xref: /linux/fs/xfs/xfs_iops.c (revision 987b741c52c7c6c68d46fbaeb95b8d1087f10b7f)
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 #include "xfs_ioctl.h"
25 
26 #include <linux/posix_acl.h>
27 #include <linux/security.h>
28 #include <linux/iversion.h>
29 #include <linux/fiemap.h>
30 
31 /*
32  * Directories have different lock order w.r.t. mmap_lock compared to regular
33  * files. This is due to readdir potentially triggering page faults on a user
34  * buffer inside filldir(), and this happens with the ilock on the directory
35  * held. For regular files, the lock order is the other way around - the
36  * mmap_lock is taken during the page fault, and then we lock the ilock to do
37  * block mapping. Hence we need a different class for the directory ilock so
38  * that lockdep can tell them apart.
39  */
40 static struct lock_class_key xfs_nondir_ilock_class;
41 static struct lock_class_key xfs_dir_ilock_class;
42 
43 static int
44 xfs_initxattrs(
45 	struct inode		*inode,
46 	const struct xattr	*xattr_array,
47 	void			*fs_info)
48 {
49 	const struct xattr	*xattr;
50 	struct xfs_inode	*ip = XFS_I(inode);
51 	int			error = 0;
52 
53 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
54 		struct xfs_da_args	args = {
55 			.dp		= ip,
56 			.attr_filter	= XFS_ATTR_SECURE,
57 			.name		= xattr->name,
58 			.namelen	= strlen(xattr->name),
59 			.value		= xattr->value,
60 			.valuelen	= xattr->value_len,
61 		};
62 		error = xfs_attr_set(&args);
63 		if (error < 0)
64 			break;
65 	}
66 	return error;
67 }
68 
69 /*
70  * Hook in SELinux.  This is not quite correct yet, what we really need
71  * here (as we do for default ACLs) is a mechanism by which creation of
72  * these attrs can be journalled at inode creation time (along with the
73  * inode, of course, such that log replay can't cause these to be lost).
74  */
75 
76 STATIC int
77 xfs_init_security(
78 	struct inode	*inode,
79 	struct inode	*dir,
80 	const struct qstr *qstr)
81 {
82 	return security_inode_init_security(inode, dir, qstr,
83 					     &xfs_initxattrs, NULL);
84 }
85 
86 static void
87 xfs_dentry_to_name(
88 	struct xfs_name	*namep,
89 	struct dentry	*dentry)
90 {
91 	namep->name = dentry->d_name.name;
92 	namep->len = dentry->d_name.len;
93 	namep->type = XFS_DIR3_FT_UNKNOWN;
94 }
95 
96 static int
97 xfs_dentry_mode_to_name(
98 	struct xfs_name	*namep,
99 	struct dentry	*dentry,
100 	int		mode)
101 {
102 	namep->name = dentry->d_name.name;
103 	namep->len = dentry->d_name.len;
104 	namep->type = xfs_mode_to_ftype(mode);
105 
106 	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
107 		return -EFSCORRUPTED;
108 
109 	return 0;
110 }
111 
112 STATIC void
113 xfs_cleanup_inode(
114 	struct inode	*dir,
115 	struct inode	*inode,
116 	struct dentry	*dentry)
117 {
118 	struct xfs_name	teardown;
119 
120 	/* Oh, the horror.
121 	 * If we can't add the ACL or we fail in
122 	 * xfs_init_security we must back out.
123 	 * ENOSPC can hit here, among other things.
124 	 */
125 	xfs_dentry_to_name(&teardown, dentry);
126 
127 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
128 }
129 
130 /*
131  * Check to see if we are likely to need an extended attribute to be added to
132  * the inode we are about to allocate. This allows the attribute fork to be
133  * created during the inode allocation, reducing the number of transactions we
134  * need to do in this fast path.
135  *
136  * The security checks are optimistic, but not guaranteed. The two LSMs that
137  * require xattrs to be added here (selinux and smack) are also the only two
138  * LSMs that add a sb->s_security structure to the superblock. Hence if security
139  * is enabled and sb->s_security is set, we have a pretty good idea that we are
140  * going to be asked to add a security xattr immediately after allocating the
141  * xfs inode and instantiating the VFS inode.
142  */
143 static inline bool
144 xfs_create_need_xattr(
145 	struct inode	*dir,
146 	struct posix_acl *default_acl,
147 	struct posix_acl *acl)
148 {
149 	if (acl)
150 		return true;
151 	if (default_acl)
152 		return true;
153 #if IS_ENABLED(CONFIG_SECURITY)
154 	if (dir->i_sb->s_security)
155 		return true;
156 #endif
157 	return false;
158 }
159 
160 
161 STATIC int
162 xfs_generic_create(
163 	struct user_namespace	*mnt_userns,
164 	struct inode	*dir,
165 	struct dentry	*dentry,
166 	umode_t		mode,
167 	dev_t		rdev,
168 	bool		tmpfile)	/* unnamed file */
169 {
170 	struct inode	*inode;
171 	struct xfs_inode *ip = NULL;
172 	struct posix_acl *default_acl, *acl;
173 	struct xfs_name	name;
174 	int		error;
175 
176 	/*
177 	 * Irix uses Missed'em'V split, but doesn't want to see
178 	 * the upper 5 bits of (14bit) major.
179 	 */
180 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
181 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
182 			return -EINVAL;
183 	} else {
184 		rdev = 0;
185 	}
186 
187 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
188 	if (error)
189 		return error;
190 
191 	/* Verify mode is valid also for tmpfile case */
192 	error = xfs_dentry_mode_to_name(&name, dentry, mode);
193 	if (unlikely(error))
194 		goto out_free_acl;
195 
196 	if (!tmpfile) {
197 		error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev,
198 				xfs_create_need_xattr(dir, default_acl, acl),
199 				&ip);
200 	} else {
201 		error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip);
202 	}
203 	if (unlikely(error))
204 		goto out_free_acl;
205 
206 	inode = VFS_I(ip);
207 
208 	error = xfs_init_security(inode, dir, &dentry->d_name);
209 	if (unlikely(error))
210 		goto out_cleanup_inode;
211 
212 #ifdef CONFIG_XFS_POSIX_ACL
213 	if (default_acl) {
214 		error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
215 		if (error)
216 			goto out_cleanup_inode;
217 	}
218 	if (acl) {
219 		error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
220 		if (error)
221 			goto out_cleanup_inode;
222 	}
223 #endif
224 
225 	xfs_setup_iops(ip);
226 
227 	if (tmpfile) {
228 		/*
229 		 * The VFS requires that any inode fed to d_tmpfile must have
230 		 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
231 		 * However, we created the temp file with nlink == 0 because
232 		 * we're not allowed to put an inode with nlink > 0 on the
233 		 * unlinked list.  Therefore we have to set nlink to 1 so that
234 		 * d_tmpfile can immediately set it back to zero.
235 		 */
236 		set_nlink(inode, 1);
237 		d_tmpfile(dentry, inode);
238 	} else
239 		d_instantiate(dentry, inode);
240 
241 	xfs_finish_inode_setup(ip);
242 
243  out_free_acl:
244 	posix_acl_release(default_acl);
245 	posix_acl_release(acl);
246 	return error;
247 
248  out_cleanup_inode:
249 	xfs_finish_inode_setup(ip);
250 	if (!tmpfile)
251 		xfs_cleanup_inode(dir, inode, dentry);
252 	xfs_irele(ip);
253 	goto out_free_acl;
254 }
255 
256 STATIC int
257 xfs_vn_mknod(
258 	struct user_namespace	*mnt_userns,
259 	struct inode		*dir,
260 	struct dentry		*dentry,
261 	umode_t			mode,
262 	dev_t			rdev)
263 {
264 	return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false);
265 }
266 
267 STATIC int
268 xfs_vn_create(
269 	struct user_namespace	*mnt_userns,
270 	struct inode		*dir,
271 	struct dentry		*dentry,
272 	umode_t			mode,
273 	bool			flags)
274 {
275 	return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false);
276 }
277 
278 STATIC int
279 xfs_vn_mkdir(
280 	struct user_namespace	*mnt_userns,
281 	struct inode		*dir,
282 	struct dentry		*dentry,
283 	umode_t			mode)
284 {
285 	return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0,
286 				  false);
287 }
288 
289 STATIC struct dentry *
290 xfs_vn_lookup(
291 	struct inode	*dir,
292 	struct dentry	*dentry,
293 	unsigned int flags)
294 {
295 	struct inode *inode;
296 	struct xfs_inode *cip;
297 	struct xfs_name	name;
298 	int		error;
299 
300 	if (dentry->d_name.len >= MAXNAMELEN)
301 		return ERR_PTR(-ENAMETOOLONG);
302 
303 	xfs_dentry_to_name(&name, dentry);
304 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
305 	if (likely(!error))
306 		inode = VFS_I(cip);
307 	else if (likely(error == -ENOENT))
308 		inode = NULL;
309 	else
310 		inode = ERR_PTR(error);
311 	return d_splice_alias(inode, dentry);
312 }
313 
314 STATIC struct dentry *
315 xfs_vn_ci_lookup(
316 	struct inode	*dir,
317 	struct dentry	*dentry,
318 	unsigned int flags)
319 {
320 	struct xfs_inode *ip;
321 	struct xfs_name	xname;
322 	struct xfs_name ci_name;
323 	struct qstr	dname;
324 	int		error;
325 
326 	if (dentry->d_name.len >= MAXNAMELEN)
327 		return ERR_PTR(-ENAMETOOLONG);
328 
329 	xfs_dentry_to_name(&xname, dentry);
330 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
331 	if (unlikely(error)) {
332 		if (unlikely(error != -ENOENT))
333 			return ERR_PTR(error);
334 		/*
335 		 * call d_add(dentry, NULL) here when d_drop_negative_children
336 		 * is called in xfs_vn_mknod (ie. allow negative dentries
337 		 * with CI filesystems).
338 		 */
339 		return NULL;
340 	}
341 
342 	/* if exact match, just splice and exit */
343 	if (!ci_name.name)
344 		return d_splice_alias(VFS_I(ip), dentry);
345 
346 	/* else case-insensitive match... */
347 	dname.name = ci_name.name;
348 	dname.len = ci_name.len;
349 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
350 	kmem_free(ci_name.name);
351 	return dentry;
352 }
353 
354 STATIC int
355 xfs_vn_link(
356 	struct dentry	*old_dentry,
357 	struct inode	*dir,
358 	struct dentry	*dentry)
359 {
360 	struct inode	*inode = d_inode(old_dentry);
361 	struct xfs_name	name;
362 	int		error;
363 
364 	error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
365 	if (unlikely(error))
366 		return error;
367 
368 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
369 	if (unlikely(error))
370 		return error;
371 
372 	ihold(inode);
373 	d_instantiate(dentry, inode);
374 	return 0;
375 }
376 
377 STATIC int
378 xfs_vn_unlink(
379 	struct inode	*dir,
380 	struct dentry	*dentry)
381 {
382 	struct xfs_name	name;
383 	int		error;
384 
385 	xfs_dentry_to_name(&name, dentry);
386 
387 	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
388 	if (error)
389 		return error;
390 
391 	/*
392 	 * With unlink, the VFS makes the dentry "negative": no inode,
393 	 * but still hashed. This is incompatible with case-insensitive
394 	 * mode, so invalidate (unhash) the dentry in CI-mode.
395 	 */
396 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
397 		d_invalidate(dentry);
398 	return 0;
399 }
400 
401 STATIC int
402 xfs_vn_symlink(
403 	struct user_namespace	*mnt_userns,
404 	struct inode		*dir,
405 	struct dentry		*dentry,
406 	const char		*symname)
407 {
408 	struct inode	*inode;
409 	struct xfs_inode *cip = NULL;
410 	struct xfs_name	name;
411 	int		error;
412 	umode_t		mode;
413 
414 	mode = S_IFLNK |
415 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
416 	error = xfs_dentry_mode_to_name(&name, dentry, mode);
417 	if (unlikely(error))
418 		goto out;
419 
420 	error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip);
421 	if (unlikely(error))
422 		goto out;
423 
424 	inode = VFS_I(cip);
425 
426 	error = xfs_init_security(inode, dir, &dentry->d_name);
427 	if (unlikely(error))
428 		goto out_cleanup_inode;
429 
430 	xfs_setup_iops(cip);
431 
432 	d_instantiate(dentry, inode);
433 	xfs_finish_inode_setup(cip);
434 	return 0;
435 
436  out_cleanup_inode:
437 	xfs_finish_inode_setup(cip);
438 	xfs_cleanup_inode(dir, inode, dentry);
439 	xfs_irele(cip);
440  out:
441 	return error;
442 }
443 
444 STATIC int
445 xfs_vn_rename(
446 	struct user_namespace	*mnt_userns,
447 	struct inode		*odir,
448 	struct dentry		*odentry,
449 	struct inode		*ndir,
450 	struct dentry		*ndentry,
451 	unsigned int		flags)
452 {
453 	struct inode	*new_inode = d_inode(ndentry);
454 	int		omode = 0;
455 	int		error;
456 	struct xfs_name	oname;
457 	struct xfs_name	nname;
458 
459 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
460 		return -EINVAL;
461 
462 	/* if we are exchanging files, we need to set i_mode of both files */
463 	if (flags & RENAME_EXCHANGE)
464 		omode = d_inode(ndentry)->i_mode;
465 
466 	error = xfs_dentry_mode_to_name(&oname, odentry, omode);
467 	if (omode && unlikely(error))
468 		return error;
469 
470 	error = xfs_dentry_mode_to_name(&nname, ndentry,
471 					d_inode(odentry)->i_mode);
472 	if (unlikely(error))
473 		return error;
474 
475 	return xfs_rename(mnt_userns, XFS_I(odir), &oname,
476 			  XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
477 			  new_inode ? XFS_I(new_inode) : NULL, flags);
478 }
479 
480 /*
481  * careful here - this function can get called recursively, so
482  * we need to be very careful about how much stack we use.
483  * uio is kmalloced for this reason...
484  */
485 STATIC const char *
486 xfs_vn_get_link(
487 	struct dentry		*dentry,
488 	struct inode		*inode,
489 	struct delayed_call	*done)
490 {
491 	char			*link;
492 	int			error = -ENOMEM;
493 
494 	if (!dentry)
495 		return ERR_PTR(-ECHILD);
496 
497 	link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
498 	if (!link)
499 		goto out_err;
500 
501 	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
502 	if (unlikely(error))
503 		goto out_kfree;
504 
505 	set_delayed_call(done, kfree_link, link);
506 	return link;
507 
508  out_kfree:
509 	kfree(link);
510  out_err:
511 	return ERR_PTR(error);
512 }
513 
514 STATIC const char *
515 xfs_vn_get_link_inline(
516 	struct dentry		*dentry,
517 	struct inode		*inode,
518 	struct delayed_call	*done)
519 {
520 	struct xfs_inode	*ip = XFS_I(inode);
521 	char			*link;
522 
523 	ASSERT(ip->i_df.if_format == XFS_DINODE_FMT_LOCAL);
524 
525 	/*
526 	 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
527 	 * if_data is junk.
528 	 */
529 	link = ip->i_df.if_u1.if_data;
530 	if (XFS_IS_CORRUPT(ip->i_mount, !link))
531 		return ERR_PTR(-EFSCORRUPTED);
532 	return link;
533 }
534 
535 static uint32_t
536 xfs_stat_blksize(
537 	struct xfs_inode	*ip)
538 {
539 	struct xfs_mount	*mp = ip->i_mount;
540 
541 	/*
542 	 * If the file blocks are being allocated from a realtime volume, then
543 	 * always return the realtime extent size.
544 	 */
545 	if (XFS_IS_REALTIME_INODE(ip))
546 		return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
547 
548 	/*
549 	 * Allow large block sizes to be reported to userspace programs if the
550 	 * "largeio" mount option is used.
551 	 *
552 	 * If compatibility mode is specified, simply return the basic unit of
553 	 * caching so that we don't get inefficient read/modify/write I/O from
554 	 * user apps. Otherwise....
555 	 *
556 	 * If the underlying volume is a stripe, then return the stripe width in
557 	 * bytes as the recommended I/O size. It is not a stripe and we've set a
558 	 * default buffered I/O size, return that, otherwise return the compat
559 	 * default.
560 	 */
561 	if (mp->m_flags & XFS_MOUNT_LARGEIO) {
562 		if (mp->m_swidth)
563 			return mp->m_swidth << mp->m_sb.sb_blocklog;
564 		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
565 			return 1U << mp->m_allocsize_log;
566 	}
567 
568 	return PAGE_SIZE;
569 }
570 
571 STATIC int
572 xfs_vn_getattr(
573 	struct user_namespace	*mnt_userns,
574 	const struct path	*path,
575 	struct kstat		*stat,
576 	u32			request_mask,
577 	unsigned int		query_flags)
578 {
579 	struct inode		*inode = d_inode(path->dentry);
580 	struct xfs_inode	*ip = XFS_I(inode);
581 	struct xfs_mount	*mp = ip->i_mount;
582 
583 	trace_xfs_getattr(ip);
584 
585 	if (XFS_FORCED_SHUTDOWN(mp))
586 		return -EIO;
587 
588 	stat->size = XFS_ISIZE(ip);
589 	stat->dev = inode->i_sb->s_dev;
590 	stat->mode = inode->i_mode;
591 	stat->nlink = inode->i_nlink;
592 	stat->uid = i_uid_into_mnt(mnt_userns, inode);
593 	stat->gid = i_gid_into_mnt(mnt_userns, inode);
594 	stat->ino = ip->i_ino;
595 	stat->atime = inode->i_atime;
596 	stat->mtime = inode->i_mtime;
597 	stat->ctime = inode->i_ctime;
598 	stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
599 
600 	if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
601 		if (request_mask & STATX_BTIME) {
602 			stat->result_mask |= STATX_BTIME;
603 			stat->btime = ip->i_crtime;
604 		}
605 	}
606 
607 	/*
608 	 * Note: If you add another clause to set an attribute flag, please
609 	 * update attributes_mask below.
610 	 */
611 	if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
612 		stat->attributes |= STATX_ATTR_IMMUTABLE;
613 	if (ip->i_diflags & XFS_DIFLAG_APPEND)
614 		stat->attributes |= STATX_ATTR_APPEND;
615 	if (ip->i_diflags & XFS_DIFLAG_NODUMP)
616 		stat->attributes |= STATX_ATTR_NODUMP;
617 
618 	stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
619 				  STATX_ATTR_APPEND |
620 				  STATX_ATTR_NODUMP);
621 
622 	switch (inode->i_mode & S_IFMT) {
623 	case S_IFBLK:
624 	case S_IFCHR:
625 		stat->blksize = BLKDEV_IOSIZE;
626 		stat->rdev = inode->i_rdev;
627 		break;
628 	default:
629 		stat->blksize = xfs_stat_blksize(ip);
630 		stat->rdev = 0;
631 		break;
632 	}
633 
634 	return 0;
635 }
636 
637 static void
638 xfs_setattr_mode(
639 	struct xfs_inode	*ip,
640 	struct iattr		*iattr)
641 {
642 	struct inode		*inode = VFS_I(ip);
643 	umode_t			mode = iattr->ia_mode;
644 
645 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
646 
647 	inode->i_mode &= S_IFMT;
648 	inode->i_mode |= mode & ~S_IFMT;
649 }
650 
651 void
652 xfs_setattr_time(
653 	struct xfs_inode	*ip,
654 	struct iattr		*iattr)
655 {
656 	struct inode		*inode = VFS_I(ip);
657 
658 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
659 
660 	if (iattr->ia_valid & ATTR_ATIME)
661 		inode->i_atime = iattr->ia_atime;
662 	if (iattr->ia_valid & ATTR_CTIME)
663 		inode->i_ctime = iattr->ia_ctime;
664 	if (iattr->ia_valid & ATTR_MTIME)
665 		inode->i_mtime = iattr->ia_mtime;
666 }
667 
668 static int
669 xfs_vn_change_ok(
670 	struct user_namespace	*mnt_userns,
671 	struct dentry		*dentry,
672 	struct iattr		*iattr)
673 {
674 	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
675 
676 	if (mp->m_flags & XFS_MOUNT_RDONLY)
677 		return -EROFS;
678 
679 	if (XFS_FORCED_SHUTDOWN(mp))
680 		return -EIO;
681 
682 	return setattr_prepare(mnt_userns, dentry, iattr);
683 }
684 
685 /*
686  * Set non-size attributes of an inode.
687  *
688  * Caution: The caller of this function is responsible for calling
689  * setattr_prepare() or otherwise verifying the change is fine.
690  */
691 static int
692 xfs_setattr_nonsize(
693 	struct user_namespace	*mnt_userns,
694 	struct xfs_inode	*ip,
695 	struct iattr		*iattr)
696 {
697 	xfs_mount_t		*mp = ip->i_mount;
698 	struct inode		*inode = VFS_I(ip);
699 	int			mask = iattr->ia_valid;
700 	xfs_trans_t		*tp;
701 	int			error;
702 	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
703 	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
704 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
705 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
706 
707 	ASSERT((mask & ATTR_SIZE) == 0);
708 
709 	/*
710 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
711 	 * before we start any other transactions. Trying to do this later
712 	 * is messy. We don't care to take a readlock to look at the ids
713 	 * in inode here, because we can't hold it across the trans_reserve.
714 	 * If the IDs do change before we take the ilock, we're covered
715 	 * because the i_*dquot fields will get updated anyway.
716 	 */
717 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
718 		uint	qflags = 0;
719 
720 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
721 			uid = iattr->ia_uid;
722 			qflags |= XFS_QMOPT_UQUOTA;
723 		} else {
724 			uid = inode->i_uid;
725 		}
726 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
727 			gid = iattr->ia_gid;
728 			qflags |= XFS_QMOPT_GQUOTA;
729 		}  else {
730 			gid = inode->i_gid;
731 		}
732 
733 		/*
734 		 * We take a reference when we initialize udqp and gdqp,
735 		 * so it is important that we never blindly double trip on
736 		 * the same variable. See xfs_create() for an example.
737 		 */
738 		ASSERT(udqp == NULL);
739 		ASSERT(gdqp == NULL);
740 		error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
741 					   qflags, &udqp, &gdqp, NULL);
742 		if (error)
743 			return error;
744 	}
745 
746 	error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
747 			capable(CAP_FOWNER), &tp);
748 	if (error)
749 		goto out_dqrele;
750 
751 	/*
752 	 * Change file ownership.  Must be the owner or privileged.
753 	 */
754 	if (mask & (ATTR_UID|ATTR_GID)) {
755 		/*
756 		 * These IDs could have changed since we last looked at them.
757 		 * But, we're assured that if the ownership did change
758 		 * while we didn't have the inode locked, inode's dquot(s)
759 		 * would have changed also.
760 		 */
761 		iuid = inode->i_uid;
762 		igid = inode->i_gid;
763 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
764 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
765 
766 		/*
767 		 * CAP_FSETID overrides the following restrictions:
768 		 *
769 		 * The set-user-ID and set-group-ID bits of a file will be
770 		 * cleared upon successful return from chown()
771 		 */
772 		if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
773 		    !capable(CAP_FSETID))
774 			inode->i_mode &= ~(S_ISUID|S_ISGID);
775 
776 		/*
777 		 * Change the ownerships and register quota modifications
778 		 * in the transaction.
779 		 */
780 		if (!uid_eq(iuid, uid)) {
781 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
782 				ASSERT(mask & ATTR_UID);
783 				ASSERT(udqp);
784 				olddquot1 = xfs_qm_vop_chown(tp, ip,
785 							&ip->i_udquot, udqp);
786 			}
787 			inode->i_uid = uid;
788 		}
789 		if (!gid_eq(igid, gid)) {
790 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
791 				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
792 				       !XFS_IS_PQUOTA_ON(mp));
793 				ASSERT(mask & ATTR_GID);
794 				ASSERT(gdqp);
795 				olddquot2 = xfs_qm_vop_chown(tp, ip,
796 							&ip->i_gdquot, gdqp);
797 			}
798 			inode->i_gid = gid;
799 		}
800 	}
801 
802 	if (mask & ATTR_MODE)
803 		xfs_setattr_mode(ip, iattr);
804 	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
805 		xfs_setattr_time(ip, iattr);
806 
807 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
808 
809 	XFS_STATS_INC(mp, xs_ig_attrchg);
810 
811 	if (mp->m_flags & XFS_MOUNT_WSYNC)
812 		xfs_trans_set_sync(tp);
813 	error = xfs_trans_commit(tp);
814 
815 	/*
816 	 * Release any dquot(s) the inode had kept before chown.
817 	 */
818 	xfs_qm_dqrele(olddquot1);
819 	xfs_qm_dqrele(olddquot2);
820 	xfs_qm_dqrele(udqp);
821 	xfs_qm_dqrele(gdqp);
822 
823 	if (error)
824 		return error;
825 
826 	/*
827 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
828 	 * 	     update.  We could avoid this with linked transactions
829 	 * 	     and passing down the transaction pointer all the way
830 	 *	     to attr_set.  No previous user of the generic
831 	 * 	     Posix ACL code seems to care about this issue either.
832 	 */
833 	if (mask & ATTR_MODE) {
834 		error = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
835 		if (error)
836 			return error;
837 	}
838 
839 	return 0;
840 
841 out_dqrele:
842 	xfs_qm_dqrele(udqp);
843 	xfs_qm_dqrele(gdqp);
844 	return error;
845 }
846 
847 /*
848  * Truncate file.  Must have write permission and not be a directory.
849  *
850  * Caution: The caller of this function is responsible for calling
851  * setattr_prepare() or otherwise verifying the change is fine.
852  */
853 STATIC int
854 xfs_setattr_size(
855 	struct user_namespace	*mnt_userns,
856 	struct xfs_inode	*ip,
857 	struct iattr		*iattr)
858 {
859 	struct xfs_mount	*mp = ip->i_mount;
860 	struct inode		*inode = VFS_I(ip);
861 	xfs_off_t		oldsize, newsize;
862 	struct xfs_trans	*tp;
863 	int			error;
864 	uint			lock_flags = 0;
865 	bool			did_zeroing = false;
866 
867 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
868 	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
869 	ASSERT(S_ISREG(inode->i_mode));
870 	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
871 		ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
872 
873 	oldsize = inode->i_size;
874 	newsize = iattr->ia_size;
875 
876 	/*
877 	 * Short circuit the truncate case for zero length files.
878 	 */
879 	if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
880 		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
881 			return 0;
882 
883 		/*
884 		 * Use the regular setattr path to update the timestamps.
885 		 */
886 		iattr->ia_valid &= ~ATTR_SIZE;
887 		return xfs_setattr_nonsize(mnt_userns, ip, iattr);
888 	}
889 
890 	/*
891 	 * Make sure that the dquots are attached to the inode.
892 	 */
893 	error = xfs_qm_dqattach(ip);
894 	if (error)
895 		return error;
896 
897 	/*
898 	 * Wait for all direct I/O to complete.
899 	 */
900 	inode_dio_wait(inode);
901 
902 	/*
903 	 * File data changes must be complete before we start the transaction to
904 	 * modify the inode.  This needs to be done before joining the inode to
905 	 * the transaction because the inode cannot be unlocked once it is a
906 	 * part of the transaction.
907 	 *
908 	 * Start with zeroing any data beyond EOF that we may expose on file
909 	 * extension, or zeroing out the rest of the block on a downward
910 	 * truncate.
911 	 */
912 	if (newsize > oldsize) {
913 		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
914 		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
915 				&did_zeroing, &xfs_buffered_write_iomap_ops);
916 	} else {
917 		/*
918 		 * iomap won't detect a dirty page over an unwritten block (or a
919 		 * cow block over a hole) and subsequently skips zeroing the
920 		 * newly post-EOF portion of the page. Flush the new EOF to
921 		 * convert the block before the pagecache truncate.
922 		 */
923 		error = filemap_write_and_wait_range(inode->i_mapping, newsize,
924 						     newsize);
925 		if (error)
926 			return error;
927 		error = iomap_truncate_page(inode, newsize, &did_zeroing,
928 				&xfs_buffered_write_iomap_ops);
929 	}
930 
931 	if (error)
932 		return error;
933 
934 	/*
935 	 * We've already locked out new page faults, so now we can safely remove
936 	 * pages from the page cache knowing they won't get refaulted until we
937 	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
938 	 * complete. The truncate_setsize() call also cleans partial EOF page
939 	 * PTEs on extending truncates and hence ensures sub-page block size
940 	 * filesystems are correctly handled, too.
941 	 *
942 	 * We have to do all the page cache truncate work outside the
943 	 * transaction context as the "lock" order is page lock->log space
944 	 * reservation as defined by extent allocation in the writeback path.
945 	 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
946 	 * having already truncated the in-memory version of the file (i.e. made
947 	 * user visible changes). There's not much we can do about this, except
948 	 * to hope that the caller sees ENOMEM and retries the truncate
949 	 * operation.
950 	 *
951 	 * And we update in-core i_size and truncate page cache beyond newsize
952 	 * before writeback the [i_disk_size, newsize] range, so we're
953 	 * guaranteed not to write stale data past the new EOF on truncate down.
954 	 */
955 	truncate_setsize(inode, newsize);
956 
957 	/*
958 	 * We are going to log the inode size change in this transaction so
959 	 * any previous writes that are beyond the on disk EOF and the new
960 	 * EOF that have not been written out need to be written here.  If we
961 	 * do not write the data out, we expose ourselves to the null files
962 	 * problem. Note that this includes any block zeroing we did above;
963 	 * otherwise those blocks may not be zeroed after a crash.
964 	 */
965 	if (did_zeroing ||
966 	    (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
967 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
968 						ip->i_disk_size, newsize - 1);
969 		if (error)
970 			return error;
971 	}
972 
973 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
974 	if (error)
975 		return error;
976 
977 	lock_flags |= XFS_ILOCK_EXCL;
978 	xfs_ilock(ip, XFS_ILOCK_EXCL);
979 	xfs_trans_ijoin(tp, ip, 0);
980 
981 	/*
982 	 * Only change the c/mtime if we are changing the size or we are
983 	 * explicitly asked to change it.  This handles the semantic difference
984 	 * between truncate() and ftruncate() as implemented in the VFS.
985 	 *
986 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
987 	 * special case where we need to update the times despite not having
988 	 * these flags set.  For all other operations the VFS set these flags
989 	 * explicitly if it wants a timestamp update.
990 	 */
991 	if (newsize != oldsize &&
992 	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
993 		iattr->ia_ctime = iattr->ia_mtime =
994 			current_time(inode);
995 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
996 	}
997 
998 	/*
999 	 * The first thing we do is set the size to new_size permanently on
1000 	 * disk.  This way we don't have to worry about anyone ever being able
1001 	 * to look at the data being freed even in the face of a crash.
1002 	 * What we're getting around here is the case where we free a block, it
1003 	 * is allocated to another file, it is written to, and then we crash.
1004 	 * If the new data gets written to the file but the log buffers
1005 	 * containing the free and reallocation don't, then we'd end up with
1006 	 * garbage in the blocks being freed.  As long as we make the new size
1007 	 * permanent before actually freeing any blocks it doesn't matter if
1008 	 * they get written to.
1009 	 */
1010 	ip->i_disk_size = newsize;
1011 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1012 
1013 	if (newsize <= oldsize) {
1014 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1015 		if (error)
1016 			goto out_trans_cancel;
1017 
1018 		/*
1019 		 * Truncated "down", so we're removing references to old data
1020 		 * here - if we delay flushing for a long time, we expose
1021 		 * ourselves unduly to the notorious NULL files problem.  So,
1022 		 * we mark this inode and flush it when the file is closed,
1023 		 * and do not wait the usual (long) time for writeout.
1024 		 */
1025 		xfs_iflags_set(ip, XFS_ITRUNCATED);
1026 
1027 		/* A truncate down always removes post-EOF blocks. */
1028 		xfs_inode_clear_eofblocks_tag(ip);
1029 	}
1030 
1031 	if (iattr->ia_valid & ATTR_MODE)
1032 		xfs_setattr_mode(ip, iattr);
1033 	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1034 		xfs_setattr_time(ip, iattr);
1035 
1036 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1037 
1038 	XFS_STATS_INC(mp, xs_ig_attrchg);
1039 
1040 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1041 		xfs_trans_set_sync(tp);
1042 
1043 	error = xfs_trans_commit(tp);
1044 out_unlock:
1045 	if (lock_flags)
1046 		xfs_iunlock(ip, lock_flags);
1047 	return error;
1048 
1049 out_trans_cancel:
1050 	xfs_trans_cancel(tp);
1051 	goto out_unlock;
1052 }
1053 
1054 int
1055 xfs_vn_setattr_size(
1056 	struct user_namespace	*mnt_userns,
1057 	struct dentry		*dentry,
1058 	struct iattr		*iattr)
1059 {
1060 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1061 	int error;
1062 
1063 	trace_xfs_setattr(ip);
1064 
1065 	error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1066 	if (error)
1067 		return error;
1068 	return xfs_setattr_size(mnt_userns, ip, iattr);
1069 }
1070 
1071 STATIC int
1072 xfs_vn_setattr(
1073 	struct user_namespace	*mnt_userns,
1074 	struct dentry		*dentry,
1075 	struct iattr		*iattr)
1076 {
1077 	struct inode		*inode = d_inode(dentry);
1078 	struct xfs_inode	*ip = XFS_I(inode);
1079 	int			error;
1080 
1081 	if (iattr->ia_valid & ATTR_SIZE) {
1082 		uint			iolock;
1083 
1084 		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1085 		iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1086 
1087 		error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1088 		if (error) {
1089 			xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1090 			return error;
1091 		}
1092 
1093 		error = xfs_vn_setattr_size(mnt_userns, dentry, iattr);
1094 		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1095 	} else {
1096 		trace_xfs_setattr(ip);
1097 
1098 		error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1099 		if (!error)
1100 			error = xfs_setattr_nonsize(mnt_userns, ip, iattr);
1101 	}
1102 
1103 	return error;
1104 }
1105 
1106 STATIC int
1107 xfs_vn_update_time(
1108 	struct inode		*inode,
1109 	struct timespec64	*now,
1110 	int			flags)
1111 {
1112 	struct xfs_inode	*ip = XFS_I(inode);
1113 	struct xfs_mount	*mp = ip->i_mount;
1114 	int			log_flags = XFS_ILOG_TIMESTAMP;
1115 	struct xfs_trans	*tp;
1116 	int			error;
1117 
1118 	trace_xfs_update_time(ip);
1119 
1120 	if (inode->i_sb->s_flags & SB_LAZYTIME) {
1121 		if (!((flags & S_VERSION) &&
1122 		      inode_maybe_inc_iversion(inode, false)))
1123 			return generic_update_time(inode, now, flags);
1124 
1125 		/* Capture the iversion update that just occurred */
1126 		log_flags |= XFS_ILOG_CORE;
1127 	}
1128 
1129 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1130 	if (error)
1131 		return error;
1132 
1133 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1134 	if (flags & S_CTIME)
1135 		inode->i_ctime = *now;
1136 	if (flags & S_MTIME)
1137 		inode->i_mtime = *now;
1138 	if (flags & S_ATIME)
1139 		inode->i_atime = *now;
1140 
1141 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1142 	xfs_trans_log_inode(tp, ip, log_flags);
1143 	return xfs_trans_commit(tp);
1144 }
1145 
1146 STATIC int
1147 xfs_vn_fiemap(
1148 	struct inode		*inode,
1149 	struct fiemap_extent_info *fieinfo,
1150 	u64			start,
1151 	u64			length)
1152 {
1153 	int			error;
1154 
1155 	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1156 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1157 		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1158 		error = iomap_fiemap(inode, fieinfo, start, length,
1159 				&xfs_xattr_iomap_ops);
1160 	} else {
1161 		error = iomap_fiemap(inode, fieinfo, start, length,
1162 				&xfs_read_iomap_ops);
1163 	}
1164 	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1165 
1166 	return error;
1167 }
1168 
1169 STATIC int
1170 xfs_vn_tmpfile(
1171 	struct user_namespace	*mnt_userns,
1172 	struct inode		*dir,
1173 	struct dentry		*dentry,
1174 	umode_t			mode)
1175 {
1176 	return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true);
1177 }
1178 
1179 static const struct inode_operations xfs_inode_operations = {
1180 	.get_acl		= xfs_get_acl,
1181 	.set_acl		= xfs_set_acl,
1182 	.getattr		= xfs_vn_getattr,
1183 	.setattr		= xfs_vn_setattr,
1184 	.listxattr		= xfs_vn_listxattr,
1185 	.fiemap			= xfs_vn_fiemap,
1186 	.update_time		= xfs_vn_update_time,
1187 	.fileattr_get		= xfs_fileattr_get,
1188 	.fileattr_set		= xfs_fileattr_set,
1189 };
1190 
1191 static const struct inode_operations xfs_dir_inode_operations = {
1192 	.create			= xfs_vn_create,
1193 	.lookup			= xfs_vn_lookup,
1194 	.link			= xfs_vn_link,
1195 	.unlink			= xfs_vn_unlink,
1196 	.symlink		= xfs_vn_symlink,
1197 	.mkdir			= xfs_vn_mkdir,
1198 	/*
1199 	 * Yes, XFS uses the same method for rmdir and unlink.
1200 	 *
1201 	 * There are some subtile differences deeper in the code,
1202 	 * but we use S_ISDIR to check for those.
1203 	 */
1204 	.rmdir			= xfs_vn_unlink,
1205 	.mknod			= xfs_vn_mknod,
1206 	.rename			= xfs_vn_rename,
1207 	.get_acl		= xfs_get_acl,
1208 	.set_acl		= xfs_set_acl,
1209 	.getattr		= xfs_vn_getattr,
1210 	.setattr		= xfs_vn_setattr,
1211 	.listxattr		= xfs_vn_listxattr,
1212 	.update_time		= xfs_vn_update_time,
1213 	.tmpfile		= xfs_vn_tmpfile,
1214 	.fileattr_get		= xfs_fileattr_get,
1215 	.fileattr_set		= xfs_fileattr_set,
1216 };
1217 
1218 static const struct inode_operations xfs_dir_ci_inode_operations = {
1219 	.create			= xfs_vn_create,
1220 	.lookup			= xfs_vn_ci_lookup,
1221 	.link			= xfs_vn_link,
1222 	.unlink			= xfs_vn_unlink,
1223 	.symlink		= xfs_vn_symlink,
1224 	.mkdir			= xfs_vn_mkdir,
1225 	/*
1226 	 * Yes, XFS uses the same method for rmdir and unlink.
1227 	 *
1228 	 * There are some subtile differences deeper in the code,
1229 	 * but we use S_ISDIR to check for those.
1230 	 */
1231 	.rmdir			= xfs_vn_unlink,
1232 	.mknod			= xfs_vn_mknod,
1233 	.rename			= xfs_vn_rename,
1234 	.get_acl		= xfs_get_acl,
1235 	.set_acl		= xfs_set_acl,
1236 	.getattr		= xfs_vn_getattr,
1237 	.setattr		= xfs_vn_setattr,
1238 	.listxattr		= xfs_vn_listxattr,
1239 	.update_time		= xfs_vn_update_time,
1240 	.tmpfile		= xfs_vn_tmpfile,
1241 	.fileattr_get		= xfs_fileattr_get,
1242 	.fileattr_set		= xfs_fileattr_set,
1243 };
1244 
1245 static const struct inode_operations xfs_symlink_inode_operations = {
1246 	.get_link		= xfs_vn_get_link,
1247 	.getattr		= xfs_vn_getattr,
1248 	.setattr		= xfs_vn_setattr,
1249 	.listxattr		= xfs_vn_listxattr,
1250 	.update_time		= xfs_vn_update_time,
1251 };
1252 
1253 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1254 	.get_link		= xfs_vn_get_link_inline,
1255 	.getattr		= xfs_vn_getattr,
1256 	.setattr		= xfs_vn_setattr,
1257 	.listxattr		= xfs_vn_listxattr,
1258 	.update_time		= xfs_vn_update_time,
1259 };
1260 
1261 /* Figure out if this file actually supports DAX. */
1262 static bool
1263 xfs_inode_supports_dax(
1264 	struct xfs_inode	*ip)
1265 {
1266 	struct xfs_mount	*mp = ip->i_mount;
1267 
1268 	/* Only supported on regular files. */
1269 	if (!S_ISREG(VFS_I(ip)->i_mode))
1270 		return false;
1271 
1272 	/* Only supported on non-reflinked files. */
1273 	if (xfs_is_reflink_inode(ip))
1274 		return false;
1275 
1276 	/* Block size must match page size */
1277 	if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1278 		return false;
1279 
1280 	/* Device has to support DAX too. */
1281 	return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1282 }
1283 
1284 static bool
1285 xfs_inode_should_enable_dax(
1286 	struct xfs_inode *ip)
1287 {
1288 	if (!IS_ENABLED(CONFIG_FS_DAX))
1289 		return false;
1290 	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1291 		return false;
1292 	if (!xfs_inode_supports_dax(ip))
1293 		return false;
1294 	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1295 		return true;
1296 	if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1297 		return true;
1298 	return false;
1299 }
1300 
1301 void
1302 xfs_diflags_to_iflags(
1303 	struct xfs_inode	*ip,
1304 	bool init)
1305 {
1306 	struct inode            *inode = VFS_I(ip);
1307 	unsigned int            xflags = xfs_ip2xflags(ip);
1308 	unsigned int            flags = 0;
1309 
1310 	ASSERT(!(IS_DAX(inode) && init));
1311 
1312 	if (xflags & FS_XFLAG_IMMUTABLE)
1313 		flags |= S_IMMUTABLE;
1314 	if (xflags & FS_XFLAG_APPEND)
1315 		flags |= S_APPEND;
1316 	if (xflags & FS_XFLAG_SYNC)
1317 		flags |= S_SYNC;
1318 	if (xflags & FS_XFLAG_NOATIME)
1319 		flags |= S_NOATIME;
1320 	if (init && xfs_inode_should_enable_dax(ip))
1321 		flags |= S_DAX;
1322 
1323 	/*
1324 	 * S_DAX can only be set during inode initialization and is never set by
1325 	 * the VFS, so we cannot mask off S_DAX in i_flags.
1326 	 */
1327 	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1328 	inode->i_flags |= flags;
1329 }
1330 
1331 /*
1332  * Initialize the Linux inode.
1333  *
1334  * When reading existing inodes from disk this is called directly from xfs_iget,
1335  * when creating a new inode it is called from xfs_ialloc after setting up the
1336  * inode. These callers have different criteria for clearing XFS_INEW, so leave
1337  * it up to the caller to deal with unlocking the inode appropriately.
1338  */
1339 void
1340 xfs_setup_inode(
1341 	struct xfs_inode	*ip)
1342 {
1343 	struct inode		*inode = &ip->i_vnode;
1344 	gfp_t			gfp_mask;
1345 
1346 	inode->i_ino = ip->i_ino;
1347 	inode->i_state = I_NEW;
1348 
1349 	inode_sb_list_add(inode);
1350 	/* make the inode look hashed for the writeback code */
1351 	inode_fake_hash(inode);
1352 
1353 	i_size_write(inode, ip->i_disk_size);
1354 	xfs_diflags_to_iflags(ip, true);
1355 
1356 	if (S_ISDIR(inode->i_mode)) {
1357 		/*
1358 		 * We set the i_rwsem class here to avoid potential races with
1359 		 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1360 		 * after a filehandle lookup has already found the inode in
1361 		 * cache before it has been unlocked via unlock_new_inode().
1362 		 */
1363 		lockdep_set_class(&inode->i_rwsem,
1364 				  &inode->i_sb->s_type->i_mutex_dir_key);
1365 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1366 	} else {
1367 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1368 	}
1369 
1370 	/*
1371 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1372 	 * prevent direct reclaim recursion back into the filesystem and blowing
1373 	 * stacks or deadlocking.
1374 	 */
1375 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1376 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1377 
1378 	/*
1379 	 * If there is no attribute fork no ACL can exist on this inode,
1380 	 * and it can't have any file capabilities attached to it either.
1381 	 */
1382 	if (!XFS_IFORK_Q(ip)) {
1383 		inode_has_no_xattr(inode);
1384 		cache_no_acl(inode);
1385 	}
1386 }
1387 
1388 void
1389 xfs_setup_iops(
1390 	struct xfs_inode	*ip)
1391 {
1392 	struct inode		*inode = &ip->i_vnode;
1393 
1394 	switch (inode->i_mode & S_IFMT) {
1395 	case S_IFREG:
1396 		inode->i_op = &xfs_inode_operations;
1397 		inode->i_fop = &xfs_file_operations;
1398 		if (IS_DAX(inode))
1399 			inode->i_mapping->a_ops = &xfs_dax_aops;
1400 		else
1401 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1402 		break;
1403 	case S_IFDIR:
1404 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1405 			inode->i_op = &xfs_dir_ci_inode_operations;
1406 		else
1407 			inode->i_op = &xfs_dir_inode_operations;
1408 		inode->i_fop = &xfs_dir_file_operations;
1409 		break;
1410 	case S_IFLNK:
1411 		if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL)
1412 			inode->i_op = &xfs_inline_symlink_inode_operations;
1413 		else
1414 			inode->i_op = &xfs_symlink_inode_operations;
1415 		break;
1416 	default:
1417 		inode->i_op = &xfs_inode_operations;
1418 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1419 		break;
1420 	}
1421 }
1422