xref: /kernel/linux/linux-5.10/drivers/block/loop.c (revision 8c2ecf20)
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 <linux/ioprio.h>
80#include <linux/blk-cgroup.h>
81
82#include "loop.h"
83
84#include <linux/uaccess.h>
85
86static DEFINE_IDR(loop_index_idr);
87static DEFINE_MUTEX(loop_ctl_mutex);
88
89static int max_part;
90static int part_shift;
91
92static int transfer_xor(struct loop_device *lo, int cmd,
93			struct page *raw_page, unsigned raw_off,
94			struct page *loop_page, unsigned loop_off,
95			int size, sector_t real_block)
96{
97	char *raw_buf = kmap_atomic(raw_page) + raw_off;
98	char *loop_buf = kmap_atomic(loop_page) + loop_off;
99	char *in, *out, *key;
100	int i, keysize;
101
102	if (cmd == READ) {
103		in = raw_buf;
104		out = loop_buf;
105	} else {
106		in = loop_buf;
107		out = raw_buf;
108	}
109
110	key = lo->lo_encrypt_key;
111	keysize = lo->lo_encrypt_key_size;
112	for (i = 0; i < size; i++)
113		*out++ = *in++ ^ key[(i & 511) % keysize];
114
115	kunmap_atomic(loop_buf);
116	kunmap_atomic(raw_buf);
117	cond_resched();
118	return 0;
119}
120
121static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122{
123	if (unlikely(info->lo_encrypt_key_size <= 0))
124		return -EINVAL;
125	return 0;
126}
127
128static struct loop_func_table none_funcs = {
129	.number = LO_CRYPT_NONE,
130};
131
132static struct loop_func_table xor_funcs = {
133	.number = LO_CRYPT_XOR,
134	.transfer = transfer_xor,
135	.init = xor_init
136};
137
138/* xfer_funcs[0] is special - its release function is never called */
139static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140	&none_funcs,
141	&xor_funcs
142};
143
144static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145{
146	loff_t loopsize;
147
148	/* Compute loopsize in bytes */
149	loopsize = i_size_read(file->f_mapping->host);
150	if (offset > 0)
151		loopsize -= offset;
152	/* offset is beyond i_size, weird but possible */
153	if (loopsize < 0)
154		return 0;
155
156	if (sizelimit > 0 && sizelimit < loopsize)
157		loopsize = sizelimit;
158	/*
159	 * Unfortunately, if we want to do I/O on the device,
160	 * the number of 512-byte sectors has to fit into a sector_t.
161	 */
162	return loopsize >> 9;
163}
164
165static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166{
167	return get_size(lo->lo_offset, lo->lo_sizelimit, file);
168}
169
170static void __loop_update_dio(struct loop_device *lo, bool dio)
171{
172	struct file *file = lo->lo_backing_file;
173	struct address_space *mapping = file->f_mapping;
174	struct inode *inode = mapping->host;
175	unsigned short sb_bsize = 0;
176	unsigned dio_align = 0;
177	bool use_dio;
178
179	if (inode->i_sb->s_bdev) {
180		sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181		dio_align = sb_bsize - 1;
182	}
183
184	/*
185	 * We support direct I/O only if lo_offset is aligned with the
186	 * logical I/O size of backing device, and the logical block
187	 * size of loop is bigger than the backing device's and the loop
188	 * needn't transform transfer.
189	 *
190	 * TODO: the above condition may be loosed in the future, and
191	 * direct I/O may be switched runtime at that time because most
192	 * of requests in sane applications should be PAGE_SIZE aligned
193	 */
194	if (dio) {
195		if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196				!(lo->lo_offset & dio_align) &&
197				mapping->a_ops->direct_IO &&
198				!lo->transfer)
199			use_dio = true;
200		else
201			use_dio = false;
202	} else {
203		use_dio = false;
204	}
205
206	if (lo->use_dio == use_dio)
207		return;
208
209	/* flush dirty pages before changing direct IO */
210	vfs_fsync(file, 0);
211
212	/*
213	 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214	 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215	 * will get updated by ioctl(LOOP_GET_STATUS)
216	 */
217	if (lo->lo_state == Lo_bound)
218		blk_mq_freeze_queue(lo->lo_queue);
219	lo->use_dio = use_dio;
220	if (use_dio) {
221		blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222		lo->lo_flags |= LO_FLAGS_DIRECT_IO;
223	} else {
224		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225		lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
226	}
227	if (lo->lo_state == Lo_bound)
228		blk_mq_unfreeze_queue(lo->lo_queue);
229}
230
231/**
232 * loop_set_size() - sets device size and notifies userspace
233 * @lo: struct loop_device to set the size for
234 * @size: new size of the loop device
235 *
236 * Callers must validate that the size passed into this function fits into
237 * a sector_t, eg using loop_validate_size()
238 */
239static void loop_set_size(struct loop_device *lo, loff_t size)
240{
241	struct block_device *bdev = lo->lo_device;
242
243	bd_set_nr_sectors(bdev, size);
244
245	if (!set_capacity_revalidate_and_notify(lo->lo_disk, size, false))
246		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
247}
248
249static inline int
250lo_do_transfer(struct loop_device *lo, int cmd,
251	       struct page *rpage, unsigned roffs,
252	       struct page *lpage, unsigned loffs,
253	       int size, sector_t rblock)
254{
255	int ret;
256
257	ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
258	if (likely(!ret))
259		return 0;
260
261	printk_ratelimited(KERN_ERR
262		"loop: Transfer error at byte offset %llu, length %i.\n",
263		(unsigned long long)rblock << 9, size);
264	return ret;
265}
266
267static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
268{
269	struct iov_iter i;
270	ssize_t bw;
271
272	iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
273
274	file_start_write(file);
275	bw = vfs_iter_write(file, &i, ppos, 0);
276	file_end_write(file);
277
278	if (likely(bw ==  bvec->bv_len))
279		return 0;
280
281	printk_ratelimited(KERN_ERR
282		"loop: Write error at byte offset %llu, length %i.\n",
283		(unsigned long long)*ppos, bvec->bv_len);
284	if (bw >= 0)
285		bw = -EIO;
286	return bw;
287}
288
289static int lo_write_simple(struct loop_device *lo, struct request *rq,
290		loff_t pos)
291{
292	struct bio_vec bvec;
293	struct req_iterator iter;
294	int ret = 0;
295
296	rq_for_each_segment(bvec, rq, iter) {
297		ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
298		if (ret < 0)
299			break;
300		cond_resched();
301	}
302
303	return ret;
304}
305
306/*
307 * This is the slow, transforming version that needs to double buffer the
308 * data as it cannot do the transformations in place without having direct
309 * access to the destination pages of the backing file.
310 */
311static int lo_write_transfer(struct loop_device *lo, struct request *rq,
312		loff_t pos)
313{
314	struct bio_vec bvec, b;
315	struct req_iterator iter;
316	struct page *page;
317	int ret = 0;
318
319	page = alloc_page(GFP_NOIO);
320	if (unlikely(!page))
321		return -ENOMEM;
322
323	rq_for_each_segment(bvec, rq, iter) {
324		ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
325			bvec.bv_offset, bvec.bv_len, pos >> 9);
326		if (unlikely(ret))
327			break;
328
329		b.bv_page = page;
330		b.bv_offset = 0;
331		b.bv_len = bvec.bv_len;
332		ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
333		if (ret < 0)
334			break;
335	}
336
337	__free_page(page);
338	return ret;
339}
340
341static int lo_read_simple(struct loop_device *lo, struct request *rq,
342		loff_t pos)
343{
344	struct bio_vec bvec;
345	struct req_iterator iter;
346	struct iov_iter i;
347	ssize_t len;
348
349	rq_for_each_segment(bvec, rq, iter) {
350		iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
351		len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
352		if (len < 0)
353			return len;
354
355		flush_dcache_page(bvec.bv_page);
356
357		if (len != bvec.bv_len) {
358			struct bio *bio;
359
360			__rq_for_each_bio(bio, rq)
361				zero_fill_bio(bio);
362			break;
363		}
364		cond_resched();
365	}
366
367	return 0;
368}
369
370static int lo_read_transfer(struct loop_device *lo, struct request *rq,
371		loff_t pos)
372{
373	struct bio_vec bvec, b;
374	struct req_iterator iter;
375	struct iov_iter i;
376	struct page *page;
377	ssize_t len;
378	int ret = 0;
379
380	page = alloc_page(GFP_NOIO);
381	if (unlikely(!page))
382		return -ENOMEM;
383
384	rq_for_each_segment(bvec, rq, iter) {
385		loff_t offset = pos;
386
387		b.bv_page = page;
388		b.bv_offset = 0;
389		b.bv_len = bvec.bv_len;
390
391		iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
392		len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
393		if (len < 0) {
394			ret = len;
395			goto out_free_page;
396		}
397
398		ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
399			bvec.bv_offset, len, offset >> 9);
400		if (ret)
401			goto out_free_page;
402
403		flush_dcache_page(bvec.bv_page);
404
405		if (len != bvec.bv_len) {
406			struct bio *bio;
407
408			__rq_for_each_bio(bio, rq)
409				zero_fill_bio(bio);
410			break;
411		}
412	}
413
414	ret = 0;
415out_free_page:
416	__free_page(page);
417	return ret;
418}
419
420static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
421			int mode)
422{
423	/*
424	 * We use fallocate to manipulate the space mappings used by the image
425	 * a.k.a. discard/zerorange. However we do not support this if
426	 * encryption is enabled, because it may give an attacker useful
427	 * information.
428	 */
429	struct file *file = lo->lo_backing_file;
430	struct request_queue *q = lo->lo_queue;
431	int ret;
432
433	mode |= FALLOC_FL_KEEP_SIZE;
434
435	if (!blk_queue_discard(q)) {
436		ret = -EOPNOTSUPP;
437		goto out;
438	}
439
440	ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
441	if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
442		ret = -EIO;
443 out:
444	return ret;
445}
446
447static int lo_req_flush(struct loop_device *lo, struct request *rq)
448{
449	struct file *file = lo->lo_backing_file;
450	int ret = vfs_fsync(file, 0);
451	if (unlikely(ret && ret != -EINVAL))
452		ret = -EIO;
453
454	return ret;
455}
456
457static void lo_complete_rq(struct request *rq)
458{
459	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
460	blk_status_t ret = BLK_STS_OK;
461
462	if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
463	    req_op(rq) != REQ_OP_READ) {
464		if (cmd->ret < 0)
465			ret = errno_to_blk_status(cmd->ret);
466		goto end_io;
467	}
468
469	/*
470	 * Short READ - if we got some data, advance our request and
471	 * retry it. If we got no data, end the rest with EIO.
472	 */
473	if (cmd->ret) {
474		blk_update_request(rq, BLK_STS_OK, cmd->ret);
475		cmd->ret = 0;
476		blk_mq_requeue_request(rq, true);
477	} else {
478		if (cmd->use_aio) {
479			struct bio *bio = rq->bio;
480
481			while (bio) {
482				zero_fill_bio(bio);
483				bio = bio->bi_next;
484			}
485		}
486		ret = BLK_STS_IOERR;
487end_io:
488		blk_mq_end_request(rq, ret);
489	}
490}
491
492static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
493{
494	struct request *rq = blk_mq_rq_from_pdu(cmd);
495
496	if (!atomic_dec_and_test(&cmd->ref))
497		return;
498	kfree(cmd->bvec);
499	cmd->bvec = NULL;
500	if (likely(!blk_should_fake_timeout(rq->q)))
501		blk_mq_complete_request(rq);
502}
503
504static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
505{
506	struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
507
508	if (cmd->css)
509		css_put(cmd->css);
510	cmd->ret = ret;
511	lo_rw_aio_do_completion(cmd);
512}
513
514static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
515		     loff_t pos, bool rw)
516{
517	struct iov_iter iter;
518	struct req_iterator rq_iter;
519	struct bio_vec *bvec;
520	struct request *rq = blk_mq_rq_from_pdu(cmd);
521	struct bio *bio = rq->bio;
522	struct file *file = lo->lo_backing_file;
523	struct bio_vec tmp;
524	unsigned int offset;
525	int nr_bvec = 0;
526	int ret;
527
528	rq_for_each_bvec(tmp, rq, rq_iter)
529		nr_bvec++;
530
531	if (rq->bio != rq->biotail) {
532
533		bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
534				     GFP_NOIO);
535		if (!bvec)
536			return -EIO;
537		cmd->bvec = bvec;
538
539		/*
540		 * The bios of the request may be started from the middle of
541		 * the 'bvec' because of bio splitting, so we can't directly
542		 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
543		 * API will take care of all details for us.
544		 */
545		rq_for_each_bvec(tmp, rq, rq_iter) {
546			*bvec = tmp;
547			bvec++;
548		}
549		bvec = cmd->bvec;
550		offset = 0;
551	} else {
552		/*
553		 * Same here, this bio may be started from the middle of the
554		 * 'bvec' because of bio splitting, so offset from the bvec
555		 * must be passed to iov iterator
556		 */
557		offset = bio->bi_iter.bi_bvec_done;
558		bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
559	}
560	atomic_set(&cmd->ref, 2);
561
562	iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
563	iter.iov_offset = offset;
564
565	cmd->iocb.ki_pos = pos;
566	cmd->iocb.ki_filp = file;
567	cmd->iocb.ki_complete = lo_rw_aio_complete;
568	cmd->iocb.ki_flags = IOCB_DIRECT;
569	cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
570	if (cmd->css)
571		kthread_associate_blkcg(cmd->css);
572
573	if (rw == WRITE)
574		ret = call_write_iter(file, &cmd->iocb, &iter);
575	else
576		ret = call_read_iter(file, &cmd->iocb, &iter);
577
578	lo_rw_aio_do_completion(cmd);
579	kthread_associate_blkcg(NULL);
580
581	if (ret != -EIOCBQUEUED)
582		cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
583	return 0;
584}
585
586static int do_req_filebacked(struct loop_device *lo, struct request *rq)
587{
588	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
589	loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
590
591	/*
592	 * lo_write_simple and lo_read_simple should have been covered
593	 * by io submit style function like lo_rw_aio(), one blocker
594	 * is that lo_read_simple() need to call flush_dcache_page after
595	 * the page is written from kernel, and it isn't easy to handle
596	 * this in io submit style function which submits all segments
597	 * of the req at one time. And direct read IO doesn't need to
598	 * run flush_dcache_page().
599	 */
600	switch (req_op(rq)) {
601	case REQ_OP_FLUSH:
602		return lo_req_flush(lo, rq);
603	case REQ_OP_WRITE_ZEROES:
604		/*
605		 * If the caller doesn't want deallocation, call zeroout to
606		 * write zeroes the range.  Otherwise, punch them out.
607		 */
608		return lo_fallocate(lo, rq, pos,
609			(rq->cmd_flags & REQ_NOUNMAP) ?
610				FALLOC_FL_ZERO_RANGE :
611				FALLOC_FL_PUNCH_HOLE);
612	case REQ_OP_DISCARD:
613		return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
614	case REQ_OP_WRITE:
615		if (lo->transfer)
616			return lo_write_transfer(lo, rq, pos);
617		else if (cmd->use_aio)
618			return lo_rw_aio(lo, cmd, pos, WRITE);
619		else
620			return lo_write_simple(lo, rq, pos);
621	case REQ_OP_READ:
622		if (lo->transfer)
623			return lo_read_transfer(lo, rq, pos);
624		else if (cmd->use_aio)
625			return lo_rw_aio(lo, cmd, pos, READ);
626		else
627			return lo_read_simple(lo, rq, pos);
628	default:
629		WARN_ON_ONCE(1);
630		return -EIO;
631	}
632}
633
634static inline void loop_update_dio(struct loop_device *lo)
635{
636	__loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
637				lo->use_dio);
638}
639
640static void loop_reread_partitions(struct loop_device *lo,
641				   struct block_device *bdev)
642{
643	int rc;
644
645	mutex_lock(&bdev->bd_mutex);
646	rc = bdev_disk_changed(bdev, false);
647	mutex_unlock(&bdev->bd_mutex);
648	if (rc)
649		pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
650			__func__, lo->lo_number, lo->lo_file_name, rc);
651}
652
653static inline int is_loop_device(struct file *file)
654{
655	struct inode *i = file->f_mapping->host;
656
657	return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
658}
659
660static int loop_validate_file(struct file *file, struct block_device *bdev)
661{
662	struct inode	*inode = file->f_mapping->host;
663	struct file	*f = file;
664
665	/* Avoid recursion */
666	while (is_loop_device(f)) {
667		struct loop_device *l;
668
669		if (f->f_mapping->host->i_bdev == bdev)
670			return -EBADF;
671
672		l = f->f_mapping->host->i_bdev->bd_disk->private_data;
673		if (l->lo_state != Lo_bound) {
674			return -EINVAL;
675		}
676		f = l->lo_backing_file;
677	}
678	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
679		return -EINVAL;
680	return 0;
681}
682
683/*
684 * loop_change_fd switched the backing store of a loopback device to
685 * a new file. This is useful for operating system installers to free up
686 * the original file and in High Availability environments to switch to
687 * an alternative location for the content in case of server meltdown.
688 * This can only work if the loop device is used read-only, and if the
689 * new backing store is the same size and type as the old backing store.
690 */
691static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
692			  unsigned int arg)
693{
694	struct file	*file = NULL, *old_file;
695	int		error;
696	bool		partscan;
697
698	error = mutex_lock_killable(&loop_ctl_mutex);
699	if (error)
700		return error;
701	error = -ENXIO;
702	if (lo->lo_state != Lo_bound)
703		goto out_err;
704
705	/* the loop device has to be read-only */
706	error = -EINVAL;
707	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
708		goto out_err;
709
710	error = -EBADF;
711	file = fget(arg);
712	if (!file)
713		goto out_err;
714
715	error = loop_validate_file(file, bdev);
716	if (error)
717		goto out_err;
718
719	old_file = lo->lo_backing_file;
720
721	error = -EINVAL;
722
723	/* size of the new backing store needs to be the same */
724	if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
725		goto out_err;
726
727	/* and ... switch */
728	blk_mq_freeze_queue(lo->lo_queue);
729	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
730	lo->lo_backing_file = file;
731	lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
732	mapping_set_gfp_mask(file->f_mapping,
733			     lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
734	loop_update_dio(lo);
735	blk_mq_unfreeze_queue(lo->lo_queue);
736	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
737	mutex_unlock(&loop_ctl_mutex);
738	/*
739	 * We must drop file reference outside of loop_ctl_mutex as dropping
740	 * the file ref can take bd_mutex which creates circular locking
741	 * dependency.
742	 */
743	fput(old_file);
744	if (partscan)
745		loop_reread_partitions(lo, bdev);
746	return 0;
747
748out_err:
749	mutex_unlock(&loop_ctl_mutex);
750	if (file)
751		fput(file);
752	return error;
753}
754
755/* loop sysfs attributes */
756
757static ssize_t loop_attr_show(struct device *dev, char *page,
758			      ssize_t (*callback)(struct loop_device *, char *))
759{
760	struct gendisk *disk = dev_to_disk(dev);
761	struct loop_device *lo = disk->private_data;
762
763	return callback(lo, page);
764}
765
766#define LOOP_ATTR_RO(_name)						\
767static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);	\
768static ssize_t loop_attr_do_show_##_name(struct device *d,		\
769				struct device_attribute *attr, char *b)	\
770{									\
771	return loop_attr_show(d, b, loop_attr_##_name##_show);		\
772}									\
773static struct device_attribute loop_attr_##_name =			\
774	__ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
775
776static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
777{
778	ssize_t ret;
779	char *p = NULL;
780
781	spin_lock_irq(&lo->lo_lock);
782	if (lo->lo_backing_file)
783		p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
784	spin_unlock_irq(&lo->lo_lock);
785
786	if (IS_ERR_OR_NULL(p))
787		ret = PTR_ERR(p);
788	else {
789		ret = strlen(p);
790		memmove(buf, p, ret);
791		buf[ret++] = '\n';
792		buf[ret] = 0;
793	}
794
795	return ret;
796}
797
798static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
799{
800	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
801}
802
803static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
804{
805	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
806}
807
808static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
809{
810	int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
811
812	return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
813}
814
815static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
816{
817	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
818
819	return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
820}
821
822static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
823{
824	int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
825
826	return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
827}
828
829LOOP_ATTR_RO(backing_file);
830LOOP_ATTR_RO(offset);
831LOOP_ATTR_RO(sizelimit);
832LOOP_ATTR_RO(autoclear);
833LOOP_ATTR_RO(partscan);
834LOOP_ATTR_RO(dio);
835
836static struct attribute *loop_attrs[] = {
837	&loop_attr_backing_file.attr,
838	&loop_attr_offset.attr,
839	&loop_attr_sizelimit.attr,
840	&loop_attr_autoclear.attr,
841	&loop_attr_partscan.attr,
842	&loop_attr_dio.attr,
843	NULL,
844};
845
846static struct attribute_group loop_attribute_group = {
847	.name = "loop",
848	.attrs= loop_attrs,
849};
850
851static void loop_sysfs_init(struct loop_device *lo)
852{
853	lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
854						&loop_attribute_group);
855}
856
857static void loop_sysfs_exit(struct loop_device *lo)
858{
859	if (lo->sysfs_inited)
860		sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
861				   &loop_attribute_group);
862}
863
864static void loop_config_discard(struct loop_device *lo)
865{
866	struct file *file = lo->lo_backing_file;
867	struct inode *inode = file->f_mapping->host;
868	struct request_queue *q = lo->lo_queue;
869	u32 granularity, max_discard_sectors;
870
871	/*
872	 * If the backing device is a block device, mirror its zeroing
873	 * capability. Set the discard sectors to the block device's zeroing
874	 * capabilities because loop discards result in blkdev_issue_zeroout(),
875	 * not blkdev_issue_discard(). This maintains consistent behavior with
876	 * file-backed loop devices: discarded regions read back as zero.
877	 */
878	if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
879		struct request_queue *backingq;
880
881		backingq = bdev_get_queue(inode->i_bdev);
882
883		max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
884		granularity = backingq->limits.discard_granularity ?:
885			queue_physical_block_size(backingq);
886
887	/*
888	 * We use punch hole to reclaim the free space used by the
889	 * image a.k.a. discard. However we do not support discard if
890	 * encryption is enabled, because it may give an attacker
891	 * useful information.
892	 */
893	} else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
894		max_discard_sectors = 0;
895		granularity = 0;
896
897	} else {
898		max_discard_sectors = UINT_MAX >> 9;
899		granularity = inode->i_sb->s_blocksize;
900	}
901
902	if (max_discard_sectors) {
903		q->limits.discard_granularity = granularity;
904		blk_queue_max_discard_sectors(q, max_discard_sectors);
905		blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
906		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
907	} else {
908		q->limits.discard_granularity = 0;
909		blk_queue_max_discard_sectors(q, 0);
910		blk_queue_max_write_zeroes_sectors(q, 0);
911		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
912	}
913	q->limits.discard_alignment = 0;
914}
915
916static void loop_unprepare_queue(struct loop_device *lo)
917{
918	kthread_flush_worker(&lo->worker);
919	kthread_stop(lo->worker_task);
920}
921
922static int loop_kthread_worker_fn(void *worker_ptr)
923{
924	current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
925	return kthread_worker_fn(worker_ptr);
926}
927
928static int loop_prepare_queue(struct loop_device *lo)
929{
930	kthread_init_worker(&lo->worker);
931	lo->worker_task = kthread_run(loop_kthread_worker_fn,
932			&lo->worker, "loop%d", lo->lo_number);
933	if (IS_ERR(lo->worker_task))
934		return -ENOMEM;
935	set_user_nice(lo->worker_task, MIN_NICE);
936	return 0;
937}
938
939static void loop_update_rotational(struct loop_device *lo)
940{
941	struct file *file = lo->lo_backing_file;
942	struct inode *file_inode = file->f_mapping->host;
943	struct block_device *file_bdev = file_inode->i_sb->s_bdev;
944	struct request_queue *q = lo->lo_queue;
945	bool nonrot = true;
946
947	/* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
948	if (file_bdev)
949		nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
950
951	if (nonrot)
952		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
953	else
954		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
955}
956
957static int
958loop_release_xfer(struct loop_device *lo)
959{
960	int err = 0;
961	struct loop_func_table *xfer = lo->lo_encryption;
962
963	if (xfer) {
964		if (xfer->release)
965			err = xfer->release(lo);
966		lo->transfer = NULL;
967		lo->lo_encryption = NULL;
968		module_put(xfer->owner);
969	}
970	return err;
971}
972
973static int
974loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
975	       const struct loop_info64 *i)
976{
977	int err = 0;
978
979	if (xfer) {
980		struct module *owner = xfer->owner;
981
982		if (!try_module_get(owner))
983			return -EINVAL;
984		if (xfer->init)
985			err = xfer->init(lo, i);
986		if (err)
987			module_put(owner);
988		else
989			lo->lo_encryption = xfer;
990	}
991	return err;
992}
993
994/**
995 * loop_set_status_from_info - configure device from loop_info
996 * @lo: struct loop_device to configure
997 * @info: struct loop_info64 to configure the device with
998 *
999 * Configures the loop device parameters according to the passed
1000 * in loop_info64 configuration.
1001 */
1002static int
1003loop_set_status_from_info(struct loop_device *lo,
1004			  const struct loop_info64 *info)
1005{
1006	int err;
1007	struct loop_func_table *xfer;
1008	kuid_t uid = current_uid();
1009
1010	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1011		return -EINVAL;
1012
1013	err = loop_release_xfer(lo);
1014	if (err)
1015		return err;
1016
1017	if (info->lo_encrypt_type) {
1018		unsigned int type = info->lo_encrypt_type;
1019
1020		if (type >= MAX_LO_CRYPT)
1021			return -EINVAL;
1022		xfer = xfer_funcs[type];
1023		if (xfer == NULL)
1024			return -EINVAL;
1025	} else
1026		xfer = NULL;
1027
1028	err = loop_init_xfer(lo, xfer, info);
1029	if (err)
1030		return err;
1031
1032	/* Avoid assigning overflow values */
1033	if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
1034		return -EOVERFLOW;
1035
1036	lo->lo_offset = info->lo_offset;
1037	lo->lo_sizelimit = info->lo_sizelimit;
1038
1039	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1040	memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1041	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1042	lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1043
1044	if (!xfer)
1045		xfer = &none_funcs;
1046	lo->transfer = xfer->transfer;
1047	lo->ioctl = xfer->ioctl;
1048
1049	lo->lo_flags = info->lo_flags;
1050
1051	lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1052	lo->lo_init[0] = info->lo_init[0];
1053	lo->lo_init[1] = info->lo_init[1];
1054	if (info->lo_encrypt_key_size) {
1055		memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1056		       info->lo_encrypt_key_size);
1057		lo->lo_key_owner = uid;
1058	}
1059
1060	return 0;
1061}
1062
1063static int loop_configure(struct loop_device *lo, fmode_t mode,
1064			  struct block_device *bdev,
1065			  const struct loop_config *config)
1066{
1067	struct file	*file;
1068	struct inode	*inode;
1069	struct address_space *mapping;
1070	struct block_device *claimed_bdev = NULL;
1071	int		error;
1072	loff_t		size;
1073	bool		partscan;
1074	unsigned short  bsize;
1075
1076	/* This is safe, since we have a reference from open(). */
1077	__module_get(THIS_MODULE);
1078
1079	error = -EBADF;
1080	file = fget(config->fd);
1081	if (!file)
1082		goto out;
1083
1084	/*
1085	 * If we don't hold exclusive handle for the device, upgrade to it
1086	 * here to avoid changing device under exclusive owner.
1087	 */
1088	if (!(mode & FMODE_EXCL)) {
1089		claimed_bdev = bdev->bd_contains;
1090		error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
1091		if (error)
1092			goto out_putf;
1093	}
1094
1095	error = mutex_lock_killable(&loop_ctl_mutex);
1096	if (error)
1097		goto out_bdev;
1098
1099	error = -EBUSY;
1100	if (lo->lo_state != Lo_unbound)
1101		goto out_unlock;
1102
1103	error = loop_validate_file(file, bdev);
1104	if (error)
1105		goto out_unlock;
1106
1107	mapping = file->f_mapping;
1108	inode = mapping->host;
1109
1110	if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1111		error = -EINVAL;
1112		goto out_unlock;
1113	}
1114
1115	if (config->block_size) {
1116		error = blk_validate_block_size(config->block_size);
1117		if (error)
1118			goto out_unlock;
1119	}
1120
1121	error = loop_set_status_from_info(lo, &config->info);
1122	if (error)
1123		goto out_unlock;
1124
1125	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1126	    !file->f_op->write_iter)
1127		lo->lo_flags |= LO_FLAGS_READ_ONLY;
1128
1129	error = loop_prepare_queue(lo);
1130	if (error)
1131		goto out_unlock;
1132
1133	set_device_ro(bdev, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1134
1135	lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1136	lo->lo_device = bdev;
1137	lo->lo_backing_file = file;
1138	lo->old_gfp_mask = mapping_gfp_mask(mapping);
1139	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1140
1141	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1142		blk_queue_write_cache(lo->lo_queue, true, false);
1143
1144	if (config->block_size)
1145		bsize = config->block_size;
1146	else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1147		/* In case of direct I/O, match underlying block size */
1148		bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1149	else
1150		bsize = 512;
1151
1152	blk_queue_logical_block_size(lo->lo_queue, bsize);
1153	blk_queue_physical_block_size(lo->lo_queue, bsize);
1154	blk_queue_io_min(lo->lo_queue, bsize);
1155
1156	loop_config_discard(lo);
1157	loop_update_rotational(lo);
1158	loop_update_dio(lo);
1159	loop_sysfs_init(lo);
1160
1161	size = get_loop_size(lo, file);
1162	loop_set_size(lo, size);
1163
1164	set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1165		      block_size(inode->i_bdev) : PAGE_SIZE);
1166
1167	lo->lo_state = Lo_bound;
1168	if (part_shift)
1169		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1170	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1171	if (partscan)
1172		lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1173
1174	/* Grab the block_device to prevent its destruction after we
1175	 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1176	 */
1177	bdgrab(bdev);
1178	mutex_unlock(&loop_ctl_mutex);
1179	if (partscan)
1180		loop_reread_partitions(lo, bdev);
1181	if (claimed_bdev)
1182		bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1183	return 0;
1184
1185out_unlock:
1186	mutex_unlock(&loop_ctl_mutex);
1187out_bdev:
1188	if (claimed_bdev)
1189		bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1190out_putf:
1191	fput(file);
1192out:
1193	/* This is safe: open() is still holding a reference. */
1194	module_put(THIS_MODULE);
1195	return error;
1196}
1197
1198static int __loop_clr_fd(struct loop_device *lo, bool release)
1199{
1200	struct file *filp = NULL;
1201	gfp_t gfp = lo->old_gfp_mask;
1202	struct block_device *bdev = lo->lo_device;
1203	int err = 0;
1204	bool partscan = false;
1205	int lo_number;
1206
1207	mutex_lock(&loop_ctl_mutex);
1208	if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1209		err = -ENXIO;
1210		goto out_unlock;
1211	}
1212
1213	filp = lo->lo_backing_file;
1214	if (filp == NULL) {
1215		err = -EINVAL;
1216		goto out_unlock;
1217	}
1218
1219	if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1220		blk_queue_write_cache(lo->lo_queue, false, false);
1221
1222	/* freeze request queue during the transition */
1223	blk_mq_freeze_queue(lo->lo_queue);
1224
1225	spin_lock_irq(&lo->lo_lock);
1226	lo->lo_backing_file = NULL;
1227	spin_unlock_irq(&lo->lo_lock);
1228
1229	loop_release_xfer(lo);
1230	lo->transfer = NULL;
1231	lo->ioctl = NULL;
1232	lo->lo_device = NULL;
1233	lo->lo_encryption = NULL;
1234	lo->lo_offset = 0;
1235	lo->lo_sizelimit = 0;
1236	lo->lo_encrypt_key_size = 0;
1237	memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1238	memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1239	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1240	blk_queue_logical_block_size(lo->lo_queue, 512);
1241	blk_queue_physical_block_size(lo->lo_queue, 512);
1242	blk_queue_io_min(lo->lo_queue, 512);
1243	if (bdev) {
1244		bdput(bdev);
1245		invalidate_bdev(bdev);
1246		bdev->bd_inode->i_mapping->wb_err = 0;
1247	}
1248	set_capacity(lo->lo_disk, 0);
1249	loop_sysfs_exit(lo);
1250	if (bdev) {
1251		bd_set_nr_sectors(bdev, 0);
1252		/* let user-space know about this change */
1253		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1254	}
1255	mapping_set_gfp_mask(filp->f_mapping, gfp);
1256	/* This is safe: open() is still holding a reference. */
1257	module_put(THIS_MODULE);
1258	blk_mq_unfreeze_queue(lo->lo_queue);
1259
1260	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1261	lo_number = lo->lo_number;
1262	loop_unprepare_queue(lo);
1263out_unlock:
1264	mutex_unlock(&loop_ctl_mutex);
1265	if (partscan) {
1266		/*
1267		 * bd_mutex has been held already in release path, so don't
1268		 * acquire it if this function is called in such case.
1269		 *
1270		 * If the reread partition isn't from release path, lo_refcnt
1271		 * must be at least one and it can only become zero when the
1272		 * current holder is released.
1273		 */
1274		if (!release)
1275			mutex_lock(&bdev->bd_mutex);
1276		err = bdev_disk_changed(bdev, false);
1277		if (!release)
1278			mutex_unlock(&bdev->bd_mutex);
1279		if (err)
1280			pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1281				__func__, lo_number, err);
1282		/* Device is gone, no point in returning error */
1283		err = 0;
1284	}
1285
1286	/*
1287	 * lo->lo_state is set to Lo_unbound here after above partscan has
1288	 * finished.
1289	 *
1290	 * There cannot be anybody else entering __loop_clr_fd() as
1291	 * lo->lo_backing_file is already cleared and Lo_rundown state
1292	 * protects us from all the other places trying to change the 'lo'
1293	 * device.
1294	 */
1295	mutex_lock(&loop_ctl_mutex);
1296	lo->lo_flags = 0;
1297	if (!part_shift)
1298		lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1299	lo->lo_state = Lo_unbound;
1300	mutex_unlock(&loop_ctl_mutex);
1301
1302	/*
1303	 * Need not hold loop_ctl_mutex to fput backing file.
1304	 * Calling fput holding loop_ctl_mutex triggers a circular
1305	 * lock dependency possibility warning as fput can take
1306	 * bd_mutex which is usually taken before loop_ctl_mutex.
1307	 */
1308	if (filp)
1309		fput(filp);
1310	return err;
1311}
1312
1313static int loop_clr_fd(struct loop_device *lo)
1314{
1315	int err;
1316
1317	err = mutex_lock_killable(&loop_ctl_mutex);
1318	if (err)
1319		return err;
1320	if (lo->lo_state != Lo_bound) {
1321		mutex_unlock(&loop_ctl_mutex);
1322		return -ENXIO;
1323	}
1324	/*
1325	 * If we've explicitly asked to tear down the loop device,
1326	 * and it has an elevated reference count, set it for auto-teardown when
1327	 * the last reference goes away. This stops $!~#$@ udev from
1328	 * preventing teardown because it decided that it needs to run blkid on
1329	 * the loopback device whenever they appear. xfstests is notorious for
1330	 * failing tests because blkid via udev races with a losetup
1331	 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1332	 * command to fail with EBUSY.
1333	 */
1334	if (atomic_read(&lo->lo_refcnt) > 1) {
1335		lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1336		mutex_unlock(&loop_ctl_mutex);
1337		return 0;
1338	}
1339	lo->lo_state = Lo_rundown;
1340	mutex_unlock(&loop_ctl_mutex);
1341
1342	return __loop_clr_fd(lo, false);
1343}
1344
1345static int
1346loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1347{
1348	int err;
1349	struct block_device *bdev;
1350	kuid_t uid = current_uid();
1351	int prev_lo_flags;
1352	bool partscan = false;
1353	bool size_changed = false;
1354
1355	err = mutex_lock_killable(&loop_ctl_mutex);
1356	if (err)
1357		return err;
1358	if (lo->lo_encrypt_key_size &&
1359	    !uid_eq(lo->lo_key_owner, uid) &&
1360	    !capable(CAP_SYS_ADMIN)) {
1361		err = -EPERM;
1362		goto out_unlock;
1363	}
1364	if (lo->lo_state != Lo_bound) {
1365		err = -ENXIO;
1366		goto out_unlock;
1367	}
1368
1369	if (lo->lo_offset != info->lo_offset ||
1370	    lo->lo_sizelimit != info->lo_sizelimit) {
1371		size_changed = true;
1372		sync_blockdev(lo->lo_device);
1373		invalidate_bdev(lo->lo_device);
1374	}
1375
1376	/* I/O need to be drained during transfer transition */
1377	blk_mq_freeze_queue(lo->lo_queue);
1378
1379	if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1380		/* If any pages were dirtied after invalidate_bdev(), try again */
1381		err = -EAGAIN;
1382		pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1383			__func__, lo->lo_number, lo->lo_file_name,
1384			lo->lo_device->bd_inode->i_mapping->nrpages);
1385		goto out_unfreeze;
1386	}
1387
1388	prev_lo_flags = lo->lo_flags;
1389
1390	err = loop_set_status_from_info(lo, info);
1391	if (err)
1392		goto out_unfreeze;
1393
1394	/* Mask out flags that can't be set using LOOP_SET_STATUS. */
1395	lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1396	/* For those flags, use the previous values instead */
1397	lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1398	/* For flags that can't be cleared, use previous values too */
1399	lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1400
1401	if (size_changed) {
1402		loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1403					   lo->lo_backing_file);
1404		loop_set_size(lo, new_size);
1405	}
1406
1407	loop_config_discard(lo);
1408
1409	/* update dio if lo_offset or transfer is changed */
1410	__loop_update_dio(lo, lo->use_dio);
1411
1412out_unfreeze:
1413	blk_mq_unfreeze_queue(lo->lo_queue);
1414
1415	if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1416	     !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1417		lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1418		bdev = lo->lo_device;
1419		partscan = true;
1420	}
1421out_unlock:
1422	mutex_unlock(&loop_ctl_mutex);
1423	if (partscan)
1424		loop_reread_partitions(lo, bdev);
1425
1426	return err;
1427}
1428
1429static int
1430loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1431{
1432	struct path path;
1433	struct kstat stat;
1434	int ret;
1435
1436	ret = mutex_lock_killable(&loop_ctl_mutex);
1437	if (ret)
1438		return ret;
1439	if (lo->lo_state != Lo_bound) {
1440		mutex_unlock(&loop_ctl_mutex);
1441		return -ENXIO;
1442	}
1443
1444	memset(info, 0, sizeof(*info));
1445	info->lo_number = lo->lo_number;
1446	info->lo_offset = lo->lo_offset;
1447	info->lo_sizelimit = lo->lo_sizelimit;
1448	info->lo_flags = lo->lo_flags;
1449	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1450	memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1451	info->lo_encrypt_type =
1452		lo->lo_encryption ? lo->lo_encryption->number : 0;
1453	if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1454		info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1455		memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1456		       lo->lo_encrypt_key_size);
1457	}
1458
1459	/* Drop loop_ctl_mutex while we call into the filesystem. */
1460	path = lo->lo_backing_file->f_path;
1461	path_get(&path);
1462	mutex_unlock(&loop_ctl_mutex);
1463	ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1464	if (!ret) {
1465		info->lo_device = huge_encode_dev(stat.dev);
1466		info->lo_inode = stat.ino;
1467		info->lo_rdevice = huge_encode_dev(stat.rdev);
1468	}
1469	path_put(&path);
1470	return ret;
1471}
1472
1473static void
1474loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1475{
1476	memset(info64, 0, sizeof(*info64));
1477	info64->lo_number = info->lo_number;
1478	info64->lo_device = info->lo_device;
1479	info64->lo_inode = info->lo_inode;
1480	info64->lo_rdevice = info->lo_rdevice;
1481	info64->lo_offset = info->lo_offset;
1482	info64->lo_sizelimit = 0;
1483	info64->lo_encrypt_type = info->lo_encrypt_type;
1484	info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1485	info64->lo_flags = info->lo_flags;
1486	info64->lo_init[0] = info->lo_init[0];
1487	info64->lo_init[1] = info->lo_init[1];
1488	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1489		memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1490	else
1491		memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1492	memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1493}
1494
1495static int
1496loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1497{
1498	memset(info, 0, sizeof(*info));
1499	info->lo_number = info64->lo_number;
1500	info->lo_device = info64->lo_device;
1501	info->lo_inode = info64->lo_inode;
1502	info->lo_rdevice = info64->lo_rdevice;
1503	info->lo_offset = info64->lo_offset;
1504	info->lo_encrypt_type = info64->lo_encrypt_type;
1505	info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1506	info->lo_flags = info64->lo_flags;
1507	info->lo_init[0] = info64->lo_init[0];
1508	info->lo_init[1] = info64->lo_init[1];
1509	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1510		memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1511	else
1512		memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1513	memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1514
1515	/* error in case values were truncated */
1516	if (info->lo_device != info64->lo_device ||
1517	    info->lo_rdevice != info64->lo_rdevice ||
1518	    info->lo_inode != info64->lo_inode ||
1519	    info->lo_offset != info64->lo_offset)
1520		return -EOVERFLOW;
1521
1522	return 0;
1523}
1524
1525static int
1526loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1527{
1528	struct loop_info info;
1529	struct loop_info64 info64;
1530
1531	if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1532		return -EFAULT;
1533	loop_info64_from_old(&info, &info64);
1534	return loop_set_status(lo, &info64);
1535}
1536
1537static int
1538loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1539{
1540	struct loop_info64 info64;
1541
1542	if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1543		return -EFAULT;
1544	return loop_set_status(lo, &info64);
1545}
1546
1547static int
1548loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1549	struct loop_info info;
1550	struct loop_info64 info64;
1551	int err;
1552
1553	if (!arg)
1554		return -EINVAL;
1555	err = loop_get_status(lo, &info64);
1556	if (!err)
1557		err = loop_info64_to_old(&info64, &info);
1558	if (!err && copy_to_user(arg, &info, sizeof(info)))
1559		err = -EFAULT;
1560
1561	return err;
1562}
1563
1564static int
1565loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1566	struct loop_info64 info64;
1567	int err;
1568
1569	if (!arg)
1570		return -EINVAL;
1571	err = loop_get_status(lo, &info64);
1572	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1573		err = -EFAULT;
1574
1575	return err;
1576}
1577
1578static int loop_set_capacity(struct loop_device *lo)
1579{
1580	loff_t size;
1581
1582	if (unlikely(lo->lo_state != Lo_bound))
1583		return -ENXIO;
1584
1585	size = get_loop_size(lo, lo->lo_backing_file);
1586	loop_set_size(lo, size);
1587
1588	return 0;
1589}
1590
1591static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1592{
1593	int error = -ENXIO;
1594	if (lo->lo_state != Lo_bound)
1595		goto out;
1596
1597	__loop_update_dio(lo, !!arg);
1598	if (lo->use_dio == !!arg)
1599		return 0;
1600	error = -EINVAL;
1601 out:
1602	return error;
1603}
1604
1605static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1606{
1607	int err = 0;
1608
1609	if (lo->lo_state != Lo_bound)
1610		return -ENXIO;
1611
1612	err = blk_validate_block_size(arg);
1613	if (err)
1614		return err;
1615
1616	if (lo->lo_queue->limits.logical_block_size == arg)
1617		return 0;
1618
1619	sync_blockdev(lo->lo_device);
1620	invalidate_bdev(lo->lo_device);
1621
1622	blk_mq_freeze_queue(lo->lo_queue);
1623
1624	/* invalidate_bdev should have truncated all the pages */
1625	if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1626		err = -EAGAIN;
1627		pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1628			__func__, lo->lo_number, lo->lo_file_name,
1629			lo->lo_device->bd_inode->i_mapping->nrpages);
1630		goto out_unfreeze;
1631	}
1632
1633	blk_queue_logical_block_size(lo->lo_queue, arg);
1634	blk_queue_physical_block_size(lo->lo_queue, arg);
1635	blk_queue_io_min(lo->lo_queue, arg);
1636	loop_update_dio(lo);
1637out_unfreeze:
1638	blk_mq_unfreeze_queue(lo->lo_queue);
1639
1640	return err;
1641}
1642
1643static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1644			   unsigned long arg)
1645{
1646	int err;
1647
1648	err = mutex_lock_killable(&loop_ctl_mutex);
1649	if (err)
1650		return err;
1651	switch (cmd) {
1652	case LOOP_SET_CAPACITY:
1653		err = loop_set_capacity(lo);
1654		break;
1655	case LOOP_SET_DIRECT_IO:
1656		err = loop_set_dio(lo, arg);
1657		break;
1658	case LOOP_SET_BLOCK_SIZE:
1659		err = loop_set_block_size(lo, arg);
1660		break;
1661	default:
1662		err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1663	}
1664	mutex_unlock(&loop_ctl_mutex);
1665	return err;
1666}
1667
1668static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1669	unsigned int cmd, unsigned long arg)
1670{
1671	struct loop_device *lo = bdev->bd_disk->private_data;
1672	void __user *argp = (void __user *) arg;
1673	int err;
1674
1675	switch (cmd) {
1676	case LOOP_SET_FD: {
1677		/*
1678		 * Legacy case - pass in a zeroed out struct loop_config with
1679		 * only the file descriptor set , which corresponds with the
1680		 * default parameters we'd have used otherwise.
1681		 */
1682		struct loop_config config;
1683
1684		memset(&config, 0, sizeof(config));
1685		config.fd = arg;
1686
1687		return loop_configure(lo, mode, bdev, &config);
1688	}
1689	case LOOP_CONFIGURE: {
1690		struct loop_config config;
1691
1692		if (copy_from_user(&config, argp, sizeof(config)))
1693			return -EFAULT;
1694
1695		return loop_configure(lo, mode, bdev, &config);
1696	}
1697	case LOOP_CHANGE_FD:
1698		return loop_change_fd(lo, bdev, arg);
1699	case LOOP_CLR_FD:
1700		return loop_clr_fd(lo);
1701	case LOOP_SET_STATUS:
1702		err = -EPERM;
1703		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1704			err = loop_set_status_old(lo, argp);
1705		}
1706		break;
1707	case LOOP_GET_STATUS:
1708		return loop_get_status_old(lo, argp);
1709	case LOOP_SET_STATUS64:
1710		err = -EPERM;
1711		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1712			err = loop_set_status64(lo, argp);
1713		}
1714		break;
1715	case LOOP_GET_STATUS64:
1716		return loop_get_status64(lo, argp);
1717	case LOOP_SET_CAPACITY:
1718	case LOOP_SET_DIRECT_IO:
1719	case LOOP_SET_BLOCK_SIZE:
1720		if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1721			return -EPERM;
1722		fallthrough;
1723	default:
1724		err = lo_simple_ioctl(lo, cmd, arg);
1725		break;
1726	}
1727
1728	return err;
1729}
1730
1731#ifdef CONFIG_COMPAT
1732struct compat_loop_info {
1733	compat_int_t	lo_number;      /* ioctl r/o */
1734	compat_dev_t	lo_device;      /* ioctl r/o */
1735	compat_ulong_t	lo_inode;       /* ioctl r/o */
1736	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1737	compat_int_t	lo_offset;
1738	compat_int_t	lo_encrypt_type;
1739	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1740	compat_int_t	lo_flags;       /* ioctl r/o */
1741	char		lo_name[LO_NAME_SIZE];
1742	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1743	compat_ulong_t	lo_init[2];
1744	char		reserved[4];
1745};
1746
1747/*
1748 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1749 * - noinlined to reduce stack space usage in main part of driver
1750 */
1751static noinline int
1752loop_info64_from_compat(const struct compat_loop_info __user *arg,
1753			struct loop_info64 *info64)
1754{
1755	struct compat_loop_info info;
1756
1757	if (copy_from_user(&info, arg, sizeof(info)))
1758		return -EFAULT;
1759
1760	memset(info64, 0, sizeof(*info64));
1761	info64->lo_number = info.lo_number;
1762	info64->lo_device = info.lo_device;
1763	info64->lo_inode = info.lo_inode;
1764	info64->lo_rdevice = info.lo_rdevice;
1765	info64->lo_offset = info.lo_offset;
1766	info64->lo_sizelimit = 0;
1767	info64->lo_encrypt_type = info.lo_encrypt_type;
1768	info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1769	info64->lo_flags = info.lo_flags;
1770	info64->lo_init[0] = info.lo_init[0];
1771	info64->lo_init[1] = info.lo_init[1];
1772	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1773		memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1774	else
1775		memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1776	memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1777	return 0;
1778}
1779
1780/*
1781 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1782 * - noinlined to reduce stack space usage in main part of driver
1783 */
1784static noinline int
1785loop_info64_to_compat(const struct loop_info64 *info64,
1786		      struct compat_loop_info __user *arg)
1787{
1788	struct compat_loop_info info;
1789
1790	memset(&info, 0, sizeof(info));
1791	info.lo_number = info64->lo_number;
1792	info.lo_device = info64->lo_device;
1793	info.lo_inode = info64->lo_inode;
1794	info.lo_rdevice = info64->lo_rdevice;
1795	info.lo_offset = info64->lo_offset;
1796	info.lo_encrypt_type = info64->lo_encrypt_type;
1797	info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1798	info.lo_flags = info64->lo_flags;
1799	info.lo_init[0] = info64->lo_init[0];
1800	info.lo_init[1] = info64->lo_init[1];
1801	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1802		memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1803	else
1804		memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1805	memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1806
1807	/* error in case values were truncated */
1808	if (info.lo_device != info64->lo_device ||
1809	    info.lo_rdevice != info64->lo_rdevice ||
1810	    info.lo_inode != info64->lo_inode ||
1811	    info.lo_offset != info64->lo_offset ||
1812	    info.lo_init[0] != info64->lo_init[0] ||
1813	    info.lo_init[1] != info64->lo_init[1])
1814		return -EOVERFLOW;
1815
1816	if (copy_to_user(arg, &info, sizeof(info)))
1817		return -EFAULT;
1818	return 0;
1819}
1820
1821static int
1822loop_set_status_compat(struct loop_device *lo,
1823		       const struct compat_loop_info __user *arg)
1824{
1825	struct loop_info64 info64;
1826	int ret;
1827
1828	ret = loop_info64_from_compat(arg, &info64);
1829	if (ret < 0)
1830		return ret;
1831	return loop_set_status(lo, &info64);
1832}
1833
1834static int
1835loop_get_status_compat(struct loop_device *lo,
1836		       struct compat_loop_info __user *arg)
1837{
1838	struct loop_info64 info64;
1839	int err;
1840
1841	if (!arg)
1842		return -EINVAL;
1843	err = loop_get_status(lo, &info64);
1844	if (!err)
1845		err = loop_info64_to_compat(&info64, arg);
1846	return err;
1847}
1848
1849static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1850			   unsigned int cmd, unsigned long arg)
1851{
1852	struct loop_device *lo = bdev->bd_disk->private_data;
1853	int err;
1854
1855	switch(cmd) {
1856	case LOOP_SET_STATUS:
1857		err = loop_set_status_compat(lo,
1858			     (const struct compat_loop_info __user *)arg);
1859		break;
1860	case LOOP_GET_STATUS:
1861		err = loop_get_status_compat(lo,
1862				     (struct compat_loop_info __user *)arg);
1863		break;
1864	case LOOP_SET_CAPACITY:
1865	case LOOP_CLR_FD:
1866	case LOOP_GET_STATUS64:
1867	case LOOP_SET_STATUS64:
1868	case LOOP_CONFIGURE:
1869		arg = (unsigned long) compat_ptr(arg);
1870		fallthrough;
1871	case LOOP_SET_FD:
1872	case LOOP_CHANGE_FD:
1873	case LOOP_SET_BLOCK_SIZE:
1874	case LOOP_SET_DIRECT_IO:
1875		err = lo_ioctl(bdev, mode, cmd, arg);
1876		break;
1877	default:
1878		err = -ENOIOCTLCMD;
1879		break;
1880	}
1881	return err;
1882}
1883#endif
1884
1885static int lo_open(struct block_device *bdev, fmode_t mode)
1886{
1887	struct loop_device *lo;
1888	int err;
1889
1890	err = mutex_lock_killable(&loop_ctl_mutex);
1891	if (err)
1892		return err;
1893	lo = bdev->bd_disk->private_data;
1894	if (!lo) {
1895		err = -ENXIO;
1896		goto out;
1897	}
1898
1899	atomic_inc(&lo->lo_refcnt);
1900out:
1901	mutex_unlock(&loop_ctl_mutex);
1902	return err;
1903}
1904
1905static void lo_release(struct gendisk *disk, fmode_t mode)
1906{
1907	struct loop_device *lo;
1908
1909	mutex_lock(&loop_ctl_mutex);
1910	lo = disk->private_data;
1911	if (atomic_dec_return(&lo->lo_refcnt))
1912		goto out_unlock;
1913
1914	if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1915		if (lo->lo_state != Lo_bound)
1916			goto out_unlock;
1917		lo->lo_state = Lo_rundown;
1918		mutex_unlock(&loop_ctl_mutex);
1919		/*
1920		 * In autoclear mode, stop the loop thread
1921		 * and remove configuration after last close.
1922		 */
1923		__loop_clr_fd(lo, true);
1924		return;
1925	} else if (lo->lo_state == Lo_bound) {
1926		/*
1927		 * Otherwise keep thread (if running) and config,
1928		 * but flush possible ongoing bios in thread.
1929		 */
1930		blk_mq_freeze_queue(lo->lo_queue);
1931		blk_mq_unfreeze_queue(lo->lo_queue);
1932	}
1933
1934out_unlock:
1935	mutex_unlock(&loop_ctl_mutex);
1936}
1937
1938static const struct block_device_operations lo_fops = {
1939	.owner =	THIS_MODULE,
1940	.open =		lo_open,
1941	.release =	lo_release,
1942	.ioctl =	lo_ioctl,
1943#ifdef CONFIG_COMPAT
1944	.compat_ioctl =	lo_compat_ioctl,
1945#endif
1946};
1947
1948/*
1949 * And now the modules code and kernel interface.
1950 */
1951static int max_loop;
1952module_param(max_loop, int, 0444);
1953MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1954module_param(max_part, int, 0444);
1955MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1956MODULE_LICENSE("GPL");
1957MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1958
1959int loop_register_transfer(struct loop_func_table *funcs)
1960{
1961	unsigned int n = funcs->number;
1962
1963	if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1964		return -EINVAL;
1965	xfer_funcs[n] = funcs;
1966	return 0;
1967}
1968
1969static int unregister_transfer_cb(int id, void *ptr, void *data)
1970{
1971	struct loop_device *lo = ptr;
1972	struct loop_func_table *xfer = data;
1973
1974	mutex_lock(&loop_ctl_mutex);
1975	if (lo->lo_encryption == xfer)
1976		loop_release_xfer(lo);
1977	mutex_unlock(&loop_ctl_mutex);
1978	return 0;
1979}
1980
1981int loop_unregister_transfer(int number)
1982{
1983	unsigned int n = number;
1984	struct loop_func_table *xfer;
1985
1986	if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1987		return -EINVAL;
1988
1989	xfer_funcs[n] = NULL;
1990	idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1991	return 0;
1992}
1993
1994EXPORT_SYMBOL(loop_register_transfer);
1995EXPORT_SYMBOL(loop_unregister_transfer);
1996
1997static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1998		const struct blk_mq_queue_data *bd)
1999{
2000	struct request *rq = bd->rq;
2001	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2002	struct loop_device *lo = rq->q->queuedata;
2003
2004	blk_mq_start_request(rq);
2005
2006	if (lo->lo_state != Lo_bound)
2007		return BLK_STS_IOERR;
2008
2009	switch (req_op(rq)) {
2010	case REQ_OP_FLUSH:
2011	case REQ_OP_DISCARD:
2012	case REQ_OP_WRITE_ZEROES:
2013		cmd->use_aio = false;
2014		break;
2015	default:
2016		cmd->use_aio = lo->use_dio;
2017		break;
2018	}
2019
2020	/* always use the first bio's css */
2021#ifdef CONFIG_BLK_CGROUP
2022	if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2023		cmd->css = &bio_blkcg(rq->bio)->css;
2024		css_get(cmd->css);
2025	} else
2026#endif
2027		cmd->css = NULL;
2028	kthread_queue_work(&lo->worker, &cmd->work);
2029
2030	return BLK_STS_OK;
2031}
2032
2033static void loop_handle_cmd(struct loop_cmd *cmd)
2034{
2035	struct request *rq = blk_mq_rq_from_pdu(cmd);
2036	const bool write = op_is_write(req_op(rq));
2037	struct loop_device *lo = rq->q->queuedata;
2038	int ret = 0;
2039
2040	if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2041		ret = -EIO;
2042		goto failed;
2043	}
2044
2045	ret = do_req_filebacked(lo, rq);
2046 failed:
2047	/* complete non-aio request */
2048	if (!cmd->use_aio || ret) {
2049		if (ret == -EOPNOTSUPP)
2050			cmd->ret = ret;
2051		else
2052			cmd->ret = ret ? -EIO : 0;
2053		if (likely(!blk_should_fake_timeout(rq->q)))
2054			blk_mq_complete_request(rq);
2055	}
2056}
2057
2058static void loop_queue_work(struct kthread_work *work)
2059{
2060	struct loop_cmd *cmd =
2061		container_of(work, struct loop_cmd, work);
2062
2063	loop_handle_cmd(cmd);
2064}
2065
2066static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2067		unsigned int hctx_idx, unsigned int numa_node)
2068{
2069	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2070
2071	kthread_init_work(&cmd->work, loop_queue_work);
2072	return 0;
2073}
2074
2075static const struct blk_mq_ops loop_mq_ops = {
2076	.queue_rq       = loop_queue_rq,
2077	.init_request	= loop_init_request,
2078	.complete	= lo_complete_rq,
2079};
2080
2081static int loop_add(struct loop_device **l, int i)
2082{
2083	struct loop_device *lo;
2084	struct gendisk *disk;
2085	int err;
2086
2087	err = -ENOMEM;
2088	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2089	if (!lo)
2090		goto out;
2091
2092	lo->lo_state = Lo_unbound;
2093
2094	/* allocate id, if @id >= 0, we're requesting that specific id */
2095	if (i >= 0) {
2096		err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2097		if (err == -ENOSPC)
2098			err = -EEXIST;
2099	} else {
2100		err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2101	}
2102	if (err < 0)
2103		goto out_free_dev;
2104	i = err;
2105
2106	err = -ENOMEM;
2107	lo->tag_set.ops = &loop_mq_ops;
2108	lo->tag_set.nr_hw_queues = 1;
2109	lo->tag_set.queue_depth = 128;
2110	lo->tag_set.numa_node = NUMA_NO_NODE;
2111	lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2112	lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
2113		BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2114	lo->tag_set.driver_data = lo;
2115
2116	err = blk_mq_alloc_tag_set(&lo->tag_set);
2117	if (err)
2118		goto out_free_idr;
2119
2120	lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2121	if (IS_ERR(lo->lo_queue)) {
2122		err = PTR_ERR(lo->lo_queue);
2123		goto out_cleanup_tags;
2124	}
2125	lo->lo_queue->queuedata = lo;
2126
2127	blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2128
2129	/*
2130	 * By default, we do buffer IO, so it doesn't make sense to enable
2131	 * merge because the I/O submitted to backing file is handled page by
2132	 * page. For directio mode, merge does help to dispatch bigger request
2133	 * to underlayer disk. We will enable merge once directio is enabled.
2134	 */
2135	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2136
2137	err = -ENOMEM;
2138	disk = lo->lo_disk = alloc_disk(1 << part_shift);
2139	if (!disk)
2140		goto out_free_queue;
2141
2142	/*
2143	 * Disable partition scanning by default. The in-kernel partition
2144	 * scanning can be requested individually per-device during its
2145	 * setup. Userspace can always add and remove partitions from all
2146	 * devices. The needed partition minors are allocated from the
2147	 * extended minor space, the main loop device numbers will continue
2148	 * to match the loop minors, regardless of the number of partitions
2149	 * used.
2150	 *
2151	 * If max_part is given, partition scanning is globally enabled for
2152	 * all loop devices. The minors for the main loop devices will be
2153	 * multiples of max_part.
2154	 *
2155	 * Note: Global-for-all-devices, set-only-at-init, read-only module
2156	 * parameteters like 'max_loop' and 'max_part' make things needlessly
2157	 * complicated, are too static, inflexible and may surprise
2158	 * userspace tools. Parameters like this in general should be avoided.
2159	 */
2160	if (!part_shift)
2161		disk->flags |= GENHD_FL_NO_PART_SCAN;
2162	disk->flags |= GENHD_FL_EXT_DEVT;
2163	atomic_set(&lo->lo_refcnt, 0);
2164	lo->lo_number		= i;
2165	spin_lock_init(&lo->lo_lock);
2166	disk->major		= LOOP_MAJOR;
2167	disk->first_minor	= i << part_shift;
2168	disk->fops		= &lo_fops;
2169	disk->private_data	= lo;
2170	disk->queue		= lo->lo_queue;
2171	sprintf(disk->disk_name, "loop%d", i);
2172	add_disk(disk);
2173	*l = lo;
2174	return lo->lo_number;
2175
2176out_free_queue:
2177	blk_cleanup_queue(lo->lo_queue);
2178out_cleanup_tags:
2179	blk_mq_free_tag_set(&lo->tag_set);
2180out_free_idr:
2181	idr_remove(&loop_index_idr, i);
2182out_free_dev:
2183	kfree(lo);
2184out:
2185	return err;
2186}
2187
2188static void loop_remove(struct loop_device *lo)
2189{
2190	del_gendisk(lo->lo_disk);
2191	blk_cleanup_queue(lo->lo_queue);
2192	blk_mq_free_tag_set(&lo->tag_set);
2193	put_disk(lo->lo_disk);
2194	kfree(lo);
2195}
2196
2197static int find_free_cb(int id, void *ptr, void *data)
2198{
2199	struct loop_device *lo = ptr;
2200	struct loop_device **l = data;
2201
2202	if (lo->lo_state == Lo_unbound) {
2203		*l = lo;
2204		return 1;
2205	}
2206	return 0;
2207}
2208
2209static int loop_lookup(struct loop_device **l, int i)
2210{
2211	struct loop_device *lo;
2212	int ret = -ENODEV;
2213
2214	if (i < 0) {
2215		int err;
2216
2217		err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2218		if (err == 1) {
2219			*l = lo;
2220			ret = lo->lo_number;
2221		}
2222		goto out;
2223	}
2224
2225	/* lookup and return a specific i */
2226	lo = idr_find(&loop_index_idr, i);
2227	if (lo) {
2228		*l = lo;
2229		ret = lo->lo_number;
2230	}
2231out:
2232	return ret;
2233}
2234
2235static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2236{
2237	struct loop_device *lo;
2238	struct kobject *kobj;
2239	int err;
2240
2241	mutex_lock(&loop_ctl_mutex);
2242	err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2243	if (err < 0)
2244		err = loop_add(&lo, MINOR(dev) >> part_shift);
2245	if (err < 0)
2246		kobj = NULL;
2247	else
2248		kobj = get_disk_and_module(lo->lo_disk);
2249	mutex_unlock(&loop_ctl_mutex);
2250
2251	*part = 0;
2252	return kobj;
2253}
2254
2255static long loop_control_ioctl(struct file *file, unsigned int cmd,
2256			       unsigned long parm)
2257{
2258	struct loop_device *lo;
2259	int ret;
2260
2261	ret = mutex_lock_killable(&loop_ctl_mutex);
2262	if (ret)
2263		return ret;
2264
2265	ret = -ENOSYS;
2266	switch (cmd) {
2267	case LOOP_CTL_ADD:
2268		ret = loop_lookup(&lo, parm);
2269		if (ret >= 0) {
2270			ret = -EEXIST;
2271			break;
2272		}
2273		ret = loop_add(&lo, parm);
2274		break;
2275	case LOOP_CTL_REMOVE:
2276		ret = loop_lookup(&lo, parm);
2277		if (ret < 0)
2278			break;
2279		if (lo->lo_state != Lo_unbound) {
2280			ret = -EBUSY;
2281			break;
2282		}
2283		if (atomic_read(&lo->lo_refcnt) > 0) {
2284			ret = -EBUSY;
2285			break;
2286		}
2287		lo->lo_disk->private_data = NULL;
2288		idr_remove(&loop_index_idr, lo->lo_number);
2289		loop_remove(lo);
2290		break;
2291	case LOOP_CTL_GET_FREE:
2292		ret = loop_lookup(&lo, -1);
2293		if (ret >= 0)
2294			break;
2295		ret = loop_add(&lo, -1);
2296	}
2297	mutex_unlock(&loop_ctl_mutex);
2298
2299	return ret;
2300}
2301
2302static const struct file_operations loop_ctl_fops = {
2303	.open		= nonseekable_open,
2304	.unlocked_ioctl	= loop_control_ioctl,
2305	.compat_ioctl	= loop_control_ioctl,
2306	.owner		= THIS_MODULE,
2307	.llseek		= noop_llseek,
2308};
2309
2310static struct miscdevice loop_misc = {
2311	.minor		= LOOP_CTRL_MINOR,
2312	.name		= "loop-control",
2313	.fops		= &loop_ctl_fops,
2314};
2315
2316MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2317MODULE_ALIAS("devname:loop-control");
2318
2319static int __init loop_init(void)
2320{
2321	int i, nr;
2322	unsigned long range;
2323	struct loop_device *lo;
2324	int err;
2325
2326	part_shift = 0;
2327	if (max_part > 0) {
2328		part_shift = fls(max_part);
2329
2330		/*
2331		 * Adjust max_part according to part_shift as it is exported
2332		 * to user space so that user can decide correct minor number
2333		 * if [s]he want to create more devices.
2334		 *
2335		 * Note that -1 is required because partition 0 is reserved
2336		 * for the whole disk.
2337		 */
2338		max_part = (1UL << part_shift) - 1;
2339	}
2340
2341	if ((1UL << part_shift) > DISK_MAX_PARTS) {
2342		err = -EINVAL;
2343		goto err_out;
2344	}
2345
2346	if (max_loop > 1UL << (MINORBITS - part_shift)) {
2347		err = -EINVAL;
2348		goto err_out;
2349	}
2350
2351	/*
2352	 * If max_loop is specified, create that many devices upfront.
2353	 * This also becomes a hard limit. If max_loop is not specified,
2354	 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2355	 * init time. Loop devices can be requested on-demand with the
2356	 * /dev/loop-control interface, or be instantiated by accessing
2357	 * a 'dead' device node.
2358	 */
2359	if (max_loop) {
2360		nr = max_loop;
2361		range = max_loop << part_shift;
2362	} else {
2363		nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2364		range = 1UL << MINORBITS;
2365	}
2366
2367	err = misc_register(&loop_misc);
2368	if (err < 0)
2369		goto err_out;
2370
2371
2372	if (register_blkdev(LOOP_MAJOR, "loop")) {
2373		err = -EIO;
2374		goto misc_out;
2375	}
2376
2377	blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2378				  THIS_MODULE, loop_probe, NULL, NULL);
2379
2380	/* pre-create number of devices given by config or max_loop */
2381	mutex_lock(&loop_ctl_mutex);
2382	for (i = 0; i < nr; i++)
2383		loop_add(&lo, i);
2384	mutex_unlock(&loop_ctl_mutex);
2385
2386	printk(KERN_INFO "loop: module loaded\n");
2387	return 0;
2388
2389misc_out:
2390	misc_deregister(&loop_misc);
2391err_out:
2392	return err;
2393}
2394
2395static int loop_exit_cb(int id, void *ptr, void *data)
2396{
2397	struct loop_device *lo = ptr;
2398
2399	loop_remove(lo);
2400	return 0;
2401}
2402
2403static void __exit loop_exit(void)
2404{
2405	unsigned long range;
2406
2407	range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2408
2409	mutex_lock(&loop_ctl_mutex);
2410
2411	idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2412	idr_destroy(&loop_index_idr);
2413
2414	blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2415	unregister_blkdev(LOOP_MAJOR, "loop");
2416
2417	misc_deregister(&loop_misc);
2418
2419	mutex_unlock(&loop_ctl_mutex);
2420}
2421
2422module_init(loop_init);
2423module_exit(loop_exit);
2424
2425#ifndef MODULE
2426static int __init max_loop_setup(char *str)
2427{
2428	max_loop = simple_strtol(str, NULL, 0);
2429	return 1;
2430}
2431
2432__setup("max_loop=", max_loop_setup);
2433#endif
2434