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_inode.h"
15 #include "xfs_quota.h"
17 #include "xfs_trans.h"
18 #include "xfs_trace.h"
19 #include "xfs_icache.h"
20 #include "xfs_symlink.h"
22 #include "xfs_iomap.h"
24 #include <linux/xattr.h>
25 #include <linux/posix_acl.h>
26 #include <linux/security.h>
27 #include <linux/iversion.h>
30 * Directories have different lock order w.r.t. mmap_sem compared to regular
31 * files. This is due to readdir potentially triggering page faults on a user
32 * buffer inside filldir(), and this happens with the ilock on the directory
33 * held. For regular files, the lock order is the other way around - the
34 * mmap_sem is taken during the page fault, and then we lock the ilock to do
35 * block mapping. Hence we need a different class for the directory ilock so
36 * that lockdep can tell them apart.
38 static struct lock_class_key xfs_nondir_ilock_class;
39 static struct lock_class_key xfs_dir_ilock_class;
44 const struct xattr *xattr_array,
47 const struct xattr *xattr;
48 struct xfs_inode *ip = XFS_I(inode);
51 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
52 error = xfs_attr_set(ip, xattr->name, xattr->value,
53 xattr->value_len, ATTR_SECURE);
61 * Hook in SELinux. This is not quite correct yet, what we really need
62 * here (as we do for default ACLs) is a mechanism by which creation of
63 * these attrs can be journalled at inode creation time (along with the
64 * inode, of course, such that log replay can't cause these to be lost).
71 const struct qstr *qstr)
73 return security_inode_init_security(inode, dir, qstr,
74 &xfs_initxattrs, NULL);
79 struct xfs_name *namep,
80 struct dentry *dentry)
82 namep->name = dentry->d_name.name;
83 namep->len = dentry->d_name.len;
84 namep->type = XFS_DIR3_FT_UNKNOWN;
88 xfs_dentry_mode_to_name(
89 struct xfs_name *namep,
90 struct dentry *dentry,
93 namep->name = dentry->d_name.name;
94 namep->len = dentry->d_name.len;
95 namep->type = xfs_mode_to_ftype(mode);
97 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
107 struct dentry *dentry)
109 struct xfs_name teardown;
112 * If we can't add the ACL or we fail in
113 * xfs_init_security we must back out.
114 * ENOSPC can hit here, among other things.
116 xfs_dentry_to_name(&teardown, dentry);
118 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
124 struct dentry *dentry,
127 bool tmpfile) /* unnamed file */
130 struct xfs_inode *ip = NULL;
131 struct posix_acl *default_acl, *acl;
132 struct xfs_name name;
136 * Irix uses Missed'em'V split, but doesn't want to see
137 * the upper 5 bits of (14bit) major.
139 if (S_ISCHR(mode) || S_ISBLK(mode)) {
140 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
146 error = posix_acl_create(dir, &mode, &default_acl, &acl);
150 /* Verify mode is valid also for tmpfile case */
151 error = xfs_dentry_mode_to_name(&name, dentry, mode);
156 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
158 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
165 error = xfs_init_security(inode, dir, &dentry->d_name);
167 goto out_cleanup_inode;
169 #ifdef CONFIG_XFS_POSIX_ACL
171 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
173 goto out_cleanup_inode;
176 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
178 goto out_cleanup_inode;
186 * The VFS requires that any inode fed to d_tmpfile must have
187 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
188 * However, we created the temp file with nlink == 0 because
189 * we're not allowed to put an inode with nlink > 0 on the
190 * unlinked list. Therefore we have to set nlink to 1 so that
191 * d_tmpfile can immediately set it back to zero.
194 d_tmpfile(dentry, inode);
196 d_instantiate(dentry, inode);
198 xfs_finish_inode_setup(ip);
202 posix_acl_release(default_acl);
204 posix_acl_release(acl);
208 xfs_finish_inode_setup(ip);
210 xfs_cleanup_inode(dir, inode, dentry);
218 struct dentry *dentry,
222 return xfs_generic_create(dir, dentry, mode, rdev, false);
228 struct dentry *dentry,
232 return xfs_vn_mknod(dir, dentry, mode, 0);
238 struct dentry *dentry,
241 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
244 STATIC struct dentry *
247 struct dentry *dentry,
251 struct xfs_inode *cip;
252 struct xfs_name name;
255 if (dentry->d_name.len >= MAXNAMELEN)
256 return ERR_PTR(-ENAMETOOLONG);
258 xfs_dentry_to_name(&name, dentry);
259 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
262 else if (likely(error == -ENOENT))
265 inode = ERR_PTR(error);
266 return d_splice_alias(inode, dentry);
269 STATIC struct dentry *
272 struct dentry *dentry,
275 struct xfs_inode *ip;
276 struct xfs_name xname;
277 struct xfs_name ci_name;
281 if (dentry->d_name.len >= MAXNAMELEN)
282 return ERR_PTR(-ENAMETOOLONG);
284 xfs_dentry_to_name(&xname, dentry);
285 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
286 if (unlikely(error)) {
287 if (unlikely(error != -ENOENT))
288 return ERR_PTR(error);
290 * call d_add(dentry, NULL) here when d_drop_negative_children
291 * is called in xfs_vn_mknod (ie. allow negative dentries
292 * with CI filesystems).
297 /* if exact match, just splice and exit */
299 return d_splice_alias(VFS_I(ip), dentry);
301 /* else case-insensitive match... */
302 dname.name = ci_name.name;
303 dname.len = ci_name.len;
304 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
305 kmem_free(ci_name.name);
311 struct dentry *old_dentry,
313 struct dentry *dentry)
315 struct inode *inode = d_inode(old_dentry);
316 struct xfs_name name;
319 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
323 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
328 d_instantiate(dentry, inode);
335 struct dentry *dentry)
337 struct xfs_name name;
340 xfs_dentry_to_name(&name, dentry);
342 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
347 * With unlink, the VFS makes the dentry "negative": no inode,
348 * but still hashed. This is incompatible with case-insensitive
349 * mode, so invalidate (unhash) the dentry in CI-mode.
351 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
352 d_invalidate(dentry);
359 struct dentry *dentry,
363 struct xfs_inode *cip = NULL;
364 struct xfs_name name;
369 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
370 error = xfs_dentry_mode_to_name(&name, dentry, mode);
374 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
380 error = xfs_init_security(inode, dir, &dentry->d_name);
382 goto out_cleanup_inode;
386 d_instantiate(dentry, inode);
387 xfs_finish_inode_setup(cip);
391 xfs_finish_inode_setup(cip);
392 xfs_cleanup_inode(dir, inode, dentry);
401 struct dentry *odentry,
403 struct dentry *ndentry,
406 struct inode *new_inode = d_inode(ndentry);
409 struct xfs_name oname;
410 struct xfs_name nname;
412 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
415 /* if we are exchanging files, we need to set i_mode of both files */
416 if (flags & RENAME_EXCHANGE)
417 omode = d_inode(ndentry)->i_mode;
419 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
420 if (omode && unlikely(error))
423 error = xfs_dentry_mode_to_name(&nname, ndentry,
424 d_inode(odentry)->i_mode);
428 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
430 new_inode ? XFS_I(new_inode) : NULL, flags);
434 * careful here - this function can get called recursively, so
435 * we need to be very careful about how much stack we use.
436 * uio is kmalloced for this reason...
440 struct dentry *dentry,
442 struct delayed_call *done)
448 return ERR_PTR(-ECHILD);
450 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
454 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
458 set_delayed_call(done, kfree_link, link);
464 return ERR_PTR(error);
468 xfs_vn_get_link_inline(
469 struct dentry *dentry,
471 struct delayed_call *done)
475 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
478 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
481 link = XFS_I(inode)->i_df.if_u1.if_data;
483 return ERR_PTR(-EFSCORRUPTED);
489 struct xfs_inode *ip)
491 struct xfs_mount *mp = ip->i_mount;
494 * If the file blocks are being allocated from a realtime volume, then
495 * always return the realtime extent size.
497 if (XFS_IS_REALTIME_INODE(ip))
498 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
501 * Allow large block sizes to be reported to userspace programs if the
502 * "largeio" mount option is used.
504 * If compatibility mode is specified, simply return the basic unit of
505 * caching so that we don't get inefficient read/modify/write I/O from
506 * user apps. Otherwise....
508 * If the underlying volume is a stripe, then return the stripe width in
509 * bytes as the recommended I/O size. It is not a stripe and we've set a
510 * default buffered I/O size, return that, otherwise return the compat
513 if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE)) {
515 return mp->m_swidth << mp->m_sb.sb_blocklog;
516 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
517 return 1U << mp->m_allocsize_log;
525 const struct path *path,
528 unsigned int query_flags)
530 struct inode *inode = d_inode(path->dentry);
531 struct xfs_inode *ip = XFS_I(inode);
532 struct xfs_mount *mp = ip->i_mount;
534 trace_xfs_getattr(ip);
536 if (XFS_FORCED_SHUTDOWN(mp))
539 stat->size = XFS_ISIZE(ip);
540 stat->dev = inode->i_sb->s_dev;
541 stat->mode = inode->i_mode;
542 stat->nlink = inode->i_nlink;
543 stat->uid = inode->i_uid;
544 stat->gid = inode->i_gid;
545 stat->ino = ip->i_ino;
546 stat->atime = inode->i_atime;
547 stat->mtime = inode->i_mtime;
548 stat->ctime = inode->i_ctime;
550 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
552 if (ip->i_d.di_version == 3) {
553 if (request_mask & STATX_BTIME) {
554 stat->result_mask |= STATX_BTIME;
555 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
556 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
561 * Note: If you add another clause to set an attribute flag, please
562 * update attributes_mask below.
564 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
565 stat->attributes |= STATX_ATTR_IMMUTABLE;
566 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
567 stat->attributes |= STATX_ATTR_APPEND;
568 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
569 stat->attributes |= STATX_ATTR_NODUMP;
571 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
575 switch (inode->i_mode & S_IFMT) {
578 stat->blksize = BLKDEV_IOSIZE;
579 stat->rdev = inode->i_rdev;
582 stat->blksize = xfs_stat_blksize(ip);
592 struct xfs_inode *ip,
595 struct inode *inode = VFS_I(ip);
596 umode_t mode = iattr->ia_mode;
598 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
600 inode->i_mode &= S_IFMT;
601 inode->i_mode |= mode & ~S_IFMT;
606 struct xfs_inode *ip,
609 struct inode *inode = VFS_I(ip);
611 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
613 if (iattr->ia_valid & ATTR_ATIME)
614 inode->i_atime = iattr->ia_atime;
615 if (iattr->ia_valid & ATTR_CTIME)
616 inode->i_ctime = iattr->ia_ctime;
617 if (iattr->ia_valid & ATTR_MTIME)
618 inode->i_mtime = iattr->ia_mtime;
623 struct dentry *dentry,
626 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
628 if (mp->m_flags & XFS_MOUNT_RDONLY)
631 if (XFS_FORCED_SHUTDOWN(mp))
634 return setattr_prepare(dentry, iattr);
638 * Set non-size attributes of an inode.
640 * Caution: The caller of this function is responsible for calling
641 * setattr_prepare() or otherwise verifying the change is fine.
645 struct xfs_inode *ip,
649 xfs_mount_t *mp = ip->i_mount;
650 struct inode *inode = VFS_I(ip);
651 int mask = iattr->ia_valid;
654 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
655 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
656 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
657 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
659 ASSERT((mask & ATTR_SIZE) == 0);
662 * If disk quotas is on, we make sure that the dquots do exist on disk,
663 * before we start any other transactions. Trying to do this later
664 * is messy. We don't care to take a readlock to look at the ids
665 * in inode here, because we can't hold it across the trans_reserve.
666 * If the IDs do change before we take the ilock, we're covered
667 * because the i_*dquot fields will get updated anyway.
669 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
672 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
674 qflags |= XFS_QMOPT_UQUOTA;
678 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
680 qflags |= XFS_QMOPT_GQUOTA;
686 * We take a reference when we initialize udqp and gdqp,
687 * so it is important that we never blindly double trip on
688 * the same variable. See xfs_create() for an example.
690 ASSERT(udqp == NULL);
691 ASSERT(gdqp == NULL);
692 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
693 xfs_kgid_to_gid(gid),
695 qflags, &udqp, &gdqp, NULL);
700 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
704 xfs_ilock(ip, XFS_ILOCK_EXCL);
705 xfs_trans_ijoin(tp, ip, 0);
708 * Change file ownership. Must be the owner or privileged.
710 if (mask & (ATTR_UID|ATTR_GID)) {
712 * These IDs could have changed since we last looked at them.
713 * But, we're assured that if the ownership did change
714 * while we didn't have the inode locked, inode's dquot(s)
715 * would have changed also.
719 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
720 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
723 * Do a quota reservation only if uid/gid is actually
726 if (XFS_IS_QUOTA_RUNNING(mp) &&
727 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
728 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
730 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
731 NULL, capable(CAP_FOWNER) ?
732 XFS_QMOPT_FORCE_RES : 0);
733 if (error) /* out of quota */
739 * Change file ownership. Must be the owner or privileged.
741 if (mask & (ATTR_UID|ATTR_GID)) {
743 * CAP_FSETID overrides the following restrictions:
745 * The set-user-ID and set-group-ID bits of a file will be
746 * cleared upon successful return from chown()
748 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
749 !capable(CAP_FSETID))
750 inode->i_mode &= ~(S_ISUID|S_ISGID);
753 * Change the ownerships and register quota modifications
754 * in the transaction.
756 if (!uid_eq(iuid, uid)) {
757 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
758 ASSERT(mask & ATTR_UID);
760 olddquot1 = xfs_qm_vop_chown(tp, ip,
761 &ip->i_udquot, udqp);
763 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
766 if (!gid_eq(igid, gid)) {
767 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
768 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
769 !XFS_IS_PQUOTA_ON(mp));
770 ASSERT(mask & ATTR_GID);
772 olddquot2 = xfs_qm_vop_chown(tp, ip,
773 &ip->i_gdquot, gdqp);
775 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
780 if (mask & ATTR_MODE)
781 xfs_setattr_mode(ip, iattr);
782 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
783 xfs_setattr_time(ip, iattr);
785 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
787 XFS_STATS_INC(mp, xs_ig_attrchg);
789 if (mp->m_flags & XFS_MOUNT_WSYNC)
790 xfs_trans_set_sync(tp);
791 error = xfs_trans_commit(tp);
793 xfs_iunlock(ip, XFS_ILOCK_EXCL);
796 * Release any dquot(s) the inode had kept before chown.
798 xfs_qm_dqrele(olddquot1);
799 xfs_qm_dqrele(olddquot2);
807 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
808 * update. We could avoid this with linked transactions
809 * and passing down the transaction pointer all the way
810 * to attr_set. No previous user of the generic
811 * Posix ACL code seems to care about this issue either.
813 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
814 error = posix_acl_chmod(inode, inode->i_mode);
822 xfs_trans_cancel(tp);
823 xfs_iunlock(ip, XFS_ILOCK_EXCL);
831 xfs_vn_setattr_nonsize(
832 struct dentry *dentry,
835 struct xfs_inode *ip = XFS_I(d_inode(dentry));
838 trace_xfs_setattr(ip);
840 error = xfs_vn_change_ok(dentry, iattr);
843 return xfs_setattr_nonsize(ip, iattr, 0);
847 * Truncate file. Must have write permission and not be a directory.
849 * Caution: The caller of this function is responsible for calling
850 * setattr_prepare() or otherwise verifying the change is fine.
854 struct xfs_inode *ip,
857 struct xfs_mount *mp = ip->i_mount;
858 struct inode *inode = VFS_I(ip);
859 xfs_off_t oldsize, newsize;
860 struct xfs_trans *tp;
863 bool did_zeroing = false;
865 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
866 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
867 ASSERT(S_ISREG(inode->i_mode));
868 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
869 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
871 oldsize = inode->i_size;
872 newsize = iattr->ia_size;
875 * Short circuit the truncate case for zero length files.
877 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
878 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
882 * Use the regular setattr path to update the timestamps.
884 iattr->ia_valid &= ~ATTR_SIZE;
885 return xfs_setattr_nonsize(ip, iattr, 0);
889 * Make sure that the dquots are attached to the inode.
891 error = xfs_qm_dqattach(ip);
896 * Wait for all direct I/O to complete.
898 inode_dio_wait(inode);
901 * File data changes must be complete before we start the transaction to
902 * modify the inode. This needs to be done before joining the inode to
903 * the transaction because the inode cannot be unlocked once it is a
904 * part of the transaction.
906 * Start with zeroing any data beyond EOF that we may expose on file
907 * extension, or zeroing out the rest of the block on a downward
910 if (newsize > oldsize) {
911 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
912 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
913 &did_zeroing, &xfs_buffered_write_iomap_ops);
915 error = iomap_truncate_page(inode, newsize, &did_zeroing,
916 &xfs_buffered_write_iomap_ops);
923 * We've already locked out new page faults, so now we can safely remove
924 * pages from the page cache knowing they won't get refaulted until we
925 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
926 * complete. The truncate_setsize() call also cleans partial EOF page
927 * PTEs on extending truncates and hence ensures sub-page block size
928 * filesystems are correctly handled, too.
930 * We have to do all the page cache truncate work outside the
931 * transaction context as the "lock" order is page lock->log space
932 * reservation as defined by extent allocation in the writeback path.
933 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
934 * having already truncated the in-memory version of the file (i.e. made
935 * user visible changes). There's not much we can do about this, except
936 * to hope that the caller sees ENOMEM and retries the truncate
939 * And we update in-core i_size and truncate page cache beyond newsize
940 * before writeback the [di_size, newsize] range, so we're guaranteed
941 * not to write stale data past the new EOF on truncate down.
943 truncate_setsize(inode, newsize);
946 * We are going to log the inode size change in this transaction so
947 * any previous writes that are beyond the on disk EOF and the new
948 * EOF that have not been written out need to be written here. If we
949 * do not write the data out, we expose ourselves to the null files
950 * problem. Note that this includes any block zeroing we did above;
951 * otherwise those blocks may not be zeroed after a crash.
954 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
955 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
956 ip->i_d.di_size, newsize - 1);
961 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
965 lock_flags |= XFS_ILOCK_EXCL;
966 xfs_ilock(ip, XFS_ILOCK_EXCL);
967 xfs_trans_ijoin(tp, ip, 0);
970 * Only change the c/mtime if we are changing the size or we are
971 * explicitly asked to change it. This handles the semantic difference
972 * between truncate() and ftruncate() as implemented in the VFS.
974 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
975 * special case where we need to update the times despite not having
976 * these flags set. For all other operations the VFS set these flags
977 * explicitly if it wants a timestamp update.
979 if (newsize != oldsize &&
980 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
981 iattr->ia_ctime = iattr->ia_mtime =
983 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
987 * The first thing we do is set the size to new_size permanently on
988 * disk. This way we don't have to worry about anyone ever being able
989 * to look at the data being freed even in the face of a crash.
990 * What we're getting around here is the case where we free a block, it
991 * is allocated to another file, it is written to, and then we crash.
992 * If the new data gets written to the file but the log buffers
993 * containing the free and reallocation don't, then we'd end up with
994 * garbage in the blocks being freed. As long as we make the new size
995 * permanent before actually freeing any blocks it doesn't matter if
996 * they get written to.
998 ip->i_d.di_size = newsize;
999 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1001 if (newsize <= oldsize) {
1002 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1004 goto out_trans_cancel;
1007 * Truncated "down", so we're removing references to old data
1008 * here - if we delay flushing for a long time, we expose
1009 * ourselves unduly to the notorious NULL files problem. So,
1010 * we mark this inode and flush it when the file is closed,
1011 * and do not wait the usual (long) time for writeout.
1013 xfs_iflags_set(ip, XFS_ITRUNCATED);
1015 /* A truncate down always removes post-EOF blocks. */
1016 xfs_inode_clear_eofblocks_tag(ip);
1019 if (iattr->ia_valid & ATTR_MODE)
1020 xfs_setattr_mode(ip, iattr);
1021 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1022 xfs_setattr_time(ip, iattr);
1024 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1026 XFS_STATS_INC(mp, xs_ig_attrchg);
1028 if (mp->m_flags & XFS_MOUNT_WSYNC)
1029 xfs_trans_set_sync(tp);
1031 error = xfs_trans_commit(tp);
1034 xfs_iunlock(ip, lock_flags);
1038 xfs_trans_cancel(tp);
1043 xfs_vn_setattr_size(
1044 struct dentry *dentry,
1045 struct iattr *iattr)
1047 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1050 trace_xfs_setattr(ip);
1052 error = xfs_vn_change_ok(dentry, iattr);
1055 return xfs_setattr_size(ip, iattr);
1060 struct dentry *dentry,
1061 struct iattr *iattr)
1065 if (iattr->ia_valid & ATTR_SIZE) {
1066 struct inode *inode = d_inode(dentry);
1067 struct xfs_inode *ip = XFS_I(inode);
1070 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1071 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1073 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1075 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1079 error = xfs_vn_setattr_size(dentry, iattr);
1080 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1082 error = xfs_vn_setattr_nonsize(dentry, iattr);
1090 struct inode *inode,
1091 struct timespec64 *now,
1094 struct xfs_inode *ip = XFS_I(inode);
1095 struct xfs_mount *mp = ip->i_mount;
1096 int log_flags = XFS_ILOG_TIMESTAMP;
1097 struct xfs_trans *tp;
1100 trace_xfs_update_time(ip);
1102 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1103 if (!((flags & S_VERSION) &&
1104 inode_maybe_inc_iversion(inode, false)))
1105 return generic_update_time(inode, now, flags);
1107 /* Capture the iversion update that just occurred */
1108 log_flags |= XFS_ILOG_CORE;
1111 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1115 xfs_ilock(ip, XFS_ILOCK_EXCL);
1116 if (flags & S_CTIME)
1117 inode->i_ctime = *now;
1118 if (flags & S_MTIME)
1119 inode->i_mtime = *now;
1120 if (flags & S_ATIME)
1121 inode->i_atime = *now;
1123 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1124 xfs_trans_log_inode(tp, ip, log_flags);
1125 return xfs_trans_commit(tp);
1130 struct inode *inode,
1131 struct fiemap_extent_info *fieinfo,
1137 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1138 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1139 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1140 error = iomap_fiemap(inode, fieinfo, start, length,
1141 &xfs_xattr_iomap_ops);
1143 error = iomap_fiemap(inode, fieinfo, start, length,
1144 &xfs_read_iomap_ops);
1146 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1154 struct dentry *dentry,
1157 return xfs_generic_create(dir, dentry, mode, 0, true);
1160 static const struct inode_operations xfs_inode_operations = {
1161 .get_acl = xfs_get_acl,
1162 .set_acl = xfs_set_acl,
1163 .getattr = xfs_vn_getattr,
1164 .setattr = xfs_vn_setattr,
1165 .listxattr = xfs_vn_listxattr,
1166 .fiemap = xfs_vn_fiemap,
1167 .update_time = xfs_vn_update_time,
1170 static const struct inode_operations xfs_dir_inode_operations = {
1171 .create = xfs_vn_create,
1172 .lookup = xfs_vn_lookup,
1173 .link = xfs_vn_link,
1174 .unlink = xfs_vn_unlink,
1175 .symlink = xfs_vn_symlink,
1176 .mkdir = xfs_vn_mkdir,
1178 * Yes, XFS uses the same method for rmdir and unlink.
1180 * There are some subtile differences deeper in the code,
1181 * but we use S_ISDIR to check for those.
1183 .rmdir = xfs_vn_unlink,
1184 .mknod = xfs_vn_mknod,
1185 .rename = xfs_vn_rename,
1186 .get_acl = xfs_get_acl,
1187 .set_acl = xfs_set_acl,
1188 .getattr = xfs_vn_getattr,
1189 .setattr = xfs_vn_setattr,
1190 .listxattr = xfs_vn_listxattr,
1191 .update_time = xfs_vn_update_time,
1192 .tmpfile = xfs_vn_tmpfile,
1195 static const struct inode_operations xfs_dir_ci_inode_operations = {
1196 .create = xfs_vn_create,
1197 .lookup = xfs_vn_ci_lookup,
1198 .link = xfs_vn_link,
1199 .unlink = xfs_vn_unlink,
1200 .symlink = xfs_vn_symlink,
1201 .mkdir = xfs_vn_mkdir,
1203 * Yes, XFS uses the same method for rmdir and unlink.
1205 * There are some subtile differences deeper in the code,
1206 * but we use S_ISDIR to check for those.
1208 .rmdir = xfs_vn_unlink,
1209 .mknod = xfs_vn_mknod,
1210 .rename = xfs_vn_rename,
1211 .get_acl = xfs_get_acl,
1212 .set_acl = xfs_set_acl,
1213 .getattr = xfs_vn_getattr,
1214 .setattr = xfs_vn_setattr,
1215 .listxattr = xfs_vn_listxattr,
1216 .update_time = xfs_vn_update_time,
1217 .tmpfile = xfs_vn_tmpfile,
1220 static const struct inode_operations xfs_symlink_inode_operations = {
1221 .get_link = xfs_vn_get_link,
1222 .getattr = xfs_vn_getattr,
1223 .setattr = xfs_vn_setattr,
1224 .listxattr = xfs_vn_listxattr,
1225 .update_time = xfs_vn_update_time,
1228 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1229 .get_link = xfs_vn_get_link_inline,
1230 .getattr = xfs_vn_getattr,
1231 .setattr = xfs_vn_setattr,
1232 .listxattr = xfs_vn_listxattr,
1233 .update_time = xfs_vn_update_time,
1236 /* Figure out if this file actually supports DAX. */
1238 xfs_inode_supports_dax(
1239 struct xfs_inode *ip)
1241 struct xfs_mount *mp = ip->i_mount;
1243 /* Only supported on non-reflinked files. */
1244 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1247 /* DAX mount option or DAX iflag must be set. */
1248 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1249 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1252 /* Block size must match page size */
1253 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1256 /* Device has to support DAX too. */
1257 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1261 xfs_diflags_to_iflags(
1262 struct inode *inode,
1263 struct xfs_inode *ip)
1265 uint16_t flags = ip->i_d.di_flags;
1267 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1270 if (flags & XFS_DIFLAG_IMMUTABLE)
1271 inode->i_flags |= S_IMMUTABLE;
1272 if (flags & XFS_DIFLAG_APPEND)
1273 inode->i_flags |= S_APPEND;
1274 if (flags & XFS_DIFLAG_SYNC)
1275 inode->i_flags |= S_SYNC;
1276 if (flags & XFS_DIFLAG_NOATIME)
1277 inode->i_flags |= S_NOATIME;
1278 if (xfs_inode_supports_dax(ip))
1279 inode->i_flags |= S_DAX;
1283 * Initialize the Linux inode.
1285 * When reading existing inodes from disk this is called directly from xfs_iget,
1286 * when creating a new inode it is called from xfs_ialloc after setting up the
1287 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1288 * it up to the caller to deal with unlocking the inode appropriately.
1292 struct xfs_inode *ip)
1294 struct inode *inode = &ip->i_vnode;
1297 inode->i_ino = ip->i_ino;
1298 inode->i_state = I_NEW;
1300 inode_sb_list_add(inode);
1301 /* make the inode look hashed for the writeback code */
1302 inode_fake_hash(inode);
1304 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1305 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1307 i_size_write(inode, ip->i_d.di_size);
1308 xfs_diflags_to_iflags(inode, ip);
1310 if (S_ISDIR(inode->i_mode)) {
1312 * We set the i_rwsem class here to avoid potential races with
1313 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1314 * after a filehandle lookup has already found the inode in
1315 * cache before it has been unlocked via unlock_new_inode().
1317 lockdep_set_class(&inode->i_rwsem,
1318 &inode->i_sb->s_type->i_mutex_dir_key);
1319 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1320 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1322 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1323 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1327 * Ensure all page cache allocations are done from GFP_NOFS context to
1328 * prevent direct reclaim recursion back into the filesystem and blowing
1329 * stacks or deadlocking.
1331 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1332 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1335 * If there is no attribute fork no ACL can exist on this inode,
1336 * and it can't have any file capabilities attached to it either.
1338 if (!XFS_IFORK_Q(ip)) {
1339 inode_has_no_xattr(inode);
1340 cache_no_acl(inode);
1346 struct xfs_inode *ip)
1348 struct inode *inode = &ip->i_vnode;
1350 switch (inode->i_mode & S_IFMT) {
1352 inode->i_op = &xfs_inode_operations;
1353 inode->i_fop = &xfs_file_operations;
1355 inode->i_mapping->a_ops = &xfs_dax_aops;
1357 inode->i_mapping->a_ops = &xfs_address_space_operations;
1360 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1361 inode->i_op = &xfs_dir_ci_inode_operations;
1363 inode->i_op = &xfs_dir_inode_operations;
1364 inode->i_fop = &xfs_dir_file_operations;
1367 if (ip->i_df.if_flags & XFS_IFINLINE)
1368 inode->i_op = &xfs_inline_symlink_inode_operations;
1370 inode->i_op = &xfs_symlink_inode_operations;
1373 inode->i_op = &xfs_inode_operations;
1374 init_special_inode(inode, inode->i_mode, inode->i_rdev);