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/posix_acl.h>
45 #include <linux/security.h>
46 #include <linux/fiemap.h>
47 #include <linux/slab.h>
50 * Directories have different lock order w.r.t. mmap_sem compared to regular
51 * files. This is due to readdir potentially triggering page faults on a user
52 * buffer inside filldir(), and this happens with the ilock on the directory
53 * held. For regular files, the lock order is the other way around - the
54 * mmap_sem is taken during the page fault, and then we lock the ilock to do
55 * block mapping. Hence we need a different class for the directory ilock so
56 * that lockdep can tell them apart.
58 static struct lock_class_key xfs_nondir_ilock_class
;
59 static struct lock_class_key xfs_dir_ilock_class
;
64 const struct xattr
*xattr_array
,
67 const struct xattr
*xattr
;
68 struct xfs_inode
*ip
= XFS_I(inode
);
71 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
72 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
73 xattr
->value_len
, ATTR_SECURE
);
81 * Hook in SELinux. This is not quite correct yet, what we really need
82 * here (as we do for default ACLs) is a mechanism by which creation of
83 * these attrs can be journalled at inode creation time (along with the
84 * inode, of course, such that log replay can't cause these to be lost).
91 const struct qstr
*qstr
)
93 return security_inode_init_security(inode
, dir
, qstr
,
94 &xfs_initxattrs
, NULL
);
99 struct xfs_name
*namep
,
100 struct dentry
*dentry
,
103 namep
->name
= dentry
->d_name
.name
;
104 namep
->len
= dentry
->d_name
.len
;
105 namep
->type
= xfs_mode_to_ftype
[(mode
& S_IFMT
) >> S_SHIFT
];
112 struct dentry
*dentry
)
114 struct xfs_name teardown
;
117 * If we can't add the ACL or we fail in
118 * xfs_init_security we must back out.
119 * ENOSPC can hit here, among other things.
121 xfs_dentry_to_name(&teardown
, dentry
, 0);
123 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
129 struct dentry
*dentry
,
132 bool tmpfile
) /* unnamed file */
135 struct xfs_inode
*ip
= NULL
;
136 struct posix_acl
*default_acl
, *acl
;
137 struct xfs_name name
;
141 * Irix uses Missed'em'V split, but doesn't want to see
142 * the upper 5 bits of (14bit) major.
144 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
145 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
147 rdev
= sysv_encode_dev(rdev
);
152 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
157 xfs_dentry_to_name(&name
, dentry
, mode
);
158 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
160 error
= xfs_create_tmpfile(XFS_I(dir
), dentry
, mode
, &ip
);
167 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
169 goto out_cleanup_inode
;
171 #ifdef CONFIG_XFS_POSIX_ACL
173 error
= xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
175 goto out_cleanup_inode
;
178 error
= xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
180 goto out_cleanup_inode
;
185 d_tmpfile(dentry
, inode
);
187 d_instantiate(dentry
, inode
);
189 xfs_finish_inode_setup(ip
);
193 posix_acl_release(default_acl
);
195 posix_acl_release(acl
);
199 xfs_finish_inode_setup(ip
);
201 xfs_cleanup_inode(dir
, inode
, dentry
);
209 struct dentry
*dentry
,
213 return xfs_generic_create(dir
, dentry
, mode
, rdev
, false);
219 struct dentry
*dentry
,
223 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
229 struct dentry
*dentry
,
232 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
235 STATIC
struct dentry
*
238 struct dentry
*dentry
,
241 struct xfs_inode
*cip
;
242 struct xfs_name name
;
245 if (dentry
->d_name
.len
>= MAXNAMELEN
)
246 return ERR_PTR(-ENAMETOOLONG
);
248 xfs_dentry_to_name(&name
, dentry
, 0);
249 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
250 if (unlikely(error
)) {
251 if (unlikely(error
!= -ENOENT
))
252 return ERR_PTR(error
);
257 return d_splice_alias(VFS_I(cip
), dentry
);
260 STATIC
struct dentry
*
263 struct dentry
*dentry
,
266 struct xfs_inode
*ip
;
267 struct xfs_name xname
;
268 struct xfs_name ci_name
;
272 if (dentry
->d_name
.len
>= MAXNAMELEN
)
273 return ERR_PTR(-ENAMETOOLONG
);
275 xfs_dentry_to_name(&xname
, dentry
, 0);
276 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
277 if (unlikely(error
)) {
278 if (unlikely(error
!= -ENOENT
))
279 return ERR_PTR(error
);
281 * call d_add(dentry, NULL) here when d_drop_negative_children
282 * is called in xfs_vn_mknod (ie. allow negative dentries
283 * with CI filesystems).
288 /* if exact match, just splice and exit */
290 return d_splice_alias(VFS_I(ip
), dentry
);
292 /* else case-insensitive match... */
293 dname
.name
= ci_name
.name
;
294 dname
.len
= ci_name
.len
;
295 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
296 kmem_free(ci_name
.name
);
302 struct dentry
*old_dentry
,
304 struct dentry
*dentry
)
306 struct inode
*inode
= d_inode(old_dentry
);
307 struct xfs_name name
;
310 xfs_dentry_to_name(&name
, dentry
, inode
->i_mode
);
312 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
317 d_instantiate(dentry
, inode
);
324 struct dentry
*dentry
)
326 struct xfs_name name
;
329 xfs_dentry_to_name(&name
, dentry
, 0);
331 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(d_inode(dentry
)));
336 * With unlink, the VFS makes the dentry "negative": no inode,
337 * but still hashed. This is incompatible with case-insensitive
338 * mode, so invalidate (unhash) the dentry in CI-mode.
340 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
341 d_invalidate(dentry
);
348 struct dentry
*dentry
,
352 struct xfs_inode
*cip
= NULL
;
353 struct xfs_name name
;
358 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
359 xfs_dentry_to_name(&name
, dentry
, mode
);
361 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
367 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
369 goto out_cleanup_inode
;
371 d_instantiate(dentry
, inode
);
372 xfs_finish_inode_setup(cip
);
376 xfs_finish_inode_setup(cip
);
377 xfs_cleanup_inode(dir
, inode
, dentry
);
386 struct dentry
*odentry
,
388 struct dentry
*ndentry
,
391 struct inode
*new_inode
= d_inode(ndentry
);
393 struct xfs_name oname
;
394 struct xfs_name nname
;
396 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
399 /* if we are exchanging files, we need to set i_mode of both files */
400 if (flags
& RENAME_EXCHANGE
)
401 omode
= d_inode(ndentry
)->i_mode
;
403 xfs_dentry_to_name(&oname
, odentry
, omode
);
404 xfs_dentry_to_name(&nname
, ndentry
, d_inode(odentry
)->i_mode
);
406 return xfs_rename(XFS_I(odir
), &oname
, XFS_I(d_inode(odentry
)),
408 new_inode
? XFS_I(new_inode
) : NULL
, flags
);
412 * careful here - this function can get called recursively, so
413 * we need to be very careful about how much stack we use.
414 * uio is kmalloced for this reason...
418 struct dentry
*dentry
,
420 struct delayed_call
*done
)
426 return ERR_PTR(-ECHILD
);
428 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
432 error
= xfs_readlink(XFS_I(d_inode(dentry
)), link
);
436 set_delayed_call(done
, kfree_link
, link
);
442 return ERR_PTR(error
);
447 struct vfsmount
*mnt
,
448 struct dentry
*dentry
,
451 struct inode
*inode
= d_inode(dentry
);
452 struct xfs_inode
*ip
= XFS_I(inode
);
453 struct xfs_mount
*mp
= ip
->i_mount
;
455 trace_xfs_getattr(ip
);
457 if (XFS_FORCED_SHUTDOWN(mp
))
460 stat
->size
= XFS_ISIZE(ip
);
461 stat
->dev
= inode
->i_sb
->s_dev
;
462 stat
->mode
= inode
->i_mode
;
463 stat
->nlink
= inode
->i_nlink
;
464 stat
->uid
= inode
->i_uid
;
465 stat
->gid
= inode
->i_gid
;
466 stat
->ino
= ip
->i_ino
;
467 stat
->atime
= inode
->i_atime
;
468 stat
->mtime
= inode
->i_mtime
;
469 stat
->ctime
= inode
->i_ctime
;
471 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
474 switch (inode
->i_mode
& S_IFMT
) {
477 stat
->blksize
= BLKDEV_IOSIZE
;
478 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
479 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
482 if (XFS_IS_REALTIME_INODE(ip
)) {
484 * If the file blocks are being allocated from a
485 * realtime volume, then return the inode's realtime
486 * extent size or the realtime volume's extent size.
489 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
491 stat
->blksize
= xfs_preferred_iosize(mp
);
501 struct xfs_inode
*ip
,
504 struct inode
*inode
= VFS_I(ip
);
505 umode_t mode
= iattr
->ia_mode
;
507 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
509 inode
->i_mode
&= S_IFMT
;
510 inode
->i_mode
|= mode
& ~S_IFMT
;
515 struct xfs_inode
*ip
,
518 struct inode
*inode
= VFS_I(ip
);
520 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
522 if (iattr
->ia_valid
& ATTR_ATIME
)
523 inode
->i_atime
= iattr
->ia_atime
;
524 if (iattr
->ia_valid
& ATTR_CTIME
)
525 inode
->i_ctime
= iattr
->ia_ctime
;
526 if (iattr
->ia_valid
& ATTR_MTIME
)
527 inode
->i_mtime
= iattr
->ia_mtime
;
532 struct xfs_inode
*ip
,
536 xfs_mount_t
*mp
= ip
->i_mount
;
537 struct inode
*inode
= VFS_I(ip
);
538 int mask
= iattr
->ia_valid
;
541 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
542 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
543 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
544 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
546 trace_xfs_setattr(ip
);
548 /* If acls are being inherited, we already have this checked */
549 if (!(flags
& XFS_ATTR_NOACL
)) {
550 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
553 if (XFS_FORCED_SHUTDOWN(mp
))
556 error
= inode_change_ok(inode
, iattr
);
561 ASSERT((mask
& ATTR_SIZE
) == 0);
564 * If disk quotas is on, we make sure that the dquots do exist on disk,
565 * before we start any other transactions. Trying to do this later
566 * is messy. We don't care to take a readlock to look at the ids
567 * in inode here, because we can't hold it across the trans_reserve.
568 * If the IDs do change before we take the ilock, we're covered
569 * because the i_*dquot fields will get updated anyway.
571 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
574 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
576 qflags
|= XFS_QMOPT_UQUOTA
;
580 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
582 qflags
|= XFS_QMOPT_GQUOTA
;
588 * We take a reference when we initialize udqp and gdqp,
589 * so it is important that we never blindly double trip on
590 * the same variable. See xfs_create() for an example.
592 ASSERT(udqp
== NULL
);
593 ASSERT(gdqp
== NULL
);
594 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
595 xfs_kgid_to_gid(gid
),
597 qflags
, &udqp
, &gdqp
, NULL
);
602 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
606 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
607 xfs_trans_ijoin(tp
, ip
, 0);
610 * Change file ownership. Must be the owner or privileged.
612 if (mask
& (ATTR_UID
|ATTR_GID
)) {
614 * These IDs could have changed since we last looked at them.
615 * But, we're assured that if the ownership did change
616 * while we didn't have the inode locked, inode's dquot(s)
617 * would have changed also.
621 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
622 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
625 * Do a quota reservation only if uid/gid is actually
628 if (XFS_IS_QUOTA_RUNNING(mp
) &&
629 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
630 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
632 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
633 NULL
, capable(CAP_FOWNER
) ?
634 XFS_QMOPT_FORCE_RES
: 0);
635 if (error
) /* out of quota */
641 * Change file ownership. Must be the owner or privileged.
643 if (mask
& (ATTR_UID
|ATTR_GID
)) {
645 * CAP_FSETID overrides the following restrictions:
647 * The set-user-ID and set-group-ID bits of a file will be
648 * cleared upon successful return from chown()
650 if ((inode
->i_mode
& (S_ISUID
|S_ISGID
)) &&
651 !capable(CAP_FSETID
))
652 inode
->i_mode
&= ~(S_ISUID
|S_ISGID
);
655 * Change the ownerships and register quota modifications
656 * in the transaction.
658 if (!uid_eq(iuid
, uid
)) {
659 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
660 ASSERT(mask
& ATTR_UID
);
662 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
663 &ip
->i_udquot
, udqp
);
665 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
668 if (!gid_eq(igid
, gid
)) {
669 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
670 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
671 !XFS_IS_PQUOTA_ON(mp
));
672 ASSERT(mask
& ATTR_GID
);
674 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
675 &ip
->i_gdquot
, gdqp
);
677 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
682 if (mask
& ATTR_MODE
)
683 xfs_setattr_mode(ip
, iattr
);
684 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
685 xfs_setattr_time(ip
, iattr
);
687 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
689 XFS_STATS_INC(mp
, xs_ig_attrchg
);
691 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
692 xfs_trans_set_sync(tp
);
693 error
= xfs_trans_commit(tp
);
695 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
698 * Release any dquot(s) the inode had kept before chown.
700 xfs_qm_dqrele(olddquot1
);
701 xfs_qm_dqrele(olddquot2
);
709 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
710 * update. We could avoid this with linked transactions
711 * and passing down the transaction pointer all the way
712 * to attr_set. No previous user of the generic
713 * Posix ACL code seems to care about this issue either.
715 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
716 error
= posix_acl_chmod(inode
, inode
->i_mode
);
724 xfs_trans_cancel(tp
);
732 * Truncate file. Must have write permission and not be a directory.
736 struct xfs_inode
*ip
,
739 struct xfs_mount
*mp
= ip
->i_mount
;
740 struct inode
*inode
= VFS_I(ip
);
741 xfs_off_t oldsize
, newsize
;
742 struct xfs_trans
*tp
;
745 bool did_zeroing
= false;
747 trace_xfs_setattr(ip
);
749 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
752 if (XFS_FORCED_SHUTDOWN(mp
))
755 error
= inode_change_ok(inode
, iattr
);
759 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
760 ASSERT(xfs_isilocked(ip
, XFS_MMAPLOCK_EXCL
));
761 ASSERT(S_ISREG(inode
->i_mode
));
762 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
763 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
765 oldsize
= inode
->i_size
;
766 newsize
= iattr
->ia_size
;
769 * Short circuit the truncate case for zero length files.
771 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
772 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
776 * Use the regular setattr path to update the timestamps.
778 iattr
->ia_valid
&= ~ATTR_SIZE
;
779 return xfs_setattr_nonsize(ip
, iattr
, 0);
783 * Make sure that the dquots are attached to the inode.
785 error
= xfs_qm_dqattach(ip
, 0);
790 * File data changes must be complete before we start the transaction to
791 * modify the inode. This needs to be done before joining the inode to
792 * the transaction because the inode cannot be unlocked once it is a
793 * part of the transaction.
795 * Start with zeroing any data block beyond EOF that we may expose on
798 if (newsize
> oldsize
) {
799 error
= xfs_zero_eof(ip
, newsize
, oldsize
, &did_zeroing
);
805 * We are going to log the inode size change in this transaction so
806 * any previous writes that are beyond the on disk EOF and the new
807 * EOF that have not been written out need to be written here. If we
808 * do not write the data out, we expose ourselves to the null files
809 * problem. Note that this includes any block zeroing we did above;
810 * otherwise those blocks may not be zeroed after a crash.
812 if (newsize
> ip
->i_d
.di_size
&&
813 (oldsize
!= ip
->i_d
.di_size
|| did_zeroing
)) {
814 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
815 ip
->i_d
.di_size
, newsize
);
820 /* Now wait for all direct I/O to complete. */
821 inode_dio_wait(inode
);
824 * We've already locked out new page faults, so now we can safely remove
825 * pages from the page cache knowing they won't get refaulted until we
826 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
827 * complete. The truncate_setsize() call also cleans partial EOF page
828 * PTEs on extending truncates and hence ensures sub-page block size
829 * filesystems are correctly handled, too.
831 * We have to do all the page cache truncate work outside the
832 * transaction context as the "lock" order is page lock->log space
833 * reservation as defined by extent allocation in the writeback path.
834 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
835 * having already truncated the in-memory version of the file (i.e. made
836 * user visible changes). There's not much we can do about this, except
837 * to hope that the caller sees ENOMEM and retries the truncate
841 error
= dax_truncate_page(inode
, newsize
, xfs_get_blocks_direct
);
843 error
= block_truncate_page(inode
->i_mapping
, newsize
,
847 truncate_setsize(inode
, newsize
);
849 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
853 lock_flags
|= XFS_ILOCK_EXCL
;
854 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
855 xfs_trans_ijoin(tp
, ip
, 0);
858 * Only change the c/mtime if we are changing the size or we are
859 * explicitly asked to change it. This handles the semantic difference
860 * between truncate() and ftruncate() as implemented in the VFS.
862 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
863 * special case where we need to update the times despite not having
864 * these flags set. For all other operations the VFS set these flags
865 * explicitly if it wants a timestamp update.
867 if (newsize
!= oldsize
&&
868 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
869 iattr
->ia_ctime
= iattr
->ia_mtime
=
870 current_fs_time(inode
->i_sb
);
871 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
875 * The first thing we do is set the size to new_size permanently on
876 * disk. This way we don't have to worry about anyone ever being able
877 * to look at the data being freed even in the face of a crash.
878 * What we're getting around here is the case where we free a block, it
879 * is allocated to another file, it is written to, and then we crash.
880 * If the new data gets written to the file but the log buffers
881 * containing the free and reallocation don't, then we'd end up with
882 * garbage in the blocks being freed. As long as we make the new size
883 * permanent before actually freeing any blocks it doesn't matter if
884 * they get written to.
886 ip
->i_d
.di_size
= newsize
;
887 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
889 if (newsize
<= oldsize
) {
890 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
892 goto out_trans_cancel
;
895 * Truncated "down", so we're removing references to old data
896 * here - if we delay flushing for a long time, we expose
897 * ourselves unduly to the notorious NULL files problem. So,
898 * we mark this inode and flush it when the file is closed,
899 * and do not wait the usual (long) time for writeout.
901 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
903 /* A truncate down always removes post-EOF blocks. */
904 xfs_inode_clear_eofblocks_tag(ip
);
907 if (iattr
->ia_valid
& ATTR_MODE
)
908 xfs_setattr_mode(ip
, iattr
);
909 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
910 xfs_setattr_time(ip
, iattr
);
912 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
914 XFS_STATS_INC(mp
, xs_ig_attrchg
);
916 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
917 xfs_trans_set_sync(tp
);
919 error
= xfs_trans_commit(tp
);
922 xfs_iunlock(ip
, lock_flags
);
926 xfs_trans_cancel(tp
);
932 struct dentry
*dentry
,
935 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
938 if (iattr
->ia_valid
& ATTR_SIZE
) {
939 uint iolock
= XFS_IOLOCK_EXCL
;
941 xfs_ilock(ip
, iolock
);
942 error
= xfs_break_layouts(d_inode(dentry
), &iolock
, true);
944 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
945 iolock
|= XFS_MMAPLOCK_EXCL
;
947 error
= xfs_setattr_size(ip
, iattr
);
949 xfs_iunlock(ip
, iolock
);
951 error
= xfs_setattr_nonsize(ip
, iattr
, 0);
960 struct timespec
*now
,
963 struct xfs_inode
*ip
= XFS_I(inode
);
964 struct xfs_mount
*mp
= ip
->i_mount
;
965 struct xfs_trans
*tp
;
968 trace_xfs_update_time(ip
);
970 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_fsyncts
, 0, 0, 0, &tp
);
974 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
976 inode
->i_ctime
= *now
;
978 inode
->i_mtime
= *now
;
980 inode
->i_atime
= *now
;
982 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
983 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
984 return xfs_trans_commit(tp
);
987 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
990 * Call fiemap helper to fill in user data.
991 * Returns positive errors to xfs_getbmap.
996 struct getbmapx
*bmv
,
1000 struct fiemap_extent_info
*fieinfo
= *arg
;
1001 u32 fiemap_flags
= 0;
1002 u64 logical
, physical
, length
;
1004 /* Do nothing for a hole */
1005 if (bmv
->bmv_block
== -1LL)
1008 logical
= BBTOB(bmv
->bmv_offset
);
1009 physical
= BBTOB(bmv
->bmv_block
);
1010 length
= BBTOB(bmv
->bmv_length
);
1012 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
1013 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
1014 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
1015 fiemap_flags
|= (FIEMAP_EXTENT_DELALLOC
|
1016 FIEMAP_EXTENT_UNKNOWN
);
1017 physical
= 0; /* no block yet */
1019 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
1020 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
1022 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
1023 length
, fiemap_flags
);
1026 *full
= 1; /* user array now full */
1034 struct inode
*inode
,
1035 struct fiemap_extent_info
*fieinfo
,
1039 xfs_inode_t
*ip
= XFS_I(inode
);
1043 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
1047 /* Set up bmap header for xfs internal routine */
1048 bm
.bmv_offset
= BTOBBT(start
);
1049 /* Special case for whole file */
1050 if (length
== FIEMAP_MAX_OFFSET
)
1051 bm
.bmv_length
= -1LL;
1053 bm
.bmv_length
= BTOBB(start
+ length
) - bm
.bmv_offset
;
1055 /* We add one because in getbmap world count includes the header */
1056 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
1057 fieinfo
->fi_extents_max
+ 1;
1058 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
1059 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
1060 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
1061 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
1062 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
1063 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
1064 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
1066 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
1076 struct dentry
*dentry
,
1079 return xfs_generic_create(dir
, dentry
, mode
, 0, true);
1082 static const struct inode_operations xfs_inode_operations
= {
1083 .get_acl
= xfs_get_acl
,
1084 .set_acl
= xfs_set_acl
,
1085 .getattr
= xfs_vn_getattr
,
1086 .setattr
= xfs_vn_setattr
,
1087 .setxattr
= generic_setxattr
,
1088 .getxattr
= generic_getxattr
,
1089 .removexattr
= generic_removexattr
,
1090 .listxattr
= xfs_vn_listxattr
,
1091 .fiemap
= xfs_vn_fiemap
,
1092 .update_time
= xfs_vn_update_time
,
1095 static const struct inode_operations xfs_dir_inode_operations
= {
1096 .create
= xfs_vn_create
,
1097 .lookup
= xfs_vn_lookup
,
1098 .link
= xfs_vn_link
,
1099 .unlink
= xfs_vn_unlink
,
1100 .symlink
= xfs_vn_symlink
,
1101 .mkdir
= xfs_vn_mkdir
,
1103 * Yes, XFS uses the same method for rmdir and unlink.
1105 * There are some subtile differences deeper in the code,
1106 * but we use S_ISDIR to check for those.
1108 .rmdir
= xfs_vn_unlink
,
1109 .mknod
= xfs_vn_mknod
,
1110 .rename2
= xfs_vn_rename
,
1111 .get_acl
= xfs_get_acl
,
1112 .set_acl
= xfs_set_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
,
1120 .tmpfile
= xfs_vn_tmpfile
,
1123 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1124 .create
= xfs_vn_create
,
1125 .lookup
= xfs_vn_ci_lookup
,
1126 .link
= xfs_vn_link
,
1127 .unlink
= xfs_vn_unlink
,
1128 .symlink
= xfs_vn_symlink
,
1129 .mkdir
= xfs_vn_mkdir
,
1131 * Yes, XFS uses the same method for rmdir and unlink.
1133 * There are some subtile differences deeper in the code,
1134 * but we use S_ISDIR to check for those.
1136 .rmdir
= xfs_vn_unlink
,
1137 .mknod
= xfs_vn_mknod
,
1138 .rename2
= xfs_vn_rename
,
1139 .get_acl
= xfs_get_acl
,
1140 .set_acl
= xfs_set_acl
,
1141 .getattr
= xfs_vn_getattr
,
1142 .setattr
= xfs_vn_setattr
,
1143 .setxattr
= generic_setxattr
,
1144 .getxattr
= generic_getxattr
,
1145 .removexattr
= generic_removexattr
,
1146 .listxattr
= xfs_vn_listxattr
,
1147 .update_time
= xfs_vn_update_time
,
1148 .tmpfile
= xfs_vn_tmpfile
,
1151 static const struct inode_operations xfs_symlink_inode_operations
= {
1152 .readlink
= generic_readlink
,
1153 .get_link
= xfs_vn_get_link
,
1154 .getattr
= xfs_vn_getattr
,
1155 .setattr
= xfs_vn_setattr
,
1156 .setxattr
= generic_setxattr
,
1157 .getxattr
= generic_getxattr
,
1158 .removexattr
= generic_removexattr
,
1159 .listxattr
= xfs_vn_listxattr
,
1160 .update_time
= xfs_vn_update_time
,
1164 xfs_diflags_to_iflags(
1165 struct inode
*inode
,
1166 struct xfs_inode
*ip
)
1168 uint16_t flags
= ip
->i_d
.di_flags
;
1170 inode
->i_flags
&= ~(S_IMMUTABLE
| S_APPEND
| S_SYNC
|
1173 if (flags
& XFS_DIFLAG_IMMUTABLE
)
1174 inode
->i_flags
|= S_IMMUTABLE
;
1175 if (flags
& XFS_DIFLAG_APPEND
)
1176 inode
->i_flags
|= S_APPEND
;
1177 if (flags
& XFS_DIFLAG_SYNC
)
1178 inode
->i_flags
|= S_SYNC
;
1179 if (flags
& XFS_DIFLAG_NOATIME
)
1180 inode
->i_flags
|= S_NOATIME
;
1181 if (S_ISREG(inode
->i_mode
) &&
1182 ip
->i_mount
->m_sb
.sb_blocksize
== PAGE_SIZE
&&
1183 (ip
->i_mount
->m_flags
& XFS_MOUNT_DAX
||
1184 ip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
))
1185 inode
->i_flags
|= S_DAX
;
1189 * Initialize the Linux inode and set up the operation vectors.
1191 * When reading existing inodes from disk this is called directly from xfs_iget,
1192 * when creating a new inode it is called from xfs_ialloc after setting up the
1193 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1194 * it up to the caller to deal with unlocking the inode appropriately.
1198 struct xfs_inode
*ip
)
1200 struct inode
*inode
= &ip
->i_vnode
;
1203 inode
->i_ino
= ip
->i_ino
;
1204 inode
->i_state
= I_NEW
;
1206 inode_sb_list_add(inode
);
1207 /* make the inode look hashed for the writeback code */
1208 hlist_add_fake(&inode
->i_hash
);
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 i_size_write(inode
, ip
->i_d
.di_size
);
1226 xfs_diflags_to_iflags(inode
, ip
);
1228 ip
->d_ops
= ip
->i_mount
->m_nondir_inode_ops
;
1229 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1230 switch (inode
->i_mode
& S_IFMT
) {
1232 inode
->i_op
= &xfs_inode_operations
;
1233 inode
->i_fop
= &xfs_file_operations
;
1234 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1237 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1238 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1239 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1241 inode
->i_op
= &xfs_dir_inode_operations
;
1242 inode
->i_fop
= &xfs_dir_file_operations
;
1243 ip
->d_ops
= ip
->i_mount
->m_dir_inode_ops
;
1246 inode
->i_op
= &xfs_symlink_inode_operations
;
1247 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1248 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1251 inode
->i_op
= &xfs_inode_operations
;
1252 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1257 * Ensure all page cache allocations are done from GFP_NOFS context to
1258 * prevent direct reclaim recursion back into the filesystem and blowing
1259 * stacks or deadlocking.
1261 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1262 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1265 * If there is no attribute fork no ACL can exist on this inode,
1266 * and it can't have any file capabilities attached to it either.
1268 if (!XFS_IFORK_Q(ip
)) {
1269 inode_has_no_xattr(inode
);
1270 cache_no_acl(inode
);