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"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_inode.h"
29 #include "xfs_bmap_util.h"
31 #include "xfs_quota.h"
32 #include "xfs_error.h"
34 #include "xfs_trans.h"
35 #include "xfs_trace.h"
36 #include "xfs_icache.h"
37 #include "xfs_symlink.h"
38 #include "xfs_da_btree.h"
39 #include "xfs_dir2_priv.h"
40 #include "xfs_trans_space.h"
42 #include <linux/capability.h>
43 #include <linux/xattr.h>
44 #include <linux/namei.h>
45 #include <linux/posix_acl.h>
46 #include <linux/security.h>
47 #include <linux/fiemap.h>
48 #include <linux/slab.h>
51 * Directories have different lock order w.r.t. mmap_sem compared to regular
52 * files. This is due to readdir potentially triggering page faults on a user
53 * buffer inside filldir(), and this happens with the ilock on the directory
54 * held. For regular files, the lock order is the other way around - the
55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
56 * block mapping. Hence we need a different class for the directory ilock so
57 * that lockdep can tell them apart.
59 static struct lock_class_key xfs_nondir_ilock_class;
60 static struct lock_class_key xfs_dir_ilock_class;
65 const struct xattr *xattr_array,
68 const struct xattr *xattr;
69 struct xfs_inode *ip = XFS_I(inode);
72 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
73 error = xfs_attr_set(ip, xattr->name, xattr->value,
74 xattr->value_len, ATTR_SECURE);
82 * Hook in SELinux. This is not quite correct yet, what we really need
83 * here (as we do for default ACLs) is a mechanism by which creation of
84 * these attrs can be journalled at inode creation time (along with the
85 * inode, of course, such that log replay can't cause these to be lost).
92 const struct qstr *qstr)
94 return security_inode_init_security(inode, dir, qstr,
95 &xfs_initxattrs, NULL);
100 struct xfs_name *namep,
101 struct dentry *dentry,
104 namep->name = dentry->d_name.name;
105 namep->len = dentry->d_name.len;
106 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
113 struct dentry *dentry)
115 struct xfs_name teardown;
118 * If we can't add the ACL or we fail in
119 * xfs_init_security we must back out.
120 * ENOSPC can hit here, among other things.
122 xfs_dentry_to_name(&teardown, dentry, 0);
124 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
130 struct dentry *dentry,
133 bool tmpfile) /* unnamed file */
136 struct xfs_inode *ip = NULL;
137 struct posix_acl *default_acl, *acl;
138 struct xfs_name name;
142 * Irix uses Missed'em'V split, but doesn't want to see
143 * the upper 5 bits of (14bit) major.
145 if (S_ISCHR(mode) || S_ISBLK(mode)) {
146 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
148 rdev = sysv_encode_dev(rdev);
153 error = posix_acl_create(dir, &mode, &default_acl, &acl);
158 xfs_dentry_to_name(&name, dentry, mode);
159 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
161 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
168 error = xfs_init_security(inode, dir, &dentry->d_name);
170 goto out_cleanup_inode;
172 #ifdef CONFIG_XFS_POSIX_ACL
174 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
176 goto out_cleanup_inode;
179 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
181 goto out_cleanup_inode;
186 d_tmpfile(dentry, inode);
188 d_instantiate(dentry, inode);
192 posix_acl_release(default_acl);
194 posix_acl_release(acl);
199 xfs_cleanup_inode(dir, inode, dentry);
207 struct dentry *dentry,
211 return xfs_generic_create(dir, dentry, mode, rdev, false);
217 struct dentry *dentry,
221 return xfs_vn_mknod(dir, dentry, mode, 0);
227 struct dentry *dentry,
230 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
233 STATIC struct dentry *
236 struct dentry *dentry,
239 struct xfs_inode *cip;
240 struct xfs_name name;
243 if (dentry->d_name.len >= MAXNAMELEN)
244 return ERR_PTR(-ENAMETOOLONG);
246 xfs_dentry_to_name(&name, dentry, 0);
247 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
248 if (unlikely(error)) {
249 if (unlikely(error != -ENOENT))
250 return ERR_PTR(error);
255 return d_splice_alias(VFS_I(cip), dentry);
258 STATIC struct dentry *
261 struct dentry *dentry,
264 struct xfs_inode *ip;
265 struct xfs_name xname;
266 struct xfs_name ci_name;
270 if (dentry->d_name.len >= MAXNAMELEN)
271 return ERR_PTR(-ENAMETOOLONG);
273 xfs_dentry_to_name(&xname, dentry, 0);
274 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
275 if (unlikely(error)) {
276 if (unlikely(error != -ENOENT))
277 return ERR_PTR(error);
279 * call d_add(dentry, NULL) here when d_drop_negative_children
280 * is called in xfs_vn_mknod (ie. allow negative dentries
281 * with CI filesystems).
286 /* if exact match, just splice and exit */
288 return d_splice_alias(VFS_I(ip), dentry);
290 /* else case-insensitive match... */
291 dname.name = ci_name.name;
292 dname.len = ci_name.len;
293 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
294 kmem_free(ci_name.name);
300 struct dentry *old_dentry,
302 struct dentry *dentry)
304 struct inode *inode = old_dentry->d_inode;
305 struct xfs_name name;
308 xfs_dentry_to_name(&name, dentry, inode->i_mode);
310 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
315 d_instantiate(dentry, inode);
322 struct dentry *dentry)
324 struct xfs_name name;
327 xfs_dentry_to_name(&name, dentry, 0);
329 error = xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
334 * With unlink, the VFS makes the dentry "negative": no inode,
335 * but still hashed. This is incompatible with case-insensitive
336 * mode, so invalidate (unhash) the dentry in CI-mode.
338 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
339 d_invalidate(dentry);
346 struct dentry *dentry,
350 struct xfs_inode *cip = NULL;
351 struct xfs_name name;
356 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
357 xfs_dentry_to_name(&name, dentry, mode);
359 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
365 error = xfs_init_security(inode, dir, &dentry->d_name);
367 goto out_cleanup_inode;
369 d_instantiate(dentry, inode);
373 xfs_cleanup_inode(dir, inode, dentry);
382 struct dentry *odentry,
384 struct dentry *ndentry)
386 struct inode *new_inode = ndentry->d_inode;
387 struct xfs_name oname;
388 struct xfs_name nname;
390 xfs_dentry_to_name(&oname, odentry, 0);
391 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
393 return xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
394 XFS_I(ndir), &nname, new_inode ?
395 XFS_I(new_inode) : NULL);
399 * careful here - this function can get called recursively, so
400 * we need to be very careful about how much stack we use.
401 * uio is kmalloced for this reason...
405 struct dentry *dentry,
406 struct nameidata *nd)
411 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
415 error = xfs_readlink(XFS_I(dentry->d_inode), link);
419 nd_set_link(nd, link);
425 nd_set_link(nd, ERR_PTR(error));
431 struct vfsmount *mnt,
432 struct dentry *dentry,
435 struct inode *inode = dentry->d_inode;
436 struct xfs_inode *ip = XFS_I(inode);
437 struct xfs_mount *mp = ip->i_mount;
439 trace_xfs_getattr(ip);
441 if (XFS_FORCED_SHUTDOWN(mp))
444 stat->size = XFS_ISIZE(ip);
445 stat->dev = inode->i_sb->s_dev;
446 stat->mode = ip->i_d.di_mode;
447 stat->nlink = ip->i_d.di_nlink;
448 stat->uid = inode->i_uid;
449 stat->gid = inode->i_gid;
450 stat->ino = ip->i_ino;
451 stat->atime = inode->i_atime;
452 stat->mtime = inode->i_mtime;
453 stat->ctime = inode->i_ctime;
455 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
458 switch (inode->i_mode & S_IFMT) {
461 stat->blksize = BLKDEV_IOSIZE;
462 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
463 sysv_minor(ip->i_df.if_u2.if_rdev));
466 if (XFS_IS_REALTIME_INODE(ip)) {
468 * If the file blocks are being allocated from a
469 * realtime volume, then return the inode's realtime
470 * extent size or the realtime volume's extent size.
473 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
475 stat->blksize = xfs_preferred_iosize(mp);
485 struct xfs_inode *ip,
488 struct inode *inode = VFS_I(ip);
489 umode_t mode = iattr->ia_mode;
491 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
493 ip->i_d.di_mode &= S_IFMT;
494 ip->i_d.di_mode |= mode & ~S_IFMT;
496 inode->i_mode &= S_IFMT;
497 inode->i_mode |= mode & ~S_IFMT;
502 struct xfs_inode *ip,
505 struct inode *inode = VFS_I(ip);
507 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
509 if (iattr->ia_valid & ATTR_ATIME) {
510 inode->i_atime = iattr->ia_atime;
511 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
512 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
514 if (iattr->ia_valid & ATTR_CTIME) {
515 inode->i_ctime = iattr->ia_ctime;
516 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
517 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
519 if (iattr->ia_valid & ATTR_MTIME) {
520 inode->i_mtime = iattr->ia_mtime;
521 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
522 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
528 struct xfs_inode *ip,
532 xfs_mount_t *mp = ip->i_mount;
533 struct inode *inode = VFS_I(ip);
534 int mask = iattr->ia_valid;
537 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
538 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
539 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
540 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
542 trace_xfs_setattr(ip);
544 /* If acls are being inherited, we already have this checked */
545 if (!(flags & XFS_ATTR_NOACL)) {
546 if (mp->m_flags & XFS_MOUNT_RDONLY)
549 if (XFS_FORCED_SHUTDOWN(mp))
552 error = inode_change_ok(inode, iattr);
557 ASSERT((mask & ATTR_SIZE) == 0);
560 * If disk quotas is on, we make sure that the dquots do exist on disk,
561 * before we start any other transactions. Trying to do this later
562 * is messy. We don't care to take a readlock to look at the ids
563 * in inode here, because we can't hold it across the trans_reserve.
564 * If the IDs do change before we take the ilock, we're covered
565 * because the i_*dquot fields will get updated anyway.
567 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
570 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
572 qflags |= XFS_QMOPT_UQUOTA;
576 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
578 qflags |= XFS_QMOPT_GQUOTA;
584 * We take a reference when we initialize udqp and gdqp,
585 * so it is important that we never blindly double trip on
586 * the same variable. See xfs_create() for an example.
588 ASSERT(udqp == NULL);
589 ASSERT(gdqp == NULL);
590 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
591 xfs_kgid_to_gid(gid),
593 qflags, &udqp, &gdqp, NULL);
598 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
599 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
603 xfs_ilock(ip, XFS_ILOCK_EXCL);
606 * Change file ownership. Must be the owner or privileged.
608 if (mask & (ATTR_UID|ATTR_GID)) {
610 * These IDs could have changed since we last looked at them.
611 * But, we're assured that if the ownership did change
612 * while we didn't have the inode locked, inode's dquot(s)
613 * would have changed also.
617 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
618 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
621 * Do a quota reservation only if uid/gid is actually
624 if (XFS_IS_QUOTA_RUNNING(mp) &&
625 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
626 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
628 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
629 NULL, capable(CAP_FOWNER) ?
630 XFS_QMOPT_FORCE_RES : 0);
631 if (error) /* out of quota */
632 goto out_trans_cancel;
636 xfs_trans_ijoin(tp, ip, 0);
639 * Change file ownership. Must be the owner or privileged.
641 if (mask & (ATTR_UID|ATTR_GID)) {
643 * CAP_FSETID overrides the following restrictions:
645 * The set-user-ID and set-group-ID bits of a file will be
646 * cleared upon successful return from chown()
648 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
649 !capable(CAP_FSETID))
650 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
653 * Change the ownerships and register quota modifications
654 * in the transaction.
656 if (!uid_eq(iuid, uid)) {
657 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
658 ASSERT(mask & ATTR_UID);
660 olddquot1 = xfs_qm_vop_chown(tp, ip,
661 &ip->i_udquot, udqp);
663 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
666 if (!gid_eq(igid, gid)) {
667 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
668 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
669 !XFS_IS_PQUOTA_ON(mp));
670 ASSERT(mask & ATTR_GID);
672 olddquot2 = xfs_qm_vop_chown(tp, ip,
673 &ip->i_gdquot, gdqp);
675 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
680 if (mask & ATTR_MODE)
681 xfs_setattr_mode(ip, iattr);
682 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
683 xfs_setattr_time(ip, iattr);
685 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
687 XFS_STATS_INC(xs_ig_attrchg);
689 if (mp->m_flags & XFS_MOUNT_WSYNC)
690 xfs_trans_set_sync(tp);
691 error = xfs_trans_commit(tp, 0);
693 xfs_iunlock(ip, XFS_ILOCK_EXCL);
696 * Release any dquot(s) the inode had kept before chown.
698 xfs_qm_dqrele(olddquot1);
699 xfs_qm_dqrele(olddquot2);
707 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
708 * update. We could avoid this with linked transactions
709 * and passing down the transaction pointer all the way
710 * to attr_set. No previous user of the generic
711 * Posix ACL code seems to care about this issue either.
713 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
714 error = posix_acl_chmod(inode, inode->i_mode);
722 xfs_trans_cancel(tp, 0);
723 xfs_iunlock(ip, XFS_ILOCK_EXCL);
731 * Truncate file. Must have write permission and not be a directory.
735 struct xfs_inode *ip,
738 struct xfs_mount *mp = ip->i_mount;
739 struct inode *inode = VFS_I(ip);
740 xfs_off_t oldsize, newsize;
741 struct xfs_trans *tp;
744 uint commit_flags = 0;
746 trace_xfs_setattr(ip);
748 if (mp->m_flags & XFS_MOUNT_RDONLY)
751 if (XFS_FORCED_SHUTDOWN(mp))
754 error = inode_change_ok(inode, iattr);
758 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
759 ASSERT(S_ISREG(ip->i_d.di_mode));
760 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
761 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
763 oldsize = inode->i_size;
764 newsize = iattr->ia_size;
767 * Short circuit the truncate case for zero length files.
769 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
770 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
774 * Use the regular setattr path to update the timestamps.
776 iattr->ia_valid &= ~ATTR_SIZE;
777 return xfs_setattr_nonsize(ip, iattr, 0);
781 * Make sure that the dquots are attached to the inode.
783 error = xfs_qm_dqattach(ip, 0);
788 * Now we can make the changes. Before we join the inode to the
789 * transaction, take care of the part of the truncation that must be
790 * done without the inode lock. This needs to be done before joining
791 * the inode to the transaction, because the inode cannot be unlocked
792 * once it is a part of the transaction.
794 if (newsize > oldsize) {
796 * Do the first part of growing a file: zero any data in the
797 * last block that is beyond the old EOF. We need to do this
798 * before the inode is joined to the transaction to modify
801 error = xfs_zero_eof(ip, newsize, oldsize);
807 * We are going to log the inode size change in this transaction so
808 * any previous writes that are beyond the on disk EOF and the new
809 * EOF that have not been written out need to be written here. If we
810 * do not write the data out, we expose ourselves to the null files
813 * Only flush from the on disk size to the smaller of the in memory
814 * file size or the new size as that's the range we really care about
815 * here and prevents waiting for other data not within the range we
818 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
819 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
820 ip->i_d.di_size, newsize);
826 * Wait for all direct I/O to complete.
828 inode_dio_wait(inode);
831 * Do all the page cache truncate work outside the transaction context
832 * as the "lock" order is page lock->log space reservation. i.e.
833 * locking pages inside the transaction can ABBA deadlock with
834 * writeback. We have to do the VFS inode size update before we truncate
835 * the pagecache, however, to avoid racing with page faults beyond the
836 * new EOF they are not serialised against truncate operations except by
837 * page locks and size updates.
839 * Hence we are in a situation where a truncate can fail with ENOMEM
840 * from xfs_trans_reserve(), but having already truncated the in-memory
841 * version of the file (i.e. made user visible changes). There's not
842 * much we can do about this, except to hope that the caller sees ENOMEM
843 * and retries the truncate operation.
845 error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
848 truncate_setsize(inode, newsize);
851 * The "we can't serialise against page faults" pain gets worse.
853 * If the file is mapped then we have to clean the page at the old EOF
854 * when extending the file. Extending the file can expose changes the
855 * underlying page mapping (e.g. from beyond EOF to a hole or
856 * unwritten), and so on the next attempt to write to that page we need
857 * to remap it for write. i.e. we need .page_mkwrite() to be called.
858 * Hence we need to clean the page to clean the pte and so a new write
859 * fault will be triggered appropriately.
861 * If we do it before we change the inode size, then we can race with a
862 * page fault that maps the page with exactly the same problem. If we do
863 * it after we change the file size, then a new page fault can come in
864 * and allocate space before we've run the rest of the truncate
865 * transaction. That's kinda grotesque, but it's better than have data
866 * over a hole, and so that's the lesser evil that has been chosen here.
868 * The real solution, however, is to have some mechanism for locking out
869 * page faults while a truncate is in progress.
871 if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
872 error = filemap_write_and_wait_range(
873 VFS_I(ip)->i_mapping,
874 round_down(oldsize, PAGE_CACHE_SIZE),
875 round_up(oldsize, PAGE_CACHE_SIZE) - 1);
880 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
881 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
883 goto out_trans_cancel;
885 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
886 lock_flags |= XFS_ILOCK_EXCL;
887 xfs_ilock(ip, XFS_ILOCK_EXCL);
888 xfs_trans_ijoin(tp, ip, 0);
891 * Only change the c/mtime if we are changing the size or we are
892 * explicitly asked to change it. This handles the semantic difference
893 * between truncate() and ftruncate() as implemented in the VFS.
895 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
896 * special case where we need to update the times despite not having
897 * these flags set. For all other operations the VFS set these flags
898 * explicitly if it wants a timestamp update.
900 if (newsize != oldsize &&
901 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
902 iattr->ia_ctime = iattr->ia_mtime =
903 current_fs_time(inode->i_sb);
904 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
908 * The first thing we do is set the size to new_size permanently on
909 * disk. This way we don't have to worry about anyone ever being able
910 * to look at the data being freed even in the face of a crash.
911 * What we're getting around here is the case where we free a block, it
912 * is allocated to another file, it is written to, and then we crash.
913 * If the new data gets written to the file but the log buffers
914 * containing the free and reallocation don't, then we'd end up with
915 * garbage in the blocks being freed. As long as we make the new size
916 * permanent before actually freeing any blocks it doesn't matter if
917 * they get written to.
919 ip->i_d.di_size = newsize;
920 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
922 if (newsize <= oldsize) {
923 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
925 goto out_trans_abort;
928 * Truncated "down", so we're removing references to old data
929 * here - if we delay flushing for a long time, we expose
930 * ourselves unduly to the notorious NULL files problem. So,
931 * we mark this inode and flush it when the file is closed,
932 * and do not wait the usual (long) time for writeout.
934 xfs_iflags_set(ip, XFS_ITRUNCATED);
936 /* A truncate down always removes post-EOF blocks. */
937 xfs_inode_clear_eofblocks_tag(ip);
940 if (iattr->ia_valid & ATTR_MODE)
941 xfs_setattr_mode(ip, iattr);
942 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
943 xfs_setattr_time(ip, iattr);
945 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
947 XFS_STATS_INC(xs_ig_attrchg);
949 if (mp->m_flags & XFS_MOUNT_WSYNC)
950 xfs_trans_set_sync(tp);
952 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
955 xfs_iunlock(ip, lock_flags);
959 commit_flags |= XFS_TRANS_ABORT;
961 xfs_trans_cancel(tp, commit_flags);
967 struct dentry *dentry,
970 struct xfs_inode *ip = XFS_I(dentry->d_inode);
973 if (iattr->ia_valid & ATTR_SIZE) {
974 xfs_ilock(ip, XFS_IOLOCK_EXCL);
975 error = xfs_setattr_size(ip, iattr);
976 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
978 error = xfs_setattr_nonsize(ip, iattr, 0);
987 struct timespec *now,
990 struct xfs_inode *ip = XFS_I(inode);
991 struct xfs_mount *mp = ip->i_mount;
992 struct xfs_trans *tp;
995 trace_xfs_update_time(ip);
997 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
998 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1000 xfs_trans_cancel(tp, 0);
1004 xfs_ilock(ip, XFS_ILOCK_EXCL);
1005 if (flags & S_CTIME) {
1006 inode->i_ctime = *now;
1007 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
1008 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
1010 if (flags & S_MTIME) {
1011 inode->i_mtime = *now;
1012 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
1013 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
1015 if (flags & S_ATIME) {
1016 inode->i_atime = *now;
1017 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
1018 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
1020 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1021 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1022 return xfs_trans_commit(tp, 0);
1025 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1028 * Call fiemap helper to fill in user data.
1029 * Returns positive errors to xfs_getbmap.
1034 struct getbmapx *bmv,
1038 struct fiemap_extent_info *fieinfo = *arg;
1039 u32 fiemap_flags = 0;
1040 u64 logical, physical, length;
1042 /* Do nothing for a hole */
1043 if (bmv->bmv_block == -1LL)
1046 logical = BBTOB(bmv->bmv_offset);
1047 physical = BBTOB(bmv->bmv_block);
1048 length = BBTOB(bmv->bmv_length);
1050 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1051 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1052 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1053 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1054 FIEMAP_EXTENT_UNKNOWN);
1055 physical = 0; /* no block yet */
1057 if (bmv->bmv_oflags & BMV_OF_LAST)
1058 fiemap_flags |= FIEMAP_EXTENT_LAST;
1060 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1061 length, fiemap_flags);
1064 *full = 1; /* user array now full */
1072 struct inode *inode,
1073 struct fiemap_extent_info *fieinfo,
1077 xfs_inode_t *ip = XFS_I(inode);
1081 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1085 /* Set up bmap header for xfs internal routine */
1086 bm.bmv_offset = BTOBBT(start);
1087 /* Special case for whole file */
1088 if (length == FIEMAP_MAX_OFFSET)
1089 bm.bmv_length = -1LL;
1091 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1093 /* We add one because in getbmap world count includes the header */
1094 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1095 fieinfo->fi_extents_max + 1;
1096 bm.bmv_count = min_t(__s32, bm.bmv_count,
1097 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1098 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1099 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1100 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1101 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1102 bm.bmv_iflags |= BMV_IF_DELALLOC;
1104 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1114 struct dentry *dentry,
1117 return xfs_generic_create(dir, dentry, mode, 0, true);
1120 static const struct inode_operations xfs_inode_operations = {
1121 .get_acl = xfs_get_acl,
1122 .set_acl = xfs_set_acl,
1123 .getattr = xfs_vn_getattr,
1124 .setattr = xfs_vn_setattr,
1125 .setxattr = generic_setxattr,
1126 .getxattr = generic_getxattr,
1127 .removexattr = generic_removexattr,
1128 .listxattr = xfs_vn_listxattr,
1129 .fiemap = xfs_vn_fiemap,
1130 .update_time = xfs_vn_update_time,
1133 static const struct inode_operations xfs_dir_inode_operations = {
1134 .create = xfs_vn_create,
1135 .lookup = xfs_vn_lookup,
1136 .link = xfs_vn_link,
1137 .unlink = xfs_vn_unlink,
1138 .symlink = xfs_vn_symlink,
1139 .mkdir = xfs_vn_mkdir,
1141 * Yes, XFS uses the same method for rmdir and unlink.
1143 * There are some subtile differences deeper in the code,
1144 * but we use S_ISDIR to check for those.
1146 .rmdir = xfs_vn_unlink,
1147 .mknod = xfs_vn_mknod,
1148 .rename = xfs_vn_rename,
1149 .get_acl = xfs_get_acl,
1150 .set_acl = xfs_set_acl,
1151 .getattr = xfs_vn_getattr,
1152 .setattr = xfs_vn_setattr,
1153 .setxattr = generic_setxattr,
1154 .getxattr = generic_getxattr,
1155 .removexattr = generic_removexattr,
1156 .listxattr = xfs_vn_listxattr,
1157 .update_time = xfs_vn_update_time,
1158 .tmpfile = xfs_vn_tmpfile,
1161 static const struct inode_operations xfs_dir_ci_inode_operations = {
1162 .create = xfs_vn_create,
1163 .lookup = xfs_vn_ci_lookup,
1164 .link = xfs_vn_link,
1165 .unlink = xfs_vn_unlink,
1166 .symlink = xfs_vn_symlink,
1167 .mkdir = xfs_vn_mkdir,
1169 * Yes, XFS uses the same method for rmdir and unlink.
1171 * There are some subtile differences deeper in the code,
1172 * but we use S_ISDIR to check for those.
1174 .rmdir = xfs_vn_unlink,
1175 .mknod = xfs_vn_mknod,
1176 .rename = xfs_vn_rename,
1177 .get_acl = xfs_get_acl,
1178 .set_acl = xfs_set_acl,
1179 .getattr = xfs_vn_getattr,
1180 .setattr = xfs_vn_setattr,
1181 .setxattr = generic_setxattr,
1182 .getxattr = generic_getxattr,
1183 .removexattr = generic_removexattr,
1184 .listxattr = xfs_vn_listxattr,
1185 .update_time = xfs_vn_update_time,
1186 .tmpfile = xfs_vn_tmpfile,
1189 static const struct inode_operations xfs_symlink_inode_operations = {
1190 .readlink = generic_readlink,
1191 .follow_link = xfs_vn_follow_link,
1192 .put_link = kfree_put_link,
1193 .getattr = xfs_vn_getattr,
1194 .setattr = xfs_vn_setattr,
1195 .setxattr = generic_setxattr,
1196 .getxattr = generic_getxattr,
1197 .removexattr = generic_removexattr,
1198 .listxattr = xfs_vn_listxattr,
1199 .update_time = xfs_vn_update_time,
1203 xfs_diflags_to_iflags(
1204 struct inode *inode,
1205 struct xfs_inode *ip)
1207 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1208 inode->i_flags |= S_IMMUTABLE;
1210 inode->i_flags &= ~S_IMMUTABLE;
1211 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1212 inode->i_flags |= S_APPEND;
1214 inode->i_flags &= ~S_APPEND;
1215 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1216 inode->i_flags |= S_SYNC;
1218 inode->i_flags &= ~S_SYNC;
1219 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1220 inode->i_flags |= S_NOATIME;
1222 inode->i_flags &= ~S_NOATIME;
1226 * Initialize the Linux inode, set up the operation vectors and
1229 * When reading existing inodes from disk this is called directly
1230 * from xfs_iget, when creating a new inode it is called from
1231 * xfs_ialloc after setting up the inode.
1233 * We are always called with an uninitialised linux inode here.
1234 * We need to initialise the necessary fields and take a reference
1239 struct xfs_inode *ip)
1241 struct inode *inode = &ip->i_vnode;
1244 inode->i_ino = ip->i_ino;
1245 inode->i_state = I_NEW;
1247 inode_sb_list_add(inode);
1248 /* make the inode look hashed for the writeback code */
1249 hlist_add_fake(&inode->i_hash);
1251 inode->i_mode = ip->i_d.di_mode;
1252 set_nlink(inode, ip->i_d.di_nlink);
1253 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1254 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1256 switch (inode->i_mode & S_IFMT) {
1260 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1261 sysv_minor(ip->i_df.if_u2.if_rdev));
1268 inode->i_generation = ip->i_d.di_gen;
1269 i_size_write(inode, ip->i_d.di_size);
1270 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1271 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1272 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1273 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1274 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1275 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1276 xfs_diflags_to_iflags(inode, ip);
1278 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1279 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1280 switch (inode->i_mode & S_IFMT) {
1282 inode->i_op = &xfs_inode_operations;
1283 inode->i_fop = &xfs_file_operations;
1284 inode->i_mapping->a_ops = &xfs_address_space_operations;
1287 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1288 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1289 inode->i_op = &xfs_dir_ci_inode_operations;
1291 inode->i_op = &xfs_dir_inode_operations;
1292 inode->i_fop = &xfs_dir_file_operations;
1293 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1296 inode->i_op = &xfs_symlink_inode_operations;
1297 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1298 inode->i_mapping->a_ops = &xfs_address_space_operations;
1301 inode->i_op = &xfs_inode_operations;
1302 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1307 * Ensure all page cache allocations are done from GFP_NOFS context to
1308 * prevent direct reclaim recursion back into the filesystem and blowing
1309 * stacks or deadlocking.
1311 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1312 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1315 * If there is no attribute fork no ACL can exist on this inode,
1316 * and it can't have any file capabilities attached to it either.
1318 if (!XFS_IFORK_Q(ip)) {
1319 inode_has_no_xattr(inode);
1320 cache_no_acl(inode);
1323 xfs_iflags_clear(ip, XFS_INEW);
1326 unlock_new_inode(inode);