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"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_trace.h"
21 #include "xfs_icache.h"
22 #include "xfs_symlink.h"
24 #include "xfs_iomap.h"
25 #include "xfs_error.h"
26 #include "xfs_ioctl.h"
27 #include "xfs_xattr.h"
29 #include <linux/posix_acl.h>
30 #include <linux/security.h>
31 #include <linux/iversion.h>
32 #include <linux/fiemap.h>
35 * Directories have different lock order w.r.t. mmap_lock compared to regular
36 * files. This is due to readdir potentially triggering page faults on a user
37 * buffer inside filldir(), and this happens with the ilock on the directory
38 * held. For regular files, the lock order is the other way around - the
39 * mmap_lock is taken during the page fault, and then we lock the ilock to do
40 * block mapping. Hence we need a different class for the directory ilock so
41 * that lockdep can tell them apart.
43 static struct lock_class_key xfs_nondir_ilock_class
;
44 static struct lock_class_key xfs_dir_ilock_class
;
49 const struct xattr
*xattr_array
,
52 const struct xattr
*xattr
;
53 struct xfs_inode
*ip
= XFS_I(inode
);
56 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
57 struct xfs_da_args args
= {
59 .attr_filter
= XFS_ATTR_SECURE
,
61 .namelen
= strlen(xattr
->name
),
62 .value
= xattr
->value
,
63 .valuelen
= xattr
->value_len
,
65 error
= xfs_attr_change(&args
);
73 * Hook in SELinux. This is not quite correct yet, what we really need
74 * here (as we do for default ACLs) is a mechanism by which creation of
75 * these attrs can be journalled at inode creation time (along with the
76 * inode, of course, such that log replay can't cause these to be lost).
79 xfs_inode_init_security(
82 const struct qstr
*qstr
)
84 return security_inode_init_security(inode
, dir
, qstr
,
85 &xfs_initxattrs
, NULL
);
90 struct xfs_name
*namep
,
91 struct dentry
*dentry
)
93 namep
->name
= dentry
->d_name
.name
;
94 namep
->len
= dentry
->d_name
.len
;
95 namep
->type
= XFS_DIR3_FT_UNKNOWN
;
99 xfs_dentry_mode_to_name(
100 struct xfs_name
*namep
,
101 struct dentry
*dentry
,
104 namep
->name
= dentry
->d_name
.name
;
105 namep
->len
= dentry
->d_name
.len
;
106 namep
->type
= xfs_mode_to_ftype(mode
);
108 if (unlikely(namep
->type
== XFS_DIR3_FT_UNKNOWN
))
109 return -EFSCORRUPTED
;
118 struct dentry
*dentry
)
120 struct xfs_name teardown
;
123 * If we can't add the ACL or we fail in
124 * xfs_inode_init_security we must back out.
125 * ENOSPC can hit here, among other things.
127 xfs_dentry_to_name(&teardown
, dentry
);
129 xfs_remove(XFS_I(dir
), &teardown
, XFS_I(inode
));
133 * Check to see if we are likely to need an extended attribute to be added to
134 * the inode we are about to allocate. This allows the attribute fork to be
135 * created during the inode allocation, reducing the number of transactions we
136 * need to do in this fast path.
138 * The security checks are optimistic, but not guaranteed. The two LSMs that
139 * require xattrs to be added here (selinux and smack) are also the only two
140 * LSMs that add a sb->s_security structure to the superblock. Hence if security
141 * is enabled and sb->s_security is set, we have a pretty good idea that we are
142 * going to be asked to add a security xattr immediately after allocating the
143 * xfs inode and instantiating the VFS inode.
146 xfs_create_need_xattr(
148 struct posix_acl
*default_acl
,
149 struct posix_acl
*acl
)
155 #if IS_ENABLED(CONFIG_SECURITY)
156 if (dir
->i_sb
->s_security
)
165 struct mnt_idmap
*idmap
,
167 struct dentry
*dentry
,
170 struct file
*tmpfile
) /* unnamed file */
173 struct xfs_inode
*ip
= NULL
;
174 struct posix_acl
*default_acl
, *acl
;
175 struct xfs_name name
;
179 * Irix uses Missed'em'V split, but doesn't want to see
180 * the upper 5 bits of (14bit) major.
182 if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
183 if (unlikely(!sysv_valid_dev(rdev
) || MAJOR(rdev
) & ~0x1ff))
189 error
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
193 /* Verify mode is valid also for tmpfile case */
194 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
199 error
= xfs_create(idmap
, XFS_I(dir
), &name
, mode
, rdev
,
200 xfs_create_need_xattr(dir
, default_acl
, acl
),
203 error
= xfs_create_tmpfile(idmap
, XFS_I(dir
), mode
, &ip
);
210 error
= xfs_inode_init_security(inode
, dir
, &dentry
->d_name
);
212 goto out_cleanup_inode
;
215 error
= __xfs_set_acl(inode
, default_acl
, ACL_TYPE_DEFAULT
);
217 goto out_cleanup_inode
;
220 error
= __xfs_set_acl(inode
, acl
, ACL_TYPE_ACCESS
);
222 goto out_cleanup_inode
;
229 * The VFS requires that any inode fed to d_tmpfile must have
230 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
231 * However, we created the temp file with nlink == 0 because
232 * we're not allowed to put an inode with nlink > 0 on the
233 * unlinked list. Therefore we have to set nlink to 1 so that
234 * d_tmpfile can immediately set it back to zero.
237 d_tmpfile(tmpfile
, inode
);
239 d_instantiate(dentry
, inode
);
241 xfs_finish_inode_setup(ip
);
244 posix_acl_release(default_acl
);
245 posix_acl_release(acl
);
249 xfs_finish_inode_setup(ip
);
251 xfs_cleanup_inode(dir
, inode
, dentry
);
258 struct mnt_idmap
*idmap
,
260 struct dentry
*dentry
,
264 return xfs_generic_create(idmap
, dir
, dentry
, mode
, rdev
, NULL
);
269 struct mnt_idmap
*idmap
,
271 struct dentry
*dentry
,
275 return xfs_generic_create(idmap
, dir
, dentry
, mode
, 0, NULL
);
280 struct mnt_idmap
*idmap
,
282 struct dentry
*dentry
,
285 return xfs_generic_create(idmap
, dir
, dentry
, mode
| S_IFDIR
, 0, NULL
);
288 STATIC
struct dentry
*
291 struct dentry
*dentry
,
295 struct xfs_inode
*cip
;
296 struct xfs_name name
;
299 if (dentry
->d_name
.len
>= MAXNAMELEN
)
300 return ERR_PTR(-ENAMETOOLONG
);
302 xfs_dentry_to_name(&name
, dentry
);
303 error
= xfs_lookup(XFS_I(dir
), &name
, &cip
, NULL
);
306 else if (likely(error
== -ENOENT
))
309 inode
= ERR_PTR(error
);
310 return d_splice_alias(inode
, dentry
);
313 STATIC
struct dentry
*
316 struct dentry
*dentry
,
319 struct xfs_inode
*ip
;
320 struct xfs_name xname
;
321 struct xfs_name ci_name
;
325 if (dentry
->d_name
.len
>= MAXNAMELEN
)
326 return ERR_PTR(-ENAMETOOLONG
);
328 xfs_dentry_to_name(&xname
, dentry
);
329 error
= xfs_lookup(XFS_I(dir
), &xname
, &ip
, &ci_name
);
330 if (unlikely(error
)) {
331 if (unlikely(error
!= -ENOENT
))
332 return ERR_PTR(error
);
334 * call d_add(dentry, NULL) here when d_drop_negative_children
335 * is called in xfs_vn_mknod (ie. allow negative dentries
336 * with CI filesystems).
341 /* if exact match, just splice and exit */
343 return d_splice_alias(VFS_I(ip
), dentry
);
345 /* else case-insensitive match... */
346 dname
.name
= ci_name
.name
;
347 dname
.len
= ci_name
.len
;
348 dentry
= d_add_ci(dentry
, VFS_I(ip
), &dname
);
349 kmem_free(ci_name
.name
);
355 struct dentry
*old_dentry
,
357 struct dentry
*dentry
)
359 struct inode
*inode
= d_inode(old_dentry
);
360 struct xfs_name name
;
363 error
= xfs_dentry_mode_to_name(&name
, dentry
, inode
->i_mode
);
367 error
= xfs_link(XFS_I(dir
), XFS_I(inode
), &name
);
372 d_instantiate(dentry
, inode
);
379 struct dentry
*dentry
)
381 struct xfs_name name
;
384 xfs_dentry_to_name(&name
, dentry
);
386 error
= xfs_remove(XFS_I(dir
), &name
, XFS_I(d_inode(dentry
)));
391 * With unlink, the VFS makes the dentry "negative": no inode,
392 * but still hashed. This is incompatible with case-insensitive
393 * mode, so invalidate (unhash) the dentry in CI-mode.
395 if (xfs_has_asciici(XFS_M(dir
->i_sb
)))
396 d_invalidate(dentry
);
402 struct mnt_idmap
*idmap
,
404 struct dentry
*dentry
,
408 struct xfs_inode
*cip
= NULL
;
409 struct xfs_name name
;
414 (irix_symlink_mode
? 0777 & ~current_umask() : S_IRWXUGO
);
415 error
= xfs_dentry_mode_to_name(&name
, dentry
, mode
);
419 error
= xfs_symlink(idmap
, XFS_I(dir
), &name
, symname
, mode
, &cip
);
425 error
= xfs_inode_init_security(inode
, dir
, &dentry
->d_name
);
427 goto out_cleanup_inode
;
431 d_instantiate(dentry
, inode
);
432 xfs_finish_inode_setup(cip
);
436 xfs_finish_inode_setup(cip
);
437 xfs_cleanup_inode(dir
, inode
, dentry
);
445 struct mnt_idmap
*idmap
,
447 struct dentry
*odentry
,
449 struct dentry
*ndentry
,
452 struct inode
*new_inode
= d_inode(ndentry
);
455 struct xfs_name oname
;
456 struct xfs_name nname
;
458 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
461 /* if we are exchanging files, we need to set i_mode of both files */
462 if (flags
& RENAME_EXCHANGE
)
463 omode
= d_inode(ndentry
)->i_mode
;
465 error
= xfs_dentry_mode_to_name(&oname
, odentry
, omode
);
466 if (omode
&& unlikely(error
))
469 error
= xfs_dentry_mode_to_name(&nname
, ndentry
,
470 d_inode(odentry
)->i_mode
);
474 return xfs_rename(idmap
, XFS_I(odir
), &oname
,
475 XFS_I(d_inode(odentry
)), XFS_I(ndir
), &nname
,
476 new_inode
? XFS_I(new_inode
) : NULL
, flags
);
480 * careful here - this function can get called recursively, so
481 * we need to be very careful about how much stack we use.
482 * uio is kmalloced for this reason...
486 struct dentry
*dentry
,
488 struct delayed_call
*done
)
494 return ERR_PTR(-ECHILD
);
496 link
= kmalloc(XFS_SYMLINK_MAXLEN
+1, GFP_KERNEL
);
500 error
= xfs_readlink(XFS_I(d_inode(dentry
)), link
);
504 set_delayed_call(done
, kfree_link
, link
);
510 return ERR_PTR(error
);
515 struct xfs_inode
*ip
)
517 struct xfs_mount
*mp
= ip
->i_mount
;
520 * If the file blocks are being allocated from a realtime volume, then
521 * always return the realtime extent size.
523 if (XFS_IS_REALTIME_INODE(ip
))
524 return XFS_FSB_TO_B(mp
, xfs_get_extsz_hint(ip
));
527 * Allow large block sizes to be reported to userspace programs if the
528 * "largeio" mount option is used.
530 * If compatibility mode is specified, simply return the basic unit of
531 * caching so that we don't get inefficient read/modify/write I/O from
532 * user apps. Otherwise....
534 * If the underlying volume is a stripe, then return the stripe width in
535 * bytes as the recommended I/O size. It is not a stripe and we've set a
536 * default buffered I/O size, return that, otherwise return the compat
539 if (xfs_has_large_iosize(mp
)) {
541 return XFS_FSB_TO_B(mp
, mp
->m_swidth
);
542 if (xfs_has_allocsize(mp
))
543 return 1U << mp
->m_allocsize_log
;
551 struct mnt_idmap
*idmap
,
552 const struct path
*path
,
555 unsigned int query_flags
)
557 struct inode
*inode
= d_inode(path
->dentry
);
558 struct xfs_inode
*ip
= XFS_I(inode
);
559 struct xfs_mount
*mp
= ip
->i_mount
;
560 vfsuid_t vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
561 vfsgid_t vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
563 trace_xfs_getattr(ip
);
565 if (xfs_is_shutdown(mp
))
568 stat
->size
= XFS_ISIZE(ip
);
569 stat
->dev
= inode
->i_sb
->s_dev
;
570 stat
->mode
= inode
->i_mode
;
571 stat
->nlink
= inode
->i_nlink
;
572 stat
->uid
= vfsuid_into_kuid(vfsuid
);
573 stat
->gid
= vfsgid_into_kgid(vfsgid
);
574 stat
->ino
= ip
->i_ino
;
575 stat
->atime
= inode
->i_atime
;
576 stat
->mtime
= inode
->i_mtime
;
577 stat
->ctime
= inode
->i_ctime
;
578 stat
->blocks
= XFS_FSB_TO_BB(mp
, ip
->i_nblocks
+ ip
->i_delayed_blks
);
580 if (xfs_has_v3inodes(mp
)) {
581 if (request_mask
& STATX_BTIME
) {
582 stat
->result_mask
|= STATX_BTIME
;
583 stat
->btime
= ip
->i_crtime
;
588 * Note: If you add another clause to set an attribute flag, please
589 * update attributes_mask below.
591 if (ip
->i_diflags
& XFS_DIFLAG_IMMUTABLE
)
592 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
593 if (ip
->i_diflags
& XFS_DIFLAG_APPEND
)
594 stat
->attributes
|= STATX_ATTR_APPEND
;
595 if (ip
->i_diflags
& XFS_DIFLAG_NODUMP
)
596 stat
->attributes
|= STATX_ATTR_NODUMP
;
598 stat
->attributes_mask
|= (STATX_ATTR_IMMUTABLE
|
602 switch (inode
->i_mode
& S_IFMT
) {
605 stat
->blksize
= BLKDEV_IOSIZE
;
606 stat
->rdev
= inode
->i_rdev
;
609 if (request_mask
& STATX_DIOALIGN
) {
610 struct xfs_buftarg
*target
= xfs_inode_buftarg(ip
);
611 struct block_device
*bdev
= target
->bt_bdev
;
613 stat
->result_mask
|= STATX_DIOALIGN
;
614 stat
->dio_mem_align
= bdev_dma_alignment(bdev
) + 1;
615 stat
->dio_offset_align
= bdev_logical_block_size(bdev
);
619 stat
->blksize
= xfs_stat_blksize(ip
);
629 struct mnt_idmap
*idmap
,
630 struct dentry
*dentry
,
633 struct xfs_mount
*mp
= XFS_I(d_inode(dentry
))->i_mount
;
635 if (xfs_is_readonly(mp
))
638 if (xfs_is_shutdown(mp
))
641 return setattr_prepare(idmap
, dentry
, iattr
);
645 * Set non-size attributes of an inode.
647 * Caution: The caller of this function is responsible for calling
648 * setattr_prepare() or otherwise verifying the change is fine.
652 struct mnt_idmap
*idmap
,
653 struct dentry
*dentry
,
654 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
;
663 kgid_t gid
= GLOBAL_ROOT_GID
;
664 struct xfs_dquot
*udqp
= NULL
, *gdqp
= NULL
;
665 struct xfs_dquot
*old_udqp
= NULL
, *old_gdqp
= 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
)) {
681 uid
= from_vfsuid(idmap
, i_user_ns(inode
),
683 qflags
|= XFS_QMOPT_UQUOTA
;
687 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
688 gid
= from_vfsgid(idmap
, i_user_ns(inode
),
690 qflags
|= XFS_QMOPT_GQUOTA
;
696 * We take a reference when we initialize udqp and gdqp,
697 * so it is important that we never blindly double trip on
698 * the same variable. See xfs_create() for an example.
700 ASSERT(udqp
== NULL
);
701 ASSERT(gdqp
== NULL
);
702 error
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, ip
->i_projid
,
703 qflags
, &udqp
, &gdqp
, NULL
);
708 error
= xfs_trans_alloc_ichange(ip
, udqp
, gdqp
, NULL
,
709 has_capability_noaudit(current
, CAP_FOWNER
), &tp
);
714 * Register quota modifications in the transaction. Must be the owner
715 * or privileged. These IDs could have changed since we last looked at
716 * them. But, we're assured that if the ownership did change while we
717 * didn't have the inode locked, inode's dquot(s) would have changed
720 if (XFS_IS_UQUOTA_ON(mp
) &&
721 i_uid_needs_update(idmap
, iattr
, inode
)) {
723 old_udqp
= xfs_qm_vop_chown(tp
, ip
, &ip
->i_udquot
, udqp
);
725 if (XFS_IS_GQUOTA_ON(mp
) &&
726 i_gid_needs_update(idmap
, iattr
, inode
)) {
727 ASSERT(xfs_has_pquotino(mp
) || !XFS_IS_PQUOTA_ON(mp
));
729 old_gdqp
= xfs_qm_vop_chown(tp
, ip
, &ip
->i_gdquot
, gdqp
);
732 setattr_copy(idmap
, inode
, iattr
);
733 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
735 XFS_STATS_INC(mp
, xs_ig_attrchg
);
737 if (xfs_has_wsync(mp
))
738 xfs_trans_set_sync(tp
);
739 error
= xfs_trans_commit(tp
);
742 * Release any dquot(s) the inode had kept before chown.
744 xfs_qm_dqrele(old_udqp
);
745 xfs_qm_dqrele(old_gdqp
);
753 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
754 * update. We could avoid this with linked transactions
755 * and passing down the transaction pointer all the way
756 * to attr_set. No previous user of the generic
757 * Posix ACL code seems to care about this issue either.
759 if (mask
& ATTR_MODE
) {
760 error
= posix_acl_chmod(idmap
, dentry
, inode
->i_mode
);
774 * Truncate file. Must have write permission and not be a directory.
776 * Caution: The caller of this function is responsible for calling
777 * setattr_prepare() or otherwise verifying the change is fine.
781 struct mnt_idmap
*idmap
,
782 struct dentry
*dentry
,
783 struct xfs_inode
*ip
,
786 struct xfs_mount
*mp
= ip
->i_mount
;
787 struct inode
*inode
= VFS_I(ip
);
788 xfs_off_t oldsize
, newsize
;
789 struct xfs_trans
*tp
;
792 bool did_zeroing
= false;
794 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
795 ASSERT(xfs_isilocked(ip
, XFS_MMAPLOCK_EXCL
));
796 ASSERT(S_ISREG(inode
->i_mode
));
797 ASSERT((iattr
->ia_valid
& (ATTR_UID
|ATTR_GID
|ATTR_ATIME
|ATTR_ATIME_SET
|
798 ATTR_MTIME_SET
|ATTR_TIMES_SET
)) == 0);
800 oldsize
= inode
->i_size
;
801 newsize
= iattr
->ia_size
;
804 * Short circuit the truncate case for zero length files.
806 if (newsize
== 0 && oldsize
== 0 && ip
->i_df
.if_nextents
== 0) {
807 if (!(iattr
->ia_valid
& (ATTR_CTIME
|ATTR_MTIME
)))
811 * Use the regular setattr path to update the timestamps.
813 iattr
->ia_valid
&= ~ATTR_SIZE
;
814 return xfs_setattr_nonsize(idmap
, dentry
, ip
, iattr
);
818 * Make sure that the dquots are attached to the inode.
820 error
= xfs_qm_dqattach(ip
);
825 * Wait for all direct I/O to complete.
827 inode_dio_wait(inode
);
830 * File data changes must be complete before we start the transaction to
831 * modify the inode. This needs to be done before joining the inode to
832 * the transaction because the inode cannot be unlocked once it is a
833 * part of the transaction.
835 * Start with zeroing any data beyond EOF that we may expose on file
836 * extension, or zeroing out the rest of the block on a downward
839 if (newsize
> oldsize
) {
840 trace_xfs_zero_eof(ip
, oldsize
, newsize
- oldsize
);
841 error
= xfs_zero_range(ip
, oldsize
, newsize
- oldsize
,
845 * iomap won't detect a dirty page over an unwritten block (or a
846 * cow block over a hole) and subsequently skips zeroing the
847 * newly post-EOF portion of the page. Flush the new EOF to
848 * convert the block before the pagecache truncate.
850 error
= filemap_write_and_wait_range(inode
->i_mapping
, newsize
,
854 error
= xfs_truncate_page(ip
, newsize
, &did_zeroing
);
861 * We've already locked out new page faults, so now we can safely remove
862 * pages from the page cache knowing they won't get refaulted until we
863 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
864 * complete. The truncate_setsize() call also cleans partial EOF page
865 * PTEs on extending truncates and hence ensures sub-page block size
866 * filesystems are correctly handled, too.
868 * We have to do all the page cache truncate work outside the
869 * transaction context as the "lock" order is page lock->log space
870 * reservation as defined by extent allocation in the writeback path.
871 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
872 * having already truncated the in-memory version of the file (i.e. made
873 * user visible changes). There's not much we can do about this, except
874 * to hope that the caller sees ENOMEM and retries the truncate
877 * And we update in-core i_size and truncate page cache beyond newsize
878 * before writeback the [i_disk_size, newsize] range, so we're
879 * guaranteed not to write stale data past the new EOF on truncate down.
881 truncate_setsize(inode
, newsize
);
884 * We are going to log the inode size change in this transaction so
885 * any previous writes that are beyond the on disk EOF and the new
886 * EOF that have not been written out need to be written here. If we
887 * do not write the data out, we expose ourselves to the null files
888 * problem. Note that this includes any block zeroing we did above;
889 * otherwise those blocks may not be zeroed after a crash.
892 (newsize
> ip
->i_disk_size
&& oldsize
!= ip
->i_disk_size
)) {
893 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
894 ip
->i_disk_size
, newsize
- 1);
899 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
903 lock_flags
|= XFS_ILOCK_EXCL
;
904 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
905 xfs_trans_ijoin(tp
, ip
, 0);
908 * Only change the c/mtime if we are changing the size or we are
909 * explicitly asked to change it. This handles the semantic difference
910 * between truncate() and ftruncate() as implemented in the VFS.
912 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
913 * special case where we need to update the times despite not having
914 * these flags set. For all other operations the VFS set these flags
915 * explicitly if it wants a timestamp update.
917 if (newsize
!= oldsize
&&
918 !(iattr
->ia_valid
& (ATTR_CTIME
| ATTR_MTIME
))) {
919 iattr
->ia_ctime
= iattr
->ia_mtime
=
921 iattr
->ia_valid
|= ATTR_CTIME
| ATTR_MTIME
;
925 * The first thing we do is set the size to new_size permanently on
926 * disk. This way we don't have to worry about anyone ever being able
927 * to look at the data being freed even in the face of a crash.
928 * What we're getting around here is the case where we free a block, it
929 * is allocated to another file, it is written to, and then we crash.
930 * If the new data gets written to the file but the log buffers
931 * containing the free and reallocation don't, then we'd end up with
932 * garbage in the blocks being freed. As long as we make the new size
933 * permanent before actually freeing any blocks it doesn't matter if
934 * they get written to.
936 ip
->i_disk_size
= newsize
;
937 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
939 if (newsize
<= oldsize
) {
940 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, newsize
);
942 goto out_trans_cancel
;
945 * Truncated "down", so we're removing references to old data
946 * here - if we delay flushing for a long time, we expose
947 * ourselves unduly to the notorious NULL files problem. So,
948 * we mark this inode and flush it when the file is closed,
949 * and do not wait the usual (long) time for writeout.
951 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
953 /* A truncate down always removes post-EOF blocks. */
954 xfs_inode_clear_eofblocks_tag(ip
);
957 ASSERT(!(iattr
->ia_valid
& (ATTR_UID
| ATTR_GID
)));
958 setattr_copy(idmap
, inode
, iattr
);
959 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
961 XFS_STATS_INC(mp
, xs_ig_attrchg
);
963 if (xfs_has_wsync(mp
))
964 xfs_trans_set_sync(tp
);
966 error
= xfs_trans_commit(tp
);
969 xfs_iunlock(ip
, lock_flags
);
973 xfs_trans_cancel(tp
);
979 struct mnt_idmap
*idmap
,
980 struct dentry
*dentry
,
983 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
986 trace_xfs_setattr(ip
);
988 error
= xfs_vn_change_ok(idmap
, dentry
, iattr
);
991 return xfs_setattr_size(idmap
, dentry
, ip
, iattr
);
996 struct mnt_idmap
*idmap
,
997 struct dentry
*dentry
,
1000 struct inode
*inode
= d_inode(dentry
);
1001 struct xfs_inode
*ip
= XFS_I(inode
);
1004 if (iattr
->ia_valid
& ATTR_SIZE
) {
1007 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1008 iolock
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1010 error
= xfs_break_layouts(inode
, &iolock
, BREAK_UNMAP
);
1012 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1016 error
= xfs_vn_setattr_size(idmap
, dentry
, iattr
);
1017 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1019 trace_xfs_setattr(ip
);
1021 error
= xfs_vn_change_ok(idmap
, dentry
, iattr
);
1023 error
= xfs_setattr_nonsize(idmap
, dentry
, ip
, iattr
);
1031 struct inode
*inode
,
1032 struct timespec64
*now
,
1035 struct xfs_inode
*ip
= XFS_I(inode
);
1036 struct xfs_mount
*mp
= ip
->i_mount
;
1037 int log_flags
= XFS_ILOG_TIMESTAMP
;
1038 struct xfs_trans
*tp
;
1041 trace_xfs_update_time(ip
);
1043 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
) {
1044 if (!((flags
& S_VERSION
) &&
1045 inode_maybe_inc_iversion(inode
, false)))
1046 return generic_update_time(inode
, now
, flags
);
1048 /* Capture the iversion update that just occurred */
1049 log_flags
|= XFS_ILOG_CORE
;
1052 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_fsyncts
, 0, 0, 0, &tp
);
1056 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1057 if (flags
& S_CTIME
)
1058 inode
->i_ctime
= *now
;
1059 if (flags
& S_MTIME
)
1060 inode
->i_mtime
= *now
;
1061 if (flags
& S_ATIME
)
1062 inode
->i_atime
= *now
;
1064 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1065 xfs_trans_log_inode(tp
, ip
, log_flags
);
1066 return xfs_trans_commit(tp
);
1071 struct inode
*inode
,
1072 struct fiemap_extent_info
*fieinfo
,
1078 xfs_ilock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1079 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
1080 fieinfo
->fi_flags
&= ~FIEMAP_FLAG_XATTR
;
1081 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1082 &xfs_xattr_iomap_ops
);
1084 error
= iomap_fiemap(inode
, fieinfo
, start
, length
,
1085 &xfs_read_iomap_ops
);
1087 xfs_iunlock(XFS_I(inode
), XFS_IOLOCK_SHARED
);
1094 struct mnt_idmap
*idmap
,
1099 int err
= xfs_generic_create(idmap
, dir
, file
->f_path
.dentry
, mode
, 0, file
);
1101 return finish_open_simple(file
, err
);
1104 static const struct inode_operations xfs_inode_operations
= {
1105 .get_inode_acl
= xfs_get_acl
,
1106 .set_acl
= xfs_set_acl
,
1107 .getattr
= xfs_vn_getattr
,
1108 .setattr
= xfs_vn_setattr
,
1109 .listxattr
= xfs_vn_listxattr
,
1110 .fiemap
= xfs_vn_fiemap
,
1111 .update_time
= xfs_vn_update_time
,
1112 .fileattr_get
= xfs_fileattr_get
,
1113 .fileattr_set
= xfs_fileattr_set
,
1116 static const struct inode_operations xfs_dir_inode_operations
= {
1117 .create
= xfs_vn_create
,
1118 .lookup
= xfs_vn_lookup
,
1119 .link
= xfs_vn_link
,
1120 .unlink
= xfs_vn_unlink
,
1121 .symlink
= xfs_vn_symlink
,
1122 .mkdir
= xfs_vn_mkdir
,
1124 * Yes, XFS uses the same method for rmdir and unlink.
1126 * There are some subtile differences deeper in the code,
1127 * but we use S_ISDIR to check for those.
1129 .rmdir
= xfs_vn_unlink
,
1130 .mknod
= xfs_vn_mknod
,
1131 .rename
= xfs_vn_rename
,
1132 .get_inode_acl
= xfs_get_acl
,
1133 .set_acl
= xfs_set_acl
,
1134 .getattr
= xfs_vn_getattr
,
1135 .setattr
= xfs_vn_setattr
,
1136 .listxattr
= xfs_vn_listxattr
,
1137 .update_time
= xfs_vn_update_time
,
1138 .tmpfile
= xfs_vn_tmpfile
,
1139 .fileattr_get
= xfs_fileattr_get
,
1140 .fileattr_set
= xfs_fileattr_set
,
1143 static const struct inode_operations xfs_dir_ci_inode_operations
= {
1144 .create
= xfs_vn_create
,
1145 .lookup
= xfs_vn_ci_lookup
,
1146 .link
= xfs_vn_link
,
1147 .unlink
= xfs_vn_unlink
,
1148 .symlink
= xfs_vn_symlink
,
1149 .mkdir
= xfs_vn_mkdir
,
1151 * Yes, XFS uses the same method for rmdir and unlink.
1153 * There are some subtile differences deeper in the code,
1154 * but we use S_ISDIR to check for those.
1156 .rmdir
= xfs_vn_unlink
,
1157 .mknod
= xfs_vn_mknod
,
1158 .rename
= xfs_vn_rename
,
1159 .get_inode_acl
= xfs_get_acl
,
1160 .set_acl
= xfs_set_acl
,
1161 .getattr
= xfs_vn_getattr
,
1162 .setattr
= xfs_vn_setattr
,
1163 .listxattr
= xfs_vn_listxattr
,
1164 .update_time
= xfs_vn_update_time
,
1165 .tmpfile
= xfs_vn_tmpfile
,
1166 .fileattr_get
= xfs_fileattr_get
,
1167 .fileattr_set
= xfs_fileattr_set
,
1170 static const struct inode_operations xfs_symlink_inode_operations
= {
1171 .get_link
= xfs_vn_get_link
,
1172 .getattr
= xfs_vn_getattr
,
1173 .setattr
= xfs_vn_setattr
,
1174 .listxattr
= xfs_vn_listxattr
,
1175 .update_time
= xfs_vn_update_time
,
1178 /* Figure out if this file actually supports DAX. */
1180 xfs_inode_supports_dax(
1181 struct xfs_inode
*ip
)
1183 struct xfs_mount
*mp
= ip
->i_mount
;
1185 /* Only supported on regular files. */
1186 if (!S_ISREG(VFS_I(ip
)->i_mode
))
1189 /* Block size must match page size */
1190 if (mp
->m_sb
.sb_blocksize
!= PAGE_SIZE
)
1193 /* Device has to support DAX too. */
1194 return xfs_inode_buftarg(ip
)->bt_daxdev
!= NULL
;
1198 xfs_inode_should_enable_dax(
1199 struct xfs_inode
*ip
)
1201 if (!IS_ENABLED(CONFIG_FS_DAX
))
1203 if (xfs_has_dax_never(ip
->i_mount
))
1205 if (!xfs_inode_supports_dax(ip
))
1207 if (xfs_has_dax_always(ip
->i_mount
))
1209 if (ip
->i_diflags2
& XFS_DIFLAG2_DAX
)
1215 xfs_diflags_to_iflags(
1216 struct xfs_inode
*ip
,
1219 struct inode
*inode
= VFS_I(ip
);
1220 unsigned int xflags
= xfs_ip2xflags(ip
);
1221 unsigned int flags
= 0;
1223 ASSERT(!(IS_DAX(inode
) && init
));
1225 if (xflags
& FS_XFLAG_IMMUTABLE
)
1226 flags
|= S_IMMUTABLE
;
1227 if (xflags
& FS_XFLAG_APPEND
)
1229 if (xflags
& FS_XFLAG_SYNC
)
1231 if (xflags
& FS_XFLAG_NOATIME
)
1233 if (init
&& xfs_inode_should_enable_dax(ip
))
1237 * S_DAX can only be set during inode initialization and is never set by
1238 * the VFS, so we cannot mask off S_DAX in i_flags.
1240 inode
->i_flags
&= ~(S_IMMUTABLE
| S_APPEND
| S_SYNC
| S_NOATIME
);
1241 inode
->i_flags
|= flags
;
1245 * Initialize the Linux inode.
1247 * When reading existing inodes from disk this is called directly from xfs_iget,
1248 * when creating a new inode it is called from xfs_init_new_inode after setting
1249 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1250 * leave it up to the caller to deal with unlocking the inode appropriately.
1254 struct xfs_inode
*ip
)
1256 struct inode
*inode
= &ip
->i_vnode
;
1259 inode
->i_ino
= ip
->i_ino
;
1260 inode
->i_state
|= I_NEW
;
1262 inode_sb_list_add(inode
);
1263 /* make the inode look hashed for the writeback code */
1264 inode_fake_hash(inode
);
1266 i_size_write(inode
, ip
->i_disk_size
);
1267 xfs_diflags_to_iflags(ip
, true);
1269 if (S_ISDIR(inode
->i_mode
)) {
1271 * We set the i_rwsem class here to avoid potential races with
1272 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1273 * after a filehandle lookup has already found the inode in
1274 * cache before it has been unlocked via unlock_new_inode().
1276 lockdep_set_class(&inode
->i_rwsem
,
1277 &inode
->i_sb
->s_type
->i_mutex_dir_key
);
1278 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_dir_ilock_class
);
1280 lockdep_set_class(&ip
->i_lock
.mr_lock
, &xfs_nondir_ilock_class
);
1284 * Ensure all page cache allocations are done from GFP_NOFS context to
1285 * prevent direct reclaim recursion back into the filesystem and blowing
1286 * stacks or deadlocking.
1288 gfp_mask
= mapping_gfp_mask(inode
->i_mapping
);
1289 mapping_set_gfp_mask(inode
->i_mapping
, (gfp_mask
& ~(__GFP_FS
)));
1292 * If there is no attribute fork no ACL can exist on this inode,
1293 * and it can't have any file capabilities attached to it either.
1295 if (!xfs_inode_has_attr_fork(ip
)) {
1296 inode_has_no_xattr(inode
);
1297 cache_no_acl(inode
);
1303 struct xfs_inode
*ip
)
1305 struct inode
*inode
= &ip
->i_vnode
;
1307 switch (inode
->i_mode
& S_IFMT
) {
1309 inode
->i_op
= &xfs_inode_operations
;
1310 inode
->i_fop
= &xfs_file_operations
;
1312 inode
->i_mapping
->a_ops
= &xfs_dax_aops
;
1314 inode
->i_mapping
->a_ops
= &xfs_address_space_operations
;
1317 if (xfs_has_asciici(XFS_M(inode
->i_sb
)))
1318 inode
->i_op
= &xfs_dir_ci_inode_operations
;
1320 inode
->i_op
= &xfs_dir_inode_operations
;
1321 inode
->i_fop
= &xfs_dir_file_operations
;
1324 inode
->i_op
= &xfs_symlink_inode_operations
;
1327 inode
->i_op
= &xfs_inode_operations
;
1328 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);