2 * Copyright (c) 2000-2005 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
24 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
36 #include "xfs_itable.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_utils.h"
41 #include "xfs_vnodeops.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_trace.h"
45 #include <linux/capability.h>
46 #include <linux/xattr.h>
47 #include <linux/namei.h>
48 #include <linux/posix_acl.h>
49 #include <linux/security.h>
50 #include <linux/fiemap.h>
51 #include <linux/slab.h>
56 const struct xattr
*xattr_array
,
59 const struct xattr
*xattr
;
60 struct xfs_inode
*ip
= XFS_I(inode
);
63 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
64 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
65 xattr
->value_len
, ATTR_SECURE
);
73 * Hook in SELinux. This is not quite correct yet, what we really need
74 * here (as we do for default ACLs) is a mechanism by which creation of
75 * these attrs can be journalled at inode creation time (along with the
76 * inode, of course, such that log replay can't cause these to be lost).
83 const struct qstr
*qstr
)
85 return security_inode_init_security(inode
, dir
, qstr
,
86 &xfs_initxattrs
, NULL
);
91 struct xfs_name
*namep
,
92 struct dentry
*dentry
)
94 namep
->name
= dentry
->d_name
.name
;
95 namep
->len
= dentry
->d_name
.len
;
102 struct dentry
*dentry
)
104 struct xfs_name teardown
;
107 * If we can't add the ACL or we fail in
108 * xfs_init_security we must back out.
109 * ENOSPC can hit here, among other things.
111 xfs_dentry_to_name(&teardown
, dentry
);
113 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
120 struct dentry
*dentry
,
125 struct xfs_inode
*ip
= NULL
;
126 struct posix_acl
*default_acl
= NULL
;
127 struct xfs_name name
;
131 * Irix uses Missed'em'V split, but doesn't want to see
132 * the upper 5 bits of (14bit) major.
134 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
135 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
137 rdev
= sysv_encode_dev(rdev
);
142 if (IS_POSIXACL(dir
)) {
143 default_acl
= xfs_get_acl(dir
, ACL_TYPE_DEFAULT
);
144 if (IS_ERR(default_acl
))
145 return PTR_ERR(default_acl
);
148 mode
&= ~current_umask();
151 xfs_dentry_to_name(&name
, dentry
);
152 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
158 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
160 goto out_cleanup_inode
;
163 error
= -xfs_inherit_acl(inode
, default_acl
);
166 goto out_cleanup_inode
;
170 d_instantiate(dentry
, inode
);
174 xfs_cleanup_inode(dir
, inode
, dentry
);
176 posix_acl_release(default_acl
);
183 struct dentry
*dentry
,
185 struct nameidata
*nd
)
187 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
193 struct dentry
*dentry
,
196 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
199 STATIC
struct dentry
*
202 struct dentry
*dentry
,
203 struct nameidata
*nd
)
205 struct xfs_inode
*cip
;
206 struct xfs_name name
;
209 if (dentry
->d_name
.len
>= MAXNAMELEN
)
210 return ERR_PTR(-ENAMETOOLONG
);
212 xfs_dentry_to_name(&name
, dentry
);
213 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
214 if (unlikely(error
)) {
215 if (unlikely(error
!= ENOENT
))
216 return ERR_PTR(-error
);
221 return d_splice_alias(VFS_I(cip
), dentry
);
224 STATIC
struct dentry
*
227 struct dentry
*dentry
,
228 struct nameidata
*nd
)
230 struct xfs_inode
*ip
;
231 struct xfs_name xname
;
232 struct xfs_name ci_name
;
236 if (dentry
->d_name
.len
>= MAXNAMELEN
)
237 return ERR_PTR(-ENAMETOOLONG
);
239 xfs_dentry_to_name(&xname
, dentry
);
240 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
241 if (unlikely(error
)) {
242 if (unlikely(error
!= ENOENT
))
243 return ERR_PTR(-error
);
245 * call d_add(dentry, NULL) here when d_drop_negative_children
246 * is called in xfs_vn_mknod (ie. allow negative dentries
247 * with CI filesystems).
252 /* if exact match, just splice and exit */
254 return d_splice_alias(VFS_I(ip
), dentry
);
256 /* else case-insensitive match... */
257 dname
.name
= ci_name
.name
;
258 dname
.len
= ci_name
.len
;
259 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
260 kmem_free(ci_name
.name
);
266 struct dentry
*old_dentry
,
268 struct dentry
*dentry
)
270 struct inode
*inode
= old_dentry
->d_inode
;
271 struct xfs_name name
;
274 xfs_dentry_to_name(&name
, dentry
);
276 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
281 d_instantiate(dentry
, inode
);
288 struct dentry
*dentry
)
290 struct xfs_name name
;
293 xfs_dentry_to_name(&name
, dentry
);
295 error
= -xfs_remove(XFS_I(dir
), &name
, XFS_I(dentry
->d_inode
));
300 * With unlink, the VFS makes the dentry "negative": no inode,
301 * but still hashed. This is incompatible with case-insensitive
302 * mode, so invalidate (unhash) the dentry in CI-mode.
304 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
305 d_invalidate(dentry
);
312 struct dentry
*dentry
,
316 struct xfs_inode
*cip
= NULL
;
317 struct xfs_name name
;
322 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
323 xfs_dentry_to_name(&name
, dentry
);
325 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
331 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
333 goto out_cleanup_inode
;
335 d_instantiate(dentry
, inode
);
339 xfs_cleanup_inode(dir
, inode
, dentry
);
347 struct dentry
*odentry
,
349 struct dentry
*ndentry
)
351 struct inode
*new_inode
= ndentry
->d_inode
;
352 struct xfs_name oname
;
353 struct xfs_name nname
;
355 xfs_dentry_to_name(&oname
, odentry
);
356 xfs_dentry_to_name(&nname
, ndentry
);
358 return -xfs_rename(XFS_I(odir
), &oname
, XFS_I(odentry
->d_inode
),
359 XFS_I(ndir
), &nname
, new_inode
?
360 XFS_I(new_inode
) : NULL
);
364 * careful here - this function can get called recursively, so
365 * we need to be very careful about how much stack we use.
366 * uio is kmalloced for this reason...
370 struct dentry
*dentry
,
371 struct nameidata
*nd
)
376 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
380 error
= -xfs_readlink(XFS_I(dentry
->d_inode
), link
);
384 nd_set_link(nd
, link
);
390 nd_set_link(nd
, ERR_PTR(error
));
396 struct dentry
*dentry
,
397 struct nameidata
*nd
,
400 char *s
= nd_get_link(nd
);
408 struct vfsmount
*mnt
,
409 struct dentry
*dentry
,
412 struct inode
*inode
= dentry
->d_inode
;
413 struct xfs_inode
*ip
= XFS_I(inode
);
414 struct xfs_mount
*mp
= ip
->i_mount
;
416 trace_xfs_getattr(ip
);
418 if (XFS_FORCED_SHUTDOWN(mp
))
419 return -XFS_ERROR(EIO
);
421 stat
->size
= XFS_ISIZE(ip
);
422 stat
->dev
= inode
->i_sb
->s_dev
;
423 stat
->mode
= ip
->i_d
.di_mode
;
424 stat
->nlink
= ip
->i_d
.di_nlink
;
425 stat
->uid
= ip
->i_d
.di_uid
;
426 stat
->gid
= ip
->i_d
.di_gid
;
427 stat
->ino
= ip
->i_ino
;
428 stat
->atime
= inode
->i_atime
;
429 stat
->mtime
= inode
->i_mtime
;
430 stat
->ctime
= inode
->i_ctime
;
432 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
435 switch (inode
->i_mode
& S_IFMT
) {
438 stat
->blksize
= BLKDEV_IOSIZE
;
439 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
440 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
443 if (XFS_IS_REALTIME_INODE(ip
)) {
445 * If the file blocks are being allocated from a
446 * realtime volume, then return the inode's realtime
447 * extent size or the realtime volume's extent size.
450 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
452 stat
->blksize
= xfs_preferred_iosize(mp
);
462 struct xfs_inode
*ip
,
466 xfs_mount_t
*mp
= ip
->i_mount
;
467 struct inode
*inode
= VFS_I(ip
);
468 int mask
= iattr
->ia_valid
;
471 uid_t uid
= 0, iuid
= 0;
472 gid_t gid
= 0, igid
= 0;
473 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
474 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
476 trace_xfs_setattr(ip
);
478 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
479 return XFS_ERROR(EROFS
);
481 if (XFS_FORCED_SHUTDOWN(mp
))
482 return XFS_ERROR(EIO
);
484 error
= -inode_change_ok(inode
, iattr
);
486 return XFS_ERROR(error
);
488 ASSERT((mask
& ATTR_SIZE
) == 0);
491 * If disk quotas is on, we make sure that the dquots do exist on disk,
492 * before we start any other transactions. Trying to do this later
493 * is messy. We don't care to take a readlock to look at the ids
494 * in inode here, because we can't hold it across the trans_reserve.
495 * If the IDs do change before we take the ilock, we're covered
496 * because the i_*dquot fields will get updated anyway.
498 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
501 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
503 qflags
|= XFS_QMOPT_UQUOTA
;
505 uid
= ip
->i_d
.di_uid
;
507 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
509 qflags
|= XFS_QMOPT_GQUOTA
;
511 gid
= ip
->i_d
.di_gid
;
515 * We take a reference when we initialize udqp and gdqp,
516 * so it is important that we never blindly double trip on
517 * the same variable. See xfs_create() for an example.
519 ASSERT(udqp
== NULL
);
520 ASSERT(gdqp
== NULL
);
521 error
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, xfs_get_projid(ip
),
522 qflags
, &udqp
, &gdqp
);
527 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
528 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES(mp
), 0, 0, 0);
532 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
535 * Change file ownership. Must be the owner or privileged.
537 if (mask
& (ATTR_UID
|ATTR_GID
)) {
539 * These IDs could have changed since we last looked at them.
540 * But, we're assured that if the ownership did change
541 * while we didn't have the inode locked, inode's dquot(s)
542 * would have changed also.
544 iuid
= ip
->i_d
.di_uid
;
545 igid
= ip
->i_d
.di_gid
;
546 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
547 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
550 * Do a quota reservation only if uid/gid is actually
553 if (XFS_IS_QUOTA_RUNNING(mp
) &&
554 ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
555 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
))) {
557 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
558 capable(CAP_FOWNER
) ?
559 XFS_QMOPT_FORCE_RES
: 0);
560 if (error
) /* out of quota */
561 goto out_trans_cancel
;
565 xfs_trans_ijoin(tp
, ip
, 0);
568 * Change file ownership. Must be the owner or privileged.
570 if (mask
& (ATTR_UID
|ATTR_GID
)) {
572 * CAP_FSETID overrides the following restrictions:
574 * The set-user-ID and set-group-ID bits of a file will be
575 * cleared upon successful return from chown()
577 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
578 !capable(CAP_FSETID
))
579 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
582 * Change the ownerships and register quota modifications
583 * in the transaction.
586 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
587 ASSERT(mask
& ATTR_UID
);
589 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
590 &ip
->i_udquot
, udqp
);
592 ip
->i_d
.di_uid
= uid
;
596 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
597 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
598 ASSERT(mask
& ATTR_GID
);
600 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
601 &ip
->i_gdquot
, gdqp
);
603 ip
->i_d
.di_gid
= gid
;
609 * Change file access modes.
611 if (mask
& ATTR_MODE
) {
612 umode_t mode
= iattr
->ia_mode
;
614 if (!in_group_p(inode
->i_gid
) && !capable(CAP_FSETID
))
617 ip
->i_d
.di_mode
&= S_IFMT
;
618 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
620 inode
->i_mode
&= S_IFMT
;
621 inode
->i_mode
|= mode
& ~S_IFMT
;
625 * Change file access or modified times.
627 if (mask
& ATTR_ATIME
) {
628 inode
->i_atime
= iattr
->ia_atime
;
629 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
630 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
632 if (mask
& ATTR_CTIME
) {
633 inode
->i_ctime
= iattr
->ia_ctime
;
634 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
635 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
637 if (mask
& ATTR_MTIME
) {
638 inode
->i_mtime
= iattr
->ia_mtime
;
639 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
640 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
643 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
645 XFS_STATS_INC(xs_ig_attrchg
);
647 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
648 xfs_trans_set_sync(tp
);
649 error
= xfs_trans_commit(tp
, 0);
651 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
654 * Release any dquot(s) the inode had kept before chown.
656 xfs_qm_dqrele(olddquot1
);
657 xfs_qm_dqrele(olddquot2
);
662 return XFS_ERROR(error
);
665 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
666 * update. We could avoid this with linked transactions
667 * and passing down the transaction pointer all the way
668 * to attr_set. No previous user of the generic
669 * Posix ACL code seems to care about this issue either.
671 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
672 error
= -xfs_acl_chmod(inode
);
674 return XFS_ERROR(error
);
680 xfs_trans_cancel(tp
, 0);
681 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
689 * Truncate file. Must have write permission and not be a directory.
693 struct xfs_inode
*ip
,
697 struct xfs_mount
*mp
= ip
->i_mount
;
698 struct inode
*inode
= VFS_I(ip
);
699 int mask
= iattr
->ia_valid
;
700 xfs_off_t oldsize
, newsize
;
701 struct xfs_trans
*tp
;
704 uint commit_flags
= 0;
706 trace_xfs_setattr(ip
);
708 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
709 return XFS_ERROR(EROFS
);
711 if (XFS_FORCED_SHUTDOWN(mp
))
712 return XFS_ERROR(EIO
);
714 error
= -inode_change_ok(inode
, iattr
);
716 return XFS_ERROR(error
);
718 ASSERT(S_ISREG(ip
->i_d
.di_mode
));
719 ASSERT((mask
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
720 ATTR_MTIME_SET
|ATTR_KILL_SUID
|ATTR_KILL_SGID
|
721 ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
723 lock_flags
= XFS_ILOCK_EXCL
;
724 if (!(flags
& XFS_ATTR_NOLOCK
))
725 lock_flags
|= XFS_IOLOCK_EXCL
;
726 xfs_ilock(ip
, lock_flags
);
728 oldsize
= inode
->i_size
;
729 newsize
= iattr
->ia_size
;
732 * Short circuit the truncate case for zero length files.
734 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
735 if (!(mask
& (ATTR_CTIME
|ATTR_MTIME
)))
739 * Use the regular setattr path to update the timestamps.
741 xfs_iunlock(ip
, lock_flags
);
742 iattr
->ia_valid
&= ~ATTR_SIZE
;
743 return xfs_setattr_nonsize(ip
, iattr
, 0);
747 * Make sure that the dquots are attached to the inode.
749 error
= xfs_qm_dqattach_locked(ip
, 0);
754 * Now we can make the changes. Before we join the inode to the
755 * transaction, take care of the part of the truncation that must be
756 * done without the inode lock. This needs to be done before joining
757 * the inode to the transaction, because the inode cannot be unlocked
758 * once it is a part of the transaction.
760 if (newsize
> oldsize
) {
762 * Do the first part of growing a file: zero any data in the
763 * last block that is beyond the old EOF. We need to do this
764 * before the inode is joined to the transaction to modify
767 error
= xfs_zero_eof(ip
, newsize
, oldsize
);
771 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
772 lock_flags
&= ~XFS_ILOCK_EXCL
;
775 * We are going to log the inode size change in this transaction so
776 * any previous writes that are beyond the on disk EOF and the new
777 * EOF that have not been written out need to be written here. If we
778 * do not write the data out, we expose ourselves to the null files
781 * Only flush from the on disk size to the smaller of the in memory
782 * file size or the new size as that's the range we really care about
783 * here and prevents waiting for other data not within the range we
786 if (oldsize
!= ip
->i_d
.di_size
&& newsize
> ip
->i_d
.di_size
) {
787 error
= xfs_flush_pages(ip
, ip
->i_d
.di_size
, newsize
, 0,
794 * Wait for all direct I/O to complete.
796 inode_dio_wait(inode
);
798 error
= -block_truncate_page(inode
->i_mapping
, newsize
, xfs_get_blocks
);
802 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
803 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
804 XFS_TRANS_PERM_LOG_RES
,
805 XFS_ITRUNCATE_LOG_COUNT
);
807 goto out_trans_cancel
;
809 truncate_setsize(inode
, newsize
);
811 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
812 lock_flags
|= XFS_ILOCK_EXCL
;
814 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
816 xfs_trans_ijoin(tp
, ip
, 0);
819 * Only change the c/mtime if we are changing the size or we are
820 * explicitly asked to change it. This handles the semantic difference
821 * between truncate() and ftruncate() as implemented in the VFS.
823 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
824 * special case where we need to update the times despite not having
825 * these flags set. For all other operations the VFS set these flags
826 * explicitly if it wants a timestamp update.
828 if (newsize
!= oldsize
&& (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))) {
829 iattr
->ia_ctime
= iattr
->ia_mtime
=
830 current_fs_time(inode
->i_sb
);
831 mask
|= ATTR_CTIME
| ATTR_MTIME
;
835 * The first thing we do is set the size to new_size permanently on
836 * disk. This way we don't have to worry about anyone ever being able
837 * to look at the data being freed even in the face of a crash.
838 * What we're getting around here is the case where we free a block, it
839 * is allocated to another file, it is written to, and then we crash.
840 * If the new data gets written to the file but the log buffers
841 * containing the free and reallocation don't, then we'd end up with
842 * garbage in the blocks being freed. As long as we make the new size
843 * permanent before actually freeing any blocks it doesn't matter if
844 * they get written to.
846 ip
->i_d
.di_size
= newsize
;
847 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
849 if (newsize
<= oldsize
) {
850 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
852 goto out_trans_abort
;
855 * Truncated "down", so we're removing references to old data
856 * here - if we delay flushing for a long time, we expose
857 * ourselves unduly to the notorious NULL files problem. So,
858 * we mark this inode and flush it when the file is closed,
859 * and do not wait the usual (long) time for writeout.
861 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
864 if (mask
& ATTR_CTIME
) {
865 inode
->i_ctime
= iattr
->ia_ctime
;
866 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
867 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
869 if (mask
& ATTR_MTIME
) {
870 inode
->i_mtime
= iattr
->ia_mtime
;
871 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
872 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
875 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
877 XFS_STATS_INC(xs_ig_attrchg
);
879 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
880 xfs_trans_set_sync(tp
);
882 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
885 xfs_iunlock(ip
, lock_flags
);
889 commit_flags
|= XFS_TRANS_ABORT
;
891 xfs_trans_cancel(tp
, commit_flags
);
897 struct dentry
*dentry
,
900 if (iattr
->ia_valid
& ATTR_SIZE
)
901 return -xfs_setattr_size(XFS_I(dentry
->d_inode
), iattr
, 0);
902 return -xfs_setattr_nonsize(XFS_I(dentry
->d_inode
), iattr
, 0);
905 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
908 * Call fiemap helper to fill in user data.
909 * Returns positive errors to xfs_getbmap.
914 struct getbmapx
*bmv
,
918 struct fiemap_extent_info
*fieinfo
= *arg
;
919 u32 fiemap_flags
= 0;
920 u64 logical
, physical
, length
;
922 /* Do nothing for a hole */
923 if (bmv
->bmv_block
== -1LL)
926 logical
= BBTOB(bmv
->bmv_offset
);
927 physical
= BBTOB(bmv
->bmv_block
);
928 length
= BBTOB(bmv
->bmv_length
);
930 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
931 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
932 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
933 fiemap_flags
|= FIEMAP_EXTENT_DELALLOC
;
934 physical
= 0; /* no block yet */
936 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
937 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
939 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
940 length
, fiemap_flags
);
943 *full
= 1; /* user array now full */
952 struct fiemap_extent_info
*fieinfo
,
956 xfs_inode_t
*ip
= XFS_I(inode
);
960 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
964 /* Set up bmap header for xfs internal routine */
965 bm
.bmv_offset
= BTOBB(start
);
966 /* Special case for whole file */
967 if (length
== FIEMAP_MAX_OFFSET
)
968 bm
.bmv_length
= -1LL;
970 bm
.bmv_length
= BTOBB(length
);
972 /* We add one because in getbmap world count includes the header */
973 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
974 fieinfo
->fi_extents_max
+ 1;
975 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
976 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
977 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
978 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
979 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
980 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
981 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
983 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
990 static const struct inode_operations xfs_inode_operations
= {
991 .get_acl
= xfs_get_acl
,
992 .getattr
= xfs_vn_getattr
,
993 .setattr
= xfs_vn_setattr
,
994 .setxattr
= generic_setxattr
,
995 .getxattr
= generic_getxattr
,
996 .removexattr
= generic_removexattr
,
997 .listxattr
= xfs_vn_listxattr
,
998 .fiemap
= xfs_vn_fiemap
,
1001 static const struct inode_operations xfs_dir_inode_operations
= {
1002 .create
= xfs_vn_create
,
1003 .lookup
= xfs_vn_lookup
,
1004 .link
= xfs_vn_link
,
1005 .unlink
= xfs_vn_unlink
,
1006 .symlink
= xfs_vn_symlink
,
1007 .mkdir
= xfs_vn_mkdir
,
1009 * Yes, XFS uses the same method for rmdir and unlink.
1011 * There are some subtile differences deeper in the code,
1012 * but we use S_ISDIR to check for those.
1014 .rmdir
= xfs_vn_unlink
,
1015 .mknod
= xfs_vn_mknod
,
1016 .rename
= xfs_vn_rename
,
1017 .get_acl
= xfs_get_acl
,
1018 .getattr
= xfs_vn_getattr
,
1019 .setattr
= xfs_vn_setattr
,
1020 .setxattr
= generic_setxattr
,
1021 .getxattr
= generic_getxattr
,
1022 .removexattr
= generic_removexattr
,
1023 .listxattr
= xfs_vn_listxattr
,
1026 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1027 .create
= xfs_vn_create
,
1028 .lookup
= xfs_vn_ci_lookup
,
1029 .link
= xfs_vn_link
,
1030 .unlink
= xfs_vn_unlink
,
1031 .symlink
= xfs_vn_symlink
,
1032 .mkdir
= xfs_vn_mkdir
,
1034 * Yes, XFS uses the same method for rmdir and unlink.
1036 * There are some subtile differences deeper in the code,
1037 * but we use S_ISDIR to check for those.
1039 .rmdir
= xfs_vn_unlink
,
1040 .mknod
= xfs_vn_mknod
,
1041 .rename
= xfs_vn_rename
,
1042 .get_acl
= xfs_get_acl
,
1043 .getattr
= xfs_vn_getattr
,
1044 .setattr
= xfs_vn_setattr
,
1045 .setxattr
= generic_setxattr
,
1046 .getxattr
= generic_getxattr
,
1047 .removexattr
= generic_removexattr
,
1048 .listxattr
= xfs_vn_listxattr
,
1051 static const struct inode_operations xfs_symlink_inode_operations
= {
1052 .readlink
= generic_readlink
,
1053 .follow_link
= xfs_vn_follow_link
,
1054 .put_link
= xfs_vn_put_link
,
1055 .get_acl
= xfs_get_acl
,
1056 .getattr
= xfs_vn_getattr
,
1057 .setattr
= xfs_vn_setattr
,
1058 .setxattr
= generic_setxattr
,
1059 .getxattr
= generic_getxattr
,
1060 .removexattr
= generic_removexattr
,
1061 .listxattr
= xfs_vn_listxattr
,
1065 xfs_diflags_to_iflags(
1066 struct inode
*inode
,
1067 struct xfs_inode
*ip
)
1069 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
1070 inode
->i_flags
|= S_IMMUTABLE
;
1072 inode
->i_flags
&= ~S_IMMUTABLE
;
1073 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
1074 inode
->i_flags
|= S_APPEND
;
1076 inode
->i_flags
&= ~S_APPEND
;
1077 if (ip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
)
1078 inode
->i_flags
|= S_SYNC
;
1080 inode
->i_flags
&= ~S_SYNC
;
1081 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
)
1082 inode
->i_flags
|= S_NOATIME
;
1084 inode
->i_flags
&= ~S_NOATIME
;
1088 * Initialize the Linux inode, set up the operation vectors and
1091 * When reading existing inodes from disk this is called directly
1092 * from xfs_iget, when creating a new inode it is called from
1093 * xfs_ialloc after setting up the inode.
1095 * We are always called with an uninitialised linux inode here.
1096 * We need to initialise the necessary fields and take a reference
1101 struct xfs_inode
*ip
)
1103 struct inode
*inode
= &ip
->i_vnode
;
1105 inode
->i_ino
= ip
->i_ino
;
1106 inode
->i_state
= I_NEW
;
1108 inode_sb_list_add(inode
);
1109 /* make the inode look hashed for the writeback code */
1110 hlist_add_fake(&inode
->i_hash
);
1112 inode
->i_mode
= ip
->i_d
.di_mode
;
1113 set_nlink(inode
, ip
->i_d
.di_nlink
);
1114 inode
->i_uid
= ip
->i_d
.di_uid
;
1115 inode
->i_gid
= ip
->i_d
.di_gid
;
1117 switch (inode
->i_mode
& S_IFMT
) {
1121 MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
1122 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
1129 inode
->i_generation
= ip
->i_d
.di_gen
;
1130 i_size_write(inode
, ip
->i_d
.di_size
);
1131 inode
->i_atime
.tv_sec
= ip
->i_d
.di_atime
.t_sec
;
1132 inode
->i_atime
.tv_nsec
= ip
->i_d
.di_atime
.t_nsec
;
1133 inode
->i_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
1134 inode
->i_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
1135 inode
->i_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
1136 inode
->i_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
1137 xfs_diflags_to_iflags(inode
, ip
);
1139 switch (inode
->i_mode
& S_IFMT
) {
1141 inode
->i_op
= &xfs_inode_operations
;
1142 inode
->i_fop
= &xfs_file_operations
;
1143 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1146 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1147 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1149 inode
->i_op
= &xfs_dir_inode_operations
;
1150 inode
->i_fop
= &xfs_dir_file_operations
;
1153 inode
->i_op
= &xfs_symlink_inode_operations
;
1154 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1155 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1158 inode
->i_op
= &xfs_inode_operations
;
1159 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1164 * If there is no attribute fork no ACL can exist on this inode,
1165 * and it can't have any file capabilities attached to it either.
1167 if (!XFS_IFORK_Q(ip
)) {
1168 inode_has_no_xattr(inode
);
1169 cache_no_acl(inode
);
1172 xfs_iflags_clear(ip
, XFS_INEW
);
1175 unlock_new_inode(inode
);