1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
15 #include "xfs_quota.h"
17 #include "xfs_trans.h"
18 #include "xfs_trace.h"
19 #include "xfs_icache.h"
20 #include "xfs_symlink.h"
22 #include "xfs_iomap.h"
23 #include "xfs_error.h"
25 #include <linux/posix_acl.h>
26 #include <linux/security.h>
27 #include <linux/iversion.h>
28 #include <linux/fiemap.h>
31 * Directories have different lock order w.r.t. mmap_lock compared to regular
32 * files. This is due to readdir potentially triggering page faults on a user
33 * buffer inside filldir(), and this happens with the ilock on the directory
34 * held. For regular files, the lock order is the other way around - the
35 * mmap_lock is taken during the page fault, and then we lock the ilock to do
36 * block mapping. Hence we need a different class for the directory ilock so
37 * that lockdep can tell them apart.
39 static struct lock_class_key xfs_nondir_ilock_class;
40 static struct lock_class_key xfs_dir_ilock_class;
45 const struct xattr *xattr_array,
48 const struct xattr *xattr;
49 struct xfs_inode *ip = XFS_I(inode);
52 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
53 struct xfs_da_args args = {
55 .attr_filter = XFS_ATTR_SECURE,
57 .namelen = strlen(xattr->name),
58 .value = xattr->value,
59 .valuelen = xattr->value_len,
61 error = xfs_attr_set(&args);
69 * Hook in SELinux. This is not quite correct yet, what we really need
70 * here (as we do for default ACLs) is a mechanism by which creation of
71 * these attrs can be journalled at inode creation time (along with the
72 * inode, of course, such that log replay can't cause these to be lost).
79 const struct qstr *qstr)
81 return security_inode_init_security(inode, dir, qstr,
82 &xfs_initxattrs, NULL);
87 struct xfs_name *namep,
88 struct dentry *dentry)
90 namep->name = dentry->d_name.name;
91 namep->len = dentry->d_name.len;
92 namep->type = XFS_DIR3_FT_UNKNOWN;
96 xfs_dentry_mode_to_name(
97 struct xfs_name *namep,
98 struct dentry *dentry,
101 namep->name = dentry->d_name.name;
102 namep->len = dentry->d_name.len;
103 namep->type = xfs_mode_to_ftype(mode);
105 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
106 return -EFSCORRUPTED;
115 struct dentry *dentry)
117 struct xfs_name teardown;
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
124 xfs_dentry_to_name(&teardown, dentry);
126 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
130 * Check to see if we are likely to need an extended attribute to be added to
131 * the inode we are about to allocate. This allows the attribute fork to be
132 * created during the inode allocation, reducing the number of transactions we
133 * need to do in this fast path.
135 * The security checks are optimistic, but not guaranteed. The two LSMs that
136 * require xattrs to be added here (selinux and smack) are also the only two
137 * LSMs that add a sb->s_security structure to the superblock. Hence if security
138 * is enabled and sb->s_security is set, we have a pretty good idea that we are
139 * going to be asked to add a security xattr immediately after allocating the
140 * xfs inode and instantiating the VFS inode.
143 xfs_create_need_xattr(
145 struct posix_acl *default_acl,
146 struct posix_acl *acl)
152 #if IS_ENABLED(CONFIG_SECURITY)
153 if (dir->i_sb->s_security)
162 struct user_namespace *mnt_userns,
164 struct dentry *dentry,
167 bool tmpfile) /* unnamed file */
170 struct xfs_inode *ip = NULL;
171 struct posix_acl *default_acl, *acl;
172 struct xfs_name name;
176 * Irix uses Missed'em'V split, but doesn't want to see
177 * the upper 5 bits of (14bit) major.
179 if (S_ISCHR(mode) || S_ISBLK(mode)) {
180 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
186 error = posix_acl_create(dir, &mode, &default_acl, &acl);
190 /* Verify mode is valid also for tmpfile case */
191 error = xfs_dentry_mode_to_name(&name, dentry, mode);
196 error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev,
197 xfs_create_need_xattr(dir, default_acl, acl),
200 error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip);
207 error = xfs_init_security(inode, dir, &dentry->d_name);
209 goto out_cleanup_inode;
211 #ifdef CONFIG_XFS_POSIX_ACL
213 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
215 goto out_cleanup_inode;
218 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
220 goto out_cleanup_inode;
228 * The VFS requires that any inode fed to d_tmpfile must have
229 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
230 * However, we created the temp file with nlink == 0 because
231 * we're not allowed to put an inode with nlink > 0 on the
232 * unlinked list. Therefore we have to set nlink to 1 so that
233 * d_tmpfile can immediately set it back to zero.
236 d_tmpfile(dentry, inode);
238 d_instantiate(dentry, inode);
240 xfs_finish_inode_setup(ip);
243 posix_acl_release(default_acl);
244 posix_acl_release(acl);
248 xfs_finish_inode_setup(ip);
250 xfs_cleanup_inode(dir, inode, dentry);
257 struct user_namespace *mnt_userns,
259 struct dentry *dentry,
263 return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false);
268 struct user_namespace *mnt_userns,
270 struct dentry *dentry,
274 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false);
279 struct user_namespace *mnt_userns,
281 struct dentry *dentry,
284 return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0,
288 STATIC struct dentry *
291 struct dentry *dentry,
295 struct xfs_inode *cip;
296 struct xfs_name name;
299 if (dentry->d_name.len >= MAXNAMELEN)
300 return ERR_PTR(-ENAMETOOLONG);
302 xfs_dentry_to_name(&name, dentry);
303 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
306 else if (likely(error == -ENOENT))
309 inode = ERR_PTR(error);
310 return d_splice_alias(inode, dentry);
313 STATIC struct dentry *
316 struct dentry *dentry,
319 struct xfs_inode *ip;
320 struct xfs_name xname;
321 struct xfs_name ci_name;
325 if (dentry->d_name.len >= MAXNAMELEN)
326 return ERR_PTR(-ENAMETOOLONG);
328 xfs_dentry_to_name(&xname, dentry);
329 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
330 if (unlikely(error)) {
331 if (unlikely(error != -ENOENT))
332 return ERR_PTR(error);
334 * call d_add(dentry, NULL) here when d_drop_negative_children
335 * is called in xfs_vn_mknod (ie. allow negative dentries
336 * with CI filesystems).
341 /* if exact match, just splice and exit */
343 return d_splice_alias(VFS_I(ip), dentry);
345 /* else case-insensitive match... */
346 dname.name = ci_name.name;
347 dname.len = ci_name.len;
348 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
349 kmem_free(ci_name.name);
355 struct dentry *old_dentry,
357 struct dentry *dentry)
359 struct inode *inode = d_inode(old_dentry);
360 struct xfs_name name;
363 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
367 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
372 d_instantiate(dentry, inode);
379 struct dentry *dentry)
381 struct xfs_name name;
384 xfs_dentry_to_name(&name, dentry);
386 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
391 * With unlink, the VFS makes the dentry "negative": no inode,
392 * but still hashed. This is incompatible with case-insensitive
393 * mode, so invalidate (unhash) the dentry in CI-mode.
395 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
396 d_invalidate(dentry);
402 struct user_namespace *mnt_userns,
404 struct dentry *dentry,
408 struct xfs_inode *cip = NULL;
409 struct xfs_name name;
414 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
415 error = xfs_dentry_mode_to_name(&name, dentry, mode);
419 error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip);
425 error = xfs_init_security(inode, dir, &dentry->d_name);
427 goto out_cleanup_inode;
431 d_instantiate(dentry, inode);
432 xfs_finish_inode_setup(cip);
436 xfs_finish_inode_setup(cip);
437 xfs_cleanup_inode(dir, inode, dentry);
445 struct user_namespace *mnt_userns,
447 struct dentry *odentry,
449 struct dentry *ndentry,
452 struct inode *new_inode = d_inode(ndentry);
455 struct xfs_name oname;
456 struct xfs_name nname;
458 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
461 /* if we are exchanging files, we need to set i_mode of both files */
462 if (flags & RENAME_EXCHANGE)
463 omode = d_inode(ndentry)->i_mode;
465 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
466 if (omode && unlikely(error))
469 error = xfs_dentry_mode_to_name(&nname, ndentry,
470 d_inode(odentry)->i_mode);
474 return xfs_rename(mnt_userns, XFS_I(odir), &oname,
475 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
476 new_inode ? XFS_I(new_inode) : NULL, flags);
480 * careful here - this function can get called recursively, so
481 * we need to be very careful about how much stack we use.
482 * uio is kmalloced for this reason...
486 struct dentry *dentry,
488 struct delayed_call *done)
494 return ERR_PTR(-ECHILD);
496 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
500 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
504 set_delayed_call(done, kfree_link, link);
510 return ERR_PTR(error);
514 xfs_vn_get_link_inline(
515 struct dentry *dentry,
517 struct delayed_call *done)
519 struct xfs_inode *ip = XFS_I(inode);
522 ASSERT(ip->i_df.if_format == XFS_DINODE_FMT_LOCAL);
525 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
528 link = ip->i_df.if_u1.if_data;
529 if (XFS_IS_CORRUPT(ip->i_mount, !link))
530 return ERR_PTR(-EFSCORRUPTED);
536 struct xfs_inode *ip)
538 struct xfs_mount *mp = ip->i_mount;
541 * If the file blocks are being allocated from a realtime volume, then
542 * always return the realtime extent size.
544 if (XFS_IS_REALTIME_INODE(ip))
545 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
548 * Allow large block sizes to be reported to userspace programs if the
549 * "largeio" mount option is used.
551 * If compatibility mode is specified, simply return the basic unit of
552 * caching so that we don't get inefficient read/modify/write I/O from
553 * user apps. Otherwise....
555 * If the underlying volume is a stripe, then return the stripe width in
556 * bytes as the recommended I/O size. It is not a stripe and we've set a
557 * default buffered I/O size, return that, otherwise return the compat
560 if (mp->m_flags & XFS_MOUNT_LARGEIO) {
562 return mp->m_swidth << mp->m_sb.sb_blocklog;
563 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
564 return 1U << mp->m_allocsize_log;
572 struct user_namespace *mnt_userns,
573 const struct path *path,
576 unsigned int query_flags)
578 struct inode *inode = d_inode(path->dentry);
579 struct xfs_inode *ip = XFS_I(inode);
580 struct xfs_mount *mp = ip->i_mount;
582 trace_xfs_getattr(ip);
584 if (XFS_FORCED_SHUTDOWN(mp))
587 stat->size = XFS_ISIZE(ip);
588 stat->dev = inode->i_sb->s_dev;
589 stat->mode = inode->i_mode;
590 stat->nlink = inode->i_nlink;
591 stat->uid = i_uid_into_mnt(mnt_userns, inode);
592 stat->gid = i_gid_into_mnt(mnt_userns, inode);
593 stat->ino = ip->i_ino;
594 stat->atime = inode->i_atime;
595 stat->mtime = inode->i_mtime;
596 stat->ctime = inode->i_ctime;
597 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
599 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
600 if (request_mask & STATX_BTIME) {
601 stat->result_mask |= STATX_BTIME;
602 stat->btime = ip->i_crtime;
607 * Note: If you add another clause to set an attribute flag, please
608 * update attributes_mask below.
610 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
611 stat->attributes |= STATX_ATTR_IMMUTABLE;
612 if (ip->i_diflags & XFS_DIFLAG_APPEND)
613 stat->attributes |= STATX_ATTR_APPEND;
614 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
615 stat->attributes |= STATX_ATTR_NODUMP;
617 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
621 switch (inode->i_mode & S_IFMT) {
624 stat->blksize = BLKDEV_IOSIZE;
625 stat->rdev = inode->i_rdev;
628 stat->blksize = xfs_stat_blksize(ip);
638 struct xfs_inode *ip,
641 struct inode *inode = VFS_I(ip);
642 umode_t mode = iattr->ia_mode;
644 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
646 inode->i_mode &= S_IFMT;
647 inode->i_mode |= mode & ~S_IFMT;
652 struct xfs_inode *ip,
655 struct inode *inode = VFS_I(ip);
657 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
659 if (iattr->ia_valid & ATTR_ATIME)
660 inode->i_atime = iattr->ia_atime;
661 if (iattr->ia_valid & ATTR_CTIME)
662 inode->i_ctime = iattr->ia_ctime;
663 if (iattr->ia_valid & ATTR_MTIME)
664 inode->i_mtime = iattr->ia_mtime;
669 struct user_namespace *mnt_userns,
670 struct dentry *dentry,
673 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
675 if (mp->m_flags & XFS_MOUNT_RDONLY)
678 if (XFS_FORCED_SHUTDOWN(mp))
681 return setattr_prepare(mnt_userns, dentry, iattr);
685 * Set non-size attributes of an inode.
687 * Caution: The caller of this function is responsible for calling
688 * setattr_prepare() or otherwise verifying the change is fine.
692 struct user_namespace *mnt_userns,
693 struct xfs_inode *ip,
696 xfs_mount_t *mp = ip->i_mount;
697 struct inode *inode = VFS_I(ip);
698 int mask = iattr->ia_valid;
701 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
702 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
703 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
704 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
706 ASSERT((mask & ATTR_SIZE) == 0);
709 * If disk quotas is on, we make sure that the dquots do exist on disk,
710 * before we start any other transactions. Trying to do this later
711 * is messy. We don't care to take a readlock to look at the ids
712 * in inode here, because we can't hold it across the trans_reserve.
713 * If the IDs do change before we take the ilock, we're covered
714 * because the i_*dquot fields will get updated anyway.
716 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
719 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
721 qflags |= XFS_QMOPT_UQUOTA;
725 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
727 qflags |= XFS_QMOPT_GQUOTA;
733 * We take a reference when we initialize udqp and gdqp,
734 * so it is important that we never blindly double trip on
735 * the same variable. See xfs_create() for an example.
737 ASSERT(udqp == NULL);
738 ASSERT(gdqp == NULL);
739 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
740 qflags, &udqp, &gdqp, NULL);
745 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
746 capable(CAP_FOWNER), &tp);
751 * Change file ownership. Must be the owner or privileged.
753 if (mask & (ATTR_UID|ATTR_GID)) {
755 * These IDs could have changed since we last looked at them.
756 * But, we're assured that if the ownership did change
757 * while we didn't have the inode locked, inode's dquot(s)
758 * would have changed also.
762 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
763 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
766 * CAP_FSETID overrides the following restrictions:
768 * The set-user-ID and set-group-ID bits of a file will be
769 * cleared upon successful return from chown()
771 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
772 !capable(CAP_FSETID))
773 inode->i_mode &= ~(S_ISUID|S_ISGID);
776 * Change the ownerships and register quota modifications
777 * in the transaction.
779 if (!uid_eq(iuid, uid)) {
780 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
781 ASSERT(mask & ATTR_UID);
783 olddquot1 = xfs_qm_vop_chown(tp, ip,
784 &ip->i_udquot, udqp);
788 if (!gid_eq(igid, gid)) {
789 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
790 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
791 !XFS_IS_PQUOTA_ON(mp));
792 ASSERT(mask & ATTR_GID);
794 olddquot2 = xfs_qm_vop_chown(tp, ip,
795 &ip->i_gdquot, gdqp);
801 if (mask & ATTR_MODE)
802 xfs_setattr_mode(ip, iattr);
803 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
804 xfs_setattr_time(ip, iattr);
806 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
808 XFS_STATS_INC(mp, xs_ig_attrchg);
810 if (mp->m_flags & XFS_MOUNT_WSYNC)
811 xfs_trans_set_sync(tp);
812 error = xfs_trans_commit(tp);
815 * Release any dquot(s) the inode had kept before chown.
817 xfs_qm_dqrele(olddquot1);
818 xfs_qm_dqrele(olddquot2);
826 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
827 * update. We could avoid this with linked transactions
828 * and passing down the transaction pointer all the way
829 * to attr_set. No previous user of the generic
830 * Posix ACL code seems to care about this issue either.
832 if (mask & ATTR_MODE) {
833 error = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
847 * Truncate file. Must have write permission and not be a directory.
849 * Caution: The caller of this function is responsible for calling
850 * setattr_prepare() or otherwise verifying the change is fine.
854 struct user_namespace *mnt_userns,
855 struct xfs_inode *ip,
858 struct xfs_mount *mp = ip->i_mount;
859 struct inode *inode = VFS_I(ip);
860 xfs_off_t oldsize, newsize;
861 struct xfs_trans *tp;
864 bool did_zeroing = false;
866 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
867 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
868 ASSERT(S_ISREG(inode->i_mode));
869 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
870 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
872 oldsize = inode->i_size;
873 newsize = iattr->ia_size;
876 * Short circuit the truncate case for zero length files.
878 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
879 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
883 * Use the regular setattr path to update the timestamps.
885 iattr->ia_valid &= ~ATTR_SIZE;
886 return xfs_setattr_nonsize(mnt_userns, ip, iattr);
890 * Make sure that the dquots are attached to the inode.
892 error = xfs_qm_dqattach(ip);
897 * Wait for all direct I/O to complete.
899 inode_dio_wait(inode);
902 * File data changes must be complete before we start the transaction to
903 * modify the inode. This needs to be done before joining the inode to
904 * the transaction because the inode cannot be unlocked once it is a
905 * part of the transaction.
907 * Start with zeroing any data beyond EOF that we may expose on file
908 * extension, or zeroing out the rest of the block on a downward
911 if (newsize > oldsize) {
912 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
913 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
914 &did_zeroing, &xfs_buffered_write_iomap_ops);
917 * iomap won't detect a dirty page over an unwritten block (or a
918 * cow block over a hole) and subsequently skips zeroing the
919 * newly post-EOF portion of the page. Flush the new EOF to
920 * convert the block before the pagecache truncate.
922 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
926 error = iomap_truncate_page(inode, newsize, &did_zeroing,
927 &xfs_buffered_write_iomap_ops);
934 * We've already locked out new page faults, so now we can safely remove
935 * pages from the page cache knowing they won't get refaulted until we
936 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
937 * complete. The truncate_setsize() call also cleans partial EOF page
938 * PTEs on extending truncates and hence ensures sub-page block size
939 * filesystems are correctly handled, too.
941 * We have to do all the page cache truncate work outside the
942 * transaction context as the "lock" order is page lock->log space
943 * reservation as defined by extent allocation in the writeback path.
944 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
945 * having already truncated the in-memory version of the file (i.e. made
946 * user visible changes). There's not much we can do about this, except
947 * to hope that the caller sees ENOMEM and retries the truncate
950 * And we update in-core i_size and truncate page cache beyond newsize
951 * before writeback the [i_disk_size, newsize] range, so we're
952 * guaranteed not to write stale data past the new EOF on truncate down.
954 truncate_setsize(inode, newsize);
957 * We are going to log the inode size change in this transaction so
958 * any previous writes that are beyond the on disk EOF and the new
959 * EOF that have not been written out need to be written here. If we
960 * do not write the data out, we expose ourselves to the null files
961 * problem. Note that this includes any block zeroing we did above;
962 * otherwise those blocks may not be zeroed after a crash.
965 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
966 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
967 ip->i_disk_size, newsize - 1);
972 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
976 lock_flags |= XFS_ILOCK_EXCL;
977 xfs_ilock(ip, XFS_ILOCK_EXCL);
978 xfs_trans_ijoin(tp, ip, 0);
981 * Only change the c/mtime if we are changing the size or we are
982 * explicitly asked to change it. This handles the semantic difference
983 * between truncate() and ftruncate() as implemented in the VFS.
985 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
986 * special case where we need to update the times despite not having
987 * these flags set. For all other operations the VFS set these flags
988 * explicitly if it wants a timestamp update.
990 if (newsize != oldsize &&
991 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
992 iattr->ia_ctime = iattr->ia_mtime =
994 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
998 * The first thing we do is set the size to new_size permanently on
999 * disk. This way we don't have to worry about anyone ever being able
1000 * to look at the data being freed even in the face of a crash.
1001 * What we're getting around here is the case where we free a block, it
1002 * is allocated to another file, it is written to, and then we crash.
1003 * If the new data gets written to the file but the log buffers
1004 * containing the free and reallocation don't, then we'd end up with
1005 * garbage in the blocks being freed. As long as we make the new size
1006 * permanent before actually freeing any blocks it doesn't matter if
1007 * they get written to.
1009 ip->i_disk_size = newsize;
1010 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1012 if (newsize <= oldsize) {
1013 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1015 goto out_trans_cancel;
1018 * Truncated "down", so we're removing references to old data
1019 * here - if we delay flushing for a long time, we expose
1020 * ourselves unduly to the notorious NULL files problem. So,
1021 * we mark this inode and flush it when the file is closed,
1022 * and do not wait the usual (long) time for writeout.
1024 xfs_iflags_set(ip, XFS_ITRUNCATED);
1026 /* A truncate down always removes post-EOF blocks. */
1027 xfs_inode_clear_eofblocks_tag(ip);
1030 if (iattr->ia_valid & ATTR_MODE)
1031 xfs_setattr_mode(ip, iattr);
1032 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1033 xfs_setattr_time(ip, iattr);
1035 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1037 XFS_STATS_INC(mp, xs_ig_attrchg);
1039 if (mp->m_flags & XFS_MOUNT_WSYNC)
1040 xfs_trans_set_sync(tp);
1042 error = xfs_trans_commit(tp);
1045 xfs_iunlock(ip, lock_flags);
1049 xfs_trans_cancel(tp);
1054 xfs_vn_setattr_size(
1055 struct user_namespace *mnt_userns,
1056 struct dentry *dentry,
1057 struct iattr *iattr)
1059 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1062 trace_xfs_setattr(ip);
1064 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1067 return xfs_setattr_size(mnt_userns, ip, iattr);
1072 struct user_namespace *mnt_userns,
1073 struct dentry *dentry,
1074 struct iattr *iattr)
1076 struct inode *inode = d_inode(dentry);
1077 struct xfs_inode *ip = XFS_I(inode);
1080 if (iattr->ia_valid & ATTR_SIZE) {
1083 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1084 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1086 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1088 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1092 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr);
1093 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1095 trace_xfs_setattr(ip);
1097 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1099 error = xfs_setattr_nonsize(mnt_userns, ip, iattr);
1107 struct inode *inode,
1108 struct timespec64 *now,
1111 struct xfs_inode *ip = XFS_I(inode);
1112 struct xfs_mount *mp = ip->i_mount;
1113 int log_flags = XFS_ILOG_TIMESTAMP;
1114 struct xfs_trans *tp;
1117 trace_xfs_update_time(ip);
1119 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1120 if (!((flags & S_VERSION) &&
1121 inode_maybe_inc_iversion(inode, false)))
1122 return generic_update_time(inode, now, flags);
1124 /* Capture the iversion update that just occurred */
1125 log_flags |= XFS_ILOG_CORE;
1128 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1132 xfs_ilock(ip, XFS_ILOCK_EXCL);
1133 if (flags & S_CTIME)
1134 inode->i_ctime = *now;
1135 if (flags & S_MTIME)
1136 inode->i_mtime = *now;
1137 if (flags & S_ATIME)
1138 inode->i_atime = *now;
1140 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1141 xfs_trans_log_inode(tp, ip, log_flags);
1142 return xfs_trans_commit(tp);
1147 struct inode *inode,
1148 struct fiemap_extent_info *fieinfo,
1154 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1155 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1156 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1157 error = iomap_fiemap(inode, fieinfo, start, length,
1158 &xfs_xattr_iomap_ops);
1160 error = iomap_fiemap(inode, fieinfo, start, length,
1161 &xfs_read_iomap_ops);
1163 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1170 struct user_namespace *mnt_userns,
1172 struct dentry *dentry,
1175 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true);
1178 static const struct inode_operations xfs_inode_operations = {
1179 .get_acl = xfs_get_acl,
1180 .set_acl = xfs_set_acl,
1181 .getattr = xfs_vn_getattr,
1182 .setattr = xfs_vn_setattr,
1183 .listxattr = xfs_vn_listxattr,
1184 .fiemap = xfs_vn_fiemap,
1185 .update_time = xfs_vn_update_time,
1188 static const struct inode_operations xfs_dir_inode_operations = {
1189 .create = xfs_vn_create,
1190 .lookup = xfs_vn_lookup,
1191 .link = xfs_vn_link,
1192 .unlink = xfs_vn_unlink,
1193 .symlink = xfs_vn_symlink,
1194 .mkdir = xfs_vn_mkdir,
1196 * Yes, XFS uses the same method for rmdir and unlink.
1198 * There are some subtile differences deeper in the code,
1199 * but we use S_ISDIR to check for those.
1201 .rmdir = xfs_vn_unlink,
1202 .mknod = xfs_vn_mknod,
1203 .rename = xfs_vn_rename,
1204 .get_acl = xfs_get_acl,
1205 .set_acl = xfs_set_acl,
1206 .getattr = xfs_vn_getattr,
1207 .setattr = xfs_vn_setattr,
1208 .listxattr = xfs_vn_listxattr,
1209 .update_time = xfs_vn_update_time,
1210 .tmpfile = xfs_vn_tmpfile,
1213 static const struct inode_operations xfs_dir_ci_inode_operations = {
1214 .create = xfs_vn_create,
1215 .lookup = xfs_vn_ci_lookup,
1216 .link = xfs_vn_link,
1217 .unlink = xfs_vn_unlink,
1218 .symlink = xfs_vn_symlink,
1219 .mkdir = xfs_vn_mkdir,
1221 * Yes, XFS uses the same method for rmdir and unlink.
1223 * There are some subtile differences deeper in the code,
1224 * but we use S_ISDIR to check for those.
1226 .rmdir = xfs_vn_unlink,
1227 .mknod = xfs_vn_mknod,
1228 .rename = xfs_vn_rename,
1229 .get_acl = xfs_get_acl,
1230 .set_acl = xfs_set_acl,
1231 .getattr = xfs_vn_getattr,
1232 .setattr = xfs_vn_setattr,
1233 .listxattr = xfs_vn_listxattr,
1234 .update_time = xfs_vn_update_time,
1235 .tmpfile = xfs_vn_tmpfile,
1238 static const struct inode_operations xfs_symlink_inode_operations = {
1239 .get_link = xfs_vn_get_link,
1240 .getattr = xfs_vn_getattr,
1241 .setattr = xfs_vn_setattr,
1242 .listxattr = xfs_vn_listxattr,
1243 .update_time = xfs_vn_update_time,
1246 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1247 .get_link = xfs_vn_get_link_inline,
1248 .getattr = xfs_vn_getattr,
1249 .setattr = xfs_vn_setattr,
1250 .listxattr = xfs_vn_listxattr,
1251 .update_time = xfs_vn_update_time,
1254 /* Figure out if this file actually supports DAX. */
1256 xfs_inode_supports_dax(
1257 struct xfs_inode *ip)
1259 struct xfs_mount *mp = ip->i_mount;
1261 /* Only supported on regular files. */
1262 if (!S_ISREG(VFS_I(ip)->i_mode))
1265 /* Only supported on non-reflinked files. */
1266 if (xfs_is_reflink_inode(ip))
1269 /* Block size must match page size */
1270 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1273 /* Device has to support DAX too. */
1274 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1278 xfs_inode_should_enable_dax(
1279 struct xfs_inode *ip)
1281 if (!IS_ENABLED(CONFIG_FS_DAX))
1283 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1285 if (!xfs_inode_supports_dax(ip))
1287 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1289 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1295 xfs_diflags_to_iflags(
1296 struct xfs_inode *ip,
1299 struct inode *inode = VFS_I(ip);
1300 unsigned int xflags = xfs_ip2xflags(ip);
1301 unsigned int flags = 0;
1303 ASSERT(!(IS_DAX(inode) && init));
1305 if (xflags & FS_XFLAG_IMMUTABLE)
1306 flags |= S_IMMUTABLE;
1307 if (xflags & FS_XFLAG_APPEND)
1309 if (xflags & FS_XFLAG_SYNC)
1311 if (xflags & FS_XFLAG_NOATIME)
1313 if (init && xfs_inode_should_enable_dax(ip))
1317 * S_DAX can only be set during inode initialization and is never set by
1318 * the VFS, so we cannot mask off S_DAX in i_flags.
1320 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1321 inode->i_flags |= flags;
1325 * Initialize the Linux inode.
1327 * When reading existing inodes from disk this is called directly from xfs_iget,
1328 * when creating a new inode it is called from xfs_ialloc after setting up the
1329 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1330 * it up to the caller to deal with unlocking the inode appropriately.
1334 struct xfs_inode *ip)
1336 struct inode *inode = &ip->i_vnode;
1339 inode->i_ino = ip->i_ino;
1340 inode->i_state = I_NEW;
1342 inode_sb_list_add(inode);
1343 /* make the inode look hashed for the writeback code */
1344 inode_fake_hash(inode);
1346 i_size_write(inode, ip->i_disk_size);
1347 xfs_diflags_to_iflags(ip, true);
1349 if (S_ISDIR(inode->i_mode)) {
1351 * We set the i_rwsem class here to avoid potential races with
1352 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1353 * after a filehandle lookup has already found the inode in
1354 * cache before it has been unlocked via unlock_new_inode().
1356 lockdep_set_class(&inode->i_rwsem,
1357 &inode->i_sb->s_type->i_mutex_dir_key);
1358 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1360 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1364 * Ensure all page cache allocations are done from GFP_NOFS context to
1365 * prevent direct reclaim recursion back into the filesystem and blowing
1366 * stacks or deadlocking.
1368 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1369 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1372 * If there is no attribute fork no ACL can exist on this inode,
1373 * and it can't have any file capabilities attached to it either.
1375 if (!XFS_IFORK_Q(ip)) {
1376 inode_has_no_xattr(inode);
1377 cache_no_acl(inode);
1383 struct xfs_inode *ip)
1385 struct inode *inode = &ip->i_vnode;
1387 switch (inode->i_mode & S_IFMT) {
1389 inode->i_op = &xfs_inode_operations;
1390 inode->i_fop = &xfs_file_operations;
1392 inode->i_mapping->a_ops = &xfs_dax_aops;
1394 inode->i_mapping->a_ops = &xfs_address_space_operations;
1397 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1398 inode->i_op = &xfs_dir_ci_inode_operations;
1400 inode->i_op = &xfs_dir_inode_operations;
1401 inode->i_fop = &xfs_dir_file_operations;
1404 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL)
1405 inode->i_op = &xfs_inline_symlink_inode_operations;
1407 inode->i_op = &xfs_symlink_inode_operations;
1410 inode->i_op = &xfs_inode_operations;
1411 init_special_inode(inode, inode->i_mode, inode->i_rdev);