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"
23 #include "xfs_error.h"
25 #include <linux/posix_acl.h>
26 #include <linux/security.h>
27 #include <linux/iversion.h>
28 #include <linux/fiemap.h>
31 * Directories have different lock order w.r.t. mmap_lock compared to regular
32 * files. This is due to readdir potentially triggering page faults on a user
33 * buffer inside filldir(), and this happens with the ilock on the directory
34 * held. For regular files, the lock order is the other way around - the
35 * mmap_lock is taken during the page fault, and then we lock the ilock to do
36 * block mapping. Hence we need a different class for the directory ilock so
37 * that lockdep can tell them apart.
39 static struct lock_class_key xfs_nondir_ilock_class;
40 static struct lock_class_key xfs_dir_ilock_class;
45 const struct xattr *xattr_array,
48 const struct xattr *xattr;
49 struct xfs_inode *ip = XFS_I(inode);
52 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
53 struct xfs_da_args args = {
55 .attr_filter = XFS_ATTR_SECURE,
57 .namelen = strlen(xattr->name),
58 .value = xattr->value,
59 .valuelen = xattr->value_len,
61 error = xfs_attr_set(&args);
69 * Hook in SELinux. This is not quite correct yet, what we really need
70 * here (as we do for default ACLs) is a mechanism by which creation of
71 * these attrs can be journalled at inode creation time (along with the
72 * inode, of course, such that log replay can't cause these to be lost).
79 const struct qstr *qstr)
81 return security_inode_init_security(inode, dir, qstr,
82 &xfs_initxattrs, NULL);
87 struct xfs_name *namep,
88 struct dentry *dentry)
90 namep->name = dentry->d_name.name;
91 namep->len = dentry->d_name.len;
92 namep->type = XFS_DIR3_FT_UNKNOWN;
96 xfs_dentry_mode_to_name(
97 struct xfs_name *namep,
98 struct dentry *dentry,
101 namep->name = dentry->d_name.name;
102 namep->len = dentry->d_name.len;
103 namep->type = xfs_mode_to_ftype(mode);
105 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
106 return -EFSCORRUPTED;
115 struct dentry *dentry)
117 struct xfs_name teardown;
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
124 xfs_dentry_to_name(&teardown, dentry);
126 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
131 struct user_namespace *mnt_userns,
133 struct dentry *dentry,
136 bool tmpfile) /* unnamed file */
139 struct xfs_inode *ip = NULL;
140 struct posix_acl *default_acl, *acl;
141 struct xfs_name name;
145 * Irix uses Missed'em'V split, but doesn't want to see
146 * the upper 5 bits of (14bit) major.
148 if (S_ISCHR(mode) || S_ISBLK(mode)) {
149 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
155 error = posix_acl_create(dir, &mode, &default_acl, &acl);
159 /* Verify mode is valid also for tmpfile case */
160 error = xfs_dentry_mode_to_name(&name, dentry, mode);
165 error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev,
168 error = xfs_create_tmpfile(mnt_userns, 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);
211 posix_acl_release(default_acl);
212 posix_acl_release(acl);
216 xfs_finish_inode_setup(ip);
218 xfs_cleanup_inode(dir, inode, dentry);
225 struct user_namespace *mnt_userns,
227 struct dentry *dentry,
231 return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false);
236 struct user_namespace *mnt_userns,
238 struct dentry *dentry,
242 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false);
247 struct user_namespace *mnt_userns,
249 struct dentry *dentry,
252 return xfs_generic_create(mnt_userns, 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);
370 struct user_namespace *mnt_userns,
372 struct dentry *dentry,
376 struct xfs_inode *cip = NULL;
377 struct xfs_name name;
382 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
383 error = xfs_dentry_mode_to_name(&name, dentry, mode);
387 error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip);
393 error = xfs_init_security(inode, dir, &dentry->d_name);
395 goto out_cleanup_inode;
399 d_instantiate(dentry, inode);
400 xfs_finish_inode_setup(cip);
404 xfs_finish_inode_setup(cip);
405 xfs_cleanup_inode(dir, inode, dentry);
413 struct user_namespace *mnt_userns,
415 struct dentry *odentry,
417 struct dentry *ndentry,
420 struct inode *new_inode = d_inode(ndentry);
423 struct xfs_name oname;
424 struct xfs_name nname;
426 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
429 /* if we are exchanging files, we need to set i_mode of both files */
430 if (flags & RENAME_EXCHANGE)
431 omode = d_inode(ndentry)->i_mode;
433 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
434 if (omode && unlikely(error))
437 error = xfs_dentry_mode_to_name(&nname, ndentry,
438 d_inode(odentry)->i_mode);
442 return xfs_rename(mnt_userns, XFS_I(odir), &oname,
443 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
444 new_inode ? XFS_I(new_inode) : NULL, flags);
448 * careful here - this function can get called recursively, so
449 * we need to be very careful about how much stack we use.
450 * uio is kmalloced for this reason...
454 struct dentry *dentry,
456 struct delayed_call *done)
462 return ERR_PTR(-ECHILD);
464 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
468 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
472 set_delayed_call(done, kfree_link, link);
478 return ERR_PTR(error);
482 xfs_vn_get_link_inline(
483 struct dentry *dentry,
485 struct delayed_call *done)
487 struct xfs_inode *ip = XFS_I(inode);
490 ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
493 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
496 link = ip->i_df.if_u1.if_data;
497 if (XFS_IS_CORRUPT(ip->i_mount, !link))
498 return ERR_PTR(-EFSCORRUPTED);
504 struct xfs_inode *ip)
506 struct xfs_mount *mp = ip->i_mount;
509 * If the file blocks are being allocated from a realtime volume, then
510 * always return the realtime extent size.
512 if (XFS_IS_REALTIME_INODE(ip))
513 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
516 * Allow large block sizes to be reported to userspace programs if the
517 * "largeio" mount option is used.
519 * If compatibility mode is specified, simply return the basic unit of
520 * caching so that we don't get inefficient read/modify/write I/O from
521 * user apps. Otherwise....
523 * If the underlying volume is a stripe, then return the stripe width in
524 * bytes as the recommended I/O size. It is not a stripe and we've set a
525 * default buffered I/O size, return that, otherwise return the compat
528 if (mp->m_flags & XFS_MOUNT_LARGEIO) {
530 return mp->m_swidth << mp->m_sb.sb_blocklog;
531 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
532 return 1U << mp->m_allocsize_log;
540 struct user_namespace *mnt_userns,
541 const struct path *path,
544 unsigned int query_flags)
546 struct inode *inode = d_inode(path->dentry);
547 struct xfs_inode *ip = XFS_I(inode);
548 struct xfs_mount *mp = ip->i_mount;
550 trace_xfs_getattr(ip);
552 if (XFS_FORCED_SHUTDOWN(mp))
555 stat->size = XFS_ISIZE(ip);
556 stat->dev = inode->i_sb->s_dev;
557 stat->mode = inode->i_mode;
558 stat->nlink = inode->i_nlink;
559 stat->uid = i_uid_into_mnt(mnt_userns, inode);
560 stat->gid = i_gid_into_mnt(mnt_userns, inode);
561 stat->ino = ip->i_ino;
562 stat->atime = inode->i_atime;
563 stat->mtime = inode->i_mtime;
564 stat->ctime = inode->i_ctime;
566 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
568 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
569 if (request_mask & STATX_BTIME) {
570 stat->result_mask |= STATX_BTIME;
571 stat->btime = ip->i_d.di_crtime;
576 * Note: If you add another clause to set an attribute flag, please
577 * update attributes_mask below.
579 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
580 stat->attributes |= STATX_ATTR_IMMUTABLE;
581 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
582 stat->attributes |= STATX_ATTR_APPEND;
583 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
584 stat->attributes |= STATX_ATTR_NODUMP;
586 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
590 switch (inode->i_mode & S_IFMT) {
593 stat->blksize = BLKDEV_IOSIZE;
594 stat->rdev = inode->i_rdev;
597 stat->blksize = xfs_stat_blksize(ip);
607 struct xfs_inode *ip,
610 struct inode *inode = VFS_I(ip);
611 umode_t mode = iattr->ia_mode;
613 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
615 inode->i_mode &= S_IFMT;
616 inode->i_mode |= mode & ~S_IFMT;
621 struct xfs_inode *ip,
624 struct inode *inode = VFS_I(ip);
626 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
628 if (iattr->ia_valid & ATTR_ATIME)
629 inode->i_atime = iattr->ia_atime;
630 if (iattr->ia_valid & ATTR_CTIME)
631 inode->i_ctime = iattr->ia_ctime;
632 if (iattr->ia_valid & ATTR_MTIME)
633 inode->i_mtime = iattr->ia_mtime;
638 struct user_namespace *mnt_userns,
639 struct dentry *dentry,
642 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
644 if (mp->m_flags & XFS_MOUNT_RDONLY)
647 if (XFS_FORCED_SHUTDOWN(mp))
650 return setattr_prepare(mnt_userns, dentry, iattr);
654 * Set non-size attributes of an inode.
656 * Caution: The caller of this function is responsible for calling
657 * setattr_prepare() or otherwise verifying the change is fine.
661 struct user_namespace *mnt_userns,
662 struct xfs_inode *ip,
665 xfs_mount_t *mp = ip->i_mount;
666 struct inode *inode = VFS_I(ip);
667 int mask = iattr->ia_valid;
670 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
671 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
672 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
673 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
675 ASSERT((mask & ATTR_SIZE) == 0);
678 * If disk quotas is on, we make sure that the dquots do exist on disk,
679 * before we start any other transactions. Trying to do this later
680 * is messy. We don't care to take a readlock to look at the ids
681 * in inode here, because we can't hold it across the trans_reserve.
682 * If the IDs do change before we take the ilock, we're covered
683 * because the i_*dquot fields will get updated anyway.
685 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
688 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
690 qflags |= XFS_QMOPT_UQUOTA;
694 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
696 qflags |= XFS_QMOPT_GQUOTA;
702 * We take a reference when we initialize udqp and gdqp,
703 * so it is important that we never blindly double trip on
704 * the same variable. See xfs_create() for an example.
706 ASSERT(udqp == NULL);
707 ASSERT(gdqp == NULL);
708 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
709 qflags, &udqp, &gdqp, NULL);
714 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
715 capable(CAP_FOWNER), &tp);
720 * Change file ownership. Must be the owner or privileged.
722 if (mask & (ATTR_UID|ATTR_GID)) {
724 * These IDs could have changed since we last looked at them.
725 * But, we're assured that if the ownership did change
726 * while we didn't have the inode locked, inode's dquot(s)
727 * would have changed also.
731 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
732 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
735 * CAP_FSETID overrides the following restrictions:
737 * The set-user-ID and set-group-ID bits of a file will be
738 * cleared upon successful return from chown()
740 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
741 !capable(CAP_FSETID))
742 inode->i_mode &= ~(S_ISUID|S_ISGID);
745 * Change the ownerships and register quota modifications
746 * in the transaction.
748 if (!uid_eq(iuid, uid)) {
749 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
750 ASSERT(mask & ATTR_UID);
752 olddquot1 = xfs_qm_vop_chown(tp, ip,
753 &ip->i_udquot, udqp);
757 if (!gid_eq(igid, gid)) {
758 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
759 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
760 !XFS_IS_PQUOTA_ON(mp));
761 ASSERT(mask & ATTR_GID);
763 olddquot2 = xfs_qm_vop_chown(tp, ip,
764 &ip->i_gdquot, gdqp);
770 if (mask & ATTR_MODE)
771 xfs_setattr_mode(ip, iattr);
772 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
773 xfs_setattr_time(ip, iattr);
775 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
777 XFS_STATS_INC(mp, xs_ig_attrchg);
779 if (mp->m_flags & XFS_MOUNT_WSYNC)
780 xfs_trans_set_sync(tp);
781 error = xfs_trans_commit(tp);
784 * Release any dquot(s) the inode had kept before chown.
786 xfs_qm_dqrele(olddquot1);
787 xfs_qm_dqrele(olddquot2);
795 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
796 * update. We could avoid this with linked transactions
797 * and passing down the transaction pointer all the way
798 * to attr_set. No previous user of the generic
799 * Posix ACL code seems to care about this issue either.
801 if (mask & ATTR_MODE) {
802 error = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
816 * Truncate file. Must have write permission and not be a directory.
818 * Caution: The caller of this function is responsible for calling
819 * setattr_prepare() or otherwise verifying the change is fine.
823 struct user_namespace *mnt_userns,
824 struct xfs_inode *ip,
827 struct xfs_mount *mp = ip->i_mount;
828 struct inode *inode = VFS_I(ip);
829 xfs_off_t oldsize, newsize;
830 struct xfs_trans *tp;
833 bool did_zeroing = false;
835 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
836 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
837 ASSERT(S_ISREG(inode->i_mode));
838 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
839 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
841 oldsize = inode->i_size;
842 newsize = iattr->ia_size;
845 * Short circuit the truncate case for zero length files.
847 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
848 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
852 * Use the regular setattr path to update the timestamps.
854 iattr->ia_valid &= ~ATTR_SIZE;
855 return xfs_setattr_nonsize(mnt_userns, ip, iattr);
859 * Make sure that the dquots are attached to the inode.
861 error = xfs_qm_dqattach(ip);
866 * Wait for all direct I/O to complete.
868 inode_dio_wait(inode);
871 * File data changes must be complete before we start the transaction to
872 * modify the inode. This needs to be done before joining the inode to
873 * the transaction because the inode cannot be unlocked once it is a
874 * part of the transaction.
876 * Start with zeroing any data beyond EOF that we may expose on file
877 * extension, or zeroing out the rest of the block on a downward
880 if (newsize > oldsize) {
881 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
882 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
883 &did_zeroing, &xfs_buffered_write_iomap_ops);
886 * iomap won't detect a dirty page over an unwritten block (or a
887 * cow block over a hole) and subsequently skips zeroing the
888 * newly post-EOF portion of the page. Flush the new EOF to
889 * convert the block before the pagecache truncate.
891 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
895 error = iomap_truncate_page(inode, newsize, &did_zeroing,
896 &xfs_buffered_write_iomap_ops);
903 * We've already locked out new page faults, so now we can safely remove
904 * pages from the page cache knowing they won't get refaulted until we
905 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
906 * complete. The truncate_setsize() call also cleans partial EOF page
907 * PTEs on extending truncates and hence ensures sub-page block size
908 * filesystems are correctly handled, too.
910 * We have to do all the page cache truncate work outside the
911 * transaction context as the "lock" order is page lock->log space
912 * reservation as defined by extent allocation in the writeback path.
913 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
914 * having already truncated the in-memory version of the file (i.e. made
915 * user visible changes). There's not much we can do about this, except
916 * to hope that the caller sees ENOMEM and retries the truncate
919 * And we update in-core i_size and truncate page cache beyond newsize
920 * before writeback the [di_size, newsize] range, so we're guaranteed
921 * not to write stale data past the new EOF on truncate down.
923 truncate_setsize(inode, newsize);
926 * We are going to log the inode size change in this transaction so
927 * any previous writes that are beyond the on disk EOF and the new
928 * EOF that have not been written out need to be written here. If we
929 * do not write the data out, we expose ourselves to the null files
930 * problem. Note that this includes any block zeroing we did above;
931 * otherwise those blocks may not be zeroed after a crash.
934 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
935 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
936 ip->i_d.di_size, newsize - 1);
941 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
945 lock_flags |= XFS_ILOCK_EXCL;
946 xfs_ilock(ip, XFS_ILOCK_EXCL);
947 xfs_trans_ijoin(tp, ip, 0);
950 * Only change the c/mtime if we are changing the size or we are
951 * explicitly asked to change it. This handles the semantic difference
952 * between truncate() and ftruncate() as implemented in the VFS.
954 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
955 * special case where we need to update the times despite not having
956 * these flags set. For all other operations the VFS set these flags
957 * explicitly if it wants a timestamp update.
959 if (newsize != oldsize &&
960 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
961 iattr->ia_ctime = iattr->ia_mtime =
963 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
967 * The first thing we do is set the size to new_size permanently on
968 * disk. This way we don't have to worry about anyone ever being able
969 * to look at the data being freed even in the face of a crash.
970 * What we're getting around here is the case where we free a block, it
971 * is allocated to another file, it is written to, and then we crash.
972 * If the new data gets written to the file but the log buffers
973 * containing the free and reallocation don't, then we'd end up with
974 * garbage in the blocks being freed. As long as we make the new size
975 * permanent before actually freeing any blocks it doesn't matter if
976 * they get written to.
978 ip->i_d.di_size = newsize;
979 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
981 if (newsize <= oldsize) {
982 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
984 goto out_trans_cancel;
987 * Truncated "down", so we're removing references to old data
988 * here - if we delay flushing for a long time, we expose
989 * ourselves unduly to the notorious NULL files problem. So,
990 * we mark this inode and flush it when the file is closed,
991 * and do not wait the usual (long) time for writeout.
993 xfs_iflags_set(ip, XFS_ITRUNCATED);
995 /* A truncate down always removes post-EOF blocks. */
996 xfs_inode_clear_eofblocks_tag(ip);
999 if (iattr->ia_valid & ATTR_MODE)
1000 xfs_setattr_mode(ip, iattr);
1001 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1002 xfs_setattr_time(ip, iattr);
1004 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1006 XFS_STATS_INC(mp, xs_ig_attrchg);
1008 if (mp->m_flags & XFS_MOUNT_WSYNC)
1009 xfs_trans_set_sync(tp);
1011 error = xfs_trans_commit(tp);
1014 xfs_iunlock(ip, lock_flags);
1018 xfs_trans_cancel(tp);
1023 xfs_vn_setattr_size(
1024 struct user_namespace *mnt_userns,
1025 struct dentry *dentry,
1026 struct iattr *iattr)
1028 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1031 trace_xfs_setattr(ip);
1033 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1036 return xfs_setattr_size(mnt_userns, ip, iattr);
1041 struct user_namespace *mnt_userns,
1042 struct dentry *dentry,
1043 struct iattr *iattr)
1045 struct inode *inode = d_inode(dentry);
1046 struct xfs_inode *ip = XFS_I(inode);
1049 if (iattr->ia_valid & ATTR_SIZE) {
1052 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1053 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1055 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1057 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1061 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr);
1062 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1064 trace_xfs_setattr(ip);
1066 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1068 error = xfs_setattr_nonsize(mnt_userns, ip, iattr);
1076 struct inode *inode,
1077 struct timespec64 *now,
1080 struct xfs_inode *ip = XFS_I(inode);
1081 struct xfs_mount *mp = ip->i_mount;
1082 int log_flags = XFS_ILOG_TIMESTAMP;
1083 struct xfs_trans *tp;
1086 trace_xfs_update_time(ip);
1088 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1089 if (!((flags & S_VERSION) &&
1090 inode_maybe_inc_iversion(inode, false)))
1091 return generic_update_time(inode, now, flags);
1093 /* Capture the iversion update that just occurred */
1094 log_flags |= XFS_ILOG_CORE;
1097 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1101 xfs_ilock(ip, XFS_ILOCK_EXCL);
1102 if (flags & S_CTIME)
1103 inode->i_ctime = *now;
1104 if (flags & S_MTIME)
1105 inode->i_mtime = *now;
1106 if (flags & S_ATIME)
1107 inode->i_atime = *now;
1109 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1110 xfs_trans_log_inode(tp, ip, log_flags);
1111 return xfs_trans_commit(tp);
1116 struct inode *inode,
1117 struct fiemap_extent_info *fieinfo,
1123 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1124 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1125 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1126 error = iomap_fiemap(inode, fieinfo, start, length,
1127 &xfs_xattr_iomap_ops);
1129 error = iomap_fiemap(inode, fieinfo, start, length,
1130 &xfs_read_iomap_ops);
1132 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1139 struct user_namespace *mnt_userns,
1141 struct dentry *dentry,
1144 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true);
1147 static const struct inode_operations xfs_inode_operations = {
1148 .get_acl = xfs_get_acl,
1149 .set_acl = xfs_set_acl,
1150 .getattr = xfs_vn_getattr,
1151 .setattr = xfs_vn_setattr,
1152 .listxattr = xfs_vn_listxattr,
1153 .fiemap = xfs_vn_fiemap,
1154 .update_time = xfs_vn_update_time,
1157 static const struct inode_operations xfs_dir_inode_operations = {
1158 .create = xfs_vn_create,
1159 .lookup = xfs_vn_lookup,
1160 .link = xfs_vn_link,
1161 .unlink = xfs_vn_unlink,
1162 .symlink = xfs_vn_symlink,
1163 .mkdir = xfs_vn_mkdir,
1165 * Yes, XFS uses the same method for rmdir and unlink.
1167 * There are some subtile differences deeper in the code,
1168 * but we use S_ISDIR to check for those.
1170 .rmdir = xfs_vn_unlink,
1171 .mknod = xfs_vn_mknod,
1172 .rename = xfs_vn_rename,
1173 .get_acl = xfs_get_acl,
1174 .set_acl = xfs_set_acl,
1175 .getattr = xfs_vn_getattr,
1176 .setattr = xfs_vn_setattr,
1177 .listxattr = xfs_vn_listxattr,
1178 .update_time = xfs_vn_update_time,
1179 .tmpfile = xfs_vn_tmpfile,
1182 static const struct inode_operations xfs_dir_ci_inode_operations = {
1183 .create = xfs_vn_create,
1184 .lookup = xfs_vn_ci_lookup,
1185 .link = xfs_vn_link,
1186 .unlink = xfs_vn_unlink,
1187 .symlink = xfs_vn_symlink,
1188 .mkdir = xfs_vn_mkdir,
1190 * Yes, XFS uses the same method for rmdir and unlink.
1192 * There are some subtile differences deeper in the code,
1193 * but we use S_ISDIR to check for those.
1195 .rmdir = xfs_vn_unlink,
1196 .mknod = xfs_vn_mknod,
1197 .rename = xfs_vn_rename,
1198 .get_acl = xfs_get_acl,
1199 .set_acl = xfs_set_acl,
1200 .getattr = xfs_vn_getattr,
1201 .setattr = xfs_vn_setattr,
1202 .listxattr = xfs_vn_listxattr,
1203 .update_time = xfs_vn_update_time,
1204 .tmpfile = xfs_vn_tmpfile,
1207 static const struct inode_operations xfs_symlink_inode_operations = {
1208 .get_link = xfs_vn_get_link,
1209 .getattr = xfs_vn_getattr,
1210 .setattr = xfs_vn_setattr,
1211 .listxattr = xfs_vn_listxattr,
1212 .update_time = xfs_vn_update_time,
1215 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1216 .get_link = xfs_vn_get_link_inline,
1217 .getattr = xfs_vn_getattr,
1218 .setattr = xfs_vn_setattr,
1219 .listxattr = xfs_vn_listxattr,
1220 .update_time = xfs_vn_update_time,
1223 /* Figure out if this file actually supports DAX. */
1225 xfs_inode_supports_dax(
1226 struct xfs_inode *ip)
1228 struct xfs_mount *mp = ip->i_mount;
1230 /* Only supported on regular files. */
1231 if (!S_ISREG(VFS_I(ip)->i_mode))
1234 /* Only supported on non-reflinked files. */
1235 if (xfs_is_reflink_inode(ip))
1238 /* Block size must match page size */
1239 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1242 /* Device has to support DAX too. */
1243 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1247 xfs_inode_should_enable_dax(
1248 struct xfs_inode *ip)
1250 if (!IS_ENABLED(CONFIG_FS_DAX))
1252 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1254 if (!xfs_inode_supports_dax(ip))
1256 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1258 if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
1264 xfs_diflags_to_iflags(
1265 struct xfs_inode *ip,
1268 struct inode *inode = VFS_I(ip);
1269 unsigned int xflags = xfs_ip2xflags(ip);
1270 unsigned int flags = 0;
1272 ASSERT(!(IS_DAX(inode) && init));
1274 if (xflags & FS_XFLAG_IMMUTABLE)
1275 flags |= S_IMMUTABLE;
1276 if (xflags & FS_XFLAG_APPEND)
1278 if (xflags & FS_XFLAG_SYNC)
1280 if (xflags & FS_XFLAG_NOATIME)
1282 if (init && xfs_inode_should_enable_dax(ip))
1286 * S_DAX can only be set during inode initialization and is never set by
1287 * the VFS, so we cannot mask off S_DAX in i_flags.
1289 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1290 inode->i_flags |= flags;
1294 * Initialize the Linux inode.
1296 * When reading existing inodes from disk this is called directly from xfs_iget,
1297 * when creating a new inode it is called from xfs_ialloc after setting up the
1298 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1299 * it up to the caller to deal with unlocking the inode appropriately.
1303 struct xfs_inode *ip)
1305 struct inode *inode = &ip->i_vnode;
1308 inode->i_ino = ip->i_ino;
1309 inode->i_state = I_NEW;
1311 inode_sb_list_add(inode);
1312 /* make the inode look hashed for the writeback code */
1313 inode_fake_hash(inode);
1315 i_size_write(inode, ip->i_d.di_size);
1316 xfs_diflags_to_iflags(ip, true);
1318 if (S_ISDIR(inode->i_mode)) {
1320 * We set the i_rwsem class here to avoid potential races with
1321 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1322 * after a filehandle lookup has already found the inode in
1323 * cache before it has been unlocked via unlock_new_inode().
1325 lockdep_set_class(&inode->i_rwsem,
1326 &inode->i_sb->s_type->i_mutex_dir_key);
1327 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1329 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1333 * Ensure all page cache allocations are done from GFP_NOFS context to
1334 * prevent direct reclaim recursion back into the filesystem and blowing
1335 * stacks or deadlocking.
1337 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1338 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1341 * If there is no attribute fork no ACL can exist on this inode,
1342 * and it can't have any file capabilities attached to it either.
1344 if (!XFS_IFORK_Q(ip)) {
1345 inode_has_no_xattr(inode);
1346 cache_no_acl(inode);
1352 struct xfs_inode *ip)
1354 struct inode *inode = &ip->i_vnode;
1356 switch (inode->i_mode & S_IFMT) {
1358 inode->i_op = &xfs_inode_operations;
1359 inode->i_fop = &xfs_file_operations;
1361 inode->i_mapping->a_ops = &xfs_dax_aops;
1363 inode->i_mapping->a_ops = &xfs_address_space_operations;
1366 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1367 inode->i_op = &xfs_dir_ci_inode_operations;
1369 inode->i_op = &xfs_dir_inode_operations;
1370 inode->i_fop = &xfs_dir_file_operations;
1373 if (ip->i_df.if_flags & XFS_IFINLINE)
1374 inode->i_op = &xfs_inline_symlink_inode_operations;
1376 inode->i_op = &xfs_symlink_inode_operations;
1379 inode->i_op = &xfs_inode_operations;
1380 init_special_inode(inode, inode->i_mode, inode->i_rdev);