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/fiemap.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 xfs_inode *ip,
551 xfs_mount_t *mp = ip->i_mount;
552 struct inode *inode = VFS_I(ip);
553 int mask = iattr->ia_valid;
556 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
557 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
558 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
559 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
561 trace_xfs_setattr(ip);
563 /* If acls are being inherited, we already have this checked */
564 if (!(flags & XFS_ATTR_NOACL)) {
565 if (mp->m_flags & XFS_MOUNT_RDONLY)
568 if (XFS_FORCED_SHUTDOWN(mp))
571 error = inode_change_ok(inode, iattr);
576 ASSERT((mask & ATTR_SIZE) == 0);
579 * If disk quotas is on, we make sure that the dquots do exist on disk,
580 * before we start any other transactions. Trying to do this later
581 * is messy. We don't care to take a readlock to look at the ids
582 * in inode here, because we can't hold it across the trans_reserve.
583 * If the IDs do change before we take the ilock, we're covered
584 * because the i_*dquot fields will get updated anyway.
586 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
589 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
591 qflags |= XFS_QMOPT_UQUOTA;
595 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
597 qflags |= XFS_QMOPT_GQUOTA;
603 * We take a reference when we initialize udqp and gdqp,
604 * so it is important that we never blindly double trip on
605 * the same variable. See xfs_create() for an example.
607 ASSERT(udqp == NULL);
608 ASSERT(gdqp == NULL);
609 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
610 xfs_kgid_to_gid(gid),
612 qflags, &udqp, &gdqp, NULL);
617 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
621 xfs_ilock(ip, XFS_ILOCK_EXCL);
622 xfs_trans_ijoin(tp, ip, 0);
625 * Change file ownership. Must be the owner or privileged.
627 if (mask & (ATTR_UID|ATTR_GID)) {
629 * These IDs could have changed since we last looked at them.
630 * But, we're assured that if the ownership did change
631 * while we didn't have the inode locked, inode's dquot(s)
632 * would have changed also.
636 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
637 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
640 * Do a quota reservation only if uid/gid is actually
643 if (XFS_IS_QUOTA_RUNNING(mp) &&
644 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
645 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
647 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
648 NULL, capable(CAP_FOWNER) ?
649 XFS_QMOPT_FORCE_RES : 0);
650 if (error) /* out of quota */
656 * Change file ownership. Must be the owner or privileged.
658 if (mask & (ATTR_UID|ATTR_GID)) {
660 * CAP_FSETID overrides the following restrictions:
662 * The set-user-ID and set-group-ID bits of a file will be
663 * cleared upon successful return from chown()
665 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
666 !capable(CAP_FSETID))
667 inode->i_mode &= ~(S_ISUID|S_ISGID);
670 * Change the ownerships and register quota modifications
671 * in the transaction.
673 if (!uid_eq(iuid, uid)) {
674 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
675 ASSERT(mask & ATTR_UID);
677 olddquot1 = xfs_qm_vop_chown(tp, ip,
678 &ip->i_udquot, udqp);
680 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
683 if (!gid_eq(igid, gid)) {
684 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
685 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
686 !XFS_IS_PQUOTA_ON(mp));
687 ASSERT(mask & ATTR_GID);
689 olddquot2 = xfs_qm_vop_chown(tp, ip,
690 &ip->i_gdquot, gdqp);
692 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
697 if (mask & ATTR_MODE)
698 xfs_setattr_mode(ip, iattr);
699 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
700 xfs_setattr_time(ip, iattr);
702 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
704 XFS_STATS_INC(mp, xs_ig_attrchg);
706 if (mp->m_flags & XFS_MOUNT_WSYNC)
707 xfs_trans_set_sync(tp);
708 error = xfs_trans_commit(tp);
710 xfs_iunlock(ip, XFS_ILOCK_EXCL);
713 * Release any dquot(s) the inode had kept before chown.
715 xfs_qm_dqrele(olddquot1);
716 xfs_qm_dqrele(olddquot2);
724 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
725 * update. We could avoid this with linked transactions
726 * and passing down the transaction pointer all the way
727 * to attr_set. No previous user of the generic
728 * Posix ACL code seems to care about this issue either.
730 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
731 error = posix_acl_chmod(inode, inode->i_mode);
739 xfs_trans_cancel(tp);
747 * Truncate file. Must have write permission and not be a directory.
751 struct xfs_inode *ip,
754 struct xfs_mount *mp = ip->i_mount;
755 struct inode *inode = VFS_I(ip);
756 xfs_off_t oldsize, newsize;
757 struct xfs_trans *tp;
760 bool did_zeroing = false;
762 trace_xfs_setattr(ip);
764 if (mp->m_flags & XFS_MOUNT_RDONLY)
767 if (XFS_FORCED_SHUTDOWN(mp))
770 error = inode_change_ok(inode, iattr);
774 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
775 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
776 ASSERT(S_ISREG(inode->i_mode));
777 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
778 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
780 oldsize = inode->i_size;
781 newsize = iattr->ia_size;
784 * Short circuit the truncate case for zero length files.
786 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
787 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
791 * Use the regular setattr path to update the timestamps.
793 iattr->ia_valid &= ~ATTR_SIZE;
794 return xfs_setattr_nonsize(ip, iattr, 0);
798 * Make sure that the dquots are attached to the inode.
800 error = xfs_qm_dqattach(ip, 0);
805 * Wait for all direct I/O to complete.
807 inode_dio_wait(inode);
810 * File data changes must be complete before we start the transaction to
811 * modify the inode. This needs to be done before joining the inode to
812 * the transaction because the inode cannot be unlocked once it is a
813 * part of the transaction.
815 * Start with zeroing any data beyond EOF that we may expose on file
816 * extension, or zeroing out the rest of the block on a downward
819 if (newsize > oldsize) {
820 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
823 error = dax_truncate_page(inode, newsize,
824 xfs_get_blocks_direct);
826 error = iomap_truncate_page(inode, newsize,
827 &did_zeroing, &xfs_iomap_ops);
835 * We are going to log the inode size change in this transaction so
836 * any previous writes that are beyond the on disk EOF and the new
837 * EOF that have not been written out need to be written here. If we
838 * do not write the data out, we expose ourselves to the null files
839 * problem. Note that this includes any block zeroing we did above;
840 * otherwise those blocks may not be zeroed after a crash.
843 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
844 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
845 ip->i_d.di_size, newsize);
851 * We've already locked out new page faults, so now we can safely remove
852 * pages from the page cache knowing they won't get refaulted until we
853 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
854 * complete. The truncate_setsize() call also cleans partial EOF page
855 * PTEs on extending truncates and hence ensures sub-page block size
856 * filesystems are correctly handled, too.
858 * We have to do all the page cache truncate work outside the
859 * transaction context as the "lock" order is page lock->log space
860 * reservation as defined by extent allocation in the writeback path.
861 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
862 * having already truncated the in-memory version of the file (i.e. made
863 * user visible changes). There's not much we can do about this, except
864 * to hope that the caller sees ENOMEM and retries the truncate
867 truncate_setsize(inode, newsize);
869 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
873 lock_flags |= XFS_ILOCK_EXCL;
874 xfs_ilock(ip, XFS_ILOCK_EXCL);
875 xfs_trans_ijoin(tp, ip, 0);
878 * Only change the c/mtime if we are changing the size or we are
879 * explicitly asked to change it. This handles the semantic difference
880 * between truncate() and ftruncate() as implemented in the VFS.
882 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
883 * special case where we need to update the times despite not having
884 * these flags set. For all other operations the VFS set these flags
885 * explicitly if it wants a timestamp update.
887 if (newsize != oldsize &&
888 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
889 iattr->ia_ctime = iattr->ia_mtime =
890 current_fs_time(inode->i_sb);
891 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
895 * The first thing we do is set the size to new_size permanently on
896 * disk. This way we don't have to worry about anyone ever being able
897 * to look at the data being freed even in the face of a crash.
898 * What we're getting around here is the case where we free a block, it
899 * is allocated to another file, it is written to, and then we crash.
900 * If the new data gets written to the file but the log buffers
901 * containing the free and reallocation don't, then we'd end up with
902 * garbage in the blocks being freed. As long as we make the new size
903 * permanent before actually freeing any blocks it doesn't matter if
904 * they get written to.
906 ip->i_d.di_size = newsize;
907 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
909 if (newsize <= oldsize) {
910 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
912 goto out_trans_cancel;
915 * Truncated "down", so we're removing references to old data
916 * here - if we delay flushing for a long time, we expose
917 * ourselves unduly to the notorious NULL files problem. So,
918 * we mark this inode and flush it when the file is closed,
919 * and do not wait the usual (long) time for writeout.
921 xfs_iflags_set(ip, XFS_ITRUNCATED);
923 /* A truncate down always removes post-EOF blocks. */
924 xfs_inode_clear_eofblocks_tag(ip);
927 if (iattr->ia_valid & ATTR_MODE)
928 xfs_setattr_mode(ip, iattr);
929 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
930 xfs_setattr_time(ip, iattr);
932 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
934 XFS_STATS_INC(mp, xs_ig_attrchg);
936 if (mp->m_flags & XFS_MOUNT_WSYNC)
937 xfs_trans_set_sync(tp);
939 error = xfs_trans_commit(tp);
942 xfs_iunlock(ip, lock_flags);
946 xfs_trans_cancel(tp);
952 struct dentry *dentry,
955 struct xfs_inode *ip = XFS_I(d_inode(dentry));
958 if (iattr->ia_valid & ATTR_SIZE) {
959 uint iolock = XFS_IOLOCK_EXCL;
961 xfs_ilock(ip, iolock);
962 error = xfs_break_layouts(d_inode(dentry), &iolock, true);
964 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
965 iolock |= XFS_MMAPLOCK_EXCL;
967 error = xfs_setattr_size(ip, iattr);
969 xfs_iunlock(ip, iolock);
971 error = xfs_setattr_nonsize(ip, iattr, 0);
980 struct timespec *now,
983 struct xfs_inode *ip = XFS_I(inode);
984 struct xfs_mount *mp = ip->i_mount;
985 struct xfs_trans *tp;
988 trace_xfs_update_time(ip);
990 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
994 xfs_ilock(ip, XFS_ILOCK_EXCL);
996 inode->i_ctime = *now;
998 inode->i_mtime = *now;
1000 inode->i_atime = *now;
1002 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1003 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1004 return xfs_trans_commit(tp);
1007 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1010 * Call fiemap helper to fill in user data.
1011 * Returns positive errors to xfs_getbmap.
1016 struct getbmapx *bmv,
1020 struct fiemap_extent_info *fieinfo = *arg;
1021 u32 fiemap_flags = 0;
1022 u64 logical, physical, length;
1024 /* Do nothing for a hole */
1025 if (bmv->bmv_block == -1LL)
1028 logical = BBTOB(bmv->bmv_offset);
1029 physical = BBTOB(bmv->bmv_block);
1030 length = BBTOB(bmv->bmv_length);
1032 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1033 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1034 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1035 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1036 FIEMAP_EXTENT_UNKNOWN);
1037 physical = 0; /* no block yet */
1039 if (bmv->bmv_oflags & BMV_OF_LAST)
1040 fiemap_flags |= FIEMAP_EXTENT_LAST;
1042 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1043 length, fiemap_flags);
1046 *full = 1; /* user array now full */
1054 struct inode *inode,
1055 struct fiemap_extent_info *fieinfo,
1059 xfs_inode_t *ip = XFS_I(inode);
1063 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1067 /* Set up bmap header for xfs internal routine */
1068 bm.bmv_offset = BTOBBT(start);
1069 /* Special case for whole file */
1070 if (length == FIEMAP_MAX_OFFSET)
1071 bm.bmv_length = -1LL;
1073 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1075 /* We add one because in getbmap world count includes the header */
1076 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1077 fieinfo->fi_extents_max + 1;
1078 bm.bmv_count = min_t(__s32, bm.bmv_count,
1079 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1080 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1081 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1082 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1083 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1084 bm.bmv_iflags |= BMV_IF_DELALLOC;
1086 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1096 struct dentry *dentry,
1099 return xfs_generic_create(dir, dentry, mode, 0, true);
1102 static const struct inode_operations xfs_inode_operations = {
1103 .get_acl = xfs_get_acl,
1104 .set_acl = xfs_set_acl,
1105 .getattr = xfs_vn_getattr,
1106 .setattr = xfs_vn_setattr,
1107 .setxattr = generic_setxattr,
1108 .getxattr = generic_getxattr,
1109 .removexattr = generic_removexattr,
1110 .listxattr = xfs_vn_listxattr,
1111 .fiemap = xfs_vn_fiemap,
1112 .update_time = xfs_vn_update_time,
1115 static const struct inode_operations xfs_dir_inode_operations = {
1116 .create = xfs_vn_create,
1117 .lookup = xfs_vn_lookup,
1118 .link = xfs_vn_link,
1119 .unlink = xfs_vn_unlink,
1120 .symlink = xfs_vn_symlink,
1121 .mkdir = xfs_vn_mkdir,
1123 * Yes, XFS uses the same method for rmdir and unlink.
1125 * There are some subtile differences deeper in the code,
1126 * but we use S_ISDIR to check for those.
1128 .rmdir = xfs_vn_unlink,
1129 .mknod = xfs_vn_mknod,
1130 .rename2 = xfs_vn_rename,
1131 .get_acl = xfs_get_acl,
1132 .set_acl = xfs_set_acl,
1133 .getattr = xfs_vn_getattr,
1134 .setattr = xfs_vn_setattr,
1135 .setxattr = generic_setxattr,
1136 .getxattr = generic_getxattr,
1137 .removexattr = generic_removexattr,
1138 .listxattr = xfs_vn_listxattr,
1139 .update_time = xfs_vn_update_time,
1140 .tmpfile = xfs_vn_tmpfile,
1143 static const struct inode_operations xfs_dir_ci_inode_operations = {
1144 .create = xfs_vn_create,
1145 .lookup = xfs_vn_ci_lookup,
1146 .link = xfs_vn_link,
1147 .unlink = xfs_vn_unlink,
1148 .symlink = xfs_vn_symlink,
1149 .mkdir = xfs_vn_mkdir,
1151 * Yes, XFS uses the same method for rmdir and unlink.
1153 * There are some subtile differences deeper in the code,
1154 * but we use S_ISDIR to check for those.
1156 .rmdir = xfs_vn_unlink,
1157 .mknod = xfs_vn_mknod,
1158 .rename2 = 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 .setxattr = generic_setxattr,
1164 .getxattr = generic_getxattr,
1165 .removexattr = generic_removexattr,
1166 .listxattr = xfs_vn_listxattr,
1167 .update_time = xfs_vn_update_time,
1168 .tmpfile = xfs_vn_tmpfile,
1171 static const struct inode_operations xfs_symlink_inode_operations = {
1172 .readlink = generic_readlink,
1173 .get_link = xfs_vn_get_link,
1174 .getattr = xfs_vn_getattr,
1175 .setattr = xfs_vn_setattr,
1176 .setxattr = generic_setxattr,
1177 .getxattr = generic_getxattr,
1178 .removexattr = generic_removexattr,
1179 .listxattr = xfs_vn_listxattr,
1180 .update_time = xfs_vn_update_time,
1183 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1184 .readlink = generic_readlink,
1185 .get_link = xfs_vn_get_link_inline,
1186 .getattr = xfs_vn_getattr,
1187 .setattr = xfs_vn_setattr,
1188 .setxattr = generic_setxattr,
1189 .getxattr = generic_getxattr,
1190 .removexattr = generic_removexattr,
1191 .listxattr = xfs_vn_listxattr,
1192 .update_time = xfs_vn_update_time,
1196 xfs_diflags_to_iflags(
1197 struct inode *inode,
1198 struct xfs_inode *ip)
1200 uint16_t flags = ip->i_d.di_flags;
1202 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1205 if (flags & XFS_DIFLAG_IMMUTABLE)
1206 inode->i_flags |= S_IMMUTABLE;
1207 if (flags & XFS_DIFLAG_APPEND)
1208 inode->i_flags |= S_APPEND;
1209 if (flags & XFS_DIFLAG_SYNC)
1210 inode->i_flags |= S_SYNC;
1211 if (flags & XFS_DIFLAG_NOATIME)
1212 inode->i_flags |= S_NOATIME;
1213 if (S_ISREG(inode->i_mode) &&
1214 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1215 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1216 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1217 inode->i_flags |= S_DAX;
1221 * Initialize the Linux inode.
1223 * When reading existing inodes from disk this is called directly from xfs_iget,
1224 * when creating a new inode it is called from xfs_ialloc after setting up the
1225 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1226 * it up to the caller to deal with unlocking the inode appropriately.
1230 struct xfs_inode *ip)
1232 struct inode *inode = &ip->i_vnode;
1235 inode->i_ino = ip->i_ino;
1236 inode->i_state = I_NEW;
1238 inode_sb_list_add(inode);
1239 /* make the inode look hashed for the writeback code */
1240 hlist_add_fake(&inode->i_hash);
1242 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1243 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1245 switch (inode->i_mode & S_IFMT) {
1249 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1250 sysv_minor(ip->i_df.if_u2.if_rdev));
1257 i_size_write(inode, ip->i_d.di_size);
1258 xfs_diflags_to_iflags(inode, ip);
1260 if (S_ISDIR(inode->i_mode)) {
1261 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1262 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1264 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1265 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1269 * Ensure all page cache allocations are done from GFP_NOFS context to
1270 * prevent direct reclaim recursion back into the filesystem and blowing
1271 * stacks or deadlocking.
1273 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1274 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1277 * If there is no attribute fork no ACL can exist on this inode,
1278 * and it can't have any file capabilities attached to it either.
1280 if (!XFS_IFORK_Q(ip)) {
1281 inode_has_no_xattr(inode);
1282 cache_no_acl(inode);
1288 struct xfs_inode *ip)
1290 struct inode *inode = &ip->i_vnode;
1292 switch (inode->i_mode & S_IFMT) {
1294 inode->i_op = &xfs_inode_operations;
1295 inode->i_fop = &xfs_file_operations;
1296 inode->i_mapping->a_ops = &xfs_address_space_operations;
1299 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1300 inode->i_op = &xfs_dir_ci_inode_operations;
1302 inode->i_op = &xfs_dir_inode_operations;
1303 inode->i_fop = &xfs_dir_file_operations;
1306 if (ip->i_df.if_flags & XFS_IFINLINE)
1307 inode->i_op = &xfs_inline_symlink_inode_operations;
1309 inode->i_op = &xfs_symlink_inode_operations;
1312 inode->i_op = &xfs_inode_operations;
1313 init_special_inode(inode, inode->i_mode, inode->i_rdev);