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"
24 #include <linux/xattr.h>
25 #include <linux/posix_acl.h>
26 #include <linux/security.h>
27 #include <linux/iversion.h>
30 * Directories have different lock order w.r.t. mmap_sem compared to regular
31 * files. This is due to readdir potentially triggering page faults on a user
32 * buffer inside filldir(), and this happens with the ilock on the directory
33 * held. For regular files, the lock order is the other way around - the
34 * mmap_sem is taken during the page fault, and then we lock the ilock to do
35 * block mapping. Hence we need a different class for the directory ilock so
36 * that lockdep can tell them apart.
38 static struct lock_class_key xfs_nondir_ilock_class;
39 static struct lock_class_key xfs_dir_ilock_class;
44 const struct xattr *xattr_array,
47 const struct xattr *xattr;
48 struct xfs_inode *ip = XFS_I(inode);
51 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
52 error = xfs_attr_set(ip, xattr->name, xattr->value,
53 xattr->value_len, ATTR_SECURE);
61 * Hook in SELinux. This is not quite correct yet, what we really need
62 * here (as we do for default ACLs) is a mechanism by which creation of
63 * these attrs can be journalled at inode creation time (along with the
64 * inode, of course, such that log replay can't cause these to be lost).
71 const struct qstr *qstr)
73 return security_inode_init_security(inode, dir, qstr,
74 &xfs_initxattrs, NULL);
79 struct xfs_name *namep,
80 struct dentry *dentry)
82 namep->name = dentry->d_name.name;
83 namep->len = dentry->d_name.len;
84 namep->type = XFS_DIR3_FT_UNKNOWN;
88 xfs_dentry_mode_to_name(
89 struct xfs_name *namep,
90 struct dentry *dentry,
93 namep->name = dentry->d_name.name;
94 namep->len = dentry->d_name.len;
95 namep->type = xfs_mode_to_ftype(mode);
97 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
107 struct dentry *dentry)
109 struct xfs_name teardown;
112 * If we can't add the ACL or we fail in
113 * xfs_init_security we must back out.
114 * ENOSPC can hit here, among other things.
116 xfs_dentry_to_name(&teardown, dentry);
118 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
124 struct dentry *dentry,
127 bool tmpfile) /* unnamed file */
130 struct xfs_inode *ip = NULL;
131 struct posix_acl *default_acl, *acl;
132 struct xfs_name name;
136 * Irix uses Missed'em'V split, but doesn't want to see
137 * the upper 5 bits of (14bit) major.
139 if (S_ISCHR(mode) || S_ISBLK(mode)) {
140 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
146 error = posix_acl_create(dir, &mode, &default_acl, &acl);
150 /* Verify mode is valid also for tmpfile case */
151 error = xfs_dentry_mode_to_name(&name, dentry, mode);
156 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
158 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
165 error = xfs_init_security(inode, dir, &dentry->d_name);
167 goto out_cleanup_inode;
169 #ifdef CONFIG_XFS_POSIX_ACL
171 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
173 goto out_cleanup_inode;
176 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
178 goto out_cleanup_inode;
186 * The VFS requires that any inode fed to d_tmpfile must have
187 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
188 * However, we created the temp file with nlink == 0 because
189 * we're not allowed to put an inode with nlink > 0 on the
190 * unlinked list. Therefore we have to set nlink to 1 so that
191 * d_tmpfile can immediately set it back to zero.
194 d_tmpfile(dentry, inode);
196 d_instantiate(dentry, inode);
198 xfs_finish_inode_setup(ip);
202 posix_acl_release(default_acl);
204 posix_acl_release(acl);
208 xfs_finish_inode_setup(ip);
210 xfs_cleanup_inode(dir, inode, dentry);
218 struct dentry *dentry,
222 return xfs_generic_create(dir, dentry, mode, rdev, false);
228 struct dentry *dentry,
232 return xfs_vn_mknod(dir, dentry, mode, 0);
238 struct dentry *dentry,
241 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
244 STATIC struct dentry *
247 struct dentry *dentry,
251 struct xfs_inode *cip;
252 struct xfs_name name;
255 if (dentry->d_name.len >= MAXNAMELEN)
256 return ERR_PTR(-ENAMETOOLONG);
258 xfs_dentry_to_name(&name, dentry);
259 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
262 else if (likely(error == -ENOENT))
265 inode = ERR_PTR(error);
266 return d_splice_alias(inode, dentry);
269 STATIC struct dentry *
272 struct dentry *dentry,
275 struct xfs_inode *ip;
276 struct xfs_name xname;
277 struct xfs_name ci_name;
281 if (dentry->d_name.len >= MAXNAMELEN)
282 return ERR_PTR(-ENAMETOOLONG);
284 xfs_dentry_to_name(&xname, dentry);
285 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
286 if (unlikely(error)) {
287 if (unlikely(error != -ENOENT))
288 return ERR_PTR(error);
290 * call d_add(dentry, NULL) here when d_drop_negative_children
291 * is called in xfs_vn_mknod (ie. allow negative dentries
292 * with CI filesystems).
297 /* if exact match, just splice and exit */
299 return d_splice_alias(VFS_I(ip), dentry);
301 /* else case-insensitive match... */
302 dname.name = ci_name.name;
303 dname.len = ci_name.len;
304 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
305 kmem_free(ci_name.name);
311 struct dentry *old_dentry,
313 struct dentry *dentry)
315 struct inode *inode = d_inode(old_dentry);
316 struct xfs_name name;
319 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
323 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
328 d_instantiate(dentry, inode);
335 struct dentry *dentry)
337 struct xfs_name name;
340 xfs_dentry_to_name(&name, dentry);
342 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
347 * With unlink, the VFS makes the dentry "negative": no inode,
348 * but still hashed. This is incompatible with case-insensitive
349 * mode, so invalidate (unhash) the dentry in CI-mode.
351 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
352 d_invalidate(dentry);
359 struct dentry *dentry,
363 struct xfs_inode *cip = NULL;
364 struct xfs_name name;
369 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
370 error = xfs_dentry_mode_to_name(&name, dentry, mode);
374 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
380 error = xfs_init_security(inode, dir, &dentry->d_name);
382 goto out_cleanup_inode;
386 d_instantiate(dentry, inode);
387 xfs_finish_inode_setup(cip);
391 xfs_finish_inode_setup(cip);
392 xfs_cleanup_inode(dir, inode, dentry);
401 struct dentry *odentry,
403 struct dentry *ndentry,
406 struct inode *new_inode = d_inode(ndentry);
409 struct xfs_name oname;
410 struct xfs_name nname;
412 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
415 /* if we are exchanging files, we need to set i_mode of both files */
416 if (flags & RENAME_EXCHANGE)
417 omode = d_inode(ndentry)->i_mode;
419 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
420 if (omode && unlikely(error))
423 error = xfs_dentry_mode_to_name(&nname, ndentry,
424 d_inode(odentry)->i_mode);
428 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
430 new_inode ? XFS_I(new_inode) : NULL, flags);
434 * careful here - this function can get called recursively, so
435 * we need to be very careful about how much stack we use.
436 * uio is kmalloced for this reason...
440 struct dentry *dentry,
442 struct delayed_call *done)
448 return ERR_PTR(-ECHILD);
450 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
454 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
458 set_delayed_call(done, kfree_link, link);
464 return ERR_PTR(error);
468 xfs_vn_get_link_inline(
469 struct dentry *dentry,
471 struct delayed_call *done)
475 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
478 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
481 link = XFS_I(inode)->i_df.if_u1.if_data;
483 return ERR_PTR(-EFSCORRUPTED);
489 const struct path *path,
492 unsigned int query_flags)
494 struct inode *inode = d_inode(path->dentry);
495 struct xfs_inode *ip = XFS_I(inode);
496 struct xfs_mount *mp = ip->i_mount;
498 trace_xfs_getattr(ip);
500 if (XFS_FORCED_SHUTDOWN(mp))
503 stat->size = XFS_ISIZE(ip);
504 stat->dev = inode->i_sb->s_dev;
505 stat->mode = inode->i_mode;
506 stat->nlink = inode->i_nlink;
507 stat->uid = inode->i_uid;
508 stat->gid = inode->i_gid;
509 stat->ino = ip->i_ino;
510 stat->atime = inode->i_atime;
511 stat->mtime = inode->i_mtime;
512 stat->ctime = inode->i_ctime;
514 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
516 if (ip->i_d.di_version == 3) {
517 if (request_mask & STATX_BTIME) {
518 stat->result_mask |= STATX_BTIME;
519 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
520 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
525 * Note: If you add another clause to set an attribute flag, please
526 * update attributes_mask below.
528 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
529 stat->attributes |= STATX_ATTR_IMMUTABLE;
530 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
531 stat->attributes |= STATX_ATTR_APPEND;
532 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
533 stat->attributes |= STATX_ATTR_NODUMP;
535 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
539 switch (inode->i_mode & S_IFMT) {
542 stat->blksize = BLKDEV_IOSIZE;
543 stat->rdev = inode->i_rdev;
546 if (XFS_IS_REALTIME_INODE(ip)) {
548 * If the file blocks are being allocated from a
549 * realtime volume, then return the inode's realtime
550 * extent size or the realtime volume's extent size.
553 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
555 stat->blksize = xfs_preferred_iosize(mp);
565 struct xfs_inode *ip,
568 struct inode *inode = VFS_I(ip);
569 umode_t mode = iattr->ia_mode;
571 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
573 inode->i_mode &= S_IFMT;
574 inode->i_mode |= mode & ~S_IFMT;
579 struct xfs_inode *ip,
582 struct inode *inode = VFS_I(ip);
584 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
586 if (iattr->ia_valid & ATTR_ATIME)
587 inode->i_atime = iattr->ia_atime;
588 if (iattr->ia_valid & ATTR_CTIME)
589 inode->i_ctime = iattr->ia_ctime;
590 if (iattr->ia_valid & ATTR_MTIME)
591 inode->i_mtime = iattr->ia_mtime;
596 struct dentry *dentry,
599 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
601 if (mp->m_flags & XFS_MOUNT_RDONLY)
604 if (XFS_FORCED_SHUTDOWN(mp))
607 return setattr_prepare(dentry, iattr);
611 * Set non-size attributes of an inode.
613 * Caution: The caller of this function is responsible for calling
614 * setattr_prepare() or otherwise verifying the change is fine.
618 struct xfs_inode *ip,
622 xfs_mount_t *mp = ip->i_mount;
623 struct inode *inode = VFS_I(ip);
624 int mask = iattr->ia_valid;
627 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
628 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
629 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
630 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
632 ASSERT((mask & ATTR_SIZE) == 0);
635 * If disk quotas is on, we make sure that the dquots do exist on disk,
636 * before we start any other transactions. Trying to do this later
637 * is messy. We don't care to take a readlock to look at the ids
638 * in inode here, because we can't hold it across the trans_reserve.
639 * If the IDs do change before we take the ilock, we're covered
640 * because the i_*dquot fields will get updated anyway.
642 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
645 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
647 qflags |= XFS_QMOPT_UQUOTA;
651 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
653 qflags |= XFS_QMOPT_GQUOTA;
659 * We take a reference when we initialize udqp and gdqp,
660 * so it is important that we never blindly double trip on
661 * the same variable. See xfs_create() for an example.
663 ASSERT(udqp == NULL);
664 ASSERT(gdqp == NULL);
665 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
666 xfs_kgid_to_gid(gid),
668 qflags, &udqp, &gdqp, NULL);
673 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
677 xfs_ilock(ip, XFS_ILOCK_EXCL);
678 xfs_trans_ijoin(tp, ip, 0);
681 * Change file ownership. Must be the owner or privileged.
683 if (mask & (ATTR_UID|ATTR_GID)) {
685 * These IDs could have changed since we last looked at them.
686 * But, we're assured that if the ownership did change
687 * while we didn't have the inode locked, inode's dquot(s)
688 * would have changed also.
692 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
693 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
696 * Do a quota reservation only if uid/gid is actually
699 if (XFS_IS_QUOTA_RUNNING(mp) &&
700 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
701 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
703 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
704 NULL, capable(CAP_FOWNER) ?
705 XFS_QMOPT_FORCE_RES : 0);
706 if (error) /* out of quota */
712 * Change file ownership. Must be the owner or privileged.
714 if (mask & (ATTR_UID|ATTR_GID)) {
716 * CAP_FSETID overrides the following restrictions:
718 * The set-user-ID and set-group-ID bits of a file will be
719 * cleared upon successful return from chown()
721 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
722 !capable(CAP_FSETID))
723 inode->i_mode &= ~(S_ISUID|S_ISGID);
726 * Change the ownerships and register quota modifications
727 * in the transaction.
729 if (!uid_eq(iuid, uid)) {
730 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
731 ASSERT(mask & ATTR_UID);
733 olddquot1 = xfs_qm_vop_chown(tp, ip,
734 &ip->i_udquot, udqp);
736 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
739 if (!gid_eq(igid, gid)) {
740 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
741 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
742 !XFS_IS_PQUOTA_ON(mp));
743 ASSERT(mask & ATTR_GID);
745 olddquot2 = xfs_qm_vop_chown(tp, ip,
746 &ip->i_gdquot, gdqp);
748 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
753 if (mask & ATTR_MODE)
754 xfs_setattr_mode(ip, iattr);
755 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
756 xfs_setattr_time(ip, iattr);
758 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
760 XFS_STATS_INC(mp, xs_ig_attrchg);
762 if (mp->m_flags & XFS_MOUNT_WSYNC)
763 xfs_trans_set_sync(tp);
764 error = xfs_trans_commit(tp);
766 xfs_iunlock(ip, XFS_ILOCK_EXCL);
769 * Release any dquot(s) the inode had kept before chown.
771 xfs_qm_dqrele(olddquot1);
772 xfs_qm_dqrele(olddquot2);
780 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
781 * update. We could avoid this with linked transactions
782 * and passing down the transaction pointer all the way
783 * to attr_set. No previous user of the generic
784 * Posix ACL code seems to care about this issue either.
786 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
787 error = posix_acl_chmod(inode, inode->i_mode);
795 xfs_trans_cancel(tp);
796 xfs_iunlock(ip, XFS_ILOCK_EXCL);
804 xfs_vn_setattr_nonsize(
805 struct dentry *dentry,
808 struct xfs_inode *ip = XFS_I(d_inode(dentry));
811 trace_xfs_setattr(ip);
813 error = xfs_vn_change_ok(dentry, iattr);
816 return xfs_setattr_nonsize(ip, iattr, 0);
820 * Truncate file. Must have write permission and not be a directory.
822 * Caution: The caller of this function is responsible for calling
823 * setattr_prepare() or otherwise verifying the change is fine.
827 struct xfs_inode *ip,
830 struct xfs_mount *mp = ip->i_mount;
831 struct inode *inode = VFS_I(ip);
832 xfs_off_t oldsize, newsize;
833 struct xfs_trans *tp;
836 bool did_zeroing = false;
838 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
839 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
840 ASSERT(S_ISREG(inode->i_mode));
841 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
842 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
844 oldsize = inode->i_size;
845 newsize = iattr->ia_size;
848 * Short circuit the truncate case for zero length files.
850 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
851 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
855 * Use the regular setattr path to update the timestamps.
857 iattr->ia_valid &= ~ATTR_SIZE;
858 return xfs_setattr_nonsize(ip, iattr, 0);
862 * Make sure that the dquots are attached to the inode.
864 error = xfs_qm_dqattach(ip);
869 * Wait for all direct I/O to complete.
871 inode_dio_wait(inode);
874 * File data changes must be complete before we start the transaction to
875 * modify the inode. This needs to be done before joining the inode to
876 * the transaction because the inode cannot be unlocked once it is a
877 * part of the transaction.
879 * Start with zeroing any data beyond EOF that we may expose on file
880 * extension, or zeroing out the rest of the block on a downward
883 if (newsize > oldsize) {
884 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
885 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
886 &did_zeroing, &xfs_iomap_ops);
888 error = iomap_truncate_page(inode, newsize, &did_zeroing,
896 * We've already locked out new page faults, so now we can safely remove
897 * pages from the page cache knowing they won't get refaulted until we
898 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
899 * complete. The truncate_setsize() call also cleans partial EOF page
900 * PTEs on extending truncates and hence ensures sub-page block size
901 * filesystems are correctly handled, too.
903 * We have to do all the page cache truncate work outside the
904 * transaction context as the "lock" order is page lock->log space
905 * reservation as defined by extent allocation in the writeback path.
906 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
907 * having already truncated the in-memory version of the file (i.e. made
908 * user visible changes). There's not much we can do about this, except
909 * to hope that the caller sees ENOMEM and retries the truncate
912 * And we update in-core i_size and truncate page cache beyond newsize
913 * before writeback the [di_size, newsize] range, so we're guaranteed
914 * not to write stale data past the new EOF on truncate down.
916 truncate_setsize(inode, newsize);
919 * We are going to log the inode size change in this transaction so
920 * any previous writes that are beyond the on disk EOF and the new
921 * EOF that have not been written out need to be written here. If we
922 * do not write the data out, we expose ourselves to the null files
923 * problem. Note that this includes any block zeroing we did above;
924 * otherwise those blocks may not be zeroed after a crash.
927 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
928 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
929 ip->i_d.di_size, newsize - 1);
934 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
938 lock_flags |= XFS_ILOCK_EXCL;
939 xfs_ilock(ip, XFS_ILOCK_EXCL);
940 xfs_trans_ijoin(tp, ip, 0);
943 * Only change the c/mtime if we are changing the size or we are
944 * explicitly asked to change it. This handles the semantic difference
945 * between truncate() and ftruncate() as implemented in the VFS.
947 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
948 * special case where we need to update the times despite not having
949 * these flags set. For all other operations the VFS set these flags
950 * explicitly if it wants a timestamp update.
952 if (newsize != oldsize &&
953 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
954 iattr->ia_ctime = iattr->ia_mtime =
956 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
960 * The first thing we do is set the size to new_size permanently on
961 * disk. This way we don't have to worry about anyone ever being able
962 * to look at the data being freed even in the face of a crash.
963 * What we're getting around here is the case where we free a block, it
964 * is allocated to another file, it is written to, and then we crash.
965 * If the new data gets written to the file but the log buffers
966 * containing the free and reallocation don't, then we'd end up with
967 * garbage in the blocks being freed. As long as we make the new size
968 * permanent before actually freeing any blocks it doesn't matter if
969 * they get written to.
971 ip->i_d.di_size = newsize;
972 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
974 if (newsize <= oldsize) {
975 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
977 goto out_trans_cancel;
980 * Truncated "down", so we're removing references to old data
981 * here - if we delay flushing for a long time, we expose
982 * ourselves unduly to the notorious NULL files problem. So,
983 * we mark this inode and flush it when the file is closed,
984 * and do not wait the usual (long) time for writeout.
986 xfs_iflags_set(ip, XFS_ITRUNCATED);
988 /* A truncate down always removes post-EOF blocks. */
989 xfs_inode_clear_eofblocks_tag(ip);
992 if (iattr->ia_valid & ATTR_MODE)
993 xfs_setattr_mode(ip, iattr);
994 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
995 xfs_setattr_time(ip, iattr);
997 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
999 XFS_STATS_INC(mp, xs_ig_attrchg);
1001 if (mp->m_flags & XFS_MOUNT_WSYNC)
1002 xfs_trans_set_sync(tp);
1004 error = xfs_trans_commit(tp);
1007 xfs_iunlock(ip, lock_flags);
1011 xfs_trans_cancel(tp);
1016 xfs_vn_setattr_size(
1017 struct dentry *dentry,
1018 struct iattr *iattr)
1020 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1023 trace_xfs_setattr(ip);
1025 error = xfs_vn_change_ok(dentry, iattr);
1028 return xfs_setattr_size(ip, iattr);
1033 struct dentry *dentry,
1034 struct iattr *iattr)
1038 if (iattr->ia_valid & ATTR_SIZE) {
1039 struct inode *inode = d_inode(dentry);
1040 struct xfs_inode *ip = XFS_I(inode);
1043 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1044 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1046 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1048 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1052 error = xfs_vn_setattr_size(dentry, iattr);
1053 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1055 error = xfs_vn_setattr_nonsize(dentry, iattr);
1063 struct inode *inode,
1064 struct timespec64 *now,
1067 struct xfs_inode *ip = XFS_I(inode);
1068 struct xfs_mount *mp = ip->i_mount;
1069 int log_flags = XFS_ILOG_TIMESTAMP;
1070 struct xfs_trans *tp;
1073 trace_xfs_update_time(ip);
1075 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1076 if (!((flags & S_VERSION) &&
1077 inode_maybe_inc_iversion(inode, false)))
1078 return generic_update_time(inode, now, flags);
1080 /* Capture the iversion update that just occurred */
1081 log_flags |= XFS_ILOG_CORE;
1084 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1088 xfs_ilock(ip, XFS_ILOCK_EXCL);
1089 if (flags & S_CTIME)
1090 inode->i_ctime = *now;
1091 if (flags & S_MTIME)
1092 inode->i_mtime = *now;
1093 if (flags & S_ATIME)
1094 inode->i_atime = *now;
1096 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1097 xfs_trans_log_inode(tp, ip, log_flags);
1098 return xfs_trans_commit(tp);
1103 struct inode *inode,
1104 struct fiemap_extent_info *fieinfo,
1110 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1111 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1112 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1113 error = iomap_fiemap(inode, fieinfo, start, length,
1114 &xfs_xattr_iomap_ops);
1116 error = iomap_fiemap(inode, fieinfo, start, length,
1119 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1127 struct dentry *dentry,
1130 return xfs_generic_create(dir, dentry, mode, 0, true);
1133 static const struct inode_operations xfs_inode_operations = {
1134 .get_acl = xfs_get_acl,
1135 .set_acl = xfs_set_acl,
1136 .getattr = xfs_vn_getattr,
1137 .setattr = xfs_vn_setattr,
1138 .listxattr = xfs_vn_listxattr,
1139 .fiemap = xfs_vn_fiemap,
1140 .update_time = xfs_vn_update_time,
1143 static const struct inode_operations xfs_dir_inode_operations = {
1144 .create = xfs_vn_create,
1145 .lookup = xfs_vn_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_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,
1168 static const struct inode_operations xfs_dir_ci_inode_operations = {
1169 .create = xfs_vn_create,
1170 .lookup = xfs_vn_ci_lookup,
1171 .link = xfs_vn_link,
1172 .unlink = xfs_vn_unlink,
1173 .symlink = xfs_vn_symlink,
1174 .mkdir = xfs_vn_mkdir,
1176 * Yes, XFS uses the same method for rmdir and unlink.
1178 * There are some subtile differences deeper in the code,
1179 * but we use S_ISDIR to check for those.
1181 .rmdir = xfs_vn_unlink,
1182 .mknod = xfs_vn_mknod,
1183 .rename = xfs_vn_rename,
1184 .get_acl = xfs_get_acl,
1185 .set_acl = xfs_set_acl,
1186 .getattr = xfs_vn_getattr,
1187 .setattr = xfs_vn_setattr,
1188 .listxattr = xfs_vn_listxattr,
1189 .update_time = xfs_vn_update_time,
1190 .tmpfile = xfs_vn_tmpfile,
1193 static const struct inode_operations xfs_symlink_inode_operations = {
1194 .get_link = xfs_vn_get_link,
1195 .getattr = xfs_vn_getattr,
1196 .setattr = xfs_vn_setattr,
1197 .listxattr = xfs_vn_listxattr,
1198 .update_time = xfs_vn_update_time,
1201 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1202 .get_link = xfs_vn_get_link_inline,
1203 .getattr = xfs_vn_getattr,
1204 .setattr = xfs_vn_setattr,
1205 .listxattr = xfs_vn_listxattr,
1206 .update_time = xfs_vn_update_time,
1209 /* Figure out if this file actually supports DAX. */
1211 xfs_inode_supports_dax(
1212 struct xfs_inode *ip)
1214 struct xfs_mount *mp = ip->i_mount;
1216 /* Only supported on non-reflinked files. */
1217 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1220 /* DAX mount option or DAX iflag must be set. */
1221 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1222 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1225 /* Block size must match page size */
1226 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1229 /* Device has to support DAX too. */
1230 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1234 xfs_diflags_to_iflags(
1235 struct inode *inode,
1236 struct xfs_inode *ip)
1238 uint16_t flags = ip->i_d.di_flags;
1240 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1243 if (flags & XFS_DIFLAG_IMMUTABLE)
1244 inode->i_flags |= S_IMMUTABLE;
1245 if (flags & XFS_DIFLAG_APPEND)
1246 inode->i_flags |= S_APPEND;
1247 if (flags & XFS_DIFLAG_SYNC)
1248 inode->i_flags |= S_SYNC;
1249 if (flags & XFS_DIFLAG_NOATIME)
1250 inode->i_flags |= S_NOATIME;
1251 if (xfs_inode_supports_dax(ip))
1252 inode->i_flags |= S_DAX;
1256 * Initialize the Linux inode.
1258 * When reading existing inodes from disk this is called directly from xfs_iget,
1259 * when creating a new inode it is called from xfs_ialloc after setting up the
1260 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1261 * it up to the caller to deal with unlocking the inode appropriately.
1265 struct xfs_inode *ip)
1267 struct inode *inode = &ip->i_vnode;
1270 inode->i_ino = ip->i_ino;
1271 inode->i_state = I_NEW;
1273 inode_sb_list_add(inode);
1274 /* make the inode look hashed for the writeback code */
1275 inode_fake_hash(inode);
1277 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1278 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1280 i_size_write(inode, ip->i_d.di_size);
1281 xfs_diflags_to_iflags(inode, ip);
1283 if (S_ISDIR(inode->i_mode)) {
1285 * We set the i_rwsem class here to avoid potential races with
1286 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1287 * after a filehandle lookup has already found the inode in
1288 * cache before it has been unlocked via unlock_new_inode().
1290 lockdep_set_class(&inode->i_rwsem,
1291 &inode->i_sb->s_type->i_mutex_dir_key);
1292 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1293 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1295 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1296 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1300 * Ensure all page cache allocations are done from GFP_NOFS context to
1301 * prevent direct reclaim recursion back into the filesystem and blowing
1302 * stacks or deadlocking.
1304 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1305 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1308 * If there is no attribute fork no ACL can exist on this inode,
1309 * and it can't have any file capabilities attached to it either.
1311 if (!XFS_IFORK_Q(ip)) {
1312 inode_has_no_xattr(inode);
1313 cache_no_acl(inode);
1319 struct xfs_inode *ip)
1321 struct inode *inode = &ip->i_vnode;
1323 switch (inode->i_mode & S_IFMT) {
1325 inode->i_op = &xfs_inode_operations;
1326 inode->i_fop = &xfs_file_operations;
1328 inode->i_mapping->a_ops = &xfs_dax_aops;
1330 inode->i_mapping->a_ops = &xfs_address_space_operations;
1333 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1334 inode->i_op = &xfs_dir_ci_inode_operations;
1336 inode->i_op = &xfs_dir_inode_operations;
1337 inode->i_fop = &xfs_dir_file_operations;
1340 if (ip->i_df.if_flags & XFS_IFINLINE)
1341 inode->i_op = &xfs_inline_symlink_inode_operations;
1343 inode->i_op = &xfs_symlink_inode_operations;
1346 inode->i_op = &xfs_inode_operations;
1347 init_special_inode(inode, inode->i_mode, inode->i_rdev);