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"
29 #include "xfs_mount.h"
30 #include "xfs_inode.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
34 #include "xfs_attr_sf.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_bmap_util.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_dinode.h"
46 #include "xfs_filestream.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
50 #include "xfs_symlink.h"
51 #include "xfs_trans_priv.h"
53 #include "xfs_bmap_btree.h"
55 kmem_zone_t *xfs_inode_zone;
58 * Used in xfs_itruncate_extents(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_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 * This is a wrapper routine around the xfs_ilock() routine used to centralize
81 * some grungy code. It is used in places that wish to lock the inode solely
82 * for reading the extents. The reason these places can't just call
83 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
84 * extents from disk for a file in b-tree format. If the inode is in b-tree
85 * format, then we need to lock the inode exclusively until the extents are read
86 * in. Locking it exclusively all the time would limit our parallelism
87 * unnecessarily, though. What we do instead is check to see if the extents
88 * have been read in yet, and only lock the inode exclusively if they have not.
90 * The function returns a value which should be given to the corresponding
94 xfs_ilock_data_map_shared(
97 uint lock_mode = XFS_ILOCK_SHARED;
99 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
101 lock_mode = XFS_ILOCK_EXCL;
102 xfs_ilock(ip, lock_mode);
107 * The xfs inode contains 2 locks: a multi-reader lock called the
108 * i_iolock and a multi-reader lock called the i_lock. This routine
109 * allows either or both of the locks to be obtained.
111 * The 2 locks should always be ordered so that the IO lock is
112 * obtained first in order to prevent deadlock.
114 * ip -- the inode being locked
115 * lock_flags -- this parameter indicates the inode's locks
116 * to be locked. It can be:
121 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
122 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
123 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
124 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
131 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
134 * You can't set both SHARED and EXCL for the same lock,
135 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
136 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
138 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
139 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
140 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
141 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
142 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
144 if (lock_flags & XFS_IOLOCK_EXCL)
145 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
146 else if (lock_flags & XFS_IOLOCK_SHARED)
147 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
149 if (lock_flags & XFS_ILOCK_EXCL)
150 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
151 else if (lock_flags & XFS_ILOCK_SHARED)
152 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
156 * This is just like xfs_ilock(), except that the caller
157 * is guaranteed not to sleep. It returns 1 if it gets
158 * the requested locks and 0 otherwise. If the IO lock is
159 * obtained but the inode lock cannot be, then the IO lock
160 * is dropped before returning.
162 * ip -- the inode being locked
163 * lock_flags -- this parameter indicates the inode's locks to be
164 * to be locked. See the comment for xfs_ilock() for a list
172 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
175 * You can't set both SHARED and EXCL for the same lock,
176 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
177 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
179 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
180 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
181 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
182 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
183 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
185 if (lock_flags & XFS_IOLOCK_EXCL) {
186 if (!mrtryupdate(&ip->i_iolock))
188 } else if (lock_flags & XFS_IOLOCK_SHARED) {
189 if (!mrtryaccess(&ip->i_iolock))
192 if (lock_flags & XFS_ILOCK_EXCL) {
193 if (!mrtryupdate(&ip->i_lock))
194 goto out_undo_iolock;
195 } else if (lock_flags & XFS_ILOCK_SHARED) {
196 if (!mrtryaccess(&ip->i_lock))
197 goto out_undo_iolock;
202 if (lock_flags & XFS_IOLOCK_EXCL)
203 mrunlock_excl(&ip->i_iolock);
204 else if (lock_flags & XFS_IOLOCK_SHARED)
205 mrunlock_shared(&ip->i_iolock);
211 * xfs_iunlock() is used to drop the inode locks acquired with
212 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
213 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
214 * that we know which locks to drop.
216 * ip -- the inode being unlocked
217 * lock_flags -- this parameter indicates the inode's locks to be
218 * to be unlocked. See the comment for xfs_ilock() for a list
219 * of valid values for this parameter.
228 * You can't set both SHARED and EXCL for the same lock,
229 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
230 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
232 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
233 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
234 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
235 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
236 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
237 ASSERT(lock_flags != 0);
239 if (lock_flags & XFS_IOLOCK_EXCL)
240 mrunlock_excl(&ip->i_iolock);
241 else if (lock_flags & XFS_IOLOCK_SHARED)
242 mrunlock_shared(&ip->i_iolock);
244 if (lock_flags & XFS_ILOCK_EXCL)
245 mrunlock_excl(&ip->i_lock);
246 else if (lock_flags & XFS_ILOCK_SHARED)
247 mrunlock_shared(&ip->i_lock);
249 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
253 * give up write locks. the i/o lock cannot be held nested
254 * if it is being demoted.
261 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
262 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
264 if (lock_flags & XFS_ILOCK_EXCL)
265 mrdemote(&ip->i_lock);
266 if (lock_flags & XFS_IOLOCK_EXCL)
267 mrdemote(&ip->i_iolock);
269 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
272 #if defined(DEBUG) || defined(XFS_WARN)
278 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
279 if (!(lock_flags & XFS_ILOCK_SHARED))
280 return !!ip->i_lock.mr_writer;
281 return rwsem_is_locked(&ip->i_lock.mr_lock);
284 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
285 if (!(lock_flags & XFS_IOLOCK_SHARED))
286 return !!ip->i_iolock.mr_writer;
287 return rwsem_is_locked(&ip->i_iolock.mr_lock);
297 int xfs_small_retries;
298 int xfs_middle_retries;
299 int xfs_lots_retries;
304 * Bump the subclass so xfs_lock_inodes() acquires each lock with
308 xfs_lock_inumorder(int lock_mode, int subclass)
310 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
311 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
312 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
313 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
319 * The following routine will lock n inodes in exclusive mode.
320 * We assume the caller calls us with the inodes in i_ino order.
322 * We need to detect deadlock where an inode that we lock
323 * is in the AIL and we start waiting for another inode that is locked
324 * by a thread in a long running transaction (such as truncate). This can
325 * result in deadlock since the long running trans might need to wait
326 * for the inode we just locked in order to push the tail and free space
335 int attempts = 0, i, j, try_lock;
338 ASSERT(ips && (inodes >= 2)); /* we need at least two */
344 for (; i < inodes; i++) {
347 if (i && (ips[i] == ips[i-1])) /* Already locked */
351 * If try_lock is not set yet, make sure all locked inodes
352 * are not in the AIL.
353 * If any are, set try_lock to be used later.
357 for (j = (i - 1); j >= 0 && !try_lock; j--) {
358 lp = (xfs_log_item_t *)ips[j]->i_itemp;
359 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
366 * If any of the previous locks we have locked is in the AIL,
367 * we must TRY to get the second and subsequent locks. If
368 * we can't get any, we must release all we have
373 /* try_lock must be 0 if i is 0. */
375 * try_lock means we have an inode locked
376 * that is in the AIL.
379 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
383 * Unlock all previous guys and try again.
384 * xfs_iunlock will try to push the tail
385 * if the inode is in the AIL.
388 for(j = i - 1; j >= 0; j--) {
391 * Check to see if we've already
393 * Not the first one going back,
394 * and the inode ptr is the same.
396 if ((j != (i - 1)) && ips[j] ==
400 xfs_iunlock(ips[j], lock_mode);
403 if ((attempts % 5) == 0) {
404 delay(1); /* Don't just spin the CPU */
414 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
420 if (attempts < 5) xfs_small_retries++;
421 else if (attempts < 100) xfs_middle_retries++;
422 else xfs_lots_retries++;
430 * xfs_lock_two_inodes() can only be used to lock one type of lock
431 * at a time - the iolock or the ilock, but not both at once. If
432 * we lock both at once, lockdep will report false positives saying
433 * we have violated locking orders.
445 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
446 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
447 ASSERT(ip0->i_ino != ip1->i_ino);
449 if (ip0->i_ino > ip1->i_ino) {
456 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
459 * If the first lock we have locked is in the AIL, we must TRY to get
460 * the second lock. If we can't get it, we must release the first one
463 lp = (xfs_log_item_t *)ip0->i_itemp;
464 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
465 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
466 xfs_iunlock(ip0, lock_mode);
467 if ((++attempts % 5) == 0)
468 delay(1); /* Don't just spin the CPU */
472 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
479 struct xfs_inode *ip)
481 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
482 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
485 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
486 if (xfs_isiflocked(ip))
488 } while (!xfs_iflock_nowait(ip));
490 finish_wait(wq, &wait.wait);
499 if (di_flags & XFS_DIFLAG_ANY) {
500 if (di_flags & XFS_DIFLAG_REALTIME)
501 flags |= XFS_XFLAG_REALTIME;
502 if (di_flags & XFS_DIFLAG_PREALLOC)
503 flags |= XFS_XFLAG_PREALLOC;
504 if (di_flags & XFS_DIFLAG_IMMUTABLE)
505 flags |= XFS_XFLAG_IMMUTABLE;
506 if (di_flags & XFS_DIFLAG_APPEND)
507 flags |= XFS_XFLAG_APPEND;
508 if (di_flags & XFS_DIFLAG_SYNC)
509 flags |= XFS_XFLAG_SYNC;
510 if (di_flags & XFS_DIFLAG_NOATIME)
511 flags |= XFS_XFLAG_NOATIME;
512 if (di_flags & XFS_DIFLAG_NODUMP)
513 flags |= XFS_XFLAG_NODUMP;
514 if (di_flags & XFS_DIFLAG_RTINHERIT)
515 flags |= XFS_XFLAG_RTINHERIT;
516 if (di_flags & XFS_DIFLAG_PROJINHERIT)
517 flags |= XFS_XFLAG_PROJINHERIT;
518 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
519 flags |= XFS_XFLAG_NOSYMLINKS;
520 if (di_flags & XFS_DIFLAG_EXTSIZE)
521 flags |= XFS_XFLAG_EXTSIZE;
522 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
523 flags |= XFS_XFLAG_EXTSZINHERIT;
524 if (di_flags & XFS_DIFLAG_NODEFRAG)
525 flags |= XFS_XFLAG_NODEFRAG;
526 if (di_flags & XFS_DIFLAG_FILESTREAM)
527 flags |= XFS_XFLAG_FILESTREAM;
537 xfs_icdinode_t *dic = &ip->i_d;
539 return _xfs_dic2xflags(dic->di_flags) |
540 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
547 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
548 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
552 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
553 * is allowed, otherwise it has to be an exact match. If a CI match is found,
554 * ci_name->name will point to a the actual name (caller must free) or
555 * will be set to NULL if an exact match is found.
560 struct xfs_name *name,
562 struct xfs_name *ci_name)
568 trace_xfs_lookup(dp, name);
570 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
571 return XFS_ERROR(EIO);
573 lock_mode = xfs_ilock_data_map_shared(dp);
574 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
575 xfs_iunlock(dp, lock_mode);
580 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
588 kmem_free(ci_name->name);
595 * Allocate an inode on disk and return a copy of its in-core version.
596 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
597 * appropriately within the inode. The uid and gid for the inode are
598 * set according to the contents of the given cred structure.
600 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
601 * has a free inode available, call xfs_iget() to obtain the in-core
602 * version of the allocated inode. Finally, fill in the inode and
603 * log its initial contents. In this case, ialloc_context would be
606 * If xfs_dialloc() does not have an available inode, it will replenish
607 * its supply by doing an allocation. Since we can only do one
608 * allocation within a transaction without deadlocks, we must commit
609 * the current transaction before returning the inode itself.
610 * In this case, therefore, we will set ialloc_context and return.
611 * The caller should then commit the current transaction, start a new
612 * transaction, and call xfs_ialloc() again to actually get the inode.
614 * To ensure that some other process does not grab the inode that
615 * was allocated during the first call to xfs_ialloc(), this routine
616 * also returns the [locked] bp pointing to the head of the freelist
617 * as ialloc_context. The caller should hold this buffer across
618 * the commit and pass it back into this routine on the second call.
620 * If we are allocating quota inodes, we do not have a parent inode
621 * to attach to or associate with (i.e. pip == NULL) because they
622 * are not linked into the directory structure - they are attached
623 * directly to the superblock - and so have no parent.
634 xfs_buf_t **ialloc_context,
637 struct xfs_mount *mp = tp->t_mountp;
646 * Call the space management code to pick
647 * the on-disk inode to be allocated.
649 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
650 ialloc_context, &ino);
653 if (*ialloc_context || ino == NULLFSINO) {
657 ASSERT(*ialloc_context == NULL);
660 * Get the in-core inode with the lock held exclusively.
661 * This is because we're setting fields here we need
662 * to prevent others from looking at until we're done.
664 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
665 XFS_ILOCK_EXCL, &ip);
670 ip->i_d.di_mode = mode;
671 ip->i_d.di_onlink = 0;
672 ip->i_d.di_nlink = nlink;
673 ASSERT(ip->i_d.di_nlink == nlink);
674 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
675 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
676 xfs_set_projid(ip, prid);
677 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
680 * If the superblock version is up to where we support new format
681 * inodes and this is currently an old format inode, then change
682 * the inode version number now. This way we only do the conversion
683 * here rather than here and in the flush/logging code.
685 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
686 ip->i_d.di_version == 1) {
687 ip->i_d.di_version = 2;
689 * We've already zeroed the old link count, the projid field,
695 * Project ids won't be stored on disk if we are using a version 1 inode.
697 if ((prid != 0) && (ip->i_d.di_version == 1))
698 xfs_bump_ino_vers2(tp, ip);
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);
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 * we can't set up filestreams until after the VFS inode
762 * is set up properly.
764 if (pip && xfs_inode_is_filestream(pip))
768 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
772 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
773 di_flags |= XFS_DIFLAG_RTINHERIT;
774 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
775 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
776 ip->i_d.di_extsize = pip->i_d.di_extsize;
778 } else if (S_ISREG(mode)) {
779 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
780 di_flags |= XFS_DIFLAG_REALTIME;
781 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
782 di_flags |= XFS_DIFLAG_EXTSIZE;
783 ip->i_d.di_extsize = pip->i_d.di_extsize;
786 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
788 di_flags |= XFS_DIFLAG_NOATIME;
789 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
791 di_flags |= XFS_DIFLAG_NODUMP;
792 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
794 di_flags |= XFS_DIFLAG_SYNC;
795 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
796 xfs_inherit_nosymlinks)
797 di_flags |= XFS_DIFLAG_NOSYMLINKS;
798 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
799 di_flags |= XFS_DIFLAG_PROJINHERIT;
800 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
801 xfs_inherit_nodefrag)
802 di_flags |= XFS_DIFLAG_NODEFRAG;
803 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
804 di_flags |= XFS_DIFLAG_FILESTREAM;
805 ip->i_d.di_flags |= di_flags;
809 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
810 ip->i_df.if_flags = XFS_IFEXTENTS;
811 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
812 ip->i_df.if_u1.if_extents = NULL;
818 * Attribute fork settings for new inode.
820 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
821 ip->i_d.di_anextents = 0;
824 * Log the new values stuffed into the inode.
826 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
827 xfs_trans_log_inode(tp, ip, flags);
829 /* now that we have an i_mode we can setup inode ops and unlock */
832 /* now we have set up the vfs inode we can associate the filestream */
834 error = xfs_filestream_associate(pip, ip);
838 xfs_iflags_set(ip, XFS_IFILESTREAM);
846 * Allocates a new inode from disk and return a pointer to the
847 * incore copy. This routine will internally commit the current
848 * transaction and allocate a new one if the Space Manager needed
849 * to do an allocation to replenish the inode free-list.
851 * This routine is designed to be called from xfs_create and
857 xfs_trans_t **tpp, /* input: current transaction;
858 output: may be a new transaction. */
859 xfs_inode_t *dp, /* directory within whose allocate
864 prid_t prid, /* project id */
865 int okalloc, /* ok to allocate new space */
866 xfs_inode_t **ipp, /* pointer to inode; it will be
874 xfs_buf_t *ialloc_context = NULL;
880 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
883 * xfs_ialloc will return a pointer to an incore inode if
884 * the Space Manager has an available inode on the free
885 * list. Otherwise, it will do an allocation and replenish
886 * the freelist. Since we can only do one allocation per
887 * transaction without deadlocks, we will need to commit the
888 * current transaction and start a new one. We will then
889 * need to call xfs_ialloc again to get the inode.
891 * If xfs_ialloc did an allocation to replenish the freelist,
892 * it returns the bp containing the head of the freelist as
893 * ialloc_context. We will hold a lock on it across the
894 * transaction commit so that no other process can steal
895 * the inode(s) that we've just allocated.
897 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
898 &ialloc_context, &ip);
901 * Return an error if we were unable to allocate a new inode.
902 * This should only happen if we run out of space on disk or
903 * encounter a disk error.
909 if (!ialloc_context && !ip) {
911 return XFS_ERROR(ENOSPC);
915 * If the AGI buffer is non-NULL, then we were unable to get an
916 * inode in one operation. We need to commit the current
917 * transaction and call xfs_ialloc() again. It is guaranteed
918 * to succeed the second time.
920 if (ialloc_context) {
921 struct xfs_trans_res tres;
924 * Normally, xfs_trans_commit releases all the locks.
925 * We call bhold to hang on to the ialloc_context across
926 * the commit. Holding this buffer prevents any other
927 * processes from doing any allocations in this
930 xfs_trans_bhold(tp, ialloc_context);
932 * Save the log reservation so we can use
933 * them in the next transaction.
935 tres.tr_logres = xfs_trans_get_log_res(tp);
936 tres.tr_logcount = xfs_trans_get_log_count(tp);
939 * We want the quota changes to be associated with the next
940 * transaction, NOT this one. So, detach the dqinfo from this
941 * and attach it to the next transaction.
946 dqinfo = (void *)tp->t_dqinfo;
948 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
949 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
952 ntp = xfs_trans_dup(tp);
953 code = xfs_trans_commit(tp, 0);
955 if (committed != NULL) {
959 * If we get an error during the commit processing,
960 * release the buffer that is still held and return
964 xfs_buf_relse(ialloc_context);
966 tp->t_dqinfo = dqinfo;
967 xfs_trans_free_dqinfo(tp);
975 * transaction commit worked ok so we can drop the extra ticket
976 * reference that we gained in xfs_trans_dup()
978 xfs_log_ticket_put(tp->t_ticket);
979 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
980 code = xfs_trans_reserve(tp, &tres, 0, 0);
983 * Re-attach the quota info that we detached from prev trx.
986 tp->t_dqinfo = dqinfo;
987 tp->t_flags |= tflags;
991 xfs_buf_relse(ialloc_context);
996 xfs_trans_bjoin(tp, ialloc_context);
999 * Call ialloc again. Since we've locked out all
1000 * other allocations in this allocation group,
1001 * this call should always succeed.
1003 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1004 okalloc, &ialloc_context, &ip);
1007 * If we get an error at this point, return to the caller
1008 * so that the current transaction can be aborted.
1015 ASSERT(!ialloc_context && ip);
1018 if (committed != NULL)
1029 * Decrement the link count on an inode & log the change.
1030 * If this causes the link count to go to zero, initiate the
1031 * logging activity required to truncate a file.
1040 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1042 ASSERT (ip->i_d.di_nlink > 0);
1044 drop_nlink(VFS_I(ip));
1045 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1048 if (ip->i_d.di_nlink == 0) {
1050 * We're dropping the last link to this file.
1051 * Move the on-disk inode to the AGI unlinked list.
1052 * From xfs_inactive() we will pull the inode from
1053 * the list and free it.
1055 error = xfs_iunlink(tp, ip);
1061 * This gets called when the inode's version needs to be changed from 1 to 2.
1062 * Currently this happens when the nlink field overflows the old 16-bit value
1063 * or when chproj is called to change the project for the first time.
1064 * As a side effect the superblock version will also get rev'd
1065 * to contain the NLINK bit.
1074 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1075 ASSERT(ip->i_d.di_version == 1);
1077 ip->i_d.di_version = 2;
1078 ip->i_d.di_onlink = 0;
1079 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1081 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1082 spin_lock(&mp->m_sb_lock);
1083 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1084 xfs_sb_version_addnlink(&mp->m_sb);
1085 spin_unlock(&mp->m_sb_lock);
1086 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1088 spin_unlock(&mp->m_sb_lock);
1091 /* Caller must log the inode */
1095 * Increment the link count on an inode & log the change.
1102 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1104 ASSERT(ip->i_d.di_nlink > 0);
1106 inc_nlink(VFS_I(ip));
1107 if ((ip->i_d.di_version == 1) &&
1108 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1110 * The inode has increased its number of links beyond
1111 * what can fit in an old format inode. It now needs
1112 * to be converted to a version 2 inode with a 32 bit
1113 * link count. If this is the first inode in the file
1114 * system to do this, then we need to bump the superblock
1115 * version number as well.
1117 xfs_bump_ino_vers2(tp, ip);
1120 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1127 struct xfs_name *name,
1132 int is_dir = S_ISDIR(mode);
1133 struct xfs_mount *mp = dp->i_mount;
1134 struct xfs_inode *ip = NULL;
1135 struct xfs_trans *tp = NULL;
1137 xfs_bmap_free_t free_list;
1138 xfs_fsblock_t first_block;
1139 bool unlock_dp_on_error = false;
1143 struct xfs_dquot *udqp = NULL;
1144 struct xfs_dquot *gdqp = NULL;
1145 struct xfs_dquot *pdqp = NULL;
1146 struct xfs_trans_res tres;
1149 trace_xfs_create(dp, name);
1151 if (XFS_FORCED_SHUTDOWN(mp))
1152 return XFS_ERROR(EIO);
1154 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1155 prid = xfs_get_projid(dp);
1157 prid = XFS_PROJID_DEFAULT;
1160 * Make sure that we have allocated dquot(s) on disk.
1162 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1163 xfs_kgid_to_gid(current_fsgid()), prid,
1164 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1165 &udqp, &gdqp, &pdqp);
1171 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1172 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1173 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1174 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1176 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1177 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1178 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1179 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1182 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1185 * Initially assume that the file does not exist and
1186 * reserve the resources for that case. If that is not
1187 * the case we'll drop the one we have and get a more
1188 * appropriate transaction later.
1190 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1191 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1192 if (error == ENOSPC) {
1193 /* flush outstanding delalloc blocks and retry */
1194 xfs_flush_inodes(mp);
1195 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1197 if (error == ENOSPC) {
1198 /* No space at all so try a "no-allocation" reservation */
1200 error = xfs_trans_reserve(tp, &tres, 0, 0);
1204 goto out_trans_cancel;
1207 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1208 unlock_dp_on_error = true;
1210 xfs_bmap_init(&free_list, &first_block);
1213 * Reserve disk quota and the inode.
1215 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1216 pdqp, resblks, 1, 0);
1218 goto out_trans_cancel;
1220 error = xfs_dir_canenter(tp, dp, name, resblks);
1222 goto out_trans_cancel;
1225 * A newly created regular or special file just has one directory
1226 * entry pointing to them, but a directory also the "." entry
1227 * pointing to itself.
1229 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1230 prid, resblks > 0, &ip, &committed);
1232 if (error == ENOSPC)
1233 goto out_trans_cancel;
1234 goto out_trans_abort;
1238 * Now we join the directory inode to the transaction. We do not do it
1239 * earlier because xfs_dir_ialloc might commit the previous transaction
1240 * (and release all the locks). An error from here on will result in
1241 * the transaction cancel unlocking dp so don't do it explicitly in the
1244 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1245 unlock_dp_on_error = false;
1247 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1248 &first_block, &free_list, resblks ?
1249 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1251 ASSERT(error != ENOSPC);
1252 goto out_trans_abort;
1254 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1255 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1258 error = xfs_dir_init(tp, ip, dp);
1260 goto out_bmap_cancel;
1262 error = xfs_bumplink(tp, dp);
1264 goto out_bmap_cancel;
1268 * If this is a synchronous mount, make sure that the
1269 * create transaction goes to disk before returning to
1272 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1273 xfs_trans_set_sync(tp);
1276 * Attach the dquot(s) to the inodes and modify them incore.
1277 * These ids of the inode couldn't have changed since the new
1278 * inode has been locked ever since it was created.
1280 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1282 error = xfs_bmap_finish(&tp, &free_list, &committed);
1284 goto out_bmap_cancel;
1286 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1288 goto out_release_inode;
1290 xfs_qm_dqrele(udqp);
1291 xfs_qm_dqrele(gdqp);
1292 xfs_qm_dqrele(pdqp);
1298 xfs_bmap_cancel(&free_list);
1300 cancel_flags |= XFS_TRANS_ABORT;
1302 xfs_trans_cancel(tp, cancel_flags);
1305 * Wait until after the current transaction is aborted to
1306 * release the inode. This prevents recursive transactions
1307 * and deadlocks from xfs_inactive.
1312 xfs_qm_dqrele(udqp);
1313 xfs_qm_dqrele(gdqp);
1314 xfs_qm_dqrele(pdqp);
1316 if (unlock_dp_on_error)
1317 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1325 struct xfs_name *target_name)
1327 xfs_mount_t *mp = tdp->i_mount;
1330 xfs_bmap_free_t free_list;
1331 xfs_fsblock_t first_block;
1336 trace_xfs_link(tdp, target_name);
1338 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1340 if (XFS_FORCED_SHUTDOWN(mp))
1341 return XFS_ERROR(EIO);
1343 error = xfs_qm_dqattach(sip, 0);
1347 error = xfs_qm_dqattach(tdp, 0);
1351 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1352 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1353 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1354 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1355 if (error == ENOSPC) {
1357 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1364 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1366 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1367 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1370 * If we are using project inheritance, we only allow hard link
1371 * creation in our tree when the project IDs are the same; else
1372 * the tree quota mechanism could be circumvented.
1374 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1375 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1376 error = XFS_ERROR(EXDEV);
1380 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1384 xfs_bmap_init(&free_list, &first_block);
1386 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1387 &first_block, &free_list, resblks);
1390 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1391 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1393 error = xfs_bumplink(tp, sip);
1398 * If this is a synchronous mount, make sure that the
1399 * link transaction goes to disk before returning to
1402 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1403 xfs_trans_set_sync(tp);
1406 error = xfs_bmap_finish (&tp, &free_list, &committed);
1408 xfs_bmap_cancel(&free_list);
1412 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1415 cancel_flags |= XFS_TRANS_ABORT;
1417 xfs_trans_cancel(tp, cancel_flags);
1423 * Free up the underlying blocks past new_size. The new size must be smaller
1424 * than the current size. This routine can be used both for the attribute and
1425 * data fork, and does not modify the inode size, which is left to the caller.
1427 * The transaction passed to this routine must have made a permanent log
1428 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1429 * given transaction and start new ones, so make sure everything involved in
1430 * the transaction is tidy before calling here. Some transaction will be
1431 * returned to the caller to be committed. The incoming transaction must
1432 * already include the inode, and both inode locks must be held exclusively.
1433 * The inode must also be "held" within the transaction. On return the inode
1434 * will be "held" within the returned transaction. This routine does NOT
1435 * require any disk space to be reserved for it within the transaction.
1437 * If we get an error, we must return with the inode locked and linked into the
1438 * current transaction. This keeps things simple for the higher level code,
1439 * because it always knows that the inode is locked and held in the transaction
1440 * that returns to it whether errors occur or not. We don't mark the inode
1441 * dirty on error so that transactions can be easily aborted if possible.
1444 xfs_itruncate_extents(
1445 struct xfs_trans **tpp,
1446 struct xfs_inode *ip,
1448 xfs_fsize_t new_size)
1450 struct xfs_mount *mp = ip->i_mount;
1451 struct xfs_trans *tp = *tpp;
1452 struct xfs_trans *ntp;
1453 xfs_bmap_free_t free_list;
1454 xfs_fsblock_t first_block;
1455 xfs_fileoff_t first_unmap_block;
1456 xfs_fileoff_t last_block;
1457 xfs_filblks_t unmap_len;
1462 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1463 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1464 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1465 ASSERT(new_size <= XFS_ISIZE(ip));
1466 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1467 ASSERT(ip->i_itemp != NULL);
1468 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1469 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1471 trace_xfs_itruncate_extents_start(ip, new_size);
1474 * Since it is possible for space to become allocated beyond
1475 * the end of the file (in a crash where the space is allocated
1476 * but the inode size is not yet updated), simply remove any
1477 * blocks which show up between the new EOF and the maximum
1478 * possible file size. If the first block to be removed is
1479 * beyond the maximum file size (ie it is the same as last_block),
1480 * then there is nothing to do.
1482 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1483 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1484 if (first_unmap_block == last_block)
1487 ASSERT(first_unmap_block < last_block);
1488 unmap_len = last_block - first_unmap_block + 1;
1490 xfs_bmap_init(&free_list, &first_block);
1491 error = xfs_bunmapi(tp, ip,
1492 first_unmap_block, unmap_len,
1493 xfs_bmapi_aflag(whichfork),
1494 XFS_ITRUNC_MAX_EXTENTS,
1495 &first_block, &free_list,
1498 goto out_bmap_cancel;
1501 * Duplicate the transaction that has the permanent
1502 * reservation and commit the old transaction.
1504 error = xfs_bmap_finish(&tp, &free_list, &committed);
1506 xfs_trans_ijoin(tp, ip, 0);
1508 goto out_bmap_cancel;
1512 * Mark the inode dirty so it will be logged and
1513 * moved forward in the log as part of every commit.
1515 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1518 ntp = xfs_trans_dup(tp);
1519 error = xfs_trans_commit(tp, 0);
1522 xfs_trans_ijoin(tp, ip, 0);
1528 * Transaction commit worked ok so we can drop the extra ticket
1529 * reference that we gained in xfs_trans_dup()
1531 xfs_log_ticket_put(tp->t_ticket);
1532 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1538 * Always re-log the inode so that our permanent transaction can keep
1539 * on rolling it forward in the log.
1541 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1543 trace_xfs_itruncate_extents_end(ip, new_size);
1550 * If the bunmapi call encounters an error, return to the caller where
1551 * the transaction can be properly aborted. We just need to make sure
1552 * we're not holding any resources that we were not when we came in.
1554 xfs_bmap_cancel(&free_list);
1562 xfs_mount_t *mp = ip->i_mount;
1565 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1568 /* If this is a read-only mount, don't do this (would generate I/O) */
1569 if (mp->m_flags & XFS_MOUNT_RDONLY)
1572 if (!XFS_FORCED_SHUTDOWN(mp)) {
1576 * If we are using filestreams, and we have an unlinked
1577 * file that we are processing the last close on, then nothing
1578 * will be able to reopen and write to this file. Purge this
1579 * inode from the filestreams cache so that it doesn't delay
1580 * teardown of the inode.
1582 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1583 xfs_filestream_deassociate(ip);
1586 * If we previously truncated this file and removed old data
1587 * in the process, we want to initiate "early" writeout on
1588 * the last close. This is an attempt to combat the notorious
1589 * NULL files problem which is particularly noticeable from a
1590 * truncate down, buffered (re-)write (delalloc), followed by
1591 * a crash. What we are effectively doing here is
1592 * significantly reducing the time window where we'd otherwise
1593 * be exposed to that problem.
1595 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1597 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1598 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1599 error = -filemap_flush(VFS_I(ip)->i_mapping);
1606 if (ip->i_d.di_nlink == 0)
1609 if (xfs_can_free_eofblocks(ip, false)) {
1612 * If we can't get the iolock just skip truncating the blocks
1613 * past EOF because we could deadlock with the mmap_sem
1614 * otherwise. We'll get another chance to drop them once the
1615 * last reference to the inode is dropped, so we'll never leak
1616 * blocks permanently.
1618 * Further, check if the inode is being opened, written and
1619 * closed frequently and we have delayed allocation blocks
1620 * outstanding (e.g. streaming writes from the NFS server),
1621 * truncating the blocks past EOF will cause fragmentation to
1624 * In this case don't do the truncation, either, but we have to
1625 * be careful how we detect this case. Blocks beyond EOF show
1626 * up as i_delayed_blks even when the inode is clean, so we
1627 * need to truncate them away first before checking for a dirty
1628 * release. Hence on the first dirty close we will still remove
1629 * the speculative allocation, but after that we will leave it
1632 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1635 error = xfs_free_eofblocks(mp, ip, true);
1636 if (error && error != EAGAIN)
1639 /* delalloc blocks after truncation means it really is dirty */
1640 if (ip->i_delayed_blks)
1641 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1647 * xfs_inactive_truncate
1649 * Called to perform a truncate when an inode becomes unlinked.
1652 xfs_inactive_truncate(
1653 struct xfs_inode *ip)
1655 struct xfs_mount *mp = ip->i_mount;
1656 struct xfs_trans *tp;
1659 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1660 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1662 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1663 xfs_trans_cancel(tp, 0);
1667 xfs_ilock(ip, XFS_ILOCK_EXCL);
1668 xfs_trans_ijoin(tp, ip, 0);
1671 * Log the inode size first to prevent stale data exposure in the event
1672 * of a system crash before the truncate completes. See the related
1673 * comment in xfs_setattr_size() for details.
1675 ip->i_d.di_size = 0;
1676 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1678 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1680 goto error_trans_cancel;
1682 ASSERT(ip->i_d.di_nextents == 0);
1684 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1688 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1692 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1694 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1699 * xfs_inactive_ifree()
1701 * Perform the inode free when an inode is unlinked.
1705 struct xfs_inode *ip)
1707 xfs_bmap_free_t free_list;
1708 xfs_fsblock_t first_block;
1710 struct xfs_mount *mp = ip->i_mount;
1711 struct xfs_trans *tp;
1714 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1715 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
1717 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1718 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1722 xfs_ilock(ip, XFS_ILOCK_EXCL);
1723 xfs_trans_ijoin(tp, ip, 0);
1725 xfs_bmap_init(&free_list, &first_block);
1726 error = xfs_ifree(tp, ip, &free_list);
1729 * If we fail to free the inode, shut down. The cancel
1730 * might do that, we need to make sure. Otherwise the
1731 * inode might be lost for a long time or forever.
1733 if (!XFS_FORCED_SHUTDOWN(mp)) {
1734 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1736 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1738 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1739 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1744 * Credit the quota account(s). The inode is gone.
1746 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1749 * Just ignore errors at this point. There is nothing we can
1750 * do except to try to keep going. Make sure it's not a silent
1753 error = xfs_bmap_finish(&tp, &free_list, &committed);
1755 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1757 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1759 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1762 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1769 * This is called when the vnode reference count for the vnode
1770 * goes to zero. If the file has been unlinked, then it must
1771 * now be truncated. Also, we clear all of the read-ahead state
1772 * kept for the inode here since the file is now closed.
1778 struct xfs_mount *mp;
1783 * If the inode is already free, then there can be nothing
1786 if (ip->i_d.di_mode == 0) {
1787 ASSERT(ip->i_df.if_real_bytes == 0);
1788 ASSERT(ip->i_df.if_broot_bytes == 0);
1794 /* If this is a read-only mount, don't do this (would generate I/O) */
1795 if (mp->m_flags & XFS_MOUNT_RDONLY)
1798 if (ip->i_d.di_nlink != 0) {
1800 * force is true because we are evicting an inode from the
1801 * cache. Post-eof blocks must be freed, lest we end up with
1802 * broken free space accounting.
1804 if (xfs_can_free_eofblocks(ip, true))
1805 xfs_free_eofblocks(mp, ip, false);
1810 if (S_ISREG(ip->i_d.di_mode) &&
1811 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1812 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1815 error = xfs_qm_dqattach(ip, 0);
1819 if (S_ISLNK(ip->i_d.di_mode))
1820 error = xfs_inactive_symlink(ip);
1822 error = xfs_inactive_truncate(ip);
1827 * If there are attributes associated with the file then blow them away
1828 * now. The code calls a routine that recursively deconstructs the
1829 * attribute fork. We need to just commit the current transaction
1830 * because we can't use it for xfs_attr_inactive().
1832 if (ip->i_d.di_anextents > 0) {
1833 ASSERT(ip->i_d.di_forkoff != 0);
1835 error = xfs_attr_inactive(ip);
1841 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1843 ASSERT(ip->i_d.di_anextents == 0);
1848 error = xfs_inactive_ifree(ip);
1853 * Release the dquots held by inode, if any.
1855 xfs_qm_dqdetach(ip);
1859 * This is called when the inode's link count goes to 0.
1860 * We place the on-disk inode on a list in the AGI. It
1861 * will be pulled from this list when the inode is freed.
1878 ASSERT(ip->i_d.di_nlink == 0);
1879 ASSERT(ip->i_d.di_mode != 0);
1884 * Get the agi buffer first. It ensures lock ordering
1887 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1890 agi = XFS_BUF_TO_AGI(agibp);
1893 * Get the index into the agi hash table for the
1894 * list this inode will go on.
1896 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1898 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1899 ASSERT(agi->agi_unlinked[bucket_index]);
1900 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1902 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1904 * There is already another inode in the bucket we need
1905 * to add ourselves to. Add us at the front of the list.
1906 * Here we put the head pointer into our next pointer,
1907 * and then we fall through to point the head at us.
1909 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1914 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1915 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1916 offset = ip->i_imap.im_boffset +
1917 offsetof(xfs_dinode_t, di_next_unlinked);
1919 /* need to recalc the inode CRC if appropriate */
1920 xfs_dinode_calc_crc(mp, dip);
1922 xfs_trans_inode_buf(tp, ibp);
1923 xfs_trans_log_buf(tp, ibp, offset,
1924 (offset + sizeof(xfs_agino_t) - 1));
1925 xfs_inobp_check(mp, ibp);
1929 * Point the bucket head pointer at the inode being inserted.
1932 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1933 offset = offsetof(xfs_agi_t, agi_unlinked) +
1934 (sizeof(xfs_agino_t) * bucket_index);
1935 xfs_trans_log_buf(tp, agibp, offset,
1936 (offset + sizeof(xfs_agino_t) - 1));
1941 * Pull the on-disk inode from the AGI unlinked list.
1954 xfs_agnumber_t agno;
1956 xfs_agino_t next_agino;
1957 xfs_buf_t *last_ibp;
1958 xfs_dinode_t *last_dip = NULL;
1960 int offset, last_offset = 0;
1964 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1967 * Get the agi buffer first. It ensures lock ordering
1970 error = xfs_read_agi(mp, tp, agno, &agibp);
1974 agi = XFS_BUF_TO_AGI(agibp);
1977 * Get the index into the agi hash table for the
1978 * list this inode will go on.
1980 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1982 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1983 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1984 ASSERT(agi->agi_unlinked[bucket_index]);
1986 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1988 * We're at the head of the list. Get the inode's on-disk
1989 * buffer to see if there is anyone after us on the list.
1990 * Only modify our next pointer if it is not already NULLAGINO.
1991 * This saves us the overhead of dealing with the buffer when
1992 * there is no need to change it.
1994 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1997 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2001 next_agino = be32_to_cpu(dip->di_next_unlinked);
2002 ASSERT(next_agino != 0);
2003 if (next_agino != NULLAGINO) {
2004 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2005 offset = ip->i_imap.im_boffset +
2006 offsetof(xfs_dinode_t, di_next_unlinked);
2008 /* need to recalc the inode CRC if appropriate */
2009 xfs_dinode_calc_crc(mp, dip);
2011 xfs_trans_inode_buf(tp, ibp);
2012 xfs_trans_log_buf(tp, ibp, offset,
2013 (offset + sizeof(xfs_agino_t) - 1));
2014 xfs_inobp_check(mp, ibp);
2016 xfs_trans_brelse(tp, ibp);
2019 * Point the bucket head pointer at the next inode.
2021 ASSERT(next_agino != 0);
2022 ASSERT(next_agino != agino);
2023 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2024 offset = offsetof(xfs_agi_t, agi_unlinked) +
2025 (sizeof(xfs_agino_t) * bucket_index);
2026 xfs_trans_log_buf(tp, agibp, offset,
2027 (offset + sizeof(xfs_agino_t) - 1));
2030 * We need to search the list for the inode being freed.
2032 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2034 while (next_agino != agino) {
2035 struct xfs_imap imap;
2038 xfs_trans_brelse(tp, last_ibp);
2041 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2043 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2046 "%s: xfs_imap returned error %d.",
2051 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2055 "%s: xfs_imap_to_bp returned error %d.",
2060 last_offset = imap.im_boffset;
2061 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2062 ASSERT(next_agino != NULLAGINO);
2063 ASSERT(next_agino != 0);
2067 * Now last_ibp points to the buffer previous to us on the
2068 * unlinked list. Pull us from the list.
2070 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2073 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2077 next_agino = be32_to_cpu(dip->di_next_unlinked);
2078 ASSERT(next_agino != 0);
2079 ASSERT(next_agino != agino);
2080 if (next_agino != NULLAGINO) {
2081 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2082 offset = ip->i_imap.im_boffset +
2083 offsetof(xfs_dinode_t, di_next_unlinked);
2085 /* need to recalc the inode CRC if appropriate */
2086 xfs_dinode_calc_crc(mp, dip);
2088 xfs_trans_inode_buf(tp, ibp);
2089 xfs_trans_log_buf(tp, ibp, offset,
2090 (offset + sizeof(xfs_agino_t) - 1));
2091 xfs_inobp_check(mp, ibp);
2093 xfs_trans_brelse(tp, ibp);
2096 * Point the previous inode on the list to the next inode.
2098 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2099 ASSERT(next_agino != 0);
2100 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2102 /* need to recalc the inode CRC if appropriate */
2103 xfs_dinode_calc_crc(mp, last_dip);
2105 xfs_trans_inode_buf(tp, last_ibp);
2106 xfs_trans_log_buf(tp, last_ibp, offset,
2107 (offset + sizeof(xfs_agino_t) - 1));
2108 xfs_inobp_check(mp, last_ibp);
2114 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2115 * inodes that are in memory - they all must be marked stale and attached to
2116 * the cluster buffer.
2120 xfs_inode_t *free_ip,
2124 xfs_mount_t *mp = free_ip->i_mount;
2125 int blks_per_cluster;
2132 xfs_inode_log_item_t *iip;
2133 xfs_log_item_t *lip;
2134 struct xfs_perag *pag;
2136 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2137 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2138 blks_per_cluster = 1;
2139 ninodes = mp->m_sb.sb_inopblock;
2140 nbufs = XFS_IALLOC_BLOCKS(mp);
2142 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2143 mp->m_sb.sb_blocksize;
2144 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2145 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2148 for (j = 0; j < nbufs; j++, inum += ninodes) {
2149 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2150 XFS_INO_TO_AGBNO(mp, inum));
2153 * We obtain and lock the backing buffer first in the process
2154 * here, as we have to ensure that any dirty inode that we
2155 * can't get the flush lock on is attached to the buffer.
2156 * If we scan the in-memory inodes first, then buffer IO can
2157 * complete before we get a lock on it, and hence we may fail
2158 * to mark all the active inodes on the buffer stale.
2160 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2161 mp->m_bsize * blks_per_cluster,
2168 * This buffer may not have been correctly initialised as we
2169 * didn't read it from disk. That's not important because we are
2170 * only using to mark the buffer as stale in the log, and to
2171 * attach stale cached inodes on it. That means it will never be
2172 * dispatched for IO. If it is, we want to know about it, and we
2173 * want it to fail. We can acheive this by adding a write
2174 * verifier to the buffer.
2176 bp->b_ops = &xfs_inode_buf_ops;
2179 * Walk the inodes already attached to the buffer and mark them
2180 * stale. These will all have the flush locks held, so an
2181 * in-memory inode walk can't lock them. By marking them all
2182 * stale first, we will not attempt to lock them in the loop
2183 * below as the XFS_ISTALE flag will be set.
2187 if (lip->li_type == XFS_LI_INODE) {
2188 iip = (xfs_inode_log_item_t *)lip;
2189 ASSERT(iip->ili_logged == 1);
2190 lip->li_cb = xfs_istale_done;
2191 xfs_trans_ail_copy_lsn(mp->m_ail,
2192 &iip->ili_flush_lsn,
2193 &iip->ili_item.li_lsn);
2194 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2196 lip = lip->li_bio_list;
2201 * For each inode in memory attempt to add it to the inode
2202 * buffer and set it up for being staled on buffer IO
2203 * completion. This is safe as we've locked out tail pushing
2204 * and flushing by locking the buffer.
2206 * We have already marked every inode that was part of a
2207 * transaction stale above, which means there is no point in
2208 * even trying to lock them.
2210 for (i = 0; i < ninodes; i++) {
2213 ip = radix_tree_lookup(&pag->pag_ici_root,
2214 XFS_INO_TO_AGINO(mp, (inum + i)));
2216 /* Inode not in memory, nothing to do */
2223 * because this is an RCU protected lookup, we could
2224 * find a recently freed or even reallocated inode
2225 * during the lookup. We need to check under the
2226 * i_flags_lock for a valid inode here. Skip it if it
2227 * is not valid, the wrong inode or stale.
2229 spin_lock(&ip->i_flags_lock);
2230 if (ip->i_ino != inum + i ||
2231 __xfs_iflags_test(ip, XFS_ISTALE)) {
2232 spin_unlock(&ip->i_flags_lock);
2236 spin_unlock(&ip->i_flags_lock);
2239 * Don't try to lock/unlock the current inode, but we
2240 * _cannot_ skip the other inodes that we did not find
2241 * in the list attached to the buffer and are not
2242 * already marked stale. If we can't lock it, back off
2245 if (ip != free_ip &&
2246 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2254 xfs_iflags_set(ip, XFS_ISTALE);
2257 * we don't need to attach clean inodes or those only
2258 * with unlogged changes (which we throw away, anyway).
2261 if (!iip || xfs_inode_clean(ip)) {
2262 ASSERT(ip != free_ip);
2264 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2268 iip->ili_last_fields = iip->ili_fields;
2269 iip->ili_fields = 0;
2270 iip->ili_logged = 1;
2271 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2272 &iip->ili_item.li_lsn);
2274 xfs_buf_attach_iodone(bp, xfs_istale_done,
2278 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2281 xfs_trans_stale_inode_buf(tp, bp);
2282 xfs_trans_binval(tp, bp);
2290 * This is called to return an inode to the inode free list.
2291 * The inode should already be truncated to 0 length and have
2292 * no pages associated with it. This routine also assumes that
2293 * the inode is already a part of the transaction.
2295 * The on-disk copy of the inode will have been added to the list
2296 * of unlinked inodes in the AGI. We need to remove the inode from
2297 * that list atomically with respect to freeing it here.
2303 xfs_bmap_free_t *flist)
2307 xfs_ino_t first_ino;
2309 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2310 ASSERT(ip->i_d.di_nlink == 0);
2311 ASSERT(ip->i_d.di_nextents == 0);
2312 ASSERT(ip->i_d.di_anextents == 0);
2313 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2314 ASSERT(ip->i_d.di_nblocks == 0);
2317 * Pull the on-disk inode from the AGI unlinked list.
2319 error = xfs_iunlink_remove(tp, ip);
2323 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2327 ip->i_d.di_mode = 0; /* mark incore inode as free */
2328 ip->i_d.di_flags = 0;
2329 ip->i_d.di_dmevmask = 0;
2330 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2331 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2332 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2334 * Bump the generation count so no one will be confused
2335 * by reincarnations of this inode.
2338 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2341 error = xfs_ifree_cluster(ip, tp, first_ino);
2347 * This is called to unpin an inode. The caller must have the inode locked
2348 * in at least shared mode so that the buffer cannot be subsequently pinned
2349 * once someone is waiting for it to be unpinned.
2353 struct xfs_inode *ip)
2355 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2357 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2359 /* Give the log a push to start the unpinning I/O */
2360 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2366 struct xfs_inode *ip)
2368 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2369 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2374 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2375 if (xfs_ipincount(ip))
2377 } while (xfs_ipincount(ip));
2378 finish_wait(wq, &wait.wait);
2383 struct xfs_inode *ip)
2385 if (xfs_ipincount(ip))
2386 __xfs_iunpin_wait(ip);
2390 * Removing an inode from the namespace involves removing the directory entry
2391 * and dropping the link count on the inode. Removing the directory entry can
2392 * result in locking an AGF (directory blocks were freed) and removing a link
2393 * count can result in placing the inode on an unlinked list which results in
2396 * The big problem here is that we have an ordering constraint on AGF and AGI
2397 * locking - inode allocation locks the AGI, then can allocate a new extent for
2398 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2399 * removes the inode from the unlinked list, requiring that we lock the AGI
2400 * first, and then freeing the inode can result in an inode chunk being freed
2401 * and hence freeing disk space requiring that we lock an AGF.
2403 * Hence the ordering that is imposed by other parts of the code is AGI before
2404 * AGF. This means we cannot remove the directory entry before we drop the inode
2405 * reference count and put it on the unlinked list as this results in a lock
2406 * order of AGF then AGI, and this can deadlock against inode allocation and
2407 * freeing. Therefore we must drop the link counts before we remove the
2410 * This is still safe from a transactional point of view - it is not until we
2411 * get to xfs_bmap_finish() that we have the possibility of multiple
2412 * transactions in this operation. Hence as long as we remove the directory
2413 * entry and drop the link count in the first transaction of the remove
2414 * operation, there are no transactional constraints on the ordering here.
2419 struct xfs_name *name,
2422 xfs_mount_t *mp = dp->i_mount;
2423 xfs_trans_t *tp = NULL;
2424 int is_dir = S_ISDIR(ip->i_d.di_mode);
2426 xfs_bmap_free_t free_list;
2427 xfs_fsblock_t first_block;
2434 trace_xfs_remove(dp, name);
2436 if (XFS_FORCED_SHUTDOWN(mp))
2437 return XFS_ERROR(EIO);
2439 error = xfs_qm_dqattach(dp, 0);
2443 error = xfs_qm_dqattach(ip, 0);
2448 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2449 log_count = XFS_DEFAULT_LOG_COUNT;
2451 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2452 log_count = XFS_REMOVE_LOG_COUNT;
2454 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2457 * We try to get the real space reservation first,
2458 * allowing for directory btree deletion(s) implying
2459 * possible bmap insert(s). If we can't get the space
2460 * reservation then we use 0 instead, and avoid the bmap
2461 * btree insert(s) in the directory code by, if the bmap
2462 * insert tries to happen, instead trimming the LAST
2463 * block from the directory.
2465 resblks = XFS_REMOVE_SPACE_RES(mp);
2466 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2467 if (error == ENOSPC) {
2469 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2472 ASSERT(error != ENOSPC);
2474 goto out_trans_cancel;
2477 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2479 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2480 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2483 * If we're removing a directory perform some additional validation.
2485 cancel_flags |= XFS_TRANS_ABORT;
2487 ASSERT(ip->i_d.di_nlink >= 2);
2488 if (ip->i_d.di_nlink != 2) {
2489 error = XFS_ERROR(ENOTEMPTY);
2490 goto out_trans_cancel;
2492 if (!xfs_dir_isempty(ip)) {
2493 error = XFS_ERROR(ENOTEMPTY);
2494 goto out_trans_cancel;
2497 /* Drop the link from ip's "..". */
2498 error = xfs_droplink(tp, dp);
2500 goto out_trans_cancel;
2502 /* Drop the "." link from ip to self. */
2503 error = xfs_droplink(tp, ip);
2505 goto out_trans_cancel;
2508 * When removing a non-directory we need to log the parent
2509 * inode here. For a directory this is done implicitly
2510 * by the xfs_droplink call for the ".." entry.
2512 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2514 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2516 /* Drop the link from dp to ip. */
2517 error = xfs_droplink(tp, ip);
2519 goto out_trans_cancel;
2521 /* Determine if this is the last link while the inode is locked */
2522 link_zero = (ip->i_d.di_nlink == 0);
2524 xfs_bmap_init(&free_list, &first_block);
2525 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2526 &first_block, &free_list, resblks);
2528 ASSERT(error != ENOENT);
2529 goto out_bmap_cancel;
2533 * If this is a synchronous mount, make sure that the
2534 * remove transaction goes to disk before returning to
2537 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2538 xfs_trans_set_sync(tp);
2540 error = xfs_bmap_finish(&tp, &free_list, &committed);
2542 goto out_bmap_cancel;
2544 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2549 * If we are using filestreams, kill the stream association.
2550 * If the file is still open it may get a new one but that
2551 * will get killed on last close in xfs_close() so we don't
2552 * have to worry about that.
2554 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2555 xfs_filestream_deassociate(ip);
2560 xfs_bmap_cancel(&free_list);
2562 xfs_trans_cancel(tp, cancel_flags);
2568 * Enter all inodes for a rename transaction into a sorted array.
2571 xfs_sort_for_rename(
2572 xfs_inode_t *dp1, /* in: old (source) directory inode */
2573 xfs_inode_t *dp2, /* in: new (target) directory inode */
2574 xfs_inode_t *ip1, /* in: inode of old entry */
2575 xfs_inode_t *ip2, /* in: inode of new entry, if it
2576 already exists, NULL otherwise. */
2577 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2578 int *num_inodes) /* out: number of inodes in array */
2584 * i_tab contains a list of pointers to inodes. We initialize
2585 * the table here & we'll sort it. We will then use it to
2586 * order the acquisition of the inode locks.
2588 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2602 * Sort the elements via bubble sort. (Remember, there are at
2603 * most 4 elements to sort, so this is adequate.)
2605 for (i = 0; i < *num_inodes; i++) {
2606 for (j = 1; j < *num_inodes; j++) {
2607 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2609 i_tab[j] = i_tab[j-1];
2621 xfs_inode_t *src_dp,
2622 struct xfs_name *src_name,
2623 xfs_inode_t *src_ip,
2624 xfs_inode_t *target_dp,
2625 struct xfs_name *target_name,
2626 xfs_inode_t *target_ip)
2628 xfs_trans_t *tp = NULL;
2629 xfs_mount_t *mp = src_dp->i_mount;
2630 int new_parent; /* moving to a new dir */
2631 int src_is_directory; /* src_name is a directory */
2633 xfs_bmap_free_t free_list;
2634 xfs_fsblock_t first_block;
2637 xfs_inode_t *inodes[4];
2641 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2643 new_parent = (src_dp != target_dp);
2644 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2646 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2647 inodes, &num_inodes);
2649 xfs_bmap_init(&free_list, &first_block);
2650 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2651 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2652 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2653 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2654 if (error == ENOSPC) {
2656 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2659 xfs_trans_cancel(tp, 0);
2664 * Attach the dquots to the inodes
2666 error = xfs_qm_vop_rename_dqattach(inodes);
2668 xfs_trans_cancel(tp, cancel_flags);
2673 * Lock all the participating inodes. Depending upon whether
2674 * the target_name exists in the target directory, and
2675 * whether the target directory is the same as the source
2676 * directory, we can lock from 2 to 4 inodes.
2678 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2681 * Join all the inodes to the transaction. From this point on,
2682 * we can rely on either trans_commit or trans_cancel to unlock
2685 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2687 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2688 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2690 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2693 * If we are using project inheritance, we only allow renames
2694 * into our tree when the project IDs are the same; else the
2695 * tree quota mechanism would be circumvented.
2697 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2698 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2699 error = XFS_ERROR(EXDEV);
2704 * Set up the target.
2706 if (target_ip == NULL) {
2708 * If there's no space reservation, check the entry will
2709 * fit before actually inserting it.
2711 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2715 * If target does not exist and the rename crosses
2716 * directories, adjust the target directory link count
2717 * to account for the ".." reference from the new entry.
2719 error = xfs_dir_createname(tp, target_dp, target_name,
2720 src_ip->i_ino, &first_block,
2721 &free_list, spaceres);
2722 if (error == ENOSPC)
2727 xfs_trans_ichgtime(tp, target_dp,
2728 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2730 if (new_parent && src_is_directory) {
2731 error = xfs_bumplink(tp, target_dp);
2735 } else { /* target_ip != NULL */
2737 * If target exists and it's a directory, check that both
2738 * target and source are directories and that target can be
2739 * destroyed, or that neither is a directory.
2741 if (S_ISDIR(target_ip->i_d.di_mode)) {
2743 * Make sure target dir is empty.
2745 if (!(xfs_dir_isempty(target_ip)) ||
2746 (target_ip->i_d.di_nlink > 2)) {
2747 error = XFS_ERROR(EEXIST);
2753 * Link the source inode under the target name.
2754 * If the source inode is a directory and we are moving
2755 * it across directories, its ".." entry will be
2756 * inconsistent until we replace that down below.
2758 * In case there is already an entry with the same
2759 * name at the destination directory, remove it first.
2761 error = xfs_dir_replace(tp, target_dp, target_name,
2763 &first_block, &free_list, spaceres);
2767 xfs_trans_ichgtime(tp, target_dp,
2768 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2771 * Decrement the link count on the target since the target
2772 * dir no longer points to it.
2774 error = xfs_droplink(tp, target_ip);
2778 if (src_is_directory) {
2780 * Drop the link from the old "." entry.
2782 error = xfs_droplink(tp, target_ip);
2786 } /* target_ip != NULL */
2789 * Remove the source.
2791 if (new_parent && src_is_directory) {
2793 * Rewrite the ".." entry to point to the new
2796 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2798 &first_block, &free_list, spaceres);
2799 ASSERT(error != EEXIST);
2805 * We always want to hit the ctime on the source inode.
2807 * This isn't strictly required by the standards since the source
2808 * inode isn't really being changed, but old unix file systems did
2809 * it and some incremental backup programs won't work without it.
2811 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2812 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2815 * Adjust the link count on src_dp. This is necessary when
2816 * renaming a directory, either within one parent when
2817 * the target existed, or across two parent directories.
2819 if (src_is_directory && (new_parent || target_ip != NULL)) {
2822 * Decrement link count on src_directory since the
2823 * entry that's moved no longer points to it.
2825 error = xfs_droplink(tp, src_dp);
2830 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2831 &first_block, &free_list, spaceres);
2835 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2836 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2838 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2841 * If this is a synchronous mount, make sure that the
2842 * rename transaction goes to disk before returning to
2845 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2846 xfs_trans_set_sync(tp);
2849 error = xfs_bmap_finish(&tp, &free_list, &committed);
2851 xfs_bmap_cancel(&free_list);
2852 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2858 * trans_commit will unlock src_ip, target_ip & decrement
2859 * the vnode references.
2861 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2864 cancel_flags |= XFS_TRANS_ABORT;
2866 xfs_bmap_cancel(&free_list);
2867 xfs_trans_cancel(tp, cancel_flags);
2877 xfs_mount_t *mp = ip->i_mount;
2878 struct xfs_perag *pag;
2879 unsigned long first_index, mask;
2880 unsigned long inodes_per_cluster;
2882 xfs_inode_t **ilist;
2889 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2891 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2892 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2893 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2897 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2898 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2900 /* really need a gang lookup range call here */
2901 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2902 first_index, inodes_per_cluster);
2906 for (i = 0; i < nr_found; i++) {
2912 * because this is an RCU protected lookup, we could find a
2913 * recently freed or even reallocated inode during the lookup.
2914 * We need to check under the i_flags_lock for a valid inode
2915 * here. Skip it if it is not valid or the wrong inode.
2917 spin_lock(&ip->i_flags_lock);
2919 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2920 spin_unlock(&ip->i_flags_lock);
2923 spin_unlock(&ip->i_flags_lock);
2926 * Do an un-protected check to see if the inode is dirty and
2927 * is a candidate for flushing. These checks will be repeated
2928 * later after the appropriate locks are acquired.
2930 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2934 * Try to get locks. If any are unavailable or it is pinned,
2935 * then this inode cannot be flushed and is skipped.
2938 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2940 if (!xfs_iflock_nowait(iq)) {
2941 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2944 if (xfs_ipincount(iq)) {
2946 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2951 * arriving here means that this inode can be flushed. First
2952 * re-check that it's dirty before flushing.
2954 if (!xfs_inode_clean(iq)) {
2956 error = xfs_iflush_int(iq, bp);
2958 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2959 goto cluster_corrupt_out;
2965 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2969 XFS_STATS_INC(xs_icluster_flushcnt);
2970 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2981 cluster_corrupt_out:
2983 * Corruption detected in the clustering loop. Invalidate the
2984 * inode buffer and shut down the filesystem.
2988 * Clean up the buffer. If it was delwri, just release it --
2989 * brelse can handle it with no problems. If not, shut down the
2990 * filesystem before releasing the buffer.
2992 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2996 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2998 if (!bufwasdelwri) {
3000 * Just like incore_relse: if we have b_iodone functions,
3001 * mark the buffer as an error and call them. Otherwise
3002 * mark it as stale and brelse.
3007 xfs_buf_ioerror(bp, EIO);
3008 xfs_buf_ioend(bp, 0);
3016 * Unlocks the flush lock
3018 xfs_iflush_abort(iq, false);
3021 return XFS_ERROR(EFSCORRUPTED);
3025 * Flush dirty inode metadata into the backing buffer.
3027 * The caller must have the inode lock and the inode flush lock held. The
3028 * inode lock will still be held upon return to the caller, and the inode
3029 * flush lock will be released after the inode has reached the disk.
3031 * The caller must write out the buffer returned in *bpp and release it.
3035 struct xfs_inode *ip,
3036 struct xfs_buf **bpp)
3038 struct xfs_mount *mp = ip->i_mount;
3040 struct xfs_dinode *dip;
3043 XFS_STATS_INC(xs_iflush_count);
3045 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3046 ASSERT(xfs_isiflocked(ip));
3047 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3048 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3052 xfs_iunpin_wait(ip);
3055 * For stale inodes we cannot rely on the backing buffer remaining
3056 * stale in cache for the remaining life of the stale inode and so
3057 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3058 * inodes below. We have to check this after ensuring the inode is
3059 * unpinned so that it is safe to reclaim the stale inode after the
3062 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3068 * This may have been unpinned because the filesystem is shutting
3069 * down forcibly. If that's the case we must not write this inode
3070 * to disk, because the log record didn't make it to disk.
3072 * We also have to remove the log item from the AIL in this case,
3073 * as we wait for an empty AIL as part of the unmount process.
3075 if (XFS_FORCED_SHUTDOWN(mp)) {
3076 error = XFS_ERROR(EIO);
3081 * Get the buffer containing the on-disk inode.
3083 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3091 * First flush out the inode that xfs_iflush was called with.
3093 error = xfs_iflush_int(ip, bp);
3098 * If the buffer is pinned then push on the log now so we won't
3099 * get stuck waiting in the write for too long.
3101 if (xfs_buf_ispinned(bp))
3102 xfs_log_force(mp, 0);
3106 * see if other inodes can be gathered into this write
3108 error = xfs_iflush_cluster(ip, bp);
3110 goto cluster_corrupt_out;
3117 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3118 cluster_corrupt_out:
3119 error = XFS_ERROR(EFSCORRUPTED);
3122 * Unlocks the flush lock
3124 xfs_iflush_abort(ip, false);
3130 struct xfs_inode *ip,
3133 struct xfs_inode_log_item *iip = ip->i_itemp;
3134 struct xfs_dinode *dip;
3135 struct xfs_mount *mp = ip->i_mount;
3137 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3138 ASSERT(xfs_isiflocked(ip));
3139 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3140 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3141 ASSERT(iip != NULL && iip->ili_fields != 0);
3143 /* set *dip = inode's place in the buffer */
3144 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3146 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3147 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3148 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3149 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3150 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3153 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3154 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3155 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3156 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3157 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3160 if (S_ISREG(ip->i_d.di_mode)) {
3162 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3163 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3164 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3165 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3166 "%s: Bad regular inode %Lu, ptr 0x%p",
3167 __func__, ip->i_ino, ip);
3170 } else if (S_ISDIR(ip->i_d.di_mode)) {
3172 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3173 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3174 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3175 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3176 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3177 "%s: Bad directory inode %Lu, ptr 0x%p",
3178 __func__, ip->i_ino, ip);
3182 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3183 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3184 XFS_RANDOM_IFLUSH_5)) {
3185 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3186 "%s: detected corrupt incore inode %Lu, "
3187 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3188 __func__, ip->i_ino,
3189 ip->i_d.di_nextents + ip->i_d.di_anextents,
3190 ip->i_d.di_nblocks, ip);
3193 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3194 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3195 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3196 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3197 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3202 * Inode item log recovery for v1/v2 inodes are dependent on the
3203 * di_flushiter count for correct sequencing. We bump the flush
3204 * iteration count so we can detect flushes which postdate a log record
3205 * during recovery. This is redundant as we now log every change and
3206 * hence this can't happen but we need to still do it to ensure
3207 * backwards compatibility with old kernels that predate logging all
3210 if (ip->i_d.di_version < 3)
3211 ip->i_d.di_flushiter++;
3214 * Copy the dirty parts of the inode into the on-disk
3215 * inode. We always copy out the core of the inode,
3216 * because if the inode is dirty at all the core must
3219 xfs_dinode_to_disk(dip, &ip->i_d);
3221 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3222 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3223 ip->i_d.di_flushiter = 0;
3226 * If this is really an old format inode and the superblock version
3227 * has not been updated to support only new format inodes, then
3228 * convert back to the old inode format. If the superblock version
3229 * has been updated, then make the conversion permanent.
3231 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3232 if (ip->i_d.di_version == 1) {
3233 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3237 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3238 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3241 * The superblock version has already been bumped,
3242 * so just make the conversion to the new inode
3245 ip->i_d.di_version = 2;
3246 dip->di_version = 2;
3247 ip->i_d.di_onlink = 0;
3249 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3250 memset(&(dip->di_pad[0]), 0,
3251 sizeof(dip->di_pad));
3252 ASSERT(xfs_get_projid(ip) == 0);
3256 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3257 if (XFS_IFORK_Q(ip))
3258 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3259 xfs_inobp_check(mp, bp);
3262 * We've recorded everything logged in the inode, so we'd like to clear
3263 * the ili_fields bits so we don't log and flush things unnecessarily.
3264 * However, we can't stop logging all this information until the data
3265 * we've copied into the disk buffer is written to disk. If we did we
3266 * might overwrite the copy of the inode in the log with all the data
3267 * after re-logging only part of it, and in the face of a crash we
3268 * wouldn't have all the data we need to recover.
3270 * What we do is move the bits to the ili_last_fields field. When
3271 * logging the inode, these bits are moved back to the ili_fields field.
3272 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3273 * know that the information those bits represent is permanently on
3274 * disk. As long as the flush completes before the inode is logged
3275 * again, then both ili_fields and ili_last_fields will be cleared.
3277 * We can play with the ili_fields bits here, because the inode lock
3278 * must be held exclusively in order to set bits there and the flush
3279 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3280 * done routine can tell whether or not to look in the AIL. Also, store
3281 * the current LSN of the inode so that we can tell whether the item has
3282 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3283 * need the AIL lock, because it is a 64 bit value that cannot be read
3286 iip->ili_last_fields = iip->ili_fields;
3287 iip->ili_fields = 0;
3288 iip->ili_logged = 1;
3290 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3291 &iip->ili_item.li_lsn);
3294 * Attach the function xfs_iflush_done to the inode's
3295 * buffer. This will remove the inode from the AIL
3296 * and unlock the inode's flush lock when the inode is
3297 * completely written to disk.
3299 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3301 /* update the lsn in the on disk inode if required */
3302 if (ip->i_d.di_version == 3)
3303 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3305 /* generate the checksum. */
3306 xfs_dinode_calc_crc(mp, dip);
3308 ASSERT(bp->b_fspriv != NULL);
3309 ASSERT(bp->b_iodone != NULL);
3313 return XFS_ERROR(EFSCORRUPTED);