xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 4e9b9de6)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_rtalloc.h"
15 #include "xfs_iwalk.h"
16 #include "xfs_itable.h"
17 #include "xfs_error.h"
18 #include "xfs_attr.h"
19 #include "xfs_bmap.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_fsops.h"
22 #include "xfs_discard.h"
23 #include "xfs_quota.h"
24 #include "xfs_export.h"
25 #include "xfs_trace.h"
26 #include "xfs_icache.h"
27 #include "xfs_trans.h"
28 #include "xfs_acl.h"
29 #include "xfs_btree.h"
30 #include <linux/fsmap.h>
31 #include "xfs_fsmap.h"
32 #include "scrub/xfs_scrub.h"
33 #include "xfs_sb.h"
34 #include "xfs_ag.h"
35 #include "xfs_health.h"
36 #include "xfs_reflink.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_da_format.h"
39 #include "xfs_da_btree.h"
40 
41 #include <linux/mount.h>
42 #include <linux/namei.h>
43 #include <linux/fileattr.h>
44 
45 /*
46  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
47  * a file or fs handle.
48  *
49  * XFS_IOC_PATH_TO_FSHANDLE
50  *    returns fs handle for a mount point or path within that mount point
51  * XFS_IOC_FD_TO_HANDLE
52  *    returns full handle for a FD opened in user space
53  * XFS_IOC_PATH_TO_HANDLE
54  *    returns full handle for a path
55  */
56 int
57 xfs_find_handle(
58 	unsigned int		cmd,
59 	xfs_fsop_handlereq_t	*hreq)
60 {
61 	int			hsize;
62 	xfs_handle_t		handle;
63 	struct inode		*inode;
64 	struct fd		f = {NULL};
65 	struct path		path;
66 	int			error;
67 	struct xfs_inode	*ip;
68 
69 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
70 		f = fdget(hreq->fd);
71 		if (!f.file)
72 			return -EBADF;
73 		inode = file_inode(f.file);
74 	} else {
75 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
76 		if (error)
77 			return error;
78 		inode = d_inode(path.dentry);
79 	}
80 	ip = XFS_I(inode);
81 
82 	/*
83 	 * We can only generate handles for inodes residing on a XFS filesystem,
84 	 * and only for regular files, directories or symbolic links.
85 	 */
86 	error = -EINVAL;
87 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
88 		goto out_put;
89 
90 	error = -EBADF;
91 	if (!S_ISREG(inode->i_mode) &&
92 	    !S_ISDIR(inode->i_mode) &&
93 	    !S_ISLNK(inode->i_mode))
94 		goto out_put;
95 
96 
97 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
98 
99 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
100 		/*
101 		 * This handle only contains an fsid, zero the rest.
102 		 */
103 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
104 		hsize = sizeof(xfs_fsid_t);
105 	} else {
106 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
107 					sizeof(handle.ha_fid.fid_len);
108 		handle.ha_fid.fid_pad = 0;
109 		handle.ha_fid.fid_gen = inode->i_generation;
110 		handle.ha_fid.fid_ino = ip->i_ino;
111 		hsize = sizeof(xfs_handle_t);
112 	}
113 
114 	error = -EFAULT;
115 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
116 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
117 		goto out_put;
118 
119 	error = 0;
120 
121  out_put:
122 	if (cmd == XFS_IOC_FD_TO_HANDLE)
123 		fdput(f);
124 	else
125 		path_put(&path);
126 	return error;
127 }
128 
129 /*
130  * No need to do permission checks on the various pathname components
131  * as the handle operations are privileged.
132  */
133 STATIC int
134 xfs_handle_acceptable(
135 	void			*context,
136 	struct dentry		*dentry)
137 {
138 	return 1;
139 }
140 
141 /*
142  * Convert userspace handle data into a dentry.
143  */
144 struct dentry *
145 xfs_handle_to_dentry(
146 	struct file		*parfilp,
147 	void __user		*uhandle,
148 	u32			hlen)
149 {
150 	xfs_handle_t		handle;
151 	struct xfs_fid64	fid;
152 
153 	/*
154 	 * Only allow handle opens under a directory.
155 	 */
156 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
157 		return ERR_PTR(-ENOTDIR);
158 
159 	if (hlen != sizeof(xfs_handle_t))
160 		return ERR_PTR(-EINVAL);
161 	if (copy_from_user(&handle, uhandle, hlen))
162 		return ERR_PTR(-EFAULT);
163 	if (handle.ha_fid.fid_len !=
164 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
165 		return ERR_PTR(-EINVAL);
166 
167 	memset(&fid, 0, sizeof(struct fid));
168 	fid.ino = handle.ha_fid.fid_ino;
169 	fid.gen = handle.ha_fid.fid_gen;
170 
171 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
172 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
173 			xfs_handle_acceptable, NULL);
174 }
175 
176 STATIC struct dentry *
177 xfs_handlereq_to_dentry(
178 	struct file		*parfilp,
179 	xfs_fsop_handlereq_t	*hreq)
180 {
181 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
182 }
183 
184 int
185 xfs_open_by_handle(
186 	struct file		*parfilp,
187 	xfs_fsop_handlereq_t	*hreq)
188 {
189 	const struct cred	*cred = current_cred();
190 	int			error;
191 	int			fd;
192 	int			permflag;
193 	struct file		*filp;
194 	struct inode		*inode;
195 	struct dentry		*dentry;
196 	fmode_t			fmode;
197 	struct path		path;
198 
199 	if (!capable(CAP_SYS_ADMIN))
200 		return -EPERM;
201 
202 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
203 	if (IS_ERR(dentry))
204 		return PTR_ERR(dentry);
205 	inode = d_inode(dentry);
206 
207 	/* Restrict xfs_open_by_handle to directories & regular files. */
208 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
209 		error = -EPERM;
210 		goto out_dput;
211 	}
212 
213 #if BITS_PER_LONG != 32
214 	hreq->oflags |= O_LARGEFILE;
215 #endif
216 
217 	permflag = hreq->oflags;
218 	fmode = OPEN_FMODE(permflag);
219 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
220 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
221 		error = -EPERM;
222 		goto out_dput;
223 	}
224 
225 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
226 		error = -EPERM;
227 		goto out_dput;
228 	}
229 
230 	/* Can't write directories. */
231 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
232 		error = -EISDIR;
233 		goto out_dput;
234 	}
235 
236 	fd = get_unused_fd_flags(0);
237 	if (fd < 0) {
238 		error = fd;
239 		goto out_dput;
240 	}
241 
242 	path.mnt = parfilp->f_path.mnt;
243 	path.dentry = dentry;
244 	filp = dentry_open(&path, hreq->oflags, cred);
245 	dput(dentry);
246 	if (IS_ERR(filp)) {
247 		put_unused_fd(fd);
248 		return PTR_ERR(filp);
249 	}
250 
251 	if (S_ISREG(inode->i_mode)) {
252 		filp->f_flags |= O_NOATIME;
253 		filp->f_mode |= FMODE_NOCMTIME;
254 	}
255 
256 	fd_install(fd, filp);
257 	return fd;
258 
259  out_dput:
260 	dput(dentry);
261 	return error;
262 }
263 
264 int
265 xfs_readlink_by_handle(
266 	struct file		*parfilp,
267 	xfs_fsop_handlereq_t	*hreq)
268 {
269 	struct dentry		*dentry;
270 	__u32			olen;
271 	int			error;
272 
273 	if (!capable(CAP_SYS_ADMIN))
274 		return -EPERM;
275 
276 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
277 	if (IS_ERR(dentry))
278 		return PTR_ERR(dentry);
279 
280 	/* Restrict this handle operation to symlinks only. */
281 	if (!d_is_symlink(dentry)) {
282 		error = -EINVAL;
283 		goto out_dput;
284 	}
285 
286 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
287 		error = -EFAULT;
288 		goto out_dput;
289 	}
290 
291 	error = vfs_readlink(dentry, hreq->ohandle, olen);
292 
293  out_dput:
294 	dput(dentry);
295 	return error;
296 }
297 
298 /*
299  * Format an attribute and copy it out to the user's buffer.
300  * Take care to check values and protect against them changing later,
301  * we may be reading them directly out of a user buffer.
302  */
303 static void
304 xfs_ioc_attr_put_listent(
305 	struct xfs_attr_list_context *context,
306 	int			flags,
307 	unsigned char		*name,
308 	int			namelen,
309 	int			valuelen)
310 {
311 	struct xfs_attrlist	*alist = context->buffer;
312 	struct xfs_attrlist_ent	*aep;
313 	int			arraytop;
314 
315 	ASSERT(!context->seen_enough);
316 	ASSERT(context->count >= 0);
317 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
318 	ASSERT(context->firstu >= sizeof(*alist));
319 	ASSERT(context->firstu <= context->bufsize);
320 
321 	/*
322 	 * Only list entries in the right namespace.
323 	 */
324 	if (context->attr_filter != (flags & XFS_ATTR_NSP_ONDISK_MASK))
325 		return;
326 
327 	arraytop = sizeof(*alist) +
328 			context->count * sizeof(alist->al_offset[0]);
329 
330 	/* decrement by the actual bytes used by the attr */
331 	context->firstu -= round_up(offsetof(struct xfs_attrlist_ent, a_name) +
332 			namelen + 1, sizeof(uint32_t));
333 	if (context->firstu < arraytop) {
334 		trace_xfs_attr_list_full(context);
335 		alist->al_more = 1;
336 		context->seen_enough = 1;
337 		return;
338 	}
339 
340 	aep = context->buffer + context->firstu;
341 	aep->a_valuelen = valuelen;
342 	memcpy(aep->a_name, name, namelen);
343 	aep->a_name[namelen] = 0;
344 	alist->al_offset[context->count++] = context->firstu;
345 	alist->al_count = context->count;
346 	trace_xfs_attr_list_add(context);
347 }
348 
349 static unsigned int
350 xfs_attr_filter(
351 	u32			ioc_flags)
352 {
353 	if (ioc_flags & XFS_IOC_ATTR_ROOT)
354 		return XFS_ATTR_ROOT;
355 	if (ioc_flags & XFS_IOC_ATTR_SECURE)
356 		return XFS_ATTR_SECURE;
357 	return 0;
358 }
359 
360 static unsigned int
361 xfs_attr_flags(
362 	u32			ioc_flags)
363 {
364 	if (ioc_flags & XFS_IOC_ATTR_CREATE)
365 		return XATTR_CREATE;
366 	if (ioc_flags & XFS_IOC_ATTR_REPLACE)
367 		return XATTR_REPLACE;
368 	return 0;
369 }
370 
371 int
372 xfs_ioc_attr_list(
373 	struct xfs_inode		*dp,
374 	void __user			*ubuf,
375 	int				bufsize,
376 	int				flags,
377 	struct xfs_attrlist_cursor __user *ucursor)
378 {
379 	struct xfs_attr_list_context	context = { };
380 	struct xfs_attrlist		*alist;
381 	void				*buffer;
382 	int				error;
383 
384 	if (bufsize < sizeof(struct xfs_attrlist) ||
385 	    bufsize > XFS_XATTR_LIST_MAX)
386 		return -EINVAL;
387 
388 	/*
389 	 * Reject flags, only allow namespaces.
390 	 */
391 	if (flags & ~(XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
392 		return -EINVAL;
393 	if (flags == (XFS_IOC_ATTR_ROOT | XFS_IOC_ATTR_SECURE))
394 		return -EINVAL;
395 
396 	/*
397 	 * Validate the cursor.
398 	 */
399 	if (copy_from_user(&context.cursor, ucursor, sizeof(context.cursor)))
400 		return -EFAULT;
401 	if (context.cursor.pad1 || context.cursor.pad2)
402 		return -EINVAL;
403 	if (!context.cursor.initted &&
404 	    (context.cursor.hashval || context.cursor.blkno ||
405 	     context.cursor.offset))
406 		return -EINVAL;
407 
408 	buffer = kvzalloc(bufsize, GFP_KERNEL);
409 	if (!buffer)
410 		return -ENOMEM;
411 
412 	/*
413 	 * Initialize the output buffer.
414 	 */
415 	context.dp = dp;
416 	context.resynch = 1;
417 	context.attr_filter = xfs_attr_filter(flags);
418 	context.buffer = buffer;
419 	context.bufsize = round_down(bufsize, sizeof(uint32_t));
420 	context.firstu = context.bufsize;
421 	context.put_listent = xfs_ioc_attr_put_listent;
422 
423 	alist = context.buffer;
424 	alist->al_count = 0;
425 	alist->al_more = 0;
426 	alist->al_offset[0] = context.bufsize;
427 
428 	error = xfs_attr_list(&context);
429 	if (error)
430 		goto out_free;
431 
432 	if (copy_to_user(ubuf, buffer, bufsize) ||
433 	    copy_to_user(ucursor, &context.cursor, sizeof(context.cursor)))
434 		error = -EFAULT;
435 out_free:
436 	kmem_free(buffer);
437 	return error;
438 }
439 
440 STATIC int
441 xfs_attrlist_by_handle(
442 	struct file		*parfilp,
443 	struct xfs_fsop_attrlist_handlereq __user *p)
444 {
445 	struct xfs_fsop_attrlist_handlereq al_hreq;
446 	struct dentry		*dentry;
447 	int			error = -ENOMEM;
448 
449 	if (!capable(CAP_SYS_ADMIN))
450 		return -EPERM;
451 	if (copy_from_user(&al_hreq, p, sizeof(al_hreq)))
452 		return -EFAULT;
453 
454 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
455 	if (IS_ERR(dentry))
456 		return PTR_ERR(dentry);
457 
458 	error = xfs_ioc_attr_list(XFS_I(d_inode(dentry)), al_hreq.buffer,
459 				  al_hreq.buflen, al_hreq.flags, &p->pos);
460 	dput(dentry);
461 	return error;
462 }
463 
464 static int
465 xfs_attrmulti_attr_get(
466 	struct inode		*inode,
467 	unsigned char		*name,
468 	unsigned char		__user *ubuf,
469 	uint32_t		*len,
470 	uint32_t		flags)
471 {
472 	struct xfs_da_args	args = {
473 		.dp		= XFS_I(inode),
474 		.attr_filter	= xfs_attr_filter(flags),
475 		.attr_flags	= xfs_attr_flags(flags),
476 		.name		= name,
477 		.namelen	= strlen(name),
478 		.valuelen	= *len,
479 	};
480 	int			error;
481 
482 	if (*len > XFS_XATTR_SIZE_MAX)
483 		return -EINVAL;
484 
485 	error = xfs_attr_get(&args);
486 	if (error)
487 		goto out_kfree;
488 
489 	*len = args.valuelen;
490 	if (copy_to_user(ubuf, args.value, args.valuelen))
491 		error = -EFAULT;
492 
493 out_kfree:
494 	kmem_free(args.value);
495 	return error;
496 }
497 
498 static int
499 xfs_attrmulti_attr_set(
500 	struct inode		*inode,
501 	unsigned char		*name,
502 	const unsigned char	__user *ubuf,
503 	uint32_t		len,
504 	uint32_t		flags)
505 {
506 	struct xfs_da_args	args = {
507 		.dp		= XFS_I(inode),
508 		.attr_filter	= xfs_attr_filter(flags),
509 		.attr_flags	= xfs_attr_flags(flags),
510 		.name		= name,
511 		.namelen	= strlen(name),
512 	};
513 	int			error;
514 
515 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
516 		return -EPERM;
517 
518 	if (ubuf) {
519 		if (len > XFS_XATTR_SIZE_MAX)
520 			return -EINVAL;
521 		args.value = memdup_user(ubuf, len);
522 		if (IS_ERR(args.value))
523 			return PTR_ERR(args.value);
524 		args.valuelen = len;
525 	}
526 
527 	error = xfs_attr_set(&args);
528 	if (!error && (flags & XFS_IOC_ATTR_ROOT))
529 		xfs_forget_acl(inode, name);
530 	kfree(args.value);
531 	return error;
532 }
533 
534 int
535 xfs_ioc_attrmulti_one(
536 	struct file		*parfilp,
537 	struct inode		*inode,
538 	uint32_t		opcode,
539 	void __user		*uname,
540 	void __user		*value,
541 	uint32_t		*len,
542 	uint32_t		flags)
543 {
544 	unsigned char		*name;
545 	int			error;
546 
547 	if ((flags & XFS_IOC_ATTR_ROOT) && (flags & XFS_IOC_ATTR_SECURE))
548 		return -EINVAL;
549 
550 	name = strndup_user(uname, MAXNAMELEN);
551 	if (IS_ERR(name))
552 		return PTR_ERR(name);
553 
554 	switch (opcode) {
555 	case ATTR_OP_GET:
556 		error = xfs_attrmulti_attr_get(inode, name, value, len, flags);
557 		break;
558 	case ATTR_OP_REMOVE:
559 		value = NULL;
560 		*len = 0;
561 		fallthrough;
562 	case ATTR_OP_SET:
563 		error = mnt_want_write_file(parfilp);
564 		if (error)
565 			break;
566 		error = xfs_attrmulti_attr_set(inode, name, value, *len, flags);
567 		mnt_drop_write_file(parfilp);
568 		break;
569 	default:
570 		error = -EINVAL;
571 		break;
572 	}
573 
574 	kfree(name);
575 	return error;
576 }
577 
578 STATIC int
579 xfs_attrmulti_by_handle(
580 	struct file		*parfilp,
581 	void			__user *arg)
582 {
583 	int			error;
584 	xfs_attr_multiop_t	*ops;
585 	xfs_fsop_attrmulti_handlereq_t am_hreq;
586 	struct dentry		*dentry;
587 	unsigned int		i, size;
588 
589 	if (!capable(CAP_SYS_ADMIN))
590 		return -EPERM;
591 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
592 		return -EFAULT;
593 
594 	/* overflow check */
595 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
596 		return -E2BIG;
597 
598 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
599 	if (IS_ERR(dentry))
600 		return PTR_ERR(dentry);
601 
602 	error = -E2BIG;
603 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
604 	if (!size || size > 16 * PAGE_SIZE)
605 		goto out_dput;
606 
607 	ops = memdup_user(am_hreq.ops, size);
608 	if (IS_ERR(ops)) {
609 		error = PTR_ERR(ops);
610 		goto out_dput;
611 	}
612 
613 	error = 0;
614 	for (i = 0; i < am_hreq.opcount; i++) {
615 		ops[i].am_error = xfs_ioc_attrmulti_one(parfilp,
616 				d_inode(dentry), ops[i].am_opcode,
617 				ops[i].am_attrname, ops[i].am_attrvalue,
618 				&ops[i].am_length, ops[i].am_flags);
619 	}
620 
621 	if (copy_to_user(am_hreq.ops, ops, size))
622 		error = -EFAULT;
623 
624 	kfree(ops);
625  out_dput:
626 	dput(dentry);
627 	return error;
628 }
629 
630 int
631 xfs_ioc_space(
632 	struct file		*filp,
633 	xfs_flock64_t		*bf)
634 {
635 	struct inode		*inode = file_inode(filp);
636 	struct xfs_inode	*ip = XFS_I(inode);
637 	struct iattr		iattr;
638 	enum xfs_prealloc_flags	flags = XFS_PREALLOC_CLEAR;
639 	uint			iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
640 	int			error;
641 
642 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND))
643 		return -EPERM;
644 
645 	if (!(filp->f_mode & FMODE_WRITE))
646 		return -EBADF;
647 
648 	if (!S_ISREG(inode->i_mode))
649 		return -EINVAL;
650 
651 	if (xfs_is_always_cow_inode(ip))
652 		return -EOPNOTSUPP;
653 
654 	if (filp->f_flags & O_DSYNC)
655 		flags |= XFS_PREALLOC_SYNC;
656 	if (filp->f_mode & FMODE_NOCMTIME)
657 		flags |= XFS_PREALLOC_INVISIBLE;
658 
659 	error = mnt_want_write_file(filp);
660 	if (error)
661 		return error;
662 
663 	xfs_ilock(ip, iolock);
664 	error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
665 	if (error)
666 		goto out_unlock;
667 	inode_dio_wait(inode);
668 
669 	switch (bf->l_whence) {
670 	case 0: /*SEEK_SET*/
671 		break;
672 	case 1: /*SEEK_CUR*/
673 		bf->l_start += filp->f_pos;
674 		break;
675 	case 2: /*SEEK_END*/
676 		bf->l_start += XFS_ISIZE(ip);
677 		break;
678 	default:
679 		error = -EINVAL;
680 		goto out_unlock;
681 	}
682 
683 	if (bf->l_start < 0 || bf->l_start > inode->i_sb->s_maxbytes) {
684 		error = -EINVAL;
685 		goto out_unlock;
686 	}
687 
688 	if (bf->l_start > XFS_ISIZE(ip)) {
689 		error = xfs_alloc_file_space(ip, XFS_ISIZE(ip),
690 				bf->l_start - XFS_ISIZE(ip), 0);
691 		if (error)
692 			goto out_unlock;
693 	}
694 
695 	iattr.ia_valid = ATTR_SIZE;
696 	iattr.ia_size = bf->l_start;
697 	error = xfs_vn_setattr_size(file_mnt_user_ns(filp), file_dentry(filp),
698 				    &iattr);
699 	if (error)
700 		goto out_unlock;
701 
702 	error = xfs_update_prealloc_flags(ip, flags);
703 
704 out_unlock:
705 	xfs_iunlock(ip, iolock);
706 	mnt_drop_write_file(filp);
707 	return error;
708 }
709 
710 /* Return 0 on success or positive error */
711 int
712 xfs_fsbulkstat_one_fmt(
713 	struct xfs_ibulk		*breq,
714 	const struct xfs_bulkstat	*bstat)
715 {
716 	struct xfs_bstat		bs1;
717 
718 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
719 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
720 		return -EFAULT;
721 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
722 }
723 
724 int
725 xfs_fsinumbers_fmt(
726 	struct xfs_ibulk		*breq,
727 	const struct xfs_inumbers	*igrp)
728 {
729 	struct xfs_inogrp		ig1;
730 
731 	xfs_inumbers_to_inogrp(&ig1, igrp);
732 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
733 		return -EFAULT;
734 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
735 }
736 
737 STATIC int
738 xfs_ioc_fsbulkstat(
739 	struct file		*file,
740 	unsigned int		cmd,
741 	void			__user *arg)
742 {
743 	struct xfs_mount	*mp = XFS_I(file_inode(file))->i_mount;
744 	struct xfs_fsop_bulkreq	bulkreq;
745 	struct xfs_ibulk	breq = {
746 		.mp		= mp,
747 		.mnt_userns	= file_mnt_user_ns(file),
748 		.ocount		= 0,
749 	};
750 	xfs_ino_t		lastino;
751 	int			error;
752 
753 	/* done = 1 if there are more stats to get and if bulkstat */
754 	/* should be called again (unused here, but used in dmapi) */
755 
756 	if (!capable(CAP_SYS_ADMIN))
757 		return -EPERM;
758 
759 	if (xfs_is_shutdown(mp))
760 		return -EIO;
761 
762 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
763 		return -EFAULT;
764 
765 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
766 		return -EFAULT;
767 
768 	if (bulkreq.icount <= 0)
769 		return -EINVAL;
770 
771 	if (bulkreq.ubuffer == NULL)
772 		return -EINVAL;
773 
774 	breq.ubuffer = bulkreq.ubuffer;
775 	breq.icount = bulkreq.icount;
776 
777 	/*
778 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
779 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
780 	 * that *lastip contains either zero or the number of the last inode to
781 	 * be examined by the previous call and return results starting with
782 	 * the next inode after that.  The new bulk request back end functions
783 	 * take the inode to start with, so we have to compute the startino
784 	 * parameter from lastino to maintain correct function.  lastino == 0
785 	 * is a special case because it has traditionally meant "first inode
786 	 * in filesystem".
787 	 */
788 	if (cmd == XFS_IOC_FSINUMBERS) {
789 		breq.startino = lastino ? lastino + 1 : 0;
790 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
791 		lastino = breq.startino - 1;
792 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
793 		breq.startino = lastino;
794 		breq.icount = 1;
795 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
796 	} else {	/* XFS_IOC_FSBULKSTAT */
797 		breq.startino = lastino ? lastino + 1 : 0;
798 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
799 		lastino = breq.startino - 1;
800 	}
801 
802 	if (error)
803 		return error;
804 
805 	if (bulkreq.lastip != NULL &&
806 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
807 		return -EFAULT;
808 
809 	if (bulkreq.ocount != NULL &&
810 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
811 		return -EFAULT;
812 
813 	return 0;
814 }
815 
816 /* Return 0 on success or positive error */
817 static int
818 xfs_bulkstat_fmt(
819 	struct xfs_ibulk		*breq,
820 	const struct xfs_bulkstat	*bstat)
821 {
822 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
823 		return -EFAULT;
824 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
825 }
826 
827 /*
828  * Check the incoming bulk request @hdr from userspace and initialize the
829  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
830  * should proceed; -ECANCELED if there's nothing to do; or the usual
831  * negative error code.
832  */
833 static int
834 xfs_bulk_ireq_setup(
835 	struct xfs_mount	*mp,
836 	struct xfs_bulk_ireq	*hdr,
837 	struct xfs_ibulk	*breq,
838 	void __user		*ubuffer)
839 {
840 	if (hdr->icount == 0 ||
841 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
842 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
843 		return -EINVAL;
844 
845 	breq->startino = hdr->ino;
846 	breq->ubuffer = ubuffer;
847 	breq->icount = hdr->icount;
848 	breq->ocount = 0;
849 	breq->flags = 0;
850 
851 	/*
852 	 * The @ino parameter is a special value, so we must look it up here.
853 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
854 	 * worth of data.
855 	 */
856 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
857 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
858 			return -EINVAL;
859 
860 		switch (hdr->ino) {
861 		case XFS_BULK_IREQ_SPECIAL_ROOT:
862 			hdr->ino = mp->m_sb.sb_rootino;
863 			break;
864 		default:
865 			return -EINVAL;
866 		}
867 		breq->icount = 1;
868 	}
869 
870 	/*
871 	 * The IREQ_AGNO flag means that we only want results from a given AG.
872 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
873 	 * beyond the specified AG then we return no results.
874 	 */
875 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
876 		if (hdr->agno >= mp->m_sb.sb_agcount)
877 			return -EINVAL;
878 
879 		if (breq->startino == 0)
880 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
881 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
882 			return -EINVAL;
883 
884 		breq->flags |= XFS_IBULK_SAME_AG;
885 
886 		/* Asking for an inode past the end of the AG?  We're done! */
887 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
888 			return -ECANCELED;
889 	} else if (hdr->agno)
890 		return -EINVAL;
891 
892 	/* Asking for an inode past the end of the FS?  We're done! */
893 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
894 		return -ECANCELED;
895 
896 	return 0;
897 }
898 
899 /*
900  * Update the userspace bulk request @hdr to reflect the end state of the
901  * internal bulk request @breq.
902  */
903 static void
904 xfs_bulk_ireq_teardown(
905 	struct xfs_bulk_ireq	*hdr,
906 	struct xfs_ibulk	*breq)
907 {
908 	hdr->ino = breq->startino;
909 	hdr->ocount = breq->ocount;
910 }
911 
912 /* Handle the v5 bulkstat ioctl. */
913 STATIC int
914 xfs_ioc_bulkstat(
915 	struct file			*file,
916 	unsigned int			cmd,
917 	struct xfs_bulkstat_req __user	*arg)
918 {
919 	struct xfs_mount		*mp = XFS_I(file_inode(file))->i_mount;
920 	struct xfs_bulk_ireq		hdr;
921 	struct xfs_ibulk		breq = {
922 		.mp			= mp,
923 		.mnt_userns		= file_mnt_user_ns(file),
924 	};
925 	int				error;
926 
927 	if (!capable(CAP_SYS_ADMIN))
928 		return -EPERM;
929 
930 	if (xfs_is_shutdown(mp))
931 		return -EIO;
932 
933 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
934 		return -EFAULT;
935 
936 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
937 	if (error == -ECANCELED)
938 		goto out_teardown;
939 	if (error < 0)
940 		return error;
941 
942 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
943 	if (error)
944 		return error;
945 
946 out_teardown:
947 	xfs_bulk_ireq_teardown(&hdr, &breq);
948 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
949 		return -EFAULT;
950 
951 	return 0;
952 }
953 
954 STATIC int
955 xfs_inumbers_fmt(
956 	struct xfs_ibulk		*breq,
957 	const struct xfs_inumbers	*igrp)
958 {
959 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
960 		return -EFAULT;
961 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
962 }
963 
964 /* Handle the v5 inumbers ioctl. */
965 STATIC int
966 xfs_ioc_inumbers(
967 	struct xfs_mount		*mp,
968 	unsigned int			cmd,
969 	struct xfs_inumbers_req __user	*arg)
970 {
971 	struct xfs_bulk_ireq		hdr;
972 	struct xfs_ibulk		breq = {
973 		.mp			= mp,
974 	};
975 	int				error;
976 
977 	if (!capable(CAP_SYS_ADMIN))
978 		return -EPERM;
979 
980 	if (xfs_is_shutdown(mp))
981 		return -EIO;
982 
983 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
984 		return -EFAULT;
985 
986 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
987 	if (error == -ECANCELED)
988 		goto out_teardown;
989 	if (error < 0)
990 		return error;
991 
992 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
993 	if (error)
994 		return error;
995 
996 out_teardown:
997 	xfs_bulk_ireq_teardown(&hdr, &breq);
998 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
999 		return -EFAULT;
1000 
1001 	return 0;
1002 }
1003 
1004 STATIC int
1005 xfs_ioc_fsgeometry(
1006 	struct xfs_mount	*mp,
1007 	void			__user *arg,
1008 	int			struct_version)
1009 {
1010 	struct xfs_fsop_geom	fsgeo;
1011 	size_t			len;
1012 
1013 	xfs_fs_geometry(mp, &fsgeo, struct_version);
1014 
1015 	if (struct_version <= 3)
1016 		len = sizeof(struct xfs_fsop_geom_v1);
1017 	else if (struct_version == 4)
1018 		len = sizeof(struct xfs_fsop_geom_v4);
1019 	else {
1020 		xfs_fsop_geom_health(mp, &fsgeo);
1021 		len = sizeof(fsgeo);
1022 	}
1023 
1024 	if (copy_to_user(arg, &fsgeo, len))
1025 		return -EFAULT;
1026 	return 0;
1027 }
1028 
1029 STATIC int
1030 xfs_ioc_ag_geometry(
1031 	struct xfs_mount	*mp,
1032 	void			__user *arg)
1033 {
1034 	struct xfs_ag_geometry	ageo;
1035 	int			error;
1036 
1037 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
1038 		return -EFAULT;
1039 	if (ageo.ag_flags)
1040 		return -EINVAL;
1041 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
1042 		return -EINVAL;
1043 
1044 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
1045 	if (error)
1046 		return error;
1047 
1048 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
1049 		return -EFAULT;
1050 	return 0;
1051 }
1052 
1053 /*
1054  * Linux extended inode flags interface.
1055  */
1056 
1057 static void
1058 xfs_fill_fsxattr(
1059 	struct xfs_inode	*ip,
1060 	int			whichfork,
1061 	struct fileattr		*fa)
1062 {
1063 	struct xfs_mount	*mp = ip->i_mount;
1064 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1065 
1066 	fileattr_fill_xflags(fa, xfs_ip2xflags(ip));
1067 
1068 	if (ip->i_diflags & XFS_DIFLAG_EXTSIZE) {
1069 		fa->fsx_extsize = XFS_FSB_TO_B(mp, ip->i_extsize);
1070 	} else if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
1071 		/*
1072 		 * Don't let a misaligned extent size hint on a directory
1073 		 * escape to userspace if it won't pass the setattr checks
1074 		 * later.
1075 		 */
1076 		if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
1077 		    ip->i_extsize % mp->m_sb.sb_rextsize > 0) {
1078 			fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE |
1079 					    FS_XFLAG_EXTSZINHERIT);
1080 			fa->fsx_extsize = 0;
1081 		} else {
1082 			fa->fsx_extsize = XFS_FSB_TO_B(mp, ip->i_extsize);
1083 		}
1084 	}
1085 
1086 	if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
1087 		fa->fsx_cowextsize = XFS_FSB_TO_B(mp, ip->i_cowextsize);
1088 	fa->fsx_projid = ip->i_projid;
1089 	if (ifp && !xfs_need_iread_extents(ifp))
1090 		fa->fsx_nextents = xfs_iext_count(ifp);
1091 	else
1092 		fa->fsx_nextents = xfs_ifork_nextents(ifp);
1093 }
1094 
1095 STATIC int
1096 xfs_ioc_fsgetxattra(
1097 	xfs_inode_t		*ip,
1098 	void			__user *arg)
1099 {
1100 	struct fileattr		fa;
1101 
1102 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1103 	xfs_fill_fsxattr(ip, XFS_ATTR_FORK, &fa);
1104 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1105 
1106 	return copy_fsxattr_to_user(&fa, arg);
1107 }
1108 
1109 int
1110 xfs_fileattr_get(
1111 	struct dentry		*dentry,
1112 	struct fileattr		*fa)
1113 {
1114 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1115 
1116 	if (d_is_special(dentry))
1117 		return -ENOTTY;
1118 
1119 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1120 	xfs_fill_fsxattr(ip, XFS_DATA_FORK, fa);
1121 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1122 
1123 	return 0;
1124 }
1125 
1126 STATIC uint16_t
1127 xfs_flags2diflags(
1128 	struct xfs_inode	*ip,
1129 	unsigned int		xflags)
1130 {
1131 	/* can't set PREALLOC this way, just preserve it */
1132 	uint16_t		di_flags =
1133 		(ip->i_diflags & XFS_DIFLAG_PREALLOC);
1134 
1135 	if (xflags & FS_XFLAG_IMMUTABLE)
1136 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1137 	if (xflags & FS_XFLAG_APPEND)
1138 		di_flags |= XFS_DIFLAG_APPEND;
1139 	if (xflags & FS_XFLAG_SYNC)
1140 		di_flags |= XFS_DIFLAG_SYNC;
1141 	if (xflags & FS_XFLAG_NOATIME)
1142 		di_flags |= XFS_DIFLAG_NOATIME;
1143 	if (xflags & FS_XFLAG_NODUMP)
1144 		di_flags |= XFS_DIFLAG_NODUMP;
1145 	if (xflags & FS_XFLAG_NODEFRAG)
1146 		di_flags |= XFS_DIFLAG_NODEFRAG;
1147 	if (xflags & FS_XFLAG_FILESTREAM)
1148 		di_flags |= XFS_DIFLAG_FILESTREAM;
1149 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1150 		if (xflags & FS_XFLAG_RTINHERIT)
1151 			di_flags |= XFS_DIFLAG_RTINHERIT;
1152 		if (xflags & FS_XFLAG_NOSYMLINKS)
1153 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1154 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1155 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1156 		if (xflags & FS_XFLAG_PROJINHERIT)
1157 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1158 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1159 		if (xflags & FS_XFLAG_REALTIME)
1160 			di_flags |= XFS_DIFLAG_REALTIME;
1161 		if (xflags & FS_XFLAG_EXTSIZE)
1162 			di_flags |= XFS_DIFLAG_EXTSIZE;
1163 	}
1164 
1165 	return di_flags;
1166 }
1167 
1168 STATIC uint64_t
1169 xfs_flags2diflags2(
1170 	struct xfs_inode	*ip,
1171 	unsigned int		xflags)
1172 {
1173 	uint64_t		di_flags2 =
1174 		(ip->i_diflags2 & (XFS_DIFLAG2_REFLINK |
1175 				   XFS_DIFLAG2_BIGTIME));
1176 
1177 	if (xflags & FS_XFLAG_DAX)
1178 		di_flags2 |= XFS_DIFLAG2_DAX;
1179 	if (xflags & FS_XFLAG_COWEXTSIZE)
1180 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1181 
1182 	return di_flags2;
1183 }
1184 
1185 static int
1186 xfs_ioctl_setattr_xflags(
1187 	struct xfs_trans	*tp,
1188 	struct xfs_inode	*ip,
1189 	struct fileattr		*fa)
1190 {
1191 	struct xfs_mount	*mp = ip->i_mount;
1192 	uint64_t		i_flags2;
1193 
1194 	/* Can't change realtime flag if any extents are allocated. */
1195 	if ((ip->i_df.if_nextents || ip->i_delayed_blks) &&
1196 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1197 		return -EINVAL;
1198 
1199 	/* If realtime flag is set then must have realtime device */
1200 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1201 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1202 		    (ip->i_extsize % mp->m_sb.sb_rextsize))
1203 			return -EINVAL;
1204 	}
1205 
1206 	/* Clear reflink if we are actually able to set the rt flag. */
1207 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1208 		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1209 
1210 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1211 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1212 		return -EINVAL;
1213 
1214 	/* diflags2 only valid for v3 inodes. */
1215 	i_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1216 	if (i_flags2 && !xfs_has_v3inodes(mp))
1217 		return -EINVAL;
1218 
1219 	ip->i_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1220 	ip->i_diflags2 = i_flags2;
1221 
1222 	xfs_diflags_to_iflags(ip, false);
1223 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1224 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1225 	XFS_STATS_INC(mp, xs_ig_attrchg);
1226 	return 0;
1227 }
1228 
1229 static void
1230 xfs_ioctl_setattr_prepare_dax(
1231 	struct xfs_inode	*ip,
1232 	struct fileattr		*fa)
1233 {
1234 	struct xfs_mount	*mp = ip->i_mount;
1235 	struct inode            *inode = VFS_I(ip);
1236 
1237 	if (S_ISDIR(inode->i_mode))
1238 		return;
1239 
1240 	if (xfs_has_dax_always(mp) || xfs_has_dax_never(mp))
1241 		return;
1242 
1243 	if (((fa->fsx_xflags & FS_XFLAG_DAX) &&
1244 	    !(ip->i_diflags2 & XFS_DIFLAG2_DAX)) ||
1245 	    (!(fa->fsx_xflags & FS_XFLAG_DAX) &&
1246 	     (ip->i_diflags2 & XFS_DIFLAG2_DAX)))
1247 		d_mark_dontcache(inode);
1248 }
1249 
1250 /*
1251  * Set up the transaction structure for the setattr operation, checking that we
1252  * have permission to do so. On success, return a clean transaction and the
1253  * inode locked exclusively ready for further operation specific checks. On
1254  * failure, return an error without modifying or locking the inode.
1255  */
1256 static struct xfs_trans *
1257 xfs_ioctl_setattr_get_trans(
1258 	struct xfs_inode	*ip,
1259 	struct xfs_dquot	*pdqp)
1260 {
1261 	struct xfs_mount	*mp = ip->i_mount;
1262 	struct xfs_trans	*tp;
1263 	int			error = -EROFS;
1264 
1265 	if (xfs_is_readonly(mp))
1266 		goto out_error;
1267 	error = -EIO;
1268 	if (xfs_is_shutdown(mp))
1269 		goto out_error;
1270 
1271 	error = xfs_trans_alloc_ichange(ip, NULL, NULL, pdqp,
1272 			capable(CAP_FOWNER), &tp);
1273 	if (error)
1274 		goto out_error;
1275 
1276 	if (xfs_has_wsync(mp))
1277 		xfs_trans_set_sync(tp);
1278 
1279 	return tp;
1280 
1281 out_error:
1282 	return ERR_PTR(error);
1283 }
1284 
1285 /*
1286  * Validate a proposed extent size hint.  For regular files, the hint can only
1287  * be changed if no extents are allocated.
1288  */
1289 static int
1290 xfs_ioctl_setattr_check_extsize(
1291 	struct xfs_inode	*ip,
1292 	struct fileattr		*fa)
1293 {
1294 	struct xfs_mount	*mp = ip->i_mount;
1295 	xfs_failaddr_t		failaddr;
1296 	uint16_t		new_diflags;
1297 
1298 	if (!fa->fsx_valid)
1299 		return 0;
1300 
1301 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_df.if_nextents &&
1302 	    XFS_FSB_TO_B(mp, ip->i_extsize) != fa->fsx_extsize)
1303 		return -EINVAL;
1304 
1305 	if (fa->fsx_extsize & mp->m_blockmask)
1306 		return -EINVAL;
1307 
1308 	new_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1309 
1310 	/*
1311 	 * Inode verifiers do not check that the extent size hint is an integer
1312 	 * multiple of the rt extent size on a directory with both rtinherit
1313 	 * and extszinherit flags set.  Don't let sysadmins misconfigure
1314 	 * directories.
1315 	 */
1316 	if ((new_diflags & XFS_DIFLAG_RTINHERIT) &&
1317 	    (new_diflags & XFS_DIFLAG_EXTSZINHERIT)) {
1318 		unsigned int	rtextsize_bytes;
1319 
1320 		rtextsize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
1321 		if (fa->fsx_extsize % rtextsize_bytes)
1322 			return -EINVAL;
1323 	}
1324 
1325 	failaddr = xfs_inode_validate_extsize(ip->i_mount,
1326 			XFS_B_TO_FSB(mp, fa->fsx_extsize),
1327 			VFS_I(ip)->i_mode, new_diflags);
1328 	return failaddr != NULL ? -EINVAL : 0;
1329 }
1330 
1331 static int
1332 xfs_ioctl_setattr_check_cowextsize(
1333 	struct xfs_inode	*ip,
1334 	struct fileattr		*fa)
1335 {
1336 	struct xfs_mount	*mp = ip->i_mount;
1337 	xfs_failaddr_t		failaddr;
1338 	uint64_t		new_diflags2;
1339 	uint16_t		new_diflags;
1340 
1341 	if (!fa->fsx_valid)
1342 		return 0;
1343 
1344 	if (fa->fsx_cowextsize & mp->m_blockmask)
1345 		return -EINVAL;
1346 
1347 	new_diflags = xfs_flags2diflags(ip, fa->fsx_xflags);
1348 	new_diflags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1349 
1350 	failaddr = xfs_inode_validate_cowextsize(ip->i_mount,
1351 			XFS_B_TO_FSB(mp, fa->fsx_cowextsize),
1352 			VFS_I(ip)->i_mode, new_diflags, new_diflags2);
1353 	return failaddr != NULL ? -EINVAL : 0;
1354 }
1355 
1356 static int
1357 xfs_ioctl_setattr_check_projid(
1358 	struct xfs_inode	*ip,
1359 	struct fileattr		*fa)
1360 {
1361 	if (!fa->fsx_valid)
1362 		return 0;
1363 
1364 	/* Disallow 32bit project ids if 32bit IDs are not enabled. */
1365 	if (fa->fsx_projid > (uint16_t)-1 &&
1366 	    !xfs_has_projid32(ip->i_mount))
1367 		return -EINVAL;
1368 	return 0;
1369 }
1370 
1371 int
1372 xfs_fileattr_set(
1373 	struct user_namespace	*mnt_userns,
1374 	struct dentry		*dentry,
1375 	struct fileattr		*fa)
1376 {
1377 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1378 	struct xfs_mount	*mp = ip->i_mount;
1379 	struct xfs_trans	*tp;
1380 	struct xfs_dquot	*pdqp = NULL;
1381 	struct xfs_dquot	*olddquot = NULL;
1382 	int			error;
1383 
1384 	trace_xfs_ioctl_setattr(ip);
1385 
1386 	if (d_is_special(dentry))
1387 		return -ENOTTY;
1388 
1389 	if (!fa->fsx_valid) {
1390 		if (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL |
1391 				  FS_NOATIME_FL | FS_NODUMP_FL |
1392 				  FS_SYNC_FL | FS_DAX_FL | FS_PROJINHERIT_FL))
1393 			return -EOPNOTSUPP;
1394 	}
1395 
1396 	error = xfs_ioctl_setattr_check_projid(ip, fa);
1397 	if (error)
1398 		return error;
1399 
1400 	/*
1401 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1402 	 * before we start any other transactions. Trying to do this later
1403 	 * is messy. We don't care to take a readlock to look at the ids
1404 	 * in inode here, because we can't hold it across the trans_reserve.
1405 	 * If the IDs do change before we take the ilock, we're covered
1406 	 * because the i_*dquot fields will get updated anyway.
1407 	 */
1408 	if (fa->fsx_valid && XFS_IS_QUOTA_ON(mp)) {
1409 		error = xfs_qm_vop_dqalloc(ip, VFS_I(ip)->i_uid,
1410 				VFS_I(ip)->i_gid, fa->fsx_projid,
1411 				XFS_QMOPT_PQUOTA, NULL, NULL, &pdqp);
1412 		if (error)
1413 			return error;
1414 	}
1415 
1416 	xfs_ioctl_setattr_prepare_dax(ip, fa);
1417 
1418 	tp = xfs_ioctl_setattr_get_trans(ip, pdqp);
1419 	if (IS_ERR(tp)) {
1420 		error = PTR_ERR(tp);
1421 		goto error_free_dquots;
1422 	}
1423 
1424 	error = xfs_ioctl_setattr_check_extsize(ip, fa);
1425 	if (error)
1426 		goto error_trans_cancel;
1427 
1428 	error = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1429 	if (error)
1430 		goto error_trans_cancel;
1431 
1432 	error = xfs_ioctl_setattr_xflags(tp, ip, fa);
1433 	if (error)
1434 		goto error_trans_cancel;
1435 
1436 	if (!fa->fsx_valid)
1437 		goto skip_xattr;
1438 	/*
1439 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1440 	 * overrides the following restrictions:
1441 	 *
1442 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1443 	 * successful return from chown()
1444 	 */
1445 
1446 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1447 	    !capable_wrt_inode_uidgid(mnt_userns, VFS_I(ip), CAP_FSETID))
1448 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1449 
1450 	/* Change the ownerships and register project quota modifications */
1451 	if (ip->i_projid != fa->fsx_projid) {
1452 		if (XFS_IS_PQUOTA_ON(mp)) {
1453 			olddquot = xfs_qm_vop_chown(tp, ip,
1454 						&ip->i_pdquot, pdqp);
1455 		}
1456 		ip->i_projid = fa->fsx_projid;
1457 	}
1458 
1459 	/*
1460 	 * Only set the extent size hint if we've already determined that the
1461 	 * extent size hint should be set on the inode. If no extent size flags
1462 	 * are set on the inode then unconditionally clear the extent size hint.
1463 	 */
1464 	if (ip->i_diflags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1465 		ip->i_extsize = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1466 	else
1467 		ip->i_extsize = 0;
1468 
1469 	if (xfs_has_v3inodes(mp)) {
1470 		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
1471 			ip->i_cowextsize = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1472 		else
1473 			ip->i_cowextsize = 0;
1474 	}
1475 
1476 skip_xattr:
1477 	error = xfs_trans_commit(tp);
1478 
1479 	/*
1480 	 * Release any dquot(s) the inode had kept before chown.
1481 	 */
1482 	xfs_qm_dqrele(olddquot);
1483 	xfs_qm_dqrele(pdqp);
1484 
1485 	return error;
1486 
1487 error_trans_cancel:
1488 	xfs_trans_cancel(tp);
1489 error_free_dquots:
1490 	xfs_qm_dqrele(pdqp);
1491 	return error;
1492 }
1493 
1494 static bool
1495 xfs_getbmap_format(
1496 	struct kgetbmap		*p,
1497 	struct getbmapx __user	*u,
1498 	size_t			recsize)
1499 {
1500 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1501 	    put_user(p->bmv_block, &u->bmv_block) ||
1502 	    put_user(p->bmv_length, &u->bmv_length) ||
1503 	    put_user(0, &u->bmv_count) ||
1504 	    put_user(0, &u->bmv_entries))
1505 		return false;
1506 	if (recsize < sizeof(struct getbmapx))
1507 		return true;
1508 	if (put_user(0, &u->bmv_iflags) ||
1509 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1510 	    put_user(0, &u->bmv_unused1) ||
1511 	    put_user(0, &u->bmv_unused2))
1512 		return false;
1513 	return true;
1514 }
1515 
1516 STATIC int
1517 xfs_ioc_getbmap(
1518 	struct file		*file,
1519 	unsigned int		cmd,
1520 	void			__user *arg)
1521 {
1522 	struct getbmapx		bmx = { 0 };
1523 	struct kgetbmap		*buf;
1524 	size_t			recsize;
1525 	int			error, i;
1526 
1527 	switch (cmd) {
1528 	case XFS_IOC_GETBMAPA:
1529 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1530 		fallthrough;
1531 	case XFS_IOC_GETBMAP:
1532 		/* struct getbmap is a strict subset of struct getbmapx. */
1533 		recsize = sizeof(struct getbmap);
1534 		break;
1535 	case XFS_IOC_GETBMAPX:
1536 		recsize = sizeof(struct getbmapx);
1537 		break;
1538 	default:
1539 		return -EINVAL;
1540 	}
1541 
1542 	if (copy_from_user(&bmx, arg, recsize))
1543 		return -EFAULT;
1544 
1545 	if (bmx.bmv_count < 2)
1546 		return -EINVAL;
1547 	if (bmx.bmv_count > ULONG_MAX / recsize)
1548 		return -ENOMEM;
1549 
1550 	buf = kvzalloc(bmx.bmv_count * sizeof(*buf), GFP_KERNEL);
1551 	if (!buf)
1552 		return -ENOMEM;
1553 
1554 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1555 	if (error)
1556 		goto out_free_buf;
1557 
1558 	error = -EFAULT;
1559 	if (copy_to_user(arg, &bmx, recsize))
1560 		goto out_free_buf;
1561 	arg += recsize;
1562 
1563 	for (i = 0; i < bmx.bmv_entries; i++) {
1564 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1565 			goto out_free_buf;
1566 		arg += recsize;
1567 	}
1568 
1569 	error = 0;
1570 out_free_buf:
1571 	kmem_free(buf);
1572 	return error;
1573 }
1574 
1575 STATIC int
1576 xfs_ioc_getfsmap(
1577 	struct xfs_inode	*ip,
1578 	struct fsmap_head	__user *arg)
1579 {
1580 	struct xfs_fsmap_head	xhead = {0};
1581 	struct fsmap_head	head;
1582 	struct fsmap		*recs;
1583 	unsigned int		count;
1584 	__u32			last_flags = 0;
1585 	bool			done = false;
1586 	int			error;
1587 
1588 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1589 		return -EFAULT;
1590 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1591 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1592 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1593 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1594 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1595 		return -EINVAL;
1596 
1597 	/*
1598 	 * Use an internal memory buffer so that we don't have to copy fsmap
1599 	 * data to userspace while holding locks.  Start by trying to allocate
1600 	 * up to 128k for the buffer, but fall back to a single page if needed.
1601 	 */
1602 	count = min_t(unsigned int, head.fmh_count,
1603 			131072 / sizeof(struct fsmap));
1604 	recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1605 	if (!recs) {
1606 		count = min_t(unsigned int, head.fmh_count,
1607 				PAGE_SIZE / sizeof(struct fsmap));
1608 		recs = kvzalloc(count * sizeof(struct fsmap), GFP_KERNEL);
1609 		if (!recs)
1610 			return -ENOMEM;
1611 	}
1612 
1613 	xhead.fmh_iflags = head.fmh_iflags;
1614 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1615 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1616 
1617 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1618 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1619 
1620 	head.fmh_entries = 0;
1621 	do {
1622 		struct fsmap __user	*user_recs;
1623 		struct fsmap		*last_rec;
1624 
1625 		user_recs = &arg->fmh_recs[head.fmh_entries];
1626 		xhead.fmh_entries = 0;
1627 		xhead.fmh_count = min_t(unsigned int, count,
1628 					head.fmh_count - head.fmh_entries);
1629 
1630 		/* Run query, record how many entries we got. */
1631 		error = xfs_getfsmap(ip->i_mount, &xhead, recs);
1632 		switch (error) {
1633 		case 0:
1634 			/*
1635 			 * There are no more records in the result set.  Copy
1636 			 * whatever we got to userspace and break out.
1637 			 */
1638 			done = true;
1639 			break;
1640 		case -ECANCELED:
1641 			/*
1642 			 * The internal memory buffer is full.  Copy whatever
1643 			 * records we got to userspace and go again if we have
1644 			 * not yet filled the userspace buffer.
1645 			 */
1646 			error = 0;
1647 			break;
1648 		default:
1649 			goto out_free;
1650 		}
1651 		head.fmh_entries += xhead.fmh_entries;
1652 		head.fmh_oflags = xhead.fmh_oflags;
1653 
1654 		/*
1655 		 * If the caller wanted a record count or there aren't any
1656 		 * new records to return, we're done.
1657 		 */
1658 		if (head.fmh_count == 0 || xhead.fmh_entries == 0)
1659 			break;
1660 
1661 		/* Copy all the records we got out to userspace. */
1662 		if (copy_to_user(user_recs, recs,
1663 				 xhead.fmh_entries * sizeof(struct fsmap))) {
1664 			error = -EFAULT;
1665 			goto out_free;
1666 		}
1667 
1668 		/* Remember the last record flags we copied to userspace. */
1669 		last_rec = &recs[xhead.fmh_entries - 1];
1670 		last_flags = last_rec->fmr_flags;
1671 
1672 		/* Set up the low key for the next iteration. */
1673 		xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec);
1674 		trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1675 	} while (!done && head.fmh_entries < head.fmh_count);
1676 
1677 	/*
1678 	 * If there are no more records in the query result set and we're not
1679 	 * in counting mode, mark the last record returned with the LAST flag.
1680 	 */
1681 	if (done && head.fmh_count > 0 && head.fmh_entries > 0) {
1682 		struct fsmap __user	*user_rec;
1683 
1684 		last_flags |= FMR_OF_LAST;
1685 		user_rec = &arg->fmh_recs[head.fmh_entries - 1];
1686 
1687 		if (copy_to_user(&user_rec->fmr_flags, &last_flags,
1688 					sizeof(last_flags))) {
1689 			error = -EFAULT;
1690 			goto out_free;
1691 		}
1692 	}
1693 
1694 	/* copy back header */
1695 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) {
1696 		error = -EFAULT;
1697 		goto out_free;
1698 	}
1699 
1700 out_free:
1701 	kmem_free(recs);
1702 	return error;
1703 }
1704 
1705 STATIC int
1706 xfs_ioc_scrub_metadata(
1707 	struct file			*file,
1708 	void				__user *arg)
1709 {
1710 	struct xfs_scrub_metadata	scrub;
1711 	int				error;
1712 
1713 	if (!capable(CAP_SYS_ADMIN))
1714 		return -EPERM;
1715 
1716 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1717 		return -EFAULT;
1718 
1719 	error = xfs_scrub_metadata(file, &scrub);
1720 	if (error)
1721 		return error;
1722 
1723 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1724 		return -EFAULT;
1725 
1726 	return 0;
1727 }
1728 
1729 int
1730 xfs_ioc_swapext(
1731 	xfs_swapext_t	*sxp)
1732 {
1733 	xfs_inode_t     *ip, *tip;
1734 	struct fd	f, tmp;
1735 	int		error = 0;
1736 
1737 	/* Pull information for the target fd */
1738 	f = fdget((int)sxp->sx_fdtarget);
1739 	if (!f.file) {
1740 		error = -EINVAL;
1741 		goto out;
1742 	}
1743 
1744 	if (!(f.file->f_mode & FMODE_WRITE) ||
1745 	    !(f.file->f_mode & FMODE_READ) ||
1746 	    (f.file->f_flags & O_APPEND)) {
1747 		error = -EBADF;
1748 		goto out_put_file;
1749 	}
1750 
1751 	tmp = fdget((int)sxp->sx_fdtmp);
1752 	if (!tmp.file) {
1753 		error = -EINVAL;
1754 		goto out_put_file;
1755 	}
1756 
1757 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1758 	    !(tmp.file->f_mode & FMODE_READ) ||
1759 	    (tmp.file->f_flags & O_APPEND)) {
1760 		error = -EBADF;
1761 		goto out_put_tmp_file;
1762 	}
1763 
1764 	if (IS_SWAPFILE(file_inode(f.file)) ||
1765 	    IS_SWAPFILE(file_inode(tmp.file))) {
1766 		error = -EINVAL;
1767 		goto out_put_tmp_file;
1768 	}
1769 
1770 	/*
1771 	 * We need to ensure that the fds passed in point to XFS inodes
1772 	 * before we cast and access them as XFS structures as we have no
1773 	 * control over what the user passes us here.
1774 	 */
1775 	if (f.file->f_op != &xfs_file_operations ||
1776 	    tmp.file->f_op != &xfs_file_operations) {
1777 		error = -EINVAL;
1778 		goto out_put_tmp_file;
1779 	}
1780 
1781 	ip = XFS_I(file_inode(f.file));
1782 	tip = XFS_I(file_inode(tmp.file));
1783 
1784 	if (ip->i_mount != tip->i_mount) {
1785 		error = -EINVAL;
1786 		goto out_put_tmp_file;
1787 	}
1788 
1789 	if (ip->i_ino == tip->i_ino) {
1790 		error = -EINVAL;
1791 		goto out_put_tmp_file;
1792 	}
1793 
1794 	if (xfs_is_shutdown(ip->i_mount)) {
1795 		error = -EIO;
1796 		goto out_put_tmp_file;
1797 	}
1798 
1799 	error = xfs_swap_extents(ip, tip, sxp);
1800 
1801  out_put_tmp_file:
1802 	fdput(tmp);
1803  out_put_file:
1804 	fdput(f);
1805  out:
1806 	return error;
1807 }
1808 
1809 static int
1810 xfs_ioc_getlabel(
1811 	struct xfs_mount	*mp,
1812 	char			__user *user_label)
1813 {
1814 	struct xfs_sb		*sbp = &mp->m_sb;
1815 	char			label[XFSLABEL_MAX + 1];
1816 
1817 	/* Paranoia */
1818 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
1819 
1820 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
1821 	memset(label, 0, sizeof(label));
1822 	spin_lock(&mp->m_sb_lock);
1823 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
1824 	spin_unlock(&mp->m_sb_lock);
1825 
1826 	if (copy_to_user(user_label, label, sizeof(label)))
1827 		return -EFAULT;
1828 	return 0;
1829 }
1830 
1831 static int
1832 xfs_ioc_setlabel(
1833 	struct file		*filp,
1834 	struct xfs_mount	*mp,
1835 	char			__user *newlabel)
1836 {
1837 	struct xfs_sb		*sbp = &mp->m_sb;
1838 	char			label[XFSLABEL_MAX + 1];
1839 	size_t			len;
1840 	int			error;
1841 
1842 	if (!capable(CAP_SYS_ADMIN))
1843 		return -EPERM;
1844 	/*
1845 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
1846 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
1847 	 * (required) NULL character to test the incoming label length.
1848 	 * NB: The on disk label doesn't need to be null terminated.
1849 	 */
1850 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
1851 		return -EFAULT;
1852 	len = strnlen(label, XFSLABEL_MAX + 1);
1853 	if (len > sizeof(sbp->sb_fname))
1854 		return -EINVAL;
1855 
1856 	error = mnt_want_write_file(filp);
1857 	if (error)
1858 		return error;
1859 
1860 	spin_lock(&mp->m_sb_lock);
1861 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
1862 	memcpy(sbp->sb_fname, label, len);
1863 	spin_unlock(&mp->m_sb_lock);
1864 
1865 	/*
1866 	 * Now we do several things to satisfy userspace.
1867 	 * In addition to normal logging of the primary superblock, we also
1868 	 * immediately write these changes to sector zero for the primary, then
1869 	 * update all backup supers (as xfs_db does for a label change), then
1870 	 * invalidate the block device page cache.  This is so that any prior
1871 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
1872 	 * and userspace will see the newly-written label.
1873 	 */
1874 	error = xfs_sync_sb_buf(mp);
1875 	if (error)
1876 		goto out;
1877 	/*
1878 	 * growfs also updates backup supers so lock against that.
1879 	 */
1880 	mutex_lock(&mp->m_growlock);
1881 	error = xfs_update_secondary_sbs(mp);
1882 	mutex_unlock(&mp->m_growlock);
1883 
1884 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
1885 
1886 out:
1887 	mnt_drop_write_file(filp);
1888 	return error;
1889 }
1890 
1891 static inline int
1892 xfs_fs_eofblocks_from_user(
1893 	struct xfs_fs_eofblocks		*src,
1894 	struct xfs_icwalk		*dst)
1895 {
1896 	if (src->eof_version != XFS_EOFBLOCKS_VERSION)
1897 		return -EINVAL;
1898 
1899 	if (src->eof_flags & ~XFS_EOF_FLAGS_VALID)
1900 		return -EINVAL;
1901 
1902 	if (memchr_inv(&src->pad32, 0, sizeof(src->pad32)) ||
1903 	    memchr_inv(src->pad64, 0, sizeof(src->pad64)))
1904 		return -EINVAL;
1905 
1906 	dst->icw_flags = 0;
1907 	if (src->eof_flags & XFS_EOF_FLAGS_SYNC)
1908 		dst->icw_flags |= XFS_ICWALK_FLAG_SYNC;
1909 	if (src->eof_flags & XFS_EOF_FLAGS_UID)
1910 		dst->icw_flags |= XFS_ICWALK_FLAG_UID;
1911 	if (src->eof_flags & XFS_EOF_FLAGS_GID)
1912 		dst->icw_flags |= XFS_ICWALK_FLAG_GID;
1913 	if (src->eof_flags & XFS_EOF_FLAGS_PRID)
1914 		dst->icw_flags |= XFS_ICWALK_FLAG_PRID;
1915 	if (src->eof_flags & XFS_EOF_FLAGS_MINFILESIZE)
1916 		dst->icw_flags |= XFS_ICWALK_FLAG_MINFILESIZE;
1917 
1918 	dst->icw_prid = src->eof_prid;
1919 	dst->icw_min_file_size = src->eof_min_file_size;
1920 
1921 	dst->icw_uid = INVALID_UID;
1922 	if (src->eof_flags & XFS_EOF_FLAGS_UID) {
1923 		dst->icw_uid = make_kuid(current_user_ns(), src->eof_uid);
1924 		if (!uid_valid(dst->icw_uid))
1925 			return -EINVAL;
1926 	}
1927 
1928 	dst->icw_gid = INVALID_GID;
1929 	if (src->eof_flags & XFS_EOF_FLAGS_GID) {
1930 		dst->icw_gid = make_kgid(current_user_ns(), src->eof_gid);
1931 		if (!gid_valid(dst->icw_gid))
1932 			return -EINVAL;
1933 	}
1934 	return 0;
1935 }
1936 
1937 /*
1938  * Note: some of the ioctl's return positive numbers as a
1939  * byte count indicating success, such as readlink_by_handle.
1940  * So we don't "sign flip" like most other routines.  This means
1941  * true errors need to be returned as a negative value.
1942  */
1943 long
1944 xfs_file_ioctl(
1945 	struct file		*filp,
1946 	unsigned int		cmd,
1947 	unsigned long		p)
1948 {
1949 	struct inode		*inode = file_inode(filp);
1950 	struct xfs_inode	*ip = XFS_I(inode);
1951 	struct xfs_mount	*mp = ip->i_mount;
1952 	void			__user *arg = (void __user *)p;
1953 	int			error;
1954 
1955 	trace_xfs_file_ioctl(ip);
1956 
1957 	switch (cmd) {
1958 	case FITRIM:
1959 		return xfs_ioc_trim(mp, arg);
1960 	case FS_IOC_GETFSLABEL:
1961 		return xfs_ioc_getlabel(mp, arg);
1962 	case FS_IOC_SETFSLABEL:
1963 		return xfs_ioc_setlabel(filp, mp, arg);
1964 	case XFS_IOC_ALLOCSP:
1965 	case XFS_IOC_FREESP:
1966 	case XFS_IOC_ALLOCSP64:
1967 	case XFS_IOC_FREESP64: {
1968 		xfs_flock64_t		bf;
1969 
1970 		if (copy_from_user(&bf, arg, sizeof(bf)))
1971 			return -EFAULT;
1972 		return xfs_ioc_space(filp, &bf);
1973 	}
1974 	case XFS_IOC_DIOINFO: {
1975 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
1976 		struct dioattr		da;
1977 
1978 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
1979 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
1980 
1981 		if (copy_to_user(arg, &da, sizeof(da)))
1982 			return -EFAULT;
1983 		return 0;
1984 	}
1985 
1986 	case XFS_IOC_FSBULKSTAT_SINGLE:
1987 	case XFS_IOC_FSBULKSTAT:
1988 	case XFS_IOC_FSINUMBERS:
1989 		return xfs_ioc_fsbulkstat(filp, cmd, arg);
1990 
1991 	case XFS_IOC_BULKSTAT:
1992 		return xfs_ioc_bulkstat(filp, cmd, arg);
1993 	case XFS_IOC_INUMBERS:
1994 		return xfs_ioc_inumbers(mp, cmd, arg);
1995 
1996 	case XFS_IOC_FSGEOMETRY_V1:
1997 		return xfs_ioc_fsgeometry(mp, arg, 3);
1998 	case XFS_IOC_FSGEOMETRY_V4:
1999 		return xfs_ioc_fsgeometry(mp, arg, 4);
2000 	case XFS_IOC_FSGEOMETRY:
2001 		return xfs_ioc_fsgeometry(mp, arg, 5);
2002 
2003 	case XFS_IOC_AG_GEOMETRY:
2004 		return xfs_ioc_ag_geometry(mp, arg);
2005 
2006 	case XFS_IOC_GETVERSION:
2007 		return put_user(inode->i_generation, (int __user *)arg);
2008 
2009 	case XFS_IOC_FSGETXATTRA:
2010 		return xfs_ioc_fsgetxattra(ip, arg);
2011 
2012 	case XFS_IOC_GETBMAP:
2013 	case XFS_IOC_GETBMAPA:
2014 	case XFS_IOC_GETBMAPX:
2015 		return xfs_ioc_getbmap(filp, cmd, arg);
2016 
2017 	case FS_IOC_GETFSMAP:
2018 		return xfs_ioc_getfsmap(ip, arg);
2019 
2020 	case XFS_IOC_SCRUB_METADATA:
2021 		return xfs_ioc_scrub_metadata(filp, arg);
2022 
2023 	case XFS_IOC_FD_TO_HANDLE:
2024 	case XFS_IOC_PATH_TO_HANDLE:
2025 	case XFS_IOC_PATH_TO_FSHANDLE: {
2026 		xfs_fsop_handlereq_t	hreq;
2027 
2028 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2029 			return -EFAULT;
2030 		return xfs_find_handle(cmd, &hreq);
2031 	}
2032 	case XFS_IOC_OPEN_BY_HANDLE: {
2033 		xfs_fsop_handlereq_t	hreq;
2034 
2035 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2036 			return -EFAULT;
2037 		return xfs_open_by_handle(filp, &hreq);
2038 	}
2039 
2040 	case XFS_IOC_READLINK_BY_HANDLE: {
2041 		xfs_fsop_handlereq_t	hreq;
2042 
2043 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2044 			return -EFAULT;
2045 		return xfs_readlink_by_handle(filp, &hreq);
2046 	}
2047 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2048 		return xfs_attrlist_by_handle(filp, arg);
2049 
2050 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2051 		return xfs_attrmulti_by_handle(filp, arg);
2052 
2053 	case XFS_IOC_SWAPEXT: {
2054 		struct xfs_swapext	sxp;
2055 
2056 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2057 			return -EFAULT;
2058 		error = mnt_want_write_file(filp);
2059 		if (error)
2060 			return error;
2061 		error = xfs_ioc_swapext(&sxp);
2062 		mnt_drop_write_file(filp);
2063 		return error;
2064 	}
2065 
2066 	case XFS_IOC_FSCOUNTS: {
2067 		xfs_fsop_counts_t out;
2068 
2069 		xfs_fs_counts(mp, &out);
2070 
2071 		if (copy_to_user(arg, &out, sizeof(out)))
2072 			return -EFAULT;
2073 		return 0;
2074 	}
2075 
2076 	case XFS_IOC_SET_RESBLKS: {
2077 		xfs_fsop_resblks_t inout;
2078 		uint64_t	   in;
2079 
2080 		if (!capable(CAP_SYS_ADMIN))
2081 			return -EPERM;
2082 
2083 		if (xfs_is_readonly(mp))
2084 			return -EROFS;
2085 
2086 		if (copy_from_user(&inout, arg, sizeof(inout)))
2087 			return -EFAULT;
2088 
2089 		error = mnt_want_write_file(filp);
2090 		if (error)
2091 			return error;
2092 
2093 		/* input parameter is passed in resblks field of structure */
2094 		in = inout.resblks;
2095 		error = xfs_reserve_blocks(mp, &in, &inout);
2096 		mnt_drop_write_file(filp);
2097 		if (error)
2098 			return error;
2099 
2100 		if (copy_to_user(arg, &inout, sizeof(inout)))
2101 			return -EFAULT;
2102 		return 0;
2103 	}
2104 
2105 	case XFS_IOC_GET_RESBLKS: {
2106 		xfs_fsop_resblks_t out;
2107 
2108 		if (!capable(CAP_SYS_ADMIN))
2109 			return -EPERM;
2110 
2111 		error = xfs_reserve_blocks(mp, NULL, &out);
2112 		if (error)
2113 			return error;
2114 
2115 		if (copy_to_user(arg, &out, sizeof(out)))
2116 			return -EFAULT;
2117 
2118 		return 0;
2119 	}
2120 
2121 	case XFS_IOC_FSGROWFSDATA: {
2122 		struct xfs_growfs_data in;
2123 
2124 		if (copy_from_user(&in, arg, sizeof(in)))
2125 			return -EFAULT;
2126 
2127 		error = mnt_want_write_file(filp);
2128 		if (error)
2129 			return error;
2130 		error = xfs_growfs_data(mp, &in);
2131 		mnt_drop_write_file(filp);
2132 		return error;
2133 	}
2134 
2135 	case XFS_IOC_FSGROWFSLOG: {
2136 		struct xfs_growfs_log in;
2137 
2138 		if (copy_from_user(&in, arg, sizeof(in)))
2139 			return -EFAULT;
2140 
2141 		error = mnt_want_write_file(filp);
2142 		if (error)
2143 			return error;
2144 		error = xfs_growfs_log(mp, &in);
2145 		mnt_drop_write_file(filp);
2146 		return error;
2147 	}
2148 
2149 	case XFS_IOC_FSGROWFSRT: {
2150 		xfs_growfs_rt_t in;
2151 
2152 		if (copy_from_user(&in, arg, sizeof(in)))
2153 			return -EFAULT;
2154 
2155 		error = mnt_want_write_file(filp);
2156 		if (error)
2157 			return error;
2158 		error = xfs_growfs_rt(mp, &in);
2159 		mnt_drop_write_file(filp);
2160 		return error;
2161 	}
2162 
2163 	case XFS_IOC_GOINGDOWN: {
2164 		uint32_t in;
2165 
2166 		if (!capable(CAP_SYS_ADMIN))
2167 			return -EPERM;
2168 
2169 		if (get_user(in, (uint32_t __user *)arg))
2170 			return -EFAULT;
2171 
2172 		return xfs_fs_goingdown(mp, in);
2173 	}
2174 
2175 	case XFS_IOC_ERROR_INJECTION: {
2176 		xfs_error_injection_t in;
2177 
2178 		if (!capable(CAP_SYS_ADMIN))
2179 			return -EPERM;
2180 
2181 		if (copy_from_user(&in, arg, sizeof(in)))
2182 			return -EFAULT;
2183 
2184 		return xfs_errortag_add(mp, in.errtag);
2185 	}
2186 
2187 	case XFS_IOC_ERROR_CLEARALL:
2188 		if (!capable(CAP_SYS_ADMIN))
2189 			return -EPERM;
2190 
2191 		return xfs_errortag_clearall(mp);
2192 
2193 	case XFS_IOC_FREE_EOFBLOCKS: {
2194 		struct xfs_fs_eofblocks	eofb;
2195 		struct xfs_icwalk	icw;
2196 
2197 		if (!capable(CAP_SYS_ADMIN))
2198 			return -EPERM;
2199 
2200 		if (xfs_is_readonly(mp))
2201 			return -EROFS;
2202 
2203 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2204 			return -EFAULT;
2205 
2206 		error = xfs_fs_eofblocks_from_user(&eofb, &icw);
2207 		if (error)
2208 			return error;
2209 
2210 		trace_xfs_ioc_free_eofblocks(mp, &icw, _RET_IP_);
2211 
2212 		sb_start_write(mp->m_super);
2213 		error = xfs_blockgc_free_space(mp, &icw);
2214 		sb_end_write(mp->m_super);
2215 		return error;
2216 	}
2217 
2218 	default:
2219 		return -ENOTTY;
2220 	}
2221 }
2222