xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 7f877908)
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 
39 #include <linux/mount.h>
40 #include <linux/namei.h>
41 
42 /*
43  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
44  * a file or fs handle.
45  *
46  * XFS_IOC_PATH_TO_FSHANDLE
47  *    returns fs handle for a mount point or path within that mount point
48  * XFS_IOC_FD_TO_HANDLE
49  *    returns full handle for a FD opened in user space
50  * XFS_IOC_PATH_TO_HANDLE
51  *    returns full handle for a path
52  */
53 int
54 xfs_find_handle(
55 	unsigned int		cmd,
56 	xfs_fsop_handlereq_t	*hreq)
57 {
58 	int			hsize;
59 	xfs_handle_t		handle;
60 	struct inode		*inode;
61 	struct fd		f = {NULL};
62 	struct path		path;
63 	int			error;
64 	struct xfs_inode	*ip;
65 
66 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
67 		f = fdget(hreq->fd);
68 		if (!f.file)
69 			return -EBADF;
70 		inode = file_inode(f.file);
71 	} else {
72 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
73 		if (error)
74 			return error;
75 		inode = d_inode(path.dentry);
76 	}
77 	ip = XFS_I(inode);
78 
79 	/*
80 	 * We can only generate handles for inodes residing on a XFS filesystem,
81 	 * and only for regular files, directories or symbolic links.
82 	 */
83 	error = -EINVAL;
84 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
85 		goto out_put;
86 
87 	error = -EBADF;
88 	if (!S_ISREG(inode->i_mode) &&
89 	    !S_ISDIR(inode->i_mode) &&
90 	    !S_ISLNK(inode->i_mode))
91 		goto out_put;
92 
93 
94 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
95 
96 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
97 		/*
98 		 * This handle only contains an fsid, zero the rest.
99 		 */
100 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
101 		hsize = sizeof(xfs_fsid_t);
102 	} else {
103 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
104 					sizeof(handle.ha_fid.fid_len);
105 		handle.ha_fid.fid_pad = 0;
106 		handle.ha_fid.fid_gen = inode->i_generation;
107 		handle.ha_fid.fid_ino = ip->i_ino;
108 		hsize = sizeof(xfs_handle_t);
109 	}
110 
111 	error = -EFAULT;
112 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
113 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
114 		goto out_put;
115 
116 	error = 0;
117 
118  out_put:
119 	if (cmd == XFS_IOC_FD_TO_HANDLE)
120 		fdput(f);
121 	else
122 		path_put(&path);
123 	return error;
124 }
125 
126 /*
127  * No need to do permission checks on the various pathname components
128  * as the handle operations are privileged.
129  */
130 STATIC int
131 xfs_handle_acceptable(
132 	void			*context,
133 	struct dentry		*dentry)
134 {
135 	return 1;
136 }
137 
138 /*
139  * Convert userspace handle data into a dentry.
140  */
141 struct dentry *
142 xfs_handle_to_dentry(
143 	struct file		*parfilp,
144 	void __user		*uhandle,
145 	u32			hlen)
146 {
147 	xfs_handle_t		handle;
148 	struct xfs_fid64	fid;
149 
150 	/*
151 	 * Only allow handle opens under a directory.
152 	 */
153 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
154 		return ERR_PTR(-ENOTDIR);
155 
156 	if (hlen != sizeof(xfs_handle_t))
157 		return ERR_PTR(-EINVAL);
158 	if (copy_from_user(&handle, uhandle, hlen))
159 		return ERR_PTR(-EFAULT);
160 	if (handle.ha_fid.fid_len !=
161 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
162 		return ERR_PTR(-EINVAL);
163 
164 	memset(&fid, 0, sizeof(struct fid));
165 	fid.ino = handle.ha_fid.fid_ino;
166 	fid.gen = handle.ha_fid.fid_gen;
167 
168 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
169 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
170 			xfs_handle_acceptable, NULL);
171 }
172 
173 STATIC struct dentry *
174 xfs_handlereq_to_dentry(
175 	struct file		*parfilp,
176 	xfs_fsop_handlereq_t	*hreq)
177 {
178 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
179 }
180 
181 int
182 xfs_open_by_handle(
183 	struct file		*parfilp,
184 	xfs_fsop_handlereq_t	*hreq)
185 {
186 	const struct cred	*cred = current_cred();
187 	int			error;
188 	int			fd;
189 	int			permflag;
190 	struct file		*filp;
191 	struct inode		*inode;
192 	struct dentry		*dentry;
193 	fmode_t			fmode;
194 	struct path		path;
195 
196 	if (!capable(CAP_SYS_ADMIN))
197 		return -EPERM;
198 
199 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
200 	if (IS_ERR(dentry))
201 		return PTR_ERR(dentry);
202 	inode = d_inode(dentry);
203 
204 	/* Restrict xfs_open_by_handle to directories & regular files. */
205 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
206 		error = -EPERM;
207 		goto out_dput;
208 	}
209 
210 #if BITS_PER_LONG != 32
211 	hreq->oflags |= O_LARGEFILE;
212 #endif
213 
214 	permflag = hreq->oflags;
215 	fmode = OPEN_FMODE(permflag);
216 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
217 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
218 		error = -EPERM;
219 		goto out_dput;
220 	}
221 
222 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
223 		error = -EPERM;
224 		goto out_dput;
225 	}
226 
227 	/* Can't write directories. */
228 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
229 		error = -EISDIR;
230 		goto out_dput;
231 	}
232 
233 	fd = get_unused_fd_flags(0);
234 	if (fd < 0) {
235 		error = fd;
236 		goto out_dput;
237 	}
238 
239 	path.mnt = parfilp->f_path.mnt;
240 	path.dentry = dentry;
241 	filp = dentry_open(&path, hreq->oflags, cred);
242 	dput(dentry);
243 	if (IS_ERR(filp)) {
244 		put_unused_fd(fd);
245 		return PTR_ERR(filp);
246 	}
247 
248 	if (S_ISREG(inode->i_mode)) {
249 		filp->f_flags |= O_NOATIME;
250 		filp->f_mode |= FMODE_NOCMTIME;
251 	}
252 
253 	fd_install(fd, filp);
254 	return fd;
255 
256  out_dput:
257 	dput(dentry);
258 	return error;
259 }
260 
261 int
262 xfs_readlink_by_handle(
263 	struct file		*parfilp,
264 	xfs_fsop_handlereq_t	*hreq)
265 {
266 	struct dentry		*dentry;
267 	__u32			olen;
268 	int			error;
269 
270 	if (!capable(CAP_SYS_ADMIN))
271 		return -EPERM;
272 
273 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
274 	if (IS_ERR(dentry))
275 		return PTR_ERR(dentry);
276 
277 	/* Restrict this handle operation to symlinks only. */
278 	if (!d_is_symlink(dentry)) {
279 		error = -EINVAL;
280 		goto out_dput;
281 	}
282 
283 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
284 		error = -EFAULT;
285 		goto out_dput;
286 	}
287 
288 	error = vfs_readlink(dentry, hreq->ohandle, olen);
289 
290  out_dput:
291 	dput(dentry);
292 	return error;
293 }
294 
295 STATIC int
296 xfs_attrlist_by_handle(
297 	struct file		*parfilp,
298 	void			__user *arg)
299 {
300 	int			error = -ENOMEM;
301 	attrlist_cursor_kern_t	*cursor;
302 	struct xfs_fsop_attrlist_handlereq __user	*p = arg;
303 	xfs_fsop_attrlist_handlereq_t al_hreq;
304 	struct dentry		*dentry;
305 	char			*kbuf;
306 
307 	if (!capable(CAP_SYS_ADMIN))
308 		return -EPERM;
309 	if (copy_from_user(&al_hreq, arg, sizeof(xfs_fsop_attrlist_handlereq_t)))
310 		return -EFAULT;
311 	if (al_hreq.buflen < sizeof(struct attrlist) ||
312 	    al_hreq.buflen > XFS_XATTR_LIST_MAX)
313 		return -EINVAL;
314 
315 	/*
316 	 * Reject flags, only allow namespaces.
317 	 */
318 	if (al_hreq.flags & ~(ATTR_ROOT | ATTR_SECURE))
319 		return -EINVAL;
320 
321 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
322 	if (IS_ERR(dentry))
323 		return PTR_ERR(dentry);
324 
325 	kbuf = kmem_zalloc_large(al_hreq.buflen, 0);
326 	if (!kbuf)
327 		goto out_dput;
328 
329 	cursor = (attrlist_cursor_kern_t *)&al_hreq.pos;
330 	error = xfs_attr_list(XFS_I(d_inode(dentry)), kbuf, al_hreq.buflen,
331 					al_hreq.flags, cursor);
332 	if (error)
333 		goto out_kfree;
334 
335 	if (copy_to_user(&p->pos, cursor, sizeof(attrlist_cursor_kern_t))) {
336 		error = -EFAULT;
337 		goto out_kfree;
338 	}
339 
340 	if (copy_to_user(al_hreq.buffer, kbuf, al_hreq.buflen))
341 		error = -EFAULT;
342 
343 out_kfree:
344 	kmem_free(kbuf);
345 out_dput:
346 	dput(dentry);
347 	return error;
348 }
349 
350 int
351 xfs_attrmulti_attr_get(
352 	struct inode		*inode,
353 	unsigned char		*name,
354 	unsigned char		__user *ubuf,
355 	uint32_t		*len,
356 	uint32_t		flags)
357 {
358 	unsigned char		*kbuf;
359 	int			error = -EFAULT;
360 	size_t			namelen;
361 
362 	if (*len > XFS_XATTR_SIZE_MAX)
363 		return -EINVAL;
364 	kbuf = kmem_zalloc_large(*len, 0);
365 	if (!kbuf)
366 		return -ENOMEM;
367 
368 	namelen = strlen(name);
369 	error = xfs_attr_get(XFS_I(inode), name, namelen, &kbuf, (int *)len,
370 			     flags);
371 	if (error)
372 		goto out_kfree;
373 
374 	if (copy_to_user(ubuf, kbuf, *len))
375 		error = -EFAULT;
376 
377 out_kfree:
378 	kmem_free(kbuf);
379 	return error;
380 }
381 
382 int
383 xfs_attrmulti_attr_set(
384 	struct inode		*inode,
385 	unsigned char		*name,
386 	const unsigned char	__user *ubuf,
387 	uint32_t		len,
388 	uint32_t		flags)
389 {
390 	unsigned char		*kbuf;
391 	int			error;
392 	size_t			namelen;
393 
394 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
395 		return -EPERM;
396 	if (len > XFS_XATTR_SIZE_MAX)
397 		return -EINVAL;
398 
399 	kbuf = memdup_user(ubuf, len);
400 	if (IS_ERR(kbuf))
401 		return PTR_ERR(kbuf);
402 
403 	namelen = strlen(name);
404 	error = xfs_attr_set(XFS_I(inode), name, namelen, kbuf, len, flags);
405 	if (!error)
406 		xfs_forget_acl(inode, name, flags);
407 	kfree(kbuf);
408 	return error;
409 }
410 
411 int
412 xfs_attrmulti_attr_remove(
413 	struct inode		*inode,
414 	unsigned char		*name,
415 	uint32_t		flags)
416 {
417 	int			error;
418 	size_t			namelen;
419 
420 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
421 		return -EPERM;
422 	namelen = strlen(name);
423 	error = xfs_attr_remove(XFS_I(inode), name, namelen, flags);
424 	if (!error)
425 		xfs_forget_acl(inode, name, flags);
426 	return error;
427 }
428 
429 STATIC int
430 xfs_attrmulti_by_handle(
431 	struct file		*parfilp,
432 	void			__user *arg)
433 {
434 	int			error;
435 	xfs_attr_multiop_t	*ops;
436 	xfs_fsop_attrmulti_handlereq_t am_hreq;
437 	struct dentry		*dentry;
438 	unsigned int		i, size;
439 	unsigned char		*attr_name;
440 
441 	if (!capable(CAP_SYS_ADMIN))
442 		return -EPERM;
443 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
444 		return -EFAULT;
445 
446 	/* overflow check */
447 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
448 		return -E2BIG;
449 
450 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
451 	if (IS_ERR(dentry))
452 		return PTR_ERR(dentry);
453 
454 	error = -E2BIG;
455 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
456 	if (!size || size > 16 * PAGE_SIZE)
457 		goto out_dput;
458 
459 	ops = memdup_user(am_hreq.ops, size);
460 	if (IS_ERR(ops)) {
461 		error = PTR_ERR(ops);
462 		goto out_dput;
463 	}
464 
465 	error = -ENOMEM;
466 	attr_name = kmalloc(MAXNAMELEN, GFP_KERNEL);
467 	if (!attr_name)
468 		goto out_kfree_ops;
469 
470 	error = 0;
471 	for (i = 0; i < am_hreq.opcount; i++) {
472 		if ((ops[i].am_flags & ATTR_ROOT) &&
473 		    (ops[i].am_flags & ATTR_SECURE)) {
474 			ops[i].am_error = -EINVAL;
475 			continue;
476 		}
477 		ops[i].am_flags &= ~ATTR_KERNEL_FLAGS;
478 
479 		ops[i].am_error = strncpy_from_user((char *)attr_name,
480 				ops[i].am_attrname, MAXNAMELEN);
481 		if (ops[i].am_error == 0 || ops[i].am_error == MAXNAMELEN)
482 			error = -ERANGE;
483 		if (ops[i].am_error < 0)
484 			break;
485 
486 		switch (ops[i].am_opcode) {
487 		case ATTR_OP_GET:
488 			ops[i].am_error = xfs_attrmulti_attr_get(
489 					d_inode(dentry), attr_name,
490 					ops[i].am_attrvalue, &ops[i].am_length,
491 					ops[i].am_flags);
492 			break;
493 		case ATTR_OP_SET:
494 			ops[i].am_error = mnt_want_write_file(parfilp);
495 			if (ops[i].am_error)
496 				break;
497 			ops[i].am_error = xfs_attrmulti_attr_set(
498 					d_inode(dentry), attr_name,
499 					ops[i].am_attrvalue, ops[i].am_length,
500 					ops[i].am_flags);
501 			mnt_drop_write_file(parfilp);
502 			break;
503 		case ATTR_OP_REMOVE:
504 			ops[i].am_error = mnt_want_write_file(parfilp);
505 			if (ops[i].am_error)
506 				break;
507 			ops[i].am_error = xfs_attrmulti_attr_remove(
508 					d_inode(dentry), attr_name,
509 					ops[i].am_flags);
510 			mnt_drop_write_file(parfilp);
511 			break;
512 		default:
513 			ops[i].am_error = -EINVAL;
514 		}
515 	}
516 
517 	if (copy_to_user(am_hreq.ops, ops, size))
518 		error = -EFAULT;
519 
520 	kfree(attr_name);
521  out_kfree_ops:
522 	kfree(ops);
523  out_dput:
524 	dput(dentry);
525 	return error;
526 }
527 
528 int
529 xfs_ioc_space(
530 	struct file		*filp,
531 	xfs_flock64_t		*bf)
532 {
533 	struct inode		*inode = file_inode(filp);
534 	struct xfs_inode	*ip = XFS_I(inode);
535 	struct iattr		iattr;
536 	enum xfs_prealloc_flags	flags = XFS_PREALLOC_CLEAR;
537 	uint			iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
538 	int			error;
539 
540 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND))
541 		return -EPERM;
542 
543 	if (!(filp->f_mode & FMODE_WRITE))
544 		return -EBADF;
545 
546 	if (!S_ISREG(inode->i_mode))
547 		return -EINVAL;
548 
549 	if (xfs_is_always_cow_inode(ip))
550 		return -EOPNOTSUPP;
551 
552 	if (filp->f_flags & O_DSYNC)
553 		flags |= XFS_PREALLOC_SYNC;
554 	if (filp->f_mode & FMODE_NOCMTIME)
555 		flags |= XFS_PREALLOC_INVISIBLE;
556 
557 	error = mnt_want_write_file(filp);
558 	if (error)
559 		return error;
560 
561 	xfs_ilock(ip, iolock);
562 	error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
563 	if (error)
564 		goto out_unlock;
565 	inode_dio_wait(inode);
566 
567 	switch (bf->l_whence) {
568 	case 0: /*SEEK_SET*/
569 		break;
570 	case 1: /*SEEK_CUR*/
571 		bf->l_start += filp->f_pos;
572 		break;
573 	case 2: /*SEEK_END*/
574 		bf->l_start += XFS_ISIZE(ip);
575 		break;
576 	default:
577 		error = -EINVAL;
578 		goto out_unlock;
579 	}
580 
581 	if (bf->l_start < 0 || bf->l_start > inode->i_sb->s_maxbytes) {
582 		error = -EINVAL;
583 		goto out_unlock;
584 	}
585 
586 	if (bf->l_start > XFS_ISIZE(ip)) {
587 		error = xfs_alloc_file_space(ip, XFS_ISIZE(ip),
588 				bf->l_start - XFS_ISIZE(ip), 0);
589 		if (error)
590 			goto out_unlock;
591 	}
592 
593 	iattr.ia_valid = ATTR_SIZE;
594 	iattr.ia_size = bf->l_start;
595 	error = xfs_vn_setattr_size(file_dentry(filp), &iattr);
596 	if (error)
597 		goto out_unlock;
598 
599 	error = xfs_update_prealloc_flags(ip, flags);
600 
601 out_unlock:
602 	xfs_iunlock(ip, iolock);
603 	mnt_drop_write_file(filp);
604 	return error;
605 }
606 
607 /* Return 0 on success or positive error */
608 int
609 xfs_fsbulkstat_one_fmt(
610 	struct xfs_ibulk		*breq,
611 	const struct xfs_bulkstat	*bstat)
612 {
613 	struct xfs_bstat		bs1;
614 
615 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
616 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
617 		return -EFAULT;
618 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
619 }
620 
621 int
622 xfs_fsinumbers_fmt(
623 	struct xfs_ibulk		*breq,
624 	const struct xfs_inumbers	*igrp)
625 {
626 	struct xfs_inogrp		ig1;
627 
628 	xfs_inumbers_to_inogrp(&ig1, igrp);
629 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
630 		return -EFAULT;
631 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
632 }
633 
634 STATIC int
635 xfs_ioc_fsbulkstat(
636 	xfs_mount_t		*mp,
637 	unsigned int		cmd,
638 	void			__user *arg)
639 {
640 	struct xfs_fsop_bulkreq	bulkreq;
641 	struct xfs_ibulk	breq = {
642 		.mp		= mp,
643 		.ocount		= 0,
644 	};
645 	xfs_ino_t		lastino;
646 	int			error;
647 
648 	/* done = 1 if there are more stats to get and if bulkstat */
649 	/* should be called again (unused here, but used in dmapi) */
650 
651 	if (!capable(CAP_SYS_ADMIN))
652 		return -EPERM;
653 
654 	if (XFS_FORCED_SHUTDOWN(mp))
655 		return -EIO;
656 
657 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
658 		return -EFAULT;
659 
660 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
661 		return -EFAULT;
662 
663 	if (bulkreq.icount <= 0)
664 		return -EINVAL;
665 
666 	if (bulkreq.ubuffer == NULL)
667 		return -EINVAL;
668 
669 	breq.ubuffer = bulkreq.ubuffer;
670 	breq.icount = bulkreq.icount;
671 
672 	/*
673 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
674 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
675 	 * that *lastip contains either zero or the number of the last inode to
676 	 * be examined by the previous call and return results starting with
677 	 * the next inode after that.  The new bulk request back end functions
678 	 * take the inode to start with, so we have to compute the startino
679 	 * parameter from lastino to maintain correct function.  lastino == 0
680 	 * is a special case because it has traditionally meant "first inode
681 	 * in filesystem".
682 	 */
683 	if (cmd == XFS_IOC_FSINUMBERS) {
684 		breq.startino = lastino ? lastino + 1 : 0;
685 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
686 		lastino = breq.startino - 1;
687 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
688 		breq.startino = lastino;
689 		breq.icount = 1;
690 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
691 	} else {	/* XFS_IOC_FSBULKSTAT */
692 		breq.startino = lastino ? lastino + 1 : 0;
693 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
694 		lastino = breq.startino - 1;
695 	}
696 
697 	if (error)
698 		return error;
699 
700 	if (bulkreq.lastip != NULL &&
701 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
702 		return -EFAULT;
703 
704 	if (bulkreq.ocount != NULL &&
705 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
706 		return -EFAULT;
707 
708 	return 0;
709 }
710 
711 /* Return 0 on success or positive error */
712 static int
713 xfs_bulkstat_fmt(
714 	struct xfs_ibulk		*breq,
715 	const struct xfs_bulkstat	*bstat)
716 {
717 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
718 		return -EFAULT;
719 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
720 }
721 
722 /*
723  * Check the incoming bulk request @hdr from userspace and initialize the
724  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
725  * should proceed; -ECANCELED if there's nothing to do; or the usual
726  * negative error code.
727  */
728 static int
729 xfs_bulk_ireq_setup(
730 	struct xfs_mount	*mp,
731 	struct xfs_bulk_ireq	*hdr,
732 	struct xfs_ibulk	*breq,
733 	void __user		*ubuffer)
734 {
735 	if (hdr->icount == 0 ||
736 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
737 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
738 		return -EINVAL;
739 
740 	breq->startino = hdr->ino;
741 	breq->ubuffer = ubuffer;
742 	breq->icount = hdr->icount;
743 	breq->ocount = 0;
744 	breq->flags = 0;
745 
746 	/*
747 	 * The @ino parameter is a special value, so we must look it up here.
748 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
749 	 * worth of data.
750 	 */
751 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
752 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
753 			return -EINVAL;
754 
755 		switch (hdr->ino) {
756 		case XFS_BULK_IREQ_SPECIAL_ROOT:
757 			hdr->ino = mp->m_sb.sb_rootino;
758 			break;
759 		default:
760 			return -EINVAL;
761 		}
762 		breq->icount = 1;
763 	}
764 
765 	/*
766 	 * The IREQ_AGNO flag means that we only want results from a given AG.
767 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
768 	 * beyond the specified AG then we return no results.
769 	 */
770 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
771 		if (hdr->agno >= mp->m_sb.sb_agcount)
772 			return -EINVAL;
773 
774 		if (breq->startino == 0)
775 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
776 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
777 			return -EINVAL;
778 
779 		breq->flags |= XFS_IBULK_SAME_AG;
780 
781 		/* Asking for an inode past the end of the AG?  We're done! */
782 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
783 			return -ECANCELED;
784 	} else if (hdr->agno)
785 		return -EINVAL;
786 
787 	/* Asking for an inode past the end of the FS?  We're done! */
788 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
789 		return -ECANCELED;
790 
791 	return 0;
792 }
793 
794 /*
795  * Update the userspace bulk request @hdr to reflect the end state of the
796  * internal bulk request @breq.
797  */
798 static void
799 xfs_bulk_ireq_teardown(
800 	struct xfs_bulk_ireq	*hdr,
801 	struct xfs_ibulk	*breq)
802 {
803 	hdr->ino = breq->startino;
804 	hdr->ocount = breq->ocount;
805 }
806 
807 /* Handle the v5 bulkstat ioctl. */
808 STATIC int
809 xfs_ioc_bulkstat(
810 	struct xfs_mount		*mp,
811 	unsigned int			cmd,
812 	struct xfs_bulkstat_req __user	*arg)
813 {
814 	struct xfs_bulk_ireq		hdr;
815 	struct xfs_ibulk		breq = {
816 		.mp			= mp,
817 	};
818 	int				error;
819 
820 	if (!capable(CAP_SYS_ADMIN))
821 		return -EPERM;
822 
823 	if (XFS_FORCED_SHUTDOWN(mp))
824 		return -EIO;
825 
826 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
827 		return -EFAULT;
828 
829 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
830 	if (error == -ECANCELED)
831 		goto out_teardown;
832 	if (error < 0)
833 		return error;
834 
835 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
836 	if (error)
837 		return error;
838 
839 out_teardown:
840 	xfs_bulk_ireq_teardown(&hdr, &breq);
841 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
842 		return -EFAULT;
843 
844 	return 0;
845 }
846 
847 STATIC int
848 xfs_inumbers_fmt(
849 	struct xfs_ibulk		*breq,
850 	const struct xfs_inumbers	*igrp)
851 {
852 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
853 		return -EFAULT;
854 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
855 }
856 
857 /* Handle the v5 inumbers ioctl. */
858 STATIC int
859 xfs_ioc_inumbers(
860 	struct xfs_mount		*mp,
861 	unsigned int			cmd,
862 	struct xfs_inumbers_req __user	*arg)
863 {
864 	struct xfs_bulk_ireq		hdr;
865 	struct xfs_ibulk		breq = {
866 		.mp			= mp,
867 	};
868 	int				error;
869 
870 	if (!capable(CAP_SYS_ADMIN))
871 		return -EPERM;
872 
873 	if (XFS_FORCED_SHUTDOWN(mp))
874 		return -EIO;
875 
876 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
877 		return -EFAULT;
878 
879 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
880 	if (error == -ECANCELED)
881 		goto out_teardown;
882 	if (error < 0)
883 		return error;
884 
885 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
886 	if (error)
887 		return error;
888 
889 out_teardown:
890 	xfs_bulk_ireq_teardown(&hdr, &breq);
891 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
892 		return -EFAULT;
893 
894 	return 0;
895 }
896 
897 STATIC int
898 xfs_ioc_fsgeometry(
899 	struct xfs_mount	*mp,
900 	void			__user *arg,
901 	int			struct_version)
902 {
903 	struct xfs_fsop_geom	fsgeo;
904 	size_t			len;
905 
906 	xfs_fs_geometry(&mp->m_sb, &fsgeo, struct_version);
907 
908 	if (struct_version <= 3)
909 		len = sizeof(struct xfs_fsop_geom_v1);
910 	else if (struct_version == 4)
911 		len = sizeof(struct xfs_fsop_geom_v4);
912 	else {
913 		xfs_fsop_geom_health(mp, &fsgeo);
914 		len = sizeof(fsgeo);
915 	}
916 
917 	if (copy_to_user(arg, &fsgeo, len))
918 		return -EFAULT;
919 	return 0;
920 }
921 
922 STATIC int
923 xfs_ioc_ag_geometry(
924 	struct xfs_mount	*mp,
925 	void			__user *arg)
926 {
927 	struct xfs_ag_geometry	ageo;
928 	int			error;
929 
930 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
931 		return -EFAULT;
932 	if (ageo.ag_flags)
933 		return -EINVAL;
934 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
935 		return -EINVAL;
936 
937 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
938 	if (error)
939 		return error;
940 
941 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
942 		return -EFAULT;
943 	return 0;
944 }
945 
946 /*
947  * Linux extended inode flags interface.
948  */
949 
950 STATIC unsigned int
951 xfs_merge_ioc_xflags(
952 	unsigned int	flags,
953 	unsigned int	start)
954 {
955 	unsigned int	xflags = start;
956 
957 	if (flags & FS_IMMUTABLE_FL)
958 		xflags |= FS_XFLAG_IMMUTABLE;
959 	else
960 		xflags &= ~FS_XFLAG_IMMUTABLE;
961 	if (flags & FS_APPEND_FL)
962 		xflags |= FS_XFLAG_APPEND;
963 	else
964 		xflags &= ~FS_XFLAG_APPEND;
965 	if (flags & FS_SYNC_FL)
966 		xflags |= FS_XFLAG_SYNC;
967 	else
968 		xflags &= ~FS_XFLAG_SYNC;
969 	if (flags & FS_NOATIME_FL)
970 		xflags |= FS_XFLAG_NOATIME;
971 	else
972 		xflags &= ~FS_XFLAG_NOATIME;
973 	if (flags & FS_NODUMP_FL)
974 		xflags |= FS_XFLAG_NODUMP;
975 	else
976 		xflags &= ~FS_XFLAG_NODUMP;
977 
978 	return xflags;
979 }
980 
981 STATIC unsigned int
982 xfs_di2lxflags(
983 	uint16_t	di_flags)
984 {
985 	unsigned int	flags = 0;
986 
987 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
988 		flags |= FS_IMMUTABLE_FL;
989 	if (di_flags & XFS_DIFLAG_APPEND)
990 		flags |= FS_APPEND_FL;
991 	if (di_flags & XFS_DIFLAG_SYNC)
992 		flags |= FS_SYNC_FL;
993 	if (di_flags & XFS_DIFLAG_NOATIME)
994 		flags |= FS_NOATIME_FL;
995 	if (di_flags & XFS_DIFLAG_NODUMP)
996 		flags |= FS_NODUMP_FL;
997 	return flags;
998 }
999 
1000 static void
1001 xfs_fill_fsxattr(
1002 	struct xfs_inode	*ip,
1003 	bool			attr,
1004 	struct fsxattr		*fa)
1005 {
1006 	simple_fill_fsxattr(fa, xfs_ip2xflags(ip));
1007 	fa->fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
1008 	fa->fsx_cowextsize = ip->i_d.di_cowextsize <<
1009 			ip->i_mount->m_sb.sb_blocklog;
1010 	fa->fsx_projid = ip->i_d.di_projid;
1011 
1012 	if (attr) {
1013 		if (ip->i_afp) {
1014 			if (ip->i_afp->if_flags & XFS_IFEXTENTS)
1015 				fa->fsx_nextents = xfs_iext_count(ip->i_afp);
1016 			else
1017 				fa->fsx_nextents = ip->i_d.di_anextents;
1018 		} else
1019 			fa->fsx_nextents = 0;
1020 	} else {
1021 		if (ip->i_df.if_flags & XFS_IFEXTENTS)
1022 			fa->fsx_nextents = xfs_iext_count(&ip->i_df);
1023 		else
1024 			fa->fsx_nextents = ip->i_d.di_nextents;
1025 	}
1026 }
1027 
1028 STATIC int
1029 xfs_ioc_fsgetxattr(
1030 	xfs_inode_t		*ip,
1031 	int			attr,
1032 	void			__user *arg)
1033 {
1034 	struct fsxattr		fa;
1035 
1036 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1037 	xfs_fill_fsxattr(ip, attr, &fa);
1038 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1039 
1040 	if (copy_to_user(arg, &fa, sizeof(fa)))
1041 		return -EFAULT;
1042 	return 0;
1043 }
1044 
1045 STATIC uint16_t
1046 xfs_flags2diflags(
1047 	struct xfs_inode	*ip,
1048 	unsigned int		xflags)
1049 {
1050 	/* can't set PREALLOC this way, just preserve it */
1051 	uint16_t		di_flags =
1052 		(ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
1053 
1054 	if (xflags & FS_XFLAG_IMMUTABLE)
1055 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1056 	if (xflags & FS_XFLAG_APPEND)
1057 		di_flags |= XFS_DIFLAG_APPEND;
1058 	if (xflags & FS_XFLAG_SYNC)
1059 		di_flags |= XFS_DIFLAG_SYNC;
1060 	if (xflags & FS_XFLAG_NOATIME)
1061 		di_flags |= XFS_DIFLAG_NOATIME;
1062 	if (xflags & FS_XFLAG_NODUMP)
1063 		di_flags |= XFS_DIFLAG_NODUMP;
1064 	if (xflags & FS_XFLAG_NODEFRAG)
1065 		di_flags |= XFS_DIFLAG_NODEFRAG;
1066 	if (xflags & FS_XFLAG_FILESTREAM)
1067 		di_flags |= XFS_DIFLAG_FILESTREAM;
1068 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1069 		if (xflags & FS_XFLAG_RTINHERIT)
1070 			di_flags |= XFS_DIFLAG_RTINHERIT;
1071 		if (xflags & FS_XFLAG_NOSYMLINKS)
1072 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1073 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1074 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1075 		if (xflags & FS_XFLAG_PROJINHERIT)
1076 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1077 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1078 		if (xflags & FS_XFLAG_REALTIME)
1079 			di_flags |= XFS_DIFLAG_REALTIME;
1080 		if (xflags & FS_XFLAG_EXTSIZE)
1081 			di_flags |= XFS_DIFLAG_EXTSIZE;
1082 	}
1083 
1084 	return di_flags;
1085 }
1086 
1087 STATIC uint64_t
1088 xfs_flags2diflags2(
1089 	struct xfs_inode	*ip,
1090 	unsigned int		xflags)
1091 {
1092 	uint64_t		di_flags2 =
1093 		(ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK);
1094 
1095 	if (xflags & FS_XFLAG_DAX)
1096 		di_flags2 |= XFS_DIFLAG2_DAX;
1097 	if (xflags & FS_XFLAG_COWEXTSIZE)
1098 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1099 
1100 	return di_flags2;
1101 }
1102 
1103 STATIC void
1104 xfs_diflags_to_linux(
1105 	struct xfs_inode	*ip)
1106 {
1107 	struct inode		*inode = VFS_I(ip);
1108 	unsigned int		xflags = xfs_ip2xflags(ip);
1109 
1110 	if (xflags & FS_XFLAG_IMMUTABLE)
1111 		inode->i_flags |= S_IMMUTABLE;
1112 	else
1113 		inode->i_flags &= ~S_IMMUTABLE;
1114 	if (xflags & FS_XFLAG_APPEND)
1115 		inode->i_flags |= S_APPEND;
1116 	else
1117 		inode->i_flags &= ~S_APPEND;
1118 	if (xflags & FS_XFLAG_SYNC)
1119 		inode->i_flags |= S_SYNC;
1120 	else
1121 		inode->i_flags &= ~S_SYNC;
1122 	if (xflags & FS_XFLAG_NOATIME)
1123 		inode->i_flags |= S_NOATIME;
1124 	else
1125 		inode->i_flags &= ~S_NOATIME;
1126 #if 0	/* disabled until the flag switching races are sorted out */
1127 	if (xflags & FS_XFLAG_DAX)
1128 		inode->i_flags |= S_DAX;
1129 	else
1130 		inode->i_flags &= ~S_DAX;
1131 #endif
1132 }
1133 
1134 static int
1135 xfs_ioctl_setattr_xflags(
1136 	struct xfs_trans	*tp,
1137 	struct xfs_inode	*ip,
1138 	struct fsxattr		*fa)
1139 {
1140 	struct xfs_mount	*mp = ip->i_mount;
1141 	uint64_t		di_flags2;
1142 
1143 	/* Can't change realtime flag if any extents are allocated. */
1144 	if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
1145 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1146 		return -EINVAL;
1147 
1148 	/* If realtime flag is set then must have realtime device */
1149 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1150 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1151 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1152 			return -EINVAL;
1153 	}
1154 
1155 	/* Clear reflink if we are actually able to set the rt flag. */
1156 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1157 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1158 
1159 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1160 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1161 		return -EINVAL;
1162 
1163 	/* diflags2 only valid for v3 inodes. */
1164 	di_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1165 	if (di_flags2 && ip->i_d.di_version < 3)
1166 		return -EINVAL;
1167 
1168 	ip->i_d.di_flags = xfs_flags2diflags(ip, fa->fsx_xflags);
1169 	ip->i_d.di_flags2 = di_flags2;
1170 
1171 	xfs_diflags_to_linux(ip);
1172 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1173 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1174 	XFS_STATS_INC(mp, xs_ig_attrchg);
1175 	return 0;
1176 }
1177 
1178 /*
1179  * If we are changing DAX flags, we have to ensure the file is clean and any
1180  * cached objects in the address space are invalidated and removed. This
1181  * requires us to lock out other IO and page faults similar to a truncate
1182  * operation. The locks need to be held until the transaction has been committed
1183  * so that the cache invalidation is atomic with respect to the DAX flag
1184  * manipulation.
1185  */
1186 static int
1187 xfs_ioctl_setattr_dax_invalidate(
1188 	struct xfs_inode	*ip,
1189 	struct fsxattr		*fa,
1190 	int			*join_flags)
1191 {
1192 	struct inode		*inode = VFS_I(ip);
1193 	struct super_block	*sb = inode->i_sb;
1194 	int			error;
1195 
1196 	*join_flags = 0;
1197 
1198 	/*
1199 	 * It is only valid to set the DAX flag on regular files and
1200 	 * directories on filesystems where the block size is equal to the page
1201 	 * size. On directories it serves as an inherited hint so we don't
1202 	 * have to check the device for dax support or flush pagecache.
1203 	 */
1204 	if (fa->fsx_xflags & FS_XFLAG_DAX) {
1205 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
1206 
1207 		if (!bdev_dax_supported(target->bt_bdev, sb->s_blocksize))
1208 			return -EINVAL;
1209 	}
1210 
1211 	/* If the DAX state is not changing, we have nothing to do here. */
1212 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && IS_DAX(inode))
1213 		return 0;
1214 	if (!(fa->fsx_xflags & FS_XFLAG_DAX) && !IS_DAX(inode))
1215 		return 0;
1216 
1217 	if (S_ISDIR(inode->i_mode))
1218 		return 0;
1219 
1220 	/* lock, flush and invalidate mapping in preparation for flag change */
1221 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1222 	error = filemap_write_and_wait(inode->i_mapping);
1223 	if (error)
1224 		goto out_unlock;
1225 	error = invalidate_inode_pages2(inode->i_mapping);
1226 	if (error)
1227 		goto out_unlock;
1228 
1229 	*join_flags = XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL;
1230 	return 0;
1231 
1232 out_unlock:
1233 	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1234 	return error;
1235 
1236 }
1237 
1238 /*
1239  * Set up the transaction structure for the setattr operation, checking that we
1240  * have permission to do so. On success, return a clean transaction and the
1241  * inode locked exclusively ready for further operation specific checks. On
1242  * failure, return an error without modifying or locking the inode.
1243  *
1244  * The inode might already be IO locked on call. If this is the case, it is
1245  * indicated in @join_flags and we take full responsibility for ensuring they
1246  * are unlocked from now on. Hence if we have an error here, we still have to
1247  * unlock them. Otherwise, once they are joined to the transaction, they will
1248  * be unlocked on commit/cancel.
1249  */
1250 static struct xfs_trans *
1251 xfs_ioctl_setattr_get_trans(
1252 	struct xfs_inode	*ip,
1253 	int			join_flags)
1254 {
1255 	struct xfs_mount	*mp = ip->i_mount;
1256 	struct xfs_trans	*tp;
1257 	int			error = -EROFS;
1258 
1259 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1260 		goto out_unlock;
1261 	error = -EIO;
1262 	if (XFS_FORCED_SHUTDOWN(mp))
1263 		goto out_unlock;
1264 
1265 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1266 	if (error)
1267 		goto out_unlock;
1268 
1269 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1270 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | join_flags);
1271 	join_flags = 0;
1272 
1273 	/*
1274 	 * CAP_FOWNER overrides the following restrictions:
1275 	 *
1276 	 * The user ID of the calling process must be equal to the file owner
1277 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1278 	 */
1279 	if (!inode_owner_or_capable(VFS_I(ip))) {
1280 		error = -EPERM;
1281 		goto out_cancel;
1282 	}
1283 
1284 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1285 		xfs_trans_set_sync(tp);
1286 
1287 	return tp;
1288 
1289 out_cancel:
1290 	xfs_trans_cancel(tp);
1291 out_unlock:
1292 	if (join_flags)
1293 		xfs_iunlock(ip, join_flags);
1294 	return ERR_PTR(error);
1295 }
1296 
1297 /*
1298  * extent size hint validation is somewhat cumbersome. Rules are:
1299  *
1300  * 1. extent size hint is only valid for directories and regular files
1301  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1302  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1303  * 4. can only be changed on regular files if no extents are allocated
1304  * 5. can be changed on directories at any time
1305  * 6. extsize hint of 0 turns off hints, clears inode flags.
1306  * 7. Extent size must be a multiple of the appropriate block size.
1307  * 8. for non-realtime files, the extent size hint must be limited
1308  *    to half the AG size to avoid alignment extending the extent beyond the
1309  *    limits of the AG.
1310  *
1311  * Please keep this function in sync with xfs_scrub_inode_extsize.
1312  */
1313 static int
1314 xfs_ioctl_setattr_check_extsize(
1315 	struct xfs_inode	*ip,
1316 	struct fsxattr		*fa)
1317 {
1318 	struct xfs_mount	*mp = ip->i_mount;
1319 	xfs_extlen_t		size;
1320 	xfs_fsblock_t		extsize_fsb;
1321 
1322 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_d.di_nextents &&
1323 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1324 		return -EINVAL;
1325 
1326 	if (fa->fsx_extsize == 0)
1327 		return 0;
1328 
1329 	extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1330 	if (extsize_fsb > MAXEXTLEN)
1331 		return -EINVAL;
1332 
1333 	if (XFS_IS_REALTIME_INODE(ip) ||
1334 	    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1335 		size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1336 	} else {
1337 		size = mp->m_sb.sb_blocksize;
1338 		if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1339 			return -EINVAL;
1340 	}
1341 
1342 	if (fa->fsx_extsize % size)
1343 		return -EINVAL;
1344 
1345 	return 0;
1346 }
1347 
1348 /*
1349  * CoW extent size hint validation rules are:
1350  *
1351  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
1352  *    The inode does not have to have any shared blocks, but it must be a v3.
1353  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
1354  *    for a directory, the hint is propagated to new files.
1355  * 3. Can be changed on files & directories at any time.
1356  * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
1357  * 5. Extent size must be a multiple of the appropriate block size.
1358  * 6. The extent size hint must be limited to half the AG size to avoid
1359  *    alignment extending the extent beyond the limits of the AG.
1360  *
1361  * Please keep this function in sync with xfs_scrub_inode_cowextsize.
1362  */
1363 static int
1364 xfs_ioctl_setattr_check_cowextsize(
1365 	struct xfs_inode	*ip,
1366 	struct fsxattr		*fa)
1367 {
1368 	struct xfs_mount	*mp = ip->i_mount;
1369 	xfs_extlen_t		size;
1370 	xfs_fsblock_t		cowextsize_fsb;
1371 
1372 	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
1373 		return 0;
1374 
1375 	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb) ||
1376 	    ip->i_d.di_version != 3)
1377 		return -EINVAL;
1378 
1379 	if (fa->fsx_cowextsize == 0)
1380 		return 0;
1381 
1382 	cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1383 	if (cowextsize_fsb > MAXEXTLEN)
1384 		return -EINVAL;
1385 
1386 	size = mp->m_sb.sb_blocksize;
1387 	if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
1388 		return -EINVAL;
1389 
1390 	if (fa->fsx_cowextsize % size)
1391 		return -EINVAL;
1392 
1393 	return 0;
1394 }
1395 
1396 static int
1397 xfs_ioctl_setattr_check_projid(
1398 	struct xfs_inode	*ip,
1399 	struct fsxattr		*fa)
1400 {
1401 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1402 	if (fa->fsx_projid > (uint16_t)-1 &&
1403 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1404 		return -EINVAL;
1405 	return 0;
1406 }
1407 
1408 STATIC int
1409 xfs_ioctl_setattr(
1410 	xfs_inode_t		*ip,
1411 	struct fsxattr		*fa)
1412 {
1413 	struct fsxattr		old_fa;
1414 	struct xfs_mount	*mp = ip->i_mount;
1415 	struct xfs_trans	*tp;
1416 	struct xfs_dquot	*udqp = NULL;
1417 	struct xfs_dquot	*pdqp = NULL;
1418 	struct xfs_dquot	*olddquot = NULL;
1419 	int			code;
1420 	int			join_flags = 0;
1421 
1422 	trace_xfs_ioctl_setattr(ip);
1423 
1424 	code = xfs_ioctl_setattr_check_projid(ip, fa);
1425 	if (code)
1426 		return code;
1427 
1428 	/*
1429 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1430 	 * before we start any other transactions. Trying to do this later
1431 	 * is messy. We don't care to take a readlock to look at the ids
1432 	 * in inode here, because we can't hold it across the trans_reserve.
1433 	 * If the IDs do change before we take the ilock, we're covered
1434 	 * because the i_*dquot fields will get updated anyway.
1435 	 */
1436 	if (XFS_IS_QUOTA_ON(mp)) {
1437 		code = xfs_qm_vop_dqalloc(ip, ip->i_d.di_uid,
1438 					 ip->i_d.di_gid, fa->fsx_projid,
1439 					 XFS_QMOPT_PQUOTA, &udqp, NULL, &pdqp);
1440 		if (code)
1441 			return code;
1442 	}
1443 
1444 	/*
1445 	 * Changing DAX config may require inode locking for mapping
1446 	 * invalidation. These need to be held all the way to transaction commit
1447 	 * or cancel time, so need to be passed through to
1448 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1449 	 * appropriately.
1450 	 */
1451 	code = xfs_ioctl_setattr_dax_invalidate(ip, fa, &join_flags);
1452 	if (code)
1453 		goto error_free_dquots;
1454 
1455 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1456 	if (IS_ERR(tp)) {
1457 		code = PTR_ERR(tp);
1458 		goto error_free_dquots;
1459 	}
1460 
1461 	if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp) &&
1462 	    ip->i_d.di_projid != fa->fsx_projid) {
1463 		code = xfs_qm_vop_chown_reserve(tp, ip, udqp, NULL, pdqp,
1464 				capable(CAP_FOWNER) ?  XFS_QMOPT_FORCE_RES : 0);
1465 		if (code)	/* out of quota */
1466 			goto error_trans_cancel;
1467 	}
1468 
1469 	xfs_fill_fsxattr(ip, false, &old_fa);
1470 	code = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, fa);
1471 	if (code)
1472 		goto error_trans_cancel;
1473 
1474 	code = xfs_ioctl_setattr_check_extsize(ip, fa);
1475 	if (code)
1476 		goto error_trans_cancel;
1477 
1478 	code = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1479 	if (code)
1480 		goto error_trans_cancel;
1481 
1482 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
1483 	if (code)
1484 		goto error_trans_cancel;
1485 
1486 	/*
1487 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1488 	 * overrides the following restrictions:
1489 	 *
1490 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1491 	 * successful return from chown()
1492 	 */
1493 
1494 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1495 	    !capable_wrt_inode_uidgid(VFS_I(ip), CAP_FSETID))
1496 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1497 
1498 	/* Change the ownerships and register project quota modifications */
1499 	if (ip->i_d.di_projid != fa->fsx_projid) {
1500 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1501 			olddquot = xfs_qm_vop_chown(tp, ip,
1502 						&ip->i_pdquot, pdqp);
1503 		}
1504 		ASSERT(ip->i_d.di_version > 1);
1505 		ip->i_d.di_projid = fa->fsx_projid;
1506 	}
1507 
1508 	/*
1509 	 * Only set the extent size hint if we've already determined that the
1510 	 * extent size hint should be set on the inode. If no extent size flags
1511 	 * are set on the inode then unconditionally clear the extent size hint.
1512 	 */
1513 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1514 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1515 	else
1516 		ip->i_d.di_extsize = 0;
1517 	if (ip->i_d.di_version == 3 &&
1518 	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1519 		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
1520 				mp->m_sb.sb_blocklog;
1521 	else
1522 		ip->i_d.di_cowextsize = 0;
1523 
1524 	code = xfs_trans_commit(tp);
1525 
1526 	/*
1527 	 * Release any dquot(s) the inode had kept before chown.
1528 	 */
1529 	xfs_qm_dqrele(olddquot);
1530 	xfs_qm_dqrele(udqp);
1531 	xfs_qm_dqrele(pdqp);
1532 
1533 	return code;
1534 
1535 error_trans_cancel:
1536 	xfs_trans_cancel(tp);
1537 error_free_dquots:
1538 	xfs_qm_dqrele(udqp);
1539 	xfs_qm_dqrele(pdqp);
1540 	return code;
1541 }
1542 
1543 STATIC int
1544 xfs_ioc_fssetxattr(
1545 	xfs_inode_t		*ip,
1546 	struct file		*filp,
1547 	void			__user *arg)
1548 {
1549 	struct fsxattr		fa;
1550 	int error;
1551 
1552 	if (copy_from_user(&fa, arg, sizeof(fa)))
1553 		return -EFAULT;
1554 
1555 	error = mnt_want_write_file(filp);
1556 	if (error)
1557 		return error;
1558 	error = xfs_ioctl_setattr(ip, &fa);
1559 	mnt_drop_write_file(filp);
1560 	return error;
1561 }
1562 
1563 STATIC int
1564 xfs_ioc_getxflags(
1565 	xfs_inode_t		*ip,
1566 	void			__user *arg)
1567 {
1568 	unsigned int		flags;
1569 
1570 	flags = xfs_di2lxflags(ip->i_d.di_flags);
1571 	if (copy_to_user(arg, &flags, sizeof(flags)))
1572 		return -EFAULT;
1573 	return 0;
1574 }
1575 
1576 STATIC int
1577 xfs_ioc_setxflags(
1578 	struct xfs_inode	*ip,
1579 	struct file		*filp,
1580 	void			__user *arg)
1581 {
1582 	struct xfs_trans	*tp;
1583 	struct fsxattr		fa;
1584 	struct fsxattr		old_fa;
1585 	unsigned int		flags;
1586 	int			join_flags = 0;
1587 	int			error;
1588 
1589 	if (copy_from_user(&flags, arg, sizeof(flags)))
1590 		return -EFAULT;
1591 
1592 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1593 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1594 		      FS_SYNC_FL))
1595 		return -EOPNOTSUPP;
1596 
1597 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1598 
1599 	error = mnt_want_write_file(filp);
1600 	if (error)
1601 		return error;
1602 
1603 	/*
1604 	 * Changing DAX config may require inode locking for mapping
1605 	 * invalidation. These need to be held all the way to transaction commit
1606 	 * or cancel time, so need to be passed through to
1607 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1608 	 * appropriately.
1609 	 */
1610 	error = xfs_ioctl_setattr_dax_invalidate(ip, &fa, &join_flags);
1611 	if (error)
1612 		goto out_drop_write;
1613 
1614 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1615 	if (IS_ERR(tp)) {
1616 		error = PTR_ERR(tp);
1617 		goto out_drop_write;
1618 	}
1619 
1620 	xfs_fill_fsxattr(ip, false, &old_fa);
1621 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, &fa);
1622 	if (error) {
1623 		xfs_trans_cancel(tp);
1624 		goto out_drop_write;
1625 	}
1626 
1627 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1628 	if (error) {
1629 		xfs_trans_cancel(tp);
1630 		goto out_drop_write;
1631 	}
1632 
1633 	error = xfs_trans_commit(tp);
1634 out_drop_write:
1635 	mnt_drop_write_file(filp);
1636 	return error;
1637 }
1638 
1639 static bool
1640 xfs_getbmap_format(
1641 	struct kgetbmap		*p,
1642 	struct getbmapx __user	*u,
1643 	size_t			recsize)
1644 {
1645 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1646 	    put_user(p->bmv_block, &u->bmv_block) ||
1647 	    put_user(p->bmv_length, &u->bmv_length) ||
1648 	    put_user(0, &u->bmv_count) ||
1649 	    put_user(0, &u->bmv_entries))
1650 		return false;
1651 	if (recsize < sizeof(struct getbmapx))
1652 		return true;
1653 	if (put_user(0, &u->bmv_iflags) ||
1654 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1655 	    put_user(0, &u->bmv_unused1) ||
1656 	    put_user(0, &u->bmv_unused2))
1657 		return false;
1658 	return true;
1659 }
1660 
1661 STATIC int
1662 xfs_ioc_getbmap(
1663 	struct file		*file,
1664 	unsigned int		cmd,
1665 	void			__user *arg)
1666 {
1667 	struct getbmapx		bmx = { 0 };
1668 	struct kgetbmap		*buf;
1669 	size_t			recsize;
1670 	int			error, i;
1671 
1672 	switch (cmd) {
1673 	case XFS_IOC_GETBMAPA:
1674 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1675 		/*FALLTHRU*/
1676 	case XFS_IOC_GETBMAP:
1677 		if (file->f_mode & FMODE_NOCMTIME)
1678 			bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1679 		/* struct getbmap is a strict subset of struct getbmapx. */
1680 		recsize = sizeof(struct getbmap);
1681 		break;
1682 	case XFS_IOC_GETBMAPX:
1683 		recsize = sizeof(struct getbmapx);
1684 		break;
1685 	default:
1686 		return -EINVAL;
1687 	}
1688 
1689 	if (copy_from_user(&bmx, arg, recsize))
1690 		return -EFAULT;
1691 
1692 	if (bmx.bmv_count < 2)
1693 		return -EINVAL;
1694 	if (bmx.bmv_count > ULONG_MAX / recsize)
1695 		return -ENOMEM;
1696 
1697 	buf = kmem_zalloc_large(bmx.bmv_count * sizeof(*buf), 0);
1698 	if (!buf)
1699 		return -ENOMEM;
1700 
1701 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1702 	if (error)
1703 		goto out_free_buf;
1704 
1705 	error = -EFAULT;
1706 	if (copy_to_user(arg, &bmx, recsize))
1707 		goto out_free_buf;
1708 	arg += recsize;
1709 
1710 	for (i = 0; i < bmx.bmv_entries; i++) {
1711 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1712 			goto out_free_buf;
1713 		arg += recsize;
1714 	}
1715 
1716 	error = 0;
1717 out_free_buf:
1718 	kmem_free(buf);
1719 	return error;
1720 }
1721 
1722 struct getfsmap_info {
1723 	struct xfs_mount	*mp;
1724 	struct fsmap_head __user *data;
1725 	unsigned int		idx;
1726 	__u32			last_flags;
1727 };
1728 
1729 STATIC int
1730 xfs_getfsmap_format(struct xfs_fsmap *xfm, void *priv)
1731 {
1732 	struct getfsmap_info	*info = priv;
1733 	struct fsmap		fm;
1734 
1735 	trace_xfs_getfsmap_mapping(info->mp, xfm);
1736 
1737 	info->last_flags = xfm->fmr_flags;
1738 	xfs_fsmap_from_internal(&fm, xfm);
1739 	if (copy_to_user(&info->data->fmh_recs[info->idx++], &fm,
1740 			sizeof(struct fsmap)))
1741 		return -EFAULT;
1742 
1743 	return 0;
1744 }
1745 
1746 STATIC int
1747 xfs_ioc_getfsmap(
1748 	struct xfs_inode	*ip,
1749 	struct fsmap_head	__user *arg)
1750 {
1751 	struct getfsmap_info	info = { NULL };
1752 	struct xfs_fsmap_head	xhead = {0};
1753 	struct fsmap_head	head;
1754 	bool			aborted = false;
1755 	int			error;
1756 
1757 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1758 		return -EFAULT;
1759 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1760 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1761 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1762 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1763 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1764 		return -EINVAL;
1765 
1766 	xhead.fmh_iflags = head.fmh_iflags;
1767 	xhead.fmh_count = head.fmh_count;
1768 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1769 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1770 
1771 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1772 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1773 
1774 	info.mp = ip->i_mount;
1775 	info.data = arg;
1776 	error = xfs_getfsmap(ip->i_mount, &xhead, xfs_getfsmap_format, &info);
1777 	if (error == -ECANCELED) {
1778 		error = 0;
1779 		aborted = true;
1780 	} else if (error)
1781 		return error;
1782 
1783 	/* If we didn't abort, set the "last" flag in the last fmx */
1784 	if (!aborted && info.idx) {
1785 		info.last_flags |= FMR_OF_LAST;
1786 		if (copy_to_user(&info.data->fmh_recs[info.idx - 1].fmr_flags,
1787 				&info.last_flags, sizeof(info.last_flags)))
1788 			return -EFAULT;
1789 	}
1790 
1791 	/* copy back header */
1792 	head.fmh_entries = xhead.fmh_entries;
1793 	head.fmh_oflags = xhead.fmh_oflags;
1794 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head)))
1795 		return -EFAULT;
1796 
1797 	return 0;
1798 }
1799 
1800 STATIC int
1801 xfs_ioc_scrub_metadata(
1802 	struct xfs_inode		*ip,
1803 	void				__user *arg)
1804 {
1805 	struct xfs_scrub_metadata	scrub;
1806 	int				error;
1807 
1808 	if (!capable(CAP_SYS_ADMIN))
1809 		return -EPERM;
1810 
1811 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1812 		return -EFAULT;
1813 
1814 	error = xfs_scrub_metadata(ip, &scrub);
1815 	if (error)
1816 		return error;
1817 
1818 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1819 		return -EFAULT;
1820 
1821 	return 0;
1822 }
1823 
1824 int
1825 xfs_ioc_swapext(
1826 	xfs_swapext_t	*sxp)
1827 {
1828 	xfs_inode_t     *ip, *tip;
1829 	struct fd	f, tmp;
1830 	int		error = 0;
1831 
1832 	/* Pull information for the target fd */
1833 	f = fdget((int)sxp->sx_fdtarget);
1834 	if (!f.file) {
1835 		error = -EINVAL;
1836 		goto out;
1837 	}
1838 
1839 	if (!(f.file->f_mode & FMODE_WRITE) ||
1840 	    !(f.file->f_mode & FMODE_READ) ||
1841 	    (f.file->f_flags & O_APPEND)) {
1842 		error = -EBADF;
1843 		goto out_put_file;
1844 	}
1845 
1846 	tmp = fdget((int)sxp->sx_fdtmp);
1847 	if (!tmp.file) {
1848 		error = -EINVAL;
1849 		goto out_put_file;
1850 	}
1851 
1852 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1853 	    !(tmp.file->f_mode & FMODE_READ) ||
1854 	    (tmp.file->f_flags & O_APPEND)) {
1855 		error = -EBADF;
1856 		goto out_put_tmp_file;
1857 	}
1858 
1859 	if (IS_SWAPFILE(file_inode(f.file)) ||
1860 	    IS_SWAPFILE(file_inode(tmp.file))) {
1861 		error = -EINVAL;
1862 		goto out_put_tmp_file;
1863 	}
1864 
1865 	/*
1866 	 * We need to ensure that the fds passed in point to XFS inodes
1867 	 * before we cast and access them as XFS structures as we have no
1868 	 * control over what the user passes us here.
1869 	 */
1870 	if (f.file->f_op != &xfs_file_operations ||
1871 	    tmp.file->f_op != &xfs_file_operations) {
1872 		error = -EINVAL;
1873 		goto out_put_tmp_file;
1874 	}
1875 
1876 	ip = XFS_I(file_inode(f.file));
1877 	tip = XFS_I(file_inode(tmp.file));
1878 
1879 	if (ip->i_mount != tip->i_mount) {
1880 		error = -EINVAL;
1881 		goto out_put_tmp_file;
1882 	}
1883 
1884 	if (ip->i_ino == tip->i_ino) {
1885 		error = -EINVAL;
1886 		goto out_put_tmp_file;
1887 	}
1888 
1889 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1890 		error = -EIO;
1891 		goto out_put_tmp_file;
1892 	}
1893 
1894 	error = xfs_swap_extents(ip, tip, sxp);
1895 
1896  out_put_tmp_file:
1897 	fdput(tmp);
1898  out_put_file:
1899 	fdput(f);
1900  out:
1901 	return error;
1902 }
1903 
1904 static int
1905 xfs_ioc_getlabel(
1906 	struct xfs_mount	*mp,
1907 	char			__user *user_label)
1908 {
1909 	struct xfs_sb		*sbp = &mp->m_sb;
1910 	char			label[XFSLABEL_MAX + 1];
1911 
1912 	/* Paranoia */
1913 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
1914 
1915 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
1916 	memset(label, 0, sizeof(label));
1917 	spin_lock(&mp->m_sb_lock);
1918 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
1919 	spin_unlock(&mp->m_sb_lock);
1920 
1921 	if (copy_to_user(user_label, label, sizeof(label)))
1922 		return -EFAULT;
1923 	return 0;
1924 }
1925 
1926 static int
1927 xfs_ioc_setlabel(
1928 	struct file		*filp,
1929 	struct xfs_mount	*mp,
1930 	char			__user *newlabel)
1931 {
1932 	struct xfs_sb		*sbp = &mp->m_sb;
1933 	char			label[XFSLABEL_MAX + 1];
1934 	size_t			len;
1935 	int			error;
1936 
1937 	if (!capable(CAP_SYS_ADMIN))
1938 		return -EPERM;
1939 	/*
1940 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
1941 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
1942 	 * (required) NULL character to test the incoming label length.
1943 	 * NB: The on disk label doesn't need to be null terminated.
1944 	 */
1945 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
1946 		return -EFAULT;
1947 	len = strnlen(label, XFSLABEL_MAX + 1);
1948 	if (len > sizeof(sbp->sb_fname))
1949 		return -EINVAL;
1950 
1951 	error = mnt_want_write_file(filp);
1952 	if (error)
1953 		return error;
1954 
1955 	spin_lock(&mp->m_sb_lock);
1956 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
1957 	memcpy(sbp->sb_fname, label, len);
1958 	spin_unlock(&mp->m_sb_lock);
1959 
1960 	/*
1961 	 * Now we do several things to satisfy userspace.
1962 	 * In addition to normal logging of the primary superblock, we also
1963 	 * immediately write these changes to sector zero for the primary, then
1964 	 * update all backup supers (as xfs_db does for a label change), then
1965 	 * invalidate the block device page cache.  This is so that any prior
1966 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
1967 	 * and userspace will see the newly-written label.
1968 	 */
1969 	error = xfs_sync_sb_buf(mp);
1970 	if (error)
1971 		goto out;
1972 	/*
1973 	 * growfs also updates backup supers so lock against that.
1974 	 */
1975 	mutex_lock(&mp->m_growlock);
1976 	error = xfs_update_secondary_sbs(mp);
1977 	mutex_unlock(&mp->m_growlock);
1978 
1979 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
1980 
1981 out:
1982 	mnt_drop_write_file(filp);
1983 	return error;
1984 }
1985 
1986 /*
1987  * Note: some of the ioctl's return positive numbers as a
1988  * byte count indicating success, such as readlink_by_handle.
1989  * So we don't "sign flip" like most other routines.  This means
1990  * true errors need to be returned as a negative value.
1991  */
1992 long
1993 xfs_file_ioctl(
1994 	struct file		*filp,
1995 	unsigned int		cmd,
1996 	unsigned long		p)
1997 {
1998 	struct inode		*inode = file_inode(filp);
1999 	struct xfs_inode	*ip = XFS_I(inode);
2000 	struct xfs_mount	*mp = ip->i_mount;
2001 	void			__user *arg = (void __user *)p;
2002 	int			error;
2003 
2004 	trace_xfs_file_ioctl(ip);
2005 
2006 	switch (cmd) {
2007 	case FITRIM:
2008 		return xfs_ioc_trim(mp, arg);
2009 	case FS_IOC_GETFSLABEL:
2010 		return xfs_ioc_getlabel(mp, arg);
2011 	case FS_IOC_SETFSLABEL:
2012 		return xfs_ioc_setlabel(filp, mp, arg);
2013 	case XFS_IOC_ALLOCSP:
2014 	case XFS_IOC_FREESP:
2015 	case XFS_IOC_ALLOCSP64:
2016 	case XFS_IOC_FREESP64: {
2017 		xfs_flock64_t		bf;
2018 
2019 		if (copy_from_user(&bf, arg, sizeof(bf)))
2020 			return -EFAULT;
2021 		return xfs_ioc_space(filp, &bf);
2022 	}
2023 	case XFS_IOC_DIOINFO: {
2024 		struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
2025 		struct dioattr		da;
2026 
2027 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
2028 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
2029 
2030 		if (copy_to_user(arg, &da, sizeof(da)))
2031 			return -EFAULT;
2032 		return 0;
2033 	}
2034 
2035 	case XFS_IOC_FSBULKSTAT_SINGLE:
2036 	case XFS_IOC_FSBULKSTAT:
2037 	case XFS_IOC_FSINUMBERS:
2038 		return xfs_ioc_fsbulkstat(mp, cmd, arg);
2039 
2040 	case XFS_IOC_BULKSTAT:
2041 		return xfs_ioc_bulkstat(mp, cmd, arg);
2042 	case XFS_IOC_INUMBERS:
2043 		return xfs_ioc_inumbers(mp, cmd, arg);
2044 
2045 	case XFS_IOC_FSGEOMETRY_V1:
2046 		return xfs_ioc_fsgeometry(mp, arg, 3);
2047 	case XFS_IOC_FSGEOMETRY_V4:
2048 		return xfs_ioc_fsgeometry(mp, arg, 4);
2049 	case XFS_IOC_FSGEOMETRY:
2050 		return xfs_ioc_fsgeometry(mp, arg, 5);
2051 
2052 	case XFS_IOC_AG_GEOMETRY:
2053 		return xfs_ioc_ag_geometry(mp, arg);
2054 
2055 	case XFS_IOC_GETVERSION:
2056 		return put_user(inode->i_generation, (int __user *)arg);
2057 
2058 	case XFS_IOC_FSGETXATTR:
2059 		return xfs_ioc_fsgetxattr(ip, 0, arg);
2060 	case XFS_IOC_FSGETXATTRA:
2061 		return xfs_ioc_fsgetxattr(ip, 1, arg);
2062 	case XFS_IOC_FSSETXATTR:
2063 		return xfs_ioc_fssetxattr(ip, filp, arg);
2064 	case XFS_IOC_GETXFLAGS:
2065 		return xfs_ioc_getxflags(ip, arg);
2066 	case XFS_IOC_SETXFLAGS:
2067 		return xfs_ioc_setxflags(ip, filp, arg);
2068 
2069 	case XFS_IOC_GETBMAP:
2070 	case XFS_IOC_GETBMAPA:
2071 	case XFS_IOC_GETBMAPX:
2072 		return xfs_ioc_getbmap(filp, cmd, arg);
2073 
2074 	case FS_IOC_GETFSMAP:
2075 		return xfs_ioc_getfsmap(ip, arg);
2076 
2077 	case XFS_IOC_SCRUB_METADATA:
2078 		return xfs_ioc_scrub_metadata(ip, arg);
2079 
2080 	case XFS_IOC_FD_TO_HANDLE:
2081 	case XFS_IOC_PATH_TO_HANDLE:
2082 	case XFS_IOC_PATH_TO_FSHANDLE: {
2083 		xfs_fsop_handlereq_t	hreq;
2084 
2085 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2086 			return -EFAULT;
2087 		return xfs_find_handle(cmd, &hreq);
2088 	}
2089 	case XFS_IOC_OPEN_BY_HANDLE: {
2090 		xfs_fsop_handlereq_t	hreq;
2091 
2092 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2093 			return -EFAULT;
2094 		return xfs_open_by_handle(filp, &hreq);
2095 	}
2096 
2097 	case XFS_IOC_READLINK_BY_HANDLE: {
2098 		xfs_fsop_handlereq_t	hreq;
2099 
2100 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2101 			return -EFAULT;
2102 		return xfs_readlink_by_handle(filp, &hreq);
2103 	}
2104 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2105 		return xfs_attrlist_by_handle(filp, arg);
2106 
2107 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2108 		return xfs_attrmulti_by_handle(filp, arg);
2109 
2110 	case XFS_IOC_SWAPEXT: {
2111 		struct xfs_swapext	sxp;
2112 
2113 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2114 			return -EFAULT;
2115 		error = mnt_want_write_file(filp);
2116 		if (error)
2117 			return error;
2118 		error = xfs_ioc_swapext(&sxp);
2119 		mnt_drop_write_file(filp);
2120 		return error;
2121 	}
2122 
2123 	case XFS_IOC_FSCOUNTS: {
2124 		xfs_fsop_counts_t out;
2125 
2126 		xfs_fs_counts(mp, &out);
2127 
2128 		if (copy_to_user(arg, &out, sizeof(out)))
2129 			return -EFAULT;
2130 		return 0;
2131 	}
2132 
2133 	case XFS_IOC_SET_RESBLKS: {
2134 		xfs_fsop_resblks_t inout;
2135 		uint64_t	   in;
2136 
2137 		if (!capable(CAP_SYS_ADMIN))
2138 			return -EPERM;
2139 
2140 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2141 			return -EROFS;
2142 
2143 		if (copy_from_user(&inout, arg, sizeof(inout)))
2144 			return -EFAULT;
2145 
2146 		error = mnt_want_write_file(filp);
2147 		if (error)
2148 			return error;
2149 
2150 		/* input parameter is passed in resblks field of structure */
2151 		in = inout.resblks;
2152 		error = xfs_reserve_blocks(mp, &in, &inout);
2153 		mnt_drop_write_file(filp);
2154 		if (error)
2155 			return error;
2156 
2157 		if (copy_to_user(arg, &inout, sizeof(inout)))
2158 			return -EFAULT;
2159 		return 0;
2160 	}
2161 
2162 	case XFS_IOC_GET_RESBLKS: {
2163 		xfs_fsop_resblks_t out;
2164 
2165 		if (!capable(CAP_SYS_ADMIN))
2166 			return -EPERM;
2167 
2168 		error = xfs_reserve_blocks(mp, NULL, &out);
2169 		if (error)
2170 			return error;
2171 
2172 		if (copy_to_user(arg, &out, sizeof(out)))
2173 			return -EFAULT;
2174 
2175 		return 0;
2176 	}
2177 
2178 	case XFS_IOC_FSGROWFSDATA: {
2179 		xfs_growfs_data_t in;
2180 
2181 		if (copy_from_user(&in, arg, sizeof(in)))
2182 			return -EFAULT;
2183 
2184 		error = mnt_want_write_file(filp);
2185 		if (error)
2186 			return error;
2187 		error = xfs_growfs_data(mp, &in);
2188 		mnt_drop_write_file(filp);
2189 		return error;
2190 	}
2191 
2192 	case XFS_IOC_FSGROWFSLOG: {
2193 		xfs_growfs_log_t in;
2194 
2195 		if (copy_from_user(&in, arg, sizeof(in)))
2196 			return -EFAULT;
2197 
2198 		error = mnt_want_write_file(filp);
2199 		if (error)
2200 			return error;
2201 		error = xfs_growfs_log(mp, &in);
2202 		mnt_drop_write_file(filp);
2203 		return error;
2204 	}
2205 
2206 	case XFS_IOC_FSGROWFSRT: {
2207 		xfs_growfs_rt_t in;
2208 
2209 		if (copy_from_user(&in, arg, sizeof(in)))
2210 			return -EFAULT;
2211 
2212 		error = mnt_want_write_file(filp);
2213 		if (error)
2214 			return error;
2215 		error = xfs_growfs_rt(mp, &in);
2216 		mnt_drop_write_file(filp);
2217 		return error;
2218 	}
2219 
2220 	case XFS_IOC_GOINGDOWN: {
2221 		uint32_t in;
2222 
2223 		if (!capable(CAP_SYS_ADMIN))
2224 			return -EPERM;
2225 
2226 		if (get_user(in, (uint32_t __user *)arg))
2227 			return -EFAULT;
2228 
2229 		return xfs_fs_goingdown(mp, in);
2230 	}
2231 
2232 	case XFS_IOC_ERROR_INJECTION: {
2233 		xfs_error_injection_t in;
2234 
2235 		if (!capable(CAP_SYS_ADMIN))
2236 			return -EPERM;
2237 
2238 		if (copy_from_user(&in, arg, sizeof(in)))
2239 			return -EFAULT;
2240 
2241 		return xfs_errortag_add(mp, in.errtag);
2242 	}
2243 
2244 	case XFS_IOC_ERROR_CLEARALL:
2245 		if (!capable(CAP_SYS_ADMIN))
2246 			return -EPERM;
2247 
2248 		return xfs_errortag_clearall(mp);
2249 
2250 	case XFS_IOC_FREE_EOFBLOCKS: {
2251 		struct xfs_fs_eofblocks eofb;
2252 		struct xfs_eofblocks keofb;
2253 
2254 		if (!capable(CAP_SYS_ADMIN))
2255 			return -EPERM;
2256 
2257 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2258 			return -EROFS;
2259 
2260 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2261 			return -EFAULT;
2262 
2263 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
2264 		if (error)
2265 			return error;
2266 
2267 		return xfs_icache_free_eofblocks(mp, &keofb);
2268 	}
2269 
2270 	default:
2271 		return -ENOTTY;
2272 	}
2273 }
2274