xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 941518d6)
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 #include <linux/fileattr.h>
44 
45 /*
46  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
47  * a file or fs handle.
48  *
49  * XFS_IOC_PATH_TO_FSHANDLE
50  *    returns fs handle for a mount point or path within that mount point
51  * XFS_IOC_FD_TO_HANDLE
52  *    returns full handle for a FD opened in user space
53  * XFS_IOC_PATH_TO_HANDLE
54  *    returns full handle for a path
55  */
56 int
57 xfs_find_handle(
58 	unsigned int		cmd,
59 	xfs_fsop_handlereq_t	*hreq)
60 {
61 	int			hsize;
62 	xfs_handle_t		handle;
63 	struct inode		*inode;
64 	struct fd		f = {NULL};
65 	struct path		path;
66 	int			error;
67 	struct xfs_inode	*ip;
68 
69 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
70 		f = fdget(hreq->fd);
71 		if (!f.file)
72 			return -EBADF;
73 		inode = file_inode(f.file);
74 	} else {
75 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
76 		if (error)
77 			return error;
78 		inode = d_inode(path.dentry);
79 	}
80 	ip = XFS_I(inode);
81 
82 	/*
83 	 * We can only generate handles for inodes residing on a XFS filesystem,
84 	 * and only for regular files, directories or symbolic links.
85 	 */
86 	error = -EINVAL;
87 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
88 		goto out_put;
89 
90 	error = -EBADF;
91 	if (!S_ISREG(inode->i_mode) &&
92 	    !S_ISDIR(inode->i_mode) &&
93 	    !S_ISLNK(inode->i_mode))
94 		goto out_put;
95 
96 
97 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
98 
99 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
100 		/*
101 		 * This handle only contains an fsid, zero the rest.
102 		 */
103 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
104 		hsize = sizeof(xfs_fsid_t);
105 	} else {
106 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
107 					sizeof(handle.ha_fid.fid_len);
108 		handle.ha_fid.fid_pad = 0;
109 		handle.ha_fid.fid_gen = inode->i_generation;
110 		handle.ha_fid.fid_ino = ip->i_ino;
111 		hsize = sizeof(xfs_handle_t);
112 	}
113 
114 	error = -EFAULT;
115 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
116 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
117 		goto out_put;
118 
119 	error = 0;
120 
121  out_put:
122 	if (cmd == XFS_IOC_FD_TO_HANDLE)
123 		fdput(f);
124 	else
125 		path_put(&path);
126 	return error;
127 }
128 
129 /*
130  * No need to do permission checks on the various pathname components
131  * as the handle operations are privileged.
132  */
133 STATIC int
134 xfs_handle_acceptable(
135 	void			*context,
136 	struct dentry		*dentry)
137 {
138 	return 1;
139 }
140 
141 /*
142  * Convert userspace handle data into a dentry.
143  */
144 struct dentry *
145 xfs_handle_to_dentry(
146 	struct file		*parfilp,
147 	void __user		*uhandle,
148 	u32			hlen)
149 {
150 	xfs_handle_t		handle;
151 	struct xfs_fid64	fid;
152 
153 	/*
154 	 * Only allow handle opens under a directory.
155 	 */
156 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
157 		return ERR_PTR(-ENOTDIR);
158 
159 	if (hlen != sizeof(xfs_handle_t))
160 		return ERR_PTR(-EINVAL);
161 	if (copy_from_user(&handle, uhandle, hlen))
162 		return ERR_PTR(-EFAULT);
163 	if (handle.ha_fid.fid_len !=
164 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
165 		return ERR_PTR(-EINVAL);
166 
167 	memset(&fid, 0, sizeof(struct fid));
168 	fid.ino = handle.ha_fid.fid_ino;
169 	fid.gen = handle.ha_fid.fid_gen;
170 
171 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
172 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
173 			xfs_handle_acceptable, NULL);
174 }
175 
176 STATIC struct dentry *
177 xfs_handlereq_to_dentry(
178 	struct file		*parfilp,
179 	xfs_fsop_handlereq_t	*hreq)
180 {
181 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
182 }
183 
184 int
185 xfs_open_by_handle(
186 	struct file		*parfilp,
187 	xfs_fsop_handlereq_t	*hreq)
188 {
189 	const struct cred	*cred = current_cred();
190 	int			error;
191 	int			fd;
192 	int			permflag;
193 	struct file		*filp;
194 	struct inode		*inode;
195 	struct dentry		*dentry;
196 	fmode_t			fmode;
197 	struct path		path;
198 
199 	if (!capable(CAP_SYS_ADMIN))
200 		return -EPERM;
201 
202 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
203 	if (IS_ERR(dentry))
204 		return PTR_ERR(dentry);
205 	inode = d_inode(dentry);
206 
207 	/* Restrict xfs_open_by_handle to directories & regular files. */
208 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
209 		error = -EPERM;
210 		goto out_dput;
211 	}
212 
213 #if BITS_PER_LONG != 32
214 	hreq->oflags |= O_LARGEFILE;
215 #endif
216 
217 	permflag = hreq->oflags;
218 	fmode = OPEN_FMODE(permflag);
219 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
220 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
221 		error = -EPERM;
222 		goto out_dput;
223 	}
224 
225 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
226 		error = -EPERM;
227 		goto out_dput;
228 	}
229 
230 	/* Can't write directories. */
231 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
232 		error = -EISDIR;
233 		goto out_dput;
234 	}
235 
236 	fd = get_unused_fd_flags(0);
237 	if (fd < 0) {
238 		error = fd;
239 		goto out_dput;
240 	}
241 
242 	path.mnt = parfilp->f_path.mnt;
243 	path.dentry = dentry;
244 	filp = dentry_open(&path, hreq->oflags, cred);
245 	dput(dentry);
246 	if (IS_ERR(filp)) {
247 		put_unused_fd(fd);
248 		return PTR_ERR(filp);
249 	}
250 
251 	if (S_ISREG(inode->i_mode)) {
252 		filp->f_flags |= O_NOATIME;
253 		filp->f_mode |= FMODE_NOCMTIME;
254 	}
255 
256 	fd_install(fd, filp);
257 	return fd;
258 
259  out_dput:
260 	dput(dentry);
261 	return error;
262 }
263 
264 int
265 xfs_readlink_by_handle(
266 	struct file		*parfilp,
267 	xfs_fsop_handlereq_t	*hreq)
268 {
269 	struct dentry		*dentry;
270 	__u32			olen;
271 	int			error;
272 
273 	if (!capable(CAP_SYS_ADMIN))
274 		return -EPERM;
275 
276 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
277 	if (IS_ERR(dentry))
278 		return PTR_ERR(dentry);
279 
280 	/* Restrict this handle operation to symlinks only. */
281 	if (!d_is_symlink(dentry)) {
282 		error = -EINVAL;
283 		goto out_dput;
284 	}
285 
286 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
287 		error = -EFAULT;
288 		goto out_dput;
289 	}
290 
291 	error = vfs_readlink(dentry, hreq->ohandle, olen);
292 
293  out_dput:
294 	dput(dentry);
295 	return error;
296 }
297 
298 /*
299  * Format an attribute and copy it out to the user's buffer.
300  * Take care to check values and protect against them changing later,
301  * we may be reading them directly out of a user buffer.
302  */
303 static void
304 xfs_ioc_attr_put_listent(
305 	struct xfs_attr_list_context *context,
306 	int			flags,
307 	unsigned char		*name,
308 	int			namelen,
309 	int			valuelen)
310 {
311 	struct xfs_attrlist	*alist = context->buffer;
312 	struct xfs_attrlist_ent	*aep;
313 	int			arraytop;
314 
315 	ASSERT(!context->seen_enough);
316 	ASSERT(context->count >= 0);
317 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
318 	ASSERT(context->firstu >= sizeof(*alist));
319 	ASSERT(context->firstu <= context->bufsize);
320 
321 	/*
322 	 * Only list entries in the right namespace.
323 	 */
324 	if (context->attr_filter != (flags & XFS_ATTR_NSP_ONDISK_MASK))
325 		return;
326 
327 	arraytop = sizeof(*alist) +
328 			context->count * sizeof(alist->al_offset[0]);
329 
330 	/* decrement by the actual bytes used by the attr */
331 	context->firstu -= round_up(offsetof(struct xfs_attrlist_ent, a_name) +
332 			namelen + 1, sizeof(uint32_t));
333 	if (context->firstu < arraytop) {
334 		trace_xfs_attr_list_full(context);
335 		alist->al_more = 1;
336 		context->seen_enough = 1;
337 		return;
338 	}
339 
340 	aep = context->buffer + context->firstu;
341 	aep->a_valuelen = valuelen;
342 	memcpy(aep->a_name, name, namelen);
343 	aep->a_name[namelen] = 0;
344 	alist->al_offset[context->count++] = context->firstu;
345 	alist->al_count = context->count;
346 	trace_xfs_attr_list_add(context);
347 }
348 
349 static unsigned int
350 xfs_attr_filter(
351 	u32			ioc_flags)
352 {
353 	if (ioc_flags & XFS_IOC_ATTR_ROOT)
354 		return XFS_ATTR_ROOT;
355 	if (ioc_flags & XFS_IOC_ATTR_SECURE)
356 		return XFS_ATTR_SECURE;
357 	return 0;
358 }
359 
360 static unsigned int
361 xfs_attr_flags(
362 	u32			ioc_flags)
363 {
364 	if (ioc_flags & XFS_IOC_ATTR_CREATE)
365 		return XATTR_CREATE;
366 	if (ioc_flags & XFS_IOC_ATTR_REPLACE)
367 		return XATTR_REPLACE;
368 	return 0;
369 }
370 
371 int
372 xfs_ioc_attr_list(
373 	struct xfs_inode		*dp,
374 	void __user			*ubuf,
375 	size_t				bufsize,
376 	int				flags,
377 	struct xfs_attrlist_cursor __user *ucursor)
378 {
379 	struct xfs_attr_list_context	context = { };
380 	struct xfs_attrlist		*alist;
381 	void				*buffer;
382 	int				error;
383 
384 	if (bufsize < sizeof(struct xfs_attrlist) ||
385 	    bufsize > XFS_XATTR_LIST_MAX)
386 		return -EINVAL;
387 
388 	/*
389 	 * Reject flags, only allow namespaces.
390 	 */
391 	if (flags & ~(XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
392 		return -EINVAL;
393 	if (flags == (XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
394 		return -EINVAL;
395 
396 	/*
397 	 * Validate the cursor.
398 	 */
399 	if (copy_from_user(&context.cursor, ucursor, sizeof(context.cursor)))
400 		return -EFAULT;
401 	if (context.cursor.pad1 || context.cursor.pad2)
402 		return -EINVAL;
403 	if (!context.cursor.initted &&
404 	    (context.cursor.hashval || context.cursor.blkno ||
405 	     context.cursor.offset))
406 		return -EINVAL;
407 
408 	buffer = kvzalloc(bufsize, GFP_KERNEL);
409 	if (!buffer)
410 		return -ENOMEM;
411 
412 	/*
413 	 * Initialize the output buffer.
414 	 */
415 	context.dp = dp;
416 	context.resynch = 1;
417 	context.attr_filter = xfs_attr_filter(flags);
418 	context.buffer = buffer;
419 	context.bufsize = round_down(bufsize, sizeof(uint32_t));
420 	context.firstu = context.bufsize;
421 	context.put_listent = xfs_ioc_attr_put_listent;
422 
423 	alist = context.buffer;
424 	alist->al_count = 0;
425 	alist->al_more = 0;
426 	alist->al_offset[0] = context.bufsize;
427 
428 	error = xfs_attr_list(&context);
429 	if (error)
430 		goto out_free;
431 
432 	if (copy_to_user(ubuf, buffer, bufsize) ||
433 	    copy_to_user(ucursor, &context.cursor, sizeof(context.cursor)))
434 		error = -EFAULT;
435 out_free:
436 	kmem_free(buffer);
437 	return error;
438 }
439 
440 STATIC int
441 xfs_attrlist_by_handle(
442 	struct file		*parfilp,
443 	struct xfs_fsop_attrlist_handlereq __user *p)
444 {
445 	struct xfs_fsop_attrlist_handlereq al_hreq;
446 	struct dentry		*dentry;
447 	int			error = -ENOMEM;
448 
449 	if (!capable(CAP_SYS_ADMIN))
450 		return -EPERM;
451 	if (copy_from_user(&al_hreq, p, sizeof(al_hreq)))
452 		return -EFAULT;
453 
454 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
455 	if (IS_ERR(dentry))
456 		return PTR_ERR(dentry);
457 
458 	error = xfs_ioc_attr_list(XFS_I(d_inode(dentry)), al_hreq.buffer,
459 				  al_hreq.buflen, al_hreq.flags, &p->pos);
460 	dput(dentry);
461 	return error;
462 }
463 
464 static int
465 xfs_attrmulti_attr_get(
466 	struct inode		*inode,
467 	unsigned char		*name,
468 	unsigned char		__user *ubuf,
469 	uint32_t		*len,
470 	uint32_t		flags)
471 {
472 	struct xfs_da_args	args = {
473 		.dp		= XFS_I(inode),
474 		.attr_filter	= xfs_attr_filter(flags),
475 		.attr_flags	= xfs_attr_flags(flags),
476 		.name		= name,
477 		.namelen	= strlen(name),
478 		.valuelen	= *len,
479 	};
480 	int			error;
481 
482 	if (*len > XFS_XATTR_SIZE_MAX)
483 		return -EINVAL;
484 
485 	error = xfs_attr_get(&args);
486 	if (error)
487 		goto out_kfree;
488 
489 	*len = args.valuelen;
490 	if (copy_to_user(ubuf, args.value, args.valuelen))
491 		error = -EFAULT;
492 
493 out_kfree:
494 	kmem_free(args.value);
495 	return error;
496 }
497 
498 static int
499 xfs_attrmulti_attr_set(
500 	struct inode		*inode,
501 	unsigned char		*name,
502 	const unsigned char	__user *ubuf,
503 	uint32_t		len,
504 	uint32_t		flags)
505 {
506 	struct xfs_da_args	args = {
507 		.dp		= XFS_I(inode),
508 		.attr_filter	= xfs_attr_filter(flags),
509 		.attr_flags	= xfs_attr_flags(flags),
510 		.name		= name,
511 		.namelen	= strlen(name),
512 	};
513 	int			error;
514 
515 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
516 		return -EPERM;
517 
518 	if (ubuf) {
519 		if (len > XFS_XATTR_SIZE_MAX)
520 			return -EINVAL;
521 		args.value = memdup_user(ubuf, len);
522 		if (IS_ERR(args.value))
523 			return PTR_ERR(args.value);
524 		args.valuelen = len;
525 	}
526 
527 	error = xfs_attr_set(&args);
528 	if (!error && (flags & XFS_IOC_ATTR_ROOT))
529 		xfs_forget_acl(inode, name);
530 	kfree(args.value);
531 	return error;
532 }
533 
534 int
535 xfs_ioc_attrmulti_one(
536 	struct file		*parfilp,
537 	struct inode		*inode,
538 	uint32_t		opcode,
539 	void __user		*uname,
540 	void __user		*value,
541 	uint32_t		*len,
542 	uint32_t		flags)
543 {
544 	unsigned char		*name;
545 	int			error;
546 
547 	if ((flags & XFS_IOC_ATTR_ROOT) && (flags & XFS_IOC_ATTR_SECURE))
548 		return -EINVAL;
549 
550 	name = strndup_user(uname, MAXNAMELEN);
551 	if (IS_ERR(name))
552 		return PTR_ERR(name);
553 
554 	switch (opcode) {
555 	case ATTR_OP_GET:
556 		error = xfs_attrmulti_attr_get(inode, name, value, len, flags);
557 		break;
558 	case ATTR_OP_REMOVE:
559 		value = NULL;
560 		*len = 0;
561 		fallthrough;
562 	case ATTR_OP_SET:
563 		error = mnt_want_write_file(parfilp);
564 		if (error)
565 			break;
566 		error = xfs_attrmulti_attr_set(inode, name, value, *len, flags);
567 		mnt_drop_write_file(parfilp);
568 		break;
569 	default:
570 		error = -EINVAL;
571 		break;
572 	}
573 
574 	kfree(name);
575 	return error;
576 }
577 
578 STATIC int
579 xfs_attrmulti_by_handle(
580 	struct file		*parfilp,
581 	void			__user *arg)
582 {
583 	int			error;
584 	xfs_attr_multiop_t	*ops;
585 	xfs_fsop_attrmulti_handlereq_t am_hreq;
586 	struct dentry		*dentry;
587 	unsigned int		i, size;
588 
589 	if (!capable(CAP_SYS_ADMIN))
590 		return -EPERM;
591 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
592 		return -EFAULT;
593 
594 	/* overflow check */
595 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
596 		return -E2BIG;
597 
598 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
599 	if (IS_ERR(dentry))
600 		return PTR_ERR(dentry);
601 
602 	error = -E2BIG;
603 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
604 	if (!size || size > 16 * PAGE_SIZE)
605 		goto out_dput;
606 
607 	ops = memdup_user(am_hreq.ops, size);
608 	if (IS_ERR(ops)) {
609 		error = PTR_ERR(ops);
610 		goto out_dput;
611 	}
612 
613 	error = 0;
614 	for (i = 0; i < am_hreq.opcount; i++) {
615 		ops[i].am_error = xfs_ioc_attrmulti_one(parfilp,
616 				d_inode(dentry), ops[i].am_opcode,
617 				ops[i].am_attrname, ops[i].am_attrvalue,
618 				&ops[i].am_length, ops[i].am_flags);
619 	}
620 
621 	if (copy_to_user(am_hreq.ops, ops, size))
622 		error = -EFAULT;
623 
624 	kfree(ops);
625  out_dput:
626 	dput(dentry);
627 	return error;
628 }
629 
630 /* Return 0 on success or positive error */
631 int
632 xfs_fsbulkstat_one_fmt(
633 	struct xfs_ibulk		*breq,
634 	const struct xfs_bulkstat	*bstat)
635 {
636 	struct xfs_bstat		bs1;
637 
638 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
639 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
640 		return -EFAULT;
641 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
642 }
643 
644 int
645 xfs_fsinumbers_fmt(
646 	struct xfs_ibulk		*breq,
647 	const struct xfs_inumbers	*igrp)
648 {
649 	struct xfs_inogrp		ig1;
650 
651 	xfs_inumbers_to_inogrp(&ig1, igrp);
652 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
653 		return -EFAULT;
654 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
655 }
656 
657 STATIC int
658 xfs_ioc_fsbulkstat(
659 	struct file		*file,
660 	unsigned int		cmd,
661 	void			__user *arg)
662 {
663 	struct xfs_mount	*mp = XFS_I(file_inode(file))->i_mount;
664 	struct xfs_fsop_bulkreq	bulkreq;
665 	struct xfs_ibulk	breq = {
666 		.mp		= mp,
667 		.mnt_userns	= file_mnt_user_ns(file),
668 		.ocount		= 0,
669 	};
670 	xfs_ino_t		lastino;
671 	int			error;
672 
673 	/* done = 1 if there are more stats to get and if bulkstat */
674 	/* should be called again (unused here, but used in dmapi) */
675 
676 	if (!capable(CAP_SYS_ADMIN))
677 		return -EPERM;
678 
679 	if (xfs_is_shutdown(mp))
680 		return -EIO;
681 
682 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
683 		return -EFAULT;
684 
685 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
686 		return -EFAULT;
687 
688 	if (bulkreq.icount <= 0)
689 		return -EINVAL;
690 
691 	if (bulkreq.ubuffer == NULL)
692 		return -EINVAL;
693 
694 	breq.ubuffer = bulkreq.ubuffer;
695 	breq.icount = bulkreq.icount;
696 
697 	/*
698 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
699 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
700 	 * that *lastip contains either zero or the number of the last inode to
701 	 * be examined by the previous call and return results starting with
702 	 * the next inode after that.  The new bulk request back end functions
703 	 * take the inode to start with, so we have to compute the startino
704 	 * parameter from lastino to maintain correct function.  lastino == 0
705 	 * is a special case because it has traditionally meant "first inode
706 	 * in filesystem".
707 	 */
708 	if (cmd == XFS_IOC_FSINUMBERS) {
709 		breq.startino = lastino ? lastino + 1 : 0;
710 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
711 		lastino = breq.startino - 1;
712 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
713 		breq.startino = lastino;
714 		breq.icount = 1;
715 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
716 	} else {	/* XFS_IOC_FSBULKSTAT */
717 		breq.startino = lastino ? lastino + 1 : 0;
718 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
719 		lastino = breq.startino - 1;
720 	}
721 
722 	if (error)
723 		return error;
724 
725 	if (bulkreq.lastip != NULL &&
726 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
727 		return -EFAULT;
728 
729 	if (bulkreq.ocount != NULL &&
730 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
731 		return -EFAULT;
732 
733 	return 0;
734 }
735 
736 /* Return 0 on success or positive error */
737 static int
738 xfs_bulkstat_fmt(
739 	struct xfs_ibulk		*breq,
740 	const struct xfs_bulkstat	*bstat)
741 {
742 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
743 		return -EFAULT;
744 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
745 }
746 
747 /*
748  * Check the incoming bulk request @hdr from userspace and initialize the
749  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
750  * should proceed; -ECANCELED if there's nothing to do; or the usual
751  * negative error code.
752  */
753 static int
754 xfs_bulk_ireq_setup(
755 	struct xfs_mount	*mp,
756 	struct xfs_bulk_ireq	*hdr,
757 	struct xfs_ibulk	*breq,
758 	void __user		*ubuffer)
759 {
760 	if (hdr->icount == 0 ||
761 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
762 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
763 		return -EINVAL;
764 
765 	breq->startino = hdr->ino;
766 	breq->ubuffer = ubuffer;
767 	breq->icount = hdr->icount;
768 	breq->ocount = 0;
769 	breq->flags = 0;
770 
771 	/*
772 	 * The @ino parameter is a special value, so we must look it up here.
773 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
774 	 * worth of data.
775 	 */
776 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
777 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
778 			return -EINVAL;
779 
780 		switch (hdr->ino) {
781 		case XFS_BULK_IREQ_SPECIAL_ROOT:
782 			hdr->ino = mp->m_sb.sb_rootino;
783 			break;
784 		default:
785 			return -EINVAL;
786 		}
787 		breq->icount = 1;
788 	}
789 
790 	/*
791 	 * The IREQ_AGNO flag means that we only want results from a given AG.
792 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
793 	 * beyond the specified AG then we return no results.
794 	 */
795 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
796 		if (hdr->agno >= mp->m_sb.sb_agcount)
797 			return -EINVAL;
798 
799 		if (breq->startino == 0)
800 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
801 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
802 			return -EINVAL;
803 
804 		breq->flags |= XFS_IBULK_SAME_AG;
805 
806 		/* Asking for an inode past the end of the AG?  We're done! */
807 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
808 			return -ECANCELED;
809 	} else if (hdr->agno)
810 		return -EINVAL;
811 
812 	/* Asking for an inode past the end of the FS?  We're done! */
813 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
814 		return -ECANCELED;
815 
816 	return 0;
817 }
818 
819 /*
820  * Update the userspace bulk request @hdr to reflect the end state of the
821  * internal bulk request @breq.
822  */
823 static void
824 xfs_bulk_ireq_teardown(
825 	struct xfs_bulk_ireq	*hdr,
826 	struct xfs_ibulk	*breq)
827 {
828 	hdr->ino = breq->startino;
829 	hdr->ocount = breq->ocount;
830 }
831 
832 /* Handle the v5 bulkstat ioctl. */
833 STATIC int
834 xfs_ioc_bulkstat(
835 	struct file			*file,
836 	unsigned int			cmd,
837 	struct xfs_bulkstat_req __user	*arg)
838 {
839 	struct xfs_mount		*mp = XFS_I(file_inode(file))->i_mount;
840 	struct xfs_bulk_ireq		hdr;
841 	struct xfs_ibulk		breq = {
842 		.mp			= mp,
843 		.mnt_userns		= file_mnt_user_ns(file),
844 	};
845 	int				error;
846 
847 	if (!capable(CAP_SYS_ADMIN))
848 		return -EPERM;
849 
850 	if (xfs_is_shutdown(mp))
851 		return -EIO;
852 
853 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
854 		return -EFAULT;
855 
856 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
857 	if (error == -ECANCELED)
858 		goto out_teardown;
859 	if (error < 0)
860 		return error;
861 
862 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
863 	if (error)
864 		return error;
865 
866 out_teardown:
867 	xfs_bulk_ireq_teardown(&hdr, &breq);
868 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
869 		return -EFAULT;
870 
871 	return 0;
872 }
873 
874 STATIC int
875 xfs_inumbers_fmt(
876 	struct xfs_ibulk		*breq,
877 	const struct xfs_inumbers	*igrp)
878 {
879 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
880 		return -EFAULT;
881 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
882 }
883 
884 /* Handle the v5 inumbers ioctl. */
885 STATIC int
886 xfs_ioc_inumbers(
887 	struct xfs_mount		*mp,
888 	unsigned int			cmd,
889 	struct xfs_inumbers_req __user	*arg)
890 {
891 	struct xfs_bulk_ireq		hdr;
892 	struct xfs_ibulk		breq = {
893 		.mp			= mp,
894 	};
895 	int				error;
896 
897 	if (!capable(CAP_SYS_ADMIN))
898 		return -EPERM;
899 
900 	if (xfs_is_shutdown(mp))
901 		return -EIO;
902 
903 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
904 		return -EFAULT;
905 
906 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
907 	if (error == -ECANCELED)
908 		goto out_teardown;
909 	if (error < 0)
910 		return error;
911 
912 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
913 	if (error)
914 		return error;
915 
916 out_teardown:
917 	xfs_bulk_ireq_teardown(&hdr, &breq);
918 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
919 		return -EFAULT;
920 
921 	return 0;
922 }
923 
924 STATIC int
925 xfs_ioc_fsgeometry(
926 	struct xfs_mount	*mp,
927 	void			__user *arg,
928 	int			struct_version)
929 {
930 	struct xfs_fsop_geom	fsgeo;
931 	size_t			len;
932 
933 	xfs_fs_geometry(mp, &fsgeo, struct_version);
934 
935 	if (struct_version <= 3)
936 		len = sizeof(struct xfs_fsop_geom_v1);
937 	else if (struct_version == 4)
938 		len = sizeof(struct xfs_fsop_geom_v4);
939 	else {
940 		xfs_fsop_geom_health(mp, &fsgeo);
941 		len = sizeof(fsgeo);
942 	}
943 
944 	if (copy_to_user(arg, &fsgeo, len))
945 		return -EFAULT;
946 	return 0;
947 }
948 
949 STATIC int
950 xfs_ioc_ag_geometry(
951 	struct xfs_mount	*mp,
952 	void			__user *arg)
953 {
954 	struct xfs_ag_geometry	ageo;
955 	int			error;
956 
957 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
958 		return -EFAULT;
959 	if (ageo.ag_flags)
960 		return -EINVAL;
961 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
962 		return -EINVAL;
963 
964 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
965 	if (error)
966 		return error;
967 
968 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
969 		return -EFAULT;
970 	return 0;
971 }
972 
973 /*
974  * Linux extended inode flags interface.
975  */
976 
977 static void
978 xfs_fill_fsxattr(
979 	struct xfs_inode	*ip,
980 	int			whichfork,
981 	struct fileattr		*fa)
982 {
983 	struct xfs_mount	*mp = ip->i_mount;
984 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
985 
986 	fileattr_fill_xflags(fa, xfs_ip2xflags(ip));
987 
988 	if (ip->i_diflags & XFS_DIFLAG_EXTSIZE) {
989 		fa->fsx_extsize = XFS_FSB_TO_B(mp, ip->i_extsize);
990 	} else if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
991 		/*
992 		 * Don't let a misaligned extent size hint on a directory
993 		 * escape to userspace if it won't pass the setattr checks
994 		 * later.
995 		 */
996 		if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
997 		    ip->i_extsize % mp->m_sb.sb_rextsize > 0) {
998 			fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE |
999 					    FS_XFLAG_EXTSZINHERIT);
1000 			fa->fsx_extsize = 0;
1001 		} else {
1002 			fa->fsx_extsize = XFS_FSB_TO_B(mp, ip->i_extsize);
1003 		}
1004 	}
1005 
1006 	if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
1007 		fa->fsx_cowextsize = XFS_FSB_TO_B(mp, ip->i_cowextsize);
1008 	fa->fsx_projid = ip->i_projid;
1009 	if (ifp && !xfs_need_iread_extents(ifp))
1010 		fa->fsx_nextents = xfs_iext_count(ifp);
1011 	else
1012 		fa->fsx_nextents = xfs_ifork_nextents(ifp);
1013 }
1014 
1015 STATIC int
1016 xfs_ioc_fsgetxattra(
1017 	xfs_inode_t		*ip,
1018 	void			__user *arg)
1019 {
1020 	struct fileattr		fa;
1021 
1022 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1023 	xfs_fill_fsxattr(ip, XFS_ATTR_FORK, &fa);
1024 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1025 
1026 	return copy_fsxattr_to_user(&fa, arg);
1027 }
1028 
1029 int
1030 xfs_fileattr_get(
1031 	struct dentry		*dentry,
1032 	struct fileattr		*fa)
1033 {
1034 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1035 
1036 	if (d_is_special(dentry))
1037 		return -ENOTTY;
1038 
1039 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1040 	xfs_fill_fsxattr(ip, XFS_DATA_FORK, fa);
1041 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1042 
1043 	return 0;
1044 }
1045 
1046 STATIC uint16_t
1047 xfs_flags2diflags(
1048 	struct xfs_inode	*ip,
1049 	unsigned int		xflags)
1050 {
1051 	/* can't set PREALLOC this way, just preserve it */
1052 	uint16_t		di_flags =
1053 		(ip->i_diflags & XFS_DIFLAG_PREALLOC);
1054 
1055 	if (xflags & FS_XFLAG_IMMUTABLE)
1056 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1057 	if (xflags & FS_XFLAG_APPEND)
1058 		di_flags |= XFS_DIFLAG_APPEND;
1059 	if (xflags & FS_XFLAG_SYNC)
1060 		di_flags |= XFS_DIFLAG_SYNC;
1061 	if (xflags & FS_XFLAG_NOATIME)
1062 		di_flags |= XFS_DIFLAG_NOATIME;
1063 	if (xflags & FS_XFLAG_NODUMP)
1064 		di_flags |= XFS_DIFLAG_NODUMP;
1065 	if (xflags & FS_XFLAG_NODEFRAG)
1066 		di_flags |= XFS_DIFLAG_NODEFRAG;
1067 	if (xflags & FS_XFLAG_FILESTREAM)
1068 		di_flags |= XFS_DIFLAG_FILESTREAM;
1069 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1070 		if (xflags & FS_XFLAG_RTINHERIT)
1071 			di_flags |= XFS_DIFLAG_RTINHERIT;
1072 		if (xflags & FS_XFLAG_NOSYMLINKS)
1073 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1074 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1075 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1076 		if (xflags & FS_XFLAG_PROJINHERIT)
1077 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1078 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1079 		if (xflags & FS_XFLAG_REALTIME)
1080 			di_flags |= XFS_DIFLAG_REALTIME;
1081 		if (xflags & FS_XFLAG_EXTSIZE)
1082 			di_flags |= XFS_DIFLAG_EXTSIZE;
1083 	}
1084 
1085 	return di_flags;
1086 }
1087 
1088 STATIC uint64_t
1089 xfs_flags2diflags2(
1090 	struct xfs_inode	*ip,
1091 	unsigned int		xflags)
1092 {
1093 	uint64_t		di_flags2 =
1094 		(ip->i_diflags2 & (XFS_DIFLAG2_REFLINK |
1095 				   XFS_DIFLAG2_BIGTIME));
1096 
1097 	if (xflags & FS_XFLAG_DAX)
1098 		di_flags2 |= XFS_DIFLAG2_DAX;
1099 	if (xflags & FS_XFLAG_COWEXTSIZE)
1100 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1101 
1102 	return di_flags2;
1103 }
1104 
1105 static int
1106 xfs_ioctl_setattr_xflags(
1107 	struct xfs_trans	*tp,
1108 	struct xfs_inode	*ip,
1109 	struct fileattr		*fa)
1110 {
1111 	struct xfs_mount	*mp = ip->i_mount;
1112 	uint64_t		i_flags2;
1113 
1114 	/* Can't change realtime flag if any extents are allocated. */
1115 	if ((ip->i_df.if_nextents || ip->i_delayed_blks) &&
1116 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1117 		return -EINVAL;
1118 
1119 	/* If realtime flag is set then must have realtime device */
1120 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1121 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1122 		    (ip->i_extsize % mp->m_sb.sb_rextsize))
1123 			return -EINVAL;
1124 	}
1125 
1126 	/* Clear reflink if we are actually able to set the rt flag. */
1127 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1128 		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1129 
1130 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1131 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1132 		return -EINVAL;
1133 
1134 	/* diflags2 only valid for v3 inodes. */
1135 	i_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1136 	if (i_flags2 && !xfs_has_v3inodes(mp))
1137 		return -EINVAL;
1138 
1139 	ip->i_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1140 	ip->i_diflags2 = i_flags2;
1141 
1142 	xfs_diflags_to_iflags(ip, false);
1143 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1144 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1145 	XFS_STATS_INC(mp, xs_ig_attrchg);
1146 	return 0;
1147 }
1148 
1149 static void
1150 xfs_ioctl_setattr_prepare_dax(
1151 	struct xfs_inode	*ip,
1152 	struct fileattr		*fa)
1153 {
1154 	struct xfs_mount	*mp = ip->i_mount;
1155 	struct inode            *inode = VFS_I(ip);
1156 
1157 	if (S_ISDIR(inode->i_mode))
1158 		return;
1159 
1160 	if (xfs_has_dax_always(mp) || xfs_has_dax_never(mp))
1161 		return;
1162 
1163 	if (((fa->fsx_xflags & FS_XFLAG_DAX) &&
1164 	    !(ip->i_diflags2 & XFS_DIFLAG2_DAX)) ||
1165 	    (!(fa->fsx_xflags & FS_XFLAG_DAX) &&
1166 	     (ip->i_diflags2 & XFS_DIFLAG2_DAX)))
1167 		d_mark_dontcache(inode);
1168 }
1169 
1170 /*
1171  * Set up the transaction structure for the setattr operation, checking that we
1172  * have permission to do so. On success, return a clean transaction and the
1173  * inode locked exclusively ready for further operation specific checks. On
1174  * failure, return an error without modifying or locking the inode.
1175  */
1176 static struct xfs_trans *
1177 xfs_ioctl_setattr_get_trans(
1178 	struct xfs_inode	*ip,
1179 	struct xfs_dquot	*pdqp)
1180 {
1181 	struct xfs_mount	*mp = ip->i_mount;
1182 	struct xfs_trans	*tp;
1183 	int			error = -EROFS;
1184 
1185 	if (xfs_is_readonly(mp))
1186 		goto out_error;
1187 	error = -EIO;
1188 	if (xfs_is_shutdown(mp))
1189 		goto out_error;
1190 
1191 	error = xfs_trans_alloc_ichange(ip, NULL, NULL, pdqp,
1192 			capable(CAP_FOWNER), &tp);
1193 	if (error)
1194 		goto out_error;
1195 
1196 	if (xfs_has_wsync(mp))
1197 		xfs_trans_set_sync(tp);
1198 
1199 	return tp;
1200 
1201 out_error:
1202 	return ERR_PTR(error);
1203 }
1204 
1205 /*
1206  * Validate a proposed extent size hint.  For regular files, the hint can only
1207  * be changed if no extents are allocated.
1208  */
1209 static int
1210 xfs_ioctl_setattr_check_extsize(
1211 	struct xfs_inode	*ip,
1212 	struct fileattr		*fa)
1213 {
1214 	struct xfs_mount	*mp = ip->i_mount;
1215 	xfs_failaddr_t		failaddr;
1216 	uint16_t		new_diflags;
1217 
1218 	if (!fa->fsx_valid)
1219 		return 0;
1220 
1221 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_df.if_nextents &&
1222 	    XFS_FSB_TO_B(mp, ip->i_extsize) != fa->fsx_extsize)
1223 		return -EINVAL;
1224 
1225 	if (fa->fsx_extsize & mp->m_blockmask)
1226 		return -EINVAL;
1227 
1228 	new_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1229 
1230 	/*
1231 	 * Inode verifiers do not check that the extent size hint is an integer
1232 	 * multiple of the rt extent size on a directory with both rtinherit
1233 	 * and extszinherit flags set.  Don't let sysadmins misconfigure
1234 	 * directories.
1235 	 */
1236 	if ((new_diflags & XFS_DIFLAG_RTINHERIT) &&
1237 	    (new_diflags & XFS_DIFLAG_EXTSZINHERIT)) {
1238 		unsigned int	rtextsize_bytes;
1239 
1240 		rtextsize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
1241 		if (fa->fsx_extsize % rtextsize_bytes)
1242 			return -EINVAL;
1243 	}
1244 
1245 	failaddr = xfs_inode_validate_extsize(ip->i_mount,
1246 			XFS_B_TO_FSB(mp, fa->fsx_extsize),
1247 			VFS_I(ip)->i_mode, new_diflags);
1248 	return failaddr != NULL ? -EINVAL : 0;
1249 }
1250 
1251 static int
1252 xfs_ioctl_setattr_check_cowextsize(
1253 	struct xfs_inode	*ip,
1254 	struct fileattr		*fa)
1255 {
1256 	struct xfs_mount	*mp = ip->i_mount;
1257 	xfs_failaddr_t		failaddr;
1258 	uint64_t		new_diflags2;
1259 	uint16_t		new_diflags;
1260 
1261 	if (!fa->fsx_valid)
1262 		return 0;
1263 
1264 	if (fa->fsx_cowextsize & mp->m_blockmask)
1265 		return -EINVAL;
1266 
1267 	new_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1268 	new_diflags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1269 
1270 	failaddr = xfs_inode_validate_cowextsize(ip->i_mount,
1271 			XFS_B_TO_FSB(mp, fa->fsx_cowextsize),
1272 			VFS_I(ip)->i_mode, new_diflags, new_diflags2);
1273 	return failaddr != NULL ? -EINVAL : 0;
1274 }
1275 
1276 static int
1277 xfs_ioctl_setattr_check_projid(
1278 	struct xfs_inode	*ip,
1279 	struct fileattr		*fa)
1280 {
1281 	if (!fa->fsx_valid)
1282 		return 0;
1283 
1284 	/* Disallow 32bit project ids if 32bit IDs are not enabled. */
1285 	if (fa->fsx_projid > (uint16_t)-1 &&
1286 	    !xfs_has_projid32(ip->i_mount))
1287 		return -EINVAL;
1288 	return 0;
1289 }
1290 
1291 int
1292 xfs_fileattr_set(
1293 	struct user_namespace	*mnt_userns,
1294 	struct dentry		*dentry,
1295 	struct fileattr		*fa)
1296 {
1297 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1298 	struct xfs_mount	*mp = ip->i_mount;
1299 	struct xfs_trans	*tp;
1300 	struct xfs_dquot	*pdqp = NULL;
1301 	struct xfs_dquot	*olddquot = NULL;
1302 	int			error;
1303 
1304 	trace_xfs_ioctl_setattr(ip);
1305 
1306 	if (d_is_special(dentry))
1307 		return -ENOTTY;
1308 
1309 	if (!fa->fsx_valid) {
1310 		if (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL |
1311 				  FS_NOATIME_FL | FS_NODUMP_FL |
1312 				  FS_SYNC_FL | FS_DAX_FL | FS_PROJINHERIT_FL))
1313 			return -EOPNOTSUPP;
1314 	}
1315 
1316 	error = xfs_ioctl_setattr_check_projid(ip, fa);
1317 	if (error)
1318 		return error;
1319 
1320 	/*
1321 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1322 	 * before we start any other transactions. Trying to do this later
1323 	 * is messy. We don't care to take a readlock to look at the ids
1324 	 * in inode here, because we can't hold it across the trans_reserve.
1325 	 * If the IDs do change before we take the ilock, we're covered
1326 	 * because the i_*dquot fields will get updated anyway.
1327 	 */
1328 	if (fa->fsx_valid && XFS_IS_QUOTA_ON(mp)) {
1329 		error = xfs_qm_vop_dqalloc(ip, VFS_I(ip)->i_uid,
1330 				VFS_I(ip)->i_gid, fa->fsx_projid,
1331 				XFS_QMOPT_PQUOTA, NULL, NULL, &pdqp);
1332 		if (error)
1333 			return error;
1334 	}
1335 
1336 	xfs_ioctl_setattr_prepare_dax(ip, fa);
1337 
1338 	tp = xfs_ioctl_setattr_get_trans(ip, pdqp);
1339 	if (IS_ERR(tp)) {
1340 		error = PTR_ERR(tp);
1341 		goto error_free_dquots;
1342 	}
1343 
1344 	error = xfs_ioctl_setattr_check_extsize(ip, fa);
1345 	if (error)
1346 		goto error_trans_cancel;
1347 
1348 	error = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1349 	if (error)
1350 		goto error_trans_cancel;
1351 
1352 	error = xfs_ioctl_setattr_xflags(tp, ip, fa);
1353 	if (error)
1354 		goto error_trans_cancel;
1355 
1356 	if (!fa->fsx_valid)
1357 		goto skip_xattr;
1358 	/*
1359 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1360 	 * overrides the following restrictions:
1361 	 *
1362 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1363 	 * successful return from chown()
1364 	 */
1365 
1366 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1367 	    !capable_wrt_inode_uidgid(mnt_userns, VFS_I(ip), CAP_FSETID))
1368 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1369 
1370 	/* Change the ownerships and register project quota modifications */
1371 	if (ip->i_projid != fa->fsx_projid) {
1372 		if (XFS_IS_PQUOTA_ON(mp)) {
1373 			olddquot = xfs_qm_vop_chown(tp, ip,
1374 						&ip->i_pdquot, pdqp);
1375 		}
1376 		ip->i_projid = fa->fsx_projid;
1377 	}
1378 
1379 	/*
1380 	 * Only set the extent size hint if we've already determined that the
1381 	 * extent size hint should be set on the inode. If no extent size flags
1382 	 * are set on the inode then unconditionally clear the extent size hint.
1383 	 */
1384 	if (ip->i_diflags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1385 		ip->i_extsize = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1386 	else
1387 		ip->i_extsize = 0;
1388 
1389 	if (xfs_has_v3inodes(mp)) {
1390 		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
1391 			ip->i_cowextsize = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1392 		else
1393 			ip->i_cowextsize = 0;
1394 	}
1395 
1396 skip_xattr:
1397 	error = xfs_trans_commit(tp);
1398 
1399 	/*
1400 	 * Release any dquot(s) the inode had kept before chown.
1401 	 */
1402 	xfs_qm_dqrele(olddquot);
1403 	xfs_qm_dqrele(pdqp);
1404 
1405 	return error;
1406 
1407 error_trans_cancel:
1408 	xfs_trans_cancel(tp);
1409 error_free_dquots:
1410 	xfs_qm_dqrele(pdqp);
1411 	return error;
1412 }
1413 
1414 static bool
1415 xfs_getbmap_format(
1416 	struct kgetbmap		*p,
1417 	struct getbmapx __user	*u,
1418 	size_t			recsize)
1419 {
1420 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1421 	    put_user(p->bmv_block, &u->bmv_block) ||
1422 	    put_user(p->bmv_length, &u->bmv_length) ||
1423 	    put_user(0, &u->bmv_count) ||
1424 	    put_user(0, &u->bmv_entries))
1425 		return false;
1426 	if (recsize < sizeof(struct getbmapx))
1427 		return true;
1428 	if (put_user(0, &u->bmv_iflags) ||
1429 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1430 	    put_user(0, &u->bmv_unused1) ||
1431 	    put_user(0, &u->bmv_unused2))
1432 		return false;
1433 	return true;
1434 }
1435 
1436 STATIC int
1437 xfs_ioc_getbmap(
1438 	struct file		*file,
1439 	unsigned int		cmd,
1440 	void			__user *arg)
1441 {
1442 	struct getbmapx		bmx = { 0 };
1443 	struct kgetbmap		*buf;
1444 	size_t			recsize;
1445 	int			error, i;
1446 
1447 	switch (cmd) {
1448 	case XFS_IOC_GETBMAPA:
1449 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1450 		fallthrough;
1451 	case XFS_IOC_GETBMAP:
1452 		/* struct getbmap is a strict subset of struct getbmapx. */
1453 		recsize = sizeof(struct getbmap);
1454 		break;
1455 	case XFS_IOC_GETBMAPX:
1456 		recsize = sizeof(struct getbmapx);
1457 		break;
1458 	default:
1459 		return -EINVAL;
1460 	}
1461 
1462 	if (copy_from_user(&bmx, arg, recsize))
1463 		return -EFAULT;
1464 
1465 	if (bmx.bmv_count < 2)
1466 		return -EINVAL;
1467 	if (bmx.bmv_count > ULONG_MAX / recsize)
1468 		return -ENOMEM;
1469 
1470 	buf = kvcalloc(bmx.bmv_count, sizeof(*buf), GFP_KERNEL);
1471 	if (!buf)
1472 		return -ENOMEM;
1473 
1474 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1475 	if (error)
1476 		goto out_free_buf;
1477 
1478 	error = -EFAULT;
1479 	if (copy_to_user(arg, &bmx, recsize))
1480 		goto out_free_buf;
1481 	arg += recsize;
1482 
1483 	for (i = 0; i < bmx.bmv_entries; i++) {
1484 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1485 			goto out_free_buf;
1486 		arg += recsize;
1487 	}
1488 
1489 	error = 0;
1490 out_free_buf:
1491 	kmem_free(buf);
1492 	return error;
1493 }
1494 
1495 STATIC int
1496 xfs_ioc_getfsmap(
1497 	struct xfs_inode	*ip,
1498 	struct fsmap_head	__user *arg)
1499 {
1500 	struct xfs_fsmap_head	xhead = {0};
1501 	struct fsmap_head	head;
1502 	struct fsmap		*recs;
1503 	unsigned int		count;
1504 	__u32			last_flags = 0;
1505 	bool			done = false;
1506 	int			error;
1507 
1508 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1509 		return -EFAULT;
1510 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1511 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1512 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1513 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1514 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1515 		return -EINVAL;
1516 
1517 	/*
1518 	 * Use an internal memory buffer so that we don't have to copy fsmap
1519 	 * data to userspace while holding locks.  Start by trying to allocate
1520 	 * up to 128k for the buffer, but fall back to a single page if needed.
1521 	 */
1522 	count = min_t(unsigned int, head.fmh_count,
1523 			131072 / sizeof(struct fsmap));
1524 	recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL);
1525 	if (!recs) {
1526 		count = min_t(unsigned int, head.fmh_count,
1527 				PAGE_SIZE / sizeof(struct fsmap));
1528 		recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL);
1529 		if (!recs)
1530 			return -ENOMEM;
1531 	}
1532 
1533 	xhead.fmh_iflags = head.fmh_iflags;
1534 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1535 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1536 
1537 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1538 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1539 
1540 	head.fmh_entries = 0;
1541 	do {
1542 		struct fsmap __user	*user_recs;
1543 		struct fsmap		*last_rec;
1544 
1545 		user_recs = &arg->fmh_recs[head.fmh_entries];
1546 		xhead.fmh_entries = 0;
1547 		xhead.fmh_count = min_t(unsigned int, count,
1548 					head.fmh_count - head.fmh_entries);
1549 
1550 		/* Run query, record how many entries we got. */
1551 		error = xfs_getfsmap(ip->i_mount, &xhead, recs);
1552 		switch (error) {
1553 		case 0:
1554 			/*
1555 			 * There are no more records in the result set.  Copy
1556 			 * whatever we got to userspace and break out.
1557 			 */
1558 			done = true;
1559 			break;
1560 		case -ECANCELED:
1561 			/*
1562 			 * The internal memory buffer is full.  Copy whatever
1563 			 * records we got to userspace and go again if we have
1564 			 * not yet filled the userspace buffer.
1565 			 */
1566 			error = 0;
1567 			break;
1568 		default:
1569 			goto out_free;
1570 		}
1571 		head.fmh_entries += xhead.fmh_entries;
1572 		head.fmh_oflags = xhead.fmh_oflags;
1573 
1574 		/*
1575 		 * If the caller wanted a record count or there aren't any
1576 		 * new records to return, we're done.
1577 		 */
1578 		if (head.fmh_count == 0 || xhead.fmh_entries == 0)
1579 			break;
1580 
1581 		/* Copy all the records we got out to userspace. */
1582 		if (copy_to_user(user_recs, recs,
1583 				 xhead.fmh_entries * sizeof(struct fsmap))) {
1584 			error = -EFAULT;
1585 			goto out_free;
1586 		}
1587 
1588 		/* Remember the last record flags we copied to userspace. */
1589 		last_rec = &recs[xhead.fmh_entries - 1];
1590 		last_flags = last_rec->fmr_flags;
1591 
1592 		/* Set up the low key for the next iteration. */
1593 		xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec);
1594 		trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1595 	} while (!done && head.fmh_entries < head.fmh_count);
1596 
1597 	/*
1598 	 * If there are no more records in the query result set and we're not
1599 	 * in counting mode, mark the last record returned with the LAST flag.
1600 	 */
1601 	if (done && head.fmh_count > 0 && head.fmh_entries > 0) {
1602 		struct fsmap __user	*user_rec;
1603 
1604 		last_flags |= FMR_OF_LAST;
1605 		user_rec = &arg->fmh_recs[head.fmh_entries - 1];
1606 
1607 		if (copy_to_user(&user_rec->fmr_flags, &last_flags,
1608 					sizeof(last_flags))) {
1609 			error = -EFAULT;
1610 			goto out_free;
1611 		}
1612 	}
1613 
1614 	/* copy back header */
1615 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) {
1616 		error = -EFAULT;
1617 		goto out_free;
1618 	}
1619 
1620 out_free:
1621 	kmem_free(recs);
1622 	return error;
1623 }
1624 
1625 STATIC int
1626 xfs_ioc_scrub_metadata(
1627 	struct file			*file,
1628 	void				__user *arg)
1629 {
1630 	struct xfs_scrub_metadata	scrub;
1631 	int				error;
1632 
1633 	if (!capable(CAP_SYS_ADMIN))
1634 		return -EPERM;
1635 
1636 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1637 		return -EFAULT;
1638 
1639 	error = xfs_scrub_metadata(file, &scrub);
1640 	if (error)
1641 		return error;
1642 
1643 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1644 		return -EFAULT;
1645 
1646 	return 0;
1647 }
1648 
1649 int
1650 xfs_ioc_swapext(
1651 	xfs_swapext_t	*sxp)
1652 {
1653 	xfs_inode_t     *ip, *tip;
1654 	struct fd	f, tmp;
1655 	int		error = 0;
1656 
1657 	/* Pull information for the target fd */
1658 	f = fdget((int)sxp->sx_fdtarget);
1659 	if (!f.file) {
1660 		error = -EINVAL;
1661 		goto out;
1662 	}
1663 
1664 	if (!(f.file->f_mode & FMODE_WRITE) ||
1665 	    !(f.file->f_mode & FMODE_READ) ||
1666 	    (f.file->f_flags & O_APPEND)) {
1667 		error = -EBADF;
1668 		goto out_put_file;
1669 	}
1670 
1671 	tmp = fdget((int)sxp->sx_fdtmp);
1672 	if (!tmp.file) {
1673 		error = -EINVAL;
1674 		goto out_put_file;
1675 	}
1676 
1677 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1678 	    !(tmp.file->f_mode & FMODE_READ) ||
1679 	    (tmp.file->f_flags & O_APPEND)) {
1680 		error = -EBADF;
1681 		goto out_put_tmp_file;
1682 	}
1683 
1684 	if (IS_SWAPFILE(file_inode(f.file)) ||
1685 	    IS_SWAPFILE(file_inode(tmp.file))) {
1686 		error = -EINVAL;
1687 		goto out_put_tmp_file;
1688 	}
1689 
1690 	/*
1691 	 * We need to ensure that the fds passed in point to XFS inodes
1692 	 * before we cast and access them as XFS structures as we have no
1693 	 * control over what the user passes us here.
1694 	 */
1695 	if (f.file->f_op != &xfs_file_operations ||
1696 	    tmp.file->f_op != &xfs_file_operations) {
1697 		error = -EINVAL;
1698 		goto out_put_tmp_file;
1699 	}
1700 
1701 	ip = XFS_I(file_inode(f.file));
1702 	tip = XFS_I(file_inode(tmp.file));
1703 
1704 	if (ip->i_mount != tip->i_mount) {
1705 		error = -EINVAL;
1706 		goto out_put_tmp_file;
1707 	}
1708 
1709 	if (ip->i_ino == tip->i_ino) {
1710 		error = -EINVAL;
1711 		goto out_put_tmp_file;
1712 	}
1713 
1714 	if (xfs_is_shutdown(ip->i_mount)) {
1715 		error = -EIO;
1716 		goto out_put_tmp_file;
1717 	}
1718 
1719 	error = xfs_swap_extents(ip, tip, sxp);
1720 
1721  out_put_tmp_file:
1722 	fdput(tmp);
1723  out_put_file:
1724 	fdput(f);
1725  out:
1726 	return error;
1727 }
1728 
1729 static int
1730 xfs_ioc_getlabel(
1731 	struct xfs_mount	*mp,
1732 	char			__user *user_label)
1733 {
1734 	struct xfs_sb		*sbp = &mp->m_sb;
1735 	char			label[XFSLABEL_MAX + 1];
1736 
1737 	/* Paranoia */
1738 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
1739 
1740 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
1741 	memset(label, 0, sizeof(label));
1742 	spin_lock(&mp->m_sb_lock);
1743 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
1744 	spin_unlock(&mp->m_sb_lock);
1745 
1746 	if (copy_to_user(user_label, label, sizeof(label)))
1747 		return -EFAULT;
1748 	return 0;
1749 }
1750 
1751 static int
1752 xfs_ioc_setlabel(
1753 	struct file		*filp,
1754 	struct xfs_mount	*mp,
1755 	char			__user *newlabel)
1756 {
1757 	struct xfs_sb		*sbp = &mp->m_sb;
1758 	char			label[XFSLABEL_MAX + 1];
1759 	size_t			len;
1760 	int			error;
1761 
1762 	if (!capable(CAP_SYS_ADMIN))
1763 		return -EPERM;
1764 	/*
1765 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
1766 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
1767 	 * (required) NULL character to test the incoming label length.
1768 	 * NB: The on disk label doesn't need to be null terminated.
1769 	 */
1770 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
1771 		return -EFAULT;
1772 	len = strnlen(label, XFSLABEL_MAX + 1);
1773 	if (len > sizeof(sbp->sb_fname))
1774 		return -EINVAL;
1775 
1776 	error = mnt_want_write_file(filp);
1777 	if (error)
1778 		return error;
1779 
1780 	spin_lock(&mp->m_sb_lock);
1781 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
1782 	memcpy(sbp->sb_fname, label, len);
1783 	spin_unlock(&mp->m_sb_lock);
1784 
1785 	/*
1786 	 * Now we do several things to satisfy userspace.
1787 	 * In addition to normal logging of the primary superblock, we also
1788 	 * immediately write these changes to sector zero for the primary, then
1789 	 * update all backup supers (as xfs_db does for a label change), then
1790 	 * invalidate the block device page cache.  This is so that any prior
1791 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
1792 	 * and userspace will see the newly-written label.
1793 	 */
1794 	error = xfs_sync_sb_buf(mp);
1795 	if (error)
1796 		goto out;
1797 	/*
1798 	 * growfs also updates backup supers so lock against that.
1799 	 */
1800 	mutex_lock(&mp->m_growlock);
1801 	error = xfs_update_secondary_sbs(mp);
1802 	mutex_unlock(&mp->m_growlock);
1803 
1804 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
1805 
1806 out:
1807 	mnt_drop_write_file(filp);
1808 	return error;
1809 }
1810 
1811 static inline int
1812 xfs_fs_eofblocks_from_user(
1813 	struct xfs_fs_eofblocks		*src,
1814 	struct xfs_icwalk		*dst)
1815 {
1816 	if (src->eof_version != XFS_EOFBLOCKS_VERSION)
1817 		return -EINVAL;
1818 
1819 	if (src->eof_flags & ~XFS_EOF_FLAGS_VALID)
1820 		return -EINVAL;
1821 
1822 	if (memchr_inv(&src->pad32, 0, sizeof(src->pad32)) ||
1823 	    memchr_inv(src->pad64, 0, sizeof(src->pad64)))
1824 		return -EINVAL;
1825 
1826 	dst->icw_flags = 0;
1827 	if (src->eof_flags & XFS_EOF_FLAGS_SYNC)
1828 		dst->icw_flags |= XFS_ICWALK_FLAG_SYNC;
1829 	if (src->eof_flags & XFS_EOF_FLAGS_UID)
1830 		dst->icw_flags |= XFS_ICWALK_FLAG_UID;
1831 	if (src->eof_flags & XFS_EOF_FLAGS_GID)
1832 		dst->icw_flags |= XFS_ICWALK_FLAG_GID;
1833 	if (src->eof_flags & XFS_EOF_FLAGS_PRID)
1834 		dst->icw_flags |= XFS_ICWALK_FLAG_PRID;
1835 	if (src->eof_flags & XFS_EOF_FLAGS_MINFILESIZE)
1836 		dst->icw_flags |= XFS_ICWALK_FLAG_MINFILESIZE;
1837 
1838 	dst->icw_prid = src->eof_prid;
1839 	dst->icw_min_file_size = src->eof_min_file_size;
1840 
1841 	dst->icw_uid = INVALID_UID;
1842 	if (src->eof_flags & XFS_EOF_FLAGS_UID) {
1843 		dst->icw_uid = make_kuid(current_user_ns(), src->eof_uid);
1844 		if (!uid_valid(dst->icw_uid))
1845 			return -EINVAL;
1846 	}
1847 
1848 	dst->icw_gid = INVALID_GID;
1849 	if (src->eof_flags & XFS_EOF_FLAGS_GID) {
1850 		dst->icw_gid = make_kgid(current_user_ns(), src->eof_gid);
1851 		if (!gid_valid(dst->icw_gid))
1852 			return -EINVAL;
1853 	}
1854 	return 0;
1855 }
1856 
1857 /*
1858  * These long-unused ioctls were removed from the official ioctl API in 5.17,
1859  * but retain these definitions so that we can log warnings about them.
1860  */
1861 #define XFS_IOC_ALLOCSP		_IOW ('X', 10, struct xfs_flock64)
1862 #define XFS_IOC_FREESP		_IOW ('X', 11, struct xfs_flock64)
1863 #define XFS_IOC_ALLOCSP64	_IOW ('X', 36, struct xfs_flock64)
1864 #define XFS_IOC_FREESP64	_IOW ('X', 37, struct xfs_flock64)
1865 
1866 /*
1867  * Note: some of the ioctl's return positive numbers as a
1868  * byte count indicating success, such as readlink_by_handle.
1869  * So we don't "sign flip" like most other routines.  This means
1870  * true errors need to be returned as a negative value.
1871  */
1872 long
1873 xfs_file_ioctl(
1874 	struct file		*filp,
1875 	unsigned int		cmd,
1876 	unsigned long		p)
1877 {
1878 	struct inode		*inode = file_inode(filp);
1879 	struct xfs_inode	*ip = XFS_I(inode);
1880 	struct xfs_mount	*mp = ip->i_mount;
1881 	void			__user *arg = (void __user *)p;
1882 	int			error;
1883 
1884 	trace_xfs_file_ioctl(ip);
1885 
1886 	switch (cmd) {
1887 	case FITRIM:
1888 		return xfs_ioc_trim(mp, arg);
1889 	case FS_IOC_GETFSLABEL:
1890 		return xfs_ioc_getlabel(mp, arg);
1891 	case FS_IOC_SETFSLABEL:
1892 		return xfs_ioc_setlabel(filp, mp, arg);
1893 	case XFS_IOC_ALLOCSP:
1894 	case XFS_IOC_FREESP:
1895 	case XFS_IOC_ALLOCSP64:
1896 	case XFS_IOC_FREESP64:
1897 		xfs_warn_once(mp,
1898 	"%s should use fallocate; XFS_IOC_{ALLOC,FREE}SP ioctl unsupported",
1899 				current->comm);
1900 		return -ENOTTY;
1901 	case XFS_IOC_DIOINFO: {
1902 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
1903 		struct dioattr		da;
1904 
1905 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
1906 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
1907 
1908 		if (copy_to_user(arg, &da, sizeof(da)))
1909 			return -EFAULT;
1910 		return 0;
1911 	}
1912 
1913 	case XFS_IOC_FSBULKSTAT_SINGLE:
1914 	case XFS_IOC_FSBULKSTAT:
1915 	case XFS_IOC_FSINUMBERS:
1916 		return xfs_ioc_fsbulkstat(filp, cmd, arg);
1917 
1918 	case XFS_IOC_BULKSTAT:
1919 		return xfs_ioc_bulkstat(filp, cmd, arg);
1920 	case XFS_IOC_INUMBERS:
1921 		return xfs_ioc_inumbers(mp, cmd, arg);
1922 
1923 	case XFS_IOC_FSGEOMETRY_V1:
1924 		return xfs_ioc_fsgeometry(mp, arg, 3);
1925 	case XFS_IOC_FSGEOMETRY_V4:
1926 		return xfs_ioc_fsgeometry(mp, arg, 4);
1927 	case XFS_IOC_FSGEOMETRY:
1928 		return xfs_ioc_fsgeometry(mp, arg, 5);
1929 
1930 	case XFS_IOC_AG_GEOMETRY:
1931 		return xfs_ioc_ag_geometry(mp, arg);
1932 
1933 	case XFS_IOC_GETVERSION:
1934 		return put_user(inode->i_generation, (int __user *)arg);
1935 
1936 	case XFS_IOC_FSGETXATTRA:
1937 		return xfs_ioc_fsgetxattra(ip, arg);
1938 
1939 	case XFS_IOC_GETBMAP:
1940 	case XFS_IOC_GETBMAPA:
1941 	case XFS_IOC_GETBMAPX:
1942 		return xfs_ioc_getbmap(filp, cmd, arg);
1943 
1944 	case FS_IOC_GETFSMAP:
1945 		return xfs_ioc_getfsmap(ip, arg);
1946 
1947 	case XFS_IOC_SCRUB_METADATA:
1948 		return xfs_ioc_scrub_metadata(filp, arg);
1949 
1950 	case XFS_IOC_FD_TO_HANDLE:
1951 	case XFS_IOC_PATH_TO_HANDLE:
1952 	case XFS_IOC_PATH_TO_FSHANDLE: {
1953 		xfs_fsop_handlereq_t	hreq;
1954 
1955 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
1956 			return -EFAULT;
1957 		return xfs_find_handle(cmd, &hreq);
1958 	}
1959 	case XFS_IOC_OPEN_BY_HANDLE: {
1960 		xfs_fsop_handlereq_t	hreq;
1961 
1962 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
1963 			return -EFAULT;
1964 		return xfs_open_by_handle(filp, &hreq);
1965 	}
1966 
1967 	case XFS_IOC_READLINK_BY_HANDLE: {
1968 		xfs_fsop_handlereq_t	hreq;
1969 
1970 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
1971 			return -EFAULT;
1972 		return xfs_readlink_by_handle(filp, &hreq);
1973 	}
1974 	case XFS_IOC_ATTRLIST_BY_HANDLE:
1975 		return xfs_attrlist_by_handle(filp, arg);
1976 
1977 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
1978 		return xfs_attrmulti_by_handle(filp, arg);
1979 
1980 	case XFS_IOC_SWAPEXT: {
1981 		struct xfs_swapext	sxp;
1982 
1983 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
1984 			return -EFAULT;
1985 		error = mnt_want_write_file(filp);
1986 		if (error)
1987 			return error;
1988 		error = xfs_ioc_swapext(&sxp);
1989 		mnt_drop_write_file(filp);
1990 		return error;
1991 	}
1992 
1993 	case XFS_IOC_FSCOUNTS: {
1994 		xfs_fsop_counts_t out;
1995 
1996 		xfs_fs_counts(mp, &out);
1997 
1998 		if (copy_to_user(arg, &out, sizeof(out)))
1999 			return -EFAULT;
2000 		return 0;
2001 	}
2002 
2003 	case XFS_IOC_SET_RESBLKS: {
2004 		xfs_fsop_resblks_t inout;
2005 		uint64_t	   in;
2006 
2007 		if (!capable(CAP_SYS_ADMIN))
2008 			return -EPERM;
2009 
2010 		if (xfs_is_readonly(mp))
2011 			return -EROFS;
2012 
2013 		if (copy_from_user(&inout, arg, sizeof(inout)))
2014 			return -EFAULT;
2015 
2016 		error = mnt_want_write_file(filp);
2017 		if (error)
2018 			return error;
2019 
2020 		/* input parameter is passed in resblks field of structure */
2021 		in = inout.resblks;
2022 		error = xfs_reserve_blocks(mp, &in, &inout);
2023 		mnt_drop_write_file(filp);
2024 		if (error)
2025 			return error;
2026 
2027 		if (copy_to_user(arg, &inout, sizeof(inout)))
2028 			return -EFAULT;
2029 		return 0;
2030 	}
2031 
2032 	case XFS_IOC_GET_RESBLKS: {
2033 		xfs_fsop_resblks_t out;
2034 
2035 		if (!capable(CAP_SYS_ADMIN))
2036 			return -EPERM;
2037 
2038 		error = xfs_reserve_blocks(mp, NULL, &out);
2039 		if (error)
2040 			return error;
2041 
2042 		if (copy_to_user(arg, &out, sizeof(out)))
2043 			return -EFAULT;
2044 
2045 		return 0;
2046 	}
2047 
2048 	case XFS_IOC_FSGROWFSDATA: {
2049 		struct xfs_growfs_data in;
2050 
2051 		if (copy_from_user(&in, arg, sizeof(in)))
2052 			return -EFAULT;
2053 
2054 		error = mnt_want_write_file(filp);
2055 		if (error)
2056 			return error;
2057 		error = xfs_growfs_data(mp, &in);
2058 		mnt_drop_write_file(filp);
2059 		return error;
2060 	}
2061 
2062 	case XFS_IOC_FSGROWFSLOG: {
2063 		struct xfs_growfs_log in;
2064 
2065 		if (copy_from_user(&in, arg, sizeof(in)))
2066 			return -EFAULT;
2067 
2068 		error = mnt_want_write_file(filp);
2069 		if (error)
2070 			return error;
2071 		error = xfs_growfs_log(mp, &in);
2072 		mnt_drop_write_file(filp);
2073 		return error;
2074 	}
2075 
2076 	case XFS_IOC_FSGROWFSRT: {
2077 		xfs_growfs_rt_t in;
2078 
2079 		if (copy_from_user(&in, arg, sizeof(in)))
2080 			return -EFAULT;
2081 
2082 		error = mnt_want_write_file(filp);
2083 		if (error)
2084 			return error;
2085 		error = xfs_growfs_rt(mp, &in);
2086 		mnt_drop_write_file(filp);
2087 		return error;
2088 	}
2089 
2090 	case XFS_IOC_GOINGDOWN: {
2091 		uint32_t in;
2092 
2093 		if (!capable(CAP_SYS_ADMIN))
2094 			return -EPERM;
2095 
2096 		if (get_user(in, (uint32_t __user *)arg))
2097 			return -EFAULT;
2098 
2099 		return xfs_fs_goingdown(mp, in);
2100 	}
2101 
2102 	case XFS_IOC_ERROR_INJECTION: {
2103 		xfs_error_injection_t in;
2104 
2105 		if (!capable(CAP_SYS_ADMIN))
2106 			return -EPERM;
2107 
2108 		if (copy_from_user(&in, arg, sizeof(in)))
2109 			return -EFAULT;
2110 
2111 		return xfs_errortag_add(mp, in.errtag);
2112 	}
2113 
2114 	case XFS_IOC_ERROR_CLEARALL:
2115 		if (!capable(CAP_SYS_ADMIN))
2116 			return -EPERM;
2117 
2118 		return xfs_errortag_clearall(mp);
2119 
2120 	case XFS_IOC_FREE_EOFBLOCKS: {
2121 		struct xfs_fs_eofblocks	eofb;
2122 		struct xfs_icwalk	icw;
2123 
2124 		if (!capable(CAP_SYS_ADMIN))
2125 			return -EPERM;
2126 
2127 		if (xfs_is_readonly(mp))
2128 			return -EROFS;
2129 
2130 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2131 			return -EFAULT;
2132 
2133 		error = xfs_fs_eofblocks_from_user(&eofb, &icw);
2134 		if (error)
2135 			return error;
2136 
2137 		trace_xfs_ioc_free_eofblocks(mp, &icw, _RET_IP_);
2138 
2139 		sb_start_write(mp->m_super);
2140 		error = xfs_blockgc_free_space(mp, &icw);
2141 		sb_end_write(mp->m_super);
2142 		return error;
2143 	}
2144 
2145 	default:
2146 		return -ENOTTY;
2147 	}
2148 }
2149