xref: /openbmc/linux/fs/xfs/xfs_ioctl.c (revision bc5aa3a0)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_inode.h"
26 #include "xfs_ioctl.h"
27 #include "xfs_alloc.h"
28 #include "xfs_rtalloc.h"
29 #include "xfs_itable.h"
30 #include "xfs_error.h"
31 #include "xfs_attr.h"
32 #include "xfs_bmap.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_fsops.h"
35 #include "xfs_discard.h"
36 #include "xfs_quota.h"
37 #include "xfs_export.h"
38 #include "xfs_trace.h"
39 #include "xfs_icache.h"
40 #include "xfs_symlink.h"
41 #include "xfs_trans.h"
42 #include "xfs_pnfs.h"
43 #include "xfs_acl.h"
44 
45 #include <linux/capability.h>
46 #include <linux/dcache.h>
47 #include <linux/mount.h>
48 #include <linux/namei.h>
49 #include <linux/pagemap.h>
50 #include <linux/slab.h>
51 #include <linux/exportfs.h>
52 
53 /*
54  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
55  * a file or fs handle.
56  *
57  * XFS_IOC_PATH_TO_FSHANDLE
58  *    returns fs handle for a mount point or path within that mount point
59  * XFS_IOC_FD_TO_HANDLE
60  *    returns full handle for a FD opened in user space
61  * XFS_IOC_PATH_TO_HANDLE
62  *    returns full handle for a path
63  */
64 int
65 xfs_find_handle(
66 	unsigned int		cmd,
67 	xfs_fsop_handlereq_t	*hreq)
68 {
69 	int			hsize;
70 	xfs_handle_t		handle;
71 	struct inode		*inode;
72 	struct fd		f = {NULL};
73 	struct path		path;
74 	int			error;
75 	struct xfs_inode	*ip;
76 
77 	if (cmd == XFS_IOC_FD_TO_HANDLE) {
78 		f = fdget(hreq->fd);
79 		if (!f.file)
80 			return -EBADF;
81 		inode = file_inode(f.file);
82 	} else {
83 		error = user_lpath((const char __user *)hreq->path, &path);
84 		if (error)
85 			return error;
86 		inode = d_inode(path.dentry);
87 	}
88 	ip = XFS_I(inode);
89 
90 	/*
91 	 * We can only generate handles for inodes residing on a XFS filesystem,
92 	 * and only for regular files, directories or symbolic links.
93 	 */
94 	error = -EINVAL;
95 	if (inode->i_sb->s_magic != XFS_SB_MAGIC)
96 		goto out_put;
97 
98 	error = -EBADF;
99 	if (!S_ISREG(inode->i_mode) &&
100 	    !S_ISDIR(inode->i_mode) &&
101 	    !S_ISLNK(inode->i_mode))
102 		goto out_put;
103 
104 
105 	memcpy(&handle.ha_fsid, ip->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
106 
107 	if (cmd == XFS_IOC_PATH_TO_FSHANDLE) {
108 		/*
109 		 * This handle only contains an fsid, zero the rest.
110 		 */
111 		memset(&handle.ha_fid, 0, sizeof(handle.ha_fid));
112 		hsize = sizeof(xfs_fsid_t);
113 	} else {
114 		handle.ha_fid.fid_len = sizeof(xfs_fid_t) -
115 					sizeof(handle.ha_fid.fid_len);
116 		handle.ha_fid.fid_pad = 0;
117 		handle.ha_fid.fid_gen = inode->i_generation;
118 		handle.ha_fid.fid_ino = ip->i_ino;
119 
120 		hsize = XFS_HSIZE(handle);
121 	}
122 
123 	error = -EFAULT;
124 	if (copy_to_user(hreq->ohandle, &handle, hsize) ||
125 	    copy_to_user(hreq->ohandlen, &hsize, sizeof(__s32)))
126 		goto out_put;
127 
128 	error = 0;
129 
130  out_put:
131 	if (cmd == XFS_IOC_FD_TO_HANDLE)
132 		fdput(f);
133 	else
134 		path_put(&path);
135 	return error;
136 }
137 
138 /*
139  * No need to do permission checks on the various pathname components
140  * as the handle operations are privileged.
141  */
142 STATIC int
143 xfs_handle_acceptable(
144 	void			*context,
145 	struct dentry		*dentry)
146 {
147 	return 1;
148 }
149 
150 /*
151  * Convert userspace handle data into a dentry.
152  */
153 struct dentry *
154 xfs_handle_to_dentry(
155 	struct file		*parfilp,
156 	void __user		*uhandle,
157 	u32			hlen)
158 {
159 	xfs_handle_t		handle;
160 	struct xfs_fid64	fid;
161 
162 	/*
163 	 * Only allow handle opens under a directory.
164 	 */
165 	if (!S_ISDIR(file_inode(parfilp)->i_mode))
166 		return ERR_PTR(-ENOTDIR);
167 
168 	if (hlen != sizeof(xfs_handle_t))
169 		return ERR_PTR(-EINVAL);
170 	if (copy_from_user(&handle, uhandle, hlen))
171 		return ERR_PTR(-EFAULT);
172 	if (handle.ha_fid.fid_len !=
173 	    sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
174 		return ERR_PTR(-EINVAL);
175 
176 	memset(&fid, 0, sizeof(struct fid));
177 	fid.ino = handle.ha_fid.fid_ino;
178 	fid.gen = handle.ha_fid.fid_gen;
179 
180 	return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
181 			FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
182 			xfs_handle_acceptable, NULL);
183 }
184 
185 STATIC struct dentry *
186 xfs_handlereq_to_dentry(
187 	struct file		*parfilp,
188 	xfs_fsop_handlereq_t	*hreq)
189 {
190 	return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
191 }
192 
193 int
194 xfs_open_by_handle(
195 	struct file		*parfilp,
196 	xfs_fsop_handlereq_t	*hreq)
197 {
198 	const struct cred	*cred = current_cred();
199 	int			error;
200 	int			fd;
201 	int			permflag;
202 	struct file		*filp;
203 	struct inode		*inode;
204 	struct dentry		*dentry;
205 	fmode_t			fmode;
206 	struct path		path;
207 
208 	if (!capable(CAP_SYS_ADMIN))
209 		return -EPERM;
210 
211 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
212 	if (IS_ERR(dentry))
213 		return PTR_ERR(dentry);
214 	inode = d_inode(dentry);
215 
216 	/* Restrict xfs_open_by_handle to directories & regular files. */
217 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
218 		error = -EPERM;
219 		goto out_dput;
220 	}
221 
222 #if BITS_PER_LONG != 32
223 	hreq->oflags |= O_LARGEFILE;
224 #endif
225 
226 	permflag = hreq->oflags;
227 	fmode = OPEN_FMODE(permflag);
228 	if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
229 	    (fmode & FMODE_WRITE) && IS_APPEND(inode)) {
230 		error = -EPERM;
231 		goto out_dput;
232 	}
233 
234 	if ((fmode & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
235 		error = -EPERM;
236 		goto out_dput;
237 	}
238 
239 	/* Can't write directories. */
240 	if (S_ISDIR(inode->i_mode) && (fmode & FMODE_WRITE)) {
241 		error = -EISDIR;
242 		goto out_dput;
243 	}
244 
245 	fd = get_unused_fd_flags(0);
246 	if (fd < 0) {
247 		error = fd;
248 		goto out_dput;
249 	}
250 
251 	path.mnt = parfilp->f_path.mnt;
252 	path.dentry = dentry;
253 	filp = dentry_open(&path, hreq->oflags, cred);
254 	dput(dentry);
255 	if (IS_ERR(filp)) {
256 		put_unused_fd(fd);
257 		return PTR_ERR(filp);
258 	}
259 
260 	if (S_ISREG(inode->i_mode)) {
261 		filp->f_flags |= O_NOATIME;
262 		filp->f_mode |= FMODE_NOCMTIME;
263 	}
264 
265 	fd_install(fd, filp);
266 	return fd;
267 
268  out_dput:
269 	dput(dentry);
270 	return error;
271 }
272 
273 int
274 xfs_readlink_by_handle(
275 	struct file		*parfilp,
276 	xfs_fsop_handlereq_t	*hreq)
277 {
278 	struct dentry		*dentry;
279 	__u32			olen;
280 	int			error;
281 
282 	if (!capable(CAP_SYS_ADMIN))
283 		return -EPERM;
284 
285 	dentry = xfs_handlereq_to_dentry(parfilp, hreq);
286 	if (IS_ERR(dentry))
287 		return PTR_ERR(dentry);
288 
289 	/* Restrict this handle operation to symlinks only. */
290 	if (!d_inode(dentry)->i_op->readlink) {
291 		error = -EINVAL;
292 		goto out_dput;
293 	}
294 
295 	if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
296 		error = -EFAULT;
297 		goto out_dput;
298 	}
299 
300 	error = d_inode(dentry)->i_op->readlink(dentry, hreq->ohandle, olen);
301 
302  out_dput:
303 	dput(dentry);
304 	return error;
305 }
306 
307 int
308 xfs_set_dmattrs(
309 	xfs_inode_t     *ip,
310 	u_int		evmask,
311 	u_int16_t	state)
312 {
313 	xfs_mount_t	*mp = ip->i_mount;
314 	xfs_trans_t	*tp;
315 	int		error;
316 
317 	if (!capable(CAP_SYS_ADMIN))
318 		return -EPERM;
319 
320 	if (XFS_FORCED_SHUTDOWN(mp))
321 		return -EIO;
322 
323 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
324 	if (error)
325 		return error;
326 
327 	xfs_ilock(ip, XFS_ILOCK_EXCL);
328 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
329 
330 	ip->i_d.di_dmevmask = evmask;
331 	ip->i_d.di_dmstate  = state;
332 
333 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
334 	error = xfs_trans_commit(tp);
335 
336 	return error;
337 }
338 
339 STATIC int
340 xfs_fssetdm_by_handle(
341 	struct file		*parfilp,
342 	void			__user *arg)
343 {
344 	int			error;
345 	struct fsdmidata	fsd;
346 	xfs_fsop_setdm_handlereq_t dmhreq;
347 	struct dentry		*dentry;
348 
349 	if (!capable(CAP_MKNOD))
350 		return -EPERM;
351 	if (copy_from_user(&dmhreq, arg, sizeof(xfs_fsop_setdm_handlereq_t)))
352 		return -EFAULT;
353 
354 	error = mnt_want_write_file(parfilp);
355 	if (error)
356 		return error;
357 
358 	dentry = xfs_handlereq_to_dentry(parfilp, &dmhreq.hreq);
359 	if (IS_ERR(dentry)) {
360 		mnt_drop_write_file(parfilp);
361 		return PTR_ERR(dentry);
362 	}
363 
364 	if (IS_IMMUTABLE(d_inode(dentry)) || IS_APPEND(d_inode(dentry))) {
365 		error = -EPERM;
366 		goto out;
367 	}
368 
369 	if (copy_from_user(&fsd, dmhreq.data, sizeof(fsd))) {
370 		error = -EFAULT;
371 		goto out;
372 	}
373 
374 	error = xfs_set_dmattrs(XFS_I(d_inode(dentry)), fsd.fsd_dmevmask,
375 				 fsd.fsd_dmstate);
376 
377  out:
378 	mnt_drop_write_file(parfilp);
379 	dput(dentry);
380 	return error;
381 }
382 
383 STATIC int
384 xfs_attrlist_by_handle(
385 	struct file		*parfilp,
386 	void			__user *arg)
387 {
388 	int			error = -ENOMEM;
389 	attrlist_cursor_kern_t	*cursor;
390 	struct xfs_fsop_attrlist_handlereq __user	*p = arg;
391 	xfs_fsop_attrlist_handlereq_t al_hreq;
392 	struct dentry		*dentry;
393 	char			*kbuf;
394 
395 	if (!capable(CAP_SYS_ADMIN))
396 		return -EPERM;
397 	if (copy_from_user(&al_hreq, arg, sizeof(xfs_fsop_attrlist_handlereq_t)))
398 		return -EFAULT;
399 	if (al_hreq.buflen < sizeof(struct attrlist) ||
400 	    al_hreq.buflen > XFS_XATTR_LIST_MAX)
401 		return -EINVAL;
402 
403 	/*
404 	 * Reject flags, only allow namespaces.
405 	 */
406 	if (al_hreq.flags & ~(ATTR_ROOT | ATTR_SECURE))
407 		return -EINVAL;
408 
409 	dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
410 	if (IS_ERR(dentry))
411 		return PTR_ERR(dentry);
412 
413 	kbuf = kmem_zalloc_large(al_hreq.buflen, KM_SLEEP);
414 	if (!kbuf)
415 		goto out_dput;
416 
417 	cursor = (attrlist_cursor_kern_t *)&al_hreq.pos;
418 	error = xfs_attr_list(XFS_I(d_inode(dentry)), kbuf, al_hreq.buflen,
419 					al_hreq.flags, cursor);
420 	if (error)
421 		goto out_kfree;
422 
423 	if (copy_to_user(&p->pos, cursor, sizeof(attrlist_cursor_kern_t))) {
424 		error = -EFAULT;
425 		goto out_kfree;
426 	}
427 
428 	if (copy_to_user(al_hreq.buffer, kbuf, al_hreq.buflen))
429 		error = -EFAULT;
430 
431 out_kfree:
432 	kmem_free(kbuf);
433 out_dput:
434 	dput(dentry);
435 	return error;
436 }
437 
438 int
439 xfs_attrmulti_attr_get(
440 	struct inode		*inode,
441 	unsigned char		*name,
442 	unsigned char		__user *ubuf,
443 	__uint32_t		*len,
444 	__uint32_t		flags)
445 {
446 	unsigned char		*kbuf;
447 	int			error = -EFAULT;
448 
449 	if (*len > XFS_XATTR_SIZE_MAX)
450 		return -EINVAL;
451 	kbuf = kmem_zalloc_large(*len, KM_SLEEP);
452 	if (!kbuf)
453 		return -ENOMEM;
454 
455 	error = xfs_attr_get(XFS_I(inode), name, kbuf, (int *)len, flags);
456 	if (error)
457 		goto out_kfree;
458 
459 	if (copy_to_user(ubuf, kbuf, *len))
460 		error = -EFAULT;
461 
462 out_kfree:
463 	kmem_free(kbuf);
464 	return error;
465 }
466 
467 int
468 xfs_attrmulti_attr_set(
469 	struct inode		*inode,
470 	unsigned char		*name,
471 	const unsigned char	__user *ubuf,
472 	__uint32_t		len,
473 	__uint32_t		flags)
474 {
475 	unsigned char		*kbuf;
476 	int			error;
477 
478 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
479 		return -EPERM;
480 	if (len > XFS_XATTR_SIZE_MAX)
481 		return -EINVAL;
482 
483 	kbuf = memdup_user(ubuf, len);
484 	if (IS_ERR(kbuf))
485 		return PTR_ERR(kbuf);
486 
487 	error = xfs_attr_set(XFS_I(inode), name, kbuf, len, flags);
488 	if (!error)
489 		xfs_forget_acl(inode, name, flags);
490 	kfree(kbuf);
491 	return error;
492 }
493 
494 int
495 xfs_attrmulti_attr_remove(
496 	struct inode		*inode,
497 	unsigned char		*name,
498 	__uint32_t		flags)
499 {
500 	int			error;
501 
502 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
503 		return -EPERM;
504 	error = xfs_attr_remove(XFS_I(inode), name, flags);
505 	if (!error)
506 		xfs_forget_acl(inode, name, flags);
507 	return error;
508 }
509 
510 STATIC int
511 xfs_attrmulti_by_handle(
512 	struct file		*parfilp,
513 	void			__user *arg)
514 {
515 	int			error;
516 	xfs_attr_multiop_t	*ops;
517 	xfs_fsop_attrmulti_handlereq_t am_hreq;
518 	struct dentry		*dentry;
519 	unsigned int		i, size;
520 	unsigned char		*attr_name;
521 
522 	if (!capable(CAP_SYS_ADMIN))
523 		return -EPERM;
524 	if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
525 		return -EFAULT;
526 
527 	/* overflow check */
528 	if (am_hreq.opcount >= INT_MAX / sizeof(xfs_attr_multiop_t))
529 		return -E2BIG;
530 
531 	dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
532 	if (IS_ERR(dentry))
533 		return PTR_ERR(dentry);
534 
535 	error = -E2BIG;
536 	size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
537 	if (!size || size > 16 * PAGE_SIZE)
538 		goto out_dput;
539 
540 	ops = memdup_user(am_hreq.ops, size);
541 	if (IS_ERR(ops)) {
542 		error = PTR_ERR(ops);
543 		goto out_dput;
544 	}
545 
546 	error = -ENOMEM;
547 	attr_name = kmalloc(MAXNAMELEN, GFP_KERNEL);
548 	if (!attr_name)
549 		goto out_kfree_ops;
550 
551 	error = 0;
552 	for (i = 0; i < am_hreq.opcount; i++) {
553 		ops[i].am_error = strncpy_from_user((char *)attr_name,
554 				ops[i].am_attrname, MAXNAMELEN);
555 		if (ops[i].am_error == 0 || ops[i].am_error == MAXNAMELEN)
556 			error = -ERANGE;
557 		if (ops[i].am_error < 0)
558 			break;
559 
560 		switch (ops[i].am_opcode) {
561 		case ATTR_OP_GET:
562 			ops[i].am_error = xfs_attrmulti_attr_get(
563 					d_inode(dentry), attr_name,
564 					ops[i].am_attrvalue, &ops[i].am_length,
565 					ops[i].am_flags);
566 			break;
567 		case ATTR_OP_SET:
568 			ops[i].am_error = mnt_want_write_file(parfilp);
569 			if (ops[i].am_error)
570 				break;
571 			ops[i].am_error = xfs_attrmulti_attr_set(
572 					d_inode(dentry), attr_name,
573 					ops[i].am_attrvalue, ops[i].am_length,
574 					ops[i].am_flags);
575 			mnt_drop_write_file(parfilp);
576 			break;
577 		case ATTR_OP_REMOVE:
578 			ops[i].am_error = mnt_want_write_file(parfilp);
579 			if (ops[i].am_error)
580 				break;
581 			ops[i].am_error = xfs_attrmulti_attr_remove(
582 					d_inode(dentry), attr_name,
583 					ops[i].am_flags);
584 			mnt_drop_write_file(parfilp);
585 			break;
586 		default:
587 			ops[i].am_error = -EINVAL;
588 		}
589 	}
590 
591 	if (copy_to_user(am_hreq.ops, ops, size))
592 		error = -EFAULT;
593 
594 	kfree(attr_name);
595  out_kfree_ops:
596 	kfree(ops);
597  out_dput:
598 	dput(dentry);
599 	return error;
600 }
601 
602 int
603 xfs_ioc_space(
604 	struct file		*filp,
605 	unsigned int		cmd,
606 	xfs_flock64_t		*bf)
607 {
608 	struct inode		*inode = file_inode(filp);
609 	struct xfs_inode	*ip = XFS_I(inode);
610 	struct iattr		iattr;
611 	enum xfs_prealloc_flags	flags = 0;
612 	uint			iolock = XFS_IOLOCK_EXCL;
613 	int			error;
614 
615 	/*
616 	 * Only allow the sys admin to reserve space unless
617 	 * unwritten extents are enabled.
618 	 */
619 	if (!xfs_sb_version_hasextflgbit(&ip->i_mount->m_sb) &&
620 	    !capable(CAP_SYS_ADMIN))
621 		return -EPERM;
622 
623 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND))
624 		return -EPERM;
625 
626 	if (!(filp->f_mode & FMODE_WRITE))
627 		return -EBADF;
628 
629 	if (!S_ISREG(inode->i_mode))
630 		return -EINVAL;
631 
632 	if (filp->f_flags & O_DSYNC)
633 		flags |= XFS_PREALLOC_SYNC;
634 	if (filp->f_mode & FMODE_NOCMTIME)
635 		flags |= XFS_PREALLOC_INVISIBLE;
636 
637 	error = mnt_want_write_file(filp);
638 	if (error)
639 		return error;
640 
641 	xfs_ilock(ip, iolock);
642 	error = xfs_break_layouts(inode, &iolock, false);
643 	if (error)
644 		goto out_unlock;
645 
646 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
647 	iolock |= XFS_MMAPLOCK_EXCL;
648 
649 	switch (bf->l_whence) {
650 	case 0: /*SEEK_SET*/
651 		break;
652 	case 1: /*SEEK_CUR*/
653 		bf->l_start += filp->f_pos;
654 		break;
655 	case 2: /*SEEK_END*/
656 		bf->l_start += XFS_ISIZE(ip);
657 		break;
658 	default:
659 		error = -EINVAL;
660 		goto out_unlock;
661 	}
662 
663 	/*
664 	 * length of <= 0 for resv/unresv/zero is invalid.  length for
665 	 * alloc/free is ignored completely and we have no idea what userspace
666 	 * might have set it to, so set it to zero to allow range
667 	 * checks to pass.
668 	 */
669 	switch (cmd) {
670 	case XFS_IOC_ZERO_RANGE:
671 	case XFS_IOC_RESVSP:
672 	case XFS_IOC_RESVSP64:
673 	case XFS_IOC_UNRESVSP:
674 	case XFS_IOC_UNRESVSP64:
675 		if (bf->l_len <= 0) {
676 			error = -EINVAL;
677 			goto out_unlock;
678 		}
679 		break;
680 	default:
681 		bf->l_len = 0;
682 		break;
683 	}
684 
685 	if (bf->l_start < 0 ||
686 	    bf->l_start > inode->i_sb->s_maxbytes ||
687 	    bf->l_start + bf->l_len < 0 ||
688 	    bf->l_start + bf->l_len >= inode->i_sb->s_maxbytes) {
689 		error = -EINVAL;
690 		goto out_unlock;
691 	}
692 
693 	switch (cmd) {
694 	case XFS_IOC_ZERO_RANGE:
695 		flags |= XFS_PREALLOC_SET;
696 		error = xfs_zero_file_space(ip, bf->l_start, bf->l_len);
697 		break;
698 	case XFS_IOC_RESVSP:
699 	case XFS_IOC_RESVSP64:
700 		flags |= XFS_PREALLOC_SET;
701 		error = xfs_alloc_file_space(ip, bf->l_start, bf->l_len,
702 						XFS_BMAPI_PREALLOC);
703 		break;
704 	case XFS_IOC_UNRESVSP:
705 	case XFS_IOC_UNRESVSP64:
706 		error = xfs_free_file_space(ip, bf->l_start, bf->l_len);
707 		break;
708 	case XFS_IOC_ALLOCSP:
709 	case XFS_IOC_ALLOCSP64:
710 	case XFS_IOC_FREESP:
711 	case XFS_IOC_FREESP64:
712 		flags |= XFS_PREALLOC_CLEAR;
713 		if (bf->l_start > XFS_ISIZE(ip)) {
714 			error = xfs_alloc_file_space(ip, XFS_ISIZE(ip),
715 					bf->l_start - XFS_ISIZE(ip), 0);
716 			if (error)
717 				goto out_unlock;
718 		}
719 
720 		iattr.ia_valid = ATTR_SIZE;
721 		iattr.ia_size = bf->l_start;
722 
723 		error = xfs_setattr_size(ip, &iattr);
724 		break;
725 	default:
726 		ASSERT(0);
727 		error = -EINVAL;
728 	}
729 
730 	if (error)
731 		goto out_unlock;
732 
733 	error = xfs_update_prealloc_flags(ip, flags);
734 
735 out_unlock:
736 	xfs_iunlock(ip, iolock);
737 	mnt_drop_write_file(filp);
738 	return error;
739 }
740 
741 STATIC int
742 xfs_ioc_bulkstat(
743 	xfs_mount_t		*mp,
744 	unsigned int		cmd,
745 	void			__user *arg)
746 {
747 	xfs_fsop_bulkreq_t	bulkreq;
748 	int			count;	/* # of records returned */
749 	xfs_ino_t		inlast;	/* last inode number */
750 	int			done;
751 	int			error;
752 
753 	/* done = 1 if there are more stats to get and if bulkstat */
754 	/* should be called again (unused here, but used in dmapi) */
755 
756 	if (!capable(CAP_SYS_ADMIN))
757 		return -EPERM;
758 
759 	if (XFS_FORCED_SHUTDOWN(mp))
760 		return -EIO;
761 
762 	if (copy_from_user(&bulkreq, arg, sizeof(xfs_fsop_bulkreq_t)))
763 		return -EFAULT;
764 
765 	if (copy_from_user(&inlast, bulkreq.lastip, sizeof(__s64)))
766 		return -EFAULT;
767 
768 	if ((count = bulkreq.icount) <= 0)
769 		return -EINVAL;
770 
771 	if (bulkreq.ubuffer == NULL)
772 		return -EINVAL;
773 
774 	if (cmd == XFS_IOC_FSINUMBERS)
775 		error = xfs_inumbers(mp, &inlast, &count,
776 					bulkreq.ubuffer, xfs_inumbers_fmt);
777 	else if (cmd == XFS_IOC_FSBULKSTAT_SINGLE)
778 		error = xfs_bulkstat_one(mp, inlast, bulkreq.ubuffer,
779 					sizeof(xfs_bstat_t), NULL, &done);
780 	else	/* XFS_IOC_FSBULKSTAT */
781 		error = xfs_bulkstat(mp, &inlast, &count, xfs_bulkstat_one,
782 				     sizeof(xfs_bstat_t), bulkreq.ubuffer,
783 				     &done);
784 
785 	if (error)
786 		return error;
787 
788 	if (bulkreq.ocount != NULL) {
789 		if (copy_to_user(bulkreq.lastip, &inlast,
790 						sizeof(xfs_ino_t)))
791 			return -EFAULT;
792 
793 		if (copy_to_user(bulkreq.ocount, &count, sizeof(count)))
794 			return -EFAULT;
795 	}
796 
797 	return 0;
798 }
799 
800 STATIC int
801 xfs_ioc_fsgeometry_v1(
802 	xfs_mount_t		*mp,
803 	void			__user *arg)
804 {
805 	xfs_fsop_geom_t         fsgeo;
806 	int			error;
807 
808 	error = xfs_fs_geometry(mp, &fsgeo, 3);
809 	if (error)
810 		return error;
811 
812 	/*
813 	 * Caller should have passed an argument of type
814 	 * xfs_fsop_geom_v1_t.  This is a proper subset of the
815 	 * xfs_fsop_geom_t that xfs_fs_geometry() fills in.
816 	 */
817 	if (copy_to_user(arg, &fsgeo, sizeof(xfs_fsop_geom_v1_t)))
818 		return -EFAULT;
819 	return 0;
820 }
821 
822 STATIC int
823 xfs_ioc_fsgeometry(
824 	xfs_mount_t		*mp,
825 	void			__user *arg)
826 {
827 	xfs_fsop_geom_t		fsgeo;
828 	int			error;
829 
830 	error = xfs_fs_geometry(mp, &fsgeo, 4);
831 	if (error)
832 		return error;
833 
834 	if (copy_to_user(arg, &fsgeo, sizeof(fsgeo)))
835 		return -EFAULT;
836 	return 0;
837 }
838 
839 /*
840  * Linux extended inode flags interface.
841  */
842 
843 STATIC unsigned int
844 xfs_merge_ioc_xflags(
845 	unsigned int	flags,
846 	unsigned int	start)
847 {
848 	unsigned int	xflags = start;
849 
850 	if (flags & FS_IMMUTABLE_FL)
851 		xflags |= FS_XFLAG_IMMUTABLE;
852 	else
853 		xflags &= ~FS_XFLAG_IMMUTABLE;
854 	if (flags & FS_APPEND_FL)
855 		xflags |= FS_XFLAG_APPEND;
856 	else
857 		xflags &= ~FS_XFLAG_APPEND;
858 	if (flags & FS_SYNC_FL)
859 		xflags |= FS_XFLAG_SYNC;
860 	else
861 		xflags &= ~FS_XFLAG_SYNC;
862 	if (flags & FS_NOATIME_FL)
863 		xflags |= FS_XFLAG_NOATIME;
864 	else
865 		xflags &= ~FS_XFLAG_NOATIME;
866 	if (flags & FS_NODUMP_FL)
867 		xflags |= FS_XFLAG_NODUMP;
868 	else
869 		xflags &= ~FS_XFLAG_NODUMP;
870 
871 	return xflags;
872 }
873 
874 STATIC unsigned int
875 xfs_di2lxflags(
876 	__uint16_t	di_flags)
877 {
878 	unsigned int	flags = 0;
879 
880 	if (di_flags & XFS_DIFLAG_IMMUTABLE)
881 		flags |= FS_IMMUTABLE_FL;
882 	if (di_flags & XFS_DIFLAG_APPEND)
883 		flags |= FS_APPEND_FL;
884 	if (di_flags & XFS_DIFLAG_SYNC)
885 		flags |= FS_SYNC_FL;
886 	if (di_flags & XFS_DIFLAG_NOATIME)
887 		flags |= FS_NOATIME_FL;
888 	if (di_flags & XFS_DIFLAG_NODUMP)
889 		flags |= FS_NODUMP_FL;
890 	return flags;
891 }
892 
893 STATIC int
894 xfs_ioc_fsgetxattr(
895 	xfs_inode_t		*ip,
896 	int			attr,
897 	void			__user *arg)
898 {
899 	struct fsxattr		fa;
900 
901 	memset(&fa, 0, sizeof(struct fsxattr));
902 
903 	xfs_ilock(ip, XFS_ILOCK_SHARED);
904 	fa.fsx_xflags = xfs_ip2xflags(ip);
905 	fa.fsx_extsize = ip->i_d.di_extsize << ip->i_mount->m_sb.sb_blocklog;
906 	fa.fsx_projid = xfs_get_projid(ip);
907 
908 	if (attr) {
909 		if (ip->i_afp) {
910 			if (ip->i_afp->if_flags & XFS_IFEXTENTS)
911 				fa.fsx_nextents = ip->i_afp->if_bytes /
912 							sizeof(xfs_bmbt_rec_t);
913 			else
914 				fa.fsx_nextents = ip->i_d.di_anextents;
915 		} else
916 			fa.fsx_nextents = 0;
917 	} else {
918 		if (ip->i_df.if_flags & XFS_IFEXTENTS)
919 			fa.fsx_nextents = ip->i_df.if_bytes /
920 						sizeof(xfs_bmbt_rec_t);
921 		else
922 			fa.fsx_nextents = ip->i_d.di_nextents;
923 	}
924 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
925 
926 	if (copy_to_user(arg, &fa, sizeof(fa)))
927 		return -EFAULT;
928 	return 0;
929 }
930 
931 STATIC void
932 xfs_set_diflags(
933 	struct xfs_inode	*ip,
934 	unsigned int		xflags)
935 {
936 	unsigned int		di_flags;
937 	uint64_t		di_flags2;
938 
939 	/* can't set PREALLOC this way, just preserve it */
940 	di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
941 	if (xflags & FS_XFLAG_IMMUTABLE)
942 		di_flags |= XFS_DIFLAG_IMMUTABLE;
943 	if (xflags & FS_XFLAG_APPEND)
944 		di_flags |= XFS_DIFLAG_APPEND;
945 	if (xflags & FS_XFLAG_SYNC)
946 		di_flags |= XFS_DIFLAG_SYNC;
947 	if (xflags & FS_XFLAG_NOATIME)
948 		di_flags |= XFS_DIFLAG_NOATIME;
949 	if (xflags & FS_XFLAG_NODUMP)
950 		di_flags |= XFS_DIFLAG_NODUMP;
951 	if (xflags & FS_XFLAG_NODEFRAG)
952 		di_flags |= XFS_DIFLAG_NODEFRAG;
953 	if (xflags & FS_XFLAG_FILESTREAM)
954 		di_flags |= XFS_DIFLAG_FILESTREAM;
955 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
956 		if (xflags & FS_XFLAG_RTINHERIT)
957 			di_flags |= XFS_DIFLAG_RTINHERIT;
958 		if (xflags & FS_XFLAG_NOSYMLINKS)
959 			di_flags |= XFS_DIFLAG_NOSYMLINKS;
960 		if (xflags & FS_XFLAG_EXTSZINHERIT)
961 			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
962 		if (xflags & FS_XFLAG_PROJINHERIT)
963 			di_flags |= XFS_DIFLAG_PROJINHERIT;
964 	} else if (S_ISREG(VFS_I(ip)->i_mode)) {
965 		if (xflags & FS_XFLAG_REALTIME)
966 			di_flags |= XFS_DIFLAG_REALTIME;
967 		if (xflags & FS_XFLAG_EXTSIZE)
968 			di_flags |= XFS_DIFLAG_EXTSIZE;
969 	}
970 	ip->i_d.di_flags = di_flags;
971 
972 	/* diflags2 only valid for v3 inodes. */
973 	if (ip->i_d.di_version < 3)
974 		return;
975 
976 	di_flags2 = 0;
977 	if (xflags & FS_XFLAG_DAX)
978 		di_flags2 |= XFS_DIFLAG2_DAX;
979 
980 	ip->i_d.di_flags2 = di_flags2;
981 
982 }
983 
984 STATIC void
985 xfs_diflags_to_linux(
986 	struct xfs_inode	*ip)
987 {
988 	struct inode		*inode = VFS_I(ip);
989 	unsigned int		xflags = xfs_ip2xflags(ip);
990 
991 	if (xflags & FS_XFLAG_IMMUTABLE)
992 		inode->i_flags |= S_IMMUTABLE;
993 	else
994 		inode->i_flags &= ~S_IMMUTABLE;
995 	if (xflags & FS_XFLAG_APPEND)
996 		inode->i_flags |= S_APPEND;
997 	else
998 		inode->i_flags &= ~S_APPEND;
999 	if (xflags & FS_XFLAG_SYNC)
1000 		inode->i_flags |= S_SYNC;
1001 	else
1002 		inode->i_flags &= ~S_SYNC;
1003 	if (xflags & FS_XFLAG_NOATIME)
1004 		inode->i_flags |= S_NOATIME;
1005 	else
1006 		inode->i_flags &= ~S_NOATIME;
1007 	if (xflags & FS_XFLAG_DAX)
1008 		inode->i_flags |= S_DAX;
1009 	else
1010 		inode->i_flags &= ~S_DAX;
1011 
1012 }
1013 
1014 static int
1015 xfs_ioctl_setattr_xflags(
1016 	struct xfs_trans	*tp,
1017 	struct xfs_inode	*ip,
1018 	struct fsxattr		*fa)
1019 {
1020 	struct xfs_mount	*mp = ip->i_mount;
1021 
1022 	/* Can't change realtime flag if any extents are allocated. */
1023 	if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
1024 	    XFS_IS_REALTIME_INODE(ip) != (fa->fsx_xflags & FS_XFLAG_REALTIME))
1025 		return -EINVAL;
1026 
1027 	/* If realtime flag is set then must have realtime device */
1028 	if (fa->fsx_xflags & FS_XFLAG_REALTIME) {
1029 		if (mp->m_sb.sb_rblocks == 0 || mp->m_sb.sb_rextsize == 0 ||
1030 		    (ip->i_d.di_extsize % mp->m_sb.sb_rextsize))
1031 			return -EINVAL;
1032 	}
1033 
1034 	/*
1035 	 * Can't modify an immutable/append-only file unless
1036 	 * we have appropriate permission.
1037 	 */
1038 	if (((ip->i_d.di_flags & (XFS_DIFLAG_IMMUTABLE | XFS_DIFLAG_APPEND)) ||
1039 	     (fa->fsx_xflags & (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND))) &&
1040 	    !capable(CAP_LINUX_IMMUTABLE))
1041 		return -EPERM;
1042 
1043 	xfs_set_diflags(ip, fa->fsx_xflags);
1044 	xfs_diflags_to_linux(ip);
1045 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1046 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1047 	XFS_STATS_INC(mp, xs_ig_attrchg);
1048 	return 0;
1049 }
1050 
1051 /*
1052  * If we are changing DAX flags, we have to ensure the file is clean and any
1053  * cached objects in the address space are invalidated and removed. This
1054  * requires us to lock out other IO and page faults similar to a truncate
1055  * operation. The locks need to be held until the transaction has been committed
1056  * so that the cache invalidation is atomic with respect to the DAX flag
1057  * manipulation.
1058  */
1059 static int
1060 xfs_ioctl_setattr_dax_invalidate(
1061 	struct xfs_inode	*ip,
1062 	struct fsxattr		*fa,
1063 	int			*join_flags)
1064 {
1065 	struct inode		*inode = VFS_I(ip);
1066 	int			error;
1067 
1068 	*join_flags = 0;
1069 
1070 	/*
1071 	 * It is only valid to set the DAX flag on regular files and
1072 	 * directories on filesystems where the block size is equal to the page
1073 	 * size. On directories it serves as an inherit hint.
1074 	 */
1075 	if (fa->fsx_xflags & FS_XFLAG_DAX) {
1076 		if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1077 			return -EINVAL;
1078 		if (ip->i_mount->m_sb.sb_blocksize != PAGE_SIZE)
1079 			return -EINVAL;
1080 	}
1081 
1082 	/* If the DAX state is not changing, we have nothing to do here. */
1083 	if ((fa->fsx_xflags & FS_XFLAG_DAX) && IS_DAX(inode))
1084 		return 0;
1085 	if (!(fa->fsx_xflags & FS_XFLAG_DAX) && !IS_DAX(inode))
1086 		return 0;
1087 
1088 	/* lock, flush and invalidate mapping in preparation for flag change */
1089 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1090 	error = filemap_write_and_wait(inode->i_mapping);
1091 	if (error)
1092 		goto out_unlock;
1093 	error = invalidate_inode_pages2(inode->i_mapping);
1094 	if (error)
1095 		goto out_unlock;
1096 
1097 	*join_flags = XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL;
1098 	return 0;
1099 
1100 out_unlock:
1101 	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL | XFS_IOLOCK_EXCL);
1102 	return error;
1103 
1104 }
1105 
1106 /*
1107  * Set up the transaction structure for the setattr operation, checking that we
1108  * have permission to do so. On success, return a clean transaction and the
1109  * inode locked exclusively ready for further operation specific checks. On
1110  * failure, return an error without modifying or locking the inode.
1111  *
1112  * The inode might already be IO locked on call. If this is the case, it is
1113  * indicated in @join_flags and we take full responsibility for ensuring they
1114  * are unlocked from now on. Hence if we have an error here, we still have to
1115  * unlock them. Otherwise, once they are joined to the transaction, they will
1116  * be unlocked on commit/cancel.
1117  */
1118 static struct xfs_trans *
1119 xfs_ioctl_setattr_get_trans(
1120 	struct xfs_inode	*ip,
1121 	int			join_flags)
1122 {
1123 	struct xfs_mount	*mp = ip->i_mount;
1124 	struct xfs_trans	*tp;
1125 	int			error = -EROFS;
1126 
1127 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1128 		goto out_unlock;
1129 	error = -EIO;
1130 	if (XFS_FORCED_SHUTDOWN(mp))
1131 		goto out_unlock;
1132 
1133 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1134 	if (error)
1135 		return ERR_PTR(error);
1136 
1137 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1138 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | join_flags);
1139 	join_flags = 0;
1140 
1141 	/*
1142 	 * CAP_FOWNER overrides the following restrictions:
1143 	 *
1144 	 * The user ID of the calling process must be equal to the file owner
1145 	 * ID, except in cases where the CAP_FSETID capability is applicable.
1146 	 */
1147 	if (!inode_owner_or_capable(VFS_I(ip))) {
1148 		error = -EPERM;
1149 		goto out_cancel;
1150 	}
1151 
1152 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1153 		xfs_trans_set_sync(tp);
1154 
1155 	return tp;
1156 
1157 out_cancel:
1158 	xfs_trans_cancel(tp);
1159 out_unlock:
1160 	if (join_flags)
1161 		xfs_iunlock(ip, join_flags);
1162 	return ERR_PTR(error);
1163 }
1164 
1165 /*
1166  * extent size hint validation is somewhat cumbersome. Rules are:
1167  *
1168  * 1. extent size hint is only valid for directories and regular files
1169  * 2. FS_XFLAG_EXTSIZE is only valid for regular files
1170  * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
1171  * 4. can only be changed on regular files if no extents are allocated
1172  * 5. can be changed on directories at any time
1173  * 6. extsize hint of 0 turns off hints, clears inode flags.
1174  * 7. Extent size must be a multiple of the appropriate block size.
1175  * 8. for non-realtime files, the extent size hint must be limited
1176  *    to half the AG size to avoid alignment extending the extent beyond the
1177  *    limits of the AG.
1178  */
1179 static int
1180 xfs_ioctl_setattr_check_extsize(
1181 	struct xfs_inode	*ip,
1182 	struct fsxattr		*fa)
1183 {
1184 	struct xfs_mount	*mp = ip->i_mount;
1185 
1186 	if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(VFS_I(ip)->i_mode))
1187 		return -EINVAL;
1188 
1189 	if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
1190 	    !S_ISDIR(VFS_I(ip)->i_mode))
1191 		return -EINVAL;
1192 
1193 	if (S_ISREG(VFS_I(ip)->i_mode) && ip->i_d.di_nextents &&
1194 	    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != fa->fsx_extsize))
1195 		return -EINVAL;
1196 
1197 	if (fa->fsx_extsize != 0) {
1198 		xfs_extlen_t    size;
1199 		xfs_fsblock_t   extsize_fsb;
1200 
1201 		extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize);
1202 		if (extsize_fsb > MAXEXTLEN)
1203 			return -EINVAL;
1204 
1205 		if (XFS_IS_REALTIME_INODE(ip) ||
1206 		    (fa->fsx_xflags & FS_XFLAG_REALTIME)) {
1207 			size = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
1208 		} else {
1209 			size = mp->m_sb.sb_blocksize;
1210 			if (extsize_fsb > mp->m_sb.sb_agblocks / 2)
1211 				return -EINVAL;
1212 		}
1213 
1214 		if (fa->fsx_extsize % size)
1215 			return -EINVAL;
1216 	} else
1217 		fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
1218 
1219 	return 0;
1220 }
1221 
1222 static int
1223 xfs_ioctl_setattr_check_projid(
1224 	struct xfs_inode	*ip,
1225 	struct fsxattr		*fa)
1226 {
1227 	/* Disallow 32bit project ids if projid32bit feature is not enabled. */
1228 	if (fa->fsx_projid > (__uint16_t)-1 &&
1229 	    !xfs_sb_version_hasprojid32bit(&ip->i_mount->m_sb))
1230 		return -EINVAL;
1231 
1232 	/*
1233 	 * Project Quota ID state is only allowed to change from within the init
1234 	 * namespace. Enforce that restriction only if we are trying to change
1235 	 * the quota ID state. Everything else is allowed in user namespaces.
1236 	 */
1237 	if (current_user_ns() == &init_user_ns)
1238 		return 0;
1239 
1240 	if (xfs_get_projid(ip) != fa->fsx_projid)
1241 		return -EINVAL;
1242 	if ((fa->fsx_xflags & FS_XFLAG_PROJINHERIT) !=
1243 	    (ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT))
1244 		return -EINVAL;
1245 
1246 	return 0;
1247 }
1248 
1249 STATIC int
1250 xfs_ioctl_setattr(
1251 	xfs_inode_t		*ip,
1252 	struct fsxattr		*fa)
1253 {
1254 	struct xfs_mount	*mp = ip->i_mount;
1255 	struct xfs_trans	*tp;
1256 	struct xfs_dquot	*udqp = NULL;
1257 	struct xfs_dquot	*pdqp = NULL;
1258 	struct xfs_dquot	*olddquot = NULL;
1259 	int			code;
1260 	int			join_flags = 0;
1261 
1262 	trace_xfs_ioctl_setattr(ip);
1263 
1264 	code = xfs_ioctl_setattr_check_projid(ip, fa);
1265 	if (code)
1266 		return code;
1267 
1268 	/*
1269 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
1270 	 * before we start any other transactions. Trying to do this later
1271 	 * is messy. We don't care to take a readlock to look at the ids
1272 	 * in inode here, because we can't hold it across the trans_reserve.
1273 	 * If the IDs do change before we take the ilock, we're covered
1274 	 * because the i_*dquot fields will get updated anyway.
1275 	 */
1276 	if (XFS_IS_QUOTA_ON(mp)) {
1277 		code = xfs_qm_vop_dqalloc(ip, ip->i_d.di_uid,
1278 					 ip->i_d.di_gid, fa->fsx_projid,
1279 					 XFS_QMOPT_PQUOTA, &udqp, NULL, &pdqp);
1280 		if (code)
1281 			return code;
1282 	}
1283 
1284 	/*
1285 	 * Changing DAX config may require inode locking for mapping
1286 	 * invalidation. These need to be held all the way to transaction commit
1287 	 * or cancel time, so need to be passed through to
1288 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1289 	 * appropriately.
1290 	 */
1291 	code = xfs_ioctl_setattr_dax_invalidate(ip, fa, &join_flags);
1292 	if (code)
1293 		goto error_free_dquots;
1294 
1295 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1296 	if (IS_ERR(tp)) {
1297 		code = PTR_ERR(tp);
1298 		goto error_free_dquots;
1299 	}
1300 
1301 
1302 	if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp) &&
1303 	    xfs_get_projid(ip) != fa->fsx_projid) {
1304 		code = xfs_qm_vop_chown_reserve(tp, ip, udqp, NULL, pdqp,
1305 				capable(CAP_FOWNER) ?  XFS_QMOPT_FORCE_RES : 0);
1306 		if (code)	/* out of quota */
1307 			goto error_trans_cancel;
1308 	}
1309 
1310 	code = xfs_ioctl_setattr_check_extsize(ip, fa);
1311 	if (code)
1312 		goto error_trans_cancel;
1313 
1314 	code = xfs_ioctl_setattr_xflags(tp, ip, fa);
1315 	if (code)
1316 		goto error_trans_cancel;
1317 
1318 	/*
1319 	 * Change file ownership.  Must be the owner or privileged.  CAP_FSETID
1320 	 * overrides the following restrictions:
1321 	 *
1322 	 * The set-user-ID and set-group-ID bits of a file will be cleared upon
1323 	 * successful return from chown()
1324 	 */
1325 
1326 	if ((VFS_I(ip)->i_mode & (S_ISUID|S_ISGID)) &&
1327 	    !capable_wrt_inode_uidgid(VFS_I(ip), CAP_FSETID))
1328 		VFS_I(ip)->i_mode &= ~(S_ISUID|S_ISGID);
1329 
1330 	/* Change the ownerships and register project quota modifications */
1331 	if (xfs_get_projid(ip) != fa->fsx_projid) {
1332 		if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_PQUOTA_ON(mp)) {
1333 			olddquot = xfs_qm_vop_chown(tp, ip,
1334 						&ip->i_pdquot, pdqp);
1335 		}
1336 		ASSERT(ip->i_d.di_version > 1);
1337 		xfs_set_projid(ip, fa->fsx_projid);
1338 	}
1339 
1340 	/*
1341 	 * Only set the extent size hint if we've already determined that the
1342 	 * extent size hint should be set on the inode. If no extent size flags
1343 	 * are set on the inode then unconditionally clear the extent size hint.
1344 	 */
1345 	if (ip->i_d.di_flags & (XFS_DIFLAG_EXTSIZE | XFS_DIFLAG_EXTSZINHERIT))
1346 		ip->i_d.di_extsize = fa->fsx_extsize >> mp->m_sb.sb_blocklog;
1347 	else
1348 		ip->i_d.di_extsize = 0;
1349 
1350 	code = xfs_trans_commit(tp);
1351 
1352 	/*
1353 	 * Release any dquot(s) the inode had kept before chown.
1354 	 */
1355 	xfs_qm_dqrele(olddquot);
1356 	xfs_qm_dqrele(udqp);
1357 	xfs_qm_dqrele(pdqp);
1358 
1359 	return code;
1360 
1361 error_trans_cancel:
1362 	xfs_trans_cancel(tp);
1363 error_free_dquots:
1364 	xfs_qm_dqrele(udqp);
1365 	xfs_qm_dqrele(pdqp);
1366 	return code;
1367 }
1368 
1369 STATIC int
1370 xfs_ioc_fssetxattr(
1371 	xfs_inode_t		*ip,
1372 	struct file		*filp,
1373 	void			__user *arg)
1374 {
1375 	struct fsxattr		fa;
1376 	int error;
1377 
1378 	if (copy_from_user(&fa, arg, sizeof(fa)))
1379 		return -EFAULT;
1380 
1381 	error = mnt_want_write_file(filp);
1382 	if (error)
1383 		return error;
1384 	error = xfs_ioctl_setattr(ip, &fa);
1385 	mnt_drop_write_file(filp);
1386 	return error;
1387 }
1388 
1389 STATIC int
1390 xfs_ioc_getxflags(
1391 	xfs_inode_t		*ip,
1392 	void			__user *arg)
1393 {
1394 	unsigned int		flags;
1395 
1396 	flags = xfs_di2lxflags(ip->i_d.di_flags);
1397 	if (copy_to_user(arg, &flags, sizeof(flags)))
1398 		return -EFAULT;
1399 	return 0;
1400 }
1401 
1402 STATIC int
1403 xfs_ioc_setxflags(
1404 	struct xfs_inode	*ip,
1405 	struct file		*filp,
1406 	void			__user *arg)
1407 {
1408 	struct xfs_trans	*tp;
1409 	struct fsxattr		fa;
1410 	unsigned int		flags;
1411 	int			join_flags = 0;
1412 	int			error;
1413 
1414 	if (copy_from_user(&flags, arg, sizeof(flags)))
1415 		return -EFAULT;
1416 
1417 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
1418 		      FS_NOATIME_FL | FS_NODUMP_FL | \
1419 		      FS_SYNC_FL))
1420 		return -EOPNOTSUPP;
1421 
1422 	fa.fsx_xflags = xfs_merge_ioc_xflags(flags, xfs_ip2xflags(ip));
1423 
1424 	error = mnt_want_write_file(filp);
1425 	if (error)
1426 		return error;
1427 
1428 	/*
1429 	 * Changing DAX config may require inode locking for mapping
1430 	 * invalidation. These need to be held all the way to transaction commit
1431 	 * or cancel time, so need to be passed through to
1432 	 * xfs_ioctl_setattr_get_trans() so it can apply them to the join call
1433 	 * appropriately.
1434 	 */
1435 	error = xfs_ioctl_setattr_dax_invalidate(ip, &fa, &join_flags);
1436 	if (error)
1437 		goto out_drop_write;
1438 
1439 	tp = xfs_ioctl_setattr_get_trans(ip, join_flags);
1440 	if (IS_ERR(tp)) {
1441 		error = PTR_ERR(tp);
1442 		goto out_drop_write;
1443 	}
1444 
1445 	error = xfs_ioctl_setattr_xflags(tp, ip, &fa);
1446 	if (error) {
1447 		xfs_trans_cancel(tp);
1448 		goto out_drop_write;
1449 	}
1450 
1451 	error = xfs_trans_commit(tp);
1452 out_drop_write:
1453 	mnt_drop_write_file(filp);
1454 	return error;
1455 }
1456 
1457 STATIC int
1458 xfs_getbmap_format(void **ap, struct getbmapx *bmv, int *full)
1459 {
1460 	struct getbmap __user	*base = (struct getbmap __user *)*ap;
1461 
1462 	/* copy only getbmap portion (not getbmapx) */
1463 	if (copy_to_user(base, bmv, sizeof(struct getbmap)))
1464 		return -EFAULT;
1465 
1466 	*ap += sizeof(struct getbmap);
1467 	return 0;
1468 }
1469 
1470 STATIC int
1471 xfs_ioc_getbmap(
1472 	struct file		*file,
1473 	unsigned int		cmd,
1474 	void			__user *arg)
1475 {
1476 	struct getbmapx		bmx;
1477 	int			error;
1478 
1479 	if (copy_from_user(&bmx, arg, sizeof(struct getbmapx)))
1480 		return -EFAULT;
1481 
1482 	if (bmx.bmv_count < 2)
1483 		return -EINVAL;
1484 
1485 	bmx.bmv_iflags = (cmd == XFS_IOC_GETBMAPA ? BMV_IF_ATTRFORK : 0);
1486 	if (file->f_mode & FMODE_NOCMTIME)
1487 		bmx.bmv_iflags |= BMV_IF_NO_DMAPI_READ;
1488 
1489 	error = xfs_getbmap(XFS_I(file_inode(file)), &bmx, xfs_getbmap_format,
1490 			    (__force struct getbmap *)arg+1);
1491 	if (error)
1492 		return error;
1493 
1494 	/* copy back header - only size of getbmap */
1495 	if (copy_to_user(arg, &bmx, sizeof(struct getbmap)))
1496 		return -EFAULT;
1497 	return 0;
1498 }
1499 
1500 STATIC int
1501 xfs_getbmapx_format(void **ap, struct getbmapx *bmv, int *full)
1502 {
1503 	struct getbmapx __user	*base = (struct getbmapx __user *)*ap;
1504 
1505 	if (copy_to_user(base, bmv, sizeof(struct getbmapx)))
1506 		return -EFAULT;
1507 
1508 	*ap += sizeof(struct getbmapx);
1509 	return 0;
1510 }
1511 
1512 STATIC int
1513 xfs_ioc_getbmapx(
1514 	struct xfs_inode	*ip,
1515 	void			__user *arg)
1516 {
1517 	struct getbmapx		bmx;
1518 	int			error;
1519 
1520 	if (copy_from_user(&bmx, arg, sizeof(bmx)))
1521 		return -EFAULT;
1522 
1523 	if (bmx.bmv_count < 2)
1524 		return -EINVAL;
1525 
1526 	if (bmx.bmv_iflags & (~BMV_IF_VALID))
1527 		return -EINVAL;
1528 
1529 	error = xfs_getbmap(ip, &bmx, xfs_getbmapx_format,
1530 			    (__force struct getbmapx *)arg+1);
1531 	if (error)
1532 		return error;
1533 
1534 	/* copy back header */
1535 	if (copy_to_user(arg, &bmx, sizeof(struct getbmapx)))
1536 		return -EFAULT;
1537 
1538 	return 0;
1539 }
1540 
1541 int
1542 xfs_ioc_swapext(
1543 	xfs_swapext_t	*sxp)
1544 {
1545 	xfs_inode_t     *ip, *tip;
1546 	struct fd	f, tmp;
1547 	int		error = 0;
1548 
1549 	/* Pull information for the target fd */
1550 	f = fdget((int)sxp->sx_fdtarget);
1551 	if (!f.file) {
1552 		error = -EINVAL;
1553 		goto out;
1554 	}
1555 
1556 	if (!(f.file->f_mode & FMODE_WRITE) ||
1557 	    !(f.file->f_mode & FMODE_READ) ||
1558 	    (f.file->f_flags & O_APPEND)) {
1559 		error = -EBADF;
1560 		goto out_put_file;
1561 	}
1562 
1563 	tmp = fdget((int)sxp->sx_fdtmp);
1564 	if (!tmp.file) {
1565 		error = -EINVAL;
1566 		goto out_put_file;
1567 	}
1568 
1569 	if (!(tmp.file->f_mode & FMODE_WRITE) ||
1570 	    !(tmp.file->f_mode & FMODE_READ) ||
1571 	    (tmp.file->f_flags & O_APPEND)) {
1572 		error = -EBADF;
1573 		goto out_put_tmp_file;
1574 	}
1575 
1576 	if (IS_SWAPFILE(file_inode(f.file)) ||
1577 	    IS_SWAPFILE(file_inode(tmp.file))) {
1578 		error = -EINVAL;
1579 		goto out_put_tmp_file;
1580 	}
1581 
1582 	/*
1583 	 * We need to ensure that the fds passed in point to XFS inodes
1584 	 * before we cast and access them as XFS structures as we have no
1585 	 * control over what the user passes us here.
1586 	 */
1587 	if (f.file->f_op != &xfs_file_operations ||
1588 	    tmp.file->f_op != &xfs_file_operations) {
1589 		error = -EINVAL;
1590 		goto out_put_tmp_file;
1591 	}
1592 
1593 	ip = XFS_I(file_inode(f.file));
1594 	tip = XFS_I(file_inode(tmp.file));
1595 
1596 	if (ip->i_mount != tip->i_mount) {
1597 		error = -EINVAL;
1598 		goto out_put_tmp_file;
1599 	}
1600 
1601 	if (ip->i_ino == tip->i_ino) {
1602 		error = -EINVAL;
1603 		goto out_put_tmp_file;
1604 	}
1605 
1606 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1607 		error = -EIO;
1608 		goto out_put_tmp_file;
1609 	}
1610 
1611 	error = xfs_swap_extents(ip, tip, sxp);
1612 
1613  out_put_tmp_file:
1614 	fdput(tmp);
1615  out_put_file:
1616 	fdput(f);
1617  out:
1618 	return error;
1619 }
1620 
1621 /*
1622  * Note: some of the ioctl's return positive numbers as a
1623  * byte count indicating success, such as readlink_by_handle.
1624  * So we don't "sign flip" like most other routines.  This means
1625  * true errors need to be returned as a negative value.
1626  */
1627 long
1628 xfs_file_ioctl(
1629 	struct file		*filp,
1630 	unsigned int		cmd,
1631 	unsigned long		p)
1632 {
1633 	struct inode		*inode = file_inode(filp);
1634 	struct xfs_inode	*ip = XFS_I(inode);
1635 	struct xfs_mount	*mp = ip->i_mount;
1636 	void			__user *arg = (void __user *)p;
1637 	int			error;
1638 
1639 	trace_xfs_file_ioctl(ip);
1640 
1641 	switch (cmd) {
1642 	case FITRIM:
1643 		return xfs_ioc_trim(mp, arg);
1644 	case XFS_IOC_ALLOCSP:
1645 	case XFS_IOC_FREESP:
1646 	case XFS_IOC_RESVSP:
1647 	case XFS_IOC_UNRESVSP:
1648 	case XFS_IOC_ALLOCSP64:
1649 	case XFS_IOC_FREESP64:
1650 	case XFS_IOC_RESVSP64:
1651 	case XFS_IOC_UNRESVSP64:
1652 	case XFS_IOC_ZERO_RANGE: {
1653 		xfs_flock64_t		bf;
1654 
1655 		if (copy_from_user(&bf, arg, sizeof(bf)))
1656 			return -EFAULT;
1657 		return xfs_ioc_space(filp, cmd, &bf);
1658 	}
1659 	case XFS_IOC_DIOINFO: {
1660 		struct dioattr	da;
1661 		xfs_buftarg_t	*target =
1662 			XFS_IS_REALTIME_INODE(ip) ?
1663 			mp->m_rtdev_targp : mp->m_ddev_targp;
1664 
1665 		da.d_mem =  da.d_miniosz = target->bt_logical_sectorsize;
1666 		da.d_maxiosz = INT_MAX & ~(da.d_miniosz - 1);
1667 
1668 		if (copy_to_user(arg, &da, sizeof(da)))
1669 			return -EFAULT;
1670 		return 0;
1671 	}
1672 
1673 	case XFS_IOC_FSBULKSTAT_SINGLE:
1674 	case XFS_IOC_FSBULKSTAT:
1675 	case XFS_IOC_FSINUMBERS:
1676 		return xfs_ioc_bulkstat(mp, cmd, arg);
1677 
1678 	case XFS_IOC_FSGEOMETRY_V1:
1679 		return xfs_ioc_fsgeometry_v1(mp, arg);
1680 
1681 	case XFS_IOC_FSGEOMETRY:
1682 		return xfs_ioc_fsgeometry(mp, arg);
1683 
1684 	case XFS_IOC_GETVERSION:
1685 		return put_user(inode->i_generation, (int __user *)arg);
1686 
1687 	case XFS_IOC_FSGETXATTR:
1688 		return xfs_ioc_fsgetxattr(ip, 0, arg);
1689 	case XFS_IOC_FSGETXATTRA:
1690 		return xfs_ioc_fsgetxattr(ip, 1, arg);
1691 	case XFS_IOC_FSSETXATTR:
1692 		return xfs_ioc_fssetxattr(ip, filp, arg);
1693 	case XFS_IOC_GETXFLAGS:
1694 		return xfs_ioc_getxflags(ip, arg);
1695 	case XFS_IOC_SETXFLAGS:
1696 		return xfs_ioc_setxflags(ip, filp, arg);
1697 
1698 	case XFS_IOC_FSSETDM: {
1699 		struct fsdmidata	dmi;
1700 
1701 		if (copy_from_user(&dmi, arg, sizeof(dmi)))
1702 			return -EFAULT;
1703 
1704 		error = mnt_want_write_file(filp);
1705 		if (error)
1706 			return error;
1707 
1708 		error = xfs_set_dmattrs(ip, dmi.fsd_dmevmask,
1709 				dmi.fsd_dmstate);
1710 		mnt_drop_write_file(filp);
1711 		return error;
1712 	}
1713 
1714 	case XFS_IOC_GETBMAP:
1715 	case XFS_IOC_GETBMAPA:
1716 		return xfs_ioc_getbmap(filp, cmd, arg);
1717 
1718 	case XFS_IOC_GETBMAPX:
1719 		return xfs_ioc_getbmapx(ip, arg);
1720 
1721 	case XFS_IOC_FD_TO_HANDLE:
1722 	case XFS_IOC_PATH_TO_HANDLE:
1723 	case XFS_IOC_PATH_TO_FSHANDLE: {
1724 		xfs_fsop_handlereq_t	hreq;
1725 
1726 		if (copy_from_user(&hreq, arg, sizeof(hreq)))
1727 			return -EFAULT;
1728 		return xfs_find_handle(cmd, &hreq);
1729 	}
1730 	case XFS_IOC_OPEN_BY_HANDLE: {
1731 		xfs_fsop_handlereq_t	hreq;
1732 
1733 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
1734 			return -EFAULT;
1735 		return xfs_open_by_handle(filp, &hreq);
1736 	}
1737 	case XFS_IOC_FSSETDM_BY_HANDLE:
1738 		return xfs_fssetdm_by_handle(filp, arg);
1739 
1740 	case XFS_IOC_READLINK_BY_HANDLE: {
1741 		xfs_fsop_handlereq_t	hreq;
1742 
1743 		if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
1744 			return -EFAULT;
1745 		return xfs_readlink_by_handle(filp, &hreq);
1746 	}
1747 	case XFS_IOC_ATTRLIST_BY_HANDLE:
1748 		return xfs_attrlist_by_handle(filp, arg);
1749 
1750 	case XFS_IOC_ATTRMULTI_BY_HANDLE:
1751 		return xfs_attrmulti_by_handle(filp, arg);
1752 
1753 	case XFS_IOC_SWAPEXT: {
1754 		struct xfs_swapext	sxp;
1755 
1756 		if (copy_from_user(&sxp, arg, sizeof(xfs_swapext_t)))
1757 			return -EFAULT;
1758 		error = mnt_want_write_file(filp);
1759 		if (error)
1760 			return error;
1761 		error = xfs_ioc_swapext(&sxp);
1762 		mnt_drop_write_file(filp);
1763 		return error;
1764 	}
1765 
1766 	case XFS_IOC_FSCOUNTS: {
1767 		xfs_fsop_counts_t out;
1768 
1769 		error = xfs_fs_counts(mp, &out);
1770 		if (error)
1771 			return error;
1772 
1773 		if (copy_to_user(arg, &out, sizeof(out)))
1774 			return -EFAULT;
1775 		return 0;
1776 	}
1777 
1778 	case XFS_IOC_SET_RESBLKS: {
1779 		xfs_fsop_resblks_t inout;
1780 		__uint64_t	   in;
1781 
1782 		if (!capable(CAP_SYS_ADMIN))
1783 			return -EPERM;
1784 
1785 		if (mp->m_flags & XFS_MOUNT_RDONLY)
1786 			return -EROFS;
1787 
1788 		if (copy_from_user(&inout, arg, sizeof(inout)))
1789 			return -EFAULT;
1790 
1791 		error = mnt_want_write_file(filp);
1792 		if (error)
1793 			return error;
1794 
1795 		/* input parameter is passed in resblks field of structure */
1796 		in = inout.resblks;
1797 		error = xfs_reserve_blocks(mp, &in, &inout);
1798 		mnt_drop_write_file(filp);
1799 		if (error)
1800 			return error;
1801 
1802 		if (copy_to_user(arg, &inout, sizeof(inout)))
1803 			return -EFAULT;
1804 		return 0;
1805 	}
1806 
1807 	case XFS_IOC_GET_RESBLKS: {
1808 		xfs_fsop_resblks_t out;
1809 
1810 		if (!capable(CAP_SYS_ADMIN))
1811 			return -EPERM;
1812 
1813 		error = xfs_reserve_blocks(mp, NULL, &out);
1814 		if (error)
1815 			return error;
1816 
1817 		if (copy_to_user(arg, &out, sizeof(out)))
1818 			return -EFAULT;
1819 
1820 		return 0;
1821 	}
1822 
1823 	case XFS_IOC_FSGROWFSDATA: {
1824 		xfs_growfs_data_t in;
1825 
1826 		if (copy_from_user(&in, arg, sizeof(in)))
1827 			return -EFAULT;
1828 
1829 		error = mnt_want_write_file(filp);
1830 		if (error)
1831 			return error;
1832 		error = xfs_growfs_data(mp, &in);
1833 		mnt_drop_write_file(filp);
1834 		return error;
1835 	}
1836 
1837 	case XFS_IOC_FSGROWFSLOG: {
1838 		xfs_growfs_log_t in;
1839 
1840 		if (copy_from_user(&in, arg, sizeof(in)))
1841 			return -EFAULT;
1842 
1843 		error = mnt_want_write_file(filp);
1844 		if (error)
1845 			return error;
1846 		error = xfs_growfs_log(mp, &in);
1847 		mnt_drop_write_file(filp);
1848 		return error;
1849 	}
1850 
1851 	case XFS_IOC_FSGROWFSRT: {
1852 		xfs_growfs_rt_t in;
1853 
1854 		if (copy_from_user(&in, arg, sizeof(in)))
1855 			return -EFAULT;
1856 
1857 		error = mnt_want_write_file(filp);
1858 		if (error)
1859 			return error;
1860 		error = xfs_growfs_rt(mp, &in);
1861 		mnt_drop_write_file(filp);
1862 		return error;
1863 	}
1864 
1865 	case XFS_IOC_GOINGDOWN: {
1866 		__uint32_t in;
1867 
1868 		if (!capable(CAP_SYS_ADMIN))
1869 			return -EPERM;
1870 
1871 		if (get_user(in, (__uint32_t __user *)arg))
1872 			return -EFAULT;
1873 
1874 		return xfs_fs_goingdown(mp, in);
1875 	}
1876 
1877 	case XFS_IOC_ERROR_INJECTION: {
1878 		xfs_error_injection_t in;
1879 
1880 		if (!capable(CAP_SYS_ADMIN))
1881 			return -EPERM;
1882 
1883 		if (copy_from_user(&in, arg, sizeof(in)))
1884 			return -EFAULT;
1885 
1886 		return xfs_errortag_add(in.errtag, mp);
1887 	}
1888 
1889 	case XFS_IOC_ERROR_CLEARALL:
1890 		if (!capable(CAP_SYS_ADMIN))
1891 			return -EPERM;
1892 
1893 		return xfs_errortag_clearall(mp, 1);
1894 
1895 	case XFS_IOC_FREE_EOFBLOCKS: {
1896 		struct xfs_fs_eofblocks eofb;
1897 		struct xfs_eofblocks keofb;
1898 
1899 		if (!capable(CAP_SYS_ADMIN))
1900 			return -EPERM;
1901 
1902 		if (mp->m_flags & XFS_MOUNT_RDONLY)
1903 			return -EROFS;
1904 
1905 		if (copy_from_user(&eofb, arg, sizeof(eofb)))
1906 			return -EFAULT;
1907 
1908 		error = xfs_fs_eofblocks_from_user(&eofb, &keofb);
1909 		if (error)
1910 			return error;
1911 
1912 		return xfs_icache_free_eofblocks(mp, &keofb);
1913 	}
1914 
1915 	default:
1916 		return -ENOTTY;
1917 	}
1918 }
1919