xref: /openbmc/linux/drivers/block/loop.c (revision b34e08d5)
1 /*
2  *  linux/drivers/block/loop.c
3  *
4  *  Written by Theodore Ts'o, 3/29/93
5  *
6  * Copyright 1993 by Theodore Ts'o.  Redistribution of this file is
7  * permitted under the GNU General Public License.
8  *
9  * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10  * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11  *
12  * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13  * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14  *
15  * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
16  *
17  * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
18  *
19  * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20  *
21  * Loadable modules and other fixes by AK, 1998
22  *
23  * Make real block number available to downstream transfer functions, enables
24  * CBC (and relatives) mode encryption requiring unique IVs per data block.
25  * Reed H. Petty, rhp@draper.net
26  *
27  * Maximum number of loop devices now dynamic via max_loop module parameter.
28  * Russell Kroll <rkroll@exploits.org> 19990701
29  *
30  * Maximum number of loop devices when compiled-in now selectable by passing
31  * max_loop=<1-255> to the kernel on boot.
32  * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
33  *
34  * Completely rewrite request handling to be make_request_fn style and
35  * non blocking, pushing work to a helper thread. Lots of fixes from
36  * Al Viro too.
37  * Jens Axboe <axboe@suse.de>, Nov 2000
38  *
39  * Support up to 256 loop devices
40  * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
41  *
42  * Support for falling back on the write file operation when the address space
43  * operations write_begin is not available on the backing filesystem.
44  * Anton Altaparmakov, 16 Feb 2005
45  *
46  * Still To Fix:
47  * - Advisory locking is ignored here.
48  * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49  *
50  */
51 
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
55 #include <linux/fs.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include "loop.h"
79 
80 #include <asm/uaccess.h>
81 
82 static DEFINE_IDR(loop_index_idr);
83 static DEFINE_MUTEX(loop_index_mutex);
84 
85 static int max_part;
86 static int part_shift;
87 
88 /*
89  * Transfer functions
90  */
91 static int transfer_none(struct loop_device *lo, int cmd,
92 			 struct page *raw_page, unsigned raw_off,
93 			 struct page *loop_page, unsigned loop_off,
94 			 int size, sector_t real_block)
95 {
96 	char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 	char *loop_buf = kmap_atomic(loop_page) + loop_off;
98 
99 	if (cmd == READ)
100 		memcpy(loop_buf, raw_buf, size);
101 	else
102 		memcpy(raw_buf, loop_buf, size);
103 
104 	kunmap_atomic(loop_buf);
105 	kunmap_atomic(raw_buf);
106 	cond_resched();
107 	return 0;
108 }
109 
110 static int transfer_xor(struct loop_device *lo, int cmd,
111 			struct page *raw_page, unsigned raw_off,
112 			struct page *loop_page, unsigned loop_off,
113 			int size, sector_t real_block)
114 {
115 	char *raw_buf = kmap_atomic(raw_page) + raw_off;
116 	char *loop_buf = kmap_atomic(loop_page) + loop_off;
117 	char *in, *out, *key;
118 	int i, keysize;
119 
120 	if (cmd == READ) {
121 		in = raw_buf;
122 		out = loop_buf;
123 	} else {
124 		in = loop_buf;
125 		out = raw_buf;
126 	}
127 
128 	key = lo->lo_encrypt_key;
129 	keysize = lo->lo_encrypt_key_size;
130 	for (i = 0; i < size; i++)
131 		*out++ = *in++ ^ key[(i & 511) % keysize];
132 
133 	kunmap_atomic(loop_buf);
134 	kunmap_atomic(raw_buf);
135 	cond_resched();
136 	return 0;
137 }
138 
139 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140 {
141 	if (unlikely(info->lo_encrypt_key_size <= 0))
142 		return -EINVAL;
143 	return 0;
144 }
145 
146 static struct loop_func_table none_funcs = {
147 	.number = LO_CRYPT_NONE,
148 	.transfer = transfer_none,
149 };
150 
151 static struct loop_func_table xor_funcs = {
152 	.number = LO_CRYPT_XOR,
153 	.transfer = transfer_xor,
154 	.init = xor_init
155 };
156 
157 /* xfer_funcs[0] is special - its release function is never called */
158 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 	&none_funcs,
160 	&xor_funcs
161 };
162 
163 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
164 {
165 	loff_t loopsize;
166 
167 	/* Compute loopsize in bytes */
168 	loopsize = i_size_read(file->f_mapping->host);
169 	if (offset > 0)
170 		loopsize -= offset;
171 	/* offset is beyond i_size, weird but possible */
172 	if (loopsize < 0)
173 		return 0;
174 
175 	if (sizelimit > 0 && sizelimit < loopsize)
176 		loopsize = sizelimit;
177 	/*
178 	 * Unfortunately, if we want to do I/O on the device,
179 	 * the number of 512-byte sectors has to fit into a sector_t.
180 	 */
181 	return loopsize >> 9;
182 }
183 
184 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
185 {
186 	return get_size(lo->lo_offset, lo->lo_sizelimit, file);
187 }
188 
189 static int
190 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
191 {
192 	loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
193 	sector_t x = (sector_t)size;
194 	struct block_device *bdev = lo->lo_device;
195 
196 	if (unlikely((loff_t)x != size))
197 		return -EFBIG;
198 	if (lo->lo_offset != offset)
199 		lo->lo_offset = offset;
200 	if (lo->lo_sizelimit != sizelimit)
201 		lo->lo_sizelimit = sizelimit;
202 	set_capacity(lo->lo_disk, x);
203 	bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
204 	/* let user-space know about the new size */
205 	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
206 	return 0;
207 }
208 
209 static inline int
210 lo_do_transfer(struct loop_device *lo, int cmd,
211 	       struct page *rpage, unsigned roffs,
212 	       struct page *lpage, unsigned loffs,
213 	       int size, sector_t rblock)
214 {
215 	if (unlikely(!lo->transfer))
216 		return 0;
217 
218 	return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
219 }
220 
221 /**
222  * __do_lo_send_write - helper for writing data to a loop device
223  *
224  * This helper just factors out common code between do_lo_send_direct_write()
225  * and do_lo_send_write().
226  */
227 static int __do_lo_send_write(struct file *file,
228 		u8 *buf, const int len, loff_t pos)
229 {
230 	ssize_t bw;
231 	mm_segment_t old_fs = get_fs();
232 
233 	file_start_write(file);
234 	set_fs(get_ds());
235 	bw = file->f_op->write(file, buf, len, &pos);
236 	set_fs(old_fs);
237 	file_end_write(file);
238 	if (likely(bw == len))
239 		return 0;
240 	printk_ratelimited(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
241 			(unsigned long long)pos, len);
242 	if (bw >= 0)
243 		bw = -EIO;
244 	return bw;
245 }
246 
247 /**
248  * do_lo_send_direct_write - helper for writing data to a loop device
249  *
250  * This is the fast, non-transforming version that does not need double
251  * buffering.
252  */
253 static int do_lo_send_direct_write(struct loop_device *lo,
254 		struct bio_vec *bvec, loff_t pos, struct page *page)
255 {
256 	ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
257 			kmap(bvec->bv_page) + bvec->bv_offset,
258 			bvec->bv_len, pos);
259 	kunmap(bvec->bv_page);
260 	cond_resched();
261 	return bw;
262 }
263 
264 /**
265  * do_lo_send_write - helper for writing data to a loop device
266  *
267  * This is the slow, transforming version that needs to double buffer the
268  * data as it cannot do the transformations in place without having direct
269  * access to the destination pages of the backing file.
270  */
271 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
272 		loff_t pos, struct page *page)
273 {
274 	int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
275 			bvec->bv_offset, bvec->bv_len, pos >> 9);
276 	if (likely(!ret))
277 		return __do_lo_send_write(lo->lo_backing_file,
278 				page_address(page), bvec->bv_len,
279 				pos);
280 	printk_ratelimited(KERN_ERR "loop: Transfer error at byte offset %llu, "
281 			"length %i.\n", (unsigned long long)pos, bvec->bv_len);
282 	if (ret > 0)
283 		ret = -EIO;
284 	return ret;
285 }
286 
287 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
288 {
289 	int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
290 			struct page *page);
291 	struct bio_vec bvec;
292 	struct bvec_iter iter;
293 	struct page *page = NULL;
294 	int ret = 0;
295 
296 	if (lo->transfer != transfer_none) {
297 		page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
298 		if (unlikely(!page))
299 			goto fail;
300 		kmap(page);
301 		do_lo_send = do_lo_send_write;
302 	} else {
303 		do_lo_send = do_lo_send_direct_write;
304 	}
305 
306 	bio_for_each_segment(bvec, bio, iter) {
307 		ret = do_lo_send(lo, &bvec, pos, page);
308 		if (ret < 0)
309 			break;
310 		pos += bvec.bv_len;
311 	}
312 	if (page) {
313 		kunmap(page);
314 		__free_page(page);
315 	}
316 out:
317 	return ret;
318 fail:
319 	printk_ratelimited(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
320 	ret = -ENOMEM;
321 	goto out;
322 }
323 
324 struct lo_read_data {
325 	struct loop_device *lo;
326 	struct page *page;
327 	unsigned offset;
328 	int bsize;
329 };
330 
331 static int
332 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
333 		struct splice_desc *sd)
334 {
335 	struct lo_read_data *p = sd->u.data;
336 	struct loop_device *lo = p->lo;
337 	struct page *page = buf->page;
338 	sector_t IV;
339 	int size;
340 
341 	IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
342 							(buf->offset >> 9);
343 	size = sd->len;
344 	if (size > p->bsize)
345 		size = p->bsize;
346 
347 	if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
348 		printk_ratelimited(KERN_ERR "loop: transfer error block %ld\n",
349 		       page->index);
350 		size = -EINVAL;
351 	}
352 
353 	flush_dcache_page(p->page);
354 
355 	if (size > 0)
356 		p->offset += size;
357 
358 	return size;
359 }
360 
361 static int
362 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
363 {
364 	return __splice_from_pipe(pipe, sd, lo_splice_actor);
365 }
366 
367 static ssize_t
368 do_lo_receive(struct loop_device *lo,
369 	      struct bio_vec *bvec, int bsize, loff_t pos)
370 {
371 	struct lo_read_data cookie;
372 	struct splice_desc sd;
373 	struct file *file;
374 	ssize_t retval;
375 
376 	cookie.lo = lo;
377 	cookie.page = bvec->bv_page;
378 	cookie.offset = bvec->bv_offset;
379 	cookie.bsize = bsize;
380 
381 	sd.len = 0;
382 	sd.total_len = bvec->bv_len;
383 	sd.flags = 0;
384 	sd.pos = pos;
385 	sd.u.data = &cookie;
386 
387 	file = lo->lo_backing_file;
388 	retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
389 
390 	return retval;
391 }
392 
393 static int
394 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
395 {
396 	struct bio_vec bvec;
397 	struct bvec_iter iter;
398 	ssize_t s;
399 
400 	bio_for_each_segment(bvec, bio, iter) {
401 		s = do_lo_receive(lo, &bvec, bsize, pos);
402 		if (s < 0)
403 			return s;
404 
405 		if (s != bvec.bv_len) {
406 			zero_fill_bio(bio);
407 			break;
408 		}
409 		pos += bvec.bv_len;
410 	}
411 	return 0;
412 }
413 
414 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
415 {
416 	loff_t pos;
417 	int ret;
418 
419 	pos = ((loff_t) bio->bi_iter.bi_sector << 9) + lo->lo_offset;
420 
421 	if (bio_rw(bio) == WRITE) {
422 		struct file *file = lo->lo_backing_file;
423 
424 		if (bio->bi_rw & REQ_FLUSH) {
425 			ret = vfs_fsync(file, 0);
426 			if (unlikely(ret && ret != -EINVAL)) {
427 				ret = -EIO;
428 				goto out;
429 			}
430 		}
431 
432 		/*
433 		 * We use punch hole to reclaim the free space used by the
434 		 * image a.k.a. discard. However we do not support discard if
435 		 * encryption is enabled, because it may give an attacker
436 		 * useful information.
437 		 */
438 		if (bio->bi_rw & REQ_DISCARD) {
439 			struct file *file = lo->lo_backing_file;
440 			int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
441 
442 			if ((!file->f_op->fallocate) ||
443 			    lo->lo_encrypt_key_size) {
444 				ret = -EOPNOTSUPP;
445 				goto out;
446 			}
447 			ret = file->f_op->fallocate(file, mode, pos,
448 						    bio->bi_iter.bi_size);
449 			if (unlikely(ret && ret != -EINVAL &&
450 				     ret != -EOPNOTSUPP))
451 				ret = -EIO;
452 			goto out;
453 		}
454 
455 		ret = lo_send(lo, bio, pos);
456 
457 		if ((bio->bi_rw & REQ_FUA) && !ret) {
458 			ret = vfs_fsync(file, 0);
459 			if (unlikely(ret && ret != -EINVAL))
460 				ret = -EIO;
461 		}
462 	} else
463 		ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
464 
465 out:
466 	return ret;
467 }
468 
469 /*
470  * Add bio to back of pending list
471  */
472 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
473 {
474 	lo->lo_bio_count++;
475 	bio_list_add(&lo->lo_bio_list, bio);
476 }
477 
478 /*
479  * Grab first pending buffer
480  */
481 static struct bio *loop_get_bio(struct loop_device *lo)
482 {
483 	lo->lo_bio_count--;
484 	return bio_list_pop(&lo->lo_bio_list);
485 }
486 
487 static void loop_make_request(struct request_queue *q, struct bio *old_bio)
488 {
489 	struct loop_device *lo = q->queuedata;
490 	int rw = bio_rw(old_bio);
491 
492 	if (rw == READA)
493 		rw = READ;
494 
495 	BUG_ON(!lo || (rw != READ && rw != WRITE));
496 
497 	spin_lock_irq(&lo->lo_lock);
498 	if (lo->lo_state != Lo_bound)
499 		goto out;
500 	if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
501 		goto out;
502 	if (lo->lo_bio_count >= q->nr_congestion_on)
503 		wait_event_lock_irq(lo->lo_req_wait,
504 				    lo->lo_bio_count < q->nr_congestion_off,
505 				    lo->lo_lock);
506 	loop_add_bio(lo, old_bio);
507 	wake_up(&lo->lo_event);
508 	spin_unlock_irq(&lo->lo_lock);
509 	return;
510 
511 out:
512 	spin_unlock_irq(&lo->lo_lock);
513 	bio_io_error(old_bio);
514 }
515 
516 struct switch_request {
517 	struct file *file;
518 	struct completion wait;
519 };
520 
521 static void do_loop_switch(struct loop_device *, struct switch_request *);
522 
523 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
524 {
525 	if (unlikely(!bio->bi_bdev)) {
526 		do_loop_switch(lo, bio->bi_private);
527 		bio_put(bio);
528 	} else {
529 		int ret = do_bio_filebacked(lo, bio);
530 		bio_endio(bio, ret);
531 	}
532 }
533 
534 /*
535  * worker thread that handles reads/writes to file backed loop devices,
536  * to avoid blocking in our make_request_fn. it also does loop decrypting
537  * on reads for block backed loop, as that is too heavy to do from
538  * b_end_io context where irqs may be disabled.
539  *
540  * Loop explanation:  loop_clr_fd() sets lo_state to Lo_rundown before
541  * calling kthread_stop().  Therefore once kthread_should_stop() is
542  * true, make_request will not place any more requests.  Therefore
543  * once kthread_should_stop() is true and lo_bio is NULL, we are
544  * done with the loop.
545  */
546 static int loop_thread(void *data)
547 {
548 	struct loop_device *lo = data;
549 	struct bio *bio;
550 
551 	set_user_nice(current, -20);
552 
553 	while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
554 
555 		wait_event_interruptible(lo->lo_event,
556 				!bio_list_empty(&lo->lo_bio_list) ||
557 				kthread_should_stop());
558 
559 		if (bio_list_empty(&lo->lo_bio_list))
560 			continue;
561 		spin_lock_irq(&lo->lo_lock);
562 		bio = loop_get_bio(lo);
563 		if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off)
564 			wake_up(&lo->lo_req_wait);
565 		spin_unlock_irq(&lo->lo_lock);
566 
567 		BUG_ON(!bio);
568 		loop_handle_bio(lo, bio);
569 	}
570 
571 	return 0;
572 }
573 
574 /*
575  * loop_switch performs the hard work of switching a backing store.
576  * First it needs to flush existing IO, it does this by sending a magic
577  * BIO down the pipe. The completion of this BIO does the actual switch.
578  */
579 static int loop_switch(struct loop_device *lo, struct file *file)
580 {
581 	struct switch_request w;
582 	struct bio *bio = bio_alloc(GFP_KERNEL, 0);
583 	if (!bio)
584 		return -ENOMEM;
585 	init_completion(&w.wait);
586 	w.file = file;
587 	bio->bi_private = &w;
588 	bio->bi_bdev = NULL;
589 	loop_make_request(lo->lo_queue, bio);
590 	wait_for_completion(&w.wait);
591 	return 0;
592 }
593 
594 /*
595  * Helper to flush the IOs in loop, but keeping loop thread running
596  */
597 static int loop_flush(struct loop_device *lo)
598 {
599 	/* loop not yet configured, no running thread, nothing to flush */
600 	if (!lo->lo_thread)
601 		return 0;
602 
603 	return loop_switch(lo, NULL);
604 }
605 
606 /*
607  * Do the actual switch; called from the BIO completion routine
608  */
609 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
610 {
611 	struct file *file = p->file;
612 	struct file *old_file = lo->lo_backing_file;
613 	struct address_space *mapping;
614 
615 	/* if no new file, only flush of queued bios requested */
616 	if (!file)
617 		goto out;
618 
619 	mapping = file->f_mapping;
620 	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
621 	lo->lo_backing_file = file;
622 	lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
623 		mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
624 	lo->old_gfp_mask = mapping_gfp_mask(mapping);
625 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
626 out:
627 	complete(&p->wait);
628 }
629 
630 
631 /*
632  * loop_change_fd switched the backing store of a loopback device to
633  * a new file. This is useful for operating system installers to free up
634  * the original file and in High Availability environments to switch to
635  * an alternative location for the content in case of server meltdown.
636  * This can only work if the loop device is used read-only, and if the
637  * new backing store is the same size and type as the old backing store.
638  */
639 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
640 			  unsigned int arg)
641 {
642 	struct file	*file, *old_file;
643 	struct inode	*inode;
644 	int		error;
645 
646 	error = -ENXIO;
647 	if (lo->lo_state != Lo_bound)
648 		goto out;
649 
650 	/* the loop device has to be read-only */
651 	error = -EINVAL;
652 	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
653 		goto out;
654 
655 	error = -EBADF;
656 	file = fget(arg);
657 	if (!file)
658 		goto out;
659 
660 	inode = file->f_mapping->host;
661 	old_file = lo->lo_backing_file;
662 
663 	error = -EINVAL;
664 
665 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
666 		goto out_putf;
667 
668 	/* size of the new backing store needs to be the same */
669 	if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
670 		goto out_putf;
671 
672 	/* and ... switch */
673 	error = loop_switch(lo, file);
674 	if (error)
675 		goto out_putf;
676 
677 	fput(old_file);
678 	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
679 		ioctl_by_bdev(bdev, BLKRRPART, 0);
680 	return 0;
681 
682  out_putf:
683 	fput(file);
684  out:
685 	return error;
686 }
687 
688 static inline int is_loop_device(struct file *file)
689 {
690 	struct inode *i = file->f_mapping->host;
691 
692 	return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
693 }
694 
695 /* loop sysfs attributes */
696 
697 static ssize_t loop_attr_show(struct device *dev, char *page,
698 			      ssize_t (*callback)(struct loop_device *, char *))
699 {
700 	struct gendisk *disk = dev_to_disk(dev);
701 	struct loop_device *lo = disk->private_data;
702 
703 	return callback(lo, page);
704 }
705 
706 #define LOOP_ATTR_RO(_name)						\
707 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);	\
708 static ssize_t loop_attr_do_show_##_name(struct device *d,		\
709 				struct device_attribute *attr, char *b)	\
710 {									\
711 	return loop_attr_show(d, b, loop_attr_##_name##_show);		\
712 }									\
713 static struct device_attribute loop_attr_##_name =			\
714 	__ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
715 
716 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
717 {
718 	ssize_t ret;
719 	char *p = NULL;
720 
721 	spin_lock_irq(&lo->lo_lock);
722 	if (lo->lo_backing_file)
723 		p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
724 	spin_unlock_irq(&lo->lo_lock);
725 
726 	if (IS_ERR_OR_NULL(p))
727 		ret = PTR_ERR(p);
728 	else {
729 		ret = strlen(p);
730 		memmove(buf, p, ret);
731 		buf[ret++] = '\n';
732 		buf[ret] = 0;
733 	}
734 
735 	return ret;
736 }
737 
738 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
739 {
740 	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
741 }
742 
743 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
744 {
745 	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
746 }
747 
748 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
749 {
750 	int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
751 
752 	return sprintf(buf, "%s\n", autoclear ? "1" : "0");
753 }
754 
755 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
756 {
757 	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
758 
759 	return sprintf(buf, "%s\n", partscan ? "1" : "0");
760 }
761 
762 LOOP_ATTR_RO(backing_file);
763 LOOP_ATTR_RO(offset);
764 LOOP_ATTR_RO(sizelimit);
765 LOOP_ATTR_RO(autoclear);
766 LOOP_ATTR_RO(partscan);
767 
768 static struct attribute *loop_attrs[] = {
769 	&loop_attr_backing_file.attr,
770 	&loop_attr_offset.attr,
771 	&loop_attr_sizelimit.attr,
772 	&loop_attr_autoclear.attr,
773 	&loop_attr_partscan.attr,
774 	NULL,
775 };
776 
777 static struct attribute_group loop_attribute_group = {
778 	.name = "loop",
779 	.attrs= loop_attrs,
780 };
781 
782 static int loop_sysfs_init(struct loop_device *lo)
783 {
784 	return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
785 				  &loop_attribute_group);
786 }
787 
788 static void loop_sysfs_exit(struct loop_device *lo)
789 {
790 	sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
791 			   &loop_attribute_group);
792 }
793 
794 static void loop_config_discard(struct loop_device *lo)
795 {
796 	struct file *file = lo->lo_backing_file;
797 	struct inode *inode = file->f_mapping->host;
798 	struct request_queue *q = lo->lo_queue;
799 
800 	/*
801 	 * We use punch hole to reclaim the free space used by the
802 	 * image a.k.a. discard. However we do not support discard if
803 	 * encryption is enabled, because it may give an attacker
804 	 * useful information.
805 	 */
806 	if ((!file->f_op->fallocate) ||
807 	    lo->lo_encrypt_key_size) {
808 		q->limits.discard_granularity = 0;
809 		q->limits.discard_alignment = 0;
810 		q->limits.max_discard_sectors = 0;
811 		q->limits.discard_zeroes_data = 0;
812 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
813 		return;
814 	}
815 
816 	q->limits.discard_granularity = inode->i_sb->s_blocksize;
817 	q->limits.discard_alignment = 0;
818 	q->limits.max_discard_sectors = UINT_MAX >> 9;
819 	q->limits.discard_zeroes_data = 1;
820 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
821 }
822 
823 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
824 		       struct block_device *bdev, unsigned int arg)
825 {
826 	struct file	*file, *f;
827 	struct inode	*inode;
828 	struct address_space *mapping;
829 	unsigned lo_blocksize;
830 	int		lo_flags = 0;
831 	int		error;
832 	loff_t		size;
833 
834 	/* This is safe, since we have a reference from open(). */
835 	__module_get(THIS_MODULE);
836 
837 	error = -EBADF;
838 	file = fget(arg);
839 	if (!file)
840 		goto out;
841 
842 	error = -EBUSY;
843 	if (lo->lo_state != Lo_unbound)
844 		goto out_putf;
845 
846 	/* Avoid recursion */
847 	f = file;
848 	while (is_loop_device(f)) {
849 		struct loop_device *l;
850 
851 		if (f->f_mapping->host->i_bdev == bdev)
852 			goto out_putf;
853 
854 		l = f->f_mapping->host->i_bdev->bd_disk->private_data;
855 		if (l->lo_state == Lo_unbound) {
856 			error = -EINVAL;
857 			goto out_putf;
858 		}
859 		f = l->lo_backing_file;
860 	}
861 
862 	mapping = file->f_mapping;
863 	inode = mapping->host;
864 
865 	error = -EINVAL;
866 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
867 		goto out_putf;
868 
869 	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
870 	    !file->f_op->write)
871 		lo_flags |= LO_FLAGS_READ_ONLY;
872 
873 	lo_blocksize = S_ISBLK(inode->i_mode) ?
874 		inode->i_bdev->bd_block_size : PAGE_SIZE;
875 
876 	error = -EFBIG;
877 	size = get_loop_size(lo, file);
878 	if ((loff_t)(sector_t)size != size)
879 		goto out_putf;
880 
881 	error = 0;
882 
883 	set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
884 
885 	lo->lo_blocksize = lo_blocksize;
886 	lo->lo_device = bdev;
887 	lo->lo_flags = lo_flags;
888 	lo->lo_backing_file = file;
889 	lo->transfer = transfer_none;
890 	lo->ioctl = NULL;
891 	lo->lo_sizelimit = 0;
892 	lo->lo_bio_count = 0;
893 	lo->old_gfp_mask = mapping_gfp_mask(mapping);
894 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
895 
896 	bio_list_init(&lo->lo_bio_list);
897 
898 	if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
899 		blk_queue_flush(lo->lo_queue, REQ_FLUSH);
900 
901 	set_capacity(lo->lo_disk, size);
902 	bd_set_size(bdev, size << 9);
903 	loop_sysfs_init(lo);
904 	/* let user-space know about the new size */
905 	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
906 
907 	set_blocksize(bdev, lo_blocksize);
908 
909 	lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
910 						lo->lo_number);
911 	if (IS_ERR(lo->lo_thread)) {
912 		error = PTR_ERR(lo->lo_thread);
913 		goto out_clr;
914 	}
915 	lo->lo_state = Lo_bound;
916 	wake_up_process(lo->lo_thread);
917 	if (part_shift)
918 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
919 	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
920 		ioctl_by_bdev(bdev, BLKRRPART, 0);
921 
922 	/* Grab the block_device to prevent its destruction after we
923 	 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
924 	 */
925 	bdgrab(bdev);
926 	return 0;
927 
928 out_clr:
929 	loop_sysfs_exit(lo);
930 	lo->lo_thread = NULL;
931 	lo->lo_device = NULL;
932 	lo->lo_backing_file = NULL;
933 	lo->lo_flags = 0;
934 	set_capacity(lo->lo_disk, 0);
935 	invalidate_bdev(bdev);
936 	bd_set_size(bdev, 0);
937 	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
938 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
939 	lo->lo_state = Lo_unbound;
940  out_putf:
941 	fput(file);
942  out:
943 	/* This is safe: open() is still holding a reference. */
944 	module_put(THIS_MODULE);
945 	return error;
946 }
947 
948 static int
949 loop_release_xfer(struct loop_device *lo)
950 {
951 	int err = 0;
952 	struct loop_func_table *xfer = lo->lo_encryption;
953 
954 	if (xfer) {
955 		if (xfer->release)
956 			err = xfer->release(lo);
957 		lo->transfer = NULL;
958 		lo->lo_encryption = NULL;
959 		module_put(xfer->owner);
960 	}
961 	return err;
962 }
963 
964 static int
965 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
966 	       const struct loop_info64 *i)
967 {
968 	int err = 0;
969 
970 	if (xfer) {
971 		struct module *owner = xfer->owner;
972 
973 		if (!try_module_get(owner))
974 			return -EINVAL;
975 		if (xfer->init)
976 			err = xfer->init(lo, i);
977 		if (err)
978 			module_put(owner);
979 		else
980 			lo->lo_encryption = xfer;
981 	}
982 	return err;
983 }
984 
985 static int loop_clr_fd(struct loop_device *lo)
986 {
987 	struct file *filp = lo->lo_backing_file;
988 	gfp_t gfp = lo->old_gfp_mask;
989 	struct block_device *bdev = lo->lo_device;
990 
991 	if (lo->lo_state != Lo_bound)
992 		return -ENXIO;
993 
994 	/*
995 	 * If we've explicitly asked to tear down the loop device,
996 	 * and it has an elevated reference count, set it for auto-teardown when
997 	 * the last reference goes away. This stops $!~#$@ udev from
998 	 * preventing teardown because it decided that it needs to run blkid on
999 	 * the loopback device whenever they appear. xfstests is notorious for
1000 	 * failing tests because blkid via udev races with a losetup
1001 	 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1002 	 * command to fail with EBUSY.
1003 	 */
1004 	if (lo->lo_refcnt > 1) {
1005 		lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1006 		mutex_unlock(&lo->lo_ctl_mutex);
1007 		return 0;
1008 	}
1009 
1010 	if (filp == NULL)
1011 		return -EINVAL;
1012 
1013 	spin_lock_irq(&lo->lo_lock);
1014 	lo->lo_state = Lo_rundown;
1015 	spin_unlock_irq(&lo->lo_lock);
1016 
1017 	kthread_stop(lo->lo_thread);
1018 
1019 	spin_lock_irq(&lo->lo_lock);
1020 	lo->lo_backing_file = NULL;
1021 	spin_unlock_irq(&lo->lo_lock);
1022 
1023 	loop_release_xfer(lo);
1024 	lo->transfer = NULL;
1025 	lo->ioctl = NULL;
1026 	lo->lo_device = NULL;
1027 	lo->lo_encryption = NULL;
1028 	lo->lo_offset = 0;
1029 	lo->lo_sizelimit = 0;
1030 	lo->lo_encrypt_key_size = 0;
1031 	lo->lo_thread = NULL;
1032 	memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1033 	memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1034 	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1035 	if (bdev) {
1036 		bdput(bdev);
1037 		invalidate_bdev(bdev);
1038 	}
1039 	set_capacity(lo->lo_disk, 0);
1040 	loop_sysfs_exit(lo);
1041 	if (bdev) {
1042 		bd_set_size(bdev, 0);
1043 		/* let user-space know about this change */
1044 		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1045 	}
1046 	mapping_set_gfp_mask(filp->f_mapping, gfp);
1047 	lo->lo_state = Lo_unbound;
1048 	/* This is safe: open() is still holding a reference. */
1049 	module_put(THIS_MODULE);
1050 	if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1051 		ioctl_by_bdev(bdev, BLKRRPART, 0);
1052 	lo->lo_flags = 0;
1053 	if (!part_shift)
1054 		lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1055 	mutex_unlock(&lo->lo_ctl_mutex);
1056 	/*
1057 	 * Need not hold lo_ctl_mutex to fput backing file.
1058 	 * Calling fput holding lo_ctl_mutex triggers a circular
1059 	 * lock dependency possibility warning as fput can take
1060 	 * bd_mutex which is usually taken before lo_ctl_mutex.
1061 	 */
1062 	fput(filp);
1063 	return 0;
1064 }
1065 
1066 static int
1067 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1068 {
1069 	int err;
1070 	struct loop_func_table *xfer;
1071 	kuid_t uid = current_uid();
1072 
1073 	if (lo->lo_encrypt_key_size &&
1074 	    !uid_eq(lo->lo_key_owner, uid) &&
1075 	    !capable(CAP_SYS_ADMIN))
1076 		return -EPERM;
1077 	if (lo->lo_state != Lo_bound)
1078 		return -ENXIO;
1079 	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1080 		return -EINVAL;
1081 
1082 	err = loop_release_xfer(lo);
1083 	if (err)
1084 		return err;
1085 
1086 	if (info->lo_encrypt_type) {
1087 		unsigned int type = info->lo_encrypt_type;
1088 
1089 		if (type >= MAX_LO_CRYPT)
1090 			return -EINVAL;
1091 		xfer = xfer_funcs[type];
1092 		if (xfer == NULL)
1093 			return -EINVAL;
1094 	} else
1095 		xfer = NULL;
1096 
1097 	err = loop_init_xfer(lo, xfer, info);
1098 	if (err)
1099 		return err;
1100 
1101 	if (lo->lo_offset != info->lo_offset ||
1102 	    lo->lo_sizelimit != info->lo_sizelimit)
1103 		if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1104 			return -EFBIG;
1105 
1106 	loop_config_discard(lo);
1107 
1108 	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1109 	memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1110 	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1111 	lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1112 
1113 	if (!xfer)
1114 		xfer = &none_funcs;
1115 	lo->transfer = xfer->transfer;
1116 	lo->ioctl = xfer->ioctl;
1117 
1118 	if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1119 	     (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1120 		lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1121 
1122 	if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1123 	     !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1124 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1125 		lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1126 		ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1127 	}
1128 
1129 	lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1130 	lo->lo_init[0] = info->lo_init[0];
1131 	lo->lo_init[1] = info->lo_init[1];
1132 	if (info->lo_encrypt_key_size) {
1133 		memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1134 		       info->lo_encrypt_key_size);
1135 		lo->lo_key_owner = uid;
1136 	}
1137 
1138 	return 0;
1139 }
1140 
1141 static int
1142 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1143 {
1144 	struct file *file = lo->lo_backing_file;
1145 	struct kstat stat;
1146 	int error;
1147 
1148 	if (lo->lo_state != Lo_bound)
1149 		return -ENXIO;
1150 	error = vfs_getattr(&file->f_path, &stat);
1151 	if (error)
1152 		return error;
1153 	memset(info, 0, sizeof(*info));
1154 	info->lo_number = lo->lo_number;
1155 	info->lo_device = huge_encode_dev(stat.dev);
1156 	info->lo_inode = stat.ino;
1157 	info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1158 	info->lo_offset = lo->lo_offset;
1159 	info->lo_sizelimit = lo->lo_sizelimit;
1160 	info->lo_flags = lo->lo_flags;
1161 	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1162 	memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1163 	info->lo_encrypt_type =
1164 		lo->lo_encryption ? lo->lo_encryption->number : 0;
1165 	if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1166 		info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1167 		memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1168 		       lo->lo_encrypt_key_size);
1169 	}
1170 	return 0;
1171 }
1172 
1173 static void
1174 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1175 {
1176 	memset(info64, 0, sizeof(*info64));
1177 	info64->lo_number = info->lo_number;
1178 	info64->lo_device = info->lo_device;
1179 	info64->lo_inode = info->lo_inode;
1180 	info64->lo_rdevice = info->lo_rdevice;
1181 	info64->lo_offset = info->lo_offset;
1182 	info64->lo_sizelimit = 0;
1183 	info64->lo_encrypt_type = info->lo_encrypt_type;
1184 	info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1185 	info64->lo_flags = info->lo_flags;
1186 	info64->lo_init[0] = info->lo_init[0];
1187 	info64->lo_init[1] = info->lo_init[1];
1188 	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1189 		memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1190 	else
1191 		memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1192 	memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1193 }
1194 
1195 static int
1196 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1197 {
1198 	memset(info, 0, sizeof(*info));
1199 	info->lo_number = info64->lo_number;
1200 	info->lo_device = info64->lo_device;
1201 	info->lo_inode = info64->lo_inode;
1202 	info->lo_rdevice = info64->lo_rdevice;
1203 	info->lo_offset = info64->lo_offset;
1204 	info->lo_encrypt_type = info64->lo_encrypt_type;
1205 	info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1206 	info->lo_flags = info64->lo_flags;
1207 	info->lo_init[0] = info64->lo_init[0];
1208 	info->lo_init[1] = info64->lo_init[1];
1209 	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1210 		memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1211 	else
1212 		memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1213 	memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1214 
1215 	/* error in case values were truncated */
1216 	if (info->lo_device != info64->lo_device ||
1217 	    info->lo_rdevice != info64->lo_rdevice ||
1218 	    info->lo_inode != info64->lo_inode ||
1219 	    info->lo_offset != info64->lo_offset)
1220 		return -EOVERFLOW;
1221 
1222 	return 0;
1223 }
1224 
1225 static int
1226 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1227 {
1228 	struct loop_info info;
1229 	struct loop_info64 info64;
1230 
1231 	if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1232 		return -EFAULT;
1233 	loop_info64_from_old(&info, &info64);
1234 	return loop_set_status(lo, &info64);
1235 }
1236 
1237 static int
1238 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1239 {
1240 	struct loop_info64 info64;
1241 
1242 	if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1243 		return -EFAULT;
1244 	return loop_set_status(lo, &info64);
1245 }
1246 
1247 static int
1248 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1249 	struct loop_info info;
1250 	struct loop_info64 info64;
1251 	int err = 0;
1252 
1253 	if (!arg)
1254 		err = -EINVAL;
1255 	if (!err)
1256 		err = loop_get_status(lo, &info64);
1257 	if (!err)
1258 		err = loop_info64_to_old(&info64, &info);
1259 	if (!err && copy_to_user(arg, &info, sizeof(info)))
1260 		err = -EFAULT;
1261 
1262 	return err;
1263 }
1264 
1265 static int
1266 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1267 	struct loop_info64 info64;
1268 	int err = 0;
1269 
1270 	if (!arg)
1271 		err = -EINVAL;
1272 	if (!err)
1273 		err = loop_get_status(lo, &info64);
1274 	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1275 		err = -EFAULT;
1276 
1277 	return err;
1278 }
1279 
1280 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1281 {
1282 	if (unlikely(lo->lo_state != Lo_bound))
1283 		return -ENXIO;
1284 
1285 	return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1286 }
1287 
1288 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1289 	unsigned int cmd, unsigned long arg)
1290 {
1291 	struct loop_device *lo = bdev->bd_disk->private_data;
1292 	int err;
1293 
1294 	mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1295 	switch (cmd) {
1296 	case LOOP_SET_FD:
1297 		err = loop_set_fd(lo, mode, bdev, arg);
1298 		break;
1299 	case LOOP_CHANGE_FD:
1300 		err = loop_change_fd(lo, bdev, arg);
1301 		break;
1302 	case LOOP_CLR_FD:
1303 		/* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1304 		err = loop_clr_fd(lo);
1305 		if (!err)
1306 			goto out_unlocked;
1307 		break;
1308 	case LOOP_SET_STATUS:
1309 		err = -EPERM;
1310 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1311 			err = loop_set_status_old(lo,
1312 					(struct loop_info __user *)arg);
1313 		break;
1314 	case LOOP_GET_STATUS:
1315 		err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1316 		break;
1317 	case LOOP_SET_STATUS64:
1318 		err = -EPERM;
1319 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1320 			err = loop_set_status64(lo,
1321 					(struct loop_info64 __user *) arg);
1322 		break;
1323 	case LOOP_GET_STATUS64:
1324 		err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1325 		break;
1326 	case LOOP_SET_CAPACITY:
1327 		err = -EPERM;
1328 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1329 			err = loop_set_capacity(lo, bdev);
1330 		break;
1331 	default:
1332 		err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1333 	}
1334 	mutex_unlock(&lo->lo_ctl_mutex);
1335 
1336 out_unlocked:
1337 	return err;
1338 }
1339 
1340 #ifdef CONFIG_COMPAT
1341 struct compat_loop_info {
1342 	compat_int_t	lo_number;      /* ioctl r/o */
1343 	compat_dev_t	lo_device;      /* ioctl r/o */
1344 	compat_ulong_t	lo_inode;       /* ioctl r/o */
1345 	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1346 	compat_int_t	lo_offset;
1347 	compat_int_t	lo_encrypt_type;
1348 	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1349 	compat_int_t	lo_flags;       /* ioctl r/o */
1350 	char		lo_name[LO_NAME_SIZE];
1351 	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1352 	compat_ulong_t	lo_init[2];
1353 	char		reserved[4];
1354 };
1355 
1356 /*
1357  * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1358  * - noinlined to reduce stack space usage in main part of driver
1359  */
1360 static noinline int
1361 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1362 			struct loop_info64 *info64)
1363 {
1364 	struct compat_loop_info info;
1365 
1366 	if (copy_from_user(&info, arg, sizeof(info)))
1367 		return -EFAULT;
1368 
1369 	memset(info64, 0, sizeof(*info64));
1370 	info64->lo_number = info.lo_number;
1371 	info64->lo_device = info.lo_device;
1372 	info64->lo_inode = info.lo_inode;
1373 	info64->lo_rdevice = info.lo_rdevice;
1374 	info64->lo_offset = info.lo_offset;
1375 	info64->lo_sizelimit = 0;
1376 	info64->lo_encrypt_type = info.lo_encrypt_type;
1377 	info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1378 	info64->lo_flags = info.lo_flags;
1379 	info64->lo_init[0] = info.lo_init[0];
1380 	info64->lo_init[1] = info.lo_init[1];
1381 	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1382 		memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1383 	else
1384 		memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1385 	memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1386 	return 0;
1387 }
1388 
1389 /*
1390  * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1391  * - noinlined to reduce stack space usage in main part of driver
1392  */
1393 static noinline int
1394 loop_info64_to_compat(const struct loop_info64 *info64,
1395 		      struct compat_loop_info __user *arg)
1396 {
1397 	struct compat_loop_info info;
1398 
1399 	memset(&info, 0, sizeof(info));
1400 	info.lo_number = info64->lo_number;
1401 	info.lo_device = info64->lo_device;
1402 	info.lo_inode = info64->lo_inode;
1403 	info.lo_rdevice = info64->lo_rdevice;
1404 	info.lo_offset = info64->lo_offset;
1405 	info.lo_encrypt_type = info64->lo_encrypt_type;
1406 	info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1407 	info.lo_flags = info64->lo_flags;
1408 	info.lo_init[0] = info64->lo_init[0];
1409 	info.lo_init[1] = info64->lo_init[1];
1410 	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1411 		memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1412 	else
1413 		memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1414 	memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1415 
1416 	/* error in case values were truncated */
1417 	if (info.lo_device != info64->lo_device ||
1418 	    info.lo_rdevice != info64->lo_rdevice ||
1419 	    info.lo_inode != info64->lo_inode ||
1420 	    info.lo_offset != info64->lo_offset ||
1421 	    info.lo_init[0] != info64->lo_init[0] ||
1422 	    info.lo_init[1] != info64->lo_init[1])
1423 		return -EOVERFLOW;
1424 
1425 	if (copy_to_user(arg, &info, sizeof(info)))
1426 		return -EFAULT;
1427 	return 0;
1428 }
1429 
1430 static int
1431 loop_set_status_compat(struct loop_device *lo,
1432 		       const struct compat_loop_info __user *arg)
1433 {
1434 	struct loop_info64 info64;
1435 	int ret;
1436 
1437 	ret = loop_info64_from_compat(arg, &info64);
1438 	if (ret < 0)
1439 		return ret;
1440 	return loop_set_status(lo, &info64);
1441 }
1442 
1443 static int
1444 loop_get_status_compat(struct loop_device *lo,
1445 		       struct compat_loop_info __user *arg)
1446 {
1447 	struct loop_info64 info64;
1448 	int err = 0;
1449 
1450 	if (!arg)
1451 		err = -EINVAL;
1452 	if (!err)
1453 		err = loop_get_status(lo, &info64);
1454 	if (!err)
1455 		err = loop_info64_to_compat(&info64, arg);
1456 	return err;
1457 }
1458 
1459 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1460 			   unsigned int cmd, unsigned long arg)
1461 {
1462 	struct loop_device *lo = bdev->bd_disk->private_data;
1463 	int err;
1464 
1465 	switch(cmd) {
1466 	case LOOP_SET_STATUS:
1467 		mutex_lock(&lo->lo_ctl_mutex);
1468 		err = loop_set_status_compat(
1469 			lo, (const struct compat_loop_info __user *) arg);
1470 		mutex_unlock(&lo->lo_ctl_mutex);
1471 		break;
1472 	case LOOP_GET_STATUS:
1473 		mutex_lock(&lo->lo_ctl_mutex);
1474 		err = loop_get_status_compat(
1475 			lo, (struct compat_loop_info __user *) arg);
1476 		mutex_unlock(&lo->lo_ctl_mutex);
1477 		break;
1478 	case LOOP_SET_CAPACITY:
1479 	case LOOP_CLR_FD:
1480 	case LOOP_GET_STATUS64:
1481 	case LOOP_SET_STATUS64:
1482 		arg = (unsigned long) compat_ptr(arg);
1483 	case LOOP_SET_FD:
1484 	case LOOP_CHANGE_FD:
1485 		err = lo_ioctl(bdev, mode, cmd, arg);
1486 		break;
1487 	default:
1488 		err = -ENOIOCTLCMD;
1489 		break;
1490 	}
1491 	return err;
1492 }
1493 #endif
1494 
1495 static int lo_open(struct block_device *bdev, fmode_t mode)
1496 {
1497 	struct loop_device *lo;
1498 	int err = 0;
1499 
1500 	mutex_lock(&loop_index_mutex);
1501 	lo = bdev->bd_disk->private_data;
1502 	if (!lo) {
1503 		err = -ENXIO;
1504 		goto out;
1505 	}
1506 
1507 	mutex_lock(&lo->lo_ctl_mutex);
1508 	lo->lo_refcnt++;
1509 	mutex_unlock(&lo->lo_ctl_mutex);
1510 out:
1511 	mutex_unlock(&loop_index_mutex);
1512 	return err;
1513 }
1514 
1515 static void lo_release(struct gendisk *disk, fmode_t mode)
1516 {
1517 	struct loop_device *lo = disk->private_data;
1518 	int err;
1519 
1520 	mutex_lock(&lo->lo_ctl_mutex);
1521 
1522 	if (--lo->lo_refcnt)
1523 		goto out;
1524 
1525 	if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1526 		/*
1527 		 * In autoclear mode, stop the loop thread
1528 		 * and remove configuration after last close.
1529 		 */
1530 		err = loop_clr_fd(lo);
1531 		if (!err)
1532 			return;
1533 	} else {
1534 		/*
1535 		 * Otherwise keep thread (if running) and config,
1536 		 * but flush possible ongoing bios in thread.
1537 		 */
1538 		loop_flush(lo);
1539 	}
1540 
1541 out:
1542 	mutex_unlock(&lo->lo_ctl_mutex);
1543 }
1544 
1545 static const struct block_device_operations lo_fops = {
1546 	.owner =	THIS_MODULE,
1547 	.open =		lo_open,
1548 	.release =	lo_release,
1549 	.ioctl =	lo_ioctl,
1550 #ifdef CONFIG_COMPAT
1551 	.compat_ioctl =	lo_compat_ioctl,
1552 #endif
1553 };
1554 
1555 /*
1556  * And now the modules code and kernel interface.
1557  */
1558 static int max_loop;
1559 module_param(max_loop, int, S_IRUGO);
1560 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1561 module_param(max_part, int, S_IRUGO);
1562 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1563 MODULE_LICENSE("GPL");
1564 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1565 
1566 int loop_register_transfer(struct loop_func_table *funcs)
1567 {
1568 	unsigned int n = funcs->number;
1569 
1570 	if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1571 		return -EINVAL;
1572 	xfer_funcs[n] = funcs;
1573 	return 0;
1574 }
1575 
1576 static int unregister_transfer_cb(int id, void *ptr, void *data)
1577 {
1578 	struct loop_device *lo = ptr;
1579 	struct loop_func_table *xfer = data;
1580 
1581 	mutex_lock(&lo->lo_ctl_mutex);
1582 	if (lo->lo_encryption == xfer)
1583 		loop_release_xfer(lo);
1584 	mutex_unlock(&lo->lo_ctl_mutex);
1585 	return 0;
1586 }
1587 
1588 int loop_unregister_transfer(int number)
1589 {
1590 	unsigned int n = number;
1591 	struct loop_func_table *xfer;
1592 
1593 	if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1594 		return -EINVAL;
1595 
1596 	xfer_funcs[n] = NULL;
1597 	idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1598 	return 0;
1599 }
1600 
1601 EXPORT_SYMBOL(loop_register_transfer);
1602 EXPORT_SYMBOL(loop_unregister_transfer);
1603 
1604 static int loop_add(struct loop_device **l, int i)
1605 {
1606 	struct loop_device *lo;
1607 	struct gendisk *disk;
1608 	int err;
1609 
1610 	err = -ENOMEM;
1611 	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1612 	if (!lo)
1613 		goto out;
1614 
1615 	lo->lo_state = Lo_unbound;
1616 
1617 	/* allocate id, if @id >= 0, we're requesting that specific id */
1618 	if (i >= 0) {
1619 		err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1620 		if (err == -ENOSPC)
1621 			err = -EEXIST;
1622 	} else {
1623 		err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1624 	}
1625 	if (err < 0)
1626 		goto out_free_dev;
1627 	i = err;
1628 
1629 	err = -ENOMEM;
1630 	lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1631 	if (!lo->lo_queue)
1632 		goto out_free_idr;
1633 
1634 	/*
1635 	 * set queue make_request_fn
1636 	 */
1637 	blk_queue_make_request(lo->lo_queue, loop_make_request);
1638 	lo->lo_queue->queuedata = lo;
1639 
1640 	disk = lo->lo_disk = alloc_disk(1 << part_shift);
1641 	if (!disk)
1642 		goto out_free_queue;
1643 
1644 	/*
1645 	 * Disable partition scanning by default. The in-kernel partition
1646 	 * scanning can be requested individually per-device during its
1647 	 * setup. Userspace can always add and remove partitions from all
1648 	 * devices. The needed partition minors are allocated from the
1649 	 * extended minor space, the main loop device numbers will continue
1650 	 * to match the loop minors, regardless of the number of partitions
1651 	 * used.
1652 	 *
1653 	 * If max_part is given, partition scanning is globally enabled for
1654 	 * all loop devices. The minors for the main loop devices will be
1655 	 * multiples of max_part.
1656 	 *
1657 	 * Note: Global-for-all-devices, set-only-at-init, read-only module
1658 	 * parameteters like 'max_loop' and 'max_part' make things needlessly
1659 	 * complicated, are too static, inflexible and may surprise
1660 	 * userspace tools. Parameters like this in general should be avoided.
1661 	 */
1662 	if (!part_shift)
1663 		disk->flags |= GENHD_FL_NO_PART_SCAN;
1664 	disk->flags |= GENHD_FL_EXT_DEVT;
1665 	mutex_init(&lo->lo_ctl_mutex);
1666 	lo->lo_number		= i;
1667 	lo->lo_thread		= NULL;
1668 	init_waitqueue_head(&lo->lo_event);
1669 	init_waitqueue_head(&lo->lo_req_wait);
1670 	spin_lock_init(&lo->lo_lock);
1671 	disk->major		= LOOP_MAJOR;
1672 	disk->first_minor	= i << part_shift;
1673 	disk->fops		= &lo_fops;
1674 	disk->private_data	= lo;
1675 	disk->queue		= lo->lo_queue;
1676 	sprintf(disk->disk_name, "loop%d", i);
1677 	add_disk(disk);
1678 	*l = lo;
1679 	return lo->lo_number;
1680 
1681 out_free_queue:
1682 	blk_cleanup_queue(lo->lo_queue);
1683 out_free_idr:
1684 	idr_remove(&loop_index_idr, i);
1685 out_free_dev:
1686 	kfree(lo);
1687 out:
1688 	return err;
1689 }
1690 
1691 static void loop_remove(struct loop_device *lo)
1692 {
1693 	del_gendisk(lo->lo_disk);
1694 	blk_cleanup_queue(lo->lo_queue);
1695 	put_disk(lo->lo_disk);
1696 	kfree(lo);
1697 }
1698 
1699 static int find_free_cb(int id, void *ptr, void *data)
1700 {
1701 	struct loop_device *lo = ptr;
1702 	struct loop_device **l = data;
1703 
1704 	if (lo->lo_state == Lo_unbound) {
1705 		*l = lo;
1706 		return 1;
1707 	}
1708 	return 0;
1709 }
1710 
1711 static int loop_lookup(struct loop_device **l, int i)
1712 {
1713 	struct loop_device *lo;
1714 	int ret = -ENODEV;
1715 
1716 	if (i < 0) {
1717 		int err;
1718 
1719 		err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1720 		if (err == 1) {
1721 			*l = lo;
1722 			ret = lo->lo_number;
1723 		}
1724 		goto out;
1725 	}
1726 
1727 	/* lookup and return a specific i */
1728 	lo = idr_find(&loop_index_idr, i);
1729 	if (lo) {
1730 		*l = lo;
1731 		ret = lo->lo_number;
1732 	}
1733 out:
1734 	return ret;
1735 }
1736 
1737 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1738 {
1739 	struct loop_device *lo;
1740 	struct kobject *kobj;
1741 	int err;
1742 
1743 	mutex_lock(&loop_index_mutex);
1744 	err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1745 	if (err < 0)
1746 		err = loop_add(&lo, MINOR(dev) >> part_shift);
1747 	if (err < 0)
1748 		kobj = NULL;
1749 	else
1750 		kobj = get_disk(lo->lo_disk);
1751 	mutex_unlock(&loop_index_mutex);
1752 
1753 	*part = 0;
1754 	return kobj;
1755 }
1756 
1757 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1758 			       unsigned long parm)
1759 {
1760 	struct loop_device *lo;
1761 	int ret = -ENOSYS;
1762 
1763 	mutex_lock(&loop_index_mutex);
1764 	switch (cmd) {
1765 	case LOOP_CTL_ADD:
1766 		ret = loop_lookup(&lo, parm);
1767 		if (ret >= 0) {
1768 			ret = -EEXIST;
1769 			break;
1770 		}
1771 		ret = loop_add(&lo, parm);
1772 		break;
1773 	case LOOP_CTL_REMOVE:
1774 		ret = loop_lookup(&lo, parm);
1775 		if (ret < 0)
1776 			break;
1777 		mutex_lock(&lo->lo_ctl_mutex);
1778 		if (lo->lo_state != Lo_unbound) {
1779 			ret = -EBUSY;
1780 			mutex_unlock(&lo->lo_ctl_mutex);
1781 			break;
1782 		}
1783 		if (lo->lo_refcnt > 0) {
1784 			ret = -EBUSY;
1785 			mutex_unlock(&lo->lo_ctl_mutex);
1786 			break;
1787 		}
1788 		lo->lo_disk->private_data = NULL;
1789 		mutex_unlock(&lo->lo_ctl_mutex);
1790 		idr_remove(&loop_index_idr, lo->lo_number);
1791 		loop_remove(lo);
1792 		break;
1793 	case LOOP_CTL_GET_FREE:
1794 		ret = loop_lookup(&lo, -1);
1795 		if (ret >= 0)
1796 			break;
1797 		ret = loop_add(&lo, -1);
1798 	}
1799 	mutex_unlock(&loop_index_mutex);
1800 
1801 	return ret;
1802 }
1803 
1804 static const struct file_operations loop_ctl_fops = {
1805 	.open		= nonseekable_open,
1806 	.unlocked_ioctl	= loop_control_ioctl,
1807 	.compat_ioctl	= loop_control_ioctl,
1808 	.owner		= THIS_MODULE,
1809 	.llseek		= noop_llseek,
1810 };
1811 
1812 static struct miscdevice loop_misc = {
1813 	.minor		= LOOP_CTRL_MINOR,
1814 	.name		= "loop-control",
1815 	.fops		= &loop_ctl_fops,
1816 };
1817 
1818 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1819 MODULE_ALIAS("devname:loop-control");
1820 
1821 static int __init loop_init(void)
1822 {
1823 	int i, nr;
1824 	unsigned long range;
1825 	struct loop_device *lo;
1826 	int err;
1827 
1828 	err = misc_register(&loop_misc);
1829 	if (err < 0)
1830 		return err;
1831 
1832 	part_shift = 0;
1833 	if (max_part > 0) {
1834 		part_shift = fls(max_part);
1835 
1836 		/*
1837 		 * Adjust max_part according to part_shift as it is exported
1838 		 * to user space so that user can decide correct minor number
1839 		 * if [s]he want to create more devices.
1840 		 *
1841 		 * Note that -1 is required because partition 0 is reserved
1842 		 * for the whole disk.
1843 		 */
1844 		max_part = (1UL << part_shift) - 1;
1845 	}
1846 
1847 	if ((1UL << part_shift) > DISK_MAX_PARTS) {
1848 		err = -EINVAL;
1849 		goto misc_out;
1850 	}
1851 
1852 	if (max_loop > 1UL << (MINORBITS - part_shift)) {
1853 		err = -EINVAL;
1854 		goto misc_out;
1855 	}
1856 
1857 	/*
1858 	 * If max_loop is specified, create that many devices upfront.
1859 	 * This also becomes a hard limit. If max_loop is not specified,
1860 	 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1861 	 * init time. Loop devices can be requested on-demand with the
1862 	 * /dev/loop-control interface, or be instantiated by accessing
1863 	 * a 'dead' device node.
1864 	 */
1865 	if (max_loop) {
1866 		nr = max_loop;
1867 		range = max_loop << part_shift;
1868 	} else {
1869 		nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1870 		range = 1UL << MINORBITS;
1871 	}
1872 
1873 	if (register_blkdev(LOOP_MAJOR, "loop")) {
1874 		err = -EIO;
1875 		goto misc_out;
1876 	}
1877 
1878 	blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1879 				  THIS_MODULE, loop_probe, NULL, NULL);
1880 
1881 	/* pre-create number of devices given by config or max_loop */
1882 	mutex_lock(&loop_index_mutex);
1883 	for (i = 0; i < nr; i++)
1884 		loop_add(&lo, i);
1885 	mutex_unlock(&loop_index_mutex);
1886 
1887 	printk(KERN_INFO "loop: module loaded\n");
1888 	return 0;
1889 
1890 misc_out:
1891 	misc_deregister(&loop_misc);
1892 	return err;
1893 }
1894 
1895 static int loop_exit_cb(int id, void *ptr, void *data)
1896 {
1897 	struct loop_device *lo = ptr;
1898 
1899 	loop_remove(lo);
1900 	return 0;
1901 }
1902 
1903 static void __exit loop_exit(void)
1904 {
1905 	unsigned long range;
1906 
1907 	range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1908 
1909 	idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1910 	idr_destroy(&loop_index_idr);
1911 
1912 	blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1913 	unregister_blkdev(LOOP_MAJOR, "loop");
1914 
1915 	misc_deregister(&loop_misc);
1916 }
1917 
1918 module_init(loop_init);
1919 module_exit(loop_exit);
1920 
1921 #ifndef MODULE
1922 static int __init max_loop_setup(char *str)
1923 {
1924 	max_loop = simple_strtol(str, NULL, 0);
1925 	return 1;
1926 }
1927 
1928 __setup("max_loop=", max_loop_setup);
1929 #endif
1930