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>
49 #include <linux/iversion.h>
52 * Directories have different lock order w.r.t. mmap_sem compared to regular
53 * files. This is due to readdir potentially triggering page faults on a user
54 * buffer inside filldir(), and this happens with the ilock on the directory
55 * held. For regular files, the lock order is the other way around - the
56 * mmap_sem is taken during the page fault, and then we lock the ilock to do
57 * block mapping. Hence we need a different class for the directory ilock so
58 * that lockdep can tell them apart.
60 static struct lock_class_key xfs_nondir_ilock_class;
61 static struct lock_class_key xfs_dir_ilock_class;
66 const struct xattr *xattr_array,
69 const struct xattr *xattr;
70 struct xfs_inode *ip = XFS_I(inode);
73 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
74 error = xfs_attr_set(ip, xattr->name, xattr->value,
75 xattr->value_len, ATTR_SECURE);
83 * Hook in SELinux. This is not quite correct yet, what we really need
84 * here (as we do for default ACLs) is a mechanism by which creation of
85 * these attrs can be journalled at inode creation time (along with the
86 * inode, of course, such that log replay can't cause these to be lost).
93 const struct qstr *qstr)
95 return security_inode_init_security(inode, dir, qstr,
96 &xfs_initxattrs, NULL);
101 struct xfs_name *namep,
102 struct dentry *dentry)
104 namep->name = dentry->d_name.name;
105 namep->len = dentry->d_name.len;
106 namep->type = XFS_DIR3_FT_UNKNOWN;
110 xfs_dentry_mode_to_name(
111 struct xfs_name *namep,
112 struct dentry *dentry,
115 namep->name = dentry->d_name.name;
116 namep->len = dentry->d_name.len;
117 namep->type = xfs_mode_to_ftype(mode);
119 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
120 return -EFSCORRUPTED;
129 struct dentry *dentry)
131 struct xfs_name teardown;
134 * If we can't add the ACL or we fail in
135 * xfs_init_security we must back out.
136 * ENOSPC can hit here, among other things.
138 xfs_dentry_to_name(&teardown, dentry);
140 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
146 struct dentry *dentry,
149 bool tmpfile) /* unnamed file */
152 struct xfs_inode *ip = NULL;
153 struct posix_acl *default_acl, *acl;
154 struct xfs_name name;
158 * Irix uses Missed'em'V split, but doesn't want to see
159 * the upper 5 bits of (14bit) major.
161 if (S_ISCHR(mode) || S_ISBLK(mode)) {
162 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
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), 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(XFS_SYMLINK_MAXLEN+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 const struct path *path,
495 unsigned int query_flags)
497 struct inode *inode = d_inode(path->dentry);
498 struct xfs_inode *ip = XFS_I(inode);
499 struct xfs_mount *mp = ip->i_mount;
501 trace_xfs_getattr(ip);
503 if (XFS_FORCED_SHUTDOWN(mp))
506 stat->size = XFS_ISIZE(ip);
507 stat->dev = inode->i_sb->s_dev;
508 stat->mode = inode->i_mode;
509 stat->nlink = inode->i_nlink;
510 stat->uid = inode->i_uid;
511 stat->gid = inode->i_gid;
512 stat->ino = ip->i_ino;
513 stat->atime = inode->i_atime;
514 stat->mtime = inode->i_mtime;
515 stat->ctime = inode->i_ctime;
517 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
519 if (ip->i_d.di_version == 3) {
520 if (request_mask & STATX_BTIME) {
521 stat->result_mask |= STATX_BTIME;
522 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
523 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
527 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
528 stat->attributes |= STATX_ATTR_IMMUTABLE;
529 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
530 stat->attributes |= STATX_ATTR_APPEND;
531 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
532 stat->attributes |= STATX_ATTR_NODUMP;
534 switch (inode->i_mode & S_IFMT) {
537 stat->blksize = BLKDEV_IOSIZE;
538 stat->rdev = inode->i_rdev;
541 if (XFS_IS_REALTIME_INODE(ip)) {
543 * If the file blocks are being allocated from a
544 * realtime volume, then return the inode's realtime
545 * extent size or the realtime volume's extent size.
548 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
550 stat->blksize = xfs_preferred_iosize(mp);
560 struct xfs_inode *ip,
563 struct inode *inode = VFS_I(ip);
564 umode_t mode = iattr->ia_mode;
566 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
568 inode->i_mode &= S_IFMT;
569 inode->i_mode |= mode & ~S_IFMT;
574 struct xfs_inode *ip,
577 struct inode *inode = VFS_I(ip);
579 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
581 if (iattr->ia_valid & ATTR_ATIME)
582 inode->i_atime = iattr->ia_atime;
583 if (iattr->ia_valid & ATTR_CTIME)
584 inode->i_ctime = iattr->ia_ctime;
585 if (iattr->ia_valid & ATTR_MTIME)
586 inode->i_mtime = iattr->ia_mtime;
591 struct dentry *dentry,
594 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
596 if (mp->m_flags & XFS_MOUNT_RDONLY)
599 if (XFS_FORCED_SHUTDOWN(mp))
602 return setattr_prepare(dentry, iattr);
606 * Set non-size attributes of an inode.
608 * Caution: The caller of this function is responsible for calling
609 * setattr_prepare() or otherwise verifying the change is fine.
613 struct xfs_inode *ip,
617 xfs_mount_t *mp = ip->i_mount;
618 struct inode *inode = VFS_I(ip);
619 int mask = iattr->ia_valid;
622 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
623 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
624 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
625 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
627 ASSERT((mask & ATTR_SIZE) == 0);
630 * If disk quotas is on, we make sure that the dquots do exist on disk,
631 * before we start any other transactions. Trying to do this later
632 * is messy. We don't care to take a readlock to look at the ids
633 * in inode here, because we can't hold it across the trans_reserve.
634 * If the IDs do change before we take the ilock, we're covered
635 * because the i_*dquot fields will get updated anyway.
637 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
640 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
642 qflags |= XFS_QMOPT_UQUOTA;
646 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
648 qflags |= XFS_QMOPT_GQUOTA;
654 * We take a reference when we initialize udqp and gdqp,
655 * so it is important that we never blindly double trip on
656 * the same variable. See xfs_create() for an example.
658 ASSERT(udqp == NULL);
659 ASSERT(gdqp == NULL);
660 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
661 xfs_kgid_to_gid(gid),
663 qflags, &udqp, &gdqp, NULL);
668 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
672 xfs_ilock(ip, XFS_ILOCK_EXCL);
673 xfs_trans_ijoin(tp, ip, 0);
676 * Change file ownership. Must be the owner or privileged.
678 if (mask & (ATTR_UID|ATTR_GID)) {
680 * These IDs could have changed since we last looked at them.
681 * But, we're assured that if the ownership did change
682 * while we didn't have the inode locked, inode's dquot(s)
683 * would have changed also.
687 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
688 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
691 * Do a quota reservation only if uid/gid is actually
694 if (XFS_IS_QUOTA_RUNNING(mp) &&
695 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
696 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
698 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
699 NULL, capable(CAP_FOWNER) ?
700 XFS_QMOPT_FORCE_RES : 0);
701 if (error) /* out of quota */
707 * Change file ownership. Must be the owner or privileged.
709 if (mask & (ATTR_UID|ATTR_GID)) {
711 * CAP_FSETID overrides the following restrictions:
713 * The set-user-ID and set-group-ID bits of a file will be
714 * cleared upon successful return from chown()
716 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
717 !capable(CAP_FSETID))
718 inode->i_mode &= ~(S_ISUID|S_ISGID);
721 * Change the ownerships and register quota modifications
722 * in the transaction.
724 if (!uid_eq(iuid, uid)) {
725 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
726 ASSERT(mask & ATTR_UID);
728 olddquot1 = xfs_qm_vop_chown(tp, ip,
729 &ip->i_udquot, udqp);
731 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
734 if (!gid_eq(igid, gid)) {
735 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
736 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
737 !XFS_IS_PQUOTA_ON(mp));
738 ASSERT(mask & ATTR_GID);
740 olddquot2 = xfs_qm_vop_chown(tp, ip,
741 &ip->i_gdquot, gdqp);
743 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
748 if (mask & ATTR_MODE)
749 xfs_setattr_mode(ip, iattr);
750 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
751 xfs_setattr_time(ip, iattr);
753 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
755 XFS_STATS_INC(mp, xs_ig_attrchg);
757 if (mp->m_flags & XFS_MOUNT_WSYNC)
758 xfs_trans_set_sync(tp);
759 error = xfs_trans_commit(tp);
761 xfs_iunlock(ip, XFS_ILOCK_EXCL);
764 * Release any dquot(s) the inode had kept before chown.
766 xfs_qm_dqrele(olddquot1);
767 xfs_qm_dqrele(olddquot2);
775 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
776 * update. We could avoid this with linked transactions
777 * and passing down the transaction pointer all the way
778 * to attr_set. No previous user of the generic
779 * Posix ACL code seems to care about this issue either.
781 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
782 error = posix_acl_chmod(inode, inode->i_mode);
790 xfs_trans_cancel(tp);
798 xfs_vn_setattr_nonsize(
799 struct dentry *dentry,
802 struct xfs_inode *ip = XFS_I(d_inode(dentry));
805 trace_xfs_setattr(ip);
807 error = xfs_vn_change_ok(dentry, iattr);
810 return xfs_setattr_nonsize(ip, iattr, 0);
814 * Truncate file. Must have write permission and not be a directory.
816 * Caution: The caller of this function is responsible for calling
817 * setattr_prepare() or otherwise verifying the change is fine.
821 struct xfs_inode *ip,
824 struct xfs_mount *mp = ip->i_mount;
825 struct inode *inode = VFS_I(ip);
826 xfs_off_t oldsize, newsize;
827 struct xfs_trans *tp;
830 bool did_zeroing = false;
832 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
833 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
834 ASSERT(S_ISREG(inode->i_mode));
835 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
836 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
838 oldsize = inode->i_size;
839 newsize = iattr->ia_size;
842 * Short circuit the truncate case for zero length files.
844 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
845 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
849 * Use the regular setattr path to update the timestamps.
851 iattr->ia_valid &= ~ATTR_SIZE;
852 return xfs_setattr_nonsize(ip, iattr, 0);
856 * Make sure that the dquots are attached to the inode.
858 error = xfs_qm_dqattach(ip);
863 * Wait for all direct I/O to complete.
865 inode_dio_wait(inode);
868 * File data changes must be complete before we start the transaction to
869 * modify the inode. This needs to be done before joining the inode to
870 * the transaction because the inode cannot be unlocked once it is a
871 * part of the transaction.
873 * Start with zeroing any data beyond EOF that we may expose on file
874 * extension, or zeroing out the rest of the block on a downward
877 if (newsize > oldsize) {
878 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
879 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
880 &did_zeroing, &xfs_iomap_ops);
882 error = iomap_truncate_page(inode, newsize, &did_zeroing,
890 * We've already locked out new page faults, so now we can safely remove
891 * pages from the page cache knowing they won't get refaulted until we
892 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
893 * complete. The truncate_setsize() call also cleans partial EOF page
894 * PTEs on extending truncates and hence ensures sub-page block size
895 * filesystems are correctly handled, too.
897 * We have to do all the page cache truncate work outside the
898 * transaction context as the "lock" order is page lock->log space
899 * reservation as defined by extent allocation in the writeback path.
900 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
901 * having already truncated the in-memory version of the file (i.e. made
902 * user visible changes). There's not much we can do about this, except
903 * to hope that the caller sees ENOMEM and retries the truncate
906 * And we update in-core i_size and truncate page cache beyond newsize
907 * before writeback the [di_size, newsize] range, so we're guaranteed
908 * not to write stale data past the new EOF on truncate down.
910 truncate_setsize(inode, newsize);
913 * We are going to log the inode size change in this transaction so
914 * any previous writes that are beyond the on disk EOF and the new
915 * EOF that have not been written out need to be written here. If we
916 * do not write the data out, we expose ourselves to the null files
917 * problem. Note that this includes any block zeroing we did above;
918 * otherwise those blocks may not be zeroed after a crash.
921 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
922 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
923 ip->i_d.di_size, newsize - 1);
928 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
932 lock_flags |= XFS_ILOCK_EXCL;
933 xfs_ilock(ip, XFS_ILOCK_EXCL);
934 xfs_trans_ijoin(tp, ip, 0);
937 * Only change the c/mtime if we are changing the size or we are
938 * explicitly asked to change it. This handles the semantic difference
939 * between truncate() and ftruncate() as implemented in the VFS.
941 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
942 * special case where we need to update the times despite not having
943 * these flags set. For all other operations the VFS set these flags
944 * explicitly if it wants a timestamp update.
946 if (newsize != oldsize &&
947 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
948 iattr->ia_ctime = iattr->ia_mtime =
950 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
954 * The first thing we do is set the size to new_size permanently on
955 * disk. This way we don't have to worry about anyone ever being able
956 * to look at the data being freed even in the face of a crash.
957 * What we're getting around here is the case where we free a block, it
958 * is allocated to another file, it is written to, and then we crash.
959 * If the new data gets written to the file but the log buffers
960 * containing the free and reallocation don't, then we'd end up with
961 * garbage in the blocks being freed. As long as we make the new size
962 * permanent before actually freeing any blocks it doesn't matter if
963 * they get written to.
965 ip->i_d.di_size = newsize;
966 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
968 if (newsize <= oldsize) {
969 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
971 goto out_trans_cancel;
974 * Truncated "down", so we're removing references to old data
975 * here - if we delay flushing for a long time, we expose
976 * ourselves unduly to the notorious NULL files problem. So,
977 * we mark this inode and flush it when the file is closed,
978 * and do not wait the usual (long) time for writeout.
980 xfs_iflags_set(ip, XFS_ITRUNCATED);
982 /* A truncate down always removes post-EOF blocks. */
983 xfs_inode_clear_eofblocks_tag(ip);
986 if (iattr->ia_valid & ATTR_MODE)
987 xfs_setattr_mode(ip, iattr);
988 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
989 xfs_setattr_time(ip, iattr);
991 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
993 XFS_STATS_INC(mp, xs_ig_attrchg);
995 if (mp->m_flags & XFS_MOUNT_WSYNC)
996 xfs_trans_set_sync(tp);
998 error = xfs_trans_commit(tp);
1001 xfs_iunlock(ip, lock_flags);
1005 xfs_trans_cancel(tp);
1010 xfs_vn_setattr_size(
1011 struct dentry *dentry,
1012 struct iattr *iattr)
1014 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1017 trace_xfs_setattr(ip);
1019 error = xfs_vn_change_ok(dentry, iattr);
1022 return xfs_setattr_size(ip, iattr);
1027 struct dentry *dentry,
1028 struct iattr *iattr)
1032 if (iattr->ia_valid & ATTR_SIZE) {
1033 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1034 uint iolock = XFS_IOLOCK_EXCL;
1036 error = xfs_break_layouts(d_inode(dentry), &iolock);
1040 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1041 error = xfs_vn_setattr_size(dentry, iattr);
1042 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1044 error = xfs_vn_setattr_nonsize(dentry, iattr);
1052 struct inode *inode,
1053 struct timespec *now,
1056 struct xfs_inode *ip = XFS_I(inode);
1057 struct xfs_mount *mp = ip->i_mount;
1058 int log_flags = XFS_ILOG_TIMESTAMP;
1059 struct xfs_trans *tp;
1062 trace_xfs_update_time(ip);
1064 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1065 if (!((flags & S_VERSION) &&
1066 inode_maybe_inc_iversion(inode, false)))
1067 return generic_update_time(inode, now, flags);
1069 /* Capture the iversion update that just occurred */
1070 log_flags |= XFS_ILOG_CORE;
1073 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1077 xfs_ilock(ip, XFS_ILOCK_EXCL);
1078 if (flags & S_CTIME)
1079 inode->i_ctime = *now;
1080 if (flags & S_MTIME)
1081 inode->i_mtime = *now;
1082 if (flags & S_ATIME)
1083 inode->i_atime = *now;
1085 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1086 xfs_trans_log_inode(tp, ip, log_flags);
1087 return xfs_trans_commit(tp);
1092 struct inode *inode,
1093 struct fiemap_extent_info *fieinfo,
1099 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1100 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1101 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1102 error = iomap_fiemap(inode, fieinfo, start, length,
1103 &xfs_xattr_iomap_ops);
1105 error = iomap_fiemap(inode, fieinfo, start, length,
1108 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1116 struct dentry *dentry,
1119 return xfs_generic_create(dir, dentry, mode, 0, true);
1122 static const struct inode_operations xfs_inode_operations = {
1123 .get_acl = xfs_get_acl,
1124 .set_acl = xfs_set_acl,
1125 .getattr = xfs_vn_getattr,
1126 .setattr = xfs_vn_setattr,
1127 .listxattr = xfs_vn_listxattr,
1128 .fiemap = xfs_vn_fiemap,
1129 .update_time = xfs_vn_update_time,
1132 static const struct inode_operations xfs_dir_inode_operations = {
1133 .create = xfs_vn_create,
1134 .lookup = xfs_vn_lookup,
1135 .link = xfs_vn_link,
1136 .unlink = xfs_vn_unlink,
1137 .symlink = xfs_vn_symlink,
1138 .mkdir = xfs_vn_mkdir,
1140 * Yes, XFS uses the same method for rmdir and unlink.
1142 * There are some subtile differences deeper in the code,
1143 * but we use S_ISDIR to check for those.
1145 .rmdir = xfs_vn_unlink,
1146 .mknod = xfs_vn_mknod,
1147 .rename = xfs_vn_rename,
1148 .get_acl = xfs_get_acl,
1149 .set_acl = xfs_set_acl,
1150 .getattr = xfs_vn_getattr,
1151 .setattr = xfs_vn_setattr,
1152 .listxattr = xfs_vn_listxattr,
1153 .update_time = xfs_vn_update_time,
1154 .tmpfile = xfs_vn_tmpfile,
1157 static const struct inode_operations xfs_dir_ci_inode_operations = {
1158 .create = xfs_vn_create,
1159 .lookup = xfs_vn_ci_lookup,
1160 .link = xfs_vn_link,
1161 .unlink = xfs_vn_unlink,
1162 .symlink = xfs_vn_symlink,
1163 .mkdir = xfs_vn_mkdir,
1165 * Yes, XFS uses the same method for rmdir and unlink.
1167 * There are some subtile differences deeper in the code,
1168 * but we use S_ISDIR to check for those.
1170 .rmdir = xfs_vn_unlink,
1171 .mknod = xfs_vn_mknod,
1172 .rename = xfs_vn_rename,
1173 .get_acl = xfs_get_acl,
1174 .set_acl = xfs_set_acl,
1175 .getattr = xfs_vn_getattr,
1176 .setattr = xfs_vn_setattr,
1177 .listxattr = xfs_vn_listxattr,
1178 .update_time = xfs_vn_update_time,
1179 .tmpfile = xfs_vn_tmpfile,
1182 static const struct inode_operations xfs_symlink_inode_operations = {
1183 .get_link = xfs_vn_get_link,
1184 .getattr = xfs_vn_getattr,
1185 .setattr = xfs_vn_setattr,
1186 .listxattr = xfs_vn_listxattr,
1187 .update_time = xfs_vn_update_time,
1190 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1191 .get_link = xfs_vn_get_link_inline,
1192 .getattr = xfs_vn_getattr,
1193 .setattr = xfs_vn_setattr,
1194 .listxattr = xfs_vn_listxattr,
1195 .update_time = xfs_vn_update_time,
1198 /* Figure out if this file actually supports DAX. */
1200 xfs_inode_supports_dax(
1201 struct xfs_inode *ip)
1203 struct xfs_mount *mp = ip->i_mount;
1205 /* Only supported on non-reflinked files. */
1206 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1209 /* DAX mount option or DAX iflag must be set. */
1210 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1211 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1214 /* Block size must match page size */
1215 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1218 /* Device has to support DAX too. */
1219 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1223 xfs_diflags_to_iflags(
1224 struct inode *inode,
1225 struct xfs_inode *ip)
1227 uint16_t flags = ip->i_d.di_flags;
1229 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1232 if (flags & XFS_DIFLAG_IMMUTABLE)
1233 inode->i_flags |= S_IMMUTABLE;
1234 if (flags & XFS_DIFLAG_APPEND)
1235 inode->i_flags |= S_APPEND;
1236 if (flags & XFS_DIFLAG_SYNC)
1237 inode->i_flags |= S_SYNC;
1238 if (flags & XFS_DIFLAG_NOATIME)
1239 inode->i_flags |= S_NOATIME;
1240 if (xfs_inode_supports_dax(ip))
1241 inode->i_flags |= S_DAX;
1245 * Initialize the Linux inode.
1247 * When reading existing inodes from disk this is called directly from xfs_iget,
1248 * when creating a new inode it is called from xfs_ialloc after setting up the
1249 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1250 * it up to the caller to deal with unlocking the inode appropriately.
1254 struct xfs_inode *ip)
1256 struct inode *inode = &ip->i_vnode;
1259 inode->i_ino = ip->i_ino;
1260 inode->i_state = I_NEW;
1262 inode_sb_list_add(inode);
1263 /* make the inode look hashed for the writeback code */
1264 hlist_add_fake(&inode->i_hash);
1266 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1267 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1269 i_size_write(inode, ip->i_d.di_size);
1270 xfs_diflags_to_iflags(inode, ip);
1272 if (S_ISDIR(inode->i_mode)) {
1273 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1274 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1276 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1277 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1281 * Ensure all page cache allocations are done from GFP_NOFS context to
1282 * prevent direct reclaim recursion back into the filesystem and blowing
1283 * stacks or deadlocking.
1285 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1286 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1289 * If there is no attribute fork no ACL can exist on this inode,
1290 * and it can't have any file capabilities attached to it either.
1292 if (!XFS_IFORK_Q(ip)) {
1293 inode_has_no_xattr(inode);
1294 cache_no_acl(inode);
1300 struct xfs_inode *ip)
1302 struct inode *inode = &ip->i_vnode;
1304 switch (inode->i_mode & S_IFMT) {
1306 inode->i_op = &xfs_inode_operations;
1307 inode->i_fop = &xfs_file_operations;
1309 inode->i_mapping->a_ops = &xfs_dax_aops;
1311 inode->i_mapping->a_ops = &xfs_address_space_operations;
1314 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1315 inode->i_op = &xfs_dir_ci_inode_operations;
1317 inode->i_op = &xfs_dir_inode_operations;
1318 inode->i_fop = &xfs_dir_file_operations;
1321 if (ip->i_df.if_flags & XFS_IFINLINE)
1322 inode->i_op = &xfs_inline_symlink_inode_operations;
1324 inode->i_op = &xfs_symlink_inode_operations;
1327 inode->i_op = &xfs_inode_operations;
1328 init_special_inode(inode, inode->i_mode, inode->i_rdev);