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_trans_space.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));
131 struct dentry *dentry,
134 bool tmpfile) /* unnamed file */
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);
159 xfs_dentry_to_name(&name, dentry, mode);
160 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
162 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
169 error = xfs_init_security(inode, dir, &dentry->d_name);
171 goto out_cleanup_inode;
173 #ifdef CONFIG_XFS_POSIX_ACL
175 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
177 goto out_cleanup_inode;
180 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
182 goto out_cleanup_inode;
187 d_tmpfile(dentry, inode);
189 d_instantiate(dentry, inode);
193 posix_acl_release(default_acl);
195 posix_acl_release(acl);
200 xfs_cleanup_inode(dir, inode, dentry);
208 struct dentry *dentry,
212 return xfs_generic_create(dir, dentry, mode, rdev, false);
218 struct dentry *dentry,
222 return xfs_vn_mknod(dir, dentry, mode, 0);
228 struct dentry *dentry,
231 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
234 STATIC struct dentry *
237 struct dentry *dentry,
240 struct xfs_inode *cip;
241 struct xfs_name name;
244 if (dentry->d_name.len >= MAXNAMELEN)
245 return ERR_PTR(-ENAMETOOLONG);
247 xfs_dentry_to_name(&name, dentry, 0);
248 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
249 if (unlikely(error)) {
250 if (unlikely(error != -ENOENT))
251 return ERR_PTR(error);
256 return d_splice_alias(VFS_I(cip), dentry);
259 STATIC struct dentry *
262 struct dentry *dentry,
265 struct xfs_inode *ip;
266 struct xfs_name xname;
267 struct xfs_name ci_name;
271 if (dentry->d_name.len >= MAXNAMELEN)
272 return ERR_PTR(-ENAMETOOLONG);
274 xfs_dentry_to_name(&xname, dentry, 0);
275 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
276 if (unlikely(error)) {
277 if (unlikely(error != -ENOENT))
278 return ERR_PTR(error);
280 * call d_add(dentry, NULL) here when d_drop_negative_children
281 * is called in xfs_vn_mknod (ie. allow negative dentries
282 * with CI filesystems).
287 /* if exact match, just splice and exit */
289 return d_splice_alias(VFS_I(ip), dentry);
291 /* else case-insensitive match... */
292 dname.name = ci_name.name;
293 dname.len = ci_name.len;
294 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
295 kmem_free(ci_name.name);
301 struct dentry *old_dentry,
303 struct dentry *dentry)
305 struct inode *inode = old_dentry->d_inode;
306 struct xfs_name name;
309 xfs_dentry_to_name(&name, dentry, inode->i_mode);
311 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
316 d_instantiate(dentry, inode);
323 struct dentry *dentry)
325 struct xfs_name name;
328 xfs_dentry_to_name(&name, dentry, 0);
330 error = xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
335 * With unlink, the VFS makes the dentry "negative": no inode,
336 * but still hashed. This is incompatible with case-insensitive
337 * mode, so invalidate (unhash) the dentry in CI-mode.
339 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
340 d_invalidate(dentry);
347 struct dentry *dentry,
351 struct xfs_inode *cip = NULL;
352 struct xfs_name name;
357 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
358 xfs_dentry_to_name(&name, dentry, mode);
360 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
366 error = xfs_init_security(inode, dir, &dentry->d_name);
368 goto out_cleanup_inode;
370 d_instantiate(dentry, inode);
374 xfs_cleanup_inode(dir, inode, dentry);
383 struct dentry *odentry,
385 struct dentry *ndentry)
387 struct inode *new_inode = ndentry->d_inode;
388 struct xfs_name oname;
389 struct xfs_name nname;
391 xfs_dentry_to_name(&oname, odentry, 0);
392 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
394 return xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
395 XFS_I(ndir), &nname, new_inode ?
396 XFS_I(new_inode) : NULL);
400 * careful here - this function can get called recursively, so
401 * we need to be very careful about how much stack we use.
402 * uio is kmalloced for this reason...
406 struct dentry *dentry,
407 struct nameidata *nd)
412 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
416 error = xfs_readlink(XFS_I(dentry->d_inode), link);
420 nd_set_link(nd, link);
426 nd_set_link(nd, ERR_PTR(error));
432 struct vfsmount *mnt,
433 struct dentry *dentry,
436 struct inode *inode = dentry->d_inode;
437 struct xfs_inode *ip = XFS_I(inode);
438 struct xfs_mount *mp = ip->i_mount;
440 trace_xfs_getattr(ip);
442 if (XFS_FORCED_SHUTDOWN(mp))
445 stat->size = XFS_ISIZE(ip);
446 stat->dev = inode->i_sb->s_dev;
447 stat->mode = ip->i_d.di_mode;
448 stat->nlink = ip->i_d.di_nlink;
449 stat->uid = inode->i_uid;
450 stat->gid = inode->i_gid;
451 stat->ino = ip->i_ino;
452 stat->atime = inode->i_atime;
453 stat->mtime = inode->i_mtime;
454 stat->ctime = inode->i_ctime;
456 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
459 switch (inode->i_mode & S_IFMT) {
462 stat->blksize = BLKDEV_IOSIZE;
463 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
464 sysv_minor(ip->i_df.if_u2.if_rdev));
467 if (XFS_IS_REALTIME_INODE(ip)) {
469 * If the file blocks are being allocated from a
470 * realtime volume, then return the inode's realtime
471 * extent size or the realtime volume's extent size.
474 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
476 stat->blksize = xfs_preferred_iosize(mp);
486 struct xfs_inode *ip,
489 struct inode *inode = VFS_I(ip);
490 umode_t mode = iattr->ia_mode;
492 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
494 ip->i_d.di_mode &= S_IFMT;
495 ip->i_d.di_mode |= mode & ~S_IFMT;
497 inode->i_mode &= S_IFMT;
498 inode->i_mode |= mode & ~S_IFMT;
503 struct xfs_inode *ip,
506 struct inode *inode = VFS_I(ip);
508 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
510 if (iattr->ia_valid & ATTR_ATIME) {
511 inode->i_atime = iattr->ia_atime;
512 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
513 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
515 if (iattr->ia_valid & ATTR_CTIME) {
516 inode->i_ctime = iattr->ia_ctime;
517 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
518 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
520 if (iattr->ia_valid & ATTR_MTIME) {
521 inode->i_mtime = iattr->ia_mtime;
522 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
523 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
529 struct xfs_inode *ip,
533 xfs_mount_t *mp = ip->i_mount;
534 struct inode *inode = VFS_I(ip);
535 int mask = iattr->ia_valid;
538 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
539 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
540 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
541 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
543 trace_xfs_setattr(ip);
545 /* If acls are being inherited, we already have this checked */
546 if (!(flags & XFS_ATTR_NOACL)) {
547 if (mp->m_flags & XFS_MOUNT_RDONLY)
550 if (XFS_FORCED_SHUTDOWN(mp))
553 error = inode_change_ok(inode, iattr);
558 ASSERT((mask & ATTR_SIZE) == 0);
561 * If disk quotas is on, we make sure that the dquots do exist on disk,
562 * before we start any other transactions. Trying to do this later
563 * is messy. We don't care to take a readlock to look at the ids
564 * in inode here, because we can't hold it across the trans_reserve.
565 * If the IDs do change before we take the ilock, we're covered
566 * because the i_*dquot fields will get updated anyway.
568 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
571 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
573 qflags |= XFS_QMOPT_UQUOTA;
577 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
579 qflags |= XFS_QMOPT_GQUOTA;
585 * We take a reference when we initialize udqp and gdqp,
586 * so it is important that we never blindly double trip on
587 * the same variable. See xfs_create() for an example.
589 ASSERT(udqp == NULL);
590 ASSERT(gdqp == NULL);
591 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
592 xfs_kgid_to_gid(gid),
594 qflags, &udqp, &gdqp, NULL);
599 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
600 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
604 xfs_ilock(ip, XFS_ILOCK_EXCL);
607 * Change file ownership. Must be the owner or privileged.
609 if (mask & (ATTR_UID|ATTR_GID)) {
611 * These IDs could have changed since we last looked at them.
612 * But, we're assured that if the ownership did change
613 * while we didn't have the inode locked, inode's dquot(s)
614 * would have changed also.
618 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
619 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
622 * Do a quota reservation only if uid/gid is actually
625 if (XFS_IS_QUOTA_RUNNING(mp) &&
626 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
627 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
629 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
630 NULL, capable(CAP_FOWNER) ?
631 XFS_QMOPT_FORCE_RES : 0);
632 if (error) /* out of quota */
633 goto out_trans_cancel;
637 xfs_trans_ijoin(tp, ip, 0);
640 * Change file ownership. Must be the owner or privileged.
642 if (mask & (ATTR_UID|ATTR_GID)) {
644 * CAP_FSETID overrides the following restrictions:
646 * The set-user-ID and set-group-ID bits of a file will be
647 * cleared upon successful return from chown()
649 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
650 !capable(CAP_FSETID))
651 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
654 * Change the ownerships and register quota modifications
655 * in the transaction.
657 if (!uid_eq(iuid, uid)) {
658 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
659 ASSERT(mask & ATTR_UID);
661 olddquot1 = xfs_qm_vop_chown(tp, ip,
662 &ip->i_udquot, udqp);
664 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
667 if (!gid_eq(igid, gid)) {
668 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
669 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
670 !XFS_IS_PQUOTA_ON(mp));
671 ASSERT(mask & ATTR_GID);
673 olddquot2 = xfs_qm_vop_chown(tp, ip,
674 &ip->i_gdquot, gdqp);
676 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
681 if (mask & ATTR_MODE)
682 xfs_setattr_mode(ip, iattr);
683 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
684 xfs_setattr_time(ip, iattr);
686 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
688 XFS_STATS_INC(xs_ig_attrchg);
690 if (mp->m_flags & XFS_MOUNT_WSYNC)
691 xfs_trans_set_sync(tp);
692 error = xfs_trans_commit(tp, 0);
694 xfs_iunlock(ip, XFS_ILOCK_EXCL);
697 * Release any dquot(s) the inode had kept before chown.
699 xfs_qm_dqrele(olddquot1);
700 xfs_qm_dqrele(olddquot2);
708 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
709 * update. We could avoid this with linked transactions
710 * and passing down the transaction pointer all the way
711 * to attr_set. No previous user of the generic
712 * Posix ACL code seems to care about this issue either.
714 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
715 error = posix_acl_chmod(inode, inode->i_mode);
723 xfs_trans_cancel(tp, 0);
724 xfs_iunlock(ip, XFS_ILOCK_EXCL);
732 * Truncate file. Must have write permission and not be a directory.
736 struct xfs_inode *ip,
739 struct xfs_mount *mp = ip->i_mount;
740 struct inode *inode = VFS_I(ip);
741 xfs_off_t oldsize, newsize;
742 struct xfs_trans *tp;
745 uint commit_flags = 0;
747 trace_xfs_setattr(ip);
749 if (mp->m_flags & XFS_MOUNT_RDONLY)
752 if (XFS_FORCED_SHUTDOWN(mp))
755 error = inode_change_ok(inode, iattr);
759 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
760 ASSERT(S_ISREG(ip->i_d.di_mode));
761 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
762 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
764 oldsize = inode->i_size;
765 newsize = iattr->ia_size;
768 * Short circuit the truncate case for zero length files.
770 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
771 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
775 * Use the regular setattr path to update the timestamps.
777 iattr->ia_valid &= ~ATTR_SIZE;
778 return xfs_setattr_nonsize(ip, iattr, 0);
782 * Make sure that the dquots are attached to the inode.
784 error = xfs_qm_dqattach(ip, 0);
789 * Now we can make the changes. Before we join the inode to the
790 * transaction, take care of the part of the truncation that must be
791 * done without the inode lock. This needs to be done before joining
792 * the inode to the transaction, because the inode cannot be unlocked
793 * once it is a part of the transaction.
795 if (newsize > oldsize) {
797 * Do the first part of growing a file: zero any data in the
798 * last block that is beyond the old EOF. We need to do this
799 * before the inode is joined to the transaction to modify
802 error = xfs_zero_eof(ip, newsize, oldsize);
808 * We are going to log the inode size change in this transaction so
809 * any previous writes that are beyond the on disk EOF and the new
810 * EOF that have not been written out need to be written here. If we
811 * do not write the data out, we expose ourselves to the null files
814 * Only flush from the on disk size to the smaller of the in memory
815 * file size or the new size as that's the range we really care about
816 * here and prevents waiting for other data not within the range we
819 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
820 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
821 ip->i_d.di_size, newsize);
827 * Wait for all direct I/O to complete.
829 inode_dio_wait(inode);
832 * Do all the page cache truncate work outside the transaction context
833 * as the "lock" order is page lock->log space reservation. i.e.
834 * locking pages inside the transaction can ABBA deadlock with
835 * writeback. We have to do the VFS inode size update before we truncate
836 * the pagecache, however, to avoid racing with page faults beyond the
837 * new EOF they are not serialised against truncate operations except by
838 * page locks and size updates.
840 * Hence we are in a situation where a truncate can fail with ENOMEM
841 * from xfs_trans_reserve(), but having already truncated the in-memory
842 * version of the file (i.e. made user visible changes). There's not
843 * much we can do about this, except to hope that the caller sees ENOMEM
844 * and retries the truncate operation.
846 error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
849 truncate_setsize(inode, newsize);
852 * The "we can't serialise against page faults" pain gets worse.
854 * If the file is mapped then we have to clean the page at the old EOF
855 * when extending the file. Extending the file can expose changes the
856 * underlying page mapping (e.g. from beyond EOF to a hole or
857 * unwritten), and so on the next attempt to write to that page we need
858 * to remap it for write. i.e. we need .page_mkwrite() to be called.
859 * Hence we need to clean the page to clean the pte and so a new write
860 * fault will be triggered appropriately.
862 * If we do it before we change the inode size, then we can race with a
863 * page fault that maps the page with exactly the same problem. If we do
864 * it after we change the file size, then a new page fault can come in
865 * and allocate space before we've run the rest of the truncate
866 * transaction. That's kinda grotesque, but it's better than have data
867 * over a hole, and so that's the lesser evil that has been chosen here.
869 * The real solution, however, is to have some mechanism for locking out
870 * page faults while a truncate is in progress.
872 if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
873 error = filemap_write_and_wait_range(
874 VFS_I(ip)->i_mapping,
875 round_down(oldsize, PAGE_CACHE_SIZE),
876 round_up(oldsize, PAGE_CACHE_SIZE) - 1);
881 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
882 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
884 goto out_trans_cancel;
886 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
887 lock_flags |= XFS_ILOCK_EXCL;
888 xfs_ilock(ip, XFS_ILOCK_EXCL);
889 xfs_trans_ijoin(tp, ip, 0);
892 * Only change the c/mtime if we are changing the size or we are
893 * explicitly asked to change it. This handles the semantic difference
894 * between truncate() and ftruncate() as implemented in the VFS.
896 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
897 * special case where we need to update the times despite not having
898 * these flags set. For all other operations the VFS set these flags
899 * explicitly if it wants a timestamp update.
901 if (newsize != oldsize &&
902 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
903 iattr->ia_ctime = iattr->ia_mtime =
904 current_fs_time(inode->i_sb);
905 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
909 * The first thing we do is set the size to new_size permanently on
910 * disk. This way we don't have to worry about anyone ever being able
911 * to look at the data being freed even in the face of a crash.
912 * What we're getting around here is the case where we free a block, it
913 * is allocated to another file, it is written to, and then we crash.
914 * If the new data gets written to the file but the log buffers
915 * containing the free and reallocation don't, then we'd end up with
916 * garbage in the blocks being freed. As long as we make the new size
917 * permanent before actually freeing any blocks it doesn't matter if
918 * they get written to.
920 ip->i_d.di_size = newsize;
921 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
923 if (newsize <= oldsize) {
924 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
926 goto out_trans_abort;
929 * Truncated "down", so we're removing references to old data
930 * here - if we delay flushing for a long time, we expose
931 * ourselves unduly to the notorious NULL files problem. So,
932 * we mark this inode and flush it when the file is closed,
933 * and do not wait the usual (long) time for writeout.
935 xfs_iflags_set(ip, XFS_ITRUNCATED);
937 /* A truncate down always removes post-EOF blocks. */
938 xfs_inode_clear_eofblocks_tag(ip);
941 if (iattr->ia_valid & ATTR_MODE)
942 xfs_setattr_mode(ip, iattr);
943 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
944 xfs_setattr_time(ip, iattr);
946 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
948 XFS_STATS_INC(xs_ig_attrchg);
950 if (mp->m_flags & XFS_MOUNT_WSYNC)
951 xfs_trans_set_sync(tp);
953 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
956 xfs_iunlock(ip, lock_flags);
960 commit_flags |= XFS_TRANS_ABORT;
962 xfs_trans_cancel(tp, commit_flags);
968 struct dentry *dentry,
971 struct xfs_inode *ip = XFS_I(dentry->d_inode);
974 if (iattr->ia_valid & ATTR_SIZE) {
975 xfs_ilock(ip, XFS_IOLOCK_EXCL);
976 error = xfs_setattr_size(ip, iattr);
977 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
979 error = xfs_setattr_nonsize(ip, iattr, 0);
988 struct timespec *now,
991 struct xfs_inode *ip = XFS_I(inode);
992 struct xfs_mount *mp = ip->i_mount;
993 struct xfs_trans *tp;
996 trace_xfs_update_time(ip);
998 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
999 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1001 xfs_trans_cancel(tp, 0);
1005 xfs_ilock(ip, XFS_ILOCK_EXCL);
1006 if (flags & S_CTIME) {
1007 inode->i_ctime = *now;
1008 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
1009 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
1011 if (flags & S_MTIME) {
1012 inode->i_mtime = *now;
1013 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
1014 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
1016 if (flags & S_ATIME) {
1017 inode->i_atime = *now;
1018 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
1019 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
1021 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1022 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1023 return xfs_trans_commit(tp, 0);
1026 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1029 * Call fiemap helper to fill in user data.
1030 * Returns positive errors to xfs_getbmap.
1035 struct getbmapx *bmv,
1039 struct fiemap_extent_info *fieinfo = *arg;
1040 u32 fiemap_flags = 0;
1041 u64 logical, physical, length;
1043 /* Do nothing for a hole */
1044 if (bmv->bmv_block == -1LL)
1047 logical = BBTOB(bmv->bmv_offset);
1048 physical = BBTOB(bmv->bmv_block);
1049 length = BBTOB(bmv->bmv_length);
1051 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1052 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1053 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1054 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1055 FIEMAP_EXTENT_UNKNOWN);
1056 physical = 0; /* no block yet */
1058 if (bmv->bmv_oflags & BMV_OF_LAST)
1059 fiemap_flags |= FIEMAP_EXTENT_LAST;
1061 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1062 length, fiemap_flags);
1065 *full = 1; /* user array now full */
1073 struct inode *inode,
1074 struct fiemap_extent_info *fieinfo,
1078 xfs_inode_t *ip = XFS_I(inode);
1082 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1086 /* Set up bmap header for xfs internal routine */
1087 bm.bmv_offset = BTOBBT(start);
1088 /* Special case for whole file */
1089 if (length == FIEMAP_MAX_OFFSET)
1090 bm.bmv_length = -1LL;
1092 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1094 /* We add one because in getbmap world count includes the header */
1095 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1096 fieinfo->fi_extents_max + 1;
1097 bm.bmv_count = min_t(__s32, bm.bmv_count,
1098 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1099 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1100 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1101 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1102 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1103 bm.bmv_iflags |= BMV_IF_DELALLOC;
1105 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1115 struct dentry *dentry,
1118 return xfs_generic_create(dir, dentry, mode, 0, true);
1121 static const struct inode_operations xfs_inode_operations = {
1122 .get_acl = xfs_get_acl,
1123 .set_acl = xfs_set_acl,
1124 .getattr = xfs_vn_getattr,
1125 .setattr = xfs_vn_setattr,
1126 .setxattr = generic_setxattr,
1127 .getxattr = generic_getxattr,
1128 .removexattr = generic_removexattr,
1129 .listxattr = xfs_vn_listxattr,
1130 .fiemap = xfs_vn_fiemap,
1131 .update_time = xfs_vn_update_time,
1134 static const struct inode_operations xfs_dir_inode_operations = {
1135 .create = xfs_vn_create,
1136 .lookup = xfs_vn_lookup,
1137 .link = xfs_vn_link,
1138 .unlink = xfs_vn_unlink,
1139 .symlink = xfs_vn_symlink,
1140 .mkdir = xfs_vn_mkdir,
1142 * Yes, XFS uses the same method for rmdir and unlink.
1144 * There are some subtile differences deeper in the code,
1145 * but we use S_ISDIR to check for those.
1147 .rmdir = xfs_vn_unlink,
1148 .mknod = xfs_vn_mknod,
1149 .rename = xfs_vn_rename,
1150 .get_acl = xfs_get_acl,
1151 .set_acl = xfs_set_acl,
1152 .getattr = xfs_vn_getattr,
1153 .setattr = xfs_vn_setattr,
1154 .setxattr = generic_setxattr,
1155 .getxattr = generic_getxattr,
1156 .removexattr = generic_removexattr,
1157 .listxattr = xfs_vn_listxattr,
1158 .update_time = xfs_vn_update_time,
1159 .tmpfile = xfs_vn_tmpfile,
1162 static const struct inode_operations xfs_dir_ci_inode_operations = {
1163 .create = xfs_vn_create,
1164 .lookup = xfs_vn_ci_lookup,
1165 .link = xfs_vn_link,
1166 .unlink = xfs_vn_unlink,
1167 .symlink = xfs_vn_symlink,
1168 .mkdir = xfs_vn_mkdir,
1170 * Yes, XFS uses the same method for rmdir and unlink.
1172 * There are some subtile differences deeper in the code,
1173 * but we use S_ISDIR to check for those.
1175 .rmdir = xfs_vn_unlink,
1176 .mknod = xfs_vn_mknod,
1177 .rename = xfs_vn_rename,
1178 .get_acl = xfs_get_acl,
1179 .set_acl = xfs_set_acl,
1180 .getattr = xfs_vn_getattr,
1181 .setattr = xfs_vn_setattr,
1182 .setxattr = generic_setxattr,
1183 .getxattr = generic_getxattr,
1184 .removexattr = generic_removexattr,
1185 .listxattr = xfs_vn_listxattr,
1186 .update_time = xfs_vn_update_time,
1187 .tmpfile = xfs_vn_tmpfile,
1190 static const struct inode_operations xfs_symlink_inode_operations = {
1191 .readlink = generic_readlink,
1192 .follow_link = xfs_vn_follow_link,
1193 .put_link = kfree_put_link,
1194 .getattr = xfs_vn_getattr,
1195 .setattr = xfs_vn_setattr,
1196 .setxattr = generic_setxattr,
1197 .getxattr = generic_getxattr,
1198 .removexattr = generic_removexattr,
1199 .listxattr = xfs_vn_listxattr,
1200 .update_time = xfs_vn_update_time,
1204 xfs_diflags_to_iflags(
1205 struct inode *inode,
1206 struct xfs_inode *ip)
1208 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1209 inode->i_flags |= S_IMMUTABLE;
1211 inode->i_flags &= ~S_IMMUTABLE;
1212 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1213 inode->i_flags |= S_APPEND;
1215 inode->i_flags &= ~S_APPEND;
1216 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1217 inode->i_flags |= S_SYNC;
1219 inode->i_flags &= ~S_SYNC;
1220 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1221 inode->i_flags |= S_NOATIME;
1223 inode->i_flags &= ~S_NOATIME;
1227 * Initialize the Linux inode, set up the operation vectors and
1230 * When reading existing inodes from disk this is called directly
1231 * from xfs_iget, when creating a new inode it is called from
1232 * xfs_ialloc after setting up the inode.
1234 * We are always called with an uninitialised linux inode here.
1235 * We need to initialise the necessary fields and take a reference
1240 struct xfs_inode *ip)
1242 struct inode *inode = &ip->i_vnode;
1245 inode->i_ino = ip->i_ino;
1246 inode->i_state = I_NEW;
1248 inode_sb_list_add(inode);
1249 /* make the inode look hashed for the writeback code */
1250 hlist_add_fake(&inode->i_hash);
1252 inode->i_mode = ip->i_d.di_mode;
1253 set_nlink(inode, ip->i_d.di_nlink);
1254 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1255 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1257 switch (inode->i_mode & S_IFMT) {
1261 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1262 sysv_minor(ip->i_df.if_u2.if_rdev));
1269 inode->i_generation = ip->i_d.di_gen;
1270 i_size_write(inode, ip->i_d.di_size);
1271 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1272 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1273 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1274 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1275 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1276 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1277 xfs_diflags_to_iflags(inode, ip);
1279 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1280 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1281 switch (inode->i_mode & S_IFMT) {
1283 inode->i_op = &xfs_inode_operations;
1284 inode->i_fop = &xfs_file_operations;
1285 inode->i_mapping->a_ops = &xfs_address_space_operations;
1288 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1289 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1290 inode->i_op = &xfs_dir_ci_inode_operations;
1292 inode->i_op = &xfs_dir_inode_operations;
1293 inode->i_fop = &xfs_dir_file_operations;
1294 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1297 inode->i_op = &xfs_symlink_inode_operations;
1298 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1299 inode->i_mapping->a_ops = &xfs_address_space_operations;
1302 inode->i_op = &xfs_inode_operations;
1303 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1308 * Ensure all page cache allocations are done from GFP_NOFS context to
1309 * prevent direct reclaim recursion back into the filesystem and blowing
1310 * stacks or deadlocking.
1312 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1313 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1316 * If there is no attribute fork no ACL can exist on this inode,
1317 * and it can't have any file capabilities attached to it either.
1319 if (!XFS_IFORK_Q(ip)) {
1320 inode_has_no_xattr(inode);
1321 cache_no_acl(inode);
1324 xfs_iflags_clear(ip, XFS_INEW);
1327 unlock_new_inode(inode);