2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_attr_sf.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
51 #include "xfs_bmap_btree.h"
53 kmem_zone_t *xfs_inode_zone;
56 * Used in xfs_itruncate_extents(). This is the maximum number of extents
57 * freed from a file in a single transaction.
59 #define XFS_ITRUNC_MAX_EXTENTS 2
61 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
63 STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
66 * helper function to extract extent size hint from inode
72 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
73 return ip->i_d.di_extsize;
74 if (XFS_IS_REALTIME_INODE(ip))
75 return ip->i_mount->m_sb.sb_rextsize;
80 * These two are wrapper routines around the xfs_ilock() routine used to
81 * centralize some grungy code. They are used in places that wish to lock the
82 * inode solely for reading the extents. The reason these places can't just
83 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
84 * bringing in of the extents from disk for a file in b-tree format. If the
85 * inode is in b-tree format, then we need to lock the inode exclusively until
86 * the extents are read in. Locking it exclusively all the time would limit
87 * our parallelism unnecessarily, though. What we do instead is check to see
88 * if the extents have been read in yet, and only lock the inode exclusively
91 * The functions return a value which should be given to the corresponding
95 xfs_ilock_data_map_shared(
98 uint lock_mode = XFS_ILOCK_SHARED;
100 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
101 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
102 lock_mode = XFS_ILOCK_EXCL;
103 xfs_ilock(ip, lock_mode);
108 xfs_ilock_attr_map_shared(
109 struct xfs_inode *ip)
111 uint lock_mode = XFS_ILOCK_SHARED;
113 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
114 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
115 lock_mode = XFS_ILOCK_EXCL;
116 xfs_ilock(ip, lock_mode);
121 * The xfs inode contains 2 locks: a multi-reader lock called the
122 * i_iolock and a multi-reader lock called the i_lock. This routine
123 * allows either or both of the locks to be obtained.
125 * The 2 locks should always be ordered so that the IO lock is
126 * obtained first in order to prevent deadlock.
128 * ip -- the inode being locked
129 * lock_flags -- this parameter indicates the inode's locks
130 * to be locked. It can be:
135 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
136 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
137 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
138 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
145 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
148 * You can't set both SHARED and EXCL for the same lock,
149 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
150 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
152 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
153 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
154 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
155 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
156 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
158 if (lock_flags & XFS_IOLOCK_EXCL)
159 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
160 else if (lock_flags & XFS_IOLOCK_SHARED)
161 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
163 if (lock_flags & XFS_ILOCK_EXCL)
164 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
165 else if (lock_flags & XFS_ILOCK_SHARED)
166 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
170 * This is just like xfs_ilock(), except that the caller
171 * is guaranteed not to sleep. It returns 1 if it gets
172 * the requested locks and 0 otherwise. If the IO lock is
173 * obtained but the inode lock cannot be, then the IO lock
174 * is dropped before returning.
176 * ip -- the inode being locked
177 * lock_flags -- this parameter indicates the inode's locks to be
178 * to be locked. See the comment for xfs_ilock() for a list
186 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
189 * You can't set both SHARED and EXCL for the same lock,
190 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
191 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
193 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
194 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
195 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
196 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
197 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
199 if (lock_flags & XFS_IOLOCK_EXCL) {
200 if (!mrtryupdate(&ip->i_iolock))
202 } else if (lock_flags & XFS_IOLOCK_SHARED) {
203 if (!mrtryaccess(&ip->i_iolock))
206 if (lock_flags & XFS_ILOCK_EXCL) {
207 if (!mrtryupdate(&ip->i_lock))
208 goto out_undo_iolock;
209 } else if (lock_flags & XFS_ILOCK_SHARED) {
210 if (!mrtryaccess(&ip->i_lock))
211 goto out_undo_iolock;
216 if (lock_flags & XFS_IOLOCK_EXCL)
217 mrunlock_excl(&ip->i_iolock);
218 else if (lock_flags & XFS_IOLOCK_SHARED)
219 mrunlock_shared(&ip->i_iolock);
225 * xfs_iunlock() is used to drop the inode locks acquired with
226 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
227 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
228 * that we know which locks to drop.
230 * ip -- the inode being unlocked
231 * lock_flags -- this parameter indicates the inode's locks to be
232 * to be unlocked. See the comment for xfs_ilock() for a list
233 * of valid values for this parameter.
242 * You can't set both SHARED and EXCL for the same lock,
243 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
244 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
246 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
247 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
248 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
249 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
250 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
251 ASSERT(lock_flags != 0);
253 if (lock_flags & XFS_IOLOCK_EXCL)
254 mrunlock_excl(&ip->i_iolock);
255 else if (lock_flags & XFS_IOLOCK_SHARED)
256 mrunlock_shared(&ip->i_iolock);
258 if (lock_flags & XFS_ILOCK_EXCL)
259 mrunlock_excl(&ip->i_lock);
260 else if (lock_flags & XFS_ILOCK_SHARED)
261 mrunlock_shared(&ip->i_lock);
263 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
267 * give up write locks. the i/o lock cannot be held nested
268 * if it is being demoted.
275 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
276 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
278 if (lock_flags & XFS_ILOCK_EXCL)
279 mrdemote(&ip->i_lock);
280 if (lock_flags & XFS_IOLOCK_EXCL)
281 mrdemote(&ip->i_iolock);
283 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
286 #if defined(DEBUG) || defined(XFS_WARN)
292 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
293 if (!(lock_flags & XFS_ILOCK_SHARED))
294 return !!ip->i_lock.mr_writer;
295 return rwsem_is_locked(&ip->i_lock.mr_lock);
298 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
299 if (!(lock_flags & XFS_IOLOCK_SHARED))
300 return !!ip->i_iolock.mr_writer;
301 return rwsem_is_locked(&ip->i_iolock.mr_lock);
311 int xfs_small_retries;
312 int xfs_middle_retries;
313 int xfs_lots_retries;
318 * Bump the subclass so xfs_lock_inodes() acquires each lock with
322 xfs_lock_inumorder(int lock_mode, int subclass)
324 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
325 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
326 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
327 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
333 * The following routine will lock n inodes in exclusive mode.
334 * We assume the caller calls us with the inodes in i_ino order.
336 * We need to detect deadlock where an inode that we lock
337 * is in the AIL and we start waiting for another inode that is locked
338 * by a thread in a long running transaction (such as truncate). This can
339 * result in deadlock since the long running trans might need to wait
340 * for the inode we just locked in order to push the tail and free space
349 int attempts = 0, i, j, try_lock;
352 ASSERT(ips && (inodes >= 2)); /* we need at least two */
358 for (; i < inodes; i++) {
361 if (i && (ips[i] == ips[i-1])) /* Already locked */
365 * If try_lock is not set yet, make sure all locked inodes
366 * are not in the AIL.
367 * If any are, set try_lock to be used later.
371 for (j = (i - 1); j >= 0 && !try_lock; j--) {
372 lp = (xfs_log_item_t *)ips[j]->i_itemp;
373 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
380 * If any of the previous locks we have locked is in the AIL,
381 * we must TRY to get the second and subsequent locks. If
382 * we can't get any, we must release all we have
387 /* try_lock must be 0 if i is 0. */
389 * try_lock means we have an inode locked
390 * that is in the AIL.
393 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
397 * Unlock all previous guys and try again.
398 * xfs_iunlock will try to push the tail
399 * if the inode is in the AIL.
402 for(j = i - 1; j >= 0; j--) {
405 * Check to see if we've already
407 * Not the first one going back,
408 * and the inode ptr is the same.
410 if ((j != (i - 1)) && ips[j] ==
414 xfs_iunlock(ips[j], lock_mode);
417 if ((attempts % 5) == 0) {
418 delay(1); /* Don't just spin the CPU */
428 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
434 if (attempts < 5) xfs_small_retries++;
435 else if (attempts < 100) xfs_middle_retries++;
436 else xfs_lots_retries++;
444 * xfs_lock_two_inodes() can only be used to lock one type of lock
445 * at a time - the iolock or the ilock, but not both at once. If
446 * we lock both at once, lockdep will report false positives saying
447 * we have violated locking orders.
459 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
460 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
461 ASSERT(ip0->i_ino != ip1->i_ino);
463 if (ip0->i_ino > ip1->i_ino) {
470 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
473 * If the first lock we have locked is in the AIL, we must TRY to get
474 * the second lock. If we can't get it, we must release the first one
477 lp = (xfs_log_item_t *)ip0->i_itemp;
478 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
479 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
480 xfs_iunlock(ip0, lock_mode);
481 if ((++attempts % 5) == 0)
482 delay(1); /* Don't just spin the CPU */
486 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
493 struct xfs_inode *ip)
495 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
496 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
499 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
500 if (xfs_isiflocked(ip))
502 } while (!xfs_iflock_nowait(ip));
504 finish_wait(wq, &wait.wait);
513 if (di_flags & XFS_DIFLAG_ANY) {
514 if (di_flags & XFS_DIFLAG_REALTIME)
515 flags |= XFS_XFLAG_REALTIME;
516 if (di_flags & XFS_DIFLAG_PREALLOC)
517 flags |= XFS_XFLAG_PREALLOC;
518 if (di_flags & XFS_DIFLAG_IMMUTABLE)
519 flags |= XFS_XFLAG_IMMUTABLE;
520 if (di_flags & XFS_DIFLAG_APPEND)
521 flags |= XFS_XFLAG_APPEND;
522 if (di_flags & XFS_DIFLAG_SYNC)
523 flags |= XFS_XFLAG_SYNC;
524 if (di_flags & XFS_DIFLAG_NOATIME)
525 flags |= XFS_XFLAG_NOATIME;
526 if (di_flags & XFS_DIFLAG_NODUMP)
527 flags |= XFS_XFLAG_NODUMP;
528 if (di_flags & XFS_DIFLAG_RTINHERIT)
529 flags |= XFS_XFLAG_RTINHERIT;
530 if (di_flags & XFS_DIFLAG_PROJINHERIT)
531 flags |= XFS_XFLAG_PROJINHERIT;
532 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
533 flags |= XFS_XFLAG_NOSYMLINKS;
534 if (di_flags & XFS_DIFLAG_EXTSIZE)
535 flags |= XFS_XFLAG_EXTSIZE;
536 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
537 flags |= XFS_XFLAG_EXTSZINHERIT;
538 if (di_flags & XFS_DIFLAG_NODEFRAG)
539 flags |= XFS_XFLAG_NODEFRAG;
540 if (di_flags & XFS_DIFLAG_FILESTREAM)
541 flags |= XFS_XFLAG_FILESTREAM;
551 xfs_icdinode_t *dic = &ip->i_d;
553 return _xfs_dic2xflags(dic->di_flags) |
554 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
561 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
562 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
566 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
567 * is allowed, otherwise it has to be an exact match. If a CI match is found,
568 * ci_name->name will point to a the actual name (caller must free) or
569 * will be set to NULL if an exact match is found.
574 struct xfs_name *name,
576 struct xfs_name *ci_name)
582 trace_xfs_lookup(dp, name);
584 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
587 lock_mode = xfs_ilock_data_map_shared(dp);
588 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
589 xfs_iunlock(dp, lock_mode);
594 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
602 kmem_free(ci_name->name);
609 * Allocate an inode on disk and return a copy of its in-core version.
610 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
611 * appropriately within the inode. The uid and gid for the inode are
612 * set according to the contents of the given cred structure.
614 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
615 * has a free inode available, call xfs_iget() to obtain the in-core
616 * version of the allocated inode. Finally, fill in the inode and
617 * log its initial contents. In this case, ialloc_context would be
620 * If xfs_dialloc() does not have an available inode, it will replenish
621 * its supply by doing an allocation. Since we can only do one
622 * allocation within a transaction without deadlocks, we must commit
623 * the current transaction before returning the inode itself.
624 * In this case, therefore, we will set ialloc_context and return.
625 * The caller should then commit the current transaction, start a new
626 * transaction, and call xfs_ialloc() again to actually get the inode.
628 * To ensure that some other process does not grab the inode that
629 * was allocated during the first call to xfs_ialloc(), this routine
630 * also returns the [locked] bp pointing to the head of the freelist
631 * as ialloc_context. The caller should hold this buffer across
632 * the commit and pass it back into this routine on the second call.
634 * If we are allocating quota inodes, we do not have a parent inode
635 * to attach to or associate with (i.e. pip == NULL) because they
636 * are not linked into the directory structure - they are attached
637 * directly to the superblock - and so have no parent.
648 xfs_buf_t **ialloc_context,
651 struct xfs_mount *mp = tp->t_mountp;
659 * Call the space management code to pick
660 * the on-disk inode to be allocated.
662 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
663 ialloc_context, &ino);
666 if (*ialloc_context || ino == NULLFSINO) {
670 ASSERT(*ialloc_context == NULL);
673 * Get the in-core inode with the lock held exclusively.
674 * This is because we're setting fields here we need
675 * to prevent others from looking at until we're done.
677 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
678 XFS_ILOCK_EXCL, &ip);
684 * We always convert v1 inodes to v2 now - we only support filesystems
685 * with >= v2 inode capability, so there is no reason for ever leaving
686 * an inode in v1 format.
688 if (ip->i_d.di_version == 1)
689 ip->i_d.di_version = 2;
691 ip->i_d.di_mode = mode;
692 ip->i_d.di_onlink = 0;
693 ip->i_d.di_nlink = nlink;
694 ASSERT(ip->i_d.di_nlink == nlink);
695 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
696 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
697 xfs_set_projid(ip, prid);
698 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
700 if (pip && XFS_INHERIT_GID(pip)) {
701 ip->i_d.di_gid = pip->i_d.di_gid;
702 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
703 ip->i_d.di_mode |= S_ISGID;
708 * If the group ID of the new file does not match the effective group
709 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
710 * (and only if the irix_sgid_inherit compatibility variable is set).
712 if ((irix_sgid_inherit) &&
713 (ip->i_d.di_mode & S_ISGID) &&
714 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
715 ip->i_d.di_mode &= ~S_ISGID;
719 ip->i_d.di_nextents = 0;
720 ASSERT(ip->i_d.di_nblocks == 0);
722 tv = current_fs_time(mp->m_super);
723 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
724 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
725 ip->i_d.di_atime = ip->i_d.di_mtime;
726 ip->i_d.di_ctime = ip->i_d.di_mtime;
729 * di_gen will have been taken care of in xfs_iread.
731 ip->i_d.di_extsize = 0;
732 ip->i_d.di_dmevmask = 0;
733 ip->i_d.di_dmstate = 0;
734 ip->i_d.di_flags = 0;
736 if (ip->i_d.di_version == 3) {
737 ASSERT(ip->i_d.di_ino == ino);
738 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
740 ip->i_d.di_changecount = 1;
742 ip->i_d.di_flags2 = 0;
743 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
744 ip->i_d.di_crtime = ip->i_d.di_mtime;
748 flags = XFS_ILOG_CORE;
749 switch (mode & S_IFMT) {
754 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
755 ip->i_df.if_u2.if_rdev = rdev;
756 ip->i_df.if_flags = 0;
757 flags |= XFS_ILOG_DEV;
761 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
765 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
766 di_flags |= XFS_DIFLAG_RTINHERIT;
767 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
768 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
769 ip->i_d.di_extsize = pip->i_d.di_extsize;
771 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
772 di_flags |= XFS_DIFLAG_PROJINHERIT;
773 } else if (S_ISREG(mode)) {
774 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
775 di_flags |= XFS_DIFLAG_REALTIME;
776 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
777 di_flags |= XFS_DIFLAG_EXTSIZE;
778 ip->i_d.di_extsize = pip->i_d.di_extsize;
781 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
783 di_flags |= XFS_DIFLAG_NOATIME;
784 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
786 di_flags |= XFS_DIFLAG_NODUMP;
787 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
789 di_flags |= XFS_DIFLAG_SYNC;
790 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
791 xfs_inherit_nosymlinks)
792 di_flags |= XFS_DIFLAG_NOSYMLINKS;
793 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
794 xfs_inherit_nodefrag)
795 di_flags |= XFS_DIFLAG_NODEFRAG;
796 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
797 di_flags |= XFS_DIFLAG_FILESTREAM;
798 ip->i_d.di_flags |= di_flags;
802 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
803 ip->i_df.if_flags = XFS_IFEXTENTS;
804 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
805 ip->i_df.if_u1.if_extents = NULL;
811 * Attribute fork settings for new inode.
813 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
814 ip->i_d.di_anextents = 0;
817 * Log the new values stuffed into the inode.
819 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
820 xfs_trans_log_inode(tp, ip, flags);
822 /* now that we have an i_mode we can setup inode ops and unlock */
830 * Allocates a new inode from disk and return a pointer to the
831 * incore copy. This routine will internally commit the current
832 * transaction and allocate a new one if the Space Manager needed
833 * to do an allocation to replenish the inode free-list.
835 * This routine is designed to be called from xfs_create and
841 xfs_trans_t **tpp, /* input: current transaction;
842 output: may be a new transaction. */
843 xfs_inode_t *dp, /* directory within whose allocate
848 prid_t prid, /* project id */
849 int okalloc, /* ok to allocate new space */
850 xfs_inode_t **ipp, /* pointer to inode; it will be
858 xfs_buf_t *ialloc_context = NULL;
864 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
867 * xfs_ialloc will return a pointer to an incore inode if
868 * the Space Manager has an available inode on the free
869 * list. Otherwise, it will do an allocation and replenish
870 * the freelist. Since we can only do one allocation per
871 * transaction without deadlocks, we will need to commit the
872 * current transaction and start a new one. We will then
873 * need to call xfs_ialloc again to get the inode.
875 * If xfs_ialloc did an allocation to replenish the freelist,
876 * it returns the bp containing the head of the freelist as
877 * ialloc_context. We will hold a lock on it across the
878 * transaction commit so that no other process can steal
879 * the inode(s) that we've just allocated.
881 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
882 &ialloc_context, &ip);
885 * Return an error if we were unable to allocate a new inode.
886 * This should only happen if we run out of space on disk or
887 * encounter a disk error.
893 if (!ialloc_context && !ip) {
899 * If the AGI buffer is non-NULL, then we were unable to get an
900 * inode in one operation. We need to commit the current
901 * transaction and call xfs_ialloc() again. It is guaranteed
902 * to succeed the second time.
904 if (ialloc_context) {
905 struct xfs_trans_res tres;
908 * Normally, xfs_trans_commit releases all the locks.
909 * We call bhold to hang on to the ialloc_context across
910 * the commit. Holding this buffer prevents any other
911 * processes from doing any allocations in this
914 xfs_trans_bhold(tp, ialloc_context);
916 * Save the log reservation so we can use
917 * them in the next transaction.
919 tres.tr_logres = xfs_trans_get_log_res(tp);
920 tres.tr_logcount = xfs_trans_get_log_count(tp);
923 * We want the quota changes to be associated with the next
924 * transaction, NOT this one. So, detach the dqinfo from this
925 * and attach it to the next transaction.
930 dqinfo = (void *)tp->t_dqinfo;
932 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
933 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
936 ntp = xfs_trans_dup(tp);
937 code = xfs_trans_commit(tp, 0);
939 if (committed != NULL) {
943 * If we get an error during the commit processing,
944 * release the buffer that is still held and return
948 xfs_buf_relse(ialloc_context);
950 tp->t_dqinfo = dqinfo;
951 xfs_trans_free_dqinfo(tp);
959 * transaction commit worked ok so we can drop the extra ticket
960 * reference that we gained in xfs_trans_dup()
962 xfs_log_ticket_put(tp->t_ticket);
963 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
964 code = xfs_trans_reserve(tp, &tres, 0, 0);
967 * Re-attach the quota info that we detached from prev trx.
970 tp->t_dqinfo = dqinfo;
971 tp->t_flags |= tflags;
975 xfs_buf_relse(ialloc_context);
980 xfs_trans_bjoin(tp, ialloc_context);
983 * Call ialloc again. Since we've locked out all
984 * other allocations in this allocation group,
985 * this call should always succeed.
987 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
988 okalloc, &ialloc_context, &ip);
991 * If we get an error at this point, return to the caller
992 * so that the current transaction can be aborted.
999 ASSERT(!ialloc_context && ip);
1002 if (committed != NULL)
1013 * Decrement the link count on an inode & log the change.
1014 * If this causes the link count to go to zero, initiate the
1015 * logging activity required to truncate a file.
1024 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1026 ASSERT (ip->i_d.di_nlink > 0);
1028 drop_nlink(VFS_I(ip));
1029 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1032 if (ip->i_d.di_nlink == 0) {
1034 * We're dropping the last link to this file.
1035 * Move the on-disk inode to the AGI unlinked list.
1036 * From xfs_inactive() we will pull the inode from
1037 * the list and free it.
1039 error = xfs_iunlink(tp, ip);
1045 * Increment the link count on an inode & log the change.
1052 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1054 ASSERT(ip->i_d.di_version > 1);
1055 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1057 inc_nlink(VFS_I(ip));
1058 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1065 struct xfs_name *name,
1070 int is_dir = S_ISDIR(mode);
1071 struct xfs_mount *mp = dp->i_mount;
1072 struct xfs_inode *ip = NULL;
1073 struct xfs_trans *tp = NULL;
1075 xfs_bmap_free_t free_list;
1076 xfs_fsblock_t first_block;
1077 bool unlock_dp_on_error = false;
1081 struct xfs_dquot *udqp = NULL;
1082 struct xfs_dquot *gdqp = NULL;
1083 struct xfs_dquot *pdqp = NULL;
1084 struct xfs_trans_res tres;
1087 trace_xfs_create(dp, name);
1089 if (XFS_FORCED_SHUTDOWN(mp))
1092 prid = xfs_get_initial_prid(dp);
1095 * Make sure that we have allocated dquot(s) on disk.
1097 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1098 xfs_kgid_to_gid(current_fsgid()), prid,
1099 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1100 &udqp, &gdqp, &pdqp);
1106 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1107 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1108 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1109 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1111 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1112 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1113 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1114 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1117 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1120 * Initially assume that the file does not exist and
1121 * reserve the resources for that case. If that is not
1122 * the case we'll drop the one we have and get a more
1123 * appropriate transaction later.
1125 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1126 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1127 if (error == -ENOSPC) {
1128 /* flush outstanding delalloc blocks and retry */
1129 xfs_flush_inodes(mp);
1130 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1132 if (error == -ENOSPC) {
1133 /* No space at all so try a "no-allocation" reservation */
1135 error = xfs_trans_reserve(tp, &tres, 0, 0);
1139 goto out_trans_cancel;
1142 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1143 unlock_dp_on_error = true;
1145 xfs_bmap_init(&free_list, &first_block);
1148 * Reserve disk quota and the inode.
1150 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1151 pdqp, resblks, 1, 0);
1153 goto out_trans_cancel;
1156 error = xfs_dir_canenter(tp, dp, name);
1158 goto out_trans_cancel;
1162 * A newly created regular or special file just has one directory
1163 * entry pointing to them, but a directory also the "." entry
1164 * pointing to itself.
1166 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1167 prid, resblks > 0, &ip, &committed);
1169 if (error == -ENOSPC)
1170 goto out_trans_cancel;
1171 goto out_trans_abort;
1175 * Now we join the directory inode to the transaction. We do not do it
1176 * earlier because xfs_dir_ialloc might commit the previous transaction
1177 * (and release all the locks). An error from here on will result in
1178 * the transaction cancel unlocking dp so don't do it explicitly in the
1181 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1182 unlock_dp_on_error = false;
1184 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1185 &first_block, &free_list, resblks ?
1186 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1188 ASSERT(error != -ENOSPC);
1189 goto out_trans_abort;
1191 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1192 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1195 error = xfs_dir_init(tp, ip, dp);
1197 goto out_bmap_cancel;
1199 error = xfs_bumplink(tp, dp);
1201 goto out_bmap_cancel;
1205 * If this is a synchronous mount, make sure that the
1206 * create transaction goes to disk before returning to
1209 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1210 xfs_trans_set_sync(tp);
1213 * Attach the dquot(s) to the inodes and modify them incore.
1214 * These ids of the inode couldn't have changed since the new
1215 * inode has been locked ever since it was created.
1217 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1219 error = xfs_bmap_finish(&tp, &free_list, &committed);
1221 goto out_bmap_cancel;
1223 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1225 goto out_release_inode;
1227 xfs_qm_dqrele(udqp);
1228 xfs_qm_dqrele(gdqp);
1229 xfs_qm_dqrele(pdqp);
1235 xfs_bmap_cancel(&free_list);
1237 cancel_flags |= XFS_TRANS_ABORT;
1239 xfs_trans_cancel(tp, cancel_flags);
1242 * Wait until after the current transaction is aborted to
1243 * release the inode. This prevents recursive transactions
1244 * and deadlocks from xfs_inactive.
1249 xfs_qm_dqrele(udqp);
1250 xfs_qm_dqrele(gdqp);
1251 xfs_qm_dqrele(pdqp);
1253 if (unlock_dp_on_error)
1254 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1260 struct xfs_inode *dp,
1261 struct dentry *dentry,
1263 struct xfs_inode **ipp)
1265 struct xfs_mount *mp = dp->i_mount;
1266 struct xfs_inode *ip = NULL;
1267 struct xfs_trans *tp = NULL;
1269 uint cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1271 struct xfs_dquot *udqp = NULL;
1272 struct xfs_dquot *gdqp = NULL;
1273 struct xfs_dquot *pdqp = NULL;
1274 struct xfs_trans_res *tres;
1277 if (XFS_FORCED_SHUTDOWN(mp))
1280 prid = xfs_get_initial_prid(dp);
1283 * Make sure that we have allocated dquot(s) on disk.
1285 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1286 xfs_kgid_to_gid(current_fsgid()), prid,
1287 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1288 &udqp, &gdqp, &pdqp);
1292 resblks = XFS_IALLOC_SPACE_RES(mp);
1293 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1295 tres = &M_RES(mp)->tr_create_tmpfile;
1296 error = xfs_trans_reserve(tp, tres, resblks, 0);
1297 if (error == -ENOSPC) {
1298 /* No space at all so try a "no-allocation" reservation */
1300 error = xfs_trans_reserve(tp, tres, 0, 0);
1304 goto out_trans_cancel;
1307 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1308 pdqp, resblks, 1, 0);
1310 goto out_trans_cancel;
1312 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1313 prid, resblks > 0, &ip, NULL);
1315 if (error == -ENOSPC)
1316 goto out_trans_cancel;
1317 goto out_trans_abort;
1320 if (mp->m_flags & XFS_MOUNT_WSYNC)
1321 xfs_trans_set_sync(tp);
1324 * Attach the dquot(s) to the inodes and modify them incore.
1325 * These ids of the inode couldn't have changed since the new
1326 * inode has been locked ever since it was created.
1328 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1331 error = xfs_iunlink(tp, ip);
1333 goto out_trans_abort;
1335 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1337 goto out_release_inode;
1339 xfs_qm_dqrele(udqp);
1340 xfs_qm_dqrele(gdqp);
1341 xfs_qm_dqrele(pdqp);
1347 cancel_flags |= XFS_TRANS_ABORT;
1349 xfs_trans_cancel(tp, cancel_flags);
1352 * Wait until after the current transaction is aborted to
1353 * release the inode. This prevents recursive transactions
1354 * and deadlocks from xfs_inactive.
1359 xfs_qm_dqrele(udqp);
1360 xfs_qm_dqrele(gdqp);
1361 xfs_qm_dqrele(pdqp);
1370 struct xfs_name *target_name)
1372 xfs_mount_t *mp = tdp->i_mount;
1375 xfs_bmap_free_t free_list;
1376 xfs_fsblock_t first_block;
1381 trace_xfs_link(tdp, target_name);
1383 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1385 if (XFS_FORCED_SHUTDOWN(mp))
1388 error = xfs_qm_dqattach(sip, 0);
1392 error = xfs_qm_dqattach(tdp, 0);
1396 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1397 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1398 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1399 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1400 if (error == -ENOSPC) {
1402 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1409 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1411 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1412 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1415 * If we are using project inheritance, we only allow hard link
1416 * creation in our tree when the project IDs are the same; else
1417 * the tree quota mechanism could be circumvented.
1419 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1420 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1426 error = xfs_dir_canenter(tp, tdp, target_name);
1431 xfs_bmap_init(&free_list, &first_block);
1433 if (sip->i_d.di_nlink == 0) {
1434 error = xfs_iunlink_remove(tp, sip);
1439 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1440 &first_block, &free_list, resblks);
1443 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1444 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1446 error = xfs_bumplink(tp, sip);
1451 * If this is a synchronous mount, make sure that the
1452 * link transaction goes to disk before returning to
1455 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1456 xfs_trans_set_sync(tp);
1459 error = xfs_bmap_finish (&tp, &free_list, &committed);
1461 xfs_bmap_cancel(&free_list);
1465 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1468 cancel_flags |= XFS_TRANS_ABORT;
1470 xfs_trans_cancel(tp, cancel_flags);
1476 * Free up the underlying blocks past new_size. The new size must be smaller
1477 * than the current size. This routine can be used both for the attribute and
1478 * data fork, and does not modify the inode size, which is left to the caller.
1480 * The transaction passed to this routine must have made a permanent log
1481 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1482 * given transaction and start new ones, so make sure everything involved in
1483 * the transaction is tidy before calling here. Some transaction will be
1484 * returned to the caller to be committed. The incoming transaction must
1485 * already include the inode, and both inode locks must be held exclusively.
1486 * The inode must also be "held" within the transaction. On return the inode
1487 * will be "held" within the returned transaction. This routine does NOT
1488 * require any disk space to be reserved for it within the transaction.
1490 * If we get an error, we must return with the inode locked and linked into the
1491 * current transaction. This keeps things simple for the higher level code,
1492 * because it always knows that the inode is locked and held in the transaction
1493 * that returns to it whether errors occur or not. We don't mark the inode
1494 * dirty on error so that transactions can be easily aborted if possible.
1497 xfs_itruncate_extents(
1498 struct xfs_trans **tpp,
1499 struct xfs_inode *ip,
1501 xfs_fsize_t new_size)
1503 struct xfs_mount *mp = ip->i_mount;
1504 struct xfs_trans *tp = *tpp;
1505 struct xfs_trans *ntp;
1506 xfs_bmap_free_t free_list;
1507 xfs_fsblock_t first_block;
1508 xfs_fileoff_t first_unmap_block;
1509 xfs_fileoff_t last_block;
1510 xfs_filblks_t unmap_len;
1515 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1516 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1517 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1518 ASSERT(new_size <= XFS_ISIZE(ip));
1519 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1520 ASSERT(ip->i_itemp != NULL);
1521 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1522 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1524 trace_xfs_itruncate_extents_start(ip, new_size);
1527 * Since it is possible for space to become allocated beyond
1528 * the end of the file (in a crash where the space is allocated
1529 * but the inode size is not yet updated), simply remove any
1530 * blocks which show up between the new EOF and the maximum
1531 * possible file size. If the first block to be removed is
1532 * beyond the maximum file size (ie it is the same as last_block),
1533 * then there is nothing to do.
1535 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1536 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1537 if (first_unmap_block == last_block)
1540 ASSERT(first_unmap_block < last_block);
1541 unmap_len = last_block - first_unmap_block + 1;
1543 xfs_bmap_init(&free_list, &first_block);
1544 error = xfs_bunmapi(tp, ip,
1545 first_unmap_block, unmap_len,
1546 xfs_bmapi_aflag(whichfork),
1547 XFS_ITRUNC_MAX_EXTENTS,
1548 &first_block, &free_list,
1551 goto out_bmap_cancel;
1554 * Duplicate the transaction that has the permanent
1555 * reservation and commit the old transaction.
1557 error = xfs_bmap_finish(&tp, &free_list, &committed);
1559 xfs_trans_ijoin(tp, ip, 0);
1561 goto out_bmap_cancel;
1565 * Mark the inode dirty so it will be logged and
1566 * moved forward in the log as part of every commit.
1568 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1571 ntp = xfs_trans_dup(tp);
1572 error = xfs_trans_commit(tp, 0);
1575 xfs_trans_ijoin(tp, ip, 0);
1581 * Transaction commit worked ok so we can drop the extra ticket
1582 * reference that we gained in xfs_trans_dup()
1584 xfs_log_ticket_put(tp->t_ticket);
1585 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1591 * Always re-log the inode so that our permanent transaction can keep
1592 * on rolling it forward in the log.
1594 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1596 trace_xfs_itruncate_extents_end(ip, new_size);
1603 * If the bunmapi call encounters an error, return to the caller where
1604 * the transaction can be properly aborted. We just need to make sure
1605 * we're not holding any resources that we were not when we came in.
1607 xfs_bmap_cancel(&free_list);
1615 xfs_mount_t *mp = ip->i_mount;
1618 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1621 /* If this is a read-only mount, don't do this (would generate I/O) */
1622 if (mp->m_flags & XFS_MOUNT_RDONLY)
1625 if (!XFS_FORCED_SHUTDOWN(mp)) {
1629 * If we previously truncated this file and removed old data
1630 * in the process, we want to initiate "early" writeout on
1631 * the last close. This is an attempt to combat the notorious
1632 * NULL files problem which is particularly noticeable from a
1633 * truncate down, buffered (re-)write (delalloc), followed by
1634 * a crash. What we are effectively doing here is
1635 * significantly reducing the time window where we'd otherwise
1636 * be exposed to that problem.
1638 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1640 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1641 if (ip->i_delayed_blks > 0) {
1642 error = filemap_flush(VFS_I(ip)->i_mapping);
1649 if (ip->i_d.di_nlink == 0)
1652 if (xfs_can_free_eofblocks(ip, false)) {
1655 * If we can't get the iolock just skip truncating the blocks
1656 * past EOF because we could deadlock with the mmap_sem
1657 * otherwise. We'll get another chance to drop them once the
1658 * last reference to the inode is dropped, so we'll never leak
1659 * blocks permanently.
1661 * Further, check if the inode is being opened, written and
1662 * closed frequently and we have delayed allocation blocks
1663 * outstanding (e.g. streaming writes from the NFS server),
1664 * truncating the blocks past EOF will cause fragmentation to
1667 * In this case don't do the truncation, either, but we have to
1668 * be careful how we detect this case. Blocks beyond EOF show
1669 * up as i_delayed_blks even when the inode is clean, so we
1670 * need to truncate them away first before checking for a dirty
1671 * release. Hence on the first dirty close we will still remove
1672 * the speculative allocation, but after that we will leave it
1675 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1678 error = xfs_free_eofblocks(mp, ip, true);
1679 if (error && error != -EAGAIN)
1682 /* delalloc blocks after truncation means it really is dirty */
1683 if (ip->i_delayed_blks)
1684 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1690 * xfs_inactive_truncate
1692 * Called to perform a truncate when an inode becomes unlinked.
1695 xfs_inactive_truncate(
1696 struct xfs_inode *ip)
1698 struct xfs_mount *mp = ip->i_mount;
1699 struct xfs_trans *tp;
1702 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1703 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1705 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1706 xfs_trans_cancel(tp, 0);
1710 xfs_ilock(ip, XFS_ILOCK_EXCL);
1711 xfs_trans_ijoin(tp, ip, 0);
1714 * Log the inode size first to prevent stale data exposure in the event
1715 * of a system crash before the truncate completes. See the related
1716 * comment in xfs_setattr_size() for details.
1718 ip->i_d.di_size = 0;
1719 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1721 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1723 goto error_trans_cancel;
1725 ASSERT(ip->i_d.di_nextents == 0);
1727 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1731 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1735 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1737 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1742 * xfs_inactive_ifree()
1744 * Perform the inode free when an inode is unlinked.
1748 struct xfs_inode *ip)
1750 xfs_bmap_free_t free_list;
1751 xfs_fsblock_t first_block;
1753 struct xfs_mount *mp = ip->i_mount;
1754 struct xfs_trans *tp;
1757 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1760 * The ifree transaction might need to allocate blocks for record
1761 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1762 * allow ifree to dip into the reserved block pool if necessary.
1764 * Freeing large sets of inodes generally means freeing inode chunks,
1765 * directory and file data blocks, so this should be relatively safe.
1766 * Only under severe circumstances should it be possible to free enough
1767 * inodes to exhaust the reserve block pool via finobt expansion while
1768 * at the same time not creating free space in the filesystem.
1770 * Send a warning if the reservation does happen to fail, as the inode
1771 * now remains allocated and sits on the unlinked list until the fs is
1774 tp->t_flags |= XFS_TRANS_RESERVE;
1775 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1776 XFS_IFREE_SPACE_RES(mp), 0);
1778 if (error == -ENOSPC) {
1779 xfs_warn_ratelimited(mp,
1780 "Failed to remove inode(s) from unlinked list. "
1781 "Please free space, unmount and run xfs_repair.");
1783 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1785 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1789 xfs_ilock(ip, XFS_ILOCK_EXCL);
1790 xfs_trans_ijoin(tp, ip, 0);
1792 xfs_bmap_init(&free_list, &first_block);
1793 error = xfs_ifree(tp, ip, &free_list);
1796 * If we fail to free the inode, shut down. The cancel
1797 * might do that, we need to make sure. Otherwise the
1798 * inode might be lost for a long time or forever.
1800 if (!XFS_FORCED_SHUTDOWN(mp)) {
1801 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1803 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1805 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1806 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1811 * Credit the quota account(s). The inode is gone.
1813 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1816 * Just ignore errors at this point. There is nothing we can
1817 * do except to try to keep going. Make sure it's not a silent
1820 error = xfs_bmap_finish(&tp, &free_list, &committed);
1822 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1824 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1826 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1829 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1836 * This is called when the vnode reference count for the vnode
1837 * goes to zero. If the file has been unlinked, then it must
1838 * now be truncated. Also, we clear all of the read-ahead state
1839 * kept for the inode here since the file is now closed.
1845 struct xfs_mount *mp;
1850 * If the inode is already free, then there can be nothing
1853 if (ip->i_d.di_mode == 0) {
1854 ASSERT(ip->i_df.if_real_bytes == 0);
1855 ASSERT(ip->i_df.if_broot_bytes == 0);
1861 /* If this is a read-only mount, don't do this (would generate I/O) */
1862 if (mp->m_flags & XFS_MOUNT_RDONLY)
1865 if (ip->i_d.di_nlink != 0) {
1867 * force is true because we are evicting an inode from the
1868 * cache. Post-eof blocks must be freed, lest we end up with
1869 * broken free space accounting.
1871 if (xfs_can_free_eofblocks(ip, true))
1872 xfs_free_eofblocks(mp, ip, false);
1877 if (S_ISREG(ip->i_d.di_mode) &&
1878 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1879 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1882 error = xfs_qm_dqattach(ip, 0);
1886 if (S_ISLNK(ip->i_d.di_mode))
1887 error = xfs_inactive_symlink(ip);
1889 error = xfs_inactive_truncate(ip);
1894 * If there are attributes associated with the file then blow them away
1895 * now. The code calls a routine that recursively deconstructs the
1896 * attribute fork. We need to just commit the current transaction
1897 * because we can't use it for xfs_attr_inactive().
1899 if (ip->i_d.di_anextents > 0) {
1900 ASSERT(ip->i_d.di_forkoff != 0);
1902 error = xfs_attr_inactive(ip);
1908 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1910 ASSERT(ip->i_d.di_anextents == 0);
1915 error = xfs_inactive_ifree(ip);
1920 * Release the dquots held by inode, if any.
1922 xfs_qm_dqdetach(ip);
1926 * This is called when the inode's link count goes to 0.
1927 * We place the on-disk inode on a list in the AGI. It
1928 * will be pulled from this list when the inode is freed.
1945 ASSERT(ip->i_d.di_nlink == 0);
1946 ASSERT(ip->i_d.di_mode != 0);
1951 * Get the agi buffer first. It ensures lock ordering
1954 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1957 agi = XFS_BUF_TO_AGI(agibp);
1960 * Get the index into the agi hash table for the
1961 * list this inode will go on.
1963 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1965 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1966 ASSERT(agi->agi_unlinked[bucket_index]);
1967 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1969 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1971 * There is already another inode in the bucket we need
1972 * to add ourselves to. Add us at the front of the list.
1973 * Here we put the head pointer into our next pointer,
1974 * and then we fall through to point the head at us.
1976 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1981 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1982 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1983 offset = ip->i_imap.im_boffset +
1984 offsetof(xfs_dinode_t, di_next_unlinked);
1986 /* need to recalc the inode CRC if appropriate */
1987 xfs_dinode_calc_crc(mp, dip);
1989 xfs_trans_inode_buf(tp, ibp);
1990 xfs_trans_log_buf(tp, ibp, offset,
1991 (offset + sizeof(xfs_agino_t) - 1));
1992 xfs_inobp_check(mp, ibp);
1996 * Point the bucket head pointer at the inode being inserted.
1999 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2000 offset = offsetof(xfs_agi_t, agi_unlinked) +
2001 (sizeof(xfs_agino_t) * bucket_index);
2002 xfs_trans_log_buf(tp, agibp, offset,
2003 (offset + sizeof(xfs_agino_t) - 1));
2008 * Pull the on-disk inode from the AGI unlinked list.
2021 xfs_agnumber_t agno;
2023 xfs_agino_t next_agino;
2024 xfs_buf_t *last_ibp;
2025 xfs_dinode_t *last_dip = NULL;
2027 int offset, last_offset = 0;
2031 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2034 * Get the agi buffer first. It ensures lock ordering
2037 error = xfs_read_agi(mp, tp, agno, &agibp);
2041 agi = XFS_BUF_TO_AGI(agibp);
2044 * Get the index into the agi hash table for the
2045 * list this inode will go on.
2047 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2049 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2050 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2051 ASSERT(agi->agi_unlinked[bucket_index]);
2053 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2055 * We're at the head of the list. Get the inode's on-disk
2056 * buffer to see if there is anyone after us on the list.
2057 * Only modify our next pointer if it is not already NULLAGINO.
2058 * This saves us the overhead of dealing with the buffer when
2059 * there is no need to change it.
2061 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2064 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2068 next_agino = be32_to_cpu(dip->di_next_unlinked);
2069 ASSERT(next_agino != 0);
2070 if (next_agino != NULLAGINO) {
2071 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2072 offset = ip->i_imap.im_boffset +
2073 offsetof(xfs_dinode_t, di_next_unlinked);
2075 /* need to recalc the inode CRC if appropriate */
2076 xfs_dinode_calc_crc(mp, dip);
2078 xfs_trans_inode_buf(tp, ibp);
2079 xfs_trans_log_buf(tp, ibp, offset,
2080 (offset + sizeof(xfs_agino_t) - 1));
2081 xfs_inobp_check(mp, ibp);
2083 xfs_trans_brelse(tp, ibp);
2086 * Point the bucket head pointer at the next inode.
2088 ASSERT(next_agino != 0);
2089 ASSERT(next_agino != agino);
2090 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2091 offset = offsetof(xfs_agi_t, agi_unlinked) +
2092 (sizeof(xfs_agino_t) * bucket_index);
2093 xfs_trans_log_buf(tp, agibp, offset,
2094 (offset + sizeof(xfs_agino_t) - 1));
2097 * We need to search the list for the inode being freed.
2099 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2101 while (next_agino != agino) {
2102 struct xfs_imap imap;
2105 xfs_trans_brelse(tp, last_ibp);
2108 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2110 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2113 "%s: xfs_imap returned error %d.",
2118 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2122 "%s: xfs_imap_to_bp returned error %d.",
2127 last_offset = imap.im_boffset;
2128 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2129 ASSERT(next_agino != NULLAGINO);
2130 ASSERT(next_agino != 0);
2134 * Now last_ibp points to the buffer previous to us on the
2135 * unlinked list. Pull us from the list.
2137 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2140 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2144 next_agino = be32_to_cpu(dip->di_next_unlinked);
2145 ASSERT(next_agino != 0);
2146 ASSERT(next_agino != agino);
2147 if (next_agino != NULLAGINO) {
2148 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2149 offset = ip->i_imap.im_boffset +
2150 offsetof(xfs_dinode_t, di_next_unlinked);
2152 /* need to recalc the inode CRC if appropriate */
2153 xfs_dinode_calc_crc(mp, dip);
2155 xfs_trans_inode_buf(tp, ibp);
2156 xfs_trans_log_buf(tp, ibp, offset,
2157 (offset + sizeof(xfs_agino_t) - 1));
2158 xfs_inobp_check(mp, ibp);
2160 xfs_trans_brelse(tp, ibp);
2163 * Point the previous inode on the list to the next inode.
2165 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2166 ASSERT(next_agino != 0);
2167 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2169 /* need to recalc the inode CRC if appropriate */
2170 xfs_dinode_calc_crc(mp, last_dip);
2172 xfs_trans_inode_buf(tp, last_ibp);
2173 xfs_trans_log_buf(tp, last_ibp, offset,
2174 (offset + sizeof(xfs_agino_t) - 1));
2175 xfs_inobp_check(mp, last_ibp);
2181 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2182 * inodes that are in memory - they all must be marked stale and attached to
2183 * the cluster buffer.
2187 xfs_inode_t *free_ip,
2191 xfs_mount_t *mp = free_ip->i_mount;
2192 int blks_per_cluster;
2193 int inodes_per_cluster;
2199 xfs_inode_log_item_t *iip;
2200 xfs_log_item_t *lip;
2201 struct xfs_perag *pag;
2203 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2204 blks_per_cluster = xfs_icluster_size_fsb(mp);
2205 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2206 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2208 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2209 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2210 XFS_INO_TO_AGBNO(mp, inum));
2213 * We obtain and lock the backing buffer first in the process
2214 * here, as we have to ensure that any dirty inode that we
2215 * can't get the flush lock on is attached to the buffer.
2216 * If we scan the in-memory inodes first, then buffer IO can
2217 * complete before we get a lock on it, and hence we may fail
2218 * to mark all the active inodes on the buffer stale.
2220 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2221 mp->m_bsize * blks_per_cluster,
2228 * This buffer may not have been correctly initialised as we
2229 * didn't read it from disk. That's not important because we are
2230 * only using to mark the buffer as stale in the log, and to
2231 * attach stale cached inodes on it. That means it will never be
2232 * dispatched for IO. If it is, we want to know about it, and we
2233 * want it to fail. We can acheive this by adding a write
2234 * verifier to the buffer.
2236 bp->b_ops = &xfs_inode_buf_ops;
2239 * Walk the inodes already attached to the buffer and mark them
2240 * stale. These will all have the flush locks held, so an
2241 * in-memory inode walk can't lock them. By marking them all
2242 * stale first, we will not attempt to lock them in the loop
2243 * below as the XFS_ISTALE flag will be set.
2247 if (lip->li_type == XFS_LI_INODE) {
2248 iip = (xfs_inode_log_item_t *)lip;
2249 ASSERT(iip->ili_logged == 1);
2250 lip->li_cb = xfs_istale_done;
2251 xfs_trans_ail_copy_lsn(mp->m_ail,
2252 &iip->ili_flush_lsn,
2253 &iip->ili_item.li_lsn);
2254 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2256 lip = lip->li_bio_list;
2261 * For each inode in memory attempt to add it to the inode
2262 * buffer and set it up for being staled on buffer IO
2263 * completion. This is safe as we've locked out tail pushing
2264 * and flushing by locking the buffer.
2266 * We have already marked every inode that was part of a
2267 * transaction stale above, which means there is no point in
2268 * even trying to lock them.
2270 for (i = 0; i < inodes_per_cluster; i++) {
2273 ip = radix_tree_lookup(&pag->pag_ici_root,
2274 XFS_INO_TO_AGINO(mp, (inum + i)));
2276 /* Inode not in memory, nothing to do */
2283 * because this is an RCU protected lookup, we could
2284 * find a recently freed or even reallocated inode
2285 * during the lookup. We need to check under the
2286 * i_flags_lock for a valid inode here. Skip it if it
2287 * is not valid, the wrong inode or stale.
2289 spin_lock(&ip->i_flags_lock);
2290 if (ip->i_ino != inum + i ||
2291 __xfs_iflags_test(ip, XFS_ISTALE)) {
2292 spin_unlock(&ip->i_flags_lock);
2296 spin_unlock(&ip->i_flags_lock);
2299 * Don't try to lock/unlock the current inode, but we
2300 * _cannot_ skip the other inodes that we did not find
2301 * in the list attached to the buffer and are not
2302 * already marked stale. If we can't lock it, back off
2305 if (ip != free_ip &&
2306 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2314 xfs_iflags_set(ip, XFS_ISTALE);
2317 * we don't need to attach clean inodes or those only
2318 * with unlogged changes (which we throw away, anyway).
2321 if (!iip || xfs_inode_clean(ip)) {
2322 ASSERT(ip != free_ip);
2324 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2328 iip->ili_last_fields = iip->ili_fields;
2329 iip->ili_fields = 0;
2330 iip->ili_logged = 1;
2331 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2332 &iip->ili_item.li_lsn);
2334 xfs_buf_attach_iodone(bp, xfs_istale_done,
2338 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2341 xfs_trans_stale_inode_buf(tp, bp);
2342 xfs_trans_binval(tp, bp);
2350 * This is called to return an inode to the inode free list.
2351 * The inode should already be truncated to 0 length and have
2352 * no pages associated with it. This routine also assumes that
2353 * the inode is already a part of the transaction.
2355 * The on-disk copy of the inode will have been added to the list
2356 * of unlinked inodes in the AGI. We need to remove the inode from
2357 * that list atomically with respect to freeing it here.
2363 xfs_bmap_free_t *flist)
2367 xfs_ino_t first_ino;
2369 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2370 ASSERT(ip->i_d.di_nlink == 0);
2371 ASSERT(ip->i_d.di_nextents == 0);
2372 ASSERT(ip->i_d.di_anextents == 0);
2373 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2374 ASSERT(ip->i_d.di_nblocks == 0);
2377 * Pull the on-disk inode from the AGI unlinked list.
2379 error = xfs_iunlink_remove(tp, ip);
2383 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2387 ip->i_d.di_mode = 0; /* mark incore inode as free */
2388 ip->i_d.di_flags = 0;
2389 ip->i_d.di_dmevmask = 0;
2390 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2391 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2392 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2394 * Bump the generation count so no one will be confused
2395 * by reincarnations of this inode.
2398 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2401 error = xfs_ifree_cluster(ip, tp, first_ino);
2407 * This is called to unpin an inode. The caller must have the inode locked
2408 * in at least shared mode so that the buffer cannot be subsequently pinned
2409 * once someone is waiting for it to be unpinned.
2413 struct xfs_inode *ip)
2415 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2417 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2419 /* Give the log a push to start the unpinning I/O */
2420 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2426 struct xfs_inode *ip)
2428 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2429 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2434 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2435 if (xfs_ipincount(ip))
2437 } while (xfs_ipincount(ip));
2438 finish_wait(wq, &wait.wait);
2443 struct xfs_inode *ip)
2445 if (xfs_ipincount(ip))
2446 __xfs_iunpin_wait(ip);
2450 * Removing an inode from the namespace involves removing the directory entry
2451 * and dropping the link count on the inode. Removing the directory entry can
2452 * result in locking an AGF (directory blocks were freed) and removing a link
2453 * count can result in placing the inode on an unlinked list which results in
2456 * The big problem here is that we have an ordering constraint on AGF and AGI
2457 * locking - inode allocation locks the AGI, then can allocate a new extent for
2458 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2459 * removes the inode from the unlinked list, requiring that we lock the AGI
2460 * first, and then freeing the inode can result in an inode chunk being freed
2461 * and hence freeing disk space requiring that we lock an AGF.
2463 * Hence the ordering that is imposed by other parts of the code is AGI before
2464 * AGF. This means we cannot remove the directory entry before we drop the inode
2465 * reference count and put it on the unlinked list as this results in a lock
2466 * order of AGF then AGI, and this can deadlock against inode allocation and
2467 * freeing. Therefore we must drop the link counts before we remove the
2470 * This is still safe from a transactional point of view - it is not until we
2471 * get to xfs_bmap_finish() that we have the possibility of multiple
2472 * transactions in this operation. Hence as long as we remove the directory
2473 * entry and drop the link count in the first transaction of the remove
2474 * operation, there are no transactional constraints on the ordering here.
2479 struct xfs_name *name,
2482 xfs_mount_t *mp = dp->i_mount;
2483 xfs_trans_t *tp = NULL;
2484 int is_dir = S_ISDIR(ip->i_d.di_mode);
2486 xfs_bmap_free_t free_list;
2487 xfs_fsblock_t first_block;
2494 trace_xfs_remove(dp, name);
2496 if (XFS_FORCED_SHUTDOWN(mp))
2499 error = xfs_qm_dqattach(dp, 0);
2503 error = xfs_qm_dqattach(ip, 0);
2508 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2509 log_count = XFS_DEFAULT_LOG_COUNT;
2511 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2512 log_count = XFS_REMOVE_LOG_COUNT;
2514 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2517 * We try to get the real space reservation first,
2518 * allowing for directory btree deletion(s) implying
2519 * possible bmap insert(s). If we can't get the space
2520 * reservation then we use 0 instead, and avoid the bmap
2521 * btree insert(s) in the directory code by, if the bmap
2522 * insert tries to happen, instead trimming the LAST
2523 * block from the directory.
2525 resblks = XFS_REMOVE_SPACE_RES(mp);
2526 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2527 if (error == -ENOSPC) {
2529 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2532 ASSERT(error != -ENOSPC);
2534 goto out_trans_cancel;
2537 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2539 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2540 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2543 * If we're removing a directory perform some additional validation.
2545 cancel_flags |= XFS_TRANS_ABORT;
2547 ASSERT(ip->i_d.di_nlink >= 2);
2548 if (ip->i_d.di_nlink != 2) {
2550 goto out_trans_cancel;
2552 if (!xfs_dir_isempty(ip)) {
2554 goto out_trans_cancel;
2557 /* Drop the link from ip's "..". */
2558 error = xfs_droplink(tp, dp);
2560 goto out_trans_cancel;
2562 /* Drop the "." link from ip to self. */
2563 error = xfs_droplink(tp, ip);
2565 goto out_trans_cancel;
2568 * When removing a non-directory we need to log the parent
2569 * inode here. For a directory this is done implicitly
2570 * by the xfs_droplink call for the ".." entry.
2572 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2574 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2576 /* Drop the link from dp to ip. */
2577 error = xfs_droplink(tp, ip);
2579 goto out_trans_cancel;
2581 /* Determine if this is the last link while the inode is locked */
2582 link_zero = (ip->i_d.di_nlink == 0);
2584 xfs_bmap_init(&free_list, &first_block);
2585 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2586 &first_block, &free_list, resblks);
2588 ASSERT(error != -ENOENT);
2589 goto out_bmap_cancel;
2593 * If this is a synchronous mount, make sure that the
2594 * remove transaction goes to disk before returning to
2597 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2598 xfs_trans_set_sync(tp);
2600 error = xfs_bmap_finish(&tp, &free_list, &committed);
2602 goto out_bmap_cancel;
2604 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2608 if (is_dir && xfs_inode_is_filestream(ip))
2609 xfs_filestream_deassociate(ip);
2614 xfs_bmap_cancel(&free_list);
2616 xfs_trans_cancel(tp, cancel_flags);
2622 * Enter all inodes for a rename transaction into a sorted array.
2625 xfs_sort_for_rename(
2626 xfs_inode_t *dp1, /* in: old (source) directory inode */
2627 xfs_inode_t *dp2, /* in: new (target) directory inode */
2628 xfs_inode_t *ip1, /* in: inode of old entry */
2629 xfs_inode_t *ip2, /* in: inode of new entry, if it
2630 already exists, NULL otherwise. */
2631 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2632 int *num_inodes) /* out: number of inodes in array */
2638 * i_tab contains a list of pointers to inodes. We initialize
2639 * the table here & we'll sort it. We will then use it to
2640 * order the acquisition of the inode locks.
2642 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2656 * Sort the elements via bubble sort. (Remember, there are at
2657 * most 4 elements to sort, so this is adequate.)
2659 for (i = 0; i < *num_inodes; i++) {
2660 for (j = 1; j < *num_inodes; j++) {
2661 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2663 i_tab[j] = i_tab[j-1];
2675 xfs_inode_t *src_dp,
2676 struct xfs_name *src_name,
2677 xfs_inode_t *src_ip,
2678 xfs_inode_t *target_dp,
2679 struct xfs_name *target_name,
2680 xfs_inode_t *target_ip)
2682 xfs_trans_t *tp = NULL;
2683 xfs_mount_t *mp = src_dp->i_mount;
2684 int new_parent; /* moving to a new dir */
2685 int src_is_directory; /* src_name is a directory */
2687 xfs_bmap_free_t free_list;
2688 xfs_fsblock_t first_block;
2691 xfs_inode_t *inodes[4];
2695 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2697 new_parent = (src_dp != target_dp);
2698 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2700 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2701 inodes, &num_inodes);
2703 xfs_bmap_init(&free_list, &first_block);
2704 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2705 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2706 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2707 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2708 if (error == -ENOSPC) {
2710 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2713 xfs_trans_cancel(tp, 0);
2718 * Attach the dquots to the inodes
2720 error = xfs_qm_vop_rename_dqattach(inodes);
2722 xfs_trans_cancel(tp, cancel_flags);
2727 * Lock all the participating inodes. Depending upon whether
2728 * the target_name exists in the target directory, and
2729 * whether the target directory is the same as the source
2730 * directory, we can lock from 2 to 4 inodes.
2732 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2735 * Join all the inodes to the transaction. From this point on,
2736 * we can rely on either trans_commit or trans_cancel to unlock
2739 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2741 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2742 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2744 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2747 * If we are using project inheritance, we only allow renames
2748 * into our tree when the project IDs are the same; else the
2749 * tree quota mechanism would be circumvented.
2751 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2752 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2758 * Set up the target.
2760 if (target_ip == NULL) {
2762 * If there's no space reservation, check the entry will
2763 * fit before actually inserting it.
2766 error = xfs_dir_canenter(tp, target_dp, target_name);
2771 * If target does not exist and the rename crosses
2772 * directories, adjust the target directory link count
2773 * to account for the ".." reference from the new entry.
2775 error = xfs_dir_createname(tp, target_dp, target_name,
2776 src_ip->i_ino, &first_block,
2777 &free_list, spaceres);
2778 if (error == -ENOSPC)
2783 xfs_trans_ichgtime(tp, target_dp,
2784 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2786 if (new_parent && src_is_directory) {
2787 error = xfs_bumplink(tp, target_dp);
2791 } else { /* target_ip != NULL */
2793 * If target exists and it's a directory, check that both
2794 * target and source are directories and that target can be
2795 * destroyed, or that neither is a directory.
2797 if (S_ISDIR(target_ip->i_d.di_mode)) {
2799 * Make sure target dir is empty.
2801 if (!(xfs_dir_isempty(target_ip)) ||
2802 (target_ip->i_d.di_nlink > 2)) {
2809 * Link the source inode under the target name.
2810 * If the source inode is a directory and we are moving
2811 * it across directories, its ".." entry will be
2812 * inconsistent until we replace that down below.
2814 * In case there is already an entry with the same
2815 * name at the destination directory, remove it first.
2817 error = xfs_dir_replace(tp, target_dp, target_name,
2819 &first_block, &free_list, spaceres);
2823 xfs_trans_ichgtime(tp, target_dp,
2824 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2827 * Decrement the link count on the target since the target
2828 * dir no longer points to it.
2830 error = xfs_droplink(tp, target_ip);
2834 if (src_is_directory) {
2836 * Drop the link from the old "." entry.
2838 error = xfs_droplink(tp, target_ip);
2842 } /* target_ip != NULL */
2845 * Remove the source.
2847 if (new_parent && src_is_directory) {
2849 * Rewrite the ".." entry to point to the new
2852 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2854 &first_block, &free_list, spaceres);
2855 ASSERT(error != -EEXIST);
2861 * We always want to hit the ctime on the source inode.
2863 * This isn't strictly required by the standards since the source
2864 * inode isn't really being changed, but old unix file systems did
2865 * it and some incremental backup programs won't work without it.
2867 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2868 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2871 * Adjust the link count on src_dp. This is necessary when
2872 * renaming a directory, either within one parent when
2873 * the target existed, or across two parent directories.
2875 if (src_is_directory && (new_parent || target_ip != NULL)) {
2878 * Decrement link count on src_directory since the
2879 * entry that's moved no longer points to it.
2881 error = xfs_droplink(tp, src_dp);
2886 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2887 &first_block, &free_list, spaceres);
2891 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2892 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2894 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2897 * If this is a synchronous mount, make sure that the
2898 * rename transaction goes to disk before returning to
2901 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2902 xfs_trans_set_sync(tp);
2905 error = xfs_bmap_finish(&tp, &free_list, &committed);
2907 xfs_bmap_cancel(&free_list);
2908 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2914 * trans_commit will unlock src_ip, target_ip & decrement
2915 * the vnode references.
2917 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2920 cancel_flags |= XFS_TRANS_ABORT;
2922 xfs_bmap_cancel(&free_list);
2923 xfs_trans_cancel(tp, cancel_flags);
2933 xfs_mount_t *mp = ip->i_mount;
2934 struct xfs_perag *pag;
2935 unsigned long first_index, mask;
2936 unsigned long inodes_per_cluster;
2938 xfs_inode_t **ilist;
2945 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2947 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
2948 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2949 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2953 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
2954 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2956 /* really need a gang lookup range call here */
2957 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2958 first_index, inodes_per_cluster);
2962 for (i = 0; i < nr_found; i++) {
2968 * because this is an RCU protected lookup, we could find a
2969 * recently freed or even reallocated inode during the lookup.
2970 * We need to check under the i_flags_lock for a valid inode
2971 * here. Skip it if it is not valid or the wrong inode.
2973 spin_lock(&ip->i_flags_lock);
2975 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2976 spin_unlock(&ip->i_flags_lock);
2979 spin_unlock(&ip->i_flags_lock);
2982 * Do an un-protected check to see if the inode is dirty and
2983 * is a candidate for flushing. These checks will be repeated
2984 * later after the appropriate locks are acquired.
2986 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2990 * Try to get locks. If any are unavailable or it is pinned,
2991 * then this inode cannot be flushed and is skipped.
2994 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2996 if (!xfs_iflock_nowait(iq)) {
2997 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3000 if (xfs_ipincount(iq)) {
3002 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3007 * arriving here means that this inode can be flushed. First
3008 * re-check that it's dirty before flushing.
3010 if (!xfs_inode_clean(iq)) {
3012 error = xfs_iflush_int(iq, bp);
3014 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3015 goto cluster_corrupt_out;
3021 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3025 XFS_STATS_INC(xs_icluster_flushcnt);
3026 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3037 cluster_corrupt_out:
3039 * Corruption detected in the clustering loop. Invalidate the
3040 * inode buffer and shut down the filesystem.
3044 * Clean up the buffer. If it was delwri, just release it --
3045 * brelse can handle it with no problems. If not, shut down the
3046 * filesystem before releasing the buffer.
3048 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3052 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3054 if (!bufwasdelwri) {
3056 * Just like incore_relse: if we have b_iodone functions,
3057 * mark the buffer as an error and call them. Otherwise
3058 * mark it as stale and brelse.
3063 xfs_buf_ioerror(bp, -EIO);
3072 * Unlocks the flush lock
3074 xfs_iflush_abort(iq, false);
3077 return -EFSCORRUPTED;
3081 * Flush dirty inode metadata into the backing buffer.
3083 * The caller must have the inode lock and the inode flush lock held. The
3084 * inode lock will still be held upon return to the caller, and the inode
3085 * flush lock will be released after the inode has reached the disk.
3087 * The caller must write out the buffer returned in *bpp and release it.
3091 struct xfs_inode *ip,
3092 struct xfs_buf **bpp)
3094 struct xfs_mount *mp = ip->i_mount;
3096 struct xfs_dinode *dip;
3099 XFS_STATS_INC(xs_iflush_count);
3101 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3102 ASSERT(xfs_isiflocked(ip));
3103 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3104 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3108 xfs_iunpin_wait(ip);
3111 * For stale inodes we cannot rely on the backing buffer remaining
3112 * stale in cache for the remaining life of the stale inode and so
3113 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3114 * inodes below. We have to check this after ensuring the inode is
3115 * unpinned so that it is safe to reclaim the stale inode after the
3118 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3124 * This may have been unpinned because the filesystem is shutting
3125 * down forcibly. If that's the case we must not write this inode
3126 * to disk, because the log record didn't make it to disk.
3128 * We also have to remove the log item from the AIL in this case,
3129 * as we wait for an empty AIL as part of the unmount process.
3131 if (XFS_FORCED_SHUTDOWN(mp)) {
3137 * Get the buffer containing the on-disk inode.
3139 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3147 * First flush out the inode that xfs_iflush was called with.
3149 error = xfs_iflush_int(ip, bp);
3154 * If the buffer is pinned then push on the log now so we won't
3155 * get stuck waiting in the write for too long.
3157 if (xfs_buf_ispinned(bp))
3158 xfs_log_force(mp, 0);
3162 * see if other inodes can be gathered into this write
3164 error = xfs_iflush_cluster(ip, bp);
3166 goto cluster_corrupt_out;
3173 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3174 cluster_corrupt_out:
3175 error = -EFSCORRUPTED;
3178 * Unlocks the flush lock
3180 xfs_iflush_abort(ip, false);
3186 struct xfs_inode *ip,
3189 struct xfs_inode_log_item *iip = ip->i_itemp;
3190 struct xfs_dinode *dip;
3191 struct xfs_mount *mp = ip->i_mount;
3193 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3194 ASSERT(xfs_isiflocked(ip));
3195 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3196 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3197 ASSERT(iip != NULL && iip->ili_fields != 0);
3198 ASSERT(ip->i_d.di_version > 1);
3200 /* set *dip = inode's place in the buffer */
3201 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3203 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3204 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3205 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3206 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3207 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3210 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3211 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3212 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3213 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3214 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3217 if (S_ISREG(ip->i_d.di_mode)) {
3219 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3220 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3221 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3222 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3223 "%s: Bad regular inode %Lu, ptr 0x%p",
3224 __func__, ip->i_ino, ip);
3227 } else if (S_ISDIR(ip->i_d.di_mode)) {
3229 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3230 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3231 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3232 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3233 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3234 "%s: Bad directory inode %Lu, ptr 0x%p",
3235 __func__, ip->i_ino, ip);
3239 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3240 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3241 XFS_RANDOM_IFLUSH_5)) {
3242 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3243 "%s: detected corrupt incore inode %Lu, "
3244 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3245 __func__, ip->i_ino,
3246 ip->i_d.di_nextents + ip->i_d.di_anextents,
3247 ip->i_d.di_nblocks, ip);
3250 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3251 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3252 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3253 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3254 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3259 * Inode item log recovery for v2 inodes are dependent on the
3260 * di_flushiter count for correct sequencing. We bump the flush
3261 * iteration count so we can detect flushes which postdate a log record
3262 * during recovery. This is redundant as we now log every change and
3263 * hence this can't happen but we need to still do it to ensure
3264 * backwards compatibility with old kernels that predate logging all
3267 if (ip->i_d.di_version < 3)
3268 ip->i_d.di_flushiter++;
3271 * Copy the dirty parts of the inode into the on-disk
3272 * inode. We always copy out the core of the inode,
3273 * because if the inode is dirty at all the core must
3276 xfs_dinode_to_disk(dip, &ip->i_d);
3278 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3279 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3280 ip->i_d.di_flushiter = 0;
3282 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3283 if (XFS_IFORK_Q(ip))
3284 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3285 xfs_inobp_check(mp, bp);
3288 * We've recorded everything logged in the inode, so we'd like to clear
3289 * the ili_fields bits so we don't log and flush things unnecessarily.
3290 * However, we can't stop logging all this information until the data
3291 * we've copied into the disk buffer is written to disk. If we did we
3292 * might overwrite the copy of the inode in the log with all the data
3293 * after re-logging only part of it, and in the face of a crash we
3294 * wouldn't have all the data we need to recover.
3296 * What we do is move the bits to the ili_last_fields field. When
3297 * logging the inode, these bits are moved back to the ili_fields field.
3298 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3299 * know that the information those bits represent is permanently on
3300 * disk. As long as the flush completes before the inode is logged
3301 * again, then both ili_fields and ili_last_fields will be cleared.
3303 * We can play with the ili_fields bits here, because the inode lock
3304 * must be held exclusively in order to set bits there and the flush
3305 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3306 * done routine can tell whether or not to look in the AIL. Also, store
3307 * the current LSN of the inode so that we can tell whether the item has
3308 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3309 * need the AIL lock, because it is a 64 bit value that cannot be read
3312 iip->ili_last_fields = iip->ili_fields;
3313 iip->ili_fields = 0;
3314 iip->ili_logged = 1;
3316 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3317 &iip->ili_item.li_lsn);
3320 * Attach the function xfs_iflush_done to the inode's
3321 * buffer. This will remove the inode from the AIL
3322 * and unlock the inode's flush lock when the inode is
3323 * completely written to disk.
3325 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3327 /* update the lsn in the on disk inode if required */
3328 if (ip->i_d.di_version == 3)
3329 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3331 /* generate the checksum. */
3332 xfs_dinode_calc_crc(mp, dip);
3334 ASSERT(bp->b_fspriv != NULL);
3335 ASSERT(bp->b_iodone != NULL);
3339 return -EFSCORRUPTED;