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"
24 #include "xfs_ioctl.h"
26 #include <linux/posix_acl.h>
27 #include <linux/security.h>
28 #include <linux/iversion.h>
29 #include <linux/fiemap.h>
32 * Directories have different lock order w.r.t. mmap_lock compared to regular
33 * files. This is due to readdir potentially triggering page faults on a user
34 * buffer inside filldir(), and this happens with the ilock on the directory
35 * held. For regular files, the lock order is the other way around - the
36 * mmap_lock is taken during the page fault, and then we lock the ilock to do
37 * block mapping. Hence we need a different class for the directory ilock so
38 * that lockdep can tell them apart.
40 static struct lock_class_key xfs_nondir_ilock_class;
41 static struct lock_class_key xfs_dir_ilock_class;
46 const struct xattr *xattr_array,
49 const struct xattr *xattr;
50 struct xfs_inode *ip = XFS_I(inode);
53 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
54 struct xfs_da_args args = {
56 .attr_filter = XFS_ATTR_SECURE,
58 .namelen = strlen(xattr->name),
59 .value = xattr->value,
60 .valuelen = xattr->value_len,
62 error = xfs_attr_set(&args);
70 * Hook in SELinux. This is not quite correct yet, what we really need
71 * here (as we do for default ACLs) is a mechanism by which creation of
72 * these attrs can be journalled at inode creation time (along with the
73 * inode, of course, such that log replay can't cause these to be lost).
80 const struct qstr *qstr)
82 return security_inode_init_security(inode, dir, qstr,
83 &xfs_initxattrs, NULL);
88 struct xfs_name *namep,
89 struct dentry *dentry)
91 namep->name = dentry->d_name.name;
92 namep->len = dentry->d_name.len;
93 namep->type = XFS_DIR3_FT_UNKNOWN;
97 xfs_dentry_mode_to_name(
98 struct xfs_name *namep,
99 struct dentry *dentry,
102 namep->name = dentry->d_name.name;
103 namep->len = dentry->d_name.len;
104 namep->type = xfs_mode_to_ftype(mode);
106 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
107 return -EFSCORRUPTED;
116 struct dentry *dentry)
118 struct xfs_name teardown;
121 * If we can't add the ACL or we fail in
122 * xfs_init_security we must back out.
123 * ENOSPC can hit here, among other things.
125 xfs_dentry_to_name(&teardown, dentry);
127 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
131 * Check to see if we are likely to need an extended attribute to be added to
132 * the inode we are about to allocate. This allows the attribute fork to be
133 * created during the inode allocation, reducing the number of transactions we
134 * need to do in this fast path.
136 * The security checks are optimistic, but not guaranteed. The two LSMs that
137 * require xattrs to be added here (selinux and smack) are also the only two
138 * LSMs that add a sb->s_security structure to the superblock. Hence if security
139 * is enabled and sb->s_security is set, we have a pretty good idea that we are
140 * going to be asked to add a security xattr immediately after allocating the
141 * xfs inode and instantiating the VFS inode.
144 xfs_create_need_xattr(
146 struct posix_acl *default_acl,
147 struct posix_acl *acl)
153 #if IS_ENABLED(CONFIG_SECURITY)
154 if (dir->i_sb->s_security)
163 struct user_namespace *mnt_userns,
165 struct dentry *dentry,
168 bool tmpfile) /* unnamed file */
171 struct xfs_inode *ip = NULL;
172 struct posix_acl *default_acl, *acl;
173 struct xfs_name name;
177 * Irix uses Missed'em'V split, but doesn't want to see
178 * the upper 5 bits of (14bit) major.
180 if (S_ISCHR(mode) || S_ISBLK(mode)) {
181 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
187 error = posix_acl_create(dir, &mode, &default_acl, &acl);
191 /* Verify mode is valid also for tmpfile case */
192 error = xfs_dentry_mode_to_name(&name, dentry, mode);
197 error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev,
198 xfs_create_need_xattr(dir, default_acl, acl),
201 error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip);
208 error = xfs_init_security(inode, dir, &dentry->d_name);
210 goto out_cleanup_inode;
212 #ifdef CONFIG_XFS_POSIX_ACL
214 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
216 goto out_cleanup_inode;
219 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
221 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(dentry, 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 user_namespace *mnt_userns,
260 struct dentry *dentry,
264 return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false);
269 struct user_namespace *mnt_userns,
271 struct dentry *dentry,
275 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false);
280 struct user_namespace *mnt_userns,
282 struct dentry *dentry,
285 return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0,
289 STATIC struct dentry *
292 struct dentry *dentry,
296 struct xfs_inode *cip;
297 struct xfs_name name;
300 if (dentry->d_name.len >= MAXNAMELEN)
301 return ERR_PTR(-ENAMETOOLONG);
303 xfs_dentry_to_name(&name, dentry);
304 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
307 else if (likely(error == -ENOENT))
310 inode = ERR_PTR(error);
311 return d_splice_alias(inode, dentry);
314 STATIC struct dentry *
317 struct dentry *dentry,
320 struct xfs_inode *ip;
321 struct xfs_name xname;
322 struct xfs_name ci_name;
326 if (dentry->d_name.len >= MAXNAMELEN)
327 return ERR_PTR(-ENAMETOOLONG);
329 xfs_dentry_to_name(&xname, dentry);
330 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
331 if (unlikely(error)) {
332 if (unlikely(error != -ENOENT))
333 return ERR_PTR(error);
335 * call d_add(dentry, NULL) here when d_drop_negative_children
336 * is called in xfs_vn_mknod (ie. allow negative dentries
337 * with CI filesystems).
342 /* if exact match, just splice and exit */
344 return d_splice_alias(VFS_I(ip), dentry);
346 /* else case-insensitive match... */
347 dname.name = ci_name.name;
348 dname.len = ci_name.len;
349 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
350 kmem_free(ci_name.name);
356 struct dentry *old_dentry,
358 struct dentry *dentry)
360 struct inode *inode = d_inode(old_dentry);
361 struct xfs_name name;
364 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
368 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
373 d_instantiate(dentry, inode);
380 struct dentry *dentry)
382 struct xfs_name name;
385 xfs_dentry_to_name(&name, dentry);
387 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
392 * With unlink, the VFS makes the dentry "negative": no inode,
393 * but still hashed. This is incompatible with case-insensitive
394 * mode, so invalidate (unhash) the dentry in CI-mode.
396 if (xfs_has_asciici(XFS_M(dir->i_sb)))
397 d_invalidate(dentry);
403 struct user_namespace *mnt_userns,
405 struct dentry *dentry,
409 struct xfs_inode *cip = NULL;
410 struct xfs_name name;
415 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
416 error = xfs_dentry_mode_to_name(&name, dentry, mode);
420 error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip);
426 error = xfs_init_security(inode, dir, &dentry->d_name);
428 goto out_cleanup_inode;
432 d_instantiate(dentry, inode);
433 xfs_finish_inode_setup(cip);
437 xfs_finish_inode_setup(cip);
438 xfs_cleanup_inode(dir, inode, dentry);
446 struct user_namespace *mnt_userns,
448 struct dentry *odentry,
450 struct dentry *ndentry,
453 struct inode *new_inode = d_inode(ndentry);
456 struct xfs_name oname;
457 struct xfs_name nname;
459 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
462 /* if we are exchanging files, we need to set i_mode of both files */
463 if (flags & RENAME_EXCHANGE)
464 omode = d_inode(ndentry)->i_mode;
466 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
467 if (omode && unlikely(error))
470 error = xfs_dentry_mode_to_name(&nname, ndentry,
471 d_inode(odentry)->i_mode);
475 return xfs_rename(mnt_userns, XFS_I(odir), &oname,
476 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
477 new_inode ? XFS_I(new_inode) : NULL, flags);
481 * careful here - this function can get called recursively, so
482 * we need to be very careful about how much stack we use.
483 * uio is kmalloced for this reason...
487 struct dentry *dentry,
489 struct delayed_call *done)
495 return ERR_PTR(-ECHILD);
497 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
501 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
505 set_delayed_call(done, kfree_link, link);
511 return ERR_PTR(error);
515 xfs_vn_get_link_inline(
516 struct dentry *dentry,
518 struct delayed_call *done)
520 struct xfs_inode *ip = XFS_I(inode);
523 ASSERT(ip->i_df.if_format == XFS_DINODE_FMT_LOCAL);
526 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
529 link = ip->i_df.if_u1.if_data;
530 if (XFS_IS_CORRUPT(ip->i_mount, !link))
531 return ERR_PTR(-EFSCORRUPTED);
537 struct xfs_inode *ip)
539 struct xfs_mount *mp = ip->i_mount;
542 * If the file blocks are being allocated from a realtime volume, then
543 * always return the realtime extent size.
545 if (XFS_IS_REALTIME_INODE(ip))
546 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip));
549 * Allow large block sizes to be reported to userspace programs if the
550 * "largeio" mount option is used.
552 * If compatibility mode is specified, simply return the basic unit of
553 * caching so that we don't get inefficient read/modify/write I/O from
554 * user apps. Otherwise....
556 * If the underlying volume is a stripe, then return the stripe width in
557 * bytes as the recommended I/O size. It is not a stripe and we've set a
558 * default buffered I/O size, return that, otherwise return the compat
561 if (xfs_has_large_iosize(mp)) {
563 return XFS_FSB_TO_B(mp, mp->m_swidth);
564 if (xfs_has_allocsize(mp))
565 return 1U << mp->m_allocsize_log;
573 struct user_namespace *mnt_userns,
574 const struct path *path,
577 unsigned int query_flags)
579 struct inode *inode = d_inode(path->dentry);
580 struct xfs_inode *ip = XFS_I(inode);
581 struct xfs_mount *mp = ip->i_mount;
583 trace_xfs_getattr(ip);
585 if (xfs_is_shutdown(mp))
588 stat->size = XFS_ISIZE(ip);
589 stat->dev = inode->i_sb->s_dev;
590 stat->mode = inode->i_mode;
591 stat->nlink = inode->i_nlink;
592 stat->uid = i_uid_into_mnt(mnt_userns, inode);
593 stat->gid = i_gid_into_mnt(mnt_userns, inode);
594 stat->ino = ip->i_ino;
595 stat->atime = inode->i_atime;
596 stat->mtime = inode->i_mtime;
597 stat->ctime = inode->i_ctime;
598 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
600 if (xfs_has_v3inodes(mp)) {
601 if (request_mask & STATX_BTIME) {
602 stat->result_mask |= STATX_BTIME;
603 stat->btime = ip->i_crtime;
608 * Note: If you add another clause to set an attribute flag, please
609 * update attributes_mask below.
611 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
612 stat->attributes |= STATX_ATTR_IMMUTABLE;
613 if (ip->i_diflags & XFS_DIFLAG_APPEND)
614 stat->attributes |= STATX_ATTR_APPEND;
615 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
616 stat->attributes |= STATX_ATTR_NODUMP;
618 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
622 switch (inode->i_mode & S_IFMT) {
625 stat->blksize = BLKDEV_IOSIZE;
626 stat->rdev = inode->i_rdev;
629 stat->blksize = xfs_stat_blksize(ip);
639 struct xfs_inode *ip,
642 struct inode *inode = VFS_I(ip);
643 umode_t mode = iattr->ia_mode;
645 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
647 inode->i_mode &= S_IFMT;
648 inode->i_mode |= mode & ~S_IFMT;
653 struct xfs_inode *ip,
656 struct inode *inode = VFS_I(ip);
658 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
660 if (iattr->ia_valid & ATTR_ATIME)
661 inode->i_atime = iattr->ia_atime;
662 if (iattr->ia_valid & ATTR_CTIME)
663 inode->i_ctime = iattr->ia_ctime;
664 if (iattr->ia_valid & ATTR_MTIME)
665 inode->i_mtime = iattr->ia_mtime;
670 struct user_namespace *mnt_userns,
671 struct dentry *dentry,
674 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
676 if (xfs_is_readonly(mp))
679 if (xfs_is_shutdown(mp))
682 return setattr_prepare(mnt_userns, dentry, iattr);
686 * Set non-size attributes of an inode.
688 * Caution: The caller of this function is responsible for calling
689 * setattr_prepare() or otherwise verifying the change is fine.
693 struct user_namespace *mnt_userns,
694 struct xfs_inode *ip,
697 xfs_mount_t *mp = ip->i_mount;
698 struct inode *inode = VFS_I(ip);
699 int mask = iattr->ia_valid;
702 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
703 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
704 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
705 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
707 ASSERT((mask & ATTR_SIZE) == 0);
710 * If disk quotas is on, we make sure that the dquots do exist on disk,
711 * before we start any other transactions. Trying to do this later
712 * is messy. We don't care to take a readlock to look at the ids
713 * in inode here, because we can't hold it across the trans_reserve.
714 * If the IDs do change before we take the ilock, we're covered
715 * because the i_*dquot fields will get updated anyway.
717 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
720 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
722 qflags |= XFS_QMOPT_UQUOTA;
726 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
728 qflags |= XFS_QMOPT_GQUOTA;
734 * We take a reference when we initialize udqp and gdqp,
735 * so it is important that we never blindly double trip on
736 * the same variable. See xfs_create() for an example.
738 ASSERT(udqp == NULL);
739 ASSERT(gdqp == NULL);
740 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
741 qflags, &udqp, &gdqp, NULL);
746 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
747 capable(CAP_FOWNER), &tp);
752 * Change file ownership. Must be the owner or privileged.
754 if (mask & (ATTR_UID|ATTR_GID)) {
756 * These IDs could have changed since we last looked at them.
757 * But, we're assured that if the ownership did change
758 * while we didn't have the inode locked, inode's dquot(s)
759 * would have changed also.
763 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
764 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
767 * CAP_FSETID overrides the following restrictions:
769 * The set-user-ID and set-group-ID bits of a file will be
770 * cleared upon successful return from chown()
772 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
773 !capable(CAP_FSETID))
774 inode->i_mode &= ~(S_ISUID|S_ISGID);
777 * Change the ownerships and register quota modifications
778 * in the transaction.
780 if (!uid_eq(iuid, uid)) {
781 if (XFS_IS_UQUOTA_ON(mp)) {
782 ASSERT(mask & ATTR_UID);
784 olddquot1 = xfs_qm_vop_chown(tp, ip,
785 &ip->i_udquot, udqp);
789 if (!gid_eq(igid, gid)) {
790 if (XFS_IS_GQUOTA_ON(mp)) {
791 ASSERT(xfs_has_pquotino(mp) ||
792 !XFS_IS_PQUOTA_ON(mp));
793 ASSERT(mask & ATTR_GID);
795 olddquot2 = xfs_qm_vop_chown(tp, ip,
796 &ip->i_gdquot, gdqp);
802 if (mask & ATTR_MODE)
803 xfs_setattr_mode(ip, iattr);
804 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
805 xfs_setattr_time(ip, iattr);
807 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
809 XFS_STATS_INC(mp, xs_ig_attrchg);
811 if (xfs_has_wsync(mp))
812 xfs_trans_set_sync(tp);
813 error = xfs_trans_commit(tp);
816 * Release any dquot(s) the inode had kept before chown.
818 xfs_qm_dqrele(olddquot1);
819 xfs_qm_dqrele(olddquot2);
827 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
828 * update. We could avoid this with linked transactions
829 * and passing down the transaction pointer all the way
830 * to attr_set. No previous user of the generic
831 * Posix ACL code seems to care about this issue either.
833 if (mask & ATTR_MODE) {
834 error = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
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 user_namespace *mnt_userns,
856 struct xfs_inode *ip,
859 struct xfs_mount *mp = ip->i_mount;
860 struct inode *inode = VFS_I(ip);
861 xfs_off_t oldsize, newsize;
862 struct xfs_trans *tp;
865 bool did_zeroing = false;
867 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
868 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
869 ASSERT(S_ISREG(inode->i_mode));
870 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
871 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
873 oldsize = inode->i_size;
874 newsize = iattr->ia_size;
877 * Short circuit the truncate case for zero length files.
879 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
880 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
884 * Use the regular setattr path to update the timestamps.
886 iattr->ia_valid &= ~ATTR_SIZE;
887 return xfs_setattr_nonsize(mnt_userns, ip, iattr);
891 * Make sure that the dquots are attached to the inode.
893 error = xfs_qm_dqattach(ip);
898 * Wait for all direct I/O to complete.
900 inode_dio_wait(inode);
903 * File data changes must be complete before we start the transaction to
904 * modify the inode. This needs to be done before joining the inode to
905 * the transaction because the inode cannot be unlocked once it is a
906 * part of the transaction.
908 * Start with zeroing any data beyond EOF that we may expose on file
909 * extension, or zeroing out the rest of the block on a downward
912 if (newsize > oldsize) {
913 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
914 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
915 &did_zeroing, &xfs_buffered_write_iomap_ops);
918 * iomap won't detect a dirty page over an unwritten block (or a
919 * cow block over a hole) and subsequently skips zeroing the
920 * newly post-EOF portion of the page. Flush the new EOF to
921 * convert the block before the pagecache truncate.
923 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
927 error = iomap_truncate_page(inode, newsize, &did_zeroing,
928 &xfs_buffered_write_iomap_ops);
935 * We've already locked out new page faults, so now we can safely remove
936 * pages from the page cache knowing they won't get refaulted until we
937 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
938 * complete. The truncate_setsize() call also cleans partial EOF page
939 * PTEs on extending truncates and hence ensures sub-page block size
940 * filesystems are correctly handled, too.
942 * We have to do all the page cache truncate work outside the
943 * transaction context as the "lock" order is page lock->log space
944 * reservation as defined by extent allocation in the writeback path.
945 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
946 * having already truncated the in-memory version of the file (i.e. made
947 * user visible changes). There's not much we can do about this, except
948 * to hope that the caller sees ENOMEM and retries the truncate
951 * And we update in-core i_size and truncate page cache beyond newsize
952 * before writeback the [i_disk_size, newsize] range, so we're
953 * guaranteed not to write stale data past the new EOF on truncate down.
955 truncate_setsize(inode, newsize);
958 * We are going to log the inode size change in this transaction so
959 * any previous writes that are beyond the on disk EOF and the new
960 * EOF that have not been written out need to be written here. If we
961 * do not write the data out, we expose ourselves to the null files
962 * problem. Note that this includes any block zeroing we did above;
963 * otherwise those blocks may not be zeroed after a crash.
966 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
967 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
968 ip->i_disk_size, newsize - 1);
973 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
977 lock_flags |= XFS_ILOCK_EXCL;
978 xfs_ilock(ip, XFS_ILOCK_EXCL);
979 xfs_trans_ijoin(tp, ip, 0);
982 * Only change the c/mtime if we are changing the size or we are
983 * explicitly asked to change it. This handles the semantic difference
984 * between truncate() and ftruncate() as implemented in the VFS.
986 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
987 * special case where we need to update the times despite not having
988 * these flags set. For all other operations the VFS set these flags
989 * explicitly if it wants a timestamp update.
991 if (newsize != oldsize &&
992 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
993 iattr->ia_ctime = iattr->ia_mtime =
995 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
999 * The first thing we do is set the size to new_size permanently on
1000 * disk. This way we don't have to worry about anyone ever being able
1001 * to look at the data being freed even in the face of a crash.
1002 * What we're getting around here is the case where we free a block, it
1003 * is allocated to another file, it is written to, and then we crash.
1004 * If the new data gets written to the file but the log buffers
1005 * containing the free and reallocation don't, then we'd end up with
1006 * garbage in the blocks being freed. As long as we make the new size
1007 * permanent before actually freeing any blocks it doesn't matter if
1008 * they get written to.
1010 ip->i_disk_size = newsize;
1011 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1013 if (newsize <= oldsize) {
1014 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1016 goto out_trans_cancel;
1019 * Truncated "down", so we're removing references to old data
1020 * here - if we delay flushing for a long time, we expose
1021 * ourselves unduly to the notorious NULL files problem. So,
1022 * we mark this inode and flush it when the file is closed,
1023 * and do not wait the usual (long) time for writeout.
1025 xfs_iflags_set(ip, XFS_ITRUNCATED);
1027 /* A truncate down always removes post-EOF blocks. */
1028 xfs_inode_clear_eofblocks_tag(ip);
1031 if (iattr->ia_valid & ATTR_MODE)
1032 xfs_setattr_mode(ip, iattr);
1033 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1034 xfs_setattr_time(ip, iattr);
1036 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1038 XFS_STATS_INC(mp, xs_ig_attrchg);
1040 if (xfs_has_wsync(mp))
1041 xfs_trans_set_sync(tp);
1043 error = xfs_trans_commit(tp);
1046 xfs_iunlock(ip, lock_flags);
1050 xfs_trans_cancel(tp);
1055 xfs_vn_setattr_size(
1056 struct user_namespace *mnt_userns,
1057 struct dentry *dentry,
1058 struct iattr *iattr)
1060 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1063 trace_xfs_setattr(ip);
1065 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1068 return xfs_setattr_size(mnt_userns, ip, iattr);
1073 struct user_namespace *mnt_userns,
1074 struct dentry *dentry,
1075 struct iattr *iattr)
1077 struct inode *inode = d_inode(dentry);
1078 struct xfs_inode *ip = XFS_I(inode);
1081 if (iattr->ia_valid & ATTR_SIZE) {
1084 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1085 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1087 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1089 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1093 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr);
1094 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1096 trace_xfs_setattr(ip);
1098 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1100 error = xfs_setattr_nonsize(mnt_userns, ip, iattr);
1108 struct inode *inode,
1109 struct timespec64 *now,
1112 struct xfs_inode *ip = XFS_I(inode);
1113 struct xfs_mount *mp = ip->i_mount;
1114 int log_flags = XFS_ILOG_TIMESTAMP;
1115 struct xfs_trans *tp;
1118 trace_xfs_update_time(ip);
1120 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1121 if (!((flags & S_VERSION) &&
1122 inode_maybe_inc_iversion(inode, false)))
1123 return generic_update_time(inode, now, flags);
1125 /* Capture the iversion update that just occurred */
1126 log_flags |= XFS_ILOG_CORE;
1129 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1133 xfs_ilock(ip, XFS_ILOCK_EXCL);
1134 if (flags & S_CTIME)
1135 inode->i_ctime = *now;
1136 if (flags & S_MTIME)
1137 inode->i_mtime = *now;
1138 if (flags & S_ATIME)
1139 inode->i_atime = *now;
1141 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1142 xfs_trans_log_inode(tp, ip, log_flags);
1143 return xfs_trans_commit(tp);
1148 struct inode *inode,
1149 struct fiemap_extent_info *fieinfo,
1155 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1156 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1157 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1158 error = iomap_fiemap(inode, fieinfo, start, length,
1159 &xfs_xattr_iomap_ops);
1161 error = iomap_fiemap(inode, fieinfo, start, length,
1162 &xfs_read_iomap_ops);
1164 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1171 struct user_namespace *mnt_userns,
1173 struct dentry *dentry,
1176 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true);
1179 static const struct inode_operations xfs_inode_operations = {
1180 .get_acl = xfs_get_acl,
1181 .set_acl = xfs_set_acl,
1182 .getattr = xfs_vn_getattr,
1183 .setattr = xfs_vn_setattr,
1184 .listxattr = xfs_vn_listxattr,
1185 .fiemap = xfs_vn_fiemap,
1186 .update_time = xfs_vn_update_time,
1187 .fileattr_get = xfs_fileattr_get,
1188 .fileattr_set = xfs_fileattr_set,
1191 static const struct inode_operations xfs_dir_inode_operations = {
1192 .create = xfs_vn_create,
1193 .lookup = xfs_vn_lookup,
1194 .link = xfs_vn_link,
1195 .unlink = xfs_vn_unlink,
1196 .symlink = xfs_vn_symlink,
1197 .mkdir = xfs_vn_mkdir,
1199 * Yes, XFS uses the same method for rmdir and unlink.
1201 * There are some subtile differences deeper in the code,
1202 * but we use S_ISDIR to check for those.
1204 .rmdir = xfs_vn_unlink,
1205 .mknod = xfs_vn_mknod,
1206 .rename = xfs_vn_rename,
1207 .get_acl = xfs_get_acl,
1208 .set_acl = xfs_set_acl,
1209 .getattr = xfs_vn_getattr,
1210 .setattr = xfs_vn_setattr,
1211 .listxattr = xfs_vn_listxattr,
1212 .update_time = xfs_vn_update_time,
1213 .tmpfile = xfs_vn_tmpfile,
1214 .fileattr_get = xfs_fileattr_get,
1215 .fileattr_set = xfs_fileattr_set,
1218 static const struct inode_operations xfs_dir_ci_inode_operations = {
1219 .create = xfs_vn_create,
1220 .lookup = xfs_vn_ci_lookup,
1221 .link = xfs_vn_link,
1222 .unlink = xfs_vn_unlink,
1223 .symlink = xfs_vn_symlink,
1224 .mkdir = xfs_vn_mkdir,
1226 * Yes, XFS uses the same method for rmdir and unlink.
1228 * There are some subtile differences deeper in the code,
1229 * but we use S_ISDIR to check for those.
1231 .rmdir = xfs_vn_unlink,
1232 .mknod = xfs_vn_mknod,
1233 .rename = xfs_vn_rename,
1234 .get_acl = xfs_get_acl,
1235 .set_acl = xfs_set_acl,
1236 .getattr = xfs_vn_getattr,
1237 .setattr = xfs_vn_setattr,
1238 .listxattr = xfs_vn_listxattr,
1239 .update_time = xfs_vn_update_time,
1240 .tmpfile = xfs_vn_tmpfile,
1241 .fileattr_get = xfs_fileattr_get,
1242 .fileattr_set = xfs_fileattr_set,
1245 static const struct inode_operations xfs_symlink_inode_operations = {
1246 .get_link = xfs_vn_get_link,
1247 .getattr = xfs_vn_getattr,
1248 .setattr = xfs_vn_setattr,
1249 .listxattr = xfs_vn_listxattr,
1250 .update_time = xfs_vn_update_time,
1253 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1254 .get_link = xfs_vn_get_link_inline,
1255 .getattr = xfs_vn_getattr,
1256 .setattr = xfs_vn_setattr,
1257 .listxattr = xfs_vn_listxattr,
1258 .update_time = xfs_vn_update_time,
1261 /* Figure out if this file actually supports DAX. */
1263 xfs_inode_supports_dax(
1264 struct xfs_inode *ip)
1266 struct xfs_mount *mp = ip->i_mount;
1268 /* Only supported on regular files. */
1269 if (!S_ISREG(VFS_I(ip)->i_mode))
1272 /* Only supported on non-reflinked files. */
1273 if (xfs_is_reflink_inode(ip))
1276 /* Block size must match page size */
1277 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1280 /* Device has to support DAX too. */
1281 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1285 xfs_inode_should_enable_dax(
1286 struct xfs_inode *ip)
1288 if (!IS_ENABLED(CONFIG_FS_DAX))
1290 if (xfs_has_dax_never(ip->i_mount))
1292 if (!xfs_inode_supports_dax(ip))
1294 if (xfs_has_dax_always(ip->i_mount))
1296 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1302 xfs_diflags_to_iflags(
1303 struct xfs_inode *ip,
1306 struct inode *inode = VFS_I(ip);
1307 unsigned int xflags = xfs_ip2xflags(ip);
1308 unsigned int flags = 0;
1310 ASSERT(!(IS_DAX(inode) && init));
1312 if (xflags & FS_XFLAG_IMMUTABLE)
1313 flags |= S_IMMUTABLE;
1314 if (xflags & FS_XFLAG_APPEND)
1316 if (xflags & FS_XFLAG_SYNC)
1318 if (xflags & FS_XFLAG_NOATIME)
1320 if (init && xfs_inode_should_enable_dax(ip))
1324 * S_DAX can only be set during inode initialization and is never set by
1325 * the VFS, so we cannot mask off S_DAX in i_flags.
1327 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1328 inode->i_flags |= flags;
1332 * Initialize the Linux inode.
1334 * When reading existing inodes from disk this is called directly from xfs_iget,
1335 * when creating a new inode it is called from xfs_ialloc after setting up the
1336 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1337 * it up to the caller to deal with unlocking the inode appropriately.
1341 struct xfs_inode *ip)
1343 struct inode *inode = &ip->i_vnode;
1346 inode->i_ino = ip->i_ino;
1347 inode->i_state |= I_NEW;
1349 inode_sb_list_add(inode);
1350 /* make the inode look hashed for the writeback code */
1351 inode_fake_hash(inode);
1353 i_size_write(inode, ip->i_disk_size);
1354 xfs_diflags_to_iflags(ip, true);
1356 if (S_ISDIR(inode->i_mode)) {
1358 * We set the i_rwsem class here to avoid potential races with
1359 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1360 * after a filehandle lookup has already found the inode in
1361 * cache before it has been unlocked via unlock_new_inode().
1363 lockdep_set_class(&inode->i_rwsem,
1364 &inode->i_sb->s_type->i_mutex_dir_key);
1365 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1367 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1371 * Ensure all page cache allocations are done from GFP_NOFS context to
1372 * prevent direct reclaim recursion back into the filesystem and blowing
1373 * stacks or deadlocking.
1375 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1376 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1379 * If there is no attribute fork no ACL can exist on this inode,
1380 * and it can't have any file capabilities attached to it either.
1382 if (!XFS_IFORK_Q(ip)) {
1383 inode_has_no_xattr(inode);
1384 cache_no_acl(inode);
1390 struct xfs_inode *ip)
1392 struct inode *inode = &ip->i_vnode;
1394 switch (inode->i_mode & S_IFMT) {
1396 inode->i_op = &xfs_inode_operations;
1397 inode->i_fop = &xfs_file_operations;
1399 inode->i_mapping->a_ops = &xfs_dax_aops;
1401 inode->i_mapping->a_ops = &xfs_address_space_operations;
1404 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1405 inode->i_op = &xfs_dir_ci_inode_operations;
1407 inode->i_op = &xfs_dir_inode_operations;
1408 inode->i_fop = &xfs_dir_file_operations;
1411 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL)
1412 inode->i_op = &xfs_inline_symlink_inode_operations;
1414 inode->i_op = &xfs_symlink_inode_operations;
1417 inode->i_op = &xfs_inode_operations;
1418 init_special_inode(inode, inode->i_mode, inode->i_rdev);