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_lock 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_lock 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);
209 posix_acl_release(default_acl);
210 posix_acl_release(acl);
214 xfs_finish_inode_setup(ip);
216 xfs_cleanup_inode(dir, inode, dentry);
224 struct dentry *dentry,
228 return xfs_generic_create(dir, dentry, mode, rdev, false);
234 struct dentry *dentry,
238 return xfs_generic_create(dir, dentry, mode, 0, false);
244 struct dentry *dentry,
247 return xfs_generic_create(dir, dentry, mode | S_IFDIR, 0, false);
250 STATIC struct dentry *
253 struct dentry *dentry,
257 struct xfs_inode *cip;
258 struct xfs_name name;
261 if (dentry->d_name.len >= MAXNAMELEN)
262 return ERR_PTR(-ENAMETOOLONG);
264 xfs_dentry_to_name(&name, dentry);
265 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
268 else if (likely(error == -ENOENT))
271 inode = ERR_PTR(error);
272 return d_splice_alias(inode, dentry);
275 STATIC struct dentry *
278 struct dentry *dentry,
281 struct xfs_inode *ip;
282 struct xfs_name xname;
283 struct xfs_name ci_name;
287 if (dentry->d_name.len >= MAXNAMELEN)
288 return ERR_PTR(-ENAMETOOLONG);
290 xfs_dentry_to_name(&xname, dentry);
291 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
292 if (unlikely(error)) {
293 if (unlikely(error != -ENOENT))
294 return ERR_PTR(error);
296 * call d_add(dentry, NULL) here when d_drop_negative_children
297 * is called in xfs_vn_mknod (ie. allow negative dentries
298 * with CI filesystems).
303 /* if exact match, just splice and exit */
305 return d_splice_alias(VFS_I(ip), dentry);
307 /* else case-insensitive match... */
308 dname.name = ci_name.name;
309 dname.len = ci_name.len;
310 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
311 kmem_free(ci_name.name);
317 struct dentry *old_dentry,
319 struct dentry *dentry)
321 struct inode *inode = d_inode(old_dentry);
322 struct xfs_name name;
325 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
329 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
334 d_instantiate(dentry, inode);
341 struct dentry *dentry)
343 struct xfs_name name;
346 xfs_dentry_to_name(&name, dentry);
348 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
353 * With unlink, the VFS makes the dentry "negative": no inode,
354 * but still hashed. This is incompatible with case-insensitive
355 * mode, so invalidate (unhash) the dentry in CI-mode.
357 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
358 d_invalidate(dentry);
365 struct dentry *dentry,
369 struct xfs_inode *cip = NULL;
370 struct xfs_name name;
375 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
376 error = xfs_dentry_mode_to_name(&name, dentry, mode);
380 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
386 error = xfs_init_security(inode, dir, &dentry->d_name);
388 goto out_cleanup_inode;
392 d_instantiate(dentry, inode);
393 xfs_finish_inode_setup(cip);
397 xfs_finish_inode_setup(cip);
398 xfs_cleanup_inode(dir, inode, dentry);
407 struct dentry *odentry,
409 struct dentry *ndentry,
412 struct inode *new_inode = d_inode(ndentry);
415 struct xfs_name oname;
416 struct xfs_name nname;
418 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
421 /* if we are exchanging files, we need to set i_mode of both files */
422 if (flags & RENAME_EXCHANGE)
423 omode = d_inode(ndentry)->i_mode;
425 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
426 if (omode && unlikely(error))
429 error = xfs_dentry_mode_to_name(&nname, ndentry,
430 d_inode(odentry)->i_mode);
434 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
436 new_inode ? XFS_I(new_inode) : NULL, flags);
440 * careful here - this function can get called recursively, so
441 * we need to be very careful about how much stack we use.
442 * uio is kmalloced for this reason...
446 struct dentry *dentry,
448 struct delayed_call *done)
454 return ERR_PTR(-ECHILD);
456 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
460 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
464 set_delayed_call(done, kfree_link, link);
470 return ERR_PTR(error);
474 xfs_vn_get_link_inline(
475 struct dentry *dentry,
477 struct delayed_call *done)
479 struct xfs_inode *ip = XFS_I(inode);
482 ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
485 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
488 link = ip->i_df.if_u1.if_data;
489 if (XFS_IS_CORRUPT(ip->i_mount, !link))
490 return ERR_PTR(-EFSCORRUPTED);
496 struct xfs_inode *ip)
498 struct xfs_mount *mp = ip->i_mount;
501 * If the file blocks are being allocated from a realtime volume, then
502 * always return the realtime extent size.
504 if (XFS_IS_REALTIME_INODE(ip))
505 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
508 * Allow large block sizes to be reported to userspace programs if the
509 * "largeio" mount option is used.
511 * If compatibility mode is specified, simply return the basic unit of
512 * caching so that we don't get inefficient read/modify/write I/O from
513 * user apps. Otherwise....
515 * If the underlying volume is a stripe, then return the stripe width in
516 * bytes as the recommended I/O size. It is not a stripe and we've set a
517 * default buffered I/O size, return that, otherwise return the compat
520 if (mp->m_flags & XFS_MOUNT_LARGEIO) {
522 return mp->m_swidth << mp->m_sb.sb_blocklog;
523 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
524 return 1U << mp->m_allocsize_log;
532 const struct path *path,
535 unsigned int query_flags)
537 struct inode *inode = d_inode(path->dentry);
538 struct xfs_inode *ip = XFS_I(inode);
539 struct xfs_mount *mp = ip->i_mount;
541 trace_xfs_getattr(ip);
543 if (XFS_FORCED_SHUTDOWN(mp))
546 stat->size = XFS_ISIZE(ip);
547 stat->dev = inode->i_sb->s_dev;
548 stat->mode = inode->i_mode;
549 stat->nlink = inode->i_nlink;
550 stat->uid = inode->i_uid;
551 stat->gid = inode->i_gid;
552 stat->ino = ip->i_ino;
553 stat->atime = inode->i_atime;
554 stat->mtime = inode->i_mtime;
555 stat->ctime = inode->i_ctime;
557 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
559 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
560 if (request_mask & STATX_BTIME) {
561 stat->result_mask |= STATX_BTIME;
562 stat->btime = ip->i_d.di_crtime;
567 * Note: If you add another clause to set an attribute flag, please
568 * update attributes_mask below.
570 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
571 stat->attributes |= STATX_ATTR_IMMUTABLE;
572 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
573 stat->attributes |= STATX_ATTR_APPEND;
574 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
575 stat->attributes |= STATX_ATTR_NODUMP;
577 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
581 switch (inode->i_mode & S_IFMT) {
584 stat->blksize = BLKDEV_IOSIZE;
585 stat->rdev = inode->i_rdev;
588 stat->blksize = xfs_stat_blksize(ip);
598 struct xfs_inode *ip,
601 struct inode *inode = VFS_I(ip);
602 umode_t mode = iattr->ia_mode;
604 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
606 inode->i_mode &= S_IFMT;
607 inode->i_mode |= mode & ~S_IFMT;
612 struct xfs_inode *ip,
615 struct inode *inode = VFS_I(ip);
617 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
619 if (iattr->ia_valid & ATTR_ATIME)
620 inode->i_atime = iattr->ia_atime;
621 if (iattr->ia_valid & ATTR_CTIME)
622 inode->i_ctime = iattr->ia_ctime;
623 if (iattr->ia_valid & ATTR_MTIME)
624 inode->i_mtime = iattr->ia_mtime;
629 struct dentry *dentry,
632 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
634 if (mp->m_flags & XFS_MOUNT_RDONLY)
637 if (XFS_FORCED_SHUTDOWN(mp))
640 return setattr_prepare(dentry, iattr);
644 * Set non-size attributes of an inode.
646 * Caution: The caller of this function is responsible for calling
647 * setattr_prepare() or otherwise verifying the change is fine.
651 struct xfs_inode *ip,
654 xfs_mount_t *mp = ip->i_mount;
655 struct inode *inode = VFS_I(ip);
656 int mask = iattr->ia_valid;
659 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
660 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
661 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
662 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
664 ASSERT((mask & ATTR_SIZE) == 0);
667 * If disk quotas is on, we make sure that the dquots do exist on disk,
668 * before we start any other transactions. Trying to do this later
669 * is messy. We don't care to take a readlock to look at the ids
670 * in inode here, because we can't hold it across the trans_reserve.
671 * If the IDs do change before we take the ilock, we're covered
672 * because the i_*dquot fields will get updated anyway.
674 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
677 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
679 qflags |= XFS_QMOPT_UQUOTA;
683 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
685 qflags |= XFS_QMOPT_GQUOTA;
691 * We take a reference when we initialize udqp and gdqp,
692 * so it is important that we never blindly double trip on
693 * the same variable. See xfs_create() for an example.
695 ASSERT(udqp == NULL);
696 ASSERT(gdqp == NULL);
697 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
698 qflags, &udqp, &gdqp, NULL);
703 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
707 xfs_ilock(ip, XFS_ILOCK_EXCL);
708 xfs_trans_ijoin(tp, ip, 0);
711 * Change file ownership. Must be the owner or privileged.
713 if (mask & (ATTR_UID|ATTR_GID)) {
715 * These IDs could have changed since we last looked at them.
716 * But, we're assured that if the ownership did change
717 * while we didn't have the inode locked, inode's dquot(s)
718 * would have changed also.
722 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
723 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
726 * Do a quota reservation only if uid/gid is actually
729 if (XFS_IS_QUOTA_RUNNING(mp) &&
730 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
731 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
733 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
734 NULL, capable(CAP_FOWNER) ?
735 XFS_QMOPT_FORCE_RES : 0);
736 if (error) /* out of quota */
741 * CAP_FSETID overrides the following restrictions:
743 * The set-user-ID and set-group-ID bits of a file will be
744 * cleared upon successful return from chown()
746 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
747 !capable(CAP_FSETID))
748 inode->i_mode &= ~(S_ISUID|S_ISGID);
751 * Change the ownerships and register quota modifications
752 * in the transaction.
754 if (!uid_eq(iuid, uid)) {
755 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
756 ASSERT(mask & ATTR_UID);
758 olddquot1 = xfs_qm_vop_chown(tp, ip,
759 &ip->i_udquot, udqp);
763 if (!gid_eq(igid, gid)) {
764 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
765 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
766 !XFS_IS_PQUOTA_ON(mp));
767 ASSERT(mask & ATTR_GID);
769 olddquot2 = xfs_qm_vop_chown(tp, ip,
770 &ip->i_gdquot, gdqp);
776 if (mask & ATTR_MODE)
777 xfs_setattr_mode(ip, iattr);
778 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
779 xfs_setattr_time(ip, iattr);
781 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
783 XFS_STATS_INC(mp, xs_ig_attrchg);
785 if (mp->m_flags & XFS_MOUNT_WSYNC)
786 xfs_trans_set_sync(tp);
787 error = xfs_trans_commit(tp);
789 xfs_iunlock(ip, XFS_ILOCK_EXCL);
792 * Release any dquot(s) the inode had kept before chown.
794 xfs_qm_dqrele(olddquot1);
795 xfs_qm_dqrele(olddquot2);
803 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
804 * update. We could avoid this with linked transactions
805 * and passing down the transaction pointer all the way
806 * to attr_set. No previous user of the generic
807 * Posix ACL code seems to care about this issue either.
809 if (mask & ATTR_MODE) {
810 error = posix_acl_chmod(inode, inode->i_mode);
818 xfs_trans_cancel(tp);
819 xfs_iunlock(ip, XFS_ILOCK_EXCL);
827 * Truncate file. Must have write permission and not be a directory.
829 * Caution: The caller of this function is responsible for calling
830 * setattr_prepare() or otherwise verifying the change is fine.
834 struct xfs_inode *ip,
837 struct xfs_mount *mp = ip->i_mount;
838 struct inode *inode = VFS_I(ip);
839 xfs_off_t oldsize, newsize;
840 struct xfs_trans *tp;
843 bool did_zeroing = false;
845 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
846 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
847 ASSERT(S_ISREG(inode->i_mode));
848 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
849 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
851 oldsize = inode->i_size;
852 newsize = iattr->ia_size;
855 * Short circuit the truncate case for zero length files.
857 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
858 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
862 * Use the regular setattr path to update the timestamps.
864 iattr->ia_valid &= ~ATTR_SIZE;
865 return xfs_setattr_nonsize(ip, iattr);
869 * Make sure that the dquots are attached to the inode.
871 error = xfs_qm_dqattach(ip);
876 * Wait for all direct I/O to complete.
878 inode_dio_wait(inode);
881 * File data changes must be complete before we start the transaction to
882 * modify the inode. This needs to be done before joining the inode to
883 * the transaction because the inode cannot be unlocked once it is a
884 * part of the transaction.
886 * Start with zeroing any data beyond EOF that we may expose on file
887 * extension, or zeroing out the rest of the block on a downward
890 if (newsize > oldsize) {
891 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
892 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
893 &did_zeroing, &xfs_buffered_write_iomap_ops);
896 * iomap won't detect a dirty page over an unwritten block (or a
897 * cow block over a hole) and subsequently skips zeroing the
898 * newly post-EOF portion of the page. Flush the new EOF to
899 * convert the block before the pagecache truncate.
901 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
905 error = iomap_truncate_page(inode, newsize, &did_zeroing,
906 &xfs_buffered_write_iomap_ops);
913 * We've already locked out new page faults, so now we can safely remove
914 * pages from the page cache knowing they won't get refaulted until we
915 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
916 * complete. The truncate_setsize() call also cleans partial EOF page
917 * PTEs on extending truncates and hence ensures sub-page block size
918 * filesystems are correctly handled, too.
920 * We have to do all the page cache truncate work outside the
921 * transaction context as the "lock" order is page lock->log space
922 * reservation as defined by extent allocation in the writeback path.
923 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
924 * having already truncated the in-memory version of the file (i.e. made
925 * user visible changes). There's not much we can do about this, except
926 * to hope that the caller sees ENOMEM and retries the truncate
929 * And we update in-core i_size and truncate page cache beyond newsize
930 * before writeback the [di_size, newsize] range, so we're guaranteed
931 * not to write stale data past the new EOF on truncate down.
933 truncate_setsize(inode, newsize);
936 * We are going to log the inode size change in this transaction so
937 * any previous writes that are beyond the on disk EOF and the new
938 * EOF that have not been written out need to be written here. If we
939 * do not write the data out, we expose ourselves to the null files
940 * problem. Note that this includes any block zeroing we did above;
941 * otherwise those blocks may not be zeroed after a crash.
944 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
945 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
946 ip->i_d.di_size, newsize - 1);
951 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
955 lock_flags |= XFS_ILOCK_EXCL;
956 xfs_ilock(ip, XFS_ILOCK_EXCL);
957 xfs_trans_ijoin(tp, ip, 0);
960 * Only change the c/mtime if we are changing the size or we are
961 * explicitly asked to change it. This handles the semantic difference
962 * between truncate() and ftruncate() as implemented in the VFS.
964 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
965 * special case where we need to update the times despite not having
966 * these flags set. For all other operations the VFS set these flags
967 * explicitly if it wants a timestamp update.
969 if (newsize != oldsize &&
970 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
971 iattr->ia_ctime = iattr->ia_mtime =
973 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
977 * The first thing we do is set the size to new_size permanently on
978 * disk. This way we don't have to worry about anyone ever being able
979 * to look at the data being freed even in the face of a crash.
980 * What we're getting around here is the case where we free a block, it
981 * is allocated to another file, it is written to, and then we crash.
982 * If the new data gets written to the file but the log buffers
983 * containing the free and reallocation don't, then we'd end up with
984 * garbage in the blocks being freed. As long as we make the new size
985 * permanent before actually freeing any blocks it doesn't matter if
986 * they get written to.
988 ip->i_d.di_size = newsize;
989 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
991 if (newsize <= oldsize) {
992 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
994 goto out_trans_cancel;
997 * Truncated "down", so we're removing references to old data
998 * here - if we delay flushing for a long time, we expose
999 * ourselves unduly to the notorious NULL files problem. So,
1000 * we mark this inode and flush it when the file is closed,
1001 * and do not wait the usual (long) time for writeout.
1003 xfs_iflags_set(ip, XFS_ITRUNCATED);
1005 /* A truncate down always removes post-EOF blocks. */
1006 xfs_inode_clear_eofblocks_tag(ip);
1009 if (iattr->ia_valid & ATTR_MODE)
1010 xfs_setattr_mode(ip, iattr);
1011 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1012 xfs_setattr_time(ip, iattr);
1014 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1016 XFS_STATS_INC(mp, xs_ig_attrchg);
1018 if (mp->m_flags & XFS_MOUNT_WSYNC)
1019 xfs_trans_set_sync(tp);
1021 error = xfs_trans_commit(tp);
1024 xfs_iunlock(ip, lock_flags);
1028 xfs_trans_cancel(tp);
1033 xfs_vn_setattr_size(
1034 struct dentry *dentry,
1035 struct iattr *iattr)
1037 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1040 trace_xfs_setattr(ip);
1042 error = xfs_vn_change_ok(dentry, iattr);
1045 return xfs_setattr_size(ip, iattr);
1050 struct dentry *dentry,
1051 struct iattr *iattr)
1053 struct inode *inode = d_inode(dentry);
1054 struct xfs_inode *ip = XFS_I(inode);
1057 if (iattr->ia_valid & ATTR_SIZE) {
1060 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1061 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1063 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1065 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1069 error = xfs_vn_setattr_size(dentry, iattr);
1070 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1072 trace_xfs_setattr(ip);
1074 error = xfs_vn_change_ok(dentry, iattr);
1076 error = xfs_setattr_nonsize(ip, iattr);
1084 struct inode *inode,
1085 struct timespec64 *now,
1088 struct xfs_inode *ip = XFS_I(inode);
1089 struct xfs_mount *mp = ip->i_mount;
1090 int log_flags = XFS_ILOG_TIMESTAMP;
1091 struct xfs_trans *tp;
1094 trace_xfs_update_time(ip);
1096 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1097 if (!((flags & S_VERSION) &&
1098 inode_maybe_inc_iversion(inode, false)))
1099 return generic_update_time(inode, now, flags);
1101 /* Capture the iversion update that just occurred */
1102 log_flags |= XFS_ILOG_CORE;
1105 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1109 xfs_ilock(ip, XFS_ILOCK_EXCL);
1110 if (flags & S_CTIME)
1111 inode->i_ctime = *now;
1112 if (flags & S_MTIME)
1113 inode->i_mtime = *now;
1114 if (flags & S_ATIME)
1115 inode->i_atime = *now;
1117 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1118 xfs_trans_log_inode(tp, ip, log_flags);
1119 return xfs_trans_commit(tp);
1124 struct inode *inode,
1125 struct fiemap_extent_info *fieinfo,
1131 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1132 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1133 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1134 error = iomap_fiemap(inode, fieinfo, start, length,
1135 &xfs_xattr_iomap_ops);
1137 error = iomap_fiemap(inode, fieinfo, start, length,
1138 &xfs_read_iomap_ops);
1140 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1148 struct dentry *dentry,
1151 return xfs_generic_create(dir, dentry, mode, 0, true);
1154 static const struct inode_operations xfs_inode_operations = {
1155 .get_acl = xfs_get_acl,
1156 .set_acl = xfs_set_acl,
1157 .getattr = xfs_vn_getattr,
1158 .setattr = xfs_vn_setattr,
1159 .listxattr = xfs_vn_listxattr,
1160 .fiemap = xfs_vn_fiemap,
1161 .update_time = xfs_vn_update_time,
1164 static const struct inode_operations xfs_dir_inode_operations = {
1165 .create = xfs_vn_create,
1166 .lookup = xfs_vn_lookup,
1167 .link = xfs_vn_link,
1168 .unlink = xfs_vn_unlink,
1169 .symlink = xfs_vn_symlink,
1170 .mkdir = xfs_vn_mkdir,
1172 * Yes, XFS uses the same method for rmdir and unlink.
1174 * There are some subtile differences deeper in the code,
1175 * but we use S_ISDIR to check for those.
1177 .rmdir = xfs_vn_unlink,
1178 .mknod = xfs_vn_mknod,
1179 .rename = xfs_vn_rename,
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 .update_time = xfs_vn_update_time,
1186 .tmpfile = xfs_vn_tmpfile,
1189 static const struct inode_operations xfs_dir_ci_inode_operations = {
1190 .create = xfs_vn_create,
1191 .lookup = xfs_vn_ci_lookup,
1192 .link = xfs_vn_link,
1193 .unlink = xfs_vn_unlink,
1194 .symlink = xfs_vn_symlink,
1195 .mkdir = xfs_vn_mkdir,
1197 * Yes, XFS uses the same method for rmdir and unlink.
1199 * There are some subtile differences deeper in the code,
1200 * but we use S_ISDIR to check for those.
1202 .rmdir = xfs_vn_unlink,
1203 .mknod = xfs_vn_mknod,
1204 .rename = xfs_vn_rename,
1205 .get_acl = xfs_get_acl,
1206 .set_acl = xfs_set_acl,
1207 .getattr = xfs_vn_getattr,
1208 .setattr = xfs_vn_setattr,
1209 .listxattr = xfs_vn_listxattr,
1210 .update_time = xfs_vn_update_time,
1211 .tmpfile = xfs_vn_tmpfile,
1214 static const struct inode_operations xfs_symlink_inode_operations = {
1215 .get_link = xfs_vn_get_link,
1216 .getattr = xfs_vn_getattr,
1217 .setattr = xfs_vn_setattr,
1218 .listxattr = xfs_vn_listxattr,
1219 .update_time = xfs_vn_update_time,
1222 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1223 .get_link = xfs_vn_get_link_inline,
1224 .getattr = xfs_vn_getattr,
1225 .setattr = xfs_vn_setattr,
1226 .listxattr = xfs_vn_listxattr,
1227 .update_time = xfs_vn_update_time,
1230 /* Figure out if this file actually supports DAX. */
1232 xfs_inode_supports_dax(
1233 struct xfs_inode *ip)
1235 struct xfs_mount *mp = ip->i_mount;
1237 /* Only supported on regular files. */
1238 if (!S_ISREG(VFS_I(ip)->i_mode))
1241 /* Only supported on non-reflinked files. */
1242 if (xfs_is_reflink_inode(ip))
1245 /* Block size must match page size */
1246 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1249 /* Device has to support DAX too. */
1250 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1254 xfs_inode_should_enable_dax(
1255 struct xfs_inode *ip)
1257 if (!IS_ENABLED(CONFIG_FS_DAX))
1259 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1261 if (!xfs_inode_supports_dax(ip))
1263 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1265 if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
1271 xfs_diflags_to_iflags(
1272 struct xfs_inode *ip,
1275 struct inode *inode = VFS_I(ip);
1276 unsigned int xflags = xfs_ip2xflags(ip);
1277 unsigned int flags = 0;
1279 ASSERT(!(IS_DAX(inode) && init));
1281 if (xflags & FS_XFLAG_IMMUTABLE)
1282 flags |= S_IMMUTABLE;
1283 if (xflags & FS_XFLAG_APPEND)
1285 if (xflags & FS_XFLAG_SYNC)
1287 if (xflags & FS_XFLAG_NOATIME)
1289 if (init && xfs_inode_should_enable_dax(ip))
1293 * S_DAX can only be set during inode initialization and is never set by
1294 * the VFS, so we cannot mask off S_DAX in i_flags.
1296 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1297 inode->i_flags |= flags;
1301 * Initialize the Linux inode.
1303 * When reading existing inodes from disk this is called directly from xfs_iget,
1304 * when creating a new inode it is called from xfs_ialloc after setting up the
1305 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1306 * it up to the caller to deal with unlocking the inode appropriately.
1310 struct xfs_inode *ip)
1312 struct inode *inode = &ip->i_vnode;
1315 inode->i_ino = ip->i_ino;
1316 inode->i_state = I_NEW;
1318 inode_sb_list_add(inode);
1319 /* make the inode look hashed for the writeback code */
1320 inode_fake_hash(inode);
1322 i_size_write(inode, ip->i_d.di_size);
1323 xfs_diflags_to_iflags(ip, true);
1325 if (S_ISDIR(inode->i_mode)) {
1327 * We set the i_rwsem class here to avoid potential races with
1328 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1329 * after a filehandle lookup has already found the inode in
1330 * cache before it has been unlocked via unlock_new_inode().
1332 lockdep_set_class(&inode->i_rwsem,
1333 &inode->i_sb->s_type->i_mutex_dir_key);
1334 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1336 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1340 * Ensure all page cache allocations are done from GFP_NOFS context to
1341 * prevent direct reclaim recursion back into the filesystem and blowing
1342 * stacks or deadlocking.
1344 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1345 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1348 * If there is no attribute fork no ACL can exist on this inode,
1349 * and it can't have any file capabilities attached to it either.
1351 if (!XFS_IFORK_Q(ip)) {
1352 inode_has_no_xattr(inode);
1353 cache_no_acl(inode);
1359 struct xfs_inode *ip)
1361 struct inode *inode = &ip->i_vnode;
1363 switch (inode->i_mode & S_IFMT) {
1365 inode->i_op = &xfs_inode_operations;
1366 inode->i_fop = &xfs_file_operations;
1368 inode->i_mapping->a_ops = &xfs_dax_aops;
1370 inode->i_mapping->a_ops = &xfs_address_space_operations;
1373 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1374 inode->i_op = &xfs_dir_ci_inode_operations;
1376 inode->i_op = &xfs_dir_inode_operations;
1377 inode->i_fop = &xfs_dir_file_operations;
1380 if (ip->i_df.if_flags & XFS_IFINLINE)
1381 inode->i_op = &xfs_inline_symlink_inode_operations;
1383 inode->i_op = &xfs_symlink_inode_operations;
1386 inode->i_op = &xfs_inode_operations;
1387 init_special_inode(inode, inode->i_mode, inode->i_rdev);