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"
25 #include "xfs_trans.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trans_priv.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_dir2_format.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
47 #include "xfs_bmap_util.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.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 = current_fsuid();
695 ip->i_d.di_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((gid_t)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);
868 struct xfs_name *name,
873 int is_dir = S_ISDIR(mode);
874 struct xfs_mount *mp = dp->i_mount;
875 struct xfs_inode *ip = NULL;
876 struct xfs_trans *tp = NULL;
878 xfs_bmap_free_t free_list;
879 xfs_fsblock_t first_block;
880 bool unlock_dp_on_error = false;
884 struct xfs_dquot *udqp = NULL;
885 struct xfs_dquot *gdqp = NULL;
886 struct xfs_dquot *pdqp = NULL;
891 trace_xfs_create(dp, name);
893 if (XFS_FORCED_SHUTDOWN(mp))
894 return XFS_ERROR(EIO);
896 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
897 prid = xfs_get_projid(dp);
899 prid = XFS_PROJID_DEFAULT;
902 * Make sure that we have allocated dquot(s) on disk.
904 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
905 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
906 &udqp, &gdqp, &pdqp);
912 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
913 log_res = XFS_MKDIR_LOG_RES(mp);
914 log_count = XFS_MKDIR_LOG_COUNT;
915 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
917 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
918 log_res = XFS_CREATE_LOG_RES(mp);
919 log_count = XFS_CREATE_LOG_COUNT;
920 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
923 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
926 * Initially assume that the file does not exist and
927 * reserve the resources for that case. If that is not
928 * the case we'll drop the one we have and get a more
929 * appropriate transaction later.
931 error = xfs_trans_reserve(tp, resblks, log_res, 0,
932 XFS_TRANS_PERM_LOG_RES, log_count);
933 if (error == ENOSPC) {
934 /* flush outstanding delalloc blocks and retry */
935 xfs_flush_inodes(mp);
936 error = xfs_trans_reserve(tp, resblks, log_res, 0,
937 XFS_TRANS_PERM_LOG_RES, log_count);
939 if (error == ENOSPC) {
940 /* No space at all so try a "no-allocation" reservation */
942 error = xfs_trans_reserve(tp, 0, log_res, 0,
943 XFS_TRANS_PERM_LOG_RES, log_count);
947 goto out_trans_cancel;
950 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
951 unlock_dp_on_error = true;
953 xfs_bmap_init(&free_list, &first_block);
956 * Reserve disk quota and the inode.
958 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
959 pdqp, resblks, 1, 0);
961 goto out_trans_cancel;
963 error = xfs_dir_canenter(tp, dp, name, resblks);
965 goto out_trans_cancel;
968 * A newly created regular or special file just has one directory
969 * entry pointing to them, but a directory also the "." entry
970 * pointing to itself.
972 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
973 prid, resblks > 0, &ip, &committed);
976 goto out_trans_cancel;
977 goto out_trans_abort;
981 * Now we join the directory inode to the transaction. We do not do it
982 * earlier because xfs_dir_ialloc might commit the previous transaction
983 * (and release all the locks). An error from here on will result in
984 * the transaction cancel unlocking dp so don't do it explicitly in the
987 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
988 unlock_dp_on_error = false;
990 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
991 &first_block, &free_list, resblks ?
992 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
994 ASSERT(error != ENOSPC);
995 goto out_trans_abort;
997 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
998 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1001 error = xfs_dir_init(tp, ip, dp);
1003 goto out_bmap_cancel;
1005 error = xfs_bumplink(tp, dp);
1007 goto out_bmap_cancel;
1011 * If this is a synchronous mount, make sure that the
1012 * create transaction goes to disk before returning to
1015 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1016 xfs_trans_set_sync(tp);
1019 * Attach the dquot(s) to the inodes and modify them incore.
1020 * These ids of the inode couldn't have changed since the new
1021 * inode has been locked ever since it was created.
1023 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1025 error = xfs_bmap_finish(&tp, &free_list, &committed);
1027 goto out_bmap_cancel;
1029 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1031 goto out_release_inode;
1033 xfs_qm_dqrele(udqp);
1034 xfs_qm_dqrele(gdqp);
1035 xfs_qm_dqrele(pdqp);
1041 xfs_bmap_cancel(&free_list);
1043 cancel_flags |= XFS_TRANS_ABORT;
1045 xfs_trans_cancel(tp, cancel_flags);
1048 * Wait until after the current transaction is aborted to
1049 * release the inode. This prevents recursive transactions
1050 * and deadlocks from xfs_inactive.
1055 xfs_qm_dqrele(udqp);
1056 xfs_qm_dqrele(gdqp);
1057 xfs_qm_dqrele(pdqp);
1059 if (unlock_dp_on_error)
1060 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1068 struct xfs_name *target_name)
1070 xfs_mount_t *mp = tdp->i_mount;
1073 xfs_bmap_free_t free_list;
1074 xfs_fsblock_t first_block;
1079 trace_xfs_link(tdp, target_name);
1081 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1083 if (XFS_FORCED_SHUTDOWN(mp))
1084 return XFS_ERROR(EIO);
1086 error = xfs_qm_dqattach(sip, 0);
1090 error = xfs_qm_dqattach(tdp, 0);
1094 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1095 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1096 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1097 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1098 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1099 if (error == ENOSPC) {
1101 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1102 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1109 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1111 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1112 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1115 * If we are using project inheritance, we only allow hard link
1116 * creation in our tree when the project IDs are the same; else
1117 * the tree quota mechanism could be circumvented.
1119 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1120 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1121 error = XFS_ERROR(EXDEV);
1125 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1129 xfs_bmap_init(&free_list, &first_block);
1131 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1132 &first_block, &free_list, resblks);
1135 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1136 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1138 error = xfs_bumplink(tp, sip);
1143 * If this is a synchronous mount, make sure that the
1144 * link transaction goes to disk before returning to
1147 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1148 xfs_trans_set_sync(tp);
1151 error = xfs_bmap_finish (&tp, &free_list, &committed);
1153 xfs_bmap_cancel(&free_list);
1157 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1160 cancel_flags |= XFS_TRANS_ABORT;
1162 xfs_trans_cancel(tp, cancel_flags);
1168 * Free up the underlying blocks past new_size. The new size must be smaller
1169 * than the current size. This routine can be used both for the attribute and
1170 * data fork, and does not modify the inode size, which is left to the caller.
1172 * The transaction passed to this routine must have made a permanent log
1173 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1174 * given transaction and start new ones, so make sure everything involved in
1175 * the transaction is tidy before calling here. Some transaction will be
1176 * returned to the caller to be committed. The incoming transaction must
1177 * already include the inode, and both inode locks must be held exclusively.
1178 * The inode must also be "held" within the transaction. On return the inode
1179 * will be "held" within the returned transaction. This routine does NOT
1180 * require any disk space to be reserved for it within the transaction.
1182 * If we get an error, we must return with the inode locked and linked into the
1183 * current transaction. This keeps things simple for the higher level code,
1184 * because it always knows that the inode is locked and held in the transaction
1185 * that returns to it whether errors occur or not. We don't mark the inode
1186 * dirty on error so that transactions can be easily aborted if possible.
1189 xfs_itruncate_extents(
1190 struct xfs_trans **tpp,
1191 struct xfs_inode *ip,
1193 xfs_fsize_t new_size)
1195 struct xfs_mount *mp = ip->i_mount;
1196 struct xfs_trans *tp = *tpp;
1197 struct xfs_trans *ntp;
1198 xfs_bmap_free_t free_list;
1199 xfs_fsblock_t first_block;
1200 xfs_fileoff_t first_unmap_block;
1201 xfs_fileoff_t last_block;
1202 xfs_filblks_t unmap_len;
1207 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1208 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1209 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1210 ASSERT(new_size <= XFS_ISIZE(ip));
1211 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1212 ASSERT(ip->i_itemp != NULL);
1213 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1214 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1216 trace_xfs_itruncate_extents_start(ip, new_size);
1219 * Since it is possible for space to become allocated beyond
1220 * the end of the file (in a crash where the space is allocated
1221 * but the inode size is not yet updated), simply remove any
1222 * blocks which show up between the new EOF and the maximum
1223 * possible file size. If the first block to be removed is
1224 * beyond the maximum file size (ie it is the same as last_block),
1225 * then there is nothing to do.
1227 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1228 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1229 if (first_unmap_block == last_block)
1232 ASSERT(first_unmap_block < last_block);
1233 unmap_len = last_block - first_unmap_block + 1;
1235 xfs_bmap_init(&free_list, &first_block);
1236 error = xfs_bunmapi(tp, ip,
1237 first_unmap_block, unmap_len,
1238 xfs_bmapi_aflag(whichfork),
1239 XFS_ITRUNC_MAX_EXTENTS,
1240 &first_block, &free_list,
1243 goto out_bmap_cancel;
1246 * Duplicate the transaction that has the permanent
1247 * reservation and commit the old transaction.
1249 error = xfs_bmap_finish(&tp, &free_list, &committed);
1251 xfs_trans_ijoin(tp, ip, 0);
1253 goto out_bmap_cancel;
1257 * Mark the inode dirty so it will be logged and
1258 * moved forward in the log as part of every commit.
1260 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1263 ntp = xfs_trans_dup(tp);
1264 error = xfs_trans_commit(tp, 0);
1267 xfs_trans_ijoin(tp, ip, 0);
1273 * Transaction commit worked ok so we can drop the extra ticket
1274 * reference that we gained in xfs_trans_dup()
1276 xfs_log_ticket_put(tp->t_ticket);
1277 error = xfs_trans_reserve(tp, 0,
1278 XFS_ITRUNCATE_LOG_RES(mp), 0,
1279 XFS_TRANS_PERM_LOG_RES,
1280 XFS_ITRUNCATE_LOG_COUNT);
1286 * Always re-log the inode so that our permanent transaction can keep
1287 * on rolling it forward in the log.
1289 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1291 trace_xfs_itruncate_extents_end(ip, new_size);
1298 * If the bunmapi call encounters an error, return to the caller where
1299 * the transaction can be properly aborted. We just need to make sure
1300 * we're not holding any resources that we were not when we came in.
1302 xfs_bmap_cancel(&free_list);
1310 xfs_mount_t *mp = ip->i_mount;
1313 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1316 /* If this is a read-only mount, don't do this (would generate I/O) */
1317 if (mp->m_flags & XFS_MOUNT_RDONLY)
1320 if (!XFS_FORCED_SHUTDOWN(mp)) {
1324 * If we are using filestreams, and we have an unlinked
1325 * file that we are processing the last close on, then nothing
1326 * will be able to reopen and write to this file. Purge this
1327 * inode from the filestreams cache so that it doesn't delay
1328 * teardown of the inode.
1330 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1331 xfs_filestream_deassociate(ip);
1334 * If we previously truncated this file and removed old data
1335 * in the process, we want to initiate "early" writeout on
1336 * the last close. This is an attempt to combat the notorious
1337 * NULL files problem which is particularly noticeable from a
1338 * truncate down, buffered (re-)write (delalloc), followed by
1339 * a crash. What we are effectively doing here is
1340 * significantly reducing the time window where we'd otherwise
1341 * be exposed to that problem.
1343 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1345 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1346 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1347 error = -filemap_flush(VFS_I(ip)->i_mapping);
1354 if (ip->i_d.di_nlink == 0)
1357 if (xfs_can_free_eofblocks(ip, false)) {
1360 * If we can't get the iolock just skip truncating the blocks
1361 * past EOF because we could deadlock with the mmap_sem
1362 * otherwise. We'll get another chance to drop them once the
1363 * last reference to the inode is dropped, so we'll never leak
1364 * blocks permanently.
1366 * Further, check if the inode is being opened, written and
1367 * closed frequently and we have delayed allocation blocks
1368 * outstanding (e.g. streaming writes from the NFS server),
1369 * truncating the blocks past EOF will cause fragmentation to
1372 * In this case don't do the truncation, either, but we have to
1373 * be careful how we detect this case. Blocks beyond EOF show
1374 * up as i_delayed_blks even when the inode is clean, so we
1375 * need to truncate them away first before checking for a dirty
1376 * release. Hence on the first dirty close we will still remove
1377 * the speculative allocation, but after that we will leave it
1380 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1383 error = xfs_free_eofblocks(mp, ip, true);
1384 if (error && error != EAGAIN)
1387 /* delalloc blocks after truncation means it really is dirty */
1388 if (ip->i_delayed_blks)
1389 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1397 * This is called when the vnode reference count for the vnode
1398 * goes to zero. If the file has been unlinked, then it must
1399 * now be truncated. Also, we clear all of the read-ahead state
1400 * kept for the inode here since the file is now closed.
1406 xfs_bmap_free_t free_list;
1407 xfs_fsblock_t first_block;
1415 * If the inode is already free, then there can be nothing
1418 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1419 ASSERT(ip->i_df.if_real_bytes == 0);
1420 ASSERT(ip->i_df.if_broot_bytes == 0);
1421 return VN_INACTIVE_CACHE;
1428 /* If this is a read-only mount, don't do this (would generate I/O) */
1429 if (mp->m_flags & XFS_MOUNT_RDONLY)
1432 if (ip->i_d.di_nlink != 0) {
1434 * force is true because we are evicting an inode from the
1435 * cache. Post-eof blocks must be freed, lest we end up with
1436 * broken free space accounting.
1438 if (xfs_can_free_eofblocks(ip, true)) {
1439 error = xfs_free_eofblocks(mp, ip, false);
1441 return VN_INACTIVE_CACHE;
1446 if (S_ISREG(ip->i_d.di_mode) &&
1447 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1448 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1451 error = xfs_qm_dqattach(ip, 0);
1453 return VN_INACTIVE_CACHE;
1455 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1456 error = xfs_trans_reserve(tp, 0,
1457 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
1458 XFS_ITRUNCATE_LOG_RES(mp) :
1459 XFS_IFREE_LOG_RES(mp),
1461 XFS_TRANS_PERM_LOG_RES,
1462 XFS_ITRUNCATE_LOG_COUNT);
1464 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1465 xfs_trans_cancel(tp, 0);
1466 return VN_INACTIVE_CACHE;
1469 xfs_ilock(ip, XFS_ILOCK_EXCL);
1470 xfs_trans_ijoin(tp, ip, 0);
1472 if (S_ISLNK(ip->i_d.di_mode)) {
1473 error = xfs_inactive_symlink(ip, &tp);
1476 } else if (truncate) {
1477 ip->i_d.di_size = 0;
1478 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1480 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1484 ASSERT(ip->i_d.di_nextents == 0);
1488 * If there are attributes associated with the file then blow them away
1489 * now. The code calls a routine that recursively deconstructs the
1490 * attribute fork. We need to just commit the current transaction
1491 * because we can't use it for xfs_attr_inactive().
1493 if (ip->i_d.di_anextents > 0) {
1494 ASSERT(ip->i_d.di_forkoff != 0);
1496 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1502 error = xfs_attr_inactive(ip);
1506 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1507 error = xfs_trans_reserve(tp, 0,
1508 XFS_IFREE_LOG_RES(mp),
1509 0, XFS_TRANS_PERM_LOG_RES,
1510 XFS_INACTIVE_LOG_COUNT);
1512 xfs_trans_cancel(tp, 0);
1516 xfs_ilock(ip, XFS_ILOCK_EXCL);
1517 xfs_trans_ijoin(tp, ip, 0);
1521 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1523 ASSERT(ip->i_d.di_anextents == 0);
1528 xfs_bmap_init(&free_list, &first_block);
1529 error = xfs_ifree(tp, ip, &free_list);
1532 * If we fail to free the inode, shut down. The cancel
1533 * might do that, we need to make sure. Otherwise the
1534 * inode might be lost for a long time or forever.
1536 if (!XFS_FORCED_SHUTDOWN(mp)) {
1537 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1539 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1541 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1544 * Credit the quota account(s). The inode is gone.
1546 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1549 * Just ignore errors at this point. There is nothing we can
1550 * do except to try to keep going. Make sure it's not a silent
1553 error = xfs_bmap_finish(&tp, &free_list, &committed);
1555 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1557 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1559 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1564 * Release the dquots held by inode, if any.
1566 xfs_qm_dqdetach(ip);
1568 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1570 return VN_INACTIVE_CACHE;
1572 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1577 * This is called when the inode's link count goes to 0.
1578 * We place the on-disk inode on a list in the AGI. It
1579 * will be pulled from this list when the inode is freed.
1596 ASSERT(ip->i_d.di_nlink == 0);
1597 ASSERT(ip->i_d.di_mode != 0);
1602 * Get the agi buffer first. It ensures lock ordering
1605 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1608 agi = XFS_BUF_TO_AGI(agibp);
1611 * Get the index into the agi hash table for the
1612 * list this inode will go on.
1614 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1616 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1617 ASSERT(agi->agi_unlinked[bucket_index]);
1618 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1620 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1622 * There is already another inode in the bucket we need
1623 * to add ourselves to. Add us at the front of the list.
1624 * Here we put the head pointer into our next pointer,
1625 * and then we fall through to point the head at us.
1627 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1632 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1633 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1634 offset = ip->i_imap.im_boffset +
1635 offsetof(xfs_dinode_t, di_next_unlinked);
1637 /* need to recalc the inode CRC if appropriate */
1638 xfs_dinode_calc_crc(mp, dip);
1640 xfs_trans_inode_buf(tp, ibp);
1641 xfs_trans_log_buf(tp, ibp, offset,
1642 (offset + sizeof(xfs_agino_t) - 1));
1643 xfs_inobp_check(mp, ibp);
1647 * Point the bucket head pointer at the inode being inserted.
1650 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1651 offset = offsetof(xfs_agi_t, agi_unlinked) +
1652 (sizeof(xfs_agino_t) * bucket_index);
1653 xfs_trans_log_buf(tp, agibp, offset,
1654 (offset + sizeof(xfs_agino_t) - 1));
1659 * Pull the on-disk inode from the AGI unlinked list.
1672 xfs_agnumber_t agno;
1674 xfs_agino_t next_agino;
1675 xfs_buf_t *last_ibp;
1676 xfs_dinode_t *last_dip = NULL;
1678 int offset, last_offset = 0;
1682 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1685 * Get the agi buffer first. It ensures lock ordering
1688 error = xfs_read_agi(mp, tp, agno, &agibp);
1692 agi = XFS_BUF_TO_AGI(agibp);
1695 * Get the index into the agi hash table for the
1696 * list this inode will go on.
1698 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1700 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1701 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1702 ASSERT(agi->agi_unlinked[bucket_index]);
1704 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1706 * We're at the head of the list. Get the inode's on-disk
1707 * buffer to see if there is anyone after us on the list.
1708 * Only modify our next pointer if it is not already NULLAGINO.
1709 * This saves us the overhead of dealing with the buffer when
1710 * there is no need to change it.
1712 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1715 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
1719 next_agino = be32_to_cpu(dip->di_next_unlinked);
1720 ASSERT(next_agino != 0);
1721 if (next_agino != NULLAGINO) {
1722 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1723 offset = ip->i_imap.im_boffset +
1724 offsetof(xfs_dinode_t, di_next_unlinked);
1726 /* need to recalc the inode CRC if appropriate */
1727 xfs_dinode_calc_crc(mp, dip);
1729 xfs_trans_inode_buf(tp, ibp);
1730 xfs_trans_log_buf(tp, ibp, offset,
1731 (offset + sizeof(xfs_agino_t) - 1));
1732 xfs_inobp_check(mp, ibp);
1734 xfs_trans_brelse(tp, ibp);
1737 * Point the bucket head pointer at the next inode.
1739 ASSERT(next_agino != 0);
1740 ASSERT(next_agino != agino);
1741 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1742 offset = offsetof(xfs_agi_t, agi_unlinked) +
1743 (sizeof(xfs_agino_t) * bucket_index);
1744 xfs_trans_log_buf(tp, agibp, offset,
1745 (offset + sizeof(xfs_agino_t) - 1));
1748 * We need to search the list for the inode being freed.
1750 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1752 while (next_agino != agino) {
1753 struct xfs_imap imap;
1756 xfs_trans_brelse(tp, last_ibp);
1759 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1761 error = xfs_imap(mp, tp, next_ino, &imap, 0);
1764 "%s: xfs_imap returned error %d.",
1769 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
1773 "%s: xfs_imap_to_bp returned error %d.",
1778 last_offset = imap.im_boffset;
1779 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1780 ASSERT(next_agino != NULLAGINO);
1781 ASSERT(next_agino != 0);
1785 * Now last_ibp points to the buffer previous to us on the
1786 * unlinked list. Pull us from the list.
1788 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1791 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
1795 next_agino = be32_to_cpu(dip->di_next_unlinked);
1796 ASSERT(next_agino != 0);
1797 ASSERT(next_agino != agino);
1798 if (next_agino != NULLAGINO) {
1799 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1800 offset = ip->i_imap.im_boffset +
1801 offsetof(xfs_dinode_t, di_next_unlinked);
1803 /* need to recalc the inode CRC if appropriate */
1804 xfs_dinode_calc_crc(mp, dip);
1806 xfs_trans_inode_buf(tp, ibp);
1807 xfs_trans_log_buf(tp, ibp, offset,
1808 (offset + sizeof(xfs_agino_t) - 1));
1809 xfs_inobp_check(mp, ibp);
1811 xfs_trans_brelse(tp, ibp);
1814 * Point the previous inode on the list to the next inode.
1816 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1817 ASSERT(next_agino != 0);
1818 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1820 /* need to recalc the inode CRC if appropriate */
1821 xfs_dinode_calc_crc(mp, last_dip);
1823 xfs_trans_inode_buf(tp, last_ibp);
1824 xfs_trans_log_buf(tp, last_ibp, offset,
1825 (offset + sizeof(xfs_agino_t) - 1));
1826 xfs_inobp_check(mp, last_ibp);
1832 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1833 * inodes that are in memory - they all must be marked stale and attached to
1834 * the cluster buffer.
1838 xfs_inode_t *free_ip,
1842 xfs_mount_t *mp = free_ip->i_mount;
1843 int blks_per_cluster;
1850 xfs_inode_log_item_t *iip;
1851 xfs_log_item_t *lip;
1852 struct xfs_perag *pag;
1854 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1855 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1856 blks_per_cluster = 1;
1857 ninodes = mp->m_sb.sb_inopblock;
1858 nbufs = XFS_IALLOC_BLOCKS(mp);
1860 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1861 mp->m_sb.sb_blocksize;
1862 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1863 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1866 for (j = 0; j < nbufs; j++, inum += ninodes) {
1867 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1868 XFS_INO_TO_AGBNO(mp, inum));
1871 * We obtain and lock the backing buffer first in the process
1872 * here, as we have to ensure that any dirty inode that we
1873 * can't get the flush lock on is attached to the buffer.
1874 * If we scan the in-memory inodes first, then buffer IO can
1875 * complete before we get a lock on it, and hence we may fail
1876 * to mark all the active inodes on the buffer stale.
1878 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1879 mp->m_bsize * blks_per_cluster,
1886 * This buffer may not have been correctly initialised as we
1887 * didn't read it from disk. That's not important because we are
1888 * only using to mark the buffer as stale in the log, and to
1889 * attach stale cached inodes on it. That means it will never be
1890 * dispatched for IO. If it is, we want to know about it, and we
1891 * want it to fail. We can acheive this by adding a write
1892 * verifier to the buffer.
1894 bp->b_ops = &xfs_inode_buf_ops;
1897 * Walk the inodes already attached to the buffer and mark them
1898 * stale. These will all have the flush locks held, so an
1899 * in-memory inode walk can't lock them. By marking them all
1900 * stale first, we will not attempt to lock them in the loop
1901 * below as the XFS_ISTALE flag will be set.
1905 if (lip->li_type == XFS_LI_INODE) {
1906 iip = (xfs_inode_log_item_t *)lip;
1907 ASSERT(iip->ili_logged == 1);
1908 lip->li_cb = xfs_istale_done;
1909 xfs_trans_ail_copy_lsn(mp->m_ail,
1910 &iip->ili_flush_lsn,
1911 &iip->ili_item.li_lsn);
1912 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1914 lip = lip->li_bio_list;
1919 * For each inode in memory attempt to add it to the inode
1920 * buffer and set it up for being staled on buffer IO
1921 * completion. This is safe as we've locked out tail pushing
1922 * and flushing by locking the buffer.
1924 * We have already marked every inode that was part of a
1925 * transaction stale above, which means there is no point in
1926 * even trying to lock them.
1928 for (i = 0; i < ninodes; i++) {
1931 ip = radix_tree_lookup(&pag->pag_ici_root,
1932 XFS_INO_TO_AGINO(mp, (inum + i)));
1934 /* Inode not in memory, nothing to do */
1941 * because this is an RCU protected lookup, we could
1942 * find a recently freed or even reallocated inode
1943 * during the lookup. We need to check under the
1944 * i_flags_lock for a valid inode here. Skip it if it
1945 * is not valid, the wrong inode or stale.
1947 spin_lock(&ip->i_flags_lock);
1948 if (ip->i_ino != inum + i ||
1949 __xfs_iflags_test(ip, XFS_ISTALE)) {
1950 spin_unlock(&ip->i_flags_lock);
1954 spin_unlock(&ip->i_flags_lock);
1957 * Don't try to lock/unlock the current inode, but we
1958 * _cannot_ skip the other inodes that we did not find
1959 * in the list attached to the buffer and are not
1960 * already marked stale. If we can't lock it, back off
1963 if (ip != free_ip &&
1964 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1972 xfs_iflags_set(ip, XFS_ISTALE);
1975 * we don't need to attach clean inodes or those only
1976 * with unlogged changes (which we throw away, anyway).
1979 if (!iip || xfs_inode_clean(ip)) {
1980 ASSERT(ip != free_ip);
1982 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1986 iip->ili_last_fields = iip->ili_fields;
1987 iip->ili_fields = 0;
1988 iip->ili_logged = 1;
1989 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1990 &iip->ili_item.li_lsn);
1992 xfs_buf_attach_iodone(bp, xfs_istale_done,
1996 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1999 xfs_trans_stale_inode_buf(tp, bp);
2000 xfs_trans_binval(tp, bp);
2008 * This is called to return an inode to the inode free list.
2009 * The inode should already be truncated to 0 length and have
2010 * no pages associated with it. This routine also assumes that
2011 * the inode is already a part of the transaction.
2013 * The on-disk copy of the inode will have been added to the list
2014 * of unlinked inodes in the AGI. We need to remove the inode from
2015 * that list atomically with respect to freeing it here.
2021 xfs_bmap_free_t *flist)
2025 xfs_ino_t first_ino;
2027 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2028 ASSERT(ip->i_d.di_nlink == 0);
2029 ASSERT(ip->i_d.di_nextents == 0);
2030 ASSERT(ip->i_d.di_anextents == 0);
2031 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2032 ASSERT(ip->i_d.di_nblocks == 0);
2035 * Pull the on-disk inode from the AGI unlinked list.
2037 error = xfs_iunlink_remove(tp, ip);
2041 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2045 ip->i_d.di_mode = 0; /* mark incore inode as free */
2046 ip->i_d.di_flags = 0;
2047 ip->i_d.di_dmevmask = 0;
2048 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2049 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2050 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2052 * Bump the generation count so no one will be confused
2053 * by reincarnations of this inode.
2056 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2059 error = xfs_ifree_cluster(ip, tp, first_ino);
2065 * This is called to unpin an inode. The caller must have the inode locked
2066 * in at least shared mode so that the buffer cannot be subsequently pinned
2067 * once someone is waiting for it to be unpinned.
2071 struct xfs_inode *ip)
2073 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2075 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2077 /* Give the log a push to start the unpinning I/O */
2078 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2084 struct xfs_inode *ip)
2086 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2087 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2092 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2093 if (xfs_ipincount(ip))
2095 } while (xfs_ipincount(ip));
2096 finish_wait(wq, &wait.wait);
2101 struct xfs_inode *ip)
2103 if (xfs_ipincount(ip))
2104 __xfs_iunpin_wait(ip);
2110 struct xfs_name *name,
2113 xfs_mount_t *mp = dp->i_mount;
2114 xfs_trans_t *tp = NULL;
2115 int is_dir = S_ISDIR(ip->i_d.di_mode);
2117 xfs_bmap_free_t free_list;
2118 xfs_fsblock_t first_block;
2125 trace_xfs_remove(dp, name);
2127 if (XFS_FORCED_SHUTDOWN(mp))
2128 return XFS_ERROR(EIO);
2130 error = xfs_qm_dqattach(dp, 0);
2134 error = xfs_qm_dqattach(ip, 0);
2139 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2140 log_count = XFS_DEFAULT_LOG_COUNT;
2142 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2143 log_count = XFS_REMOVE_LOG_COUNT;
2145 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2148 * We try to get the real space reservation first,
2149 * allowing for directory btree deletion(s) implying
2150 * possible bmap insert(s). If we can't get the space
2151 * reservation then we use 0 instead, and avoid the bmap
2152 * btree insert(s) in the directory code by, if the bmap
2153 * insert tries to happen, instead trimming the LAST
2154 * block from the directory.
2156 resblks = XFS_REMOVE_SPACE_RES(mp);
2157 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2158 XFS_TRANS_PERM_LOG_RES, log_count);
2159 if (error == ENOSPC) {
2161 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2162 XFS_TRANS_PERM_LOG_RES, log_count);
2165 ASSERT(error != ENOSPC);
2167 goto out_trans_cancel;
2170 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2172 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2173 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2176 * If we're removing a directory perform some additional validation.
2179 ASSERT(ip->i_d.di_nlink >= 2);
2180 if (ip->i_d.di_nlink != 2) {
2181 error = XFS_ERROR(ENOTEMPTY);
2182 goto out_trans_cancel;
2184 if (!xfs_dir_isempty(ip)) {
2185 error = XFS_ERROR(ENOTEMPTY);
2186 goto out_trans_cancel;
2190 xfs_bmap_init(&free_list, &first_block);
2191 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2192 &first_block, &free_list, resblks);
2194 ASSERT(error != ENOENT);
2195 goto out_bmap_cancel;
2197 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2201 * Drop the link from ip's "..".
2203 error = xfs_droplink(tp, dp);
2205 goto out_bmap_cancel;
2208 * Drop the "." link from ip to self.
2210 error = xfs_droplink(tp, ip);
2212 goto out_bmap_cancel;
2215 * When removing a non-directory we need to log the parent
2216 * inode here. For a directory this is done implicitly
2217 * by the xfs_droplink call for the ".." entry.
2219 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2223 * Drop the link from dp to ip.
2225 error = xfs_droplink(tp, ip);
2227 goto out_bmap_cancel;
2230 * Determine if this is the last link while
2231 * we are in the transaction.
2233 link_zero = (ip->i_d.di_nlink == 0);
2236 * If this is a synchronous mount, make sure that the
2237 * remove transaction goes to disk before returning to
2240 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2241 xfs_trans_set_sync(tp);
2243 error = xfs_bmap_finish(&tp, &free_list, &committed);
2245 goto out_bmap_cancel;
2247 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2252 * If we are using filestreams, kill the stream association.
2253 * If the file is still open it may get a new one but that
2254 * will get killed on last close in xfs_close() so we don't
2255 * have to worry about that.
2257 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2258 xfs_filestream_deassociate(ip);
2263 xfs_bmap_cancel(&free_list);
2264 cancel_flags |= XFS_TRANS_ABORT;
2266 xfs_trans_cancel(tp, cancel_flags);
2272 * Enter all inodes for a rename transaction into a sorted array.
2275 xfs_sort_for_rename(
2276 xfs_inode_t *dp1, /* in: old (source) directory inode */
2277 xfs_inode_t *dp2, /* in: new (target) directory inode */
2278 xfs_inode_t *ip1, /* in: inode of old entry */
2279 xfs_inode_t *ip2, /* in: inode of new entry, if it
2280 already exists, NULL otherwise. */
2281 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2282 int *num_inodes) /* out: number of inodes in array */
2288 * i_tab contains a list of pointers to inodes. We initialize
2289 * the table here & we'll sort it. We will then use it to
2290 * order the acquisition of the inode locks.
2292 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2306 * Sort the elements via bubble sort. (Remember, there are at
2307 * most 4 elements to sort, so this is adequate.)
2309 for (i = 0; i < *num_inodes; i++) {
2310 for (j = 1; j < *num_inodes; j++) {
2311 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2313 i_tab[j] = i_tab[j-1];
2325 xfs_inode_t *src_dp,
2326 struct xfs_name *src_name,
2327 xfs_inode_t *src_ip,
2328 xfs_inode_t *target_dp,
2329 struct xfs_name *target_name,
2330 xfs_inode_t *target_ip)
2332 xfs_trans_t *tp = NULL;
2333 xfs_mount_t *mp = src_dp->i_mount;
2334 int new_parent; /* moving to a new dir */
2335 int src_is_directory; /* src_name is a directory */
2337 xfs_bmap_free_t free_list;
2338 xfs_fsblock_t first_block;
2341 xfs_inode_t *inodes[4];
2345 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2347 new_parent = (src_dp != target_dp);
2348 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2350 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2351 inodes, &num_inodes);
2353 xfs_bmap_init(&free_list, &first_block);
2354 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2355 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2356 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2357 error = xfs_trans_reserve(tp, spaceres, XFS_RENAME_LOG_RES(mp), 0,
2358 XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
2359 if (error == ENOSPC) {
2361 error = xfs_trans_reserve(tp, 0, XFS_RENAME_LOG_RES(mp), 0,
2362 XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
2365 xfs_trans_cancel(tp, 0);
2370 * Attach the dquots to the inodes
2372 error = xfs_qm_vop_rename_dqattach(inodes);
2374 xfs_trans_cancel(tp, cancel_flags);
2379 * Lock all the participating inodes. Depending upon whether
2380 * the target_name exists in the target directory, and
2381 * whether the target directory is the same as the source
2382 * directory, we can lock from 2 to 4 inodes.
2384 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2387 * Join all the inodes to the transaction. From this point on,
2388 * we can rely on either trans_commit or trans_cancel to unlock
2391 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2393 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2394 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2396 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2399 * If we are using project inheritance, we only allow renames
2400 * into our tree when the project IDs are the same; else the
2401 * tree quota mechanism would be circumvented.
2403 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2404 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2405 error = XFS_ERROR(EXDEV);
2410 * Set up the target.
2412 if (target_ip == NULL) {
2414 * If there's no space reservation, check the entry will
2415 * fit before actually inserting it.
2417 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2421 * If target does not exist and the rename crosses
2422 * directories, adjust the target directory link count
2423 * to account for the ".." reference from the new entry.
2425 error = xfs_dir_createname(tp, target_dp, target_name,
2426 src_ip->i_ino, &first_block,
2427 &free_list, spaceres);
2428 if (error == ENOSPC)
2433 xfs_trans_ichgtime(tp, target_dp,
2434 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2436 if (new_parent && src_is_directory) {
2437 error = xfs_bumplink(tp, target_dp);
2441 } else { /* target_ip != NULL */
2443 * If target exists and it's a directory, check that both
2444 * target and source are directories and that target can be
2445 * destroyed, or that neither is a directory.
2447 if (S_ISDIR(target_ip->i_d.di_mode)) {
2449 * Make sure target dir is empty.
2451 if (!(xfs_dir_isempty(target_ip)) ||
2452 (target_ip->i_d.di_nlink > 2)) {
2453 error = XFS_ERROR(EEXIST);
2459 * Link the source inode under the target name.
2460 * If the source inode is a directory and we are moving
2461 * it across directories, its ".." entry will be
2462 * inconsistent until we replace that down below.
2464 * In case there is already an entry with the same
2465 * name at the destination directory, remove it first.
2467 error = xfs_dir_replace(tp, target_dp, target_name,
2469 &first_block, &free_list, spaceres);
2473 xfs_trans_ichgtime(tp, target_dp,
2474 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2477 * Decrement the link count on the target since the target
2478 * dir no longer points to it.
2480 error = xfs_droplink(tp, target_ip);
2484 if (src_is_directory) {
2486 * Drop the link from the old "." entry.
2488 error = xfs_droplink(tp, target_ip);
2492 } /* target_ip != NULL */
2495 * Remove the source.
2497 if (new_parent && src_is_directory) {
2499 * Rewrite the ".." entry to point to the new
2502 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2504 &first_block, &free_list, spaceres);
2505 ASSERT(error != EEXIST);
2511 * We always want to hit the ctime on the source inode.
2513 * This isn't strictly required by the standards since the source
2514 * inode isn't really being changed, but old unix file systems did
2515 * it and some incremental backup programs won't work without it.
2517 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2518 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2521 * Adjust the link count on src_dp. This is necessary when
2522 * renaming a directory, either within one parent when
2523 * the target existed, or across two parent directories.
2525 if (src_is_directory && (new_parent || target_ip != NULL)) {
2528 * Decrement link count on src_directory since the
2529 * entry that's moved no longer points to it.
2531 error = xfs_droplink(tp, src_dp);
2536 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2537 &first_block, &free_list, spaceres);
2541 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2542 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2544 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2547 * If this is a synchronous mount, make sure that the
2548 * rename transaction goes to disk before returning to
2551 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2552 xfs_trans_set_sync(tp);
2555 error = xfs_bmap_finish(&tp, &free_list, &committed);
2557 xfs_bmap_cancel(&free_list);
2558 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2564 * trans_commit will unlock src_ip, target_ip & decrement
2565 * the vnode references.
2567 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2570 cancel_flags |= XFS_TRANS_ABORT;
2572 xfs_bmap_cancel(&free_list);
2573 xfs_trans_cancel(tp, cancel_flags);
2583 xfs_mount_t *mp = ip->i_mount;
2584 struct xfs_perag *pag;
2585 unsigned long first_index, mask;
2586 unsigned long inodes_per_cluster;
2588 xfs_inode_t **ilist;
2595 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2597 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2598 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2599 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2603 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2604 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2606 /* really need a gang lookup range call here */
2607 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2608 first_index, inodes_per_cluster);
2612 for (i = 0; i < nr_found; i++) {
2618 * because this is an RCU protected lookup, we could find a
2619 * recently freed or even reallocated inode during the lookup.
2620 * We need to check under the i_flags_lock for a valid inode
2621 * here. Skip it if it is not valid or the wrong inode.
2623 spin_lock(&ip->i_flags_lock);
2625 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2626 spin_unlock(&ip->i_flags_lock);
2629 spin_unlock(&ip->i_flags_lock);
2632 * Do an un-protected check to see if the inode is dirty and
2633 * is a candidate for flushing. These checks will be repeated
2634 * later after the appropriate locks are acquired.
2636 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2640 * Try to get locks. If any are unavailable or it is pinned,
2641 * then this inode cannot be flushed and is skipped.
2644 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2646 if (!xfs_iflock_nowait(iq)) {
2647 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2650 if (xfs_ipincount(iq)) {
2652 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2657 * arriving here means that this inode can be flushed. First
2658 * re-check that it's dirty before flushing.
2660 if (!xfs_inode_clean(iq)) {
2662 error = xfs_iflush_int(iq, bp);
2664 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2665 goto cluster_corrupt_out;
2671 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2675 XFS_STATS_INC(xs_icluster_flushcnt);
2676 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2687 cluster_corrupt_out:
2689 * Corruption detected in the clustering loop. Invalidate the
2690 * inode buffer and shut down the filesystem.
2694 * Clean up the buffer. If it was delwri, just release it --
2695 * brelse can handle it with no problems. If not, shut down the
2696 * filesystem before releasing the buffer.
2698 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2702 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2704 if (!bufwasdelwri) {
2706 * Just like incore_relse: if we have b_iodone functions,
2707 * mark the buffer as an error and call them. Otherwise
2708 * mark it as stale and brelse.
2713 xfs_buf_ioerror(bp, EIO);
2714 xfs_buf_ioend(bp, 0);
2722 * Unlocks the flush lock
2724 xfs_iflush_abort(iq, false);
2727 return XFS_ERROR(EFSCORRUPTED);
2731 * Flush dirty inode metadata into the backing buffer.
2733 * The caller must have the inode lock and the inode flush lock held. The
2734 * inode lock will still be held upon return to the caller, and the inode
2735 * flush lock will be released after the inode has reached the disk.
2737 * The caller must write out the buffer returned in *bpp and release it.
2741 struct xfs_inode *ip,
2742 struct xfs_buf **bpp)
2744 struct xfs_mount *mp = ip->i_mount;
2746 struct xfs_dinode *dip;
2749 XFS_STATS_INC(xs_iflush_count);
2751 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2752 ASSERT(xfs_isiflocked(ip));
2753 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2754 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2758 xfs_iunpin_wait(ip);
2761 * For stale inodes we cannot rely on the backing buffer remaining
2762 * stale in cache for the remaining life of the stale inode and so
2763 * xfs_imap_to_bp() below may give us a buffer that no longer contains
2764 * inodes below. We have to check this after ensuring the inode is
2765 * unpinned so that it is safe to reclaim the stale inode after the
2768 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2774 * This may have been unpinned because the filesystem is shutting
2775 * down forcibly. If that's the case we must not write this inode
2776 * to disk, because the log record didn't make it to disk.
2778 * We also have to remove the log item from the AIL in this case,
2779 * as we wait for an empty AIL as part of the unmount process.
2781 if (XFS_FORCED_SHUTDOWN(mp)) {
2782 error = XFS_ERROR(EIO);
2787 * Get the buffer containing the on-disk inode.
2789 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
2797 * First flush out the inode that xfs_iflush was called with.
2799 error = xfs_iflush_int(ip, bp);
2804 * If the buffer is pinned then push on the log now so we won't
2805 * get stuck waiting in the write for too long.
2807 if (xfs_buf_ispinned(bp))
2808 xfs_log_force(mp, 0);
2812 * see if other inodes can be gathered into this write
2814 error = xfs_iflush_cluster(ip, bp);
2816 goto cluster_corrupt_out;
2823 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2824 cluster_corrupt_out:
2825 error = XFS_ERROR(EFSCORRUPTED);
2828 * Unlocks the flush lock
2830 xfs_iflush_abort(ip, false);
2836 struct xfs_inode *ip,
2839 struct xfs_inode_log_item *iip = ip->i_itemp;
2840 struct xfs_dinode *dip;
2841 struct xfs_mount *mp = ip->i_mount;
2843 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2844 ASSERT(xfs_isiflocked(ip));
2845 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2846 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2847 ASSERT(iip != NULL && iip->ili_fields != 0);
2849 /* set *dip = inode's place in the buffer */
2850 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2852 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2853 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2854 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2855 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2856 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2859 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2860 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2861 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2862 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2863 __func__, ip->i_ino, ip, ip->i_d.di_magic);
2866 if (S_ISREG(ip->i_d.di_mode)) {
2868 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2869 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2870 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2871 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2872 "%s: Bad regular inode %Lu, ptr 0x%p",
2873 __func__, ip->i_ino, ip);
2876 } else if (S_ISDIR(ip->i_d.di_mode)) {
2878 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2879 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2880 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2881 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2882 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2883 "%s: Bad directory inode %Lu, ptr 0x%p",
2884 __func__, ip->i_ino, ip);
2888 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2889 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2890 XFS_RANDOM_IFLUSH_5)) {
2891 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2892 "%s: detected corrupt incore inode %Lu, "
2893 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2894 __func__, ip->i_ino,
2895 ip->i_d.di_nextents + ip->i_d.di_anextents,
2896 ip->i_d.di_nblocks, ip);
2899 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2900 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2901 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2902 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2903 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2908 * Inode item log recovery for v1/v2 inodes are dependent on the
2909 * di_flushiter count for correct sequencing. We bump the flush
2910 * iteration count so we can detect flushes which postdate a log record
2911 * during recovery. This is redundant as we now log every change and
2912 * hence this can't happen but we need to still do it to ensure
2913 * backwards compatibility with old kernels that predate logging all
2916 if (ip->i_d.di_version < 3)
2917 ip->i_d.di_flushiter++;
2920 * Copy the dirty parts of the inode into the on-disk
2921 * inode. We always copy out the core of the inode,
2922 * because if the inode is dirty at all the core must
2925 xfs_dinode_to_disk(dip, &ip->i_d);
2927 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2928 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2929 ip->i_d.di_flushiter = 0;
2932 * If this is really an old format inode and the superblock version
2933 * has not been updated to support only new format inodes, then
2934 * convert back to the old inode format. If the superblock version
2935 * has been updated, then make the conversion permanent.
2937 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2938 if (ip->i_d.di_version == 1) {
2939 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2943 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2944 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2947 * The superblock version has already been bumped,
2948 * so just make the conversion to the new inode
2951 ip->i_d.di_version = 2;
2952 dip->di_version = 2;
2953 ip->i_d.di_onlink = 0;
2955 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2956 memset(&(dip->di_pad[0]), 0,
2957 sizeof(dip->di_pad));
2958 ASSERT(xfs_get_projid(ip) == 0);
2962 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2963 if (XFS_IFORK_Q(ip))
2964 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2965 xfs_inobp_check(mp, bp);
2968 * We've recorded everything logged in the inode, so we'd like to clear
2969 * the ili_fields bits so we don't log and flush things unnecessarily.
2970 * However, we can't stop logging all this information until the data
2971 * we've copied into the disk buffer is written to disk. If we did we
2972 * might overwrite the copy of the inode in the log with all the data
2973 * after re-logging only part of it, and in the face of a crash we
2974 * wouldn't have all the data we need to recover.
2976 * What we do is move the bits to the ili_last_fields field. When
2977 * logging the inode, these bits are moved back to the ili_fields field.
2978 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2979 * know that the information those bits represent is permanently on
2980 * disk. As long as the flush completes before the inode is logged
2981 * again, then both ili_fields and ili_last_fields will be cleared.
2983 * We can play with the ili_fields bits here, because the inode lock
2984 * must be held exclusively in order to set bits there and the flush
2985 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2986 * done routine can tell whether or not to look in the AIL. Also, store
2987 * the current LSN of the inode so that we can tell whether the item has
2988 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2989 * need the AIL lock, because it is a 64 bit value that cannot be read
2992 iip->ili_last_fields = iip->ili_fields;
2993 iip->ili_fields = 0;
2994 iip->ili_logged = 1;
2996 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2997 &iip->ili_item.li_lsn);
3000 * Attach the function xfs_iflush_done to the inode's
3001 * buffer. This will remove the inode from the AIL
3002 * and unlock the inode's flush lock when the inode is
3003 * completely written to disk.
3005 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3007 /* update the lsn in the on disk inode if required */
3008 if (ip->i_d.di_version == 3)
3009 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3011 /* generate the checksum. */
3012 xfs_dinode_calc_crc(mp, dip);
3014 ASSERT(bp->b_fspriv != NULL);
3015 ASSERT(bp->b_iodone != NULL);
3019 return XFS_ERROR(EFSCORRUPTED);