xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision 6cc23ed2)
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 
37 #include <linux/mount.h>
38 #include <linux/namei.h>
39 
40 /*
41  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
42  * a file or fs handle.
43  *
44  * XFS_IOC_PATH_TO_FSHANDLE
45  *    returns fs handle for a mount point or path within that mount point
46  * XFS_IOC_FD_TO_HANDLE
47  *    returns full handle for a FD opened in user space
48  * XFS_IOC_PATH_TO_HANDLE
49  *    returns full handle for a path
50  */
51 int
52 xfs_find_handle(
53 	unsigned int		cmd,
54 	xfs_fsop_handlereq_t	*hreq)
55 {
56 	int			hsize;
57 	xfs_handle_t		handle;
58 	struct inode		*inode;
59 	struct fd		f = {NULL};
60 	struct path		path;
61 	int			error;
62 	struct xfs_inode	*ip;
63 
64 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
65 		f = fdget(hreq->fd);
66 		if (!f.file)
67 			return -EBADF;
68 		inode = file_inode(f.file);
69 	} else {
70 		error = user_path_at(AT_FDCWD, hreq->path, 0, &path);
71 		if (error)
72 			return error;
73 		inode = d_inode(path.dentry);
74 	}
75 	ip = XFS_I(inode);
76 
77 	/*
78 	 * We can only generate handles for inodes residing on a XFS filesystem,
79 	 * and only for regular files, directories or symbolic links.
80 	 */
81 	error = -EINVAL;
82 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
83 		goto out_put;
84 
85 	error = -EBADF;
86 	if (!S_ISREG(inode->i_mode) &&
87 	    !S_ISDIR(inode->i_mode) &&
88 	    !S_ISLNK(inode->i_mode))
89 		goto out_put;
90 
91 
92 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
93 
94 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
95 		/*
96 		 * This handle only contains an fsid, zero the rest.
97 		 */
98 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
99 		hsize = sizeof(xfs_fsid_t);
100 	} else {
101 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
102 					sizeof(handle.ha_fid.fid_len);
103 		handle.ha_fid.fid_pad = 0;
104 		handle.ha_fid.fid_gen = inode->i_generation;
105 		handle.ha_fid.fid_ino = ip->i_ino;
106 		hsize = sizeof(xfs_handle_t);
107 	}
108 
109 	error = -EFAULT;
110 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
111 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
112 		goto out_put;
113 
114 	error = 0;
115 
116  out_put:
117 	if (cmd == XFS_IOC_FD_TO_HANDLE)
118 		fdput(f);
119 	else
120 		path_put(&path);
121 	return error;
122 }
123 
124 /*
125  * No need to do permission checks on the various pathname components
126  * as the handle operations are privileged.
127  */
128 STATIC int
129 xfs_handle_acceptable(
130 	void			*context,
131 	struct dentry		*dentry)
132 {
133 	return 1;
134 }
135 
136 /*
137  * Convert userspace handle data into a dentry.
138  */
139 struct dentry *
140 xfs_handle_to_dentry(
141 	struct file		*parfilp,
142 	void __user		*uhandle,
143 	u32			hlen)
144 {
145 	xfs_handle_t		handle;
146 	struct xfs_fid64	fid;
147 
148 	/*
149 	 * Only allow handle opens under a directory.
150 	 */
151 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
152 		return ERR_PTR(-ENOTDIR);
153 
154 	if (hlen != sizeof(xfs_handle_t))
155 		return ERR_PTR(-EINVAL);
156 	if (copy_from_user(&handle, uhandle, hlen))
157 		return ERR_PTR(-EFAULT);
158 	if (handle.ha_fid.fid_len !=
159 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
160 		return ERR_PTR(-EINVAL);
161 
162 	memset(&fid, 0, sizeof(struct fid));
163 	fid.ino = handle.ha_fid.fid_ino;
164 	fid.gen = handle.ha_fid.fid_gen;
165 
166 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
167 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
168 			xfs_handle_acceptable, NULL);
169 }
170 
171 STATIC struct dentry *
172 xfs_handlereq_to_dentry(
173 	struct file		*parfilp,
174 	xfs_fsop_handlereq_t	*hreq)
175 {
176 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
177 }
178 
179 int
180 xfs_open_by_handle(
181 	struct file		*parfilp,
182 	xfs_fsop_handlereq_t	*hreq)
183 {
184 	const struct cred	*cred = current_cred();
185 	int			error;
186 	int			fd;
187 	int			permflag;
188 	struct file		*filp;
189 	struct inode		*inode;
190 	struct dentry		*dentry;
191 	fmode_t			fmode;
192 	struct path		path;
193 
194 	if (!capable(CAP_SYS_ADMIN))
195 		return -EPERM;
196 
197 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
198 	if (IS_ERR(dentry))
199 		return PTR_ERR(dentry);
200 	inode = d_inode(dentry);
201 
202 	/* Restrict xfs_open_by_handle to directories & regular files. */
203 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
204 		error = -EPERM;
205 		goto out_dput;
206 	}
207 
208 #if BITS_PER_LONG != 32
209 	hreq->oflags |= O_LARGEFILE;
210 #endif
211 
212 	permflag = hreq->oflags;
213 	fmode = OPEN_FMODE(permflag);
214 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
215 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
216 		error = -EPERM;
217 		goto out_dput;
218 	}
219 
220 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
221 		error = -EPERM;
222 		goto out_dput;
223 	}
224 
225 	/* Can't write directories. */
226 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
227 		error = -EISDIR;
228 		goto out_dput;
229 	}
230 
231 	fd = get_unused_fd_flags(0);
232 	if (fd < 0) {
233 		error = fd;
234 		goto out_dput;
235 	}
236 
237 	path.mnt = parfilp->f_path.mnt;
238 	path.dentry = dentry;
239 	filp = dentry_open(&path, hreq->oflags, cred);
240 	dput(dentry);
241 	if (IS_ERR(filp)) {
242 		put_unused_fd(fd);
243 		return PTR_ERR(filp);
244 	}
245 
246 	if (S_ISREG(inode->i_mode)) {
247 		filp->f_flags |= O_NOATIME;
248 		filp->f_mode |= FMODE_NOCMTIME;
249 	}
250 
251 	fd_install(fd, filp);
252 	return fd;
253 
254  out_dput:
255 	dput(dentry);
256 	return error;
257 }
258 
259 int
260 xfs_readlink_by_handle(
261 	struct file		*parfilp,
262 	xfs_fsop_handlereq_t	*hreq)
263 {
264 	struct dentry		*dentry;
265 	__u32			olen;
266 	int			error;
267 
268 	if (!capable(CAP_SYS_ADMIN))
269 		return -EPERM;
270 
271 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
272 	if (IS_ERR(dentry))
273 		return PTR_ERR(dentry);
274 
275 	/* Restrict this handle operation to symlinks only. */
276 	if (!d_is_symlink(dentry)) {
277 		error = -EINVAL;
278 		goto out_dput;
279 	}
280 
281 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
282 		error = -EFAULT;
283 		goto out_dput;
284 	}
285 
286 	error = vfs_readlink(dentry, hreq->ohandle, olen);
287 
288  out_dput:
289 	dput(dentry);
290 	return error;
291 }
292 
293 int
294 xfs_set_dmattrs(
295 	xfs_inode_t     *ip,
296 	uint		evmask,
297 	uint16_t	state)
298 {
299 	xfs_mount_t	*mp = ip->i_mount;
300 	xfs_trans_t	*tp;
301 	int		error;
302 
303 	if (!capable(CAP_SYS_ADMIN))
304 		return -EPERM;
305 
306 	if (XFS_FORCED_SHUTDOWN(mp))
307 		return -EIO;
308 
309 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
310 	if (error)
311 		return error;
312 
313 	xfs_ilock(ip, XFS_ILOCK_EXCL);
314 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
315 
316 	ip->i_d.di_dmevmask = evmask;
317 	ip->i_d.di_dmstate  = state;
318 
319 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
320 	error = xfs_trans_commit(tp);
321 
322 	return error;
323 }
324 
325 STATIC int
326 xfs_fssetdm_by_handle(
327 	struct file		*parfilp,
328 	void			__user *arg)
329 {
330 	int			error;
331 	struct fsdmidata	fsd;
332 	xfs_fsop_setdm_handlereq_t dmhreq;
333 	struct dentry		*dentry;
334 
335 	if (!capable(CAP_MKNOD))
336 		return -EPERM;
337 	if (copy_from_user(&dmhreq, arg, sizeof(xfs_fsop_setdm_handlereq_t)))
338 		return -EFAULT;
339 
340 	error = mnt_want_write_file(parfilp);
341 	if (error)
342 		return error;
343 
344 	dentry = xfs_handlereq_to_dentry(parfilp, &dmhreq.hreq);
345 	if (IS_ERR(dentry)) {
346 		mnt_drop_write_file(parfilp);
347 		return PTR_ERR(dentry);
348 	}
349 
350 	if (IS_IMMUTABLE(d_inode(dentry)) || IS_APPEND(d_inode(dentry))) {
351 		error = -EPERM;
352 		goto out;
353 	}
354 
355 	if (copy_from_user(&fsd, dmhreq.data, sizeof(fsd))) {
356 		error = -EFAULT;
357 		goto out;
358 	}
359 
360 	error = xfs_set_dmattrs(XFS_I(d_inode(dentry)), fsd.fsd_dmevmask,
361 				 fsd.fsd_dmstate);
362 
363  out:
364 	mnt_drop_write_file(parfilp);
365 	dput(dentry);
366 	return error;
367 }
368 
369 STATIC int
370 xfs_attrlist_by_handle(
371 	struct file		*parfilp,
372 	void			__user *arg)
373 {
374 	int			error = -ENOMEM;
375 	attrlist_cursor_kern_t	*cursor;
376 	struct xfs_fsop_attrlist_handlereq __user	*p = arg;
377 	xfs_fsop_attrlist_handlereq_t al_hreq;
378 	struct dentry		*dentry;
379 	char			*kbuf;
380 
381 	if (!capable(CAP_SYS_ADMIN))
382 		return -EPERM;
383 	if (copy_from_user(&al_hreq, arg, sizeof(xfs_fsop_attrlist_handlereq_t)))
384 		return -EFAULT;
385 	if (al_hreq.buflen < sizeof(struct attrlist) ||
386 	    al_hreq.buflen > XFS_XATTR_LIST_MAX)
387 		return -EINVAL;
388 
389 	/*
390 	 * Reject flags, only allow namespaces.
391 	 */
392 	if (al_hreq.flags & ~(ATTR_ROOT | ATTR_SECURE))
393 		return -EINVAL;
394 
395 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
396 	if (IS_ERR(dentry))
397 		return PTR_ERR(dentry);
398 
399 	kbuf = kmem_zalloc_large(al_hreq.buflen, 0);
400 	if (!kbuf)
401 		goto out_dput;
402 
403 	cursor = (attrlist_cursor_kern_t *)&al_hreq.pos;
404 	error = xfs_attr_list(XFS_I(d_inode(dentry)), kbuf, al_hreq.buflen,
405 					al_hreq.flags, cursor);
406 	if (error)
407 		goto out_kfree;
408 
409 	if (copy_to_user(&p->pos, cursor, sizeof(attrlist_cursor_kern_t))) {
410 		error = -EFAULT;
411 		goto out_kfree;
412 	}
413 
414 	if (copy_to_user(al_hreq.buffer, kbuf, al_hreq.buflen))
415 		error = -EFAULT;
416 
417 out_kfree:
418 	kmem_free(kbuf);
419 out_dput:
420 	dput(dentry);
421 	return error;
422 }
423 
424 int
425 xfs_attrmulti_attr_get(
426 	struct inode		*inode,
427 	unsigned char		*name,
428 	unsigned char		__user *ubuf,
429 	uint32_t		*len,
430 	uint32_t		flags)
431 {
432 	unsigned char		*kbuf;
433 	int			error = -EFAULT;
434 
435 	if (*len > XFS_XATTR_SIZE_MAX)
436 		return -EINVAL;
437 	kbuf = kmem_zalloc_large(*len, 0);
438 	if (!kbuf)
439 		return -ENOMEM;
440 
441 	error = xfs_attr_get(XFS_I(inode), name, &kbuf, (int *)len, flags);
442 	if (error)
443 		goto out_kfree;
444 
445 	if (copy_to_user(ubuf, kbuf, *len))
446 		error = -EFAULT;
447 
448 out_kfree:
449 	kmem_free(kbuf);
450 	return error;
451 }
452 
453 int
454 xfs_attrmulti_attr_set(
455 	struct inode		*inode,
456 	unsigned char		*name,
457 	const unsigned char	__user *ubuf,
458 	uint32_t		len,
459 	uint32_t		flags)
460 {
461 	unsigned char		*kbuf;
462 	int			error;
463 
464 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
465 		return -EPERM;
466 	if (len > XFS_XATTR_SIZE_MAX)
467 		return -EINVAL;
468 
469 	kbuf = memdup_user(ubuf, len);
470 	if (IS_ERR(kbuf))
471 		return PTR_ERR(kbuf);
472 
473 	error = xfs_attr_set(XFS_I(inode), name, kbuf, len, flags);
474 	if (!error)
475 		xfs_forget_acl(inode, name, flags);
476 	kfree(kbuf);
477 	return error;
478 }
479 
480 int
481 xfs_attrmulti_attr_remove(
482 	struct inode		*inode,
483 	unsigned char		*name,
484 	uint32_t		flags)
485 {
486 	int			error;
487 
488 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
489 		return -EPERM;
490 	error = xfs_attr_remove(XFS_I(inode), name, flags);
491 	if (!error)
492 		xfs_forget_acl(inode, name, flags);
493 	return error;
494 }
495 
496 STATIC int
497 xfs_attrmulti_by_handle(
498 	struct file		*parfilp,
499 	void			__user *arg)
500 {
501 	int			error;
502 	xfs_attr_multiop_t	*ops;
503 	xfs_fsop_attrmulti_handlereq_t am_hreq;
504 	struct dentry		*dentry;
505 	unsigned int		i, size;
506 	unsigned char		*attr_name;
507 
508 	if (!capable(CAP_SYS_ADMIN))
509 		return -EPERM;
510 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
511 		return -EFAULT;
512 
513 	/* overflow check */
514 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
515 		return -E2BIG;
516 
517 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
518 	if (IS_ERR(dentry))
519 		return PTR_ERR(dentry);
520 
521 	error = -E2BIG;
522 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
523 	if (!size || size > 16 * PAGE_SIZE)
524 		goto out_dput;
525 
526 	ops = memdup_user(am_hreq.ops, size);
527 	if (IS_ERR(ops)) {
528 		error = PTR_ERR(ops);
529 		goto out_dput;
530 	}
531 
532 	error = -ENOMEM;
533 	attr_name = kmalloc(MAXNAMELEN, GFP_KERNEL);
534 	if (!attr_name)
535 		goto out_kfree_ops;
536 
537 	error = 0;
538 	for (i = 0; i < am_hreq.opcount; i++) {
539 		ops[i].am_error = strncpy_from_user((char *)attr_name,
540 				ops[i].am_attrname, MAXNAMELEN);
541 		if (ops[i].am_error == 0 || ops[i].am_error == MAXNAMELEN)
542 			error = -ERANGE;
543 		if (ops[i].am_error < 0)
544 			break;
545 
546 		switch (ops[i].am_opcode) {
547 		case ATTR_OP_GET:
548 			ops[i].am_error = xfs_attrmulti_attr_get(
549 					d_inode(dentry), attr_name,
550 					ops[i].am_attrvalue, &ops[i].am_length,
551 					ops[i].am_flags);
552 			break;
553 		case ATTR_OP_SET:
554 			ops[i].am_error = mnt_want_write_file(parfilp);
555 			if (ops[i].am_error)
556 				break;
557 			ops[i].am_error = xfs_attrmulti_attr_set(
558 					d_inode(dentry), attr_name,
559 					ops[i].am_attrvalue, ops[i].am_length,
560 					ops[i].am_flags);
561 			mnt_drop_write_file(parfilp);
562 			break;
563 		case ATTR_OP_REMOVE:
564 			ops[i].am_error = mnt_want_write_file(parfilp);
565 			if (ops[i].am_error)
566 				break;
567 			ops[i].am_error = xfs_attrmulti_attr_remove(
568 					d_inode(dentry), attr_name,
569 					ops[i].am_flags);
570 			mnt_drop_write_file(parfilp);
571 			break;
572 		default:
573 			ops[i].am_error = -EINVAL;
574 		}
575 	}
576 
577 	if (copy_to_user(am_hreq.ops, ops, size))
578 		error = -EFAULT;
579 
580 	kfree(attr_name);
581  out_kfree_ops:
582 	kfree(ops);
583  out_dput:
584 	dput(dentry);
585 	return error;
586 }
587 
588 int
589 xfs_ioc_space(
590 	struct file		*filp,
591 	unsigned int		cmd,
592 	xfs_flock64_t		*bf)
593 {
594 	struct inode		*inode = file_inode(filp);
595 	struct xfs_inode	*ip = XFS_I(inode);
596 	struct iattr		iattr;
597 	enum xfs_prealloc_flags	flags = 0;
598 	uint			iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
599 	int			error;
600 
601 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND))
602 		return -EPERM;
603 
604 	if (!(filp->f_mode & FMODE_WRITE))
605 		return -EBADF;
606 
607 	if (!S_ISREG(inode->i_mode))
608 		return -EINVAL;
609 
610 	if (filp->f_flags & O_DSYNC)
611 		flags |= XFS_PREALLOC_SYNC;
612 	if (filp->f_mode & FMODE_NOCMTIME)
613 		flags |= XFS_PREALLOC_INVISIBLE;
614 
615 	error = mnt_want_write_file(filp);
616 	if (error)
617 		return error;
618 
619 	xfs_ilock(ip, iolock);
620 	error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
621 	if (error)
622 		goto out_unlock;
623 
624 	switch (bf->l_whence) {
625 	case 0: /*SEEK_SET*/
626 		break;
627 	case 1: /*SEEK_CUR*/
628 		bf->l_start += filp->f_pos;
629 		break;
630 	case 2: /*SEEK_END*/
631 		bf->l_start += XFS_ISIZE(ip);
632 		break;
633 	default:
634 		error = -EINVAL;
635 		goto out_unlock;
636 	}
637 
638 	/*
639 	 * length of <= 0 for resv/unresv/zero is invalid.  length for
640 	 * alloc/free is ignored completely and we have no idea what userspace
641 	 * might have set it to, so set it to zero to allow range
642 	 * checks to pass.
643 	 */
644 	switch (cmd) {
645 	case XFS_IOC_ZERO_RANGE:
646 	case XFS_IOC_RESVSP:
647 	case XFS_IOC_RESVSP64:
648 	case XFS_IOC_UNRESVSP:
649 	case XFS_IOC_UNRESVSP64:
650 		if (bf->l_len <= 0) {
651 			error = -EINVAL;
652 			goto out_unlock;
653 		}
654 		break;
655 	default:
656 		bf->l_len = 0;
657 		break;
658 	}
659 
660 	if (bf->l_start < 0 ||
661 	    bf->l_start > inode->i_sb->s_maxbytes ||
662 	    bf->l_start + bf->l_len < 0 ||
663 	    bf->l_start + bf->l_len >= inode->i_sb->s_maxbytes) {
664 		error = -EINVAL;
665 		goto out_unlock;
666 	}
667 
668 	switch (cmd) {
669 	case XFS_IOC_ZERO_RANGE:
670 		flags |= XFS_PREALLOC_SET;
671 		error = xfs_zero_file_space(ip, bf->l_start, bf->l_len);
672 		break;
673 	case XFS_IOC_RESVSP:
674 	case XFS_IOC_RESVSP64:
675 		flags |= XFS_PREALLOC_SET;
676 		error = xfs_alloc_file_space(ip, bf->l_start, bf->l_len,
677 						XFS_BMAPI_PREALLOC);
678 		break;
679 	case XFS_IOC_UNRESVSP:
680 	case XFS_IOC_UNRESVSP64:
681 		error = xfs_free_file_space(ip, bf->l_start, bf->l_len);
682 		break;
683 	case XFS_IOC_ALLOCSP:
684 	case XFS_IOC_ALLOCSP64:
685 	case XFS_IOC_FREESP:
686 	case XFS_IOC_FREESP64:
687 		flags |= XFS_PREALLOC_CLEAR;
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 
698 		error = xfs_vn_setattr_size(file_dentry(filp), &iattr);
699 		break;
700 	default:
701 		ASSERT(0);
702 		error = -EINVAL;
703 	}
704 
705 	if (error)
706 		goto out_unlock;
707 
708 	error = xfs_update_prealloc_flags(ip, flags);
709 
710 out_unlock:
711 	xfs_iunlock(ip, iolock);
712 	mnt_drop_write_file(filp);
713 	return error;
714 }
715 
716 /* Return 0 on success or positive error */
717 int
718 xfs_fsbulkstat_one_fmt(
719 	struct xfs_ibulk		*breq,
720 	const struct xfs_bulkstat	*bstat)
721 {
722 	struct xfs_bstat		bs1;
723 
724 	xfs_bulkstat_to_bstat(breq->mp, &bs1, bstat);
725 	if (copy_to_user(breq->ubuffer, &bs1, sizeof(bs1)))
726 		return -EFAULT;
727 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bstat));
728 }
729 
730 int
731 xfs_fsinumbers_fmt(
732 	struct xfs_ibulk		*breq,
733 	const struct xfs_inumbers	*igrp)
734 {
735 	struct xfs_inogrp		ig1;
736 
737 	xfs_inumbers_to_inogrp(&ig1, igrp);
738 	if (copy_to_user(breq->ubuffer, &ig1, sizeof(struct xfs_inogrp)))
739 		return -EFAULT;
740 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inogrp));
741 }
742 
743 STATIC int
744 xfs_ioc_fsbulkstat(
745 	xfs_mount_t		*mp,
746 	unsigned int		cmd,
747 	void			__user *arg)
748 {
749 	struct xfs_fsop_bulkreq	bulkreq;
750 	struct xfs_ibulk	breq = {
751 		.mp		= mp,
752 		.ocount		= 0,
753 	};
754 	xfs_ino_t		lastino;
755 	int			error;
756 
757 	/* done = 1 if there are more stats to get and if bulkstat */
758 	/* should be called again (unused here, but used in dmapi) */
759 
760 	if (!capable(CAP_SYS_ADMIN))
761 		return -EPERM;
762 
763 	if (XFS_FORCED_SHUTDOWN(mp))
764 		return -EIO;
765 
766 	if (copy_from_user(&bulkreq, arg, sizeof(struct xfs_fsop_bulkreq)))
767 		return -EFAULT;
768 
769 	if (copy_from_user(&lastino, bulkreq.lastip, sizeof(__s64)))
770 		return -EFAULT;
771 
772 	if (bulkreq.icount <= 0)
773 		return -EINVAL;
774 
775 	if (bulkreq.ubuffer == NULL)
776 		return -EINVAL;
777 
778 	breq.ubuffer = bulkreq.ubuffer;
779 	breq.icount = bulkreq.icount;
780 
781 	/*
782 	 * FSBULKSTAT_SINGLE expects that *lastip contains the inode number
783 	 * that we want to stat.  However, FSINUMBERS and FSBULKSTAT expect
784 	 * that *lastip contains either zero or the number of the last inode to
785 	 * be examined by the previous call and return results starting with
786 	 * the next inode after that.  The new bulk request back end functions
787 	 * take the inode to start with, so we have to compute the startino
788 	 * parameter from lastino to maintain correct function.  lastino == 0
789 	 * is a special case because it has traditionally meant "first inode
790 	 * in filesystem".
791 	 */
792 	if (cmd == XFS_IOC_FSINUMBERS) {
793 		breq.startino = lastino ? lastino + 1 : 0;
794 		error = xfs_inumbers(&breq, xfs_fsinumbers_fmt);
795 		lastino = breq.startino - 1;
796 	} else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE) {
797 		breq.startino = lastino;
798 		breq.icount = 1;
799 		error = xfs_bulkstat_one(&breq, xfs_fsbulkstat_one_fmt);
800 	} else {	/* XFS_IOC_FSBULKSTAT */
801 		breq.startino = lastino ? lastino + 1 : 0;
802 		error = xfs_bulkstat(&breq, xfs_fsbulkstat_one_fmt);
803 		lastino = breq.startino - 1;
804 	}
805 
806 	if (error)
807 		return error;
808 
809 	if (bulkreq.lastip != NULL &&
810 	    copy_to_user(bulkreq.lastip, &lastino, sizeof(xfs_ino_t)))
811 		return -EFAULT;
812 
813 	if (bulkreq.ocount != NULL &&
814 	    copy_to_user(bulkreq.ocount, &breq.ocount, sizeof(__s32)))
815 		return -EFAULT;
816 
817 	return 0;
818 }
819 
820 /* Return 0 on success or positive error */
821 static int
822 xfs_bulkstat_fmt(
823 	struct xfs_ibulk		*breq,
824 	const struct xfs_bulkstat	*bstat)
825 {
826 	if (copy_to_user(breq->ubuffer, bstat, sizeof(struct xfs_bulkstat)))
827 		return -EFAULT;
828 	return xfs_ibulk_advance(breq, sizeof(struct xfs_bulkstat));
829 }
830 
831 /*
832  * Check the incoming bulk request @hdr from userspace and initialize the
833  * internal @breq bulk request appropriately.  Returns 0 if the bulk request
834  * should proceed; -ECANCELED if there's nothing to do; or the usual
835  * negative error code.
836  */
837 static int
838 xfs_bulk_ireq_setup(
839 	struct xfs_mount	*mp,
840 	struct xfs_bulk_ireq	*hdr,
841 	struct xfs_ibulk	*breq,
842 	void __user		*ubuffer)
843 {
844 	if (hdr->icount == 0 ||
845 	    (hdr->flags & ~XFS_BULK_IREQ_FLAGS_ALL) ||
846 	    memchr_inv(hdr->reserved, 0, sizeof(hdr->reserved)))
847 		return -EINVAL;
848 
849 	breq->startino = hdr->ino;
850 	breq->ubuffer = ubuffer;
851 	breq->icount = hdr->icount;
852 	breq->ocount = 0;
853 	breq->flags = 0;
854 
855 	/*
856 	 * The @ino parameter is a special value, so we must look it up here.
857 	 * We're not allowed to have IREQ_AGNO, and we only return one inode
858 	 * worth of data.
859 	 */
860 	if (hdr->flags & XFS_BULK_IREQ_SPECIAL) {
861 		if (hdr->flags & XFS_BULK_IREQ_AGNO)
862 			return -EINVAL;
863 
864 		switch (hdr->ino) {
865 		case XFS_BULK_IREQ_SPECIAL_ROOT:
866 			hdr->ino = mp->m_sb.sb_rootino;
867 			break;
868 		default:
869 			return -EINVAL;
870 		}
871 		breq->icount = 1;
872 	}
873 
874 	/*
875 	 * The IREQ_AGNO flag means that we only want results from a given AG.
876 	 * If @hdr->ino is zero, we start iterating in that AG.  If @hdr->ino is
877 	 * beyond the specified AG then we return no results.
878 	 */
879 	if (hdr->flags & XFS_BULK_IREQ_AGNO) {
880 		if (hdr->agno >= mp->m_sb.sb_agcount)
881 			return -EINVAL;
882 
883 		if (breq->startino == 0)
884 			breq->startino = XFS_AGINO_TO_INO(mp, hdr->agno, 0);
885 		else if (XFS_INO_TO_AGNO(mp, breq->startino) < hdr->agno)
886 			return -EINVAL;
887 
888 		breq->flags |= XFS_IBULK_SAME_AG;
889 
890 		/* Asking for an inode past the end of the AG?  We're done! */
891 		if (XFS_INO_TO_AGNO(mp, breq->startino) > hdr->agno)
892 			return -ECANCELED;
893 	} else if (hdr->agno)
894 		return -EINVAL;
895 
896 	/* Asking for an inode past the end of the FS?  We're done! */
897 	if (XFS_INO_TO_AGNO(mp, breq->startino) >= mp->m_sb.sb_agcount)
898 		return -ECANCELED;
899 
900 	return 0;
901 }
902 
903 /*
904  * Update the userspace bulk request @hdr to reflect the end state of the
905  * internal bulk request @breq.
906  */
907 static void
908 xfs_bulk_ireq_teardown(
909 	struct xfs_bulk_ireq	*hdr,
910 	struct xfs_ibulk	*breq)
911 {
912 	hdr->ino = breq->startino;
913 	hdr->ocount = breq->ocount;
914 }
915 
916 /* Handle the v5 bulkstat ioctl. */
917 STATIC int
918 xfs_ioc_bulkstat(
919 	struct xfs_mount		*mp,
920 	unsigned int			cmd,
921 	struct xfs_bulkstat_req __user	*arg)
922 {
923 	struct xfs_bulk_ireq		hdr;
924 	struct xfs_ibulk		breq = {
925 		.mp			= mp,
926 	};
927 	int				error;
928 
929 	if (!capable(CAP_SYS_ADMIN))
930 		return -EPERM;
931 
932 	if (XFS_FORCED_SHUTDOWN(mp))
933 		return -EIO;
934 
935 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
936 		return -EFAULT;
937 
938 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->bulkstat);
939 	if (error == -ECANCELED)
940 		goto out_teardown;
941 	if (error < 0)
942 		return error;
943 
944 	error = xfs_bulkstat(&breq, xfs_bulkstat_fmt);
945 	if (error)
946 		return error;
947 
948 out_teardown:
949 	xfs_bulk_ireq_teardown(&hdr, &breq);
950 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
951 		return -EFAULT;
952 
953 	return 0;
954 }
955 
956 STATIC int
957 xfs_inumbers_fmt(
958 	struct xfs_ibulk		*breq,
959 	const struct xfs_inumbers	*igrp)
960 {
961 	if (copy_to_user(breq->ubuffer, igrp, sizeof(struct xfs_inumbers)))
962 		return -EFAULT;
963 	return xfs_ibulk_advance(breq, sizeof(struct xfs_inumbers));
964 }
965 
966 /* Handle the v5 inumbers ioctl. */
967 STATIC int
968 xfs_ioc_inumbers(
969 	struct xfs_mount		*mp,
970 	unsigned int			cmd,
971 	struct xfs_inumbers_req __user	*arg)
972 {
973 	struct xfs_bulk_ireq		hdr;
974 	struct xfs_ibulk		breq = {
975 		.mp			= mp,
976 	};
977 	int				error;
978 
979 	if (!capable(CAP_SYS_ADMIN))
980 		return -EPERM;
981 
982 	if (XFS_FORCED_SHUTDOWN(mp))
983 		return -EIO;
984 
985 	if (copy_from_user(&hdr, &arg->hdr, sizeof(hdr)))
986 		return -EFAULT;
987 
988 	error = xfs_bulk_ireq_setup(mp, &hdr, &breq, arg->inumbers);
989 	if (error == -ECANCELED)
990 		goto out_teardown;
991 	if (error < 0)
992 		return error;
993 
994 	error = xfs_inumbers(&breq, xfs_inumbers_fmt);
995 	if (error)
996 		return error;
997 
998 out_teardown:
999 	xfs_bulk_ireq_teardown(&hdr, &breq);
1000 	if (copy_to_user(&arg->hdr, &hdr, sizeof(hdr)))
1001 		return -EFAULT;
1002 
1003 	return 0;
1004 }
1005 
1006 STATIC int
1007 xfs_ioc_fsgeometry(
1008 	struct xfs_mount	*mp,
1009 	void			__user *arg,
1010 	int			struct_version)
1011 {
1012 	struct xfs_fsop_geom	fsgeo;
1013 	size_t			len;
1014 
1015 	xfs_fs_geometry(&mp->m_sb, &fsgeo, struct_version);
1016 
1017 	if (struct_version <= 3)
1018 		len = sizeof(struct xfs_fsop_geom_v1);
1019 	else if (struct_version == 4)
1020 		len = sizeof(struct xfs_fsop_geom_v4);
1021 	else {
1022 		xfs_fsop_geom_health(mp, &fsgeo);
1023 		len = sizeof(fsgeo);
1024 	}
1025 
1026 	if (copy_to_user(arg, &fsgeo, len))
1027 		return -EFAULT;
1028 	return 0;
1029 }
1030 
1031 STATIC int
1032 xfs_ioc_ag_geometry(
1033 	struct xfs_mount	*mp,
1034 	void			__user *arg)
1035 {
1036 	struct xfs_ag_geometry	ageo;
1037 	int			error;
1038 
1039 	if (copy_from_user(&ageo, arg, sizeof(ageo)))
1040 		return -EFAULT;
1041 	if (ageo.ag_flags)
1042 		return -EINVAL;
1043 	if (memchr_inv(&ageo.ag_reserved, 0, sizeof(ageo.ag_reserved)))
1044 		return -EINVAL;
1045 
1046 	error = xfs_ag_get_geometry(mp, ageo.ag_number, &ageo);
1047 	if (error)
1048 		return error;
1049 
1050 	if (copy_to_user(arg, &ageo, sizeof(ageo)))
1051 		return -EFAULT;
1052 	return 0;
1053 }
1054 
1055 /*
1056  * Linux extended inode flags interface.
1057  */
1058 
1059 STATIC unsigned int
1060 xfs_merge_ioc_xflags(
1061 	unsigned int	flags,
1062 	unsigned int	start)
1063 {
1064 	unsigned int	xflags = start;
1065 
1066 	if (flags & FS_IMMUTABLE_FL)
1067 		xflags |= FS_XFLAG_IMMUTABLE;
1068 	else
1069 		xflags &= ~FS_XFLAG_IMMUTABLE;
1070 	if (flags & FS_APPEND_FL)
1071 		xflags |= FS_XFLAG_APPEND;
1072 	else
1073 		xflags &= ~FS_XFLAG_APPEND;
1074 	if (flags & FS_SYNC_FL)
1075 		xflags |= FS_XFLAG_SYNC;
1076 	else
1077 		xflags &= ~FS_XFLAG_SYNC;
1078 	if (flags & FS_NOATIME_FL)
1079 		xflags |= FS_XFLAG_NOATIME;
1080 	else
1081 		xflags &= ~FS_XFLAG_NOATIME;
1082 	if (flags & FS_NODUMP_FL)
1083 		xflags |= FS_XFLAG_NODUMP;
1084 	else
1085 		xflags &= ~FS_XFLAG_NODUMP;
1086 
1087 	return xflags;
1088 }
1089 
1090 STATIC unsigned int
1091 xfs_di2lxflags(
1092 	uint16_t	di_flags)
1093 {
1094 	unsigned int	flags = 0;
1095 
1096 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
1097 		flags |= FS_IMMUTABLE_FL;
1098 	if (di_flags & XFS_DIFLAG_APPEND)
1099 		flags |= FS_APPEND_FL;
1100 	if (di_flags & XFS_DIFLAG_SYNC)
1101 		flags |= FS_SYNC_FL;
1102 	if (di_flags & XFS_DIFLAG_NOATIME)
1103 		flags |= FS_NOATIME_FL;
1104 	if (di_flags & XFS_DIFLAG_NODUMP)
1105 		flags |= FS_NODUMP_FL;
1106 	return flags;
1107 }
1108 
1109 static void
1110 xfs_fill_fsxattr(
1111 	struct xfs_inode	*ip,
1112 	bool			attr,
1113 	struct fsxattr		*fa)
1114 {
1115 	simple_fill_fsxattr(fa, xfs_ip2xflags(ip));
1116 	fa->fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
1117 	fa->fsx_cowextsize = ip->i_d.di_cowextsize <<
1118 			ip->i_mount->m_sb.sb_blocklog;
1119 	fa->fsx_projid = xfs_get_projid(ip);
1120 
1121 	if (attr) {
1122 		if (ip->i_afp) {
1123 			if (ip->i_afp->if_flags & XFS_IFEXTENTS)
1124 				fa->fsx_nextents = xfs_iext_count(ip->i_afp);
1125 			else
1126 				fa->fsx_nextents = ip->i_d.di_anextents;
1127 		} else
1128 			fa->fsx_nextents = 0;
1129 	} else {
1130 		if (ip->i_df.if_flags & XFS_IFEXTENTS)
1131 			fa->fsx_nextents = xfs_iext_count(&ip->i_df);
1132 		else
1133 			fa->fsx_nextents = ip->i_d.di_nextents;
1134 	}
1135 }
1136 
1137 STATIC int
1138 xfs_ioc_fsgetxattr(
1139 	xfs_inode_t		*ip,
1140 	int			attr,
1141 	void			__user *arg)
1142 {
1143 	struct fsxattr		fa;
1144 
1145 	xfs_ilock(ip, XFS_ILOCK_SHARED);
1146 	xfs_fill_fsxattr(ip, attr, &fa);
1147 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1148 
1149 	if (copy_to_user(arg, &fa, sizeof(fa)))
1150 		return -EFAULT;
1151 	return 0;
1152 }
1153 
1154 STATIC uint16_t
1155 xfs_flags2diflags(
1156 	struct xfs_inode	*ip,
1157 	unsigned int		xflags)
1158 {
1159 	/* can't set PREALLOC this way, just preserve it */
1160 	uint16_t		di_flags =
1161 		(ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
1162 
1163 	if (xflags & FS_XFLAG_IMMUTABLE)
1164 		di_flags |= XFS_DIFLAG_IMMUTABLE;
1165 	if (xflags & FS_XFLAG_APPEND)
1166 		di_flags |= XFS_DIFLAG_APPEND;
1167 	if (xflags & FS_XFLAG_SYNC)
1168 		di_flags |= XFS_DIFLAG_SYNC;
1169 	if (xflags & FS_XFLAG_NOATIME)
1170 		di_flags |= XFS_DIFLAG_NOATIME;
1171 	if (xflags & FS_XFLAG_NODUMP)
1172 		di_flags |= XFS_DIFLAG_NODUMP;
1173 	if (xflags & FS_XFLAG_NODEFRAG)
1174 		di_flags |= XFS_DIFLAG_NODEFRAG;
1175 	if (xflags & FS_XFLAG_FILESTREAM)
1176 		di_flags |= XFS_DIFLAG_FILESTREAM;
1177 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1178 		if (xflags & FS_XFLAG_RTINHERIT)
1179 			di_flags |= XFS_DIFLAG_RTINHERIT;
1180 		if (xflags & FS_XFLAG_NOSYMLINKS)
1181 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
1182 		if (xflags & FS_XFLAG_EXTSZINHERIT)
1183 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1184 		if (xflags & FS_XFLAG_PROJINHERIT)
1185 			di_flags |= XFS_DIFLAG_PROJINHERIT;
1186 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
1187 		if (xflags & FS_XFLAG_REALTIME)
1188 			di_flags |= XFS_DIFLAG_REALTIME;
1189 		if (xflags & FS_XFLAG_EXTSIZE)
1190 			di_flags |= XFS_DIFLAG_EXTSIZE;
1191 	}
1192 
1193 	return di_flags;
1194 }
1195 
1196 STATIC uint64_t
1197 xfs_flags2diflags2(
1198 	struct xfs_inode	*ip,
1199 	unsigned int		xflags)
1200 {
1201 	uint64_t		di_flags2 =
1202 		(ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK);
1203 
1204 	if (xflags & FS_XFLAG_DAX)
1205 		di_flags2 |= XFS_DIFLAG2_DAX;
1206 	if (xflags & FS_XFLAG_COWEXTSIZE)
1207 		di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
1208 
1209 	return di_flags2;
1210 }
1211 
1212 STATIC void
1213 xfs_diflags_to_linux(
1214 	struct xfs_inode	*ip)
1215 {
1216 	struct inode		*inode = VFS_I(ip);
1217 	unsigned int		xflags = xfs_ip2xflags(ip);
1218 
1219 	if (xflags & FS_XFLAG_IMMUTABLE)
1220 		inode->i_flags |= S_IMMUTABLE;
1221 	else
1222 		inode->i_flags &= ~S_IMMUTABLE;
1223 	if (xflags & FS_XFLAG_APPEND)
1224 		inode->i_flags |= S_APPEND;
1225 	else
1226 		inode->i_flags &= ~S_APPEND;
1227 	if (xflags & FS_XFLAG_SYNC)
1228 		inode->i_flags |= S_SYNC;
1229 	else
1230 		inode->i_flags &= ~S_SYNC;
1231 	if (xflags & FS_XFLAG_NOATIME)
1232 		inode->i_flags |= S_NOATIME;
1233 	else
1234 		inode->i_flags &= ~S_NOATIME;
1235 #if 0	/* disabled until the flag switching races are sorted out */
1236 	if (xflags & FS_XFLAG_DAX)
1237 		inode->i_flags |= S_DAX;
1238 	else
1239 		inode->i_flags &= ~S_DAX;
1240 #endif
1241 }
1242 
1243 static int
1244 xfs_ioctl_setattr_xflags(
1245 	struct xfs_trans	*tp,
1246 	struct xfs_inode	*ip,
1247 	struct fsxattr		*fa)
1248 {
1249 	struct xfs_mount	*mp = ip->i_mount;
1250 	uint64_t		di_flags2;
1251 
1252 	/* Can't change realtime flag if any extents are allocated. */
1253 	if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
1254 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1255 		return -EINVAL;
1256 
1257 	/* If realtime flag is set then must have realtime device */
1258 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1259 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1260 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1261 			return -EINVAL;
1262 	}
1263 
1264 	/* Clear reflink if we are actually able to set the rt flag. */
1265 	if ((fa->fsx_xflags & FS_XFLAG_REALTIME) && xfs_is_reflink_inode(ip))
1266 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1267 
1268 	/* Don't allow us to set DAX mode for a reflinked file for now. */
1269 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && xfs_is_reflink_inode(ip))
1270 		return -EINVAL;
1271 
1272 	/* diflags2 only valid for v3 inodes. */
1273 	di_flags2 = xfs_flags2diflags2(ip, fa->fsx_xflags);
1274 	if (di_flags2 && ip->i_d.di_version < 3)
1275 		return -EINVAL;
1276 
1277 	ip->i_d.di_flags = xfs_flags2diflags(ip, fa->fsx_xflags);
1278 	ip->i_d.di_flags2 = di_flags2;
1279 
1280 	xfs_diflags_to_linux(ip);
1281 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1282 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1283 	XFS_STATS_INC(mp, xs_ig_attrchg);
1284 	return 0;
1285 }
1286 
1287 /*
1288  * If we are changing DAX flags, we have to ensure the file is clean and any
1289  * cached objects in the address space are invalidated and removed. This
1290  * requires us to lock out other IO and page faults similar to a truncate
1291  * operation. The locks need to be held until the transaction has been committed
1292  * so that the cache invalidation is atomic with respect to the DAX flag
1293  * manipulation.
1294  */
1295 static int
1296 xfs_ioctl_setattr_dax_invalidate(
1297 	struct xfs_inode	*ip,
1298 	struct fsxattr		*fa,
1299 	int			*join_flags)
1300 {
1301 	struct inode		*inode = VFS_I(ip);
1302 	struct super_block	*sb = inode->i_sb;
1303 	int			error;
1304 
1305 	*join_flags = 0;
1306 
1307 	/*
1308 	 * It is only valid to set the DAX flag on regular files and
1309 	 * directories on filesystems where the block size is equal to the page
1310 	 * size. On directories it serves as an inherited hint so we don't
1311 	 * have to check the device for dax support or flush pagecache.
1312 	 */
1313 	if (fa->fsx_xflags & FS_XFLAG_DAX) {
1314 		if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1315 			return -EINVAL;
1316 		if (!bdev_dax_supported(xfs_find_bdev_for_inode(VFS_I(ip)),
1317 				sb->s_blocksize))
1318 			return -EINVAL;
1319 	}
1320 
1321 	/* If the DAX state is not changing, we have nothing to do here. */
1322 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && IS_DAX(inode))
1323 		return 0;
1324 	if (!(fa->fsx_xflags & FS_XFLAG_DAX) && !IS_DAX(inode))
1325 		return 0;
1326 
1327 	if (S_ISDIR(inode->i_mode))
1328 		return 0;
1329 
1330 	/* lock, flush and invalidate mapping in preparation for flag change */
1331 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1332 	error = filemap_write_and_wait(inode->i_mapping);
1333 	if (error)
1334 		goto out_unlock;
1335 	error = invalidate_inode_pages2(inode->i_mapping);
1336 	if (error)
1337 		goto out_unlock;
1338 
1339 	*join_flags = XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL;
1340 	return 0;
1341 
1342 out_unlock:
1343 	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1344 	return error;
1345 
1346 }
1347 
1348 /*
1349  * Set up the transaction structure for the setattr operation, checking that we
1350  * have permission to do so. On success, return a clean transaction and the
1351  * inode locked exclusively ready for further operation specific checks. On
1352  * failure, return an error without modifying or locking the inode.
1353  *
1354  * The inode might already be IO locked on call. If this is the case, it is
1355  * indicated in @join_flags and we take full responsibility for ensuring they
1356  * are unlocked from now on. Hence if we have an error here, we still have to
1357  * unlock them. Otherwise, once they are joined to the transaction, they will
1358  * be unlocked on commit/cancel.
1359  */
1360 static struct xfs_trans *
1361 xfs_ioctl_setattr_get_trans(
1362 	struct xfs_inode	*ip,
1363 	int			join_flags)
1364 {
1365 	struct xfs_mount	*mp = ip->i_mount;
1366 	struct xfs_trans	*tp;
1367 	int			error = -EROFS;
1368 
1369 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1370 		goto out_unlock;
1371 	error = -EIO;
1372 	if (XFS_FORCED_SHUTDOWN(mp))
1373 		goto out_unlock;
1374 
1375 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1376 	if (error)
1377 		goto out_unlock;
1378 
1379 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1380 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | join_flags);
1381 	join_flags = 0;
1382 
1383 	/*
1384 	 * CAP_FOWNER overrides the following restrictions:
1385 	 *
1386 	 * The user ID of the calling process must be equal to the file owner
1387 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1388 	 */
1389 	if (!inode_owner_or_capable(VFS_I(ip))) {
1390 		error = -EPERM;
1391 		goto out_cancel;
1392 	}
1393 
1394 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1395 		xfs_trans_set_sync(tp);
1396 
1397 	return tp;
1398 
1399 out_cancel:
1400 	xfs_trans_cancel(tp);
1401 out_unlock:
1402 	if (join_flags)
1403 		xfs_iunlock(ip, join_flags);
1404 	return ERR_PTR(error);
1405 }
1406 
1407 /*
1408  * extent size hint validation is somewhat cumbersome. Rules are:
1409  *
1410  * 1. extent size hint is only valid for directories and regular files
1411  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1412  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1413  * 4. can only be changed on regular files if no extents are allocated
1414  * 5. can be changed on directories at any time
1415  * 6. extsize hint of 0 turns off hints, clears inode flags.
1416  * 7. Extent size must be a multiple of the appropriate block size.
1417  * 8. for non-realtime files, the extent size hint must be limited
1418  *    to half the AG size to avoid alignment extending the extent beyond the
1419  *    limits of the AG.
1420  *
1421  * Please keep this function in sync with xfs_scrub_inode_extsize.
1422  */
1423 static int
1424 xfs_ioctl_setattr_check_extsize(
1425 	struct xfs_inode	*ip,
1426 	struct fsxattr		*fa)
1427 {
1428 	struct xfs_mount	*mp = ip->i_mount;
1429 	xfs_extlen_t		size;
1430 	xfs_fsblock_t		extsize_fsb;
1431 
1432 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_d.di_nextents &&
1433 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1434 		return -EINVAL;
1435 
1436 	if (fa->fsx_extsize == 0)
1437 		return 0;
1438 
1439 	extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1440 	if (extsize_fsb > MAXEXTLEN)
1441 		return -EINVAL;
1442 
1443 	if (XFS_IS_REALTIME_INODE(ip) ||
1444 	    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1445 		size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1446 	} else {
1447 		size = mp->m_sb.sb_blocksize;
1448 		if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1449 			return -EINVAL;
1450 	}
1451 
1452 	if (fa->fsx_extsize % size)
1453 		return -EINVAL;
1454 
1455 	return 0;
1456 }
1457 
1458 /*
1459  * CoW extent size hint validation rules are:
1460  *
1461  * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
1462  *    The inode does not have to have any shared blocks, but it must be a v3.
1463  * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
1464  *    for a directory, the hint is propagated to new files.
1465  * 3. Can be changed on files & directories at any time.
1466  * 4. CoW extsize hint of 0 turns off hints, clears inode flags.
1467  * 5. Extent size must be a multiple of the appropriate block size.
1468  * 6. The extent size hint must be limited to half the AG size to avoid
1469  *    alignment extending the extent beyond the limits of the AG.
1470  *
1471  * Please keep this function in sync with xfs_scrub_inode_cowextsize.
1472  */
1473 static int
1474 xfs_ioctl_setattr_check_cowextsize(
1475 	struct xfs_inode	*ip,
1476 	struct fsxattr		*fa)
1477 {
1478 	struct xfs_mount	*mp = ip->i_mount;
1479 	xfs_extlen_t		size;
1480 	xfs_fsblock_t		cowextsize_fsb;
1481 
1482 	if (!(fa->fsx_xflags & FS_XFLAG_COWEXTSIZE))
1483 		return 0;
1484 
1485 	if (!xfs_sb_version_hasreflink(&ip->i_mount->m_sb) ||
1486 	    ip->i_d.di_version != 3)
1487 		return -EINVAL;
1488 
1489 	if (fa->fsx_cowextsize == 0)
1490 		return 0;
1491 
1492 	cowextsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_cowextsize);
1493 	if (cowextsize_fsb > MAXEXTLEN)
1494 		return -EINVAL;
1495 
1496 	size = mp->m_sb.sb_blocksize;
1497 	if (cowextsize_fsb > mp->m_sb.sb_agblocks / 2)
1498 		return -EINVAL;
1499 
1500 	if (fa->fsx_cowextsize % size)
1501 		return -EINVAL;
1502 
1503 	return 0;
1504 }
1505 
1506 static int
1507 xfs_ioctl_setattr_check_projid(
1508 	struct xfs_inode	*ip,
1509 	struct fsxattr		*fa)
1510 {
1511 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1512 	if (fa->fsx_projid > (uint16_t)-1 &&
1513 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1514 		return -EINVAL;
1515 	return 0;
1516 }
1517 
1518 STATIC int
1519 xfs_ioctl_setattr(
1520 	xfs_inode_t		*ip,
1521 	struct fsxattr		*fa)
1522 {
1523 	struct fsxattr		old_fa;
1524 	struct xfs_mount	*mp = ip->i_mount;
1525 	struct xfs_trans	*tp;
1526 	struct xfs_dquot	*udqp = NULL;
1527 	struct xfs_dquot	*pdqp = NULL;
1528 	struct xfs_dquot	*olddquot = NULL;
1529 	int			code;
1530 	int			join_flags = 0;
1531 
1532 	trace_xfs_ioctl_setattr(ip);
1533 
1534 	code = xfs_ioctl_setattr_check_projid(ip, fa);
1535 	if (code)
1536 		return code;
1537 
1538 	/*
1539 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1540 	 * before we start any other transactions. Trying to do this later
1541 	 * is messy. We don't care to take a readlock to look at the ids
1542 	 * in inode here, because we can't hold it across the trans_reserve.
1543 	 * If the IDs do change before we take the ilock, we're covered
1544 	 * because the i_*dquot fields will get updated anyway.
1545 	 */
1546 	if (XFS_IS_QUOTA_ON(mp)) {
1547 		code = xfs_qm_vop_dqalloc(ip, ip->i_d.di_uid,
1548 					 ip->i_d.di_gid, fa->fsx_projid,
1549 					 XFS_QMOPT_PQUOTA, &udqp, NULL, &pdqp);
1550 		if (code)
1551 			return code;
1552 	}
1553 
1554 	/*
1555 	 * Changing DAX config may require inode locking for mapping
1556 	 * invalidation. These need to be held all the way to transaction commit
1557 	 * or cancel time, so need to be passed through to
1558 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1559 	 * appropriately.
1560 	 */
1561 	code = xfs_ioctl_setattr_dax_invalidate(ip, fa, &join_flags);
1562 	if (code)
1563 		goto error_free_dquots;
1564 
1565 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1566 	if (IS_ERR(tp)) {
1567 		code = PTR_ERR(tp);
1568 		goto error_free_dquots;
1569 	}
1570 
1571 	if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp) &&
1572 	    xfs_get_projid(ip) != fa->fsx_projid) {
1573 		code = xfs_qm_vop_chown_reserve(tp, ip, udqp, NULL, pdqp,
1574 				capable(CAP_FOWNER) ?  XFS_QMOPT_FORCE_RES : 0);
1575 		if (code)	/* out of quota */
1576 			goto error_trans_cancel;
1577 	}
1578 
1579 	xfs_fill_fsxattr(ip, false, &old_fa);
1580 	code = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, fa);
1581 	if (code)
1582 		goto error_trans_cancel;
1583 
1584 	code = xfs_ioctl_setattr_check_extsize(ip, fa);
1585 	if (code)
1586 		goto error_trans_cancel;
1587 
1588 	code = xfs_ioctl_setattr_check_cowextsize(ip, fa);
1589 	if (code)
1590 		goto error_trans_cancel;
1591 
1592 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
1593 	if (code)
1594 		goto error_trans_cancel;
1595 
1596 	/*
1597 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1598 	 * overrides the following restrictions:
1599 	 *
1600 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1601 	 * successful return from chown()
1602 	 */
1603 
1604 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1605 	    !capable_wrt_inode_uidgid(VFS_I(ip), CAP_FSETID))
1606 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1607 
1608 	/* Change the ownerships and register project quota modifications */
1609 	if (xfs_get_projid(ip) != fa->fsx_projid) {
1610 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1611 			olddquot = xfs_qm_vop_chown(tp, ip,
1612 						&ip->i_pdquot, pdqp);
1613 		}
1614 		ASSERT(ip->i_d.di_version > 1);
1615 		xfs_set_projid(ip, fa->fsx_projid);
1616 	}
1617 
1618 	/*
1619 	 * Only set the extent size hint if we've already determined that the
1620 	 * extent size hint should be set on the inode. If no extent size flags
1621 	 * are set on the inode then unconditionally clear the extent size hint.
1622 	 */
1623 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1624 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1625 	else
1626 		ip->i_d.di_extsize = 0;
1627 	if (ip->i_d.di_version == 3 &&
1628 	    (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1629 		ip->i_d.di_cowextsize = fa->fsx_cowextsize >>
1630 				mp->m_sb.sb_blocklog;
1631 	else
1632 		ip->i_d.di_cowextsize = 0;
1633 
1634 	code = xfs_trans_commit(tp);
1635 
1636 	/*
1637 	 * Release any dquot(s) the inode had kept before chown.
1638 	 */
1639 	xfs_qm_dqrele(olddquot);
1640 	xfs_qm_dqrele(udqp);
1641 	xfs_qm_dqrele(pdqp);
1642 
1643 	return code;
1644 
1645 error_trans_cancel:
1646 	xfs_trans_cancel(tp);
1647 error_free_dquots:
1648 	xfs_qm_dqrele(udqp);
1649 	xfs_qm_dqrele(pdqp);
1650 	return code;
1651 }
1652 
1653 STATIC int
1654 xfs_ioc_fssetxattr(
1655 	xfs_inode_t		*ip,
1656 	struct file		*filp,
1657 	void			__user *arg)
1658 {
1659 	struct fsxattr		fa;
1660 	int error;
1661 
1662 	if (copy_from_user(&fa, arg, sizeof(fa)))
1663 		return -EFAULT;
1664 
1665 	error = mnt_want_write_file(filp);
1666 	if (error)
1667 		return error;
1668 	error = xfs_ioctl_setattr(ip, &fa);
1669 	mnt_drop_write_file(filp);
1670 	return error;
1671 }
1672 
1673 STATIC int
1674 xfs_ioc_getxflags(
1675 	xfs_inode_t		*ip,
1676 	void			__user *arg)
1677 {
1678 	unsigned int		flags;
1679 
1680 	flags = xfs_di2lxflags(ip->i_d.di_flags);
1681 	if (copy_to_user(arg, &flags, sizeof(flags)))
1682 		return -EFAULT;
1683 	return 0;
1684 }
1685 
1686 STATIC int
1687 xfs_ioc_setxflags(
1688 	struct xfs_inode	*ip,
1689 	struct file		*filp,
1690 	void			__user *arg)
1691 {
1692 	struct xfs_trans	*tp;
1693 	struct fsxattr		fa;
1694 	struct fsxattr		old_fa;
1695 	unsigned int		flags;
1696 	int			join_flags = 0;
1697 	int			error;
1698 
1699 	if (copy_from_user(&flags, arg, sizeof(flags)))
1700 		return -EFAULT;
1701 
1702 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1703 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1704 		      FS_SYNC_FL))
1705 		return -EOPNOTSUPP;
1706 
1707 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1708 
1709 	error = mnt_want_write_file(filp);
1710 	if (error)
1711 		return error;
1712 
1713 	/*
1714 	 * Changing DAX config may require inode locking for mapping
1715 	 * invalidation. These need to be held all the way to transaction commit
1716 	 * or cancel time, so need to be passed through to
1717 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1718 	 * appropriately.
1719 	 */
1720 	error = xfs_ioctl_setattr_dax_invalidate(ip, &fa, &join_flags);
1721 	if (error)
1722 		goto out_drop_write;
1723 
1724 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1725 	if (IS_ERR(tp)) {
1726 		error = PTR_ERR(tp);
1727 		goto out_drop_write;
1728 	}
1729 
1730 	xfs_fill_fsxattr(ip, false, &old_fa);
1731 	error = vfs_ioc_fssetxattr_check(VFS_I(ip), &old_fa, &fa);
1732 	if (error) {
1733 		xfs_trans_cancel(tp);
1734 		goto out_drop_write;
1735 	}
1736 
1737 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1738 	if (error) {
1739 		xfs_trans_cancel(tp);
1740 		goto out_drop_write;
1741 	}
1742 
1743 	error = xfs_trans_commit(tp);
1744 out_drop_write:
1745 	mnt_drop_write_file(filp);
1746 	return error;
1747 }
1748 
1749 static bool
1750 xfs_getbmap_format(
1751 	struct kgetbmap		*p,
1752 	struct getbmapx __user	*u,
1753 	size_t			recsize)
1754 {
1755 	if (put_user(p->bmv_offset, &u->bmv_offset) ||
1756 	    put_user(p->bmv_block, &u->bmv_block) ||
1757 	    put_user(p->bmv_length, &u->bmv_length) ||
1758 	    put_user(0, &u->bmv_count) ||
1759 	    put_user(0, &u->bmv_entries))
1760 		return false;
1761 	if (recsize < sizeof(struct getbmapx))
1762 		return true;
1763 	if (put_user(0, &u->bmv_iflags) ||
1764 	    put_user(p->bmv_oflags, &u->bmv_oflags) ||
1765 	    put_user(0, &u->bmv_unused1) ||
1766 	    put_user(0, &u->bmv_unused2))
1767 		return false;
1768 	return true;
1769 }
1770 
1771 STATIC int
1772 xfs_ioc_getbmap(
1773 	struct file		*file,
1774 	unsigned int		cmd,
1775 	void			__user *arg)
1776 {
1777 	struct getbmapx		bmx = { 0 };
1778 	struct kgetbmap		*buf;
1779 	size_t			recsize;
1780 	int			error, i;
1781 
1782 	switch (cmd) {
1783 	case XFS_IOC_GETBMAPA:
1784 		bmx.bmv_iflags = BMV_IF_ATTRFORK;
1785 		/*FALLTHRU*/
1786 	case XFS_IOC_GETBMAP:
1787 		if (file->f_mode & FMODE_NOCMTIME)
1788 			bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1789 		/* struct getbmap is a strict subset of struct getbmapx. */
1790 		recsize = sizeof(struct getbmap);
1791 		break;
1792 	case XFS_IOC_GETBMAPX:
1793 		recsize = sizeof(struct getbmapx);
1794 		break;
1795 	default:
1796 		return -EINVAL;
1797 	}
1798 
1799 	if (copy_from_user(&bmx, arg, recsize))
1800 		return -EFAULT;
1801 
1802 	if (bmx.bmv_count < 2)
1803 		return -EINVAL;
1804 	if (bmx.bmv_count > ULONG_MAX / recsize)
1805 		return -ENOMEM;
1806 
1807 	buf = kmem_zalloc_large(bmx.bmv_count * sizeof(*buf), 0);
1808 	if (!buf)
1809 		return -ENOMEM;
1810 
1811 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, buf);
1812 	if (error)
1813 		goto out_free_buf;
1814 
1815 	error = -EFAULT;
1816 	if (copy_to_user(arg, &bmx, recsize))
1817 		goto out_free_buf;
1818 	arg += recsize;
1819 
1820 	for (i = 0; i < bmx.bmv_entries; i++) {
1821 		if (!xfs_getbmap_format(buf + i, arg, recsize))
1822 			goto out_free_buf;
1823 		arg += recsize;
1824 	}
1825 
1826 	error = 0;
1827 out_free_buf:
1828 	kmem_free(buf);
1829 	return error;
1830 }
1831 
1832 struct getfsmap_info {
1833 	struct xfs_mount	*mp;
1834 	struct fsmap_head __user *data;
1835 	unsigned int		idx;
1836 	__u32			last_flags;
1837 };
1838 
1839 STATIC int
1840 xfs_getfsmap_format(struct xfs_fsmap *xfm, void *priv)
1841 {
1842 	struct getfsmap_info	*info = priv;
1843 	struct fsmap		fm;
1844 
1845 	trace_xfs_getfsmap_mapping(info->mp, xfm);
1846 
1847 	info->last_flags = xfm->fmr_flags;
1848 	xfs_fsmap_from_internal(&fm, xfm);
1849 	if (copy_to_user(&info->data->fmh_recs[info->idx++], &fm,
1850 			sizeof(struct fsmap)))
1851 		return -EFAULT;
1852 
1853 	return 0;
1854 }
1855 
1856 STATIC int
1857 xfs_ioc_getfsmap(
1858 	struct xfs_inode	*ip,
1859 	struct fsmap_head	__user *arg)
1860 {
1861 	struct getfsmap_info	info = { NULL };
1862 	struct xfs_fsmap_head	xhead = {0};
1863 	struct fsmap_head	head;
1864 	bool			aborted = false;
1865 	int			error;
1866 
1867 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1868 		return -EFAULT;
1869 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1870 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1871 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1872 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1873 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1874 		return -EINVAL;
1875 
1876 	xhead.fmh_iflags = head.fmh_iflags;
1877 	xhead.fmh_count = head.fmh_count;
1878 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1879 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1880 
1881 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1882 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1883 
1884 	info.mp = ip->i_mount;
1885 	info.data = arg;
1886 	error = xfs_getfsmap(ip->i_mount, &xhead, xfs_getfsmap_format, &info);
1887 	if (error == -ECANCELED) {
1888 		error = 0;
1889 		aborted = true;
1890 	} else if (error)
1891 		return error;
1892 
1893 	/* If we didn't abort, set the "last" flag in the last fmx */
1894 	if (!aborted && info.idx) {
1895 		info.last_flags |= FMR_OF_LAST;
1896 		if (copy_to_user(&info.data->fmh_recs[info.idx - 1].fmr_flags,
1897 				&info.last_flags, sizeof(info.last_flags)))
1898 			return -EFAULT;
1899 	}
1900 
1901 	/* copy back header */
1902 	head.fmh_entries = xhead.fmh_entries;
1903 	head.fmh_oflags = xhead.fmh_oflags;
1904 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head)))
1905 		return -EFAULT;
1906 
1907 	return 0;
1908 }
1909 
1910 STATIC int
1911 xfs_ioc_scrub_metadata(
1912 	struct xfs_inode		*ip,
1913 	void				__user *arg)
1914 {
1915 	struct xfs_scrub_metadata	scrub;
1916 	int				error;
1917 
1918 	if (!capable(CAP_SYS_ADMIN))
1919 		return -EPERM;
1920 
1921 	if (copy_from_user(&scrub, arg, sizeof(scrub)))
1922 		return -EFAULT;
1923 
1924 	error = xfs_scrub_metadata(ip, &scrub);
1925 	if (error)
1926 		return error;
1927 
1928 	if (copy_to_user(arg, &scrub, sizeof(scrub)))
1929 		return -EFAULT;
1930 
1931 	return 0;
1932 }
1933 
1934 int
1935 xfs_ioc_swapext(
1936 	xfs_swapext_t	*sxp)
1937 {
1938 	xfs_inode_t     *ip, *tip;
1939 	struct fd	f, tmp;
1940 	int		error = 0;
1941 
1942 	/* Pull information for the target fd */
1943 	f = fdget((int)sxp->sx_fdtarget);
1944 	if (!f.file) {
1945 		error = -EINVAL;
1946 		goto out;
1947 	}
1948 
1949 	if (!(f.file->f_mode & FMODE_WRITE) ||
1950 	    !(f.file->f_mode & FMODE_READ) ||
1951 	    (f.file->f_flags & O_APPEND)) {
1952 		error = -EBADF;
1953 		goto out_put_file;
1954 	}
1955 
1956 	tmp = fdget((int)sxp->sx_fdtmp);
1957 	if (!tmp.file) {
1958 		error = -EINVAL;
1959 		goto out_put_file;
1960 	}
1961 
1962 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1963 	    !(tmp.file->f_mode & FMODE_READ) ||
1964 	    (tmp.file->f_flags & O_APPEND)) {
1965 		error = -EBADF;
1966 		goto out_put_tmp_file;
1967 	}
1968 
1969 	if (IS_SWAPFILE(file_inode(f.file)) ||
1970 	    IS_SWAPFILE(file_inode(tmp.file))) {
1971 		error = -EINVAL;
1972 		goto out_put_tmp_file;
1973 	}
1974 
1975 	/*
1976 	 * We need to ensure that the fds passed in point to XFS inodes
1977 	 * before we cast and access them as XFS structures as we have no
1978 	 * control over what the user passes us here.
1979 	 */
1980 	if (f.file->f_op != &xfs_file_operations ||
1981 	    tmp.file->f_op != &xfs_file_operations) {
1982 		error = -EINVAL;
1983 		goto out_put_tmp_file;
1984 	}
1985 
1986 	ip = XFS_I(file_inode(f.file));
1987 	tip = XFS_I(file_inode(tmp.file));
1988 
1989 	if (ip->i_mount != tip->i_mount) {
1990 		error = -EINVAL;
1991 		goto out_put_tmp_file;
1992 	}
1993 
1994 	if (ip->i_ino == tip->i_ino) {
1995 		error = -EINVAL;
1996 		goto out_put_tmp_file;
1997 	}
1998 
1999 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
2000 		error = -EIO;
2001 		goto out_put_tmp_file;
2002 	}
2003 
2004 	error = xfs_swap_extents(ip, tip, sxp);
2005 
2006  out_put_tmp_file:
2007 	fdput(tmp);
2008  out_put_file:
2009 	fdput(f);
2010  out:
2011 	return error;
2012 }
2013 
2014 static int
2015 xfs_ioc_getlabel(
2016 	struct xfs_mount	*mp,
2017 	char			__user *user_label)
2018 {
2019 	struct xfs_sb		*sbp = &mp->m_sb;
2020 	char			label[XFSLABEL_MAX + 1];
2021 
2022 	/* Paranoia */
2023 	BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
2024 
2025 	/* 1 larger than sb_fname, so this ensures a trailing NUL char */
2026 	memset(label, 0, sizeof(label));
2027 	spin_lock(&mp->m_sb_lock);
2028 	strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
2029 	spin_unlock(&mp->m_sb_lock);
2030 
2031 	if (copy_to_user(user_label, label, sizeof(label)))
2032 		return -EFAULT;
2033 	return 0;
2034 }
2035 
2036 static int
2037 xfs_ioc_setlabel(
2038 	struct file		*filp,
2039 	struct xfs_mount	*mp,
2040 	char			__user *newlabel)
2041 {
2042 	struct xfs_sb		*sbp = &mp->m_sb;
2043 	char			label[XFSLABEL_MAX + 1];
2044 	size_t			len;
2045 	int			error;
2046 
2047 	if (!capable(CAP_SYS_ADMIN))
2048 		return -EPERM;
2049 	/*
2050 	 * The generic ioctl allows up to FSLABEL_MAX chars, but XFS is much
2051 	 * smaller, at 12 bytes.  We copy one more to be sure we find the
2052 	 * (required) NULL character to test the incoming label length.
2053 	 * NB: The on disk label doesn't need to be null terminated.
2054 	 */
2055 	if (copy_from_user(label, newlabel, XFSLABEL_MAX + 1))
2056 		return -EFAULT;
2057 	len = strnlen(label, XFSLABEL_MAX + 1);
2058 	if (len > sizeof(sbp->sb_fname))
2059 		return -EINVAL;
2060 
2061 	error = mnt_want_write_file(filp);
2062 	if (error)
2063 		return error;
2064 
2065 	spin_lock(&mp->m_sb_lock);
2066 	memset(sbp->sb_fname, 0, sizeof(sbp->sb_fname));
2067 	memcpy(sbp->sb_fname, label, len);
2068 	spin_unlock(&mp->m_sb_lock);
2069 
2070 	/*
2071 	 * Now we do several things to satisfy userspace.
2072 	 * In addition to normal logging of the primary superblock, we also
2073 	 * immediately write these changes to sector zero for the primary, then
2074 	 * update all backup supers (as xfs_db does for a label change), then
2075 	 * invalidate the block device page cache.  This is so that any prior
2076 	 * buffered reads from userspace (i.e. from blkid) are invalidated,
2077 	 * and userspace will see the newly-written label.
2078 	 */
2079 	error = xfs_sync_sb_buf(mp);
2080 	if (error)
2081 		goto out;
2082 	/*
2083 	 * growfs also updates backup supers so lock against that.
2084 	 */
2085 	mutex_lock(&mp->m_growlock);
2086 	error = xfs_update_secondary_sbs(mp);
2087 	mutex_unlock(&mp->m_growlock);
2088 
2089 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
2090 
2091 out:
2092 	mnt_drop_write_file(filp);
2093 	return error;
2094 }
2095 
2096 /*
2097  * Note: some of the ioctl's return positive numbers as a
2098  * byte count indicating success, such as readlink_by_handle.
2099  * So we don't "sign flip" like most other routines.  This means
2100  * true errors need to be returned as a negative value.
2101  */
2102 long
2103 xfs_file_ioctl(
2104 	struct file		*filp,
2105 	unsigned int		cmd,
2106 	unsigned long		p)
2107 {
2108 	struct inode		*inode = file_inode(filp);
2109 	struct xfs_inode	*ip = XFS_I(inode);
2110 	struct xfs_mount	*mp = ip->i_mount;
2111 	void			__user *arg = (void __user *)p;
2112 	int			error;
2113 
2114 	trace_xfs_file_ioctl(ip);
2115 
2116 	switch (cmd) {
2117 	case FITRIM:
2118 		return xfs_ioc_trim(mp, arg);
2119 	case FS_IOC_GETFSLABEL:
2120 		return xfs_ioc_getlabel(mp, arg);
2121 	case FS_IOC_SETFSLABEL:
2122 		return xfs_ioc_setlabel(filp, mp, arg);
2123 	case XFS_IOC_ALLOCSP:
2124 	case XFS_IOC_FREESP:
2125 	case XFS_IOC_RESVSP:
2126 	case XFS_IOC_UNRESVSP:
2127 	case XFS_IOC_ALLOCSP64:
2128 	case XFS_IOC_FREESP64:
2129 	case XFS_IOC_RESVSP64:
2130 	case XFS_IOC_UNRESVSP64:
2131 	case XFS_IOC_ZERO_RANGE: {
2132 		xfs_flock64_t		bf;
2133 
2134 		if (copy_from_user(&bf, arg, sizeof(bf)))
2135 			return -EFAULT;
2136 		return xfs_ioc_space(filp, cmd, &bf);
2137 	}
2138 	case XFS_IOC_DIOINFO: {
2139 		struct dioattr	da;
2140 		xfs_buftarg_t	*target =
2141 			XFS_IS_REALTIME_INODE(ip) ?
2142 			mp->m_rtdev_targp : mp->m_ddev_targp;
2143 
2144 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
2145 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
2146 
2147 		if (copy_to_user(arg, &da, sizeof(da)))
2148 			return -EFAULT;
2149 		return 0;
2150 	}
2151 
2152 	case XFS_IOC_FSBULKSTAT_SINGLE:
2153 	case XFS_IOC_FSBULKSTAT:
2154 	case XFS_IOC_FSINUMBERS:
2155 		return xfs_ioc_fsbulkstat(mp, cmd, arg);
2156 
2157 	case XFS_IOC_BULKSTAT:
2158 		return xfs_ioc_bulkstat(mp, cmd, arg);
2159 	case XFS_IOC_INUMBERS:
2160 		return xfs_ioc_inumbers(mp, cmd, arg);
2161 
2162 	case XFS_IOC_FSGEOMETRY_V1:
2163 		return xfs_ioc_fsgeometry(mp, arg, 3);
2164 	case XFS_IOC_FSGEOMETRY_V4:
2165 		return xfs_ioc_fsgeometry(mp, arg, 4);
2166 	case XFS_IOC_FSGEOMETRY:
2167 		return xfs_ioc_fsgeometry(mp, arg, 5);
2168 
2169 	case XFS_IOC_AG_GEOMETRY:
2170 		return xfs_ioc_ag_geometry(mp, arg);
2171 
2172 	case XFS_IOC_GETVERSION:
2173 		return put_user(inode->i_generation, (int __user *)arg);
2174 
2175 	case XFS_IOC_FSGETXATTR:
2176 		return xfs_ioc_fsgetxattr(ip, 0, arg);
2177 	case XFS_IOC_FSGETXATTRA:
2178 		return xfs_ioc_fsgetxattr(ip, 1, arg);
2179 	case XFS_IOC_FSSETXATTR:
2180 		return xfs_ioc_fssetxattr(ip, filp, arg);
2181 	case XFS_IOC_GETXFLAGS:
2182 		return xfs_ioc_getxflags(ip, arg);
2183 	case XFS_IOC_SETXFLAGS:
2184 		return xfs_ioc_setxflags(ip, filp, arg);
2185 
2186 	case XFS_IOC_FSSETDM: {
2187 		struct fsdmidata	dmi;
2188 
2189 		if (copy_from_user(&dmi, arg, sizeof(dmi)))
2190 			return -EFAULT;
2191 
2192 		error = mnt_want_write_file(filp);
2193 		if (error)
2194 			return error;
2195 
2196 		error = xfs_set_dmattrs(ip, dmi.fsd_dmevmask,
2197 				dmi.fsd_dmstate);
2198 		mnt_drop_write_file(filp);
2199 		return error;
2200 	}
2201 
2202 	case XFS_IOC_GETBMAP:
2203 	case XFS_IOC_GETBMAPA:
2204 	case XFS_IOC_GETBMAPX:
2205 		return xfs_ioc_getbmap(filp, cmd, arg);
2206 
2207 	case FS_IOC_GETFSMAP:
2208 		return xfs_ioc_getfsmap(ip, arg);
2209 
2210 	case XFS_IOC_SCRUB_METADATA:
2211 		return xfs_ioc_scrub_metadata(ip, arg);
2212 
2213 	case XFS_IOC_FD_TO_HANDLE:
2214 	case XFS_IOC_PATH_TO_HANDLE:
2215 	case XFS_IOC_PATH_TO_FSHANDLE: {
2216 		xfs_fsop_handlereq_t	hreq;
2217 
2218 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
2219 			return -EFAULT;
2220 		return xfs_find_handle(cmd, &hreq);
2221 	}
2222 	case XFS_IOC_OPEN_BY_HANDLE: {
2223 		xfs_fsop_handlereq_t	hreq;
2224 
2225 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2226 			return -EFAULT;
2227 		return xfs_open_by_handle(filp, &hreq);
2228 	}
2229 	case XFS_IOC_FSSETDM_BY_HANDLE:
2230 		return xfs_fssetdm_by_handle(filp, arg);
2231 
2232 	case XFS_IOC_READLINK_BY_HANDLE: {
2233 		xfs_fsop_handlereq_t	hreq;
2234 
2235 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
2236 			return -EFAULT;
2237 		return xfs_readlink_by_handle(filp, &hreq);
2238 	}
2239 	case XFS_IOC_ATTRLIST_BY_HANDLE:
2240 		return xfs_attrlist_by_handle(filp, arg);
2241 
2242 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
2243 		return xfs_attrmulti_by_handle(filp, arg);
2244 
2245 	case XFS_IOC_SWAPEXT: {
2246 		struct xfs_swapext	sxp;
2247 
2248 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
2249 			return -EFAULT;
2250 		error = mnt_want_write_file(filp);
2251 		if (error)
2252 			return error;
2253 		error = xfs_ioc_swapext(&sxp);
2254 		mnt_drop_write_file(filp);
2255 		return error;
2256 	}
2257 
2258 	case XFS_IOC_FSCOUNTS: {
2259 		xfs_fsop_counts_t out;
2260 
2261 		xfs_fs_counts(mp, &out);
2262 
2263 		if (copy_to_user(arg, &out, sizeof(out)))
2264 			return -EFAULT;
2265 		return 0;
2266 	}
2267 
2268 	case XFS_IOC_SET_RESBLKS: {
2269 		xfs_fsop_resblks_t inout;
2270 		uint64_t	   in;
2271 
2272 		if (!capable(CAP_SYS_ADMIN))
2273 			return -EPERM;
2274 
2275 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2276 			return -EROFS;
2277 
2278 		if (copy_from_user(&inout, arg, sizeof(inout)))
2279 			return -EFAULT;
2280 
2281 		error = mnt_want_write_file(filp);
2282 		if (error)
2283 			return error;
2284 
2285 		/* input parameter is passed in resblks field of structure */
2286 		in = inout.resblks;
2287 		error = xfs_reserve_blocks(mp, &in, &inout);
2288 		mnt_drop_write_file(filp);
2289 		if (error)
2290 			return error;
2291 
2292 		if (copy_to_user(arg, &inout, sizeof(inout)))
2293 			return -EFAULT;
2294 		return 0;
2295 	}
2296 
2297 	case XFS_IOC_GET_RESBLKS: {
2298 		xfs_fsop_resblks_t out;
2299 
2300 		if (!capable(CAP_SYS_ADMIN))
2301 			return -EPERM;
2302 
2303 		error = xfs_reserve_blocks(mp, NULL, &out);
2304 		if (error)
2305 			return error;
2306 
2307 		if (copy_to_user(arg, &out, sizeof(out)))
2308 			return -EFAULT;
2309 
2310 		return 0;
2311 	}
2312 
2313 	case XFS_IOC_FSGROWFSDATA: {
2314 		xfs_growfs_data_t in;
2315 
2316 		if (copy_from_user(&in, arg, sizeof(in)))
2317 			return -EFAULT;
2318 
2319 		error = mnt_want_write_file(filp);
2320 		if (error)
2321 			return error;
2322 		error = xfs_growfs_data(mp, &in);
2323 		mnt_drop_write_file(filp);
2324 		return error;
2325 	}
2326 
2327 	case XFS_IOC_FSGROWFSLOG: {
2328 		xfs_growfs_log_t in;
2329 
2330 		if (copy_from_user(&in, arg, sizeof(in)))
2331 			return -EFAULT;
2332 
2333 		error = mnt_want_write_file(filp);
2334 		if (error)
2335 			return error;
2336 		error = xfs_growfs_log(mp, &in);
2337 		mnt_drop_write_file(filp);
2338 		return error;
2339 	}
2340 
2341 	case XFS_IOC_FSGROWFSRT: {
2342 		xfs_growfs_rt_t in;
2343 
2344 		if (copy_from_user(&in, arg, sizeof(in)))
2345 			return -EFAULT;
2346 
2347 		error = mnt_want_write_file(filp);
2348 		if (error)
2349 			return error;
2350 		error = xfs_growfs_rt(mp, &in);
2351 		mnt_drop_write_file(filp);
2352 		return error;
2353 	}
2354 
2355 	case XFS_IOC_GOINGDOWN: {
2356 		uint32_t in;
2357 
2358 		if (!capable(CAP_SYS_ADMIN))
2359 			return -EPERM;
2360 
2361 		if (get_user(in, (uint32_t __user *)arg))
2362 			return -EFAULT;
2363 
2364 		return xfs_fs_goingdown(mp, in);
2365 	}
2366 
2367 	case XFS_IOC_ERROR_INJECTION: {
2368 		xfs_error_injection_t in;
2369 
2370 		if (!capable(CAP_SYS_ADMIN))
2371 			return -EPERM;
2372 
2373 		if (copy_from_user(&in, arg, sizeof(in)))
2374 			return -EFAULT;
2375 
2376 		return xfs_errortag_add(mp, in.errtag);
2377 	}
2378 
2379 	case XFS_IOC_ERROR_CLEARALL:
2380 		if (!capable(CAP_SYS_ADMIN))
2381 			return -EPERM;
2382 
2383 		return xfs_errortag_clearall(mp);
2384 
2385 	case XFS_IOC_FREE_EOFBLOCKS: {
2386 		struct xfs_fs_eofblocks eofb;
2387 		struct xfs_eofblocks keofb;
2388 
2389 		if (!capable(CAP_SYS_ADMIN))
2390 			return -EPERM;
2391 
2392 		if (mp->m_flags & XFS_MOUNT_RDONLY)
2393 			return -EROFS;
2394 
2395 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
2396 			return -EFAULT;
2397 
2398 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
2399 		if (error)
2400 			return error;
2401 
2402 		return xfs_icache_free_eofblocks(mp, &keofb);
2403 	}
2404 
2405 	default:
2406 		return -ENOTTY;
2407 	}
2408 }
2409