2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_inode.h"
28 #include "xfs_bmap_util.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_symlink.h"
37 #include "xfs_da_btree.h"
39 #include "xfs_trans_space.h"
41 #include "xfs_iomap.h"
43 #include <linux/capability.h>
44 #include <linux/xattr.h>
45 #include <linux/posix_acl.h>
46 #include <linux/security.h>
47 #include <linux/iomap.h>
48 #include <linux/slab.h>
51 * Directories have different lock order w.r.t. mmap_sem compared to regular
52 * files. This is due to readdir potentially triggering page faults on a user
53 * buffer inside filldir(), and this happens with the ilock on the directory
54 * held. For regular files, the lock order is the other way around - the
55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
56 * block mapping. Hence we need a different class for the directory ilock so
57 * that lockdep can tell them apart.
59 static struct lock_class_key xfs_nondir_ilock_class;
60 static struct lock_class_key xfs_dir_ilock_class;
65 const struct xattr *xattr_array,
68 const struct xattr *xattr;
69 struct xfs_inode *ip = XFS_I(inode);
72 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
73 error = xfs_attr_set(ip, xattr->name, xattr->value,
74 xattr->value_len, ATTR_SECURE);
82 * Hook in SELinux. This is not quite correct yet, what we really need
83 * here (as we do for default ACLs) is a mechanism by which creation of
84 * these attrs can be journalled at inode creation time (along with the
85 * inode, of course, such that log replay can't cause these to be lost).
92 const struct qstr *qstr)
94 return security_inode_init_security(inode, dir, qstr,
95 &xfs_initxattrs, NULL);
100 struct xfs_name *namep,
101 struct dentry *dentry,
104 namep->name = dentry->d_name.name;
105 namep->len = dentry->d_name.len;
106 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
113 struct dentry *dentry)
115 struct xfs_name teardown;
118 * If we can't add the ACL or we fail in
119 * xfs_init_security we must back out.
120 * ENOSPC can hit here, among other things.
122 xfs_dentry_to_name(&teardown, dentry, 0);
124 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
130 struct dentry *dentry,
133 bool tmpfile) /* unnamed file */
136 struct xfs_inode *ip = NULL;
137 struct posix_acl *default_acl, *acl;
138 struct xfs_name name;
142 * Irix uses Missed'em'V split, but doesn't want to see
143 * the upper 5 bits of (14bit) major.
145 if (S_ISCHR(mode) || S_ISBLK(mode)) {
146 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
148 rdev = sysv_encode_dev(rdev);
153 error = posix_acl_create(dir, &mode, &default_acl, &acl);
158 xfs_dentry_to_name(&name, dentry, mode);
159 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
161 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
168 error = xfs_init_security(inode, dir, &dentry->d_name);
170 goto out_cleanup_inode;
172 #ifdef CONFIG_XFS_POSIX_ACL
174 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
176 goto out_cleanup_inode;
179 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
181 goto out_cleanup_inode;
188 d_tmpfile(dentry, inode);
190 d_instantiate(dentry, inode);
192 xfs_finish_inode_setup(ip);
196 posix_acl_release(default_acl);
198 posix_acl_release(acl);
202 xfs_finish_inode_setup(ip);
204 xfs_cleanup_inode(dir, inode, dentry);
212 struct dentry *dentry,
216 return xfs_generic_create(dir, dentry, mode, rdev, false);
222 struct dentry *dentry,
226 return xfs_vn_mknod(dir, dentry, mode, 0);
232 struct dentry *dentry,
235 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
238 STATIC struct dentry *
241 struct dentry *dentry,
244 struct xfs_inode *cip;
245 struct xfs_name name;
248 if (dentry->d_name.len >= MAXNAMELEN)
249 return ERR_PTR(-ENAMETOOLONG);
251 xfs_dentry_to_name(&name, dentry, 0);
252 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
253 if (unlikely(error)) {
254 if (unlikely(error != -ENOENT))
255 return ERR_PTR(error);
260 return d_splice_alias(VFS_I(cip), dentry);
263 STATIC struct dentry *
266 struct dentry *dentry,
269 struct xfs_inode *ip;
270 struct xfs_name xname;
271 struct xfs_name ci_name;
275 if (dentry->d_name.len >= MAXNAMELEN)
276 return ERR_PTR(-ENAMETOOLONG);
278 xfs_dentry_to_name(&xname, dentry, 0);
279 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
280 if (unlikely(error)) {
281 if (unlikely(error != -ENOENT))
282 return ERR_PTR(error);
284 * call d_add(dentry, NULL) here when d_drop_negative_children
285 * is called in xfs_vn_mknod (ie. allow negative dentries
286 * with CI filesystems).
291 /* if exact match, just splice and exit */
293 return d_splice_alias(VFS_I(ip), dentry);
295 /* else case-insensitive match... */
296 dname.name = ci_name.name;
297 dname.len = ci_name.len;
298 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
299 kmem_free(ci_name.name);
305 struct dentry *old_dentry,
307 struct dentry *dentry)
309 struct inode *inode = d_inode(old_dentry);
310 struct xfs_name name;
313 xfs_dentry_to_name(&name, dentry, inode->i_mode);
315 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
320 d_instantiate(dentry, inode);
327 struct dentry *dentry)
329 struct xfs_name name;
332 xfs_dentry_to_name(&name, dentry, 0);
334 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
339 * With unlink, the VFS makes the dentry "negative": no inode,
340 * but still hashed. This is incompatible with case-insensitive
341 * mode, so invalidate (unhash) the dentry in CI-mode.
343 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
344 d_invalidate(dentry);
351 struct dentry *dentry,
355 struct xfs_inode *cip = NULL;
356 struct xfs_name name;
361 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
362 xfs_dentry_to_name(&name, dentry, mode);
364 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
370 error = xfs_init_security(inode, dir, &dentry->d_name);
372 goto out_cleanup_inode;
376 d_instantiate(dentry, inode);
377 xfs_finish_inode_setup(cip);
381 xfs_finish_inode_setup(cip);
382 xfs_cleanup_inode(dir, inode, dentry);
391 struct dentry *odentry,
393 struct dentry *ndentry,
396 struct inode *new_inode = d_inode(ndentry);
398 struct xfs_name oname;
399 struct xfs_name nname;
401 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
404 /* if we are exchanging files, we need to set i_mode of both files */
405 if (flags & RENAME_EXCHANGE)
406 omode = d_inode(ndentry)->i_mode;
408 xfs_dentry_to_name(&oname, odentry, omode);
409 xfs_dentry_to_name(&nname, ndentry, d_inode(odentry)->i_mode);
411 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
413 new_inode ? XFS_I(new_inode) : NULL, flags);
417 * careful here - this function can get called recursively, so
418 * we need to be very careful about how much stack we use.
419 * uio is kmalloced for this reason...
423 struct dentry *dentry,
425 struct delayed_call *done)
431 return ERR_PTR(-ECHILD);
433 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
437 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
441 set_delayed_call(done, kfree_link, link);
447 return ERR_PTR(error);
451 xfs_vn_get_link_inline(
452 struct dentry *dentry,
454 struct delayed_call *done)
456 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
457 return XFS_I(inode)->i_df.if_u1.if_data;
462 struct vfsmount *mnt,
463 struct dentry *dentry,
466 struct inode *inode = d_inode(dentry);
467 struct xfs_inode *ip = XFS_I(inode);
468 struct xfs_mount *mp = ip->i_mount;
470 trace_xfs_getattr(ip);
472 if (XFS_FORCED_SHUTDOWN(mp))
475 stat->size = XFS_ISIZE(ip);
476 stat->dev = inode->i_sb->s_dev;
477 stat->mode = inode->i_mode;
478 stat->nlink = inode->i_nlink;
479 stat->uid = inode->i_uid;
480 stat->gid = inode->i_gid;
481 stat->ino = ip->i_ino;
482 stat->atime = inode->i_atime;
483 stat->mtime = inode->i_mtime;
484 stat->ctime = inode->i_ctime;
486 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
489 switch (inode->i_mode & S_IFMT) {
492 stat->blksize = BLKDEV_IOSIZE;
493 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
494 sysv_minor(ip->i_df.if_u2.if_rdev));
497 if (XFS_IS_REALTIME_INODE(ip)) {
499 * If the file blocks are being allocated from a
500 * realtime volume, then return the inode's realtime
501 * extent size or the realtime volume's extent size.
504 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
506 stat->blksize = xfs_preferred_iosize(mp);
516 struct xfs_inode *ip,
519 struct inode *inode = VFS_I(ip);
520 umode_t mode = iattr->ia_mode;
522 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
524 inode->i_mode &= S_IFMT;
525 inode->i_mode |= mode & ~S_IFMT;
530 struct xfs_inode *ip,
533 struct inode *inode = VFS_I(ip);
535 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
537 if (iattr->ia_valid & ATTR_ATIME)
538 inode->i_atime = iattr->ia_atime;
539 if (iattr->ia_valid & ATTR_CTIME)
540 inode->i_ctime = iattr->ia_ctime;
541 if (iattr->ia_valid & ATTR_MTIME)
542 inode->i_mtime = iattr->ia_mtime;
547 struct dentry *dentry,
550 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
552 if (mp->m_flags & XFS_MOUNT_RDONLY)
555 if (XFS_FORCED_SHUTDOWN(mp))
558 return setattr_prepare(dentry, iattr);
562 * Set non-size attributes of an inode.
564 * Caution: The caller of this function is responsible for calling
565 * setattr_prepare() or otherwise verifying the change is fine.
569 struct xfs_inode *ip,
573 xfs_mount_t *mp = ip->i_mount;
574 struct inode *inode = VFS_I(ip);
575 int mask = iattr->ia_valid;
578 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
579 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
580 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
581 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
583 ASSERT((mask & ATTR_SIZE) == 0);
586 * If disk quotas is on, we make sure that the dquots do exist on disk,
587 * before we start any other transactions. Trying to do this later
588 * is messy. We don't care to take a readlock to look at the ids
589 * in inode here, because we can't hold it across the trans_reserve.
590 * If the IDs do change before we take the ilock, we're covered
591 * because the i_*dquot fields will get updated anyway.
593 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
596 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
598 qflags |= XFS_QMOPT_UQUOTA;
602 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
604 qflags |= XFS_QMOPT_GQUOTA;
610 * We take a reference when we initialize udqp and gdqp,
611 * so it is important that we never blindly double trip on
612 * the same variable. See xfs_create() for an example.
614 ASSERT(udqp == NULL);
615 ASSERT(gdqp == NULL);
616 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
617 xfs_kgid_to_gid(gid),
619 qflags, &udqp, &gdqp, NULL);
624 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
628 xfs_ilock(ip, XFS_ILOCK_EXCL);
629 xfs_trans_ijoin(tp, ip, 0);
632 * Change file ownership. Must be the owner or privileged.
634 if (mask & (ATTR_UID|ATTR_GID)) {
636 * These IDs could have changed since we last looked at them.
637 * But, we're assured that if the ownership did change
638 * while we didn't have the inode locked, inode's dquot(s)
639 * would have changed also.
643 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
644 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
647 * Do a quota reservation only if uid/gid is actually
650 if (XFS_IS_QUOTA_RUNNING(mp) &&
651 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
652 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
654 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
655 NULL, capable(CAP_FOWNER) ?
656 XFS_QMOPT_FORCE_RES : 0);
657 if (error) /* out of quota */
663 * Change file ownership. Must be the owner or privileged.
665 if (mask & (ATTR_UID|ATTR_GID)) {
667 * CAP_FSETID overrides the following restrictions:
669 * The set-user-ID and set-group-ID bits of a file will be
670 * cleared upon successful return from chown()
672 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
673 !capable(CAP_FSETID))
674 inode->i_mode &= ~(S_ISUID|S_ISGID);
677 * Change the ownerships and register quota modifications
678 * in the transaction.
680 if (!uid_eq(iuid, uid)) {
681 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
682 ASSERT(mask & ATTR_UID);
684 olddquot1 = xfs_qm_vop_chown(tp, ip,
685 &ip->i_udquot, udqp);
687 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
690 if (!gid_eq(igid, gid)) {
691 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
692 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
693 !XFS_IS_PQUOTA_ON(mp));
694 ASSERT(mask & ATTR_GID);
696 olddquot2 = xfs_qm_vop_chown(tp, ip,
697 &ip->i_gdquot, gdqp);
699 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
704 if (mask & ATTR_MODE)
705 xfs_setattr_mode(ip, iattr);
706 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
707 xfs_setattr_time(ip, iattr);
709 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
711 XFS_STATS_INC(mp, xs_ig_attrchg);
713 if (mp->m_flags & XFS_MOUNT_WSYNC)
714 xfs_trans_set_sync(tp);
715 error = xfs_trans_commit(tp);
717 xfs_iunlock(ip, XFS_ILOCK_EXCL);
720 * Release any dquot(s) the inode had kept before chown.
722 xfs_qm_dqrele(olddquot1);
723 xfs_qm_dqrele(olddquot2);
731 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
732 * update. We could avoid this with linked transactions
733 * and passing down the transaction pointer all the way
734 * to attr_set. No previous user of the generic
735 * Posix ACL code seems to care about this issue either.
737 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
738 error = posix_acl_chmod(inode, inode->i_mode);
746 xfs_trans_cancel(tp);
754 xfs_vn_setattr_nonsize(
755 struct dentry *dentry,
758 struct xfs_inode *ip = XFS_I(d_inode(dentry));
761 trace_xfs_setattr(ip);
763 error = xfs_vn_change_ok(dentry, iattr);
766 return xfs_setattr_nonsize(ip, iattr, 0);
770 * Truncate file. Must have write permission and not be a directory.
772 * Caution: The caller of this function is responsible for calling
773 * setattr_prepare() or otherwise verifying the change is fine.
777 struct xfs_inode *ip,
780 struct xfs_mount *mp = ip->i_mount;
781 struct inode *inode = VFS_I(ip);
782 xfs_off_t oldsize, newsize;
783 struct xfs_trans *tp;
786 bool did_zeroing = false;
788 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
789 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
790 ASSERT(S_ISREG(inode->i_mode));
791 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
792 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
794 oldsize = inode->i_size;
795 newsize = iattr->ia_size;
798 * Short circuit the truncate case for zero length files.
800 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
801 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
805 * Use the regular setattr path to update the timestamps.
807 iattr->ia_valid &= ~ATTR_SIZE;
808 return xfs_setattr_nonsize(ip, iattr, 0);
812 * Make sure that the dquots are attached to the inode.
814 error = xfs_qm_dqattach(ip, 0);
819 * Wait for all direct I/O to complete.
821 inode_dio_wait(inode);
824 * File data changes must be complete before we start the transaction to
825 * modify the inode. This needs to be done before joining the inode to
826 * the transaction because the inode cannot be unlocked once it is a
827 * part of the transaction.
829 * Start with zeroing any data beyond EOF that we may expose on file
830 * extension, or zeroing out the rest of the block on a downward
833 if (newsize > oldsize) {
834 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
836 error = iomap_truncate_page(inode, newsize, &did_zeroing,
844 * We are going to log the inode size change in this transaction so
845 * any previous writes that are beyond the on disk EOF and the new
846 * EOF that have not been written out need to be written here. If we
847 * do not write the data out, we expose ourselves to the null files
848 * problem. Note that this includes any block zeroing we did above;
849 * otherwise those blocks may not be zeroed after a crash.
852 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
853 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
854 ip->i_d.di_size, newsize);
860 * We've already locked out new page faults, so now we can safely remove
861 * pages from the page cache knowing they won't get refaulted until we
862 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
863 * complete. The truncate_setsize() call also cleans partial EOF page
864 * PTEs on extending truncates and hence ensures sub-page block size
865 * filesystems are correctly handled, too.
867 * We have to do all the page cache truncate work outside the
868 * transaction context as the "lock" order is page lock->log space
869 * reservation as defined by extent allocation in the writeback path.
870 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
871 * having already truncated the in-memory version of the file (i.e. made
872 * user visible changes). There's not much we can do about this, except
873 * to hope that the caller sees ENOMEM and retries the truncate
876 truncate_setsize(inode, newsize);
878 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
882 lock_flags |= XFS_ILOCK_EXCL;
883 xfs_ilock(ip, XFS_ILOCK_EXCL);
884 xfs_trans_ijoin(tp, ip, 0);
887 * Only change the c/mtime if we are changing the size or we are
888 * explicitly asked to change it. This handles the semantic difference
889 * between truncate() and ftruncate() as implemented in the VFS.
891 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
892 * special case where we need to update the times despite not having
893 * these flags set. For all other operations the VFS set these flags
894 * explicitly if it wants a timestamp update.
896 if (newsize != oldsize &&
897 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
898 iattr->ia_ctime = iattr->ia_mtime =
900 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
904 * The first thing we do is set the size to new_size permanently on
905 * disk. This way we don't have to worry about anyone ever being able
906 * to look at the data being freed even in the face of a crash.
907 * What we're getting around here is the case where we free a block, it
908 * is allocated to another file, it is written to, and then we crash.
909 * If the new data gets written to the file but the log buffers
910 * containing the free and reallocation don't, then we'd end up with
911 * garbage in the blocks being freed. As long as we make the new size
912 * permanent before actually freeing any blocks it doesn't matter if
913 * they get written to.
915 ip->i_d.di_size = newsize;
916 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
918 if (newsize <= oldsize) {
919 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
921 goto out_trans_cancel;
924 * Truncated "down", so we're removing references to old data
925 * here - if we delay flushing for a long time, we expose
926 * ourselves unduly to the notorious NULL files problem. So,
927 * we mark this inode and flush it when the file is closed,
928 * and do not wait the usual (long) time for writeout.
930 xfs_iflags_set(ip, XFS_ITRUNCATED);
932 /* A truncate down always removes post-EOF blocks. */
933 xfs_inode_clear_eofblocks_tag(ip);
936 if (iattr->ia_valid & ATTR_MODE)
937 xfs_setattr_mode(ip, iattr);
938 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
939 xfs_setattr_time(ip, iattr);
941 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
943 XFS_STATS_INC(mp, xs_ig_attrchg);
945 if (mp->m_flags & XFS_MOUNT_WSYNC)
946 xfs_trans_set_sync(tp);
948 error = xfs_trans_commit(tp);
951 xfs_iunlock(ip, lock_flags);
955 xfs_trans_cancel(tp);
961 struct dentry *dentry,
964 struct xfs_inode *ip = XFS_I(d_inode(dentry));
967 trace_xfs_setattr(ip);
969 error = xfs_vn_change_ok(dentry, iattr);
972 return xfs_setattr_size(ip, iattr);
977 struct dentry *dentry,
982 if (iattr->ia_valid & ATTR_SIZE) {
983 struct xfs_inode *ip = XFS_I(d_inode(dentry));
984 uint iolock = XFS_IOLOCK_EXCL;
986 error = xfs_break_layouts(d_inode(dentry), &iolock);
990 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
991 error = xfs_vn_setattr_size(dentry, iattr);
992 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
994 error = xfs_vn_setattr_nonsize(dentry, iattr);
1002 struct inode *inode,
1003 struct timespec *now,
1006 struct xfs_inode *ip = XFS_I(inode);
1007 struct xfs_mount *mp = ip->i_mount;
1008 struct xfs_trans *tp;
1011 trace_xfs_update_time(ip);
1013 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1017 xfs_ilock(ip, XFS_ILOCK_EXCL);
1018 if (flags & S_CTIME)
1019 inode->i_ctime = *now;
1020 if (flags & S_MTIME)
1021 inode->i_mtime = *now;
1022 if (flags & S_ATIME)
1023 inode->i_atime = *now;
1025 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1026 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1027 return xfs_trans_commit(tp);
1032 struct inode *inode,
1033 struct fiemap_extent_info *fieinfo,
1039 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1040 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1041 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1042 error = iomap_fiemap(inode, fieinfo, start, length,
1043 &xfs_xattr_iomap_ops);
1045 error = iomap_fiemap(inode, fieinfo, start, length,
1048 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1056 struct dentry *dentry,
1059 return xfs_generic_create(dir, dentry, mode, 0, true);
1062 static const struct inode_operations xfs_inode_operations = {
1063 .get_acl = xfs_get_acl,
1064 .set_acl = xfs_set_acl,
1065 .getattr = xfs_vn_getattr,
1066 .setattr = xfs_vn_setattr,
1067 .listxattr = xfs_vn_listxattr,
1068 .fiemap = xfs_vn_fiemap,
1069 .update_time = xfs_vn_update_time,
1072 static const struct inode_operations xfs_dir_inode_operations = {
1073 .create = xfs_vn_create,
1074 .lookup = xfs_vn_lookup,
1075 .link = xfs_vn_link,
1076 .unlink = xfs_vn_unlink,
1077 .symlink = xfs_vn_symlink,
1078 .mkdir = xfs_vn_mkdir,
1080 * Yes, XFS uses the same method for rmdir and unlink.
1082 * There are some subtile differences deeper in the code,
1083 * but we use S_ISDIR to check for those.
1085 .rmdir = xfs_vn_unlink,
1086 .mknod = xfs_vn_mknod,
1087 .rename = xfs_vn_rename,
1088 .get_acl = xfs_get_acl,
1089 .set_acl = xfs_set_acl,
1090 .getattr = xfs_vn_getattr,
1091 .setattr = xfs_vn_setattr,
1092 .listxattr = xfs_vn_listxattr,
1093 .update_time = xfs_vn_update_time,
1094 .tmpfile = xfs_vn_tmpfile,
1097 static const struct inode_operations xfs_dir_ci_inode_operations = {
1098 .create = xfs_vn_create,
1099 .lookup = xfs_vn_ci_lookup,
1100 .link = xfs_vn_link,
1101 .unlink = xfs_vn_unlink,
1102 .symlink = xfs_vn_symlink,
1103 .mkdir = xfs_vn_mkdir,
1105 * Yes, XFS uses the same method for rmdir and unlink.
1107 * There are some subtile differences deeper in the code,
1108 * but we use S_ISDIR to check for those.
1110 .rmdir = xfs_vn_unlink,
1111 .mknod = xfs_vn_mknod,
1112 .rename = xfs_vn_rename,
1113 .get_acl = xfs_get_acl,
1114 .set_acl = xfs_set_acl,
1115 .getattr = xfs_vn_getattr,
1116 .setattr = xfs_vn_setattr,
1117 .listxattr = xfs_vn_listxattr,
1118 .update_time = xfs_vn_update_time,
1119 .tmpfile = xfs_vn_tmpfile,
1122 static const struct inode_operations xfs_symlink_inode_operations = {
1123 .readlink = generic_readlink,
1124 .get_link = xfs_vn_get_link,
1125 .getattr = xfs_vn_getattr,
1126 .setattr = xfs_vn_setattr,
1127 .listxattr = xfs_vn_listxattr,
1128 .update_time = xfs_vn_update_time,
1131 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1132 .readlink = generic_readlink,
1133 .get_link = xfs_vn_get_link_inline,
1134 .getattr = xfs_vn_getattr,
1135 .setattr = xfs_vn_setattr,
1136 .listxattr = xfs_vn_listxattr,
1137 .update_time = xfs_vn_update_time,
1141 xfs_diflags_to_iflags(
1142 struct inode *inode,
1143 struct xfs_inode *ip)
1145 uint16_t flags = ip->i_d.di_flags;
1147 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1150 if (flags & XFS_DIFLAG_IMMUTABLE)
1151 inode->i_flags |= S_IMMUTABLE;
1152 if (flags & XFS_DIFLAG_APPEND)
1153 inode->i_flags |= S_APPEND;
1154 if (flags & XFS_DIFLAG_SYNC)
1155 inode->i_flags |= S_SYNC;
1156 if (flags & XFS_DIFLAG_NOATIME)
1157 inode->i_flags |= S_NOATIME;
1158 if (S_ISREG(inode->i_mode) &&
1159 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1160 !xfs_is_reflink_inode(ip) &&
1161 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1162 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1163 inode->i_flags |= S_DAX;
1167 * Initialize the Linux inode.
1169 * When reading existing inodes from disk this is called directly from xfs_iget,
1170 * when creating a new inode it is called from xfs_ialloc after setting up the
1171 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1172 * it up to the caller to deal with unlocking the inode appropriately.
1176 struct xfs_inode *ip)
1178 struct inode *inode = &ip->i_vnode;
1181 inode->i_ino = ip->i_ino;
1182 inode->i_state = I_NEW;
1184 inode_sb_list_add(inode);
1185 /* make the inode look hashed for the writeback code */
1186 hlist_add_fake(&inode->i_hash);
1188 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1189 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1191 switch (inode->i_mode & S_IFMT) {
1195 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1196 sysv_minor(ip->i_df.if_u2.if_rdev));
1203 i_size_write(inode, ip->i_d.di_size);
1204 xfs_diflags_to_iflags(inode, ip);
1206 if (S_ISDIR(inode->i_mode)) {
1207 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1208 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1210 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1211 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1215 * Ensure all page cache allocations are done from GFP_NOFS context to
1216 * prevent direct reclaim recursion back into the filesystem and blowing
1217 * stacks or deadlocking.
1219 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1220 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1223 * If there is no attribute fork no ACL can exist on this inode,
1224 * and it can't have any file capabilities attached to it either.
1226 if (!XFS_IFORK_Q(ip)) {
1227 inode_has_no_xattr(inode);
1228 cache_no_acl(inode);
1234 struct xfs_inode *ip)
1236 struct inode *inode = &ip->i_vnode;
1238 switch (inode->i_mode & S_IFMT) {
1240 inode->i_op = &xfs_inode_operations;
1241 inode->i_fop = &xfs_file_operations;
1242 inode->i_mapping->a_ops = &xfs_address_space_operations;
1245 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1246 inode->i_op = &xfs_dir_ci_inode_operations;
1248 inode->i_op = &xfs_dir_inode_operations;
1249 inode->i_fop = &xfs_dir_file_operations;
1252 if (ip->i_df.if_flags & XFS_IFINLINE)
1253 inode->i_op = &xfs_inline_symlink_inode_operations;
1255 inode->i_op = &xfs_symlink_inode_operations;
1258 inode->i_op = &xfs_inode_operations;
1259 init_special_inode(inode, inode->i_mode, inode->i_rdev);