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>
54 const struct xattr *xattr_array,
57 const struct xattr *xattr;
58 struct xfs_inode *ip = XFS_I(inode);
61 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
62 error = xfs_attr_set(ip, xattr->name, xattr->value,
63 xattr->value_len, ATTR_SECURE);
71 * Hook in SELinux. This is not quite correct yet, what we really need
72 * here (as we do for default ACLs) is a mechanism by which creation of
73 * these attrs can be journalled at inode creation time (along with the
74 * inode, of course, such that log replay can't cause these to be lost).
81 const struct qstr *qstr)
83 return security_inode_init_security(inode, dir, qstr,
84 &xfs_initxattrs, NULL);
89 struct xfs_name *namep,
90 struct dentry *dentry,
93 namep->name = dentry->d_name.name;
94 namep->len = dentry->d_name.len;
95 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
102 struct dentry *dentry)
104 struct xfs_name teardown;
107 * If we can't add the ACL or we fail in
108 * xfs_init_security we must back out.
109 * ENOSPC can hit here, among other things.
111 xfs_dentry_to_name(&teardown, dentry, 0);
113 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
120 struct dentry *dentry,
125 struct xfs_inode *ip = NULL;
126 struct posix_acl *default_acl = NULL;
127 struct xfs_name name;
131 * Irix uses Missed'em'V split, but doesn't want to see
132 * the upper 5 bits of (14bit) major.
134 if (S_ISCHR(mode) || S_ISBLK(mode)) {
135 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
137 rdev = sysv_encode_dev(rdev);
142 if (IS_POSIXACL(dir)) {
143 default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
144 if (IS_ERR(default_acl))
145 return PTR_ERR(default_acl);
148 mode &= ~current_umask();
151 xfs_dentry_to_name(&name, dentry, mode);
152 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
158 error = xfs_init_security(inode, dir, &dentry->d_name);
160 goto out_cleanup_inode;
163 error = -xfs_inherit_acl(inode, default_acl);
166 goto out_cleanup_inode;
170 d_instantiate(dentry, inode);
174 xfs_cleanup_inode(dir, inode, dentry);
176 posix_acl_release(default_acl);
183 struct dentry *dentry,
187 return xfs_vn_mknod(dir, dentry, mode, 0);
193 struct dentry *dentry,
196 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
199 STATIC struct dentry *
202 struct dentry *dentry,
205 struct xfs_inode *cip;
206 struct xfs_name name;
209 if (dentry->d_name.len >= MAXNAMELEN)
210 return ERR_PTR(-ENAMETOOLONG);
212 xfs_dentry_to_name(&name, dentry, 0);
213 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
214 if (unlikely(error)) {
215 if (unlikely(error != ENOENT))
216 return ERR_PTR(-error);
221 return d_splice_alias(VFS_I(cip), dentry);
224 STATIC struct dentry *
227 struct dentry *dentry,
230 struct xfs_inode *ip;
231 struct xfs_name xname;
232 struct xfs_name ci_name;
236 if (dentry->d_name.len >= MAXNAMELEN)
237 return ERR_PTR(-ENAMETOOLONG);
239 xfs_dentry_to_name(&xname, dentry, 0);
240 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
241 if (unlikely(error)) {
242 if (unlikely(error != ENOENT))
243 return ERR_PTR(-error);
245 * call d_add(dentry, NULL) here when d_drop_negative_children
246 * is called in xfs_vn_mknod (ie. allow negative dentries
247 * with CI filesystems).
252 /* if exact match, just splice and exit */
254 return d_splice_alias(VFS_I(ip), dentry);
256 /* else case-insensitive match... */
257 dname.name = ci_name.name;
258 dname.len = ci_name.len;
259 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
260 kmem_free(ci_name.name);
266 struct dentry *old_dentry,
268 struct dentry *dentry)
270 struct inode *inode = old_dentry->d_inode;
271 struct xfs_name name;
274 xfs_dentry_to_name(&name, dentry, inode->i_mode);
276 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
281 d_instantiate(dentry, inode);
288 struct dentry *dentry)
290 struct xfs_name name;
293 xfs_dentry_to_name(&name, dentry, 0);
295 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
300 * With unlink, the VFS makes the dentry "negative": no inode,
301 * but still hashed. This is incompatible with case-insensitive
302 * mode, so invalidate (unhash) the dentry in CI-mode.
304 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
305 d_invalidate(dentry);
312 struct dentry *dentry,
316 struct xfs_inode *cip = NULL;
317 struct xfs_name name;
322 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
323 xfs_dentry_to_name(&name, dentry, mode);
325 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
331 error = xfs_init_security(inode, dir, &dentry->d_name);
333 goto out_cleanup_inode;
335 d_instantiate(dentry, inode);
339 xfs_cleanup_inode(dir, inode, dentry);
347 struct dentry *odentry,
349 struct dentry *ndentry)
351 struct inode *new_inode = ndentry->d_inode;
352 struct xfs_name oname;
353 struct xfs_name nname;
355 xfs_dentry_to_name(&oname, odentry, 0);
356 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
358 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
359 XFS_I(ndir), &nname, new_inode ?
360 XFS_I(new_inode) : NULL);
364 * careful here - this function can get called recursively, so
365 * we need to be very careful about how much stack we use.
366 * uio is kmalloced for this reason...
370 struct dentry *dentry,
371 struct nameidata *nd)
376 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
380 error = -xfs_readlink(XFS_I(dentry->d_inode), link);
384 nd_set_link(nd, link);
390 nd_set_link(nd, ERR_PTR(error));
396 struct vfsmount *mnt,
397 struct dentry *dentry,
400 struct inode *inode = dentry->d_inode;
401 struct xfs_inode *ip = XFS_I(inode);
402 struct xfs_mount *mp = ip->i_mount;
404 trace_xfs_getattr(ip);
406 if (XFS_FORCED_SHUTDOWN(mp))
407 return -XFS_ERROR(EIO);
409 stat->size = XFS_ISIZE(ip);
410 stat->dev = inode->i_sb->s_dev;
411 stat->mode = ip->i_d.di_mode;
412 stat->nlink = ip->i_d.di_nlink;
413 stat->uid = inode->i_uid;
414 stat->gid = inode->i_gid;
415 stat->ino = ip->i_ino;
416 stat->atime = inode->i_atime;
417 stat->mtime = inode->i_mtime;
418 stat->ctime = inode->i_ctime;
420 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
423 switch (inode->i_mode & S_IFMT) {
426 stat->blksize = BLKDEV_IOSIZE;
427 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
428 sysv_minor(ip->i_df.if_u2.if_rdev));
431 if (XFS_IS_REALTIME_INODE(ip)) {
433 * If the file blocks are being allocated from a
434 * realtime volume, then return the inode's realtime
435 * extent size or the realtime volume's extent size.
438 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
440 stat->blksize = xfs_preferred_iosize(mp);
450 struct xfs_trans *tp,
451 struct xfs_inode *ip,
454 struct inode *inode = VFS_I(ip);
455 umode_t mode = iattr->ia_mode;
458 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
460 ip->i_d.di_mode &= S_IFMT;
461 ip->i_d.di_mode |= mode & ~S_IFMT;
463 inode->i_mode &= S_IFMT;
464 inode->i_mode |= mode & ~S_IFMT;
469 struct xfs_inode *ip,
473 xfs_mount_t *mp = ip->i_mount;
474 struct inode *inode = VFS_I(ip);
475 int mask = iattr->ia_valid;
478 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
479 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
480 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
481 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
483 trace_xfs_setattr(ip);
485 /* If acls are being inherited, we already have this checked */
486 if (!(flags & XFS_ATTR_NOACL)) {
487 if (mp->m_flags & XFS_MOUNT_RDONLY)
488 return XFS_ERROR(EROFS);
490 if (XFS_FORCED_SHUTDOWN(mp))
491 return XFS_ERROR(EIO);
493 error = -inode_change_ok(inode, iattr);
495 return XFS_ERROR(error);
498 ASSERT((mask & ATTR_SIZE) == 0);
501 * If disk quotas is on, we make sure that the dquots do exist on disk,
502 * before we start any other transactions. Trying to do this later
503 * is messy. We don't care to take a readlock to look at the ids
504 * in inode here, because we can't hold it across the trans_reserve.
505 * If the IDs do change before we take the ilock, we're covered
506 * because the i_*dquot fields will get updated anyway.
508 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
511 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
513 qflags |= XFS_QMOPT_UQUOTA;
517 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
519 qflags |= XFS_QMOPT_GQUOTA;
525 * We take a reference when we initialize udqp and gdqp,
526 * so it is important that we never blindly double trip on
527 * the same variable. See xfs_create() for an example.
529 ASSERT(udqp == NULL);
530 ASSERT(gdqp == NULL);
531 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
532 xfs_kgid_to_gid(gid),
534 qflags, &udqp, &gdqp, NULL);
539 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
540 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
544 xfs_ilock(ip, XFS_ILOCK_EXCL);
547 * Change file ownership. Must be the owner or privileged.
549 if (mask & (ATTR_UID|ATTR_GID)) {
551 * These IDs could have changed since we last looked at them.
552 * But, we're assured that if the ownership did change
553 * while we didn't have the inode locked, inode's dquot(s)
554 * would have changed also.
558 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
559 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
562 * Do a quota reservation only if uid/gid is actually
565 if (XFS_IS_QUOTA_RUNNING(mp) &&
566 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
567 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
569 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
570 NULL, capable(CAP_FOWNER) ?
571 XFS_QMOPT_FORCE_RES : 0);
572 if (error) /* out of quota */
573 goto out_trans_cancel;
577 xfs_trans_ijoin(tp, ip, 0);
580 * Change file ownership. Must be the owner or privileged.
582 if (mask & (ATTR_UID|ATTR_GID)) {
584 * CAP_FSETID overrides the following restrictions:
586 * The set-user-ID and set-group-ID bits of a file will be
587 * cleared upon successful return from chown()
589 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
590 !capable(CAP_FSETID))
591 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
594 * Change the ownerships and register quota modifications
595 * in the transaction.
597 if (!uid_eq(iuid, uid)) {
598 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
599 ASSERT(mask & ATTR_UID);
601 olddquot1 = xfs_qm_vop_chown(tp, ip,
602 &ip->i_udquot, udqp);
604 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
607 if (!gid_eq(igid, gid)) {
608 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
609 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
610 !XFS_IS_PQUOTA_ON(mp));
611 ASSERT(mask & ATTR_GID);
613 olddquot2 = xfs_qm_vop_chown(tp, ip,
614 &ip->i_gdquot, gdqp);
616 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
622 * Change file access modes.
624 if (mask & ATTR_MODE)
625 xfs_setattr_mode(tp, ip, iattr);
628 * Change file access or modified times.
630 if (mask & ATTR_ATIME) {
631 inode->i_atime = iattr->ia_atime;
632 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
633 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
635 if (mask & ATTR_CTIME) {
636 inode->i_ctime = iattr->ia_ctime;
637 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
638 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
640 if (mask & ATTR_MTIME) {
641 inode->i_mtime = iattr->ia_mtime;
642 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
643 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
646 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
648 XFS_STATS_INC(xs_ig_attrchg);
650 if (mp->m_flags & XFS_MOUNT_WSYNC)
651 xfs_trans_set_sync(tp);
652 error = xfs_trans_commit(tp, 0);
654 xfs_iunlock(ip, XFS_ILOCK_EXCL);
657 * Release any dquot(s) the inode had kept before chown.
659 xfs_qm_dqrele(olddquot1);
660 xfs_qm_dqrele(olddquot2);
665 return XFS_ERROR(error);
668 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
669 * update. We could avoid this with linked transactions
670 * and passing down the transaction pointer all the way
671 * to attr_set. No previous user of the generic
672 * Posix ACL code seems to care about this issue either.
674 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
675 error = -xfs_acl_chmod(inode);
677 return XFS_ERROR(error);
683 xfs_trans_cancel(tp, 0);
684 xfs_iunlock(ip, XFS_ILOCK_EXCL);
692 * Truncate file. Must have write permission and not be a directory.
696 struct xfs_inode *ip,
699 struct xfs_mount *mp = ip->i_mount;
700 struct inode *inode = VFS_I(ip);
701 int mask = iattr->ia_valid;
702 xfs_off_t oldsize, newsize;
703 struct xfs_trans *tp;
706 uint commit_flags = 0;
708 trace_xfs_setattr(ip);
710 if (mp->m_flags & XFS_MOUNT_RDONLY)
711 return XFS_ERROR(EROFS);
713 if (XFS_FORCED_SHUTDOWN(mp))
714 return XFS_ERROR(EIO);
716 error = -inode_change_ok(inode, iattr);
718 return XFS_ERROR(error);
720 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
721 ASSERT(S_ISREG(ip->i_d.di_mode));
722 ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
723 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
725 oldsize = inode->i_size;
726 newsize = iattr->ia_size;
729 * Short circuit the truncate case for zero length files.
731 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
732 if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
736 * Use the regular setattr path to update the timestamps.
738 iattr->ia_valid &= ~ATTR_SIZE;
739 return xfs_setattr_nonsize(ip, iattr, 0);
743 * Make sure that the dquots are attached to the inode.
745 error = xfs_qm_dqattach(ip, 0);
750 * Now we can make the changes. Before we join the inode to the
751 * transaction, take care of the part of the truncation that must be
752 * done without the inode lock. This needs to be done before joining
753 * the inode to the transaction, because the inode cannot be unlocked
754 * once it is a part of the transaction.
756 if (newsize > oldsize) {
758 * Do the first part of growing a file: zero any data in the
759 * last block that is beyond the old EOF. We need to do this
760 * before the inode is joined to the transaction to modify
763 error = xfs_zero_eof(ip, newsize, oldsize);
769 * We are going to log the inode size change in this transaction so
770 * any previous writes that are beyond the on disk EOF and the new
771 * EOF that have not been written out need to be written here. If we
772 * do not write the data out, we expose ourselves to the null files
775 * Only flush from the on disk size to the smaller of the in memory
776 * file size or the new size as that's the range we really care about
777 * here and prevents waiting for other data not within the range we
780 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
781 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
782 ip->i_d.di_size, newsize);
788 * Wait for all direct I/O to complete.
790 inode_dio_wait(inode);
792 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
796 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
797 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
799 goto out_trans_cancel;
801 truncate_setsize(inode, newsize);
803 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
804 lock_flags |= XFS_ILOCK_EXCL;
806 xfs_ilock(ip, XFS_ILOCK_EXCL);
808 xfs_trans_ijoin(tp, ip, 0);
811 * Only change the c/mtime if we are changing the size or we are
812 * explicitly asked to change it. This handles the semantic difference
813 * between truncate() and ftruncate() as implemented in the VFS.
815 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
816 * special case where we need to update the times despite not having
817 * these flags set. For all other operations the VFS set these flags
818 * explicitly if it wants a timestamp update.
820 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
821 iattr->ia_ctime = iattr->ia_mtime =
822 current_fs_time(inode->i_sb);
823 mask |= ATTR_CTIME | ATTR_MTIME;
827 * The first thing we do is set the size to new_size permanently on
828 * disk. This way we don't have to worry about anyone ever being able
829 * to look at the data being freed even in the face of a crash.
830 * What we're getting around here is the case where we free a block, it
831 * is allocated to another file, it is written to, and then we crash.
832 * If the new data gets written to the file but the log buffers
833 * containing the free and reallocation don't, then we'd end up with
834 * garbage in the blocks being freed. As long as we make the new size
835 * permanent before actually freeing any blocks it doesn't matter if
836 * they get written to.
838 ip->i_d.di_size = newsize;
839 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
841 if (newsize <= oldsize) {
842 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
844 goto out_trans_abort;
847 * Truncated "down", so we're removing references to old data
848 * here - if we delay flushing for a long time, we expose
849 * ourselves unduly to the notorious NULL files problem. So,
850 * we mark this inode and flush it when the file is closed,
851 * and do not wait the usual (long) time for writeout.
853 xfs_iflags_set(ip, XFS_ITRUNCATED);
855 /* A truncate down always removes post-EOF blocks. */
856 xfs_inode_clear_eofblocks_tag(ip);
860 * Change file access modes.
862 if (mask & ATTR_MODE)
863 xfs_setattr_mode(tp, ip, iattr);
865 if (mask & ATTR_CTIME) {
866 inode->i_ctime = iattr->ia_ctime;
867 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
868 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
870 if (mask & ATTR_MTIME) {
871 inode->i_mtime = iattr->ia_mtime;
872 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
873 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
876 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
878 XFS_STATS_INC(xs_ig_attrchg);
880 if (mp->m_flags & XFS_MOUNT_WSYNC)
881 xfs_trans_set_sync(tp);
883 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
886 xfs_iunlock(ip, lock_flags);
890 commit_flags |= XFS_TRANS_ABORT;
892 xfs_trans_cancel(tp, commit_flags);
898 struct dentry *dentry,
901 struct xfs_inode *ip = XFS_I(dentry->d_inode);
904 if (iattr->ia_valid & ATTR_SIZE) {
905 xfs_ilock(ip, XFS_IOLOCK_EXCL);
906 error = xfs_setattr_size(ip, iattr);
907 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
909 error = xfs_setattr_nonsize(ip, iattr, 0);
918 struct timespec *now,
921 struct xfs_inode *ip = XFS_I(inode);
922 struct xfs_mount *mp = ip->i_mount;
923 struct xfs_trans *tp;
926 trace_xfs_update_time(ip);
928 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
929 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
931 xfs_trans_cancel(tp, 0);
935 xfs_ilock(ip, XFS_ILOCK_EXCL);
936 if (flags & S_CTIME) {
937 inode->i_ctime = *now;
938 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
939 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
941 if (flags & S_MTIME) {
942 inode->i_mtime = *now;
943 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
944 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
946 if (flags & S_ATIME) {
947 inode->i_atime = *now;
948 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
949 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
951 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
952 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
953 return -xfs_trans_commit(tp, 0);
956 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
959 * Call fiemap helper to fill in user data.
960 * Returns positive errors to xfs_getbmap.
965 struct getbmapx *bmv,
969 struct fiemap_extent_info *fieinfo = *arg;
970 u32 fiemap_flags = 0;
971 u64 logical, physical, length;
973 /* Do nothing for a hole */
974 if (bmv->bmv_block == -1LL)
977 logical = BBTOB(bmv->bmv_offset);
978 physical = BBTOB(bmv->bmv_block);
979 length = BBTOB(bmv->bmv_length);
981 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
982 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
983 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
984 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
985 FIEMAP_EXTENT_UNKNOWN);
986 physical = 0; /* no block yet */
988 if (bmv->bmv_oflags & BMV_OF_LAST)
989 fiemap_flags |= FIEMAP_EXTENT_LAST;
991 error = fiemap_fill_next_extent(fieinfo, logical, physical,
992 length, fiemap_flags);
995 *full = 1; /* user array now full */
1003 struct inode *inode,
1004 struct fiemap_extent_info *fieinfo,
1008 xfs_inode_t *ip = XFS_I(inode);
1012 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1016 /* Set up bmap header for xfs internal routine */
1017 bm.bmv_offset = BTOBB(start);
1018 /* Special case for whole file */
1019 if (length == FIEMAP_MAX_OFFSET)
1020 bm.bmv_length = -1LL;
1022 bm.bmv_length = BTOBB(length);
1024 /* We add one because in getbmap world count includes the header */
1025 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1026 fieinfo->fi_extents_max + 1;
1027 bm.bmv_count = min_t(__s32, bm.bmv_count,
1028 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1029 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1030 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1031 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1032 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1033 bm.bmv_iflags |= BMV_IF_DELALLOC;
1035 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1042 static const struct inode_operations xfs_inode_operations = {
1043 .get_acl = xfs_get_acl,
1044 .getattr = xfs_vn_getattr,
1045 .setattr = xfs_vn_setattr,
1046 .setxattr = generic_setxattr,
1047 .getxattr = generic_getxattr,
1048 .removexattr = generic_removexattr,
1049 .listxattr = xfs_vn_listxattr,
1050 .fiemap = xfs_vn_fiemap,
1051 .update_time = xfs_vn_update_time,
1054 static const struct inode_operations xfs_dir_inode_operations = {
1055 .create = xfs_vn_create,
1056 .lookup = xfs_vn_lookup,
1057 .link = xfs_vn_link,
1058 .unlink = xfs_vn_unlink,
1059 .symlink = xfs_vn_symlink,
1060 .mkdir = xfs_vn_mkdir,
1062 * Yes, XFS uses the same method for rmdir and unlink.
1064 * There are some subtile differences deeper in the code,
1065 * but we use S_ISDIR to check for those.
1067 .rmdir = xfs_vn_unlink,
1068 .mknod = xfs_vn_mknod,
1069 .rename = xfs_vn_rename,
1070 .get_acl = xfs_get_acl,
1071 .getattr = xfs_vn_getattr,
1072 .setattr = xfs_vn_setattr,
1073 .setxattr = generic_setxattr,
1074 .getxattr = generic_getxattr,
1075 .removexattr = generic_removexattr,
1076 .listxattr = xfs_vn_listxattr,
1077 .update_time = xfs_vn_update_time,
1080 static const struct inode_operations xfs_dir_ci_inode_operations = {
1081 .create = xfs_vn_create,
1082 .lookup = xfs_vn_ci_lookup,
1083 .link = xfs_vn_link,
1084 .unlink = xfs_vn_unlink,
1085 .symlink = xfs_vn_symlink,
1086 .mkdir = xfs_vn_mkdir,
1088 * Yes, XFS uses the same method for rmdir and unlink.
1090 * There are some subtile differences deeper in the code,
1091 * but we use S_ISDIR to check for those.
1093 .rmdir = xfs_vn_unlink,
1094 .mknod = xfs_vn_mknod,
1095 .rename = xfs_vn_rename,
1096 .get_acl = xfs_get_acl,
1097 .getattr = xfs_vn_getattr,
1098 .setattr = xfs_vn_setattr,
1099 .setxattr = generic_setxattr,
1100 .getxattr = generic_getxattr,
1101 .removexattr = generic_removexattr,
1102 .listxattr = xfs_vn_listxattr,
1103 .update_time = xfs_vn_update_time,
1106 static const struct inode_operations xfs_symlink_inode_operations = {
1107 .readlink = generic_readlink,
1108 .follow_link = xfs_vn_follow_link,
1109 .put_link = kfree_put_link,
1110 .get_acl = xfs_get_acl,
1111 .getattr = xfs_vn_getattr,
1112 .setattr = xfs_vn_setattr,
1113 .setxattr = generic_setxattr,
1114 .getxattr = generic_getxattr,
1115 .removexattr = generic_removexattr,
1116 .listxattr = xfs_vn_listxattr,
1117 .update_time = xfs_vn_update_time,
1121 xfs_diflags_to_iflags(
1122 struct inode *inode,
1123 struct xfs_inode *ip)
1125 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1126 inode->i_flags |= S_IMMUTABLE;
1128 inode->i_flags &= ~S_IMMUTABLE;
1129 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1130 inode->i_flags |= S_APPEND;
1132 inode->i_flags &= ~S_APPEND;
1133 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1134 inode->i_flags |= S_SYNC;
1136 inode->i_flags &= ~S_SYNC;
1137 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1138 inode->i_flags |= S_NOATIME;
1140 inode->i_flags &= ~S_NOATIME;
1144 * Initialize the Linux inode, set up the operation vectors and
1147 * When reading existing inodes from disk this is called directly
1148 * from xfs_iget, when creating a new inode it is called from
1149 * xfs_ialloc after setting up the inode.
1151 * We are always called with an uninitialised linux inode here.
1152 * We need to initialise the necessary fields and take a reference
1157 struct xfs_inode *ip)
1159 struct inode *inode = &ip->i_vnode;
1162 inode->i_ino = ip->i_ino;
1163 inode->i_state = I_NEW;
1165 inode_sb_list_add(inode);
1166 /* make the inode look hashed for the writeback code */
1167 hlist_add_fake(&inode->i_hash);
1169 inode->i_mode = ip->i_d.di_mode;
1170 set_nlink(inode, ip->i_d.di_nlink);
1171 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1172 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1174 switch (inode->i_mode & S_IFMT) {
1178 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1179 sysv_minor(ip->i_df.if_u2.if_rdev));
1186 inode->i_generation = ip->i_d.di_gen;
1187 i_size_write(inode, ip->i_d.di_size);
1188 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1189 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1190 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1191 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1192 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1193 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1194 xfs_diflags_to_iflags(inode, ip);
1196 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1197 switch (inode->i_mode & S_IFMT) {
1199 inode->i_op = &xfs_inode_operations;
1200 inode->i_fop = &xfs_file_operations;
1201 inode->i_mapping->a_ops = &xfs_address_space_operations;
1204 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1205 inode->i_op = &xfs_dir_ci_inode_operations;
1207 inode->i_op = &xfs_dir_inode_operations;
1208 inode->i_fop = &xfs_dir_file_operations;
1209 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1212 inode->i_op = &xfs_symlink_inode_operations;
1213 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1214 inode->i_mapping->a_ops = &xfs_address_space_operations;
1217 inode->i_op = &xfs_inode_operations;
1218 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1223 * Ensure all page cache allocations are done from GFP_NOFS context to
1224 * prevent direct reclaim recursion back into the filesystem and blowing
1225 * stacks or deadlocking.
1227 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1228 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1231 * If there is no attribute fork no ACL can exist on this inode,
1232 * and it can't have any file capabilities attached to it either.
1234 if (!XFS_IFORK_Q(ip)) {
1235 inode_has_no_xattr(inode);
1236 cache_no_acl(inode);
1239 xfs_iflags_clear(ip, XFS_INEW);
1242 unlock_new_inode(inode);