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/xattr.h>
26 #include <linux/posix_acl.h>
27 #include <linux/security.h>
28 #include <linux/iversion.h>
31 * Directories have different lock order w.r.t. mmap_sem 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_sem 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 error = xfs_attr_set(ip, xattr->name, xattr->value,
54 xattr->value_len, ATTR_SECURE);
62 * Hook in SELinux. This is not quite correct yet, what we really need
63 * here (as we do for default ACLs) is a mechanism by which creation of
64 * these attrs can be journalled at inode creation time (along with the
65 * inode, of course, such that log replay can't cause these to be lost).
72 const struct qstr *qstr)
74 return security_inode_init_security(inode, dir, qstr,
75 &xfs_initxattrs, NULL);
80 struct xfs_name *namep,
81 struct dentry *dentry)
83 namep->name = dentry->d_name.name;
84 namep->len = dentry->d_name.len;
85 namep->type = XFS_DIR3_FT_UNKNOWN;
89 xfs_dentry_mode_to_name(
90 struct xfs_name *namep,
91 struct dentry *dentry,
94 namep->name = dentry->d_name.name;
95 namep->len = dentry->d_name.len;
96 namep->type = xfs_mode_to_ftype(mode);
98 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
108 struct dentry *dentry)
110 struct xfs_name teardown;
113 * If we can't add the ACL or we fail in
114 * xfs_init_security we must back out.
115 * ENOSPC can hit here, among other things.
117 xfs_dentry_to_name(&teardown, dentry);
119 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
125 struct dentry *dentry,
128 bool tmpfile) /* unnamed file */
131 struct xfs_inode *ip = NULL;
132 struct posix_acl *default_acl, *acl;
133 struct xfs_name name;
137 * Irix uses Missed'em'V split, but doesn't want to see
138 * the upper 5 bits of (14bit) major.
140 if (S_ISCHR(mode) || S_ISBLK(mode)) {
141 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
147 error = posix_acl_create(dir, &mode, &default_acl, &acl);
151 /* Verify mode is valid also for tmpfile case */
152 error = xfs_dentry_mode_to_name(&name, dentry, mode);
157 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
159 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
166 error = xfs_init_security(inode, dir, &dentry->d_name);
168 goto out_cleanup_inode;
170 #ifdef CONFIG_XFS_POSIX_ACL
172 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
174 goto out_cleanup_inode;
177 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
179 goto out_cleanup_inode;
187 * The VFS requires that any inode fed to d_tmpfile must have
188 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
189 * However, we created the temp file with nlink == 0 because
190 * we're not allowed to put an inode with nlink > 0 on the
191 * unlinked list. Therefore we have to set nlink to 1 so that
192 * d_tmpfile can immediately set it back to zero.
195 d_tmpfile(dentry, inode);
197 d_instantiate(dentry, inode);
199 xfs_finish_inode_setup(ip);
203 posix_acl_release(default_acl);
205 posix_acl_release(acl);
209 xfs_finish_inode_setup(ip);
211 xfs_cleanup_inode(dir, inode, dentry);
219 struct dentry *dentry,
223 return xfs_generic_create(dir, dentry, mode, rdev, false);
229 struct dentry *dentry,
233 return xfs_vn_mknod(dir, dentry, mode, 0);
239 struct dentry *dentry,
242 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
245 STATIC struct dentry *
248 struct dentry *dentry,
252 struct xfs_inode *cip;
253 struct xfs_name name;
256 if (dentry->d_name.len >= MAXNAMELEN)
257 return ERR_PTR(-ENAMETOOLONG);
259 xfs_dentry_to_name(&name, dentry);
260 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
263 else if (likely(error == -ENOENT))
266 inode = ERR_PTR(error);
267 return d_splice_alias(inode, dentry);
270 STATIC struct dentry *
273 struct dentry *dentry,
276 struct xfs_inode *ip;
277 struct xfs_name xname;
278 struct xfs_name ci_name;
282 if (dentry->d_name.len >= MAXNAMELEN)
283 return ERR_PTR(-ENAMETOOLONG);
285 xfs_dentry_to_name(&xname, dentry);
286 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
287 if (unlikely(error)) {
288 if (unlikely(error != -ENOENT))
289 return ERR_PTR(error);
291 * call d_add(dentry, NULL) here when d_drop_negative_children
292 * is called in xfs_vn_mknod (ie. allow negative dentries
293 * with CI filesystems).
298 /* if exact match, just splice and exit */
300 return d_splice_alias(VFS_I(ip), dentry);
302 /* else case-insensitive match... */
303 dname.name = ci_name.name;
304 dname.len = ci_name.len;
305 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
306 kmem_free(ci_name.name);
312 struct dentry *old_dentry,
314 struct dentry *dentry)
316 struct inode *inode = d_inode(old_dentry);
317 struct xfs_name name;
320 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
324 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
329 d_instantiate(dentry, inode);
336 struct dentry *dentry)
338 struct xfs_name name;
341 xfs_dentry_to_name(&name, dentry);
343 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
348 * With unlink, the VFS makes the dentry "negative": no inode,
349 * but still hashed. This is incompatible with case-insensitive
350 * mode, so invalidate (unhash) the dentry in CI-mode.
352 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
353 d_invalidate(dentry);
360 struct dentry *dentry,
364 struct xfs_inode *cip = NULL;
365 struct xfs_name name;
370 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
371 error = xfs_dentry_mode_to_name(&name, dentry, mode);
375 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
381 error = xfs_init_security(inode, dir, &dentry->d_name);
383 goto out_cleanup_inode;
387 d_instantiate(dentry, inode);
388 xfs_finish_inode_setup(cip);
392 xfs_finish_inode_setup(cip);
393 xfs_cleanup_inode(dir, inode, dentry);
402 struct dentry *odentry,
404 struct dentry *ndentry,
407 struct inode *new_inode = d_inode(ndentry);
410 struct xfs_name oname;
411 struct xfs_name nname;
413 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
416 /* if we are exchanging files, we need to set i_mode of both files */
417 if (flags & RENAME_EXCHANGE)
418 omode = d_inode(ndentry)->i_mode;
420 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
421 if (omode && unlikely(error))
424 error = xfs_dentry_mode_to_name(&nname, ndentry,
425 d_inode(odentry)->i_mode);
429 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
431 new_inode ? XFS_I(new_inode) : NULL, flags);
435 * careful here - this function can get called recursively, so
436 * we need to be very careful about how much stack we use.
437 * uio is kmalloced for this reason...
441 struct dentry *dentry,
443 struct delayed_call *done)
449 return ERR_PTR(-ECHILD);
451 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
455 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
459 set_delayed_call(done, kfree_link, link);
465 return ERR_PTR(error);
469 xfs_vn_get_link_inline(
470 struct dentry *dentry,
472 struct delayed_call *done)
474 struct xfs_inode *ip = XFS_I(inode);
477 ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
480 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
483 link = ip->i_df.if_u1.if_data;
485 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, ip->i_mount);
486 return ERR_PTR(-EFSCORRUPTED);
493 struct xfs_inode *ip)
495 struct xfs_mount *mp = ip->i_mount;
498 * If the file blocks are being allocated from a realtime volume, then
499 * always return the realtime extent size.
501 if (XFS_IS_REALTIME_INODE(ip))
502 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
505 * Allow large block sizes to be reported to userspace programs if the
506 * "largeio" mount option is used.
508 * If compatibility mode is specified, simply return the basic unit of
509 * caching so that we don't get inefficient read/modify/write I/O from
510 * user apps. Otherwise....
512 * If the underlying volume is a stripe, then return the stripe width in
513 * bytes as the recommended I/O size. It is not a stripe and we've set a
514 * default buffered I/O size, return that, otherwise return the compat
517 if (mp->m_flags & XFS_MOUNT_LARGEIO) {
519 return mp->m_swidth << mp->m_sb.sb_blocklog;
520 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
521 return 1U << mp->m_allocsize_log;
529 const struct path *path,
532 unsigned int query_flags)
534 struct inode *inode = d_inode(path->dentry);
535 struct xfs_inode *ip = XFS_I(inode);
536 struct xfs_mount *mp = ip->i_mount;
538 trace_xfs_getattr(ip);
540 if (XFS_FORCED_SHUTDOWN(mp))
543 stat->size = XFS_ISIZE(ip);
544 stat->dev = inode->i_sb->s_dev;
545 stat->mode = inode->i_mode;
546 stat->nlink = inode->i_nlink;
547 stat->uid = inode->i_uid;
548 stat->gid = inode->i_gid;
549 stat->ino = ip->i_ino;
550 stat->atime = inode->i_atime;
551 stat->mtime = inode->i_mtime;
552 stat->ctime = inode->i_ctime;
554 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
556 if (ip->i_d.di_version == 3) {
557 if (request_mask & STATX_BTIME) {
558 stat->result_mask |= STATX_BTIME;
559 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
560 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
565 * Note: If you add another clause to set an attribute flag, please
566 * update attributes_mask below.
568 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
569 stat->attributes |= STATX_ATTR_IMMUTABLE;
570 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
571 stat->attributes |= STATX_ATTR_APPEND;
572 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
573 stat->attributes |= STATX_ATTR_NODUMP;
575 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
579 switch (inode->i_mode & S_IFMT) {
582 stat->blksize = BLKDEV_IOSIZE;
583 stat->rdev = inode->i_rdev;
586 stat->blksize = xfs_stat_blksize(ip);
596 struct xfs_inode *ip,
599 struct inode *inode = VFS_I(ip);
600 umode_t mode = iattr->ia_mode;
602 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
604 inode->i_mode &= S_IFMT;
605 inode->i_mode |= mode & ~S_IFMT;
610 struct xfs_inode *ip,
613 struct inode *inode = VFS_I(ip);
615 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
617 if (iattr->ia_valid & ATTR_ATIME)
618 inode->i_atime = iattr->ia_atime;
619 if (iattr->ia_valid & ATTR_CTIME)
620 inode->i_ctime = iattr->ia_ctime;
621 if (iattr->ia_valid & ATTR_MTIME)
622 inode->i_mtime = iattr->ia_mtime;
627 struct dentry *dentry,
630 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
632 if (mp->m_flags & XFS_MOUNT_RDONLY)
635 if (XFS_FORCED_SHUTDOWN(mp))
638 return setattr_prepare(dentry, iattr);
642 * Set non-size attributes of an inode.
644 * Caution: The caller of this function is responsible for calling
645 * setattr_prepare() or otherwise verifying the change is fine.
649 struct xfs_inode *ip,
653 xfs_mount_t *mp = ip->i_mount;
654 struct inode *inode = VFS_I(ip);
655 int mask = iattr->ia_valid;
658 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
659 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
660 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
661 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
663 ASSERT((mask & ATTR_SIZE) == 0);
666 * If disk quotas is on, we make sure that the dquots do exist on disk,
667 * before we start any other transactions. Trying to do this later
668 * is messy. We don't care to take a readlock to look at the ids
669 * in inode here, because we can't hold it across the trans_reserve.
670 * If the IDs do change before we take the ilock, we're covered
671 * because the i_*dquot fields will get updated anyway.
673 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
676 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
678 qflags |= XFS_QMOPT_UQUOTA;
682 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
684 qflags |= XFS_QMOPT_GQUOTA;
690 * We take a reference when we initialize udqp and gdqp,
691 * so it is important that we never blindly double trip on
692 * the same variable. See xfs_create() for an example.
694 ASSERT(udqp == NULL);
695 ASSERT(gdqp == NULL);
696 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
697 xfs_kgid_to_gid(gid),
699 qflags, &udqp, &gdqp, NULL);
704 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
708 xfs_ilock(ip, XFS_ILOCK_EXCL);
709 xfs_trans_ijoin(tp, ip, 0);
712 * Change file ownership. Must be the owner or privileged.
714 if (mask & (ATTR_UID|ATTR_GID)) {
716 * These IDs could have changed since we last looked at them.
717 * But, we're assured that if the ownership did change
718 * while we didn't have the inode locked, inode's dquot(s)
719 * would have changed also.
723 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
724 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
727 * Do a quota reservation only if uid/gid is actually
730 if (XFS_IS_QUOTA_RUNNING(mp) &&
731 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
732 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
734 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
735 NULL, capable(CAP_FOWNER) ?
736 XFS_QMOPT_FORCE_RES : 0);
737 if (error) /* out of quota */
743 * Change file ownership. Must be the owner or privileged.
745 if (mask & (ATTR_UID|ATTR_GID)) {
747 * CAP_FSETID overrides the following restrictions:
749 * The set-user-ID and set-group-ID bits of a file will be
750 * cleared upon successful return from chown()
752 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
753 !capable(CAP_FSETID))
754 inode->i_mode &= ~(S_ISUID|S_ISGID);
757 * Change the ownerships and register quota modifications
758 * in the transaction.
760 if (!uid_eq(iuid, uid)) {
761 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
762 ASSERT(mask & ATTR_UID);
764 olddquot1 = xfs_qm_vop_chown(tp, ip,
765 &ip->i_udquot, udqp);
767 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
770 if (!gid_eq(igid, gid)) {
771 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
772 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
773 !XFS_IS_PQUOTA_ON(mp));
774 ASSERT(mask & ATTR_GID);
776 olddquot2 = xfs_qm_vop_chown(tp, ip,
777 &ip->i_gdquot, gdqp);
779 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
784 if (mask & ATTR_MODE)
785 xfs_setattr_mode(ip, iattr);
786 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
787 xfs_setattr_time(ip, iattr);
789 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
791 XFS_STATS_INC(mp, xs_ig_attrchg);
793 if (mp->m_flags & XFS_MOUNT_WSYNC)
794 xfs_trans_set_sync(tp);
795 error = xfs_trans_commit(tp);
797 xfs_iunlock(ip, XFS_ILOCK_EXCL);
800 * Release any dquot(s) the inode had kept before chown.
802 xfs_qm_dqrele(olddquot1);
803 xfs_qm_dqrele(olddquot2);
811 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
812 * update. We could avoid this with linked transactions
813 * and passing down the transaction pointer all the way
814 * to attr_set. No previous user of the generic
815 * Posix ACL code seems to care about this issue either.
817 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
818 error = posix_acl_chmod(inode, inode->i_mode);
826 xfs_trans_cancel(tp);
827 xfs_iunlock(ip, XFS_ILOCK_EXCL);
835 xfs_vn_setattr_nonsize(
836 struct dentry *dentry,
839 struct xfs_inode *ip = XFS_I(d_inode(dentry));
842 trace_xfs_setattr(ip);
844 error = xfs_vn_change_ok(dentry, iattr);
847 return xfs_setattr_nonsize(ip, iattr, 0);
851 * Truncate file. Must have write permission and not be a directory.
853 * Caution: The caller of this function is responsible for calling
854 * setattr_prepare() or otherwise verifying the change is fine.
858 struct xfs_inode *ip,
861 struct xfs_mount *mp = ip->i_mount;
862 struct inode *inode = VFS_I(ip);
863 xfs_off_t oldsize, newsize;
864 struct xfs_trans *tp;
867 bool did_zeroing = false;
869 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
870 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
871 ASSERT(S_ISREG(inode->i_mode));
872 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
873 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
875 oldsize = inode->i_size;
876 newsize = iattr->ia_size;
879 * Short circuit the truncate case for zero length files.
881 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
882 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
886 * Use the regular setattr path to update the timestamps.
888 iattr->ia_valid &= ~ATTR_SIZE;
889 return xfs_setattr_nonsize(ip, iattr, 0);
893 * Make sure that the dquots are attached to the inode.
895 error = xfs_qm_dqattach(ip);
900 * Wait for all direct I/O to complete.
902 inode_dio_wait(inode);
905 * File data changes must be complete before we start the transaction to
906 * modify the inode. This needs to be done before joining the inode to
907 * the transaction because the inode cannot be unlocked once it is a
908 * part of the transaction.
910 * Start with zeroing any data beyond EOF that we may expose on file
911 * extension, or zeroing out the rest of the block on a downward
914 if (newsize > oldsize) {
915 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
916 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
917 &did_zeroing, &xfs_buffered_write_iomap_ops);
919 error = iomap_truncate_page(inode, newsize, &did_zeroing,
920 &xfs_buffered_write_iomap_ops);
927 * We've already locked out new page faults, so now we can safely remove
928 * pages from the page cache knowing they won't get refaulted until we
929 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
930 * complete. The truncate_setsize() call also cleans partial EOF page
931 * PTEs on extending truncates and hence ensures sub-page block size
932 * filesystems are correctly handled, too.
934 * We have to do all the page cache truncate work outside the
935 * transaction context as the "lock" order is page lock->log space
936 * reservation as defined by extent allocation in the writeback path.
937 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
938 * having already truncated the in-memory version of the file (i.e. made
939 * user visible changes). There's not much we can do about this, except
940 * to hope that the caller sees ENOMEM and retries the truncate
943 * And we update in-core i_size and truncate page cache beyond newsize
944 * before writeback the [di_size, newsize] range, so we're guaranteed
945 * not to write stale data past the new EOF on truncate down.
947 truncate_setsize(inode, newsize);
950 * We are going to log the inode size change in this transaction so
951 * any previous writes that are beyond the on disk EOF and the new
952 * EOF that have not been written out need to be written here. If we
953 * do not write the data out, we expose ourselves to the null files
954 * problem. Note that this includes any block zeroing we did above;
955 * otherwise those blocks may not be zeroed after a crash.
958 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
959 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
960 ip->i_d.di_size, newsize - 1);
965 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
969 lock_flags |= XFS_ILOCK_EXCL;
970 xfs_ilock(ip, XFS_ILOCK_EXCL);
971 xfs_trans_ijoin(tp, ip, 0);
974 * Only change the c/mtime if we are changing the size or we are
975 * explicitly asked to change it. This handles the semantic difference
976 * between truncate() and ftruncate() as implemented in the VFS.
978 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
979 * special case where we need to update the times despite not having
980 * these flags set. For all other operations the VFS set these flags
981 * explicitly if it wants a timestamp update.
983 if (newsize != oldsize &&
984 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
985 iattr->ia_ctime = iattr->ia_mtime =
987 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
991 * The first thing we do is set the size to new_size permanently on
992 * disk. This way we don't have to worry about anyone ever being able
993 * to look at the data being freed even in the face of a crash.
994 * What we're getting around here is the case where we free a block, it
995 * is allocated to another file, it is written to, and then we crash.
996 * If the new data gets written to the file but the log buffers
997 * containing the free and reallocation don't, then we'd end up with
998 * garbage in the blocks being freed. As long as we make the new size
999 * permanent before actually freeing any blocks it doesn't matter if
1000 * they get written to.
1002 ip->i_d.di_size = newsize;
1003 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1005 if (newsize <= oldsize) {
1006 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1008 goto out_trans_cancel;
1011 * Truncated "down", so we're removing references to old data
1012 * here - if we delay flushing for a long time, we expose
1013 * ourselves unduly to the notorious NULL files problem. So,
1014 * we mark this inode and flush it when the file is closed,
1015 * and do not wait the usual (long) time for writeout.
1017 xfs_iflags_set(ip, XFS_ITRUNCATED);
1019 /* A truncate down always removes post-EOF blocks. */
1020 xfs_inode_clear_eofblocks_tag(ip);
1023 if (iattr->ia_valid & ATTR_MODE)
1024 xfs_setattr_mode(ip, iattr);
1025 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1026 xfs_setattr_time(ip, iattr);
1028 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1030 XFS_STATS_INC(mp, xs_ig_attrchg);
1032 if (mp->m_flags & XFS_MOUNT_WSYNC)
1033 xfs_trans_set_sync(tp);
1035 error = xfs_trans_commit(tp);
1038 xfs_iunlock(ip, lock_flags);
1042 xfs_trans_cancel(tp);
1047 xfs_vn_setattr_size(
1048 struct dentry *dentry,
1049 struct iattr *iattr)
1051 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1054 trace_xfs_setattr(ip);
1056 error = xfs_vn_change_ok(dentry, iattr);
1059 return xfs_setattr_size(ip, iattr);
1064 struct dentry *dentry,
1065 struct iattr *iattr)
1069 if (iattr->ia_valid & ATTR_SIZE) {
1070 struct inode *inode = d_inode(dentry);
1071 struct xfs_inode *ip = XFS_I(inode);
1074 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1075 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1077 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1079 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1083 error = xfs_vn_setattr_size(dentry, iattr);
1084 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1086 error = xfs_vn_setattr_nonsize(dentry, iattr);
1094 struct inode *inode,
1095 struct timespec64 *now,
1098 struct xfs_inode *ip = XFS_I(inode);
1099 struct xfs_mount *mp = ip->i_mount;
1100 int log_flags = XFS_ILOG_TIMESTAMP;
1101 struct xfs_trans *tp;
1104 trace_xfs_update_time(ip);
1106 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1107 if (!((flags & S_VERSION) &&
1108 inode_maybe_inc_iversion(inode, false)))
1109 return generic_update_time(inode, now, flags);
1111 /* Capture the iversion update that just occurred */
1112 log_flags |= XFS_ILOG_CORE;
1115 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1119 xfs_ilock(ip, XFS_ILOCK_EXCL);
1120 if (flags & S_CTIME)
1121 inode->i_ctime = *now;
1122 if (flags & S_MTIME)
1123 inode->i_mtime = *now;
1124 if (flags & S_ATIME)
1125 inode->i_atime = *now;
1127 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1128 xfs_trans_log_inode(tp, ip, log_flags);
1129 return xfs_trans_commit(tp);
1134 struct inode *inode,
1135 struct fiemap_extent_info *fieinfo,
1141 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1142 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1143 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1144 error = iomap_fiemap(inode, fieinfo, start, length,
1145 &xfs_xattr_iomap_ops);
1147 error = iomap_fiemap(inode, fieinfo, start, length,
1148 &xfs_read_iomap_ops);
1150 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1158 struct dentry *dentry,
1161 return xfs_generic_create(dir, dentry, mode, 0, true);
1164 static const struct inode_operations xfs_inode_operations = {
1165 .get_acl = xfs_get_acl,
1166 .set_acl = xfs_set_acl,
1167 .getattr = xfs_vn_getattr,
1168 .setattr = xfs_vn_setattr,
1169 .listxattr = xfs_vn_listxattr,
1170 .fiemap = xfs_vn_fiemap,
1171 .update_time = xfs_vn_update_time,
1174 static const struct inode_operations xfs_dir_inode_operations = {
1175 .create = xfs_vn_create,
1176 .lookup = xfs_vn_lookup,
1177 .link = xfs_vn_link,
1178 .unlink = xfs_vn_unlink,
1179 .symlink = xfs_vn_symlink,
1180 .mkdir = xfs_vn_mkdir,
1182 * Yes, XFS uses the same method for rmdir and unlink.
1184 * There are some subtile differences deeper in the code,
1185 * but we use S_ISDIR to check for those.
1187 .rmdir = xfs_vn_unlink,
1188 .mknod = xfs_vn_mknod,
1189 .rename = xfs_vn_rename,
1190 .get_acl = xfs_get_acl,
1191 .set_acl = xfs_set_acl,
1192 .getattr = xfs_vn_getattr,
1193 .setattr = xfs_vn_setattr,
1194 .listxattr = xfs_vn_listxattr,
1195 .update_time = xfs_vn_update_time,
1196 .tmpfile = xfs_vn_tmpfile,
1199 static const struct inode_operations xfs_dir_ci_inode_operations = {
1200 .create = xfs_vn_create,
1201 .lookup = xfs_vn_ci_lookup,
1202 .link = xfs_vn_link,
1203 .unlink = xfs_vn_unlink,
1204 .symlink = xfs_vn_symlink,
1205 .mkdir = xfs_vn_mkdir,
1207 * Yes, XFS uses the same method for rmdir and unlink.
1209 * There are some subtile differences deeper in the code,
1210 * but we use S_ISDIR to check for those.
1212 .rmdir = xfs_vn_unlink,
1213 .mknod = xfs_vn_mknod,
1214 .rename = xfs_vn_rename,
1215 .get_acl = xfs_get_acl,
1216 .set_acl = xfs_set_acl,
1217 .getattr = xfs_vn_getattr,
1218 .setattr = xfs_vn_setattr,
1219 .listxattr = xfs_vn_listxattr,
1220 .update_time = xfs_vn_update_time,
1221 .tmpfile = xfs_vn_tmpfile,
1224 static const struct inode_operations xfs_symlink_inode_operations = {
1225 .get_link = xfs_vn_get_link,
1226 .getattr = xfs_vn_getattr,
1227 .setattr = xfs_vn_setattr,
1228 .listxattr = xfs_vn_listxattr,
1229 .update_time = xfs_vn_update_time,
1232 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1233 .get_link = xfs_vn_get_link_inline,
1234 .getattr = xfs_vn_getattr,
1235 .setattr = xfs_vn_setattr,
1236 .listxattr = xfs_vn_listxattr,
1237 .update_time = xfs_vn_update_time,
1240 /* Figure out if this file actually supports DAX. */
1242 xfs_inode_supports_dax(
1243 struct xfs_inode *ip)
1245 struct xfs_mount *mp = ip->i_mount;
1247 /* Only supported on non-reflinked files. */
1248 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1251 /* DAX mount option or DAX iflag must be set. */
1252 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1253 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1256 /* Block size must match page size */
1257 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1260 /* Device has to support DAX too. */
1261 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1265 xfs_diflags_to_iflags(
1266 struct inode *inode,
1267 struct xfs_inode *ip)
1269 uint16_t flags = ip->i_d.di_flags;
1271 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1274 if (flags & XFS_DIFLAG_IMMUTABLE)
1275 inode->i_flags |= S_IMMUTABLE;
1276 if (flags & XFS_DIFLAG_APPEND)
1277 inode->i_flags |= S_APPEND;
1278 if (flags & XFS_DIFLAG_SYNC)
1279 inode->i_flags |= S_SYNC;
1280 if (flags & XFS_DIFLAG_NOATIME)
1281 inode->i_flags |= S_NOATIME;
1282 if (xfs_inode_supports_dax(ip))
1283 inode->i_flags |= S_DAX;
1287 * Initialize the Linux inode.
1289 * When reading existing inodes from disk this is called directly from xfs_iget,
1290 * when creating a new inode it is called from xfs_ialloc after setting up the
1291 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1292 * it up to the caller to deal with unlocking the inode appropriately.
1296 struct xfs_inode *ip)
1298 struct inode *inode = &ip->i_vnode;
1301 inode->i_ino = ip->i_ino;
1302 inode->i_state = I_NEW;
1304 inode_sb_list_add(inode);
1305 /* make the inode look hashed for the writeback code */
1306 inode_fake_hash(inode);
1308 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1309 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1311 i_size_write(inode, ip->i_d.di_size);
1312 xfs_diflags_to_iflags(inode, ip);
1314 if (S_ISDIR(inode->i_mode)) {
1316 * We set the i_rwsem class here to avoid potential races with
1317 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1318 * after a filehandle lookup has already found the inode in
1319 * cache before it has been unlocked via unlock_new_inode().
1321 lockdep_set_class(&inode->i_rwsem,
1322 &inode->i_sb->s_type->i_mutex_dir_key);
1323 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1325 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1329 * Ensure all page cache allocations are done from GFP_NOFS context to
1330 * prevent direct reclaim recursion back into the filesystem and blowing
1331 * stacks or deadlocking.
1333 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1334 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1337 * If there is no attribute fork no ACL can exist on this inode,
1338 * and it can't have any file capabilities attached to it either.
1340 if (!XFS_IFORK_Q(ip)) {
1341 inode_has_no_xattr(inode);
1342 cache_no_acl(inode);
1348 struct xfs_inode *ip)
1350 struct inode *inode = &ip->i_vnode;
1352 switch (inode->i_mode & S_IFMT) {
1354 inode->i_op = &xfs_inode_operations;
1355 inode->i_fop = &xfs_file_operations;
1357 inode->i_mapping->a_ops = &xfs_dax_aops;
1359 inode->i_mapping->a_ops = &xfs_address_space_operations;
1362 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1363 inode->i_op = &xfs_dir_ci_inode_operations;
1365 inode->i_op = &xfs_dir_inode_operations;
1366 inode->i_fop = &xfs_dir_file_operations;
1369 if (ip->i_df.if_flags & XFS_IFINLINE)
1370 inode->i_op = &xfs_inline_symlink_inode_operations;
1372 inode->i_op = &xfs_symlink_inode_operations;
1375 inode->i_op = &xfs_inode_operations;
1376 init_special_inode(inode, inode->i_mode, inode->i_rdev);