xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 70a59dd8)
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_dentry(filp), &iattr);
697 	if (error)
698 		goto out_unlock;
699 
700 	error = xfs_update_prealloc_flags(ip, flags);
701 
702 out_unlock:
703 	xfs_iunlock(ip, iolock);
704 	mnt_drop_write_file(filp);
705 	return error;
706 }
707 
708 /* Return 0 on success or positive error */
709 int
710 xfs_fsbulkstat_one_fmt(
711 	struct xfs_ibulk		*breq,
712 	const struct xfs_bulkstat	*bstat)
713 {
714 	struct xfs_bstat		bs1;
715 
716 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
717 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
718 		return -EFAULT;
719 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
720 }
721 
722 int
723 xfs_fsinumbers_fmt(
724 	struct xfs_ibulk		*breq,
725 	const struct xfs_inumbers	*igrp)
726 {
727 	struct xfs_inogrp		ig1;
728 
729 	xfs_inumbers_to_inogrp(&ig1, igrp);
730 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
731 		return -EFAULT;
732 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
733 }
734 
735 STATIC int
736 xfs_ioc_fsbulkstat(
737 	xfs_mount_t		*mp,
738 	unsigned int		cmd,
739 	void			__user *arg)
740 {
741 	struct xfs_fsop_bulkreq	bulkreq;
742 	struct xfs_ibulk	breq = {
743 		.mp		= mp,
744 		.ocount		= 0,
745 	};
746 	xfs_ino_t		lastino;
747 	int			error;
748 
749 	/* done = 1 if there are more stats to get and if bulkstat */
750 	/* should be called again (unused here, but used in dmapi) */
751 
752 	if (!capable(CAP_SYS_ADMIN))
753 		return -EPERM;
754 
755 	if (XFS_FORCED_SHUTDOWN(mp))
756 		return -EIO;
757 
758 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
759 		return -EFAULT;
760 
761 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
762 		return -EFAULT;
763 
764 	if (bulkreq.icount <= 0)
765 		return -EINVAL;
766 
767 	if (bulkreq.ubuffer == NULL)
768 		return -EINVAL;
769 
770 	breq.ubuffer = bulkreq.ubuffer;
771 	breq.icount = bulkreq.icount;
772 
773 	/*
774 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
775 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
776 	 * that *lastip contains either zero or the number of the last inode to
777 	 * be examined by the previous call and return results starting with
778 	 * the next inode after that.  The new bulk request back end functions
779 	 * take the inode to start with, so we have to compute the startino
780 	 * parameter from lastino to maintain correct function.  lastino == 0
781 	 * is a special case because it has traditionally meant "first inode
782 	 * in filesystem".
783 	 */
784 	if (cmd == XFS_IOC_FSINUMBERS) {
785 		breq.startino = lastino ? lastino + 1 : 0;
786 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
787 		lastino = breq.startino - 1;
788 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
789 		breq.startino = lastino;
790 		breq.icount = 1;
791 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
792 	} else {	/* XFS_IOC_FSBULKSTAT */
793 		breq.startino = lastino ? lastino + 1 : 0;
794 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
795 		lastino = breq.startino - 1;
796 	}
797 
798 	if (error)
799 		return error;
800 
801 	if (bulkreq.lastip != NULL &&
802 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
803 		return -EFAULT;
804 
805 	if (bulkreq.ocount != NULL &&
806 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
807 		return -EFAULT;
808 
809 	return 0;
810 }
811 
812 /* Return 0 on success or positive error */
813 static int
814 xfs_bulkstat_fmt(
815 	struct xfs_ibulk		*breq,
816 	const struct xfs_bulkstat	*bstat)
817 {
818 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
819 		return -EFAULT;
820 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
821 }
822 
823 /*
824  * Check the incoming bulk request @hdr from userspace and initialize the
825  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
826  * should proceed; -ECANCELED if there's nothing to do; or the usual
827  * negative error code.
828  */
829 static int
830 xfs_bulk_ireq_setup(
831 	struct xfs_mount	*mp,
832 	struct xfs_bulk_ireq	*hdr,
833 	struct xfs_ibulk	*breq,
834 	void __user		*ubuffer)
835 {
836 	if (hdr->icount == 0 ||
837 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
838 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
839 		return -EINVAL;
840 
841 	breq->startino = hdr->ino;
842 	breq->ubuffer = ubuffer;
843 	breq->icount = hdr->icount;
844 	breq->ocount = 0;
845 	breq->flags = 0;
846 
847 	/*
848 	 * The @ino parameter is a special value, so we must look it up here.
849 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
850 	 * worth of data.
851 	 */
852 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
853 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
854 			return -EINVAL;
855 
856 		switch (hdr->ino) {
857 		case XFS_BULK_IREQ_SPECIAL_ROOT:
858 			hdr->ino = mp->m_sb.sb_rootino;
859 			break;
860 		default:
861 			return -EINVAL;
862 		}
863 		breq->icount = 1;
864 	}
865 
866 	/*
867 	 * The IREQ_AGNO flag means that we only want results from a given AG.
868 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
869 	 * beyond the specified AG then we return no results.
870 	 */
871 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
872 		if (hdr->agno >= mp->m_sb.sb_agcount)
873 			return -EINVAL;
874 
875 		if (breq->startino == 0)
876 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
877 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
878 			return -EINVAL;
879 
880 		breq->flags |= XFS_IBULK_SAME_AG;
881 
882 		/* Asking for an inode past the end of the AG?  We're done! */
883 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
884 			return -ECANCELED;
885 	} else if (hdr->agno)
886 		return -EINVAL;
887 
888 	/* Asking for an inode past the end of the FS?  We're done! */
889 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
890 		return -ECANCELED;
891 
892 	return 0;
893 }
894 
895 /*
896  * Update the userspace bulk request @hdr to reflect the end state of the
897  * internal bulk request @breq.
898  */
899 static void
900 xfs_bulk_ireq_teardown(
901 	struct xfs_bulk_ireq	*hdr,
902 	struct xfs_ibulk	*breq)
903 {
904 	hdr->ino = breq->startino;
905 	hdr->ocount = breq->ocount;
906 }
907 
908 /* Handle the v5 bulkstat ioctl. */
909 STATIC int
910 xfs_ioc_bulkstat(
911 	struct xfs_mount		*mp,
912 	unsigned int			cmd,
913 	struct xfs_bulkstat_req __user	*arg)
914 {
915 	struct xfs_bulk_ireq		hdr;
916 	struct xfs_ibulk		breq = {
917 		.mp			= mp,
918 	};
919 	int				error;
920 
921 	if (!capable(CAP_SYS_ADMIN))
922 		return -EPERM;
923 
924 	if (XFS_FORCED_SHUTDOWN(mp))
925 		return -EIO;
926 
927 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
928 		return -EFAULT;
929 
930 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
931 	if (error == -ECANCELED)
932 		goto out_teardown;
933 	if (error < 0)
934 		return error;
935 
936 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
937 	if (error)
938 		return error;
939 
940 out_teardown:
941 	xfs_bulk_ireq_teardown(&hdr, &breq);
942 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
943 		return -EFAULT;
944 
945 	return 0;
946 }
947 
948 STATIC int
949 xfs_inumbers_fmt(
950 	struct xfs_ibulk		*breq,
951 	const struct xfs_inumbers	*igrp)
952 {
953 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
954 		return -EFAULT;
955 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
956 }
957 
958 /* Handle the v5 inumbers ioctl. */
959 STATIC int
960 xfs_ioc_inumbers(
961 	struct xfs_mount		*mp,
962 	unsigned int			cmd,
963 	struct xfs_inumbers_req __user	*arg)
964 {
965 	struct xfs_bulk_ireq		hdr;
966 	struct xfs_ibulk		breq = {
967 		.mp			= mp,
968 	};
969 	int				error;
970 
971 	if (!capable(CAP_SYS_ADMIN))
972 		return -EPERM;
973 
974 	if (XFS_FORCED_SHUTDOWN(mp))
975 		return -EIO;
976 
977 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
978 		return -EFAULT;
979 
980 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
981 	if (error == -ECANCELED)
982 		goto out_teardown;
983 	if (error < 0)
984 		return error;
985 
986 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
987 	if (error)
988 		return error;
989 
990 out_teardown:
991 	xfs_bulk_ireq_teardown(&hdr, &breq);
992 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
993 		return -EFAULT;
994 
995 	return 0;
996 }
997 
998 STATIC int
999 xfs_ioc_fsgeometry(
1000 	struct xfs_mount	*mp,
1001 	void			__user *arg,
1002 	int			struct_version)
1003 {
1004 	struct xfs_fsop_geom	fsgeo;
1005 	size_t			len;
1006 
1007 	xfs_fs_geometry(&mp->m_sb, &fsgeo, struct_version);
1008 
1009 	if (struct_version <= 3)
1010 		len = sizeof(struct xfs_fsop_geom_v1);
1011 	else if (struct_version == 4)
1012 		len = sizeof(struct xfs_fsop_geom_v4);
1013 	else {
1014 		xfs_fsop_geom_health(mp, &fsgeo);
1015 		len = sizeof(fsgeo);
1016 	}
1017 
1018 	if (copy_to_user(arg, &fsgeo, len))
1019 		return -EFAULT;
1020 	return 0;
1021 }
1022 
1023 STATIC int
1024 xfs_ioc_ag_geometry(
1025 	struct xfs_mount	*mp,
1026 	void			__user *arg)
1027 {
1028 	struct xfs_ag_geometry	ageo;
1029 	int			error;
1030 
1031 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
1032 		return -EFAULT;
1033 	if (ageo.ag_flags)
1034 		return -EINVAL;
1035 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
1036 		return -EINVAL;
1037 
1038 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
1039 	if (error)
1040 		return error;
1041 
1042 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
1043 		return -EFAULT;
1044 	return 0;
1045 }
1046 
1047 /*
1048  * Linux extended inode flags interface.
1049  */
1050 
1051 STATIC unsigned int
1052 xfs_merge_ioc_xflags(
1053 	unsigned int	flags,
1054 	unsigned int	start)
1055 {
1056 	unsigned int	xflags = start;
1057 
1058 	if (flags & FS_IMMUTABLE_FL)
1059 		xflags |= FS_XFLAG_IMMUTABLE;
1060 	else
1061 		xflags &= ~FS_XFLAG_IMMUTABLE;
1062 	if (flags & FS_APPEND_FL)
1063 		xflags |= FS_XFLAG_APPEND;
1064 	else
1065 		xflags &= ~FS_XFLAG_APPEND;
1066 	if (flags & FS_SYNC_FL)
1067 		xflags |= FS_XFLAG_SYNC;
1068 	else
1069 		xflags &= ~FS_XFLAG_SYNC;
1070 	if (flags & FS_NOATIME_FL)
1071 		xflags |= FS_XFLAG_NOATIME;
1072 	else
1073 		xflags &= ~FS_XFLAG_NOATIME;
1074 	if (flags & FS_NODUMP_FL)
1075 		xflags |= FS_XFLAG_NODUMP;
1076 	else
1077 		xflags &= ~FS_XFLAG_NODUMP;
1078 	if (flags & FS_DAX_FL)
1079 		xflags |= FS_XFLAG_DAX;
1080 	else
1081 		xflags &= ~FS_XFLAG_DAX;
1082 
1083 	return xflags;
1084 }
1085 
1086 STATIC unsigned int
1087 xfs_di2lxflags(
1088 	uint16_t	di_flags,
1089 	uint64_t	di_flags2)
1090 {
1091 	unsigned int	flags = 0;
1092 
1093 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
1094 		flags |= FS_IMMUTABLE_FL;
1095 	if (di_flags & XFS_DIFLAG_APPEND)
1096 		flags |= FS_APPEND_FL;
1097 	if (di_flags & XFS_DIFLAG_SYNC)
1098 		flags |= FS_SYNC_FL;
1099 	if (di_flags & XFS_DIFLAG_NOATIME)
1100 		flags |= FS_NOATIME_FL;
1101 	if (di_flags & XFS_DIFLAG_NODUMP)
1102 		flags |= FS_NODUMP_FL;
1103 	if (di_flags2 & XFS_DIFLAG2_DAX) {
1104 		flags |= FS_DAX_FL;
1105 	}
1106 	return flags;
1107 }
1108 
1109 static void
1110 xfs_fill_fsxattr(
1111 	struct xfs_inode	*ip,
1112 	bool			attr,
1113 	struct fsxattr		*fa)
1114 {
1115 	struct xfs_ifork	*ifp = attr ? ip->i_afp : &ip->i_df;
1116 
1117 	simple_fill_fsxattr(fa, xfs_ip2xflags(ip));
1118 	fa->fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
1119 	fa->fsx_cowextsize = ip->i_d.di_cowextsize <<
1120 			ip->i_mount->m_sb.sb_blocklog;
1121 	fa->fsx_projid = ip->i_d.di_projid;
1122 	if (ifp && (ifp->if_flags & XFS_IFEXTENTS))
1123 		fa->fsx_nextents = xfs_iext_count(ifp);
1124 	else
1125 		fa->fsx_nextents = xfs_ifork_nextents(ifp);
1126 }
1127 
1128 STATIC int
1129 xfs_ioc_fsgetxattr(
1130 	xfs_inode_t		*ip,
1131 	int			attr,
1132 	void			__user *arg)
1133 {
1134 	struct fsxattr		fa;
1135 
1136 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1137 	xfs_fill_fsxattr(ip, attr, &fa);
1138 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1139 
1140 	if (copy_to_user(arg, &fa, sizeof(fa)))
1141 		return -EFAULT;
1142 	return 0;
1143 }
1144 
1145 STATIC uint16_t
1146 xfs_flags2diflags(
1147 	struct xfs_inode	*ip,
1148 	unsigned int		xflags)
1149 {
1150 	/* can't set PREALLOC this way, just preserve it */
1151 	uint16_t		di_flags =
1152 		(ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
1153 
1154 	if (xflags & FS_XFLAG_IMMUTABLE)
1155 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1156 	if (xflags & FS_XFLAG_APPEND)
1157 		di_flags |= XFS_DIFLAG_APPEND;
1158 	if (xflags & FS_XFLAG_SYNC)
1159 		di_flags |= XFS_DIFLAG_SYNC;
1160 	if (xflags & FS_XFLAG_NOATIME)
1161 		di_flags |= XFS_DIFLAG_NOATIME;
1162 	if (xflags & FS_XFLAG_NODUMP)
1163 		di_flags |= XFS_DIFLAG_NODUMP;
1164 	if (xflags & FS_XFLAG_NODEFRAG)
1165 		di_flags |= XFS_DIFLAG_NODEFRAG;
1166 	if (xflags & FS_XFLAG_FILESTREAM)
1167 		di_flags |= XFS_DIFLAG_FILESTREAM;
1168 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1169 		if (xflags & FS_XFLAG_RTINHERIT)
1170 			di_flags |= XFS_DIFLAG_RTINHERIT;
1171 		if (xflags & FS_XFLAG_NOSYMLINKS)
1172 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1173 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1174 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1175 		if (xflags & FS_XFLAG_PROJINHERIT)
1176 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1177 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1178 		if (xflags & FS_XFLAG_REALTIME)
1179 			di_flags |= XFS_DIFLAG_REALTIME;
1180 		if (xflags & FS_XFLAG_EXTSIZE)
1181 			di_flags |= XFS_DIFLAG_EXTSIZE;
1182 	}
1183 
1184 	return di_flags;
1185 }
1186 
1187 STATIC uint64_t
1188 xfs_flags2diflags2(
1189 	struct xfs_inode	*ip,
1190 	unsigned int		xflags)
1191 {
1192 	uint64_t		di_flags2 =
1193 		(ip->i_d.di_flags2 & (XFS_DIFLAG2_REFLINK |
1194 				      XFS_DIFLAG2_BIGTIME));
1195 
1196 	if (xflags & FS_XFLAG_DAX)
1197 		di_flags2 |= XFS_DIFLAG2_DAX;
1198 	if (xflags & FS_XFLAG_COWEXTSIZE)
1199 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1200 
1201 	return di_flags2;
1202 }
1203 
1204 static int
1205 xfs_ioctl_setattr_xflags(
1206 	struct xfs_trans	*tp,
1207 	struct xfs_inode	*ip,
1208 	struct fsxattr		*fa)
1209 {
1210 	struct xfs_mount	*mp = ip->i_mount;
1211 	uint64_t		di_flags2;
1212 
1213 	/* Can't change realtime flag if any extents are allocated. */
1214 	if ((ip->i_df.if_nextents || ip->i_delayed_blks) &&
1215 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1216 		return -EINVAL;
1217 
1218 	/* If realtime flag is set then must have realtime device */
1219 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1220 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1221 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1222 			return -EINVAL;
1223 	}
1224 
1225 	/* Clear reflink if we are actually able to set the rt flag. */
1226 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1227 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1228 
1229 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1230 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1231 		return -EINVAL;
1232 
1233 	/* diflags2 only valid for v3 inodes. */
1234 	di_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1235 	if (di_flags2 && !xfs_sb_version_has_v3inode(&mp->m_sb))
1236 		return -EINVAL;
1237 
1238 	ip->i_d.di_flags = xfs_flags2diflags(ip, fa->fsx_xflags);
1239 	ip->i_d.di_flags2 = di_flags2;
1240 
1241 	xfs_diflags_to_iflags(ip, false);
1242 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1243 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1244 	XFS_STATS_INC(mp, xs_ig_attrchg);
1245 	return 0;
1246 }
1247 
1248 static void
1249 xfs_ioctl_setattr_prepare_dax(
1250 	struct xfs_inode	*ip,
1251 	struct fsxattr		*fa)
1252 {
1253 	struct xfs_mount	*mp = ip->i_mount;
1254 	struct inode            *inode = VFS_I(ip);
1255 
1256 	if (S_ISDIR(inode->i_mode))
1257 		return;
1258 
1259 	if ((mp->m_flags & XFS_MOUNT_DAX_ALWAYS) ||
1260 	    (mp->m_flags & XFS_MOUNT_DAX_NEVER))
1261 		return;
1262 
1263 	if (((fa->fsx_xflags & FS_XFLAG_DAX) &&
1264 	    !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)) ||
1265 	    (!(fa->fsx_xflags & FS_XFLAG_DAX) &&
1266 	     (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)))
1267 		d_mark_dontcache(inode);
1268 }
1269 
1270 /*
1271  * Set up the transaction structure for the setattr operation, checking that we
1272  * have permission to do so. On success, return a clean transaction and the
1273  * inode locked exclusively ready for further operation specific checks. On
1274  * failure, return an error without modifying or locking the inode.
1275  */
1276 static struct xfs_trans *
1277 xfs_ioctl_setattr_get_trans(
1278 	struct xfs_inode	*ip)
1279 {
1280 	struct xfs_mount	*mp = ip->i_mount;
1281 	struct xfs_trans	*tp;
1282 	int			error = -EROFS;
1283 
1284 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1285 		goto out_unlock;
1286 	error = -EIO;
1287 	if (XFS_FORCED_SHUTDOWN(mp))
1288 		goto out_unlock;
1289 
1290 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1291 	if (error)
1292 		goto out_unlock;
1293 
1294 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1295 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1296 
1297 	/*
1298 	 * CAP_FOWNER overrides the following restrictions:
1299 	 *
1300 	 * The user ID of the calling process must be equal to the file owner
1301 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1302 	 */
1303 	if (!inode_owner_or_capable(VFS_I(ip))) {
1304 		error = -EPERM;
1305 		goto out_cancel;
1306 	}
1307 
1308 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1309 		xfs_trans_set_sync(tp);
1310 
1311 	return tp;
1312 
1313 out_cancel:
1314 	xfs_trans_cancel(tp);
1315 out_unlock:
1316 	return ERR_PTR(error);
1317 }
1318 
1319 /*
1320  * extent size hint validation is somewhat cumbersome. Rules are:
1321  *
1322  * 1. extent size hint is only valid for directories and regular files
1323  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1324  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1325  * 4. can only be changed on regular files if no extents are allocated
1326  * 5. can be changed on directories at any time
1327  * 6. extsize hint of 0 turns off hints, clears inode flags.
1328  * 7. Extent size must be a multiple of the appropriate block size.
1329  * 8. for non-realtime files, the extent size hint must be limited
1330  *    to half the AG size to avoid alignment extending the extent beyond the
1331  *    limits of the AG.
1332  *
1333  * Please keep this function in sync with xfs_scrub_inode_extsize.
1334  */
1335 static int
1336 xfs_ioctl_setattr_check_extsize(
1337 	struct xfs_inode	*ip,
1338 	struct fsxattr		*fa)
1339 {
1340 	struct xfs_mount	*mp = ip->i_mount;
1341 	xfs_extlen_t		size;
1342 	xfs_fsblock_t		extsize_fsb;
1343 
1344 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_df.if_nextents &&
1345 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1346 		return -EINVAL;
1347 
1348 	if (fa->fsx_extsize == 0)
1349 		return 0;
1350 
1351 	extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1352 	if (extsize_fsb > MAXEXTLEN)
1353 		return -EINVAL;
1354 
1355 	if (XFS_IS_REALTIME_INODE(ip) ||
1356 	    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1357 		size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1358 	} else {
1359 		size = mp->m_sb.sb_blocksize;
1360 		if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1361 			return -EINVAL;
1362 	}
1363 
1364 	if (fa->fsx_extsize % size)
1365 		return -EINVAL;
1366 
1367 	return 0;
1368 }
1369 
1370 /*
1371  * CoW extent size hint validation rules are:
1372  *
1373  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
1374  *    The inode does not have to have any shared blocks, but it must be a v3.
1375  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
1376  *    for a directory, the hint is propagated to new files.
1377  * 3. Can be changed on files & directories at any time.
1378  * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
1379  * 5. Extent size must be a multiple of the appropriate block size.
1380  * 6. The extent size hint must be limited to half the AG size to avoid
1381  *    alignment extending the extent beyond the limits of the AG.
1382  *
1383  * Please keep this function in sync with xfs_scrub_inode_cowextsize.
1384  */
1385 static int
1386 xfs_ioctl_setattr_check_cowextsize(
1387 	struct xfs_inode	*ip,
1388 	struct fsxattr		*fa)
1389 {
1390 	struct xfs_mount	*mp = ip->i_mount;
1391 	xfs_extlen_t		size;
1392 	xfs_fsblock_t		cowextsize_fsb;
1393 
1394 	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
1395 		return 0;
1396 
1397 	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb))
1398 		return -EINVAL;
1399 
1400 	if (fa->fsx_cowextsize == 0)
1401 		return 0;
1402 
1403 	cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1404 	if (cowextsize_fsb > MAXEXTLEN)
1405 		return -EINVAL;
1406 
1407 	size = mp->m_sb.sb_blocksize;
1408 	if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
1409 		return -EINVAL;
1410 
1411 	if (fa->fsx_cowextsize % size)
1412 		return -EINVAL;
1413 
1414 	return 0;
1415 }
1416 
1417 static int
1418 xfs_ioctl_setattr_check_projid(
1419 	struct xfs_inode	*ip,
1420 	struct fsxattr		*fa)
1421 {
1422 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1423 	if (fa->fsx_projid > (uint16_t)-1 &&
1424 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1425 		return -EINVAL;
1426 	return 0;
1427 }
1428 
1429 STATIC int
1430 xfs_ioctl_setattr(
1431 	xfs_inode_t		*ip,
1432 	struct fsxattr		*fa)
1433 {
1434 	struct fsxattr		old_fa;
1435 	struct xfs_mount	*mp = ip->i_mount;
1436 	struct xfs_trans	*tp;
1437 	struct xfs_dquot	*pdqp = NULL;
1438 	struct xfs_dquot	*olddquot = NULL;
1439 	int			code;
1440 
1441 	trace_xfs_ioctl_setattr(ip);
1442 
1443 	code = xfs_ioctl_setattr_check_projid(ip, fa);
1444 	if (code)
1445 		return code;
1446 
1447 	/*
1448 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1449 	 * before we start any other transactions. Trying to do this later
1450 	 * is messy. We don't care to take a readlock to look at the ids
1451 	 * in inode here, because we can't hold it across the trans_reserve.
1452 	 * If the IDs do change before we take the ilock, we're covered
1453 	 * because the i_*dquot fields will get updated anyway.
1454 	 */
1455 	if (XFS_IS_QUOTA_ON(mp)) {
1456 		code = xfs_qm_vop_dqalloc(ip, VFS_I(ip)->i_uid,
1457 				VFS_I(ip)->i_gid, fa->fsx_projid,
1458 				XFS_QMOPT_PQUOTA, NULL, NULL, &pdqp);
1459 		if (code)
1460 			return code;
1461 	}
1462 
1463 	xfs_ioctl_setattr_prepare_dax(ip, fa);
1464 
1465 	tp = xfs_ioctl_setattr_get_trans(ip);
1466 	if (IS_ERR(tp)) {
1467 		code = PTR_ERR(tp);
1468 		goto error_free_dquots;
1469 	}
1470 
1471 	if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp) &&
1472 	    ip->i_d.di_projid != fa->fsx_projid) {
1473 		code = xfs_qm_vop_chown_reserve(tp, ip, NULL, NULL, pdqp,
1474 				capable(CAP_FOWNER) ?  XFS_QMOPT_FORCE_RES : 0);
1475 		if (code)	/* out of quota */
1476 			goto error_trans_cancel;
1477 	}
1478 
1479 	xfs_fill_fsxattr(ip, false, &old_fa);
1480 	code = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, fa);
1481 	if (code)
1482 		goto error_trans_cancel;
1483 
1484 	code = xfs_ioctl_setattr_check_extsize(ip, fa);
1485 	if (code)
1486 		goto error_trans_cancel;
1487 
1488 	code = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1489 	if (code)
1490 		goto error_trans_cancel;
1491 
1492 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
1493 	if (code)
1494 		goto error_trans_cancel;
1495 
1496 	/*
1497 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1498 	 * overrides the following restrictions:
1499 	 *
1500 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1501 	 * successful return from chown()
1502 	 */
1503 
1504 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1505 	    !capable_wrt_inode_uidgid(VFS_I(ip), CAP_FSETID))
1506 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1507 
1508 	/* Change the ownerships and register project quota modifications */
1509 	if (ip->i_d.di_projid != fa->fsx_projid) {
1510 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1511 			olddquot = xfs_qm_vop_chown(tp, ip,
1512 						&ip->i_pdquot, pdqp);
1513 		}
1514 		ip->i_d.di_projid = fa->fsx_projid;
1515 	}
1516 
1517 	/*
1518 	 * Only set the extent size hint if we've already determined that the
1519 	 * extent size hint should be set on the inode. If no extent size flags
1520 	 * are set on the inode then unconditionally clear the extent size hint.
1521 	 */
1522 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1523 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1524 	else
1525 		ip->i_d.di_extsize = 0;
1526 	if (xfs_sb_version_has_v3inode(&mp->m_sb) &&
1527 	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1528 		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
1529 				mp->m_sb.sb_blocklog;
1530 	else
1531 		ip->i_d.di_cowextsize = 0;
1532 
1533 	code = xfs_trans_commit(tp);
1534 
1535 	/*
1536 	 * Release any dquot(s) the inode had kept before chown.
1537 	 */
1538 	xfs_qm_dqrele(olddquot);
1539 	xfs_qm_dqrele(pdqp);
1540 
1541 	return code;
1542 
1543 error_trans_cancel:
1544 	xfs_trans_cancel(tp);
1545 error_free_dquots:
1546 	xfs_qm_dqrele(pdqp);
1547 	return code;
1548 }
1549 
1550 STATIC int
1551 xfs_ioc_fssetxattr(
1552 	xfs_inode_t		*ip,
1553 	struct file		*filp,
1554 	void			__user *arg)
1555 {
1556 	struct fsxattr		fa;
1557 	int error;
1558 
1559 	if (copy_from_user(&fa, arg, sizeof(fa)))
1560 		return -EFAULT;
1561 
1562 	error = mnt_want_write_file(filp);
1563 	if (error)
1564 		return error;
1565 	error = xfs_ioctl_setattr(ip, &fa);
1566 	mnt_drop_write_file(filp);
1567 	return error;
1568 }
1569 
1570 STATIC int
1571 xfs_ioc_getxflags(
1572 	xfs_inode_t		*ip,
1573 	void			__user *arg)
1574 {
1575 	unsigned int		flags;
1576 
1577 	flags = xfs_di2lxflags(ip->i_d.di_flags, ip->i_d.di_flags2);
1578 	if (copy_to_user(arg, &flags, sizeof(flags)))
1579 		return -EFAULT;
1580 	return 0;
1581 }
1582 
1583 STATIC int
1584 xfs_ioc_setxflags(
1585 	struct xfs_inode	*ip,
1586 	struct file		*filp,
1587 	void			__user *arg)
1588 {
1589 	struct xfs_trans	*tp;
1590 	struct fsxattr		fa;
1591 	struct fsxattr		old_fa;
1592 	unsigned int		flags;
1593 	int			error;
1594 
1595 	if (copy_from_user(&flags, arg, sizeof(flags)))
1596 		return -EFAULT;
1597 
1598 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1599 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1600 		      FS_SYNC_FL | FS_DAX_FL))
1601 		return -EOPNOTSUPP;
1602 
1603 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1604 
1605 	error = mnt_want_write_file(filp);
1606 	if (error)
1607 		return error;
1608 
1609 	xfs_ioctl_setattr_prepare_dax(ip, &fa);
1610 
1611 	tp = xfs_ioctl_setattr_get_trans(ip);
1612 	if (IS_ERR(tp)) {
1613 		error = PTR_ERR(tp);
1614 		goto out_drop_write;
1615 	}
1616 
1617 	xfs_fill_fsxattr(ip, false, &old_fa);
1618 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, &fa);
1619 	if (error) {
1620 		xfs_trans_cancel(tp);
1621 		goto out_drop_write;
1622 	}
1623 
1624 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1625 	if (error) {
1626 		xfs_trans_cancel(tp);
1627 		goto out_drop_write;
1628 	}
1629 
1630 	error = xfs_trans_commit(tp);
1631 out_drop_write:
1632 	mnt_drop_write_file(filp);
1633 	return error;
1634 }
1635 
1636 static bool
1637 xfs_getbmap_format(
1638 	struct kgetbmap		*p,
1639 	struct getbmapx __user	*u,
1640 	size_t			recsize)
1641 {
1642 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1643 	    put_user(p->bmv_block, &u->bmv_block) ||
1644 	    put_user(p->bmv_length, &u->bmv_length) ||
1645 	    put_user(0, &u->bmv_count) ||
1646 	    put_user(0, &u->bmv_entries))
1647 		return false;
1648 	if (recsize < sizeof(struct getbmapx))
1649 		return true;
1650 	if (put_user(0, &u->bmv_iflags) ||
1651 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1652 	    put_user(0, &u->bmv_unused1) ||
1653 	    put_user(0, &u->bmv_unused2))
1654 		return false;
1655 	return true;
1656 }
1657 
1658 STATIC int
1659 xfs_ioc_getbmap(
1660 	struct file		*file,
1661 	unsigned int		cmd,
1662 	void			__user *arg)
1663 {
1664 	struct getbmapx		bmx = { 0 };
1665 	struct kgetbmap		*buf;
1666 	size_t			recsize;
1667 	int			error, i;
1668 
1669 	switch (cmd) {
1670 	case XFS_IOC_GETBMAPA:
1671 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1672 		/*FALLTHRU*/
1673 	case XFS_IOC_GETBMAP:
1674 		if (file->f_mode & FMODE_NOCMTIME)
1675 			bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1676 		/* struct getbmap is a strict subset of struct getbmapx. */
1677 		recsize = sizeof(struct getbmap);
1678 		break;
1679 	case XFS_IOC_GETBMAPX:
1680 		recsize = sizeof(struct getbmapx);
1681 		break;
1682 	default:
1683 		return -EINVAL;
1684 	}
1685 
1686 	if (copy_from_user(&bmx, arg, recsize))
1687 		return -EFAULT;
1688 
1689 	if (bmx.bmv_count < 2)
1690 		return -EINVAL;
1691 	if (bmx.bmv_count > ULONG_MAX / recsize)
1692 		return -ENOMEM;
1693 
1694 	buf = kvzalloc(bmx.bmv_count * sizeof(*buf), GFP_KERNEL);
1695 	if (!buf)
1696 		return -ENOMEM;
1697 
1698 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1699 	if (error)
1700 		goto out_free_buf;
1701 
1702 	error = -EFAULT;
1703 	if (copy_to_user(arg, &bmx, recsize))
1704 		goto out_free_buf;
1705 	arg += recsize;
1706 
1707 	for (i = 0; i < bmx.bmv_entries; i++) {
1708 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1709 			goto out_free_buf;
1710 		arg += recsize;
1711 	}
1712 
1713 	error = 0;
1714 out_free_buf:
1715 	kmem_free(buf);
1716 	return error;
1717 }
1718 
1719 STATIC int
1720 xfs_ioc_getfsmap(
1721 	struct xfs_inode	*ip,
1722 	struct fsmap_head	__user *arg)
1723 {
1724 	struct xfs_fsmap_head	xhead = {0};
1725 	struct fsmap_head	head;
1726 	struct fsmap		*recs;
1727 	unsigned int		count;
1728 	__u32			last_flags = 0;
1729 	bool			done = false;
1730 	int			error;
1731 
1732 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1733 		return -EFAULT;
1734 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1735 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1736 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1737 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1738 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1739 		return -EINVAL;
1740 
1741 	/*
1742 	 * Use an internal memory buffer so that we don't have to copy fsmap
1743 	 * data to userspace while holding locks.  Start by trying to allocate
1744 	 * up to 128k for the buffer, but fall back to a single page if needed.
1745 	 */
1746 	count = min_t(unsigned int, head.fmh_count,
1747 			131072 / sizeof(struct fsmap));
1748 	recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1749 	if (!recs) {
1750 		count = min_t(unsigned int, head.fmh_count,
1751 				PAGE_SIZE / sizeof(struct fsmap));
1752 		recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1753 		if (!recs)
1754 			return -ENOMEM;
1755 	}
1756 
1757 	xhead.fmh_iflags = head.fmh_iflags;
1758 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1759 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1760 
1761 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1762 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1763 
1764 	head.fmh_entries = 0;
1765 	do {
1766 		struct fsmap __user	*user_recs;
1767 		struct fsmap		*last_rec;
1768 
1769 		user_recs = &arg->fmh_recs[head.fmh_entries];
1770 		xhead.fmh_entries = 0;
1771 		xhead.fmh_count = min_t(unsigned int, count,
1772 					head.fmh_count - head.fmh_entries);
1773 
1774 		/* Run query, record how many entries we got. */
1775 		error = xfs_getfsmap(ip->i_mount, &xhead, recs);
1776 		switch (error) {
1777 		case 0:
1778 			/*
1779 			 * There are no more records in the result set.  Copy
1780 			 * whatever we got to userspace and break out.
1781 			 */
1782 			done = true;
1783 			break;
1784 		case -ECANCELED:
1785 			/*
1786 			 * The internal memory buffer is full.  Copy whatever
1787 			 * records we got to userspace and go again if we have
1788 			 * not yet filled the userspace buffer.
1789 			 */
1790 			error = 0;
1791 			break;
1792 		default:
1793 			goto out_free;
1794 		}
1795 		head.fmh_entries += xhead.fmh_entries;
1796 		head.fmh_oflags = xhead.fmh_oflags;
1797 
1798 		/*
1799 		 * If the caller wanted a record count or there aren't any
1800 		 * new records to return, we're done.
1801 		 */
1802 		if (head.fmh_count == 0 || xhead.fmh_entries == 0)
1803 			break;
1804 
1805 		/* Copy all the records we got out to userspace. */
1806 		if (copy_to_user(user_recs, recs,
1807 				 xhead.fmh_entries * sizeof(struct fsmap))) {
1808 			error = -EFAULT;
1809 			goto out_free;
1810 		}
1811 
1812 		/* Remember the last record flags we copied to userspace. */
1813 		last_rec = &recs[xhead.fmh_entries - 1];
1814 		last_flags = last_rec->fmr_flags;
1815 
1816 		/* Set up the low key for the next iteration. */
1817 		xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec);
1818 		trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1819 	} while (!done && head.fmh_entries < head.fmh_count);
1820 
1821 	/*
1822 	 * If there are no more records in the query result set and we're not
1823 	 * in counting mode, mark the last record returned with the LAST flag.
1824 	 */
1825 	if (done && head.fmh_count > 0 && head.fmh_entries > 0) {
1826 		struct fsmap __user	*user_rec;
1827 
1828 		last_flags |= FMR_OF_LAST;
1829 		user_rec = &arg->fmh_recs[head.fmh_entries - 1];
1830 
1831 		if (copy_to_user(&user_rec->fmr_flags, &last_flags,
1832 					sizeof(last_flags))) {
1833 			error = -EFAULT;
1834 			goto out_free;
1835 		}
1836 	}
1837 
1838 	/* copy back header */
1839 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) {
1840 		error = -EFAULT;
1841 		goto out_free;
1842 	}
1843 
1844 out_free:
1845 	kmem_free(recs);
1846 	return error;
1847 }
1848 
1849 STATIC int
1850 xfs_ioc_scrub_metadata(
1851 	struct xfs_inode		*ip,
1852 	void				__user *arg)
1853 {
1854 	struct xfs_scrub_metadata	scrub;
1855 	int				error;
1856 
1857 	if (!capable(CAP_SYS_ADMIN))
1858 		return -EPERM;
1859 
1860 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1861 		return -EFAULT;
1862 
1863 	error = xfs_scrub_metadata(ip, &scrub);
1864 	if (error)
1865 		return error;
1866 
1867 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1868 		return -EFAULT;
1869 
1870 	return 0;
1871 }
1872 
1873 int
1874 xfs_ioc_swapext(
1875 	xfs_swapext_t	*sxp)
1876 {
1877 	xfs_inode_t     *ip, *tip;
1878 	struct fd	f, tmp;
1879 	int		error = 0;
1880 
1881 	/* Pull information for the target fd */
1882 	f = fdget((int)sxp->sx_fdtarget);
1883 	if (!f.file) {
1884 		error = -EINVAL;
1885 		goto out;
1886 	}
1887 
1888 	if (!(f.file->f_mode & FMODE_WRITE) ||
1889 	    !(f.file->f_mode & FMODE_READ) ||
1890 	    (f.file->f_flags & O_APPEND)) {
1891 		error = -EBADF;
1892 		goto out_put_file;
1893 	}
1894 
1895 	tmp = fdget((int)sxp->sx_fdtmp);
1896 	if (!tmp.file) {
1897 		error = -EINVAL;
1898 		goto out_put_file;
1899 	}
1900 
1901 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1902 	    !(tmp.file->f_mode & FMODE_READ) ||
1903 	    (tmp.file->f_flags & O_APPEND)) {
1904 		error = -EBADF;
1905 		goto out_put_tmp_file;
1906 	}
1907 
1908 	if (IS_SWAPFILE(file_inode(f.file)) ||
1909 	    IS_SWAPFILE(file_inode(tmp.file))) {
1910 		error = -EINVAL;
1911 		goto out_put_tmp_file;
1912 	}
1913 
1914 	/*
1915 	 * We need to ensure that the fds passed in point to XFS inodes
1916 	 * before we cast and access them as XFS structures as we have no
1917 	 * control over what the user passes us here.
1918 	 */
1919 	if (f.file->f_op != &xfs_file_operations ||
1920 	    tmp.file->f_op != &xfs_file_operations) {
1921 		error = -EINVAL;
1922 		goto out_put_tmp_file;
1923 	}
1924 
1925 	ip = XFS_I(file_inode(f.file));
1926 	tip = XFS_I(file_inode(tmp.file));
1927 
1928 	if (ip->i_mount != tip->i_mount) {
1929 		error = -EINVAL;
1930 		goto out_put_tmp_file;
1931 	}
1932 
1933 	if (ip->i_ino == tip->i_ino) {
1934 		error = -EINVAL;
1935 		goto out_put_tmp_file;
1936 	}
1937 
1938 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1939 		error = -EIO;
1940 		goto out_put_tmp_file;
1941 	}
1942 
1943 	error = xfs_swap_extents(ip, tip, sxp);
1944 
1945  out_put_tmp_file:
1946 	fdput(tmp);
1947  out_put_file:
1948 	fdput(f);
1949  out:
1950 	return error;
1951 }
1952 
1953 static int
1954 xfs_ioc_getlabel(
1955 	struct xfs_mount	*mp,
1956 	char			__user *user_label)
1957 {
1958 	struct xfs_sb		*sbp = &mp->m_sb;
1959 	char			label[XFSLABEL_MAX + 1];
1960 
1961 	/* Paranoia */
1962 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
1963 
1964 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
1965 	memset(label, 0, sizeof(label));
1966 	spin_lock(&mp->m_sb_lock);
1967 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
1968 	spin_unlock(&mp->m_sb_lock);
1969 
1970 	if (copy_to_user(user_label, label, sizeof(label)))
1971 		return -EFAULT;
1972 	return 0;
1973 }
1974 
1975 static int
1976 xfs_ioc_setlabel(
1977 	struct file		*filp,
1978 	struct xfs_mount	*mp,
1979 	char			__user *newlabel)
1980 {
1981 	struct xfs_sb		*sbp = &mp->m_sb;
1982 	char			label[XFSLABEL_MAX + 1];
1983 	size_t			len;
1984 	int			error;
1985 
1986 	if (!capable(CAP_SYS_ADMIN))
1987 		return -EPERM;
1988 	/*
1989 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
1990 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
1991 	 * (required) NULL character to test the incoming label length.
1992 	 * NB: The on disk label doesn't need to be null terminated.
1993 	 */
1994 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
1995 		return -EFAULT;
1996 	len = strnlen(label, XFSLABEL_MAX + 1);
1997 	if (len > sizeof(sbp->sb_fname))
1998 		return -EINVAL;
1999 
2000 	error = mnt_want_write_file(filp);
2001 	if (error)
2002 		return error;
2003 
2004 	spin_lock(&mp->m_sb_lock);
2005 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
2006 	memcpy(sbp->sb_fname, label, len);
2007 	spin_unlock(&mp->m_sb_lock);
2008 
2009 	/*
2010 	 * Now we do several things to satisfy userspace.
2011 	 * In addition to normal logging of the primary superblock, we also
2012 	 * immediately write these changes to sector zero for the primary, then
2013 	 * update all backup supers (as xfs_db does for a label change), then
2014 	 * invalidate the block device page cache.  This is so that any prior
2015 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
2016 	 * and userspace will see the newly-written label.
2017 	 */
2018 	error = xfs_sync_sb_buf(mp);
2019 	if (error)
2020 		goto out;
2021 	/*
2022 	 * growfs also updates backup supers so lock against that.
2023 	 */
2024 	mutex_lock(&mp->m_growlock);
2025 	error = xfs_update_secondary_sbs(mp);
2026 	mutex_unlock(&mp->m_growlock);
2027 
2028 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
2029 
2030 out:
2031 	mnt_drop_write_file(filp);
2032 	return error;
2033 }
2034 
2035 static inline int
2036 xfs_fs_eofblocks_from_user(
2037 	struct xfs_fs_eofblocks		*src,
2038 	struct xfs_eofblocks		*dst)
2039 {
2040 	if (src->eof_version != XFS_EOFBLOCKS_VERSION)
2041 		return -EINVAL;
2042 
2043 	if (src->eof_flags & ~XFS_EOF_FLAGS_VALID)
2044 		return -EINVAL;
2045 
2046 	if (memchr_inv(&src->pad32, 0, sizeof(src->pad32)) ||
2047 	    memchr_inv(src->pad64, 0, sizeof(src->pad64)))
2048 		return -EINVAL;
2049 
2050 	dst->eof_flags = src->eof_flags;
2051 	dst->eof_prid = src->eof_prid;
2052 	dst->eof_min_file_size = src->eof_min_file_size;
2053 
2054 	dst->eof_uid = INVALID_UID;
2055 	if (src->eof_flags & XFS_EOF_FLAGS_UID) {
2056 		dst->eof_uid = make_kuid(current_user_ns(), src->eof_uid);
2057 		if (!uid_valid(dst->eof_uid))
2058 			return -EINVAL;
2059 	}
2060 
2061 	dst->eof_gid = INVALID_GID;
2062 	if (src->eof_flags & XFS_EOF_FLAGS_GID) {
2063 		dst->eof_gid = make_kgid(current_user_ns(), src->eof_gid);
2064 		if (!gid_valid(dst->eof_gid))
2065 			return -EINVAL;
2066 	}
2067 	return 0;
2068 }
2069 
2070 /*
2071  * Note: some of the ioctl's return positive numbers as a
2072  * byte count indicating success, such as readlink_by_handle.
2073  * So we don't "sign flip" like most other routines.  This means
2074  * true errors need to be returned as a negative value.
2075  */
2076 long
2077 xfs_file_ioctl(
2078 	struct file		*filp,
2079 	unsigned int		cmd,
2080 	unsigned long		p)
2081 {
2082 	struct inode		*inode = file_inode(filp);
2083 	struct xfs_inode	*ip = XFS_I(inode);
2084 	struct xfs_mount	*mp = ip->i_mount;
2085 	void			__user *arg = (void __user *)p;
2086 	int			error;
2087 
2088 	trace_xfs_file_ioctl(ip);
2089 
2090 	switch (cmd) {
2091 	case FITRIM:
2092 		return xfs_ioc_trim(mp, arg);
2093 	case FS_IOC_GETFSLABEL:
2094 		return xfs_ioc_getlabel(mp, arg);
2095 	case FS_IOC_SETFSLABEL:
2096 		return xfs_ioc_setlabel(filp, mp, arg);
2097 	case XFS_IOC_ALLOCSP:
2098 	case XFS_IOC_FREESP:
2099 	case XFS_IOC_ALLOCSP64:
2100 	case XFS_IOC_FREESP64: {
2101 		xfs_flock64_t		bf;
2102 
2103 		if (copy_from_user(&bf, arg, sizeof(bf)))
2104 			return -EFAULT;
2105 		return xfs_ioc_space(filp, &bf);
2106 	}
2107 	case XFS_IOC_DIOINFO: {
2108 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
2109 		struct dioattr		da;
2110 
2111 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
2112 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
2113 
2114 		if (copy_to_user(arg, &da, sizeof(da)))
2115 			return -EFAULT;
2116 		return 0;
2117 	}
2118 
2119 	case XFS_IOC_FSBULKSTAT_SINGLE:
2120 	case XFS_IOC_FSBULKSTAT:
2121 	case XFS_IOC_FSINUMBERS:
2122 		return xfs_ioc_fsbulkstat(mp, cmd, arg);
2123 
2124 	case XFS_IOC_BULKSTAT:
2125 		return xfs_ioc_bulkstat(mp, cmd, arg);
2126 	case XFS_IOC_INUMBERS:
2127 		return xfs_ioc_inumbers(mp, cmd, arg);
2128 
2129 	case XFS_IOC_FSGEOMETRY_V1:
2130 		return xfs_ioc_fsgeometry(mp, arg, 3);
2131 	case XFS_IOC_FSGEOMETRY_V4:
2132 		return xfs_ioc_fsgeometry(mp, arg, 4);
2133 	case XFS_IOC_FSGEOMETRY:
2134 		return xfs_ioc_fsgeometry(mp, arg, 5);
2135 
2136 	case XFS_IOC_AG_GEOMETRY:
2137 		return xfs_ioc_ag_geometry(mp, arg);
2138 
2139 	case XFS_IOC_GETVERSION:
2140 		return put_user(inode->i_generation, (int __user *)arg);
2141 
2142 	case XFS_IOC_FSGETXATTR:
2143 		return xfs_ioc_fsgetxattr(ip, 0, arg);
2144 	case XFS_IOC_FSGETXATTRA:
2145 		return xfs_ioc_fsgetxattr(ip, 1, arg);
2146 	case XFS_IOC_FSSETXATTR:
2147 		return xfs_ioc_fssetxattr(ip, filp, arg);
2148 	case XFS_IOC_GETXFLAGS:
2149 		return xfs_ioc_getxflags(ip, arg);
2150 	case XFS_IOC_SETXFLAGS:
2151 		return xfs_ioc_setxflags(ip, filp, arg);
2152 
2153 	case XFS_IOC_GETBMAP:
2154 	case XFS_IOC_GETBMAPA:
2155 	case XFS_IOC_GETBMAPX:
2156 		return xfs_ioc_getbmap(filp, cmd, arg);
2157 
2158 	case FS_IOC_GETFSMAP:
2159 		return xfs_ioc_getfsmap(ip, arg);
2160 
2161 	case XFS_IOC_SCRUB_METADATA:
2162 		return xfs_ioc_scrub_metadata(ip, arg);
2163 
2164 	case XFS_IOC_FD_TO_HANDLE:
2165 	case XFS_IOC_PATH_TO_HANDLE:
2166 	case XFS_IOC_PATH_TO_FSHANDLE: {
2167 		xfs_fsop_handlereq_t	hreq;
2168 
2169 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2170 			return -EFAULT;
2171 		return xfs_find_handle(cmd, &hreq);
2172 	}
2173 	case XFS_IOC_OPEN_BY_HANDLE: {
2174 		xfs_fsop_handlereq_t	hreq;
2175 
2176 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2177 			return -EFAULT;
2178 		return xfs_open_by_handle(filp, &hreq);
2179 	}
2180 
2181 	case XFS_IOC_READLINK_BY_HANDLE: {
2182 		xfs_fsop_handlereq_t	hreq;
2183 
2184 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2185 			return -EFAULT;
2186 		return xfs_readlink_by_handle(filp, &hreq);
2187 	}
2188 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2189 		return xfs_attrlist_by_handle(filp, arg);
2190 
2191 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2192 		return xfs_attrmulti_by_handle(filp, arg);
2193 
2194 	case XFS_IOC_SWAPEXT: {
2195 		struct xfs_swapext	sxp;
2196 
2197 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2198 			return -EFAULT;
2199 		error = mnt_want_write_file(filp);
2200 		if (error)
2201 			return error;
2202 		error = xfs_ioc_swapext(&sxp);
2203 		mnt_drop_write_file(filp);
2204 		return error;
2205 	}
2206 
2207 	case XFS_IOC_FSCOUNTS: {
2208 		xfs_fsop_counts_t out;
2209 
2210 		xfs_fs_counts(mp, &out);
2211 
2212 		if (copy_to_user(arg, &out, sizeof(out)))
2213 			return -EFAULT;
2214 		return 0;
2215 	}
2216 
2217 	case XFS_IOC_SET_RESBLKS: {
2218 		xfs_fsop_resblks_t inout;
2219 		uint64_t	   in;
2220 
2221 		if (!capable(CAP_SYS_ADMIN))
2222 			return -EPERM;
2223 
2224 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2225 			return -EROFS;
2226 
2227 		if (copy_from_user(&inout, arg, sizeof(inout)))
2228 			return -EFAULT;
2229 
2230 		error = mnt_want_write_file(filp);
2231 		if (error)
2232 			return error;
2233 
2234 		/* input parameter is passed in resblks field of structure */
2235 		in = inout.resblks;
2236 		error = xfs_reserve_blocks(mp, &in, &inout);
2237 		mnt_drop_write_file(filp);
2238 		if (error)
2239 			return error;
2240 
2241 		if (copy_to_user(arg, &inout, sizeof(inout)))
2242 			return -EFAULT;
2243 		return 0;
2244 	}
2245 
2246 	case XFS_IOC_GET_RESBLKS: {
2247 		xfs_fsop_resblks_t out;
2248 
2249 		if (!capable(CAP_SYS_ADMIN))
2250 			return -EPERM;
2251 
2252 		error = xfs_reserve_blocks(mp, NULL, &out);
2253 		if (error)
2254 			return error;
2255 
2256 		if (copy_to_user(arg, &out, sizeof(out)))
2257 			return -EFAULT;
2258 
2259 		return 0;
2260 	}
2261 
2262 	case XFS_IOC_FSGROWFSDATA: {
2263 		xfs_growfs_data_t in;
2264 
2265 		if (copy_from_user(&in, arg, sizeof(in)))
2266 			return -EFAULT;
2267 
2268 		error = mnt_want_write_file(filp);
2269 		if (error)
2270 			return error;
2271 		error = xfs_growfs_data(mp, &in);
2272 		mnt_drop_write_file(filp);
2273 		return error;
2274 	}
2275 
2276 	case XFS_IOC_FSGROWFSLOG: {
2277 		xfs_growfs_log_t in;
2278 
2279 		if (copy_from_user(&in, arg, sizeof(in)))
2280 			return -EFAULT;
2281 
2282 		error = mnt_want_write_file(filp);
2283 		if (error)
2284 			return error;
2285 		error = xfs_growfs_log(mp, &in);
2286 		mnt_drop_write_file(filp);
2287 		return error;
2288 	}
2289 
2290 	case XFS_IOC_FSGROWFSRT: {
2291 		xfs_growfs_rt_t in;
2292 
2293 		if (copy_from_user(&in, arg, sizeof(in)))
2294 			return -EFAULT;
2295 
2296 		error = mnt_want_write_file(filp);
2297 		if (error)
2298 			return error;
2299 		error = xfs_growfs_rt(mp, &in);
2300 		mnt_drop_write_file(filp);
2301 		return error;
2302 	}
2303 
2304 	case XFS_IOC_GOINGDOWN: {
2305 		uint32_t in;
2306 
2307 		if (!capable(CAP_SYS_ADMIN))
2308 			return -EPERM;
2309 
2310 		if (get_user(in, (uint32_t __user *)arg))
2311 			return -EFAULT;
2312 
2313 		return xfs_fs_goingdown(mp, in);
2314 	}
2315 
2316 	case XFS_IOC_ERROR_INJECTION: {
2317 		xfs_error_injection_t in;
2318 
2319 		if (!capable(CAP_SYS_ADMIN))
2320 			return -EPERM;
2321 
2322 		if (copy_from_user(&in, arg, sizeof(in)))
2323 			return -EFAULT;
2324 
2325 		return xfs_errortag_add(mp, in.errtag);
2326 	}
2327 
2328 	case XFS_IOC_ERROR_CLEARALL:
2329 		if (!capable(CAP_SYS_ADMIN))
2330 			return -EPERM;
2331 
2332 		return xfs_errortag_clearall(mp);
2333 
2334 	case XFS_IOC_FREE_EOFBLOCKS: {
2335 		struct xfs_fs_eofblocks eofb;
2336 		struct xfs_eofblocks keofb;
2337 
2338 		if (!capable(CAP_SYS_ADMIN))
2339 			return -EPERM;
2340 
2341 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2342 			return -EROFS;
2343 
2344 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2345 			return -EFAULT;
2346 
2347 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
2348 		if (error)
2349 			return error;
2350 
2351 		sb_start_write(mp->m_super);
2352 		error = xfs_icache_free_eofblocks(mp, &keofb);
2353 		sb_end_write(mp->m_super);
2354 		return error;
2355 	}
2356 
2357 	default:
2358 		return -ENOTTY;
2359 	}
2360 }
2361