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_format.h"
23 #include "xfs_shared.h"
26 #include "xfs_trans.h"
27 #include "xfs_trans_space.h"
28 #include "xfs_trans_priv.h"
31 #include "xfs_mount.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_dir2_format.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38 #include "xfs_attr_sf.h"
40 #include "xfs_dinode.h"
41 #include "xfs_inode.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_btree.h"
45 #include "xfs_alloc.h"
46 #include "xfs_ialloc.h"
48 #include "xfs_bmap_util.h"
49 #include "xfs_error.h"
50 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
52 #include "xfs_cksum.h"
53 #include "xfs_trace.h"
54 #include "xfs_icache.h"
55 #include "xfs_symlink.h"
57 kmem_zone_t *xfs_inode_zone;
60 * Used in xfs_itruncate_extents(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
68 * helper function to extract extent size hint from inode
74 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
75 return ip->i_d.di_extsize;
76 if (XFS_IS_REALTIME_INODE(ip))
77 return ip->i_mount->m_sb.sb_rextsize;
82 * This is a wrapper routine around the xfs_ilock() routine used to centralize
83 * some grungy code. It is used in places that wish to lock the inode solely
84 * for reading the extents. The reason these places can't just call
85 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
86 * extents from disk for a file in b-tree format. If the inode is in b-tree
87 * format, then we need to lock the inode exclusively until the extents are read
88 * in. Locking it exclusively all the time would limit our parallelism
89 * unnecessarily, though. What we do instead is check to see if the extents
90 * have been read in yet, and only lock the inode exclusively if they have not.
92 * The function returns a value which should be given to the corresponding
93 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
102 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
103 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
104 lock_mode = XFS_ILOCK_EXCL;
106 lock_mode = XFS_ILOCK_SHARED;
109 xfs_ilock(ip, lock_mode);
115 * This is simply the unlock routine to go with xfs_ilock_map_shared().
116 * All it does is call xfs_iunlock() with the given lock_mode.
119 xfs_iunlock_map_shared(
121 unsigned int lock_mode)
123 xfs_iunlock(ip, lock_mode);
127 * The xfs inode contains 2 locks: a multi-reader lock called the
128 * i_iolock and a multi-reader lock called the i_lock. This routine
129 * allows either or both of the locks to be obtained.
131 * The 2 locks should always be ordered so that the IO lock is
132 * obtained first in order to prevent deadlock.
134 * ip -- the inode being locked
135 * lock_flags -- this parameter indicates the inode's locks
136 * to be locked. It can be:
141 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
142 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
143 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
144 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
151 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
154 * You can't set both SHARED and EXCL for the same lock,
155 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
156 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
158 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
159 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
160 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
161 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
162 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
164 if (lock_flags & XFS_IOLOCK_EXCL)
165 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
166 else if (lock_flags & XFS_IOLOCK_SHARED)
167 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
169 if (lock_flags & XFS_ILOCK_EXCL)
170 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
171 else if (lock_flags & XFS_ILOCK_SHARED)
172 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
176 * This is just like xfs_ilock(), except that the caller
177 * is guaranteed not to sleep. It returns 1 if it gets
178 * the requested locks and 0 otherwise. If the IO lock is
179 * obtained but the inode lock cannot be, then the IO lock
180 * is dropped before returning.
182 * ip -- the inode being locked
183 * lock_flags -- this parameter indicates the inode's locks to be
184 * to be locked. See the comment for xfs_ilock() for a list
192 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
195 * You can't set both SHARED and EXCL for the same lock,
196 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
197 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
199 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
200 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
201 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
202 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
203 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
205 if (lock_flags & XFS_IOLOCK_EXCL) {
206 if (!mrtryupdate(&ip->i_iolock))
208 } else if (lock_flags & XFS_IOLOCK_SHARED) {
209 if (!mrtryaccess(&ip->i_iolock))
212 if (lock_flags & XFS_ILOCK_EXCL) {
213 if (!mrtryupdate(&ip->i_lock))
214 goto out_undo_iolock;
215 } else if (lock_flags & XFS_ILOCK_SHARED) {
216 if (!mrtryaccess(&ip->i_lock))
217 goto out_undo_iolock;
222 if (lock_flags & XFS_IOLOCK_EXCL)
223 mrunlock_excl(&ip->i_iolock);
224 else if (lock_flags & XFS_IOLOCK_SHARED)
225 mrunlock_shared(&ip->i_iolock);
231 * xfs_iunlock() is used to drop the inode locks acquired with
232 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
233 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
234 * that we know which locks to drop.
236 * ip -- the inode being unlocked
237 * lock_flags -- this parameter indicates the inode's locks to be
238 * to be unlocked. See the comment for xfs_ilock() for a list
239 * of valid values for this parameter.
248 * You can't set both SHARED and EXCL for the same lock,
249 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
250 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
252 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
253 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
254 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
255 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
256 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
257 ASSERT(lock_flags != 0);
259 if (lock_flags & XFS_IOLOCK_EXCL)
260 mrunlock_excl(&ip->i_iolock);
261 else if (lock_flags & XFS_IOLOCK_SHARED)
262 mrunlock_shared(&ip->i_iolock);
264 if (lock_flags & XFS_ILOCK_EXCL)
265 mrunlock_excl(&ip->i_lock);
266 else if (lock_flags & XFS_ILOCK_SHARED)
267 mrunlock_shared(&ip->i_lock);
269 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
273 * give up write locks. the i/o lock cannot be held nested
274 * if it is being demoted.
281 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
282 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
284 if (lock_flags & XFS_ILOCK_EXCL)
285 mrdemote(&ip->i_lock);
286 if (lock_flags & XFS_IOLOCK_EXCL)
287 mrdemote(&ip->i_iolock);
289 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
292 #if defined(DEBUG) || defined(XFS_WARN)
298 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
299 if (!(lock_flags & XFS_ILOCK_SHARED))
300 return !!ip->i_lock.mr_writer;
301 return rwsem_is_locked(&ip->i_lock.mr_lock);
304 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
305 if (!(lock_flags & XFS_IOLOCK_SHARED))
306 return !!ip->i_iolock.mr_writer;
307 return rwsem_is_locked(&ip->i_iolock.mr_lock);
317 int xfs_small_retries;
318 int xfs_middle_retries;
319 int xfs_lots_retries;
324 * Bump the subclass so xfs_lock_inodes() acquires each lock with
328 xfs_lock_inumorder(int lock_mode, int subclass)
330 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
331 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
332 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
333 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
339 * The following routine will lock n inodes in exclusive mode.
340 * We assume the caller calls us with the inodes in i_ino order.
342 * We need to detect deadlock where an inode that we lock
343 * is in the AIL and we start waiting for another inode that is locked
344 * by a thread in a long running transaction (such as truncate). This can
345 * result in deadlock since the long running trans might need to wait
346 * for the inode we just locked in order to push the tail and free space
355 int attempts = 0, i, j, try_lock;
358 ASSERT(ips && (inodes >= 2)); /* we need at least two */
364 for (; i < inodes; i++) {
367 if (i && (ips[i] == ips[i-1])) /* Already locked */
371 * If try_lock is not set yet, make sure all locked inodes
372 * are not in the AIL.
373 * If any are, set try_lock to be used later.
377 for (j = (i - 1); j >= 0 && !try_lock; j--) {
378 lp = (xfs_log_item_t *)ips[j]->i_itemp;
379 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
386 * If any of the previous locks we have locked is in the AIL,
387 * we must TRY to get the second and subsequent locks. If
388 * we can't get any, we must release all we have
393 /* try_lock must be 0 if i is 0. */
395 * try_lock means we have an inode locked
396 * that is in the AIL.
399 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
403 * Unlock all previous guys and try again.
404 * xfs_iunlock will try to push the tail
405 * if the inode is in the AIL.
408 for(j = i - 1; j >= 0; j--) {
411 * Check to see if we've already
413 * Not the first one going back,
414 * and the inode ptr is the same.
416 if ((j != (i - 1)) && ips[j] ==
420 xfs_iunlock(ips[j], lock_mode);
423 if ((attempts % 5) == 0) {
424 delay(1); /* Don't just spin the CPU */
434 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
440 if (attempts < 5) xfs_small_retries++;
441 else if (attempts < 100) xfs_middle_retries++;
442 else xfs_lots_retries++;
450 * xfs_lock_two_inodes() can only be used to lock one type of lock
451 * at a time - the iolock or the ilock, but not both at once. If
452 * we lock both at once, lockdep will report false positives saying
453 * we have violated locking orders.
465 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
466 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
467 ASSERT(ip0->i_ino != ip1->i_ino);
469 if (ip0->i_ino > ip1->i_ino) {
476 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
479 * If the first lock we have locked is in the AIL, we must TRY to get
480 * the second lock. If we can't get it, we must release the first one
483 lp = (xfs_log_item_t *)ip0->i_itemp;
484 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
485 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
486 xfs_iunlock(ip0, lock_mode);
487 if ((++attempts % 5) == 0)
488 delay(1); /* Don't just spin the CPU */
492 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
499 struct xfs_inode *ip)
501 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
502 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
505 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
506 if (xfs_isiflocked(ip))
508 } while (!xfs_iflock_nowait(ip));
510 finish_wait(wq, &wait.wait);
519 if (di_flags & XFS_DIFLAG_ANY) {
520 if (di_flags & XFS_DIFLAG_REALTIME)
521 flags |= XFS_XFLAG_REALTIME;
522 if (di_flags & XFS_DIFLAG_PREALLOC)
523 flags |= XFS_XFLAG_PREALLOC;
524 if (di_flags & XFS_DIFLAG_IMMUTABLE)
525 flags |= XFS_XFLAG_IMMUTABLE;
526 if (di_flags & XFS_DIFLAG_APPEND)
527 flags |= XFS_XFLAG_APPEND;
528 if (di_flags & XFS_DIFLAG_SYNC)
529 flags |= XFS_XFLAG_SYNC;
530 if (di_flags & XFS_DIFLAG_NOATIME)
531 flags |= XFS_XFLAG_NOATIME;
532 if (di_flags & XFS_DIFLAG_NODUMP)
533 flags |= XFS_XFLAG_NODUMP;
534 if (di_flags & XFS_DIFLAG_RTINHERIT)
535 flags |= XFS_XFLAG_RTINHERIT;
536 if (di_flags & XFS_DIFLAG_PROJINHERIT)
537 flags |= XFS_XFLAG_PROJINHERIT;
538 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
539 flags |= XFS_XFLAG_NOSYMLINKS;
540 if (di_flags & XFS_DIFLAG_EXTSIZE)
541 flags |= XFS_XFLAG_EXTSIZE;
542 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
543 flags |= XFS_XFLAG_EXTSZINHERIT;
544 if (di_flags & XFS_DIFLAG_NODEFRAG)
545 flags |= XFS_XFLAG_NODEFRAG;
546 if (di_flags & XFS_DIFLAG_FILESTREAM)
547 flags |= XFS_XFLAG_FILESTREAM;
557 xfs_icdinode_t *dic = &ip->i_d;
559 return _xfs_dic2xflags(dic->di_flags) |
560 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
567 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
568 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
572 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
573 * is allowed, otherwise it has to be an exact match. If a CI match is found,
574 * ci_name->name will point to a the actual name (caller must free) or
575 * will be set to NULL if an exact match is found.
580 struct xfs_name *name,
582 struct xfs_name *ci_name)
588 trace_xfs_lookup(dp, name);
590 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
591 return XFS_ERROR(EIO);
593 lock_mode = xfs_ilock_map_shared(dp);
594 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
595 xfs_iunlock_map_shared(dp, lock_mode);
600 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
608 kmem_free(ci_name->name);
615 * Allocate an inode on disk and return a copy of its in-core version.
616 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
617 * appropriately within the inode. The uid and gid for the inode are
618 * set according to the contents of the given cred structure.
620 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
621 * has a free inode available, call xfs_iget() to obtain the in-core
622 * version of the allocated inode. Finally, fill in the inode and
623 * log its initial contents. In this case, ialloc_context would be
626 * If xfs_dialloc() does not have an available inode, it will replenish
627 * its supply by doing an allocation. Since we can only do one
628 * allocation within a transaction without deadlocks, we must commit
629 * the current transaction before returning the inode itself.
630 * In this case, therefore, we will set ialloc_context and return.
631 * The caller should then commit the current transaction, start a new
632 * transaction, and call xfs_ialloc() again to actually get the inode.
634 * To ensure that some other process does not grab the inode that
635 * was allocated during the first call to xfs_ialloc(), this routine
636 * also returns the [locked] bp pointing to the head of the freelist
637 * as ialloc_context. The caller should hold this buffer across
638 * the commit and pass it back into this routine on the second call.
640 * If we are allocating quota inodes, we do not have a parent inode
641 * to attach to or associate with (i.e. pip == NULL) because they
642 * are not linked into the directory structure - they are attached
643 * directly to the superblock - and so have no parent.
654 xfs_buf_t **ialloc_context,
657 struct xfs_mount *mp = tp->t_mountp;
666 * Call the space management code to pick
667 * the on-disk inode to be allocated.
669 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
670 ialloc_context, &ino);
673 if (*ialloc_context || ino == NULLFSINO) {
677 ASSERT(*ialloc_context == NULL);
680 * Get the in-core inode with the lock held exclusively.
681 * This is because we're setting fields here we need
682 * to prevent others from looking at until we're done.
684 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
685 XFS_ILOCK_EXCL, &ip);
690 ip->i_d.di_mode = mode;
691 ip->i_d.di_onlink = 0;
692 ip->i_d.di_nlink = nlink;
693 ASSERT(ip->i_d.di_nlink == nlink);
694 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
695 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
696 xfs_set_projid(ip, prid);
697 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
700 * If the superblock version is up to where we support new format
701 * inodes and this is currently an old format inode, then change
702 * the inode version number now. This way we only do the conversion
703 * here rather than here and in the flush/logging code.
705 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
706 ip->i_d.di_version == 1) {
707 ip->i_d.di_version = 2;
709 * We've already zeroed the old link count, the projid field,
715 * Project ids won't be stored on disk if we are using a version 1 inode.
717 if ((prid != 0) && (ip->i_d.di_version == 1))
718 xfs_bump_ino_vers2(tp, ip);
720 if (pip && XFS_INHERIT_GID(pip)) {
721 ip->i_d.di_gid = pip->i_d.di_gid;
722 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
723 ip->i_d.di_mode |= S_ISGID;
728 * If the group ID of the new file does not match the effective group
729 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
730 * (and only if the irix_sgid_inherit compatibility variable is set).
732 if ((irix_sgid_inherit) &&
733 (ip->i_d.di_mode & S_ISGID) &&
734 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
735 ip->i_d.di_mode &= ~S_ISGID;
739 ip->i_d.di_nextents = 0;
740 ASSERT(ip->i_d.di_nblocks == 0);
743 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
744 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
745 ip->i_d.di_atime = ip->i_d.di_mtime;
746 ip->i_d.di_ctime = ip->i_d.di_mtime;
749 * di_gen will have been taken care of in xfs_iread.
751 ip->i_d.di_extsize = 0;
752 ip->i_d.di_dmevmask = 0;
753 ip->i_d.di_dmstate = 0;
754 ip->i_d.di_flags = 0;
756 if (ip->i_d.di_version == 3) {
757 ASSERT(ip->i_d.di_ino == ino);
758 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
760 ip->i_d.di_changecount = 1;
762 ip->i_d.di_flags2 = 0;
763 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
764 ip->i_d.di_crtime = ip->i_d.di_mtime;
768 flags = XFS_ILOG_CORE;
769 switch (mode & S_IFMT) {
774 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
775 ip->i_df.if_u2.if_rdev = rdev;
776 ip->i_df.if_flags = 0;
777 flags |= XFS_ILOG_DEV;
781 * we can't set up filestreams until after the VFS inode
782 * is set up properly.
784 if (pip && xfs_inode_is_filestream(pip))
788 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
792 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
793 di_flags |= XFS_DIFLAG_RTINHERIT;
794 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
795 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
796 ip->i_d.di_extsize = pip->i_d.di_extsize;
798 } else if (S_ISREG(mode)) {
799 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
800 di_flags |= XFS_DIFLAG_REALTIME;
801 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
802 di_flags |= XFS_DIFLAG_EXTSIZE;
803 ip->i_d.di_extsize = pip->i_d.di_extsize;
806 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
808 di_flags |= XFS_DIFLAG_NOATIME;
809 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
811 di_flags |= XFS_DIFLAG_NODUMP;
812 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
814 di_flags |= XFS_DIFLAG_SYNC;
815 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
816 xfs_inherit_nosymlinks)
817 di_flags |= XFS_DIFLAG_NOSYMLINKS;
818 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
819 di_flags |= XFS_DIFLAG_PROJINHERIT;
820 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
821 xfs_inherit_nodefrag)
822 di_flags |= XFS_DIFLAG_NODEFRAG;
823 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
824 di_flags |= XFS_DIFLAG_FILESTREAM;
825 ip->i_d.di_flags |= di_flags;
829 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
830 ip->i_df.if_flags = XFS_IFEXTENTS;
831 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
832 ip->i_df.if_u1.if_extents = NULL;
838 * Attribute fork settings for new inode.
840 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
841 ip->i_d.di_anextents = 0;
844 * Log the new values stuffed into the inode.
846 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
847 xfs_trans_log_inode(tp, ip, flags);
849 /* now that we have an i_mode we can setup inode ops and unlock */
852 /* now we have set up the vfs inode we can associate the filestream */
854 error = xfs_filestream_associate(pip, ip);
858 xfs_iflags_set(ip, XFS_IFILESTREAM);
866 * Allocates a new inode from disk and return a pointer to the
867 * incore copy. This routine will internally commit the current
868 * transaction and allocate a new one if the Space Manager needed
869 * to do an allocation to replenish the inode free-list.
871 * This routine is designed to be called from xfs_create and
877 xfs_trans_t **tpp, /* input: current transaction;
878 output: may be a new transaction. */
879 xfs_inode_t *dp, /* directory within whose allocate
884 prid_t prid, /* project id */
885 int okalloc, /* ok to allocate new space */
886 xfs_inode_t **ipp, /* pointer to inode; it will be
894 xfs_buf_t *ialloc_context = NULL;
900 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
903 * xfs_ialloc will return a pointer to an incore inode if
904 * the Space Manager has an available inode on the free
905 * list. Otherwise, it will do an allocation and replenish
906 * the freelist. Since we can only do one allocation per
907 * transaction without deadlocks, we will need to commit the
908 * current transaction and start a new one. We will then
909 * need to call xfs_ialloc again to get the inode.
911 * If xfs_ialloc did an allocation to replenish the freelist,
912 * it returns the bp containing the head of the freelist as
913 * ialloc_context. We will hold a lock on it across the
914 * transaction commit so that no other process can steal
915 * the inode(s) that we've just allocated.
917 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
918 &ialloc_context, &ip);
921 * Return an error if we were unable to allocate a new inode.
922 * This should only happen if we run out of space on disk or
923 * encounter a disk error.
929 if (!ialloc_context && !ip) {
931 return XFS_ERROR(ENOSPC);
935 * If the AGI buffer is non-NULL, then we were unable to get an
936 * inode in one operation. We need to commit the current
937 * transaction and call xfs_ialloc() again. It is guaranteed
938 * to succeed the second time.
940 if (ialloc_context) {
941 struct xfs_trans_res tres;
944 * Normally, xfs_trans_commit releases all the locks.
945 * We call bhold to hang on to the ialloc_context across
946 * the commit. Holding this buffer prevents any other
947 * processes from doing any allocations in this
950 xfs_trans_bhold(tp, ialloc_context);
952 * Save the log reservation so we can use
953 * them in the next transaction.
955 tres.tr_logres = xfs_trans_get_log_res(tp);
956 tres.tr_logcount = xfs_trans_get_log_count(tp);
959 * We want the quota changes to be associated with the next
960 * transaction, NOT this one. So, detach the dqinfo from this
961 * and attach it to the next transaction.
966 dqinfo = (void *)tp->t_dqinfo;
968 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
969 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
972 ntp = xfs_trans_dup(tp);
973 code = xfs_trans_commit(tp, 0);
975 if (committed != NULL) {
979 * If we get an error during the commit processing,
980 * release the buffer that is still held and return
984 xfs_buf_relse(ialloc_context);
986 tp->t_dqinfo = dqinfo;
987 xfs_trans_free_dqinfo(tp);
995 * transaction commit worked ok so we can drop the extra ticket
996 * reference that we gained in xfs_trans_dup()
998 xfs_log_ticket_put(tp->t_ticket);
999 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1000 code = xfs_trans_reserve(tp, &tres, 0, 0);
1003 * Re-attach the quota info that we detached from prev trx.
1006 tp->t_dqinfo = dqinfo;
1007 tp->t_flags |= tflags;
1011 xfs_buf_relse(ialloc_context);
1016 xfs_trans_bjoin(tp, ialloc_context);
1019 * Call ialloc again. Since we've locked out all
1020 * other allocations in this allocation group,
1021 * this call should always succeed.
1023 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1024 okalloc, &ialloc_context, &ip);
1027 * If we get an error at this point, return to the caller
1028 * so that the current transaction can be aborted.
1035 ASSERT(!ialloc_context && ip);
1038 if (committed != NULL)
1049 * Decrement the link count on an inode & log the change.
1050 * If this causes the link count to go to zero, initiate the
1051 * logging activity required to truncate a file.
1060 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1062 ASSERT (ip->i_d.di_nlink > 0);
1064 drop_nlink(VFS_I(ip));
1065 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1068 if (ip->i_d.di_nlink == 0) {
1070 * We're dropping the last link to this file.
1071 * Move the on-disk inode to the AGI unlinked list.
1072 * From xfs_inactive() we will pull the inode from
1073 * the list and free it.
1075 error = xfs_iunlink(tp, ip);
1081 * This gets called when the inode's version needs to be changed from 1 to 2.
1082 * Currently this happens when the nlink field overflows the old 16-bit value
1083 * or when chproj is called to change the project for the first time.
1084 * As a side effect the superblock version will also get rev'd
1085 * to contain the NLINK bit.
1094 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1095 ASSERT(ip->i_d.di_version == 1);
1097 ip->i_d.di_version = 2;
1098 ip->i_d.di_onlink = 0;
1099 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1101 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1102 spin_lock(&mp->m_sb_lock);
1103 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1104 xfs_sb_version_addnlink(&mp->m_sb);
1105 spin_unlock(&mp->m_sb_lock);
1106 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1108 spin_unlock(&mp->m_sb_lock);
1111 /* Caller must log the inode */
1115 * Increment the link count on an inode & log the change.
1122 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1124 ASSERT(ip->i_d.di_nlink > 0);
1126 inc_nlink(VFS_I(ip));
1127 if ((ip->i_d.di_version == 1) &&
1128 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1130 * The inode has increased its number of links beyond
1131 * what can fit in an old format inode. It now needs
1132 * to be converted to a version 2 inode with a 32 bit
1133 * link count. If this is the first inode in the file
1134 * system to do this, then we need to bump the superblock
1135 * version number as well.
1137 xfs_bump_ino_vers2(tp, ip);
1140 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1147 struct xfs_name *name,
1152 int is_dir = S_ISDIR(mode);
1153 struct xfs_mount *mp = dp->i_mount;
1154 struct xfs_inode *ip = NULL;
1155 struct xfs_trans *tp = NULL;
1157 xfs_bmap_free_t free_list;
1158 xfs_fsblock_t first_block;
1159 bool unlock_dp_on_error = false;
1163 struct xfs_dquot *udqp = NULL;
1164 struct xfs_dquot *gdqp = NULL;
1165 struct xfs_dquot *pdqp = NULL;
1166 struct xfs_trans_res tres;
1169 trace_xfs_create(dp, name);
1171 if (XFS_FORCED_SHUTDOWN(mp))
1172 return XFS_ERROR(EIO);
1174 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1175 prid = xfs_get_projid(dp);
1177 prid = XFS_PROJID_DEFAULT;
1180 * Make sure that we have allocated dquot(s) on disk.
1182 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1183 xfs_kgid_to_gid(current_fsgid()), prid,
1184 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1185 &udqp, &gdqp, &pdqp);
1191 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1192 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1193 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1194 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1196 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1197 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1198 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1199 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1202 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1205 * Initially assume that the file does not exist and
1206 * reserve the resources for that case. If that is not
1207 * the case we'll drop the one we have and get a more
1208 * appropriate transaction later.
1210 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1211 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1212 if (error == ENOSPC) {
1213 /* flush outstanding delalloc blocks and retry */
1214 xfs_flush_inodes(mp);
1215 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1217 if (error == ENOSPC) {
1218 /* No space at all so try a "no-allocation" reservation */
1220 error = xfs_trans_reserve(tp, &tres, 0, 0);
1224 goto out_trans_cancel;
1227 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1228 unlock_dp_on_error = true;
1230 xfs_bmap_init(&free_list, &first_block);
1233 * Reserve disk quota and the inode.
1235 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1236 pdqp, resblks, 1, 0);
1238 goto out_trans_cancel;
1240 error = xfs_dir_canenter(tp, dp, name, resblks);
1242 goto out_trans_cancel;
1245 * A newly created regular or special file just has one directory
1246 * entry pointing to them, but a directory also the "." entry
1247 * pointing to itself.
1249 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1250 prid, resblks > 0, &ip, &committed);
1252 if (error == ENOSPC)
1253 goto out_trans_cancel;
1254 goto out_trans_abort;
1258 * Now we join the directory inode to the transaction. We do not do it
1259 * earlier because xfs_dir_ialloc might commit the previous transaction
1260 * (and release all the locks). An error from here on will result in
1261 * the transaction cancel unlocking dp so don't do it explicitly in the
1264 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1265 unlock_dp_on_error = false;
1267 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1268 &first_block, &free_list, resblks ?
1269 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1271 ASSERT(error != ENOSPC);
1272 goto out_trans_abort;
1274 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1275 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1278 error = xfs_dir_init(tp, ip, dp);
1280 goto out_bmap_cancel;
1282 error = xfs_bumplink(tp, dp);
1284 goto out_bmap_cancel;
1288 * If this is a synchronous mount, make sure that the
1289 * create transaction goes to disk before returning to
1292 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1293 xfs_trans_set_sync(tp);
1296 * Attach the dquot(s) to the inodes and modify them incore.
1297 * These ids of the inode couldn't have changed since the new
1298 * inode has been locked ever since it was created.
1300 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1302 error = xfs_bmap_finish(&tp, &free_list, &committed);
1304 goto out_bmap_cancel;
1306 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1308 goto out_release_inode;
1310 xfs_qm_dqrele(udqp);
1311 xfs_qm_dqrele(gdqp);
1312 xfs_qm_dqrele(pdqp);
1318 xfs_bmap_cancel(&free_list);
1320 cancel_flags |= XFS_TRANS_ABORT;
1322 xfs_trans_cancel(tp, cancel_flags);
1325 * Wait until after the current transaction is aborted to
1326 * release the inode. This prevents recursive transactions
1327 * and deadlocks from xfs_inactive.
1332 xfs_qm_dqrele(udqp);
1333 xfs_qm_dqrele(gdqp);
1334 xfs_qm_dqrele(pdqp);
1336 if (unlock_dp_on_error)
1337 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1345 struct xfs_name *target_name)
1347 xfs_mount_t *mp = tdp->i_mount;
1350 xfs_bmap_free_t free_list;
1351 xfs_fsblock_t first_block;
1356 trace_xfs_link(tdp, target_name);
1358 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1360 if (XFS_FORCED_SHUTDOWN(mp))
1361 return XFS_ERROR(EIO);
1363 error = xfs_qm_dqattach(sip, 0);
1367 error = xfs_qm_dqattach(tdp, 0);
1371 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1372 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1373 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1374 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1375 if (error == ENOSPC) {
1377 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1384 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1386 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1387 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1390 * If we are using project inheritance, we only allow hard link
1391 * creation in our tree when the project IDs are the same; else
1392 * the tree quota mechanism could be circumvented.
1394 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1395 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1396 error = XFS_ERROR(EXDEV);
1400 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1404 xfs_bmap_init(&free_list, &first_block);
1406 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1407 &first_block, &free_list, resblks);
1410 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1411 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1413 error = xfs_bumplink(tp, sip);
1418 * If this is a synchronous mount, make sure that the
1419 * link transaction goes to disk before returning to
1422 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1423 xfs_trans_set_sync(tp);
1426 error = xfs_bmap_finish (&tp, &free_list, &committed);
1428 xfs_bmap_cancel(&free_list);
1432 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1435 cancel_flags |= XFS_TRANS_ABORT;
1437 xfs_trans_cancel(tp, cancel_flags);
1443 * Free up the underlying blocks past new_size. The new size must be smaller
1444 * than the current size. This routine can be used both for the attribute and
1445 * data fork, and does not modify the inode size, which is left to the caller.
1447 * The transaction passed to this routine must have made a permanent log
1448 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1449 * given transaction and start new ones, so make sure everything involved in
1450 * the transaction is tidy before calling here. Some transaction will be
1451 * returned to the caller to be committed. The incoming transaction must
1452 * already include the inode, and both inode locks must be held exclusively.
1453 * The inode must also be "held" within the transaction. On return the inode
1454 * will be "held" within the returned transaction. This routine does NOT
1455 * require any disk space to be reserved for it within the transaction.
1457 * If we get an error, we must return with the inode locked and linked into the
1458 * current transaction. This keeps things simple for the higher level code,
1459 * because it always knows that the inode is locked and held in the transaction
1460 * that returns to it whether errors occur or not. We don't mark the inode
1461 * dirty on error so that transactions can be easily aborted if possible.
1464 xfs_itruncate_extents(
1465 struct xfs_trans **tpp,
1466 struct xfs_inode *ip,
1468 xfs_fsize_t new_size)
1470 struct xfs_mount *mp = ip->i_mount;
1471 struct xfs_trans *tp = *tpp;
1472 struct xfs_trans *ntp;
1473 xfs_bmap_free_t free_list;
1474 xfs_fsblock_t first_block;
1475 xfs_fileoff_t first_unmap_block;
1476 xfs_fileoff_t last_block;
1477 xfs_filblks_t unmap_len;
1482 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1483 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1484 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1485 ASSERT(new_size <= XFS_ISIZE(ip));
1486 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1487 ASSERT(ip->i_itemp != NULL);
1488 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1489 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1491 trace_xfs_itruncate_extents_start(ip, new_size);
1494 * Since it is possible for space to become allocated beyond
1495 * the end of the file (in a crash where the space is allocated
1496 * but the inode size is not yet updated), simply remove any
1497 * blocks which show up between the new EOF and the maximum
1498 * possible file size. If the first block to be removed is
1499 * beyond the maximum file size (ie it is the same as last_block),
1500 * then there is nothing to do.
1502 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1503 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1504 if (first_unmap_block == last_block)
1507 ASSERT(first_unmap_block < last_block);
1508 unmap_len = last_block - first_unmap_block + 1;
1510 xfs_bmap_init(&free_list, &first_block);
1511 error = xfs_bunmapi(tp, ip,
1512 first_unmap_block, unmap_len,
1513 xfs_bmapi_aflag(whichfork),
1514 XFS_ITRUNC_MAX_EXTENTS,
1515 &first_block, &free_list,
1518 goto out_bmap_cancel;
1521 * Duplicate the transaction that has the permanent
1522 * reservation and commit the old transaction.
1524 error = xfs_bmap_finish(&tp, &free_list, &committed);
1526 xfs_trans_ijoin(tp, ip, 0);
1528 goto out_bmap_cancel;
1532 * Mark the inode dirty so it will be logged and
1533 * moved forward in the log as part of every commit.
1535 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1538 ntp = xfs_trans_dup(tp);
1539 error = xfs_trans_commit(tp, 0);
1542 xfs_trans_ijoin(tp, ip, 0);
1548 * Transaction commit worked ok so we can drop the extra ticket
1549 * reference that we gained in xfs_trans_dup()
1551 xfs_log_ticket_put(tp->t_ticket);
1552 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1558 * Always re-log the inode so that our permanent transaction can keep
1559 * on rolling it forward in the log.
1561 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1563 trace_xfs_itruncate_extents_end(ip, new_size);
1570 * If the bunmapi call encounters an error, return to the caller where
1571 * the transaction can be properly aborted. We just need to make sure
1572 * we're not holding any resources that we were not when we came in.
1574 xfs_bmap_cancel(&free_list);
1582 xfs_mount_t *mp = ip->i_mount;
1585 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1588 /* If this is a read-only mount, don't do this (would generate I/O) */
1589 if (mp->m_flags & XFS_MOUNT_RDONLY)
1592 if (!XFS_FORCED_SHUTDOWN(mp)) {
1596 * If we are using filestreams, and we have an unlinked
1597 * file that we are processing the last close on, then nothing
1598 * will be able to reopen and write to this file. Purge this
1599 * inode from the filestreams cache so that it doesn't delay
1600 * teardown of the inode.
1602 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1603 xfs_filestream_deassociate(ip);
1606 * If we previously truncated this file and removed old data
1607 * in the process, we want to initiate "early" writeout on
1608 * the last close. This is an attempt to combat the notorious
1609 * NULL files problem which is particularly noticeable from a
1610 * truncate down, buffered (re-)write (delalloc), followed by
1611 * a crash. What we are effectively doing here is
1612 * significantly reducing the time window where we'd otherwise
1613 * be exposed to that problem.
1615 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1617 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1618 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1619 error = -filemap_flush(VFS_I(ip)->i_mapping);
1626 if (ip->i_d.di_nlink == 0)
1629 if (xfs_can_free_eofblocks(ip, false)) {
1632 * If we can't get the iolock just skip truncating the blocks
1633 * past EOF because we could deadlock with the mmap_sem
1634 * otherwise. We'll get another chance to drop them once the
1635 * last reference to the inode is dropped, so we'll never leak
1636 * blocks permanently.
1638 * Further, check if the inode is being opened, written and
1639 * closed frequently and we have delayed allocation blocks
1640 * outstanding (e.g. streaming writes from the NFS server),
1641 * truncating the blocks past EOF will cause fragmentation to
1644 * In this case don't do the truncation, either, but we have to
1645 * be careful how we detect this case. Blocks beyond EOF show
1646 * up as i_delayed_blks even when the inode is clean, so we
1647 * need to truncate them away first before checking for a dirty
1648 * release. Hence on the first dirty close we will still remove
1649 * the speculative allocation, but after that we will leave it
1652 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1655 error = xfs_free_eofblocks(mp, ip, true);
1656 if (error && error != EAGAIN)
1659 /* delalloc blocks after truncation means it really is dirty */
1660 if (ip->i_delayed_blks)
1661 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1667 * xfs_inactive_truncate
1669 * Called to perform a truncate when an inode becomes unlinked.
1672 xfs_inactive_truncate(
1673 struct xfs_inode *ip)
1675 struct xfs_mount *mp = ip->i_mount;
1676 struct xfs_trans *tp;
1679 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1680 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1682 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1683 xfs_trans_cancel(tp, 0);
1687 xfs_ilock(ip, XFS_ILOCK_EXCL);
1688 xfs_trans_ijoin(tp, ip, 0);
1691 * Log the inode size first to prevent stale data exposure in the event
1692 * of a system crash before the truncate completes. See the related
1693 * comment in xfs_setattr_size() for details.
1695 ip->i_d.di_size = 0;
1696 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1698 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1700 goto error_trans_cancel;
1702 ASSERT(ip->i_d.di_nextents == 0);
1704 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1708 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1712 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1714 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1719 * xfs_inactive_ifree()
1721 * Perform the inode free when an inode is unlinked.
1725 struct xfs_inode *ip)
1727 xfs_bmap_free_t free_list;
1728 xfs_fsblock_t first_block;
1730 struct xfs_mount *mp = ip->i_mount;
1731 struct xfs_trans *tp;
1734 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1735 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
1737 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1738 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1742 xfs_ilock(ip, XFS_ILOCK_EXCL);
1743 xfs_trans_ijoin(tp, ip, 0);
1745 xfs_bmap_init(&free_list, &first_block);
1746 error = xfs_ifree(tp, ip, &free_list);
1749 * If we fail to free the inode, shut down. The cancel
1750 * might do that, we need to make sure. Otherwise the
1751 * inode might be lost for a long time or forever.
1753 if (!XFS_FORCED_SHUTDOWN(mp)) {
1754 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1756 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1758 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1759 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1764 * Credit the quota account(s). The inode is gone.
1766 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1769 * Just ignore errors at this point. There is nothing we can
1770 * do except to try to keep going. Make sure it's not a silent
1773 error = xfs_bmap_finish(&tp, &free_list, &committed);
1775 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1777 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1779 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1782 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1789 * This is called when the vnode reference count for the vnode
1790 * goes to zero. If the file has been unlinked, then it must
1791 * now be truncated. Also, we clear all of the read-ahead state
1792 * kept for the inode here since the file is now closed.
1798 struct xfs_mount *mp;
1803 * If the inode is already free, then there can be nothing
1806 if (ip->i_d.di_mode == 0) {
1807 ASSERT(ip->i_df.if_real_bytes == 0);
1808 ASSERT(ip->i_df.if_broot_bytes == 0);
1814 /* If this is a read-only mount, don't do this (would generate I/O) */
1815 if (mp->m_flags & XFS_MOUNT_RDONLY)
1818 if (ip->i_d.di_nlink != 0) {
1820 * force is true because we are evicting an inode from the
1821 * cache. Post-eof blocks must be freed, lest we end up with
1822 * broken free space accounting.
1824 if (xfs_can_free_eofblocks(ip, true))
1825 xfs_free_eofblocks(mp, ip, false);
1830 if (S_ISREG(ip->i_d.di_mode) &&
1831 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1832 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1835 error = xfs_qm_dqattach(ip, 0);
1839 if (S_ISLNK(ip->i_d.di_mode))
1840 error = xfs_inactive_symlink(ip);
1842 error = xfs_inactive_truncate(ip);
1847 * If there are attributes associated with the file then blow them away
1848 * now. The code calls a routine that recursively deconstructs the
1849 * attribute fork. We need to just commit the current transaction
1850 * because we can't use it for xfs_attr_inactive().
1852 if (ip->i_d.di_anextents > 0) {
1853 ASSERT(ip->i_d.di_forkoff != 0);
1855 error = xfs_attr_inactive(ip);
1861 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1863 ASSERT(ip->i_d.di_anextents == 0);
1868 error = xfs_inactive_ifree(ip);
1873 * Release the dquots held by inode, if any.
1875 xfs_qm_dqdetach(ip);
1879 * This is called when the inode's link count goes to 0.
1880 * We place the on-disk inode on a list in the AGI. It
1881 * will be pulled from this list when the inode is freed.
1898 ASSERT(ip->i_d.di_nlink == 0);
1899 ASSERT(ip->i_d.di_mode != 0);
1904 * Get the agi buffer first. It ensures lock ordering
1907 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1910 agi = XFS_BUF_TO_AGI(agibp);
1913 * Get the index into the agi hash table for the
1914 * list this inode will go on.
1916 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1918 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1919 ASSERT(agi->agi_unlinked[bucket_index]);
1920 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1922 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1924 * There is already another inode in the bucket we need
1925 * to add ourselves to. Add us at the front of the list.
1926 * Here we put the head pointer into our next pointer,
1927 * and then we fall through to point the head at us.
1929 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1934 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1935 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1936 offset = ip->i_imap.im_boffset +
1937 offsetof(xfs_dinode_t, di_next_unlinked);
1939 /* need to recalc the inode CRC if appropriate */
1940 xfs_dinode_calc_crc(mp, dip);
1942 xfs_trans_inode_buf(tp, ibp);
1943 xfs_trans_log_buf(tp, ibp, offset,
1944 (offset + sizeof(xfs_agino_t) - 1));
1945 xfs_inobp_check(mp, ibp);
1949 * Point the bucket head pointer at the inode being inserted.
1952 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1953 offset = offsetof(xfs_agi_t, agi_unlinked) +
1954 (sizeof(xfs_agino_t) * bucket_index);
1955 xfs_trans_log_buf(tp, agibp, offset,
1956 (offset + sizeof(xfs_agino_t) - 1));
1961 * Pull the on-disk inode from the AGI unlinked list.
1974 xfs_agnumber_t agno;
1976 xfs_agino_t next_agino;
1977 xfs_buf_t *last_ibp;
1978 xfs_dinode_t *last_dip = NULL;
1980 int offset, last_offset = 0;
1984 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1987 * Get the agi buffer first. It ensures lock ordering
1990 error = xfs_read_agi(mp, tp, agno, &agibp);
1994 agi = XFS_BUF_TO_AGI(agibp);
1997 * Get the index into the agi hash table for the
1998 * list this inode will go on.
2000 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2002 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2003 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2004 ASSERT(agi->agi_unlinked[bucket_index]);
2006 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2008 * We're at the head of the list. Get the inode's on-disk
2009 * buffer to see if there is anyone after us on the list.
2010 * Only modify our next pointer if it is not already NULLAGINO.
2011 * This saves us the overhead of dealing with the buffer when
2012 * there is no need to change it.
2014 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2017 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2021 next_agino = be32_to_cpu(dip->di_next_unlinked);
2022 ASSERT(next_agino != 0);
2023 if (next_agino != NULLAGINO) {
2024 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2025 offset = ip->i_imap.im_boffset +
2026 offsetof(xfs_dinode_t, di_next_unlinked);
2028 /* need to recalc the inode CRC if appropriate */
2029 xfs_dinode_calc_crc(mp, dip);
2031 xfs_trans_inode_buf(tp, ibp);
2032 xfs_trans_log_buf(tp, ibp, offset,
2033 (offset + sizeof(xfs_agino_t) - 1));
2034 xfs_inobp_check(mp, ibp);
2036 xfs_trans_brelse(tp, ibp);
2039 * Point the bucket head pointer at the next inode.
2041 ASSERT(next_agino != 0);
2042 ASSERT(next_agino != agino);
2043 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2044 offset = offsetof(xfs_agi_t, agi_unlinked) +
2045 (sizeof(xfs_agino_t) * bucket_index);
2046 xfs_trans_log_buf(tp, agibp, offset,
2047 (offset + sizeof(xfs_agino_t) - 1));
2050 * We need to search the list for the inode being freed.
2052 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2054 while (next_agino != agino) {
2055 struct xfs_imap imap;
2058 xfs_trans_brelse(tp, last_ibp);
2061 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2063 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2066 "%s: xfs_imap returned error %d.",
2071 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2075 "%s: xfs_imap_to_bp returned error %d.",
2080 last_offset = imap.im_boffset;
2081 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2082 ASSERT(next_agino != NULLAGINO);
2083 ASSERT(next_agino != 0);
2087 * Now last_ibp points to the buffer previous to us on the
2088 * unlinked list. Pull us from the list.
2090 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2093 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2097 next_agino = be32_to_cpu(dip->di_next_unlinked);
2098 ASSERT(next_agino != 0);
2099 ASSERT(next_agino != agino);
2100 if (next_agino != NULLAGINO) {
2101 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2102 offset = ip->i_imap.im_boffset +
2103 offsetof(xfs_dinode_t, di_next_unlinked);
2105 /* need to recalc the inode CRC if appropriate */
2106 xfs_dinode_calc_crc(mp, dip);
2108 xfs_trans_inode_buf(tp, ibp);
2109 xfs_trans_log_buf(tp, ibp, offset,
2110 (offset + sizeof(xfs_agino_t) - 1));
2111 xfs_inobp_check(mp, ibp);
2113 xfs_trans_brelse(tp, ibp);
2116 * Point the previous inode on the list to the next inode.
2118 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2119 ASSERT(next_agino != 0);
2120 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2122 /* need to recalc the inode CRC if appropriate */
2123 xfs_dinode_calc_crc(mp, last_dip);
2125 xfs_trans_inode_buf(tp, last_ibp);
2126 xfs_trans_log_buf(tp, last_ibp, offset,
2127 (offset + sizeof(xfs_agino_t) - 1));
2128 xfs_inobp_check(mp, last_ibp);
2134 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2135 * inodes that are in memory - they all must be marked stale and attached to
2136 * the cluster buffer.
2140 xfs_inode_t *free_ip,
2144 xfs_mount_t *mp = free_ip->i_mount;
2145 int blks_per_cluster;
2152 xfs_inode_log_item_t *iip;
2153 xfs_log_item_t *lip;
2154 struct xfs_perag *pag;
2156 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2157 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2158 blks_per_cluster = 1;
2159 ninodes = mp->m_sb.sb_inopblock;
2160 nbufs = XFS_IALLOC_BLOCKS(mp);
2162 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2163 mp->m_sb.sb_blocksize;
2164 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2165 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2168 for (j = 0; j < nbufs; j++, inum += ninodes) {
2169 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2170 XFS_INO_TO_AGBNO(mp, inum));
2173 * We obtain and lock the backing buffer first in the process
2174 * here, as we have to ensure that any dirty inode that we
2175 * can't get the flush lock on is attached to the buffer.
2176 * If we scan the in-memory inodes first, then buffer IO can
2177 * complete before we get a lock on it, and hence we may fail
2178 * to mark all the active inodes on the buffer stale.
2180 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2181 mp->m_bsize * blks_per_cluster,
2188 * This buffer may not have been correctly initialised as we
2189 * didn't read it from disk. That's not important because we are
2190 * only using to mark the buffer as stale in the log, and to
2191 * attach stale cached inodes on it. That means it will never be
2192 * dispatched for IO. If it is, we want to know about it, and we
2193 * want it to fail. We can acheive this by adding a write
2194 * verifier to the buffer.
2196 bp->b_ops = &xfs_inode_buf_ops;
2199 * Walk the inodes already attached to the buffer and mark them
2200 * stale. These will all have the flush locks held, so an
2201 * in-memory inode walk can't lock them. By marking them all
2202 * stale first, we will not attempt to lock them in the loop
2203 * below as the XFS_ISTALE flag will be set.
2207 if (lip->li_type == XFS_LI_INODE) {
2208 iip = (xfs_inode_log_item_t *)lip;
2209 ASSERT(iip->ili_logged == 1);
2210 lip->li_cb = xfs_istale_done;
2211 xfs_trans_ail_copy_lsn(mp->m_ail,
2212 &iip->ili_flush_lsn,
2213 &iip->ili_item.li_lsn);
2214 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2216 lip = lip->li_bio_list;
2221 * For each inode in memory attempt to add it to the inode
2222 * buffer and set it up for being staled on buffer IO
2223 * completion. This is safe as we've locked out tail pushing
2224 * and flushing by locking the buffer.
2226 * We have already marked every inode that was part of a
2227 * transaction stale above, which means there is no point in
2228 * even trying to lock them.
2230 for (i = 0; i < ninodes; i++) {
2233 ip = radix_tree_lookup(&pag->pag_ici_root,
2234 XFS_INO_TO_AGINO(mp, (inum + i)));
2236 /* Inode not in memory, nothing to do */
2243 * because this is an RCU protected lookup, we could
2244 * find a recently freed or even reallocated inode
2245 * during the lookup. We need to check under the
2246 * i_flags_lock for a valid inode here. Skip it if it
2247 * is not valid, the wrong inode or stale.
2249 spin_lock(&ip->i_flags_lock);
2250 if (ip->i_ino != inum + i ||
2251 __xfs_iflags_test(ip, XFS_ISTALE)) {
2252 spin_unlock(&ip->i_flags_lock);
2256 spin_unlock(&ip->i_flags_lock);
2259 * Don't try to lock/unlock the current inode, but we
2260 * _cannot_ skip the other inodes that we did not find
2261 * in the list attached to the buffer and are not
2262 * already marked stale. If we can't lock it, back off
2265 if (ip != free_ip &&
2266 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2274 xfs_iflags_set(ip, XFS_ISTALE);
2277 * we don't need to attach clean inodes or those only
2278 * with unlogged changes (which we throw away, anyway).
2281 if (!iip || xfs_inode_clean(ip)) {
2282 ASSERT(ip != free_ip);
2284 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2288 iip->ili_last_fields = iip->ili_fields;
2289 iip->ili_fields = 0;
2290 iip->ili_logged = 1;
2291 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2292 &iip->ili_item.li_lsn);
2294 xfs_buf_attach_iodone(bp, xfs_istale_done,
2298 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2301 xfs_trans_stale_inode_buf(tp, bp);
2302 xfs_trans_binval(tp, bp);
2310 * This is called to return an inode to the inode free list.
2311 * The inode should already be truncated to 0 length and have
2312 * no pages associated with it. This routine also assumes that
2313 * the inode is already a part of the transaction.
2315 * The on-disk copy of the inode will have been added to the list
2316 * of unlinked inodes in the AGI. We need to remove the inode from
2317 * that list atomically with respect to freeing it here.
2323 xfs_bmap_free_t *flist)
2327 xfs_ino_t first_ino;
2329 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2330 ASSERT(ip->i_d.di_nlink == 0);
2331 ASSERT(ip->i_d.di_nextents == 0);
2332 ASSERT(ip->i_d.di_anextents == 0);
2333 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2334 ASSERT(ip->i_d.di_nblocks == 0);
2337 * Pull the on-disk inode from the AGI unlinked list.
2339 error = xfs_iunlink_remove(tp, ip);
2343 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2347 ip->i_d.di_mode = 0; /* mark incore inode as free */
2348 ip->i_d.di_flags = 0;
2349 ip->i_d.di_dmevmask = 0;
2350 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2351 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2352 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2354 * Bump the generation count so no one will be confused
2355 * by reincarnations of this inode.
2358 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2361 error = xfs_ifree_cluster(ip, tp, first_ino);
2367 * This is called to unpin an inode. The caller must have the inode locked
2368 * in at least shared mode so that the buffer cannot be subsequently pinned
2369 * once someone is waiting for it to be unpinned.
2373 struct xfs_inode *ip)
2375 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2377 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2379 /* Give the log a push to start the unpinning I/O */
2380 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2386 struct xfs_inode *ip)
2388 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2389 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2394 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2395 if (xfs_ipincount(ip))
2397 } while (xfs_ipincount(ip));
2398 finish_wait(wq, &wait.wait);
2403 struct xfs_inode *ip)
2405 if (xfs_ipincount(ip))
2406 __xfs_iunpin_wait(ip);
2412 struct xfs_name *name,
2415 xfs_mount_t *mp = dp->i_mount;
2416 xfs_trans_t *tp = NULL;
2417 int is_dir = S_ISDIR(ip->i_d.di_mode);
2419 xfs_bmap_free_t free_list;
2420 xfs_fsblock_t first_block;
2427 trace_xfs_remove(dp, name);
2429 if (XFS_FORCED_SHUTDOWN(mp))
2430 return XFS_ERROR(EIO);
2432 error = xfs_qm_dqattach(dp, 0);
2436 error = xfs_qm_dqattach(ip, 0);
2441 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2442 log_count = XFS_DEFAULT_LOG_COUNT;
2444 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2445 log_count = XFS_REMOVE_LOG_COUNT;
2447 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2450 * We try to get the real space reservation first,
2451 * allowing for directory btree deletion(s) implying
2452 * possible bmap insert(s). If we can't get the space
2453 * reservation then we use 0 instead, and avoid the bmap
2454 * btree insert(s) in the directory code by, if the bmap
2455 * insert tries to happen, instead trimming the LAST
2456 * block from the directory.
2458 resblks = XFS_REMOVE_SPACE_RES(mp);
2459 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2460 if (error == ENOSPC) {
2462 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2465 ASSERT(error != ENOSPC);
2467 goto out_trans_cancel;
2470 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2472 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2473 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2476 * If we're removing a directory perform some additional validation.
2479 ASSERT(ip->i_d.di_nlink >= 2);
2480 if (ip->i_d.di_nlink != 2) {
2481 error = XFS_ERROR(ENOTEMPTY);
2482 goto out_trans_cancel;
2484 if (!xfs_dir_isempty(ip)) {
2485 error = XFS_ERROR(ENOTEMPTY);
2486 goto out_trans_cancel;
2490 xfs_bmap_init(&free_list, &first_block);
2491 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2492 &first_block, &free_list, resblks);
2494 ASSERT(error != ENOENT);
2495 goto out_bmap_cancel;
2497 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2501 * Drop the link from ip's "..".
2503 error = xfs_droplink(tp, dp);
2505 goto out_bmap_cancel;
2508 * Drop the "." link from ip to self.
2510 error = xfs_droplink(tp, ip);
2512 goto out_bmap_cancel;
2515 * When removing a non-directory we need to log the parent
2516 * inode here. For a directory this is done implicitly
2517 * by the xfs_droplink call for the ".." entry.
2519 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2523 * Drop the link from dp to ip.
2525 error = xfs_droplink(tp, ip);
2527 goto out_bmap_cancel;
2530 * Determine if this is the last link while
2531 * we are in the transaction.
2533 link_zero = (ip->i_d.di_nlink == 0);
2536 * If this is a synchronous mount, make sure that the
2537 * remove transaction goes to disk before returning to
2540 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2541 xfs_trans_set_sync(tp);
2543 error = xfs_bmap_finish(&tp, &free_list, &committed);
2545 goto out_bmap_cancel;
2547 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2552 * If we are using filestreams, kill the stream association.
2553 * If the file is still open it may get a new one but that
2554 * will get killed on last close in xfs_close() so we don't
2555 * have to worry about that.
2557 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2558 xfs_filestream_deassociate(ip);
2563 xfs_bmap_cancel(&free_list);
2564 cancel_flags |= XFS_TRANS_ABORT;
2566 xfs_trans_cancel(tp, cancel_flags);
2572 * Enter all inodes for a rename transaction into a sorted array.
2575 xfs_sort_for_rename(
2576 xfs_inode_t *dp1, /* in: old (source) directory inode */
2577 xfs_inode_t *dp2, /* in: new (target) directory inode */
2578 xfs_inode_t *ip1, /* in: inode of old entry */
2579 xfs_inode_t *ip2, /* in: inode of new entry, if it
2580 already exists, NULL otherwise. */
2581 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2582 int *num_inodes) /* out: number of inodes in array */
2588 * i_tab contains a list of pointers to inodes. We initialize
2589 * the table here & we'll sort it. We will then use it to
2590 * order the acquisition of the inode locks.
2592 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2606 * Sort the elements via bubble sort. (Remember, there are at
2607 * most 4 elements to sort, so this is adequate.)
2609 for (i = 0; i < *num_inodes; i++) {
2610 for (j = 1; j < *num_inodes; j++) {
2611 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2613 i_tab[j] = i_tab[j-1];
2625 xfs_inode_t *src_dp,
2626 struct xfs_name *src_name,
2627 xfs_inode_t *src_ip,
2628 xfs_inode_t *target_dp,
2629 struct xfs_name *target_name,
2630 xfs_inode_t *target_ip)
2632 xfs_trans_t *tp = NULL;
2633 xfs_mount_t *mp = src_dp->i_mount;
2634 int new_parent; /* moving to a new dir */
2635 int src_is_directory; /* src_name is a directory */
2637 xfs_bmap_free_t free_list;
2638 xfs_fsblock_t first_block;
2641 xfs_inode_t *inodes[4];
2645 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2647 new_parent = (src_dp != target_dp);
2648 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2650 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2651 inodes, &num_inodes);
2653 xfs_bmap_init(&free_list, &first_block);
2654 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2655 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2656 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2657 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2658 if (error == ENOSPC) {
2660 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2663 xfs_trans_cancel(tp, 0);
2668 * Attach the dquots to the inodes
2670 error = xfs_qm_vop_rename_dqattach(inodes);
2672 xfs_trans_cancel(tp, cancel_flags);
2677 * Lock all the participating inodes. Depending upon whether
2678 * the target_name exists in the target directory, and
2679 * whether the target directory is the same as the source
2680 * directory, we can lock from 2 to 4 inodes.
2682 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2685 * Join all the inodes to the transaction. From this point on,
2686 * we can rely on either trans_commit or trans_cancel to unlock
2689 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2691 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2692 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2694 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2697 * If we are using project inheritance, we only allow renames
2698 * into our tree when the project IDs are the same; else the
2699 * tree quota mechanism would be circumvented.
2701 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2702 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2703 error = XFS_ERROR(EXDEV);
2708 * Set up the target.
2710 if (target_ip == NULL) {
2712 * If there's no space reservation, check the entry will
2713 * fit before actually inserting it.
2715 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2719 * If target does not exist and the rename crosses
2720 * directories, adjust the target directory link count
2721 * to account for the ".." reference from the new entry.
2723 error = xfs_dir_createname(tp, target_dp, target_name,
2724 src_ip->i_ino, &first_block,
2725 &free_list, spaceres);
2726 if (error == ENOSPC)
2731 xfs_trans_ichgtime(tp, target_dp,
2732 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2734 if (new_parent && src_is_directory) {
2735 error = xfs_bumplink(tp, target_dp);
2739 } else { /* target_ip != NULL */
2741 * If target exists and it's a directory, check that both
2742 * target and source are directories and that target can be
2743 * destroyed, or that neither is a directory.
2745 if (S_ISDIR(target_ip->i_d.di_mode)) {
2747 * Make sure target dir is empty.
2749 if (!(xfs_dir_isempty(target_ip)) ||
2750 (target_ip->i_d.di_nlink > 2)) {
2751 error = XFS_ERROR(EEXIST);
2757 * Link the source inode under the target name.
2758 * If the source inode is a directory and we are moving
2759 * it across directories, its ".." entry will be
2760 * inconsistent until we replace that down below.
2762 * In case there is already an entry with the same
2763 * name at the destination directory, remove it first.
2765 error = xfs_dir_replace(tp, target_dp, target_name,
2767 &first_block, &free_list, spaceres);
2771 xfs_trans_ichgtime(tp, target_dp,
2772 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2775 * Decrement the link count on the target since the target
2776 * dir no longer points to it.
2778 error = xfs_droplink(tp, target_ip);
2782 if (src_is_directory) {
2784 * Drop the link from the old "." entry.
2786 error = xfs_droplink(tp, target_ip);
2790 } /* target_ip != NULL */
2793 * Remove the source.
2795 if (new_parent && src_is_directory) {
2797 * Rewrite the ".." entry to point to the new
2800 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2802 &first_block, &free_list, spaceres);
2803 ASSERT(error != EEXIST);
2809 * We always want to hit the ctime on the source inode.
2811 * This isn't strictly required by the standards since the source
2812 * inode isn't really being changed, but old unix file systems did
2813 * it and some incremental backup programs won't work without it.
2815 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2816 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2819 * Adjust the link count on src_dp. This is necessary when
2820 * renaming a directory, either within one parent when
2821 * the target existed, or across two parent directories.
2823 if (src_is_directory && (new_parent || target_ip != NULL)) {
2826 * Decrement link count on src_directory since the
2827 * entry that's moved no longer points to it.
2829 error = xfs_droplink(tp, src_dp);
2834 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2835 &first_block, &free_list, spaceres);
2839 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2840 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2842 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2845 * If this is a synchronous mount, make sure that the
2846 * rename transaction goes to disk before returning to
2849 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2850 xfs_trans_set_sync(tp);
2853 error = xfs_bmap_finish(&tp, &free_list, &committed);
2855 xfs_bmap_cancel(&free_list);
2856 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2862 * trans_commit will unlock src_ip, target_ip & decrement
2863 * the vnode references.
2865 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2868 cancel_flags |= XFS_TRANS_ABORT;
2870 xfs_bmap_cancel(&free_list);
2871 xfs_trans_cancel(tp, cancel_flags);
2881 xfs_mount_t *mp = ip->i_mount;
2882 struct xfs_perag *pag;
2883 unsigned long first_index, mask;
2884 unsigned long inodes_per_cluster;
2886 xfs_inode_t **ilist;
2893 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2895 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2896 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2897 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2901 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2902 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2904 /* really need a gang lookup range call here */
2905 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2906 first_index, inodes_per_cluster);
2910 for (i = 0; i < nr_found; i++) {
2916 * because this is an RCU protected lookup, we could find a
2917 * recently freed or even reallocated inode during the lookup.
2918 * We need to check under the i_flags_lock for a valid inode
2919 * here. Skip it if it is not valid or the wrong inode.
2921 spin_lock(&ip->i_flags_lock);
2923 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2924 spin_unlock(&ip->i_flags_lock);
2927 spin_unlock(&ip->i_flags_lock);
2930 * Do an un-protected check to see if the inode is dirty and
2931 * is a candidate for flushing. These checks will be repeated
2932 * later after the appropriate locks are acquired.
2934 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2938 * Try to get locks. If any are unavailable or it is pinned,
2939 * then this inode cannot be flushed and is skipped.
2942 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2944 if (!xfs_iflock_nowait(iq)) {
2945 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2948 if (xfs_ipincount(iq)) {
2950 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2955 * arriving here means that this inode can be flushed. First
2956 * re-check that it's dirty before flushing.
2958 if (!xfs_inode_clean(iq)) {
2960 error = xfs_iflush_int(iq, bp);
2962 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2963 goto cluster_corrupt_out;
2969 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2973 XFS_STATS_INC(xs_icluster_flushcnt);
2974 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2985 cluster_corrupt_out:
2987 * Corruption detected in the clustering loop. Invalidate the
2988 * inode buffer and shut down the filesystem.
2992 * Clean up the buffer. If it was delwri, just release it --
2993 * brelse can handle it with no problems. If not, shut down the
2994 * filesystem before releasing the buffer.
2996 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3000 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3002 if (!bufwasdelwri) {
3004 * Just like incore_relse: if we have b_iodone functions,
3005 * mark the buffer as an error and call them. Otherwise
3006 * mark it as stale and brelse.
3011 xfs_buf_ioerror(bp, EIO);
3012 xfs_buf_ioend(bp, 0);
3020 * Unlocks the flush lock
3022 xfs_iflush_abort(iq, false);
3025 return XFS_ERROR(EFSCORRUPTED);
3029 * Flush dirty inode metadata into the backing buffer.
3031 * The caller must have the inode lock and the inode flush lock held. The
3032 * inode lock will still be held upon return to the caller, and the inode
3033 * flush lock will be released after the inode has reached the disk.
3035 * The caller must write out the buffer returned in *bpp and release it.
3039 struct xfs_inode *ip,
3040 struct xfs_buf **bpp)
3042 struct xfs_mount *mp = ip->i_mount;
3044 struct xfs_dinode *dip;
3047 XFS_STATS_INC(xs_iflush_count);
3049 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3050 ASSERT(xfs_isiflocked(ip));
3051 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3052 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3056 xfs_iunpin_wait(ip);
3059 * For stale inodes we cannot rely on the backing buffer remaining
3060 * stale in cache for the remaining life of the stale inode and so
3061 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3062 * inodes below. We have to check this after ensuring the inode is
3063 * unpinned so that it is safe to reclaim the stale inode after the
3066 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3072 * This may have been unpinned because the filesystem is shutting
3073 * down forcibly. If that's the case we must not write this inode
3074 * to disk, because the log record didn't make it to disk.
3076 * We also have to remove the log item from the AIL in this case,
3077 * as we wait for an empty AIL as part of the unmount process.
3079 if (XFS_FORCED_SHUTDOWN(mp)) {
3080 error = XFS_ERROR(EIO);
3085 * Get the buffer containing the on-disk inode.
3087 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3095 * First flush out the inode that xfs_iflush was called with.
3097 error = xfs_iflush_int(ip, bp);
3102 * If the buffer is pinned then push on the log now so we won't
3103 * get stuck waiting in the write for too long.
3105 if (xfs_buf_ispinned(bp))
3106 xfs_log_force(mp, 0);
3110 * see if other inodes can be gathered into this write
3112 error = xfs_iflush_cluster(ip, bp);
3114 goto cluster_corrupt_out;
3121 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3122 cluster_corrupt_out:
3123 error = XFS_ERROR(EFSCORRUPTED);
3126 * Unlocks the flush lock
3128 xfs_iflush_abort(ip, false);
3134 struct xfs_inode *ip,
3137 struct xfs_inode_log_item *iip = ip->i_itemp;
3138 struct xfs_dinode *dip;
3139 struct xfs_mount *mp = ip->i_mount;
3141 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3142 ASSERT(xfs_isiflocked(ip));
3143 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3144 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3145 ASSERT(iip != NULL && iip->ili_fields != 0);
3147 /* set *dip = inode's place in the buffer */
3148 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3150 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3151 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3152 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3153 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3154 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3157 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3158 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3159 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3160 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3161 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3164 if (S_ISREG(ip->i_d.di_mode)) {
3166 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3167 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3168 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3169 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3170 "%s: Bad regular inode %Lu, ptr 0x%p",
3171 __func__, ip->i_ino, ip);
3174 } else if (S_ISDIR(ip->i_d.di_mode)) {
3176 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3177 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3178 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3179 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3180 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3181 "%s: Bad directory inode %Lu, ptr 0x%p",
3182 __func__, ip->i_ino, ip);
3186 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3187 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3188 XFS_RANDOM_IFLUSH_5)) {
3189 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3190 "%s: detected corrupt incore inode %Lu, "
3191 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3192 __func__, ip->i_ino,
3193 ip->i_d.di_nextents + ip->i_d.di_anextents,
3194 ip->i_d.di_nblocks, ip);
3197 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3198 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3199 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3200 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3201 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3206 * Inode item log recovery for v1/v2 inodes are dependent on the
3207 * di_flushiter count for correct sequencing. We bump the flush
3208 * iteration count so we can detect flushes which postdate a log record
3209 * during recovery. This is redundant as we now log every change and
3210 * hence this can't happen but we need to still do it to ensure
3211 * backwards compatibility with old kernels that predate logging all
3214 if (ip->i_d.di_version < 3)
3215 ip->i_d.di_flushiter++;
3218 * Copy the dirty parts of the inode into the on-disk
3219 * inode. We always copy out the core of the inode,
3220 * because if the inode is dirty at all the core must
3223 xfs_dinode_to_disk(dip, &ip->i_d);
3225 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3226 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3227 ip->i_d.di_flushiter = 0;
3230 * If this is really an old format inode and the superblock version
3231 * has not been updated to support only new format inodes, then
3232 * convert back to the old inode format. If the superblock version
3233 * has been updated, then make the conversion permanent.
3235 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3236 if (ip->i_d.di_version == 1) {
3237 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3241 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3242 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3245 * The superblock version has already been bumped,
3246 * so just make the conversion to the new inode
3249 ip->i_d.di_version = 2;
3250 dip->di_version = 2;
3251 ip->i_d.di_onlink = 0;
3253 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3254 memset(&(dip->di_pad[0]), 0,
3255 sizeof(dip->di_pad));
3256 ASSERT(xfs_get_projid(ip) == 0);
3260 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3261 if (XFS_IFORK_Q(ip))
3262 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3263 xfs_inobp_check(mp, bp);
3266 * We've recorded everything logged in the inode, so we'd like to clear
3267 * the ili_fields bits so we don't log and flush things unnecessarily.
3268 * However, we can't stop logging all this information until the data
3269 * we've copied into the disk buffer is written to disk. If we did we
3270 * might overwrite the copy of the inode in the log with all the data
3271 * after re-logging only part of it, and in the face of a crash we
3272 * wouldn't have all the data we need to recover.
3274 * What we do is move the bits to the ili_last_fields field. When
3275 * logging the inode, these bits are moved back to the ili_fields field.
3276 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3277 * know that the information those bits represent is permanently on
3278 * disk. As long as the flush completes before the inode is logged
3279 * again, then both ili_fields and ili_last_fields will be cleared.
3281 * We can play with the ili_fields bits here, because the inode lock
3282 * must be held exclusively in order to set bits there and the flush
3283 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3284 * done routine can tell whether or not to look in the AIL. Also, store
3285 * the current LSN of the inode so that we can tell whether the item has
3286 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3287 * need the AIL lock, because it is a 64 bit value that cannot be read
3290 iip->ili_last_fields = iip->ili_fields;
3291 iip->ili_fields = 0;
3292 iip->ili_logged = 1;
3294 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3295 &iip->ili_item.li_lsn);
3298 * Attach the function xfs_iflush_done to the inode's
3299 * buffer. This will remove the inode from the AIL
3300 * and unlock the inode's flush lock when the inode is
3301 * completely written to disk.
3303 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3305 /* update the lsn in the on disk inode if required */
3306 if (ip->i_d.di_version == 3)
3307 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3309 /* generate the checksum. */
3310 xfs_dinode_calc_crc(mp, dip);
3312 ASSERT(bp->b_fspriv != NULL);
3313 ASSERT(bp->b_iodone != NULL);
3317 return XFS_ERROR(EFSCORRUPTED);