xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision b58c6630)
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_rtalloc.h"
15 #include "xfs_iwalk.h"
16 #include "xfs_itable.h"
17 #include "xfs_error.h"
18 #include "xfs_attr.h"
19 #include "xfs_bmap.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_fsops.h"
22 #include "xfs_discard.h"
23 #include "xfs_quota.h"
24 #include "xfs_export.h"
25 #include "xfs_trace.h"
26 #include "xfs_icache.h"
27 #include "xfs_trans.h"
28 #include "xfs_acl.h"
29 #include "xfs_btree.h"
30 #include <linux/fsmap.h>
31 #include "xfs_fsmap.h"
32 #include "scrub/xfs_scrub.h"
33 #include "xfs_sb.h"
34 #include "xfs_ag.h"
35 #include "xfs_health.h"
36 #include "xfs_reflink.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_da_format.h"
39 #include "xfs_da_btree.h"
40 
41 #include <linux/mount.h>
42 #include <linux/namei.h>
43 
44 /*
45  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
46  * a file or fs handle.
47  *
48  * XFS_IOC_PATH_TO_FSHANDLE
49  *    returns fs handle for a mount point or path within that mount point
50  * XFS_IOC_FD_TO_HANDLE
51  *    returns full handle for a FD opened in user space
52  * XFS_IOC_PATH_TO_HANDLE
53  *    returns full handle for a path
54  */
55 int
56 xfs_find_handle(
57 	unsigned int		cmd,
58 	xfs_fsop_handlereq_t	*hreq)
59 {
60 	int			hsize;
61 	xfs_handle_t		handle;
62 	struct inode		*inode;
63 	struct fd		f = {NULL};
64 	struct path		path;
65 	int			error;
66 	struct xfs_inode	*ip;
67 
68 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
69 		f = fdget(hreq->fd);
70 		if (!f.file)
71 			return -EBADF;
72 		inode = file_inode(f.file);
73 	} else {
74 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
75 		if (error)
76 			return error;
77 		inode = d_inode(path.dentry);
78 	}
79 	ip = XFS_I(inode);
80 
81 	/*
82 	 * We can only generate handles for inodes residing on a XFS filesystem,
83 	 * and only for regular files, directories or symbolic links.
84 	 */
85 	error = -EINVAL;
86 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
87 		goto out_put;
88 
89 	error = -EBADF;
90 	if (!S_ISREG(inode->i_mode) &&
91 	    !S_ISDIR(inode->i_mode) &&
92 	    !S_ISLNK(inode->i_mode))
93 		goto out_put;
94 
95 
96 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
97 
98 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
99 		/*
100 		 * This handle only contains an fsid, zero the rest.
101 		 */
102 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
103 		hsize = sizeof(xfs_fsid_t);
104 	} else {
105 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
106 					sizeof(handle.ha_fid.fid_len);
107 		handle.ha_fid.fid_pad = 0;
108 		handle.ha_fid.fid_gen = inode->i_generation;
109 		handle.ha_fid.fid_ino = ip->i_ino;
110 		hsize = sizeof(xfs_handle_t);
111 	}
112 
113 	error = -EFAULT;
114 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
115 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
116 		goto out_put;
117 
118 	error = 0;
119 
120  out_put:
121 	if (cmd == XFS_IOC_FD_TO_HANDLE)
122 		fdput(f);
123 	else
124 		path_put(&path);
125 	return error;
126 }
127 
128 /*
129  * No need to do permission checks on the various pathname components
130  * as the handle operations are privileged.
131  */
132 STATIC int
133 xfs_handle_acceptable(
134 	void			*context,
135 	struct dentry		*dentry)
136 {
137 	return 1;
138 }
139 
140 /*
141  * Convert userspace handle data into a dentry.
142  */
143 struct dentry *
144 xfs_handle_to_dentry(
145 	struct file		*parfilp,
146 	void __user		*uhandle,
147 	u32			hlen)
148 {
149 	xfs_handle_t		handle;
150 	struct xfs_fid64	fid;
151 
152 	/*
153 	 * Only allow handle opens under a directory.
154 	 */
155 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
156 		return ERR_PTR(-ENOTDIR);
157 
158 	if (hlen != sizeof(xfs_handle_t))
159 		return ERR_PTR(-EINVAL);
160 	if (copy_from_user(&handle, uhandle, hlen))
161 		return ERR_PTR(-EFAULT);
162 	if (handle.ha_fid.fid_len !=
163 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
164 		return ERR_PTR(-EINVAL);
165 
166 	memset(&fid, 0, sizeof(struct fid));
167 	fid.ino = handle.ha_fid.fid_ino;
168 	fid.gen = handle.ha_fid.fid_gen;
169 
170 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
171 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
172 			xfs_handle_acceptable, NULL);
173 }
174 
175 STATIC struct dentry *
176 xfs_handlereq_to_dentry(
177 	struct file		*parfilp,
178 	xfs_fsop_handlereq_t	*hreq)
179 {
180 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
181 }
182 
183 int
184 xfs_open_by_handle(
185 	struct file		*parfilp,
186 	xfs_fsop_handlereq_t	*hreq)
187 {
188 	const struct cred	*cred = current_cred();
189 	int			error;
190 	int			fd;
191 	int			permflag;
192 	struct file		*filp;
193 	struct inode		*inode;
194 	struct dentry		*dentry;
195 	fmode_t			fmode;
196 	struct path		path;
197 
198 	if (!capable(CAP_SYS_ADMIN))
199 		return -EPERM;
200 
201 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
202 	if (IS_ERR(dentry))
203 		return PTR_ERR(dentry);
204 	inode = d_inode(dentry);
205 
206 	/* Restrict xfs_open_by_handle to directories & regular files. */
207 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
208 		error = -EPERM;
209 		goto out_dput;
210 	}
211 
212 #if BITS_PER_LONG != 32
213 	hreq->oflags |= O_LARGEFILE;
214 #endif
215 
216 	permflag = hreq->oflags;
217 	fmode = OPEN_FMODE(permflag);
218 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
219 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
220 		error = -EPERM;
221 		goto out_dput;
222 	}
223 
224 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
225 		error = -EPERM;
226 		goto out_dput;
227 	}
228 
229 	/* Can't write directories. */
230 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
231 		error = -EISDIR;
232 		goto out_dput;
233 	}
234 
235 	fd = get_unused_fd_flags(0);
236 	if (fd < 0) {
237 		error = fd;
238 		goto out_dput;
239 	}
240 
241 	path.mnt = parfilp->f_path.mnt;
242 	path.dentry = dentry;
243 	filp = dentry_open(&path, hreq->oflags, cred);
244 	dput(dentry);
245 	if (IS_ERR(filp)) {
246 		put_unused_fd(fd);
247 		return PTR_ERR(filp);
248 	}
249 
250 	if (S_ISREG(inode->i_mode)) {
251 		filp->f_flags |= O_NOATIME;
252 		filp->f_mode |= FMODE_NOCMTIME;
253 	}
254 
255 	fd_install(fd, filp);
256 	return fd;
257 
258  out_dput:
259 	dput(dentry);
260 	return error;
261 }
262 
263 int
264 xfs_readlink_by_handle(
265 	struct file		*parfilp,
266 	xfs_fsop_handlereq_t	*hreq)
267 {
268 	struct dentry		*dentry;
269 	__u32			olen;
270 	int			error;
271 
272 	if (!capable(CAP_SYS_ADMIN))
273 		return -EPERM;
274 
275 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
276 	if (IS_ERR(dentry))
277 		return PTR_ERR(dentry);
278 
279 	/* Restrict this handle operation to symlinks only. */
280 	if (!d_is_symlink(dentry)) {
281 		error = -EINVAL;
282 		goto out_dput;
283 	}
284 
285 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
286 		error = -EFAULT;
287 		goto out_dput;
288 	}
289 
290 	error = vfs_readlink(dentry, hreq->ohandle, olen);
291 
292  out_dput:
293 	dput(dentry);
294 	return error;
295 }
296 
297 /*
298  * Format an attribute and copy it out to the user's buffer.
299  * Take care to check values and protect against them changing later,
300  * we may be reading them directly out of a user buffer.
301  */
302 static void
303 xfs_ioc_attr_put_listent(
304 	struct xfs_attr_list_context *context,
305 	int			flags,
306 	unsigned char		*name,
307 	int			namelen,
308 	int			valuelen)
309 {
310 	struct xfs_attrlist	*alist = context->buffer;
311 	struct xfs_attrlist_ent	*aep;
312 	int			arraytop;
313 
314 	ASSERT(!context->seen_enough);
315 	ASSERT(context->count >= 0);
316 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
317 	ASSERT(context->firstu >= sizeof(*alist));
318 	ASSERT(context->firstu <= context->bufsize);
319 
320 	/*
321 	 * Only list entries in the right namespace.
322 	 */
323 	if (context->attr_filter != (flags & XFS_ATTR_NSP_ONDISK_MASK))
324 		return;
325 
326 	arraytop = sizeof(*alist) +
327 			context->count * sizeof(alist->al_offset[0]);
328 
329 	/* decrement by the actual bytes used by the attr */
330 	context->firstu -= round_up(offsetof(struct xfs_attrlist_ent, a_name) +
331 			namelen + 1, sizeof(uint32_t));
332 	if (context->firstu < arraytop) {
333 		trace_xfs_attr_list_full(context);
334 		alist->al_more = 1;
335 		context->seen_enough = 1;
336 		return;
337 	}
338 
339 	aep = context->buffer + context->firstu;
340 	aep->a_valuelen = valuelen;
341 	memcpy(aep->a_name, name, namelen);
342 	aep->a_name[namelen] = 0;
343 	alist->al_offset[context->count++] = context->firstu;
344 	alist->al_count = context->count;
345 	trace_xfs_attr_list_add(context);
346 }
347 
348 static unsigned int
349 xfs_attr_filter(
350 	u32			ioc_flags)
351 {
352 	if (ioc_flags & XFS_IOC_ATTR_ROOT)
353 		return XFS_ATTR_ROOT;
354 	if (ioc_flags & XFS_IOC_ATTR_SECURE)
355 		return XFS_ATTR_SECURE;
356 	return 0;
357 }
358 
359 static unsigned int
360 xfs_attr_flags(
361 	u32			ioc_flags)
362 {
363 	if (ioc_flags & XFS_IOC_ATTR_CREATE)
364 		return XATTR_CREATE;
365 	if (ioc_flags & XFS_IOC_ATTR_REPLACE)
366 		return XATTR_REPLACE;
367 	return 0;
368 }
369 
370 int
371 xfs_ioc_attr_list(
372 	struct xfs_inode		*dp,
373 	void __user			*ubuf,
374 	int				bufsize,
375 	int				flags,
376 	struct xfs_attrlist_cursor __user *ucursor)
377 {
378 	struct xfs_attr_list_context	context = { };
379 	struct xfs_attrlist		*alist;
380 	void				*buffer;
381 	int				error;
382 
383 	if (bufsize < sizeof(struct xfs_attrlist) ||
384 	    bufsize > XFS_XATTR_LIST_MAX)
385 		return -EINVAL;
386 
387 	/*
388 	 * Reject flags, only allow namespaces.
389 	 */
390 	if (flags & ~(XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
391 		return -EINVAL;
392 	if (flags == (XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
393 		return -EINVAL;
394 
395 	/*
396 	 * Validate the cursor.
397 	 */
398 	if (copy_from_user(&context.cursor, ucursor, sizeof(context.cursor)))
399 		return -EFAULT;
400 	if (context.cursor.pad1 || context.cursor.pad2)
401 		return -EINVAL;
402 	if (!context.cursor.initted &&
403 	    (context.cursor.hashval || context.cursor.blkno ||
404 	     context.cursor.offset))
405 		return -EINVAL;
406 
407 	buffer = kmem_zalloc_large(bufsize, 0);
408 	if (!buffer)
409 		return -ENOMEM;
410 
411 	/*
412 	 * Initialize the output buffer.
413 	 */
414 	context.dp = dp;
415 	context.resynch = 1;
416 	context.attr_filter = xfs_attr_filter(flags);
417 	context.buffer = buffer;
418 	context.bufsize = round_down(bufsize, sizeof(uint32_t));
419 	context.firstu = context.bufsize;
420 	context.put_listent = xfs_ioc_attr_put_listent;
421 
422 	alist = context.buffer;
423 	alist->al_count = 0;
424 	alist->al_more = 0;
425 	alist->al_offset[0] = context.bufsize;
426 
427 	error = xfs_attr_list(&context);
428 	if (error)
429 		goto out_free;
430 
431 	if (copy_to_user(ubuf, buffer, bufsize) ||
432 	    copy_to_user(ucursor, &context.cursor, sizeof(context.cursor)))
433 		error = -EFAULT;
434 out_free:
435 	kmem_free(buffer);
436 	return error;
437 }
438 
439 STATIC int
440 xfs_attrlist_by_handle(
441 	struct file		*parfilp,
442 	struct xfs_fsop_attrlist_handlereq __user *p)
443 {
444 	struct xfs_fsop_attrlist_handlereq al_hreq;
445 	struct dentry		*dentry;
446 	int			error = -ENOMEM;
447 
448 	if (!capable(CAP_SYS_ADMIN))
449 		return -EPERM;
450 	if (copy_from_user(&al_hreq, p, sizeof(al_hreq)))
451 		return -EFAULT;
452 
453 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
454 	if (IS_ERR(dentry))
455 		return PTR_ERR(dentry);
456 
457 	error = xfs_ioc_attr_list(XFS_I(d_inode(dentry)), al_hreq.buffer,
458 				  al_hreq.buflen, al_hreq.flags, &p->pos);
459 	dput(dentry);
460 	return error;
461 }
462 
463 static int
464 xfs_attrmulti_attr_get(
465 	struct inode		*inode,
466 	unsigned char		*name,
467 	unsigned char		__user *ubuf,
468 	uint32_t		*len,
469 	uint32_t		flags)
470 {
471 	struct xfs_da_args	args = {
472 		.dp		= XFS_I(inode),
473 		.attr_filter	= xfs_attr_filter(flags),
474 		.attr_flags	= xfs_attr_flags(flags),
475 		.name		= name,
476 		.namelen	= strlen(name),
477 		.valuelen	= *len,
478 	};
479 	int			error;
480 
481 	if (*len > XFS_XATTR_SIZE_MAX)
482 		return -EINVAL;
483 
484 	error = xfs_attr_get(&args);
485 	if (error)
486 		goto out_kfree;
487 
488 	*len = args.valuelen;
489 	if (copy_to_user(ubuf, args.value, args.valuelen))
490 		error = -EFAULT;
491 
492 out_kfree:
493 	kmem_free(args.value);
494 	return error;
495 }
496 
497 static int
498 xfs_attrmulti_attr_set(
499 	struct inode		*inode,
500 	unsigned char		*name,
501 	const unsigned char	__user *ubuf,
502 	uint32_t		len,
503 	uint32_t		flags)
504 {
505 	struct xfs_da_args	args = {
506 		.dp		= XFS_I(inode),
507 		.attr_filter	= xfs_attr_filter(flags),
508 		.attr_flags	= xfs_attr_flags(flags),
509 		.name		= name,
510 		.namelen	= strlen(name),
511 	};
512 	int			error;
513 
514 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
515 		return -EPERM;
516 
517 	if (ubuf) {
518 		if (len > XFS_XATTR_SIZE_MAX)
519 			return -EINVAL;
520 		args.value = memdup_user(ubuf, len);
521 		if (IS_ERR(args.value))
522 			return PTR_ERR(args.value);
523 		args.valuelen = len;
524 	}
525 
526 	error = xfs_attr_set(&args);
527 	if (!error && (flags & XFS_IOC_ATTR_ROOT))
528 		xfs_forget_acl(inode, name);
529 	kfree(args.value);
530 	return error;
531 }
532 
533 int
534 xfs_ioc_attrmulti_one(
535 	struct file		*parfilp,
536 	struct inode		*inode,
537 	uint32_t		opcode,
538 	void __user		*uname,
539 	void __user		*value,
540 	uint32_t		*len,
541 	uint32_t		flags)
542 {
543 	unsigned char		*name;
544 	int			error;
545 
546 	if ((flags & XFS_IOC_ATTR_ROOT) && (flags & XFS_IOC_ATTR_SECURE))
547 		return -EINVAL;
548 
549 	name = strndup_user(uname, MAXNAMELEN);
550 	if (IS_ERR(name))
551 		return PTR_ERR(name);
552 
553 	switch (opcode) {
554 	case ATTR_OP_GET:
555 		error = xfs_attrmulti_attr_get(inode, name, value, len, flags);
556 		break;
557 	case ATTR_OP_REMOVE:
558 		value = NULL;
559 		*len = 0;
560 		/* fall through */
561 	case ATTR_OP_SET:
562 		error = mnt_want_write_file(parfilp);
563 		if (error)
564 			break;
565 		error = xfs_attrmulti_attr_set(inode, name, value, *len, flags);
566 		mnt_drop_write_file(parfilp);
567 		break;
568 	default:
569 		error = -EINVAL;
570 		break;
571 	}
572 
573 	kfree(name);
574 	return error;
575 }
576 
577 STATIC int
578 xfs_attrmulti_by_handle(
579 	struct file		*parfilp,
580 	void			__user *arg)
581 {
582 	int			error;
583 	xfs_attr_multiop_t	*ops;
584 	xfs_fsop_attrmulti_handlereq_t am_hreq;
585 	struct dentry		*dentry;
586 	unsigned int		i, size;
587 
588 	if (!capable(CAP_SYS_ADMIN))
589 		return -EPERM;
590 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
591 		return -EFAULT;
592 
593 	/* overflow check */
594 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
595 		return -E2BIG;
596 
597 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
598 	if (IS_ERR(dentry))
599 		return PTR_ERR(dentry);
600 
601 	error = -E2BIG;
602 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
603 	if (!size || size > 16 * PAGE_SIZE)
604 		goto out_dput;
605 
606 	ops = memdup_user(am_hreq.ops, size);
607 	if (IS_ERR(ops)) {
608 		error = PTR_ERR(ops);
609 		goto out_dput;
610 	}
611 
612 	error = 0;
613 	for (i = 0; i < am_hreq.opcount; i++) {
614 		ops[i].am_error = xfs_ioc_attrmulti_one(parfilp,
615 				d_inode(dentry), ops[i].am_opcode,
616 				ops[i].am_attrname, ops[i].am_attrvalue,
617 				&ops[i].am_length, ops[i].am_flags);
618 	}
619 
620 	if (copy_to_user(am_hreq.ops, ops, size))
621 		error = -EFAULT;
622 
623 	kfree(ops);
624  out_dput:
625 	dput(dentry);
626 	return error;
627 }
628 
629 int
630 xfs_ioc_space(
631 	struct file		*filp,
632 	xfs_flock64_t		*bf)
633 {
634 	struct inode		*inode = file_inode(filp);
635 	struct xfs_inode	*ip = XFS_I(inode);
636 	struct iattr		iattr;
637 	enum xfs_prealloc_flags	flags = XFS_PREALLOC_CLEAR;
638 	uint			iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
639 	int			error;
640 
641 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND))
642 		return -EPERM;
643 
644 	if (!(filp->f_mode & FMODE_WRITE))
645 		return -EBADF;
646 
647 	if (!S_ISREG(inode->i_mode))
648 		return -EINVAL;
649 
650 	if (xfs_is_always_cow_inode(ip))
651 		return -EOPNOTSUPP;
652 
653 	if (filp->f_flags & O_DSYNC)
654 		flags |= XFS_PREALLOC_SYNC;
655 	if (filp->f_mode & FMODE_NOCMTIME)
656 		flags |= XFS_PREALLOC_INVISIBLE;
657 
658 	error = mnt_want_write_file(filp);
659 	if (error)
660 		return error;
661 
662 	xfs_ilock(ip, iolock);
663 	error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
664 	if (error)
665 		goto out_unlock;
666 	inode_dio_wait(inode);
667 
668 	switch (bf->l_whence) {
669 	case 0: /*SEEK_SET*/
670 		break;
671 	case 1: /*SEEK_CUR*/
672 		bf->l_start += filp->f_pos;
673 		break;
674 	case 2: /*SEEK_END*/
675 		bf->l_start += XFS_ISIZE(ip);
676 		break;
677 	default:
678 		error = -EINVAL;
679 		goto out_unlock;
680 	}
681 
682 	if (bf->l_start < 0 || bf->l_start > inode->i_sb->s_maxbytes) {
683 		error = -EINVAL;
684 		goto out_unlock;
685 	}
686 
687 	if (bf->l_start > XFS_ISIZE(ip)) {
688 		error = xfs_alloc_file_space(ip, XFS_ISIZE(ip),
689 				bf->l_start - XFS_ISIZE(ip), 0);
690 		if (error)
691 			goto out_unlock;
692 	}
693 
694 	iattr.ia_valid = ATTR_SIZE;
695 	iattr.ia_size = bf->l_start;
696 	error = xfs_vn_setattr_size(file_dentry(filp), &iattr);
697 	if (error)
698 		goto out_unlock;
699 
700 	error = xfs_update_prealloc_flags(ip, flags);
701 
702 out_unlock:
703 	xfs_iunlock(ip, iolock);
704 	mnt_drop_write_file(filp);
705 	return error;
706 }
707 
708 /* Return 0 on success or positive error */
709 int
710 xfs_fsbulkstat_one_fmt(
711 	struct xfs_ibulk		*breq,
712 	const struct xfs_bulkstat	*bstat)
713 {
714 	struct xfs_bstat		bs1;
715 
716 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
717 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
718 		return -EFAULT;
719 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
720 }
721 
722 int
723 xfs_fsinumbers_fmt(
724 	struct xfs_ibulk		*breq,
725 	const struct xfs_inumbers	*igrp)
726 {
727 	struct xfs_inogrp		ig1;
728 
729 	xfs_inumbers_to_inogrp(&ig1, igrp);
730 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
731 		return -EFAULT;
732 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
733 }
734 
735 STATIC int
736 xfs_ioc_fsbulkstat(
737 	xfs_mount_t		*mp,
738 	unsigned int		cmd,
739 	void			__user *arg)
740 {
741 	struct xfs_fsop_bulkreq	bulkreq;
742 	struct xfs_ibulk	breq = {
743 		.mp		= mp,
744 		.ocount		= 0,
745 	};
746 	xfs_ino_t		lastino;
747 	int			error;
748 
749 	/* done = 1 if there are more stats to get and if bulkstat */
750 	/* should be called again (unused here, but used in dmapi) */
751 
752 	if (!capable(CAP_SYS_ADMIN))
753 		return -EPERM;
754 
755 	if (XFS_FORCED_SHUTDOWN(mp))
756 		return -EIO;
757 
758 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
759 		return -EFAULT;
760 
761 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
762 		return -EFAULT;
763 
764 	if (bulkreq.icount <= 0)
765 		return -EINVAL;
766 
767 	if (bulkreq.ubuffer == NULL)
768 		return -EINVAL;
769 
770 	breq.ubuffer = bulkreq.ubuffer;
771 	breq.icount = bulkreq.icount;
772 
773 	/*
774 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
775 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
776 	 * that *lastip contains either zero or the number of the last inode to
777 	 * be examined by the previous call and return results starting with
778 	 * the next inode after that.  The new bulk request back end functions
779 	 * take the inode to start with, so we have to compute the startino
780 	 * parameter from lastino to maintain correct function.  lastino == 0
781 	 * is a special case because it has traditionally meant "first inode
782 	 * in filesystem".
783 	 */
784 	if (cmd == XFS_IOC_FSINUMBERS) {
785 		breq.startino = lastino ? lastino + 1 : 0;
786 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
787 		lastino = breq.startino - 1;
788 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
789 		breq.startino = lastino;
790 		breq.icount = 1;
791 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
792 	} else {	/* XFS_IOC_FSBULKSTAT */
793 		breq.startino = lastino ? lastino + 1 : 0;
794 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
795 		lastino = breq.startino - 1;
796 	}
797 
798 	if (error)
799 		return error;
800 
801 	if (bulkreq.lastip != NULL &&
802 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
803 		return -EFAULT;
804 
805 	if (bulkreq.ocount != NULL &&
806 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
807 		return -EFAULT;
808 
809 	return 0;
810 }
811 
812 /* Return 0 on success or positive error */
813 static int
814 xfs_bulkstat_fmt(
815 	struct xfs_ibulk		*breq,
816 	const struct xfs_bulkstat	*bstat)
817 {
818 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
819 		return -EFAULT;
820 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
821 }
822 
823 /*
824  * Check the incoming bulk request @hdr from userspace and initialize the
825  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
826  * should proceed; -ECANCELED if there's nothing to do; or the usual
827  * negative error code.
828  */
829 static int
830 xfs_bulk_ireq_setup(
831 	struct xfs_mount	*mp,
832 	struct xfs_bulk_ireq	*hdr,
833 	struct xfs_ibulk	*breq,
834 	void __user		*ubuffer)
835 {
836 	if (hdr->icount == 0 ||
837 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
838 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
839 		return -EINVAL;
840 
841 	breq->startino = hdr->ino;
842 	breq->ubuffer = ubuffer;
843 	breq->icount = hdr->icount;
844 	breq->ocount = 0;
845 	breq->flags = 0;
846 
847 	/*
848 	 * The @ino parameter is a special value, so we must look it up here.
849 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
850 	 * worth of data.
851 	 */
852 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
853 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
854 			return -EINVAL;
855 
856 		switch (hdr->ino) {
857 		case XFS_BULK_IREQ_SPECIAL_ROOT:
858 			hdr->ino = mp->m_sb.sb_rootino;
859 			break;
860 		default:
861 			return -EINVAL;
862 		}
863 		breq->icount = 1;
864 	}
865 
866 	/*
867 	 * The IREQ_AGNO flag means that we only want results from a given AG.
868 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
869 	 * beyond the specified AG then we return no results.
870 	 */
871 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
872 		if (hdr->agno >= mp->m_sb.sb_agcount)
873 			return -EINVAL;
874 
875 		if (breq->startino == 0)
876 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
877 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
878 			return -EINVAL;
879 
880 		breq->flags |= XFS_IBULK_SAME_AG;
881 
882 		/* Asking for an inode past the end of the AG?  We're done! */
883 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
884 			return -ECANCELED;
885 	} else if (hdr->agno)
886 		return -EINVAL;
887 
888 	/* Asking for an inode past the end of the FS?  We're done! */
889 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
890 		return -ECANCELED;
891 
892 	return 0;
893 }
894 
895 /*
896  * Update the userspace bulk request @hdr to reflect the end state of the
897  * internal bulk request @breq.
898  */
899 static void
900 xfs_bulk_ireq_teardown(
901 	struct xfs_bulk_ireq	*hdr,
902 	struct xfs_ibulk	*breq)
903 {
904 	hdr->ino = breq->startino;
905 	hdr->ocount = breq->ocount;
906 }
907 
908 /* Handle the v5 bulkstat ioctl. */
909 STATIC int
910 xfs_ioc_bulkstat(
911 	struct xfs_mount		*mp,
912 	unsigned int			cmd,
913 	struct xfs_bulkstat_req __user	*arg)
914 {
915 	struct xfs_bulk_ireq		hdr;
916 	struct xfs_ibulk		breq = {
917 		.mp			= mp,
918 	};
919 	int				error;
920 
921 	if (!capable(CAP_SYS_ADMIN))
922 		return -EPERM;
923 
924 	if (XFS_FORCED_SHUTDOWN(mp))
925 		return -EIO;
926 
927 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
928 		return -EFAULT;
929 
930 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
931 	if (error == -ECANCELED)
932 		goto out_teardown;
933 	if (error < 0)
934 		return error;
935 
936 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
937 	if (error)
938 		return error;
939 
940 out_teardown:
941 	xfs_bulk_ireq_teardown(&hdr, &breq);
942 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
943 		return -EFAULT;
944 
945 	return 0;
946 }
947 
948 STATIC int
949 xfs_inumbers_fmt(
950 	struct xfs_ibulk		*breq,
951 	const struct xfs_inumbers	*igrp)
952 {
953 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
954 		return -EFAULT;
955 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
956 }
957 
958 /* Handle the v5 inumbers ioctl. */
959 STATIC int
960 xfs_ioc_inumbers(
961 	struct xfs_mount		*mp,
962 	unsigned int			cmd,
963 	struct xfs_inumbers_req __user	*arg)
964 {
965 	struct xfs_bulk_ireq		hdr;
966 	struct xfs_ibulk		breq = {
967 		.mp			= mp,
968 	};
969 	int				error;
970 
971 	if (!capable(CAP_SYS_ADMIN))
972 		return -EPERM;
973 
974 	if (XFS_FORCED_SHUTDOWN(mp))
975 		return -EIO;
976 
977 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
978 		return -EFAULT;
979 
980 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
981 	if (error == -ECANCELED)
982 		goto out_teardown;
983 	if (error < 0)
984 		return error;
985 
986 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
987 	if (error)
988 		return error;
989 
990 out_teardown:
991 	xfs_bulk_ireq_teardown(&hdr, &breq);
992 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
993 		return -EFAULT;
994 
995 	return 0;
996 }
997 
998 STATIC int
999 xfs_ioc_fsgeometry(
1000 	struct xfs_mount	*mp,
1001 	void			__user *arg,
1002 	int			struct_version)
1003 {
1004 	struct xfs_fsop_geom	fsgeo;
1005 	size_t			len;
1006 
1007 	xfs_fs_geometry(&mp->m_sb, &fsgeo, struct_version);
1008 
1009 	if (struct_version <= 3)
1010 		len = sizeof(struct xfs_fsop_geom_v1);
1011 	else if (struct_version == 4)
1012 		len = sizeof(struct xfs_fsop_geom_v4);
1013 	else {
1014 		xfs_fsop_geom_health(mp, &fsgeo);
1015 		len = sizeof(fsgeo);
1016 	}
1017 
1018 	if (copy_to_user(arg, &fsgeo, len))
1019 		return -EFAULT;
1020 	return 0;
1021 }
1022 
1023 STATIC int
1024 xfs_ioc_ag_geometry(
1025 	struct xfs_mount	*mp,
1026 	void			__user *arg)
1027 {
1028 	struct xfs_ag_geometry	ageo;
1029 	int			error;
1030 
1031 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
1032 		return -EFAULT;
1033 	if (ageo.ag_flags)
1034 		return -EINVAL;
1035 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
1036 		return -EINVAL;
1037 
1038 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
1039 	if (error)
1040 		return error;
1041 
1042 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
1043 		return -EFAULT;
1044 	return 0;
1045 }
1046 
1047 /*
1048  * Linux extended inode flags interface.
1049  */
1050 
1051 STATIC unsigned int
1052 xfs_merge_ioc_xflags(
1053 	unsigned int	flags,
1054 	unsigned int	start)
1055 {
1056 	unsigned int	xflags = start;
1057 
1058 	if (flags & FS_IMMUTABLE_FL)
1059 		xflags |= FS_XFLAG_IMMUTABLE;
1060 	else
1061 		xflags &= ~FS_XFLAG_IMMUTABLE;
1062 	if (flags & FS_APPEND_FL)
1063 		xflags |= FS_XFLAG_APPEND;
1064 	else
1065 		xflags &= ~FS_XFLAG_APPEND;
1066 	if (flags & FS_SYNC_FL)
1067 		xflags |= FS_XFLAG_SYNC;
1068 	else
1069 		xflags &= ~FS_XFLAG_SYNC;
1070 	if (flags & FS_NOATIME_FL)
1071 		xflags |= FS_XFLAG_NOATIME;
1072 	else
1073 		xflags &= ~FS_XFLAG_NOATIME;
1074 	if (flags & FS_NODUMP_FL)
1075 		xflags |= FS_XFLAG_NODUMP;
1076 	else
1077 		xflags &= ~FS_XFLAG_NODUMP;
1078 
1079 	return xflags;
1080 }
1081 
1082 STATIC unsigned int
1083 xfs_di2lxflags(
1084 	uint16_t	di_flags)
1085 {
1086 	unsigned int	flags = 0;
1087 
1088 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
1089 		flags |= FS_IMMUTABLE_FL;
1090 	if (di_flags & XFS_DIFLAG_APPEND)
1091 		flags |= FS_APPEND_FL;
1092 	if (di_flags & XFS_DIFLAG_SYNC)
1093 		flags |= FS_SYNC_FL;
1094 	if (di_flags & XFS_DIFLAG_NOATIME)
1095 		flags |= FS_NOATIME_FL;
1096 	if (di_flags & XFS_DIFLAG_NODUMP)
1097 		flags |= FS_NODUMP_FL;
1098 	return flags;
1099 }
1100 
1101 static void
1102 xfs_fill_fsxattr(
1103 	struct xfs_inode	*ip,
1104 	bool			attr,
1105 	struct fsxattr		*fa)
1106 {
1107 	simple_fill_fsxattr(fa, xfs_ip2xflags(ip));
1108 	fa->fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
1109 	fa->fsx_cowextsize = ip->i_d.di_cowextsize <<
1110 			ip->i_mount->m_sb.sb_blocklog;
1111 	fa->fsx_projid = ip->i_d.di_projid;
1112 
1113 	if (attr) {
1114 		if (ip->i_afp) {
1115 			if (ip->i_afp->if_flags & XFS_IFEXTENTS)
1116 				fa->fsx_nextents = xfs_iext_count(ip->i_afp);
1117 			else
1118 				fa->fsx_nextents = ip->i_d.di_anextents;
1119 		} else
1120 			fa->fsx_nextents = 0;
1121 	} else {
1122 		if (ip->i_df.if_flags & XFS_IFEXTENTS)
1123 			fa->fsx_nextents = xfs_iext_count(&ip->i_df);
1124 		else
1125 			fa->fsx_nextents = ip->i_d.di_nextents;
1126 	}
1127 }
1128 
1129 STATIC int
1130 xfs_ioc_fsgetxattr(
1131 	xfs_inode_t		*ip,
1132 	int			attr,
1133 	void			__user *arg)
1134 {
1135 	struct fsxattr		fa;
1136 
1137 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1138 	xfs_fill_fsxattr(ip, attr, &fa);
1139 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1140 
1141 	if (copy_to_user(arg, &fa, sizeof(fa)))
1142 		return -EFAULT;
1143 	return 0;
1144 }
1145 
1146 STATIC uint16_t
1147 xfs_flags2diflags(
1148 	struct xfs_inode	*ip,
1149 	unsigned int		xflags)
1150 {
1151 	/* can't set PREALLOC this way, just preserve it */
1152 	uint16_t		di_flags =
1153 		(ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
1154 
1155 	if (xflags & FS_XFLAG_IMMUTABLE)
1156 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1157 	if (xflags & FS_XFLAG_APPEND)
1158 		di_flags |= XFS_DIFLAG_APPEND;
1159 	if (xflags & FS_XFLAG_SYNC)
1160 		di_flags |= XFS_DIFLAG_SYNC;
1161 	if (xflags & FS_XFLAG_NOATIME)
1162 		di_flags |= XFS_DIFLAG_NOATIME;
1163 	if (xflags & FS_XFLAG_NODUMP)
1164 		di_flags |= XFS_DIFLAG_NODUMP;
1165 	if (xflags & FS_XFLAG_NODEFRAG)
1166 		di_flags |= XFS_DIFLAG_NODEFRAG;
1167 	if (xflags & FS_XFLAG_FILESTREAM)
1168 		di_flags |= XFS_DIFLAG_FILESTREAM;
1169 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1170 		if (xflags & FS_XFLAG_RTINHERIT)
1171 			di_flags |= XFS_DIFLAG_RTINHERIT;
1172 		if (xflags & FS_XFLAG_NOSYMLINKS)
1173 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1174 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1175 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1176 		if (xflags & FS_XFLAG_PROJINHERIT)
1177 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1178 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1179 		if (xflags & FS_XFLAG_REALTIME)
1180 			di_flags |= XFS_DIFLAG_REALTIME;
1181 		if (xflags & FS_XFLAG_EXTSIZE)
1182 			di_flags |= XFS_DIFLAG_EXTSIZE;
1183 	}
1184 
1185 	return di_flags;
1186 }
1187 
1188 STATIC uint64_t
1189 xfs_flags2diflags2(
1190 	struct xfs_inode	*ip,
1191 	unsigned int		xflags)
1192 {
1193 	uint64_t		di_flags2 =
1194 		(ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK);
1195 
1196 	if (xflags & FS_XFLAG_DAX)
1197 		di_flags2 |= XFS_DIFLAG2_DAX;
1198 	if (xflags & FS_XFLAG_COWEXTSIZE)
1199 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1200 
1201 	return di_flags2;
1202 }
1203 
1204 STATIC void
1205 xfs_diflags_to_linux(
1206 	struct xfs_inode	*ip)
1207 {
1208 	struct inode		*inode = VFS_I(ip);
1209 	unsigned int		xflags = xfs_ip2xflags(ip);
1210 
1211 	if (xflags & FS_XFLAG_IMMUTABLE)
1212 		inode->i_flags |= S_IMMUTABLE;
1213 	else
1214 		inode->i_flags &= ~S_IMMUTABLE;
1215 	if (xflags & FS_XFLAG_APPEND)
1216 		inode->i_flags |= S_APPEND;
1217 	else
1218 		inode->i_flags &= ~S_APPEND;
1219 	if (xflags & FS_XFLAG_SYNC)
1220 		inode->i_flags |= S_SYNC;
1221 	else
1222 		inode->i_flags &= ~S_SYNC;
1223 	if (xflags & FS_XFLAG_NOATIME)
1224 		inode->i_flags |= S_NOATIME;
1225 	else
1226 		inode->i_flags &= ~S_NOATIME;
1227 #if 0	/* disabled until the flag switching races are sorted out */
1228 	if (xflags & FS_XFLAG_DAX)
1229 		inode->i_flags |= S_DAX;
1230 	else
1231 		inode->i_flags &= ~S_DAX;
1232 #endif
1233 }
1234 
1235 static int
1236 xfs_ioctl_setattr_xflags(
1237 	struct xfs_trans	*tp,
1238 	struct xfs_inode	*ip,
1239 	struct fsxattr		*fa)
1240 {
1241 	struct xfs_mount	*mp = ip->i_mount;
1242 	uint64_t		di_flags2;
1243 
1244 	/* Can't change realtime flag if any extents are allocated. */
1245 	if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
1246 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1247 		return -EINVAL;
1248 
1249 	/* If realtime flag is set then must have realtime device */
1250 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1251 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1252 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1253 			return -EINVAL;
1254 	}
1255 
1256 	/* Clear reflink if we are actually able to set the rt flag. */
1257 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1258 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1259 
1260 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1261 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1262 		return -EINVAL;
1263 
1264 	/* diflags2 only valid for v3 inodes. */
1265 	di_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1266 	if (di_flags2 && !xfs_sb_version_has_v3inode(&mp->m_sb))
1267 		return -EINVAL;
1268 
1269 	ip->i_d.di_flags = xfs_flags2diflags(ip, fa->fsx_xflags);
1270 	ip->i_d.di_flags2 = di_flags2;
1271 
1272 	xfs_diflags_to_linux(ip);
1273 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1274 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1275 	XFS_STATS_INC(mp, xs_ig_attrchg);
1276 	return 0;
1277 }
1278 
1279 /*
1280  * If we are changing DAX flags, we have to ensure the file is clean and any
1281  * cached objects in the address space are invalidated and removed. This
1282  * requires us to lock out other IO and page faults similar to a truncate
1283  * operation. The locks need to be held until the transaction has been committed
1284  * so that the cache invalidation is atomic with respect to the DAX flag
1285  * manipulation.
1286  */
1287 static int
1288 xfs_ioctl_setattr_dax_invalidate(
1289 	struct xfs_inode	*ip,
1290 	struct fsxattr		*fa,
1291 	int			*join_flags)
1292 {
1293 	struct inode		*inode = VFS_I(ip);
1294 	struct super_block	*sb = inode->i_sb;
1295 	int			error;
1296 
1297 	*join_flags = 0;
1298 
1299 	/*
1300 	 * It is only valid to set the DAX flag on regular files and
1301 	 * directories on filesystems where the block size is equal to the page
1302 	 * size. On directories it serves as an inherited hint so we don't
1303 	 * have to check the device for dax support or flush pagecache.
1304 	 */
1305 	if (fa->fsx_xflags & FS_XFLAG_DAX) {
1306 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
1307 
1308 		if (!bdev_dax_supported(target->bt_bdev, sb->s_blocksize))
1309 			return -EINVAL;
1310 	}
1311 
1312 	/* If the DAX state is not changing, we have nothing to do here. */
1313 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && IS_DAX(inode))
1314 		return 0;
1315 	if (!(fa->fsx_xflags & FS_XFLAG_DAX) && !IS_DAX(inode))
1316 		return 0;
1317 
1318 	if (S_ISDIR(inode->i_mode))
1319 		return 0;
1320 
1321 	/* lock, flush and invalidate mapping in preparation for flag change */
1322 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1323 	error = filemap_write_and_wait(inode->i_mapping);
1324 	if (error)
1325 		goto out_unlock;
1326 	error = invalidate_inode_pages2(inode->i_mapping);
1327 	if (error)
1328 		goto out_unlock;
1329 
1330 	*join_flags = XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL;
1331 	return 0;
1332 
1333 out_unlock:
1334 	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1335 	return error;
1336 
1337 }
1338 
1339 /*
1340  * Set up the transaction structure for the setattr operation, checking that we
1341  * have permission to do so. On success, return a clean transaction and the
1342  * inode locked exclusively ready for further operation specific checks. On
1343  * failure, return an error without modifying or locking the inode.
1344  *
1345  * The inode might already be IO locked on call. If this is the case, it is
1346  * indicated in @join_flags and we take full responsibility for ensuring they
1347  * are unlocked from now on. Hence if we have an error here, we still have to
1348  * unlock them. Otherwise, once they are joined to the transaction, they will
1349  * be unlocked on commit/cancel.
1350  */
1351 static struct xfs_trans *
1352 xfs_ioctl_setattr_get_trans(
1353 	struct xfs_inode	*ip,
1354 	int			join_flags)
1355 {
1356 	struct xfs_mount	*mp = ip->i_mount;
1357 	struct xfs_trans	*tp;
1358 	int			error = -EROFS;
1359 
1360 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1361 		goto out_unlock;
1362 	error = -EIO;
1363 	if (XFS_FORCED_SHUTDOWN(mp))
1364 		goto out_unlock;
1365 
1366 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1367 	if (error)
1368 		goto out_unlock;
1369 
1370 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1371 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | join_flags);
1372 	join_flags = 0;
1373 
1374 	/*
1375 	 * CAP_FOWNER overrides the following restrictions:
1376 	 *
1377 	 * The user ID of the calling process must be equal to the file owner
1378 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1379 	 */
1380 	if (!inode_owner_or_capable(VFS_I(ip))) {
1381 		error = -EPERM;
1382 		goto out_cancel;
1383 	}
1384 
1385 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1386 		xfs_trans_set_sync(tp);
1387 
1388 	return tp;
1389 
1390 out_cancel:
1391 	xfs_trans_cancel(tp);
1392 out_unlock:
1393 	if (join_flags)
1394 		xfs_iunlock(ip, join_flags);
1395 	return ERR_PTR(error);
1396 }
1397 
1398 /*
1399  * extent size hint validation is somewhat cumbersome. Rules are:
1400  *
1401  * 1. extent size hint is only valid for directories and regular files
1402  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1403  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1404  * 4. can only be changed on regular files if no extents are allocated
1405  * 5. can be changed on directories at any time
1406  * 6. extsize hint of 0 turns off hints, clears inode flags.
1407  * 7. Extent size must be a multiple of the appropriate block size.
1408  * 8. for non-realtime files, the extent size hint must be limited
1409  *    to half the AG size to avoid alignment extending the extent beyond the
1410  *    limits of the AG.
1411  *
1412  * Please keep this function in sync with xfs_scrub_inode_extsize.
1413  */
1414 static int
1415 xfs_ioctl_setattr_check_extsize(
1416 	struct xfs_inode	*ip,
1417 	struct fsxattr		*fa)
1418 {
1419 	struct xfs_mount	*mp = ip->i_mount;
1420 	xfs_extlen_t		size;
1421 	xfs_fsblock_t		extsize_fsb;
1422 
1423 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_d.di_nextents &&
1424 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1425 		return -EINVAL;
1426 
1427 	if (fa->fsx_extsize == 0)
1428 		return 0;
1429 
1430 	extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1431 	if (extsize_fsb > MAXEXTLEN)
1432 		return -EINVAL;
1433 
1434 	if (XFS_IS_REALTIME_INODE(ip) ||
1435 	    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1436 		size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1437 	} else {
1438 		size = mp->m_sb.sb_blocksize;
1439 		if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1440 			return -EINVAL;
1441 	}
1442 
1443 	if (fa->fsx_extsize % size)
1444 		return -EINVAL;
1445 
1446 	return 0;
1447 }
1448 
1449 /*
1450  * CoW extent size hint validation rules are:
1451  *
1452  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
1453  *    The inode does not have to have any shared blocks, but it must be a v3.
1454  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
1455  *    for a directory, the hint is propagated to new files.
1456  * 3. Can be changed on files & directories at any time.
1457  * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
1458  * 5. Extent size must be a multiple of the appropriate block size.
1459  * 6. The extent size hint must be limited to half the AG size to avoid
1460  *    alignment extending the extent beyond the limits of the AG.
1461  *
1462  * Please keep this function in sync with xfs_scrub_inode_cowextsize.
1463  */
1464 static int
1465 xfs_ioctl_setattr_check_cowextsize(
1466 	struct xfs_inode	*ip,
1467 	struct fsxattr		*fa)
1468 {
1469 	struct xfs_mount	*mp = ip->i_mount;
1470 	xfs_extlen_t		size;
1471 	xfs_fsblock_t		cowextsize_fsb;
1472 
1473 	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
1474 		return 0;
1475 
1476 	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb))
1477 		return -EINVAL;
1478 
1479 	if (fa->fsx_cowextsize == 0)
1480 		return 0;
1481 
1482 	cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1483 	if (cowextsize_fsb > MAXEXTLEN)
1484 		return -EINVAL;
1485 
1486 	size = mp->m_sb.sb_blocksize;
1487 	if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
1488 		return -EINVAL;
1489 
1490 	if (fa->fsx_cowextsize % size)
1491 		return -EINVAL;
1492 
1493 	return 0;
1494 }
1495 
1496 static int
1497 xfs_ioctl_setattr_check_projid(
1498 	struct xfs_inode	*ip,
1499 	struct fsxattr		*fa)
1500 {
1501 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1502 	if (fa->fsx_projid > (uint16_t)-1 &&
1503 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1504 		return -EINVAL;
1505 	return 0;
1506 }
1507 
1508 STATIC int
1509 xfs_ioctl_setattr(
1510 	xfs_inode_t		*ip,
1511 	struct fsxattr		*fa)
1512 {
1513 	struct fsxattr		old_fa;
1514 	struct xfs_mount	*mp = ip->i_mount;
1515 	struct xfs_trans	*tp;
1516 	struct xfs_dquot	*udqp = NULL;
1517 	struct xfs_dquot	*pdqp = NULL;
1518 	struct xfs_dquot	*olddquot = NULL;
1519 	int			code;
1520 	int			join_flags = 0;
1521 
1522 	trace_xfs_ioctl_setattr(ip);
1523 
1524 	code = xfs_ioctl_setattr_check_projid(ip, fa);
1525 	if (code)
1526 		return code;
1527 
1528 	/*
1529 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1530 	 * before we start any other transactions. Trying to do this later
1531 	 * is messy. We don't care to take a readlock to look at the ids
1532 	 * in inode here, because we can't hold it across the trans_reserve.
1533 	 * If the IDs do change before we take the ilock, we're covered
1534 	 * because the i_*dquot fields will get updated anyway.
1535 	 */
1536 	if (XFS_IS_QUOTA_ON(mp)) {
1537 		code = xfs_qm_vop_dqalloc(ip, VFS_I(ip)->i_uid,
1538 				VFS_I(ip)->i_gid, fa->fsx_projid,
1539 				XFS_QMOPT_PQUOTA, &udqp, NULL, &pdqp);
1540 		if (code)
1541 			return code;
1542 	}
1543 
1544 	/*
1545 	 * Changing DAX config may require inode locking for mapping
1546 	 * invalidation. These need to be held all the way to transaction commit
1547 	 * or cancel time, so need to be passed through to
1548 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1549 	 * appropriately.
1550 	 */
1551 	code = xfs_ioctl_setattr_dax_invalidate(ip, fa, &join_flags);
1552 	if (code)
1553 		goto error_free_dquots;
1554 
1555 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1556 	if (IS_ERR(tp)) {
1557 		code = PTR_ERR(tp);
1558 		goto error_free_dquots;
1559 	}
1560 
1561 	if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp) &&
1562 	    ip->i_d.di_projid != fa->fsx_projid) {
1563 		code = xfs_qm_vop_chown_reserve(tp, ip, udqp, NULL, pdqp,
1564 				capable(CAP_FOWNER) ?  XFS_QMOPT_FORCE_RES : 0);
1565 		if (code)	/* out of quota */
1566 			goto error_trans_cancel;
1567 	}
1568 
1569 	xfs_fill_fsxattr(ip, false, &old_fa);
1570 	code = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, fa);
1571 	if (code)
1572 		goto error_trans_cancel;
1573 
1574 	code = xfs_ioctl_setattr_check_extsize(ip, fa);
1575 	if (code)
1576 		goto error_trans_cancel;
1577 
1578 	code = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1579 	if (code)
1580 		goto error_trans_cancel;
1581 
1582 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
1583 	if (code)
1584 		goto error_trans_cancel;
1585 
1586 	/*
1587 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1588 	 * overrides the following restrictions:
1589 	 *
1590 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1591 	 * successful return from chown()
1592 	 */
1593 
1594 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1595 	    !capable_wrt_inode_uidgid(VFS_I(ip), CAP_FSETID))
1596 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1597 
1598 	/* Change the ownerships and register project quota modifications */
1599 	if (ip->i_d.di_projid != fa->fsx_projid) {
1600 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1601 			olddquot = xfs_qm_vop_chown(tp, ip,
1602 						&ip->i_pdquot, pdqp);
1603 		}
1604 		ip->i_d.di_projid = fa->fsx_projid;
1605 	}
1606 
1607 	/*
1608 	 * Only set the extent size hint if we've already determined that the
1609 	 * extent size hint should be set on the inode. If no extent size flags
1610 	 * are set on the inode then unconditionally clear the extent size hint.
1611 	 */
1612 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1613 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1614 	else
1615 		ip->i_d.di_extsize = 0;
1616 	if (xfs_sb_version_has_v3inode(&mp->m_sb) &&
1617 	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1618 		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
1619 				mp->m_sb.sb_blocklog;
1620 	else
1621 		ip->i_d.di_cowextsize = 0;
1622 
1623 	code = xfs_trans_commit(tp);
1624 
1625 	/*
1626 	 * Release any dquot(s) the inode had kept before chown.
1627 	 */
1628 	xfs_qm_dqrele(olddquot);
1629 	xfs_qm_dqrele(udqp);
1630 	xfs_qm_dqrele(pdqp);
1631 
1632 	return code;
1633 
1634 error_trans_cancel:
1635 	xfs_trans_cancel(tp);
1636 error_free_dquots:
1637 	xfs_qm_dqrele(udqp);
1638 	xfs_qm_dqrele(pdqp);
1639 	return code;
1640 }
1641 
1642 STATIC int
1643 xfs_ioc_fssetxattr(
1644 	xfs_inode_t		*ip,
1645 	struct file		*filp,
1646 	void			__user *arg)
1647 {
1648 	struct fsxattr		fa;
1649 	int error;
1650 
1651 	if (copy_from_user(&fa, arg, sizeof(fa)))
1652 		return -EFAULT;
1653 
1654 	error = mnt_want_write_file(filp);
1655 	if (error)
1656 		return error;
1657 	error = xfs_ioctl_setattr(ip, &fa);
1658 	mnt_drop_write_file(filp);
1659 	return error;
1660 }
1661 
1662 STATIC int
1663 xfs_ioc_getxflags(
1664 	xfs_inode_t		*ip,
1665 	void			__user *arg)
1666 {
1667 	unsigned int		flags;
1668 
1669 	flags = xfs_di2lxflags(ip->i_d.di_flags);
1670 	if (copy_to_user(arg, &flags, sizeof(flags)))
1671 		return -EFAULT;
1672 	return 0;
1673 }
1674 
1675 STATIC int
1676 xfs_ioc_setxflags(
1677 	struct xfs_inode	*ip,
1678 	struct file		*filp,
1679 	void			__user *arg)
1680 {
1681 	struct xfs_trans	*tp;
1682 	struct fsxattr		fa;
1683 	struct fsxattr		old_fa;
1684 	unsigned int		flags;
1685 	int			join_flags = 0;
1686 	int			error;
1687 
1688 	if (copy_from_user(&flags, arg, sizeof(flags)))
1689 		return -EFAULT;
1690 
1691 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1692 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1693 		      FS_SYNC_FL))
1694 		return -EOPNOTSUPP;
1695 
1696 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1697 
1698 	error = mnt_want_write_file(filp);
1699 	if (error)
1700 		return error;
1701 
1702 	/*
1703 	 * Changing DAX config may require inode locking for mapping
1704 	 * invalidation. These need to be held all the way to transaction commit
1705 	 * or cancel time, so need to be passed through to
1706 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1707 	 * appropriately.
1708 	 */
1709 	error = xfs_ioctl_setattr_dax_invalidate(ip, &fa, &join_flags);
1710 	if (error)
1711 		goto out_drop_write;
1712 
1713 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1714 	if (IS_ERR(tp)) {
1715 		error = PTR_ERR(tp);
1716 		goto out_drop_write;
1717 	}
1718 
1719 	xfs_fill_fsxattr(ip, false, &old_fa);
1720 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, &fa);
1721 	if (error) {
1722 		xfs_trans_cancel(tp);
1723 		goto out_drop_write;
1724 	}
1725 
1726 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1727 	if (error) {
1728 		xfs_trans_cancel(tp);
1729 		goto out_drop_write;
1730 	}
1731 
1732 	error = xfs_trans_commit(tp);
1733 out_drop_write:
1734 	mnt_drop_write_file(filp);
1735 	return error;
1736 }
1737 
1738 static bool
1739 xfs_getbmap_format(
1740 	struct kgetbmap		*p,
1741 	struct getbmapx __user	*u,
1742 	size_t			recsize)
1743 {
1744 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1745 	    put_user(p->bmv_block, &u->bmv_block) ||
1746 	    put_user(p->bmv_length, &u->bmv_length) ||
1747 	    put_user(0, &u->bmv_count) ||
1748 	    put_user(0, &u->bmv_entries))
1749 		return false;
1750 	if (recsize < sizeof(struct getbmapx))
1751 		return true;
1752 	if (put_user(0, &u->bmv_iflags) ||
1753 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1754 	    put_user(0, &u->bmv_unused1) ||
1755 	    put_user(0, &u->bmv_unused2))
1756 		return false;
1757 	return true;
1758 }
1759 
1760 STATIC int
1761 xfs_ioc_getbmap(
1762 	struct file		*file,
1763 	unsigned int		cmd,
1764 	void			__user *arg)
1765 {
1766 	struct getbmapx		bmx = { 0 };
1767 	struct kgetbmap		*buf;
1768 	size_t			recsize;
1769 	int			error, i;
1770 
1771 	switch (cmd) {
1772 	case XFS_IOC_GETBMAPA:
1773 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1774 		/*FALLTHRU*/
1775 	case XFS_IOC_GETBMAP:
1776 		if (file->f_mode & FMODE_NOCMTIME)
1777 			bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1778 		/* struct getbmap is a strict subset of struct getbmapx. */
1779 		recsize = sizeof(struct getbmap);
1780 		break;
1781 	case XFS_IOC_GETBMAPX:
1782 		recsize = sizeof(struct getbmapx);
1783 		break;
1784 	default:
1785 		return -EINVAL;
1786 	}
1787 
1788 	if (copy_from_user(&bmx, arg, recsize))
1789 		return -EFAULT;
1790 
1791 	if (bmx.bmv_count < 2)
1792 		return -EINVAL;
1793 	if (bmx.bmv_count > ULONG_MAX / recsize)
1794 		return -ENOMEM;
1795 
1796 	buf = kmem_zalloc_large(bmx.bmv_count * sizeof(*buf), 0);
1797 	if (!buf)
1798 		return -ENOMEM;
1799 
1800 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1801 	if (error)
1802 		goto out_free_buf;
1803 
1804 	error = -EFAULT;
1805 	if (copy_to_user(arg, &bmx, recsize))
1806 		goto out_free_buf;
1807 	arg += recsize;
1808 
1809 	for (i = 0; i < bmx.bmv_entries; i++) {
1810 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1811 			goto out_free_buf;
1812 		arg += recsize;
1813 	}
1814 
1815 	error = 0;
1816 out_free_buf:
1817 	kmem_free(buf);
1818 	return error;
1819 }
1820 
1821 struct getfsmap_info {
1822 	struct xfs_mount	*mp;
1823 	struct fsmap_head __user *data;
1824 	unsigned int		idx;
1825 	__u32			last_flags;
1826 };
1827 
1828 STATIC int
1829 xfs_getfsmap_format(struct xfs_fsmap *xfm, void *priv)
1830 {
1831 	struct getfsmap_info	*info = priv;
1832 	struct fsmap		fm;
1833 
1834 	trace_xfs_getfsmap_mapping(info->mp, xfm);
1835 
1836 	info->last_flags = xfm->fmr_flags;
1837 	xfs_fsmap_from_internal(&fm, xfm);
1838 	if (copy_to_user(&info->data->fmh_recs[info->idx++], &fm,
1839 			sizeof(struct fsmap)))
1840 		return -EFAULT;
1841 
1842 	return 0;
1843 }
1844 
1845 STATIC int
1846 xfs_ioc_getfsmap(
1847 	struct xfs_inode	*ip,
1848 	struct fsmap_head	__user *arg)
1849 {
1850 	struct getfsmap_info	info = { NULL };
1851 	struct xfs_fsmap_head	xhead = {0};
1852 	struct fsmap_head	head;
1853 	bool			aborted = false;
1854 	int			error;
1855 
1856 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1857 		return -EFAULT;
1858 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1859 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1860 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1861 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1862 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1863 		return -EINVAL;
1864 
1865 	xhead.fmh_iflags = head.fmh_iflags;
1866 	xhead.fmh_count = head.fmh_count;
1867 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1868 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1869 
1870 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1871 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1872 
1873 	info.mp = ip->i_mount;
1874 	info.data = arg;
1875 	error = xfs_getfsmap(ip->i_mount, &xhead, xfs_getfsmap_format, &info);
1876 	if (error == -ECANCELED) {
1877 		error = 0;
1878 		aborted = true;
1879 	} else if (error)
1880 		return error;
1881 
1882 	/* If we didn't abort, set the "last" flag in the last fmx */
1883 	if (!aborted && info.idx) {
1884 		info.last_flags |= FMR_OF_LAST;
1885 		if (copy_to_user(&info.data->fmh_recs[info.idx - 1].fmr_flags,
1886 				&info.last_flags, sizeof(info.last_flags)))
1887 			return -EFAULT;
1888 	}
1889 
1890 	/* copy back header */
1891 	head.fmh_entries = xhead.fmh_entries;
1892 	head.fmh_oflags = xhead.fmh_oflags;
1893 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head)))
1894 		return -EFAULT;
1895 
1896 	return 0;
1897 }
1898 
1899 STATIC int
1900 xfs_ioc_scrub_metadata(
1901 	struct xfs_inode		*ip,
1902 	void				__user *arg)
1903 {
1904 	struct xfs_scrub_metadata	scrub;
1905 	int				error;
1906 
1907 	if (!capable(CAP_SYS_ADMIN))
1908 		return -EPERM;
1909 
1910 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1911 		return -EFAULT;
1912 
1913 	error = xfs_scrub_metadata(ip, &scrub);
1914 	if (error)
1915 		return error;
1916 
1917 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1918 		return -EFAULT;
1919 
1920 	return 0;
1921 }
1922 
1923 int
1924 xfs_ioc_swapext(
1925 	xfs_swapext_t	*sxp)
1926 {
1927 	xfs_inode_t     *ip, *tip;
1928 	struct fd	f, tmp;
1929 	int		error = 0;
1930 
1931 	/* Pull information for the target fd */
1932 	f = fdget((int)sxp->sx_fdtarget);
1933 	if (!f.file) {
1934 		error = -EINVAL;
1935 		goto out;
1936 	}
1937 
1938 	if (!(f.file->f_mode & FMODE_WRITE) ||
1939 	    !(f.file->f_mode & FMODE_READ) ||
1940 	    (f.file->f_flags & O_APPEND)) {
1941 		error = -EBADF;
1942 		goto out_put_file;
1943 	}
1944 
1945 	tmp = fdget((int)sxp->sx_fdtmp);
1946 	if (!tmp.file) {
1947 		error = -EINVAL;
1948 		goto out_put_file;
1949 	}
1950 
1951 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1952 	    !(tmp.file->f_mode & FMODE_READ) ||
1953 	    (tmp.file->f_flags & O_APPEND)) {
1954 		error = -EBADF;
1955 		goto out_put_tmp_file;
1956 	}
1957 
1958 	if (IS_SWAPFILE(file_inode(f.file)) ||
1959 	    IS_SWAPFILE(file_inode(tmp.file))) {
1960 		error = -EINVAL;
1961 		goto out_put_tmp_file;
1962 	}
1963 
1964 	/*
1965 	 * We need to ensure that the fds passed in point to XFS inodes
1966 	 * before we cast and access them as XFS structures as we have no
1967 	 * control over what the user passes us here.
1968 	 */
1969 	if (f.file->f_op != &xfs_file_operations ||
1970 	    tmp.file->f_op != &xfs_file_operations) {
1971 		error = -EINVAL;
1972 		goto out_put_tmp_file;
1973 	}
1974 
1975 	ip = XFS_I(file_inode(f.file));
1976 	tip = XFS_I(file_inode(tmp.file));
1977 
1978 	if (ip->i_mount != tip->i_mount) {
1979 		error = -EINVAL;
1980 		goto out_put_tmp_file;
1981 	}
1982 
1983 	if (ip->i_ino == tip->i_ino) {
1984 		error = -EINVAL;
1985 		goto out_put_tmp_file;
1986 	}
1987 
1988 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1989 		error = -EIO;
1990 		goto out_put_tmp_file;
1991 	}
1992 
1993 	error = xfs_swap_extents(ip, tip, sxp);
1994 
1995  out_put_tmp_file:
1996 	fdput(tmp);
1997  out_put_file:
1998 	fdput(f);
1999  out:
2000 	return error;
2001 }
2002 
2003 static int
2004 xfs_ioc_getlabel(
2005 	struct xfs_mount	*mp,
2006 	char			__user *user_label)
2007 {
2008 	struct xfs_sb		*sbp = &mp->m_sb;
2009 	char			label[XFSLABEL_MAX + 1];
2010 
2011 	/* Paranoia */
2012 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
2013 
2014 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
2015 	memset(label, 0, sizeof(label));
2016 	spin_lock(&mp->m_sb_lock);
2017 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
2018 	spin_unlock(&mp->m_sb_lock);
2019 
2020 	if (copy_to_user(user_label, label, sizeof(label)))
2021 		return -EFAULT;
2022 	return 0;
2023 }
2024 
2025 static int
2026 xfs_ioc_setlabel(
2027 	struct file		*filp,
2028 	struct xfs_mount	*mp,
2029 	char			__user *newlabel)
2030 {
2031 	struct xfs_sb		*sbp = &mp->m_sb;
2032 	char			label[XFSLABEL_MAX + 1];
2033 	size_t			len;
2034 	int			error;
2035 
2036 	if (!capable(CAP_SYS_ADMIN))
2037 		return -EPERM;
2038 	/*
2039 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
2040 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
2041 	 * (required) NULL character to test the incoming label length.
2042 	 * NB: The on disk label doesn't need to be null terminated.
2043 	 */
2044 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
2045 		return -EFAULT;
2046 	len = strnlen(label, XFSLABEL_MAX + 1);
2047 	if (len > sizeof(sbp->sb_fname))
2048 		return -EINVAL;
2049 
2050 	error = mnt_want_write_file(filp);
2051 	if (error)
2052 		return error;
2053 
2054 	spin_lock(&mp->m_sb_lock);
2055 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
2056 	memcpy(sbp->sb_fname, label, len);
2057 	spin_unlock(&mp->m_sb_lock);
2058 
2059 	/*
2060 	 * Now we do several things to satisfy userspace.
2061 	 * In addition to normal logging of the primary superblock, we also
2062 	 * immediately write these changes to sector zero for the primary, then
2063 	 * update all backup supers (as xfs_db does for a label change), then
2064 	 * invalidate the block device page cache.  This is so that any prior
2065 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
2066 	 * and userspace will see the newly-written label.
2067 	 */
2068 	error = xfs_sync_sb_buf(mp);
2069 	if (error)
2070 		goto out;
2071 	/*
2072 	 * growfs also updates backup supers so lock against that.
2073 	 */
2074 	mutex_lock(&mp->m_growlock);
2075 	error = xfs_update_secondary_sbs(mp);
2076 	mutex_unlock(&mp->m_growlock);
2077 
2078 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
2079 
2080 out:
2081 	mnt_drop_write_file(filp);
2082 	return error;
2083 }
2084 
2085 /*
2086  * Note: some of the ioctl's return positive numbers as a
2087  * byte count indicating success, such as readlink_by_handle.
2088  * So we don't "sign flip" like most other routines.  This means
2089  * true errors need to be returned as a negative value.
2090  */
2091 long
2092 xfs_file_ioctl(
2093 	struct file		*filp,
2094 	unsigned int		cmd,
2095 	unsigned long		p)
2096 {
2097 	struct inode		*inode = file_inode(filp);
2098 	struct xfs_inode	*ip = XFS_I(inode);
2099 	struct xfs_mount	*mp = ip->i_mount;
2100 	void			__user *arg = (void __user *)p;
2101 	int			error;
2102 
2103 	trace_xfs_file_ioctl(ip);
2104 
2105 	switch (cmd) {
2106 	case FITRIM:
2107 		return xfs_ioc_trim(mp, arg);
2108 	case FS_IOC_GETFSLABEL:
2109 		return xfs_ioc_getlabel(mp, arg);
2110 	case FS_IOC_SETFSLABEL:
2111 		return xfs_ioc_setlabel(filp, mp, arg);
2112 	case XFS_IOC_ALLOCSP:
2113 	case XFS_IOC_FREESP:
2114 	case XFS_IOC_ALLOCSP64:
2115 	case XFS_IOC_FREESP64: {
2116 		xfs_flock64_t		bf;
2117 
2118 		if (copy_from_user(&bf, arg, sizeof(bf)))
2119 			return -EFAULT;
2120 		return xfs_ioc_space(filp, &bf);
2121 	}
2122 	case XFS_IOC_DIOINFO: {
2123 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
2124 		struct dioattr		da;
2125 
2126 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
2127 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
2128 
2129 		if (copy_to_user(arg, &da, sizeof(da)))
2130 			return -EFAULT;
2131 		return 0;
2132 	}
2133 
2134 	case XFS_IOC_FSBULKSTAT_SINGLE:
2135 	case XFS_IOC_FSBULKSTAT:
2136 	case XFS_IOC_FSINUMBERS:
2137 		return xfs_ioc_fsbulkstat(mp, cmd, arg);
2138 
2139 	case XFS_IOC_BULKSTAT:
2140 		return xfs_ioc_bulkstat(mp, cmd, arg);
2141 	case XFS_IOC_INUMBERS:
2142 		return xfs_ioc_inumbers(mp, cmd, arg);
2143 
2144 	case XFS_IOC_FSGEOMETRY_V1:
2145 		return xfs_ioc_fsgeometry(mp, arg, 3);
2146 	case XFS_IOC_FSGEOMETRY_V4:
2147 		return xfs_ioc_fsgeometry(mp, arg, 4);
2148 	case XFS_IOC_FSGEOMETRY:
2149 		return xfs_ioc_fsgeometry(mp, arg, 5);
2150 
2151 	case XFS_IOC_AG_GEOMETRY:
2152 		return xfs_ioc_ag_geometry(mp, arg);
2153 
2154 	case XFS_IOC_GETVERSION:
2155 		return put_user(inode->i_generation, (int __user *)arg);
2156 
2157 	case XFS_IOC_FSGETXATTR:
2158 		return xfs_ioc_fsgetxattr(ip, 0, arg);
2159 	case XFS_IOC_FSGETXATTRA:
2160 		return xfs_ioc_fsgetxattr(ip, 1, arg);
2161 	case XFS_IOC_FSSETXATTR:
2162 		return xfs_ioc_fssetxattr(ip, filp, arg);
2163 	case XFS_IOC_GETXFLAGS:
2164 		return xfs_ioc_getxflags(ip, arg);
2165 	case XFS_IOC_SETXFLAGS:
2166 		return xfs_ioc_setxflags(ip, filp, arg);
2167 
2168 	case XFS_IOC_GETBMAP:
2169 	case XFS_IOC_GETBMAPA:
2170 	case XFS_IOC_GETBMAPX:
2171 		return xfs_ioc_getbmap(filp, cmd, arg);
2172 
2173 	case FS_IOC_GETFSMAP:
2174 		return xfs_ioc_getfsmap(ip, arg);
2175 
2176 	case XFS_IOC_SCRUB_METADATA:
2177 		return xfs_ioc_scrub_metadata(ip, arg);
2178 
2179 	case XFS_IOC_FD_TO_HANDLE:
2180 	case XFS_IOC_PATH_TO_HANDLE:
2181 	case XFS_IOC_PATH_TO_FSHANDLE: {
2182 		xfs_fsop_handlereq_t	hreq;
2183 
2184 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2185 			return -EFAULT;
2186 		return xfs_find_handle(cmd, &hreq);
2187 	}
2188 	case XFS_IOC_OPEN_BY_HANDLE: {
2189 		xfs_fsop_handlereq_t	hreq;
2190 
2191 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2192 			return -EFAULT;
2193 		return xfs_open_by_handle(filp, &hreq);
2194 	}
2195 
2196 	case XFS_IOC_READLINK_BY_HANDLE: {
2197 		xfs_fsop_handlereq_t	hreq;
2198 
2199 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2200 			return -EFAULT;
2201 		return xfs_readlink_by_handle(filp, &hreq);
2202 	}
2203 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2204 		return xfs_attrlist_by_handle(filp, arg);
2205 
2206 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2207 		return xfs_attrmulti_by_handle(filp, arg);
2208 
2209 	case XFS_IOC_SWAPEXT: {
2210 		struct xfs_swapext	sxp;
2211 
2212 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2213 			return -EFAULT;
2214 		error = mnt_want_write_file(filp);
2215 		if (error)
2216 			return error;
2217 		error = xfs_ioc_swapext(&sxp);
2218 		mnt_drop_write_file(filp);
2219 		return error;
2220 	}
2221 
2222 	case XFS_IOC_FSCOUNTS: {
2223 		xfs_fsop_counts_t out;
2224 
2225 		xfs_fs_counts(mp, &out);
2226 
2227 		if (copy_to_user(arg, &out, sizeof(out)))
2228 			return -EFAULT;
2229 		return 0;
2230 	}
2231 
2232 	case XFS_IOC_SET_RESBLKS: {
2233 		xfs_fsop_resblks_t inout;
2234 		uint64_t	   in;
2235 
2236 		if (!capable(CAP_SYS_ADMIN))
2237 			return -EPERM;
2238 
2239 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2240 			return -EROFS;
2241 
2242 		if (copy_from_user(&inout, arg, sizeof(inout)))
2243 			return -EFAULT;
2244 
2245 		error = mnt_want_write_file(filp);
2246 		if (error)
2247 			return error;
2248 
2249 		/* input parameter is passed in resblks field of structure */
2250 		in = inout.resblks;
2251 		error = xfs_reserve_blocks(mp, &in, &inout);
2252 		mnt_drop_write_file(filp);
2253 		if (error)
2254 			return error;
2255 
2256 		if (copy_to_user(arg, &inout, sizeof(inout)))
2257 			return -EFAULT;
2258 		return 0;
2259 	}
2260 
2261 	case XFS_IOC_GET_RESBLKS: {
2262 		xfs_fsop_resblks_t out;
2263 
2264 		if (!capable(CAP_SYS_ADMIN))
2265 			return -EPERM;
2266 
2267 		error = xfs_reserve_blocks(mp, NULL, &out);
2268 		if (error)
2269 			return error;
2270 
2271 		if (copy_to_user(arg, &out, sizeof(out)))
2272 			return -EFAULT;
2273 
2274 		return 0;
2275 	}
2276 
2277 	case XFS_IOC_FSGROWFSDATA: {
2278 		xfs_growfs_data_t in;
2279 
2280 		if (copy_from_user(&in, arg, sizeof(in)))
2281 			return -EFAULT;
2282 
2283 		error = mnt_want_write_file(filp);
2284 		if (error)
2285 			return error;
2286 		error = xfs_growfs_data(mp, &in);
2287 		mnt_drop_write_file(filp);
2288 		return error;
2289 	}
2290 
2291 	case XFS_IOC_FSGROWFSLOG: {
2292 		xfs_growfs_log_t in;
2293 
2294 		if (copy_from_user(&in, arg, sizeof(in)))
2295 			return -EFAULT;
2296 
2297 		error = mnt_want_write_file(filp);
2298 		if (error)
2299 			return error;
2300 		error = xfs_growfs_log(mp, &in);
2301 		mnt_drop_write_file(filp);
2302 		return error;
2303 	}
2304 
2305 	case XFS_IOC_FSGROWFSRT: {
2306 		xfs_growfs_rt_t in;
2307 
2308 		if (copy_from_user(&in, arg, sizeof(in)))
2309 			return -EFAULT;
2310 
2311 		error = mnt_want_write_file(filp);
2312 		if (error)
2313 			return error;
2314 		error = xfs_growfs_rt(mp, &in);
2315 		mnt_drop_write_file(filp);
2316 		return error;
2317 	}
2318 
2319 	case XFS_IOC_GOINGDOWN: {
2320 		uint32_t in;
2321 
2322 		if (!capable(CAP_SYS_ADMIN))
2323 			return -EPERM;
2324 
2325 		if (get_user(in, (uint32_t __user *)arg))
2326 			return -EFAULT;
2327 
2328 		return xfs_fs_goingdown(mp, in);
2329 	}
2330 
2331 	case XFS_IOC_ERROR_INJECTION: {
2332 		xfs_error_injection_t in;
2333 
2334 		if (!capable(CAP_SYS_ADMIN))
2335 			return -EPERM;
2336 
2337 		if (copy_from_user(&in, arg, sizeof(in)))
2338 			return -EFAULT;
2339 
2340 		return xfs_errortag_add(mp, in.errtag);
2341 	}
2342 
2343 	case XFS_IOC_ERROR_CLEARALL:
2344 		if (!capable(CAP_SYS_ADMIN))
2345 			return -EPERM;
2346 
2347 		return xfs_errortag_clearall(mp);
2348 
2349 	case XFS_IOC_FREE_EOFBLOCKS: {
2350 		struct xfs_fs_eofblocks eofb;
2351 		struct xfs_eofblocks keofb;
2352 
2353 		if (!capable(CAP_SYS_ADMIN))
2354 			return -EPERM;
2355 
2356 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2357 			return -EROFS;
2358 
2359 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2360 			return -EFAULT;
2361 
2362 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
2363 		if (error)
2364 			return error;
2365 
2366 		sb_start_write(mp->m_super);
2367 		error = xfs_icache_free_eofblocks(mp, &keofb);
2368 		sb_end_write(mp->m_super);
2369 		return error;
2370 	}
2371 
2372 	default:
2373 		return -ENOTTY;
2374 	}
2375 }
2376