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)
160 if (xfs_has_parent(XFS_I(dir)->i_mount))
168 struct mnt_idmap *idmap,
170 struct dentry *dentry,
173 struct file *tmpfile) /* unnamed file */
176 struct xfs_inode *ip = NULL;
177 struct posix_acl *default_acl, *acl;
178 struct xfs_name name;
182 * Irix uses Missed'em'V split, but doesn't want to see
183 * the upper 5 bits of (14bit) major.
185 if (S_ISCHR(mode) || S_ISBLK(mode)) {
186 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
192 error = posix_acl_create(dir, &mode, &default_acl, &acl);
196 /* Verify mode is valid also for tmpfile case */
197 error = xfs_dentry_mode_to_name(&name, dentry, mode);
202 error = xfs_create(idmap, XFS_I(dir), &name, mode, rdev,
203 xfs_create_need_xattr(dir, default_acl, acl),
206 bool init_xattrs = false;
209 * If this temporary file will be linkable, set up the file
210 * with an attr fork to receive a parent pointer.
212 if (!(tmpfile->f_flags & O_EXCL) &&
213 xfs_has_parent(XFS_I(dir)->i_mount))
216 error = xfs_create_tmpfile(idmap, XFS_I(dir), mode,
224 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
226 goto out_cleanup_inode;
229 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
231 goto out_cleanup_inode;
234 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
236 goto out_cleanup_inode;
243 * The VFS requires that any inode fed to d_tmpfile must have
244 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
245 * However, we created the temp file with nlink == 0 because
246 * we're not allowed to put an inode with nlink > 0 on the
247 * unlinked list. Therefore we have to set nlink to 1 so that
248 * d_tmpfile can immediately set it back to zero.
251 d_tmpfile(tmpfile, inode);
253 d_instantiate(dentry, inode);
255 xfs_finish_inode_setup(ip);
258 posix_acl_release(default_acl);
259 posix_acl_release(acl);
263 xfs_finish_inode_setup(ip);
265 xfs_cleanup_inode(dir, inode, dentry);
272 struct mnt_idmap *idmap,
274 struct dentry *dentry,
278 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
283 struct mnt_idmap *idmap,
285 struct dentry *dentry,
289 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
294 struct mnt_idmap *idmap,
296 struct dentry *dentry,
299 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
302 STATIC struct dentry *
305 struct dentry *dentry,
309 struct xfs_inode *cip;
310 struct xfs_name name;
313 if (dentry->d_name.len >= MAXNAMELEN)
314 return ERR_PTR(-ENAMETOOLONG);
316 xfs_dentry_to_name(&name, dentry);
317 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
320 else if (likely(error == -ENOENT))
323 inode = ERR_PTR(error);
324 return d_splice_alias(inode, dentry);
327 STATIC struct dentry *
330 struct dentry *dentry,
333 struct xfs_inode *ip;
334 struct xfs_name xname;
335 struct xfs_name ci_name;
339 if (dentry->d_name.len >= MAXNAMELEN)
340 return ERR_PTR(-ENAMETOOLONG);
342 xfs_dentry_to_name(&xname, dentry);
343 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
344 if (unlikely(error)) {
345 if (unlikely(error != -ENOENT))
346 return ERR_PTR(error);
348 * call d_add(dentry, NULL) here when d_drop_negative_children
349 * is called in xfs_vn_mknod (ie. allow negative dentries
350 * with CI filesystems).
355 /* if exact match, just splice and exit */
357 return d_splice_alias(VFS_I(ip), dentry);
359 /* else case-insensitive match... */
360 dname.name = ci_name.name;
361 dname.len = ci_name.len;
362 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
369 struct dentry *old_dentry,
371 struct dentry *dentry)
373 struct inode *inode = d_inode(old_dentry);
374 struct xfs_name name;
377 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
381 if (IS_PRIVATE(inode))
384 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
389 d_instantiate(dentry, inode);
396 struct dentry *dentry)
398 struct xfs_name name;
401 xfs_dentry_to_name(&name, dentry);
403 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
408 * With unlink, the VFS makes the dentry "negative": no inode,
409 * but still hashed. This is incompatible with case-insensitive
410 * mode, so invalidate (unhash) the dentry in CI-mode.
412 if (xfs_has_asciici(XFS_M(dir->i_sb)))
413 d_invalidate(dentry);
419 struct mnt_idmap *idmap,
421 struct dentry *dentry,
425 struct xfs_inode *cip = NULL;
426 struct xfs_name name;
431 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
432 error = xfs_dentry_mode_to_name(&name, dentry, mode);
436 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
442 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
444 goto out_cleanup_inode;
448 d_instantiate(dentry, inode);
449 xfs_finish_inode_setup(cip);
453 xfs_finish_inode_setup(cip);
454 xfs_cleanup_inode(dir, inode, dentry);
462 struct mnt_idmap *idmap,
464 struct dentry *odentry,
466 struct dentry *ndentry,
469 struct inode *new_inode = d_inode(ndentry);
472 struct xfs_name oname;
473 struct xfs_name nname;
475 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
478 /* if we are exchanging files, we need to set i_mode of both files */
479 if (flags & RENAME_EXCHANGE)
480 omode = d_inode(ndentry)->i_mode;
482 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
483 if (omode && unlikely(error))
486 error = xfs_dentry_mode_to_name(&nname, ndentry,
487 d_inode(odentry)->i_mode);
491 return xfs_rename(idmap, XFS_I(odir), &oname,
492 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
493 new_inode ? XFS_I(new_inode) : NULL, flags);
497 * careful here - this function can get called recursively, so
498 * we need to be very careful about how much stack we use.
499 * uio is kmalloced for this reason...
503 struct dentry *dentry,
505 struct delayed_call *done)
511 return ERR_PTR(-ECHILD);
513 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
517 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
521 set_delayed_call(done, kfree_link, link);
527 return ERR_PTR(error);
532 struct xfs_inode *ip)
534 struct xfs_mount *mp = ip->i_mount;
537 * If the file blocks are being allocated from a realtime volume, then
538 * always return the realtime extent size.
540 if (XFS_IS_REALTIME_INODE(ip))
541 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
544 * Allow large block sizes to be reported to userspace programs if the
545 * "largeio" mount option is used.
547 * If compatibility mode is specified, simply return the basic unit of
548 * caching so that we don't get inefficient read/modify/write I/O from
549 * user apps. Otherwise....
551 * If the underlying volume is a stripe, then return the stripe width in
552 * bytes as the recommended I/O size. It is not a stripe and we've set a
553 * default buffered I/O size, return that, otherwise return the compat
556 if (xfs_has_large_iosize(mp)) {
558 return XFS_FSB_TO_B(mp, mp->m_swidth);
559 if (xfs_has_allocsize(mp))
560 return 1U << mp->m_allocsize_log;
568 struct mnt_idmap *idmap,
569 const struct path *path,
572 unsigned int query_flags)
574 struct inode *inode = d_inode(path->dentry);
575 struct xfs_inode *ip = XFS_I(inode);
576 struct xfs_mount *mp = ip->i_mount;
577 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
578 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
580 trace_xfs_getattr(ip);
582 if (xfs_is_shutdown(mp))
585 stat->size = XFS_ISIZE(ip);
586 stat->dev = inode->i_sb->s_dev;
587 stat->mode = inode->i_mode;
588 stat->nlink = inode->i_nlink;
589 stat->uid = vfsuid_into_kuid(vfsuid);
590 stat->gid = vfsgid_into_kgid(vfsgid);
591 stat->ino = ip->i_ino;
592 stat->atime = inode_get_atime(inode);
593 stat->mtime = inode_get_mtime(inode);
594 stat->ctime = inode_get_ctime(inode);
595 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
597 if (xfs_has_v3inodes(mp)) {
598 if (request_mask & STATX_BTIME) {
599 stat->result_mask |= STATX_BTIME;
600 stat->btime = ip->i_crtime;
604 if ((request_mask & STATX_CHANGE_COOKIE) && IS_I_VERSION(inode)) {
605 stat->change_cookie = inode_query_iversion(inode);
606 stat->result_mask |= STATX_CHANGE_COOKIE;
610 * Note: If you add another clause to set an attribute flag, please
611 * update attributes_mask below.
613 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
614 stat->attributes |= STATX_ATTR_IMMUTABLE;
615 if (ip->i_diflags & XFS_DIFLAG_APPEND)
616 stat->attributes |= STATX_ATTR_APPEND;
617 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
618 stat->attributes |= STATX_ATTR_NODUMP;
620 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
624 switch (inode->i_mode & S_IFMT) {
627 stat->blksize = BLKDEV_IOSIZE;
628 stat->rdev = inode->i_rdev;
631 if (request_mask & STATX_DIOALIGN) {
632 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
633 struct block_device *bdev = target->bt_bdev;
635 stat->result_mask |= STATX_DIOALIGN;
636 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
637 stat->dio_offset_align = bdev_logical_block_size(bdev);
641 stat->blksize = xfs_stat_blksize(ip);
651 struct mnt_idmap *idmap,
652 struct dentry *dentry,
655 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
657 if (xfs_is_readonly(mp))
660 if (xfs_is_shutdown(mp))
663 return setattr_prepare(idmap, dentry, iattr);
667 * Set non-size attributes of an inode.
669 * Caution: The caller of this function is responsible for calling
670 * setattr_prepare() or otherwise verifying the change is fine.
674 struct mnt_idmap *idmap,
675 struct dentry *dentry,
676 struct xfs_inode *ip,
679 xfs_mount_t *mp = ip->i_mount;
680 struct inode *inode = VFS_I(ip);
681 int mask = iattr->ia_valid;
684 kuid_t uid = GLOBAL_ROOT_UID;
685 kgid_t gid = GLOBAL_ROOT_GID;
686 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
687 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
689 ASSERT((mask & ATTR_SIZE) == 0);
692 * If disk quotas is on, we make sure that the dquots do exist on disk,
693 * before we start any other transactions. Trying to do this later
694 * is messy. We don't care to take a readlock to look at the ids
695 * in inode here, because we can't hold it across the trans_reserve.
696 * If the IDs do change before we take the ilock, we're covered
697 * because the i_*dquot fields will get updated anyway.
699 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
702 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
703 uid = from_vfsuid(idmap, i_user_ns(inode),
705 qflags |= XFS_QMOPT_UQUOTA;
709 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
710 gid = from_vfsgid(idmap, i_user_ns(inode),
712 qflags |= XFS_QMOPT_GQUOTA;
718 * We take a reference when we initialize udqp and gdqp,
719 * so it is important that we never blindly double trip on
720 * the same variable. See xfs_create() for an example.
722 ASSERT(udqp == NULL);
723 ASSERT(gdqp == NULL);
724 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
725 qflags, &udqp, &gdqp, NULL);
730 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
731 has_capability_noaudit(current, CAP_FOWNER), &tp);
736 * Register quota modifications in the transaction. Must be the owner
737 * or privileged. These IDs could have changed since we last looked at
738 * them. But, we're assured that if the ownership did change while we
739 * didn't have the inode locked, inode's dquot(s) would have changed
742 if (XFS_IS_UQUOTA_ON(mp) &&
743 i_uid_needs_update(idmap, iattr, inode)) {
745 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
747 if (XFS_IS_GQUOTA_ON(mp) &&
748 i_gid_needs_update(idmap, iattr, inode)) {
749 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
751 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
754 setattr_copy(idmap, inode, iattr);
755 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
757 XFS_STATS_INC(mp, xs_ig_attrchg);
759 if (xfs_has_wsync(mp))
760 xfs_trans_set_sync(tp);
761 error = xfs_trans_commit(tp);
764 * Release any dquot(s) the inode had kept before chown.
766 xfs_qm_dqrele(old_udqp);
767 xfs_qm_dqrele(old_gdqp);
775 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
776 * update. We could avoid this with linked transactions
777 * and passing down the transaction pointer all the way
778 * to attr_set. No previous user of the generic
779 * Posix ACL code seems to care about this issue either.
781 if (mask & ATTR_MODE) {
782 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
796 * Truncate file. Must have write permission and not be a directory.
798 * Caution: The caller of this function is responsible for calling
799 * setattr_prepare() or otherwise verifying the change is fine.
803 struct mnt_idmap *idmap,
804 struct dentry *dentry,
805 struct xfs_inode *ip,
808 struct xfs_mount *mp = ip->i_mount;
809 struct inode *inode = VFS_I(ip);
810 xfs_off_t oldsize, newsize;
811 struct xfs_trans *tp;
814 bool did_zeroing = false;
816 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
817 ASSERT(S_ISREG(inode->i_mode));
818 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
819 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
821 oldsize = inode->i_size;
822 newsize = iattr->ia_size;
825 * Short circuit the truncate case for zero length files.
827 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
828 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
832 * Use the regular setattr path to update the timestamps.
834 iattr->ia_valid &= ~ATTR_SIZE;
835 return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
839 * Make sure that the dquots are attached to the inode.
841 error = xfs_qm_dqattach(ip);
846 * Wait for all direct I/O to complete.
848 inode_dio_wait(inode);
851 * File data changes must be complete before we start the transaction to
852 * modify the inode. This needs to be done before joining the inode to
853 * the transaction because the inode cannot be unlocked once it is a
854 * part of the transaction.
856 * Start with zeroing any data beyond EOF that we may expose on file
857 * extension, or zeroing out the rest of the block on a downward
860 if (newsize > oldsize) {
861 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
862 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
866 * iomap won't detect a dirty page over an unwritten block (or a
867 * cow block over a hole) and subsequently skips zeroing the
868 * newly post-EOF portion of the page. Flush the new EOF to
869 * convert the block before the pagecache truncate.
871 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
875 error = xfs_truncate_page(ip, newsize, &did_zeroing);
882 * We've already locked out new page faults, so now we can safely remove
883 * pages from the page cache knowing they won't get refaulted until we
884 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
885 * complete. The truncate_setsize() call also cleans partial EOF page
886 * PTEs on extending truncates and hence ensures sub-page block size
887 * filesystems are correctly handled, too.
889 * We have to do all the page cache truncate work outside the
890 * transaction context as the "lock" order is page lock->log space
891 * reservation as defined by extent allocation in the writeback path.
892 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
893 * having already truncated the in-memory version of the file (i.e. made
894 * user visible changes). There's not much we can do about this, except
895 * to hope that the caller sees ENOMEM and retries the truncate
898 * And we update in-core i_size and truncate page cache beyond newsize
899 * before writeback the [i_disk_size, newsize] range, so we're
900 * guaranteed not to write stale data past the new EOF on truncate down.
902 truncate_setsize(inode, newsize);
905 * We are going to log the inode size change in this transaction so
906 * any previous writes that are beyond the on disk EOF and the new
907 * EOF that have not been written out need to be written here. If we
908 * do not write the data out, we expose ourselves to the null files
909 * problem. Note that this includes any block zeroing we did above;
910 * otherwise those blocks may not be zeroed after a crash.
913 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
914 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
915 ip->i_disk_size, newsize - 1);
920 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
924 lock_flags |= XFS_ILOCK_EXCL;
925 xfs_ilock(ip, XFS_ILOCK_EXCL);
926 xfs_trans_ijoin(tp, ip, 0);
929 * Only change the c/mtime if we are changing the size or we are
930 * explicitly asked to change it. This handles the semantic difference
931 * between truncate() and ftruncate() as implemented in the VFS.
933 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
934 * special case where we need to update the times despite not having
935 * these flags set. For all other operations the VFS set these flags
936 * explicitly if it wants a timestamp update.
938 if (newsize != oldsize &&
939 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
940 iattr->ia_ctime = iattr->ia_mtime =
942 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
946 * The first thing we do is set the size to new_size permanently on
947 * disk. This way we don't have to worry about anyone ever being able
948 * to look at the data being freed even in the face of a crash.
949 * What we're getting around here is the case where we free a block, it
950 * is allocated to another file, it is written to, and then we crash.
951 * If the new data gets written to the file but the log buffers
952 * containing the free and reallocation don't, then we'd end up with
953 * garbage in the blocks being freed. As long as we make the new size
954 * permanent before actually freeing any blocks it doesn't matter if
955 * they get written to.
957 ip->i_disk_size = newsize;
958 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
960 if (newsize <= oldsize) {
961 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
963 goto out_trans_cancel;
966 * Truncated "down", so we're removing references to old data
967 * here - if we delay flushing for a long time, we expose
968 * ourselves unduly to the notorious NULL files problem. So,
969 * we mark this inode and flush it when the file is closed,
970 * and do not wait the usual (long) time for writeout.
972 xfs_iflags_set(ip, XFS_ITRUNCATED);
974 /* A truncate down always removes post-EOF blocks. */
975 xfs_inode_clear_eofblocks_tag(ip);
978 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
979 setattr_copy(idmap, inode, iattr);
980 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
982 XFS_STATS_INC(mp, xs_ig_attrchg);
984 if (xfs_has_wsync(mp))
985 xfs_trans_set_sync(tp);
987 error = xfs_trans_commit(tp);
990 xfs_iunlock(ip, lock_flags);
994 xfs_trans_cancel(tp);
1000 struct mnt_idmap *idmap,
1001 struct dentry *dentry,
1002 struct iattr *iattr)
1004 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1007 trace_xfs_setattr(ip);
1009 error = xfs_vn_change_ok(idmap, dentry, iattr);
1012 return xfs_setattr_size(idmap, dentry, ip, iattr);
1017 struct mnt_idmap *idmap,
1018 struct dentry *dentry,
1019 struct iattr *iattr)
1021 struct inode *inode = d_inode(dentry);
1022 struct xfs_inode *ip = XFS_I(inode);
1025 if (iattr->ia_valid & ATTR_SIZE) {
1028 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1029 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1031 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1033 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1037 error = xfs_vn_setattr_size(idmap, dentry, iattr);
1038 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1040 trace_xfs_setattr(ip);
1042 error = xfs_vn_change_ok(idmap, dentry, iattr);
1044 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1052 struct inode *inode,
1055 struct xfs_inode *ip = XFS_I(inode);
1056 struct xfs_mount *mp = ip->i_mount;
1057 int log_flags = XFS_ILOG_TIMESTAMP;
1058 struct xfs_trans *tp;
1060 struct timespec64 now;
1062 trace_xfs_update_time(ip);
1064 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1065 if (!((flags & S_VERSION) &&
1066 inode_maybe_inc_iversion(inode, false))) {
1067 generic_update_time(inode, flags);
1071 /* Capture the iversion update that just occurred */
1072 log_flags |= XFS_ILOG_CORE;
1075 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1079 xfs_ilock(ip, XFS_ILOCK_EXCL);
1080 if (flags & (S_CTIME|S_MTIME))
1081 now = inode_set_ctime_current(inode);
1083 now = current_time(inode);
1085 if (flags & S_MTIME)
1086 inode_set_mtime_to_ts(inode, now);
1087 if (flags & S_ATIME)
1088 inode_set_atime_to_ts(inode, now);
1090 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1091 xfs_trans_log_inode(tp, ip, log_flags);
1092 return xfs_trans_commit(tp);
1097 struct inode *inode,
1098 struct fiemap_extent_info *fieinfo,
1104 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1105 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1106 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1107 error = iomap_fiemap(inode, fieinfo, start, length,
1108 &xfs_xattr_iomap_ops);
1110 error = iomap_fiemap(inode, fieinfo, start, length,
1111 &xfs_read_iomap_ops);
1113 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1120 struct mnt_idmap *idmap,
1125 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1127 return finish_open_simple(file, err);
1130 static const struct inode_operations xfs_inode_operations = {
1131 .get_inode_acl = xfs_get_acl,
1132 .set_acl = xfs_set_acl,
1133 .getattr = xfs_vn_getattr,
1134 .setattr = xfs_vn_setattr,
1135 .listxattr = xfs_vn_listxattr,
1136 .fiemap = xfs_vn_fiemap,
1137 .update_time = xfs_vn_update_time,
1138 .fileattr_get = xfs_fileattr_get,
1139 .fileattr_set = xfs_fileattr_set,
1142 static const struct inode_operations xfs_dir_inode_operations = {
1143 .create = xfs_vn_create,
1144 .lookup = xfs_vn_lookup,
1145 .link = xfs_vn_link,
1146 .unlink = xfs_vn_unlink,
1147 .symlink = xfs_vn_symlink,
1148 .mkdir = xfs_vn_mkdir,
1150 * Yes, XFS uses the same method for rmdir and unlink.
1152 * There are some subtile differences deeper in the code,
1153 * but we use S_ISDIR to check for those.
1155 .rmdir = xfs_vn_unlink,
1156 .mknod = xfs_vn_mknod,
1157 .rename = xfs_vn_rename,
1158 .get_inode_acl = xfs_get_acl,
1159 .set_acl = xfs_set_acl,
1160 .getattr = xfs_vn_getattr,
1161 .setattr = xfs_vn_setattr,
1162 .listxattr = xfs_vn_listxattr,
1163 .update_time = xfs_vn_update_time,
1164 .tmpfile = xfs_vn_tmpfile,
1165 .fileattr_get = xfs_fileattr_get,
1166 .fileattr_set = xfs_fileattr_set,
1169 static const struct inode_operations xfs_dir_ci_inode_operations = {
1170 .create = xfs_vn_create,
1171 .lookup = xfs_vn_ci_lookup,
1172 .link = xfs_vn_link,
1173 .unlink = xfs_vn_unlink,
1174 .symlink = xfs_vn_symlink,
1175 .mkdir = xfs_vn_mkdir,
1177 * Yes, XFS uses the same method for rmdir and unlink.
1179 * There are some subtile differences deeper in the code,
1180 * but we use S_ISDIR to check for those.
1182 .rmdir = xfs_vn_unlink,
1183 .mknod = xfs_vn_mknod,
1184 .rename = xfs_vn_rename,
1185 .get_inode_acl = xfs_get_acl,
1186 .set_acl = xfs_set_acl,
1187 .getattr = xfs_vn_getattr,
1188 .setattr = xfs_vn_setattr,
1189 .listxattr = xfs_vn_listxattr,
1190 .update_time = xfs_vn_update_time,
1191 .tmpfile = xfs_vn_tmpfile,
1192 .fileattr_get = xfs_fileattr_get,
1193 .fileattr_set = xfs_fileattr_set,
1196 static const struct inode_operations xfs_symlink_inode_operations = {
1197 .get_link = xfs_vn_get_link,
1198 .getattr = xfs_vn_getattr,
1199 .setattr = xfs_vn_setattr,
1200 .listxattr = xfs_vn_listxattr,
1201 .update_time = xfs_vn_update_time,
1204 /* Figure out if this file actually supports DAX. */
1206 xfs_inode_supports_dax(
1207 struct xfs_inode *ip)
1209 struct xfs_mount *mp = ip->i_mount;
1211 /* Only supported on regular files. */
1212 if (!S_ISREG(VFS_I(ip)->i_mode))
1215 /* Block size must match page size */
1216 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1219 /* Device has to support DAX too. */
1220 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1224 xfs_inode_should_enable_dax(
1225 struct xfs_inode *ip)
1227 if (!IS_ENABLED(CONFIG_FS_DAX))
1229 if (xfs_has_dax_never(ip->i_mount))
1231 if (!xfs_inode_supports_dax(ip))
1233 if (xfs_has_dax_always(ip->i_mount))
1235 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1241 xfs_diflags_to_iflags(
1242 struct xfs_inode *ip,
1245 struct inode *inode = VFS_I(ip);
1246 unsigned int xflags = xfs_ip2xflags(ip);
1247 unsigned int flags = 0;
1249 ASSERT(!(IS_DAX(inode) && init));
1251 if (xflags & FS_XFLAG_IMMUTABLE)
1252 flags |= S_IMMUTABLE;
1253 if (xflags & FS_XFLAG_APPEND)
1255 if (xflags & FS_XFLAG_SYNC)
1257 if (xflags & FS_XFLAG_NOATIME)
1259 if (init && xfs_inode_should_enable_dax(ip))
1263 * S_DAX can only be set during inode initialization and is never set by
1264 * the VFS, so we cannot mask off S_DAX in i_flags.
1266 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1267 inode->i_flags |= flags;
1271 * Initialize the Linux inode.
1273 * When reading existing inodes from disk this is called directly from xfs_iget,
1274 * when creating a new inode it is called from xfs_init_new_inode after setting
1275 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1276 * leave it up to the caller to deal with unlocking the inode appropriately.
1280 struct xfs_inode *ip)
1282 struct inode *inode = &ip->i_vnode;
1285 inode->i_ino = ip->i_ino;
1286 inode->i_state |= I_NEW;
1288 inode_sb_list_add(inode);
1289 /* make the inode look hashed for the writeback code */
1290 inode_fake_hash(inode);
1292 i_size_write(inode, ip->i_disk_size);
1293 xfs_diflags_to_iflags(ip, true);
1295 if (S_ISDIR(inode->i_mode)) {
1297 * We set the i_rwsem class here to avoid potential races with
1298 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1299 * after a filehandle lookup has already found the inode in
1300 * cache before it has been unlocked via unlock_new_inode().
1302 lockdep_set_class(&inode->i_rwsem,
1303 &inode->i_sb->s_type->i_mutex_dir_key);
1304 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1306 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1310 * Ensure all page cache allocations are done from GFP_NOFS context to
1311 * prevent direct reclaim recursion back into the filesystem and blowing
1312 * stacks or deadlocking.
1314 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1315 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1318 * For real-time inodes update the stable write flags to that of the RT
1319 * device instead of the data device.
1321 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1322 xfs_update_stable_writes(ip);
1325 * If there is no attribute fork no ACL can exist on this inode,
1326 * and it can't have any file capabilities attached to it either.
1328 if (!xfs_inode_has_attr_fork(ip)) {
1329 inode_has_no_xattr(inode);
1330 cache_no_acl(inode);
1336 struct xfs_inode *ip)
1338 struct inode *inode = &ip->i_vnode;
1340 switch (inode->i_mode & S_IFMT) {
1342 inode->i_op = &xfs_inode_operations;
1343 inode->i_fop = &xfs_file_operations;
1345 inode->i_mapping->a_ops = &xfs_dax_aops;
1347 inode->i_mapping->a_ops = &xfs_address_space_operations;
1350 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1351 inode->i_op = &xfs_dir_ci_inode_operations;
1353 inode->i_op = &xfs_dir_inode_operations;
1354 inode->i_fop = &xfs_dir_file_operations;
1357 inode->i_op = &xfs_symlink_inode_operations;
1360 inode->i_op = &xfs_inode_operations;
1361 init_special_inode(inode, inode->i_mode, inode->i_rdev);