1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_inode.h"
16 #include "xfs_bmap_util.h"
18 #include "xfs_quota.h"
19 #include "xfs_error.h"
21 #include "xfs_trans.h"
22 #include "xfs_trace.h"
23 #include "xfs_icache.h"
24 #include "xfs_symlink.h"
25 #include "xfs_da_btree.h"
27 #include "xfs_trans_space.h"
28 #include "xfs_iomap.h"
29 #include "xfs_defer.h"
31 #include <linux/capability.h>
32 #include <linux/xattr.h>
33 #include <linux/posix_acl.h>
34 #include <linux/security.h>
35 #include <linux/iomap.h>
36 #include <linux/slab.h>
37 #include <linux/iversion.h>
40 * Directories have different lock order w.r.t. mmap_sem compared to regular
41 * files. This is due to readdir potentially triggering page faults on a user
42 * buffer inside filldir(), and this happens with the ilock on the directory
43 * held. For regular files, the lock order is the other way around - the
44 * mmap_sem is taken during the page fault, and then we lock the ilock to do
45 * block mapping. Hence we need a different class for the directory ilock so
46 * that lockdep can tell them apart.
48 static struct lock_class_key xfs_nondir_ilock_class;
49 static struct lock_class_key xfs_dir_ilock_class;
54 const struct xattr *xattr_array,
57 const struct xattr *xattr;
58 struct xfs_inode *ip = XFS_I(inode);
61 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
62 error = xfs_attr_set(ip, xattr->name, xattr->value,
63 xattr->value_len, ATTR_SECURE);
71 * Hook in SELinux. This is not quite correct yet, what we really need
72 * here (as we do for default ACLs) is a mechanism by which creation of
73 * these attrs can be journalled at inode creation time (along with the
74 * inode, of course, such that log replay can't cause these to be lost).
81 const struct qstr *qstr)
83 return security_inode_init_security(inode, dir, qstr,
84 &xfs_initxattrs, NULL);
89 struct xfs_name *namep,
90 struct dentry *dentry)
92 namep->name = dentry->d_name.name;
93 namep->len = dentry->d_name.len;
94 namep->type = XFS_DIR3_FT_UNKNOWN;
98 xfs_dentry_mode_to_name(
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);
107 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
108 return -EFSCORRUPTED;
117 struct dentry *dentry)
119 struct xfs_name teardown;
122 * If we can't add the ACL or we fail in
123 * xfs_init_security we must back out.
124 * ENOSPC can hit here, among other things.
126 xfs_dentry_to_name(&teardown, dentry);
128 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
134 struct dentry *dentry,
137 bool tmpfile) /* unnamed file */
140 struct xfs_inode *ip = NULL;
141 struct posix_acl *default_acl, *acl;
142 struct xfs_name name;
146 * Irix uses Missed'em'V split, but doesn't want to see
147 * the upper 5 bits of (14bit) major.
149 if (S_ISCHR(mode) || S_ISBLK(mode)) {
150 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
156 error = posix_acl_create(dir, &mode, &default_acl, &acl);
160 /* Verify mode is valid also for tmpfile case */
161 error = xfs_dentry_mode_to_name(&name, dentry, mode);
166 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
168 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
175 error = xfs_init_security(inode, dir, &dentry->d_name);
177 goto out_cleanup_inode;
179 #ifdef CONFIG_XFS_POSIX_ACL
181 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
183 goto out_cleanup_inode;
186 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
188 goto out_cleanup_inode;
196 * The VFS requires that any inode fed to d_tmpfile must have
197 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
198 * However, we created the temp file with nlink == 0 because
199 * we're not allowed to put an inode with nlink > 0 on the
200 * unlinked list. Therefore we have to set nlink to 1 so that
201 * d_tmpfile can immediately set it back to zero.
204 d_tmpfile(dentry, inode);
206 d_instantiate(dentry, inode);
208 xfs_finish_inode_setup(ip);
212 posix_acl_release(default_acl);
214 posix_acl_release(acl);
218 xfs_finish_inode_setup(ip);
220 xfs_cleanup_inode(dir, inode, dentry);
228 struct dentry *dentry,
232 return xfs_generic_create(dir, dentry, mode, rdev, false);
238 struct dentry *dentry,
242 return xfs_vn_mknod(dir, dentry, mode, 0);
248 struct dentry *dentry,
251 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
254 STATIC struct dentry *
257 struct dentry *dentry,
261 struct xfs_inode *cip;
262 struct xfs_name name;
265 if (dentry->d_name.len >= MAXNAMELEN)
266 return ERR_PTR(-ENAMETOOLONG);
268 xfs_dentry_to_name(&name, dentry);
269 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
272 else if (likely(error == -ENOENT))
275 inode = ERR_PTR(error);
276 return d_splice_alias(inode, dentry);
279 STATIC struct dentry *
282 struct dentry *dentry,
285 struct xfs_inode *ip;
286 struct xfs_name xname;
287 struct xfs_name ci_name;
291 if (dentry->d_name.len >= MAXNAMELEN)
292 return ERR_PTR(-ENAMETOOLONG);
294 xfs_dentry_to_name(&xname, dentry);
295 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
296 if (unlikely(error)) {
297 if (unlikely(error != -ENOENT))
298 return ERR_PTR(error);
300 * call d_add(dentry, NULL) here when d_drop_negative_children
301 * is called in xfs_vn_mknod (ie. allow negative dentries
302 * with CI filesystems).
307 /* if exact match, just splice and exit */
309 return d_splice_alias(VFS_I(ip), dentry);
311 /* else case-insensitive match... */
312 dname.name = ci_name.name;
313 dname.len = ci_name.len;
314 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
315 kmem_free(ci_name.name);
321 struct dentry *old_dentry,
323 struct dentry *dentry)
325 struct inode *inode = d_inode(old_dentry);
326 struct xfs_name name;
329 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
333 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
338 d_instantiate(dentry, inode);
345 struct dentry *dentry)
347 struct xfs_name name;
350 xfs_dentry_to_name(&name, dentry);
352 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
357 * With unlink, the VFS makes the dentry "negative": no inode,
358 * but still hashed. This is incompatible with case-insensitive
359 * mode, so invalidate (unhash) the dentry in CI-mode.
361 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
362 d_invalidate(dentry);
369 struct dentry *dentry,
373 struct xfs_inode *cip = NULL;
374 struct xfs_name name;
379 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
380 error = xfs_dentry_mode_to_name(&name, dentry, mode);
384 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
390 error = xfs_init_security(inode, dir, &dentry->d_name);
392 goto out_cleanup_inode;
396 d_instantiate(dentry, inode);
397 xfs_finish_inode_setup(cip);
401 xfs_finish_inode_setup(cip);
402 xfs_cleanup_inode(dir, inode, dentry);
411 struct dentry *odentry,
413 struct dentry *ndentry,
416 struct inode *new_inode = d_inode(ndentry);
419 struct xfs_name oname;
420 struct xfs_name nname;
422 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
425 /* if we are exchanging files, we need to set i_mode of both files */
426 if (flags & RENAME_EXCHANGE)
427 omode = d_inode(ndentry)->i_mode;
429 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
430 if (omode && unlikely(error))
433 error = xfs_dentry_mode_to_name(&nname, ndentry,
434 d_inode(odentry)->i_mode);
438 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
440 new_inode ? XFS_I(new_inode) : NULL, flags);
444 * careful here - this function can get called recursively, so
445 * we need to be very careful about how much stack we use.
446 * uio is kmalloced for this reason...
450 struct dentry *dentry,
452 struct delayed_call *done)
458 return ERR_PTR(-ECHILD);
460 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
464 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
468 set_delayed_call(done, kfree_link, link);
474 return ERR_PTR(error);
478 xfs_vn_get_link_inline(
479 struct dentry *dentry,
481 struct delayed_call *done)
485 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
488 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
491 link = XFS_I(inode)->i_df.if_u1.if_data;
493 return ERR_PTR(-EFSCORRUPTED);
499 const struct path *path,
502 unsigned int query_flags)
504 struct inode *inode = d_inode(path->dentry);
505 struct xfs_inode *ip = XFS_I(inode);
506 struct xfs_mount *mp = ip->i_mount;
508 trace_xfs_getattr(ip);
510 if (XFS_FORCED_SHUTDOWN(mp))
513 stat->size = XFS_ISIZE(ip);
514 stat->dev = inode->i_sb->s_dev;
515 stat->mode = inode->i_mode;
516 stat->nlink = inode->i_nlink;
517 stat->uid = inode->i_uid;
518 stat->gid = inode->i_gid;
519 stat->ino = ip->i_ino;
520 stat->atime = inode->i_atime;
521 stat->mtime = inode->i_mtime;
522 stat->ctime = inode->i_ctime;
524 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
526 if (ip->i_d.di_version == 3) {
527 if (request_mask & STATX_BTIME) {
528 stat->result_mask |= STATX_BTIME;
529 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
530 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
534 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
535 stat->attributes |= STATX_ATTR_IMMUTABLE;
536 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
537 stat->attributes |= STATX_ATTR_APPEND;
538 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
539 stat->attributes |= STATX_ATTR_NODUMP;
541 switch (inode->i_mode & S_IFMT) {
544 stat->blksize = BLKDEV_IOSIZE;
545 stat->rdev = inode->i_rdev;
548 if (XFS_IS_REALTIME_INODE(ip)) {
550 * If the file blocks are being allocated from a
551 * realtime volume, then return the inode's realtime
552 * extent size or the realtime volume's extent size.
555 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
557 stat->blksize = xfs_preferred_iosize(mp);
567 struct xfs_inode *ip,
570 struct inode *inode = VFS_I(ip);
571 umode_t mode = iattr->ia_mode;
573 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
575 inode->i_mode &= S_IFMT;
576 inode->i_mode |= mode & ~S_IFMT;
581 struct xfs_inode *ip,
584 struct inode *inode = VFS_I(ip);
586 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
588 if (iattr->ia_valid & ATTR_ATIME)
589 inode->i_atime = iattr->ia_atime;
590 if (iattr->ia_valid & ATTR_CTIME)
591 inode->i_ctime = iattr->ia_ctime;
592 if (iattr->ia_valid & ATTR_MTIME)
593 inode->i_mtime = iattr->ia_mtime;
598 struct dentry *dentry,
601 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
603 if (mp->m_flags & XFS_MOUNT_RDONLY)
606 if (XFS_FORCED_SHUTDOWN(mp))
609 return setattr_prepare(dentry, iattr);
613 * Set non-size attributes of an inode.
615 * Caution: The caller of this function is responsible for calling
616 * setattr_prepare() or otherwise verifying the change is fine.
620 struct xfs_inode *ip,
624 xfs_mount_t *mp = ip->i_mount;
625 struct inode *inode = VFS_I(ip);
626 int mask = iattr->ia_valid;
629 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
630 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
631 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
632 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
634 ASSERT((mask & ATTR_SIZE) == 0);
637 * If disk quotas is on, we make sure that the dquots do exist on disk,
638 * before we start any other transactions. Trying to do this later
639 * is messy. We don't care to take a readlock to look at the ids
640 * in inode here, because we can't hold it across the trans_reserve.
641 * If the IDs do change before we take the ilock, we're covered
642 * because the i_*dquot fields will get updated anyway.
644 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
647 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
649 qflags |= XFS_QMOPT_UQUOTA;
653 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
655 qflags |= XFS_QMOPT_GQUOTA;
661 * We take a reference when we initialize udqp and gdqp,
662 * so it is important that we never blindly double trip on
663 * the same variable. See xfs_create() for an example.
665 ASSERT(udqp == NULL);
666 ASSERT(gdqp == NULL);
667 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
668 xfs_kgid_to_gid(gid),
670 qflags, &udqp, &gdqp, NULL);
675 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
679 xfs_ilock(ip, XFS_ILOCK_EXCL);
680 xfs_trans_ijoin(tp, ip, 0);
683 * Change file ownership. Must be the owner or privileged.
685 if (mask & (ATTR_UID|ATTR_GID)) {
687 * These IDs could have changed since we last looked at them.
688 * But, we're assured that if the ownership did change
689 * while we didn't have the inode locked, inode's dquot(s)
690 * would have changed also.
694 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
695 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
698 * Do a quota reservation only if uid/gid is actually
701 if (XFS_IS_QUOTA_RUNNING(mp) &&
702 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
703 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
705 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
706 NULL, capable(CAP_FOWNER) ?
707 XFS_QMOPT_FORCE_RES : 0);
708 if (error) /* out of quota */
714 * Change file ownership. Must be the owner or privileged.
716 if (mask & (ATTR_UID|ATTR_GID)) {
718 * CAP_FSETID overrides the following restrictions:
720 * The set-user-ID and set-group-ID bits of a file will be
721 * cleared upon successful return from chown()
723 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
724 !capable(CAP_FSETID))
725 inode->i_mode &= ~(S_ISUID|S_ISGID);
728 * Change the ownerships and register quota modifications
729 * in the transaction.
731 if (!uid_eq(iuid, uid)) {
732 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
733 ASSERT(mask & ATTR_UID);
735 olddquot1 = xfs_qm_vop_chown(tp, ip,
736 &ip->i_udquot, udqp);
738 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
741 if (!gid_eq(igid, gid)) {
742 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
743 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
744 !XFS_IS_PQUOTA_ON(mp));
745 ASSERT(mask & ATTR_GID);
747 olddquot2 = xfs_qm_vop_chown(tp, ip,
748 &ip->i_gdquot, gdqp);
750 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
755 if (mask & ATTR_MODE)
756 xfs_setattr_mode(ip, iattr);
757 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
758 xfs_setattr_time(ip, iattr);
760 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
762 XFS_STATS_INC(mp, xs_ig_attrchg);
764 if (mp->m_flags & XFS_MOUNT_WSYNC)
765 xfs_trans_set_sync(tp);
766 error = xfs_trans_commit(tp);
768 xfs_iunlock(ip, XFS_ILOCK_EXCL);
771 * Release any dquot(s) the inode had kept before chown.
773 xfs_qm_dqrele(olddquot1);
774 xfs_qm_dqrele(olddquot2);
782 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
783 * update. We could avoid this with linked transactions
784 * and passing down the transaction pointer all the way
785 * to attr_set. No previous user of the generic
786 * Posix ACL code seems to care about this issue either.
788 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
789 error = posix_acl_chmod(inode, inode->i_mode);
797 xfs_trans_cancel(tp);
805 xfs_vn_setattr_nonsize(
806 struct dentry *dentry,
809 struct xfs_inode *ip = XFS_I(d_inode(dentry));
812 trace_xfs_setattr(ip);
814 error = xfs_vn_change_ok(dentry, iattr);
817 return xfs_setattr_nonsize(ip, iattr, 0);
821 * Truncate file. Must have write permission and not be a directory.
823 * Caution: The caller of this function is responsible for calling
824 * setattr_prepare() or otherwise verifying the change is fine.
828 struct xfs_inode *ip,
831 struct xfs_mount *mp = ip->i_mount;
832 struct inode *inode = VFS_I(ip);
833 xfs_off_t oldsize, newsize;
834 struct xfs_trans *tp;
837 bool did_zeroing = false;
839 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
840 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
841 ASSERT(S_ISREG(inode->i_mode));
842 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
843 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
845 oldsize = inode->i_size;
846 newsize = iattr->ia_size;
849 * Short circuit the truncate case for zero length files.
851 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
852 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
856 * Use the regular setattr path to update the timestamps.
858 iattr->ia_valid &= ~ATTR_SIZE;
859 return xfs_setattr_nonsize(ip, iattr, 0);
863 * Make sure that the dquots are attached to the inode.
865 error = xfs_qm_dqattach(ip);
870 * Wait for all direct I/O to complete.
872 inode_dio_wait(inode);
875 * File data changes must be complete before we start the transaction to
876 * modify the inode. This needs to be done before joining the inode to
877 * the transaction because the inode cannot be unlocked once it is a
878 * part of the transaction.
880 * Start with zeroing any data beyond EOF that we may expose on file
881 * extension, or zeroing out the rest of the block on a downward
884 if (newsize > oldsize) {
885 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
886 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
887 &did_zeroing, &xfs_iomap_ops);
889 error = iomap_truncate_page(inode, newsize, &did_zeroing,
897 * We've already locked out new page faults, so now we can safely remove
898 * pages from the page cache knowing they won't get refaulted until we
899 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
900 * complete. The truncate_setsize() call also cleans partial EOF page
901 * PTEs on extending truncates and hence ensures sub-page block size
902 * filesystems are correctly handled, too.
904 * We have to do all the page cache truncate work outside the
905 * transaction context as the "lock" order is page lock->log space
906 * reservation as defined by extent allocation in the writeback path.
907 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
908 * having already truncated the in-memory version of the file (i.e. made
909 * user visible changes). There's not much we can do about this, except
910 * to hope that the caller sees ENOMEM and retries the truncate
913 * And we update in-core i_size and truncate page cache beyond newsize
914 * before writeback the [di_size, newsize] range, so we're guaranteed
915 * not to write stale data past the new EOF on truncate down.
917 truncate_setsize(inode, newsize);
920 * We are going to log the inode size change in this transaction so
921 * any previous writes that are beyond the on disk EOF and the new
922 * EOF that have not been written out need to be written here. If we
923 * do not write the data out, we expose ourselves to the null files
924 * problem. Note that this includes any block zeroing we did above;
925 * otherwise those blocks may not be zeroed after a crash.
928 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
929 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
930 ip->i_d.di_size, newsize - 1);
935 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
939 lock_flags |= XFS_ILOCK_EXCL;
940 xfs_ilock(ip, XFS_ILOCK_EXCL);
941 xfs_trans_ijoin(tp, ip, 0);
944 * Only change the c/mtime if we are changing the size or we are
945 * explicitly asked to change it. This handles the semantic difference
946 * between truncate() and ftruncate() as implemented in the VFS.
948 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
949 * special case where we need to update the times despite not having
950 * these flags set. For all other operations the VFS set these flags
951 * explicitly if it wants a timestamp update.
953 if (newsize != oldsize &&
954 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
955 iattr->ia_ctime = iattr->ia_mtime =
957 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
961 * The first thing we do is set the size to new_size permanently on
962 * disk. This way we don't have to worry about anyone ever being able
963 * to look at the data being freed even in the face of a crash.
964 * What we're getting around here is the case where we free a block, it
965 * is allocated to another file, it is written to, and then we crash.
966 * If the new data gets written to the file but the log buffers
967 * containing the free and reallocation don't, then we'd end up with
968 * garbage in the blocks being freed. As long as we make the new size
969 * permanent before actually freeing any blocks it doesn't matter if
970 * they get written to.
972 ip->i_d.di_size = newsize;
973 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
975 if (newsize <= oldsize) {
976 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
978 goto out_trans_cancel;
981 * Truncated "down", so we're removing references to old data
982 * here - if we delay flushing for a long time, we expose
983 * ourselves unduly to the notorious NULL files problem. So,
984 * we mark this inode and flush it when the file is closed,
985 * and do not wait the usual (long) time for writeout.
987 xfs_iflags_set(ip, XFS_ITRUNCATED);
989 /* A truncate down always removes post-EOF blocks. */
990 xfs_inode_clear_eofblocks_tag(ip);
993 if (iattr->ia_valid & ATTR_MODE)
994 xfs_setattr_mode(ip, iattr);
995 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
996 xfs_setattr_time(ip, iattr);
998 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1000 XFS_STATS_INC(mp, xs_ig_attrchg);
1002 if (mp->m_flags & XFS_MOUNT_WSYNC)
1003 xfs_trans_set_sync(tp);
1005 error = xfs_trans_commit(tp);
1008 xfs_iunlock(ip, lock_flags);
1012 xfs_trans_cancel(tp);
1017 xfs_vn_setattr_size(
1018 struct dentry *dentry,
1019 struct iattr *iattr)
1021 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1024 trace_xfs_setattr(ip);
1026 error = xfs_vn_change_ok(dentry, iattr);
1029 return xfs_setattr_size(ip, iattr);
1034 struct dentry *dentry,
1035 struct iattr *iattr)
1039 if (iattr->ia_valid & ATTR_SIZE) {
1040 struct inode *inode = d_inode(dentry);
1041 struct xfs_inode *ip = XFS_I(inode);
1044 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1045 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1047 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1049 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1053 error = xfs_vn_setattr_size(dentry, iattr);
1054 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1056 error = xfs_vn_setattr_nonsize(dentry, iattr);
1064 struct inode *inode,
1065 struct timespec64 *now,
1068 struct xfs_inode *ip = XFS_I(inode);
1069 struct xfs_mount *mp = ip->i_mount;
1070 int log_flags = XFS_ILOG_TIMESTAMP;
1071 struct xfs_trans *tp;
1074 trace_xfs_update_time(ip);
1076 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1077 if (!((flags & S_VERSION) &&
1078 inode_maybe_inc_iversion(inode, false)))
1079 return generic_update_time(inode, now, flags);
1081 /* Capture the iversion update that just occurred */
1082 log_flags |= XFS_ILOG_CORE;
1085 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1089 xfs_ilock(ip, XFS_ILOCK_EXCL);
1090 if (flags & S_CTIME)
1091 inode->i_ctime = *now;
1092 if (flags & S_MTIME)
1093 inode->i_mtime = *now;
1094 if (flags & S_ATIME)
1095 inode->i_atime = *now;
1097 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1098 xfs_trans_log_inode(tp, ip, log_flags);
1099 return xfs_trans_commit(tp);
1104 struct inode *inode,
1105 struct fiemap_extent_info *fieinfo,
1111 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1112 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1113 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1114 error = iomap_fiemap(inode, fieinfo, start, length,
1115 &xfs_xattr_iomap_ops);
1117 error = iomap_fiemap(inode, fieinfo, start, length,
1120 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1128 struct dentry *dentry,
1131 return xfs_generic_create(dir, dentry, mode, 0, true);
1134 static const struct inode_operations xfs_inode_operations = {
1135 .get_acl = xfs_get_acl,
1136 .set_acl = xfs_set_acl,
1137 .getattr = xfs_vn_getattr,
1138 .setattr = xfs_vn_setattr,
1139 .listxattr = xfs_vn_listxattr,
1140 .fiemap = xfs_vn_fiemap,
1141 .update_time = xfs_vn_update_time,
1144 static const struct inode_operations xfs_dir_inode_operations = {
1145 .create = xfs_vn_create,
1146 .lookup = xfs_vn_lookup,
1147 .link = xfs_vn_link,
1148 .unlink = xfs_vn_unlink,
1149 .symlink = xfs_vn_symlink,
1150 .mkdir = xfs_vn_mkdir,
1152 * Yes, XFS uses the same method for rmdir and unlink.
1154 * There are some subtile differences deeper in the code,
1155 * but we use S_ISDIR to check for those.
1157 .rmdir = xfs_vn_unlink,
1158 .mknod = xfs_vn_mknod,
1159 .rename = xfs_vn_rename,
1160 .get_acl = xfs_get_acl,
1161 .set_acl = xfs_set_acl,
1162 .getattr = xfs_vn_getattr,
1163 .setattr = xfs_vn_setattr,
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 .rename = 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 .listxattr = xfs_vn_listxattr,
1190 .update_time = xfs_vn_update_time,
1191 .tmpfile = xfs_vn_tmpfile,
1194 static const struct inode_operations xfs_symlink_inode_operations = {
1195 .get_link = xfs_vn_get_link,
1196 .getattr = xfs_vn_getattr,
1197 .setattr = xfs_vn_setattr,
1198 .listxattr = xfs_vn_listxattr,
1199 .update_time = xfs_vn_update_time,
1202 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1203 .get_link = xfs_vn_get_link_inline,
1204 .getattr = xfs_vn_getattr,
1205 .setattr = xfs_vn_setattr,
1206 .listxattr = xfs_vn_listxattr,
1207 .update_time = xfs_vn_update_time,
1210 /* Figure out if this file actually supports DAX. */
1212 xfs_inode_supports_dax(
1213 struct xfs_inode *ip)
1215 struct xfs_mount *mp = ip->i_mount;
1217 /* Only supported on non-reflinked files. */
1218 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1221 /* DAX mount option or DAX iflag must be set. */
1222 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1223 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1226 /* Block size must match page size */
1227 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1230 /* Device has to support DAX too. */
1231 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1235 xfs_diflags_to_iflags(
1236 struct inode *inode,
1237 struct xfs_inode *ip)
1239 uint16_t flags = ip->i_d.di_flags;
1241 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1244 if (flags & XFS_DIFLAG_IMMUTABLE)
1245 inode->i_flags |= S_IMMUTABLE;
1246 if (flags & XFS_DIFLAG_APPEND)
1247 inode->i_flags |= S_APPEND;
1248 if (flags & XFS_DIFLAG_SYNC)
1249 inode->i_flags |= S_SYNC;
1250 if (flags & XFS_DIFLAG_NOATIME)
1251 inode->i_flags |= S_NOATIME;
1252 if (xfs_inode_supports_dax(ip))
1253 inode->i_flags |= S_DAX;
1257 * Initialize the Linux inode.
1259 * When reading existing inodes from disk this is called directly from xfs_iget,
1260 * when creating a new inode it is called from xfs_ialloc after setting up the
1261 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1262 * it up to the caller to deal with unlocking the inode appropriately.
1266 struct xfs_inode *ip)
1268 struct inode *inode = &ip->i_vnode;
1271 inode->i_ino = ip->i_ino;
1272 inode->i_state = I_NEW;
1274 inode_sb_list_add(inode);
1275 /* make the inode look hashed for the writeback code */
1276 inode_fake_hash(inode);
1278 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1279 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1281 i_size_write(inode, ip->i_d.di_size);
1282 xfs_diflags_to_iflags(inode, ip);
1284 if (S_ISDIR(inode->i_mode)) {
1286 * We set the i_rwsem class here to avoid potential races with
1287 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1288 * after a filehandle lookup has already found the inode in
1289 * cache before it has been unlocked via unlock_new_inode().
1291 lockdep_set_class(&inode->i_rwsem,
1292 &inode->i_sb->s_type->i_mutex_dir_key);
1293 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1294 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1296 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1297 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1301 * Ensure all page cache allocations are done from GFP_NOFS context to
1302 * prevent direct reclaim recursion back into the filesystem and blowing
1303 * stacks or deadlocking.
1305 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1306 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1309 * If there is no attribute fork no ACL can exist on this inode,
1310 * and it can't have any file capabilities attached to it either.
1312 if (!XFS_IFORK_Q(ip)) {
1313 inode_has_no_xattr(inode);
1314 cache_no_acl(inode);
1320 struct xfs_inode *ip)
1322 struct inode *inode = &ip->i_vnode;
1324 switch (inode->i_mode & S_IFMT) {
1326 inode->i_op = &xfs_inode_operations;
1327 inode->i_fop = &xfs_file_operations;
1329 inode->i_mapping->a_ops = &xfs_dax_aops;
1331 inode->i_mapping->a_ops = &xfs_address_space_operations;
1334 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1335 inode->i_op = &xfs_dir_ci_inode_operations;
1337 inode->i_op = &xfs_dir_inode_operations;
1338 inode->i_fop = &xfs_dir_file_operations;
1341 if (ip->i_df.if_flags & XFS_IFINLINE)
1342 inode->i_op = &xfs_inline_symlink_inode_operations;
1344 inode->i_op = &xfs_symlink_inode_operations;
1347 inode->i_op = &xfs_inode_operations;
1348 init_special_inode(inode, inode->i_mode, inode->i_rdev);