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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_inode.h"
30 #include "xfs_bmap_util.h"
32 #include "xfs_quota.h"
33 #include "xfs_error.h"
35 #include "xfs_trans.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_symlink.h"
39 #include "xfs_da_btree.h"
40 #include "xfs_dir2_priv.h"
41 #include "xfs_dinode.h"
42 #include "xfs_trans_space.h"
44 #include <linux/capability.h>
45 #include <linux/xattr.h>
46 #include <linux/namei.h>
47 #include <linux/posix_acl.h>
48 #include <linux/security.h>
49 #include <linux/fiemap.h>
50 #include <linux/slab.h>
53 * Directories have different lock order w.r.t. mmap_sem compared to regular
54 * files. This is due to readdir potentially triggering page faults on a user
55 * buffer inside filldir(), and this happens with the ilock on the directory
56 * held. For regular files, the lock order is the other way around - the
57 * mmap_sem is taken during the page fault, and then we lock the ilock to do
58 * block mapping. Hence we need a different class for the directory ilock so
59 * that lockdep can tell them apart.
61 static struct lock_class_key xfs_nondir_ilock_class
;
62 static struct lock_class_key xfs_dir_ilock_class
;
67 const struct xattr
*xattr_array
,
70 const struct xattr
*xattr
;
71 struct xfs_inode
*ip
= XFS_I(inode
);
74 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
75 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
76 xattr
->value_len
, ATTR_SECURE
);
84 * Hook in SELinux. This is not quite correct yet, what we really need
85 * here (as we do for default ACLs) is a mechanism by which creation of
86 * these attrs can be journalled at inode creation time (along with the
87 * inode, of course, such that log replay can't cause these to be lost).
94 const struct qstr
*qstr
)
96 return security_inode_init_security(inode
, dir
, qstr
,
97 &xfs_initxattrs
, NULL
);
102 struct xfs_name
*namep
,
103 struct dentry
*dentry
,
106 namep
->name
= dentry
->d_name
.name
;
107 namep
->len
= dentry
->d_name
.len
;
108 namep
->type
= xfs_mode_to_ftype
[(mode
& S_IFMT
) >> S_SHIFT
];
115 struct dentry
*dentry
)
117 struct xfs_name teardown
;
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
124 xfs_dentry_to_name(&teardown
, dentry
, 0);
126 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
133 struct dentry
*dentry
,
138 struct xfs_inode
*ip
= NULL
;
139 struct posix_acl
*default_acl
, *acl
;
140 struct xfs_name name
;
144 * Irix uses Missed'em'V split, but doesn't want to see
145 * the upper 5 bits of (14bit) major.
147 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
148 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
150 rdev
= sysv_encode_dev(rdev
);
155 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
159 xfs_dentry_to_name(&name
, dentry
, mode
);
160 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
166 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
168 goto out_cleanup_inode
;
170 #ifdef CONFIG_XFS_POSIX_ACL
172 error
= xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
174 goto out_cleanup_inode
;
177 error
= xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
179 goto out_cleanup_inode
;
183 d_instantiate(dentry
, inode
);
186 posix_acl_release(default_acl
);
188 posix_acl_release(acl
);
192 xfs_cleanup_inode(dir
, inode
, dentry
);
199 struct dentry
*dentry
,
203 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
209 struct dentry
*dentry
,
212 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
215 STATIC
struct dentry
*
218 struct dentry
*dentry
,
221 struct xfs_inode
*cip
;
222 struct xfs_name name
;
225 if (dentry
->d_name
.len
>= MAXNAMELEN
)
226 return ERR_PTR(-ENAMETOOLONG
);
228 xfs_dentry_to_name(&name
, dentry
, 0);
229 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
230 if (unlikely(error
)) {
231 if (unlikely(error
!= ENOENT
))
232 return ERR_PTR(-error
);
237 return d_splice_alias(VFS_I(cip
), dentry
);
240 STATIC
struct dentry
*
243 struct dentry
*dentry
,
246 struct xfs_inode
*ip
;
247 struct xfs_name xname
;
248 struct xfs_name ci_name
;
252 if (dentry
->d_name
.len
>= MAXNAMELEN
)
253 return ERR_PTR(-ENAMETOOLONG
);
255 xfs_dentry_to_name(&xname
, dentry
, 0);
256 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
257 if (unlikely(error
)) {
258 if (unlikely(error
!= ENOENT
))
259 return ERR_PTR(-error
);
261 * call d_add(dentry, NULL) here when d_drop_negative_children
262 * is called in xfs_vn_mknod (ie. allow negative dentries
263 * with CI filesystems).
268 /* if exact match, just splice and exit */
270 return d_splice_alias(VFS_I(ip
), dentry
);
272 /* else case-insensitive match... */
273 dname
.name
= ci_name
.name
;
274 dname
.len
= ci_name
.len
;
275 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
276 kmem_free(ci_name
.name
);
282 struct dentry
*old_dentry
,
284 struct dentry
*dentry
)
286 struct inode
*inode
= old_dentry
->d_inode
;
287 struct xfs_name name
;
290 xfs_dentry_to_name(&name
, dentry
, inode
->i_mode
);
292 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
297 d_instantiate(dentry
, inode
);
304 struct dentry
*dentry
)
306 struct xfs_name name
;
309 xfs_dentry_to_name(&name
, dentry
, 0);
311 error
= -xfs_remove(XFS_I(dir
), &name
, XFS_I(dentry
->d_inode
));
316 * With unlink, the VFS makes the dentry "negative": no inode,
317 * but still hashed. This is incompatible with case-insensitive
318 * mode, so invalidate (unhash) the dentry in CI-mode.
320 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
321 d_invalidate(dentry
);
328 struct dentry
*dentry
,
332 struct xfs_inode
*cip
= NULL
;
333 struct xfs_name name
;
338 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
339 xfs_dentry_to_name(&name
, dentry
, mode
);
341 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
347 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
349 goto out_cleanup_inode
;
351 d_instantiate(dentry
, inode
);
355 xfs_cleanup_inode(dir
, inode
, dentry
);
363 struct dentry
*odentry
,
365 struct dentry
*ndentry
)
367 struct inode
*new_inode
= ndentry
->d_inode
;
368 struct xfs_name oname
;
369 struct xfs_name nname
;
371 xfs_dentry_to_name(&oname
, odentry
, 0);
372 xfs_dentry_to_name(&nname
, ndentry
, odentry
->d_inode
->i_mode
);
374 return -xfs_rename(XFS_I(odir
), &oname
, XFS_I(odentry
->d_inode
),
375 XFS_I(ndir
), &nname
, new_inode
?
376 XFS_I(new_inode
) : NULL
);
380 * careful here - this function can get called recursively, so
381 * we need to be very careful about how much stack we use.
382 * uio is kmalloced for this reason...
386 struct dentry
*dentry
,
387 struct nameidata
*nd
)
392 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
396 error
= -xfs_readlink(XFS_I(dentry
->d_inode
), link
);
400 nd_set_link(nd
, link
);
406 nd_set_link(nd
, ERR_PTR(error
));
412 struct vfsmount
*mnt
,
413 struct dentry
*dentry
,
416 struct inode
*inode
= dentry
->d_inode
;
417 struct xfs_inode
*ip
= XFS_I(inode
);
418 struct xfs_mount
*mp
= ip
->i_mount
;
420 trace_xfs_getattr(ip
);
422 if (XFS_FORCED_SHUTDOWN(mp
))
423 return -XFS_ERROR(EIO
);
425 stat
->size
= XFS_ISIZE(ip
);
426 stat
->dev
= inode
->i_sb
->s_dev
;
427 stat
->mode
= ip
->i_d
.di_mode
;
428 stat
->nlink
= ip
->i_d
.di_nlink
;
429 stat
->uid
= inode
->i_uid
;
430 stat
->gid
= inode
->i_gid
;
431 stat
->ino
= ip
->i_ino
;
432 stat
->atime
= inode
->i_atime
;
433 stat
->mtime
= inode
->i_mtime
;
434 stat
->ctime
= inode
->i_ctime
;
436 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
439 switch (inode
->i_mode
& S_IFMT
) {
442 stat
->blksize
= BLKDEV_IOSIZE
;
443 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
444 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
447 if (XFS_IS_REALTIME_INODE(ip
)) {
449 * If the file blocks are being allocated from a
450 * realtime volume, then return the inode's realtime
451 * extent size or the realtime volume's extent size.
454 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
456 stat
->blksize
= xfs_preferred_iosize(mp
);
466 struct xfs_inode
*ip
,
469 struct inode
*inode
= VFS_I(ip
);
470 umode_t mode
= iattr
->ia_mode
;
472 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
474 ip
->i_d
.di_mode
&= S_IFMT
;
475 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
477 inode
->i_mode
&= S_IFMT
;
478 inode
->i_mode
|= mode
& ~S_IFMT
;
483 struct xfs_inode
*ip
,
486 struct inode
*inode
= VFS_I(ip
);
488 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
490 if (iattr
->ia_valid
& ATTR_ATIME
) {
491 inode
->i_atime
= iattr
->ia_atime
;
492 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
493 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
495 if (iattr
->ia_valid
& ATTR_CTIME
) {
496 inode
->i_ctime
= iattr
->ia_ctime
;
497 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
498 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
500 if (iattr
->ia_valid
& ATTR_MTIME
) {
501 inode
->i_mtime
= iattr
->ia_mtime
;
502 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
503 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
509 struct xfs_inode
*ip
,
513 xfs_mount_t
*mp
= ip
->i_mount
;
514 struct inode
*inode
= VFS_I(ip
);
515 int mask
= iattr
->ia_valid
;
518 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
519 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
520 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
521 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
523 trace_xfs_setattr(ip
);
525 /* If acls are being inherited, we already have this checked */
526 if (!(flags
& XFS_ATTR_NOACL
)) {
527 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
528 return XFS_ERROR(EROFS
);
530 if (XFS_FORCED_SHUTDOWN(mp
))
531 return XFS_ERROR(EIO
);
533 error
= -inode_change_ok(inode
, iattr
);
535 return XFS_ERROR(error
);
538 ASSERT((mask
& ATTR_SIZE
) == 0);
541 * If disk quotas is on, we make sure that the dquots do exist on disk,
542 * before we start any other transactions. Trying to do this later
543 * is messy. We don't care to take a readlock to look at the ids
544 * in inode here, because we can't hold it across the trans_reserve.
545 * If the IDs do change before we take the ilock, we're covered
546 * because the i_*dquot fields will get updated anyway.
548 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
551 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
553 qflags
|= XFS_QMOPT_UQUOTA
;
557 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
559 qflags
|= XFS_QMOPT_GQUOTA
;
565 * We take a reference when we initialize udqp and gdqp,
566 * so it is important that we never blindly double trip on
567 * the same variable. See xfs_create() for an example.
569 ASSERT(udqp
== NULL
);
570 ASSERT(gdqp
== NULL
);
571 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
572 xfs_kgid_to_gid(gid
),
574 qflags
, &udqp
, &gdqp
, NULL
);
579 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
580 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
584 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
587 * Change file ownership. Must be the owner or privileged.
589 if (mask
& (ATTR_UID
|ATTR_GID
)) {
591 * These IDs could have changed since we last looked at them.
592 * But, we're assured that if the ownership did change
593 * while we didn't have the inode locked, inode's dquot(s)
594 * would have changed also.
598 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
599 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
602 * Do a quota reservation only if uid/gid is actually
605 if (XFS_IS_QUOTA_RUNNING(mp
) &&
606 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
607 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
609 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
610 NULL
, capable(CAP_FOWNER
) ?
611 XFS_QMOPT_FORCE_RES
: 0);
612 if (error
) /* out of quota */
613 goto out_trans_cancel
;
617 xfs_trans_ijoin(tp
, ip
, 0);
620 * Change file ownership. Must be the owner or privileged.
622 if (mask
& (ATTR_UID
|ATTR_GID
)) {
624 * CAP_FSETID overrides the following restrictions:
626 * The set-user-ID and set-group-ID bits of a file will be
627 * cleared upon successful return from chown()
629 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
630 !capable(CAP_FSETID
))
631 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
634 * Change the ownerships and register quota modifications
635 * in the transaction.
637 if (!uid_eq(iuid
, uid
)) {
638 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
639 ASSERT(mask
& ATTR_UID
);
641 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
642 &ip
->i_udquot
, udqp
);
644 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
647 if (!gid_eq(igid
, gid
)) {
648 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
649 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
650 !XFS_IS_PQUOTA_ON(mp
));
651 ASSERT(mask
& ATTR_GID
);
653 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
654 &ip
->i_gdquot
, gdqp
);
656 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
661 if (mask
& ATTR_MODE
)
662 xfs_setattr_mode(ip
, iattr
);
663 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
664 xfs_setattr_time(ip
, iattr
);
666 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
668 XFS_STATS_INC(xs_ig_attrchg
);
670 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
671 xfs_trans_set_sync(tp
);
672 error
= xfs_trans_commit(tp
, 0);
674 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
677 * Release any dquot(s) the inode had kept before chown.
679 xfs_qm_dqrele(olddquot1
);
680 xfs_qm_dqrele(olddquot2
);
685 return XFS_ERROR(error
);
688 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
689 * update. We could avoid this with linked transactions
690 * and passing down the transaction pointer all the way
691 * to attr_set. No previous user of the generic
692 * Posix ACL code seems to care about this issue either.
694 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
695 error
= -posix_acl_chmod(inode
, inode
->i_mode
);
697 return XFS_ERROR(error
);
703 xfs_trans_cancel(tp
, 0);
704 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
712 * Truncate file. Must have write permission and not be a directory.
716 struct xfs_inode
*ip
,
719 struct xfs_mount
*mp
= ip
->i_mount
;
720 struct inode
*inode
= VFS_I(ip
);
721 xfs_off_t oldsize
, newsize
;
722 struct xfs_trans
*tp
;
725 uint commit_flags
= 0;
727 trace_xfs_setattr(ip
);
729 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
730 return XFS_ERROR(EROFS
);
732 if (XFS_FORCED_SHUTDOWN(mp
))
733 return XFS_ERROR(EIO
);
735 error
= -inode_change_ok(inode
, iattr
);
737 return XFS_ERROR(error
);
739 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
740 ASSERT(S_ISREG(ip
->i_d
.di_mode
));
741 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
742 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
744 oldsize
= inode
->i_size
;
745 newsize
= iattr
->ia_size
;
748 * Short circuit the truncate case for zero length files.
750 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
751 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
755 * Use the regular setattr path to update the timestamps.
757 iattr
->ia_valid
&= ~ATTR_SIZE
;
758 return xfs_setattr_nonsize(ip
, iattr
, 0);
762 * Make sure that the dquots are attached to the inode.
764 error
= xfs_qm_dqattach(ip
, 0);
769 * Now we can make the changes. Before we join the inode to the
770 * transaction, take care of the part of the truncation that must be
771 * done without the inode lock. This needs to be done before joining
772 * the inode to the transaction, because the inode cannot be unlocked
773 * once it is a part of the transaction.
775 if (newsize
> oldsize
) {
777 * Do the first part of growing a file: zero any data in the
778 * last block that is beyond the old EOF. We need to do this
779 * before the inode is joined to the transaction to modify
782 error
= xfs_zero_eof(ip
, newsize
, oldsize
);
788 * We are going to log the inode size change in this transaction so
789 * any previous writes that are beyond the on disk EOF and the new
790 * EOF that have not been written out need to be written here. If we
791 * do not write the data out, we expose ourselves to the null files
794 * Only flush from the on disk size to the smaller of the in memory
795 * file size or the new size as that's the range we really care about
796 * here and prevents waiting for other data not within the range we
799 if (oldsize
!= ip
->i_d
.di_size
&& newsize
> ip
->i_d
.di_size
) {
800 error
= -filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
801 ip
->i_d
.di_size
, newsize
);
807 * Wait for all direct I/O to complete.
809 inode_dio_wait(inode
);
811 error
= -block_truncate_page(inode
->i_mapping
, newsize
, xfs_get_blocks
);
815 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
816 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
818 goto out_trans_cancel
;
820 truncate_setsize(inode
, newsize
);
822 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
823 lock_flags
|= XFS_ILOCK_EXCL
;
825 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
827 xfs_trans_ijoin(tp
, ip
, 0);
830 * Only change the c/mtime if we are changing the size or we are
831 * explicitly asked to change it. This handles the semantic difference
832 * between truncate() and ftruncate() as implemented in the VFS.
834 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
835 * special case where we need to update the times despite not having
836 * these flags set. For all other operations the VFS set these flags
837 * explicitly if it wants a timestamp update.
839 if (newsize
!= oldsize
&&
840 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
841 iattr
->ia_ctime
= iattr
->ia_mtime
=
842 current_fs_time(inode
->i_sb
);
843 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
847 * The first thing we do is set the size to new_size permanently on
848 * disk. This way we don't have to worry about anyone ever being able
849 * to look at the data being freed even in the face of a crash.
850 * What we're getting around here is the case where we free a block, it
851 * is allocated to another file, it is written to, and then we crash.
852 * If the new data gets written to the file but the log buffers
853 * containing the free and reallocation don't, then we'd end up with
854 * garbage in the blocks being freed. As long as we make the new size
855 * permanent before actually freeing any blocks it doesn't matter if
856 * they get written to.
858 ip
->i_d
.di_size
= newsize
;
859 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
861 if (newsize
<= oldsize
) {
862 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
864 goto out_trans_abort
;
867 * Truncated "down", so we're removing references to old data
868 * here - if we delay flushing for a long time, we expose
869 * ourselves unduly to the notorious NULL files problem. So,
870 * we mark this inode and flush it when the file is closed,
871 * and do not wait the usual (long) time for writeout.
873 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
875 /* A truncate down always removes post-EOF blocks. */
876 xfs_inode_clear_eofblocks_tag(ip
);
879 if (iattr
->ia_valid
& ATTR_MODE
)
880 xfs_setattr_mode(ip
, iattr
);
881 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
882 xfs_setattr_time(ip
, iattr
);
884 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
886 XFS_STATS_INC(xs_ig_attrchg
);
888 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
889 xfs_trans_set_sync(tp
);
891 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
894 xfs_iunlock(ip
, lock_flags
);
898 commit_flags
|= XFS_TRANS_ABORT
;
900 xfs_trans_cancel(tp
, commit_flags
);
906 struct dentry
*dentry
,
909 struct xfs_inode
*ip
= XFS_I(dentry
->d_inode
);
912 if (iattr
->ia_valid
& ATTR_SIZE
) {
913 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
914 error
= xfs_setattr_size(ip
, iattr
);
915 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
917 error
= xfs_setattr_nonsize(ip
, iattr
, 0);
926 struct timespec
*now
,
929 struct xfs_inode
*ip
= XFS_I(inode
);
930 struct xfs_mount
*mp
= ip
->i_mount
;
931 struct xfs_trans
*tp
;
934 trace_xfs_update_time(ip
);
936 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
937 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_fsyncts
, 0, 0);
939 xfs_trans_cancel(tp
, 0);
943 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
944 if (flags
& S_CTIME
) {
945 inode
->i_ctime
= *now
;
946 ip
->i_d
.di_ctime
.t_sec
= (__int32_t
)now
->tv_sec
;
947 ip
->i_d
.di_ctime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
949 if (flags
& S_MTIME
) {
950 inode
->i_mtime
= *now
;
951 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)now
->tv_sec
;
952 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
954 if (flags
& S_ATIME
) {
955 inode
->i_atime
= *now
;
956 ip
->i_d
.di_atime
.t_sec
= (__int32_t
)now
->tv_sec
;
957 ip
->i_d
.di_atime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
959 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
960 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
961 return -xfs_trans_commit(tp
, 0);
964 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
967 * Call fiemap helper to fill in user data.
968 * Returns positive errors to xfs_getbmap.
973 struct getbmapx
*bmv
,
977 struct fiemap_extent_info
*fieinfo
= *arg
;
978 u32 fiemap_flags
= 0;
979 u64 logical
, physical
, length
;
981 /* Do nothing for a hole */
982 if (bmv
->bmv_block
== -1LL)
985 logical
= BBTOB(bmv
->bmv_offset
);
986 physical
= BBTOB(bmv
->bmv_block
);
987 length
= BBTOB(bmv
->bmv_length
);
989 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
990 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
991 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
992 fiemap_flags
|= (FIEMAP_EXTENT_DELALLOC
|
993 FIEMAP_EXTENT_UNKNOWN
);
994 physical
= 0; /* no block yet */
996 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
997 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
999 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
1000 length
, fiemap_flags
);
1003 *full
= 1; /* user array now full */
1011 struct inode
*inode
,
1012 struct fiemap_extent_info
*fieinfo
,
1016 xfs_inode_t
*ip
= XFS_I(inode
);
1020 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
1024 /* Set up bmap header for xfs internal routine */
1025 bm
.bmv_offset
= BTOBB(start
);
1026 /* Special case for whole file */
1027 if (length
== FIEMAP_MAX_OFFSET
)
1028 bm
.bmv_length
= -1LL;
1030 bm
.bmv_length
= BTOBB(length
);
1032 /* We add one because in getbmap world count includes the header */
1033 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
1034 fieinfo
->fi_extents_max
+ 1;
1035 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
1036 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
1037 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
1038 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
1039 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
1040 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
1041 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
1043 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
1053 struct dentry
*dentry
,
1057 struct xfs_inode
*ip
;
1058 struct inode
*inode
;
1060 error
= xfs_create_tmpfile(XFS_I(dir
), dentry
, mode
, &ip
);
1061 if (unlikely(error
))
1066 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
1067 if (unlikely(error
)) {
1072 d_tmpfile(dentry
, inode
);
1077 static const struct inode_operations xfs_inode_operations
= {
1078 .get_acl
= xfs_get_acl
,
1079 .set_acl
= xfs_set_acl
,
1080 .getattr
= xfs_vn_getattr
,
1081 .setattr
= xfs_vn_setattr
,
1082 .setxattr
= generic_setxattr
,
1083 .getxattr
= generic_getxattr
,
1084 .removexattr
= generic_removexattr
,
1085 .listxattr
= xfs_vn_listxattr
,
1086 .fiemap
= xfs_vn_fiemap
,
1087 .update_time
= xfs_vn_update_time
,
1090 static const struct inode_operations xfs_dir_inode_operations
= {
1091 .create
= xfs_vn_create
,
1092 .lookup
= xfs_vn_lookup
,
1093 .link
= xfs_vn_link
,
1094 .unlink
= xfs_vn_unlink
,
1095 .symlink
= xfs_vn_symlink
,
1096 .mkdir
= xfs_vn_mkdir
,
1098 * Yes, XFS uses the same method for rmdir and unlink.
1100 * There are some subtile differences deeper in the code,
1101 * but we use S_ISDIR to check for those.
1103 .rmdir
= xfs_vn_unlink
,
1104 .mknod
= xfs_vn_mknod
,
1105 .rename
= xfs_vn_rename
,
1106 .get_acl
= xfs_get_acl
,
1107 .set_acl
= xfs_set_acl
,
1108 .getattr
= xfs_vn_getattr
,
1109 .setattr
= xfs_vn_setattr
,
1110 .setxattr
= generic_setxattr
,
1111 .getxattr
= generic_getxattr
,
1112 .removexattr
= generic_removexattr
,
1113 .listxattr
= xfs_vn_listxattr
,
1114 .update_time
= xfs_vn_update_time
,
1115 .tmpfile
= xfs_vn_tmpfile
,
1118 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1119 .create
= xfs_vn_create
,
1120 .lookup
= xfs_vn_ci_lookup
,
1121 .link
= xfs_vn_link
,
1122 .unlink
= xfs_vn_unlink
,
1123 .symlink
= xfs_vn_symlink
,
1124 .mkdir
= xfs_vn_mkdir
,
1126 * Yes, XFS uses the same method for rmdir and unlink.
1128 * There are some subtile differences deeper in the code,
1129 * but we use S_ISDIR to check for those.
1131 .rmdir
= xfs_vn_unlink
,
1132 .mknod
= xfs_vn_mknod
,
1133 .rename
= xfs_vn_rename
,
1134 .get_acl
= xfs_get_acl
,
1135 .set_acl
= xfs_set_acl
,
1136 .getattr
= xfs_vn_getattr
,
1137 .setattr
= xfs_vn_setattr
,
1138 .setxattr
= generic_setxattr
,
1139 .getxattr
= generic_getxattr
,
1140 .removexattr
= generic_removexattr
,
1141 .listxattr
= xfs_vn_listxattr
,
1142 .update_time
= xfs_vn_update_time
,
1143 .tmpfile
= xfs_vn_tmpfile
,
1146 static const struct inode_operations xfs_symlink_inode_operations
= {
1147 .readlink
= generic_readlink
,
1148 .follow_link
= xfs_vn_follow_link
,
1149 .put_link
= kfree_put_link
,
1150 .getattr
= xfs_vn_getattr
,
1151 .setattr
= xfs_vn_setattr
,
1152 .setxattr
= generic_setxattr
,
1153 .getxattr
= generic_getxattr
,
1154 .removexattr
= generic_removexattr
,
1155 .listxattr
= xfs_vn_listxattr
,
1156 .update_time
= xfs_vn_update_time
,
1160 xfs_diflags_to_iflags(
1161 struct inode
*inode
,
1162 struct xfs_inode
*ip
)
1164 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
1165 inode
->i_flags
|= S_IMMUTABLE
;
1167 inode
->i_flags
&= ~S_IMMUTABLE
;
1168 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
1169 inode
->i_flags
|= S_APPEND
;
1171 inode
->i_flags
&= ~S_APPEND
;
1172 if (ip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
)
1173 inode
->i_flags
|= S_SYNC
;
1175 inode
->i_flags
&= ~S_SYNC
;
1176 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
)
1177 inode
->i_flags
|= S_NOATIME
;
1179 inode
->i_flags
&= ~S_NOATIME
;
1183 * Initialize the Linux inode, set up the operation vectors and
1186 * When reading existing inodes from disk this is called directly
1187 * from xfs_iget, when creating a new inode it is called from
1188 * xfs_ialloc after setting up the inode.
1190 * We are always called with an uninitialised linux inode here.
1191 * We need to initialise the necessary fields and take a reference
1196 struct xfs_inode
*ip
)
1198 struct inode
*inode
= &ip
->i_vnode
;
1201 inode
->i_ino
= ip
->i_ino
;
1202 inode
->i_state
= I_NEW
;
1204 inode_sb_list_add(inode
);
1205 /* make the inode look hashed for the writeback code */
1206 hlist_add_fake(&inode
->i_hash
);
1208 inode
->i_mode
= ip
->i_d
.di_mode
;
1209 set_nlink(inode
, ip
->i_d
.di_nlink
);
1210 inode
->i_uid
= xfs_uid_to_kuid(ip
->i_d
.di_uid
);
1211 inode
->i_gid
= xfs_gid_to_kgid(ip
->i_d
.di_gid
);
1213 switch (inode
->i_mode
& S_IFMT
) {
1217 MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
1218 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
1225 inode
->i_generation
= ip
->i_d
.di_gen
;
1226 i_size_write(inode
, ip
->i_d
.di_size
);
1227 inode
->i_atime
.tv_sec
= ip
->i_d
.di_atime
.t_sec
;
1228 inode
->i_atime
.tv_nsec
= ip
->i_d
.di_atime
.t_nsec
;
1229 inode
->i_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
1230 inode
->i_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
1231 inode
->i_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
1232 inode
->i_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
1233 xfs_diflags_to_iflags(inode
, ip
);
1235 ip
->d_ops
= ip
->i_mount
->m_nondir_inode_ops
;
1236 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1237 switch (inode
->i_mode
& S_IFMT
) {
1239 inode
->i_op
= &xfs_inode_operations
;
1240 inode
->i_fop
= &xfs_file_operations
;
1241 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1244 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1245 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1246 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1248 inode
->i_op
= &xfs_dir_inode_operations
;
1249 inode
->i_fop
= &xfs_dir_file_operations
;
1250 ip
->d_ops
= ip
->i_mount
->m_dir_inode_ops
;
1253 inode
->i_op
= &xfs_symlink_inode_operations
;
1254 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1255 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1258 inode
->i_op
= &xfs_inode_operations
;
1259 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1264 * Ensure all page cache allocations are done from GFP_NOFS context to
1265 * prevent direct reclaim recursion back into the filesystem and blowing
1266 * stacks or deadlocking.
1268 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1269 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1272 * If there is no attribute fork no ACL can exist on this inode,
1273 * and it can't have any file capabilities attached to it either.
1275 if (!XFS_IFORK_Q(ip
)) {
1276 inode_has_no_xattr(inode
);
1277 cache_no_acl(inode
);
1280 xfs_iflags_clear(ip
, XFS_INEW
);
1283 unlock_new_inode(inode
);