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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_inode.h"
30 #include "xfs_bmap_util.h"
32 #include "xfs_quota.h"
33 #include "xfs_error.h"
35 #include "xfs_trans.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_symlink.h"
39 #include "xfs_da_btree.h"
40 #include "xfs_dir2_priv.h"
41 #include "xfs_dinode.h"
43 #include <linux/capability.h>
44 #include <linux/xattr.h>
45 #include <linux/namei.h>
46 #include <linux/posix_acl.h>
47 #include <linux/security.h>
48 #include <linux/fiemap.h>
49 #include <linux/slab.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,
105 namep->name = dentry->d_name.name;
106 namep->len = dentry->d_name.len;
107 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
114 struct dentry *dentry)
116 struct xfs_name teardown;
119 * If we can't add the ACL or we fail in
120 * xfs_init_security we must back out.
121 * ENOSPC can hit here, among other things.
123 xfs_dentry_to_name(&teardown, dentry, 0);
125 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
132 struct dentry *dentry,
137 struct xfs_inode *ip = NULL;
138 struct posix_acl *default_acl, *acl;
139 struct xfs_name name;
143 * Irix uses Missed'em'V split, but doesn't want to see
144 * the upper 5 bits of (14bit) major.
146 if (S_ISCHR(mode) || S_ISBLK(mode)) {
147 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
149 rdev = sysv_encode_dev(rdev);
154 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);
165 error = xfs_init_security(inode, dir, &dentry->d_name);
167 goto out_cleanup_inode;
169 #ifdef CONFIG_XFS_POSIX_ACL
171 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
173 goto out_cleanup_inode;
176 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
178 goto out_cleanup_inode;
182 d_instantiate(dentry, inode);
185 posix_acl_release(default_acl);
187 posix_acl_release(acl);
191 xfs_cleanup_inode(dir, inode, dentry);
198 struct dentry *dentry,
202 return xfs_vn_mknod(dir, dentry, mode, 0);
208 struct dentry *dentry,
211 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
214 STATIC struct dentry *
217 struct dentry *dentry,
220 struct xfs_inode *cip;
221 struct xfs_name name;
224 if (dentry->d_name.len >= MAXNAMELEN)
225 return ERR_PTR(-ENAMETOOLONG);
227 xfs_dentry_to_name(&name, dentry, 0);
228 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
229 if (unlikely(error)) {
230 if (unlikely(error != ENOENT))
231 return ERR_PTR(-error);
236 return d_splice_alias(VFS_I(cip), dentry);
239 STATIC struct dentry *
242 struct dentry *dentry,
245 struct xfs_inode *ip;
246 struct xfs_name xname;
247 struct xfs_name ci_name;
251 if (dentry->d_name.len >= MAXNAMELEN)
252 return ERR_PTR(-ENAMETOOLONG);
254 xfs_dentry_to_name(&xname, dentry, 0);
255 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
256 if (unlikely(error)) {
257 if (unlikely(error != ENOENT))
258 return ERR_PTR(-error);
260 * call d_add(dentry, NULL) here when d_drop_negative_children
261 * is called in xfs_vn_mknod (ie. allow negative dentries
262 * with CI filesystems).
267 /* if exact match, just splice and exit */
269 return d_splice_alias(VFS_I(ip), dentry);
271 /* else case-insensitive match... */
272 dname.name = ci_name.name;
273 dname.len = ci_name.len;
274 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
275 kmem_free(ci_name.name);
281 struct dentry *old_dentry,
283 struct dentry *dentry)
285 struct inode *inode = old_dentry->d_inode;
286 struct xfs_name name;
289 xfs_dentry_to_name(&name, dentry, inode->i_mode);
291 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
296 d_instantiate(dentry, inode);
303 struct dentry *dentry)
305 struct xfs_name name;
308 xfs_dentry_to_name(&name, dentry, 0);
310 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
315 * With unlink, the VFS makes the dentry "negative": no inode,
316 * but still hashed. This is incompatible with case-insensitive
317 * mode, so invalidate (unhash) the dentry in CI-mode.
319 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
320 d_invalidate(dentry);
327 struct dentry *dentry,
331 struct xfs_inode *cip = NULL;
332 struct xfs_name name;
337 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
338 xfs_dentry_to_name(&name, dentry, mode);
340 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
346 error = xfs_init_security(inode, dir, &dentry->d_name);
348 goto out_cleanup_inode;
350 d_instantiate(dentry, inode);
354 xfs_cleanup_inode(dir, inode, dentry);
362 struct dentry *odentry,
364 struct dentry *ndentry)
366 struct inode *new_inode = ndentry->d_inode;
367 struct xfs_name oname;
368 struct xfs_name nname;
370 xfs_dentry_to_name(&oname, odentry, 0);
371 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
373 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
374 XFS_I(ndir), &nname, new_inode ?
375 XFS_I(new_inode) : NULL);
379 * careful here - this function can get called recursively, so
380 * we need to be very careful about how much stack we use.
381 * uio is kmalloced for this reason...
385 struct dentry *dentry,
386 struct nameidata *nd)
391 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
395 error = -xfs_readlink(XFS_I(dentry->d_inode), link);
399 nd_set_link(nd, link);
405 nd_set_link(nd, ERR_PTR(error));
411 struct vfsmount *mnt,
412 struct dentry *dentry,
415 struct inode *inode = dentry->d_inode;
416 struct xfs_inode *ip = XFS_I(inode);
417 struct xfs_mount *mp = ip->i_mount;
419 trace_xfs_getattr(ip);
421 if (XFS_FORCED_SHUTDOWN(mp))
422 return -XFS_ERROR(EIO);
424 stat->size = XFS_ISIZE(ip);
425 stat->dev = inode->i_sb->s_dev;
426 stat->mode = ip->i_d.di_mode;
427 stat->nlink = ip->i_d.di_nlink;
428 stat->uid = inode->i_uid;
429 stat->gid = inode->i_gid;
430 stat->ino = ip->i_ino;
431 stat->atime = inode->i_atime;
432 stat->mtime = inode->i_mtime;
433 stat->ctime = inode->i_ctime;
435 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
438 switch (inode->i_mode & S_IFMT) {
441 stat->blksize = BLKDEV_IOSIZE;
442 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
443 sysv_minor(ip->i_df.if_u2.if_rdev));
446 if (XFS_IS_REALTIME_INODE(ip)) {
448 * If the file blocks are being allocated from a
449 * realtime volume, then return the inode's realtime
450 * extent size or the realtime volume's extent size.
453 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
455 stat->blksize = xfs_preferred_iosize(mp);
465 struct xfs_inode *ip,
468 struct inode *inode = VFS_I(ip);
469 umode_t mode = iattr->ia_mode;
471 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
473 ip->i_d.di_mode &= S_IFMT;
474 ip->i_d.di_mode |= mode & ~S_IFMT;
476 inode->i_mode &= S_IFMT;
477 inode->i_mode |= mode & ~S_IFMT;
482 struct xfs_inode *ip,
485 struct inode *inode = VFS_I(ip);
487 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
489 if (iattr->ia_valid & ATTR_ATIME) {
490 inode->i_atime = iattr->ia_atime;
491 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
492 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
494 if (iattr->ia_valid & ATTR_CTIME) {
495 inode->i_ctime = iattr->ia_ctime;
496 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
497 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
499 if (iattr->ia_valid & ATTR_MTIME) {
500 inode->i_mtime = iattr->ia_mtime;
501 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
502 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
508 struct xfs_inode *ip,
512 xfs_mount_t *mp = ip->i_mount;
513 struct inode *inode = VFS_I(ip);
514 int mask = iattr->ia_valid;
517 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
518 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
519 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
520 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
522 trace_xfs_setattr(ip);
524 /* If acls are being inherited, we already have this checked */
525 if (!(flags & XFS_ATTR_NOACL)) {
526 if (mp->m_flags & XFS_MOUNT_RDONLY)
527 return XFS_ERROR(EROFS);
529 if (XFS_FORCED_SHUTDOWN(mp))
530 return XFS_ERROR(EIO);
532 error = -inode_change_ok(inode, iattr);
534 return XFS_ERROR(error);
537 ASSERT((mask & ATTR_SIZE) == 0);
540 * If disk quotas is on, we make sure that the dquots do exist on disk,
541 * before we start any other transactions. Trying to do this later
542 * is messy. We don't care to take a readlock to look at the ids
543 * in inode here, because we can't hold it across the trans_reserve.
544 * If the IDs do change before we take the ilock, we're covered
545 * because the i_*dquot fields will get updated anyway.
547 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
550 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
552 qflags |= XFS_QMOPT_UQUOTA;
556 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
558 qflags |= XFS_QMOPT_GQUOTA;
564 * We take a reference when we initialize udqp and gdqp,
565 * so it is important that we never blindly double trip on
566 * the same variable. See xfs_create() for an example.
568 ASSERT(udqp == NULL);
569 ASSERT(gdqp == NULL);
570 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
571 xfs_kgid_to_gid(gid),
573 qflags, &udqp, &gdqp, NULL);
578 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
579 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
583 xfs_ilock(ip, XFS_ILOCK_EXCL);
586 * Change file ownership. Must be the owner or privileged.
588 if (mask & (ATTR_UID|ATTR_GID)) {
590 * These IDs could have changed since we last looked at them.
591 * But, we're assured that if the ownership did change
592 * while we didn't have the inode locked, inode's dquot(s)
593 * would have changed also.
597 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
598 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
601 * Do a quota reservation only if uid/gid is actually
604 if (XFS_IS_QUOTA_RUNNING(mp) &&
605 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
606 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
608 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
609 NULL, capable(CAP_FOWNER) ?
610 XFS_QMOPT_FORCE_RES : 0);
611 if (error) /* out of quota */
612 goto out_trans_cancel;
616 xfs_trans_ijoin(tp, ip, 0);
619 * Change file ownership. Must be the owner or privileged.
621 if (mask & (ATTR_UID|ATTR_GID)) {
623 * CAP_FSETID overrides the following restrictions:
625 * The set-user-ID and set-group-ID bits of a file will be
626 * cleared upon successful return from chown()
628 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
629 !capable(CAP_FSETID))
630 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
633 * Change the ownerships and register quota modifications
634 * in the transaction.
636 if (!uid_eq(iuid, uid)) {
637 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
638 ASSERT(mask & ATTR_UID);
640 olddquot1 = xfs_qm_vop_chown(tp, ip,
641 &ip->i_udquot, udqp);
643 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
646 if (!gid_eq(igid, gid)) {
647 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
648 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
649 !XFS_IS_PQUOTA_ON(mp));
650 ASSERT(mask & ATTR_GID);
652 olddquot2 = xfs_qm_vop_chown(tp, ip,
653 &ip->i_gdquot, gdqp);
655 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
660 if (mask & ATTR_MODE)
661 xfs_setattr_mode(ip, iattr);
662 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
663 xfs_setattr_time(ip, iattr);
665 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
667 XFS_STATS_INC(xs_ig_attrchg);
669 if (mp->m_flags & XFS_MOUNT_WSYNC)
670 xfs_trans_set_sync(tp);
671 error = xfs_trans_commit(tp, 0);
673 xfs_iunlock(ip, XFS_ILOCK_EXCL);
676 * Release any dquot(s) the inode had kept before chown.
678 xfs_qm_dqrele(olddquot1);
679 xfs_qm_dqrele(olddquot2);
684 return XFS_ERROR(error);
687 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
688 * update. We could avoid this with linked transactions
689 * and passing down the transaction pointer all the way
690 * to attr_set. No previous user of the generic
691 * Posix ACL code seems to care about this issue either.
693 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
694 error = -posix_acl_chmod(inode, inode->i_mode);
696 return XFS_ERROR(error);
702 xfs_trans_cancel(tp, 0);
703 xfs_iunlock(ip, XFS_ILOCK_EXCL);
711 * Truncate file. Must have write permission and not be a directory.
715 struct xfs_inode *ip,
718 struct xfs_mount *mp = ip->i_mount;
719 struct inode *inode = VFS_I(ip);
720 xfs_off_t oldsize, newsize;
721 struct xfs_trans *tp;
724 uint commit_flags = 0;
726 trace_xfs_setattr(ip);
728 if (mp->m_flags & XFS_MOUNT_RDONLY)
729 return XFS_ERROR(EROFS);
731 if (XFS_FORCED_SHUTDOWN(mp))
732 return XFS_ERROR(EIO);
734 error = -inode_change_ok(inode, iattr);
736 return XFS_ERROR(error);
738 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
739 ASSERT(S_ISREG(ip->i_d.di_mode));
740 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
741 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
743 oldsize = inode->i_size;
744 newsize = iattr->ia_size;
747 * Short circuit the truncate case for zero length files.
749 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
750 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
754 * Use the regular setattr path to update the timestamps.
756 iattr->ia_valid &= ~ATTR_SIZE;
757 return xfs_setattr_nonsize(ip, iattr, 0);
761 * Make sure that the dquots are attached to the inode.
763 error = xfs_qm_dqattach(ip, 0);
768 * Now we can make the changes. Before we join the inode to the
769 * transaction, take care of the part of the truncation that must be
770 * done without the inode lock. This needs to be done before joining
771 * the inode to the transaction, because the inode cannot be unlocked
772 * once it is a part of the transaction.
774 if (newsize > oldsize) {
776 * Do the first part of growing a file: zero any data in the
777 * last block that is beyond the old EOF. We need to do this
778 * before the inode is joined to the transaction to modify
781 error = xfs_zero_eof(ip, newsize, oldsize);
787 * We are going to log the inode size change in this transaction so
788 * any previous writes that are beyond the on disk EOF and the new
789 * EOF that have not been written out need to be written here. If we
790 * do not write the data out, we expose ourselves to the null files
793 * Only flush from the on disk size to the smaller of the in memory
794 * file size or the new size as that's the range we really care about
795 * here and prevents waiting for other data not within the range we
798 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
799 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
800 ip->i_d.di_size, newsize);
806 * Wait for all direct I/O to complete.
808 inode_dio_wait(inode);
810 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
814 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
815 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
817 goto out_trans_cancel;
819 truncate_setsize(inode, newsize);
821 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
822 lock_flags |= XFS_ILOCK_EXCL;
824 xfs_ilock(ip, XFS_ILOCK_EXCL);
826 xfs_trans_ijoin(tp, ip, 0);
829 * Only change the c/mtime if we are changing the size or we are
830 * explicitly asked to change it. This handles the semantic difference
831 * between truncate() and ftruncate() as implemented in the VFS.
833 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
834 * special case where we need to update the times despite not having
835 * these flags set. For all other operations the VFS set these flags
836 * explicitly if it wants a timestamp update.
838 if (newsize != oldsize &&
839 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
840 iattr->ia_ctime = iattr->ia_mtime =
841 current_fs_time(inode->i_sb);
842 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
846 * The first thing we do is set the size to new_size permanently on
847 * disk. This way we don't have to worry about anyone ever being able
848 * to look at the data being freed even in the face of a crash.
849 * What we're getting around here is the case where we free a block, it
850 * is allocated to another file, it is written to, and then we crash.
851 * If the new data gets written to the file but the log buffers
852 * containing the free and reallocation don't, then we'd end up with
853 * garbage in the blocks being freed. As long as we make the new size
854 * permanent before actually freeing any blocks it doesn't matter if
855 * they get written to.
857 ip->i_d.di_size = newsize;
858 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
860 if (newsize <= oldsize) {
861 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
863 goto out_trans_abort;
866 * Truncated "down", so we're removing references to old data
867 * here - if we delay flushing for a long time, we expose
868 * ourselves unduly to the notorious NULL files problem. So,
869 * we mark this inode and flush it when the file is closed,
870 * and do not wait the usual (long) time for writeout.
872 xfs_iflags_set(ip, XFS_ITRUNCATED);
874 /* A truncate down always removes post-EOF blocks. */
875 xfs_inode_clear_eofblocks_tag(ip);
878 if (iattr->ia_valid & ATTR_MODE)
879 xfs_setattr_mode(ip, iattr);
880 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
881 xfs_setattr_time(ip, iattr);
883 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
885 XFS_STATS_INC(xs_ig_attrchg);
887 if (mp->m_flags & XFS_MOUNT_WSYNC)
888 xfs_trans_set_sync(tp);
890 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
893 xfs_iunlock(ip, lock_flags);
897 commit_flags |= XFS_TRANS_ABORT;
899 xfs_trans_cancel(tp, commit_flags);
905 struct dentry *dentry,
908 struct xfs_inode *ip = XFS_I(dentry->d_inode);
911 if (iattr->ia_valid & ATTR_SIZE) {
912 xfs_ilock(ip, XFS_IOLOCK_EXCL);
913 error = xfs_setattr_size(ip, iattr);
914 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
916 error = xfs_setattr_nonsize(ip, iattr, 0);
925 struct timespec *now,
928 struct xfs_inode *ip = XFS_I(inode);
929 struct xfs_mount *mp = ip->i_mount;
930 struct xfs_trans *tp;
933 trace_xfs_update_time(ip);
935 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
936 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
938 xfs_trans_cancel(tp, 0);
942 xfs_ilock(ip, XFS_ILOCK_EXCL);
943 if (flags & S_CTIME) {
944 inode->i_ctime = *now;
945 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
946 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
948 if (flags & S_MTIME) {
949 inode->i_mtime = *now;
950 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
951 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
953 if (flags & S_ATIME) {
954 inode->i_atime = *now;
955 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
956 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
958 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
959 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
960 return -xfs_trans_commit(tp, 0);
963 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
966 * Call fiemap helper to fill in user data.
967 * Returns positive errors to xfs_getbmap.
972 struct getbmapx *bmv,
976 struct fiemap_extent_info *fieinfo = *arg;
977 u32 fiemap_flags = 0;
978 u64 logical, physical, length;
980 /* Do nothing for a hole */
981 if (bmv->bmv_block == -1LL)
984 logical = BBTOB(bmv->bmv_offset);
985 physical = BBTOB(bmv->bmv_block);
986 length = BBTOB(bmv->bmv_length);
988 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
989 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
990 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
991 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
992 FIEMAP_EXTENT_UNKNOWN);
993 physical = 0; /* no block yet */
995 if (bmv->bmv_oflags & BMV_OF_LAST)
996 fiemap_flags |= FIEMAP_EXTENT_LAST;
998 error = fiemap_fill_next_extent(fieinfo, logical, physical,
999 length, fiemap_flags);
1002 *full = 1; /* user array now full */
1010 struct inode *inode,
1011 struct fiemap_extent_info *fieinfo,
1015 xfs_inode_t *ip = XFS_I(inode);
1019 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1023 /* Set up bmap header for xfs internal routine */
1024 bm.bmv_offset = BTOBB(start);
1025 /* Special case for whole file */
1026 if (length == FIEMAP_MAX_OFFSET)
1027 bm.bmv_length = -1LL;
1029 bm.bmv_length = BTOBB(length);
1031 /* We add one because in getbmap world count includes the header */
1032 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1033 fieinfo->fi_extents_max + 1;
1034 bm.bmv_count = min_t(__s32, bm.bmv_count,
1035 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1036 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1037 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1038 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1039 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1040 bm.bmv_iflags |= BMV_IF_DELALLOC;
1042 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1049 static const struct inode_operations xfs_inode_operations = {
1050 .get_acl = xfs_get_acl,
1051 .set_acl = xfs_set_acl,
1052 .getattr = xfs_vn_getattr,
1053 .setattr = xfs_vn_setattr,
1054 .setxattr = generic_setxattr,
1055 .getxattr = generic_getxattr,
1056 .removexattr = generic_removexattr,
1057 .listxattr = xfs_vn_listxattr,
1058 .fiemap = xfs_vn_fiemap,
1059 .update_time = xfs_vn_update_time,
1062 static const struct inode_operations xfs_dir_inode_operations = {
1063 .create = xfs_vn_create,
1064 .lookup = xfs_vn_lookup,
1065 .link = xfs_vn_link,
1066 .unlink = xfs_vn_unlink,
1067 .symlink = xfs_vn_symlink,
1068 .mkdir = xfs_vn_mkdir,
1070 * Yes, XFS uses the same method for rmdir and unlink.
1072 * There are some subtile differences deeper in the code,
1073 * but we use S_ISDIR to check for those.
1075 .rmdir = xfs_vn_unlink,
1076 .mknod = xfs_vn_mknod,
1077 .rename = xfs_vn_rename,
1078 .get_acl = xfs_get_acl,
1079 .set_acl = xfs_set_acl,
1080 .getattr = xfs_vn_getattr,
1081 .setattr = xfs_vn_setattr,
1082 .setxattr = generic_setxattr,
1083 .getxattr = generic_getxattr,
1084 .removexattr = generic_removexattr,
1085 .listxattr = xfs_vn_listxattr,
1086 .update_time = xfs_vn_update_time,
1089 static const struct inode_operations xfs_dir_ci_inode_operations = {
1090 .create = xfs_vn_create,
1091 .lookup = xfs_vn_ci_lookup,
1092 .link = xfs_vn_link,
1093 .unlink = xfs_vn_unlink,
1094 .symlink = xfs_vn_symlink,
1095 .mkdir = xfs_vn_mkdir,
1097 * Yes, XFS uses the same method for rmdir and unlink.
1099 * There are some subtile differences deeper in the code,
1100 * but we use S_ISDIR to check for those.
1102 .rmdir = xfs_vn_unlink,
1103 .mknod = xfs_vn_mknod,
1104 .rename = xfs_vn_rename,
1105 .get_acl = xfs_get_acl,
1106 .set_acl = xfs_set_acl,
1107 .getattr = xfs_vn_getattr,
1108 .setattr = xfs_vn_setattr,
1109 .setxattr = generic_setxattr,
1110 .getxattr = generic_getxattr,
1111 .removexattr = generic_removexattr,
1112 .listxattr = xfs_vn_listxattr,
1113 .update_time = xfs_vn_update_time,
1116 static const struct inode_operations xfs_symlink_inode_operations = {
1117 .readlink = generic_readlink,
1118 .follow_link = xfs_vn_follow_link,
1119 .put_link = kfree_put_link,
1120 .getattr = xfs_vn_getattr,
1121 .setattr = xfs_vn_setattr,
1122 .setxattr = generic_setxattr,
1123 .getxattr = generic_getxattr,
1124 .removexattr = generic_removexattr,
1125 .listxattr = xfs_vn_listxattr,
1126 .update_time = xfs_vn_update_time,
1130 xfs_diflags_to_iflags(
1131 struct inode *inode,
1132 struct xfs_inode *ip)
1134 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1135 inode->i_flags |= S_IMMUTABLE;
1137 inode->i_flags &= ~S_IMMUTABLE;
1138 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1139 inode->i_flags |= S_APPEND;
1141 inode->i_flags &= ~S_APPEND;
1142 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1143 inode->i_flags |= S_SYNC;
1145 inode->i_flags &= ~S_SYNC;
1146 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1147 inode->i_flags |= S_NOATIME;
1149 inode->i_flags &= ~S_NOATIME;
1153 * Initialize the Linux inode, set up the operation vectors and
1156 * When reading existing inodes from disk this is called directly
1157 * from xfs_iget, when creating a new inode it is called from
1158 * xfs_ialloc after setting up the inode.
1160 * We are always called with an uninitialised linux inode here.
1161 * We need to initialise the necessary fields and take a reference
1166 struct xfs_inode *ip)
1168 struct inode *inode = &ip->i_vnode;
1171 inode->i_ino = ip->i_ino;
1172 inode->i_state = I_NEW;
1174 inode_sb_list_add(inode);
1175 /* make the inode look hashed for the writeback code */
1176 hlist_add_fake(&inode->i_hash);
1178 inode->i_mode = ip->i_d.di_mode;
1179 set_nlink(inode, ip->i_d.di_nlink);
1180 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1181 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1183 switch (inode->i_mode & S_IFMT) {
1187 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1188 sysv_minor(ip->i_df.if_u2.if_rdev));
1195 inode->i_generation = ip->i_d.di_gen;
1196 i_size_write(inode, ip->i_d.di_size);
1197 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1198 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1199 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1200 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1201 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1202 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1203 xfs_diflags_to_iflags(inode, ip);
1205 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1206 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1207 switch (inode->i_mode & S_IFMT) {
1209 inode->i_op = &xfs_inode_operations;
1210 inode->i_fop = &xfs_file_operations;
1211 inode->i_mapping->a_ops = &xfs_address_space_operations;
1214 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1215 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1216 inode->i_op = &xfs_dir_ci_inode_operations;
1218 inode->i_op = &xfs_dir_inode_operations;
1219 inode->i_fop = &xfs_dir_file_operations;
1220 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1223 inode->i_op = &xfs_symlink_inode_operations;
1224 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1225 inode->i_mapping->a_ops = &xfs_address_space_operations;
1228 inode->i_op = &xfs_inode_operations;
1229 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1234 * Ensure all page cache allocations are done from GFP_NOFS context to
1235 * prevent direct reclaim recursion back into the filesystem and blowing
1236 * stacks or deadlocking.
1238 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1239 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1242 * If there is no attribute fork no ACL can exist on this inode,
1243 * and it can't have any file capabilities attached to it either.
1245 if (!XFS_IFORK_Q(ip)) {
1246 inode_has_no_xattr(inode);
1247 cache_no_acl(inode);
1250 xfs_iflags_clear(ip, XFS_INEW);
1253 unlock_new_inode(inode);