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"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_inode.h"
28 #include "xfs_bmap_util.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_symlink.h"
37 #include "xfs_da_btree.h"
39 #include "xfs_trans_space.h"
42 #include <linux/capability.h>
43 #include <linux/xattr.h>
44 #include <linux/namei.h>
45 #include <linux/posix_acl.h>
46 #include <linux/security.h>
47 #include <linux/fiemap.h>
48 #include <linux/slab.h>
51 * Directories have different lock order w.r.t. mmap_sem compared to regular
52 * files. This is due to readdir potentially triggering page faults on a user
53 * buffer inside filldir(), and this happens with the ilock on the directory
54 * held. For regular files, the lock order is the other way around - the
55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
56 * block mapping. Hence we need a different class for the directory ilock so
57 * that lockdep can tell them apart.
59 static struct lock_class_key xfs_nondir_ilock_class
;
60 static struct lock_class_key xfs_dir_ilock_class
;
65 const struct xattr
*xattr_array
,
68 const struct xattr
*xattr
;
69 struct xfs_inode
*ip
= XFS_I(inode
);
72 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
73 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
74 xattr
->value_len
, ATTR_SECURE
);
82 * Hook in SELinux. This is not quite correct yet, what we really need
83 * here (as we do for default ACLs) is a mechanism by which creation of
84 * these attrs can be journalled at inode creation time (along with the
85 * inode, of course, such that log replay can't cause these to be lost).
92 const struct qstr
*qstr
)
94 return security_inode_init_security(inode
, dir
, qstr
,
95 &xfs_initxattrs
, NULL
);
100 struct xfs_name
*namep
,
101 struct dentry
*dentry
,
104 namep
->name
= dentry
->d_name
.name
;
105 namep
->len
= dentry
->d_name
.len
;
106 namep
->type
= xfs_mode_to_ftype
[(mode
& S_IFMT
) >> S_SHIFT
];
113 struct dentry
*dentry
)
115 struct xfs_name teardown
;
118 * If we can't add the ACL or we fail in
119 * xfs_init_security we must back out.
120 * ENOSPC can hit here, among other things.
122 xfs_dentry_to_name(&teardown
, dentry
, 0);
124 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
130 struct dentry
*dentry
,
133 bool tmpfile
) /* unnamed file */
136 struct xfs_inode
*ip
= NULL
;
137 struct posix_acl
*default_acl
, *acl
;
138 struct xfs_name name
;
142 * Irix uses Missed'em'V split, but doesn't want to see
143 * the upper 5 bits of (14bit) major.
145 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
146 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
148 rdev
= sysv_encode_dev(rdev
);
153 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
158 xfs_dentry_to_name(&name
, dentry
, mode
);
159 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
161 error
= xfs_create_tmpfile(XFS_I(dir
), dentry
, mode
, &ip
);
168 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
170 goto out_cleanup_inode
;
172 #ifdef CONFIG_XFS_POSIX_ACL
174 error
= xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
176 goto out_cleanup_inode
;
179 error
= xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
181 goto out_cleanup_inode
;
186 d_tmpfile(dentry
, inode
);
188 d_instantiate(dentry
, inode
);
192 posix_acl_release(default_acl
);
194 posix_acl_release(acl
);
199 xfs_cleanup_inode(dir
, inode
, dentry
);
207 struct dentry
*dentry
,
211 return xfs_generic_create(dir
, dentry
, mode
, rdev
, false);
217 struct dentry
*dentry
,
221 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
227 struct dentry
*dentry
,
230 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
233 STATIC
struct dentry
*
236 struct dentry
*dentry
,
239 struct xfs_inode
*cip
;
240 struct xfs_name name
;
243 if (dentry
->d_name
.len
>= MAXNAMELEN
)
244 return ERR_PTR(-ENAMETOOLONG
);
246 xfs_dentry_to_name(&name
, dentry
, 0);
247 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
248 if (unlikely(error
)) {
249 if (unlikely(error
!= -ENOENT
))
250 return ERR_PTR(error
);
255 return d_splice_alias(VFS_I(cip
), dentry
);
258 STATIC
struct dentry
*
261 struct dentry
*dentry
,
264 struct xfs_inode
*ip
;
265 struct xfs_name xname
;
266 struct xfs_name ci_name
;
270 if (dentry
->d_name
.len
>= MAXNAMELEN
)
271 return ERR_PTR(-ENAMETOOLONG
);
273 xfs_dentry_to_name(&xname
, dentry
, 0);
274 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
275 if (unlikely(error
)) {
276 if (unlikely(error
!= -ENOENT
))
277 return ERR_PTR(error
);
279 * call d_add(dentry, NULL) here when d_drop_negative_children
280 * is called in xfs_vn_mknod (ie. allow negative dentries
281 * with CI filesystems).
286 /* if exact match, just splice and exit */
288 return d_splice_alias(VFS_I(ip
), dentry
);
290 /* else case-insensitive match... */
291 dname
.name
= ci_name
.name
;
292 dname
.len
= ci_name
.len
;
293 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
294 kmem_free(ci_name
.name
);
300 struct dentry
*old_dentry
,
302 struct dentry
*dentry
)
304 struct inode
*inode
= d_inode(old_dentry
);
305 struct xfs_name name
;
308 xfs_dentry_to_name(&name
, dentry
, inode
->i_mode
);
310 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
315 d_instantiate(dentry
, inode
);
322 struct dentry
*dentry
)
324 struct xfs_name name
;
327 xfs_dentry_to_name(&name
, dentry
, 0);
329 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(d_inode(dentry
)));
334 * With unlink, the VFS makes the dentry "negative": no inode,
335 * but still hashed. This is incompatible with case-insensitive
336 * mode, so invalidate (unhash) the dentry in CI-mode.
338 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
339 d_invalidate(dentry
);
346 struct dentry
*dentry
,
350 struct xfs_inode
*cip
= NULL
;
351 struct xfs_name name
;
356 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
357 xfs_dentry_to_name(&name
, dentry
, mode
);
359 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
365 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
367 goto out_cleanup_inode
;
369 d_instantiate(dentry
, inode
);
373 xfs_cleanup_inode(dir
, inode
, dentry
);
382 struct dentry
*odentry
,
384 struct dentry
*ndentry
,
387 struct inode
*new_inode
= d_inode(ndentry
);
389 struct xfs_name oname
;
390 struct xfs_name nname
;
392 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
))
395 /* if we are exchanging files, we need to set i_mode of both files */
396 if (flags
& RENAME_EXCHANGE
)
397 omode
= d_inode(ndentry
)->i_mode
;
399 xfs_dentry_to_name(&oname
, odentry
, omode
);
400 xfs_dentry_to_name(&nname
, ndentry
, d_inode(odentry
)->i_mode
);
402 return xfs_rename(XFS_I(odir
), &oname
, XFS_I(d_inode(odentry
)),
404 new_inode
? XFS_I(new_inode
) : NULL
, flags
);
408 * careful here - this function can get called recursively, so
409 * we need to be very careful about how much stack we use.
410 * uio is kmalloced for this reason...
414 struct dentry
*dentry
,
415 struct nameidata
*nd
)
420 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
424 error
= xfs_readlink(XFS_I(d_inode(dentry
)), link
);
428 nd_set_link(nd
, link
);
434 nd_set_link(nd
, ERR_PTR(error
));
440 struct vfsmount
*mnt
,
441 struct dentry
*dentry
,
444 struct inode
*inode
= d_inode(dentry
);
445 struct xfs_inode
*ip
= XFS_I(inode
);
446 struct xfs_mount
*mp
= ip
->i_mount
;
448 trace_xfs_getattr(ip
);
450 if (XFS_FORCED_SHUTDOWN(mp
))
453 stat
->size
= XFS_ISIZE(ip
);
454 stat
->dev
= inode
->i_sb
->s_dev
;
455 stat
->mode
= ip
->i_d
.di_mode
;
456 stat
->nlink
= ip
->i_d
.di_nlink
;
457 stat
->uid
= inode
->i_uid
;
458 stat
->gid
= inode
->i_gid
;
459 stat
->ino
= ip
->i_ino
;
460 stat
->atime
= inode
->i_atime
;
461 stat
->mtime
= inode
->i_mtime
;
462 stat
->ctime
= inode
->i_ctime
;
464 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
467 switch (inode
->i_mode
& S_IFMT
) {
470 stat
->blksize
= BLKDEV_IOSIZE
;
471 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
472 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
475 if (XFS_IS_REALTIME_INODE(ip
)) {
477 * If the file blocks are being allocated from a
478 * realtime volume, then return the inode's realtime
479 * extent size or the realtime volume's extent size.
482 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
484 stat
->blksize
= xfs_preferred_iosize(mp
);
494 struct xfs_inode
*ip
,
497 struct inode
*inode
= VFS_I(ip
);
498 umode_t mode
= iattr
->ia_mode
;
500 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
502 ip
->i_d
.di_mode
&= S_IFMT
;
503 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
505 inode
->i_mode
&= S_IFMT
;
506 inode
->i_mode
|= mode
& ~S_IFMT
;
511 struct xfs_inode
*ip
,
514 struct inode
*inode
= VFS_I(ip
);
516 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
518 if (iattr
->ia_valid
& ATTR_ATIME
) {
519 inode
->i_atime
= iattr
->ia_atime
;
520 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
521 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
523 if (iattr
->ia_valid
& ATTR_CTIME
) {
524 inode
->i_ctime
= iattr
->ia_ctime
;
525 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
526 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
528 if (iattr
->ia_valid
& ATTR_MTIME
) {
529 inode
->i_mtime
= iattr
->ia_mtime
;
530 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
531 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
537 struct xfs_inode
*ip
,
541 xfs_mount_t
*mp
= ip
->i_mount
;
542 struct inode
*inode
= VFS_I(ip
);
543 int mask
= iattr
->ia_valid
;
546 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
547 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
548 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
549 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
551 trace_xfs_setattr(ip
);
553 /* If acls are being inherited, we already have this checked */
554 if (!(flags
& XFS_ATTR_NOACL
)) {
555 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
558 if (XFS_FORCED_SHUTDOWN(mp
))
561 error
= inode_change_ok(inode
, iattr
);
566 ASSERT((mask
& ATTR_SIZE
) == 0);
569 * If disk quotas is on, we make sure that the dquots do exist on disk,
570 * before we start any other transactions. Trying to do this later
571 * is messy. We don't care to take a readlock to look at the ids
572 * in inode here, because we can't hold it across the trans_reserve.
573 * If the IDs do change before we take the ilock, we're covered
574 * because the i_*dquot fields will get updated anyway.
576 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
579 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
581 qflags
|= XFS_QMOPT_UQUOTA
;
585 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
587 qflags
|= XFS_QMOPT_GQUOTA
;
593 * We take a reference when we initialize udqp and gdqp,
594 * so it is important that we never blindly double trip on
595 * the same variable. See xfs_create() for an example.
597 ASSERT(udqp
== NULL
);
598 ASSERT(gdqp
== NULL
);
599 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
600 xfs_kgid_to_gid(gid
),
602 qflags
, &udqp
, &gdqp
, NULL
);
607 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
608 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
612 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
615 * Change file ownership. Must be the owner or privileged.
617 if (mask
& (ATTR_UID
|ATTR_GID
)) {
619 * These IDs could have changed since we last looked at them.
620 * But, we're assured that if the ownership did change
621 * while we didn't have the inode locked, inode's dquot(s)
622 * would have changed also.
626 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
627 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
630 * Do a quota reservation only if uid/gid is actually
633 if (XFS_IS_QUOTA_RUNNING(mp
) &&
634 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
635 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
637 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
638 NULL
, capable(CAP_FOWNER
) ?
639 XFS_QMOPT_FORCE_RES
: 0);
640 if (error
) /* out of quota */
641 goto out_trans_cancel
;
645 xfs_trans_ijoin(tp
, ip
, 0);
648 * Change file ownership. Must be the owner or privileged.
650 if (mask
& (ATTR_UID
|ATTR_GID
)) {
652 * CAP_FSETID overrides the following restrictions:
654 * The set-user-ID and set-group-ID bits of a file will be
655 * cleared upon successful return from chown()
657 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
658 !capable(CAP_FSETID
))
659 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
662 * Change the ownerships and register quota modifications
663 * in the transaction.
665 if (!uid_eq(iuid
, uid
)) {
666 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
667 ASSERT(mask
& ATTR_UID
);
669 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
670 &ip
->i_udquot
, udqp
);
672 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
675 if (!gid_eq(igid
, gid
)) {
676 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
677 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
678 !XFS_IS_PQUOTA_ON(mp
));
679 ASSERT(mask
& ATTR_GID
);
681 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
682 &ip
->i_gdquot
, gdqp
);
684 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
689 if (mask
& ATTR_MODE
)
690 xfs_setattr_mode(ip
, iattr
);
691 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
692 xfs_setattr_time(ip
, iattr
);
694 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
696 XFS_STATS_INC(xs_ig_attrchg
);
698 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
699 xfs_trans_set_sync(tp
);
700 error
= xfs_trans_commit(tp
, 0);
702 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
705 * Release any dquot(s) the inode had kept before chown.
707 xfs_qm_dqrele(olddquot1
);
708 xfs_qm_dqrele(olddquot2
);
716 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
717 * update. We could avoid this with linked transactions
718 * and passing down the transaction pointer all the way
719 * to attr_set. No previous user of the generic
720 * Posix ACL code seems to care about this issue either.
722 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
723 error
= posix_acl_chmod(inode
, inode
->i_mode
);
731 xfs_trans_cancel(tp
, 0);
732 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
740 * Truncate file. Must have write permission and not be a directory.
744 struct xfs_inode
*ip
,
747 struct xfs_mount
*mp
= ip
->i_mount
;
748 struct inode
*inode
= VFS_I(ip
);
749 xfs_off_t oldsize
, newsize
;
750 struct xfs_trans
*tp
;
753 uint commit_flags
= 0;
754 bool did_zeroing
= false;
756 trace_xfs_setattr(ip
);
758 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
761 if (XFS_FORCED_SHUTDOWN(mp
))
764 error
= inode_change_ok(inode
, iattr
);
768 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
769 ASSERT(S_ISREG(ip
->i_d
.di_mode
));
770 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
771 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
773 oldsize
= inode
->i_size
;
774 newsize
= iattr
->ia_size
;
777 * Short circuit the truncate case for zero length files.
779 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
780 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
784 * Use the regular setattr path to update the timestamps.
786 iattr
->ia_valid
&= ~ATTR_SIZE
;
787 return xfs_setattr_nonsize(ip
, iattr
, 0);
791 * Make sure that the dquots are attached to the inode.
793 error
= xfs_qm_dqattach(ip
, 0);
798 * File data changes must be complete before we start the transaction to
799 * modify the inode. This needs to be done before joining the inode to
800 * the transaction because the inode cannot be unlocked once it is a
801 * part of the transaction.
803 * Start with zeroing any data block beyond EOF that we may expose on
806 if (newsize
> oldsize
) {
807 error
= xfs_zero_eof(ip
, newsize
, oldsize
, &did_zeroing
);
813 * We are going to log the inode size change in this transaction so
814 * any previous writes that are beyond the on disk EOF and the new
815 * EOF that have not been written out need to be written here. If we
816 * do not write the data out, we expose ourselves to the null files
817 * problem. Note that this includes any block zeroing we did above;
818 * otherwise those blocks may not be zeroed after a crash.
820 if (newsize
> ip
->i_d
.di_size
&&
821 (oldsize
!= ip
->i_d
.di_size
|| did_zeroing
)) {
822 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
823 ip
->i_d
.di_size
, newsize
);
828 /* Now wait for all direct I/O to complete. */
829 inode_dio_wait(inode
);
832 * Do all the page cache truncate work outside the transaction context
833 * as the "lock" order is page lock->log space reservation. i.e.
834 * locking pages inside the transaction can ABBA deadlock with
835 * writeback. We have to do the VFS inode size update before we truncate
836 * the pagecache, however, to avoid racing with page faults beyond the
837 * new EOF they are not serialised against truncate operations except by
838 * page locks and size updates.
840 * Hence we are in a situation where a truncate can fail with ENOMEM
841 * from xfs_trans_reserve(), but having already truncated the in-memory
842 * version of the file (i.e. made user visible changes). There's not
843 * much we can do about this, except to hope that the caller sees ENOMEM
844 * and retries the truncate operation.
846 error
= block_truncate_page(inode
->i_mapping
, newsize
, xfs_get_blocks
);
849 truncate_setsize(inode
, newsize
);
852 * The "we can't serialise against page faults" pain gets worse.
854 * If the file is mapped then we have to clean the page at the old EOF
855 * when extending the file. Extending the file can expose changes the
856 * underlying page mapping (e.g. from beyond EOF to a hole or
857 * unwritten), and so on the next attempt to write to that page we need
858 * to remap it for write. i.e. we need .page_mkwrite() to be called.
859 * Hence we need to clean the page to clean the pte and so a new write
860 * fault will be triggered appropriately.
862 * If we do it before we change the inode size, then we can race with a
863 * page fault that maps the page with exactly the same problem. If we do
864 * it after we change the file size, then a new page fault can come in
865 * and allocate space before we've run the rest of the truncate
866 * transaction. That's kinda grotesque, but it's better than have data
867 * over a hole, and so that's the lesser evil that has been chosen here.
869 * The real solution, however, is to have some mechanism for locking out
870 * page faults while a truncate is in progress.
872 if (newsize
> oldsize
&& mapping_mapped(VFS_I(ip
)->i_mapping
)) {
873 error
= filemap_write_and_wait_range(
874 VFS_I(ip
)->i_mapping
,
875 round_down(oldsize
, PAGE_CACHE_SIZE
),
876 round_up(oldsize
, PAGE_CACHE_SIZE
) - 1);
881 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
882 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
884 goto out_trans_cancel
;
886 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
887 lock_flags
|= XFS_ILOCK_EXCL
;
888 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
889 xfs_trans_ijoin(tp
, ip
, 0);
892 * Only change the c/mtime if we are changing the size or we are
893 * explicitly asked to change it. This handles the semantic difference
894 * between truncate() and ftruncate() as implemented in the VFS.
896 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
897 * special case where we need to update the times despite not having
898 * these flags set. For all other operations the VFS set these flags
899 * explicitly if it wants a timestamp update.
901 if (newsize
!= oldsize
&&
902 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
903 iattr
->ia_ctime
= iattr
->ia_mtime
=
904 current_fs_time(inode
->i_sb
);
905 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
909 * The first thing we do is set the size to new_size permanently on
910 * disk. This way we don't have to worry about anyone ever being able
911 * to look at the data being freed even in the face of a crash.
912 * What we're getting around here is the case where we free a block, it
913 * is allocated to another file, it is written to, and then we crash.
914 * If the new data gets written to the file but the log buffers
915 * containing the free and reallocation don't, then we'd end up with
916 * garbage in the blocks being freed. As long as we make the new size
917 * permanent before actually freeing any blocks it doesn't matter if
918 * they get written to.
920 ip
->i_d
.di_size
= newsize
;
921 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
923 if (newsize
<= oldsize
) {
924 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
926 goto out_trans_abort
;
929 * Truncated "down", so we're removing references to old data
930 * here - if we delay flushing for a long time, we expose
931 * ourselves unduly to the notorious NULL files problem. So,
932 * we mark this inode and flush it when the file is closed,
933 * and do not wait the usual (long) time for writeout.
935 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
937 /* A truncate down always removes post-EOF blocks. */
938 xfs_inode_clear_eofblocks_tag(ip
);
941 if (iattr
->ia_valid
& ATTR_MODE
)
942 xfs_setattr_mode(ip
, iattr
);
943 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
944 xfs_setattr_time(ip
, iattr
);
946 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
948 XFS_STATS_INC(xs_ig_attrchg
);
950 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
951 xfs_trans_set_sync(tp
);
953 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
956 xfs_iunlock(ip
, lock_flags
);
960 commit_flags
|= XFS_TRANS_ABORT
;
962 xfs_trans_cancel(tp
, commit_flags
);
968 struct dentry
*dentry
,
971 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
974 if (iattr
->ia_valid
& ATTR_SIZE
) {
975 uint iolock
= XFS_IOLOCK_EXCL
;
977 xfs_ilock(ip
, iolock
);
978 error
= xfs_break_layouts(d_inode(dentry
), &iolock
);
980 error
= xfs_setattr_size(ip
, iattr
);
981 xfs_iunlock(ip
, iolock
);
983 error
= xfs_setattr_nonsize(ip
, iattr
, 0);
992 struct timespec
*now
,
995 struct xfs_inode
*ip
= XFS_I(inode
);
996 struct xfs_mount
*mp
= ip
->i_mount
;
997 struct xfs_trans
*tp
;
1000 trace_xfs_update_time(ip
);
1002 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
1003 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_fsyncts
, 0, 0);
1005 xfs_trans_cancel(tp
, 0);
1009 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1010 if (flags
& S_CTIME
) {
1011 inode
->i_ctime
= *now
;
1012 ip
->i_d
.di_ctime
.t_sec
= (__int32_t
)now
->tv_sec
;
1013 ip
->i_d
.di_ctime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1015 if (flags
& S_MTIME
) {
1016 inode
->i_mtime
= *now
;
1017 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)now
->tv_sec
;
1018 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1020 if (flags
& S_ATIME
) {
1021 inode
->i_atime
= *now
;
1022 ip
->i_d
.di_atime
.t_sec
= (__int32_t
)now
->tv_sec
;
1023 ip
->i_d
.di_atime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1025 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1026 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
1027 return xfs_trans_commit(tp
, 0);
1030 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1033 * Call fiemap helper to fill in user data.
1034 * Returns positive errors to xfs_getbmap.
1039 struct getbmapx
*bmv
,
1043 struct fiemap_extent_info
*fieinfo
= *arg
;
1044 u32 fiemap_flags
= 0;
1045 u64 logical
, physical
, length
;
1047 /* Do nothing for a hole */
1048 if (bmv
->bmv_block
== -1LL)
1051 logical
= BBTOB(bmv
->bmv_offset
);
1052 physical
= BBTOB(bmv
->bmv_block
);
1053 length
= BBTOB(bmv
->bmv_length
);
1055 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
1056 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
1057 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
1058 fiemap_flags
|= (FIEMAP_EXTENT_DELALLOC
|
1059 FIEMAP_EXTENT_UNKNOWN
);
1060 physical
= 0; /* no block yet */
1062 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
1063 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
1065 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
1066 length
, fiemap_flags
);
1069 *full
= 1; /* user array now full */
1077 struct inode
*inode
,
1078 struct fiemap_extent_info
*fieinfo
,
1082 xfs_inode_t
*ip
= XFS_I(inode
);
1086 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
1090 /* Set up bmap header for xfs internal routine */
1091 bm
.bmv_offset
= BTOBBT(start
);
1092 /* Special case for whole file */
1093 if (length
== FIEMAP_MAX_OFFSET
)
1094 bm
.bmv_length
= -1LL;
1096 bm
.bmv_length
= BTOBB(start
+ length
) - bm
.bmv_offset
;
1098 /* We add one because in getbmap world count includes the header */
1099 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
1100 fieinfo
->fi_extents_max
+ 1;
1101 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
1102 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
1103 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
1104 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
1105 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
1106 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
1107 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
1109 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
1119 struct dentry
*dentry
,
1122 return xfs_generic_create(dir
, dentry
, mode
, 0, true);
1125 static const struct inode_operations xfs_inode_operations
= {
1126 .get_acl
= xfs_get_acl
,
1127 .set_acl
= xfs_set_acl
,
1128 .getattr
= xfs_vn_getattr
,
1129 .setattr
= xfs_vn_setattr
,
1130 .setxattr
= generic_setxattr
,
1131 .getxattr
= generic_getxattr
,
1132 .removexattr
= generic_removexattr
,
1133 .listxattr
= xfs_vn_listxattr
,
1134 .fiemap
= xfs_vn_fiemap
,
1135 .update_time
= xfs_vn_update_time
,
1138 static const struct inode_operations xfs_dir_inode_operations
= {
1139 .create
= xfs_vn_create
,
1140 .lookup
= xfs_vn_lookup
,
1141 .link
= xfs_vn_link
,
1142 .unlink
= xfs_vn_unlink
,
1143 .symlink
= xfs_vn_symlink
,
1144 .mkdir
= xfs_vn_mkdir
,
1146 * Yes, XFS uses the same method for rmdir and unlink.
1148 * There are some subtile differences deeper in the code,
1149 * but we use S_ISDIR to check for those.
1151 .rmdir
= xfs_vn_unlink
,
1152 .mknod
= xfs_vn_mknod
,
1153 .rename2
= xfs_vn_rename
,
1154 .get_acl
= xfs_get_acl
,
1155 .set_acl
= xfs_set_acl
,
1156 .getattr
= xfs_vn_getattr
,
1157 .setattr
= xfs_vn_setattr
,
1158 .setxattr
= generic_setxattr
,
1159 .getxattr
= generic_getxattr
,
1160 .removexattr
= generic_removexattr
,
1161 .listxattr
= xfs_vn_listxattr
,
1162 .update_time
= xfs_vn_update_time
,
1163 .tmpfile
= xfs_vn_tmpfile
,
1166 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1167 .create
= xfs_vn_create
,
1168 .lookup
= xfs_vn_ci_lookup
,
1169 .link
= xfs_vn_link
,
1170 .unlink
= xfs_vn_unlink
,
1171 .symlink
= xfs_vn_symlink
,
1172 .mkdir
= xfs_vn_mkdir
,
1174 * Yes, XFS uses the same method for rmdir and unlink.
1176 * There are some subtile differences deeper in the code,
1177 * but we use S_ISDIR to check for those.
1179 .rmdir
= xfs_vn_unlink
,
1180 .mknod
= xfs_vn_mknod
,
1181 .rename2
= xfs_vn_rename
,
1182 .get_acl
= xfs_get_acl
,
1183 .set_acl
= xfs_set_acl
,
1184 .getattr
= xfs_vn_getattr
,
1185 .setattr
= xfs_vn_setattr
,
1186 .setxattr
= generic_setxattr
,
1187 .getxattr
= generic_getxattr
,
1188 .removexattr
= generic_removexattr
,
1189 .listxattr
= xfs_vn_listxattr
,
1190 .update_time
= xfs_vn_update_time
,
1191 .tmpfile
= xfs_vn_tmpfile
,
1194 static const struct inode_operations xfs_symlink_inode_operations
= {
1195 .readlink
= generic_readlink
,
1196 .follow_link
= xfs_vn_follow_link
,
1197 .put_link
= kfree_put_link
,
1198 .getattr
= xfs_vn_getattr
,
1199 .setattr
= xfs_vn_setattr
,
1200 .setxattr
= generic_setxattr
,
1201 .getxattr
= generic_getxattr
,
1202 .removexattr
= generic_removexattr
,
1203 .listxattr
= xfs_vn_listxattr
,
1204 .update_time
= xfs_vn_update_time
,
1208 xfs_diflags_to_iflags(
1209 struct inode
*inode
,
1210 struct xfs_inode
*ip
)
1212 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
1213 inode
->i_flags
|= S_IMMUTABLE
;
1215 inode
->i_flags
&= ~S_IMMUTABLE
;
1216 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
1217 inode
->i_flags
|= S_APPEND
;
1219 inode
->i_flags
&= ~S_APPEND
;
1220 if (ip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
)
1221 inode
->i_flags
|= S_SYNC
;
1223 inode
->i_flags
&= ~S_SYNC
;
1224 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
)
1225 inode
->i_flags
|= S_NOATIME
;
1227 inode
->i_flags
&= ~S_NOATIME
;
1231 * Initialize the Linux inode, set up the operation vectors and
1234 * When reading existing inodes from disk this is called directly
1235 * from xfs_iget, when creating a new inode it is called from
1236 * xfs_ialloc after setting up the inode.
1238 * We are always called with an uninitialised linux inode here.
1239 * We need to initialise the necessary fields and take a reference
1244 struct xfs_inode
*ip
)
1246 struct inode
*inode
= &ip
->i_vnode
;
1249 inode
->i_ino
= ip
->i_ino
;
1250 inode
->i_state
= I_NEW
;
1252 inode_sb_list_add(inode
);
1253 /* make the inode look hashed for the writeback code */
1254 hlist_add_fake(&inode
->i_hash
);
1256 inode
->i_mode
= ip
->i_d
.di_mode
;
1257 set_nlink(inode
, ip
->i_d
.di_nlink
);
1258 inode
->i_uid
= xfs_uid_to_kuid(ip
->i_d
.di_uid
);
1259 inode
->i_gid
= xfs_gid_to_kgid(ip
->i_d
.di_gid
);
1261 switch (inode
->i_mode
& S_IFMT
) {
1265 MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
1266 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
1273 inode
->i_generation
= ip
->i_d
.di_gen
;
1274 i_size_write(inode
, ip
->i_d
.di_size
);
1275 inode
->i_atime
.tv_sec
= ip
->i_d
.di_atime
.t_sec
;
1276 inode
->i_atime
.tv_nsec
= ip
->i_d
.di_atime
.t_nsec
;
1277 inode
->i_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
1278 inode
->i_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
1279 inode
->i_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
1280 inode
->i_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
1281 xfs_diflags_to_iflags(inode
, ip
);
1283 ip
->d_ops
= ip
->i_mount
->m_nondir_inode_ops
;
1284 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1285 switch (inode
->i_mode
& S_IFMT
) {
1287 inode
->i_op
= &xfs_inode_operations
;
1288 inode
->i_fop
= &xfs_file_operations
;
1289 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1292 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1293 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1294 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1296 inode
->i_op
= &xfs_dir_inode_operations
;
1297 inode
->i_fop
= &xfs_dir_file_operations
;
1298 ip
->d_ops
= ip
->i_mount
->m_dir_inode_ops
;
1301 inode
->i_op
= &xfs_symlink_inode_operations
;
1302 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1303 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1306 inode
->i_op
= &xfs_inode_operations
;
1307 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1312 * Ensure all page cache allocations are done from GFP_NOFS context to
1313 * prevent direct reclaim recursion back into the filesystem and blowing
1314 * stacks or deadlocking.
1316 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1317 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1320 * If there is no attribute fork no ACL can exist on this inode,
1321 * and it can't have any file capabilities attached to it either.
1323 if (!XFS_IFORK_Q(ip
)) {
1324 inode_has_no_xattr(inode
);
1325 cache_no_acl(inode
);
1328 xfs_iflags_clear(ip
, XFS_INEW
);
1331 unlock_new_inode(inode
);