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
20 #include "xfs_format.h"
21 #include "xfs_shared.h"
24 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_format.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_itable.h"
40 #include "xfs_buf_item.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_symlink.h"
45 #include "xfs_da_btree.h"
46 #include "xfs_dir2_priv.h"
48 #include <linux/capability.h>
49 #include <linux/xattr.h>
50 #include <linux/namei.h>
51 #include <linux/posix_acl.h>
52 #include <linux/security.h>
53 #include <linux/fiemap.h>
54 #include <linux/slab.h>
59 const struct xattr
*xattr_array
,
62 const struct xattr
*xattr
;
63 struct xfs_inode
*ip
= XFS_I(inode
);
66 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
67 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
68 xattr
->value_len
, ATTR_SECURE
);
76 * Hook in SELinux. This is not quite correct yet, what we really need
77 * here (as we do for default ACLs) is a mechanism by which creation of
78 * these attrs can be journalled at inode creation time (along with the
79 * inode, of course, such that log replay can't cause these to be lost).
86 const struct qstr
*qstr
)
88 return security_inode_init_security(inode
, dir
, qstr
,
89 &xfs_initxattrs
, NULL
);
94 struct xfs_name
*namep
,
95 struct dentry
*dentry
,
98 namep
->name
= dentry
->d_name
.name
;
99 namep
->len
= dentry
->d_name
.len
;
100 namep
->type
= xfs_mode_to_ftype
[(mode
& S_IFMT
) >> S_SHIFT
];
107 struct dentry
*dentry
)
109 struct xfs_name teardown
;
112 * If we can't add the ACL or we fail in
113 * xfs_init_security we must back out.
114 * ENOSPC can hit here, among other things.
116 xfs_dentry_to_name(&teardown
, dentry
, 0);
118 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
125 struct dentry
*dentry
,
130 struct xfs_inode
*ip
= NULL
;
131 struct posix_acl
*default_acl
= NULL
;
132 struct xfs_name name
;
136 * Irix uses Missed'em'V split, but doesn't want to see
137 * the upper 5 bits of (14bit) major.
139 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
140 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
142 rdev
= sysv_encode_dev(rdev
);
147 if (IS_POSIXACL(dir
)) {
148 default_acl
= xfs_get_acl(dir
, ACL_TYPE_DEFAULT
);
149 if (IS_ERR(default_acl
))
150 return PTR_ERR(default_acl
);
153 mode
&= ~current_umask();
156 xfs_dentry_to_name(&name
, dentry
, mode
);
157 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
163 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
165 goto out_cleanup_inode
;
168 error
= -xfs_inherit_acl(inode
, default_acl
);
171 goto out_cleanup_inode
;
175 d_instantiate(dentry
, inode
);
179 xfs_cleanup_inode(dir
, inode
, dentry
);
181 posix_acl_release(default_acl
);
188 struct dentry
*dentry
,
192 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
198 struct dentry
*dentry
,
201 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
204 STATIC
struct dentry
*
207 struct dentry
*dentry
,
210 struct xfs_inode
*cip
;
211 struct xfs_name name
;
214 if (dentry
->d_name
.len
>= MAXNAMELEN
)
215 return ERR_PTR(-ENAMETOOLONG
);
217 xfs_dentry_to_name(&name
, dentry
, 0);
218 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
219 if (unlikely(error
)) {
220 if (unlikely(error
!= ENOENT
))
221 return ERR_PTR(-error
);
226 return d_splice_alias(VFS_I(cip
), dentry
);
229 STATIC
struct dentry
*
232 struct dentry
*dentry
,
235 struct xfs_inode
*ip
;
236 struct xfs_name xname
;
237 struct xfs_name ci_name
;
241 if (dentry
->d_name
.len
>= MAXNAMELEN
)
242 return ERR_PTR(-ENAMETOOLONG
);
244 xfs_dentry_to_name(&xname
, dentry
, 0);
245 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
246 if (unlikely(error
)) {
247 if (unlikely(error
!= ENOENT
))
248 return ERR_PTR(-error
);
250 * call d_add(dentry, NULL) here when d_drop_negative_children
251 * is called in xfs_vn_mknod (ie. allow negative dentries
252 * with CI filesystems).
257 /* if exact match, just splice and exit */
259 return d_splice_alias(VFS_I(ip
), dentry
);
261 /* else case-insensitive match... */
262 dname
.name
= ci_name
.name
;
263 dname
.len
= ci_name
.len
;
264 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
265 kmem_free(ci_name
.name
);
271 struct dentry
*old_dentry
,
273 struct dentry
*dentry
)
275 struct inode
*inode
= old_dentry
->d_inode
;
276 struct xfs_name name
;
279 xfs_dentry_to_name(&name
, dentry
, inode
->i_mode
);
281 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
286 d_instantiate(dentry
, inode
);
293 struct dentry
*dentry
)
295 struct xfs_name name
;
298 xfs_dentry_to_name(&name
, dentry
, 0);
300 error
= -xfs_remove(XFS_I(dir
), &name
, XFS_I(dentry
->d_inode
));
305 * With unlink, the VFS makes the dentry "negative": no inode,
306 * but still hashed. This is incompatible with case-insensitive
307 * mode, so invalidate (unhash) the dentry in CI-mode.
309 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
310 d_invalidate(dentry
);
317 struct dentry
*dentry
,
321 struct xfs_inode
*cip
= NULL
;
322 struct xfs_name name
;
327 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
328 xfs_dentry_to_name(&name
, dentry
, mode
);
330 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
336 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
338 goto out_cleanup_inode
;
340 d_instantiate(dentry
, inode
);
344 xfs_cleanup_inode(dir
, inode
, dentry
);
352 struct dentry
*odentry
,
354 struct dentry
*ndentry
)
356 struct inode
*new_inode
= ndentry
->d_inode
;
357 struct xfs_name oname
;
358 struct xfs_name nname
;
360 xfs_dentry_to_name(&oname
, odentry
, 0);
361 xfs_dentry_to_name(&nname
, ndentry
, odentry
->d_inode
->i_mode
);
363 return -xfs_rename(XFS_I(odir
), &oname
, XFS_I(odentry
->d_inode
),
364 XFS_I(ndir
), &nname
, new_inode
?
365 XFS_I(new_inode
) : NULL
);
369 * careful here - this function can get called recursively, so
370 * we need to be very careful about how much stack we use.
371 * uio is kmalloced for this reason...
375 struct dentry
*dentry
,
376 struct nameidata
*nd
)
381 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
385 error
= -xfs_readlink(XFS_I(dentry
->d_inode
), link
);
389 nd_set_link(nd
, link
);
395 nd_set_link(nd
, ERR_PTR(error
));
401 struct dentry
*dentry
,
402 struct nameidata
*nd
,
405 char *s
= nd_get_link(nd
);
413 struct vfsmount
*mnt
,
414 struct dentry
*dentry
,
417 struct inode
*inode
= dentry
->d_inode
;
418 struct xfs_inode
*ip
= XFS_I(inode
);
419 struct xfs_mount
*mp
= ip
->i_mount
;
421 trace_xfs_getattr(ip
);
423 if (XFS_FORCED_SHUTDOWN(mp
))
424 return -XFS_ERROR(EIO
);
426 stat
->size
= XFS_ISIZE(ip
);
427 stat
->dev
= inode
->i_sb
->s_dev
;
428 stat
->mode
= ip
->i_d
.di_mode
;
429 stat
->nlink
= ip
->i_d
.di_nlink
;
430 stat
->uid
= inode
->i_uid
;
431 stat
->gid
= inode
->i_gid
;
432 stat
->ino
= ip
->i_ino
;
433 stat
->atime
= inode
->i_atime
;
434 stat
->mtime
= inode
->i_mtime
;
435 stat
->ctime
= inode
->i_ctime
;
437 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
440 switch (inode
->i_mode
& S_IFMT
) {
443 stat
->blksize
= BLKDEV_IOSIZE
;
444 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
445 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
448 if (XFS_IS_REALTIME_INODE(ip
)) {
450 * If the file blocks are being allocated from a
451 * realtime volume, then return the inode's realtime
452 * extent size or the realtime volume's extent size.
455 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
457 stat
->blksize
= xfs_preferred_iosize(mp
);
467 struct xfs_trans
*tp
,
468 struct xfs_inode
*ip
,
471 struct inode
*inode
= VFS_I(ip
);
472 umode_t mode
= iattr
->ia_mode
;
475 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
477 ip
->i_d
.di_mode
&= S_IFMT
;
478 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
480 inode
->i_mode
&= S_IFMT
;
481 inode
->i_mode
|= mode
& ~S_IFMT
;
486 struct xfs_inode
*ip
,
490 xfs_mount_t
*mp
= ip
->i_mount
;
491 struct inode
*inode
= VFS_I(ip
);
492 int mask
= iattr
->ia_valid
;
495 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
496 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
497 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
498 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
500 trace_xfs_setattr(ip
);
502 /* If acls are being inherited, we already have this checked */
503 if (!(flags
& XFS_ATTR_NOACL
)) {
504 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
505 return XFS_ERROR(EROFS
);
507 if (XFS_FORCED_SHUTDOWN(mp
))
508 return XFS_ERROR(EIO
);
510 error
= -inode_change_ok(inode
, iattr
);
512 return XFS_ERROR(error
);
515 ASSERT((mask
& ATTR_SIZE
) == 0);
518 * If disk quotas is on, we make sure that the dquots do exist on disk,
519 * before we start any other transactions. Trying to do this later
520 * is messy. We don't care to take a readlock to look at the ids
521 * in inode here, because we can't hold it across the trans_reserve.
522 * If the IDs do change before we take the ilock, we're covered
523 * because the i_*dquot fields will get updated anyway.
525 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
528 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
530 qflags
|= XFS_QMOPT_UQUOTA
;
534 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
536 qflags
|= XFS_QMOPT_GQUOTA
;
542 * We take a reference when we initialize udqp and gdqp,
543 * so it is important that we never blindly double trip on
544 * the same variable. See xfs_create() for an example.
546 ASSERT(udqp
== NULL
);
547 ASSERT(gdqp
== NULL
);
548 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
549 xfs_kgid_to_gid(gid
),
551 qflags
, &udqp
, &gdqp
, NULL
);
556 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
557 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
561 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
564 * Change file ownership. Must be the owner or privileged.
566 if (mask
& (ATTR_UID
|ATTR_GID
)) {
568 * These IDs could have changed since we last looked at them.
569 * But, we're assured that if the ownership did change
570 * while we didn't have the inode locked, inode's dquot(s)
571 * would have changed also.
575 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
576 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
579 * Do a quota reservation only if uid/gid is actually
582 if (XFS_IS_QUOTA_RUNNING(mp
) &&
583 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
584 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
586 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
587 NULL
, capable(CAP_FOWNER
) ?
588 XFS_QMOPT_FORCE_RES
: 0);
589 if (error
) /* out of quota */
590 goto out_trans_cancel
;
594 xfs_trans_ijoin(tp
, ip
, 0);
597 * Change file ownership. Must be the owner or privileged.
599 if (mask
& (ATTR_UID
|ATTR_GID
)) {
601 * CAP_FSETID overrides the following restrictions:
603 * The set-user-ID and set-group-ID bits of a file will be
604 * cleared upon successful return from chown()
606 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
607 !capable(CAP_FSETID
))
608 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
611 * Change the ownerships and register quota modifications
612 * in the transaction.
614 if (!uid_eq(iuid
, uid
)) {
615 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
616 ASSERT(mask
& ATTR_UID
);
618 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
619 &ip
->i_udquot
, udqp
);
621 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
624 if (!gid_eq(igid
, gid
)) {
625 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
626 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
627 ASSERT(mask
& ATTR_GID
);
629 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
630 &ip
->i_gdquot
, gdqp
);
632 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
638 * Change file access modes.
640 if (mask
& ATTR_MODE
)
641 xfs_setattr_mode(tp
, ip
, iattr
);
644 * Change file access or modified times.
646 if (mask
& ATTR_ATIME
) {
647 inode
->i_atime
= iattr
->ia_atime
;
648 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
649 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
651 if (mask
& ATTR_CTIME
) {
652 inode
->i_ctime
= iattr
->ia_ctime
;
653 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
654 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
656 if (mask
& ATTR_MTIME
) {
657 inode
->i_mtime
= iattr
->ia_mtime
;
658 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
659 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
662 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
664 XFS_STATS_INC(xs_ig_attrchg
);
666 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
667 xfs_trans_set_sync(tp
);
668 error
= xfs_trans_commit(tp
, 0);
670 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
673 * Release any dquot(s) the inode had kept before chown.
675 xfs_qm_dqrele(olddquot1
);
676 xfs_qm_dqrele(olddquot2
);
681 return XFS_ERROR(error
);
684 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
685 * update. We could avoid this with linked transactions
686 * and passing down the transaction pointer all the way
687 * to attr_set. No previous user of the generic
688 * Posix ACL code seems to care about this issue either.
690 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
691 error
= -xfs_acl_chmod(inode
);
693 return XFS_ERROR(error
);
699 xfs_trans_cancel(tp
, 0);
700 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
708 * Truncate file. Must have write permission and not be a directory.
712 struct xfs_inode
*ip
,
715 struct xfs_mount
*mp
= ip
->i_mount
;
716 struct inode
*inode
= VFS_I(ip
);
717 int mask
= iattr
->ia_valid
;
718 xfs_off_t oldsize
, newsize
;
719 struct xfs_trans
*tp
;
722 uint commit_flags
= 0;
724 trace_xfs_setattr(ip
);
726 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
727 return XFS_ERROR(EROFS
);
729 if (XFS_FORCED_SHUTDOWN(mp
))
730 return XFS_ERROR(EIO
);
732 error
= -inode_change_ok(inode
, iattr
);
734 return XFS_ERROR(error
);
736 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
737 ASSERT(S_ISREG(ip
->i_d
.di_mode
));
738 ASSERT((mask
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
739 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
741 oldsize
= inode
->i_size
;
742 newsize
= iattr
->ia_size
;
745 * Short circuit the truncate case for zero length files.
747 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
748 if (!(mask
& (ATTR_CTIME
|ATTR_MTIME
)))
752 * Use the regular setattr path to update the timestamps.
754 iattr
->ia_valid
&= ~ATTR_SIZE
;
755 return xfs_setattr_nonsize(ip
, iattr
, 0);
759 * Make sure that the dquots are attached to the inode.
761 error
= xfs_qm_dqattach(ip
, 0);
766 * Now we can make the changes. Before we join the inode to the
767 * transaction, take care of the part of the truncation that must be
768 * done without the inode lock. This needs to be done before joining
769 * the inode to the transaction, because the inode cannot be unlocked
770 * once it is a part of the transaction.
772 if (newsize
> oldsize
) {
774 * Do the first part of growing a file: zero any data in the
775 * last block that is beyond the old EOF. We need to do this
776 * before the inode is joined to the transaction to modify
779 error
= xfs_zero_eof(ip
, newsize
, oldsize
);
785 * We are going to log the inode size change in this transaction so
786 * any previous writes that are beyond the on disk EOF and the new
787 * EOF that have not been written out need to be written here. If we
788 * do not write the data out, we expose ourselves to the null files
791 * Only flush from the on disk size to the smaller of the in memory
792 * file size or the new size as that's the range we really care about
793 * here and prevents waiting for other data not within the range we
796 if (oldsize
!= ip
->i_d
.di_size
&& newsize
> ip
->i_d
.di_size
) {
797 error
= -filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
798 ip
->i_d
.di_size
, newsize
);
804 * Wait for all direct I/O to complete.
806 inode_dio_wait(inode
);
808 error
= -block_truncate_page(inode
->i_mapping
, newsize
, xfs_get_blocks
);
812 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
813 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
815 goto out_trans_cancel
;
817 truncate_setsize(inode
, newsize
);
819 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
820 lock_flags
|= XFS_ILOCK_EXCL
;
822 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
824 xfs_trans_ijoin(tp
, ip
, 0);
827 * Only change the c/mtime if we are changing the size or we are
828 * explicitly asked to change it. This handles the semantic difference
829 * between truncate() and ftruncate() as implemented in the VFS.
831 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
832 * special case where we need to update the times despite not having
833 * these flags set. For all other operations the VFS set these flags
834 * explicitly if it wants a timestamp update.
836 if (newsize
!= oldsize
&& (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))) {
837 iattr
->ia_ctime
= iattr
->ia_mtime
=
838 current_fs_time(inode
->i_sb
);
839 mask
|= ATTR_CTIME
| ATTR_MTIME
;
843 * The first thing we do is set the size to new_size permanently on
844 * disk. This way we don't have to worry about anyone ever being able
845 * to look at the data being freed even in the face of a crash.
846 * What we're getting around here is the case where we free a block, it
847 * is allocated to another file, it is written to, and then we crash.
848 * If the new data gets written to the file but the log buffers
849 * containing the free and reallocation don't, then we'd end up with
850 * garbage in the blocks being freed. As long as we make the new size
851 * permanent before actually freeing any blocks it doesn't matter if
852 * they get written to.
854 ip
->i_d
.di_size
= newsize
;
855 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
857 if (newsize
<= oldsize
) {
858 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
860 goto out_trans_abort
;
863 * Truncated "down", so we're removing references to old data
864 * here - if we delay flushing for a long time, we expose
865 * ourselves unduly to the notorious NULL files problem. So,
866 * we mark this inode and flush it when the file is closed,
867 * and do not wait the usual (long) time for writeout.
869 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
871 /* A truncate down always removes post-EOF blocks. */
872 xfs_inode_clear_eofblocks_tag(ip
);
876 * Change file access modes.
878 if (mask
& ATTR_MODE
)
879 xfs_setattr_mode(tp
, ip
, iattr
);
881 if (mask
& ATTR_CTIME
) {
882 inode
->i_ctime
= iattr
->ia_ctime
;
883 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
884 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
886 if (mask
& ATTR_MTIME
) {
887 inode
->i_mtime
= iattr
->ia_mtime
;
888 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
889 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
892 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
894 XFS_STATS_INC(xs_ig_attrchg
);
896 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
897 xfs_trans_set_sync(tp
);
899 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
902 xfs_iunlock(ip
, lock_flags
);
906 commit_flags
|= XFS_TRANS_ABORT
;
908 xfs_trans_cancel(tp
, commit_flags
);
914 struct dentry
*dentry
,
917 struct xfs_inode
*ip
= XFS_I(dentry
->d_inode
);
920 if (iattr
->ia_valid
& ATTR_SIZE
) {
921 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
922 error
= xfs_setattr_size(ip
, iattr
);
923 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
925 error
= xfs_setattr_nonsize(ip
, iattr
, 0);
934 struct timespec
*now
,
937 struct xfs_inode
*ip
= XFS_I(inode
);
938 struct xfs_mount
*mp
= ip
->i_mount
;
939 struct xfs_trans
*tp
;
942 trace_xfs_update_time(ip
);
944 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
945 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_fsyncts
, 0, 0);
947 xfs_trans_cancel(tp
, 0);
951 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
952 if (flags
& S_CTIME
) {
953 inode
->i_ctime
= *now
;
954 ip
->i_d
.di_ctime
.t_sec
= (__int32_t
)now
->tv_sec
;
955 ip
->i_d
.di_ctime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
957 if (flags
& S_MTIME
) {
958 inode
->i_mtime
= *now
;
959 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)now
->tv_sec
;
960 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
962 if (flags
& S_ATIME
) {
963 inode
->i_atime
= *now
;
964 ip
->i_d
.di_atime
.t_sec
= (__int32_t
)now
->tv_sec
;
965 ip
->i_d
.di_atime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
967 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
968 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
969 return -xfs_trans_commit(tp
, 0);
972 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
975 * Call fiemap helper to fill in user data.
976 * Returns positive errors to xfs_getbmap.
981 struct getbmapx
*bmv
,
985 struct fiemap_extent_info
*fieinfo
= *arg
;
986 u32 fiemap_flags
= 0;
987 u64 logical
, physical
, length
;
989 /* Do nothing for a hole */
990 if (bmv
->bmv_block
== -1LL)
993 logical
= BBTOB(bmv
->bmv_offset
);
994 physical
= BBTOB(bmv
->bmv_block
);
995 length
= BBTOB(bmv
->bmv_length
);
997 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
998 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
999 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
1000 fiemap_flags
|= (FIEMAP_EXTENT_DELALLOC
|
1001 FIEMAP_EXTENT_UNKNOWN
);
1002 physical
= 0; /* no block yet */
1004 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
1005 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
1007 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
1008 length
, fiemap_flags
);
1011 *full
= 1; /* user array now full */
1019 struct inode
*inode
,
1020 struct fiemap_extent_info
*fieinfo
,
1024 xfs_inode_t
*ip
= XFS_I(inode
);
1028 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
1032 /* Set up bmap header for xfs internal routine */
1033 bm
.bmv_offset
= BTOBB(start
);
1034 /* Special case for whole file */
1035 if (length
== FIEMAP_MAX_OFFSET
)
1036 bm
.bmv_length
= -1LL;
1038 bm
.bmv_length
= BTOBB(length
);
1040 /* We add one because in getbmap world count includes the header */
1041 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
1042 fieinfo
->fi_extents_max
+ 1;
1043 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
1044 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
1045 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
1046 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
1047 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
1048 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
1049 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
1051 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
1058 static const struct inode_operations xfs_inode_operations
= {
1059 .get_acl
= xfs_get_acl
,
1060 .getattr
= xfs_vn_getattr
,
1061 .setattr
= xfs_vn_setattr
,
1062 .setxattr
= generic_setxattr
,
1063 .getxattr
= generic_getxattr
,
1064 .removexattr
= generic_removexattr
,
1065 .listxattr
= xfs_vn_listxattr
,
1066 .fiemap
= xfs_vn_fiemap
,
1067 .update_time
= xfs_vn_update_time
,
1070 static const struct inode_operations xfs_dir_inode_operations
= {
1071 .create
= xfs_vn_create
,
1072 .lookup
= xfs_vn_lookup
,
1073 .link
= xfs_vn_link
,
1074 .unlink
= xfs_vn_unlink
,
1075 .symlink
= xfs_vn_symlink
,
1076 .mkdir
= xfs_vn_mkdir
,
1078 * Yes, XFS uses the same method for rmdir and unlink.
1080 * There are some subtile differences deeper in the code,
1081 * but we use S_ISDIR to check for those.
1083 .rmdir
= xfs_vn_unlink
,
1084 .mknod
= xfs_vn_mknod
,
1085 .rename
= xfs_vn_rename
,
1086 .get_acl
= xfs_get_acl
,
1087 .getattr
= xfs_vn_getattr
,
1088 .setattr
= xfs_vn_setattr
,
1089 .setxattr
= generic_setxattr
,
1090 .getxattr
= generic_getxattr
,
1091 .removexattr
= generic_removexattr
,
1092 .listxattr
= xfs_vn_listxattr
,
1093 .update_time
= xfs_vn_update_time
,
1096 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1097 .create
= xfs_vn_create
,
1098 .lookup
= xfs_vn_ci_lookup
,
1099 .link
= xfs_vn_link
,
1100 .unlink
= xfs_vn_unlink
,
1101 .symlink
= xfs_vn_symlink
,
1102 .mkdir
= xfs_vn_mkdir
,
1104 * Yes, XFS uses the same method for rmdir and unlink.
1106 * There are some subtile differences deeper in the code,
1107 * but we use S_ISDIR to check for those.
1109 .rmdir
= xfs_vn_unlink
,
1110 .mknod
= xfs_vn_mknod
,
1111 .rename
= xfs_vn_rename
,
1112 .get_acl
= xfs_get_acl
,
1113 .getattr
= xfs_vn_getattr
,
1114 .setattr
= xfs_vn_setattr
,
1115 .setxattr
= generic_setxattr
,
1116 .getxattr
= generic_getxattr
,
1117 .removexattr
= generic_removexattr
,
1118 .listxattr
= xfs_vn_listxattr
,
1119 .update_time
= xfs_vn_update_time
,
1122 static const struct inode_operations xfs_symlink_inode_operations
= {
1123 .readlink
= generic_readlink
,
1124 .follow_link
= xfs_vn_follow_link
,
1125 .put_link
= xfs_vn_put_link
,
1126 .get_acl
= xfs_get_acl
,
1127 .getattr
= xfs_vn_getattr
,
1128 .setattr
= xfs_vn_setattr
,
1129 .setxattr
= generic_setxattr
,
1130 .getxattr
= generic_getxattr
,
1131 .removexattr
= generic_removexattr
,
1132 .listxattr
= xfs_vn_listxattr
,
1133 .update_time
= xfs_vn_update_time
,
1137 xfs_diflags_to_iflags(
1138 struct inode
*inode
,
1139 struct xfs_inode
*ip
)
1141 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
1142 inode
->i_flags
|= S_IMMUTABLE
;
1144 inode
->i_flags
&= ~S_IMMUTABLE
;
1145 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
1146 inode
->i_flags
|= S_APPEND
;
1148 inode
->i_flags
&= ~S_APPEND
;
1149 if (ip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
)
1150 inode
->i_flags
|= S_SYNC
;
1152 inode
->i_flags
&= ~S_SYNC
;
1153 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
)
1154 inode
->i_flags
|= S_NOATIME
;
1156 inode
->i_flags
&= ~S_NOATIME
;
1160 * Initialize the Linux inode, set up the operation vectors and
1163 * When reading existing inodes from disk this is called directly
1164 * from xfs_iget, when creating a new inode it is called from
1165 * xfs_ialloc after setting up the inode.
1167 * We are always called with an uninitialised linux inode here.
1168 * We need to initialise the necessary fields and take a reference
1173 struct xfs_inode
*ip
)
1175 struct inode
*inode
= &ip
->i_vnode
;
1177 inode
->i_ino
= ip
->i_ino
;
1178 inode
->i_state
= I_NEW
;
1180 inode_sb_list_add(inode
);
1181 /* make the inode look hashed for the writeback code */
1182 hlist_add_fake(&inode
->i_hash
);
1184 inode
->i_mode
= ip
->i_d
.di_mode
;
1185 set_nlink(inode
, ip
->i_d
.di_nlink
);
1186 inode
->i_uid
= xfs_uid_to_kuid(ip
->i_d
.di_uid
);
1187 inode
->i_gid
= xfs_gid_to_kgid(ip
->i_d
.di_gid
);
1189 switch (inode
->i_mode
& S_IFMT
) {
1193 MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
1194 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
1201 inode
->i_generation
= ip
->i_d
.di_gen
;
1202 i_size_write(inode
, ip
->i_d
.di_size
);
1203 inode
->i_atime
.tv_sec
= ip
->i_d
.di_atime
.t_sec
;
1204 inode
->i_atime
.tv_nsec
= ip
->i_d
.di_atime
.t_nsec
;
1205 inode
->i_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
1206 inode
->i_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
1207 inode
->i_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
1208 inode
->i_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
1209 xfs_diflags_to_iflags(inode
, ip
);
1211 switch (inode
->i_mode
& S_IFMT
) {
1213 inode
->i_op
= &xfs_inode_operations
;
1214 inode
->i_fop
= &xfs_file_operations
;
1215 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1218 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1219 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1221 inode
->i_op
= &xfs_dir_inode_operations
;
1222 inode
->i_fop
= &xfs_dir_file_operations
;
1225 inode
->i_op
= &xfs_symlink_inode_operations
;
1226 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1227 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1230 inode
->i_op
= &xfs_inode_operations
;
1231 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1236 * If there is no attribute fork no ACL can exist on this inode,
1237 * and it can't have any file capabilities attached to it either.
1239 if (!XFS_IFORK_Q(ip
)) {
1240 inode_has_no_xattr(inode
);
1241 cache_no_acl(inode
);
1244 xfs_iflags_clear(ip
, XFS_INEW
);
1247 unlock_new_inode(inode
);