xfs_fsblock_t first_block;
1566 xfs_fileoff_t first_unmap_block;
1567 xfs_fileoff_t last_block;
1568 xfs_filblks_t unmap_len;
1572 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1573 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1574 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1575 ASSERT(new_size <= XFS_ISIZE(ip));
1576 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1577 ASSERT(ip->i_itemp != NULL);
1578 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1579 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1581 trace_xfs_itruncate_extents_start(ip, new_size);
1584 * Since it is possible for space to become allocated beyond
1585 * the end of the file (in a crash where the space is allocated
1586 * but the inode size is not yet updated), simply remove any
1587 * blocks which show up between the new EOF and the maximum
1588 * possible file size. If the first block to be removed is
1589 * beyond the maximum file size (ie it is the same as last_block),
1590 * then there is nothing to do.
1592 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1593 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1594 if (first_unmap_block == last_block)
1597 ASSERT(first_unmap_block < last_block);
1598 unmap_len = last_block - first_unmap_block + 1;
1600 xfs_defer_init(&dfops, &first_block);
1601 error = xfs_bunmapi(tp, ip,
1602 first_unmap_block, unmap_len,
1603 xfs_bmapi_aflag(whichfork),
1604 XFS_ITRUNC_MAX_EXTENTS,
1605 &first_block, &dfops,
1608 goto out_bmap_cancel;
1611 * Duplicate the transaction that has the permanent
1612 * reservation and commit the old transaction.
1614 error = xfs_defer_finish(&tp, &dfops, ip);
1616 goto out_bmap_cancel;
1618 error = xfs_trans_roll(&tp, ip);
1623 /* Remove all pending CoW reservations. */
1624 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1630 * Clear the reflink flag if we truncated everything.
1632 if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip))
1633 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1636 * Always re-log the inode so that our permanent transaction can keep
1637 * on rolling it forward in the log.
1639 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1641 trace_xfs_itruncate_extents_end(ip, new_size);
1648 * If the bunmapi call encounters an error, return to the caller where
1649 * the transaction can be properly aborted. We just need to make sure
1650 * we're not holding any resources that we were not when we came in.
1652 xfs_defer_cancel(&dfops);
1660 xfs_mount_t *mp = ip->i_mount;
1663 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1666 /* If this is a read-only mount, don't do this (would generate I/O) */
1667 if (mp->m_flags & XFS_MOUNT_RDONLY)
1670 if (!XFS_FORCED_SHUTDOWN(mp)) {
1674 * If we previously truncated this file and removed old data
1675 * in the process, we want to initiate "early" writeout on
1676 * the last close. This is an attempt to combat the notorious
1677 * NULL files problem which is particularly noticeable from a
1678 * truncate down, buffered (re-)write (delalloc), followed by
1679 * a crash. What we are effectively doing here is
1680 * significantly reducing the time window where we'd otherwise
1681 * be exposed to that problem.
1683 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1685 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1686 if (ip->i_delayed_blks > 0) {
1687 error = filemap_flush(VFS_I(ip)->i_mapping);
1694 if (VFS_I(ip)->i_nlink == 0)
1697 if (xfs_can_free_eofblocks(ip, false)) {
1700 * If we can't get the iolock just skip truncating the blocks
1701 * past EOF because we could deadlock with the mmap_sem
1702 * otherwise. We'll get another chance to drop them once the
1703 * last reference to the inode is dropped, so we'll never leak
1704 * blocks permanently.
1706 * Further, check if the inode is being opened, written and
1707 * closed frequently and we have delayed allocation blocks
1708 * outstanding (e.g. streaming writes from the NFS server),
1709 * truncating the blocks past EOF will cause fragmentation to
1712 * In this case don't do the truncation, either, but we have to
1713 * be careful how we detect this case. Blocks beyond EOF show
1714 * up as i_delayed_blks even when the inode is clean, so we
1715 * need to truncate them away first before checking for a dirty
1716 * release. Hence on the first dirty close we will still remove
1717 * the speculative allocation, but after that we will leave it
1720 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1723 error = xfs_free_eofblocks(mp, ip, true);
1724 if (error && error != -EAGAIN)
1727 /* delalloc blocks after truncation means it really is dirty */
1728 if (ip->i_delayed_blks)
1729 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1735 * xfs_inactive_truncate
1737 * Called to perform a truncate when an inode becomes unlinked.
1740 xfs_inactive_truncate(
1741 struct xfs_inode *ip)
1743 struct xfs_mount *mp = ip->i_mount;
1744 struct xfs_trans *tp;
1747 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1749 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1753 xfs_ilock(ip, XFS_ILOCK_EXCL);
1754 xfs_trans_ijoin(tp, ip, 0);
1757 * Log the inode size first to prevent stale data exposure in the event
1758 * of a system crash before the truncate completes. See the related
1759 * comment in xfs_setattr_size() for details.
1761 ip->i_d.di_size = 0;
1762 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1764 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1766 goto error_trans_cancel;
1768 ASSERT(ip->i_d.di_nextents == 0);
1770 error = xfs_trans_commit(tp);
1774 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1778 xfs_trans_cancel(tp);
1780 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1785 * xfs_inactive_ifree()
1787 * Perform the inode free when an inode is unlinked.
1791 struct xfs_inode *ip)
1793 struct xfs_defer_ops dfops;
1794 xfs_fsblock_t first_block;
1795 struct xfs_mount *mp = ip->i_mount;
1796 struct xfs_trans *tp;
1800 * The ifree transaction might need to allocate blocks for record
1801 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1802 * allow ifree to dip into the reserved block pool if necessary.
1804 * Freeing large sets of inodes generally means freeing inode chunks,
1805 * directory and file data blocks, so this should be relatively safe.
1806 * Only under severe circumstances should it be possible to free enough
1807 * inodes to exhaust the reserve block pool via finobt expansion while
1808 * at the same time not creating free space in the filesystem.
1810 * Send a warning if the reservation does happen to fail, as the inode
1811 * now remains allocated and sits on the unlinked list until the fs is
1814 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1815 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
1817 if (error == -ENOSPC) {
1818 xfs_warn_ratelimited(mp,
1819 "Failed to remove inode(s) from unlinked list. "
1820 "Please free space, unmount and run xfs_repair.");
1822 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1827 xfs_ilock(ip, XFS_ILOCK_EXCL);
1828 xfs_trans_ijoin(tp, ip, 0);
1830 xfs_defer_init(&dfops, &first_block);
1831 error = xfs_ifree(tp, ip, &dfops);
1834 * If we fail to free the inode, shut down. The cancel
1835 * might do that, we need to make sure. Otherwise the
1836 * inode might be lost for a long time or forever.
1838 if (!XFS_FORCED_SHUTDOWN(mp)) {
1839 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1841 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1843 xfs_trans_cancel(tp);
1844 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1849 * Credit the quota account(s). The inode is gone.
1851 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1854 * Just ignore errors at this point. There is nothing we can do except
1855 * to try to keep going. Make sure it's not a silent error.
1857 error = xfs_defer_finish(&tp, &dfops, NULL);
1859 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1861 xfs_defer_cancel(&dfops);
1863 error = xfs_trans_commit(tp);
1865 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1868 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1875 * This is called when the vnode reference count for the vnode
1876 * goes to zero. If the file has been unlinked, then it must
1877 * now be truncated. Also, we clear all of the read-ahead state
1878 * kept for the inode here since the file is now closed.
1884 struct xfs_mount *mp;
1889 * If the inode is already free, then there can be nothing
1892 if (VFS_I(ip)->i_mode == 0) {
1893 ASSERT(ip->i_df.if_real_bytes == 0);
1894 ASSERT(ip->i_df.if_broot_bytes == 0);
1899 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1901 /* If this is a read-only mount, don't do this (would generate I/O) */
1902 if (mp->m_flags & XFS_MOUNT_RDONLY)
1905 if (VFS_I(ip)->i_nlink != 0) {
1907 * force is true because we are evicting an inode from the
1908 * cache. Post-eof blocks must be freed, lest we end up with
1909 * broken free space accounting.
1911 if (xfs_can_free_eofblocks(ip, true))
1912 xfs_free_eofblocks(mp, ip, false);
1917 if (S_ISREG(VFS_I(ip)->i_mode) &&
1918 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1919 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1922 error = xfs_qm_dqattach(ip, 0);
1926 if (S_ISLNK(VFS_I(ip)->i_mode))
1927 error = xfs_inactive_symlink(ip);
1929 error = xfs_inactive_truncate(ip);
1934 * If there are attributes associated with the file then blow them away
1935 * now. The code calls a routine that recursively deconstructs the
1936 * attribute fork. If also blows away the in-core attribute fork.
1938 if (XFS_IFORK_Q(ip)) {
1939 error = xfs_attr_inactive(ip);
1945 ASSERT(ip->i_d.di_anextents == 0);
1946 ASSERT(ip->i_d.di_forkoff == 0);
1951 error = xfs_inactive_ifree(ip);
1956 * Release the dquots held by inode, if any.
1958 xfs_qm_dqdetach(ip);
1962 * This is called when the inode's link count goes to 0 or we are creating a
1963 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1964 * set to true as the link count is dropped to zero by the VFS after we've
1965 * created the file successfully, so we have to add it to the unlinked list
1966 * while the link count is non-zero.
1968 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1969 * list when the inode is freed.
1973 struct xfs_trans *tp,
1974 struct xfs_inode *ip)
1976 xfs_mount_t *mp = tp->t_mountp;
1986 ASSERT(VFS_I(ip)->i_mode != 0);
1989 * Get the agi buffer first. It ensures lock ordering
1992 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1995 agi = XFS_BUF_TO_AGI(agibp);
1998 * Get the index into the agi hash table for the
1999 * list this inode will go on.
2001 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2003 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2004 ASSERT(agi->agi_unlinked[bucket_index]);
2005 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2007 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2009 * There is already another inode in the bucket we need
2010 * to add ourselves to. Add us at the front of the list.
2011 * Here we put the head pointer into our next pointer,
2012 * and then we fall through to point the head at us.
2014 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2019 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2020 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2021 offset = ip->i_imap.im_boffset +
2022 offsetof(xfs_dinode_t, di_next_unlinked);
2024 /* need to recalc the inode CRC if appropriate */
2025 xfs_dinode_calc_crc(mp, dip);
2027 xfs_trans_inode_buf(tp, ibp);
2028 xfs_trans_log_buf(tp, ibp, offset,
2029 (offset + sizeof(xfs_agino_t) - 1));
2030 xfs_inobp_check(mp, ibp);
2034 * Point the bucket head pointer at the inode being inserted.
2037 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2038 offset = offsetof(xfs_agi_t, agi_unlinked) +
2039 (sizeof(xfs_agino_t) * bucket_index);
2040 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
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_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2133 xfs_trans_log_buf(tp, agibp, offset,
2134 (offset + sizeof(xfs_agino_t) - 1));
2137 * We need to search the list for the inode being freed.
2139 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2141 while (next_agino != agino) {
2142 struct xfs_imap imap;
2145 xfs_trans_brelse(tp, last_ibp);
2148 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2150 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2153 "%s: xfs_imap returned error %d.",
2158 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2162 "%s: xfs_imap_to_bp returned error %d.",
2167 last_offset = imap.im_boffset;
2168 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2169 ASSERT(next_agino != NULLAGINO);
2170 ASSERT(next_agino != 0);
2174 * Now last_ibp points to the buffer previous to us on the
2175 * unlinked list. Pull us from the list.
2177 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2180 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2184 next_agino = be32_to_cpu(dip->di_next_unlinked);
2185 ASSERT(next_agino != 0);
2186 ASSERT(next_agino != agino);
2187 if (next_agino != NULLAGINO) {
2188 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2189 offset = ip->i_imap.im_boffset +
2190 offsetof(xfs_dinode_t, di_next_unlinked);
2192 /* need to recalc the inode CRC if appropriate */
2193 xfs_dinode_calc_crc(mp, dip);
2195 xfs_trans_inode_buf(tp, ibp);
2196 xfs_trans_log_buf(tp, ibp, offset,
2197 (offset + sizeof(xfs_agino_t) - 1));
2198 xfs_inobp_check(mp, ibp);
2200 xfs_trans_brelse(tp, ibp);
2203 * Point the previous inode on the list to the next inode.
2205 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2206 ASSERT(next_agino != 0);
2207 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2209 /* need to recalc the inode CRC if appropriate */
2210 xfs_dinode_calc_crc(mp, last_dip);
2212 xfs_trans_inode_buf(tp, last_ibp);
2213 xfs_trans_log_buf(tp, last_ibp, offset,
2214 (offset + sizeof(xfs_agino_t) - 1));
2215 xfs_inobp_check(mp, last_ibp);
2221 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2222 * inodes that are in memory - they all must be marked stale and attached to
2223 * the cluster buffer.
2227 xfs_inode_t *free_ip,
2229 struct xfs_icluster *xic)
2231 xfs_mount_t *mp = free_ip->i_mount;
2232 int blks_per_cluster;
2233 int inodes_per_cluster;
2240 xfs_inode_log_item_t *iip;
2241 xfs_log_item_t *lip;
2242 struct xfs_perag *pag;
2245 inum = xic->first_ino;
2246 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2247 blks_per_cluster = xfs_icluster_size_fsb(mp);
2248 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2249 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2251 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2253 * The allocation bitmap tells us which inodes of the chunk were
2254 * physically allocated. Skip the cluster if an inode falls into
2257 ioffset = inum - xic->first_ino;
2258 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2259 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2263 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2264 XFS_INO_TO_AGBNO(mp, inum));
2267 * We obtain and lock the backing buffer first in the process
2268 * here, as we have to ensure that any dirty inode that we
2269 * can't get the flush lock on is attached to the buffer.
2270 * If we scan the in-memory inodes first, then buffer IO can
2271 * complete before we get a lock on it, and hence we may fail
2272 * to mark all the active inodes on the buffer stale.
2274 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2275 mp->m_bsize * blks_per_cluster,
2282 * This buffer may not have been correctly initialised as we
2283 * didn't read it from disk. That's not important because we are
2284 * only using to mark the buffer as stale in the log, and to
2285 * attach stale cached inodes on it. That means it will never be
2286 * dispatched for IO. If it is, we want to know about it, and we
2287 * want it to fail. We can acheive this by adding a write
2288 * verifier to the buffer.
2290 bp->b_ops = &xfs_inode_buf_ops;
2293 * Walk the inodes already attached to the buffer and mark them
2294 * stale. These will all have the flush locks held, so an
2295 * in-memory inode walk can't lock them. By marking them all
2296 * stale first, we will not attempt to lock them in the loop
2297 * below as the XFS_ISTALE flag will be set.
2301 if (lip->li_type == XFS_LI_INODE) {
2302 iip = (xfs_inode_log_item_t *)lip;
2303 ASSERT(iip->ili_logged == 1);
2304 lip->li_cb = xfs_istale_done;
2305 xfs_trans_ail_copy_lsn(mp->m_ail,
2306 &iip->ili_flush_lsn,
2307 &iip->ili_item.li_lsn);
2308 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2310 lip = lip->li_bio_list;
2315 * For each inode in memory attempt to add it to the inode
2316 * buffer and set it up for being staled on buffer IO
2317 * completion. This is safe as we've locked out tail pushing
2318 * and flushing by locking the buffer.
2320 * We have already marked every inode that was part of a
2321 * transaction stale above, which means there is no point in
2322 * even trying to lock them.
2324 for (i = 0; i < inodes_per_cluster; i++) {
2327 ip = radix_tree_lookup(&pag->pag_ici_root,
2328 XFS_INO_TO_AGINO(mp, (inum + i)));
2330 /* Inode not in memory, nothing to do */
2337 * because this is an RCU protected lookup, we could
2338 * find a recently freed or even reallocated inode
2339 * during the lookup. We need to check under the
2340 * i_flags_lock for a valid inode here. Skip it if it
2341 * is not valid, the wrong inode or stale.
2343 spin_lock(&ip->i_flags_lock);
2344 if (ip->i_ino != inum + i ||
2345 __xfs_iflags_test(ip, XFS_ISTALE)) {
2346 spin_unlock(&ip->i_flags_lock);
2350 spin_unlock(&ip->i_flags_lock);
2353 * Don't try to lock/unlock the current inode, but we
2354 * _cannot_ skip the other inodes that we did not find
2355 * in the list attached to the buffer and are not
2356 * already marked stale. If we can't lock it, back off
2359 if (ip != free_ip &&
2360 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2368 xfs_iflags_set(ip, XFS_ISTALE);
2371 * we don't need to attach clean inodes or those only
2372 * with unlogged changes (which we throw away, anyway).
2375 if (!iip || xfs_inode_clean(ip)) {
2376 ASSERT(ip != free_ip);
2378 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2382 iip->ili_last_fields = iip->ili_fields;
2383 iip->ili_fields = 0;
2384 iip->ili_fsync_fields = 0;
2385 iip->ili_logged = 1;
2386 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2387 &iip->ili_item.li_lsn);
2389 xfs_buf_attach_iodone(bp, xfs_istale_done,
2393 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2396 xfs_trans_stale_inode_buf(tp, bp);
2397 xfs_trans_binval(tp, bp);
2405 * This is called to return an inode to the inode free list.
2406 * The inode should already be truncated to 0 length and have
2407 * no pages associated with it. This routine also assumes that
2408 * the inode is already a part of the transaction.
2410 * The on-disk copy of the inode will have been added to the list
2411 * of unlinked inodes in the AGI. We need to remove the inode from
2412 * that list atomically with respect to freeing it here.
2418 struct xfs_defer_ops *dfops)
2421 struct xfs_icluster xic = { 0 };
2423 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2424 ASSERT(VFS_I(ip)->i_nlink == 0);
2425 ASSERT(ip->i_d.di_nextents == 0);
2426 ASSERT(ip->i_d.di_anextents == 0);
2427 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2428 ASSERT(ip->i_d.di_nblocks == 0);
2431 * Pull the on-disk inode from the AGI unlinked list.
2433 error = xfs_iunlink_remove(tp, ip);
2437 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2441 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2442 ip->i_d.di_flags = 0;
2443 ip->i_d.di_dmevmask = 0;
2444 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2445 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2446 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2448 * Bump the generation count so no one will be confused
2449 * by reincarnations of this inode.
2451 VFS_I(ip)->i_generation++;
2452 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2455 error = xfs_ifree_cluster(ip, tp, &xic);
2461 * This is called to unpin an inode. The caller must have the inode locked
2462 * in at least shared mode so that the buffer cannot be subsequently pinned
2463 * once someone is waiting for it to be unpinned.
2467 struct xfs_inode *ip)
2469 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2471 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2473 /* Give the log a push to start the unpinning I/O */
2474 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2480 struct xfs_inode *ip)
2482 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2483 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2488 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2489 if (xfs_ipincount(ip))
2491 } while (xfs_ipincount(ip));
2492 finish_wait(wq, &wait.wait);
2497 struct xfs_inode *ip)
2499 if (xfs_ipincount(ip))
2500 __xfs_iunpin_wait(ip);
2504 * Removing an inode from the namespace involves removing the directory entry
2505 * and dropping the link count on the inode. Removing the directory entry can
2506 * result in locking an AGF (directory blocks were freed) and removing a link
2507 * count can result in placing the inode on an unlinked list which results in
2510 * The big problem here is that we have an ordering constraint on AGF and AGI
2511 * locking - inode allocation locks the AGI, then can allocate a new extent for
2512 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2513 * removes the inode from the unlinked list, requiring that we lock the AGI
2514 * first, and then freeing the inode can result in an inode chunk being freed
2515 * and hence freeing disk space requiring that we lock an AGF.
2517 * Hence the ordering that is imposed by other parts of the code is AGI before
2518 * AGF. This means we cannot remove the directory entry before we drop the inode
2519 * reference count and put it on the unlinked list as this results in a lock
2520 * order of AGF then AGI, and this can deadlock against inode allocation and
2521 * freeing. Therefore we must drop the link counts before we remove the
2524 * This is still safe from a transactional point of view - it is not until we
2525 * get to xfs_defer_finish() that we have the possibility of multiple
2526 * transactions in this operation. Hence as long as we remove the directory
2527 * entry and drop the link count in the first transaction of the remove
2528 * operation, there are no transactional constraints on the ordering here.
2533 struct xfs_name *name,
2536 xfs_mount_t *mp = dp->i_mount;
2537 xfs_trans_t *tp = NULL;
2538 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2540 struct xfs_defer_ops dfops;
2541 xfs_fsblock_t first_block;
2544 trace_xfs_remove(dp, name);
2546 if (XFS_FORCED_SHUTDOWN(mp))
2549 error = xfs_qm_dqattach(dp, 0);
2553 error = xfs_qm_dqattach(ip, 0);
2558 * We try to get the real space reservation first,
2559 * allowing for directory btree deletion(s) implying
2560 * possible bmap insert(s). If we can't get the space
2561 * reservation then we use 0 instead, and avoid the bmap
2562 * btree insert(s) in the directory code by, if the bmap
2563 * insert tries to happen, instead trimming the LAST
2564 * block from the directory.
2566 resblks = XFS_REMOVE_SPACE_RES(mp);
2567 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2568 if (error == -ENOSPC) {
2570 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2574 ASSERT(error != -ENOSPC);
2578 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2579 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2581 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2582 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2585 * If we're removing a directory perform some additional validation.
2588 ASSERT(VFS_I(ip)->i_nlink >= 2);
2589 if (VFS_I(ip)->i_nlink != 2) {
2591 goto out_trans_cancel;
2593 if (!xfs_dir_isempty(ip)) {
2595 goto out_trans_cancel;
2598 /* Drop the link from ip's "..". */
2599 error = xfs_droplink(tp, dp);
2601 goto out_trans_cancel;
2603 /* Drop the "." link from ip to self. */
2604 error = xfs_droplink(tp, ip);
2606 goto out_trans_cancel;
2609 * When removing a non-directory we need to log the parent
2610 * inode here. For a directory this is done implicitly
2611 * by the xfs_droplink call for the ".." entry.
2613 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2615 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2617 /* Drop the link from dp to ip. */
2618 error = xfs_droplink(tp, ip);
2620 goto out_trans_cancel;
2622 xfs_defer_init(&dfops, &first_block);
2623 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2624 &first_block, &dfops, resblks);
2626 ASSERT(error != -ENOENT);
2627 goto out_bmap_cancel;
2631 * If this is a synchronous mount, make sure that the
2632 * remove transaction goes to disk before returning to
2635 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2636 xfs_trans_set_sync(tp);
2638 error = xfs_defer_finish(&tp, &dfops, NULL);
2640 goto out_bmap_cancel;
2642 error = xfs_trans_commit(tp);
2646 if (is_dir && xfs_inode_is_filestream(ip))
2647 xfs_filestream_deassociate(ip);
2652 xfs_defer_cancel(&dfops);
2654 xfs_trans_cancel(tp);
2660 * Enter all inodes for a rename transaction into a sorted array.
2662 #define __XFS_SORT_INODES 5
2664 xfs_sort_for_rename(
2665 struct xfs_inode *dp1, /* in: old (source) directory inode */
2666 struct xfs_inode *dp2, /* in: new (target) directory inode */
2667 struct xfs_inode *ip1, /* in: inode of old entry */
2668 struct xfs_inode *ip2, /* in: inode of new entry */
2669 struct xfs_inode *wip, /* in: whiteout inode */
2670 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2671 int *num_inodes) /* in/out: inodes in array */
2675 ASSERT(*num_inodes == __XFS_SORT_INODES);
2676 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2679 * i_tab contains a list of pointers to inodes. We initialize
2680 * the table here & we'll sort it. We will then use it to
2681 * order the acquisition of the inode locks.
2683 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2696 * Sort the elements via bubble sort. (Remember, there are at
2697 * most 5 elements to sort, so this is adequate.)
2699 for (i = 0; i < *num_inodes; i++) {
2700 for (j = 1; j < *num_inodes; j++) {
2701 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2702 struct xfs_inode *temp = i_tab[j];
2703 i_tab[j] = i_tab[j-1];
2712 struct xfs_trans *tp,
2713 struct xfs_defer_ops *dfops)
2718 * If this is a synchronous mount, make sure that the rename transaction
2719 * goes to disk before returning to the user.
2721 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2722 xfs_trans_set_sync(tp);
2724 error = xfs_defer_finish(&tp, dfops, NULL);
2726 xfs_defer_cancel(dfops);
2727 xfs_trans_cancel(tp);
2731 return xfs_trans_commit(tp);
2735 * xfs_cross_rename()
2737 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2741 struct xfs_trans *tp,
2742 struct xfs_inode *dp1,
2743 struct xfs_name *name1,
2744 struct xfs_inode *ip1,
2745 struct xfs_inode *dp2,
2746 struct xfs_name *name2,
2747 struct xfs_inode *ip2,
2748 struct xfs_defer_ops *dfops,
2749 xfs_fsblock_t *first_block,
2757 /* Swap inode number for dirent in first parent */
2758 error = xfs_dir_replace(tp, dp1, name1,
2760 first_block, dfops, spaceres);
2762 goto out_trans_abort;
2764 /* Swap inode number for dirent in second parent */
2765 error = xfs_dir_replace(tp, dp2, name2,
2767 first_block, dfops, spaceres);
2769 goto out_trans_abort;
2772 * If we're renaming one or more directories across different parents,
2773 * update the respective ".." entries (and link counts) to match the new
2777 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2779 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2780 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2781 dp1->i_ino, first_block,
2784 goto out_trans_abort;
2786 /* transfer ip2 ".." reference to dp1 */
2787 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2788 error = xfs_droplink(tp, dp2);
2790 goto out_trans_abort;
2791 error = xfs_bumplink(tp, dp1);
2793 goto out_trans_abort;
2797 * Although ip1 isn't changed here, userspace needs
2798 * to be warned about the change, so that applications
2799 * relying on it (like backup ones), will properly
2802 ip1_flags |= XFS_ICHGTIME_CHG;
2803 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2806 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2807 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2808 dp2->i_ino, first_block,
2811 goto out_trans_abort;
2813 /* transfer ip1 ".." reference to dp2 */
2814 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2815 error = xfs_droplink(tp, dp1);
2817 goto out_trans_abort;
2818 error = xfs_bumplink(tp, dp2);
2820 goto out_trans_abort;
2824 * Although ip2 isn't changed here, userspace needs
2825 * to be warned about the change, so that applications
2826 * relying on it (like backup ones), will properly
2829 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2830 ip2_flags |= XFS_ICHGTIME_CHG;
2835 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2836 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2839 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2840 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2843 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2844 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2846 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2847 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2848 return xfs_finish_rename(tp, dfops);
2851 xfs_defer_cancel(dfops);
2852 xfs_trans_cancel(tp);
2857 * xfs_rename_alloc_whiteout()
2859 * Return a referenced, unlinked, unlocked inode that that can be used as a
2860 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2861 * crash between allocating the inode and linking it into the rename transaction
2862 * recovery will free the inode and we won't leak it.
2865 xfs_rename_alloc_whiteout(
2866 struct xfs_inode *dp,
2867 struct xfs_inode **wip)
2869 struct xfs_inode *tmpfile;
2872 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2877 * Prepare the tmpfile inode as if it were created through the VFS.
2878 * Otherwise, the link increment paths will complain about nlink 0->1.
2879 * Drop the link count as done by d_tmpfile(), complete the inode setup
2880 * and flag it as linkable.
2882 drop_nlink(VFS_I(tmpfile));
2883 xfs_setup_iops(tmpfile);
2884 xfs_finish_inode_setup(tmpfile);
2885 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2896 struct xfs_inode *src_dp,
2897 struct xfs_name *src_name,
2898 struct xfs_inode *src_ip,
2899 struct xfs_inode *target_dp,
2900 struct xfs_name *target_name,
2901 struct xfs_inode *target_ip,
2904 struct xfs_mount *mp = src_dp->i_mount;
2905 struct xfs_trans *tp;
2906 struct xfs_defer_ops dfops;
2907 xfs_fsblock_t first_block;
2908 struct xfs_inode *wip = NULL; /* whiteout inode */
2909 struct xfs_inode *inodes[__XFS_SORT_INODES];
2910 int num_inodes = __XFS_SORT_INODES;
2911 bool new_parent = (src_dp != target_dp);
2912 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2916 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2918 if ((flags & RENAME_EXCHANGE) && !target_ip)
2922 * If we are doing a whiteout operation, allocate the whiteout inode
2923 * we will be placing at the target and ensure the type is set
2926 if (flags & RENAME_WHITEOUT) {
2927 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2928 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2932 /* setup target dirent info as whiteout */
2933 src_name->type = XFS_DIR3_FT_CHRDEV;
2936 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2937 inodes, &num_inodes);
2939 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2940 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2941 if (error == -ENOSPC) {
2943 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2947 goto out_release_wip;
2950 * Attach the dquots to the inodes
2952 error = xfs_qm_vop_rename_dqattach(inodes);
2954 goto out_trans_cancel;
2957 * Lock all the participating inodes. Depending upon whether
2958 * the target_name exists in the target directory, and
2959 * whether the target directory is the same as the source
2960 * directory, we can lock from 2 to 4 inodes.
2963 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2965 xfs_lock_two_inodes(src_dp, target_dp,
2966 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2968 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2971 * Join all the inodes to the transaction. From this point on,
2972 * we can rely on either trans_commit or trans_cancel to unlock
2975 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2977 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2978 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2980 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2982 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2985 * If we are using project inheritance, we only allow renames
2986 * into our tree when the project IDs are the same; else the
2987 * tree quota mechanism would be circumvented.
2989 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2990 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2992 goto out_trans_cancel;
2995 xfs_defer_init(&dfops, &first_block);
2997 /* RENAME_EXCHANGE is unique from here on. */
2998 if (flags & RENAME_EXCHANGE)
2999 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3000 target_dp, target_name, target_ip,
3001 &dfops, &first_block, spaceres);
3004 * Set up the target.
3006 if (target_ip == NULL) {
3008 * If there's no space reservation, check the entry will
3009 * fit before actually inserting it.
3012 error = xfs_dir_canenter(tp, target_dp, target_name);
3014 goto out_trans_cancel;
3017 * If target does not exist and the rename crosses
3018 * directories, adjust the target directory link count
3019 * to account for the ".." reference from the new entry.
3021 error = xfs_dir_createname(tp, target_dp, target_name,
3022 src_ip->i_ino, &first_block,
3025 goto out_bmap_cancel;
3027 xfs_trans_ichgtime(tp, target_dp,
3028 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3030 if (new_parent && src_is_directory) {
3031 error = xfs_bumplink(tp, target_dp);
3033 goto out_bmap_cancel;
3035 } else { /* target_ip != NULL */
3037 * If target exists and it's a directory, check that both
3038 * target and source are directories and that target can be
3039 * destroyed, or that neither is a directory.
3041 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3043 * Make sure target dir is empty.
3045 if (!(xfs_dir_isempty(target_ip)) ||
3046 (VFS_I(target_ip)->i_nlink > 2)) {
3048 goto out_trans_cancel;
3053 * Link the source inode under the target name.
3054 * If the source inode is a directory and we are moving
3055 * it across directories, its ".." entry will be
3056 * inconsistent until we replace that down below.
3058 * In case there is already an entry with the same
3059 * name at the destination directory, remove it first.
3061 error = xfs_dir_replace(tp, target_dp, target_name,
3063 &first_block, &dfops, spaceres);
3065 goto out_bmap_cancel;
3067 xfs_trans_ichgtime(tp, target_dp,
3068 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3071 * Decrement the link count on the target since the target
3072 * dir no longer points to it.
3074 error = xfs_droplink(tp, target_ip);
3076 goto out_bmap_cancel;
3078 if (src_is_directory) {
3080 * Drop the link from the old "." entry.
3082 error = xfs_droplink(tp, target_ip);
3084 goto out_bmap_cancel;
3086 } /* target_ip != NULL */
3089 * Remove the source.
3091 if (new_parent && src_is_directory) {
3093 * Rewrite the ".." entry to point to the new
3096 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3098 &first_block, &dfops, spaceres);
3099 ASSERT(error != -EEXIST);
3101 goto out_bmap_cancel;
3105 * We always want to hit the ctime on the source inode.
3107 * This isn't strictly required by the standards since the source
3108 * inode isn't really being changed, but old unix file systems did
3109 * it and some incremental backup programs won't work without it.
3111 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3112 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3115 * Adjust the link count on src_dp. This is necessary when
3116 * renaming a directory, either within one parent when
3117 * the target existed, or across two parent directories.
3119 if (src_is_directory && (new_parent || target_ip != NULL)) {
3122 * Decrement link count on src_directory since the
3123 * entry that's moved no longer points to it.
3125 error = xfs_droplink(tp, src_dp);
3127 goto out_bmap_cancel;
3131 * For whiteouts, we only need to update the source dirent with the
3132 * inode number of the whiteout inode rather than removing it
3136 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3137 &first_block, &dfops, spaceres);
3139 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3140 &first_block, &dfops, spaceres);
3142 goto out_bmap_cancel;
3145 * For whiteouts, we need to bump the link count on the whiteout inode.
3146 * This means that failures all the way up to this point leave the inode
3147 * on the unlinked list and so cleanup is a simple matter of dropping
3148 * the remaining reference to it. If we fail here after bumping the link
3149 * count, we're shutting down the filesystem so we'll never see the
3150 * intermediate state on disk.
3153 ASSERT(VFS_I(wip)->i_nlink == 0);
3154 error = xfs_bumplink(tp, wip);
3156 goto out_bmap_cancel;
3157 error = xfs_iunlink_remove(tp, wip);
3159 goto out_bmap_cancel;
3160 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3163 * Now we have a real link, clear the "I'm a tmpfile" state
3164 * flag from the inode so it doesn't accidentally get misused in
3167 VFS_I(wip)->i_state &= ~I_LINKABLE;
3170 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3171 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3173 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3175 error = xfs_finish_rename(tp, &dfops);
3181 xfs_defer_cancel(&dfops);
3183 xfs_trans_cancel(tp);
3192 struct xfs_inode *ip,
3195 struct xfs_mount *mp = ip->i_mount;
3196 struct xfs_perag *pag;
3197 unsigned long first_index, mask;
3198 unsigned long inodes_per_cluster;
3200 struct xfs_inode **cilist;
3201 struct xfs_inode *cip;
3207 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3209 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3210 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3211 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3215 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3216 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3218 /* really need a gang lookup range call here */
3219 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3220 first_index, inodes_per_cluster);
3224 for (i = 0; i < nr_found; i++) {
3230 * because this is an RCU protected lookup, we could find a
3231 * recently freed or even reallocated inode during the lookup.
3232 * We need to check under the i_flags_lock for a valid inode
3233 * here. Skip it if it is not valid or the wrong inode.
3235 spin_lock(&cip->i_flags_lock);
3237 __xfs_iflags_test(cip, XFS_ISTALE)) {
3238 spin_unlock(&cip->i_flags_lock);
3243 * Once we fall off the end of the cluster, no point checking
3244 * any more inodes in the list because they will also all be
3245 * outside the cluster.
3247 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3248 spin_unlock(&cip->i_flags_lock);
3251 spin_unlock(&cip->i_flags_lock);
3254 * Do an un-protected check to see if the inode is dirty and
3255 * is a candidate for flushing. These checks will be repeated
3256 * later after the appropriate locks are acquired.
3258 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3262 * Try to get locks. If any are unavailable or it is pinned,
3263 * then this inode cannot be flushed and is skipped.
3266 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3268 if (!xfs_iflock_nowait(cip)) {
3269 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3272 if (xfs_ipincount(cip)) {
3274 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3280 * Check the inode number again, just to be certain we are not
3281 * racing with freeing in xfs_reclaim_inode(). See the comments
3282 * in that function for more information as to why the initial
3283 * check is not sufficient.
3287 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3292 * arriving here means that this inode can be flushed. First
3293 * re-check that it's dirty before flushing.
3295 if (!xfs_inode_clean(cip)) {
3297 error = xfs_iflush_int(cip, bp);
3299 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3300 goto cluster_corrupt_out;
3306 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3310 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3311 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3322 cluster_corrupt_out:
3324 * Corruption detected in the clustering loop. Invalidate the
3325 * inode buffer and shut down the filesystem.
3329 * Clean up the buffer. If it was delwri, just release it --
3330 * brelse can handle it with no problems. If not, shut down the
3331 * filesystem before releasing the buffer.
3333 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3337 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3339 if (!bufwasdelwri) {
3341 * Just like incore_relse: if we have b_iodone functions,
3342 * mark the buffer as an error and call them. Otherwise
3343 * mark it as stale and brelse.
3346 bp->b_flags &= ~XBF_DONE;
3348 xfs_buf_ioerror(bp, -EIO);
3357 * Unlocks the flush lock
3359 xfs_iflush_abort(cip, false);
3362 return -EFSCORRUPTED;
3366 * Flush dirty inode metadata into the backing buffer.
3368 * The caller must have the inode lock and the inode flush lock held. The
3369 * inode lock will still be held upon return to the caller, and the inode
3370 * flush lock will be released after the inode has reached the disk.
3372 * The caller must write out the buffer returned in *bpp and release it.
3376 struct xfs_inode *ip,
3377 struct xfs_buf **bpp)
3379 struct xfs_mount *mp = ip->i_mount;
3380 struct xfs_buf *bp = NULL;
3381 struct xfs_dinode *dip;
3384 XFS_STATS_INC(mp, xs_iflush_count);
3386 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3387 ASSERT(xfs_isiflocked(ip));
3388 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3389 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3393 xfs_iunpin_wait(ip);
3396 * For stale inodes we cannot rely on the backing buffer remaining
3397 * stale in cache for the remaining life of the stale inode and so
3398 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3399 * inodes below. We have to check this after ensuring the inode is
3400 * unpinned so that it is safe to reclaim the stale inode after the
3403 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3409 * This may have been unpinned because the filesystem is shutting
3410 * down forcibly. If that's the case we must not write this inode
3411 * to disk, because the log record didn't make it to disk.
3413 * We also have to remove the log item from the AIL in this case,
3414 * as we wait for an empty AIL as part of the unmount process.
3416 if (XFS_FORCED_SHUTDOWN(mp)) {
3422 * Get the buffer containing the on-disk inode. We are doing a try-lock
3423 * operation here, so we may get an EAGAIN error. In that case, we
3424 * simply want to return with the inode still dirty.
3426 * If we get any other error, we effectively have a corruption situation
3427 * and we cannot flush the inode, so we treat it the same as failing
3430 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3432 if (error == -EAGAIN) {
3440 * First flush out the inode that xfs_iflush was called with.
3442 error = xfs_iflush_int(ip, bp);
3447 * If the buffer is pinned then push on the log now so we won't
3448 * get stuck waiting in the write for too long.
3450 if (xfs_buf_ispinned(bp))
3451 xfs_log_force(mp, 0);
3455 * see if other inodes can be gathered into this write
3457 error = xfs_iflush_cluster(ip, bp);
3459 goto cluster_corrupt_out;
3467 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3468 cluster_corrupt_out:
3469 error = -EFSCORRUPTED;
3472 * Unlocks the flush lock
3474 xfs_iflush_abort(ip, false);
3480 struct xfs_inode *ip,
3483 struct xfs_inode_log_item *iip = ip->i_itemp;
3484 struct xfs_dinode *dip;
3485 struct xfs_mount *mp = ip->i_mount;
3487 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3488 ASSERT(xfs_isiflocked(ip));
3489 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3490 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3491 ASSERT(iip != NULL && iip->ili_fields != 0);
3492 ASSERT(ip->i_d.di_version > 1);
3494 /* set *dip = inode's place in the buffer */
3495 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3497 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3498 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3499 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3500 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3501 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3504 if (S_ISREG(VFS_I(ip)->i_mode)) {
3506 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3507 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3508 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3509 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3510 "%s: Bad regular inode %Lu, ptr 0x%p",
3511 __func__, ip->i_ino, ip);
3514 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3516 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3517 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3518 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3519 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3520 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3521 "%s: Bad directory inode %Lu, ptr 0x%p",
3522 __func__, ip->i_ino, ip);
3526 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3527 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3528 XFS_RANDOM_IFLUSH_5)) {
3529 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3530 "%s: detected corrupt incore inode %Lu, "
3531 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3532 __func__, ip->i_ino,
3533 ip->i_d.di_nextents + ip->i_d.di_anextents,
3534 ip->i_d.di_nblocks, ip);
3537 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3538 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3539 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3540 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3541 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3546 * Inode item log recovery for v2 inodes are dependent on the
3547 * di_flushiter count for correct sequencing. We bump the flush
3548 * iteration count so we can detect flushes which postdate a log record
3549 * during recovery. This is redundant as we now log every change and
3550 * hence this can't happen but we need to still do it to ensure
3551 * backwards compatibility with old kernels that predate logging all
3554 if (ip->i_d.di_version < 3)
3555 ip->i_d.di_flushiter++;
3558 * Copy the dirty parts of the inode into the on-disk inode. We always
3559 * copy out the core of the inode, because if the inode is dirty at all
3562 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3564 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3565 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3566 ip->i_d.di_flushiter = 0;
3568 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3569 if (XFS_IFORK_Q(ip))
3570 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3571 xfs_inobp_check(mp, bp);
3574 * We've recorded everything logged in the inode, so we'd like to clear
3575 * the ili_fields bits so we don't log and flush things unnecessarily.
3576 * However, we can't stop logging all this information until the data
3577 * we've copied into the disk buffer is written to disk. If we did we
3578 * might overwrite the copy of the inode in the log with all the data
3579 * after re-logging only part of it, and in the face of a crash we
3580 * wouldn't have all the data we need to recover.
3582 * What we do is move the bits to the ili_last_fields field. When
3583 * logging the inode, these bits are moved back to the ili_fields field.
3584 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3585 * know that the information those bits represent is permanently on
3586 * disk. As long as the flush completes before the inode is logged
3587 * again, then both ili_fields and ili_last_fields will be cleared.
3589 * We can play with the ili_fields bits here, because the inode lock
3590 * must be held exclusively in order to set bits there and the flush
3591 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3592 * done routine can tell whether or not to look in the AIL. Also, store
3593 * the current LSN of the inode so that we can tell whether the item has
3594 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3595 * need the AIL lock, because it is a 64 bit value that cannot be read
3598 iip->ili_last_fields = iip->ili_fields;
3599 iip->ili_fields = 0;
3600 iip->ili_fsync_fields = 0;
3601 iip->ili_logged = 1;
3603 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3604 &iip->ili_item.li_lsn);
3607 * Attach the function xfs_iflush_done to the inode's
3608 * buffer. This will remove the inode from the AIL
3609 * and unlock the inode's flush lock when the inode is
3610 * completely written to disk.
3612 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3614 /* generate the checksum. */
3615 xfs_dinode_calc_crc(mp, dip);
3617 ASSERT(bp->b_fspriv != NULL);
3618 ASSERT(bp->b_iodone != NULL);
3622 return -EFSCORRUPTED;