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"
41 #include <linux/capability.h>
42 #include <linux/xattr.h>
43 #include <linux/namei.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
);
191 posix_acl_release(default_acl
);
193 posix_acl_release(acl
);
198 xfs_cleanup_inode(dir
, inode
, dentry
);
206 struct dentry
*dentry
,
210 return xfs_generic_create(dir
, dentry
, mode
, rdev
, false);
216 struct dentry
*dentry
,
220 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
226 struct dentry
*dentry
,
229 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
232 STATIC
struct dentry
*
235 struct dentry
*dentry
,
238 struct xfs_inode
*cip
;
239 struct xfs_name name
;
242 if (dentry
->d_name
.len
>= MAXNAMELEN
)
243 return ERR_PTR(-ENAMETOOLONG
);
245 xfs_dentry_to_name(&name
, dentry
, 0);
246 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
247 if (unlikely(error
)) {
248 if (unlikely(error
!= -ENOENT
))
249 return ERR_PTR(error
);
254 return d_splice_alias(VFS_I(cip
), dentry
);
257 STATIC
struct dentry
*
260 struct dentry
*dentry
,
263 struct xfs_inode
*ip
;
264 struct xfs_name xname
;
265 struct xfs_name ci_name
;
269 if (dentry
->d_name
.len
>= MAXNAMELEN
)
270 return ERR_PTR(-ENAMETOOLONG
);
272 xfs_dentry_to_name(&xname
, dentry
, 0);
273 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
274 if (unlikely(error
)) {
275 if (unlikely(error
!= -ENOENT
))
276 return ERR_PTR(error
);
278 * call d_add(dentry, NULL) here when d_drop_negative_children
279 * is called in xfs_vn_mknod (ie. allow negative dentries
280 * with CI filesystems).
285 /* if exact match, just splice and exit */
287 return d_splice_alias(VFS_I(ip
), dentry
);
289 /* else case-insensitive match... */
290 dname
.name
= ci_name
.name
;
291 dname
.len
= ci_name
.len
;
292 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
293 kmem_free(ci_name
.name
);
299 struct dentry
*old_dentry
,
301 struct dentry
*dentry
)
303 struct inode
*inode
= old_dentry
->d_inode
;
304 struct xfs_name name
;
307 xfs_dentry_to_name(&name
, dentry
, inode
->i_mode
);
309 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
314 d_instantiate(dentry
, inode
);
321 struct dentry
*dentry
)
323 struct xfs_name name
;
326 xfs_dentry_to_name(&name
, dentry
, 0);
328 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(dentry
->d_inode
));
333 * With unlink, the VFS makes the dentry "negative": no inode,
334 * but still hashed. This is incompatible with case-insensitive
335 * mode, so invalidate (unhash) the dentry in CI-mode.
337 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
338 d_invalidate(dentry
);
345 struct dentry
*dentry
,
349 struct xfs_inode
*cip
= NULL
;
350 struct xfs_name name
;
355 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
356 xfs_dentry_to_name(&name
, dentry
, mode
);
358 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
364 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
366 goto out_cleanup_inode
;
368 d_instantiate(dentry
, inode
);
372 xfs_cleanup_inode(dir
, inode
, dentry
);
381 struct dentry
*odentry
,
383 struct dentry
*ndentry
)
385 struct inode
*new_inode
= ndentry
->d_inode
;
386 struct xfs_name oname
;
387 struct xfs_name nname
;
389 xfs_dentry_to_name(&oname
, odentry
, 0);
390 xfs_dentry_to_name(&nname
, ndentry
, odentry
->d_inode
->i_mode
);
392 return xfs_rename(XFS_I(odir
), &oname
, XFS_I(odentry
->d_inode
),
393 XFS_I(ndir
), &nname
, new_inode
?
394 XFS_I(new_inode
) : NULL
);
398 * careful here - this function can get called recursively, so
399 * we need to be very careful about how much stack we use.
400 * uio is kmalloced for this reason...
404 struct dentry
*dentry
,
405 struct nameidata
*nd
)
410 link
= kmalloc(MAXPATHLEN
+1, GFP_KERNEL
);
414 error
= xfs_readlink(XFS_I(dentry
->d_inode
), link
);
418 nd_set_link(nd
, link
);
424 nd_set_link(nd
, ERR_PTR(error
));
430 struct vfsmount
*mnt
,
431 struct dentry
*dentry
,
434 struct inode
*inode
= dentry
->d_inode
;
435 struct xfs_inode
*ip
= XFS_I(inode
);
436 struct xfs_mount
*mp
= ip
->i_mount
;
438 trace_xfs_getattr(ip
);
440 if (XFS_FORCED_SHUTDOWN(mp
))
443 stat
->size
= XFS_ISIZE(ip
);
444 stat
->dev
= inode
->i_sb
->s_dev
;
445 stat
->mode
= ip
->i_d
.di_mode
;
446 stat
->nlink
= ip
->i_d
.di_nlink
;
447 stat
->uid
= inode
->i_uid
;
448 stat
->gid
= inode
->i_gid
;
449 stat
->ino
= ip
->i_ino
;
450 stat
->atime
= inode
->i_atime
;
451 stat
->mtime
= inode
->i_mtime
;
452 stat
->ctime
= inode
->i_ctime
;
454 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
457 switch (inode
->i_mode
& S_IFMT
) {
460 stat
->blksize
= BLKDEV_IOSIZE
;
461 stat
->rdev
= MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
462 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
465 if (XFS_IS_REALTIME_INODE(ip
)) {
467 * If the file blocks are being allocated from a
468 * realtime volume, then return the inode's realtime
469 * extent size or the realtime volume's extent size.
472 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
474 stat
->blksize
= xfs_preferred_iosize(mp
);
484 struct xfs_inode
*ip
,
487 struct inode
*inode
= VFS_I(ip
);
488 umode_t mode
= iattr
->ia_mode
;
490 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
492 ip
->i_d
.di_mode
&= S_IFMT
;
493 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
495 inode
->i_mode
&= S_IFMT
;
496 inode
->i_mode
|= mode
& ~S_IFMT
;
501 struct xfs_inode
*ip
,
504 struct inode
*inode
= VFS_I(ip
);
506 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
508 if (iattr
->ia_valid
& ATTR_ATIME
) {
509 inode
->i_atime
= iattr
->ia_atime
;
510 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
511 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
513 if (iattr
->ia_valid
& ATTR_CTIME
) {
514 inode
->i_ctime
= iattr
->ia_ctime
;
515 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
516 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
518 if (iattr
->ia_valid
& ATTR_MTIME
) {
519 inode
->i_mtime
= iattr
->ia_mtime
;
520 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
521 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
527 struct xfs_inode
*ip
,
531 xfs_mount_t
*mp
= ip
->i_mount
;
532 struct inode
*inode
= VFS_I(ip
);
533 int mask
= iattr
->ia_valid
;
536 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
537 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
538 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
539 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
541 trace_xfs_setattr(ip
);
543 /* If acls are being inherited, we already have this checked */
544 if (!(flags
& XFS_ATTR_NOACL
)) {
545 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
548 if (XFS_FORCED_SHUTDOWN(mp
))
551 error
= inode_change_ok(inode
, iattr
);
556 ASSERT((mask
& ATTR_SIZE
) == 0);
559 * If disk quotas is on, we make sure that the dquots do exist on disk,
560 * before we start any other transactions. Trying to do this later
561 * is messy. We don't care to take a readlock to look at the ids
562 * in inode here, because we can't hold it across the trans_reserve.
563 * If the IDs do change before we take the ilock, we're covered
564 * because the i_*dquot fields will get updated anyway.
566 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
569 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
571 qflags
|= XFS_QMOPT_UQUOTA
;
575 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
577 qflags
|= XFS_QMOPT_GQUOTA
;
583 * We take a reference when we initialize udqp and gdqp,
584 * so it is important that we never blindly double trip on
585 * the same variable. See xfs_create() for an example.
587 ASSERT(udqp
== NULL
);
588 ASSERT(gdqp
== NULL
);
589 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
590 xfs_kgid_to_gid(gid
),
592 qflags
, &udqp
, &gdqp
, NULL
);
597 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
598 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
602 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
605 * Change file ownership. Must be the owner or privileged.
607 if (mask
& (ATTR_UID
|ATTR_GID
)) {
609 * These IDs could have changed since we last looked at them.
610 * But, we're assured that if the ownership did change
611 * while we didn't have the inode locked, inode's dquot(s)
612 * would have changed also.
616 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
617 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
620 * Do a quota reservation only if uid/gid is actually
623 if (XFS_IS_QUOTA_RUNNING(mp
) &&
624 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
625 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
627 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
628 NULL
, capable(CAP_FOWNER
) ?
629 XFS_QMOPT_FORCE_RES
: 0);
630 if (error
) /* out of quota */
631 goto out_trans_cancel
;
635 xfs_trans_ijoin(tp
, ip
, 0);
638 * Change file ownership. Must be the owner or privileged.
640 if (mask
& (ATTR_UID
|ATTR_GID
)) {
642 * CAP_FSETID overrides the following restrictions:
644 * The set-user-ID and set-group-ID bits of a file will be
645 * cleared upon successful return from chown()
647 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
648 !capable(CAP_FSETID
))
649 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
652 * Change the ownerships and register quota modifications
653 * in the transaction.
655 if (!uid_eq(iuid
, uid
)) {
656 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
657 ASSERT(mask
& ATTR_UID
);
659 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
660 &ip
->i_udquot
, udqp
);
662 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
665 if (!gid_eq(igid
, gid
)) {
666 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
667 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
668 !XFS_IS_PQUOTA_ON(mp
));
669 ASSERT(mask
& ATTR_GID
);
671 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
672 &ip
->i_gdquot
, gdqp
);
674 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
679 if (mask
& ATTR_MODE
)
680 xfs_setattr_mode(ip
, iattr
);
681 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
682 xfs_setattr_time(ip
, iattr
);
684 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
686 XFS_STATS_INC(xs_ig_attrchg
);
688 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
689 xfs_trans_set_sync(tp
);
690 error
= xfs_trans_commit(tp
, 0);
692 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
695 * Release any dquot(s) the inode had kept before chown.
697 xfs_qm_dqrele(olddquot1
);
698 xfs_qm_dqrele(olddquot2
);
706 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
707 * update. We could avoid this with linked transactions
708 * and passing down the transaction pointer all the way
709 * to attr_set. No previous user of the generic
710 * Posix ACL code seems to care about this issue either.
712 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
713 error
= posix_acl_chmod(inode
, inode
->i_mode
);
721 xfs_trans_cancel(tp
, 0);
722 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
730 * Truncate file. Must have write permission and not be a directory.
734 struct xfs_inode
*ip
,
737 struct xfs_mount
*mp
= ip
->i_mount
;
738 struct inode
*inode
= VFS_I(ip
);
739 xfs_off_t oldsize
, newsize
;
740 struct xfs_trans
*tp
;
743 uint commit_flags
= 0;
745 trace_xfs_setattr(ip
);
747 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
750 if (XFS_FORCED_SHUTDOWN(mp
))
753 error
= inode_change_ok(inode
, iattr
);
757 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
758 ASSERT(S_ISREG(ip
->i_d
.di_mode
));
759 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
760 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
762 oldsize
= inode
->i_size
;
763 newsize
= iattr
->ia_size
;
766 * Short circuit the truncate case for zero length files.
768 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
769 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
773 * Use the regular setattr path to update the timestamps.
775 iattr
->ia_valid
&= ~ATTR_SIZE
;
776 return xfs_setattr_nonsize(ip
, iattr
, 0);
780 * Make sure that the dquots are attached to the inode.
782 error
= xfs_qm_dqattach(ip
, 0);
787 * Now we can make the changes. Before we join the inode to the
788 * transaction, take care of the part of the truncation that must be
789 * done without the inode lock. This needs to be done before joining
790 * the inode to the transaction, because the inode cannot be unlocked
791 * once it is a part of the transaction.
793 if (newsize
> oldsize
) {
795 * Do the first part of growing a file: zero any data in the
796 * last block that is beyond the old EOF. We need to do this
797 * before the inode is joined to the transaction to modify
800 error
= xfs_zero_eof(ip
, newsize
, oldsize
);
806 * We are going to log the inode size change in this transaction so
807 * any previous writes that are beyond the on disk EOF and the new
808 * EOF that have not been written out need to be written here. If we
809 * do not write the data out, we expose ourselves to the null files
812 * Only flush from the on disk size to the smaller of the in memory
813 * file size or the new size as that's the range we really care about
814 * here and prevents waiting for other data not within the range we
817 if (oldsize
!= ip
->i_d
.di_size
&& newsize
> ip
->i_d
.di_size
) {
818 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
819 ip
->i_d
.di_size
, newsize
);
825 * Wait for all direct I/O to complete.
827 inode_dio_wait(inode
);
830 * Do all the page cache truncate work outside the transaction context
831 * as the "lock" order is page lock->log space reservation. i.e.
832 * locking pages inside the transaction can ABBA deadlock with
833 * writeback. We have to do the VFS inode size update before we truncate
834 * the pagecache, however, to avoid racing with page faults beyond the
835 * new EOF they are not serialised against truncate operations except by
836 * page locks and size updates.
838 * Hence we are in a situation where a truncate can fail with ENOMEM
839 * from xfs_trans_reserve(), but having already truncated the in-memory
840 * version of the file (i.e. made user visible changes). There's not
841 * much we can do about this, except to hope that the caller sees ENOMEM
842 * and retries the truncate operation.
844 error
= block_truncate_page(inode
->i_mapping
, newsize
, xfs_get_blocks
);
847 truncate_setsize(inode
, newsize
);
850 * The "we can't serialise against page faults" pain gets worse.
852 * If the file is mapped then we have to clean the page at the old EOF
853 * when extending the file. Extending the file can expose changes the
854 * underlying page mapping (e.g. from beyond EOF to a hole or
855 * unwritten), and so on the next attempt to write to that page we need
856 * to remap it for write. i.e. we need .page_mkwrite() to be called.
857 * Hence we need to clean the page to clean the pte and so a new write
858 * fault will be triggered appropriately.
860 * If we do it before we change the inode size, then we can race with a
861 * page fault that maps the page with exactly the same problem. If we do
862 * it after we change the file size, then a new page fault can come in
863 * and allocate space before we've run the rest of the truncate
864 * transaction. That's kinda grotesque, but it's better than have data
865 * over a hole, and so that's the lesser evil that has been chosen here.
867 * The real solution, however, is to have some mechanism for locking out
868 * page faults while a truncate is in progress.
870 if (newsize
> oldsize
&& mapping_mapped(VFS_I(ip
)->i_mapping
)) {
871 error
= filemap_write_and_wait_range(
872 VFS_I(ip
)->i_mapping
,
873 round_down(oldsize
, PAGE_CACHE_SIZE
),
874 round_up(oldsize
, PAGE_CACHE_SIZE
) - 1);
879 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
880 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
882 goto out_trans_cancel
;
884 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
885 lock_flags
|= XFS_ILOCK_EXCL
;
886 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
887 xfs_trans_ijoin(tp
, ip
, 0);
890 * Only change the c/mtime if we are changing the size or we are
891 * explicitly asked to change it. This handles the semantic difference
892 * between truncate() and ftruncate() as implemented in the VFS.
894 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
895 * special case where we need to update the times despite not having
896 * these flags set. For all other operations the VFS set these flags
897 * explicitly if it wants a timestamp update.
899 if (newsize
!= oldsize
&&
900 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
901 iattr
->ia_ctime
= iattr
->ia_mtime
=
902 current_fs_time(inode
->i_sb
);
903 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
907 * The first thing we do is set the size to new_size permanently on
908 * disk. This way we don't have to worry about anyone ever being able
909 * to look at the data being freed even in the face of a crash.
910 * What we're getting around here is the case where we free a block, it
911 * is allocated to another file, it is written to, and then we crash.
912 * If the new data gets written to the file but the log buffers
913 * containing the free and reallocation don't, then we'd end up with
914 * garbage in the blocks being freed. As long as we make the new size
915 * permanent before actually freeing any blocks it doesn't matter if
916 * they get written to.
918 ip
->i_d
.di_size
= newsize
;
919 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
921 if (newsize
<= oldsize
) {
922 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
924 goto out_trans_abort
;
927 * Truncated "down", so we're removing references to old data
928 * here - if we delay flushing for a long time, we expose
929 * ourselves unduly to the notorious NULL files problem. So,
930 * we mark this inode and flush it when the file is closed,
931 * and do not wait the usual (long) time for writeout.
933 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
935 /* A truncate down always removes post-EOF blocks. */
936 xfs_inode_clear_eofblocks_tag(ip
);
939 if (iattr
->ia_valid
& ATTR_MODE
)
940 xfs_setattr_mode(ip
, iattr
);
941 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
942 xfs_setattr_time(ip
, iattr
);
944 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
946 XFS_STATS_INC(xs_ig_attrchg
);
948 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
949 xfs_trans_set_sync(tp
);
951 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
954 xfs_iunlock(ip
, lock_flags
);
958 commit_flags
|= XFS_TRANS_ABORT
;
960 xfs_trans_cancel(tp
, commit_flags
);
966 struct dentry
*dentry
,
969 struct xfs_inode
*ip
= XFS_I(dentry
->d_inode
);
972 if (iattr
->ia_valid
& ATTR_SIZE
) {
973 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
974 error
= xfs_setattr_size(ip
, iattr
);
975 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
977 error
= xfs_setattr_nonsize(ip
, iattr
, 0);
986 struct timespec
*now
,
989 struct xfs_inode
*ip
= XFS_I(inode
);
990 struct xfs_mount
*mp
= ip
->i_mount
;
991 struct xfs_trans
*tp
;
994 trace_xfs_update_time(ip
);
996 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
997 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_fsyncts
, 0, 0);
999 xfs_trans_cancel(tp
, 0);
1003 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1004 if (flags
& S_CTIME
) {
1005 inode
->i_ctime
= *now
;
1006 ip
->i_d
.di_ctime
.t_sec
= (__int32_t
)now
->tv_sec
;
1007 ip
->i_d
.di_ctime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1009 if (flags
& S_MTIME
) {
1010 inode
->i_mtime
= *now
;
1011 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)now
->tv_sec
;
1012 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1014 if (flags
& S_ATIME
) {
1015 inode
->i_atime
= *now
;
1016 ip
->i_d
.di_atime
.t_sec
= (__int32_t
)now
->tv_sec
;
1017 ip
->i_d
.di_atime
.t_nsec
= (__int32_t
)now
->tv_nsec
;
1019 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1020 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
1021 return xfs_trans_commit(tp
, 0);
1024 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1027 * Call fiemap helper to fill in user data.
1028 * Returns positive errors to xfs_getbmap.
1033 struct getbmapx
*bmv
,
1037 struct fiemap_extent_info
*fieinfo
= *arg
;
1038 u32 fiemap_flags
= 0;
1039 u64 logical
, physical
, length
;
1041 /* Do nothing for a hole */
1042 if (bmv
->bmv_block
== -1LL)
1045 logical
= BBTOB(bmv
->bmv_offset
);
1046 physical
= BBTOB(bmv
->bmv_block
);
1047 length
= BBTOB(bmv
->bmv_length
);
1049 if (bmv
->bmv_oflags
& BMV_OF_PREALLOC
)
1050 fiemap_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
1051 else if (bmv
->bmv_oflags
& BMV_OF_DELALLOC
) {
1052 fiemap_flags
|= (FIEMAP_EXTENT_DELALLOC
|
1053 FIEMAP_EXTENT_UNKNOWN
);
1054 physical
= 0; /* no block yet */
1056 if (bmv
->bmv_oflags
& BMV_OF_LAST
)
1057 fiemap_flags
|= FIEMAP_EXTENT_LAST
;
1059 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
1060 length
, fiemap_flags
);
1063 *full
= 1; /* user array now full */
1071 struct inode
*inode
,
1072 struct fiemap_extent_info
*fieinfo
,
1076 xfs_inode_t
*ip
= XFS_I(inode
);
1080 error
= fiemap_check_flags(fieinfo
, XFS_FIEMAP_FLAGS
);
1084 /* Set up bmap header for xfs internal routine */
1085 bm
.bmv_offset
= BTOBBT(start
);
1086 /* Special case for whole file */
1087 if (length
== FIEMAP_MAX_OFFSET
)
1088 bm
.bmv_length
= -1LL;
1090 bm
.bmv_length
= BTOBB(start
+ length
) - bm
.bmv_offset
;
1092 /* We add one because in getbmap world count includes the header */
1093 bm
.bmv_count
= !fieinfo
->fi_extents_max
? MAXEXTNUM
:
1094 fieinfo
->fi_extents_max
+ 1;
1095 bm
.bmv_count
= min_t(__s32
, bm
.bmv_count
,
1096 (PAGE_SIZE
* 16 / sizeof(struct getbmapx
)));
1097 bm
.bmv_iflags
= BMV_IF_PREALLOC
| BMV_IF_NO_HOLES
;
1098 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
)
1099 bm
.bmv_iflags
|= BMV_IF_ATTRFORK
;
1100 if (!(fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
))
1101 bm
.bmv_iflags
|= BMV_IF_DELALLOC
;
1103 error
= xfs_getbmap(ip
, &bm
, xfs_fiemap_format
, fieinfo
);
1113 struct dentry
*dentry
,
1116 return xfs_generic_create(dir
, dentry
, mode
, 0, true);
1119 static const struct inode_operations xfs_inode_operations
= {
1120 .get_acl
= xfs_get_acl
,
1121 .set_acl
= xfs_set_acl
,
1122 .getattr
= xfs_vn_getattr
,
1123 .setattr
= xfs_vn_setattr
,
1124 .setxattr
= generic_setxattr
,
1125 .getxattr
= generic_getxattr
,
1126 .removexattr
= generic_removexattr
,
1127 .listxattr
= xfs_vn_listxattr
,
1128 .fiemap
= xfs_vn_fiemap
,
1129 .update_time
= xfs_vn_update_time
,
1132 static const struct inode_operations xfs_dir_inode_operations
= {
1133 .create
= xfs_vn_create
,
1134 .lookup
= xfs_vn_lookup
,
1135 .link
= xfs_vn_link
,
1136 .unlink
= xfs_vn_unlink
,
1137 .symlink
= xfs_vn_symlink
,
1138 .mkdir
= xfs_vn_mkdir
,
1140 * Yes, XFS uses the same method for rmdir and unlink.
1142 * There are some subtile differences deeper in the code,
1143 * but we use S_ISDIR to check for those.
1145 .rmdir
= xfs_vn_unlink
,
1146 .mknod
= xfs_vn_mknod
,
1147 .rename
= xfs_vn_rename
,
1148 .get_acl
= xfs_get_acl
,
1149 .set_acl
= xfs_set_acl
,
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
,
1157 .tmpfile
= xfs_vn_tmpfile
,
1160 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1161 .create
= xfs_vn_create
,
1162 .lookup
= xfs_vn_ci_lookup
,
1163 .link
= xfs_vn_link
,
1164 .unlink
= xfs_vn_unlink
,
1165 .symlink
= xfs_vn_symlink
,
1166 .mkdir
= xfs_vn_mkdir
,
1168 * Yes, XFS uses the same method for rmdir and unlink.
1170 * There are some subtile differences deeper in the code,
1171 * but we use S_ISDIR to check for those.
1173 .rmdir
= xfs_vn_unlink
,
1174 .mknod
= xfs_vn_mknod
,
1175 .rename
= xfs_vn_rename
,
1176 .get_acl
= xfs_get_acl
,
1177 .set_acl
= xfs_set_acl
,
1178 .getattr
= xfs_vn_getattr
,
1179 .setattr
= xfs_vn_setattr
,
1180 .setxattr
= generic_setxattr
,
1181 .getxattr
= generic_getxattr
,
1182 .removexattr
= generic_removexattr
,
1183 .listxattr
= xfs_vn_listxattr
,
1184 .update_time
= xfs_vn_update_time
,
1185 .tmpfile
= xfs_vn_tmpfile
,
1188 static const struct inode_operations xfs_symlink_inode_operations
= {
1189 .readlink
= generic_readlink
,
1190 .follow_link
= xfs_vn_follow_link
,
1191 .put_link
= kfree_put_link
,
1192 .getattr
= xfs_vn_getattr
,
1193 .setattr
= xfs_vn_setattr
,
1194 .setxattr
= generic_setxattr
,
1195 .getxattr
= generic_getxattr
,
1196 .removexattr
= generic_removexattr
,
1197 .listxattr
= xfs_vn_listxattr
,
1198 .update_time
= xfs_vn_update_time
,
1202 xfs_diflags_to_iflags(
1203 struct inode
*inode
,
1204 struct xfs_inode
*ip
)
1206 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
1207 inode
->i_flags
|= S_IMMUTABLE
;
1209 inode
->i_flags
&= ~S_IMMUTABLE
;
1210 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
1211 inode
->i_flags
|= S_APPEND
;
1213 inode
->i_flags
&= ~S_APPEND
;
1214 if (ip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
)
1215 inode
->i_flags
|= S_SYNC
;
1217 inode
->i_flags
&= ~S_SYNC
;
1218 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
)
1219 inode
->i_flags
|= S_NOATIME
;
1221 inode
->i_flags
&= ~S_NOATIME
;
1225 * Initialize the Linux inode, set up the operation vectors and
1228 * When reading existing inodes from disk this is called directly
1229 * from xfs_iget, when creating a new inode it is called from
1230 * xfs_ialloc after setting up the inode.
1232 * We are always called with an uninitialised linux inode here.
1233 * We need to initialise the necessary fields and take a reference
1238 struct xfs_inode
*ip
)
1240 struct inode
*inode
= &ip
->i_vnode
;
1243 inode
->i_ino
= ip
->i_ino
;
1244 inode
->i_state
= I_NEW
;
1246 inode_sb_list_add(inode
);
1247 /* make the inode look hashed for the writeback code */
1248 hlist_add_fake(&inode
->i_hash
);
1250 inode
->i_mode
= ip
->i_d
.di_mode
;
1251 set_nlink(inode
, ip
->i_d
.di_nlink
);
1252 inode
->i_uid
= xfs_uid_to_kuid(ip
->i_d
.di_uid
);
1253 inode
->i_gid
= xfs_gid_to_kgid(ip
->i_d
.di_gid
);
1255 switch (inode
->i_mode
& S_IFMT
) {
1259 MKDEV(sysv_major(ip
->i_df
.if_u2
.if_rdev
) & 0x1ff,
1260 sysv_minor(ip
->i_df
.if_u2
.if_rdev
));
1267 inode
->i_generation
= ip
->i_d
.di_gen
;
1268 i_size_write(inode
, ip
->i_d
.di_size
);
1269 inode
->i_atime
.tv_sec
= ip
->i_d
.di_atime
.t_sec
;
1270 inode
->i_atime
.tv_nsec
= ip
->i_d
.di_atime
.t_nsec
;
1271 inode
->i_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
1272 inode
->i_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
1273 inode
->i_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
1274 inode
->i_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
1275 xfs_diflags_to_iflags(inode
, ip
);
1277 ip
->d_ops
= ip
->i_mount
->m_nondir_inode_ops
;
1278 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1279 switch (inode
->i_mode
& S_IFMT
) {
1281 inode
->i_op
= &xfs_inode_operations
;
1282 inode
->i_fop
= &xfs_file_operations
;
1283 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1286 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1287 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1288 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1290 inode
->i_op
= &xfs_dir_inode_operations
;
1291 inode
->i_fop
= &xfs_dir_file_operations
;
1292 ip
->d_ops
= ip
->i_mount
->m_dir_inode_ops
;
1295 inode
->i_op
= &xfs_symlink_inode_operations
;
1296 if (!(ip
->i_df
.if_flags
& XFS_IFINLINE
))
1297 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1300 inode
->i_op
= &xfs_inode_operations
;
1301 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
1306 * Ensure all page cache allocations are done from GFP_NOFS context to
1307 * prevent direct reclaim recursion back into the filesystem and blowing
1308 * stacks or deadlocking.
1310 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1311 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1314 * If there is no attribute fork no ACL can exist on this inode,
1315 * and it can't have any file capabilities attached to it either.
1317 if (!XFS_IFORK_Q(ip
)) {
1318 inode_has_no_xattr(inode
);
1319 cache_no_acl(inode
);
1322 xfs_iflags_clear(ip
, XFS_INEW
);
1325 unlock_new_inode(inode
);