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/posix_acl.h>
26 #include <linux/security.h>
27 #include <linux/iversion.h>
28 #include <linux/fiemap.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 struct xfs_da_args args
= {
55 .attr_filter
= XFS_ATTR_SECURE
,
57 .namelen
= strlen(xattr
->name
),
58 .value
= xattr
->value
,
59 .valuelen
= xattr
->value_len
,
61 error
= xfs_attr_set(&args
);
69 * Hook in SELinux. This is not quite correct yet, what we really need
70 * here (as we do for default ACLs) is a mechanism by which creation of
71 * these attrs can be journalled at inode creation time (along with the
72 * inode, of course, such that log replay can't cause these to be lost).
79 const struct qstr
*qstr
)
81 return security_inode_init_security(inode
, dir
, qstr
,
82 &xfs_initxattrs
, NULL
);
87 struct xfs_name
*namep
,
88 struct dentry
*dentry
)
90 namep
->name
= dentry
->d_name
.name
;
91 namep
->len
= dentry
->d_name
.len
;
92 namep
->type
= XFS_DIR3_FT_UNKNOWN
;
96 xfs_dentry_mode_to_name(
97 struct xfs_name
*namep
,
98 struct dentry
*dentry
,
101 namep
->name
= dentry
->d_name
.name
;
102 namep
->len
= dentry
->d_name
.len
;
103 namep
->type
= xfs_mode_to_ftype(mode
);
105 if (unlikely(namep
->type
== XFS_DIR3_FT_UNKNOWN
))
106 return -EFSCORRUPTED
;
115 struct dentry
*dentry
)
117 struct xfs_name teardown
;
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
124 xfs_dentry_to_name(&teardown
, dentry
);
126 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
132 struct dentry
*dentry
,
135 bool tmpfile
) /* unnamed file */
138 struct xfs_inode
*ip
= NULL
;
139 struct posix_acl
*default_acl
, *acl
;
140 struct xfs_name name
;
144 * Irix uses Missed'em'V split, but doesn't want to see
145 * the upper 5 bits of (14bit) major.
147 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
148 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
154 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
158 /* Verify mode is valid also for tmpfile case */
159 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
164 error
= xfs_create(XFS_I(dir
), &name
, mode
, rdev
, &ip
);
166 error
= xfs_create_tmpfile(XFS_I(dir
), mode
, &ip
);
173 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
175 goto out_cleanup_inode
;
177 #ifdef CONFIG_XFS_POSIX_ACL
179 error
= __xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
181 goto out_cleanup_inode
;
184 error
= __xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
186 goto out_cleanup_inode
;
194 * The VFS requires that any inode fed to d_tmpfile must have
195 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
196 * However, we created the temp file with nlink == 0 because
197 * we're not allowed to put an inode with nlink > 0 on the
198 * unlinked list. Therefore we have to set nlink to 1 so that
199 * d_tmpfile can immediately set it back to zero.
202 d_tmpfile(dentry
, inode
);
204 d_instantiate(dentry
, inode
);
206 xfs_finish_inode_setup(ip
);
210 posix_acl_release(default_acl
);
212 posix_acl_release(acl
);
216 xfs_finish_inode_setup(ip
);
218 xfs_cleanup_inode(dir
, inode
, dentry
);
226 struct dentry
*dentry
,
230 return xfs_generic_create(dir
, dentry
, mode
, rdev
, false);
236 struct dentry
*dentry
,
240 return xfs_vn_mknod(dir
, dentry
, mode
, 0);
246 struct dentry
*dentry
,
249 return xfs_vn_mknod(dir
, dentry
, mode
|S_IFDIR
, 0);
252 STATIC
struct dentry
*
255 struct dentry
*dentry
,
259 struct xfs_inode
*cip
;
260 struct xfs_name name
;
263 if (dentry
->d_name
.len
>= MAXNAMELEN
)
264 return ERR_PTR(-ENAMETOOLONG
);
266 xfs_dentry_to_name(&name
, dentry
);
267 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
270 else if (likely(error
== -ENOENT
))
273 inode
= ERR_PTR(error
);
274 return d_splice_alias(inode
, dentry
);
277 STATIC
struct dentry
*
280 struct dentry
*dentry
,
283 struct xfs_inode
*ip
;
284 struct xfs_name xname
;
285 struct xfs_name ci_name
;
289 if (dentry
->d_name
.len
>= MAXNAMELEN
)
290 return ERR_PTR(-ENAMETOOLONG
);
292 xfs_dentry_to_name(&xname
, dentry
);
293 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
294 if (unlikely(error
)) {
295 if (unlikely(error
!= -ENOENT
))
296 return ERR_PTR(error
);
298 * call d_add(dentry, NULL) here when d_drop_negative_children
299 * is called in xfs_vn_mknod (ie. allow negative dentries
300 * with CI filesystems).
305 /* if exact match, just splice and exit */
307 return d_splice_alias(VFS_I(ip
), dentry
);
309 /* else case-insensitive match... */
310 dname
.name
= ci_name
.name
;
311 dname
.len
= ci_name
.len
;
312 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
313 kmem_free(ci_name
.name
);
319 struct dentry
*old_dentry
,
321 struct dentry
*dentry
)
323 struct inode
*inode
= d_inode(old_dentry
);
324 struct xfs_name name
;
327 error
= xfs_dentry_mode_to_name(&name
, dentry
, inode
->i_mode
);
331 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
336 d_instantiate(dentry
, inode
);
343 struct dentry
*dentry
)
345 struct xfs_name name
;
348 xfs_dentry_to_name(&name
, dentry
);
350 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(d_inode(dentry
)));
355 * With unlink, the VFS makes the dentry "negative": no inode,
356 * but still hashed. This is incompatible with case-insensitive
357 * mode, so invalidate (unhash) the dentry in CI-mode.
359 if (xfs_sb_version_hasasciici(&XFS_M(dir
->i_sb
)->m_sb
))
360 d_invalidate(dentry
);
367 struct dentry
*dentry
,
371 struct xfs_inode
*cip
= NULL
;
372 struct xfs_name name
;
377 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
378 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
382 error
= xfs_symlink(XFS_I(dir
), &name
, symname
, mode
, &cip
);
388 error
= xfs_init_security(inode
, dir
, &dentry
->d_name
);
390 goto out_cleanup_inode
;
394 d_instantiate(dentry
, inode
);
395 xfs_finish_inode_setup(cip
);
399 xfs_finish_inode_setup(cip
);
400 xfs_cleanup_inode(dir
, inode
, dentry
);
409 struct dentry
*odentry
,
411 struct dentry
*ndentry
,
414 struct inode
*new_inode
= d_inode(ndentry
);
417 struct xfs_name oname
;
418 struct xfs_name nname
;
420 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
423 /* if we are exchanging files, we need to set i_mode of both files */
424 if (flags
& RENAME_EXCHANGE
)
425 omode
= d_inode(ndentry
)->i_mode
;
427 error
= xfs_dentry_mode_to_name(&oname
, odentry
, omode
);
428 if (omode
&& unlikely(error
))
431 error
= xfs_dentry_mode_to_name(&nname
, ndentry
,
432 d_inode(odentry
)->i_mode
);
436 return xfs_rename(XFS_I(odir
), &oname
, XFS_I(d_inode(odentry
)),
438 new_inode
? XFS_I(new_inode
) : NULL
, flags
);
442 * careful here - this function can get called recursively, so
443 * we need to be very careful about how much stack we use.
444 * uio is kmalloced for this reason...
448 struct dentry
*dentry
,
450 struct delayed_call
*done
)
456 return ERR_PTR(-ECHILD
);
458 link
= kmalloc(XFS_SYMLINK_MAXLEN
+1, GFP_KERNEL
);
462 error
= xfs_readlink(XFS_I(d_inode(dentry
)), link
);
466 set_delayed_call(done
, kfree_link
, link
);
472 return ERR_PTR(error
);
476 xfs_vn_get_link_inline(
477 struct dentry
*dentry
,
479 struct delayed_call
*done
)
481 struct xfs_inode
*ip
= XFS_I(inode
);
484 ASSERT(ip
->i_df
.if_flags
& XFS_IFINLINE
);
487 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
490 link
= ip
->i_df
.if_u1
.if_data
;
491 if (XFS_IS_CORRUPT(ip
->i_mount
, !link
))
492 return ERR_PTR(-EFSCORRUPTED
);
498 struct xfs_inode
*ip
)
500 struct xfs_mount
*mp
= ip
->i_mount
;
503 * If the file blocks are being allocated from a realtime volume, then
504 * always return the realtime extent size.
506 if (XFS_IS_REALTIME_INODE(ip
))
507 return xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
510 * Allow large block sizes to be reported to userspace programs if the
511 * "largeio" mount option is used.
513 * If compatibility mode is specified, simply return the basic unit of
514 * caching so that we don't get inefficient read/modify/write I/O from
515 * user apps. Otherwise....
517 * If the underlying volume is a stripe, then return the stripe width in
518 * bytes as the recommended I/O size. It is not a stripe and we've set a
519 * default buffered I/O size, return that, otherwise return the compat
522 if (mp
->m_flags
& XFS_MOUNT_LARGEIO
) {
524 return mp
->m_swidth
<< mp
->m_sb
.sb_blocklog
;
525 if (mp
->m_flags
& XFS_MOUNT_ALLOCSIZE
)
526 return 1U << mp
->m_allocsize_log
;
534 const struct path
*path
,
537 unsigned int query_flags
)
539 struct inode
*inode
= d_inode(path
->dentry
);
540 struct xfs_inode
*ip
= XFS_I(inode
);
541 struct xfs_mount
*mp
= ip
->i_mount
;
543 trace_xfs_getattr(ip
);
545 if (XFS_FORCED_SHUTDOWN(mp
))
548 stat
->size
= XFS_ISIZE(ip
);
549 stat
->dev
= inode
->i_sb
->s_dev
;
550 stat
->mode
= inode
->i_mode
;
551 stat
->nlink
= inode
->i_nlink
;
552 stat
->uid
= inode
->i_uid
;
553 stat
->gid
= inode
->i_gid
;
554 stat
->ino
= ip
->i_ino
;
555 stat
->atime
= inode
->i_atime
;
556 stat
->mtime
= inode
->i_mtime
;
557 stat
->ctime
= inode
->i_ctime
;
559 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
561 if (xfs_sb_version_has_v3inode(&mp
->m_sb
)) {
562 if (request_mask
& STATX_BTIME
) {
563 stat
->result_mask
|= STATX_BTIME
;
564 stat
->btime
= ip
->i_d
.di_crtime
;
569 * Note: If you add another clause to set an attribute flag, please
570 * update attributes_mask below.
572 if (ip
->i_d
.di_flags
& XFS_DIFLAG_IMMUTABLE
)
573 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
574 if (ip
->i_d
.di_flags
& XFS_DIFLAG_APPEND
)
575 stat
->attributes
|= STATX_ATTR_APPEND
;
576 if (ip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
)
577 stat
->attributes
|= STATX_ATTR_NODUMP
;
579 stat
->attributes_mask
|= (STATX_ATTR_IMMUTABLE
|
583 switch (inode
->i_mode
& S_IFMT
) {
586 stat
->blksize
= BLKDEV_IOSIZE
;
587 stat
->rdev
= inode
->i_rdev
;
590 stat
->blksize
= xfs_stat_blksize(ip
);
600 struct xfs_inode
*ip
,
603 struct inode
*inode
= VFS_I(ip
);
604 umode_t mode
= iattr
->ia_mode
;
606 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
608 inode
->i_mode
&= S_IFMT
;
609 inode
->i_mode
|= mode
& ~S_IFMT
;
614 struct xfs_inode
*ip
,
617 struct inode
*inode
= VFS_I(ip
);
619 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
621 if (iattr
->ia_valid
& ATTR_ATIME
)
622 inode
->i_atime
= iattr
->ia_atime
;
623 if (iattr
->ia_valid
& ATTR_CTIME
)
624 inode
->i_ctime
= iattr
->ia_ctime
;
625 if (iattr
->ia_valid
& ATTR_MTIME
)
626 inode
->i_mtime
= iattr
->ia_mtime
;
631 struct dentry
*dentry
,
634 struct xfs_mount
*mp
= XFS_I(d_inode(dentry
))->i_mount
;
636 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
639 if (XFS_FORCED_SHUTDOWN(mp
))
642 return setattr_prepare(dentry
, iattr
);
646 * Set non-size attributes of an inode.
648 * Caution: The caller of this function is responsible for calling
649 * setattr_prepare() or otherwise verifying the change is fine.
653 struct xfs_inode
*ip
,
657 xfs_mount_t
*mp
= ip
->i_mount
;
658 struct inode
*inode
= VFS_I(ip
);
659 int mask
= iattr
->ia_valid
;
662 kuid_t uid
= GLOBAL_ROOT_UID
, iuid
= GLOBAL_ROOT_UID
;
663 kgid_t gid
= GLOBAL_ROOT_GID
, igid
= GLOBAL_ROOT_GID
;
664 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
665 struct xfs_dquot
*olddquot1
= NULL
, *olddquot2
= NULL
;
667 ASSERT((mask
& ATTR_SIZE
) == 0);
670 * If disk quotas is on, we make sure that the dquots do exist on disk,
671 * before we start any other transactions. Trying to do this later
672 * is messy. We don't care to take a readlock to look at the ids
673 * in inode here, because we can't hold it across the trans_reserve.
674 * If the IDs do change before we take the ilock, we're covered
675 * because the i_*dquot fields will get updated anyway.
677 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
680 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
682 qflags
|= XFS_QMOPT_UQUOTA
;
686 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
688 qflags
|= XFS_QMOPT_GQUOTA
;
694 * We take a reference when we initialize udqp and gdqp,
695 * so it is important that we never blindly double trip on
696 * the same variable. See xfs_create() for an example.
698 ASSERT(udqp
== NULL
);
699 ASSERT(gdqp
== NULL
);
700 error
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, ip
->i_d
.di_projid
,
701 qflags
, &udqp
, &gdqp
, NULL
);
706 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
710 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
711 xfs_trans_ijoin(tp
, ip
, 0);
714 * Change file ownership. Must be the owner or privileged.
716 if (mask
& (ATTR_UID
|ATTR_GID
)) {
718 * These IDs could have changed since we last looked at them.
719 * But, we're assured that if the ownership did change
720 * while we didn't have the inode locked, inode's dquot(s)
721 * would have changed also.
725 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
726 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
729 * Do a quota reservation only if uid/gid is actually
732 if (XFS_IS_QUOTA_RUNNING(mp
) &&
733 ((XFS_IS_UQUOTA_ON(mp
) && !uid_eq(iuid
, uid
)) ||
734 (XFS_IS_GQUOTA_ON(mp
) && !gid_eq(igid
, gid
)))) {
736 error
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
737 NULL
, capable(CAP_FOWNER
) ?
738 XFS_QMOPT_FORCE_RES
: 0);
739 if (error
) /* out of quota */
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
);
766 if (!gid_eq(igid
, gid
)) {
767 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
768 ASSERT(xfs_sb_version_has_pquotino(&mp
->m_sb
) ||
769 !XFS_IS_PQUOTA_ON(mp
));
770 ASSERT(mask
& ATTR_GID
);
772 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
773 &ip
->i_gdquot
, gdqp
);
779 if (mask
& ATTR_MODE
)
780 xfs_setattr_mode(ip
, iattr
);
781 if (mask
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
782 xfs_setattr_time(ip
, iattr
);
784 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
786 XFS_STATS_INC(mp
, xs_ig_attrchg
);
788 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
789 xfs_trans_set_sync(tp
);
790 error
= xfs_trans_commit(tp
);
792 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
795 * Release any dquot(s) the inode had kept before chown.
797 xfs_qm_dqrele(olddquot1
);
798 xfs_qm_dqrele(olddquot2
);
806 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
807 * update. We could avoid this with linked transactions
808 * and passing down the transaction pointer all the way
809 * to attr_set. No previous user of the generic
810 * Posix ACL code seems to care about this issue either.
812 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
813 error
= posix_acl_chmod(inode
, inode
->i_mode
);
821 xfs_trans_cancel(tp
);
822 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
830 xfs_vn_setattr_nonsize(
831 struct dentry
*dentry
,
834 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
837 trace_xfs_setattr(ip
);
839 error
= xfs_vn_change_ok(dentry
, iattr
);
842 return xfs_setattr_nonsize(ip
, iattr
, 0);
846 * Truncate file. Must have write permission and not be a directory.
848 * Caution: The caller of this function is responsible for calling
849 * setattr_prepare() or otherwise verifying the change is fine.
853 struct xfs_inode
*ip
,
856 struct xfs_mount
*mp
= ip
->i_mount
;
857 struct inode
*inode
= VFS_I(ip
);
858 xfs_off_t oldsize
, newsize
;
859 struct xfs_trans
*tp
;
862 bool did_zeroing
= false;
864 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
865 ASSERT(xfs_isilocked(ip
, XFS_MMAPLOCK_EXCL
));
866 ASSERT(S_ISREG(inode
->i_mode
));
867 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
868 ATTR_MTIME_SET
|ATTR_KILL_PRIV
|ATTR_TIMES_SET
)) == 0);
870 oldsize
= inode
->i_size
;
871 newsize
= iattr
->ia_size
;
874 * Short circuit the truncate case for zero length files.
876 if (newsize
== 0 && oldsize
== 0 && ip
->i_df
.if_nextents
== 0) {
877 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
881 * Use the regular setattr path to update the timestamps.
883 iattr
->ia_valid
&= ~ATTR_SIZE
;
884 return xfs_setattr_nonsize(ip
, iattr
, 0);
888 * Make sure that the dquots are attached to the inode.
890 error
= xfs_qm_dqattach(ip
);
895 * Wait for all direct I/O to complete.
897 inode_dio_wait(inode
);
900 * File data changes must be complete before we start the transaction to
901 * modify the inode. This needs to be done before joining the inode to
902 * the transaction because the inode cannot be unlocked once it is a
903 * part of the transaction.
905 * Start with zeroing any data beyond EOF that we may expose on file
906 * extension, or zeroing out the rest of the block on a downward
909 if (newsize
> oldsize
) {
910 trace_xfs_zero_eof(ip
, oldsize
, newsize
- oldsize
);
911 error
= iomap_zero_range(inode
, oldsize
, newsize
- oldsize
,
912 &did_zeroing
, &xfs_buffered_write_iomap_ops
);
914 error
= iomap_truncate_page(inode
, newsize
, &did_zeroing
,
915 &xfs_buffered_write_iomap_ops
);
922 * We've already locked out new page faults, so now we can safely remove
923 * pages from the page cache knowing they won't get refaulted until we
924 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
925 * complete. The truncate_setsize() call also cleans partial EOF page
926 * PTEs on extending truncates and hence ensures sub-page block size
927 * filesystems are correctly handled, too.
929 * We have to do all the page cache truncate work outside the
930 * transaction context as the "lock" order is page lock->log space
931 * reservation as defined by extent allocation in the writeback path.
932 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
933 * having already truncated the in-memory version of the file (i.e. made
934 * user visible changes). There's not much we can do about this, except
935 * to hope that the caller sees ENOMEM and retries the truncate
938 * And we update in-core i_size and truncate page cache beyond newsize
939 * before writeback the [di_size, newsize] range, so we're guaranteed
940 * not to write stale data past the new EOF on truncate down.
942 truncate_setsize(inode
, newsize
);
945 * We are going to log the inode size change in this transaction so
946 * any previous writes that are beyond the on disk EOF and the new
947 * EOF that have not been written out need to be written here. If we
948 * do not write the data out, we expose ourselves to the null files
949 * problem. Note that this includes any block zeroing we did above;
950 * otherwise those blocks may not be zeroed after a crash.
953 (newsize
> ip
->i_d
.di_size
&& oldsize
!= ip
->i_d
.di_size
)) {
954 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
955 ip
->i_d
.di_size
, newsize
- 1);
960 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
964 lock_flags
|= XFS_ILOCK_EXCL
;
965 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
966 xfs_trans_ijoin(tp
, ip
, 0);
969 * Only change the c/mtime if we are changing the size or we are
970 * explicitly asked to change it. This handles the semantic difference
971 * between truncate() and ftruncate() as implemented in the VFS.
973 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
974 * special case where we need to update the times despite not having
975 * these flags set. For all other operations the VFS set these flags
976 * explicitly if it wants a timestamp update.
978 if (newsize
!= oldsize
&&
979 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
980 iattr
->ia_ctime
= iattr
->ia_mtime
=
982 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
986 * The first thing we do is set the size to new_size permanently on
987 * disk. This way we don't have to worry about anyone ever being able
988 * to look at the data being freed even in the face of a crash.
989 * What we're getting around here is the case where we free a block, it
990 * is allocated to another file, it is written to, and then we crash.
991 * If the new data gets written to the file but the log buffers
992 * containing the free and reallocation don't, then we'd end up with
993 * garbage in the blocks being freed. As long as we make the new size
994 * permanent before actually freeing any blocks it doesn't matter if
995 * they get written to.
997 ip
->i_d
.di_size
= newsize
;
998 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1000 if (newsize
<= oldsize
) {
1001 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
1003 goto out_trans_cancel
;
1006 * Truncated "down", so we're removing references to old data
1007 * here - if we delay flushing for a long time, we expose
1008 * ourselves unduly to the notorious NULL files problem. So,
1009 * we mark this inode and flush it when the file is closed,
1010 * and do not wait the usual (long) time for writeout.
1012 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
1014 /* A truncate down always removes post-EOF blocks. */
1015 xfs_inode_clear_eofblocks_tag(ip
);
1018 if (iattr
->ia_valid
& ATTR_MODE
)
1019 xfs_setattr_mode(ip
, iattr
);
1020 if (iattr
->ia_valid
& (ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
1021 xfs_setattr_time(ip
, iattr
);
1023 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1025 XFS_STATS_INC(mp
, xs_ig_attrchg
);
1027 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1028 xfs_trans_set_sync(tp
);
1030 error
= xfs_trans_commit(tp
);
1033 xfs_iunlock(ip
, lock_flags
);
1037 xfs_trans_cancel(tp
);
1042 xfs_vn_setattr_size(
1043 struct dentry
*dentry
,
1044 struct iattr
*iattr
)
1046 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
1049 trace_xfs_setattr(ip
);
1051 error
= xfs_vn_change_ok(dentry
, iattr
);
1054 return xfs_setattr_size(ip
, iattr
);
1059 struct dentry
*dentry
,
1060 struct iattr
*iattr
)
1064 if (iattr
->ia_valid
& ATTR_SIZE
) {
1065 struct inode
*inode
= d_inode(dentry
);
1066 struct xfs_inode
*ip
= XFS_I(inode
);
1069 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1070 iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1072 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
1074 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1078 error
= xfs_vn_setattr_size(dentry
, iattr
);
1079 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1081 error
= xfs_vn_setattr_nonsize(dentry
, iattr
);
1089 struct inode
*inode
,
1090 struct timespec64
*now
,
1093 struct xfs_inode
*ip
= XFS_I(inode
);
1094 struct xfs_mount
*mp
= ip
->i_mount
;
1095 int log_flags
= XFS_ILOG_TIMESTAMP
;
1096 struct xfs_trans
*tp
;
1099 trace_xfs_update_time(ip
);
1101 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
) {
1102 if (!((flags
& S_VERSION
) &&
1103 inode_maybe_inc_iversion(inode
, false)))
1104 return generic_update_time(inode
, now
, flags
);
1106 /* Capture the iversion update that just occurred */
1107 log_flags
|= XFS_ILOG_CORE
;
1110 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_fsyncts
, 0, 0, 0, &tp
);
1114 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1115 if (flags
& S_CTIME
)
1116 inode
->i_ctime
= *now
;
1117 if (flags
& S_MTIME
)
1118 inode
->i_mtime
= *now
;
1119 if (flags
& S_ATIME
)
1120 inode
->i_atime
= *now
;
1122 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1123 xfs_trans_log_inode(tp
, ip
, log_flags
);
1124 return xfs_trans_commit(tp
);
1129 struct inode
*inode
,
1130 struct fiemap_extent_info
*fieinfo
,
1136 xfs_ilock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1137 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
1138 fieinfo
->fi_flags
&= ~FIEMAP_FLAG_XATTR
;
1139 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1140 &xfs_xattr_iomap_ops
);
1142 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1143 &xfs_read_iomap_ops
);
1145 xfs_iunlock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1153 struct dentry
*dentry
,
1156 return xfs_generic_create(dir
, dentry
, mode
, 0, true);
1159 static const struct inode_operations xfs_inode_operations
= {
1160 .get_acl
= xfs_get_acl
,
1161 .set_acl
= xfs_set_acl
,
1162 .getattr
= xfs_vn_getattr
,
1163 .setattr
= xfs_vn_setattr
,
1164 .listxattr
= xfs_vn_listxattr
,
1165 .fiemap
= xfs_vn_fiemap
,
1166 .update_time
= xfs_vn_update_time
,
1169 static const struct inode_operations xfs_dir_inode_operations
= {
1170 .create
= xfs_vn_create
,
1171 .lookup
= xfs_vn_lookup
,
1172 .link
= xfs_vn_link
,
1173 .unlink
= xfs_vn_unlink
,
1174 .symlink
= xfs_vn_symlink
,
1175 .mkdir
= xfs_vn_mkdir
,
1177 * Yes, XFS uses the same method for rmdir and unlink.
1179 * There are some subtile differences deeper in the code,
1180 * but we use S_ISDIR to check for those.
1182 .rmdir
= xfs_vn_unlink
,
1183 .mknod
= xfs_vn_mknod
,
1184 .rename
= xfs_vn_rename
,
1185 .get_acl
= xfs_get_acl
,
1186 .set_acl
= xfs_set_acl
,
1187 .getattr
= xfs_vn_getattr
,
1188 .setattr
= xfs_vn_setattr
,
1189 .listxattr
= xfs_vn_listxattr
,
1190 .update_time
= xfs_vn_update_time
,
1191 .tmpfile
= xfs_vn_tmpfile
,
1194 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1195 .create
= xfs_vn_create
,
1196 .lookup
= xfs_vn_ci_lookup
,
1197 .link
= xfs_vn_link
,
1198 .unlink
= xfs_vn_unlink
,
1199 .symlink
= xfs_vn_symlink
,
1200 .mkdir
= xfs_vn_mkdir
,
1202 * Yes, XFS uses the same method for rmdir and unlink.
1204 * There are some subtile differences deeper in the code,
1205 * but we use S_ISDIR to check for those.
1207 .rmdir
= xfs_vn_unlink
,
1208 .mknod
= xfs_vn_mknod
,
1209 .rename
= xfs_vn_rename
,
1210 .get_acl
= xfs_get_acl
,
1211 .set_acl
= xfs_set_acl
,
1212 .getattr
= xfs_vn_getattr
,
1213 .setattr
= xfs_vn_setattr
,
1214 .listxattr
= xfs_vn_listxattr
,
1215 .update_time
= xfs_vn_update_time
,
1216 .tmpfile
= xfs_vn_tmpfile
,
1219 static const struct inode_operations xfs_symlink_inode_operations
= {
1220 .get_link
= xfs_vn_get_link
,
1221 .getattr
= xfs_vn_getattr
,
1222 .setattr
= xfs_vn_setattr
,
1223 .listxattr
= xfs_vn_listxattr
,
1224 .update_time
= xfs_vn_update_time
,
1227 static const struct inode_operations xfs_inline_symlink_inode_operations
= {
1228 .get_link
= xfs_vn_get_link_inline
,
1229 .getattr
= xfs_vn_getattr
,
1230 .setattr
= xfs_vn_setattr
,
1231 .listxattr
= xfs_vn_listxattr
,
1232 .update_time
= xfs_vn_update_time
,
1235 /* Figure out if this file actually supports DAX. */
1237 xfs_inode_supports_dax(
1238 struct xfs_inode
*ip
)
1240 struct xfs_mount
*mp
= ip
->i_mount
;
1242 /* Only supported on regular files. */
1243 if (!S_ISREG(VFS_I(ip
)->i_mode
))
1246 /* Only supported on non-reflinked files. */
1247 if (xfs_is_reflink_inode(ip
))
1250 /* Block size must match page size */
1251 if (mp
->m_sb
.sb_blocksize
!= PAGE_SIZE
)
1254 /* Device has to support DAX too. */
1255 return xfs_inode_buftarg(ip
)->bt_daxdev
!= NULL
;
1259 xfs_inode_should_enable_dax(
1260 struct xfs_inode
*ip
)
1262 if (!IS_ENABLED(CONFIG_FS_DAX
))
1264 if (ip
->i_mount
->m_flags
& XFS_MOUNT_DAX_NEVER
)
1266 if (!xfs_inode_supports_dax(ip
))
1268 if (ip
->i_mount
->m_flags
& XFS_MOUNT_DAX_ALWAYS
)
1270 if (ip
->i_d
.di_flags2
& XFS_DIFLAG2_DAX
)
1276 xfs_diflags_to_iflags(
1277 struct xfs_inode
*ip
,
1280 struct inode
*inode
= VFS_I(ip
);
1281 unsigned int xflags
= xfs_ip2xflags(ip
);
1282 unsigned int flags
= 0;
1284 ASSERT(!(IS_DAX(inode
) && init
));
1286 if (xflags
& FS_XFLAG_IMMUTABLE
)
1287 flags
|= S_IMMUTABLE
;
1288 if (xflags
& FS_XFLAG_APPEND
)
1290 if (xflags
& FS_XFLAG_SYNC
)
1292 if (xflags
& FS_XFLAG_NOATIME
)
1294 if (init
&& xfs_inode_should_enable_dax(ip
))
1298 * S_DAX can only be set during inode initialization and is never set by
1299 * the VFS, so we cannot mask off S_DAX in i_flags.
1301 inode
->i_flags
&= ~(S_IMMUTABLE
| S_APPEND
| S_SYNC
| S_NOATIME
);
1302 inode
->i_flags
|= flags
;
1306 * Initialize the Linux inode.
1308 * When reading existing inodes from disk this is called directly from xfs_iget,
1309 * when creating a new inode it is called from xfs_ialloc after setting up the
1310 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1311 * it up to the caller to deal with unlocking the inode appropriately.
1315 struct xfs_inode
*ip
)
1317 struct inode
*inode
= &ip
->i_vnode
;
1320 inode
->i_ino
= ip
->i_ino
;
1321 inode
->i_state
= I_NEW
;
1323 inode_sb_list_add(inode
);
1324 /* make the inode look hashed for the writeback code */
1325 inode_fake_hash(inode
);
1327 i_size_write(inode
, ip
->i_d
.di_size
);
1328 xfs_diflags_to_iflags(ip
, true);
1330 if (S_ISDIR(inode
->i_mode
)) {
1332 * We set the i_rwsem class here to avoid potential races with
1333 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1334 * after a filehandle lookup has already found the inode in
1335 * cache before it has been unlocked via unlock_new_inode().
1337 lockdep_set_class(&inode
->i_rwsem
,
1338 &inode
->i_sb
->s_type
->i_mutex_dir_key
);
1339 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1341 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1345 * Ensure all page cache allocations are done from GFP_NOFS context to
1346 * prevent direct reclaim recursion back into the filesystem and blowing
1347 * stacks or deadlocking.
1349 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1350 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1353 * If there is no attribute fork no ACL can exist on this inode,
1354 * and it can't have any file capabilities attached to it either.
1356 if (!XFS_IFORK_Q(ip
)) {
1357 inode_has_no_xattr(inode
);
1358 cache_no_acl(inode
);
1364 struct xfs_inode
*ip
)
1366 struct inode
*inode
= &ip
->i_vnode
;
1368 switch (inode
->i_mode
& S_IFMT
) {
1370 inode
->i_op
= &xfs_inode_operations
;
1371 inode
->i_fop
= &xfs_file_operations
;
1373 inode
->i_mapping
->a_ops
= &xfs_dax_aops
;
1375 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1378 if (xfs_sb_version_hasasciici(&XFS_M(inode
->i_sb
)->m_sb
))
1379 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1381 inode
->i_op
= &xfs_dir_inode_operations
;
1382 inode
->i_fop
= &xfs_dir_file_operations
;
1385 if (ip
->i_df
.if_flags
& XFS_IFINLINE
)
1386 inode
->i_op
= &xfs_inline_symlink_inode_operations
;
1388 inode
->i_op
= &xfs_symlink_inode_operations
;
1391 inode
->i_op
= &xfs_inode_operations
;
1392 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);