xref: /linux/fs/xfs/xfs_iops.c (revision 572af9f284669d31d9175122bbef9bc62cea8ded)
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_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_attr.h"
19 #include "xfs_trans.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_trace.h"
23 #include "xfs_icache.h"
24 #include "xfs_symlink.h"
25 #include "xfs_dir2.h"
26 #include "xfs_iomap.h"
27 #include "xfs_error.h"
28 #include "xfs_ioctl.h"
29 #include "xfs_xattr.h"
30 #include "xfs_file.h"
31 #include "xfs_bmap.h"
32 
33 #include <linux/posix_acl.h>
34 #include <linux/security.h>
35 #include <linux/iversion.h>
36 #include <linux/fiemap.h>
37 
38 /*
39  * Directories have different lock order w.r.t. mmap_lock compared to regular
40  * files. This is due to readdir potentially triggering page faults on a user
41  * buffer inside filldir(), and this happens with the ilock on the directory
42  * held. For regular files, the lock order is the other way around - the
43  * mmap_lock is taken during the page fault, and then we lock the ilock to do
44  * block mapping. Hence we need a different class for the directory ilock so
45  * that lockdep can tell them apart.  Directories in the metadata directory
46  * tree get a separate class so that lockdep reports will warn us if someone
47  * ever tries to lock regular directories after locking metadata directories.
48  */
49 static struct lock_class_key xfs_nondir_ilock_class;
50 static struct lock_class_key xfs_dir_ilock_class;
51 
52 static int
53 xfs_initxattrs(
54 	struct inode		*inode,
55 	const struct xattr	*xattr_array,
56 	void			*fs_info)
57 {
58 	const struct xattr	*xattr;
59 	struct xfs_inode	*ip = XFS_I(inode);
60 	int			error = 0;
61 
62 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
63 		struct xfs_da_args	args = {
64 			.dp		= ip,
65 			.attr_filter	= XFS_ATTR_SECURE,
66 			.name		= xattr->name,
67 			.namelen	= strlen(xattr->name),
68 			.value		= xattr->value,
69 			.valuelen	= xattr->value_len,
70 		};
71 		error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT);
72 		if (error < 0)
73 			break;
74 	}
75 	return error;
76 }
77 
78 /*
79  * Hook in SELinux.  This is not quite correct yet, what we really need
80  * here (as we do for default ACLs) is a mechanism by which creation of
81  * these attrs can be journalled at inode creation time (along with the
82  * inode, of course, such that log replay can't cause these to be lost).
83  */
84 int
85 xfs_inode_init_security(
86 	struct inode	*inode,
87 	struct inode	*dir,
88 	const struct qstr *qstr)
89 {
90 	return security_inode_init_security(inode, dir, qstr,
91 					     &xfs_initxattrs, NULL);
92 }
93 
94 static void
95 xfs_dentry_to_name(
96 	struct xfs_name	*namep,
97 	struct dentry	*dentry)
98 {
99 	namep->name = dentry->d_name.name;
100 	namep->len = dentry->d_name.len;
101 	namep->type = XFS_DIR3_FT_UNKNOWN;
102 }
103 
104 static int
105 xfs_dentry_mode_to_name(
106 	struct xfs_name	*namep,
107 	struct dentry	*dentry,
108 	int		mode)
109 {
110 	namep->name = dentry->d_name.name;
111 	namep->len = dentry->d_name.len;
112 	namep->type = xfs_mode_to_ftype(mode);
113 
114 	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
115 		return -EFSCORRUPTED;
116 
117 	return 0;
118 }
119 
120 STATIC void
121 xfs_cleanup_inode(
122 	struct inode	*dir,
123 	struct inode	*inode,
124 	struct dentry	*dentry)
125 {
126 	struct xfs_name	teardown;
127 
128 	/* Oh, the horror.
129 	 * If we can't add the ACL or we fail in
130 	 * xfs_inode_init_security we must back out.
131 	 * ENOSPC can hit here, among other things.
132 	 */
133 	xfs_dentry_to_name(&teardown, dentry);
134 
135 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
136 }
137 
138 /*
139  * Check to see if we are likely to need an extended attribute to be added to
140  * the inode we are about to allocate. This allows the attribute fork to be
141  * created during the inode allocation, reducing the number of transactions we
142  * need to do in this fast path.
143  *
144  * The security checks are optimistic, but not guaranteed. The two LSMs that
145  * require xattrs to be added here (selinux and smack) are also the only two
146  * LSMs that add a sb->s_security structure to the superblock. Hence if security
147  * is enabled and sb->s_security is set, we have a pretty good idea that we are
148  * going to be asked to add a security xattr immediately after allocating the
149  * xfs inode and instantiating the VFS inode.
150  */
151 static inline bool
152 xfs_create_need_xattr(
153 	struct inode	*dir,
154 	struct posix_acl *default_acl,
155 	struct posix_acl *acl)
156 {
157 	if (acl)
158 		return true;
159 	if (default_acl)
160 		return true;
161 #if IS_ENABLED(CONFIG_SECURITY)
162 	if (dir->i_sb->s_security)
163 		return true;
164 #endif
165 	return false;
166 }
167 
168 
169 STATIC int
170 xfs_generic_create(
171 	struct mnt_idmap	*idmap,
172 	struct inode		*dir,
173 	struct dentry		*dentry,
174 	umode_t			mode,
175 	dev_t			rdev,
176 	struct file		*tmpfile)	/* unnamed file */
177 {
178 	struct xfs_icreate_args	args = {
179 		.idmap		= idmap,
180 		.pip		= XFS_I(dir),
181 		.rdev		= rdev,
182 		.mode		= mode,
183 	};
184 	struct inode		*inode;
185 	struct xfs_inode	*ip = NULL;
186 	struct posix_acl	*default_acl, *acl;
187 	struct xfs_name		name;
188 	int			error;
189 
190 	/*
191 	 * Irix uses Missed'em'V split, but doesn't want to see
192 	 * the upper 5 bits of (14bit) major.
193 	 */
194 	if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) {
195 		if (unlikely(!sysv_valid_dev(args.rdev) ||
196 			     MAJOR(args.rdev) & ~0x1ff))
197 			return -EINVAL;
198 	} else {
199 		args.rdev = 0;
200 	}
201 
202 	error = posix_acl_create(dir, &args.mode, &default_acl, &acl);
203 	if (error)
204 		return error;
205 
206 	/* Verify mode is valid also for tmpfile case */
207 	error = xfs_dentry_mode_to_name(&name, dentry, args.mode);
208 	if (unlikely(error))
209 		goto out_free_acl;
210 
211 	if (!tmpfile) {
212 		if (xfs_create_need_xattr(dir, default_acl, acl))
213 			args.flags |= XFS_ICREATE_INIT_XATTRS;
214 
215 		error = xfs_create(&args, &name, &ip);
216 	} else {
217 		args.flags |= XFS_ICREATE_TMPFILE;
218 
219 		/*
220 		 * If this temporary file will not be linkable, don't bother
221 		 * creating an attr fork to receive a parent pointer.
222 		 */
223 		if (tmpfile->f_flags & O_EXCL)
224 			args.flags |= XFS_ICREATE_UNLINKABLE;
225 
226 		error = xfs_create_tmpfile(&args, &ip);
227 	}
228 	if (unlikely(error))
229 		goto out_free_acl;
230 
231 	inode = VFS_I(ip);
232 
233 	error = xfs_inode_init_security(inode, dir, &dentry->d_name);
234 	if (unlikely(error))
235 		goto out_cleanup_inode;
236 
237 	if (default_acl) {
238 		error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
239 		if (error)
240 			goto out_cleanup_inode;
241 	}
242 	if (acl) {
243 		error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
244 		if (error)
245 			goto out_cleanup_inode;
246 	}
247 
248 	xfs_setup_iops(ip);
249 
250 	if (tmpfile) {
251 		/*
252 		 * The VFS requires that any inode fed to d_tmpfile must have
253 		 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
254 		 * However, we created the temp file with nlink == 0 because
255 		 * we're not allowed to put an inode with nlink > 0 on the
256 		 * unlinked list.  Therefore we have to set nlink to 1 so that
257 		 * d_tmpfile can immediately set it back to zero.
258 		 */
259 		set_nlink(inode, 1);
260 		d_tmpfile(tmpfile, inode);
261 	} else
262 		d_instantiate(dentry, inode);
263 
264 	xfs_finish_inode_setup(ip);
265 
266  out_free_acl:
267 	posix_acl_release(default_acl);
268 	posix_acl_release(acl);
269 	return error;
270 
271  out_cleanup_inode:
272 	xfs_finish_inode_setup(ip);
273 	if (!tmpfile)
274 		xfs_cleanup_inode(dir, inode, dentry);
275 	xfs_irele(ip);
276 	goto out_free_acl;
277 }
278 
279 STATIC int
280 xfs_vn_mknod(
281 	struct mnt_idmap	*idmap,
282 	struct inode		*dir,
283 	struct dentry		*dentry,
284 	umode_t			mode,
285 	dev_t			rdev)
286 {
287 	return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
288 }
289 
290 STATIC int
291 xfs_vn_create(
292 	struct mnt_idmap	*idmap,
293 	struct inode		*dir,
294 	struct dentry		*dentry,
295 	umode_t			mode,
296 	bool			flags)
297 {
298 	return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
299 }
300 
301 STATIC int
302 xfs_vn_mkdir(
303 	struct mnt_idmap	*idmap,
304 	struct inode		*dir,
305 	struct dentry		*dentry,
306 	umode_t			mode)
307 {
308 	return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
309 }
310 
311 STATIC struct dentry *
312 xfs_vn_lookup(
313 	struct inode	*dir,
314 	struct dentry	*dentry,
315 	unsigned int flags)
316 {
317 	struct inode *inode;
318 	struct xfs_inode *cip;
319 	struct xfs_name	name;
320 	int		error;
321 
322 	if (dentry->d_name.len >= MAXNAMELEN)
323 		return ERR_PTR(-ENAMETOOLONG);
324 
325 	xfs_dentry_to_name(&name, dentry);
326 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
327 	if (likely(!error))
328 		inode = VFS_I(cip);
329 	else if (likely(error == -ENOENT))
330 		inode = NULL;
331 	else
332 		inode = ERR_PTR(error);
333 	return d_splice_alias(inode, dentry);
334 }
335 
336 STATIC struct dentry *
337 xfs_vn_ci_lookup(
338 	struct inode	*dir,
339 	struct dentry	*dentry,
340 	unsigned int flags)
341 {
342 	struct xfs_inode *ip;
343 	struct xfs_name	xname;
344 	struct xfs_name ci_name;
345 	struct qstr	dname;
346 	int		error;
347 
348 	if (dentry->d_name.len >= MAXNAMELEN)
349 		return ERR_PTR(-ENAMETOOLONG);
350 
351 	xfs_dentry_to_name(&xname, dentry);
352 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
353 	if (unlikely(error)) {
354 		if (unlikely(error != -ENOENT))
355 			return ERR_PTR(error);
356 		/*
357 		 * call d_add(dentry, NULL) here when d_drop_negative_children
358 		 * is called in xfs_vn_mknod (ie. allow negative dentries
359 		 * with CI filesystems).
360 		 */
361 		return NULL;
362 	}
363 
364 	/* if exact match, just splice and exit */
365 	if (!ci_name.name)
366 		return d_splice_alias(VFS_I(ip), dentry);
367 
368 	/* else case-insensitive match... */
369 	dname.name = ci_name.name;
370 	dname.len = ci_name.len;
371 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
372 	kfree(ci_name.name);
373 	return dentry;
374 }
375 
376 STATIC int
377 xfs_vn_link(
378 	struct dentry	*old_dentry,
379 	struct inode	*dir,
380 	struct dentry	*dentry)
381 {
382 	struct inode	*inode = d_inode(old_dentry);
383 	struct xfs_name	name;
384 	int		error;
385 
386 	error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
387 	if (unlikely(error))
388 		return error;
389 
390 	if (IS_PRIVATE(inode))
391 		return -EPERM;
392 
393 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
394 	if (unlikely(error))
395 		return error;
396 
397 	ihold(inode);
398 	d_instantiate(dentry, inode);
399 	return 0;
400 }
401 
402 STATIC int
403 xfs_vn_unlink(
404 	struct inode	*dir,
405 	struct dentry	*dentry)
406 {
407 	struct xfs_name	name;
408 	int		error;
409 
410 	xfs_dentry_to_name(&name, dentry);
411 
412 	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
413 	if (error)
414 		return error;
415 
416 	/*
417 	 * With unlink, the VFS makes the dentry "negative": no inode,
418 	 * but still hashed. This is incompatible with case-insensitive
419 	 * mode, so invalidate (unhash) the dentry in CI-mode.
420 	 */
421 	if (xfs_has_asciici(XFS_M(dir->i_sb)))
422 		d_invalidate(dentry);
423 	return 0;
424 }
425 
426 STATIC int
427 xfs_vn_symlink(
428 	struct mnt_idmap	*idmap,
429 	struct inode		*dir,
430 	struct dentry		*dentry,
431 	const char		*symname)
432 {
433 	struct inode	*inode;
434 	struct xfs_inode *cip = NULL;
435 	struct xfs_name	name;
436 	int		error;
437 	umode_t		mode;
438 
439 	mode = S_IFLNK |
440 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
441 	error = xfs_dentry_mode_to_name(&name, dentry, mode);
442 	if (unlikely(error))
443 		goto out;
444 
445 	error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
446 	if (unlikely(error))
447 		goto out;
448 
449 	inode = VFS_I(cip);
450 
451 	error = xfs_inode_init_security(inode, dir, &dentry->d_name);
452 	if (unlikely(error))
453 		goto out_cleanup_inode;
454 
455 	xfs_setup_iops(cip);
456 
457 	d_instantiate(dentry, inode);
458 	xfs_finish_inode_setup(cip);
459 	return 0;
460 
461  out_cleanup_inode:
462 	xfs_finish_inode_setup(cip);
463 	xfs_cleanup_inode(dir, inode, dentry);
464 	xfs_irele(cip);
465  out:
466 	return error;
467 }
468 
469 STATIC int
470 xfs_vn_rename(
471 	struct mnt_idmap	*idmap,
472 	struct inode		*odir,
473 	struct dentry		*odentry,
474 	struct inode		*ndir,
475 	struct dentry		*ndentry,
476 	unsigned int		flags)
477 {
478 	struct inode	*new_inode = d_inode(ndentry);
479 	int		omode = 0;
480 	int		error;
481 	struct xfs_name	oname;
482 	struct xfs_name	nname;
483 
484 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
485 		return -EINVAL;
486 
487 	/* if we are exchanging files, we need to set i_mode of both files */
488 	if (flags & RENAME_EXCHANGE)
489 		omode = d_inode(ndentry)->i_mode;
490 
491 	error = xfs_dentry_mode_to_name(&oname, odentry, omode);
492 	if (omode && unlikely(error))
493 		return error;
494 
495 	error = xfs_dentry_mode_to_name(&nname, ndentry,
496 					d_inode(odentry)->i_mode);
497 	if (unlikely(error))
498 		return error;
499 
500 	return xfs_rename(idmap, XFS_I(odir), &oname,
501 			  XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
502 			  new_inode ? XFS_I(new_inode) : NULL, flags);
503 }
504 
505 /*
506  * careful here - this function can get called recursively, so
507  * we need to be very careful about how much stack we use.
508  * uio is kmalloced for this reason...
509  */
510 STATIC const char *
511 xfs_vn_get_link(
512 	struct dentry		*dentry,
513 	struct inode		*inode,
514 	struct delayed_call	*done)
515 {
516 	char			*link;
517 	int			error = -ENOMEM;
518 
519 	if (!dentry)
520 		return ERR_PTR(-ECHILD);
521 
522 	link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
523 	if (!link)
524 		goto out_err;
525 
526 	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
527 	if (unlikely(error))
528 		goto out_kfree;
529 
530 	set_delayed_call(done, kfree_link, link);
531 	return link;
532 
533  out_kfree:
534 	kfree(link);
535  out_err:
536 	return ERR_PTR(error);
537 }
538 
539 static uint32_t
540 xfs_stat_blksize(
541 	struct xfs_inode	*ip)
542 {
543 	struct xfs_mount	*mp = ip->i_mount;
544 
545 	/*
546 	 * If the file blocks are being allocated from a realtime volume, then
547 	 * always return the realtime extent size.
548 	 */
549 	if (XFS_IS_REALTIME_INODE(ip))
550 		return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
551 
552 	/*
553 	 * Allow large block sizes to be reported to userspace programs if the
554 	 * "largeio" mount option is used.
555 	 *
556 	 * If compatibility mode is specified, simply return the basic unit of
557 	 * caching so that we don't get inefficient read/modify/write I/O from
558 	 * user apps. Otherwise....
559 	 *
560 	 * If the underlying volume is a stripe, then return the stripe width in
561 	 * bytes as the recommended I/O size. It is not a stripe and we've set a
562 	 * default buffered I/O size, return that, otherwise return the compat
563 	 * default.
564 	 */
565 	if (xfs_has_large_iosize(mp)) {
566 		if (mp->m_swidth)
567 			return XFS_FSB_TO_B(mp, mp->m_swidth);
568 		if (xfs_has_allocsize(mp))
569 			return 1U << mp->m_allocsize_log;
570 	}
571 
572 	return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize);
573 }
574 
575 static void
576 xfs_get_atomic_write_attr(
577 	struct xfs_inode	*ip,
578 	unsigned int		*unit_min,
579 	unsigned int		*unit_max)
580 {
581 	if (!xfs_inode_can_atomicwrite(ip)) {
582 		*unit_min = *unit_max = 0;
583 		return;
584 	}
585 
586 	*unit_min = *unit_max = ip->i_mount->m_sb.sb_blocksize;
587 }
588 
589 STATIC int
590 xfs_vn_getattr(
591 	struct mnt_idmap	*idmap,
592 	const struct path	*path,
593 	struct kstat		*stat,
594 	u32			request_mask,
595 	unsigned int		query_flags)
596 {
597 	struct inode		*inode = d_inode(path->dentry);
598 	struct xfs_inode	*ip = XFS_I(inode);
599 	struct xfs_mount	*mp = ip->i_mount;
600 	vfsuid_t		vfsuid = i_uid_into_vfsuid(idmap, inode);
601 	vfsgid_t		vfsgid = i_gid_into_vfsgid(idmap, inode);
602 
603 	trace_xfs_getattr(ip);
604 
605 	if (xfs_is_shutdown(mp))
606 		return -EIO;
607 
608 	stat->size = XFS_ISIZE(ip);
609 	stat->dev = inode->i_sb->s_dev;
610 	stat->mode = inode->i_mode;
611 	stat->nlink = inode->i_nlink;
612 	stat->uid = vfsuid_into_kuid(vfsuid);
613 	stat->gid = vfsgid_into_kgid(vfsgid);
614 	stat->ino = ip->i_ino;
615 	stat->atime = inode_get_atime(inode);
616 
617 	fill_mg_cmtime(stat, request_mask, inode);
618 
619 	stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
620 
621 	if (xfs_has_v3inodes(mp)) {
622 		if (request_mask & STATX_BTIME) {
623 			stat->result_mask |= STATX_BTIME;
624 			stat->btime = ip->i_crtime;
625 		}
626 	}
627 
628 	/*
629 	 * Note: If you add another clause to set an attribute flag, please
630 	 * update attributes_mask below.
631 	 */
632 	if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
633 		stat->attributes |= STATX_ATTR_IMMUTABLE;
634 	if (ip->i_diflags & XFS_DIFLAG_APPEND)
635 		stat->attributes |= STATX_ATTR_APPEND;
636 	if (ip->i_diflags & XFS_DIFLAG_NODUMP)
637 		stat->attributes |= STATX_ATTR_NODUMP;
638 
639 	stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
640 				  STATX_ATTR_APPEND |
641 				  STATX_ATTR_NODUMP);
642 
643 	switch (inode->i_mode & S_IFMT) {
644 	case S_IFBLK:
645 	case S_IFCHR:
646 		stat->blksize = BLKDEV_IOSIZE;
647 		stat->rdev = inode->i_rdev;
648 		break;
649 	case S_IFREG:
650 		if (request_mask & STATX_DIOALIGN) {
651 			struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
652 			struct block_device	*bdev = target->bt_bdev;
653 
654 			stat->result_mask |= STATX_DIOALIGN;
655 			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
656 			stat->dio_offset_align = bdev_logical_block_size(bdev);
657 		}
658 		if (request_mask & STATX_WRITE_ATOMIC) {
659 			unsigned int	unit_min, unit_max;
660 
661 			xfs_get_atomic_write_attr(ip, &unit_min,
662 					&unit_max);
663 			generic_fill_statx_atomic_writes(stat,
664 					unit_min, unit_max);
665 		}
666 		fallthrough;
667 	default:
668 		stat->blksize = xfs_stat_blksize(ip);
669 		stat->rdev = 0;
670 		break;
671 	}
672 
673 	return 0;
674 }
675 
676 static int
677 xfs_vn_change_ok(
678 	struct mnt_idmap	*idmap,
679 	struct dentry		*dentry,
680 	struct iattr		*iattr)
681 {
682 	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
683 
684 	if (xfs_is_readonly(mp))
685 		return -EROFS;
686 
687 	if (xfs_is_shutdown(mp))
688 		return -EIO;
689 
690 	return setattr_prepare(idmap, dentry, iattr);
691 }
692 
693 /*
694  * Set non-size attributes of an inode.
695  *
696  * Caution: The caller of this function is responsible for calling
697  * setattr_prepare() or otherwise verifying the change is fine.
698  */
699 static int
700 xfs_setattr_nonsize(
701 	struct mnt_idmap	*idmap,
702 	struct dentry		*dentry,
703 	struct xfs_inode	*ip,
704 	struct iattr		*iattr)
705 {
706 	xfs_mount_t		*mp = ip->i_mount;
707 	struct inode		*inode = VFS_I(ip);
708 	int			mask = iattr->ia_valid;
709 	xfs_trans_t		*tp;
710 	int			error;
711 	kuid_t			uid = GLOBAL_ROOT_UID;
712 	kgid_t			gid = GLOBAL_ROOT_GID;
713 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
714 	struct xfs_dquot	*old_udqp = NULL, *old_gdqp = NULL;
715 
716 	ASSERT((mask & ATTR_SIZE) == 0);
717 
718 	/*
719 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
720 	 * before we start any other transactions. Trying to do this later
721 	 * is messy. We don't care to take a readlock to look at the ids
722 	 * in inode here, because we can't hold it across the trans_reserve.
723 	 * If the IDs do change before we take the ilock, we're covered
724 	 * because the i_*dquot fields will get updated anyway.
725 	 */
726 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
727 		uint	qflags = 0;
728 
729 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
730 			uid = from_vfsuid(idmap, i_user_ns(inode),
731 					  iattr->ia_vfsuid);
732 			qflags |= XFS_QMOPT_UQUOTA;
733 		} else {
734 			uid = inode->i_uid;
735 		}
736 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
737 			gid = from_vfsgid(idmap, i_user_ns(inode),
738 					  iattr->ia_vfsgid);
739 			qflags |= XFS_QMOPT_GQUOTA;
740 		}  else {
741 			gid = inode->i_gid;
742 		}
743 
744 		/*
745 		 * We take a reference when we initialize udqp and gdqp,
746 		 * so it is important that we never blindly double trip on
747 		 * the same variable. See xfs_create() for an example.
748 		 */
749 		ASSERT(udqp == NULL);
750 		ASSERT(gdqp == NULL);
751 		error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
752 					   qflags, &udqp, &gdqp, NULL);
753 		if (error)
754 			return error;
755 	}
756 
757 	error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
758 			has_capability_noaudit(current, CAP_FOWNER), &tp);
759 	if (error)
760 		goto out_dqrele;
761 
762 	/*
763 	 * Register quota modifications in the transaction.  Must be the owner
764 	 * or privileged.  These IDs could have changed since we last looked at
765 	 * them.  But, we're assured that if the ownership did change while we
766 	 * didn't have the inode locked, inode's dquot(s) would have changed
767 	 * also.
768 	 */
769 	if (XFS_IS_UQUOTA_ON(mp) &&
770 	    i_uid_needs_update(idmap, iattr, inode)) {
771 		ASSERT(udqp);
772 		old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
773 	}
774 	if (XFS_IS_GQUOTA_ON(mp) &&
775 	    i_gid_needs_update(idmap, iattr, inode)) {
776 		ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
777 		ASSERT(gdqp);
778 		old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
779 	}
780 
781 	setattr_copy(idmap, inode, iattr);
782 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
783 
784 	XFS_STATS_INC(mp, xs_ig_attrchg);
785 
786 	if (xfs_has_wsync(mp))
787 		xfs_trans_set_sync(tp);
788 	error = xfs_trans_commit(tp);
789 
790 	/*
791 	 * Release any dquot(s) the inode had kept before chown.
792 	 */
793 	xfs_qm_dqrele(old_udqp);
794 	xfs_qm_dqrele(old_gdqp);
795 	xfs_qm_dqrele(udqp);
796 	xfs_qm_dqrele(gdqp);
797 
798 	if (error)
799 		return error;
800 
801 	/*
802 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
803 	 * 	     update.  We could avoid this with linked transactions
804 	 * 	     and passing down the transaction pointer all the way
805 	 *	     to attr_set.  No previous user of the generic
806 	 * 	     Posix ACL code seems to care about this issue either.
807 	 */
808 	if (mask & ATTR_MODE) {
809 		error = posix_acl_chmod(idmap, dentry, inode->i_mode);
810 		if (error)
811 			return error;
812 	}
813 
814 	return 0;
815 
816 out_dqrele:
817 	xfs_qm_dqrele(udqp);
818 	xfs_qm_dqrele(gdqp);
819 	return error;
820 }
821 
822 /*
823  * Truncate file.  Must have write permission and not be a directory.
824  *
825  * Caution: The caller of this function is responsible for calling
826  * setattr_prepare() or otherwise verifying the change is fine.
827  */
828 STATIC int
829 xfs_setattr_size(
830 	struct mnt_idmap	*idmap,
831 	struct dentry		*dentry,
832 	struct xfs_inode	*ip,
833 	struct iattr		*iattr)
834 {
835 	struct xfs_mount	*mp = ip->i_mount;
836 	struct inode		*inode = VFS_I(ip);
837 	xfs_off_t		oldsize, newsize;
838 	struct xfs_trans	*tp;
839 	int			error;
840 	uint			lock_flags = 0;
841 	uint			resblks = 0;
842 	bool			did_zeroing = false;
843 
844 	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
845 	ASSERT(S_ISREG(inode->i_mode));
846 	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
847 		ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
848 
849 	oldsize = inode->i_size;
850 	newsize = iattr->ia_size;
851 
852 	/*
853 	 * Short circuit the truncate case for zero length files.
854 	 */
855 	if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
856 		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
857 			return 0;
858 
859 		/*
860 		 * Use the regular setattr path to update the timestamps.
861 		 */
862 		iattr->ia_valid &= ~ATTR_SIZE;
863 		return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
864 	}
865 
866 	/*
867 	 * Make sure that the dquots are attached to the inode.
868 	 */
869 	error = xfs_qm_dqattach(ip);
870 	if (error)
871 		return error;
872 
873 	/*
874 	 * Wait for all direct I/O to complete.
875 	 */
876 	inode_dio_wait(inode);
877 
878 	/*
879 	 * File data changes must be complete before we start the transaction to
880 	 * modify the inode.  This needs to be done before joining the inode to
881 	 * the transaction because the inode cannot be unlocked once it is a
882 	 * part of the transaction.
883 	 *
884 	 * Start with zeroing any data beyond EOF that we may expose on file
885 	 * extension, or zeroing out the rest of the block on a downward
886 	 * truncate.
887 	 */
888 	if (newsize > oldsize) {
889 		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
890 		error = xfs_zero_range(ip, oldsize, newsize - oldsize,
891 				&did_zeroing);
892 	} else {
893 		error = xfs_truncate_page(ip, newsize, &did_zeroing);
894 	}
895 
896 	if (error)
897 		return error;
898 
899 	/*
900 	 * We've already locked out new page faults, so now we can safely remove
901 	 * pages from the page cache knowing they won't get refaulted until we
902 	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
903 	 * complete. The truncate_setsize() call also cleans partial EOF page
904 	 * PTEs on extending truncates and hence ensures sub-page block size
905 	 * filesystems are correctly handled, too.
906 	 *
907 	 * We have to do all the page cache truncate work outside the
908 	 * transaction context as the "lock" order is page lock->log space
909 	 * reservation as defined by extent allocation in the writeback path.
910 	 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
911 	 * having already truncated the in-memory version of the file (i.e. made
912 	 * user visible changes). There's not much we can do about this, except
913 	 * to hope that the caller sees ENOMEM and retries the truncate
914 	 * operation.
915 	 *
916 	 * And we update in-core i_size and truncate page cache beyond newsize
917 	 * before writeback the [i_disk_size, newsize] range, so we're
918 	 * guaranteed not to write stale data past the new EOF on truncate down.
919 	 */
920 	truncate_setsize(inode, newsize);
921 
922 	/*
923 	 * We are going to log the inode size change in this transaction so
924 	 * any previous writes that are beyond the on disk EOF and the new
925 	 * EOF that have not been written out need to be written here.  If we
926 	 * do not write the data out, we expose ourselves to the null files
927 	 * problem. Note that this includes any block zeroing we did above;
928 	 * otherwise those blocks may not be zeroed after a crash.
929 	 */
930 	if (did_zeroing ||
931 	    (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
932 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
933 						ip->i_disk_size, newsize - 1);
934 		if (error)
935 			return error;
936 	}
937 
938 	/*
939 	 * For realtime inode with more than one block rtextsize, we need the
940 	 * block reservation for bmap btree block allocations/splits that can
941 	 * happen since it could split the tail written extent and convert the
942 	 * right beyond EOF one to unwritten.
943 	 */
944 	if (xfs_inode_has_bigrtalloc(ip))
945 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
946 
947 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
948 				0, 0, &tp);
949 	if (error)
950 		return error;
951 
952 	lock_flags |= XFS_ILOCK_EXCL;
953 	xfs_ilock(ip, XFS_ILOCK_EXCL);
954 	xfs_trans_ijoin(tp, ip, 0);
955 
956 	/*
957 	 * Only change the c/mtime if we are changing the size or we are
958 	 * explicitly asked to change it.  This handles the semantic difference
959 	 * between truncate() and ftruncate() as implemented in the VFS.
960 	 *
961 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
962 	 * special case where we need to update the times despite not having
963 	 * these flags set.  For all other operations the VFS set these flags
964 	 * explicitly if it wants a timestamp update.
965 	 */
966 	if (newsize != oldsize &&
967 	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
968 		iattr->ia_ctime = iattr->ia_mtime =
969 			current_time(inode);
970 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
971 	}
972 
973 	/*
974 	 * The first thing we do is set the size to new_size permanently on
975 	 * disk.  This way we don't have to worry about anyone ever being able
976 	 * to look at the data being freed even in the face of a crash.
977 	 * What we're getting around here is the case where we free a block, it
978 	 * is allocated to another file, it is written to, and then we crash.
979 	 * If the new data gets written to the file but the log buffers
980 	 * containing the free and reallocation don't, then we'd end up with
981 	 * garbage in the blocks being freed.  As long as we make the new size
982 	 * permanent before actually freeing any blocks it doesn't matter if
983 	 * they get written to.
984 	 */
985 	ip->i_disk_size = newsize;
986 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
987 
988 	if (newsize <= oldsize) {
989 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
990 		if (error)
991 			goto out_trans_cancel;
992 
993 		/*
994 		 * Truncated "down", so we're removing references to old data
995 		 * here - if we delay flushing for a long time, we expose
996 		 * ourselves unduly to the notorious NULL files problem.  So,
997 		 * we mark this inode and flush it when the file is closed,
998 		 * and do not wait the usual (long) time for writeout.
999 		 */
1000 		xfs_iflags_set(ip, XFS_ITRUNCATED);
1001 
1002 		/* A truncate down always removes post-EOF blocks. */
1003 		xfs_inode_clear_eofblocks_tag(ip);
1004 	}
1005 
1006 	ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
1007 	setattr_copy(idmap, inode, iattr);
1008 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1009 
1010 	XFS_STATS_INC(mp, xs_ig_attrchg);
1011 
1012 	if (xfs_has_wsync(mp))
1013 		xfs_trans_set_sync(tp);
1014 
1015 	error = xfs_trans_commit(tp);
1016 out_unlock:
1017 	if (lock_flags)
1018 		xfs_iunlock(ip, lock_flags);
1019 	return error;
1020 
1021 out_trans_cancel:
1022 	xfs_trans_cancel(tp);
1023 	goto out_unlock;
1024 }
1025 
1026 int
1027 xfs_vn_setattr_size(
1028 	struct mnt_idmap	*idmap,
1029 	struct dentry		*dentry,
1030 	struct iattr		*iattr)
1031 {
1032 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1033 	int error;
1034 
1035 	trace_xfs_setattr(ip);
1036 
1037 	error = xfs_vn_change_ok(idmap, dentry, iattr);
1038 	if (error)
1039 		return error;
1040 	return xfs_setattr_size(idmap, dentry, ip, iattr);
1041 }
1042 
1043 STATIC int
1044 xfs_vn_setattr(
1045 	struct mnt_idmap	*idmap,
1046 	struct dentry		*dentry,
1047 	struct iattr		*iattr)
1048 {
1049 	struct inode		*inode = d_inode(dentry);
1050 	struct xfs_inode	*ip = XFS_I(inode);
1051 	int			error;
1052 
1053 	if (iattr->ia_valid & ATTR_SIZE) {
1054 		uint			iolock;
1055 
1056 		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1057 		iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1058 
1059 		error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1060 		if (error) {
1061 			xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1062 			return error;
1063 		}
1064 
1065 		error = xfs_vn_setattr_size(idmap, dentry, iattr);
1066 		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1067 	} else {
1068 		trace_xfs_setattr(ip);
1069 
1070 		error = xfs_vn_change_ok(idmap, dentry, iattr);
1071 		if (!error)
1072 			error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1073 	}
1074 
1075 	return error;
1076 }
1077 
1078 STATIC int
1079 xfs_vn_update_time(
1080 	struct inode		*inode,
1081 	int			flags)
1082 {
1083 	struct xfs_inode	*ip = XFS_I(inode);
1084 	struct xfs_mount	*mp = ip->i_mount;
1085 	int			log_flags = XFS_ILOG_TIMESTAMP;
1086 	struct xfs_trans	*tp;
1087 	int			error;
1088 	struct timespec64	now;
1089 
1090 	trace_xfs_update_time(ip);
1091 
1092 	if (inode->i_sb->s_flags & SB_LAZYTIME) {
1093 		if (!((flags & S_VERSION) &&
1094 		      inode_maybe_inc_iversion(inode, false))) {
1095 			generic_update_time(inode, flags);
1096 			return 0;
1097 		}
1098 
1099 		/* Capture the iversion update that just occurred */
1100 		log_flags |= XFS_ILOG_CORE;
1101 	}
1102 
1103 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1104 	if (error)
1105 		return error;
1106 
1107 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1108 	if (flags & (S_CTIME|S_MTIME))
1109 		now = inode_set_ctime_current(inode);
1110 	else
1111 		now = current_time(inode);
1112 
1113 	if (flags & S_MTIME)
1114 		inode_set_mtime_to_ts(inode, now);
1115 	if (flags & S_ATIME)
1116 		inode_set_atime_to_ts(inode, now);
1117 
1118 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1119 	xfs_trans_log_inode(tp, ip, log_flags);
1120 	return xfs_trans_commit(tp);
1121 }
1122 
1123 STATIC int
1124 xfs_vn_fiemap(
1125 	struct inode		*inode,
1126 	struct fiemap_extent_info *fieinfo,
1127 	u64			start,
1128 	u64			length)
1129 {
1130 	int			error;
1131 
1132 	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1133 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1134 		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1135 		error = iomap_fiemap(inode, fieinfo, start, length,
1136 				&xfs_xattr_iomap_ops);
1137 	} else {
1138 		error = iomap_fiemap(inode, fieinfo, start, length,
1139 				&xfs_read_iomap_ops);
1140 	}
1141 	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1142 
1143 	return error;
1144 }
1145 
1146 STATIC int
1147 xfs_vn_tmpfile(
1148 	struct mnt_idmap	*idmap,
1149 	struct inode		*dir,
1150 	struct file		*file,
1151 	umode_t			mode)
1152 {
1153 	int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1154 
1155 	return finish_open_simple(file, err);
1156 }
1157 
1158 static const struct inode_operations xfs_inode_operations = {
1159 	.get_inode_acl		= xfs_get_acl,
1160 	.set_acl		= xfs_set_acl,
1161 	.getattr		= xfs_vn_getattr,
1162 	.setattr		= xfs_vn_setattr,
1163 	.listxattr		= xfs_vn_listxattr,
1164 	.fiemap			= xfs_vn_fiemap,
1165 	.update_time		= xfs_vn_update_time,
1166 	.fileattr_get		= xfs_fileattr_get,
1167 	.fileattr_set		= xfs_fileattr_set,
1168 };
1169 
1170 static const struct inode_operations xfs_dir_inode_operations = {
1171 	.create			= xfs_vn_create,
1172 	.lookup			= xfs_vn_lookup,
1173 	.link			= xfs_vn_link,
1174 	.unlink			= xfs_vn_unlink,
1175 	.symlink		= xfs_vn_symlink,
1176 	.mkdir			= xfs_vn_mkdir,
1177 	/*
1178 	 * Yes, XFS uses the same method for rmdir and unlink.
1179 	 *
1180 	 * There are some subtile differences deeper in the code,
1181 	 * but we use S_ISDIR to check for those.
1182 	 */
1183 	.rmdir			= xfs_vn_unlink,
1184 	.mknod			= xfs_vn_mknod,
1185 	.rename			= xfs_vn_rename,
1186 	.get_inode_acl		= xfs_get_acl,
1187 	.set_acl		= xfs_set_acl,
1188 	.getattr		= xfs_vn_getattr,
1189 	.setattr		= xfs_vn_setattr,
1190 	.listxattr		= xfs_vn_listxattr,
1191 	.update_time		= xfs_vn_update_time,
1192 	.tmpfile		= xfs_vn_tmpfile,
1193 	.fileattr_get		= xfs_fileattr_get,
1194 	.fileattr_set		= xfs_fileattr_set,
1195 };
1196 
1197 static const struct inode_operations xfs_dir_ci_inode_operations = {
1198 	.create			= xfs_vn_create,
1199 	.lookup			= xfs_vn_ci_lookup,
1200 	.link			= xfs_vn_link,
1201 	.unlink			= xfs_vn_unlink,
1202 	.symlink		= xfs_vn_symlink,
1203 	.mkdir			= xfs_vn_mkdir,
1204 	/*
1205 	 * Yes, XFS uses the same method for rmdir and unlink.
1206 	 *
1207 	 * There are some subtile differences deeper in the code,
1208 	 * but we use S_ISDIR to check for those.
1209 	 */
1210 	.rmdir			= xfs_vn_unlink,
1211 	.mknod			= xfs_vn_mknod,
1212 	.rename			= xfs_vn_rename,
1213 	.get_inode_acl		= xfs_get_acl,
1214 	.set_acl		= xfs_set_acl,
1215 	.getattr		= xfs_vn_getattr,
1216 	.setattr		= xfs_vn_setattr,
1217 	.listxattr		= xfs_vn_listxattr,
1218 	.update_time		= xfs_vn_update_time,
1219 	.tmpfile		= xfs_vn_tmpfile,
1220 	.fileattr_get		= xfs_fileattr_get,
1221 	.fileattr_set		= xfs_fileattr_set,
1222 };
1223 
1224 static const struct inode_operations xfs_symlink_inode_operations = {
1225 	.get_link		= xfs_vn_get_link,
1226 	.getattr		= xfs_vn_getattr,
1227 	.setattr		= xfs_vn_setattr,
1228 	.listxattr		= xfs_vn_listxattr,
1229 	.update_time		= xfs_vn_update_time,
1230 };
1231 
1232 /* Figure out if this file actually supports DAX. */
1233 static bool
1234 xfs_inode_supports_dax(
1235 	struct xfs_inode	*ip)
1236 {
1237 	struct xfs_mount	*mp = ip->i_mount;
1238 
1239 	/* Only supported on regular files. */
1240 	if (!S_ISREG(VFS_I(ip)->i_mode))
1241 		return false;
1242 
1243 	/* Block size must match page size */
1244 	if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1245 		return false;
1246 
1247 	/* Device has to support DAX too. */
1248 	return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1249 }
1250 
1251 static bool
1252 xfs_inode_should_enable_dax(
1253 	struct xfs_inode *ip)
1254 {
1255 	if (!IS_ENABLED(CONFIG_FS_DAX))
1256 		return false;
1257 	if (xfs_has_dax_never(ip->i_mount))
1258 		return false;
1259 	if (!xfs_inode_supports_dax(ip))
1260 		return false;
1261 	if (xfs_has_dax_always(ip->i_mount))
1262 		return true;
1263 	if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1264 		return true;
1265 	return false;
1266 }
1267 
1268 void
1269 xfs_diflags_to_iflags(
1270 	struct xfs_inode	*ip,
1271 	bool init)
1272 {
1273 	struct inode            *inode = VFS_I(ip);
1274 	unsigned int            xflags = xfs_ip2xflags(ip);
1275 	unsigned int            flags = 0;
1276 
1277 	ASSERT(!(IS_DAX(inode) && init));
1278 
1279 	if (xflags & FS_XFLAG_IMMUTABLE)
1280 		flags |= S_IMMUTABLE;
1281 	if (xflags & FS_XFLAG_APPEND)
1282 		flags |= S_APPEND;
1283 	if (xflags & FS_XFLAG_SYNC)
1284 		flags |= S_SYNC;
1285 	if (xflags & FS_XFLAG_NOATIME)
1286 		flags |= S_NOATIME;
1287 	if (init && xfs_inode_should_enable_dax(ip))
1288 		flags |= S_DAX;
1289 
1290 	/*
1291 	 * S_DAX can only be set during inode initialization and is never set by
1292 	 * the VFS, so we cannot mask off S_DAX in i_flags.
1293 	 */
1294 	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1295 	inode->i_flags |= flags;
1296 }
1297 
1298 /*
1299  * Initialize the Linux inode.
1300  *
1301  * When reading existing inodes from disk this is called directly from xfs_iget,
1302  * when creating a new inode it is called from xfs_init_new_inode after setting
1303  * up the inode. These callers have different criteria for clearing XFS_INEW, so
1304  * leave it up to the caller to deal with unlocking the inode appropriately.
1305  */
1306 void
1307 xfs_setup_inode(
1308 	struct xfs_inode	*ip)
1309 {
1310 	struct inode		*inode = &ip->i_vnode;
1311 	gfp_t			gfp_mask;
1312 	bool			is_meta = xfs_is_internal_inode(ip);
1313 
1314 	inode->i_ino = ip->i_ino;
1315 	inode->i_state |= I_NEW;
1316 
1317 	inode_sb_list_add(inode);
1318 	/* make the inode look hashed for the writeback code */
1319 	inode_fake_hash(inode);
1320 
1321 	i_size_write(inode, ip->i_disk_size);
1322 	xfs_diflags_to_iflags(ip, true);
1323 
1324 	/*
1325 	 * Mark our metadata files as private so that LSMs and the ACL code
1326 	 * don't try to add their own metadata or reason about these files,
1327 	 * and users cannot ever obtain file handles to them.
1328 	 */
1329 	if (is_meta) {
1330 		inode->i_flags |= S_PRIVATE;
1331 		inode->i_opflags &= ~IOP_XATTR;
1332 	}
1333 
1334 	if (S_ISDIR(inode->i_mode)) {
1335 		/*
1336 		 * We set the i_rwsem class here to avoid potential races with
1337 		 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1338 		 * after a filehandle lookup has already found the inode in
1339 		 * cache before it has been unlocked via unlock_new_inode().
1340 		 */
1341 		lockdep_set_class(&inode->i_rwsem,
1342 				  &inode->i_sb->s_type->i_mutex_dir_key);
1343 		lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1344 	} else {
1345 		lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1346 	}
1347 
1348 	/*
1349 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1350 	 * prevent direct reclaim recursion back into the filesystem and blowing
1351 	 * stacks or deadlocking.
1352 	 */
1353 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1354 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1355 
1356 	/*
1357 	 * For real-time inodes update the stable write flags to that of the RT
1358 	 * device instead of the data device.
1359 	 */
1360 	if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1361 		xfs_update_stable_writes(ip);
1362 
1363 	/*
1364 	 * If there is no attribute fork no ACL can exist on this inode,
1365 	 * and it can't have any file capabilities attached to it either.
1366 	 */
1367 	if (!xfs_inode_has_attr_fork(ip)) {
1368 		inode_has_no_xattr(inode);
1369 		cache_no_acl(inode);
1370 	}
1371 }
1372 
1373 void
1374 xfs_setup_iops(
1375 	struct xfs_inode	*ip)
1376 {
1377 	struct inode		*inode = &ip->i_vnode;
1378 
1379 	switch (inode->i_mode & S_IFMT) {
1380 	case S_IFREG:
1381 		inode->i_op = &xfs_inode_operations;
1382 		inode->i_fop = &xfs_file_operations;
1383 		if (IS_DAX(inode))
1384 			inode->i_mapping->a_ops = &xfs_dax_aops;
1385 		else
1386 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1387 		break;
1388 	case S_IFDIR:
1389 		if (xfs_has_asciici(XFS_M(inode->i_sb)))
1390 			inode->i_op = &xfs_dir_ci_inode_operations;
1391 		else
1392 			inode->i_op = &xfs_dir_inode_operations;
1393 		inode->i_fop = &xfs_dir_file_operations;
1394 		break;
1395 	case S_IFLNK:
1396 		inode->i_op = &xfs_symlink_inode_operations;
1397 		break;
1398 	default:
1399 		inode->i_op = &xfs_inode_operations;
1400 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1401 		break;
1402 	}
1403 }
1404