2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_inode.h"
28 #include "xfs_bmap_util.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_symlink.h"
37 #include "xfs_da_btree.h"
39 #include "xfs_trans_space.h"
41 #include <linux/capability.h>
42 #include <linux/xattr.h>
43 #include <linux/namei.h>
44 #include <linux/posix_acl.h>
45 #include <linux/security.h>
46 #include <linux/fiemap.h>
47 #include <linux/slab.h>
50 * Directories have different lock order w.r.t. mmap_sem compared to regular
51 * files. This is due to readdir potentially triggering page faults on a user
52 * buffer inside filldir(), and this happens with the ilock on the directory
53 * held. For regular files, the lock order is the other way around - the
54 * mmap_sem is taken during the page fault, and then we lock the ilock to do
55 * block mapping. Hence we need a different class for the directory ilock so
56 * that lockdep can tell them apart.
58 static struct lock_class_key xfs_nondir_ilock_class;
59 static struct lock_class_key xfs_dir_ilock_class;
64 const struct xattr *xattr_array,
67 const struct xattr *xattr;
68 struct xfs_inode *ip = XFS_I(inode);
71 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
72 error = xfs_attr_set(ip, xattr->name, xattr->value,
73 xattr->value_len, ATTR_SECURE);
81 * Hook in SELinux. This is not quite correct yet, what we really need
82 * here (as we do for default ACLs) is a mechanism by which creation of
83 * these attrs can be journalled at inode creation time (along with the
84 * inode, of course, such that log replay can't cause these to be lost).
91 const struct qstr *qstr)
93 return security_inode_init_security(inode, dir, qstr,
94 &xfs_initxattrs, NULL);
99 struct xfs_name *namep,
100 struct dentry *dentry,
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
112 struct dentry *dentry)
114 struct xfs_name teardown;
117 * If we can't add the ACL or we fail in
118 * xfs_init_security we must back out.
119 * ENOSPC can hit here, among other things.
121 xfs_dentry_to_name(&teardown, dentry, 0);
123 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
129 struct dentry *dentry,
132 bool tmpfile) /* unnamed file */
135 struct xfs_inode *ip = NULL;
136 struct posix_acl *default_acl, *acl;
137 struct xfs_name name;
141 * Irix uses Missed'em'V split, but doesn't want to see
142 * the upper 5 bits of (14bit) major.
144 if (S_ISCHR(mode) || S_ISBLK(mode)) {
145 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
147 rdev = sysv_encode_dev(rdev);
152 error = posix_acl_create(dir, &mode, &default_acl, &acl);
157 xfs_dentry_to_name(&name, dentry, mode);
158 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
160 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
167 error = xfs_init_security(inode, dir, &dentry->d_name);
169 goto out_cleanup_inode;
171 #ifdef CONFIG_XFS_POSIX_ACL
173 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
175 goto out_cleanup_inode;
178 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
180 goto out_cleanup_inode;
185 d_tmpfile(dentry, inode);
187 d_instantiate(dentry, inode);
191 posix_acl_release(default_acl);
193 posix_acl_release(acl);
198 xfs_cleanup_inode(dir, inode, dentry);
206 struct dentry *dentry,
210 return xfs_generic_create(dir, dentry, mode, rdev, false);
216 struct dentry *dentry,
220 return xfs_vn_mknod(dir, dentry, mode, 0);
226 struct dentry *dentry,
229 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
232 STATIC struct dentry *
235 struct dentry *dentry,
238 struct xfs_inode *cip;
239 struct xfs_name name;
242 if (dentry->d_name.len >= MAXNAMELEN)
243 return ERR_PTR(-ENAMETOOLONG);
245 xfs_dentry_to_name(&name, dentry, 0);
246 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
247 if (unlikely(error)) {
248 if (unlikely(error != -ENOENT))
249 return ERR_PTR(error);
254 return d_splice_alias(VFS_I(cip), dentry);
257 STATIC struct dentry *
260 struct dentry *dentry,
263 struct xfs_inode *ip;
264 struct xfs_name xname;
265 struct xfs_name ci_name;
269 if (dentry->d_name.len >= MAXNAMELEN)
270 return ERR_PTR(-ENAMETOOLONG);
272 xfs_dentry_to_name(&xname, dentry, 0);
273 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
274 if (unlikely(error)) {
275 if (unlikely(error != -ENOENT))
276 return ERR_PTR(error);
278 * call d_add(dentry, NULL) here when d_drop_negative_children
279 * is called in xfs_vn_mknod (ie. allow negative dentries
280 * with CI filesystems).
285 /* if exact match, just splice and exit */
287 return d_splice_alias(VFS_I(ip), dentry);
289 /* else case-insensitive match... */
290 dname.name = ci_name.name;
291 dname.len = ci_name.len;
292 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
293 kmem_free(ci_name.name);
299 struct dentry *old_dentry,
301 struct dentry *dentry)
303 struct inode *inode = old_dentry->d_inode;
304 struct xfs_name name;
307 xfs_dentry_to_name(&name, dentry, inode->i_mode);
309 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
314 d_instantiate(dentry, inode);
321 struct dentry *dentry)
323 struct xfs_name name;
326 xfs_dentry_to_name(&name, dentry, 0);
328 error = xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
333 * With unlink, the VFS makes the dentry "negative": no inode,
334 * but still hashed. This is incompatible with case-insensitive
335 * mode, so invalidate (unhash) the dentry in CI-mode.
337 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
338 d_invalidate(dentry);
345 struct dentry *dentry,
349 struct xfs_inode *cip = NULL;
350 struct xfs_name name;
355 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
356 xfs_dentry_to_name(&name, dentry, mode);
358 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
364 error = xfs_init_security(inode, dir, &dentry->d_name);
366 goto out_cleanup_inode;
368 d_instantiate(dentry, inode);
372 xfs_cleanup_inode(dir, inode, dentry);
381 struct dentry *odentry,
383 struct dentry *ndentry,
386 struct inode *new_inode = ndentry->d_inode;
388 struct xfs_name oname;
389 struct xfs_name nname;
391 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
394 /* if we are exchanging files, we need to set i_mode of both files */
395 if (flags & RENAME_EXCHANGE)
396 omode = ndentry->d_inode->i_mode;
398 xfs_dentry_to_name(&oname, odentry, omode);
399 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
401 return xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
403 new_inode ? XFS_I(new_inode) : NULL, flags);
407 * careful here - this function can get called recursively, so
408 * we need to be very careful about how much stack we use.
409 * uio is kmalloced for this reason...
413 struct dentry *dentry,
414 struct nameidata *nd)
419 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
423 error = xfs_readlink(XFS_I(dentry->d_inode), link);
427 nd_set_link(nd, link);
433 nd_set_link(nd, ERR_PTR(error));
439 struct vfsmount *mnt,
440 struct dentry *dentry,
443 struct inode *inode = dentry->d_inode;
444 struct xfs_inode *ip = XFS_I(inode);
445 struct xfs_mount *mp = ip->i_mount;
447 trace_xfs_getattr(ip);
449 if (XFS_FORCED_SHUTDOWN(mp))
452 stat->size = XFS_ISIZE(ip);
453 stat->dev = inode->i_sb->s_dev;
454 stat->mode = ip->i_d.di_mode;
455 stat->nlink = ip->i_d.di_nlink;
456 stat->uid = inode->i_uid;
457 stat->gid = inode->i_gid;
458 stat->ino = ip->i_ino;
459 stat->atime = inode->i_atime;
460 stat->mtime = inode->i_mtime;
461 stat->ctime = inode->i_ctime;
463 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
466 switch (inode->i_mode & S_IFMT) {
469 stat->blksize = BLKDEV_IOSIZE;
470 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
471 sysv_minor(ip->i_df.if_u2.if_rdev));
474 if (XFS_IS_REALTIME_INODE(ip)) {
476 * If the file blocks are being allocated from a
477 * realtime volume, then return the inode's realtime
478 * extent size or the realtime volume's extent size.
481 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
483 stat->blksize = xfs_preferred_iosize(mp);
493 struct xfs_inode *ip,
496 struct inode *inode = VFS_I(ip);
497 umode_t mode = iattr->ia_mode;
499 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
501 ip->i_d.di_mode &= S_IFMT;
502 ip->i_d.di_mode |= mode & ~S_IFMT;
504 inode->i_mode &= S_IFMT;
505 inode->i_mode |= mode & ~S_IFMT;
510 struct xfs_inode *ip,
513 struct inode *inode = VFS_I(ip);
515 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
517 if (iattr->ia_valid & ATTR_ATIME) {
518 inode->i_atime = iattr->ia_atime;
519 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
520 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
522 if (iattr->ia_valid & ATTR_CTIME) {
523 inode->i_ctime = iattr->ia_ctime;
524 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
525 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
527 if (iattr->ia_valid & ATTR_MTIME) {
528 inode->i_mtime = iattr->ia_mtime;
529 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
530 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
536 struct xfs_inode *ip,
540 xfs_mount_t *mp = ip->i_mount;
541 struct inode *inode = VFS_I(ip);
542 int mask = iattr->ia_valid;
545 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
546 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
547 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
548 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
550 trace_xfs_setattr(ip);
552 /* If acls are being inherited, we already have this checked */
553 if (!(flags & XFS_ATTR_NOACL)) {
554 if (mp->m_flags & XFS_MOUNT_RDONLY)
557 if (XFS_FORCED_SHUTDOWN(mp))
560 error = inode_change_ok(inode, iattr);
565 ASSERT((mask & ATTR_SIZE) == 0);
568 * If disk quotas is on, we make sure that the dquots do exist on disk,
569 * before we start any other transactions. Trying to do this later
570 * is messy. We don't care to take a readlock to look at the ids
571 * in inode here, because we can't hold it across the trans_reserve.
572 * If the IDs do change before we take the ilock, we're covered
573 * because the i_*dquot fields will get updated anyway.
575 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
578 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
580 qflags |= XFS_QMOPT_UQUOTA;
584 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
586 qflags |= XFS_QMOPT_GQUOTA;
592 * We take a reference when we initialize udqp and gdqp,
593 * so it is important that we never blindly double trip on
594 * the same variable. See xfs_create() for an example.
596 ASSERT(udqp == NULL);
597 ASSERT(gdqp == NULL);
598 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
599 xfs_kgid_to_gid(gid),
601 qflags, &udqp, &gdqp, NULL);
606 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
607 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
611 xfs_ilock(ip, XFS_ILOCK_EXCL);
614 * Change file ownership. Must be the owner or privileged.
616 if (mask & (ATTR_UID|ATTR_GID)) {
618 * These IDs could have changed since we last looked at them.
619 * But, we're assured that if the ownership did change
620 * while we didn't have the inode locked, inode's dquot(s)
621 * would have changed also.
625 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
626 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
629 * Do a quota reservation only if uid/gid is actually
632 if (XFS_IS_QUOTA_RUNNING(mp) &&
633 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
634 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
636 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
637 NULL, capable(CAP_FOWNER) ?
638 XFS_QMOPT_FORCE_RES : 0);
639 if (error) /* out of quota */
640 goto out_trans_cancel;
644 xfs_trans_ijoin(tp, ip, 0);
647 * Change file ownership. Must be the owner or privileged.
649 if (mask & (ATTR_UID|ATTR_GID)) {
651 * CAP_FSETID overrides the following restrictions:
653 * The set-user-ID and set-group-ID bits of a file will be
654 * cleared upon successful return from chown()
656 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
657 !capable(CAP_FSETID))
658 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
661 * Change the ownerships and register quota modifications
662 * in the transaction.
664 if (!uid_eq(iuid, uid)) {
665 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
666 ASSERT(mask & ATTR_UID);
668 olddquot1 = xfs_qm_vop_chown(tp, ip,
669 &ip->i_udquot, udqp);
671 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
674 if (!gid_eq(igid, gid)) {
675 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
676 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
677 !XFS_IS_PQUOTA_ON(mp));
678 ASSERT(mask & ATTR_GID);
680 olddquot2 = xfs_qm_vop_chown(tp, ip,
681 &ip->i_gdquot, gdqp);
683 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
688 if (mask & ATTR_MODE)
689 xfs_setattr_mode(ip, iattr);
690 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
691 xfs_setattr_time(ip, iattr);
693 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
695 XFS_STATS_INC(xs_ig_attrchg);
697 if (mp->m_flags & XFS_MOUNT_WSYNC)
698 xfs_trans_set_sync(tp);
699 error = xfs_trans_commit(tp, 0);
701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
704 * Release any dquot(s) the inode had kept before chown.
706 xfs_qm_dqrele(olddquot1);
707 xfs_qm_dqrele(olddquot2);
715 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
716 * update. We could avoid this with linked transactions
717 * and passing down the transaction pointer all the way
718 * to attr_set. No previous user of the generic
719 * Posix ACL code seems to care about this issue either.
721 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
722 error = posix_acl_chmod(inode, inode->i_mode);
730 xfs_trans_cancel(tp, 0);
731 xfs_iunlock(ip, XFS_ILOCK_EXCL);
739 * Truncate file. Must have write permission and not be a directory.
743 struct xfs_inode *ip,
746 struct xfs_mount *mp = ip->i_mount;
747 struct inode *inode = VFS_I(ip);
748 xfs_off_t oldsize, newsize;
749 struct xfs_trans *tp;
752 uint commit_flags = 0;
754 trace_xfs_setattr(ip);
756 if (mp->m_flags & XFS_MOUNT_RDONLY)
759 if (XFS_FORCED_SHUTDOWN(mp))
762 error = inode_change_ok(inode, iattr);
766 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
767 ASSERT(S_ISREG(ip->i_d.di_mode));
768 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
769 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
771 oldsize = inode->i_size;
772 newsize = iattr->ia_size;
775 * Short circuit the truncate case for zero length files.
777 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
778 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
782 * Use the regular setattr path to update the timestamps.
784 iattr->ia_valid &= ~ATTR_SIZE;
785 return xfs_setattr_nonsize(ip, iattr, 0);
789 * Make sure that the dquots are attached to the inode.
791 error = xfs_qm_dqattach(ip, 0);
796 * Now we can make the changes. Before we join the inode to the
797 * transaction, take care of the part of the truncation that must be
798 * done without the inode lock. This needs to be done before joining
799 * the inode to the transaction, because the inode cannot be unlocked
800 * once it is a part of the transaction.
802 if (newsize > oldsize) {
804 * Do the first part of growing a file: zero any data in the
805 * last block that is beyond the old EOF. We need to do this
806 * before the inode is joined to the transaction to modify
809 error = xfs_zero_eof(ip, newsize, oldsize);
815 * We are going to log the inode size change in this transaction so
816 * any previous writes that are beyond the on disk EOF and the new
817 * EOF that have not been written out need to be written here. If we
818 * do not write the data out, we expose ourselves to the null files
821 * Only flush from the on disk size to the smaller of the in memory
822 * file size or the new size as that's the range we really care about
823 * here and prevents waiting for other data not within the range we
826 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
827 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
828 ip->i_d.di_size, newsize);
834 * Wait for all direct I/O to complete.
836 inode_dio_wait(inode);
839 * Do all the page cache truncate work outside the transaction context
840 * as the "lock" order is page lock->log space reservation. i.e.
841 * locking pages inside the transaction can ABBA deadlock with
842 * writeback. We have to do the VFS inode size update before we truncate
843 * the pagecache, however, to avoid racing with page faults beyond the
844 * new EOF they are not serialised against truncate operations except by
845 * page locks and size updates.
847 * Hence we are in a situation where a truncate can fail with ENOMEM
848 * from xfs_trans_reserve(), but having already truncated the in-memory
849 * version of the file (i.e. made user visible changes). There's not
850 * much we can do about this, except to hope that the caller sees ENOMEM
851 * and retries the truncate operation.
853 error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
856 truncate_setsize(inode, newsize);
859 * The "we can't serialise against page faults" pain gets worse.
861 * If the file is mapped then we have to clean the page at the old EOF
862 * when extending the file. Extending the file can expose changes the
863 * underlying page mapping (e.g. from beyond EOF to a hole or
864 * unwritten), and so on the next attempt to write to that page we need
865 * to remap it for write. i.e. we need .page_mkwrite() to be called.
866 * Hence we need to clean the page to clean the pte and so a new write
867 * fault will be triggered appropriately.
869 * If we do it before we change the inode size, then we can race with a
870 * page fault that maps the page with exactly the same problem. If we do
871 * it after we change the file size, then a new page fault can come in
872 * and allocate space before we've run the rest of the truncate
873 * transaction. That's kinda grotesque, but it's better than have data
874 * over a hole, and so that's the lesser evil that has been chosen here.
876 * The real solution, however, is to have some mechanism for locking out
877 * page faults while a truncate is in progress.
879 if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
880 error = filemap_write_and_wait_range(
881 VFS_I(ip)->i_mapping,
882 round_down(oldsize, PAGE_CACHE_SIZE),
883 round_up(oldsize, PAGE_CACHE_SIZE) - 1);
888 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
889 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
891 goto out_trans_cancel;
893 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
894 lock_flags |= XFS_ILOCK_EXCL;
895 xfs_ilock(ip, XFS_ILOCK_EXCL);
896 xfs_trans_ijoin(tp, ip, 0);
899 * Only change the c/mtime if we are changing the size or we are
900 * explicitly asked to change it. This handles the semantic difference
901 * between truncate() and ftruncate() as implemented in the VFS.
903 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
904 * special case where we need to update the times despite not having
905 * these flags set. For all other operations the VFS set these flags
906 * explicitly if it wants a timestamp update.
908 if (newsize != oldsize &&
909 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
910 iattr->ia_ctime = iattr->ia_mtime =
911 current_fs_time(inode->i_sb);
912 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
916 * The first thing we do is set the size to new_size permanently on
917 * disk. This way we don't have to worry about anyone ever being able
918 * to look at the data being freed even in the face of a crash.
919 * What we're getting around here is the case where we free a block, it
920 * is allocated to another file, it is written to, and then we crash.
921 * If the new data gets written to the file but the log buffers
922 * containing the free and reallocation don't, then we'd end up with
923 * garbage in the blocks being freed. As long as we make the new size
924 * permanent before actually freeing any blocks it doesn't matter if
925 * they get written to.
927 ip->i_d.di_size = newsize;
928 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
930 if (newsize <= oldsize) {
931 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
933 goto out_trans_abort;
936 * Truncated "down", so we're removing references to old data
937 * here - if we delay flushing for a long time, we expose
938 * ourselves unduly to the notorious NULL files problem. So,
939 * we mark this inode and flush it when the file is closed,
940 * and do not wait the usual (long) time for writeout.
942 xfs_iflags_set(ip, XFS_ITRUNCATED);
944 /* A truncate down always removes post-EOF blocks. */
945 xfs_inode_clear_eofblocks_tag(ip);
948 if (iattr->ia_valid & ATTR_MODE)
949 xfs_setattr_mode(ip, iattr);
950 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
951 xfs_setattr_time(ip, iattr);
953 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
955 XFS_STATS_INC(xs_ig_attrchg);
957 if (mp->m_flags & XFS_MOUNT_WSYNC)
958 xfs_trans_set_sync(tp);
960 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
963 xfs_iunlock(ip, lock_flags);
967 commit_flags |= XFS_TRANS_ABORT;
969 xfs_trans_cancel(tp, commit_flags);
975 struct dentry *dentry,
978 struct xfs_inode *ip = XFS_I(dentry->d_inode);
981 if (iattr->ia_valid & ATTR_SIZE) {
982 xfs_ilock(ip, XFS_IOLOCK_EXCL);
983 error = xfs_setattr_size(ip, iattr);
984 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
986 error = xfs_setattr_nonsize(ip, iattr, 0);
995 struct timespec *now,
998 struct xfs_inode *ip = XFS_I(inode);
999 struct xfs_mount *mp = ip->i_mount;
1000 struct xfs_trans *tp;
1003 trace_xfs_update_time(ip);
1005 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
1006 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1008 xfs_trans_cancel(tp, 0);
1012 xfs_ilock(ip, XFS_ILOCK_EXCL);
1013 if (flags & S_CTIME) {
1014 inode->i_ctime = *now;
1015 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
1016 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
1018 if (flags & S_MTIME) {
1019 inode->i_mtime = *now;
1020 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
1021 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
1023 if (flags & S_ATIME) {
1024 inode->i_atime = *now;
1025 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
1026 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
1028 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1029 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1030 return xfs_trans_commit(tp, 0);
1033 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1036 * Call fiemap helper to fill in user data.
1037 * Returns positive errors to xfs_getbmap.
1042 struct getbmapx *bmv,
1046 struct fiemap_extent_info *fieinfo = *arg;
1047 u32 fiemap_flags = 0;
1048 u64 logical, physical, length;
1050 /* Do nothing for a hole */
1051 if (bmv->bmv_block == -1LL)
1054 logical = BBTOB(bmv->bmv_offset);
1055 physical = BBTOB(bmv->bmv_block);
1056 length = BBTOB(bmv->bmv_length);
1058 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1059 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1060 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1061 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1062 FIEMAP_EXTENT_UNKNOWN);
1063 physical = 0; /* no block yet */
1065 if (bmv->bmv_oflags & BMV_OF_LAST)
1066 fiemap_flags |= FIEMAP_EXTENT_LAST;
1068 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1069 length, fiemap_flags);
1072 *full = 1; /* user array now full */
1080 struct inode *inode,
1081 struct fiemap_extent_info *fieinfo,
1085 xfs_inode_t *ip = XFS_I(inode);
1089 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1093 /* Set up bmap header for xfs internal routine */
1094 bm.bmv_offset = BTOBBT(start);
1095 /* Special case for whole file */
1096 if (length == FIEMAP_MAX_OFFSET)
1097 bm.bmv_length = -1LL;
1099 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1101 /* We add one because in getbmap world count includes the header */
1102 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1103 fieinfo->fi_extents_max + 1;
1104 bm.bmv_count = min_t(__s32, bm.bmv_count,
1105 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1106 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1107 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1108 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1109 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1110 bm.bmv_iflags |= BMV_IF_DELALLOC;
1112 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1122 struct dentry *dentry,
1125 return xfs_generic_create(dir, dentry, mode, 0, true);
1128 static const struct inode_operations xfs_inode_operations = {
1129 .get_acl = xfs_get_acl,
1130 .set_acl = xfs_set_acl,
1131 .getattr = xfs_vn_getattr,
1132 .setattr = xfs_vn_setattr,
1133 .setxattr = generic_setxattr,
1134 .getxattr = generic_getxattr,
1135 .removexattr = generic_removexattr,
1136 .listxattr = xfs_vn_listxattr,
1137 .fiemap = xfs_vn_fiemap,
1138 .update_time = xfs_vn_update_time,
1141 static const struct inode_operations xfs_dir_inode_operations = {
1142 .create = xfs_vn_create,
1143 .lookup = xfs_vn_lookup,
1144 .link = xfs_vn_link,
1145 .unlink = xfs_vn_unlink,
1146 .symlink = xfs_vn_symlink,
1147 .mkdir = xfs_vn_mkdir,
1149 * Yes, XFS uses the same method for rmdir and unlink.
1151 * There are some subtile differences deeper in the code,
1152 * but we use S_ISDIR to check for those.
1154 .rmdir = xfs_vn_unlink,
1155 .mknod = xfs_vn_mknod,
1156 .rename2 = xfs_vn_rename,
1157 .get_acl = xfs_get_acl,
1158 .set_acl = xfs_set_acl,
1159 .getattr = xfs_vn_getattr,
1160 .setattr = xfs_vn_setattr,
1161 .setxattr = generic_setxattr,
1162 .getxattr = generic_getxattr,
1163 .removexattr = generic_removexattr,
1164 .listxattr = xfs_vn_listxattr,
1165 .update_time = xfs_vn_update_time,
1166 .tmpfile = xfs_vn_tmpfile,
1169 static const struct inode_operations xfs_dir_ci_inode_operations = {
1170 .create = xfs_vn_create,
1171 .lookup = xfs_vn_ci_lookup,
1172 .link = xfs_vn_link,
1173 .unlink = xfs_vn_unlink,
1174 .symlink = xfs_vn_symlink,
1175 .mkdir = xfs_vn_mkdir,
1177 * Yes, XFS uses the same method for rmdir and unlink.
1179 * There are some subtile differences deeper in the code,
1180 * but we use S_ISDIR to check for those.
1182 .rmdir = xfs_vn_unlink,
1183 .mknod = xfs_vn_mknod,
1184 .rename2 = xfs_vn_rename,
1185 .get_acl = xfs_get_acl,
1186 .set_acl = xfs_set_acl,
1187 .getattr = xfs_vn_getattr,
1188 .setattr = xfs_vn_setattr,
1189 .setxattr = generic_setxattr,
1190 .getxattr = generic_getxattr,
1191 .removexattr = generic_removexattr,
1192 .listxattr = xfs_vn_listxattr,
1193 .update_time = xfs_vn_update_time,
1194 .tmpfile = xfs_vn_tmpfile,
1197 static const struct inode_operations xfs_symlink_inode_operations = {
1198 .readlink = generic_readlink,
1199 .follow_link = xfs_vn_follow_link,
1200 .put_link = kfree_put_link,
1201 .getattr = xfs_vn_getattr,
1202 .setattr = xfs_vn_setattr,
1203 .setxattr = generic_setxattr,
1204 .getxattr = generic_getxattr,
1205 .removexattr = generic_removexattr,
1206 .listxattr = xfs_vn_listxattr,
1207 .update_time = xfs_vn_update_time,
1211 xfs_diflags_to_iflags(
1212 struct inode *inode,
1213 struct xfs_inode *ip)
1215 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1216 inode->i_flags |= S_IMMUTABLE;
1218 inode->i_flags &= ~S_IMMUTABLE;
1219 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1220 inode->i_flags |= S_APPEND;
1222 inode->i_flags &= ~S_APPEND;
1223 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1224 inode->i_flags |= S_SYNC;
1226 inode->i_flags &= ~S_SYNC;
1227 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1228 inode->i_flags |= S_NOATIME;
1230 inode->i_flags &= ~S_NOATIME;
1234 * Initialize the Linux inode, set up the operation vectors and
1237 * When reading existing inodes from disk this is called directly
1238 * from xfs_iget, when creating a new inode it is called from
1239 * xfs_ialloc after setting up the inode.
1241 * We are always called with an uninitialised linux inode here.
1242 * We need to initialise the necessary fields and take a reference
1247 struct xfs_inode *ip)
1249 struct inode *inode = &ip->i_vnode;
1252 inode->i_ino = ip->i_ino;
1253 inode->i_state = I_NEW;
1255 inode_sb_list_add(inode);
1256 /* make the inode look hashed for the writeback code */
1257 hlist_add_fake(&inode->i_hash);
1259 inode->i_mode = ip->i_d.di_mode;
1260 set_nlink(inode, ip->i_d.di_nlink);
1261 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1262 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1264 switch (inode->i_mode & S_IFMT) {
1268 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1269 sysv_minor(ip->i_df.if_u2.if_rdev));
1276 inode->i_generation = ip->i_d.di_gen;
1277 i_size_write(inode, ip->i_d.di_size);
1278 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1279 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1280 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1281 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1282 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1283 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1284 xfs_diflags_to_iflags(inode, ip);
1286 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1287 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1288 switch (inode->i_mode & S_IFMT) {
1290 inode->i_op = &xfs_inode_operations;
1291 inode->i_fop = &xfs_file_operations;
1292 inode->i_mapping->a_ops = &xfs_address_space_operations;
1295 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1296 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1297 inode->i_op = &xfs_dir_ci_inode_operations;
1299 inode->i_op = &xfs_dir_inode_operations;
1300 inode->i_fop = &xfs_dir_file_operations;
1301 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1304 inode->i_op = &xfs_symlink_inode_operations;
1305 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1306 inode->i_mapping->a_ops = &xfs_address_space_operations;
1309 inode->i_op = &xfs_inode_operations;
1310 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1315 * Ensure all page cache allocations are done from GFP_NOFS context to
1316 * prevent direct reclaim recursion back into the filesystem and blowing
1317 * stacks or deadlocking.
1319 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1320 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1323 * If there is no attribute fork no ACL can exist on this inode,
1324 * and it can't have any file capabilities attached to it either.
1326 if (!XFS_IFORK_Q(ip)) {
1327 inode_has_no_xattr(inode);
1328 cache_no_acl(inode);
1331 xfs_iflags_clear(ip, XFS_INEW);
1334 unlock_new_inode(inode);