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"
40 #include "xfs_iomap.h"
42 #include <linux/capability.h>
43 #include <linux/xattr.h>
44 #include <linux/posix_acl.h>
45 #include <linux/security.h>
46 #include <linux/iomap.h>
47 #include <linux/slab.h>
48 #include <linux/iversion.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)
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = XFS_DIR3_FT_UNKNOWN;
109 xfs_dentry_mode_to_name(
110 struct xfs_name *namep,
111 struct dentry *dentry,
114 namep->name = dentry->d_name.name;
115 namep->len = dentry->d_name.len;
116 namep->type = xfs_mode_to_ftype(mode);
118 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
119 return -EFSCORRUPTED;
128 struct dentry *dentry)
130 struct xfs_name teardown;
133 * If we can't add the ACL or we fail in
134 * xfs_init_security we must back out.
135 * ENOSPC can hit here, among other things.
137 xfs_dentry_to_name(&teardown, dentry);
139 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
145 struct dentry *dentry,
148 bool tmpfile) /* unnamed file */
151 struct xfs_inode *ip = NULL;
152 struct posix_acl *default_acl, *acl;
153 struct xfs_name name;
157 * Irix uses Missed'em'V split, but doesn't want to see
158 * the upper 5 bits of (14bit) major.
160 if (S_ISCHR(mode) || S_ISBLK(mode)) {
161 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
167 error = posix_acl_create(dir, &mode, &default_acl, &acl);
171 /* Verify mode is valid also for tmpfile case */
172 error = xfs_dentry_mode_to_name(&name, dentry, mode);
177 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
179 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
186 error = xfs_init_security(inode, dir, &dentry->d_name);
188 goto out_cleanup_inode;
190 #ifdef CONFIG_XFS_POSIX_ACL
192 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
194 goto out_cleanup_inode;
197 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
199 goto out_cleanup_inode;
206 d_tmpfile(dentry, inode);
208 d_instantiate(dentry, inode);
210 xfs_finish_inode_setup(ip);
214 posix_acl_release(default_acl);
216 posix_acl_release(acl);
220 xfs_finish_inode_setup(ip);
222 xfs_cleanup_inode(dir, inode, dentry);
230 struct dentry *dentry,
234 return xfs_generic_create(dir, dentry, mode, rdev, false);
240 struct dentry *dentry,
244 return xfs_vn_mknod(dir, dentry, mode, 0);
250 struct dentry *dentry,
253 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
256 STATIC struct dentry *
259 struct dentry *dentry,
263 struct xfs_inode *cip;
264 struct xfs_name name;
267 if (dentry->d_name.len >= MAXNAMELEN)
268 return ERR_PTR(-ENAMETOOLONG);
270 xfs_dentry_to_name(&name, dentry);
271 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
274 else if (likely(error == -ENOENT))
277 inode = ERR_PTR(error);
278 return d_splice_alias(inode, dentry);
281 STATIC struct dentry *
284 struct dentry *dentry,
287 struct xfs_inode *ip;
288 struct xfs_name xname;
289 struct xfs_name ci_name;
293 if (dentry->d_name.len >= MAXNAMELEN)
294 return ERR_PTR(-ENAMETOOLONG);
296 xfs_dentry_to_name(&xname, dentry);
297 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
298 if (unlikely(error)) {
299 if (unlikely(error != -ENOENT))
300 return ERR_PTR(error);
302 * call d_add(dentry, NULL) here when d_drop_negative_children
303 * is called in xfs_vn_mknod (ie. allow negative dentries
304 * with CI filesystems).
309 /* if exact match, just splice and exit */
311 return d_splice_alias(VFS_I(ip), dentry);
313 /* else case-insensitive match... */
314 dname.name = ci_name.name;
315 dname.len = ci_name.len;
316 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
317 kmem_free(ci_name.name);
323 struct dentry *old_dentry,
325 struct dentry *dentry)
327 struct inode *inode = d_inode(old_dentry);
328 struct xfs_name name;
331 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
335 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
340 d_instantiate(dentry, inode);
347 struct dentry *dentry)
349 struct xfs_name name;
352 xfs_dentry_to_name(&name, dentry);
354 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
359 * With unlink, the VFS makes the dentry "negative": no inode,
360 * but still hashed. This is incompatible with case-insensitive
361 * mode, so invalidate (unhash) the dentry in CI-mode.
363 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
364 d_invalidate(dentry);
371 struct dentry *dentry,
375 struct xfs_inode *cip = NULL;
376 struct xfs_name name;
381 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
382 error = xfs_dentry_mode_to_name(&name, dentry, mode);
386 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
392 error = xfs_init_security(inode, dir, &dentry->d_name);
394 goto out_cleanup_inode;
398 d_instantiate(dentry, inode);
399 xfs_finish_inode_setup(cip);
403 xfs_finish_inode_setup(cip);
404 xfs_cleanup_inode(dir, inode, dentry);
413 struct dentry *odentry,
415 struct dentry *ndentry,
418 struct inode *new_inode = d_inode(ndentry);
421 struct xfs_name oname;
422 struct xfs_name nname;
424 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
427 /* if we are exchanging files, we need to set i_mode of both files */
428 if (flags & RENAME_EXCHANGE)
429 omode = d_inode(ndentry)->i_mode;
431 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
432 if (omode && unlikely(error))
435 error = xfs_dentry_mode_to_name(&nname, ndentry,
436 d_inode(odentry)->i_mode);
440 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
442 new_inode ? XFS_I(new_inode) : NULL, flags);
446 * careful here - this function can get called recursively, so
447 * we need to be very careful about how much stack we use.
448 * uio is kmalloced for this reason...
452 struct dentry *dentry,
454 struct delayed_call *done)
460 return ERR_PTR(-ECHILD);
462 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
466 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
470 set_delayed_call(done, kfree_link, link);
476 return ERR_PTR(error);
480 xfs_vn_get_link_inline(
481 struct dentry *dentry,
483 struct delayed_call *done)
485 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
486 return XFS_I(inode)->i_df.if_u1.if_data;
491 const struct path *path,
494 unsigned int query_flags)
496 struct inode *inode = d_inode(path->dentry);
497 struct xfs_inode *ip = XFS_I(inode);
498 struct xfs_mount *mp = ip->i_mount;
500 trace_xfs_getattr(ip);
502 if (XFS_FORCED_SHUTDOWN(mp))
505 stat->size = XFS_ISIZE(ip);
506 stat->dev = inode->i_sb->s_dev;
507 stat->mode = inode->i_mode;
508 stat->nlink = inode->i_nlink;
509 stat->uid = inode->i_uid;
510 stat->gid = inode->i_gid;
511 stat->ino = ip->i_ino;
512 stat->atime = inode->i_atime;
513 stat->mtime = inode->i_mtime;
514 stat->ctime = inode->i_ctime;
516 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
518 if (ip->i_d.di_version == 3) {
519 if (request_mask & STATX_BTIME) {
520 stat->result_mask |= STATX_BTIME;
521 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
522 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
526 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
527 stat->attributes |= STATX_ATTR_IMMUTABLE;
528 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
529 stat->attributes |= STATX_ATTR_APPEND;
530 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
531 stat->attributes |= STATX_ATTR_NODUMP;
533 switch (inode->i_mode & S_IFMT) {
536 stat->blksize = BLKDEV_IOSIZE;
537 stat->rdev = inode->i_rdev;
540 if (XFS_IS_REALTIME_INODE(ip)) {
542 * If the file blocks are being allocated from a
543 * realtime volume, then return the inode's realtime
544 * extent size or the realtime volume's extent size.
547 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
549 stat->blksize = xfs_preferred_iosize(mp);
559 struct xfs_inode *ip,
562 struct inode *inode = VFS_I(ip);
563 umode_t mode = iattr->ia_mode;
565 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
567 inode->i_mode &= S_IFMT;
568 inode->i_mode |= mode & ~S_IFMT;
573 struct xfs_inode *ip,
576 struct inode *inode = VFS_I(ip);
578 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
580 if (iattr->ia_valid & ATTR_ATIME)
581 inode->i_atime = iattr->ia_atime;
582 if (iattr->ia_valid & ATTR_CTIME)
583 inode->i_ctime = iattr->ia_ctime;
584 if (iattr->ia_valid & ATTR_MTIME)
585 inode->i_mtime = iattr->ia_mtime;
590 struct dentry *dentry,
593 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
595 if (mp->m_flags & XFS_MOUNT_RDONLY)
598 if (XFS_FORCED_SHUTDOWN(mp))
601 return setattr_prepare(dentry, iattr);
605 * Set non-size attributes of an inode.
607 * Caution: The caller of this function is responsible for calling
608 * setattr_prepare() or otherwise verifying the change is fine.
612 struct xfs_inode *ip,
616 xfs_mount_t *mp = ip->i_mount;
617 struct inode *inode = VFS_I(ip);
618 int mask = iattr->ia_valid;
621 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
622 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
623 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
624 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
626 ASSERT((mask & ATTR_SIZE) == 0);
629 * If disk quotas is on, we make sure that the dquots do exist on disk,
630 * before we start any other transactions. Trying to do this later
631 * is messy. We don't care to take a readlock to look at the ids
632 * in inode here, because we can't hold it across the trans_reserve.
633 * If the IDs do change before we take the ilock, we're covered
634 * because the i_*dquot fields will get updated anyway.
636 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
639 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
641 qflags |= XFS_QMOPT_UQUOTA;
645 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
647 qflags |= XFS_QMOPT_GQUOTA;
653 * We take a reference when we initialize udqp and gdqp,
654 * so it is important that we never blindly double trip on
655 * the same variable. See xfs_create() for an example.
657 ASSERT(udqp == NULL);
658 ASSERT(gdqp == NULL);
659 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
660 xfs_kgid_to_gid(gid),
662 qflags, &udqp, &gdqp, NULL);
667 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
671 xfs_ilock(ip, XFS_ILOCK_EXCL);
672 xfs_trans_ijoin(tp, ip, 0);
675 * Change file ownership. Must be the owner or privileged.
677 if (mask & (ATTR_UID|ATTR_GID)) {
679 * These IDs could have changed since we last looked at them.
680 * But, we're assured that if the ownership did change
681 * while we didn't have the inode locked, inode's dquot(s)
682 * would have changed also.
686 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
687 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
690 * Do a quota reservation only if uid/gid is actually
693 if (XFS_IS_QUOTA_RUNNING(mp) &&
694 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
695 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
697 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
698 NULL, capable(CAP_FOWNER) ?
699 XFS_QMOPT_FORCE_RES : 0);
700 if (error) /* out of quota */
706 * Change file ownership. Must be the owner or privileged.
708 if (mask & (ATTR_UID|ATTR_GID)) {
710 * CAP_FSETID overrides the following restrictions:
712 * The set-user-ID and set-group-ID bits of a file will be
713 * cleared upon successful return from chown()
715 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
716 !capable(CAP_FSETID))
717 inode->i_mode &= ~(S_ISUID|S_ISGID);
720 * Change the ownerships and register quota modifications
721 * in the transaction.
723 if (!uid_eq(iuid, uid)) {
724 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
725 ASSERT(mask & ATTR_UID);
727 olddquot1 = xfs_qm_vop_chown(tp, ip,
728 &ip->i_udquot, udqp);
730 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
733 if (!gid_eq(igid, gid)) {
734 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
735 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
736 !XFS_IS_PQUOTA_ON(mp));
737 ASSERT(mask & ATTR_GID);
739 olddquot2 = xfs_qm_vop_chown(tp, ip,
740 &ip->i_gdquot, gdqp);
742 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
747 if (mask & ATTR_MODE)
748 xfs_setattr_mode(ip, iattr);
749 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
750 xfs_setattr_time(ip, iattr);
752 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
754 XFS_STATS_INC(mp, xs_ig_attrchg);
756 if (mp->m_flags & XFS_MOUNT_WSYNC)
757 xfs_trans_set_sync(tp);
758 error = xfs_trans_commit(tp);
760 xfs_iunlock(ip, XFS_ILOCK_EXCL);
763 * Release any dquot(s) the inode had kept before chown.
765 xfs_qm_dqrele(olddquot1);
766 xfs_qm_dqrele(olddquot2);
774 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
775 * update. We could avoid this with linked transactions
776 * and passing down the transaction pointer all the way
777 * to attr_set. No previous user of the generic
778 * Posix ACL code seems to care about this issue either.
780 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
781 error = posix_acl_chmod(inode, inode->i_mode);
789 xfs_trans_cancel(tp);
797 xfs_vn_setattr_nonsize(
798 struct dentry *dentry,
801 struct xfs_inode *ip = XFS_I(d_inode(dentry));
804 trace_xfs_setattr(ip);
806 error = xfs_vn_change_ok(dentry, iattr);
809 return xfs_setattr_nonsize(ip, iattr, 0);
813 * Truncate file. Must have write permission and not be a directory.
815 * Caution: The caller of this function is responsible for calling
816 * setattr_prepare() or otherwise verifying the change is fine.
820 struct xfs_inode *ip,
823 struct xfs_mount *mp = ip->i_mount;
824 struct inode *inode = VFS_I(ip);
825 xfs_off_t oldsize, newsize;
826 struct xfs_trans *tp;
829 bool did_zeroing = false;
831 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
832 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
833 ASSERT(S_ISREG(inode->i_mode));
834 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
835 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
837 oldsize = inode->i_size;
838 newsize = iattr->ia_size;
841 * Short circuit the truncate case for zero length files.
843 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
844 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
848 * Use the regular setattr path to update the timestamps.
850 iattr->ia_valid &= ~ATTR_SIZE;
851 return xfs_setattr_nonsize(ip, iattr, 0);
855 * Make sure that the dquots are attached to the inode.
857 error = xfs_qm_dqattach(ip);
862 * Wait for all direct I/O to complete.
864 inode_dio_wait(inode);
867 * File data changes must be complete before we start the transaction to
868 * modify the inode. This needs to be done before joining the inode to
869 * the transaction because the inode cannot be unlocked once it is a
870 * part of the transaction.
872 * Start with zeroing any data beyond EOF that we may expose on file
873 * extension, or zeroing out the rest of the block on a downward
876 if (newsize > oldsize) {
877 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
878 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
879 &did_zeroing, &xfs_iomap_ops);
881 error = iomap_truncate_page(inode, newsize, &did_zeroing,
889 * We've already locked out new page faults, so now we can safely remove
890 * pages from the page cache knowing they won't get refaulted until we
891 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
892 * complete. The truncate_setsize() call also cleans partial EOF page
893 * PTEs on extending truncates and hence ensures sub-page block size
894 * filesystems are correctly handled, too.
896 * We have to do all the page cache truncate work outside the
897 * transaction context as the "lock" order is page lock->log space
898 * reservation as defined by extent allocation in the writeback path.
899 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
900 * having already truncated the in-memory version of the file (i.e. made
901 * user visible changes). There's not much we can do about this, except
902 * to hope that the caller sees ENOMEM and retries the truncate
905 * And we update in-core i_size and truncate page cache beyond newsize
906 * before writeback the [di_size, newsize] range, so we're guaranteed
907 * not to write stale data past the new EOF on truncate down.
909 truncate_setsize(inode, newsize);
912 * We are going to log the inode size change in this transaction so
913 * any previous writes that are beyond the on disk EOF and the new
914 * EOF that have not been written out need to be written here. If we
915 * do not write the data out, we expose ourselves to the null files
916 * problem. Note that this includes any block zeroing we did above;
917 * otherwise those blocks may not be zeroed after a crash.
920 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
921 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
922 ip->i_d.di_size, newsize - 1);
927 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
931 lock_flags |= XFS_ILOCK_EXCL;
932 xfs_ilock(ip, XFS_ILOCK_EXCL);
933 xfs_trans_ijoin(tp, ip, 0);
936 * Only change the c/mtime if we are changing the size or we are
937 * explicitly asked to change it. This handles the semantic difference
938 * between truncate() and ftruncate() as implemented in the VFS.
940 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
941 * special case where we need to update the times despite not having
942 * these flags set. For all other operations the VFS set these flags
943 * explicitly if it wants a timestamp update.
945 if (newsize != oldsize &&
946 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
947 iattr->ia_ctime = iattr->ia_mtime =
949 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
953 * The first thing we do is set the size to new_size permanently on
954 * disk. This way we don't have to worry about anyone ever being able
955 * to look at the data being freed even in the face of a crash.
956 * What we're getting around here is the case where we free a block, it
957 * is allocated to another file, it is written to, and then we crash.
958 * If the new data gets written to the file but the log buffers
959 * containing the free and reallocation don't, then we'd end up with
960 * garbage in the blocks being freed. As long as we make the new size
961 * permanent before actually freeing any blocks it doesn't matter if
962 * they get written to.
964 ip->i_d.di_size = newsize;
965 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
967 if (newsize <= oldsize) {
968 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
970 goto out_trans_cancel;
973 * Truncated "down", so we're removing references to old data
974 * here - if we delay flushing for a long time, we expose
975 * ourselves unduly to the notorious NULL files problem. So,
976 * we mark this inode and flush it when the file is closed,
977 * and do not wait the usual (long) time for writeout.
979 xfs_iflags_set(ip, XFS_ITRUNCATED);
981 /* A truncate down always removes post-EOF blocks. */
982 xfs_inode_clear_eofblocks_tag(ip);
985 if (iattr->ia_valid & ATTR_MODE)
986 xfs_setattr_mode(ip, iattr);
987 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
988 xfs_setattr_time(ip, iattr);
990 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
992 XFS_STATS_INC(mp, xs_ig_attrchg);
994 if (mp->m_flags & XFS_MOUNT_WSYNC)
995 xfs_trans_set_sync(tp);
997 error = xfs_trans_commit(tp);
1000 xfs_iunlock(ip, lock_flags);
1004 xfs_trans_cancel(tp);
1009 xfs_vn_setattr_size(
1010 struct dentry *dentry,
1011 struct iattr *iattr)
1013 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1016 trace_xfs_setattr(ip);
1018 error = xfs_vn_change_ok(dentry, iattr);
1021 return xfs_setattr_size(ip, iattr);
1026 struct dentry *dentry,
1027 struct iattr *iattr)
1031 if (iattr->ia_valid & ATTR_SIZE) {
1032 struct inode *inode = d_inode(dentry);
1033 struct xfs_inode *ip = XFS_I(inode);
1036 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1037 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1039 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1041 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1045 error = xfs_vn_setattr_size(dentry, iattr);
1046 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1048 error = xfs_vn_setattr_nonsize(dentry, iattr);
1056 struct inode *inode,
1057 struct timespec *now,
1060 struct xfs_inode *ip = XFS_I(inode);
1061 struct xfs_mount *mp = ip->i_mount;
1062 int log_flags = XFS_ILOG_TIMESTAMP;
1063 struct xfs_trans *tp;
1066 trace_xfs_update_time(ip);
1068 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1069 if (!((flags & S_VERSION) &&
1070 inode_maybe_inc_iversion(inode, false)))
1071 return generic_update_time(inode, now, flags);
1073 /* Capture the iversion update that just occurred */
1074 log_flags |= XFS_ILOG_CORE;
1077 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1081 xfs_ilock(ip, XFS_ILOCK_EXCL);
1082 if (flags & S_CTIME)
1083 inode->i_ctime = *now;
1084 if (flags & S_MTIME)
1085 inode->i_mtime = *now;
1086 if (flags & S_ATIME)
1087 inode->i_atime = *now;
1089 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1090 xfs_trans_log_inode(tp, ip, log_flags);
1091 return xfs_trans_commit(tp);
1096 struct inode *inode,
1097 struct fiemap_extent_info *fieinfo,
1103 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1104 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1105 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1106 error = iomap_fiemap(inode, fieinfo, start, length,
1107 &xfs_xattr_iomap_ops);
1109 error = iomap_fiemap(inode, fieinfo, start, length,
1112 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1120 struct dentry *dentry,
1123 return xfs_generic_create(dir, dentry, mode, 0, true);
1126 static const struct inode_operations xfs_inode_operations = {
1127 .get_acl = xfs_get_acl,
1128 .set_acl = xfs_set_acl,
1129 .getattr = xfs_vn_getattr,
1130 .setattr = xfs_vn_setattr,
1131 .listxattr = xfs_vn_listxattr,
1132 .fiemap = xfs_vn_fiemap,
1133 .update_time = xfs_vn_update_time,
1136 static const struct inode_operations xfs_dir_inode_operations = {
1137 .create = xfs_vn_create,
1138 .lookup = xfs_vn_lookup,
1139 .link = xfs_vn_link,
1140 .unlink = xfs_vn_unlink,
1141 .symlink = xfs_vn_symlink,
1142 .mkdir = xfs_vn_mkdir,
1144 * Yes, XFS uses the same method for rmdir and unlink.
1146 * There are some subtile differences deeper in the code,
1147 * but we use S_ISDIR to check for those.
1149 .rmdir = xfs_vn_unlink,
1150 .mknod = xfs_vn_mknod,
1151 .rename = xfs_vn_rename,
1152 .get_acl = xfs_get_acl,
1153 .set_acl = xfs_set_acl,
1154 .getattr = xfs_vn_getattr,
1155 .setattr = xfs_vn_setattr,
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 .listxattr = xfs_vn_listxattr,
1182 .update_time = xfs_vn_update_time,
1183 .tmpfile = xfs_vn_tmpfile,
1186 static const struct inode_operations xfs_symlink_inode_operations = {
1187 .get_link = xfs_vn_get_link,
1188 .getattr = xfs_vn_getattr,
1189 .setattr = xfs_vn_setattr,
1190 .listxattr = xfs_vn_listxattr,
1191 .update_time = xfs_vn_update_time,
1194 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1195 .get_link = xfs_vn_get_link_inline,
1196 .getattr = xfs_vn_getattr,
1197 .setattr = xfs_vn_setattr,
1198 .listxattr = xfs_vn_listxattr,
1199 .update_time = xfs_vn_update_time,
1202 /* Figure out if this file actually supports DAX. */
1204 xfs_inode_supports_dax(
1205 struct xfs_inode *ip)
1207 struct xfs_mount *mp = ip->i_mount;
1209 /* Only supported on non-reflinked files. */
1210 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1213 /* DAX mount option or DAX iflag must be set. */
1214 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1215 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1218 /* Block size must match page size */
1219 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1222 /* Device has to support DAX too. */
1223 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1227 xfs_diflags_to_iflags(
1228 struct inode *inode,
1229 struct xfs_inode *ip)
1231 uint16_t flags = ip->i_d.di_flags;
1233 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1236 if (flags & XFS_DIFLAG_IMMUTABLE)
1237 inode->i_flags |= S_IMMUTABLE;
1238 if (flags & XFS_DIFLAG_APPEND)
1239 inode->i_flags |= S_APPEND;
1240 if (flags & XFS_DIFLAG_SYNC)
1241 inode->i_flags |= S_SYNC;
1242 if (flags & XFS_DIFLAG_NOATIME)
1243 inode->i_flags |= S_NOATIME;
1244 if (xfs_inode_supports_dax(ip))
1245 inode->i_flags |= S_DAX;
1249 * Initialize the Linux inode.
1251 * When reading existing inodes from disk this is called directly from xfs_iget,
1252 * when creating a new inode it is called from xfs_ialloc after setting up the
1253 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1254 * it up to the caller to deal with unlocking the inode appropriately.
1258 struct xfs_inode *ip)
1260 struct inode *inode = &ip->i_vnode;
1263 inode->i_ino = ip->i_ino;
1264 inode->i_state = I_NEW;
1266 inode_sb_list_add(inode);
1267 /* make the inode look hashed for the writeback code */
1268 hlist_add_fake(&inode->i_hash);
1270 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1271 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1273 i_size_write(inode, ip->i_d.di_size);
1274 xfs_diflags_to_iflags(inode, ip);
1276 if (S_ISDIR(inode->i_mode)) {
1277 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1278 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1280 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1281 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1285 * Ensure all page cache allocations are done from GFP_NOFS context to
1286 * prevent direct reclaim recursion back into the filesystem and blowing
1287 * stacks or deadlocking.
1289 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1290 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1293 * If there is no attribute fork no ACL can exist on this inode,
1294 * and it can't have any file capabilities attached to it either.
1296 if (!XFS_IFORK_Q(ip)) {
1297 inode_has_no_xattr(inode);
1298 cache_no_acl(inode);
1304 struct xfs_inode *ip)
1306 struct inode *inode = &ip->i_vnode;
1308 switch (inode->i_mode & S_IFMT) {
1310 inode->i_op = &xfs_inode_operations;
1311 inode->i_fop = &xfs_file_operations;
1313 inode->i_mapping->a_ops = &xfs_dax_aops;
1315 inode->i_mapping->a_ops = &xfs_address_space_operations;
1318 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1319 inode->i_op = &xfs_dir_ci_inode_operations;
1321 inode->i_op = &xfs_dir_inode_operations;
1322 inode->i_fop = &xfs_dir_file_operations;
1325 if (ip->i_df.if_flags & XFS_IFINLINE)
1326 inode->i_op = &xfs_inline_symlink_inode_operations;
1328 inode->i_op = &xfs_symlink_inode_operations;
1331 inode->i_op = &xfs_inode_operations;
1332 init_special_inode(inode, inode->i_mode, inode->i_rdev);