xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 40ddb537)
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 = kvzalloc(bufsize, GFP_KERNEL);
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_mnt_user_ns(filp), file_dentry(filp),
697 				    &iattr);
698 	if (error)
699 		goto out_unlock;
700 
701 	error = xfs_update_prealloc_flags(ip, flags);
702 
703 out_unlock:
704 	xfs_iunlock(ip, iolock);
705 	mnt_drop_write_file(filp);
706 	return error;
707 }
708 
709 /* Return 0 on success or positive error */
710 int
711 xfs_fsbulkstat_one_fmt(
712 	struct xfs_ibulk		*breq,
713 	const struct xfs_bulkstat	*bstat)
714 {
715 	struct xfs_bstat		bs1;
716 
717 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
718 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
719 		return -EFAULT;
720 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
721 }
722 
723 int
724 xfs_fsinumbers_fmt(
725 	struct xfs_ibulk		*breq,
726 	const struct xfs_inumbers	*igrp)
727 {
728 	struct xfs_inogrp		ig1;
729 
730 	xfs_inumbers_to_inogrp(&ig1, igrp);
731 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
732 		return -EFAULT;
733 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
734 }
735 
736 STATIC int
737 xfs_ioc_fsbulkstat(
738 	struct file		*file,
739 	unsigned int		cmd,
740 	void			__user *arg)
741 {
742 	struct xfs_mount	*mp = XFS_I(file_inode(file))->i_mount;
743 	struct xfs_fsop_bulkreq	bulkreq;
744 	struct xfs_ibulk	breq = {
745 		.mp		= mp,
746 		.mnt_userns	= file_mnt_user_ns(file),
747 		.ocount		= 0,
748 	};
749 	xfs_ino_t		lastino;
750 	int			error;
751 
752 	/* done = 1 if there are more stats to get and if bulkstat */
753 	/* should be called again (unused here, but used in dmapi) */
754 
755 	if (!capable(CAP_SYS_ADMIN))
756 		return -EPERM;
757 
758 	if (XFS_FORCED_SHUTDOWN(mp))
759 		return -EIO;
760 
761 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
762 		return -EFAULT;
763 
764 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
765 		return -EFAULT;
766 
767 	if (bulkreq.icount <= 0)
768 		return -EINVAL;
769 
770 	if (bulkreq.ubuffer == NULL)
771 		return -EINVAL;
772 
773 	breq.ubuffer = bulkreq.ubuffer;
774 	breq.icount = bulkreq.icount;
775 
776 	/*
777 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
778 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
779 	 * that *lastip contains either zero or the number of the last inode to
780 	 * be examined by the previous call and return results starting with
781 	 * the next inode after that.  The new bulk request back end functions
782 	 * take the inode to start with, so we have to compute the startino
783 	 * parameter from lastino to maintain correct function.  lastino == 0
784 	 * is a special case because it has traditionally meant "first inode
785 	 * in filesystem".
786 	 */
787 	if (cmd == XFS_IOC_FSINUMBERS) {
788 		breq.startino = lastino ? lastino + 1 : 0;
789 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
790 		lastino = breq.startino - 1;
791 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
792 		breq.startino = lastino;
793 		breq.icount = 1;
794 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
795 	} else {	/* XFS_IOC_FSBULKSTAT */
796 		breq.startino = lastino ? lastino + 1 : 0;
797 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
798 		lastino = breq.startino - 1;
799 	}
800 
801 	if (error)
802 		return error;
803 
804 	if (bulkreq.lastip != NULL &&
805 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
806 		return -EFAULT;
807 
808 	if (bulkreq.ocount != NULL &&
809 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
810 		return -EFAULT;
811 
812 	return 0;
813 }
814 
815 /* Return 0 on success or positive error */
816 static int
817 xfs_bulkstat_fmt(
818 	struct xfs_ibulk		*breq,
819 	const struct xfs_bulkstat	*bstat)
820 {
821 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
822 		return -EFAULT;
823 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
824 }
825 
826 /*
827  * Check the incoming bulk request @hdr from userspace and initialize the
828  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
829  * should proceed; -ECANCELED if there's nothing to do; or the usual
830  * negative error code.
831  */
832 static int
833 xfs_bulk_ireq_setup(
834 	struct xfs_mount	*mp,
835 	struct xfs_bulk_ireq	*hdr,
836 	struct xfs_ibulk	*breq,
837 	void __user		*ubuffer)
838 {
839 	if (hdr->icount == 0 ||
840 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
841 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
842 		return -EINVAL;
843 
844 	breq->startino = hdr->ino;
845 	breq->ubuffer = ubuffer;
846 	breq->icount = hdr->icount;
847 	breq->ocount = 0;
848 	breq->flags = 0;
849 
850 	/*
851 	 * The @ino parameter is a special value, so we must look it up here.
852 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
853 	 * worth of data.
854 	 */
855 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
856 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
857 			return -EINVAL;
858 
859 		switch (hdr->ino) {
860 		case XFS_BULK_IREQ_SPECIAL_ROOT:
861 			hdr->ino = mp->m_sb.sb_rootino;
862 			break;
863 		default:
864 			return -EINVAL;
865 		}
866 		breq->icount = 1;
867 	}
868 
869 	/*
870 	 * The IREQ_AGNO flag means that we only want results from a given AG.
871 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
872 	 * beyond the specified AG then we return no results.
873 	 */
874 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
875 		if (hdr->agno >= mp->m_sb.sb_agcount)
876 			return -EINVAL;
877 
878 		if (breq->startino == 0)
879 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
880 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
881 			return -EINVAL;
882 
883 		breq->flags |= XFS_IBULK_SAME_AG;
884 
885 		/* Asking for an inode past the end of the AG?  We're done! */
886 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
887 			return -ECANCELED;
888 	} else if (hdr->agno)
889 		return -EINVAL;
890 
891 	/* Asking for an inode past the end of the FS?  We're done! */
892 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
893 		return -ECANCELED;
894 
895 	return 0;
896 }
897 
898 /*
899  * Update the userspace bulk request @hdr to reflect the end state of the
900  * internal bulk request @breq.
901  */
902 static void
903 xfs_bulk_ireq_teardown(
904 	struct xfs_bulk_ireq	*hdr,
905 	struct xfs_ibulk	*breq)
906 {
907 	hdr->ino = breq->startino;
908 	hdr->ocount = breq->ocount;
909 }
910 
911 /* Handle the v5 bulkstat ioctl. */
912 STATIC int
913 xfs_ioc_bulkstat(
914 	struct file			*file,
915 	unsigned int			cmd,
916 	struct xfs_bulkstat_req __user	*arg)
917 {
918 	struct xfs_mount		*mp = XFS_I(file_inode(file))->i_mount;
919 	struct xfs_bulk_ireq		hdr;
920 	struct xfs_ibulk		breq = {
921 		.mp			= mp,
922 		.mnt_userns		= file_mnt_user_ns(file),
923 	};
924 	int				error;
925 
926 	if (!capable(CAP_SYS_ADMIN))
927 		return -EPERM;
928 
929 	if (XFS_FORCED_SHUTDOWN(mp))
930 		return -EIO;
931 
932 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
933 		return -EFAULT;
934 
935 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
936 	if (error == -ECANCELED)
937 		goto out_teardown;
938 	if (error < 0)
939 		return error;
940 
941 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
942 	if (error)
943 		return error;
944 
945 out_teardown:
946 	xfs_bulk_ireq_teardown(&hdr, &breq);
947 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
948 		return -EFAULT;
949 
950 	return 0;
951 }
952 
953 STATIC int
954 xfs_inumbers_fmt(
955 	struct xfs_ibulk		*breq,
956 	const struct xfs_inumbers	*igrp)
957 {
958 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
959 		return -EFAULT;
960 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
961 }
962 
963 /* Handle the v5 inumbers ioctl. */
964 STATIC int
965 xfs_ioc_inumbers(
966 	struct xfs_mount		*mp,
967 	unsigned int			cmd,
968 	struct xfs_inumbers_req __user	*arg)
969 {
970 	struct xfs_bulk_ireq		hdr;
971 	struct xfs_ibulk		breq = {
972 		.mp			= mp,
973 	};
974 	int				error;
975 
976 	if (!capable(CAP_SYS_ADMIN))
977 		return -EPERM;
978 
979 	if (XFS_FORCED_SHUTDOWN(mp))
980 		return -EIO;
981 
982 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
983 		return -EFAULT;
984 
985 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
986 	if (error == -ECANCELED)
987 		goto out_teardown;
988 	if (error < 0)
989 		return error;
990 
991 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
992 	if (error)
993 		return error;
994 
995 out_teardown:
996 	xfs_bulk_ireq_teardown(&hdr, &breq);
997 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
998 		return -EFAULT;
999 
1000 	return 0;
1001 }
1002 
1003 STATIC int
1004 xfs_ioc_fsgeometry(
1005 	struct xfs_mount	*mp,
1006 	void			__user *arg,
1007 	int			struct_version)
1008 {
1009 	struct xfs_fsop_geom	fsgeo;
1010 	size_t			len;
1011 
1012 	xfs_fs_geometry(&mp->m_sb, &fsgeo, struct_version);
1013 
1014 	if (struct_version <= 3)
1015 		len = sizeof(struct xfs_fsop_geom_v1);
1016 	else if (struct_version == 4)
1017 		len = sizeof(struct xfs_fsop_geom_v4);
1018 	else {
1019 		xfs_fsop_geom_health(mp, &fsgeo);
1020 		len = sizeof(fsgeo);
1021 	}
1022 
1023 	if (copy_to_user(arg, &fsgeo, len))
1024 		return -EFAULT;
1025 	return 0;
1026 }
1027 
1028 STATIC int
1029 xfs_ioc_ag_geometry(
1030 	struct xfs_mount	*mp,
1031 	void			__user *arg)
1032 {
1033 	struct xfs_ag_geometry	ageo;
1034 	int			error;
1035 
1036 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
1037 		return -EFAULT;
1038 	if (ageo.ag_flags)
1039 		return -EINVAL;
1040 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
1041 		return -EINVAL;
1042 
1043 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
1044 	if (error)
1045 		return error;
1046 
1047 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
1048 		return -EFAULT;
1049 	return 0;
1050 }
1051 
1052 /*
1053  * Linux extended inode flags interface.
1054  */
1055 
1056 STATIC unsigned int
1057 xfs_merge_ioc_xflags(
1058 	unsigned int	flags,
1059 	unsigned int	start)
1060 {
1061 	unsigned int	xflags = start;
1062 
1063 	if (flags & FS_IMMUTABLE_FL)
1064 		xflags |= FS_XFLAG_IMMUTABLE;
1065 	else
1066 		xflags &= ~FS_XFLAG_IMMUTABLE;
1067 	if (flags & FS_APPEND_FL)
1068 		xflags |= FS_XFLAG_APPEND;
1069 	else
1070 		xflags &= ~FS_XFLAG_APPEND;
1071 	if (flags & FS_SYNC_FL)
1072 		xflags |= FS_XFLAG_SYNC;
1073 	else
1074 		xflags &= ~FS_XFLAG_SYNC;
1075 	if (flags & FS_NOATIME_FL)
1076 		xflags |= FS_XFLAG_NOATIME;
1077 	else
1078 		xflags &= ~FS_XFLAG_NOATIME;
1079 	if (flags & FS_NODUMP_FL)
1080 		xflags |= FS_XFLAG_NODUMP;
1081 	else
1082 		xflags &= ~FS_XFLAG_NODUMP;
1083 	if (flags & FS_DAX_FL)
1084 		xflags |= FS_XFLAG_DAX;
1085 	else
1086 		xflags &= ~FS_XFLAG_DAX;
1087 
1088 	return xflags;
1089 }
1090 
1091 STATIC unsigned int
1092 xfs_di2lxflags(
1093 	uint16_t	di_flags,
1094 	uint64_t	di_flags2)
1095 {
1096 	unsigned int	flags = 0;
1097 
1098 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
1099 		flags |= FS_IMMUTABLE_FL;
1100 	if (di_flags & XFS_DIFLAG_APPEND)
1101 		flags |= FS_APPEND_FL;
1102 	if (di_flags & XFS_DIFLAG_SYNC)
1103 		flags |= FS_SYNC_FL;
1104 	if (di_flags & XFS_DIFLAG_NOATIME)
1105 		flags |= FS_NOATIME_FL;
1106 	if (di_flags & XFS_DIFLAG_NODUMP)
1107 		flags |= FS_NODUMP_FL;
1108 	if (di_flags2 & XFS_DIFLAG2_DAX) {
1109 		flags |= FS_DAX_FL;
1110 	}
1111 	return flags;
1112 }
1113 
1114 static void
1115 xfs_fill_fsxattr(
1116 	struct xfs_inode	*ip,
1117 	bool			attr,
1118 	struct fsxattr		*fa)
1119 {
1120 	struct xfs_ifork	*ifp = attr ? ip->i_afp : &ip->i_df;
1121 
1122 	simple_fill_fsxattr(fa, xfs_ip2xflags(ip));
1123 	fa->fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
1124 	fa->fsx_cowextsize = ip->i_d.di_cowextsize <<
1125 			ip->i_mount->m_sb.sb_blocklog;
1126 	fa->fsx_projid = ip->i_d.di_projid;
1127 	if (ifp && (ifp->if_flags & XFS_IFEXTENTS))
1128 		fa->fsx_nextents = xfs_iext_count(ifp);
1129 	else
1130 		fa->fsx_nextents = xfs_ifork_nextents(ifp);
1131 }
1132 
1133 STATIC int
1134 xfs_ioc_fsgetxattr(
1135 	xfs_inode_t		*ip,
1136 	int			attr,
1137 	void			__user *arg)
1138 {
1139 	struct fsxattr		fa;
1140 
1141 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1142 	xfs_fill_fsxattr(ip, attr, &fa);
1143 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1144 
1145 	if (copy_to_user(arg, &fa, sizeof(fa)))
1146 		return -EFAULT;
1147 	return 0;
1148 }
1149 
1150 STATIC uint16_t
1151 xfs_flags2diflags(
1152 	struct xfs_inode	*ip,
1153 	unsigned int		xflags)
1154 {
1155 	/* can't set PREALLOC this way, just preserve it */
1156 	uint16_t		di_flags =
1157 		(ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
1158 
1159 	if (xflags & FS_XFLAG_IMMUTABLE)
1160 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1161 	if (xflags & FS_XFLAG_APPEND)
1162 		di_flags |= XFS_DIFLAG_APPEND;
1163 	if (xflags & FS_XFLAG_SYNC)
1164 		di_flags |= XFS_DIFLAG_SYNC;
1165 	if (xflags & FS_XFLAG_NOATIME)
1166 		di_flags |= XFS_DIFLAG_NOATIME;
1167 	if (xflags & FS_XFLAG_NODUMP)
1168 		di_flags |= XFS_DIFLAG_NODUMP;
1169 	if (xflags & FS_XFLAG_NODEFRAG)
1170 		di_flags |= XFS_DIFLAG_NODEFRAG;
1171 	if (xflags & FS_XFLAG_FILESTREAM)
1172 		di_flags |= XFS_DIFLAG_FILESTREAM;
1173 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1174 		if (xflags & FS_XFLAG_RTINHERIT)
1175 			di_flags |= XFS_DIFLAG_RTINHERIT;
1176 		if (xflags & FS_XFLAG_NOSYMLINKS)
1177 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1178 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1179 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1180 		if (xflags & FS_XFLAG_PROJINHERIT)
1181 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1182 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1183 		if (xflags & FS_XFLAG_REALTIME)
1184 			di_flags |= XFS_DIFLAG_REALTIME;
1185 		if (xflags & FS_XFLAG_EXTSIZE)
1186 			di_flags |= XFS_DIFLAG_EXTSIZE;
1187 	}
1188 
1189 	return di_flags;
1190 }
1191 
1192 STATIC uint64_t
1193 xfs_flags2diflags2(
1194 	struct xfs_inode	*ip,
1195 	unsigned int		xflags)
1196 {
1197 	uint64_t		di_flags2 =
1198 		(ip->i_d.di_flags2 & (XFS_DIFLAG2_REFLINK |
1199 				      XFS_DIFLAG2_BIGTIME));
1200 
1201 	if (xflags & FS_XFLAG_DAX)
1202 		di_flags2 |= XFS_DIFLAG2_DAX;
1203 	if (xflags & FS_XFLAG_COWEXTSIZE)
1204 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1205 
1206 	return di_flags2;
1207 }
1208 
1209 static int
1210 xfs_ioctl_setattr_xflags(
1211 	struct xfs_trans	*tp,
1212 	struct xfs_inode	*ip,
1213 	struct fsxattr		*fa)
1214 {
1215 	struct xfs_mount	*mp = ip->i_mount;
1216 	uint64_t		di_flags2;
1217 
1218 	/* Can't change realtime flag if any extents are allocated. */
1219 	if ((ip->i_df.if_nextents || ip->i_delayed_blks) &&
1220 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1221 		return -EINVAL;
1222 
1223 	/* If realtime flag is set then must have realtime device */
1224 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1225 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1226 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1227 			return -EINVAL;
1228 	}
1229 
1230 	/* Clear reflink if we are actually able to set the rt flag. */
1231 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1232 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1233 
1234 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1235 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1236 		return -EINVAL;
1237 
1238 	/* diflags2 only valid for v3 inodes. */
1239 	di_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1240 	if (di_flags2 && !xfs_sb_version_has_v3inode(&mp->m_sb))
1241 		return -EINVAL;
1242 
1243 	ip->i_d.di_flags = xfs_flags2diflags(ip, fa->fsx_xflags);
1244 	ip->i_d.di_flags2 = di_flags2;
1245 
1246 	xfs_diflags_to_iflags(ip, false);
1247 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1248 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1249 	XFS_STATS_INC(mp, xs_ig_attrchg);
1250 	return 0;
1251 }
1252 
1253 static void
1254 xfs_ioctl_setattr_prepare_dax(
1255 	struct xfs_inode	*ip,
1256 	struct fsxattr		*fa)
1257 {
1258 	struct xfs_mount	*mp = ip->i_mount;
1259 	struct inode            *inode = VFS_I(ip);
1260 
1261 	if (S_ISDIR(inode->i_mode))
1262 		return;
1263 
1264 	if ((mp->m_flags & XFS_MOUNT_DAX_ALWAYS) ||
1265 	    (mp->m_flags & XFS_MOUNT_DAX_NEVER))
1266 		return;
1267 
1268 	if (((fa->fsx_xflags & FS_XFLAG_DAX) &&
1269 	    !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)) ||
1270 	    (!(fa->fsx_xflags & FS_XFLAG_DAX) &&
1271 	     (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)))
1272 		d_mark_dontcache(inode);
1273 }
1274 
1275 /*
1276  * Set up the transaction structure for the setattr operation, checking that we
1277  * have permission to do so. On success, return a clean transaction and the
1278  * inode locked exclusively ready for further operation specific checks. On
1279  * failure, return an error without modifying or locking the inode.
1280  */
1281 static struct xfs_trans *
1282 xfs_ioctl_setattr_get_trans(
1283 	struct file		*file,
1284 	struct xfs_dquot	*pdqp)
1285 {
1286 	struct xfs_inode	*ip = XFS_I(file_inode(file));
1287 	struct xfs_mount	*mp = ip->i_mount;
1288 	struct xfs_trans	*tp;
1289 	int			error = -EROFS;
1290 
1291 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1292 		goto out_error;
1293 	error = -EIO;
1294 	if (XFS_FORCED_SHUTDOWN(mp))
1295 		goto out_error;
1296 
1297 	error = xfs_trans_alloc_ichange(ip, NULL, NULL, pdqp,
1298 			capable(CAP_FOWNER), &tp);
1299 	if (error)
1300 		goto out_error;
1301 
1302 	/*
1303 	 * CAP_FOWNER overrides the following restrictions:
1304 	 *
1305 	 * The user ID of the calling process must be equal to the file owner
1306 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1307 	 */
1308 	if (!inode_owner_or_capable(file_mnt_user_ns(file), VFS_I(ip))) {
1309 		error = -EPERM;
1310 		goto out_cancel;
1311 	}
1312 
1313 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1314 		xfs_trans_set_sync(tp);
1315 
1316 	return tp;
1317 
1318 out_cancel:
1319 	xfs_trans_cancel(tp);
1320 out_error:
1321 	return ERR_PTR(error);
1322 }
1323 
1324 /*
1325  * extent size hint validation is somewhat cumbersome. Rules are:
1326  *
1327  * 1. extent size hint is only valid for directories and regular files
1328  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1329  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1330  * 4. can only be changed on regular files if no extents are allocated
1331  * 5. can be changed on directories at any time
1332  * 6. extsize hint of 0 turns off hints, clears inode flags.
1333  * 7. Extent size must be a multiple of the appropriate block size.
1334  * 8. for non-realtime files, the extent size hint must be limited
1335  *    to half the AG size to avoid alignment extending the extent beyond the
1336  *    limits of the AG.
1337  *
1338  * Please keep this function in sync with xfs_scrub_inode_extsize.
1339  */
1340 static int
1341 xfs_ioctl_setattr_check_extsize(
1342 	struct xfs_inode	*ip,
1343 	struct fsxattr		*fa)
1344 {
1345 	struct xfs_mount	*mp = ip->i_mount;
1346 	xfs_extlen_t		size;
1347 	xfs_fsblock_t		extsize_fsb;
1348 
1349 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_df.if_nextents &&
1350 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1351 		return -EINVAL;
1352 
1353 	if (fa->fsx_extsize == 0)
1354 		return 0;
1355 
1356 	extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1357 	if (extsize_fsb > MAXEXTLEN)
1358 		return -EINVAL;
1359 
1360 	if (XFS_IS_REALTIME_INODE(ip) ||
1361 	    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1362 		size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1363 	} else {
1364 		size = mp->m_sb.sb_blocksize;
1365 		if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1366 			return -EINVAL;
1367 	}
1368 
1369 	if (fa->fsx_extsize % size)
1370 		return -EINVAL;
1371 
1372 	return 0;
1373 }
1374 
1375 /*
1376  * CoW extent size hint validation rules are:
1377  *
1378  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
1379  *    The inode does not have to have any shared blocks, but it must be a v3.
1380  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
1381  *    for a directory, the hint is propagated to new files.
1382  * 3. Can be changed on files & directories at any time.
1383  * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
1384  * 5. Extent size must be a multiple of the appropriate block size.
1385  * 6. The extent size hint must be limited to half the AG size to avoid
1386  *    alignment extending the extent beyond the limits of the AG.
1387  *
1388  * Please keep this function in sync with xfs_scrub_inode_cowextsize.
1389  */
1390 static int
1391 xfs_ioctl_setattr_check_cowextsize(
1392 	struct xfs_inode	*ip,
1393 	struct fsxattr		*fa)
1394 {
1395 	struct xfs_mount	*mp = ip->i_mount;
1396 	xfs_extlen_t		size;
1397 	xfs_fsblock_t		cowextsize_fsb;
1398 
1399 	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
1400 		return 0;
1401 
1402 	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb))
1403 		return -EINVAL;
1404 
1405 	if (fa->fsx_cowextsize == 0)
1406 		return 0;
1407 
1408 	cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1409 	if (cowextsize_fsb > MAXEXTLEN)
1410 		return -EINVAL;
1411 
1412 	size = mp->m_sb.sb_blocksize;
1413 	if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
1414 		return -EINVAL;
1415 
1416 	if (fa->fsx_cowextsize % size)
1417 		return -EINVAL;
1418 
1419 	return 0;
1420 }
1421 
1422 static int
1423 xfs_ioctl_setattr_check_projid(
1424 	struct xfs_inode	*ip,
1425 	struct fsxattr		*fa)
1426 {
1427 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1428 	if (fa->fsx_projid > (uint16_t)-1 &&
1429 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1430 		return -EINVAL;
1431 	return 0;
1432 }
1433 
1434 STATIC int
1435 xfs_ioctl_setattr(
1436 	struct file		*file,
1437 	struct fsxattr		*fa)
1438 {
1439 	struct user_namespace	*mnt_userns = file_mnt_user_ns(file);
1440 	struct xfs_inode	*ip = XFS_I(file_inode(file));
1441 	struct fsxattr		old_fa;
1442 	struct xfs_mount	*mp = ip->i_mount;
1443 	struct xfs_trans	*tp;
1444 	struct xfs_dquot	*pdqp = NULL;
1445 	struct xfs_dquot	*olddquot = NULL;
1446 	int			error;
1447 
1448 	trace_xfs_ioctl_setattr(ip);
1449 
1450 	error = xfs_ioctl_setattr_check_projid(ip, fa);
1451 	if (error)
1452 		return error;
1453 
1454 	/*
1455 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1456 	 * before we start any other transactions. Trying to do this later
1457 	 * is messy. We don't care to take a readlock to look at the ids
1458 	 * in inode here, because we can't hold it across the trans_reserve.
1459 	 * If the IDs do change before we take the ilock, we're covered
1460 	 * because the i_*dquot fields will get updated anyway.
1461 	 */
1462 	if (XFS_IS_QUOTA_ON(mp)) {
1463 		error = xfs_qm_vop_dqalloc(ip, VFS_I(ip)->i_uid,
1464 				VFS_I(ip)->i_gid, fa->fsx_projid,
1465 				XFS_QMOPT_PQUOTA, NULL, NULL, &pdqp);
1466 		if (error)
1467 			return error;
1468 	}
1469 
1470 	xfs_ioctl_setattr_prepare_dax(ip, fa);
1471 
1472 	tp = xfs_ioctl_setattr_get_trans(file, pdqp);
1473 	if (IS_ERR(tp)) {
1474 		error = PTR_ERR(tp);
1475 		goto error_free_dquots;
1476 	}
1477 
1478 	xfs_fill_fsxattr(ip, false, &old_fa);
1479 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, fa);
1480 	if (error)
1481 		goto error_trans_cancel;
1482 
1483 	error = xfs_ioctl_setattr_check_extsize(ip, fa);
1484 	if (error)
1485 		goto error_trans_cancel;
1486 
1487 	error = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1488 	if (error)
1489 		goto error_trans_cancel;
1490 
1491 	error = xfs_ioctl_setattr_xflags(tp, ip, fa);
1492 	if (error)
1493 		goto error_trans_cancel;
1494 
1495 	/*
1496 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1497 	 * overrides the following restrictions:
1498 	 *
1499 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1500 	 * successful return from chown()
1501 	 */
1502 
1503 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1504 	    !capable_wrt_inode_uidgid(mnt_userns, VFS_I(ip), CAP_FSETID))
1505 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1506 
1507 	/* Change the ownerships and register project quota modifications */
1508 	if (ip->i_d.di_projid != fa->fsx_projid) {
1509 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1510 			olddquot = xfs_qm_vop_chown(tp, ip,
1511 						&ip->i_pdquot, pdqp);
1512 		}
1513 		ip->i_d.di_projid = fa->fsx_projid;
1514 	}
1515 
1516 	/*
1517 	 * Only set the extent size hint if we've already determined that the
1518 	 * extent size hint should be set on the inode. If no extent size flags
1519 	 * are set on the inode then unconditionally clear the extent size hint.
1520 	 */
1521 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1522 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1523 	else
1524 		ip->i_d.di_extsize = 0;
1525 	if (xfs_sb_version_has_v3inode(&mp->m_sb) &&
1526 	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1527 		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
1528 				mp->m_sb.sb_blocklog;
1529 	else
1530 		ip->i_d.di_cowextsize = 0;
1531 
1532 	error = xfs_trans_commit(tp);
1533 
1534 	/*
1535 	 * Release any dquot(s) the inode had kept before chown.
1536 	 */
1537 	xfs_qm_dqrele(olddquot);
1538 	xfs_qm_dqrele(pdqp);
1539 
1540 	return error;
1541 
1542 error_trans_cancel:
1543 	xfs_trans_cancel(tp);
1544 error_free_dquots:
1545 	xfs_qm_dqrele(pdqp);
1546 	return error;
1547 }
1548 
1549 STATIC int
1550 xfs_ioc_fssetxattr(
1551 	struct file		*filp,
1552 	void			__user *arg)
1553 {
1554 	struct fsxattr		fa;
1555 	int error;
1556 
1557 	if (copy_from_user(&fa, arg, sizeof(fa)))
1558 		return -EFAULT;
1559 
1560 	error = mnt_want_write_file(filp);
1561 	if (error)
1562 		return error;
1563 	error = xfs_ioctl_setattr(filp, &fa);
1564 	mnt_drop_write_file(filp);
1565 	return error;
1566 }
1567 
1568 STATIC int
1569 xfs_ioc_getxflags(
1570 	xfs_inode_t		*ip,
1571 	void			__user *arg)
1572 {
1573 	unsigned int		flags;
1574 
1575 	flags = xfs_di2lxflags(ip->i_d.di_flags, ip->i_d.di_flags2);
1576 	if (copy_to_user(arg, &flags, sizeof(flags)))
1577 		return -EFAULT;
1578 	return 0;
1579 }
1580 
1581 STATIC int
1582 xfs_ioc_setxflags(
1583 	struct xfs_inode	*ip,
1584 	struct file		*filp,
1585 	void			__user *arg)
1586 {
1587 	struct xfs_trans	*tp;
1588 	struct fsxattr		fa;
1589 	struct fsxattr		old_fa;
1590 	unsigned int		flags;
1591 	int			error;
1592 
1593 	if (copy_from_user(&flags, arg, sizeof(flags)))
1594 		return -EFAULT;
1595 
1596 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1597 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1598 		      FS_SYNC_FL | FS_DAX_FL))
1599 		return -EOPNOTSUPP;
1600 
1601 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1602 
1603 	error = mnt_want_write_file(filp);
1604 	if (error)
1605 		return error;
1606 
1607 	xfs_ioctl_setattr_prepare_dax(ip, &fa);
1608 
1609 	tp = xfs_ioctl_setattr_get_trans(filp, NULL);
1610 	if (IS_ERR(tp)) {
1611 		error = PTR_ERR(tp);
1612 		goto out_drop_write;
1613 	}
1614 
1615 	xfs_fill_fsxattr(ip, false, &old_fa);
1616 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, &fa);
1617 	if (error) {
1618 		xfs_trans_cancel(tp);
1619 		goto out_drop_write;
1620 	}
1621 
1622 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1623 	if (error) {
1624 		xfs_trans_cancel(tp);
1625 		goto out_drop_write;
1626 	}
1627 
1628 	error = xfs_trans_commit(tp);
1629 out_drop_write:
1630 	mnt_drop_write_file(filp);
1631 	return error;
1632 }
1633 
1634 static bool
1635 xfs_getbmap_format(
1636 	struct kgetbmap		*p,
1637 	struct getbmapx __user	*u,
1638 	size_t			recsize)
1639 {
1640 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1641 	    put_user(p->bmv_block, &u->bmv_block) ||
1642 	    put_user(p->bmv_length, &u->bmv_length) ||
1643 	    put_user(0, &u->bmv_count) ||
1644 	    put_user(0, &u->bmv_entries))
1645 		return false;
1646 	if (recsize < sizeof(struct getbmapx))
1647 		return true;
1648 	if (put_user(0, &u->bmv_iflags) ||
1649 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1650 	    put_user(0, &u->bmv_unused1) ||
1651 	    put_user(0, &u->bmv_unused2))
1652 		return false;
1653 	return true;
1654 }
1655 
1656 STATIC int
1657 xfs_ioc_getbmap(
1658 	struct file		*file,
1659 	unsigned int		cmd,
1660 	void			__user *arg)
1661 {
1662 	struct getbmapx		bmx = { 0 };
1663 	struct kgetbmap		*buf;
1664 	size_t			recsize;
1665 	int			error, i;
1666 
1667 	switch (cmd) {
1668 	case XFS_IOC_GETBMAPA:
1669 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1670 		/*FALLTHRU*/
1671 	case XFS_IOC_GETBMAP:
1672 		if (file->f_mode & FMODE_NOCMTIME)
1673 			bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1674 		/* struct getbmap is a strict subset of struct getbmapx. */
1675 		recsize = sizeof(struct getbmap);
1676 		break;
1677 	case XFS_IOC_GETBMAPX:
1678 		recsize = sizeof(struct getbmapx);
1679 		break;
1680 	default:
1681 		return -EINVAL;
1682 	}
1683 
1684 	if (copy_from_user(&bmx, arg, recsize))
1685 		return -EFAULT;
1686 
1687 	if (bmx.bmv_count < 2)
1688 		return -EINVAL;
1689 	if (bmx.bmv_count > ULONG_MAX / recsize)
1690 		return -ENOMEM;
1691 
1692 	buf = kvzalloc(bmx.bmv_count * sizeof(*buf), GFP_KERNEL);
1693 	if (!buf)
1694 		return -ENOMEM;
1695 
1696 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1697 	if (error)
1698 		goto out_free_buf;
1699 
1700 	error = -EFAULT;
1701 	if (copy_to_user(arg, &bmx, recsize))
1702 		goto out_free_buf;
1703 	arg += recsize;
1704 
1705 	for (i = 0; i < bmx.bmv_entries; i++) {
1706 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1707 			goto out_free_buf;
1708 		arg += recsize;
1709 	}
1710 
1711 	error = 0;
1712 out_free_buf:
1713 	kmem_free(buf);
1714 	return error;
1715 }
1716 
1717 STATIC int
1718 xfs_ioc_getfsmap(
1719 	struct xfs_inode	*ip,
1720 	struct fsmap_head	__user *arg)
1721 {
1722 	struct xfs_fsmap_head	xhead = {0};
1723 	struct fsmap_head	head;
1724 	struct fsmap		*recs;
1725 	unsigned int		count;
1726 	__u32			last_flags = 0;
1727 	bool			done = false;
1728 	int			error;
1729 
1730 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1731 		return -EFAULT;
1732 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1733 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1734 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1735 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1736 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1737 		return -EINVAL;
1738 
1739 	/*
1740 	 * Use an internal memory buffer so that we don't have to copy fsmap
1741 	 * data to userspace while holding locks.  Start by trying to allocate
1742 	 * up to 128k for the buffer, but fall back to a single page if needed.
1743 	 */
1744 	count = min_t(unsigned int, head.fmh_count,
1745 			131072 / sizeof(struct fsmap));
1746 	recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1747 	if (!recs) {
1748 		count = min_t(unsigned int, head.fmh_count,
1749 				PAGE_SIZE / sizeof(struct fsmap));
1750 		recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1751 		if (!recs)
1752 			return -ENOMEM;
1753 	}
1754 
1755 	xhead.fmh_iflags = head.fmh_iflags;
1756 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1757 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1758 
1759 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1760 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1761 
1762 	head.fmh_entries = 0;
1763 	do {
1764 		struct fsmap __user	*user_recs;
1765 		struct fsmap		*last_rec;
1766 
1767 		user_recs = &arg->fmh_recs[head.fmh_entries];
1768 		xhead.fmh_entries = 0;
1769 		xhead.fmh_count = min_t(unsigned int, count,
1770 					head.fmh_count - head.fmh_entries);
1771 
1772 		/* Run query, record how many entries we got. */
1773 		error = xfs_getfsmap(ip->i_mount, &xhead, recs);
1774 		switch (error) {
1775 		case 0:
1776 			/*
1777 			 * There are no more records in the result set.  Copy
1778 			 * whatever we got to userspace and break out.
1779 			 */
1780 			done = true;
1781 			break;
1782 		case -ECANCELED:
1783 			/*
1784 			 * The internal memory buffer is full.  Copy whatever
1785 			 * records we got to userspace and go again if we have
1786 			 * not yet filled the userspace buffer.
1787 			 */
1788 			error = 0;
1789 			break;
1790 		default:
1791 			goto out_free;
1792 		}
1793 		head.fmh_entries += xhead.fmh_entries;
1794 		head.fmh_oflags = xhead.fmh_oflags;
1795 
1796 		/*
1797 		 * If the caller wanted a record count or there aren't any
1798 		 * new records to return, we're done.
1799 		 */
1800 		if (head.fmh_count == 0 || xhead.fmh_entries == 0)
1801 			break;
1802 
1803 		/* Copy all the records we got out to userspace. */
1804 		if (copy_to_user(user_recs, recs,
1805 				 xhead.fmh_entries * sizeof(struct fsmap))) {
1806 			error = -EFAULT;
1807 			goto out_free;
1808 		}
1809 
1810 		/* Remember the last record flags we copied to userspace. */
1811 		last_rec = &recs[xhead.fmh_entries - 1];
1812 		last_flags = last_rec->fmr_flags;
1813 
1814 		/* Set up the low key for the next iteration. */
1815 		xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec);
1816 		trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1817 	} while (!done && head.fmh_entries < head.fmh_count);
1818 
1819 	/*
1820 	 * If there are no more records in the query result set and we're not
1821 	 * in counting mode, mark the last record returned with the LAST flag.
1822 	 */
1823 	if (done && head.fmh_count > 0 && head.fmh_entries > 0) {
1824 		struct fsmap __user	*user_rec;
1825 
1826 		last_flags |= FMR_OF_LAST;
1827 		user_rec = &arg->fmh_recs[head.fmh_entries - 1];
1828 
1829 		if (copy_to_user(&user_rec->fmr_flags, &last_flags,
1830 					sizeof(last_flags))) {
1831 			error = -EFAULT;
1832 			goto out_free;
1833 		}
1834 	}
1835 
1836 	/* copy back header */
1837 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) {
1838 		error = -EFAULT;
1839 		goto out_free;
1840 	}
1841 
1842 out_free:
1843 	kmem_free(recs);
1844 	return error;
1845 }
1846 
1847 STATIC int
1848 xfs_ioc_scrub_metadata(
1849 	struct xfs_inode		*ip,
1850 	void				__user *arg)
1851 {
1852 	struct xfs_scrub_metadata	scrub;
1853 	int				error;
1854 
1855 	if (!capable(CAP_SYS_ADMIN))
1856 		return -EPERM;
1857 
1858 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1859 		return -EFAULT;
1860 
1861 	error = xfs_scrub_metadata(ip, &scrub);
1862 	if (error)
1863 		return error;
1864 
1865 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1866 		return -EFAULT;
1867 
1868 	return 0;
1869 }
1870 
1871 int
1872 xfs_ioc_swapext(
1873 	xfs_swapext_t	*sxp)
1874 {
1875 	xfs_inode_t     *ip, *tip;
1876 	struct fd	f, tmp;
1877 	int		error = 0;
1878 
1879 	/* Pull information for the target fd */
1880 	f = fdget((int)sxp->sx_fdtarget);
1881 	if (!f.file) {
1882 		error = -EINVAL;
1883 		goto out;
1884 	}
1885 
1886 	if (!(f.file->f_mode & FMODE_WRITE) ||
1887 	    !(f.file->f_mode & FMODE_READ) ||
1888 	    (f.file->f_flags & O_APPEND)) {
1889 		error = -EBADF;
1890 		goto out_put_file;
1891 	}
1892 
1893 	tmp = fdget((int)sxp->sx_fdtmp);
1894 	if (!tmp.file) {
1895 		error = -EINVAL;
1896 		goto out_put_file;
1897 	}
1898 
1899 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1900 	    !(tmp.file->f_mode & FMODE_READ) ||
1901 	    (tmp.file->f_flags & O_APPEND)) {
1902 		error = -EBADF;
1903 		goto out_put_tmp_file;
1904 	}
1905 
1906 	if (IS_SWAPFILE(file_inode(f.file)) ||
1907 	    IS_SWAPFILE(file_inode(tmp.file))) {
1908 		error = -EINVAL;
1909 		goto out_put_tmp_file;
1910 	}
1911 
1912 	/*
1913 	 * We need to ensure that the fds passed in point to XFS inodes
1914 	 * before we cast and access them as XFS structures as we have no
1915 	 * control over what the user passes us here.
1916 	 */
1917 	if (f.file->f_op != &xfs_file_operations ||
1918 	    tmp.file->f_op != &xfs_file_operations) {
1919 		error = -EINVAL;
1920 		goto out_put_tmp_file;
1921 	}
1922 
1923 	ip = XFS_I(file_inode(f.file));
1924 	tip = XFS_I(file_inode(tmp.file));
1925 
1926 	if (ip->i_mount != tip->i_mount) {
1927 		error = -EINVAL;
1928 		goto out_put_tmp_file;
1929 	}
1930 
1931 	if (ip->i_ino == tip->i_ino) {
1932 		error = -EINVAL;
1933 		goto out_put_tmp_file;
1934 	}
1935 
1936 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1937 		error = -EIO;
1938 		goto out_put_tmp_file;
1939 	}
1940 
1941 	error = xfs_swap_extents(ip, tip, sxp);
1942 
1943  out_put_tmp_file:
1944 	fdput(tmp);
1945  out_put_file:
1946 	fdput(f);
1947  out:
1948 	return error;
1949 }
1950 
1951 static int
1952 xfs_ioc_getlabel(
1953 	struct xfs_mount	*mp,
1954 	char			__user *user_label)
1955 {
1956 	struct xfs_sb		*sbp = &mp->m_sb;
1957 	char			label[XFSLABEL_MAX + 1];
1958 
1959 	/* Paranoia */
1960 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
1961 
1962 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
1963 	memset(label, 0, sizeof(label));
1964 	spin_lock(&mp->m_sb_lock);
1965 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
1966 	spin_unlock(&mp->m_sb_lock);
1967 
1968 	if (copy_to_user(user_label, label, sizeof(label)))
1969 		return -EFAULT;
1970 	return 0;
1971 }
1972 
1973 static int
1974 xfs_ioc_setlabel(
1975 	struct file		*filp,
1976 	struct xfs_mount	*mp,
1977 	char			__user *newlabel)
1978 {
1979 	struct xfs_sb		*sbp = &mp->m_sb;
1980 	char			label[XFSLABEL_MAX + 1];
1981 	size_t			len;
1982 	int			error;
1983 
1984 	if (!capable(CAP_SYS_ADMIN))
1985 		return -EPERM;
1986 	/*
1987 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
1988 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
1989 	 * (required) NULL character to test the incoming label length.
1990 	 * NB: The on disk label doesn't need to be null terminated.
1991 	 */
1992 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
1993 		return -EFAULT;
1994 	len = strnlen(label, XFSLABEL_MAX + 1);
1995 	if (len > sizeof(sbp->sb_fname))
1996 		return -EINVAL;
1997 
1998 	error = mnt_want_write_file(filp);
1999 	if (error)
2000 		return error;
2001 
2002 	spin_lock(&mp->m_sb_lock);
2003 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
2004 	memcpy(sbp->sb_fname, label, len);
2005 	spin_unlock(&mp->m_sb_lock);
2006 
2007 	/*
2008 	 * Now we do several things to satisfy userspace.
2009 	 * In addition to normal logging of the primary superblock, we also
2010 	 * immediately write these changes to sector zero for the primary, then
2011 	 * update all backup supers (as xfs_db does for a label change), then
2012 	 * invalidate the block device page cache.  This is so that any prior
2013 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
2014 	 * and userspace will see the newly-written label.
2015 	 */
2016 	error = xfs_sync_sb_buf(mp);
2017 	if (error)
2018 		goto out;
2019 	/*
2020 	 * growfs also updates backup supers so lock against that.
2021 	 */
2022 	mutex_lock(&mp->m_growlock);
2023 	error = xfs_update_secondary_sbs(mp);
2024 	mutex_unlock(&mp->m_growlock);
2025 
2026 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
2027 
2028 out:
2029 	mnt_drop_write_file(filp);
2030 	return error;
2031 }
2032 
2033 static inline int
2034 xfs_fs_eofblocks_from_user(
2035 	struct xfs_fs_eofblocks		*src,
2036 	struct xfs_eofblocks		*dst)
2037 {
2038 	if (src->eof_version != XFS_EOFBLOCKS_VERSION)
2039 		return -EINVAL;
2040 
2041 	if (src->eof_flags & ~XFS_EOF_FLAGS_VALID)
2042 		return -EINVAL;
2043 
2044 	if (memchr_inv(&src->pad32, 0, sizeof(src->pad32)) ||
2045 	    memchr_inv(src->pad64, 0, sizeof(src->pad64)))
2046 		return -EINVAL;
2047 
2048 	dst->eof_flags = src->eof_flags;
2049 	dst->eof_prid = src->eof_prid;
2050 	dst->eof_min_file_size = src->eof_min_file_size;
2051 
2052 	dst->eof_uid = INVALID_UID;
2053 	if (src->eof_flags & XFS_EOF_FLAGS_UID) {
2054 		dst->eof_uid = make_kuid(current_user_ns(), src->eof_uid);
2055 		if (!uid_valid(dst->eof_uid))
2056 			return -EINVAL;
2057 	}
2058 
2059 	dst->eof_gid = INVALID_GID;
2060 	if (src->eof_flags & XFS_EOF_FLAGS_GID) {
2061 		dst->eof_gid = make_kgid(current_user_ns(), src->eof_gid);
2062 		if (!gid_valid(dst->eof_gid))
2063 			return -EINVAL;
2064 	}
2065 	return 0;
2066 }
2067 
2068 /*
2069  * Note: some of the ioctl's return positive numbers as a
2070  * byte count indicating success, such as readlink_by_handle.
2071  * So we don't "sign flip" like most other routines.  This means
2072  * true errors need to be returned as a negative value.
2073  */
2074 long
2075 xfs_file_ioctl(
2076 	struct file		*filp,
2077 	unsigned int		cmd,
2078 	unsigned long		p)
2079 {
2080 	struct inode		*inode = file_inode(filp);
2081 	struct xfs_inode	*ip = XFS_I(inode);
2082 	struct xfs_mount	*mp = ip->i_mount;
2083 	void			__user *arg = (void __user *)p;
2084 	int			error;
2085 
2086 	trace_xfs_file_ioctl(ip);
2087 
2088 	switch (cmd) {
2089 	case FITRIM:
2090 		return xfs_ioc_trim(mp, arg);
2091 	case FS_IOC_GETFSLABEL:
2092 		return xfs_ioc_getlabel(mp, arg);
2093 	case FS_IOC_SETFSLABEL:
2094 		return xfs_ioc_setlabel(filp, mp, arg);
2095 	case XFS_IOC_ALLOCSP:
2096 	case XFS_IOC_FREESP:
2097 	case XFS_IOC_ALLOCSP64:
2098 	case XFS_IOC_FREESP64: {
2099 		xfs_flock64_t		bf;
2100 
2101 		if (copy_from_user(&bf, arg, sizeof(bf)))
2102 			return -EFAULT;
2103 		return xfs_ioc_space(filp, &bf);
2104 	}
2105 	case XFS_IOC_DIOINFO: {
2106 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
2107 		struct dioattr		da;
2108 
2109 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
2110 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
2111 
2112 		if (copy_to_user(arg, &da, sizeof(da)))
2113 			return -EFAULT;
2114 		return 0;
2115 	}
2116 
2117 	case XFS_IOC_FSBULKSTAT_SINGLE:
2118 	case XFS_IOC_FSBULKSTAT:
2119 	case XFS_IOC_FSINUMBERS:
2120 		return xfs_ioc_fsbulkstat(filp, cmd, arg);
2121 
2122 	case XFS_IOC_BULKSTAT:
2123 		return xfs_ioc_bulkstat(filp, cmd, arg);
2124 	case XFS_IOC_INUMBERS:
2125 		return xfs_ioc_inumbers(mp, cmd, arg);
2126 
2127 	case XFS_IOC_FSGEOMETRY_V1:
2128 		return xfs_ioc_fsgeometry(mp, arg, 3);
2129 	case XFS_IOC_FSGEOMETRY_V4:
2130 		return xfs_ioc_fsgeometry(mp, arg, 4);
2131 	case XFS_IOC_FSGEOMETRY:
2132 		return xfs_ioc_fsgeometry(mp, arg, 5);
2133 
2134 	case XFS_IOC_AG_GEOMETRY:
2135 		return xfs_ioc_ag_geometry(mp, arg);
2136 
2137 	case XFS_IOC_GETVERSION:
2138 		return put_user(inode->i_generation, (int __user *)arg);
2139 
2140 	case XFS_IOC_FSGETXATTR:
2141 		return xfs_ioc_fsgetxattr(ip, 0, arg);
2142 	case XFS_IOC_FSGETXATTRA:
2143 		return xfs_ioc_fsgetxattr(ip, 1, arg);
2144 	case XFS_IOC_FSSETXATTR:
2145 		return xfs_ioc_fssetxattr(filp, arg);
2146 	case XFS_IOC_GETXFLAGS:
2147 		return xfs_ioc_getxflags(ip, arg);
2148 	case XFS_IOC_SETXFLAGS:
2149 		return xfs_ioc_setxflags(ip, filp, arg);
2150 
2151 	case XFS_IOC_GETBMAP:
2152 	case XFS_IOC_GETBMAPA:
2153 	case XFS_IOC_GETBMAPX:
2154 		return xfs_ioc_getbmap(filp, cmd, arg);
2155 
2156 	case FS_IOC_GETFSMAP:
2157 		return xfs_ioc_getfsmap(ip, arg);
2158 
2159 	case XFS_IOC_SCRUB_METADATA:
2160 		return xfs_ioc_scrub_metadata(ip, arg);
2161 
2162 	case XFS_IOC_FD_TO_HANDLE:
2163 	case XFS_IOC_PATH_TO_HANDLE:
2164 	case XFS_IOC_PATH_TO_FSHANDLE: {
2165 		xfs_fsop_handlereq_t	hreq;
2166 
2167 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2168 			return -EFAULT;
2169 		return xfs_find_handle(cmd, &hreq);
2170 	}
2171 	case XFS_IOC_OPEN_BY_HANDLE: {
2172 		xfs_fsop_handlereq_t	hreq;
2173 
2174 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2175 			return -EFAULT;
2176 		return xfs_open_by_handle(filp, &hreq);
2177 	}
2178 
2179 	case XFS_IOC_READLINK_BY_HANDLE: {
2180 		xfs_fsop_handlereq_t	hreq;
2181 
2182 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2183 			return -EFAULT;
2184 		return xfs_readlink_by_handle(filp, &hreq);
2185 	}
2186 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2187 		return xfs_attrlist_by_handle(filp, arg);
2188 
2189 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2190 		return xfs_attrmulti_by_handle(filp, arg);
2191 
2192 	case XFS_IOC_SWAPEXT: {
2193 		struct xfs_swapext	sxp;
2194 
2195 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2196 			return -EFAULT;
2197 		error = mnt_want_write_file(filp);
2198 		if (error)
2199 			return error;
2200 		error = xfs_ioc_swapext(&sxp);
2201 		mnt_drop_write_file(filp);
2202 		return error;
2203 	}
2204 
2205 	case XFS_IOC_FSCOUNTS: {
2206 		xfs_fsop_counts_t out;
2207 
2208 		xfs_fs_counts(mp, &out);
2209 
2210 		if (copy_to_user(arg, &out, sizeof(out)))
2211 			return -EFAULT;
2212 		return 0;
2213 	}
2214 
2215 	case XFS_IOC_SET_RESBLKS: {
2216 		xfs_fsop_resblks_t inout;
2217 		uint64_t	   in;
2218 
2219 		if (!capable(CAP_SYS_ADMIN))
2220 			return -EPERM;
2221 
2222 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2223 			return -EROFS;
2224 
2225 		if (copy_from_user(&inout, arg, sizeof(inout)))
2226 			return -EFAULT;
2227 
2228 		error = mnt_want_write_file(filp);
2229 		if (error)
2230 			return error;
2231 
2232 		/* input parameter is passed in resblks field of structure */
2233 		in = inout.resblks;
2234 		error = xfs_reserve_blocks(mp, &in, &inout);
2235 		mnt_drop_write_file(filp);
2236 		if (error)
2237 			return error;
2238 
2239 		if (copy_to_user(arg, &inout, sizeof(inout)))
2240 			return -EFAULT;
2241 		return 0;
2242 	}
2243 
2244 	case XFS_IOC_GET_RESBLKS: {
2245 		xfs_fsop_resblks_t out;
2246 
2247 		if (!capable(CAP_SYS_ADMIN))
2248 			return -EPERM;
2249 
2250 		error = xfs_reserve_blocks(mp, NULL, &out);
2251 		if (error)
2252 			return error;
2253 
2254 		if (copy_to_user(arg, &out, sizeof(out)))
2255 			return -EFAULT;
2256 
2257 		return 0;
2258 	}
2259 
2260 	case XFS_IOC_FSGROWFSDATA: {
2261 		struct xfs_growfs_data in;
2262 
2263 		if (copy_from_user(&in, arg, sizeof(in)))
2264 			return -EFAULT;
2265 
2266 		error = mnt_want_write_file(filp);
2267 		if (error)
2268 			return error;
2269 		error = xfs_growfs_data(mp, &in);
2270 		mnt_drop_write_file(filp);
2271 		return error;
2272 	}
2273 
2274 	case XFS_IOC_FSGROWFSLOG: {
2275 		struct xfs_growfs_log in;
2276 
2277 		if (copy_from_user(&in, arg, sizeof(in)))
2278 			return -EFAULT;
2279 
2280 		error = mnt_want_write_file(filp);
2281 		if (error)
2282 			return error;
2283 		error = xfs_growfs_log(mp, &in);
2284 		mnt_drop_write_file(filp);
2285 		return error;
2286 	}
2287 
2288 	case XFS_IOC_FSGROWFSRT: {
2289 		xfs_growfs_rt_t in;
2290 
2291 		if (copy_from_user(&in, arg, sizeof(in)))
2292 			return -EFAULT;
2293 
2294 		error = mnt_want_write_file(filp);
2295 		if (error)
2296 			return error;
2297 		error = xfs_growfs_rt(mp, &in);
2298 		mnt_drop_write_file(filp);
2299 		return error;
2300 	}
2301 
2302 	case XFS_IOC_GOINGDOWN: {
2303 		uint32_t in;
2304 
2305 		if (!capable(CAP_SYS_ADMIN))
2306 			return -EPERM;
2307 
2308 		if (get_user(in, (uint32_t __user *)arg))
2309 			return -EFAULT;
2310 
2311 		return xfs_fs_goingdown(mp, in);
2312 	}
2313 
2314 	case XFS_IOC_ERROR_INJECTION: {
2315 		xfs_error_injection_t in;
2316 
2317 		if (!capable(CAP_SYS_ADMIN))
2318 			return -EPERM;
2319 
2320 		if (copy_from_user(&in, arg, sizeof(in)))
2321 			return -EFAULT;
2322 
2323 		return xfs_errortag_add(mp, in.errtag);
2324 	}
2325 
2326 	case XFS_IOC_ERROR_CLEARALL:
2327 		if (!capable(CAP_SYS_ADMIN))
2328 			return -EPERM;
2329 
2330 		return xfs_errortag_clearall(mp);
2331 
2332 	case XFS_IOC_FREE_EOFBLOCKS: {
2333 		struct xfs_fs_eofblocks eofb;
2334 		struct xfs_eofblocks keofb;
2335 
2336 		if (!capable(CAP_SYS_ADMIN))
2337 			return -EPERM;
2338 
2339 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2340 			return -EROFS;
2341 
2342 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2343 			return -EFAULT;
2344 
2345 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
2346 		if (error)
2347 			return error;
2348 
2349 		trace_xfs_ioc_free_eofblocks(mp, &keofb, _RET_IP_);
2350 
2351 		sb_start_write(mp->m_super);
2352 		error = xfs_blockgc_free_space(mp, &keofb);
2353 		sb_end_write(mp->m_super);
2354 		return error;
2355 	}
2356 
2357 	default:
2358 		return -ENOTTY;
2359 	}
2360 }
2361