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"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_trace.h"
21 #include "xfs_icache.h"
22 #include "xfs_symlink.h"
24 #include "xfs_iomap.h"
25 #include "xfs_error.h"
26 #include "xfs_ioctl.h"
27 #include "xfs_xattr.h"
30 #include <linux/posix_acl.h>
31 #include <linux/security.h>
32 #include <linux/iversion.h>
33 #include <linux/fiemap.h>
36 * Directories have different lock order w.r.t. mmap_lock compared to regular
37 * files. This is due to readdir potentially triggering page faults on a user
38 * buffer inside filldir(), and this happens with the ilock on the directory
39 * held. For regular files, the lock order is the other way around - the
40 * mmap_lock is taken during the page fault, and then we lock the ilock to do
41 * block mapping. Hence we need a different class for the directory ilock so
42 * that lockdep can tell them apart.
44 static struct lock_class_key xfs_nondir_ilock_class;
45 static struct lock_class_key xfs_dir_ilock_class;
50 const struct xattr *xattr_array,
53 const struct xattr *xattr;
54 struct xfs_inode *ip = XFS_I(inode);
57 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
58 struct xfs_da_args args = {
60 .attr_filter = XFS_ATTR_SECURE,
62 .namelen = strlen(xattr->name),
63 .value = xattr->value,
64 .valuelen = xattr->value_len,
66 error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT);
74 * Hook in SELinux. This is not quite correct yet, what we really need
75 * here (as we do for default ACLs) is a mechanism by which creation of
76 * these attrs can be journalled at inode creation time (along with the
77 * inode, of course, such that log replay can't cause these to be lost).
80 xfs_inode_init_security(
83 const struct qstr *qstr)
85 return security_inode_init_security(inode, dir, qstr,
86 &xfs_initxattrs, NULL);
91 struct xfs_name *namep,
92 struct dentry *dentry)
94 namep->name = dentry->d_name.name;
95 namep->len = dentry->d_name.len;
96 namep->type = XFS_DIR3_FT_UNKNOWN;
100 xfs_dentry_mode_to_name(
101 struct xfs_name *namep,
102 struct dentry *dentry,
105 namep->name = dentry->d_name.name;
106 namep->len = dentry->d_name.len;
107 namep->type = xfs_mode_to_ftype(mode);
109 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
110 return -EFSCORRUPTED;
119 struct dentry *dentry)
121 struct xfs_name teardown;
124 * If we can't add the ACL or we fail in
125 * xfs_inode_init_security we must back out.
126 * ENOSPC can hit here, among other things.
128 xfs_dentry_to_name(&teardown, dentry);
130 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
134 * Check to see if we are likely to need an extended attribute to be added to
135 * the inode we are about to allocate. This allows the attribute fork to be
136 * created during the inode allocation, reducing the number of transactions we
137 * need to do in this fast path.
139 * The security checks are optimistic, but not guaranteed. The two LSMs that
140 * require xattrs to be added here (selinux and smack) are also the only two
141 * LSMs that add a sb->s_security structure to the superblock. Hence if security
142 * is enabled and sb->s_security is set, we have a pretty good idea that we are
143 * going to be asked to add a security xattr immediately after allocating the
144 * xfs inode and instantiating the VFS inode.
147 xfs_create_need_xattr(
149 struct posix_acl *default_acl,
150 struct posix_acl *acl)
156 #if IS_ENABLED(CONFIG_SECURITY)
157 if (dir->i_sb->s_security)
166 struct mnt_idmap *idmap,
168 struct dentry *dentry,
171 struct file *tmpfile) /* unnamed file */
174 struct xfs_inode *ip = NULL;
175 struct posix_acl *default_acl, *acl;
176 struct xfs_name name;
180 * Irix uses Missed'em'V split, but doesn't want to see
181 * the upper 5 bits of (14bit) major.
183 if (S_ISCHR(mode) || S_ISBLK(mode)) {
184 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
190 error = posix_acl_create(dir, &mode, &default_acl, &acl);
194 /* Verify mode is valid also for tmpfile case */
195 error = xfs_dentry_mode_to_name(&name, dentry, mode);
200 error = xfs_create(idmap, XFS_I(dir), &name, mode, rdev,
201 xfs_create_need_xattr(dir, default_acl, acl),
204 error = xfs_create_tmpfile(idmap, XFS_I(dir), mode, false, &ip);
211 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
213 goto out_cleanup_inode;
216 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
218 goto out_cleanup_inode;
221 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
223 goto out_cleanup_inode;
230 * The VFS requires that any inode fed to d_tmpfile must have
231 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
232 * However, we created the temp file with nlink == 0 because
233 * we're not allowed to put an inode with nlink > 0 on the
234 * unlinked list. Therefore we have to set nlink to 1 so that
235 * d_tmpfile can immediately set it back to zero.
238 d_tmpfile(tmpfile, inode);
240 d_instantiate(dentry, inode);
242 xfs_finish_inode_setup(ip);
245 posix_acl_release(default_acl);
246 posix_acl_release(acl);
250 xfs_finish_inode_setup(ip);
252 xfs_cleanup_inode(dir, inode, dentry);
259 struct mnt_idmap *idmap,
261 struct dentry *dentry,
265 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
270 struct mnt_idmap *idmap,
272 struct dentry *dentry,
276 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
281 struct mnt_idmap *idmap,
283 struct dentry *dentry,
286 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
289 STATIC struct dentry *
292 struct dentry *dentry,
296 struct xfs_inode *cip;
297 struct xfs_name name;
300 if (dentry->d_name.len >= MAXNAMELEN)
301 return ERR_PTR(-ENAMETOOLONG);
303 xfs_dentry_to_name(&name, dentry);
304 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
307 else if (likely(error == -ENOENT))
310 inode = ERR_PTR(error);
311 return d_splice_alias(inode, dentry);
314 STATIC struct dentry *
317 struct dentry *dentry,
320 struct xfs_inode *ip;
321 struct xfs_name xname;
322 struct xfs_name ci_name;
326 if (dentry->d_name.len >= MAXNAMELEN)
327 return ERR_PTR(-ENAMETOOLONG);
329 xfs_dentry_to_name(&xname, dentry);
330 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
331 if (unlikely(error)) {
332 if (unlikely(error != -ENOENT))
333 return ERR_PTR(error);
335 * call d_add(dentry, NULL) here when d_drop_negative_children
336 * is called in xfs_vn_mknod (ie. allow negative dentries
337 * with CI filesystems).
342 /* if exact match, just splice and exit */
344 return d_splice_alias(VFS_I(ip), dentry);
346 /* else case-insensitive match... */
347 dname.name = ci_name.name;
348 dname.len = ci_name.len;
349 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
356 struct dentry *old_dentry,
358 struct dentry *dentry)
360 struct inode *inode = d_inode(old_dentry);
361 struct xfs_name name;
364 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
368 if (IS_PRIVATE(inode))
371 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
376 d_instantiate(dentry, inode);
383 struct dentry *dentry)
385 struct xfs_name name;
388 xfs_dentry_to_name(&name, dentry);
390 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
395 * With unlink, the VFS makes the dentry "negative": no inode,
396 * but still hashed. This is incompatible with case-insensitive
397 * mode, so invalidate (unhash) the dentry in CI-mode.
399 if (xfs_has_asciici(XFS_M(dir->i_sb)))
400 d_invalidate(dentry);
406 struct mnt_idmap *idmap,
408 struct dentry *dentry,
412 struct xfs_inode *cip = NULL;
413 struct xfs_name name;
418 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
419 error = xfs_dentry_mode_to_name(&name, dentry, mode);
423 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
429 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
431 goto out_cleanup_inode;
435 d_instantiate(dentry, inode);
436 xfs_finish_inode_setup(cip);
440 xfs_finish_inode_setup(cip);
441 xfs_cleanup_inode(dir, inode, dentry);
449 struct mnt_idmap *idmap,
451 struct dentry *odentry,
453 struct dentry *ndentry,
456 struct inode *new_inode = d_inode(ndentry);
459 struct xfs_name oname;
460 struct xfs_name nname;
462 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
465 /* if we are exchanging files, we need to set i_mode of both files */
466 if (flags & RENAME_EXCHANGE)
467 omode = d_inode(ndentry)->i_mode;
469 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
470 if (omode && unlikely(error))
473 error = xfs_dentry_mode_to_name(&nname, ndentry,
474 d_inode(odentry)->i_mode);
478 return xfs_rename(idmap, XFS_I(odir), &oname,
479 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
480 new_inode ? XFS_I(new_inode) : NULL, flags);
484 * careful here - this function can get called recursively, so
485 * we need to be very careful about how much stack we use.
486 * uio is kmalloced for this reason...
490 struct dentry *dentry,
492 struct delayed_call *done)
498 return ERR_PTR(-ECHILD);
500 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
504 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
508 set_delayed_call(done, kfree_link, link);
514 return ERR_PTR(error);
519 struct xfs_inode *ip)
521 struct xfs_mount *mp = ip->i_mount;
524 * If the file blocks are being allocated from a realtime volume, then
525 * always return the realtime extent size.
527 if (XFS_IS_REALTIME_INODE(ip))
528 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
531 * Allow large block sizes to be reported to userspace programs if the
532 * "largeio" mount option is used.
534 * If compatibility mode is specified, simply return the basic unit of
535 * caching so that we don't get inefficient read/modify/write I/O from
536 * user apps. Otherwise....
538 * If the underlying volume is a stripe, then return the stripe width in
539 * bytes as the recommended I/O size. It is not a stripe and we've set a
540 * default buffered I/O size, return that, otherwise return the compat
543 if (xfs_has_large_iosize(mp)) {
545 return XFS_FSB_TO_B(mp, mp->m_swidth);
546 if (xfs_has_allocsize(mp))
547 return 1U << mp->m_allocsize_log;
555 struct mnt_idmap *idmap,
556 const struct path *path,
559 unsigned int query_flags)
561 struct inode *inode = d_inode(path->dentry);
562 struct xfs_inode *ip = XFS_I(inode);
563 struct xfs_mount *mp = ip->i_mount;
564 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
565 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
567 trace_xfs_getattr(ip);
569 if (xfs_is_shutdown(mp))
572 stat->size = XFS_ISIZE(ip);
573 stat->dev = inode->i_sb->s_dev;
574 stat->mode = inode->i_mode;
575 stat->nlink = inode->i_nlink;
576 stat->uid = vfsuid_into_kuid(vfsuid);
577 stat->gid = vfsgid_into_kgid(vfsgid);
578 stat->ino = ip->i_ino;
579 stat->atime = inode_get_atime(inode);
580 stat->mtime = inode_get_mtime(inode);
581 stat->ctime = inode_get_ctime(inode);
582 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
584 if (xfs_has_v3inodes(mp)) {
585 if (request_mask & STATX_BTIME) {
586 stat->result_mask |= STATX_BTIME;
587 stat->btime = ip->i_crtime;
591 if ((request_mask & STATX_CHANGE_COOKIE) && IS_I_VERSION(inode)) {
592 stat->change_cookie = inode_query_iversion(inode);
593 stat->result_mask |= STATX_CHANGE_COOKIE;
597 * Note: If you add another clause to set an attribute flag, please
598 * update attributes_mask below.
600 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
601 stat->attributes |= STATX_ATTR_IMMUTABLE;
602 if (ip->i_diflags & XFS_DIFLAG_APPEND)
603 stat->attributes |= STATX_ATTR_APPEND;
604 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
605 stat->attributes |= STATX_ATTR_NODUMP;
607 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
611 switch (inode->i_mode & S_IFMT) {
614 stat->blksize = BLKDEV_IOSIZE;
615 stat->rdev = inode->i_rdev;
618 if (request_mask & STATX_DIOALIGN) {
619 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
620 struct block_device *bdev = target->bt_bdev;
622 stat->result_mask |= STATX_DIOALIGN;
623 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
624 stat->dio_offset_align = bdev_logical_block_size(bdev);
628 stat->blksize = xfs_stat_blksize(ip);
638 struct mnt_idmap *idmap,
639 struct dentry *dentry,
642 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
644 if (xfs_is_readonly(mp))
647 if (xfs_is_shutdown(mp))
650 return setattr_prepare(idmap, 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 mnt_idmap *idmap,
662 struct dentry *dentry,
663 struct xfs_inode *ip,
666 xfs_mount_t *mp = ip->i_mount;
667 struct inode *inode = VFS_I(ip);
668 int mask = iattr->ia_valid;
671 kuid_t uid = GLOBAL_ROOT_UID;
672 kgid_t gid = GLOBAL_ROOT_GID;
673 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
674 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
676 ASSERT((mask & ATTR_SIZE) == 0);
679 * If disk quotas is on, we make sure that the dquots do exist on disk,
680 * before we start any other transactions. Trying to do this later
681 * is messy. We don't care to take a readlock to look at the ids
682 * in inode here, because we can't hold it across the trans_reserve.
683 * If the IDs do change before we take the ilock, we're covered
684 * because the i_*dquot fields will get updated anyway.
686 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
689 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
690 uid = from_vfsuid(idmap, i_user_ns(inode),
692 qflags |= XFS_QMOPT_UQUOTA;
696 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
697 gid = from_vfsgid(idmap, i_user_ns(inode),
699 qflags |= XFS_QMOPT_GQUOTA;
705 * We take a reference when we initialize udqp and gdqp,
706 * so it is important that we never blindly double trip on
707 * the same variable. See xfs_create() for an example.
709 ASSERT(udqp == NULL);
710 ASSERT(gdqp == NULL);
711 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
712 qflags, &udqp, &gdqp, NULL);
717 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
718 has_capability_noaudit(current, CAP_FOWNER), &tp);
723 * Register quota modifications in the transaction. Must be the owner
724 * or privileged. These IDs could have changed since we last looked at
725 * them. But, we're assured that if the ownership did change while we
726 * didn't have the inode locked, inode's dquot(s) would have changed
729 if (XFS_IS_UQUOTA_ON(mp) &&
730 i_uid_needs_update(idmap, iattr, inode)) {
732 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
734 if (XFS_IS_GQUOTA_ON(mp) &&
735 i_gid_needs_update(idmap, iattr, inode)) {
736 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
738 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
741 setattr_copy(idmap, inode, iattr);
742 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
744 XFS_STATS_INC(mp, xs_ig_attrchg);
746 if (xfs_has_wsync(mp))
747 xfs_trans_set_sync(tp);
748 error = xfs_trans_commit(tp);
751 * Release any dquot(s) the inode had kept before chown.
753 xfs_qm_dqrele(old_udqp);
754 xfs_qm_dqrele(old_gdqp);
762 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
763 * update. We could avoid this with linked transactions
764 * and passing down the transaction pointer all the way
765 * to attr_set. No previous user of the generic
766 * Posix ACL code seems to care about this issue either.
768 if (mask & ATTR_MODE) {
769 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
783 * Truncate file. Must have write permission and not be a directory.
785 * Caution: The caller of this function is responsible for calling
786 * setattr_prepare() or otherwise verifying the change is fine.
790 struct mnt_idmap *idmap,
791 struct dentry *dentry,
792 struct xfs_inode *ip,
795 struct xfs_mount *mp = ip->i_mount;
796 struct inode *inode = VFS_I(ip);
797 xfs_off_t oldsize, newsize;
798 struct xfs_trans *tp;
801 bool did_zeroing = false;
803 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
804 ASSERT(S_ISREG(inode->i_mode));
805 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
806 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
808 oldsize = inode->i_size;
809 newsize = iattr->ia_size;
812 * Short circuit the truncate case for zero length files.
814 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
815 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
819 * Use the regular setattr path to update the timestamps.
821 iattr->ia_valid &= ~ATTR_SIZE;
822 return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
826 * Make sure that the dquots are attached to the inode.
828 error = xfs_qm_dqattach(ip);
833 * Wait for all direct I/O to complete.
835 inode_dio_wait(inode);
838 * File data changes must be complete before we start the transaction to
839 * modify the inode. This needs to be done before joining the inode to
840 * the transaction because the inode cannot be unlocked once it is a
841 * part of the transaction.
843 * Start with zeroing any data beyond EOF that we may expose on file
844 * extension, or zeroing out the rest of the block on a downward
847 if (newsize > oldsize) {
848 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
849 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
853 * iomap won't detect a dirty page over an unwritten block (or a
854 * cow block over a hole) and subsequently skips zeroing the
855 * newly post-EOF portion of the page. Flush the new EOF to
856 * convert the block before the pagecache truncate.
858 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
862 error = xfs_truncate_page(ip, newsize, &did_zeroing);
869 * We've already locked out new page faults, so now we can safely remove
870 * pages from the page cache knowing they won't get refaulted until we
871 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
872 * complete. The truncate_setsize() call also cleans partial EOF page
873 * PTEs on extending truncates and hence ensures sub-page block size
874 * filesystems are correctly handled, too.
876 * We have to do all the page cache truncate work outside the
877 * transaction context as the "lock" order is page lock->log space
878 * reservation as defined by extent allocation in the writeback path.
879 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
880 * having already truncated the in-memory version of the file (i.e. made
881 * user visible changes). There's not much we can do about this, except
882 * to hope that the caller sees ENOMEM and retries the truncate
885 * And we update in-core i_size and truncate page cache beyond newsize
886 * before writeback the [i_disk_size, newsize] range, so we're
887 * guaranteed not to write stale data past the new EOF on truncate down.
889 truncate_setsize(inode, newsize);
892 * We are going to log the inode size change in this transaction so
893 * any previous writes that are beyond the on disk EOF and the new
894 * EOF that have not been written out need to be written here. If we
895 * do not write the data out, we expose ourselves to the null files
896 * problem. Note that this includes any block zeroing we did above;
897 * otherwise those blocks may not be zeroed after a crash.
900 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
901 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
902 ip->i_disk_size, newsize - 1);
907 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
911 lock_flags |= XFS_ILOCK_EXCL;
912 xfs_ilock(ip, XFS_ILOCK_EXCL);
913 xfs_trans_ijoin(tp, ip, 0);
916 * Only change the c/mtime if we are changing the size or we are
917 * explicitly asked to change it. This handles the semantic difference
918 * between truncate() and ftruncate() as implemented in the VFS.
920 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
921 * special case where we need to update the times despite not having
922 * these flags set. For all other operations the VFS set these flags
923 * explicitly if it wants a timestamp update.
925 if (newsize != oldsize &&
926 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
927 iattr->ia_ctime = iattr->ia_mtime =
929 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
933 * The first thing we do is set the size to new_size permanently on
934 * disk. This way we don't have to worry about anyone ever being able
935 * to look at the data being freed even in the face of a crash.
936 * What we're getting around here is the case where we free a block, it
937 * is allocated to another file, it is written to, and then we crash.
938 * If the new data gets written to the file but the log buffers
939 * containing the free and reallocation don't, then we'd end up with
940 * garbage in the blocks being freed. As long as we make the new size
941 * permanent before actually freeing any blocks it doesn't matter if
942 * they get written to.
944 ip->i_disk_size = newsize;
945 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
947 if (newsize <= oldsize) {
948 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
950 goto out_trans_cancel;
953 * Truncated "down", so we're removing references to old data
954 * here - if we delay flushing for a long time, we expose
955 * ourselves unduly to the notorious NULL files problem. So,
956 * we mark this inode and flush it when the file is closed,
957 * and do not wait the usual (long) time for writeout.
959 xfs_iflags_set(ip, XFS_ITRUNCATED);
961 /* A truncate down always removes post-EOF blocks. */
962 xfs_inode_clear_eofblocks_tag(ip);
965 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
966 setattr_copy(idmap, inode, iattr);
967 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
969 XFS_STATS_INC(mp, xs_ig_attrchg);
971 if (xfs_has_wsync(mp))
972 xfs_trans_set_sync(tp);
974 error = xfs_trans_commit(tp);
977 xfs_iunlock(ip, lock_flags);
981 xfs_trans_cancel(tp);
987 struct mnt_idmap *idmap,
988 struct dentry *dentry,
991 struct xfs_inode *ip = XFS_I(d_inode(dentry));
994 trace_xfs_setattr(ip);
996 error = xfs_vn_change_ok(idmap, dentry, iattr);
999 return xfs_setattr_size(idmap, dentry, ip, iattr);
1004 struct mnt_idmap *idmap,
1005 struct dentry *dentry,
1006 struct iattr *iattr)
1008 struct inode *inode = d_inode(dentry);
1009 struct xfs_inode *ip = XFS_I(inode);
1012 if (iattr->ia_valid & ATTR_SIZE) {
1015 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1016 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1018 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1020 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1024 error = xfs_vn_setattr_size(idmap, dentry, iattr);
1025 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1027 trace_xfs_setattr(ip);
1029 error = xfs_vn_change_ok(idmap, dentry, iattr);
1031 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1039 struct inode *inode,
1042 struct xfs_inode *ip = XFS_I(inode);
1043 struct xfs_mount *mp = ip->i_mount;
1044 int log_flags = XFS_ILOG_TIMESTAMP;
1045 struct xfs_trans *tp;
1047 struct timespec64 now;
1049 trace_xfs_update_time(ip);
1051 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1052 if (!((flags & S_VERSION) &&
1053 inode_maybe_inc_iversion(inode, false))) {
1054 generic_update_time(inode, flags);
1058 /* Capture the iversion update that just occurred */
1059 log_flags |= XFS_ILOG_CORE;
1062 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1066 xfs_ilock(ip, XFS_ILOCK_EXCL);
1067 if (flags & (S_CTIME|S_MTIME))
1068 now = inode_set_ctime_current(inode);
1070 now = current_time(inode);
1072 if (flags & S_MTIME)
1073 inode_set_mtime_to_ts(inode, now);
1074 if (flags & S_ATIME)
1075 inode_set_atime_to_ts(inode, now);
1077 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1078 xfs_trans_log_inode(tp, ip, log_flags);
1079 return xfs_trans_commit(tp);
1084 struct inode *inode,
1085 struct fiemap_extent_info *fieinfo,
1091 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1092 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1093 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1094 error = iomap_fiemap(inode, fieinfo, start, length,
1095 &xfs_xattr_iomap_ops);
1097 error = iomap_fiemap(inode, fieinfo, start, length,
1098 &xfs_read_iomap_ops);
1100 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1107 struct mnt_idmap *idmap,
1112 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1114 return finish_open_simple(file, err);
1117 static const struct inode_operations xfs_inode_operations = {
1118 .get_inode_acl = xfs_get_acl,
1119 .set_acl = xfs_set_acl,
1120 .getattr = xfs_vn_getattr,
1121 .setattr = xfs_vn_setattr,
1122 .listxattr = xfs_vn_listxattr,
1123 .fiemap = xfs_vn_fiemap,
1124 .update_time = xfs_vn_update_time,
1125 .fileattr_get = xfs_fileattr_get,
1126 .fileattr_set = xfs_fileattr_set,
1129 static const struct inode_operations xfs_dir_inode_operations = {
1130 .create = xfs_vn_create,
1131 .lookup = xfs_vn_lookup,
1132 .link = xfs_vn_link,
1133 .unlink = xfs_vn_unlink,
1134 .symlink = xfs_vn_symlink,
1135 .mkdir = xfs_vn_mkdir,
1137 * Yes, XFS uses the same method for rmdir and unlink.
1139 * There are some subtile differences deeper in the code,
1140 * but we use S_ISDIR to check for those.
1142 .rmdir = xfs_vn_unlink,
1143 .mknod = xfs_vn_mknod,
1144 .rename = xfs_vn_rename,
1145 .get_inode_acl = xfs_get_acl,
1146 .set_acl = xfs_set_acl,
1147 .getattr = xfs_vn_getattr,
1148 .setattr = xfs_vn_setattr,
1149 .listxattr = xfs_vn_listxattr,
1150 .update_time = xfs_vn_update_time,
1151 .tmpfile = xfs_vn_tmpfile,
1152 .fileattr_get = xfs_fileattr_get,
1153 .fileattr_set = xfs_fileattr_set,
1156 static const struct inode_operations xfs_dir_ci_inode_operations = {
1157 .create = xfs_vn_create,
1158 .lookup = xfs_vn_ci_lookup,
1159 .link = xfs_vn_link,
1160 .unlink = xfs_vn_unlink,
1161 .symlink = xfs_vn_symlink,
1162 .mkdir = xfs_vn_mkdir,
1164 * Yes, XFS uses the same method for rmdir and unlink.
1166 * There are some subtile differences deeper in the code,
1167 * but we use S_ISDIR to check for those.
1169 .rmdir = xfs_vn_unlink,
1170 .mknod = xfs_vn_mknod,
1171 .rename = xfs_vn_rename,
1172 .get_inode_acl = xfs_get_acl,
1173 .set_acl = xfs_set_acl,
1174 .getattr = xfs_vn_getattr,
1175 .setattr = xfs_vn_setattr,
1176 .listxattr = xfs_vn_listxattr,
1177 .update_time = xfs_vn_update_time,
1178 .tmpfile = xfs_vn_tmpfile,
1179 .fileattr_get = xfs_fileattr_get,
1180 .fileattr_set = xfs_fileattr_set,
1183 static const struct inode_operations xfs_symlink_inode_operations = {
1184 .get_link = xfs_vn_get_link,
1185 .getattr = xfs_vn_getattr,
1186 .setattr = xfs_vn_setattr,
1187 .listxattr = xfs_vn_listxattr,
1188 .update_time = xfs_vn_update_time,
1191 /* Figure out if this file actually supports DAX. */
1193 xfs_inode_supports_dax(
1194 struct xfs_inode *ip)
1196 struct xfs_mount *mp = ip->i_mount;
1198 /* Only supported on regular files. */
1199 if (!S_ISREG(VFS_I(ip)->i_mode))
1202 /* Block size must match page size */
1203 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1206 /* Device has to support DAX too. */
1207 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1211 xfs_inode_should_enable_dax(
1212 struct xfs_inode *ip)
1214 if (!IS_ENABLED(CONFIG_FS_DAX))
1216 if (xfs_has_dax_never(ip->i_mount))
1218 if (!xfs_inode_supports_dax(ip))
1220 if (xfs_has_dax_always(ip->i_mount))
1222 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1228 xfs_diflags_to_iflags(
1229 struct xfs_inode *ip,
1232 struct inode *inode = VFS_I(ip);
1233 unsigned int xflags = xfs_ip2xflags(ip);
1234 unsigned int flags = 0;
1236 ASSERT(!(IS_DAX(inode) && init));
1238 if (xflags & FS_XFLAG_IMMUTABLE)
1239 flags |= S_IMMUTABLE;
1240 if (xflags & FS_XFLAG_APPEND)
1242 if (xflags & FS_XFLAG_SYNC)
1244 if (xflags & FS_XFLAG_NOATIME)
1246 if (init && xfs_inode_should_enable_dax(ip))
1250 * S_DAX can only be set during inode initialization and is never set by
1251 * the VFS, so we cannot mask off S_DAX in i_flags.
1253 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1254 inode->i_flags |= flags;
1258 * Initialize the Linux inode.
1260 * When reading existing inodes from disk this is called directly from xfs_iget,
1261 * when creating a new inode it is called from xfs_init_new_inode after setting
1262 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1263 * leave it up to the caller to deal with unlocking the inode appropriately.
1267 struct xfs_inode *ip)
1269 struct inode *inode = &ip->i_vnode;
1272 inode->i_ino = ip->i_ino;
1273 inode->i_state |= I_NEW;
1275 inode_sb_list_add(inode);
1276 /* make the inode look hashed for the writeback code */
1277 inode_fake_hash(inode);
1279 i_size_write(inode, ip->i_disk_size);
1280 xfs_diflags_to_iflags(ip, true);
1282 if (S_ISDIR(inode->i_mode)) {
1284 * We set the i_rwsem class here to avoid potential races with
1285 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1286 * after a filehandle lookup has already found the inode in
1287 * cache before it has been unlocked via unlock_new_inode().
1289 lockdep_set_class(&inode->i_rwsem,
1290 &inode->i_sb->s_type->i_mutex_dir_key);
1291 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1293 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1297 * Ensure all page cache allocations are done from GFP_NOFS context to
1298 * prevent direct reclaim recursion back into the filesystem and blowing
1299 * stacks or deadlocking.
1301 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1302 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1305 * For real-time inodes update the stable write flags to that of the RT
1306 * device instead of the data device.
1308 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1309 xfs_update_stable_writes(ip);
1312 * If there is no attribute fork no ACL can exist on this inode,
1313 * and it can't have any file capabilities attached to it either.
1315 if (!xfs_inode_has_attr_fork(ip)) {
1316 inode_has_no_xattr(inode);
1317 cache_no_acl(inode);
1323 struct xfs_inode *ip)
1325 struct inode *inode = &ip->i_vnode;
1327 switch (inode->i_mode & S_IFMT) {
1329 inode->i_op = &xfs_inode_operations;
1330 inode->i_fop = &xfs_file_operations;
1332 inode->i_mapping->a_ops = &xfs_dax_aops;
1334 inode->i_mapping->a_ops = &xfs_address_space_operations;
1337 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1338 inode->i_op = &xfs_dir_ci_inode_operations;
1340 inode->i_op = &xfs_dir_inode_operations;
1341 inode->i_fop = &xfs_dir_file_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);