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)
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = XFS_DIR3_FT_UNKNOWN;
109 xfs_dentry_mode_to_name(
110 struct xfs_name *namep,
111 struct dentry *dentry,
114 namep->name = dentry->d_name.name;
115 namep->len = dentry->d_name.len;
116 namep->type = xfs_mode_to_ftype(mode);
118 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
119 return -EFSCORRUPTED;
128 struct dentry *dentry)
130 struct xfs_name teardown;
133 * If we can't add the ACL or we fail in
134 * xfs_init_security we must back out.
135 * ENOSPC can hit here, among other things.
137 xfs_dentry_to_name(&teardown, dentry);
139 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
145 struct dentry *dentry,
148 bool tmpfile) /* unnamed file */
151 struct xfs_inode *ip = NULL;
152 struct posix_acl *default_acl, *acl;
153 struct xfs_name name;
157 * Irix uses Missed'em'V split, but doesn't want to see
158 * the upper 5 bits of (14bit) major.
160 if (S_ISCHR(mode) || S_ISBLK(mode)) {
161 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
163 rdev = sysv_encode_dev(rdev);
168 error = posix_acl_create(dir, &mode, &default_acl, &acl);
172 /* Verify mode is valid also for tmpfile case */
173 error = xfs_dentry_mode_to_name(&name, dentry, mode);
178 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
180 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
187 error = xfs_init_security(inode, dir, &dentry->d_name);
189 goto out_cleanup_inode;
191 #ifdef CONFIG_XFS_POSIX_ACL
193 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
195 goto out_cleanup_inode;
198 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
200 goto out_cleanup_inode;
207 d_tmpfile(dentry, inode);
209 d_instantiate(dentry, inode);
211 xfs_finish_inode_setup(ip);
215 posix_acl_release(default_acl);
217 posix_acl_release(acl);
221 xfs_finish_inode_setup(ip);
223 xfs_cleanup_inode(dir, inode, dentry);
231 struct dentry *dentry,
235 return xfs_generic_create(dir, dentry, mode, rdev, false);
241 struct dentry *dentry,
245 return xfs_vn_mknod(dir, dentry, mode, 0);
251 struct dentry *dentry,
254 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
257 STATIC struct dentry *
260 struct dentry *dentry,
263 struct xfs_inode *cip;
264 struct xfs_name name;
267 if (dentry->d_name.len >= MAXNAMELEN)
268 return ERR_PTR(-ENAMETOOLONG);
270 xfs_dentry_to_name(&name, dentry);
271 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
272 if (unlikely(error)) {
273 if (unlikely(error != -ENOENT))
274 return ERR_PTR(error);
279 return d_splice_alias(VFS_I(cip), dentry);
282 STATIC struct dentry *
285 struct dentry *dentry,
288 struct xfs_inode *ip;
289 struct xfs_name xname;
290 struct xfs_name ci_name;
294 if (dentry->d_name.len >= MAXNAMELEN)
295 return ERR_PTR(-ENAMETOOLONG);
297 xfs_dentry_to_name(&xname, dentry);
298 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
299 if (unlikely(error)) {
300 if (unlikely(error != -ENOENT))
301 return ERR_PTR(error);
303 * call d_add(dentry, NULL) here when d_drop_negative_children
304 * is called in xfs_vn_mknod (ie. allow negative dentries
305 * with CI filesystems).
310 /* if exact match, just splice and exit */
312 return d_splice_alias(VFS_I(ip), dentry);
314 /* else case-insensitive match... */
315 dname.name = ci_name.name;
316 dname.len = ci_name.len;
317 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
318 kmem_free(ci_name.name);
324 struct dentry *old_dentry,
326 struct dentry *dentry)
328 struct inode *inode = d_inode(old_dentry);
329 struct xfs_name name;
332 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
336 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
341 d_instantiate(dentry, inode);
348 struct dentry *dentry)
350 struct xfs_name name;
353 xfs_dentry_to_name(&name, dentry);
355 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
360 * With unlink, the VFS makes the dentry "negative": no inode,
361 * but still hashed. This is incompatible with case-insensitive
362 * mode, so invalidate (unhash) the dentry in CI-mode.
364 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
365 d_invalidate(dentry);
372 struct dentry *dentry,
376 struct xfs_inode *cip = NULL;
377 struct xfs_name name;
382 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
383 error = xfs_dentry_mode_to_name(&name, dentry, mode);
387 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
393 error = xfs_init_security(inode, dir, &dentry->d_name);
395 goto out_cleanup_inode;
399 d_instantiate(dentry, inode);
400 xfs_finish_inode_setup(cip);
404 xfs_finish_inode_setup(cip);
405 xfs_cleanup_inode(dir, inode, dentry);
414 struct dentry *odentry,
416 struct dentry *ndentry,
419 struct inode *new_inode = d_inode(ndentry);
422 struct xfs_name oname;
423 struct xfs_name nname;
425 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
428 /* if we are exchanging files, we need to set i_mode of both files */
429 if (flags & RENAME_EXCHANGE)
430 omode = d_inode(ndentry)->i_mode;
432 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
433 if (omode && unlikely(error))
436 error = xfs_dentry_mode_to_name(&nname, ndentry,
437 d_inode(odentry)->i_mode);
441 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
443 new_inode ? XFS_I(new_inode) : NULL, flags);
447 * careful here - this function can get called recursively, so
448 * we need to be very careful about how much stack we use.
449 * uio is kmalloced for this reason...
453 struct dentry *dentry,
455 struct delayed_call *done)
461 return ERR_PTR(-ECHILD);
463 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
467 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
471 set_delayed_call(done, kfree_link, link);
477 return ERR_PTR(error);
481 xfs_vn_get_link_inline(
482 struct dentry *dentry,
484 struct delayed_call *done)
486 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
487 return XFS_I(inode)->i_df.if_u1.if_data;
492 struct vfsmount *mnt,
493 struct dentry *dentry,
496 struct inode *inode = d_inode(dentry);
497 struct xfs_inode *ip = XFS_I(inode);
498 struct xfs_mount *mp = ip->i_mount;
500 trace_xfs_getattr(ip);
502 if (XFS_FORCED_SHUTDOWN(mp))
505 stat->size = XFS_ISIZE(ip);
506 stat->dev = inode->i_sb->s_dev;
507 stat->mode = inode->i_mode;
508 stat->nlink = inode->i_nlink;
509 stat->uid = inode->i_uid;
510 stat->gid = inode->i_gid;
511 stat->ino = ip->i_ino;
512 stat->atime = inode->i_atime;
513 stat->mtime = inode->i_mtime;
514 stat->ctime = inode->i_ctime;
516 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
519 switch (inode->i_mode & S_IFMT) {
522 stat->blksize = BLKDEV_IOSIZE;
523 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
524 sysv_minor(ip->i_df.if_u2.if_rdev));
527 if (XFS_IS_REALTIME_INODE(ip)) {
529 * If the file blocks are being allocated from a
530 * realtime volume, then return the inode's realtime
531 * extent size or the realtime volume's extent size.
534 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
536 stat->blksize = xfs_preferred_iosize(mp);
546 struct xfs_inode *ip,
549 struct inode *inode = VFS_I(ip);
550 umode_t mode = iattr->ia_mode;
552 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
554 inode->i_mode &= S_IFMT;
555 inode->i_mode |= mode & ~S_IFMT;
560 struct xfs_inode *ip,
563 struct inode *inode = VFS_I(ip);
565 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
567 if (iattr->ia_valid & ATTR_ATIME)
568 inode->i_atime = iattr->ia_atime;
569 if (iattr->ia_valid & ATTR_CTIME)
570 inode->i_ctime = iattr->ia_ctime;
571 if (iattr->ia_valid & ATTR_MTIME)
572 inode->i_mtime = iattr->ia_mtime;
577 struct dentry *dentry,
580 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
582 if (mp->m_flags & XFS_MOUNT_RDONLY)
585 if (XFS_FORCED_SHUTDOWN(mp))
588 return setattr_prepare(dentry, iattr);
592 * Set non-size attributes of an inode.
594 * Caution: The caller of this function is responsible for calling
595 * setattr_prepare() or otherwise verifying the change is fine.
599 struct xfs_inode *ip,
603 xfs_mount_t *mp = ip->i_mount;
604 struct inode *inode = VFS_I(ip);
605 int mask = iattr->ia_valid;
608 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
609 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
610 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
611 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
613 ASSERT((mask & ATTR_SIZE) == 0);
616 * If disk quotas is on, we make sure that the dquots do exist on disk,
617 * before we start any other transactions. Trying to do this later
618 * is messy. We don't care to take a readlock to look at the ids
619 * in inode here, because we can't hold it across the trans_reserve.
620 * If the IDs do change before we take the ilock, we're covered
621 * because the i_*dquot fields will get updated anyway.
623 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
626 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
628 qflags |= XFS_QMOPT_UQUOTA;
632 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
634 qflags |= XFS_QMOPT_GQUOTA;
640 * We take a reference when we initialize udqp and gdqp,
641 * so it is important that we never blindly double trip on
642 * the same variable. See xfs_create() for an example.
644 ASSERT(udqp == NULL);
645 ASSERT(gdqp == NULL);
646 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
647 xfs_kgid_to_gid(gid),
649 qflags, &udqp, &gdqp, NULL);
654 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
658 xfs_ilock(ip, XFS_ILOCK_EXCL);
659 xfs_trans_ijoin(tp, ip, 0);
662 * Change file ownership. Must be the owner or privileged.
664 if (mask & (ATTR_UID|ATTR_GID)) {
666 * These IDs could have changed since we last looked at them.
667 * But, we're assured that if the ownership did change
668 * while we didn't have the inode locked, inode's dquot(s)
669 * would have changed also.
673 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
674 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
677 * Do a quota reservation only if uid/gid is actually
680 if (XFS_IS_QUOTA_RUNNING(mp) &&
681 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
682 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
684 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
685 NULL, capable(CAP_FOWNER) ?
686 XFS_QMOPT_FORCE_RES : 0);
687 if (error) /* out of quota */
693 * Change file ownership. Must be the owner or privileged.
695 if (mask & (ATTR_UID|ATTR_GID)) {
697 * CAP_FSETID overrides the following restrictions:
699 * The set-user-ID and set-group-ID bits of a file will be
700 * cleared upon successful return from chown()
702 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
703 !capable(CAP_FSETID))
704 inode->i_mode &= ~(S_ISUID|S_ISGID);
707 * Change the ownerships and register quota modifications
708 * in the transaction.
710 if (!uid_eq(iuid, uid)) {
711 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
712 ASSERT(mask & ATTR_UID);
714 olddquot1 = xfs_qm_vop_chown(tp, ip,
715 &ip->i_udquot, udqp);
717 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
720 if (!gid_eq(igid, gid)) {
721 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
722 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
723 !XFS_IS_PQUOTA_ON(mp));
724 ASSERT(mask & ATTR_GID);
726 olddquot2 = xfs_qm_vop_chown(tp, ip,
727 &ip->i_gdquot, gdqp);
729 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
734 if (mask & ATTR_MODE)
735 xfs_setattr_mode(ip, iattr);
736 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
737 xfs_setattr_time(ip, iattr);
739 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
741 XFS_STATS_INC(mp, xs_ig_attrchg);
743 if (mp->m_flags & XFS_MOUNT_WSYNC)
744 xfs_trans_set_sync(tp);
745 error = xfs_trans_commit(tp);
747 xfs_iunlock(ip, XFS_ILOCK_EXCL);
750 * Release any dquot(s) the inode had kept before chown.
752 xfs_qm_dqrele(olddquot1);
753 xfs_qm_dqrele(olddquot2);
761 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
762 * update. We could avoid this with linked transactions
763 * and passing down the transaction pointer all the way
764 * to attr_set. No previous user of the generic
765 * Posix ACL code seems to care about this issue either.
767 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
768 error = posix_acl_chmod(inode, inode->i_mode);
776 xfs_trans_cancel(tp);
777 xfs_iunlock(ip, XFS_ILOCK_EXCL);
785 xfs_vn_setattr_nonsize(
786 struct dentry *dentry,
789 struct xfs_inode *ip = XFS_I(d_inode(dentry));
792 trace_xfs_setattr(ip);
794 error = xfs_vn_change_ok(dentry, iattr);
797 return xfs_setattr_nonsize(ip, iattr, 0);
801 * Truncate file. Must have write permission and not be a directory.
803 * Caution: The caller of this function is responsible for calling
804 * setattr_prepare() or otherwise verifying the change is fine.
808 struct xfs_inode *ip,
811 struct xfs_mount *mp = ip->i_mount;
812 struct inode *inode = VFS_I(ip);
813 xfs_off_t oldsize, newsize;
814 struct xfs_trans *tp;
817 bool did_zeroing = false;
819 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
820 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
821 ASSERT(S_ISREG(inode->i_mode));
822 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
823 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
825 oldsize = inode->i_size;
826 newsize = iattr->ia_size;
829 * Short circuit the truncate case for zero length files.
831 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
832 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
836 * Use the regular setattr path to update the timestamps.
838 iattr->ia_valid &= ~ATTR_SIZE;
839 return xfs_setattr_nonsize(ip, iattr, 0);
843 * Make sure that the dquots are attached to the inode.
845 error = xfs_qm_dqattach(ip, 0);
850 * Wait for all direct I/O to complete.
852 inode_dio_wait(inode);
855 * File data changes must be complete before we start the transaction to
856 * modify the inode. This needs to be done before joining the inode to
857 * the transaction because the inode cannot be unlocked once it is a
858 * part of the transaction.
860 * Start with zeroing any data beyond EOF that we may expose on file
861 * extension, or zeroing out the rest of the block on a downward
864 if (newsize > oldsize) {
865 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
868 * iomap won't detect a dirty page over an unwritten block (or a
869 * cow block over a hole) and subsequently skips zeroing the
870 * newly post-EOF portion of the page. Flush the new EOF to
871 * convert the block before the pagecache truncate.
873 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
877 error = iomap_truncate_page(inode, newsize, &did_zeroing,
885 * We've already locked out new page faults, so now we can safely remove
886 * pages from the page cache knowing they won't get refaulted until we
887 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
888 * complete. The truncate_setsize() call also cleans partial EOF page
889 * PTEs on extending truncates and hence ensures sub-page block size
890 * filesystems are correctly handled, too.
892 * We have to do all the page cache truncate work outside the
893 * transaction context as the "lock" order is page lock->log space
894 * reservation as defined by extent allocation in the writeback path.
895 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
896 * having already truncated the in-memory version of the file (i.e. made
897 * user visible changes). There's not much we can do about this, except
898 * to hope that the caller sees ENOMEM and retries the truncate
901 * And we update in-core i_size and truncate page cache beyond newsize
902 * before writeback the [di_size, newsize] range, so we're guaranteed
903 * not to write stale data past the new EOF on truncate down.
905 truncate_setsize(inode, newsize);
908 * We are going to log the inode size change in this transaction so
909 * any previous writes that are beyond the on disk EOF and the new
910 * EOF that have not been written out need to be written here. If we
911 * do not write the data out, we expose ourselves to the null files
912 * problem. Note that this includes any block zeroing we did above;
913 * otherwise those blocks may not be zeroed after a crash.
916 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
917 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
918 ip->i_d.di_size, newsize - 1);
923 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
927 lock_flags |= XFS_ILOCK_EXCL;
928 xfs_ilock(ip, XFS_ILOCK_EXCL);
929 xfs_trans_ijoin(tp, ip, 0);
932 * Only change the c/mtime if we are changing the size or we are
933 * explicitly asked to change it. This handles the semantic difference
934 * between truncate() and ftruncate() as implemented in the VFS.
936 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
937 * special case where we need to update the times despite not having
938 * these flags set. For all other operations the VFS set these flags
939 * explicitly if it wants a timestamp update.
941 if (newsize != oldsize &&
942 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
943 iattr->ia_ctime = iattr->ia_mtime =
945 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
949 * The first thing we do is set the size to new_size permanently on
950 * disk. This way we don't have to worry about anyone ever being able
951 * to look at the data being freed even in the face of a crash.
952 * What we're getting around here is the case where we free a block, it
953 * is allocated to another file, it is written to, and then we crash.
954 * If the new data gets written to the file but the log buffers
955 * containing the free and reallocation don't, then we'd end up with
956 * garbage in the blocks being freed. As long as we make the new size
957 * permanent before actually freeing any blocks it doesn't matter if
958 * they get written to.
960 ip->i_d.di_size = newsize;
961 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
963 if (newsize <= oldsize) {
964 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
966 goto out_trans_cancel;
969 * Truncated "down", so we're removing references to old data
970 * here - if we delay flushing for a long time, we expose
971 * ourselves unduly to the notorious NULL files problem. So,
972 * we mark this inode and flush it when the file is closed,
973 * and do not wait the usual (long) time for writeout.
975 xfs_iflags_set(ip, XFS_ITRUNCATED);
977 /* A truncate down always removes post-EOF blocks. */
978 xfs_inode_clear_eofblocks_tag(ip);
981 if (iattr->ia_valid & ATTR_MODE)
982 xfs_setattr_mode(ip, iattr);
983 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
984 xfs_setattr_time(ip, iattr);
986 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
988 XFS_STATS_INC(mp, xs_ig_attrchg);
990 if (mp->m_flags & XFS_MOUNT_WSYNC)
991 xfs_trans_set_sync(tp);
993 error = xfs_trans_commit(tp);
996 xfs_iunlock(ip, lock_flags);
1000 xfs_trans_cancel(tp);
1005 xfs_vn_setattr_size(
1006 struct dentry *dentry,
1007 struct iattr *iattr)
1009 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1012 trace_xfs_setattr(ip);
1014 error = xfs_vn_change_ok(dentry, iattr);
1017 return xfs_setattr_size(ip, iattr);
1022 struct dentry *dentry,
1023 struct iattr *iattr)
1027 if (iattr->ia_valid & ATTR_SIZE) {
1028 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1029 uint iolock = XFS_IOLOCK_EXCL;
1031 xfs_ilock(ip, iolock);
1032 error = xfs_break_layouts(d_inode(dentry), &iolock, true);
1034 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1035 iolock |= XFS_MMAPLOCK_EXCL;
1037 error = xfs_vn_setattr_size(dentry, iattr);
1039 xfs_iunlock(ip, iolock);
1041 error = xfs_vn_setattr_nonsize(dentry, iattr);
1049 struct inode *inode,
1050 struct timespec *now,
1053 struct xfs_inode *ip = XFS_I(inode);
1054 struct xfs_mount *mp = ip->i_mount;
1055 struct xfs_trans *tp;
1058 trace_xfs_update_time(ip);
1060 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1064 xfs_ilock(ip, XFS_ILOCK_EXCL);
1065 if (flags & S_CTIME)
1066 inode->i_ctime = *now;
1067 if (flags & S_MTIME)
1068 inode->i_mtime = *now;
1069 if (flags & S_ATIME)
1070 inode->i_atime = *now;
1072 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1073 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1074 return xfs_trans_commit(tp);
1079 struct inode *inode,
1080 struct fiemap_extent_info *fieinfo,
1086 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1087 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1088 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1089 error = iomap_fiemap(inode, fieinfo, start, length,
1090 &xfs_xattr_iomap_ops);
1092 error = iomap_fiemap(inode, fieinfo, start, length,
1095 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1103 struct dentry *dentry,
1106 return xfs_generic_create(dir, dentry, mode, 0, true);
1109 static const struct inode_operations xfs_inode_operations = {
1110 .get_acl = xfs_get_acl,
1111 .set_acl = xfs_set_acl,
1112 .getattr = xfs_vn_getattr,
1113 .setattr = xfs_vn_setattr,
1114 .listxattr = xfs_vn_listxattr,
1115 .fiemap = xfs_vn_fiemap,
1116 .update_time = xfs_vn_update_time,
1119 static const struct inode_operations xfs_dir_inode_operations = {
1120 .create = xfs_vn_create,
1121 .lookup = xfs_vn_lookup,
1122 .link = xfs_vn_link,
1123 .unlink = xfs_vn_unlink,
1124 .symlink = xfs_vn_symlink,
1125 .mkdir = xfs_vn_mkdir,
1127 * Yes, XFS uses the same method for rmdir and unlink.
1129 * There are some subtile differences deeper in the code,
1130 * but we use S_ISDIR to check for those.
1132 .rmdir = xfs_vn_unlink,
1133 .mknod = xfs_vn_mknod,
1134 .rename = xfs_vn_rename,
1135 .get_acl = xfs_get_acl,
1136 .set_acl = xfs_set_acl,
1137 .getattr = xfs_vn_getattr,
1138 .setattr = xfs_vn_setattr,
1139 .listxattr = xfs_vn_listxattr,
1140 .update_time = xfs_vn_update_time,
1141 .tmpfile = xfs_vn_tmpfile,
1144 static const struct inode_operations xfs_dir_ci_inode_operations = {
1145 .create = xfs_vn_create,
1146 .lookup = xfs_vn_ci_lookup,
1147 .link = xfs_vn_link,
1148 .unlink = xfs_vn_unlink,
1149 .symlink = xfs_vn_symlink,
1150 .mkdir = xfs_vn_mkdir,
1152 * Yes, XFS uses the same method for rmdir and unlink.
1154 * There are some subtile differences deeper in the code,
1155 * but we use S_ISDIR to check for those.
1157 .rmdir = xfs_vn_unlink,
1158 .mknod = xfs_vn_mknod,
1159 .rename = xfs_vn_rename,
1160 .get_acl = xfs_get_acl,
1161 .set_acl = xfs_set_acl,
1162 .getattr = xfs_vn_getattr,
1163 .setattr = xfs_vn_setattr,
1164 .listxattr = xfs_vn_listxattr,
1165 .update_time = xfs_vn_update_time,
1166 .tmpfile = xfs_vn_tmpfile,
1169 static const struct inode_operations xfs_symlink_inode_operations = {
1170 .readlink = generic_readlink,
1171 .get_link = xfs_vn_get_link,
1172 .getattr = xfs_vn_getattr,
1173 .setattr = xfs_vn_setattr,
1174 .listxattr = xfs_vn_listxattr,
1175 .update_time = xfs_vn_update_time,
1178 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1179 .readlink = generic_readlink,
1180 .get_link = xfs_vn_get_link_inline,
1181 .getattr = xfs_vn_getattr,
1182 .setattr = xfs_vn_setattr,
1183 .listxattr = xfs_vn_listxattr,
1184 .update_time = xfs_vn_update_time,
1188 xfs_diflags_to_iflags(
1189 struct inode *inode,
1190 struct xfs_inode *ip)
1192 uint16_t flags = ip->i_d.di_flags;
1194 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1197 if (flags & XFS_DIFLAG_IMMUTABLE)
1198 inode->i_flags |= S_IMMUTABLE;
1199 if (flags & XFS_DIFLAG_APPEND)
1200 inode->i_flags |= S_APPEND;
1201 if (flags & XFS_DIFLAG_SYNC)
1202 inode->i_flags |= S_SYNC;
1203 if (flags & XFS_DIFLAG_NOATIME)
1204 inode->i_flags |= S_NOATIME;
1205 if (S_ISREG(inode->i_mode) &&
1206 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1207 !xfs_is_reflink_inode(ip) &&
1208 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1209 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1210 inode->i_flags |= S_DAX;
1214 * Initialize the Linux inode.
1216 * When reading existing inodes from disk this is called directly from xfs_iget,
1217 * when creating a new inode it is called from xfs_ialloc after setting up the
1218 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1219 * it up to the caller to deal with unlocking the inode appropriately.
1223 struct xfs_inode *ip)
1225 struct inode *inode = &ip->i_vnode;
1228 inode->i_ino = ip->i_ino;
1229 inode->i_state = I_NEW;
1231 inode_sb_list_add(inode);
1232 /* make the inode look hashed for the writeback code */
1233 hlist_add_fake(&inode->i_hash);
1235 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1236 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1238 switch (inode->i_mode & S_IFMT) {
1242 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1243 sysv_minor(ip->i_df.if_u2.if_rdev));
1250 i_size_write(inode, ip->i_d.di_size);
1251 xfs_diflags_to_iflags(inode, ip);
1253 if (S_ISDIR(inode->i_mode)) {
1254 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1255 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1257 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1258 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1262 * Ensure all page cache allocations are done from GFP_NOFS context to
1263 * prevent direct reclaim recursion back into the filesystem and blowing
1264 * stacks or deadlocking.
1266 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1267 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1270 * If there is no attribute fork no ACL can exist on this inode,
1271 * and it can't have any file capabilities attached to it either.
1273 if (!XFS_IFORK_Q(ip)) {
1274 inode_has_no_xattr(inode);
1275 cache_no_acl(inode);
1281 struct xfs_inode *ip)
1283 struct inode *inode = &ip->i_vnode;
1285 switch (inode->i_mode & S_IFMT) {
1287 inode->i_op = &xfs_inode_operations;
1288 inode->i_fop = &xfs_file_operations;
1289 inode->i_mapping->a_ops = &xfs_address_space_operations;
1292 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1293 inode->i_op = &xfs_dir_ci_inode_operations;
1295 inode->i_op = &xfs_dir_inode_operations;
1296 inode->i_fop = &xfs_dir_file_operations;
1299 if (ip->i_df.if_flags & XFS_IFINLINE)
1300 inode->i_op = &xfs_inline_symlink_inode_operations;
1302 inode->i_op = &xfs_symlink_inode_operations;
1305 inode->i_op = &xfs_inode_operations;
1306 init_special_inode(inode, inode->i_mode, inode->i_rdev);