xref: /openbmc/linux/drivers/block/loop.c (revision a36954f5)
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 <linux/uio.h>
79 #include "loop.h"
80 
81 #include <linux/uaccess.h>
82 
83 static DEFINE_IDR(loop_index_idr);
84 static DEFINE_MUTEX(loop_index_mutex);
85 
86 static int max_part;
87 static int part_shift;
88 
89 static int transfer_xor(struct loop_device *lo, int cmd,
90 			struct page *raw_page, unsigned raw_off,
91 			struct page *loop_page, unsigned loop_off,
92 			int size, sector_t real_block)
93 {
94 	char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 	char *loop_buf = kmap_atomic(loop_page) + loop_off;
96 	char *in, *out, *key;
97 	int i, keysize;
98 
99 	if (cmd == READ) {
100 		in = raw_buf;
101 		out = loop_buf;
102 	} else {
103 		in = loop_buf;
104 		out = raw_buf;
105 	}
106 
107 	key = lo->lo_encrypt_key;
108 	keysize = lo->lo_encrypt_key_size;
109 	for (i = 0; i < size; i++)
110 		*out++ = *in++ ^ key[(i & 511) % keysize];
111 
112 	kunmap_atomic(loop_buf);
113 	kunmap_atomic(raw_buf);
114 	cond_resched();
115 	return 0;
116 }
117 
118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
119 {
120 	if (unlikely(info->lo_encrypt_key_size <= 0))
121 		return -EINVAL;
122 	return 0;
123 }
124 
125 static struct loop_func_table none_funcs = {
126 	.number = LO_CRYPT_NONE,
127 };
128 
129 static struct loop_func_table xor_funcs = {
130 	.number = LO_CRYPT_XOR,
131 	.transfer = transfer_xor,
132 	.init = xor_init
133 };
134 
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
137 	&none_funcs,
138 	&xor_funcs
139 };
140 
141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
142 {
143 	loff_t loopsize;
144 
145 	/* Compute loopsize in bytes */
146 	loopsize = i_size_read(file->f_mapping->host);
147 	if (offset > 0)
148 		loopsize -= offset;
149 	/* offset is beyond i_size, weird but possible */
150 	if (loopsize < 0)
151 		return 0;
152 
153 	if (sizelimit > 0 && sizelimit < loopsize)
154 		loopsize = sizelimit;
155 	/*
156 	 * Unfortunately, if we want to do I/O on the device,
157 	 * the number of 512-byte sectors has to fit into a sector_t.
158 	 */
159 	return loopsize >> 9;
160 }
161 
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163 {
164 	return get_size(lo->lo_offset, lo->lo_sizelimit, file);
165 }
166 
167 static void __loop_update_dio(struct loop_device *lo, bool dio)
168 {
169 	struct file *file = lo->lo_backing_file;
170 	struct address_space *mapping = file->f_mapping;
171 	struct inode *inode = mapping->host;
172 	unsigned short sb_bsize = 0;
173 	unsigned dio_align = 0;
174 	bool use_dio;
175 
176 	if (inode->i_sb->s_bdev) {
177 		sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 		dio_align = sb_bsize - 1;
179 	}
180 
181 	/*
182 	 * We support direct I/O only if lo_offset is aligned with the
183 	 * logical I/O size of backing device, and the logical block
184 	 * size of loop is bigger than the backing device's and the loop
185 	 * needn't transform transfer.
186 	 *
187 	 * TODO: the above condition may be loosed in the future, and
188 	 * direct I/O may be switched runtime at that time because most
189 	 * of requests in sane applications should be PAGE_SIZE aligned
190 	 */
191 	if (dio) {
192 		if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 				!(lo->lo_offset & dio_align) &&
194 				mapping->a_ops->direct_IO &&
195 				!lo->transfer)
196 			use_dio = true;
197 		else
198 			use_dio = false;
199 	} else {
200 		use_dio = false;
201 	}
202 
203 	if (lo->use_dio == use_dio)
204 		return;
205 
206 	/* flush dirty pages before changing direct IO */
207 	vfs_fsync(file, 0);
208 
209 	/*
210 	 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 	 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 	 * will get updated by ioctl(LOOP_GET_STATUS)
213 	 */
214 	blk_mq_freeze_queue(lo->lo_queue);
215 	lo->use_dio = use_dio;
216 	if (use_dio)
217 		lo->lo_flags |= LO_FLAGS_DIRECT_IO;
218 	else
219 		lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 	blk_mq_unfreeze_queue(lo->lo_queue);
221 }
222 
223 static int
224 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
225 {
226 	loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
227 	sector_t x = (sector_t)size;
228 	struct block_device *bdev = lo->lo_device;
229 
230 	if (unlikely((loff_t)x != size))
231 		return -EFBIG;
232 	if (lo->lo_offset != offset)
233 		lo->lo_offset = offset;
234 	if (lo->lo_sizelimit != sizelimit)
235 		lo->lo_sizelimit = sizelimit;
236 	set_capacity(lo->lo_disk, x);
237 	bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
238 	/* let user-space know about the new size */
239 	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
240 	return 0;
241 }
242 
243 static inline int
244 lo_do_transfer(struct loop_device *lo, int cmd,
245 	       struct page *rpage, unsigned roffs,
246 	       struct page *lpage, unsigned loffs,
247 	       int size, sector_t rblock)
248 {
249 	int ret;
250 
251 	ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
252 	if (likely(!ret))
253 		return 0;
254 
255 	printk_ratelimited(KERN_ERR
256 		"loop: Transfer error at byte offset %llu, length %i.\n",
257 		(unsigned long long)rblock << 9, size);
258 	return ret;
259 }
260 
261 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
262 {
263 	struct iov_iter i;
264 	ssize_t bw;
265 
266 	iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
267 
268 	file_start_write(file);
269 	bw = vfs_iter_write(file, &i, ppos);
270 	file_end_write(file);
271 
272 	if (likely(bw ==  bvec->bv_len))
273 		return 0;
274 
275 	printk_ratelimited(KERN_ERR
276 		"loop: Write error at byte offset %llu, length %i.\n",
277 		(unsigned long long)*ppos, bvec->bv_len);
278 	if (bw >= 0)
279 		bw = -EIO;
280 	return bw;
281 }
282 
283 static int lo_write_simple(struct loop_device *lo, struct request *rq,
284 		loff_t pos)
285 {
286 	struct bio_vec bvec;
287 	struct req_iterator iter;
288 	int ret = 0;
289 
290 	rq_for_each_segment(bvec, rq, iter) {
291 		ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
292 		if (ret < 0)
293 			break;
294 		cond_resched();
295 	}
296 
297 	return ret;
298 }
299 
300 /*
301  * This is the slow, transforming version that needs to double buffer the
302  * data as it cannot do the transformations in place without having direct
303  * access to the destination pages of the backing file.
304  */
305 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
306 		loff_t pos)
307 {
308 	struct bio_vec bvec, b;
309 	struct req_iterator iter;
310 	struct page *page;
311 	int ret = 0;
312 
313 	page = alloc_page(GFP_NOIO);
314 	if (unlikely(!page))
315 		return -ENOMEM;
316 
317 	rq_for_each_segment(bvec, rq, iter) {
318 		ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
319 			bvec.bv_offset, bvec.bv_len, pos >> 9);
320 		if (unlikely(ret))
321 			break;
322 
323 		b.bv_page = page;
324 		b.bv_offset = 0;
325 		b.bv_len = bvec.bv_len;
326 		ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
327 		if (ret < 0)
328 			break;
329 	}
330 
331 	__free_page(page);
332 	return ret;
333 }
334 
335 static int lo_read_simple(struct loop_device *lo, struct request *rq,
336 		loff_t pos)
337 {
338 	struct bio_vec bvec;
339 	struct req_iterator iter;
340 	struct iov_iter i;
341 	ssize_t len;
342 
343 	rq_for_each_segment(bvec, rq, iter) {
344 		iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
345 		len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
346 		if (len < 0)
347 			return len;
348 
349 		flush_dcache_page(bvec.bv_page);
350 
351 		if (len != bvec.bv_len) {
352 			struct bio *bio;
353 
354 			__rq_for_each_bio(bio, rq)
355 				zero_fill_bio(bio);
356 			break;
357 		}
358 		cond_resched();
359 	}
360 
361 	return 0;
362 }
363 
364 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
365 		loff_t pos)
366 {
367 	struct bio_vec bvec, b;
368 	struct req_iterator iter;
369 	struct iov_iter i;
370 	struct page *page;
371 	ssize_t len;
372 	int ret = 0;
373 
374 	page = alloc_page(GFP_NOIO);
375 	if (unlikely(!page))
376 		return -ENOMEM;
377 
378 	rq_for_each_segment(bvec, rq, iter) {
379 		loff_t offset = pos;
380 
381 		b.bv_page = page;
382 		b.bv_offset = 0;
383 		b.bv_len = bvec.bv_len;
384 
385 		iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
386 		len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
387 		if (len < 0) {
388 			ret = len;
389 			goto out_free_page;
390 		}
391 
392 		ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
393 			bvec.bv_offset, len, offset >> 9);
394 		if (ret)
395 			goto out_free_page;
396 
397 		flush_dcache_page(bvec.bv_page);
398 
399 		if (len != bvec.bv_len) {
400 			struct bio *bio;
401 
402 			__rq_for_each_bio(bio, rq)
403 				zero_fill_bio(bio);
404 			break;
405 		}
406 	}
407 
408 	ret = 0;
409 out_free_page:
410 	__free_page(page);
411 	return ret;
412 }
413 
414 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
415 {
416 	/*
417 	 * We use punch hole to reclaim the free space used by the
418 	 * image a.k.a. discard. However we do not support discard if
419 	 * encryption is enabled, because it may give an attacker
420 	 * useful information.
421 	 */
422 	struct file *file = lo->lo_backing_file;
423 	int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
424 	int ret;
425 
426 	if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
427 		ret = -EOPNOTSUPP;
428 		goto out;
429 	}
430 
431 	ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
432 	if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
433 		ret = -EIO;
434  out:
435 	return ret;
436 }
437 
438 static int lo_req_flush(struct loop_device *lo, struct request *rq)
439 {
440 	struct file *file = lo->lo_backing_file;
441 	int ret = vfs_fsync(file, 0);
442 	if (unlikely(ret && ret != -EINVAL))
443 		ret = -EIO;
444 
445 	return ret;
446 }
447 
448 static void lo_complete_rq(struct request *rq)
449 {
450 	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
451 
452 	if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio &&
453 		     cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) {
454 		struct bio *bio = cmd->rq->bio;
455 
456 		bio_advance(bio, cmd->ret);
457 		zero_fill_bio(bio);
458 	}
459 
460 	blk_mq_end_request(rq, cmd->ret < 0 ? -EIO : 0);
461 }
462 
463 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
464 {
465 	struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
466 
467 	cmd->ret = ret;
468 	blk_mq_complete_request(cmd->rq);
469 }
470 
471 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
472 		     loff_t pos, bool rw)
473 {
474 	struct iov_iter iter;
475 	struct bio_vec *bvec;
476 	struct bio *bio = cmd->rq->bio;
477 	struct file *file = lo->lo_backing_file;
478 	int ret;
479 
480 	/* nomerge for loop request queue */
481 	WARN_ON(cmd->rq->bio != cmd->rq->biotail);
482 
483 	bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
484 	iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
485 		      bio_segments(bio), blk_rq_bytes(cmd->rq));
486 	/*
487 	 * This bio may be started from the middle of the 'bvec'
488 	 * because of bio splitting, so offset from the bvec must
489 	 * be passed to iov iterator
490 	 */
491 	iter.iov_offset = bio->bi_iter.bi_bvec_done;
492 
493 	cmd->iocb.ki_pos = pos;
494 	cmd->iocb.ki_filp = file;
495 	cmd->iocb.ki_complete = lo_rw_aio_complete;
496 	cmd->iocb.ki_flags = IOCB_DIRECT;
497 
498 	if (rw == WRITE)
499 		ret = call_write_iter(file, &cmd->iocb, &iter);
500 	else
501 		ret = call_read_iter(file, &cmd->iocb, &iter);
502 
503 	if (ret != -EIOCBQUEUED)
504 		cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
505 	return 0;
506 }
507 
508 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
509 {
510 	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
511 	loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
512 
513 	/*
514 	 * lo_write_simple and lo_read_simple should have been covered
515 	 * by io submit style function like lo_rw_aio(), one blocker
516 	 * is that lo_read_simple() need to call flush_dcache_page after
517 	 * the page is written from kernel, and it isn't easy to handle
518 	 * this in io submit style function which submits all segments
519 	 * of the req at one time. And direct read IO doesn't need to
520 	 * run flush_dcache_page().
521 	 */
522 	switch (req_op(rq)) {
523 	case REQ_OP_FLUSH:
524 		return lo_req_flush(lo, rq);
525 	case REQ_OP_DISCARD:
526 	case REQ_OP_WRITE_ZEROES:
527 		return lo_discard(lo, rq, pos);
528 	case REQ_OP_WRITE:
529 		if (lo->transfer)
530 			return lo_write_transfer(lo, rq, pos);
531 		else if (cmd->use_aio)
532 			return lo_rw_aio(lo, cmd, pos, WRITE);
533 		else
534 			return lo_write_simple(lo, rq, pos);
535 	case REQ_OP_READ:
536 		if (lo->transfer)
537 			return lo_read_transfer(lo, rq, pos);
538 		else if (cmd->use_aio)
539 			return lo_rw_aio(lo, cmd, pos, READ);
540 		else
541 			return lo_read_simple(lo, rq, pos);
542 	default:
543 		WARN_ON_ONCE(1);
544 		return -EIO;
545 		break;
546 	}
547 }
548 
549 struct switch_request {
550 	struct file *file;
551 	struct completion wait;
552 };
553 
554 static inline void loop_update_dio(struct loop_device *lo)
555 {
556 	__loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
557 			lo->use_dio);
558 }
559 
560 /*
561  * Do the actual switch; called from the BIO completion routine
562  */
563 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
564 {
565 	struct file *file = p->file;
566 	struct file *old_file = lo->lo_backing_file;
567 	struct address_space *mapping;
568 
569 	/* if no new file, only flush of queued bios requested */
570 	if (!file)
571 		return;
572 
573 	mapping = file->f_mapping;
574 	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
575 	lo->lo_backing_file = file;
576 	lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
577 		mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
578 	lo->old_gfp_mask = mapping_gfp_mask(mapping);
579 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
580 	loop_update_dio(lo);
581 }
582 
583 /*
584  * loop_switch performs the hard work of switching a backing store.
585  * First it needs to flush existing IO, it does this by sending a magic
586  * BIO down the pipe. The completion of this BIO does the actual switch.
587  */
588 static int loop_switch(struct loop_device *lo, struct file *file)
589 {
590 	struct switch_request w;
591 
592 	w.file = file;
593 
594 	/* freeze queue and wait for completion of scheduled requests */
595 	blk_mq_freeze_queue(lo->lo_queue);
596 
597 	/* do the switch action */
598 	do_loop_switch(lo, &w);
599 
600 	/* unfreeze */
601 	blk_mq_unfreeze_queue(lo->lo_queue);
602 
603 	return 0;
604 }
605 
606 /*
607  * Helper to flush the IOs in loop, but keeping loop thread running
608  */
609 static int loop_flush(struct loop_device *lo)
610 {
611 	return loop_switch(lo, NULL);
612 }
613 
614 static void loop_reread_partitions(struct loop_device *lo,
615 				   struct block_device *bdev)
616 {
617 	int rc;
618 
619 	/*
620 	 * bd_mutex has been held already in release path, so don't
621 	 * acquire it if this function is called in such case.
622 	 *
623 	 * If the reread partition isn't from release path, lo_refcnt
624 	 * must be at least one and it can only become zero when the
625 	 * current holder is released.
626 	 */
627 	if (!atomic_read(&lo->lo_refcnt))
628 		rc = __blkdev_reread_part(bdev);
629 	else
630 		rc = blkdev_reread_part(bdev);
631 	if (rc)
632 		pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
633 			__func__, lo->lo_number, lo->lo_file_name, rc);
634 }
635 
636 /*
637  * loop_change_fd switched the backing store of a loopback device to
638  * a new file. This is useful for operating system installers to free up
639  * the original file and in High Availability environments to switch to
640  * an alternative location for the content in case of server meltdown.
641  * This can only work if the loop device is used read-only, and if the
642  * new backing store is the same size and type as the old backing store.
643  */
644 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
645 			  unsigned int arg)
646 {
647 	struct file	*file, *old_file;
648 	struct inode	*inode;
649 	int		error;
650 
651 	error = -ENXIO;
652 	if (lo->lo_state != Lo_bound)
653 		goto out;
654 
655 	/* the loop device has to be read-only */
656 	error = -EINVAL;
657 	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
658 		goto out;
659 
660 	error = -EBADF;
661 	file = fget(arg);
662 	if (!file)
663 		goto out;
664 
665 	inode = file->f_mapping->host;
666 	old_file = lo->lo_backing_file;
667 
668 	error = -EINVAL;
669 
670 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
671 		goto out_putf;
672 
673 	/* size of the new backing store needs to be the same */
674 	if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
675 		goto out_putf;
676 
677 	/* and ... switch */
678 	error = loop_switch(lo, file);
679 	if (error)
680 		goto out_putf;
681 
682 	fput(old_file);
683 	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
684 		loop_reread_partitions(lo, bdev);
685 	return 0;
686 
687  out_putf:
688 	fput(file);
689  out:
690 	return error;
691 }
692 
693 static inline int is_loop_device(struct file *file)
694 {
695 	struct inode *i = file->f_mapping->host;
696 
697 	return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
698 }
699 
700 /* loop sysfs attributes */
701 
702 static ssize_t loop_attr_show(struct device *dev, char *page,
703 			      ssize_t (*callback)(struct loop_device *, char *))
704 {
705 	struct gendisk *disk = dev_to_disk(dev);
706 	struct loop_device *lo = disk->private_data;
707 
708 	return callback(lo, page);
709 }
710 
711 #define LOOP_ATTR_RO(_name)						\
712 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);	\
713 static ssize_t loop_attr_do_show_##_name(struct device *d,		\
714 				struct device_attribute *attr, char *b)	\
715 {									\
716 	return loop_attr_show(d, b, loop_attr_##_name##_show);		\
717 }									\
718 static struct device_attribute loop_attr_##_name =			\
719 	__ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
720 
721 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
722 {
723 	ssize_t ret;
724 	char *p = NULL;
725 
726 	spin_lock_irq(&lo->lo_lock);
727 	if (lo->lo_backing_file)
728 		p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
729 	spin_unlock_irq(&lo->lo_lock);
730 
731 	if (IS_ERR_OR_NULL(p))
732 		ret = PTR_ERR(p);
733 	else {
734 		ret = strlen(p);
735 		memmove(buf, p, ret);
736 		buf[ret++] = '\n';
737 		buf[ret] = 0;
738 	}
739 
740 	return ret;
741 }
742 
743 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
744 {
745 	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
746 }
747 
748 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
749 {
750 	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
751 }
752 
753 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
754 {
755 	int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
756 
757 	return sprintf(buf, "%s\n", autoclear ? "1" : "0");
758 }
759 
760 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
761 {
762 	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
763 
764 	return sprintf(buf, "%s\n", partscan ? "1" : "0");
765 }
766 
767 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
768 {
769 	int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
770 
771 	return sprintf(buf, "%s\n", dio ? "1" : "0");
772 }
773 
774 LOOP_ATTR_RO(backing_file);
775 LOOP_ATTR_RO(offset);
776 LOOP_ATTR_RO(sizelimit);
777 LOOP_ATTR_RO(autoclear);
778 LOOP_ATTR_RO(partscan);
779 LOOP_ATTR_RO(dio);
780 
781 static struct attribute *loop_attrs[] = {
782 	&loop_attr_backing_file.attr,
783 	&loop_attr_offset.attr,
784 	&loop_attr_sizelimit.attr,
785 	&loop_attr_autoclear.attr,
786 	&loop_attr_partscan.attr,
787 	&loop_attr_dio.attr,
788 	NULL,
789 };
790 
791 static struct attribute_group loop_attribute_group = {
792 	.name = "loop",
793 	.attrs= loop_attrs,
794 };
795 
796 static int loop_sysfs_init(struct loop_device *lo)
797 {
798 	return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
799 				  &loop_attribute_group);
800 }
801 
802 static void loop_sysfs_exit(struct loop_device *lo)
803 {
804 	sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
805 			   &loop_attribute_group);
806 }
807 
808 static void loop_config_discard(struct loop_device *lo)
809 {
810 	struct file *file = lo->lo_backing_file;
811 	struct inode *inode = file->f_mapping->host;
812 	struct request_queue *q = lo->lo_queue;
813 
814 	/*
815 	 * We use punch hole to reclaim the free space used by the
816 	 * image a.k.a. discard. However we do not support discard if
817 	 * encryption is enabled, because it may give an attacker
818 	 * useful information.
819 	 */
820 	if ((!file->f_op->fallocate) ||
821 	    lo->lo_encrypt_key_size) {
822 		q->limits.discard_granularity = 0;
823 		q->limits.discard_alignment = 0;
824 		blk_queue_max_discard_sectors(q, 0);
825 		blk_queue_max_write_zeroes_sectors(q, 0);
826 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
827 		return;
828 	}
829 
830 	q->limits.discard_granularity = inode->i_sb->s_blocksize;
831 	q->limits.discard_alignment = 0;
832 	blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
833 	blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
834 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
835 }
836 
837 static void loop_unprepare_queue(struct loop_device *lo)
838 {
839 	kthread_flush_worker(&lo->worker);
840 	kthread_stop(lo->worker_task);
841 }
842 
843 static int loop_prepare_queue(struct loop_device *lo)
844 {
845 	kthread_init_worker(&lo->worker);
846 	lo->worker_task = kthread_run(kthread_worker_fn,
847 			&lo->worker, "loop%d", lo->lo_number);
848 	if (IS_ERR(lo->worker_task))
849 		return -ENOMEM;
850 	set_user_nice(lo->worker_task, MIN_NICE);
851 	return 0;
852 }
853 
854 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
855 		       struct block_device *bdev, unsigned int arg)
856 {
857 	struct file	*file, *f;
858 	struct inode	*inode;
859 	struct address_space *mapping;
860 	unsigned lo_blocksize;
861 	int		lo_flags = 0;
862 	int		error;
863 	loff_t		size;
864 
865 	/* This is safe, since we have a reference from open(). */
866 	__module_get(THIS_MODULE);
867 
868 	error = -EBADF;
869 	file = fget(arg);
870 	if (!file)
871 		goto out;
872 
873 	error = -EBUSY;
874 	if (lo->lo_state != Lo_unbound)
875 		goto out_putf;
876 
877 	/* Avoid recursion */
878 	f = file;
879 	while (is_loop_device(f)) {
880 		struct loop_device *l;
881 
882 		if (f->f_mapping->host->i_bdev == bdev)
883 			goto out_putf;
884 
885 		l = f->f_mapping->host->i_bdev->bd_disk->private_data;
886 		if (l->lo_state == Lo_unbound) {
887 			error = -EINVAL;
888 			goto out_putf;
889 		}
890 		f = l->lo_backing_file;
891 	}
892 
893 	mapping = file->f_mapping;
894 	inode = mapping->host;
895 
896 	error = -EINVAL;
897 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
898 		goto out_putf;
899 
900 	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
901 	    !file->f_op->write_iter)
902 		lo_flags |= LO_FLAGS_READ_ONLY;
903 
904 	lo_blocksize = S_ISBLK(inode->i_mode) ?
905 		inode->i_bdev->bd_block_size : PAGE_SIZE;
906 
907 	error = -EFBIG;
908 	size = get_loop_size(lo, file);
909 	if ((loff_t)(sector_t)size != size)
910 		goto out_putf;
911 	error = loop_prepare_queue(lo);
912 	if (error)
913 		goto out_putf;
914 
915 	error = 0;
916 
917 	set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
918 
919 	lo->use_dio = false;
920 	lo->lo_blocksize = lo_blocksize;
921 	lo->lo_device = bdev;
922 	lo->lo_flags = lo_flags;
923 	lo->lo_backing_file = file;
924 	lo->transfer = NULL;
925 	lo->ioctl = NULL;
926 	lo->lo_sizelimit = 0;
927 	lo->old_gfp_mask = mapping_gfp_mask(mapping);
928 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
929 
930 	if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
931 		blk_queue_write_cache(lo->lo_queue, true, false);
932 
933 	loop_update_dio(lo);
934 	set_capacity(lo->lo_disk, size);
935 	bd_set_size(bdev, size << 9);
936 	loop_sysfs_init(lo);
937 	/* let user-space know about the new size */
938 	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
939 
940 	set_blocksize(bdev, lo_blocksize);
941 
942 	lo->lo_state = Lo_bound;
943 	if (part_shift)
944 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
945 	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
946 		loop_reread_partitions(lo, bdev);
947 
948 	/* Grab the block_device to prevent its destruction after we
949 	 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
950 	 */
951 	bdgrab(bdev);
952 	return 0;
953 
954  out_putf:
955 	fput(file);
956  out:
957 	/* This is safe: open() is still holding a reference. */
958 	module_put(THIS_MODULE);
959 	return error;
960 }
961 
962 static int
963 loop_release_xfer(struct loop_device *lo)
964 {
965 	int err = 0;
966 	struct loop_func_table *xfer = lo->lo_encryption;
967 
968 	if (xfer) {
969 		if (xfer->release)
970 			err = xfer->release(lo);
971 		lo->transfer = NULL;
972 		lo->lo_encryption = NULL;
973 		module_put(xfer->owner);
974 	}
975 	return err;
976 }
977 
978 static int
979 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
980 	       const struct loop_info64 *i)
981 {
982 	int err = 0;
983 
984 	if (xfer) {
985 		struct module *owner = xfer->owner;
986 
987 		if (!try_module_get(owner))
988 			return -EINVAL;
989 		if (xfer->init)
990 			err = xfer->init(lo, i);
991 		if (err)
992 			module_put(owner);
993 		else
994 			lo->lo_encryption = xfer;
995 	}
996 	return err;
997 }
998 
999 static int loop_clr_fd(struct loop_device *lo)
1000 {
1001 	struct file *filp = lo->lo_backing_file;
1002 	gfp_t gfp = lo->old_gfp_mask;
1003 	struct block_device *bdev = lo->lo_device;
1004 
1005 	if (lo->lo_state != Lo_bound)
1006 		return -ENXIO;
1007 
1008 	/*
1009 	 * If we've explicitly asked to tear down the loop device,
1010 	 * and it has an elevated reference count, set it for auto-teardown when
1011 	 * the last reference goes away. This stops $!~#$@ udev from
1012 	 * preventing teardown because it decided that it needs to run blkid on
1013 	 * the loopback device whenever they appear. xfstests is notorious for
1014 	 * failing tests because blkid via udev races with a losetup
1015 	 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1016 	 * command to fail with EBUSY.
1017 	 */
1018 	if (atomic_read(&lo->lo_refcnt) > 1) {
1019 		lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1020 		mutex_unlock(&lo->lo_ctl_mutex);
1021 		return 0;
1022 	}
1023 
1024 	if (filp == NULL)
1025 		return -EINVAL;
1026 
1027 	/* freeze request queue during the transition */
1028 	blk_mq_freeze_queue(lo->lo_queue);
1029 
1030 	spin_lock_irq(&lo->lo_lock);
1031 	lo->lo_state = Lo_rundown;
1032 	lo->lo_backing_file = NULL;
1033 	spin_unlock_irq(&lo->lo_lock);
1034 
1035 	loop_release_xfer(lo);
1036 	lo->transfer = NULL;
1037 	lo->ioctl = NULL;
1038 	lo->lo_device = NULL;
1039 	lo->lo_encryption = NULL;
1040 	lo->lo_offset = 0;
1041 	lo->lo_sizelimit = 0;
1042 	lo->lo_encrypt_key_size = 0;
1043 	memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1044 	memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1045 	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1046 	if (bdev) {
1047 		bdput(bdev);
1048 		invalidate_bdev(bdev);
1049 	}
1050 	set_capacity(lo->lo_disk, 0);
1051 	loop_sysfs_exit(lo);
1052 	if (bdev) {
1053 		bd_set_size(bdev, 0);
1054 		/* let user-space know about this change */
1055 		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1056 	}
1057 	mapping_set_gfp_mask(filp->f_mapping, gfp);
1058 	lo->lo_state = Lo_unbound;
1059 	/* This is safe: open() is still holding a reference. */
1060 	module_put(THIS_MODULE);
1061 	blk_mq_unfreeze_queue(lo->lo_queue);
1062 
1063 	if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1064 		loop_reread_partitions(lo, bdev);
1065 	lo->lo_flags = 0;
1066 	if (!part_shift)
1067 		lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1068 	loop_unprepare_queue(lo);
1069 	mutex_unlock(&lo->lo_ctl_mutex);
1070 	/*
1071 	 * Need not hold lo_ctl_mutex to fput backing file.
1072 	 * Calling fput holding lo_ctl_mutex triggers a circular
1073 	 * lock dependency possibility warning as fput can take
1074 	 * bd_mutex which is usually taken before lo_ctl_mutex.
1075 	 */
1076 	fput(filp);
1077 	return 0;
1078 }
1079 
1080 static int
1081 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1082 {
1083 	int err;
1084 	struct loop_func_table *xfer;
1085 	kuid_t uid = current_uid();
1086 
1087 	if (lo->lo_encrypt_key_size &&
1088 	    !uid_eq(lo->lo_key_owner, uid) &&
1089 	    !capable(CAP_SYS_ADMIN))
1090 		return -EPERM;
1091 	if (lo->lo_state != Lo_bound)
1092 		return -ENXIO;
1093 	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1094 		return -EINVAL;
1095 
1096 	/* I/O need to be drained during transfer transition */
1097 	blk_mq_freeze_queue(lo->lo_queue);
1098 
1099 	err = loop_release_xfer(lo);
1100 	if (err)
1101 		goto exit;
1102 
1103 	if (info->lo_encrypt_type) {
1104 		unsigned int type = info->lo_encrypt_type;
1105 
1106 		if (type >= MAX_LO_CRYPT)
1107 			return -EINVAL;
1108 		xfer = xfer_funcs[type];
1109 		if (xfer == NULL)
1110 			return -EINVAL;
1111 	} else
1112 		xfer = NULL;
1113 
1114 	err = loop_init_xfer(lo, xfer, info);
1115 	if (err)
1116 		goto exit;
1117 
1118 	if (lo->lo_offset != info->lo_offset ||
1119 	    lo->lo_sizelimit != info->lo_sizelimit)
1120 		if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1121 			err = -EFBIG;
1122 			goto exit;
1123 		}
1124 
1125 	loop_config_discard(lo);
1126 
1127 	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1128 	memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1129 	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1130 	lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1131 
1132 	if (!xfer)
1133 		xfer = &none_funcs;
1134 	lo->transfer = xfer->transfer;
1135 	lo->ioctl = xfer->ioctl;
1136 
1137 	if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1138 	     (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1139 		lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1140 
1141 	lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1142 	lo->lo_init[0] = info->lo_init[0];
1143 	lo->lo_init[1] = info->lo_init[1];
1144 	if (info->lo_encrypt_key_size) {
1145 		memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1146 		       info->lo_encrypt_key_size);
1147 		lo->lo_key_owner = uid;
1148 	}
1149 
1150 	/* update dio if lo_offset or transfer is changed */
1151 	__loop_update_dio(lo, lo->use_dio);
1152 
1153  exit:
1154 	blk_mq_unfreeze_queue(lo->lo_queue);
1155 
1156 	if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1157 	     !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1158 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1159 		lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1160 		loop_reread_partitions(lo, lo->lo_device);
1161 	}
1162 
1163 	return err;
1164 }
1165 
1166 static int
1167 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1168 {
1169 	struct file *file = lo->lo_backing_file;
1170 	struct kstat stat;
1171 	int error;
1172 
1173 	if (lo->lo_state != Lo_bound)
1174 		return -ENXIO;
1175 	error = vfs_getattr(&file->f_path, &stat,
1176 			    STATX_INO, AT_STATX_SYNC_AS_STAT);
1177 	if (error)
1178 		return error;
1179 	memset(info, 0, sizeof(*info));
1180 	info->lo_number = lo->lo_number;
1181 	info->lo_device = huge_encode_dev(stat.dev);
1182 	info->lo_inode = stat.ino;
1183 	info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1184 	info->lo_offset = lo->lo_offset;
1185 	info->lo_sizelimit = lo->lo_sizelimit;
1186 	info->lo_flags = lo->lo_flags;
1187 	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1188 	memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1189 	info->lo_encrypt_type =
1190 		lo->lo_encryption ? lo->lo_encryption->number : 0;
1191 	if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1192 		info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1193 		memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1194 		       lo->lo_encrypt_key_size);
1195 	}
1196 	return 0;
1197 }
1198 
1199 static void
1200 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1201 {
1202 	memset(info64, 0, sizeof(*info64));
1203 	info64->lo_number = info->lo_number;
1204 	info64->lo_device = info->lo_device;
1205 	info64->lo_inode = info->lo_inode;
1206 	info64->lo_rdevice = info->lo_rdevice;
1207 	info64->lo_offset = info->lo_offset;
1208 	info64->lo_sizelimit = 0;
1209 	info64->lo_encrypt_type = info->lo_encrypt_type;
1210 	info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1211 	info64->lo_flags = info->lo_flags;
1212 	info64->lo_init[0] = info->lo_init[0];
1213 	info64->lo_init[1] = info->lo_init[1];
1214 	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1215 		memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1216 	else
1217 		memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1218 	memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1219 }
1220 
1221 static int
1222 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1223 {
1224 	memset(info, 0, sizeof(*info));
1225 	info->lo_number = info64->lo_number;
1226 	info->lo_device = info64->lo_device;
1227 	info->lo_inode = info64->lo_inode;
1228 	info->lo_rdevice = info64->lo_rdevice;
1229 	info->lo_offset = info64->lo_offset;
1230 	info->lo_encrypt_type = info64->lo_encrypt_type;
1231 	info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1232 	info->lo_flags = info64->lo_flags;
1233 	info->lo_init[0] = info64->lo_init[0];
1234 	info->lo_init[1] = info64->lo_init[1];
1235 	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1236 		memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1237 	else
1238 		memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1239 	memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1240 
1241 	/* error in case values were truncated */
1242 	if (info->lo_device != info64->lo_device ||
1243 	    info->lo_rdevice != info64->lo_rdevice ||
1244 	    info->lo_inode != info64->lo_inode ||
1245 	    info->lo_offset != info64->lo_offset)
1246 		return -EOVERFLOW;
1247 
1248 	return 0;
1249 }
1250 
1251 static int
1252 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1253 {
1254 	struct loop_info info;
1255 	struct loop_info64 info64;
1256 
1257 	if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1258 		return -EFAULT;
1259 	loop_info64_from_old(&info, &info64);
1260 	return loop_set_status(lo, &info64);
1261 }
1262 
1263 static int
1264 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1265 {
1266 	struct loop_info64 info64;
1267 
1268 	if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1269 		return -EFAULT;
1270 	return loop_set_status(lo, &info64);
1271 }
1272 
1273 static int
1274 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1275 	struct loop_info info;
1276 	struct loop_info64 info64;
1277 	int err = 0;
1278 
1279 	if (!arg)
1280 		err = -EINVAL;
1281 	if (!err)
1282 		err = loop_get_status(lo, &info64);
1283 	if (!err)
1284 		err = loop_info64_to_old(&info64, &info);
1285 	if (!err && copy_to_user(arg, &info, sizeof(info)))
1286 		err = -EFAULT;
1287 
1288 	return err;
1289 }
1290 
1291 static int
1292 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1293 	struct loop_info64 info64;
1294 	int err = 0;
1295 
1296 	if (!arg)
1297 		err = -EINVAL;
1298 	if (!err)
1299 		err = loop_get_status(lo, &info64);
1300 	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1301 		err = -EFAULT;
1302 
1303 	return err;
1304 }
1305 
1306 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1307 {
1308 	if (unlikely(lo->lo_state != Lo_bound))
1309 		return -ENXIO;
1310 
1311 	return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1312 }
1313 
1314 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1315 {
1316 	int error = -ENXIO;
1317 	if (lo->lo_state != Lo_bound)
1318 		goto out;
1319 
1320 	__loop_update_dio(lo, !!arg);
1321 	if (lo->use_dio == !!arg)
1322 		return 0;
1323 	error = -EINVAL;
1324  out:
1325 	return error;
1326 }
1327 
1328 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1329 	unsigned int cmd, unsigned long arg)
1330 {
1331 	struct loop_device *lo = bdev->bd_disk->private_data;
1332 	int err;
1333 
1334 	mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1335 	switch (cmd) {
1336 	case LOOP_SET_FD:
1337 		err = loop_set_fd(lo, mode, bdev, arg);
1338 		break;
1339 	case LOOP_CHANGE_FD:
1340 		err = loop_change_fd(lo, bdev, arg);
1341 		break;
1342 	case LOOP_CLR_FD:
1343 		/* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1344 		err = loop_clr_fd(lo);
1345 		if (!err)
1346 			goto out_unlocked;
1347 		break;
1348 	case LOOP_SET_STATUS:
1349 		err = -EPERM;
1350 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1351 			err = loop_set_status_old(lo,
1352 					(struct loop_info __user *)arg);
1353 		break;
1354 	case LOOP_GET_STATUS:
1355 		err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1356 		break;
1357 	case LOOP_SET_STATUS64:
1358 		err = -EPERM;
1359 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1360 			err = loop_set_status64(lo,
1361 					(struct loop_info64 __user *) arg);
1362 		break;
1363 	case LOOP_GET_STATUS64:
1364 		err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1365 		break;
1366 	case LOOP_SET_CAPACITY:
1367 		err = -EPERM;
1368 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1369 			err = loop_set_capacity(lo, bdev);
1370 		break;
1371 	case LOOP_SET_DIRECT_IO:
1372 		err = -EPERM;
1373 		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1374 			err = loop_set_dio(lo, arg);
1375 		break;
1376 	default:
1377 		err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1378 	}
1379 	mutex_unlock(&lo->lo_ctl_mutex);
1380 
1381 out_unlocked:
1382 	return err;
1383 }
1384 
1385 #ifdef CONFIG_COMPAT
1386 struct compat_loop_info {
1387 	compat_int_t	lo_number;      /* ioctl r/o */
1388 	compat_dev_t	lo_device;      /* ioctl r/o */
1389 	compat_ulong_t	lo_inode;       /* ioctl r/o */
1390 	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1391 	compat_int_t	lo_offset;
1392 	compat_int_t	lo_encrypt_type;
1393 	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1394 	compat_int_t	lo_flags;       /* ioctl r/o */
1395 	char		lo_name[LO_NAME_SIZE];
1396 	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1397 	compat_ulong_t	lo_init[2];
1398 	char		reserved[4];
1399 };
1400 
1401 /*
1402  * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1403  * - noinlined to reduce stack space usage in main part of driver
1404  */
1405 static noinline int
1406 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1407 			struct loop_info64 *info64)
1408 {
1409 	struct compat_loop_info info;
1410 
1411 	if (copy_from_user(&info, arg, sizeof(info)))
1412 		return -EFAULT;
1413 
1414 	memset(info64, 0, sizeof(*info64));
1415 	info64->lo_number = info.lo_number;
1416 	info64->lo_device = info.lo_device;
1417 	info64->lo_inode = info.lo_inode;
1418 	info64->lo_rdevice = info.lo_rdevice;
1419 	info64->lo_offset = info.lo_offset;
1420 	info64->lo_sizelimit = 0;
1421 	info64->lo_encrypt_type = info.lo_encrypt_type;
1422 	info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1423 	info64->lo_flags = info.lo_flags;
1424 	info64->lo_init[0] = info.lo_init[0];
1425 	info64->lo_init[1] = info.lo_init[1];
1426 	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1427 		memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1428 	else
1429 		memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1430 	memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1431 	return 0;
1432 }
1433 
1434 /*
1435  * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1436  * - noinlined to reduce stack space usage in main part of driver
1437  */
1438 static noinline int
1439 loop_info64_to_compat(const struct loop_info64 *info64,
1440 		      struct compat_loop_info __user *arg)
1441 {
1442 	struct compat_loop_info info;
1443 
1444 	memset(&info, 0, sizeof(info));
1445 	info.lo_number = info64->lo_number;
1446 	info.lo_device = info64->lo_device;
1447 	info.lo_inode = info64->lo_inode;
1448 	info.lo_rdevice = info64->lo_rdevice;
1449 	info.lo_offset = info64->lo_offset;
1450 	info.lo_encrypt_type = info64->lo_encrypt_type;
1451 	info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1452 	info.lo_flags = info64->lo_flags;
1453 	info.lo_init[0] = info64->lo_init[0];
1454 	info.lo_init[1] = info64->lo_init[1];
1455 	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1456 		memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1457 	else
1458 		memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1459 	memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1460 
1461 	/* error in case values were truncated */
1462 	if (info.lo_device != info64->lo_device ||
1463 	    info.lo_rdevice != info64->lo_rdevice ||
1464 	    info.lo_inode != info64->lo_inode ||
1465 	    info.lo_offset != info64->lo_offset ||
1466 	    info.lo_init[0] != info64->lo_init[0] ||
1467 	    info.lo_init[1] != info64->lo_init[1])
1468 		return -EOVERFLOW;
1469 
1470 	if (copy_to_user(arg, &info, sizeof(info)))
1471 		return -EFAULT;
1472 	return 0;
1473 }
1474 
1475 static int
1476 loop_set_status_compat(struct loop_device *lo,
1477 		       const struct compat_loop_info __user *arg)
1478 {
1479 	struct loop_info64 info64;
1480 	int ret;
1481 
1482 	ret = loop_info64_from_compat(arg, &info64);
1483 	if (ret < 0)
1484 		return ret;
1485 	return loop_set_status(lo, &info64);
1486 }
1487 
1488 static int
1489 loop_get_status_compat(struct loop_device *lo,
1490 		       struct compat_loop_info __user *arg)
1491 {
1492 	struct loop_info64 info64;
1493 	int err = 0;
1494 
1495 	if (!arg)
1496 		err = -EINVAL;
1497 	if (!err)
1498 		err = loop_get_status(lo, &info64);
1499 	if (!err)
1500 		err = loop_info64_to_compat(&info64, arg);
1501 	return err;
1502 }
1503 
1504 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1505 			   unsigned int cmd, unsigned long arg)
1506 {
1507 	struct loop_device *lo = bdev->bd_disk->private_data;
1508 	int err;
1509 
1510 	switch(cmd) {
1511 	case LOOP_SET_STATUS:
1512 		mutex_lock(&lo->lo_ctl_mutex);
1513 		err = loop_set_status_compat(
1514 			lo, (const struct compat_loop_info __user *) arg);
1515 		mutex_unlock(&lo->lo_ctl_mutex);
1516 		break;
1517 	case LOOP_GET_STATUS:
1518 		mutex_lock(&lo->lo_ctl_mutex);
1519 		err = loop_get_status_compat(
1520 			lo, (struct compat_loop_info __user *) arg);
1521 		mutex_unlock(&lo->lo_ctl_mutex);
1522 		break;
1523 	case LOOP_SET_CAPACITY:
1524 	case LOOP_CLR_FD:
1525 	case LOOP_GET_STATUS64:
1526 	case LOOP_SET_STATUS64:
1527 		arg = (unsigned long) compat_ptr(arg);
1528 	case LOOP_SET_FD:
1529 	case LOOP_CHANGE_FD:
1530 		err = lo_ioctl(bdev, mode, cmd, arg);
1531 		break;
1532 	default:
1533 		err = -ENOIOCTLCMD;
1534 		break;
1535 	}
1536 	return err;
1537 }
1538 #endif
1539 
1540 static int lo_open(struct block_device *bdev, fmode_t mode)
1541 {
1542 	struct loop_device *lo;
1543 	int err = 0;
1544 
1545 	mutex_lock(&loop_index_mutex);
1546 	lo = bdev->bd_disk->private_data;
1547 	if (!lo) {
1548 		err = -ENXIO;
1549 		goto out;
1550 	}
1551 
1552 	atomic_inc(&lo->lo_refcnt);
1553 out:
1554 	mutex_unlock(&loop_index_mutex);
1555 	return err;
1556 }
1557 
1558 static void lo_release(struct gendisk *disk, fmode_t mode)
1559 {
1560 	struct loop_device *lo = disk->private_data;
1561 	int err;
1562 
1563 	if (atomic_dec_return(&lo->lo_refcnt))
1564 		return;
1565 
1566 	mutex_lock(&lo->lo_ctl_mutex);
1567 	if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1568 		/*
1569 		 * In autoclear mode, stop the loop thread
1570 		 * and remove configuration after last close.
1571 		 */
1572 		err = loop_clr_fd(lo);
1573 		if (!err)
1574 			return;
1575 	} else {
1576 		/*
1577 		 * Otherwise keep thread (if running) and config,
1578 		 * but flush possible ongoing bios in thread.
1579 		 */
1580 		loop_flush(lo);
1581 	}
1582 
1583 	mutex_unlock(&lo->lo_ctl_mutex);
1584 }
1585 
1586 static const struct block_device_operations lo_fops = {
1587 	.owner =	THIS_MODULE,
1588 	.open =		lo_open,
1589 	.release =	lo_release,
1590 	.ioctl =	lo_ioctl,
1591 #ifdef CONFIG_COMPAT
1592 	.compat_ioctl =	lo_compat_ioctl,
1593 #endif
1594 };
1595 
1596 /*
1597  * And now the modules code and kernel interface.
1598  */
1599 static int max_loop;
1600 module_param(max_loop, int, S_IRUGO);
1601 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1602 module_param(max_part, int, S_IRUGO);
1603 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1604 MODULE_LICENSE("GPL");
1605 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1606 
1607 int loop_register_transfer(struct loop_func_table *funcs)
1608 {
1609 	unsigned int n = funcs->number;
1610 
1611 	if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1612 		return -EINVAL;
1613 	xfer_funcs[n] = funcs;
1614 	return 0;
1615 }
1616 
1617 static int unregister_transfer_cb(int id, void *ptr, void *data)
1618 {
1619 	struct loop_device *lo = ptr;
1620 	struct loop_func_table *xfer = data;
1621 
1622 	mutex_lock(&lo->lo_ctl_mutex);
1623 	if (lo->lo_encryption == xfer)
1624 		loop_release_xfer(lo);
1625 	mutex_unlock(&lo->lo_ctl_mutex);
1626 	return 0;
1627 }
1628 
1629 int loop_unregister_transfer(int number)
1630 {
1631 	unsigned int n = number;
1632 	struct loop_func_table *xfer;
1633 
1634 	if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1635 		return -EINVAL;
1636 
1637 	xfer_funcs[n] = NULL;
1638 	idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1639 	return 0;
1640 }
1641 
1642 EXPORT_SYMBOL(loop_register_transfer);
1643 EXPORT_SYMBOL(loop_unregister_transfer);
1644 
1645 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1646 		const struct blk_mq_queue_data *bd)
1647 {
1648 	struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1649 	struct loop_device *lo = cmd->rq->q->queuedata;
1650 
1651 	blk_mq_start_request(bd->rq);
1652 
1653 	if (lo->lo_state != Lo_bound)
1654 		return BLK_MQ_RQ_QUEUE_ERROR;
1655 
1656 	switch (req_op(cmd->rq)) {
1657 	case REQ_OP_FLUSH:
1658 	case REQ_OP_DISCARD:
1659 	case REQ_OP_WRITE_ZEROES:
1660 		cmd->use_aio = false;
1661 		break;
1662 	default:
1663 		cmd->use_aio = lo->use_dio;
1664 		break;
1665 	}
1666 
1667 	kthread_queue_work(&lo->worker, &cmd->work);
1668 
1669 	return BLK_MQ_RQ_QUEUE_OK;
1670 }
1671 
1672 static void loop_handle_cmd(struct loop_cmd *cmd)
1673 {
1674 	const bool write = op_is_write(req_op(cmd->rq));
1675 	struct loop_device *lo = cmd->rq->q->queuedata;
1676 	int ret = 0;
1677 
1678 	if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1679 		ret = -EIO;
1680 		goto failed;
1681 	}
1682 
1683 	ret = do_req_filebacked(lo, cmd->rq);
1684  failed:
1685 	/* complete non-aio request */
1686 	if (!cmd->use_aio || ret) {
1687 		cmd->ret = ret ? -EIO : 0;
1688 		blk_mq_complete_request(cmd->rq);
1689 	}
1690 }
1691 
1692 static void loop_queue_work(struct kthread_work *work)
1693 {
1694 	struct loop_cmd *cmd =
1695 		container_of(work, struct loop_cmd, work);
1696 
1697 	loop_handle_cmd(cmd);
1698 }
1699 
1700 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1701 		unsigned int hctx_idx, unsigned int numa_node)
1702 {
1703 	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1704 
1705 	cmd->rq = rq;
1706 	kthread_init_work(&cmd->work, loop_queue_work);
1707 
1708 	return 0;
1709 }
1710 
1711 static const struct blk_mq_ops loop_mq_ops = {
1712 	.queue_rq       = loop_queue_rq,
1713 	.init_request	= loop_init_request,
1714 	.complete	= lo_complete_rq,
1715 };
1716 
1717 static int loop_add(struct loop_device **l, int i)
1718 {
1719 	struct loop_device *lo;
1720 	struct gendisk *disk;
1721 	int err;
1722 
1723 	err = -ENOMEM;
1724 	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1725 	if (!lo)
1726 		goto out;
1727 
1728 	lo->lo_state = Lo_unbound;
1729 
1730 	/* allocate id, if @id >= 0, we're requesting that specific id */
1731 	if (i >= 0) {
1732 		err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1733 		if (err == -ENOSPC)
1734 			err = -EEXIST;
1735 	} else {
1736 		err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1737 	}
1738 	if (err < 0)
1739 		goto out_free_dev;
1740 	i = err;
1741 
1742 	err = -ENOMEM;
1743 	lo->tag_set.ops = &loop_mq_ops;
1744 	lo->tag_set.nr_hw_queues = 1;
1745 	lo->tag_set.queue_depth = 128;
1746 	lo->tag_set.numa_node = NUMA_NO_NODE;
1747 	lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1748 	lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1749 	lo->tag_set.driver_data = lo;
1750 
1751 	err = blk_mq_alloc_tag_set(&lo->tag_set);
1752 	if (err)
1753 		goto out_free_idr;
1754 
1755 	lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1756 	if (IS_ERR_OR_NULL(lo->lo_queue)) {
1757 		err = PTR_ERR(lo->lo_queue);
1758 		goto out_cleanup_tags;
1759 	}
1760 	lo->lo_queue->queuedata = lo;
1761 
1762 	/*
1763 	 * It doesn't make sense to enable merge because the I/O
1764 	 * submitted to backing file is handled page by page.
1765 	 */
1766 	queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1767 
1768 	err = -ENOMEM;
1769 	disk = lo->lo_disk = alloc_disk(1 << part_shift);
1770 	if (!disk)
1771 		goto out_free_queue;
1772 
1773 	/*
1774 	 * Disable partition scanning by default. The in-kernel partition
1775 	 * scanning can be requested individually per-device during its
1776 	 * setup. Userspace can always add and remove partitions from all
1777 	 * devices. The needed partition minors are allocated from the
1778 	 * extended minor space, the main loop device numbers will continue
1779 	 * to match the loop minors, regardless of the number of partitions
1780 	 * used.
1781 	 *
1782 	 * If max_part is given, partition scanning is globally enabled for
1783 	 * all loop devices. The minors for the main loop devices will be
1784 	 * multiples of max_part.
1785 	 *
1786 	 * Note: Global-for-all-devices, set-only-at-init, read-only module
1787 	 * parameteters like 'max_loop' and 'max_part' make things needlessly
1788 	 * complicated, are too static, inflexible and may surprise
1789 	 * userspace tools. Parameters like this in general should be avoided.
1790 	 */
1791 	if (!part_shift)
1792 		disk->flags |= GENHD_FL_NO_PART_SCAN;
1793 	disk->flags |= GENHD_FL_EXT_DEVT;
1794 	mutex_init(&lo->lo_ctl_mutex);
1795 	atomic_set(&lo->lo_refcnt, 0);
1796 	lo->lo_number		= i;
1797 	spin_lock_init(&lo->lo_lock);
1798 	disk->major		= LOOP_MAJOR;
1799 	disk->first_minor	= i << part_shift;
1800 	disk->fops		= &lo_fops;
1801 	disk->private_data	= lo;
1802 	disk->queue		= lo->lo_queue;
1803 	sprintf(disk->disk_name, "loop%d", i);
1804 	add_disk(disk);
1805 	*l = lo;
1806 	return lo->lo_number;
1807 
1808 out_free_queue:
1809 	blk_cleanup_queue(lo->lo_queue);
1810 out_cleanup_tags:
1811 	blk_mq_free_tag_set(&lo->tag_set);
1812 out_free_idr:
1813 	idr_remove(&loop_index_idr, i);
1814 out_free_dev:
1815 	kfree(lo);
1816 out:
1817 	return err;
1818 }
1819 
1820 static void loop_remove(struct loop_device *lo)
1821 {
1822 	blk_cleanup_queue(lo->lo_queue);
1823 	del_gendisk(lo->lo_disk);
1824 	blk_mq_free_tag_set(&lo->tag_set);
1825 	put_disk(lo->lo_disk);
1826 	kfree(lo);
1827 }
1828 
1829 static int find_free_cb(int id, void *ptr, void *data)
1830 {
1831 	struct loop_device *lo = ptr;
1832 	struct loop_device **l = data;
1833 
1834 	if (lo->lo_state == Lo_unbound) {
1835 		*l = lo;
1836 		return 1;
1837 	}
1838 	return 0;
1839 }
1840 
1841 static int loop_lookup(struct loop_device **l, int i)
1842 {
1843 	struct loop_device *lo;
1844 	int ret = -ENODEV;
1845 
1846 	if (i < 0) {
1847 		int err;
1848 
1849 		err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1850 		if (err == 1) {
1851 			*l = lo;
1852 			ret = lo->lo_number;
1853 		}
1854 		goto out;
1855 	}
1856 
1857 	/* lookup and return a specific i */
1858 	lo = idr_find(&loop_index_idr, i);
1859 	if (lo) {
1860 		*l = lo;
1861 		ret = lo->lo_number;
1862 	}
1863 out:
1864 	return ret;
1865 }
1866 
1867 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1868 {
1869 	struct loop_device *lo;
1870 	struct kobject *kobj;
1871 	int err;
1872 
1873 	mutex_lock(&loop_index_mutex);
1874 	err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1875 	if (err < 0)
1876 		err = loop_add(&lo, MINOR(dev) >> part_shift);
1877 	if (err < 0)
1878 		kobj = NULL;
1879 	else
1880 		kobj = get_disk(lo->lo_disk);
1881 	mutex_unlock(&loop_index_mutex);
1882 
1883 	*part = 0;
1884 	return kobj;
1885 }
1886 
1887 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1888 			       unsigned long parm)
1889 {
1890 	struct loop_device *lo;
1891 	int ret = -ENOSYS;
1892 
1893 	mutex_lock(&loop_index_mutex);
1894 	switch (cmd) {
1895 	case LOOP_CTL_ADD:
1896 		ret = loop_lookup(&lo, parm);
1897 		if (ret >= 0) {
1898 			ret = -EEXIST;
1899 			break;
1900 		}
1901 		ret = loop_add(&lo, parm);
1902 		break;
1903 	case LOOP_CTL_REMOVE:
1904 		ret = loop_lookup(&lo, parm);
1905 		if (ret < 0)
1906 			break;
1907 		mutex_lock(&lo->lo_ctl_mutex);
1908 		if (lo->lo_state != Lo_unbound) {
1909 			ret = -EBUSY;
1910 			mutex_unlock(&lo->lo_ctl_mutex);
1911 			break;
1912 		}
1913 		if (atomic_read(&lo->lo_refcnt) > 0) {
1914 			ret = -EBUSY;
1915 			mutex_unlock(&lo->lo_ctl_mutex);
1916 			break;
1917 		}
1918 		lo->lo_disk->private_data = NULL;
1919 		mutex_unlock(&lo->lo_ctl_mutex);
1920 		idr_remove(&loop_index_idr, lo->lo_number);
1921 		loop_remove(lo);
1922 		break;
1923 	case LOOP_CTL_GET_FREE:
1924 		ret = loop_lookup(&lo, -1);
1925 		if (ret >= 0)
1926 			break;
1927 		ret = loop_add(&lo, -1);
1928 	}
1929 	mutex_unlock(&loop_index_mutex);
1930 
1931 	return ret;
1932 }
1933 
1934 static const struct file_operations loop_ctl_fops = {
1935 	.open		= nonseekable_open,
1936 	.unlocked_ioctl	= loop_control_ioctl,
1937 	.compat_ioctl	= loop_control_ioctl,
1938 	.owner		= THIS_MODULE,
1939 	.llseek		= noop_llseek,
1940 };
1941 
1942 static struct miscdevice loop_misc = {
1943 	.minor		= LOOP_CTRL_MINOR,
1944 	.name		= "loop-control",
1945 	.fops		= &loop_ctl_fops,
1946 };
1947 
1948 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1949 MODULE_ALIAS("devname:loop-control");
1950 
1951 static int __init loop_init(void)
1952 {
1953 	int i, nr;
1954 	unsigned long range;
1955 	struct loop_device *lo;
1956 	int err;
1957 
1958 	err = misc_register(&loop_misc);
1959 	if (err < 0)
1960 		return err;
1961 
1962 	part_shift = 0;
1963 	if (max_part > 0) {
1964 		part_shift = fls(max_part);
1965 
1966 		/*
1967 		 * Adjust max_part according to part_shift as it is exported
1968 		 * to user space so that user can decide correct minor number
1969 		 * if [s]he want to create more devices.
1970 		 *
1971 		 * Note that -1 is required because partition 0 is reserved
1972 		 * for the whole disk.
1973 		 */
1974 		max_part = (1UL << part_shift) - 1;
1975 	}
1976 
1977 	if ((1UL << part_shift) > DISK_MAX_PARTS) {
1978 		err = -EINVAL;
1979 		goto misc_out;
1980 	}
1981 
1982 	if (max_loop > 1UL << (MINORBITS - part_shift)) {
1983 		err = -EINVAL;
1984 		goto misc_out;
1985 	}
1986 
1987 	/*
1988 	 * If max_loop is specified, create that many devices upfront.
1989 	 * This also becomes a hard limit. If max_loop is not specified,
1990 	 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1991 	 * init time. Loop devices can be requested on-demand with the
1992 	 * /dev/loop-control interface, or be instantiated by accessing
1993 	 * a 'dead' device node.
1994 	 */
1995 	if (max_loop) {
1996 		nr = max_loop;
1997 		range = max_loop << part_shift;
1998 	} else {
1999 		nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2000 		range = 1UL << MINORBITS;
2001 	}
2002 
2003 	if (register_blkdev(LOOP_MAJOR, "loop")) {
2004 		err = -EIO;
2005 		goto misc_out;
2006 	}
2007 
2008 	blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2009 				  THIS_MODULE, loop_probe, NULL, NULL);
2010 
2011 	/* pre-create number of devices given by config or max_loop */
2012 	mutex_lock(&loop_index_mutex);
2013 	for (i = 0; i < nr; i++)
2014 		loop_add(&lo, i);
2015 	mutex_unlock(&loop_index_mutex);
2016 
2017 	printk(KERN_INFO "loop: module loaded\n");
2018 	return 0;
2019 
2020 misc_out:
2021 	misc_deregister(&loop_misc);
2022 	return err;
2023 }
2024 
2025 static int loop_exit_cb(int id, void *ptr, void *data)
2026 {
2027 	struct loop_device *lo = ptr;
2028 
2029 	loop_remove(lo);
2030 	return 0;
2031 }
2032 
2033 static void __exit loop_exit(void)
2034 {
2035 	unsigned long range;
2036 
2037 	range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2038 
2039 	idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2040 	idr_destroy(&loop_index_idr);
2041 
2042 	blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2043 	unregister_blkdev(LOOP_MAJOR, "loop");
2044 
2045 	misc_deregister(&loop_misc);
2046 }
2047 
2048 module_init(loop_init);
2049 module_exit(loop_exit);
2050 
2051 #ifndef MODULE
2052 static int __init max_loop_setup(char *str)
2053 {
2054 	max_loop = simple_strtol(str, NULL, 0);
2055 	return 1;
2056 }
2057 
2058 __setup("max_loop=", max_loop_setup);
2059 #endif
2060