xref: /kernel/linux/linux-5.10/fs/xfs/xfs_iops.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
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"
14#include "xfs_acl.h"
15#include "xfs_quota.h"
16#include "xfs_attr.h"
17#include "xfs_trans.h"
18#include "xfs_trace.h"
19#include "xfs_icache.h"
20#include "xfs_symlink.h"
21#include "xfs_dir2.h"
22#include "xfs_iomap.h"
23#include "xfs_error.h"
24
25#include <linux/posix_acl.h>
26#include <linux/security.h>
27#include <linux/iversion.h>
28#include <linux/fiemap.h>
29
30/*
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.
38 */
39static struct lock_class_key xfs_nondir_ilock_class;
40static struct lock_class_key xfs_dir_ilock_class;
41
42static int
43xfs_initxattrs(
44	struct inode		*inode,
45	const struct xattr	*xattr_array,
46	void			*fs_info)
47{
48	const struct xattr	*xattr;
49	struct xfs_inode	*ip = XFS_I(inode);
50	int			error = 0;
51
52	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
53		struct xfs_da_args	args = {
54			.dp		= ip,
55			.attr_filter	= XFS_ATTR_SECURE,
56			.name		= xattr->name,
57			.namelen	= strlen(xattr->name),
58			.value		= xattr->value,
59			.valuelen	= xattr->value_len,
60		};
61		error = xfs_attr_set(&args);
62		if (error < 0)
63			break;
64	}
65	return error;
66}
67
68/*
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).
73 */
74
75STATIC int
76xfs_init_security(
77	struct inode	*inode,
78	struct inode	*dir,
79	const struct qstr *qstr)
80{
81	return security_inode_init_security(inode, dir, qstr,
82					     &xfs_initxattrs, NULL);
83}
84
85static void
86xfs_dentry_to_name(
87	struct xfs_name	*namep,
88	struct dentry	*dentry)
89{
90	namep->name = dentry->d_name.name;
91	namep->len = dentry->d_name.len;
92	namep->type = XFS_DIR3_FT_UNKNOWN;
93}
94
95static int
96xfs_dentry_mode_to_name(
97	struct xfs_name	*namep,
98	struct dentry	*dentry,
99	int		mode)
100{
101	namep->name = dentry->d_name.name;
102	namep->len = dentry->d_name.len;
103	namep->type = xfs_mode_to_ftype(mode);
104
105	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
106		return -EFSCORRUPTED;
107
108	return 0;
109}
110
111STATIC void
112xfs_cleanup_inode(
113	struct inode	*dir,
114	struct inode	*inode,
115	struct dentry	*dentry)
116{
117	struct xfs_name	teardown;
118
119	/* Oh, the horror.
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.
123	 */
124	xfs_dentry_to_name(&teardown, dentry);
125
126	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
127}
128
129STATIC int
130xfs_generic_create(
131	struct inode	*dir,
132	struct dentry	*dentry,
133	umode_t		mode,
134	dev_t		rdev,
135	bool		tmpfile)	/* unnamed file */
136{
137	struct inode	*inode;
138	struct xfs_inode *ip = NULL;
139	struct posix_acl *default_acl, *acl;
140	struct xfs_name	name;
141	int		error;
142
143	/*
144	 * Irix uses Missed'em'V split, but doesn't want to see
145	 * the upper 5 bits of (14bit) major.
146	 */
147	if (S_ISCHR(mode) || S_ISBLK(mode)) {
148		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
149			return -EINVAL;
150	} else {
151		rdev = 0;
152	}
153
154	error = posix_acl_create(dir, &mode, &default_acl, &acl);
155	if (error)
156		return error;
157
158	/* Verify mode is valid also for tmpfile case */
159	error = xfs_dentry_mode_to_name(&name, dentry, mode);
160	if (unlikely(error))
161		goto out_free_acl;
162
163	if (!tmpfile) {
164		error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
165	} else {
166		error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
167	}
168	if (unlikely(error))
169		goto out_free_acl;
170
171	inode = VFS_I(ip);
172
173	error = xfs_init_security(inode, dir, &dentry->d_name);
174	if (unlikely(error))
175		goto out_cleanup_inode;
176
177#ifdef CONFIG_XFS_POSIX_ACL
178	if (default_acl) {
179		error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
180		if (error)
181			goto out_cleanup_inode;
182	}
183	if (acl) {
184		error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
185		if (error)
186			goto out_cleanup_inode;
187	}
188#endif
189
190	xfs_setup_iops(ip);
191
192	if (tmpfile) {
193		/*
194		 * The VFS requires that any inode fed to d_tmpfile must have
195		 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
196		 * However, we created the temp file with nlink == 0 because
197		 * we're not allowed to put an inode with nlink > 0 on the
198		 * unlinked list.  Therefore we have to set nlink to 1 so that
199		 * d_tmpfile can immediately set it back to zero.
200		 */
201		set_nlink(inode, 1);
202		d_tmpfile(dentry, inode);
203	} else
204		d_instantiate(dentry, inode);
205
206	xfs_finish_inode_setup(ip);
207
208 out_free_acl:
209	if (default_acl)
210		posix_acl_release(default_acl);
211	if (acl)
212		posix_acl_release(acl);
213	return error;
214
215 out_cleanup_inode:
216	xfs_finish_inode_setup(ip);
217	if (!tmpfile)
218		xfs_cleanup_inode(dir, inode, dentry);
219	xfs_irele(ip);
220	goto out_free_acl;
221}
222
223STATIC int
224xfs_vn_mknod(
225	struct inode	*dir,
226	struct dentry	*dentry,
227	umode_t		mode,
228	dev_t		rdev)
229{
230	return xfs_generic_create(dir, dentry, mode, rdev, false);
231}
232
233STATIC int
234xfs_vn_create(
235	struct inode	*dir,
236	struct dentry	*dentry,
237	umode_t		mode,
238	bool		flags)
239{
240	return xfs_generic_create(dir, dentry, mode, 0, false);
241}
242
243STATIC int
244xfs_vn_mkdir(
245	struct inode	*dir,
246	struct dentry	*dentry,
247	umode_t		mode)
248{
249	return xfs_generic_create(dir, dentry, mode | S_IFDIR, 0, false);
250}
251
252STATIC struct dentry *
253xfs_vn_lookup(
254	struct inode	*dir,
255	struct dentry	*dentry,
256	unsigned int flags)
257{
258	struct inode *inode;
259	struct xfs_inode *cip;
260	struct xfs_name	name;
261	int		error;
262
263	if (dentry->d_name.len >= MAXNAMELEN)
264		return ERR_PTR(-ENAMETOOLONG);
265
266	xfs_dentry_to_name(&name, dentry);
267	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
268	if (likely(!error))
269		inode = VFS_I(cip);
270	else if (likely(error == -ENOENT))
271		inode = NULL;
272	else
273		inode = ERR_PTR(error);
274	return d_splice_alias(inode, dentry);
275}
276
277STATIC struct dentry *
278xfs_vn_ci_lookup(
279	struct inode	*dir,
280	struct dentry	*dentry,
281	unsigned int flags)
282{
283	struct xfs_inode *ip;
284	struct xfs_name	xname;
285	struct xfs_name ci_name;
286	struct qstr	dname;
287	int		error;
288
289	if (dentry->d_name.len >= MAXNAMELEN)
290		return ERR_PTR(-ENAMETOOLONG);
291
292	xfs_dentry_to_name(&xname, dentry);
293	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
294	if (unlikely(error)) {
295		if (unlikely(error != -ENOENT))
296			return ERR_PTR(error);
297		/*
298		 * call d_add(dentry, NULL) here when d_drop_negative_children
299		 * is called in xfs_vn_mknod (ie. allow negative dentries
300		 * with CI filesystems).
301		 */
302		return NULL;
303	}
304
305	/* if exact match, just splice and exit */
306	if (!ci_name.name)
307		return d_splice_alias(VFS_I(ip), dentry);
308
309	/* else case-insensitive match... */
310	dname.name = ci_name.name;
311	dname.len = ci_name.len;
312	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
313	kmem_free(ci_name.name);
314	return dentry;
315}
316
317STATIC int
318xfs_vn_link(
319	struct dentry	*old_dentry,
320	struct inode	*dir,
321	struct dentry	*dentry)
322{
323	struct inode	*inode = d_inode(old_dentry);
324	struct xfs_name	name;
325	int		error;
326
327	error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
328	if (unlikely(error))
329		return error;
330
331	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
332	if (unlikely(error))
333		return error;
334
335	ihold(inode);
336	d_instantiate(dentry, inode);
337	return 0;
338}
339
340STATIC int
341xfs_vn_unlink(
342	struct inode	*dir,
343	struct dentry	*dentry)
344{
345	struct xfs_name	name;
346	int		error;
347
348	xfs_dentry_to_name(&name, dentry);
349
350	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
351	if (error)
352		return error;
353
354	/*
355	 * With unlink, the VFS makes the dentry "negative": no inode,
356	 * but still hashed. This is incompatible with case-insensitive
357	 * mode, so invalidate (unhash) the dentry in CI-mode.
358	 */
359	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
360		d_invalidate(dentry);
361	return 0;
362}
363
364STATIC int
365xfs_vn_symlink(
366	struct inode	*dir,
367	struct dentry	*dentry,
368	const char	*symname)
369{
370	struct inode	*inode;
371	struct xfs_inode *cip = NULL;
372	struct xfs_name	name;
373	int		error;
374	umode_t		mode;
375
376	mode = S_IFLNK |
377		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
378	error = xfs_dentry_mode_to_name(&name, dentry, mode);
379	if (unlikely(error))
380		goto out;
381
382	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
383	if (unlikely(error))
384		goto out;
385
386	inode = VFS_I(cip);
387
388	error = xfs_init_security(inode, dir, &dentry->d_name);
389	if (unlikely(error))
390		goto out_cleanup_inode;
391
392	xfs_setup_iops(cip);
393
394	d_instantiate(dentry, inode);
395	xfs_finish_inode_setup(cip);
396	return 0;
397
398 out_cleanup_inode:
399	xfs_finish_inode_setup(cip);
400	xfs_cleanup_inode(dir, inode, dentry);
401	xfs_irele(cip);
402 out:
403	return error;
404}
405
406STATIC int
407xfs_vn_rename(
408	struct inode	*odir,
409	struct dentry	*odentry,
410	struct inode	*ndir,
411	struct dentry	*ndentry,
412	unsigned int	flags)
413{
414	struct inode	*new_inode = d_inode(ndentry);
415	int		omode = 0;
416	int		error;
417	struct xfs_name	oname;
418	struct xfs_name	nname;
419
420	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
421		return -EINVAL;
422
423	/* if we are exchanging files, we need to set i_mode of both files */
424	if (flags & RENAME_EXCHANGE)
425		omode = d_inode(ndentry)->i_mode;
426
427	error = xfs_dentry_mode_to_name(&oname, odentry, omode);
428	if (omode && unlikely(error))
429		return error;
430
431	error = xfs_dentry_mode_to_name(&nname, ndentry,
432					d_inode(odentry)->i_mode);
433	if (unlikely(error))
434		return error;
435
436	return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
437			  XFS_I(ndir), &nname,
438			  new_inode ? XFS_I(new_inode) : NULL, flags);
439}
440
441/*
442 * careful here - this function can get called recursively, so
443 * we need to be very careful about how much stack we use.
444 * uio is kmalloced for this reason...
445 */
446STATIC const char *
447xfs_vn_get_link(
448	struct dentry		*dentry,
449	struct inode		*inode,
450	struct delayed_call	*done)
451{
452	char			*link;
453	int			error = -ENOMEM;
454
455	if (!dentry)
456		return ERR_PTR(-ECHILD);
457
458	link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
459	if (!link)
460		goto out_err;
461
462	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
463	if (unlikely(error))
464		goto out_kfree;
465
466	set_delayed_call(done, kfree_link, link);
467	return link;
468
469 out_kfree:
470	kfree(link);
471 out_err:
472	return ERR_PTR(error);
473}
474
475STATIC const char *
476xfs_vn_get_link_inline(
477	struct dentry		*dentry,
478	struct inode		*inode,
479	struct delayed_call	*done)
480{
481	struct xfs_inode	*ip = XFS_I(inode);
482	char			*link;
483
484	ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
485
486	/*
487	 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
488	 * if_data is junk.
489	 */
490	link = ip->i_df.if_u1.if_data;
491	if (XFS_IS_CORRUPT(ip->i_mount, !link))
492		return ERR_PTR(-EFSCORRUPTED);
493	return link;
494}
495
496static uint32_t
497xfs_stat_blksize(
498	struct xfs_inode	*ip)
499{
500	struct xfs_mount	*mp = ip->i_mount;
501
502	/*
503	 * If the file blocks are being allocated from a realtime volume, then
504	 * always return the realtime extent size.
505	 */
506	if (XFS_IS_REALTIME_INODE(ip))
507		return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
508
509	/*
510	 * Allow large block sizes to be reported to userspace programs if the
511	 * "largeio" mount option is used.
512	 *
513	 * If compatibility mode is specified, simply return the basic unit of
514	 * caching so that we don't get inefficient read/modify/write I/O from
515	 * user apps. Otherwise....
516	 *
517	 * If the underlying volume is a stripe, then return the stripe width in
518	 * bytes as the recommended I/O size. It is not a stripe and we've set a
519	 * default buffered I/O size, return that, otherwise return the compat
520	 * default.
521	 */
522	if (mp->m_flags & XFS_MOUNT_LARGEIO) {
523		if (mp->m_swidth)
524			return mp->m_swidth << mp->m_sb.sb_blocklog;
525		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
526			return 1U << mp->m_allocsize_log;
527	}
528
529	return PAGE_SIZE;
530}
531
532STATIC int
533xfs_vn_getattr(
534	const struct path	*path,
535	struct kstat		*stat,
536	u32			request_mask,
537	unsigned int		query_flags)
538{
539	struct inode		*inode = d_inode(path->dentry);
540	struct xfs_inode	*ip = XFS_I(inode);
541	struct xfs_mount	*mp = ip->i_mount;
542
543	trace_xfs_getattr(ip);
544
545	if (XFS_FORCED_SHUTDOWN(mp))
546		return -EIO;
547
548	stat->size = XFS_ISIZE(ip);
549	stat->dev = inode->i_sb->s_dev;
550	stat->mode = inode->i_mode;
551	stat->nlink = inode->i_nlink;
552	stat->uid = inode->i_uid;
553	stat->gid = inode->i_gid;
554	stat->ino = ip->i_ino;
555	stat->atime = inode->i_atime;
556	stat->mtime = inode->i_mtime;
557	stat->ctime = inode->i_ctime;
558	stat->blocks =
559		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
560
561	if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
562		if (request_mask & STATX_BTIME) {
563			stat->result_mask |= STATX_BTIME;
564			stat->btime = ip->i_d.di_crtime;
565		}
566	}
567
568	/*
569	 * Note: If you add another clause to set an attribute flag, please
570	 * update attributes_mask below.
571	 */
572	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
573		stat->attributes |= STATX_ATTR_IMMUTABLE;
574	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
575		stat->attributes |= STATX_ATTR_APPEND;
576	if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
577		stat->attributes |= STATX_ATTR_NODUMP;
578
579	stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
580				  STATX_ATTR_APPEND |
581				  STATX_ATTR_NODUMP);
582
583	switch (inode->i_mode & S_IFMT) {
584	case S_IFBLK:
585	case S_IFCHR:
586		stat->blksize = BLKDEV_IOSIZE;
587		stat->rdev = inode->i_rdev;
588		break;
589	default:
590		stat->blksize = xfs_stat_blksize(ip);
591		stat->rdev = 0;
592		break;
593	}
594
595	return 0;
596}
597
598static int
599xfs_vn_change_ok(
600	struct dentry	*dentry,
601	struct iattr	*iattr)
602{
603	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
604
605	if (mp->m_flags & XFS_MOUNT_RDONLY)
606		return -EROFS;
607
608	if (XFS_FORCED_SHUTDOWN(mp))
609		return -EIO;
610
611	return setattr_prepare(dentry, iattr);
612}
613
614/*
615 * Set non-size attributes of an inode.
616 *
617 * Caution: The caller of this function is responsible for calling
618 * setattr_prepare() or otherwise verifying the change is fine.
619 */
620int
621xfs_setattr_nonsize(
622	struct xfs_inode	*ip,
623	struct iattr		*iattr,
624	int			flags)
625{
626	xfs_mount_t		*mp = ip->i_mount;
627	struct inode		*inode = VFS_I(ip);
628	int			mask = iattr->ia_valid;
629	xfs_trans_t		*tp;
630	int			error;
631	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
632	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
633	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
634	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
635
636	ASSERT((mask & ATTR_SIZE) == 0);
637
638	/*
639	 * If disk quotas is on, we make sure that the dquots do exist on disk,
640	 * before we start any other transactions. Trying to do this later
641	 * is messy. We don't care to take a readlock to look at the ids
642	 * in inode here, because we can't hold it across the trans_reserve.
643	 * If the IDs do change before we take the ilock, we're covered
644	 * because the i_*dquot fields will get updated anyway.
645	 */
646	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
647		uint	qflags = 0;
648
649		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
650			uid = iattr->ia_uid;
651			qflags |= XFS_QMOPT_UQUOTA;
652		} else {
653			uid = inode->i_uid;
654		}
655		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
656			gid = iattr->ia_gid;
657			qflags |= XFS_QMOPT_GQUOTA;
658		}  else {
659			gid = inode->i_gid;
660		}
661
662		/*
663		 * We take a reference when we initialize udqp and gdqp,
664		 * so it is important that we never blindly double trip on
665		 * the same variable. See xfs_create() for an example.
666		 */
667		ASSERT(udqp == NULL);
668		ASSERT(gdqp == NULL);
669		error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
670					   qflags, &udqp, &gdqp, NULL);
671		if (error)
672			return error;
673	}
674
675	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
676	if (error)
677		goto out_dqrele;
678
679	xfs_ilock(ip, XFS_ILOCK_EXCL);
680	xfs_trans_ijoin(tp, ip, 0);
681
682	/*
683	 * Change file ownership.  Must be the owner or privileged.
684	 */
685	if (mask & (ATTR_UID|ATTR_GID)) {
686		/*
687		 * These IDs could have changed since we last looked at them.
688		 * But, we're assured that if the ownership did change
689		 * while we didn't have the inode locked, inode's dquot(s)
690		 * would have changed also.
691		 */
692		iuid = inode->i_uid;
693		igid = inode->i_gid;
694		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
695		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
696
697		/*
698		 * Do a quota reservation only if uid/gid is actually
699		 * going to change.
700		 */
701		if (XFS_IS_QUOTA_RUNNING(mp) &&
702		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
703		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
704			ASSERT(tp);
705			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
706						NULL, capable(CAP_FOWNER) ?
707						XFS_QMOPT_FORCE_RES : 0);
708			if (error)	/* out of quota */
709				goto out_cancel;
710		}
711
712		/*
713		 * Change the ownerships and register quota modifications
714		 * in the transaction.
715		 */
716		if (!uid_eq(iuid, uid)) {
717			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
718				ASSERT(mask & ATTR_UID);
719				ASSERT(udqp);
720				olddquot1 = xfs_qm_vop_chown(tp, ip,
721							&ip->i_udquot, udqp);
722			}
723		}
724		if (!gid_eq(igid, gid)) {
725			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
726				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
727				       !XFS_IS_PQUOTA_ON(mp));
728				ASSERT(mask & ATTR_GID);
729				ASSERT(gdqp);
730				olddquot2 = xfs_qm_vop_chown(tp, ip,
731							&ip->i_gdquot, gdqp);
732			}
733		}
734	}
735
736	setattr_copy(inode, iattr);
737	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
738
739	XFS_STATS_INC(mp, xs_ig_attrchg);
740
741	if (mp->m_flags & XFS_MOUNT_WSYNC)
742		xfs_trans_set_sync(tp);
743	error = xfs_trans_commit(tp);
744
745	xfs_iunlock(ip, XFS_ILOCK_EXCL);
746
747	/*
748	 * Release any dquot(s) the inode had kept before chown.
749	 */
750	xfs_qm_dqrele(olddquot1);
751	xfs_qm_dqrele(olddquot2);
752	xfs_qm_dqrele(udqp);
753	xfs_qm_dqrele(gdqp);
754
755	if (error)
756		return error;
757
758	/*
759	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
760	 * 	     update.  We could avoid this with linked transactions
761	 * 	     and passing down the transaction pointer all the way
762	 *	     to attr_set.  No previous user of the generic
763	 * 	     Posix ACL code seems to care about this issue either.
764	 */
765	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
766		error = posix_acl_chmod(inode, inode->i_mode);
767		if (error)
768			return error;
769	}
770
771	return 0;
772
773out_cancel:
774	xfs_trans_cancel(tp);
775	xfs_iunlock(ip, XFS_ILOCK_EXCL);
776out_dqrele:
777	xfs_qm_dqrele(udqp);
778	xfs_qm_dqrele(gdqp);
779	return error;
780}
781
782int
783xfs_vn_setattr_nonsize(
784	struct dentry		*dentry,
785	struct iattr		*iattr)
786{
787	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
788	int error;
789
790	trace_xfs_setattr(ip);
791
792	error = xfs_vn_change_ok(dentry, iattr);
793	if (error)
794		return error;
795	return xfs_setattr_nonsize(ip, iattr, 0);
796}
797
798/*
799 * Truncate file.  Must have write permission and not be a directory.
800 *
801 * Caution: The caller of this function is responsible for calling
802 * setattr_prepare() or otherwise verifying the change is fine.
803 */
804STATIC int
805xfs_setattr_size(
806	struct xfs_inode	*ip,
807	struct iattr		*iattr)
808{
809	struct xfs_mount	*mp = ip->i_mount;
810	struct inode		*inode = VFS_I(ip);
811	xfs_off_t		oldsize, newsize;
812	struct xfs_trans	*tp;
813	int			error;
814	uint			lock_flags = 0;
815	bool			did_zeroing = false;
816
817	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
818	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
819	ASSERT(S_ISREG(inode->i_mode));
820	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
821		ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
822
823	oldsize = inode->i_size;
824	newsize = iattr->ia_size;
825
826	/*
827	 * Short circuit the truncate case for zero length files.
828	 */
829	if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
830		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
831			return 0;
832
833		/*
834		 * Use the regular setattr path to update the timestamps.
835		 */
836		iattr->ia_valid &= ~ATTR_SIZE;
837		return xfs_setattr_nonsize(ip, iattr, 0);
838	}
839
840	/*
841	 * Make sure that the dquots are attached to the inode.
842	 */
843	error = xfs_qm_dqattach(ip);
844	if (error)
845		return error;
846
847	/*
848	 * Wait for all direct I/O to complete.
849	 */
850	inode_dio_wait(inode);
851
852	/*
853	 * File data changes must be complete before we start the transaction to
854	 * modify the inode.  This needs to be done before joining the inode to
855	 * the transaction because the inode cannot be unlocked once it is a
856	 * part of the transaction.
857	 *
858	 * Start with zeroing any data beyond EOF that we may expose on file
859	 * extension, or zeroing out the rest of the block on a downward
860	 * truncate.
861	 */
862	if (newsize > oldsize) {
863		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
864		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
865				&did_zeroing, &xfs_buffered_write_iomap_ops);
866	} else {
867		/*
868		 * iomap won't detect a dirty page over an unwritten block (or a
869		 * cow block over a hole) and subsequently skips zeroing the
870		 * newly post-EOF portion of the page. Flush the new EOF to
871		 * convert the block before the pagecache truncate.
872		 */
873		error = filemap_write_and_wait_range(inode->i_mapping, newsize,
874						     newsize);
875		if (error)
876			return error;
877		error = iomap_truncate_page(inode, newsize, &did_zeroing,
878				&xfs_buffered_write_iomap_ops);
879	}
880
881	if (error)
882		return error;
883
884	/*
885	 * We've already locked out new page faults, so now we can safely remove
886	 * pages from the page cache knowing they won't get refaulted until we
887	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
888	 * complete. The truncate_setsize() call also cleans partial EOF page
889	 * PTEs on extending truncates and hence ensures sub-page block size
890	 * filesystems are correctly handled, too.
891	 *
892	 * We have to do all the page cache truncate work outside the
893	 * transaction context as the "lock" order is page lock->log space
894	 * reservation as defined by extent allocation in the writeback path.
895	 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
896	 * having already truncated the in-memory version of the file (i.e. made
897	 * user visible changes). There's not much we can do about this, except
898	 * to hope that the caller sees ENOMEM and retries the truncate
899	 * operation.
900	 *
901	 * And we update in-core i_size and truncate page cache beyond newsize
902	 * before writeback the [di_size, newsize] range, so we're guaranteed
903	 * not to write stale data past the new EOF on truncate down.
904	 */
905	truncate_setsize(inode, newsize);
906
907	/*
908	 * We are going to log the inode size change in this transaction so
909	 * any previous writes that are beyond the on disk EOF and the new
910	 * EOF that have not been written out need to be written here.  If we
911	 * do not write the data out, we expose ourselves to the null files
912	 * problem. Note that this includes any block zeroing we did above;
913	 * otherwise those blocks may not be zeroed after a crash.
914	 */
915	if (did_zeroing ||
916	    (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
917		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
918						ip->i_d.di_size, newsize - 1);
919		if (error)
920			return error;
921	}
922
923	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
924	if (error)
925		return error;
926
927	lock_flags |= XFS_ILOCK_EXCL;
928	xfs_ilock(ip, XFS_ILOCK_EXCL);
929	xfs_trans_ijoin(tp, ip, 0);
930
931	/*
932	 * Only change the c/mtime if we are changing the size or we are
933	 * explicitly asked to change it.  This handles the semantic difference
934	 * between truncate() and ftruncate() as implemented in the VFS.
935	 *
936	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
937	 * special case where we need to update the times despite not having
938	 * these flags set.  For all other operations the VFS set these flags
939	 * explicitly if it wants a timestamp update.
940	 */
941	if (newsize != oldsize &&
942	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
943		iattr->ia_ctime = iattr->ia_mtime =
944			current_time(inode);
945		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
946	}
947
948	/*
949	 * The first thing we do is set the size to new_size permanently on
950	 * disk.  This way we don't have to worry about anyone ever being able
951	 * to look at the data being freed even in the face of a crash.
952	 * What we're getting around here is the case where we free a block, it
953	 * is allocated to another file, it is written to, and then we crash.
954	 * If the new data gets written to the file but the log buffers
955	 * containing the free and reallocation don't, then we'd end up with
956	 * garbage in the blocks being freed.  As long as we make the new size
957	 * permanent before actually freeing any blocks it doesn't matter if
958	 * they get written to.
959	 */
960	ip->i_d.di_size = newsize;
961	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
962
963	if (newsize <= oldsize) {
964		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
965		if (error)
966			goto out_trans_cancel;
967
968		/*
969		 * Truncated "down", so we're removing references to old data
970		 * here - if we delay flushing for a long time, we expose
971		 * ourselves unduly to the notorious NULL files problem.  So,
972		 * we mark this inode and flush it when the file is closed,
973		 * and do not wait the usual (long) time for writeout.
974		 */
975		xfs_iflags_set(ip, XFS_ITRUNCATED);
976
977		/* A truncate down always removes post-EOF blocks. */
978		xfs_inode_clear_eofblocks_tag(ip);
979	}
980
981	ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
982	setattr_copy(inode, iattr);
983	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
984
985	XFS_STATS_INC(mp, xs_ig_attrchg);
986
987	if (mp->m_flags & XFS_MOUNT_WSYNC)
988		xfs_trans_set_sync(tp);
989
990	error = xfs_trans_commit(tp);
991out_unlock:
992	if (lock_flags)
993		xfs_iunlock(ip, lock_flags);
994	return error;
995
996out_trans_cancel:
997	xfs_trans_cancel(tp);
998	goto out_unlock;
999}
1000
1001int
1002xfs_vn_setattr_size(
1003	struct dentry		*dentry,
1004	struct iattr		*iattr)
1005{
1006	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1007	int error;
1008
1009	trace_xfs_setattr(ip);
1010
1011	error = xfs_vn_change_ok(dentry, iattr);
1012	if (error)
1013		return error;
1014	return xfs_setattr_size(ip, iattr);
1015}
1016
1017STATIC int
1018xfs_vn_setattr(
1019	struct dentry		*dentry,
1020	struct iattr		*iattr)
1021{
1022	int			error;
1023
1024	if (iattr->ia_valid & ATTR_SIZE) {
1025		struct inode		*inode = d_inode(dentry);
1026		struct xfs_inode	*ip = XFS_I(inode);
1027		uint			iolock;
1028
1029		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1030		iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1031
1032		error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1033		if (error) {
1034			xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1035			return error;
1036		}
1037
1038		error = xfs_vn_setattr_size(dentry, iattr);
1039		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1040	} else {
1041		error = xfs_vn_setattr_nonsize(dentry, iattr);
1042	}
1043
1044	return error;
1045}
1046
1047STATIC int
1048xfs_vn_update_time(
1049	struct inode		*inode,
1050	struct timespec64	*now,
1051	int			flags)
1052{
1053	struct xfs_inode	*ip = XFS_I(inode);
1054	struct xfs_mount	*mp = ip->i_mount;
1055	int			log_flags = XFS_ILOG_TIMESTAMP;
1056	struct xfs_trans	*tp;
1057	int			error;
1058
1059	trace_xfs_update_time(ip);
1060
1061	if (inode->i_sb->s_flags & SB_LAZYTIME) {
1062		if (!((flags & S_VERSION) &&
1063		      inode_maybe_inc_iversion(inode, false)))
1064			return generic_update_time(inode, now, flags);
1065
1066		/* Capture the iversion update that just occurred */
1067		log_flags |= XFS_ILOG_CORE;
1068	}
1069
1070	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1071	if (error)
1072		return error;
1073
1074	xfs_ilock(ip, XFS_ILOCK_EXCL);
1075	if (flags & S_CTIME)
1076		inode->i_ctime = *now;
1077	if (flags & S_MTIME)
1078		inode->i_mtime = *now;
1079	if (flags & S_ATIME)
1080		inode->i_atime = *now;
1081
1082	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1083	xfs_trans_log_inode(tp, ip, log_flags);
1084	return xfs_trans_commit(tp);
1085}
1086
1087STATIC int
1088xfs_vn_fiemap(
1089	struct inode		*inode,
1090	struct fiemap_extent_info *fieinfo,
1091	u64			start,
1092	u64			length)
1093{
1094	int			error;
1095
1096	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1097	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1098		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1099		error = iomap_fiemap(inode, fieinfo, start, length,
1100				&xfs_xattr_iomap_ops);
1101	} else {
1102		error = iomap_fiemap(inode, fieinfo, start, length,
1103				&xfs_read_iomap_ops);
1104	}
1105	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1106
1107	return error;
1108}
1109
1110STATIC int
1111xfs_vn_tmpfile(
1112	struct inode	*dir,
1113	struct dentry	*dentry,
1114	umode_t		mode)
1115{
1116	return xfs_generic_create(dir, dentry, mode, 0, true);
1117}
1118
1119static const struct inode_operations xfs_inode_operations = {
1120	.get_acl		= xfs_get_acl,
1121	.set_acl		= xfs_set_acl,
1122	.getattr		= xfs_vn_getattr,
1123	.setattr		= xfs_vn_setattr,
1124	.listxattr		= xfs_vn_listxattr,
1125	.fiemap			= xfs_vn_fiemap,
1126	.update_time		= xfs_vn_update_time,
1127};
1128
1129static 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,
1136	/*
1137	 * Yes, XFS uses the same method for rmdir and unlink.
1138	 *
1139	 * There are some subtile differences deeper in the code,
1140	 * but we use S_ISDIR to check for those.
1141	 */
1142	.rmdir			= xfs_vn_unlink,
1143	.mknod			= xfs_vn_mknod,
1144	.rename			= xfs_vn_rename,
1145	.get_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};
1153
1154static const struct inode_operations xfs_dir_ci_inode_operations = {
1155	.create			= xfs_vn_create,
1156	.lookup			= xfs_vn_ci_lookup,
1157	.link			= xfs_vn_link,
1158	.unlink			= xfs_vn_unlink,
1159	.symlink		= xfs_vn_symlink,
1160	.mkdir			= xfs_vn_mkdir,
1161	/*
1162	 * Yes, XFS uses the same method for rmdir and unlink.
1163	 *
1164	 * There are some subtile differences deeper in the code,
1165	 * but we use S_ISDIR to check for those.
1166	 */
1167	.rmdir			= xfs_vn_unlink,
1168	.mknod			= xfs_vn_mknod,
1169	.rename			= xfs_vn_rename,
1170	.get_acl		= xfs_get_acl,
1171	.set_acl		= xfs_set_acl,
1172	.getattr		= xfs_vn_getattr,
1173	.setattr		= xfs_vn_setattr,
1174	.listxattr		= xfs_vn_listxattr,
1175	.update_time		= xfs_vn_update_time,
1176	.tmpfile		= xfs_vn_tmpfile,
1177};
1178
1179static const struct inode_operations xfs_symlink_inode_operations = {
1180	.get_link		= xfs_vn_get_link,
1181	.getattr		= xfs_vn_getattr,
1182	.setattr		= xfs_vn_setattr,
1183	.listxattr		= xfs_vn_listxattr,
1184	.update_time		= xfs_vn_update_time,
1185};
1186
1187static const struct inode_operations xfs_inline_symlink_inode_operations = {
1188	.get_link		= xfs_vn_get_link_inline,
1189	.getattr		= xfs_vn_getattr,
1190	.setattr		= xfs_vn_setattr,
1191	.listxattr		= xfs_vn_listxattr,
1192	.update_time		= xfs_vn_update_time,
1193};
1194
1195/* Figure out if this file actually supports DAX. */
1196static bool
1197xfs_inode_supports_dax(
1198	struct xfs_inode	*ip)
1199{
1200	struct xfs_mount	*mp = ip->i_mount;
1201
1202	/* Only supported on regular files. */
1203	if (!S_ISREG(VFS_I(ip)->i_mode))
1204		return false;
1205
1206	/* Only supported on non-reflinked files. */
1207	if (xfs_is_reflink_inode(ip))
1208		return false;
1209
1210	/* Block size must match page size */
1211	if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1212		return false;
1213
1214	/* Device has to support DAX too. */
1215	return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1216}
1217
1218static bool
1219xfs_inode_should_enable_dax(
1220	struct xfs_inode *ip)
1221{
1222	if (!IS_ENABLED(CONFIG_FS_DAX))
1223		return false;
1224	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1225		return false;
1226	if (!xfs_inode_supports_dax(ip))
1227		return false;
1228	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1229		return true;
1230	if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
1231		return true;
1232	return false;
1233}
1234
1235void
1236xfs_diflags_to_iflags(
1237	struct xfs_inode	*ip,
1238	bool init)
1239{
1240	struct inode            *inode = VFS_I(ip);
1241	unsigned int            xflags = xfs_ip2xflags(ip);
1242	unsigned int            flags = 0;
1243
1244	ASSERT(!(IS_DAX(inode) && init));
1245
1246	if (xflags & FS_XFLAG_IMMUTABLE)
1247		flags |= S_IMMUTABLE;
1248	if (xflags & FS_XFLAG_APPEND)
1249		flags |= S_APPEND;
1250	if (xflags & FS_XFLAG_SYNC)
1251		flags |= S_SYNC;
1252	if (xflags & FS_XFLAG_NOATIME)
1253		flags |= S_NOATIME;
1254	if (init && xfs_inode_should_enable_dax(ip))
1255		flags |= S_DAX;
1256
1257	/*
1258	 * S_DAX can only be set during inode initialization and is never set by
1259	 * the VFS, so we cannot mask off S_DAX in i_flags.
1260	 */
1261	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1262	inode->i_flags |= flags;
1263}
1264
1265/*
1266 * Initialize the Linux inode.
1267 *
1268 * When reading existing inodes from disk this is called directly from xfs_iget,
1269 * when creating a new inode it is called from xfs_ialloc after setting up the
1270 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1271 * it up to the caller to deal with unlocking the inode appropriately.
1272 */
1273void
1274xfs_setup_inode(
1275	struct xfs_inode	*ip)
1276{
1277	struct inode		*inode = &ip->i_vnode;
1278	gfp_t			gfp_mask;
1279
1280	inode->i_ino = ip->i_ino;
1281	inode->i_state |= I_NEW;
1282
1283	inode_sb_list_add(inode);
1284	/* make the inode look hashed for the writeback code */
1285	inode_fake_hash(inode);
1286
1287	i_size_write(inode, ip->i_d.di_size);
1288	xfs_diflags_to_iflags(ip, true);
1289
1290	if (S_ISDIR(inode->i_mode)) {
1291		/*
1292		 * We set the i_rwsem class here to avoid potential races with
1293		 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1294		 * after a filehandle lookup has already found the inode in
1295		 * cache before it has been unlocked via unlock_new_inode().
1296		 */
1297		lockdep_set_class(&inode->i_rwsem,
1298				  &inode->i_sb->s_type->i_mutex_dir_key);
1299		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1300	} else {
1301		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1302	}
1303
1304	/*
1305	 * Ensure all page cache allocations are done from GFP_NOFS context to
1306	 * prevent direct reclaim recursion back into the filesystem and blowing
1307	 * stacks or deadlocking.
1308	 */
1309	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1310	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1311
1312	/*
1313	 * If there is no attribute fork no ACL can exist on this inode,
1314	 * and it can't have any file capabilities attached to it either.
1315	 */
1316	if (!XFS_IFORK_Q(ip)) {
1317		inode_has_no_xattr(inode);
1318		cache_no_acl(inode);
1319	}
1320}
1321
1322void
1323xfs_setup_iops(
1324	struct xfs_inode	*ip)
1325{
1326	struct inode		*inode = &ip->i_vnode;
1327
1328	switch (inode->i_mode & S_IFMT) {
1329	case S_IFREG:
1330		inode->i_op = &xfs_inode_operations;
1331		inode->i_fop = &xfs_file_operations;
1332		if (IS_DAX(inode))
1333			inode->i_mapping->a_ops = &xfs_dax_aops;
1334		else
1335			inode->i_mapping->a_ops = &xfs_address_space_operations;
1336		break;
1337	case S_IFDIR:
1338		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1339			inode->i_op = &xfs_dir_ci_inode_operations;
1340		else
1341			inode->i_op = &xfs_dir_inode_operations;
1342		inode->i_fop = &xfs_dir_file_operations;
1343		break;
1344	case S_IFLNK:
1345		if (ip->i_df.if_flags & XFS_IFINLINE)
1346			inode->i_op = &xfs_inline_symlink_inode_operations;
1347		else
1348			inode->i_op = &xfs_symlink_inode_operations;
1349		break;
1350	default:
1351		inode->i_op = &xfs_inode_operations;
1352		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1353		break;
1354	}
1355}
1356