1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
15 #include "xfs_quota.h"
17 #include "xfs_trans.h"
18 #include "xfs_trace.h"
19 #include "xfs_icache.h"
20 #include "xfs_symlink.h"
22 #include "xfs_iomap.h"
23 #include "xfs_error.h"
25 #include <linux/xattr.h>
26 #include <linux/posix_acl.h>
27 #include <linux/security.h>
28 #include <linux/iversion.h>
31 * Directories have different lock order w.r.t. mmap_sem compared to regular
32 * files. This is due to readdir potentially triggering page faults on a user
33 * buffer inside filldir(), and this happens with the ilock on the directory
34 * held. For regular files, the lock order is the other way around - the
35 * mmap_sem is taken during the page fault, and then we lock the ilock to do
36 * block mapping. Hence we need a different class for the directory ilock so
37 * that lockdep can tell them apart.
39 static struct lock_class_key xfs_nondir_ilock_class
;
40 static struct lock_class_key xfs_dir_ilock_class
;
45 const struct xattr
*xattr_array
,
48 const struct xattr
*xattr
;
49 struct xfs_inode
*ip
= XFS_I(inode
);
52 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
53 error
= xfs_attr_set(ip
, xattr
->name
, xattr
->value
,
54 xattr
->value_len
, ATTR_SECURE
);
62 * Hook in SELinux. This is not quite correct yet, what we really need
63 * here (as we do for default ACLs) is a mechanism by which creation of
64 * these attrs can be journalled at inode creation time (along with the
65 * inode, of course, such that log replay can't cause these to be lost).
72 const struct qstr
*qstr
)
74 return security_inode_init_security(inode
, dir
, qstr
,
75 &xfs_initxattrs
, NULL
);
80 struct xfs_name
*namep
,
81 struct dentry
*dentry
)
83 namep
->name
= dentry
->d_name
.name
;
84 namep
->len
= dentry
->d_name
.len
;
85 namep
->type
= XFS_DIR3_FT_UNKNOWN
;
89 xfs_dentry_mode_to_name(
90 struct xfs_name
*namep
,
91 struct dentry
*dentry
,
94 namep
->name
= dentry
->d_name
.name
;
95 namep
->len
= dentry
->d_name
.len
;
96 namep
->type
= xfs_mode_to_ftype(mode
);
98 if (unlikely(namep
->type
== XFS_DIR3_FT_UNKNOWN
))
108 struct dentry
*dentry
)
110 struct xfs_name teardown
;
113 * If we can't add the ACL or we fail in
114 * xfs_init_security we must back out.
115 * ENOSPC can hit here, among other things.
117 xfs_dentry_to_name(&teardown
, dentry
);
119 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
125 struct dentry
*dentry
,
128 bool tmpfile
) /* unnamed file */
131 struct xfs_inode
*ip
= NULL
;
132 struct posix_acl
*default_acl
, *acl
;
133 struct xfs_name name
;
137 * Irix uses Missed'em'V split, but doesn't want to see
138 * the upper 5 bits of (14bit) major.
140 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
141 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
147 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
151 /* Verify mode is valid also for tmpfile case */
152 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
157 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
159 error
= xfs_create_tmpfile(XFS_I(dir
), mode
, &ip
);
166 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
168 goto out_cleanup_inode
;
170 #ifdef CONFIG_XFS_POSIX_ACL
172 error
= __xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
174 goto out_cleanup_inode
;
177 error
= __xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
179 goto out_cleanup_inode
;
187 * The VFS requires that any inode fed to d_tmpfile must have
188 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
189 * However, we created the temp file with nlink == 0 because
190 * we're not allowed to put an inode with nlink > 0 on the
191 * unlinked list. Therefore we have to set nlink to 1 so that
192 * d_tmpfile can immediately set it back to zero.
195 d_tmpfile(dentry
, inode
);
197 d_instantiate(dentry
, inode
);
199 xfs_finish_inode_setup(ip
);
203 posix_acl_release(default_acl
);
205 posix_acl_release(acl
);
209 xfs_finish_inode_setup(ip
);
211 xfs_cleanup_inode(dir
, inode
, dentry
);
219 struct dentry
*dentry
,
223 return xfs_generic_create(dir
, dentry
, mode
, rdev
, false);
229 struct dentry
*dentry
,
233 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
239 struct dentry
*dentry
,
242 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
245 STATIC
struct dentry
*
248 struct dentry
*dentry
,
252 struct xfs_inode
*cip
;
253 struct xfs_name name
;
256 if (dentry
->d_name
.len
>= MAXNAMELEN
)
257 return ERR_PTR(-ENAMETOOLONG
);
259 xfs_dentry_to_name(&name
, dentry
);
260 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
263 else if (likely(error
== -ENOENT
))
266 inode
= ERR_PTR(error
);
267 return d_splice_alias(inode
, dentry
);
270 STATIC
struct dentry
*
273 struct dentry
*dentry
,
276 struct xfs_inode
*ip
;
277 struct xfs_name xname
;
278 struct xfs_name ci_name
;
282 if (dentry
->d_name
.len
>= MAXNAMELEN
)
283 return ERR_PTR(-ENAMETOOLONG
);
285 xfs_dentry_to_name(&xname
, dentry
);
286 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
287 if (unlikely(error
)) {
288 if (unlikely(error
!= -ENOENT
))
289 return ERR_PTR(error
);
291 * call d_add(dentry, NULL) here when d_drop_negative_children
292 * is called in xfs_vn_mknod (ie. allow negative dentries
293 * with CI filesystems).
298 /* if exact match, just splice and exit */
300 return d_splice_alias(VFS_I(ip
), dentry
);
302 /* else case-insensitive match... */
303 dname
.name
= ci_name
.name
;
304 dname
.len
= ci_name
.len
;
305 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
306 kmem_free(ci_name
.name
);
312 struct dentry
*old_dentry
,
314 struct dentry
*dentry
)
316 struct inode
*inode
= d_inode(old_dentry
);
317 struct xfs_name name
;
320 error
= xfs_dentry_mode_to_name(&name
, dentry
, inode
->i_mode
);
324 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
329 d_instantiate(dentry
, inode
);
336 struct dentry
*dentry
)
338 struct xfs_name name
;
341 xfs_dentry_to_name(&name
, dentry
);
343 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(d_inode(dentry
)));
348 * With unlink, the VFS makes the dentry "negative": no inode,
349 * but still hashed. This is incompatible with case-insensitive
350 * mode, so invalidate (unhash) the dentry in CI-mode.
352 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
353 d_invalidate(dentry
);
360 struct dentry
*dentry
,
364 struct xfs_inode
*cip
= NULL
;
365 struct xfs_name name
;
370 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
371 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
375 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
381 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
383 goto out_cleanup_inode
;
387 d_instantiate(dentry
, inode
);
388 xfs_finish_inode_setup(cip
);
392 xfs_finish_inode_setup(cip
);
393 xfs_cleanup_inode(dir
, inode
, dentry
);
402 struct dentry
*odentry
,
404 struct dentry
*ndentry
,
407 struct inode
*new_inode
= d_inode(ndentry
);
410 struct xfs_name oname
;
411 struct xfs_name nname
;
413 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
416 /* if we are exchanging files, we need to set i_mode of both files */
417 if (flags
& RENAME_EXCHANGE
)
418 omode
= d_inode(ndentry
)->i_mode
;
420 error
= xfs_dentry_mode_to_name(&oname
, odentry
, omode
);
421 if (omode
&& unlikely(error
))
424 error
= xfs_dentry_mode_to_name(&nname
, ndentry
,
425 d_inode(odentry
)->i_mode
);
429 return xfs_rename(XFS_I(odir
), &oname
, XFS_I(d_inode(odentry
)),
431 new_inode
? XFS_I(new_inode
) : NULL
, flags
);
435 * careful here - this function can get called recursively, so
436 * we need to be very careful about how much stack we use.
437 * uio is kmalloced for this reason...
441 struct dentry
*dentry
,
443 struct delayed_call
*done
)
449 return ERR_PTR(-ECHILD
);
451 link
= kmalloc(XFS_SYMLINK_MAXLEN
+1, GFP_KERNEL
);
455 error
= xfs_readlink(XFS_I(d_inode(dentry
)), link
);
459 set_delayed_call(done
, kfree_link
, link
);
465 return ERR_PTR(error
);
469 xfs_vn_get_link_inline(
470 struct dentry
*dentry
,
472 struct delayed_call
*done
)
474 struct xfs_inode
*ip
= XFS_I(inode
);
477 ASSERT(ip
->i_df
.if_flags
& XFS_IFINLINE
);
480 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
483 link
= ip
->i_df
.if_u1
.if_data
;
484 if (XFS_IS_CORRUPT(ip
->i_mount
, !link
))
485 return ERR_PTR(-EFSCORRUPTED
);
491 struct xfs_inode
*ip
)
493 struct xfs_mount
*mp
= ip
->i_mount
;
496 * If the file blocks are being allocated from a realtime volume, then
497 * always return the realtime extent size.
499 if (XFS_IS_REALTIME_INODE(ip
))
500 return xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
503 * Allow large block sizes to be reported to userspace programs if the
504 * "largeio" mount option is used.
506 * If compatibility mode is specified, simply return the basic unit of
507 * caching so that we don't get inefficient read/modify/write I/O from
508 * user apps. Otherwise....
510 * If the underlying volume is a stripe, then return the stripe width in
511 * bytes as the recommended I/O size. It is not a stripe and we've set a
512 * default buffered I/O size, return that, otherwise return the compat
515 if (mp
->m_flags
& XFS_MOUNT_LARGEIO
) {
517 return mp
->m_swidth
<< mp
->m_sb
.sb_blocklog
;
518 if (mp
->m_flags
& XFS_MOUNT_ALLOCSIZE
)
519 return 1U << mp
->m_allocsize_log
;
527 const struct path
*path
,
530 unsigned int query_flags
)
532 struct inode
*inode
= d_inode(path
->dentry
);
533 struct xfs_inode
*ip
= XFS_I(inode
);
534 struct xfs_mount
*mp
= ip
->i_mount
;
536 trace_xfs_getattr(ip
);
538 if (XFS_FORCED_SHUTDOWN(mp
))
541 stat
->size
= XFS_ISIZE(ip
);
542 stat
->dev
= inode
->i_sb
->s_dev
;
543 stat
->mode
= inode
->i_mode
;
544 stat
->nlink
= inode
->i_nlink
;
545 stat
->uid
= inode
->i_uid
;
546 stat
->gid
= inode
->i_gid
;
547 stat
->ino
= ip
->i_ino
;
548 stat
->atime
= inode
->i_atime
;
549 stat
->mtime
= inode
->i_mtime
;
550 stat
->ctime
= inode
->i_ctime
;
552 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
554 if (ip
->i_d
.di_version
== 3) {
555 if (request_mask
& STATX_BTIME
) {
556 stat
->result_mask
|= STATX_BTIME
;
557 stat
->btime
= ip
->i_d
.di_crtime
;
562 * Note: If you add another clause to set an attribute flag, please
563 * update attributes_mask below.
565 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
566 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
567 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
568 stat
->attributes
|= STATX_ATTR_APPEND
;
569 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
)
570 stat
->attributes
|= STATX_ATTR_NODUMP
;
572 stat
->attributes_mask
|= (STATX_ATTR_IMMUTABLE
|
576 switch (inode
->i_mode
& S_IFMT
) {
579 stat
->blksize
= BLKDEV_IOSIZE
;
580 stat
->rdev
= inode
->i_rdev
;
583 stat
->blksize
= xfs_stat_blksize(ip
);
593 struct xfs_inode
*ip
,
596 struct inode
*inode
= VFS_I(ip
);
597 umode_t mode
= iattr
->ia_mode
;
599 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
601 inode
->i_mode
&= S_IFMT
;
602 inode
->i_mode
|= mode
& ~S_IFMT
;
607 struct xfs_inode
*ip
,
610 struct inode
*inode
= VFS_I(ip
);
612 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
614 if (iattr
->ia_valid
& ATTR_ATIME
)
615 inode
->i_atime
= iattr
->ia_atime
;
616 if (iattr
->ia_valid
& ATTR_CTIME
)
617 inode
->i_ctime
= iattr
->ia_ctime
;
618 if (iattr
->ia_valid
& ATTR_MTIME
)
619 inode
->i_mtime
= iattr
->ia_mtime
;
624 struct dentry
*dentry
,
627 struct xfs_mount
*mp
= XFS_I(d_inode(dentry
))->i_mount
;
629 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
632 if (XFS_FORCED_SHUTDOWN(mp
))
635 return setattr_prepare(dentry
, iattr
);
639 * Set non-size attributes of an inode.
641 * Caution: The caller of this function is responsible for calling
642 * setattr_prepare() or otherwise verifying the change is fine.
646 struct xfs_inode
*ip
,
650 xfs_mount_t
*mp
= ip
->i_mount
;
651 struct inode
*inode
= VFS_I(ip
);
652 int mask
= iattr
->ia_valid
;
655 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
656 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
657 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
658 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
660 ASSERT((mask
& ATTR_SIZE
) == 0);
663 * If disk quotas is on, we make sure that the dquots do exist on disk,
664 * before we start any other transactions. Trying to do this later
665 * is messy. We don't care to take a readlock to look at the ids
666 * in inode here, because we can't hold it across the trans_reserve.
667 * If the IDs do change before we take the ilock, we're covered
668 * because the i_*dquot fields will get updated anyway.
670 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
673 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
675 qflags
|= XFS_QMOPT_UQUOTA
;
679 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
681 qflags
|= XFS_QMOPT_GQUOTA
;
687 * We take a reference when we initialize udqp and gdqp,
688 * so it is important that we never blindly double trip on
689 * the same variable. See xfs_create() for an example.
691 ASSERT(udqp
== NULL
);
692 ASSERT(gdqp
== NULL
);
693 error
= xfs_qm_vop_dqalloc(ip
, xfs_kuid_to_uid(uid
),
694 xfs_kgid_to_gid(gid
),
696 qflags
, &udqp
, &gdqp
, NULL
);
701 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
705 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
706 xfs_trans_ijoin(tp
, ip
, 0);
709 * Change file ownership. Must be the owner or privileged.
711 if (mask
& (ATTR_UID
|ATTR_GID
)) {
713 * These IDs could have changed since we last looked at them.
714 * But, we're assured that if the ownership did change
715 * while we didn't have the inode locked, inode's dquot(s)
716 * would have changed also.
720 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
721 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
724 * Do a quota reservation only if uid/gid is actually
727 if (XFS_IS_QUOTA_RUNNING(mp
) &&
728 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
729 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
731 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
732 NULL
, capable(CAP_FOWNER
) ?
733 XFS_QMOPT_FORCE_RES
: 0);
734 if (error
) /* out of quota */
740 * Change file ownership. Must be the owner or privileged.
742 if (mask
& (ATTR_UID
|ATTR_GID
)) {
744 * CAP_FSETID overrides the following restrictions:
746 * The set-user-ID and set-group-ID bits of a file will be
747 * cleared upon successful return from chown()
749 if ((inode
->i_mode
& (S_ISUID
|S_ISGID
)) &&
750 !capable(CAP_FSETID
))
751 inode
->i_mode
&= ~(S_ISUID
|S_ISGID
);
754 * Change the ownerships and register quota modifications
755 * in the transaction.
757 if (!uid_eq(iuid
, uid
)) {
758 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
759 ASSERT(mask
& ATTR_UID
);
761 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
762 &ip
->i_udquot
, udqp
);
764 ip
->i_d
.di_uid
= xfs_kuid_to_uid(uid
);
767 if (!gid_eq(igid
, gid
)) {
768 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
769 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
770 !XFS_IS_PQUOTA_ON(mp
));
771 ASSERT(mask
& ATTR_GID
);
773 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
774 &ip
->i_gdquot
, gdqp
);
776 ip
->i_d
.di_gid
= xfs_kgid_to_gid(gid
);
781 if (mask
& ATTR_MODE
)
782 xfs_setattr_mode(ip
, iattr
);
783 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
784 xfs_setattr_time(ip
, iattr
);
786 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
788 XFS_STATS_INC(mp
, xs_ig_attrchg
);
790 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
791 xfs_trans_set_sync(tp
);
792 error
= xfs_trans_commit(tp
);
794 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
797 * Release any dquot(s) the inode had kept before chown.
799 xfs_qm_dqrele(olddquot1
);
800 xfs_qm_dqrele(olddquot2
);
808 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
809 * update. We could avoid this with linked transactions
810 * and passing down the transaction pointer all the way
811 * to attr_set. No previous user of the generic
812 * Posix ACL code seems to care about this issue either.
814 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
815 error
= posix_acl_chmod(inode
, inode
->i_mode
);
823 xfs_trans_cancel(tp
);
824 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
832 xfs_vn_setattr_nonsize(
833 struct dentry
*dentry
,
836 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
839 trace_xfs_setattr(ip
);
841 error
= xfs_vn_change_ok(dentry
, iattr
);
844 return xfs_setattr_nonsize(ip
, iattr
, 0);
848 * Truncate file. Must have write permission and not be a directory.
850 * Caution: The caller of this function is responsible for calling
851 * setattr_prepare() or otherwise verifying the change is fine.
855 struct xfs_inode
*ip
,
858 struct xfs_mount
*mp
= ip
->i_mount
;
859 struct inode
*inode
= VFS_I(ip
);
860 xfs_off_t oldsize
, newsize
;
861 struct xfs_trans
*tp
;
864 bool did_zeroing
= false;
866 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
867 ASSERT(xfs_isilocked(ip
, XFS_MMAPLOCK_EXCL
));
868 ASSERT(S_ISREG(inode
->i_mode
));
869 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
870 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
872 oldsize
= inode
->i_size
;
873 newsize
= iattr
->ia_size
;
876 * Short circuit the truncate case for zero length files.
878 if (newsize
== 0 && oldsize
== 0 && ip
->i_d
.di_nextents
== 0) {
879 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
883 * Use the regular setattr path to update the timestamps.
885 iattr
->ia_valid
&= ~ATTR_SIZE
;
886 return xfs_setattr_nonsize(ip
, iattr
, 0);
890 * Make sure that the dquots are attached to the inode.
892 error
= xfs_qm_dqattach(ip
);
897 * Wait for all direct I/O to complete.
899 inode_dio_wait(inode
);
902 * File data changes must be complete before we start the transaction to
903 * modify the inode. This needs to be done before joining the inode to
904 * the transaction because the inode cannot be unlocked once it is a
905 * part of the transaction.
907 * Start with zeroing any data beyond EOF that we may expose on file
908 * extension, or zeroing out the rest of the block on a downward
911 if (newsize
> oldsize
) {
912 trace_xfs_zero_eof(ip
, oldsize
, newsize
- oldsize
);
913 error
= iomap_zero_range(inode
, oldsize
, newsize
- oldsize
,
914 &did_zeroing
, &xfs_buffered_write_iomap_ops
);
916 error
= iomap_truncate_page(inode
, newsize
, &did_zeroing
,
917 &xfs_buffered_write_iomap_ops
);
924 * We've already locked out new page faults, so now we can safely remove
925 * pages from the page cache knowing they won't get refaulted until we
926 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
927 * complete. The truncate_setsize() call also cleans partial EOF page
928 * PTEs on extending truncates and hence ensures sub-page block size
929 * filesystems are correctly handled, too.
931 * We have to do all the page cache truncate work outside the
932 * transaction context as the "lock" order is page lock->log space
933 * reservation as defined by extent allocation in the writeback path.
934 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
935 * having already truncated the in-memory version of the file (i.e. made
936 * user visible changes). There's not much we can do about this, except
937 * to hope that the caller sees ENOMEM and retries the truncate
940 * And we update in-core i_size and truncate page cache beyond newsize
941 * before writeback the [di_size, newsize] range, so we're guaranteed
942 * not to write stale data past the new EOF on truncate down.
944 truncate_setsize(inode
, newsize
);
947 * We are going to log the inode size change in this transaction so
948 * any previous writes that are beyond the on disk EOF and the new
949 * EOF that have not been written out need to be written here. If we
950 * do not write the data out, we expose ourselves to the null files
951 * problem. Note that this includes any block zeroing we did above;
952 * otherwise those blocks may not be zeroed after a crash.
955 (newsize
> ip
->i_d
.di_size
&& oldsize
!= ip
->i_d
.di_size
)) {
956 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
957 ip
->i_d
.di_size
, newsize
- 1);
962 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
966 lock_flags
|= XFS_ILOCK_EXCL
;
967 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
968 xfs_trans_ijoin(tp
, ip
, 0);
971 * Only change the c/mtime if we are changing the size or we are
972 * explicitly asked to change it. This handles the semantic difference
973 * between truncate() and ftruncate() as implemented in the VFS.
975 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
976 * special case where we need to update the times despite not having
977 * these flags set. For all other operations the VFS set these flags
978 * explicitly if it wants a timestamp update.
980 if (newsize
!= oldsize
&&
981 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
982 iattr
->ia_ctime
= iattr
->ia_mtime
=
984 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
988 * The first thing we do is set the size to new_size permanently on
989 * disk. This way we don't have to worry about anyone ever being able
990 * to look at the data being freed even in the face of a crash.
991 * What we're getting around here is the case where we free a block, it
992 * is allocated to another file, it is written to, and then we crash.
993 * If the new data gets written to the file but the log buffers
994 * containing the free and reallocation don't, then we'd end up with
995 * garbage in the blocks being freed. As long as we make the new size
996 * permanent before actually freeing any blocks it doesn't matter if
997 * they get written to.
999 ip
->i_d
.di_size
= newsize
;
1000 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1002 if (newsize
<= oldsize
) {
1003 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
1005 goto out_trans_cancel
;
1008 * Truncated "down", so we're removing references to old data
1009 * here - if we delay flushing for a long time, we expose
1010 * ourselves unduly to the notorious NULL files problem. So,
1011 * we mark this inode and flush it when the file is closed,
1012 * and do not wait the usual (long) time for writeout.
1014 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
1016 /* A truncate down always removes post-EOF blocks. */
1017 xfs_inode_clear_eofblocks_tag(ip
);
1020 if (iattr
->ia_valid
& ATTR_MODE
)
1021 xfs_setattr_mode(ip
, iattr
);
1022 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
1023 xfs_setattr_time(ip
, iattr
);
1025 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1027 XFS_STATS_INC(mp
, xs_ig_attrchg
);
1029 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1030 xfs_trans_set_sync(tp
);
1032 error
= xfs_trans_commit(tp
);
1035 xfs_iunlock(ip
, lock_flags
);
1039 xfs_trans_cancel(tp
);
1044 xfs_vn_setattr_size(
1045 struct dentry
*dentry
,
1046 struct iattr
*iattr
)
1048 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
1051 trace_xfs_setattr(ip
);
1053 error
= xfs_vn_change_ok(dentry
, iattr
);
1056 return xfs_setattr_size(ip
, iattr
);
1061 struct dentry
*dentry
,
1062 struct iattr
*iattr
)
1066 if (iattr
->ia_valid
& ATTR_SIZE
) {
1067 struct inode
*inode
= d_inode(dentry
);
1068 struct xfs_inode
*ip
= XFS_I(inode
);
1071 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1072 iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1074 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
1076 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1080 error
= xfs_vn_setattr_size(dentry
, iattr
);
1081 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1083 error
= xfs_vn_setattr_nonsize(dentry
, iattr
);
1091 struct inode
*inode
,
1092 struct timespec64
*now
,
1095 struct xfs_inode
*ip
= XFS_I(inode
);
1096 struct xfs_mount
*mp
= ip
->i_mount
;
1097 int log_flags
= XFS_ILOG_TIMESTAMP
;
1098 struct xfs_trans
*tp
;
1101 trace_xfs_update_time(ip
);
1103 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
) {
1104 if (!((flags
& S_VERSION
) &&
1105 inode_maybe_inc_iversion(inode
, false)))
1106 return generic_update_time(inode
, now
, flags
);
1108 /* Capture the iversion update that just occurred */
1109 log_flags
|= XFS_ILOG_CORE
;
1112 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_fsyncts
, 0, 0, 0, &tp
);
1116 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1117 if (flags
& S_CTIME
)
1118 inode
->i_ctime
= *now
;
1119 if (flags
& S_MTIME
)
1120 inode
->i_mtime
= *now
;
1121 if (flags
& S_ATIME
)
1122 inode
->i_atime
= *now
;
1124 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1125 xfs_trans_log_inode(tp
, ip
, log_flags
);
1126 return xfs_trans_commit(tp
);
1131 struct inode
*inode
,
1132 struct fiemap_extent_info
*fieinfo
,
1138 xfs_ilock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1139 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
1140 fieinfo
->fi_flags
&= ~FIEMAP_FLAG_XATTR
;
1141 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1142 &xfs_xattr_iomap_ops
);
1144 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1145 &xfs_read_iomap_ops
);
1147 xfs_iunlock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1155 struct dentry
*dentry
,
1158 return xfs_generic_create(dir
, dentry
, mode
, 0, true);
1161 static const struct inode_operations xfs_inode_operations
= {
1162 .get_acl
= xfs_get_acl
,
1163 .set_acl
= xfs_set_acl
,
1164 .getattr
= xfs_vn_getattr
,
1165 .setattr
= xfs_vn_setattr
,
1166 .listxattr
= xfs_vn_listxattr
,
1167 .fiemap
= xfs_vn_fiemap
,
1168 .update_time
= xfs_vn_update_time
,
1171 static const struct inode_operations xfs_dir_inode_operations
= {
1172 .create
= xfs_vn_create
,
1173 .lookup
= xfs_vn_lookup
,
1174 .link
= xfs_vn_link
,
1175 .unlink
= xfs_vn_unlink
,
1176 .symlink
= xfs_vn_symlink
,
1177 .mkdir
= xfs_vn_mkdir
,
1179 * Yes, XFS uses the same method for rmdir and unlink.
1181 * There are some subtile differences deeper in the code,
1182 * but we use S_ISDIR to check for those.
1184 .rmdir
= xfs_vn_unlink
,
1185 .mknod
= xfs_vn_mknod
,
1186 .rename
= xfs_vn_rename
,
1187 .get_acl
= xfs_get_acl
,
1188 .set_acl
= xfs_set_acl
,
1189 .getattr
= xfs_vn_getattr
,
1190 .setattr
= xfs_vn_setattr
,
1191 .listxattr
= xfs_vn_listxattr
,
1192 .update_time
= xfs_vn_update_time
,
1193 .tmpfile
= xfs_vn_tmpfile
,
1196 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1197 .create
= xfs_vn_create
,
1198 .lookup
= xfs_vn_ci_lookup
,
1199 .link
= xfs_vn_link
,
1200 .unlink
= xfs_vn_unlink
,
1201 .symlink
= xfs_vn_symlink
,
1202 .mkdir
= xfs_vn_mkdir
,
1204 * Yes, XFS uses the same method for rmdir and unlink.
1206 * There are some subtile differences deeper in the code,
1207 * but we use S_ISDIR to check for those.
1209 .rmdir
= xfs_vn_unlink
,
1210 .mknod
= xfs_vn_mknod
,
1211 .rename
= xfs_vn_rename
,
1212 .get_acl
= xfs_get_acl
,
1213 .set_acl
= xfs_set_acl
,
1214 .getattr
= xfs_vn_getattr
,
1215 .setattr
= xfs_vn_setattr
,
1216 .listxattr
= xfs_vn_listxattr
,
1217 .update_time
= xfs_vn_update_time
,
1218 .tmpfile
= xfs_vn_tmpfile
,
1221 static const struct inode_operations xfs_symlink_inode_operations
= {
1222 .get_link
= xfs_vn_get_link
,
1223 .getattr
= xfs_vn_getattr
,
1224 .setattr
= xfs_vn_setattr
,
1225 .listxattr
= xfs_vn_listxattr
,
1226 .update_time
= xfs_vn_update_time
,
1229 static const struct inode_operations xfs_inline_symlink_inode_operations
= {
1230 .get_link
= xfs_vn_get_link_inline
,
1231 .getattr
= xfs_vn_getattr
,
1232 .setattr
= xfs_vn_setattr
,
1233 .listxattr
= xfs_vn_listxattr
,
1234 .update_time
= xfs_vn_update_time
,
1237 /* Figure out if this file actually supports DAX. */
1239 xfs_inode_supports_dax(
1240 struct xfs_inode
*ip
)
1242 struct xfs_mount
*mp
= ip
->i_mount
;
1244 /* Only supported on non-reflinked files. */
1245 if (!S_ISREG(VFS_I(ip
)->i_mode
) || xfs_is_reflink_inode(ip
))
1248 /* DAX mount option or DAX iflag must be set. */
1249 if (!(mp
->m_flags
& XFS_MOUNT_DAX
) &&
1250 !(ip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
))
1253 /* Block size must match page size */
1254 if (mp
->m_sb
.sb_blocksize
!= PAGE_SIZE
)
1257 /* Device has to support DAX too. */
1258 return xfs_inode_buftarg(ip
)->bt_daxdev
!= NULL
;
1262 xfs_diflags_to_iflags(
1263 struct inode
*inode
,
1264 struct xfs_inode
*ip
)
1266 uint16_t flags
= ip
->i_d
.di_flags
;
1268 inode
->i_flags
&= ~(S_IMMUTABLE
| S_APPEND
| S_SYNC
|
1271 if (flags
& XFS_DIFLAG_IMMUTABLE
)
1272 inode
->i_flags
|= S_IMMUTABLE
;
1273 if (flags
& XFS_DIFLAG_APPEND
)
1274 inode
->i_flags
|= S_APPEND
;
1275 if (flags
& XFS_DIFLAG_SYNC
)
1276 inode
->i_flags
|= S_SYNC
;
1277 if (flags
& XFS_DIFLAG_NOATIME
)
1278 inode
->i_flags
|= S_NOATIME
;
1279 if (xfs_inode_supports_dax(ip
))
1280 inode
->i_flags
|= S_DAX
;
1284 * Initialize the Linux inode.
1286 * When reading existing inodes from disk this is called directly from xfs_iget,
1287 * when creating a new inode it is called from xfs_ialloc after setting up the
1288 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1289 * it up to the caller to deal with unlocking the inode appropriately.
1293 struct xfs_inode
*ip
)
1295 struct inode
*inode
= &ip
->i_vnode
;
1298 inode
->i_ino
= ip
->i_ino
;
1299 inode
->i_state
= I_NEW
;
1301 inode_sb_list_add(inode
);
1302 /* make the inode look hashed for the writeback code */
1303 inode_fake_hash(inode
);
1305 inode
->i_uid
= xfs_uid_to_kuid(ip
->i_d
.di_uid
);
1306 inode
->i_gid
= xfs_gid_to_kgid(ip
->i_d
.di_gid
);
1308 i_size_write(inode
, ip
->i_d
.di_size
);
1309 xfs_diflags_to_iflags(inode
, ip
);
1311 if (S_ISDIR(inode
->i_mode
)) {
1313 * We set the i_rwsem class here to avoid potential races with
1314 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1315 * after a filehandle lookup has already found the inode in
1316 * cache before it has been unlocked via unlock_new_inode().
1318 lockdep_set_class(&inode
->i_rwsem
,
1319 &inode
->i_sb
->s_type
->i_mutex_dir_key
);
1320 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1322 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1326 * Ensure all page cache allocations are done from GFP_NOFS context to
1327 * prevent direct reclaim recursion back into the filesystem and blowing
1328 * stacks or deadlocking.
1330 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1331 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1334 * If there is no attribute fork no ACL can exist on this inode,
1335 * and it can't have any file capabilities attached to it either.
1337 if (!XFS_IFORK_Q(ip
)) {
1338 inode_has_no_xattr(inode
);
1339 cache_no_acl(inode
);
1345 struct xfs_inode
*ip
)
1347 struct inode
*inode
= &ip
->i_vnode
;
1349 switch (inode
->i_mode
& S_IFMT
) {
1351 inode
->i_op
= &xfs_inode_operations
;
1352 inode
->i_fop
= &xfs_file_operations
;
1354 inode
->i_mapping
->a_ops
= &xfs_dax_aops
;
1356 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1359 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1360 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1362 inode
->i_op
= &xfs_dir_inode_operations
;
1363 inode
->i_fop
= &xfs_dir_file_operations
;
1366 if (ip
->i_df
.if_flags
& XFS_IFINLINE
)
1367 inode
->i_op
= &xfs_inline_symlink_inode_operations
;
1369 inode
->i_op
= &xfs_symlink_inode_operations
;
1372 inode
->i_op
= &xfs_inode_operations
;
1373 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);