blob: f1d7a4fe30fcd1721eeccc0b86f0d30c12077dd6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
Jan Engelhardt96de0e22007-10-19 23:21:04 +020032 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 *
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
Nick Piggin4e02ed42008-10-29 14:00:55 -070043 * operations write_begin is not available on the backing filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/swap.h>
65#include <linux/slab.h>
David Howells863d5b822006-08-29 19:06:14 +010066#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070068#include <linux/freezer.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020069#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <linux/completion.h>
72#include <linux/highmem.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070073#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020074#include <linux/splice.h>
Milan Brozee862732010-08-23 15:16:00 +020075#include <linux/sysfs.h>
Kay Sievers770fe302011-07-31 22:08:04 +020076#include <linux/miscdevice.h>
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +020077#include <linux/falloc.h>
Al Viro283e7e52015-04-03 15:21:59 -040078#include <linux/uio.h>
Adam Manzanaresd9a08a92018-05-22 10:52:19 -070079#include <linux/ioprio.h>
80
Al Viro83a87612013-05-12 10:14:07 -040081#include "loop.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080083#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Kay Sievers34dd82a2011-07-31 22:08:04 +020085static DEFINE_IDR(loop_index_idr);
Tetsuo Handa310ca162018-11-08 14:01:02 +010086static DEFINE_MUTEX(loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Laurent Vivier476a4812008-03-26 12:11:53 +010088static int max_part;
89static int part_shift;
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091static int transfer_xor(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
Cong Wangcfd80052011-11-25 23:14:18 +080096 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 char *in, *out, *key;
99 int i, keysize;
100
101 if (cmd == READ) {
102 in = raw_buf;
103 out = loop_buf;
104 } else {
105 in = loop_buf;
106 out = raw_buf;
107 }
108
109 key = lo->lo_encrypt_key;
110 keysize = lo->lo_encrypt_key_size;
111 for (i = 0; i < size; i++)
112 *out++ = *in++ ^ key[(i & 511) % keysize];
113
Cong Wangcfd80052011-11-25 23:14:18 +0800114 kunmap_atomic(loop_buf);
115 kunmap_atomic(raw_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 cond_resched();
117 return 0;
118}
119
120static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
121{
122 if (unlikely(info->lo_encrypt_key_size <= 0))
123 return -EINVAL;
124 return 0;
125}
126
127static struct loop_func_table none_funcs = {
128 .number = LO_CRYPT_NONE,
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200129};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131static struct loop_func_table xor_funcs = {
132 .number = LO_CRYPT_XOR,
133 .transfer = transfer_xor,
134 .init = xor_init
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200135};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
137/* xfer_funcs[0] is special - its release function is never called */
138static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
139 &none_funcs,
140 &xor_funcs
141};
142
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100143static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
Guo Chaob7a1da62013-02-21 15:16:50 -0800145 loff_t loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 /* Compute loopsize in bytes */
Guo Chaob7a1da62013-02-21 15:16:50 -0800148 loopsize = i_size_read(file->f_mapping->host);
149 if (offset > 0)
150 loopsize -= offset;
151 /* offset is beyond i_size, weird but possible */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100152 if (loopsize < 0)
153 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100155 if (sizelimit > 0 && sizelimit < loopsize)
156 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 /*
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
160 */
161 return loopsize >> 9;
162}
163
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100164static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100166 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167}
168
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800169static void __loop_update_dio(struct loop_device *lo, bool dio)
170{
171 struct file *file = lo->lo_backing_file;
172 struct address_space *mapping = file->f_mapping;
173 struct inode *inode = mapping->host;
174 unsigned short sb_bsize = 0;
175 unsigned dio_align = 0;
176 bool use_dio;
177
178 if (inode->i_sb->s_bdev) {
179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
180 dio_align = sb_bsize - 1;
181 }
182
183 /*
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
188 *
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
Masahiro Yamada89d790a2017-02-27 14:29:01 -0800191 * of requests in sane applications should be PAGE_SIZE aligned
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800192 */
193 if (dio) {
194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
195 !(lo->lo_offset & dio_align) &&
196 mapping->a_ops->direct_IO &&
197 !lo->transfer)
198 use_dio = true;
199 else
200 use_dio = false;
201 } else {
202 use_dio = false;
203 }
204
205 if (lo->use_dio == use_dio)
206 return;
207
208 /* flush dirty pages before changing direct IO */
209 vfs_fsync(file, 0);
210
211 /*
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
215 */
216 blk_mq_freeze_queue(lo->lo_queue);
217 lo->use_dio = use_dio;
Shaohua Li40326d82017-08-31 22:09:46 -0700218 if (use_dio) {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800220 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700221 } else {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700224 }
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800225 blk_mq_unfreeze_queue(lo->lo_queue);
226}
227
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100228static int
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -0700229figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100230{
231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 sector_t x = (sector_t)size;
Guo Chao7b0576a2013-02-21 15:16:47 -0800233 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 if (unlikely((loff_t)x != size))
236 return -EFBIG;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100237 if (lo->lo_offset != offset)
238 lo->lo_offset = offset;
239 if (lo->lo_sizelimit != sizelimit)
240 lo->lo_sizelimit = sizelimit;
Ken Chen73285082007-05-08 00:28:20 -0700241 set_capacity(lo->lo_disk, x);
Guo Chao7b0576a2013-02-21 15:16:47 -0800242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
248static inline int
249lo_do_transfer(struct loop_device *lo, int cmd,
250 struct page *rpage, unsigned roffs,
251 struct page *lpage, unsigned loffs,
252 int size, sector_t rblock)
253{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200254 int ret;
255
256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
257 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return 0;
259
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock << 9, size);
263 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200266static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200268 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 ssize_t bw;
Al Viro283e7e52015-04-03 15:21:59 -0400270
David Howellsaa563d72018-10-20 00:57:56 +0100271 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Al Viro03d95eb2013-03-20 13:04:20 -0400273 file_start_write(file);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300274 bw = vfs_iter_write(file, &i, ppos, 0);
Al Viro03d95eb2013-03-20 13:04:20 -0400275 file_end_write(file);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200276
277 if (likely(bw == bvec->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return 0;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200279
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 if (bw >= 0)
284 bw = -EIO;
285 return bw;
286}
287
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200288static int lo_write_simple(struct loop_device *lo, struct request *rq,
289 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200291 struct bio_vec bvec;
292 struct req_iterator iter;
293 int ret = 0;
294
295 rq_for_each_segment(bvec, rq, iter) {
296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
297 if (ret < 0)
298 break;
299 cond_resched();
300 }
301
302 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303}
304
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200305/*
Christoph Hellwig456be142011-10-17 12:57:20 +0200306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 */
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200310static int lo_write_transfer(struct loop_device *lo, struct request *rq,
311 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200313 struct bio_vec bvec, b;
Ming Lei30112012014-12-31 13:22:58 +0000314 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200315 struct page *page;
Kent Overstreet79886132013-11-23 17:19:00 -0800316 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200318 page = alloc_page(GFP_NOIO);
319 if (unlikely(!page))
320 return -ENOMEM;
Christoph Hellwig456be142011-10-17 12:57:20 +0200321
Ming Lei30112012014-12-31 13:22:58 +0000322 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
324 bvec.bv_offset, bvec.bv_len, pos >> 9);
325 if (unlikely(ret))
326 break;
327
328 b.bv_page = page;
329 b.bv_offset = 0;
330 b.bv_len = bvec.bv_len;
331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (ret < 0)
333 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200335
336 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200340static int lo_read_simple(struct loop_device *lo, struct request *rq,
341 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Kent Overstreet79886132013-11-23 17:19:00 -0800343 struct bio_vec bvec;
Ming Lei30112012014-12-31 13:22:58 +0000344 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200345 struct iov_iter i;
346 ssize_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Ming Lei30112012014-12-31 13:22:58 +0000348 rq_for_each_segment(bvec, rq, iter) {
David Howellsaa563d72018-10-20 00:57:56 +0100349 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200351 if (len < 0)
352 return len;
Dave Young306df072012-02-08 22:07:19 +0100353
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200354 flush_dcache_page(bvec.bv_page);
355
356 if (len != bvec.bv_len) {
Ming Lei30112012014-12-31 13:22:58 +0000357 struct bio *bio;
358
359 __rq_for_each_bio(bio, rq)
360 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 break;
Dave Young306df072012-02-08 22:07:19 +0100362 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200363 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200365
Dave Young306df072012-02-08 22:07:19 +0100366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200369static int lo_read_transfer(struct loop_device *lo, struct request *rq,
370 loff_t pos)
371{
372 struct bio_vec bvec, b;
373 struct req_iterator iter;
374 struct iov_iter i;
375 struct page *page;
376 ssize_t len;
377 int ret = 0;
378
379 page = alloc_page(GFP_NOIO);
380 if (unlikely(!page))
381 return -ENOMEM;
382
383 rq_for_each_segment(bvec, rq, iter) {
384 loff_t offset = pos;
385
386 b.bv_page = page;
387 b.bv_offset = 0;
388 b.bv_len = bvec.bv_len;
389
David Howellsaa563d72018-10-20 00:57:56 +0100390 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200392 if (len < 0) {
393 ret = len;
394 goto out_free_page;
395 }
396
397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
398 bvec.bv_offset, len, offset >> 9);
399 if (ret)
400 goto out_free_page;
401
402 flush_dcache_page(bvec.bv_page);
403
404 if (len != bvec.bv_len) {
405 struct bio *bio;
406
407 __rq_for_each_bio(bio, rq)
408 zero_fill_bio(bio);
409 break;
410 }
411 }
412
413 ret = 0;
414out_free_page:
415 __free_page(page);
416 return ret;
417}
418
Ming Leicf655d92014-12-31 13:22:59 +0000419static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
420{
421 /*
422 * We use punch hole to reclaim the free space used by the
423 * image a.k.a. discard. However we do not support discard if
424 * encryption is enabled, because it may give an attacker
425 * useful information.
426 */
427 struct file *file = lo->lo_backing_file;
428 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
429 int ret;
430
431 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
432 ret = -EOPNOTSUPP;
433 goto out;
434 }
435
436 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
437 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
438 ret = -EIO;
439 out:
440 return ret;
441}
442
443static int lo_req_flush(struct loop_device *lo, struct request *rq)
444{
445 struct file *file = lo->lo_backing_file;
446 int ret = vfs_fsync(file, 0);
447 if (unlikely(ret && ret != -EINVAL))
448 ret = -EIO;
449
450 return ret;
451}
452
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200453static void lo_complete_rq(struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800454{
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200455 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Jens Axboef9de14b2018-04-13 16:25:57 -0600456 blk_status_t ret = BLK_STS_OK;
Ming Leibc07c102015-08-17 10:31:51 +0800457
Jens Axboef9de14b2018-04-13 16:25:57 -0600458 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
459 req_op(rq) != REQ_OP_READ) {
460 if (cmd->ret < 0)
461 ret = BLK_STS_IOERR;
462 goto end_io;
Ming Leibc07c102015-08-17 10:31:51 +0800463 }
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200464
Jens Axboef9de14b2018-04-13 16:25:57 -0600465 /*
466 * Short READ - if we got some data, advance our request and
467 * retry it. If we got no data, end the rest with EIO.
468 */
469 if (cmd->ret) {
470 blk_update_request(rq, BLK_STS_OK, cmd->ret);
471 cmd->ret = 0;
472 blk_mq_requeue_request(rq, true);
473 } else {
474 if (cmd->use_aio) {
475 struct bio *bio = rq->bio;
476
477 while (bio) {
478 zero_fill_bio(bio);
479 bio = bio->bi_next;
480 }
481 }
482 ret = BLK_STS_IOERR;
483end_io:
484 blk_mq_end_request(rq, ret);
485 }
Ming Leibc07c102015-08-17 10:31:51 +0800486}
487
Shaohua Li92d77332017-09-01 11:15:17 -0700488static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
489{
Jens Axboe1894e912018-04-13 16:24:29 -0600490 struct request *rq = blk_mq_rq_from_pdu(cmd);
491
Shaohua Li92d77332017-09-01 11:15:17 -0700492 if (!atomic_dec_and_test(&cmd->ref))
493 return;
494 kfree(cmd->bvec);
495 cmd->bvec = NULL;
Jens Axboe1894e912018-04-13 16:24:29 -0600496 blk_mq_complete_request(rq);
Shaohua Li92d77332017-09-01 11:15:17 -0700497}
498
Ming Leibc07c102015-08-17 10:31:51 +0800499static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
500{
501 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
Ming Leibc07c102015-08-17 10:31:51 +0800502
Shaohua Lid4478e92017-09-25 13:07:22 -0600503 if (cmd->css)
504 css_put(cmd->css);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200505 cmd->ret = ret;
Shaohua Li92d77332017-09-01 11:15:17 -0700506 lo_rw_aio_do_completion(cmd);
Ming Leibc07c102015-08-17 10:31:51 +0800507}
508
509static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
510 loff_t pos, bool rw)
511{
512 struct iov_iter iter;
513 struct bio_vec *bvec;
Jens Axboe1894e912018-04-13 16:24:29 -0600514 struct request *rq = blk_mq_rq_from_pdu(cmd);
Shaohua Li40326d82017-08-31 22:09:46 -0700515 struct bio *bio = rq->bio;
Ming Leibc07c102015-08-17 10:31:51 +0800516 struct file *file = lo->lo_backing_file;
Shaohua Li40326d82017-08-31 22:09:46 -0700517 unsigned int offset;
518 int segments = 0;
Ming Leibc07c102015-08-17 10:31:51 +0800519 int ret;
520
Shaohua Li40326d82017-08-31 22:09:46 -0700521 if (rq->bio != rq->biotail) {
522 struct req_iterator iter;
523 struct bio_vec tmp;
Ming Leibc07c102015-08-17 10:31:51 +0800524
Shaohua Li40326d82017-08-31 22:09:46 -0700525 __rq_for_each_bio(bio, rq)
526 segments += bio_segments(bio);
Kees Cook6da2ec52018-06-12 13:55:00 -0700527 bvec = kmalloc_array(segments, sizeof(struct bio_vec),
528 GFP_NOIO);
Shaohua Li40326d82017-08-31 22:09:46 -0700529 if (!bvec)
530 return -EIO;
531 cmd->bvec = bvec;
532
533 /*
534 * The bios of the request may be started from the middle of
535 * the 'bvec' because of bio splitting, so we can't directly
536 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
537 * API will take care of all details for us.
538 */
539 rq_for_each_segment(tmp, rq, iter) {
540 *bvec = tmp;
541 bvec++;
542 }
543 bvec = cmd->bvec;
544 offset = 0;
545 } else {
546 /*
547 * Same here, this bio may be started from the middle of the
548 * 'bvec' because of bio splitting, so offset from the bvec
549 * must be passed to iov iterator
550 */
551 offset = bio->bi_iter.bi_bvec_done;
552 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
553 segments = bio_segments(bio);
554 }
Shaohua Li92d77332017-09-01 11:15:17 -0700555 atomic_set(&cmd->ref, 2);
Shaohua Li40326d82017-08-31 22:09:46 -0700556
David Howellsaa563d72018-10-20 00:57:56 +0100557 iov_iter_bvec(&iter, rw, bvec, segments, blk_rq_bytes(rq));
Shaohua Li40326d82017-08-31 22:09:46 -0700558 iter.iov_offset = offset;
Ming Leibc07c102015-08-17 10:31:51 +0800559
560 cmd->iocb.ki_pos = pos;
561 cmd->iocb.ki_filp = file;
562 cmd->iocb.ki_complete = lo_rw_aio_complete;
563 cmd->iocb.ki_flags = IOCB_DIRECT;
Adam Manzanaresd9a08a92018-05-22 10:52:19 -0700564 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
Shaohua Lid4478e92017-09-25 13:07:22 -0600565 if (cmd->css)
566 kthread_associate_blkcg(cmd->css);
Ming Leibc07c102015-08-17 10:31:51 +0800567
568 if (rw == WRITE)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100569 ret = call_write_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800570 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100571 ret = call_read_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800572
Shaohua Li92d77332017-09-01 11:15:17 -0700573 lo_rw_aio_do_completion(cmd);
Shaohua Lid4478e92017-09-25 13:07:22 -0600574 kthread_associate_blkcg(NULL);
Shaohua Li92d77332017-09-01 11:15:17 -0700575
Ming Leibc07c102015-08-17 10:31:51 +0800576 if (ret != -EIOCBQUEUED)
577 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
578 return 0;
579}
580
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200581static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800582{
583 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200584 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800585
586 /*
587 * lo_write_simple and lo_read_simple should have been covered
588 * by io submit style function like lo_rw_aio(), one blocker
589 * is that lo_read_simple() need to call flush_dcache_page after
590 * the page is written from kernel, and it isn't easy to handle
591 * this in io submit style function which submits all segments
592 * of the req at one time. And direct read IO doesn't need to
593 * run flush_dcache_page().
594 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200595 switch (req_op(rq)) {
596 case REQ_OP_FLUSH:
597 return lo_req_flush(lo, rq);
598 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +0200599 case REQ_OP_WRITE_ZEROES:
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200600 return lo_discard(lo, rq, pos);
601 case REQ_OP_WRITE:
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200602 if (lo->transfer)
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200603 return lo_write_transfer(lo, rq, pos);
604 else if (cmd->use_aio)
605 return lo_rw_aio(lo, cmd, pos, WRITE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200606 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200607 return lo_write_simple(lo, rq, pos);
608 case REQ_OP_READ:
609 if (lo->transfer)
610 return lo_read_transfer(lo, rq, pos);
611 else if (cmd->use_aio)
612 return lo_rw_aio(lo, cmd, pos, READ);
613 else
614 return lo_read_simple(lo, rq, pos);
615 default:
616 WARN_ON_ONCE(1);
617 return -EIO;
618 break;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800622static inline void loop_update_dio(struct loop_device *lo)
623{
624 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
625 lo->use_dio);
626}
627
Ming Lei06f0e9e2015-05-06 12:26:24 +0800628static void loop_reread_partitions(struct loop_device *lo,
629 struct block_device *bdev)
630{
631 int rc;
632
Jan Karad57f3372018-11-08 14:01:12 +0100633 rc = blkdev_reread_part(bdev);
Ming Lei06f0e9e2015-05-06 12:26:24 +0800634 if (rc)
635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
636 __func__, lo->lo_number, lo->lo_file_name, rc);
637}
638
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400639static inline int is_loop_device(struct file *file)
640{
641 struct inode *i = file->f_mapping->host;
642
643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
644}
645
646static int loop_validate_file(struct file *file, struct block_device *bdev)
647{
648 struct inode *inode = file->f_mapping->host;
649 struct file *f = file;
650
651 /* Avoid recursion */
652 while (is_loop_device(f)) {
653 struct loop_device *l;
654
655 if (f->f_mapping->host->i_bdev == bdev)
656 return -EBADF;
657
658 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
659 if (l->lo_state == Lo_unbound) {
660 return -EINVAL;
661 }
662 f = l->lo_backing_file;
663 }
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
665 return -EINVAL;
666 return 0;
667}
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669/*
670 * loop_change_fd switched the backing store of a loopback device to
671 * a new file. This is useful for operating system installers to free up
672 * the original file and in High Availability environments to switch to
673 * an alternative location for the content in case of server meltdown.
674 * This can only work if the loop device is used read-only, and if the
675 * new backing store is the same size and type as the old backing store.
676 */
Al Virobb214882008-03-02 09:29:48 -0500677static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
678 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 struct file *file, *old_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 int error;
682
Jan Karac3710772018-11-08 14:01:11 +0100683 error = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
684 if (error)
685 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 error = -ENXIO;
687 if (lo->lo_state != Lo_bound)
Jan Karac3710772018-11-08 14:01:11 +0100688 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 /* the loop device has to be read-only */
691 error = -EINVAL;
692 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
Jan Karac3710772018-11-08 14:01:11 +0100693 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695 error = -EBADF;
696 file = fget(arg);
697 if (!file)
Jan Karac3710772018-11-08 14:01:11 +0100698 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400700 error = loop_validate_file(file, bdev);
701 if (error)
702 goto out_putf;
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 old_file = lo->lo_backing_file;
705
706 error = -EINVAL;
707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 /* size of the new backing store needs to be the same */
709 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
710 goto out_putf;
711
712 /* and ... switch */
Omar Sandoval43cade802017-08-24 00:03:44 -0700713 blk_mq_freeze_queue(lo->lo_queue);
714 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
715 lo->lo_backing_file = file;
716 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
717 mapping_set_gfp_mask(file->f_mapping,
718 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
719 loop_update_dio(lo);
720 blk_mq_unfreeze_queue(lo->lo_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200723 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800724 loop_reread_partitions(lo, bdev);
Jan Karac3710772018-11-08 14:01:11 +0100725 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return 0;
727
Jan Karac3710772018-11-08 14:01:11 +0100728out_putf:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 fput(file);
Jan Karac3710772018-11-08 14:01:11 +0100730out_unlock:
731 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return error;
733}
734
Milan Brozee862732010-08-23 15:16:00 +0200735/* loop sysfs attributes */
736
737static ssize_t loop_attr_show(struct device *dev, char *page,
738 ssize_t (*callback)(struct loop_device *, char *))
739{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200740 struct gendisk *disk = dev_to_disk(dev);
741 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200742
Kay Sievers34dd82a2011-07-31 22:08:04 +0200743 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200744}
745
746#define LOOP_ATTR_RO(_name) \
747static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
748static ssize_t loop_attr_do_show_##_name(struct device *d, \
749 struct device_attribute *attr, char *b) \
750{ \
751 return loop_attr_show(d, b, loop_attr_##_name##_show); \
752} \
753static struct device_attribute loop_attr_##_name = \
Joe Perches5657a812018-05-24 13:38:59 -0600754 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
Milan Brozee862732010-08-23 15:16:00 +0200755
756static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
757{
758 ssize_t ret;
759 char *p = NULL;
760
Kay Sievers05eb0f22011-07-31 22:21:35 +0200761 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200762 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200763 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200764 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200765
766 if (IS_ERR_OR_NULL(p))
767 ret = PTR_ERR(p);
768 else {
769 ret = strlen(p);
770 memmove(buf, p, ret);
771 buf[ret++] = '\n';
772 buf[ret] = 0;
773 }
774
775 return ret;
776}
777
778static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
779{
780 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
781}
782
783static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
784{
785 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
786}
787
788static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
789{
790 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
791
792 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
793}
794
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200795static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
796{
797 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
798
799 return sprintf(buf, "%s\n", partscan ? "1" : "0");
800}
801
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800802static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
803{
804 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
805
806 return sprintf(buf, "%s\n", dio ? "1" : "0");
807}
808
Milan Brozee862732010-08-23 15:16:00 +0200809LOOP_ATTR_RO(backing_file);
810LOOP_ATTR_RO(offset);
811LOOP_ATTR_RO(sizelimit);
812LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200813LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800814LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200815
816static struct attribute *loop_attrs[] = {
817 &loop_attr_backing_file.attr,
818 &loop_attr_offset.attr,
819 &loop_attr_sizelimit.attr,
820 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200821 &loop_attr_partscan.attr,
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800822 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200823 NULL,
824};
825
826static struct attribute_group loop_attribute_group = {
827 .name = "loop",
828 .attrs= loop_attrs,
829};
830
Tetsuo Handad3349b62018-05-04 10:58:09 -0600831static void loop_sysfs_init(struct loop_device *lo)
Milan Brozee862732010-08-23 15:16:00 +0200832{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600833 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
834 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200835}
836
837static void loop_sysfs_exit(struct loop_device *lo)
838{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600839 if (lo->sysfs_inited)
840 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
841 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200842}
843
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200844static void loop_config_discard(struct loop_device *lo)
845{
846 struct file *file = lo->lo_backing_file;
847 struct inode *inode = file->f_mapping->host;
848 struct request_queue *q = lo->lo_queue;
849
850 /*
851 * We use punch hole to reclaim the free space used by the
Olaf Hering12a64d22014-01-21 14:39:24 -0800852 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200853 * encryption is enabled, because it may give an attacker
854 * useful information.
855 */
856 if ((!file->f_op->fallocate) ||
857 lo->lo_encrypt_key_size) {
858 q->limits.discard_granularity = 0;
859 q->limits.discard_alignment = 0;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600860 blk_queue_max_discard_sectors(q, 0);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200861 blk_queue_max_write_zeroes_sectors(q, 0);
Bart Van Assche8b904b52018-03-07 17:10:10 -0800862 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200863 return;
864 }
865
866 q->limits.discard_granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaf3c02011-12-02 14:47:03 +0100867 q->limits.discard_alignment = 0;
Hannes Reineckef2c6df72017-06-08 13:46:45 +0200868
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -0700869 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
870 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
Bart Van Assche8b904b52018-03-07 17:10:10 -0800871 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200872}
873
Ming Leie03a3d72015-08-17 10:31:48 +0800874static void loop_unprepare_queue(struct loop_device *lo)
875{
Petr Mladek39891442016-10-11 13:55:20 -0700876 kthread_flush_worker(&lo->worker);
Ming Leie03a3d72015-08-17 10:31:48 +0800877 kthread_stop(lo->worker_task);
878}
879
NeilBrownb2ee7d42017-06-16 15:02:09 +1000880static int loop_kthread_worker_fn(void *worker_ptr)
881{
882 current->flags |= PF_LESS_THROTTLE;
883 return kthread_worker_fn(worker_ptr);
884}
885
Ming Leie03a3d72015-08-17 10:31:48 +0800886static int loop_prepare_queue(struct loop_device *lo)
887{
Petr Mladek39891442016-10-11 13:55:20 -0700888 kthread_init_worker(&lo->worker);
NeilBrownb2ee7d42017-06-16 15:02:09 +1000889 lo->worker_task = kthread_run(loop_kthread_worker_fn,
Ming Leie03a3d72015-08-17 10:31:48 +0800890 &lo->worker, "loop%d", lo->lo_number);
891 if (IS_ERR(lo->worker_task))
892 return -ENOMEM;
893 set_user_nice(lo->worker_task, MIN_NICE);
894 return 0;
895}
896
Al Virobb214882008-03-02 09:29:48 -0500897static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 struct block_device *bdev, unsigned int arg)
899{
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400900 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 struct inode *inode;
902 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 int lo_flags = 0;
904 int error;
905 loff_t size;
906
907 /* This is safe, since we have a reference from open(). */
908 __module_get(THIS_MODULE);
909
910 error = -EBADF;
911 file = fget(arg);
912 if (!file)
913 goto out;
914
Jan Kara757ecf42018-11-08 14:01:10 +0100915 error = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
916 if (error)
917 goto out_putf;
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 error = -EBUSY;
920 if (lo->lo_state != Lo_unbound)
Jan Kara757ecf42018-11-08 14:01:10 +0100921 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400923 error = loop_validate_file(file, bdev);
924 if (error)
Jan Kara757ecf42018-11-08 14:01:10 +0100925 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 mapping = file->f_mapping;
928 inode = mapping->host;
929
Christoph Hellwig456be142011-10-17 12:57:20 +0200930 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
Al Viro283e7e52015-04-03 15:21:59 -0400931 !file->f_op->write_iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 lo_flags |= LO_FLAGS_READ_ONLY;
933
Christoph Hellwig456be142011-10-17 12:57:20 +0200934 error = -EFBIG;
935 size = get_loop_size(lo, file);
936 if ((loff_t)(sector_t)size != size)
Jan Kara757ecf42018-11-08 14:01:10 +0100937 goto out_unlock;
Ming Leie03a3d72015-08-17 10:31:48 +0800938 error = loop_prepare_queue(lo);
939 if (error)
Jan Kara757ecf42018-11-08 14:01:10 +0100940 goto out_unlock;
Christoph Hellwig456be142011-10-17 12:57:20 +0200941
942 error = 0;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
945
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800946 lo->use_dio = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 lo->lo_device = bdev;
948 lo->lo_flags = lo_flags;
949 lo->lo_backing_file = file;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200950 lo->transfer = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 lo->ioctl = NULL;
952 lo->lo_sizelimit = 0;
953 lo->old_gfp_mask = mapping_gfp_mask(mapping);
954 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
955
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100956 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Jens Axboe21d07272016-03-30 10:09:35 -0600957 blk_queue_write_cache(lo->lo_queue, true, false);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100958
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800959 loop_update_dio(lo);
Ken Chen73285082007-05-08 00:28:20 -0700960 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200962 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200963 /* let user-space know about the new size */
964 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Omar Sandoval8a0740c2017-08-24 00:03:41 -0700966 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
967 block_size(inode->i_bdev) : PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700969 lo->lo_state = Lo_bound;
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200970 if (part_shift)
971 lo->lo_flags |= LO_FLAGS_PARTSCAN;
972 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800973 loop_reread_partitions(lo, bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700974
975 /* Grab the block_device to prevent its destruction after we
Jan Karaa2505b72018-11-08 14:01:06 +0100976 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700977 */
978 bdgrab(bdev);
Jan Kara757ecf42018-11-08 14:01:10 +0100979 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 return 0;
981
Jan Kara757ecf42018-11-08 14:01:10 +0100982out_unlock:
983 mutex_unlock(&loop_ctl_mutex);
984out_putf:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 fput(file);
Jan Kara757ecf42018-11-08 14:01:10 +0100986out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /* This is safe: open() is still holding a reference. */
988 module_put(THIS_MODULE);
989 return error;
990}
991
992static int
993loop_release_xfer(struct loop_device *lo)
994{
995 int err = 0;
996 struct loop_func_table *xfer = lo->lo_encryption;
997
998 if (xfer) {
999 if (xfer->release)
1000 err = xfer->release(lo);
1001 lo->transfer = NULL;
1002 lo->lo_encryption = NULL;
1003 module_put(xfer->owner);
1004 }
1005 return err;
1006}
1007
1008static int
1009loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1010 const struct loop_info64 *i)
1011{
1012 int err = 0;
1013
1014 if (xfer) {
1015 struct module *owner = xfer->owner;
1016
1017 if (!try_module_get(owner))
1018 return -EINVAL;
1019 if (xfer->init)
1020 err = xfer->init(lo, i);
1021 if (err)
1022 module_put(owner);
1023 else
1024 lo->lo_encryption = xfer;
1025 }
1026 return err;
1027}
1028
Jan Karaa2505b72018-11-08 14:01:06 +01001029static int __loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
Jan Kara7ccd0792018-11-08 14:01:07 +01001031 struct file *filp = NULL;
Al Virob4e3ca12005-10-21 03:22:34 -04001032 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +02001033 struct block_device *bdev = lo->lo_device;
Jan Kara7ccd0792018-11-08 14:01:07 +01001034 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Jan Kara7ccd0792018-11-08 14:01:07 +01001036 mutex_lock(&loop_ctl_mutex);
1037 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1038 err = -ENXIO;
1039 goto out_unlock;
1040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Jan Kara7ccd0792018-11-08 14:01:07 +01001042 filp = lo->lo_backing_file;
1043 if (filp == NULL) {
1044 err = -EINVAL;
1045 goto out_unlock;
1046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Ming Leif8933662015-05-06 12:26:23 +08001048 /* freeze request queue during the transition */
1049 blk_mq_freeze_queue(lo->lo_queue);
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001053 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 loop_release_xfer(lo);
1056 lo->transfer = NULL;
1057 lo->ioctl = NULL;
1058 lo->lo_device = NULL;
1059 lo->lo_encryption = NULL;
1060 lo->lo_offset = 0;
1061 lo->lo_sizelimit = 0;
1062 lo->lo_encrypt_key_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1064 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1065 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001066 blk_queue_logical_block_size(lo->lo_queue, 512);
Omar Sandovalbf093752017-09-05 14:24:47 -07001067 blk_queue_physical_block_size(lo->lo_queue, 512);
1068 blk_queue_io_min(lo->lo_queue, 512);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001069 if (bdev) {
1070 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -05001071 invalidate_bdev(bdev);
Jeff Laytoneedffa22018-05-21 14:35:03 -04001072 bdev->bd_inode->i_mapping->wb_err = 0;
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001073 }
Ken Chen73285082007-05-08 00:28:20 -07001074 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001075 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001076 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001077 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001078 /* let user-space know about this change */
1079 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1080 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 mapping_set_gfp_mask(filp->f_mapping, gfp);
1082 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 /* This is safe: open() is still holding a reference. */
1084 module_put(THIS_MODULE);
Ming Leif8933662015-05-06 12:26:23 +08001085 blk_mq_unfreeze_queue(lo->lo_queue);
1086
Jan Karad57f3372018-11-08 14:01:12 +01001087 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) {
1088 /*
1089 * bd_mutex has been held already in release path, so don't
1090 * acquire it if this function is called in such case.
1091 *
1092 * If the reread partition isn't from release path, lo_refcnt
1093 * must be at least one and it can only become zero when the
1094 * current holder is released.
1095 */
1096 if (!atomic_read(&lo->lo_refcnt))
1097 err = __blkdev_reread_part(bdev);
1098 else
1099 err = blkdev_reread_part(bdev);
1100 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1101 __func__, lo->lo_number, err);
1102 /* Device is gone, no point in returning error */
1103 err = 0;
1104 }
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001105 lo->lo_flags = 0;
1106 if (!part_shift)
1107 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Ming Leie03a3d72015-08-17 10:31:48 +08001108 loop_unprepare_queue(lo);
Jan Kara7ccd0792018-11-08 14:01:07 +01001109out_unlock:
Tetsuo Handa310ca162018-11-08 14:01:02 +01001110 mutex_unlock(&loop_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001111 /*
Tetsuo Handa310ca162018-11-08 14:01:02 +01001112 * Need not hold loop_ctl_mutex to fput backing file.
1113 * Calling fput holding loop_ctl_mutex triggers a circular
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001114 * lock dependency possibility warning as fput can take
Tetsuo Handa310ca162018-11-08 14:01:02 +01001115 * bd_mutex which is usually taken before loop_ctl_mutex.
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001116 */
Jan Kara7ccd0792018-11-08 14:01:07 +01001117 if (filp)
1118 fput(filp);
1119 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120}
1121
Jan Karaa2505b72018-11-08 14:01:06 +01001122static int loop_clr_fd(struct loop_device *lo)
1123{
Jan Kara7ccd0792018-11-08 14:01:07 +01001124 int err;
1125
1126 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1127 if (err)
1128 return err;
1129 if (lo->lo_state != Lo_bound) {
1130 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001131 return -ENXIO;
Jan Kara7ccd0792018-11-08 14:01:07 +01001132 }
Jan Karaa2505b72018-11-08 14:01:06 +01001133 /*
1134 * If we've explicitly asked to tear down the loop device,
1135 * and it has an elevated reference count, set it for auto-teardown when
1136 * the last reference goes away. This stops $!~#$@ udev from
1137 * preventing teardown because it decided that it needs to run blkid on
1138 * the loopback device whenever they appear. xfstests is notorious for
1139 * failing tests because blkid via udev races with a losetup
1140 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1141 * command to fail with EBUSY.
1142 */
1143 if (atomic_read(&lo->lo_refcnt) > 1) {
1144 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1145 mutex_unlock(&loop_ctl_mutex);
1146 return 0;
1147 }
1148 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001149 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001150
1151 return __loop_clr_fd(lo);
1152}
1153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154static int
1155loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1156{
1157 int err;
1158 struct loop_func_table *xfer;
Eric W. Biedermane4849732012-02-11 11:23:51 -08001159 kuid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Jan Kara550df5fd2018-11-08 14:01:09 +01001161 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1162 if (err)
1163 return err;
David Howellsb0fafa82008-11-14 10:38:41 +11001164 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001165 !uid_eq(lo->lo_key_owner, uid) &&
Jan Kara550df5fd2018-11-08 14:01:09 +01001166 !capable(CAP_SYS_ADMIN)) {
1167 err = -EPERM;
1168 goto out_unlock;
1169 }
1170 if (lo->lo_state != Lo_bound) {
1171 err = -ENXIO;
1172 goto out_unlock;
1173 }
1174 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1175 err = -EINVAL;
1176 goto out_unlock;
1177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Ming Leiecdd0952017-02-11 11:40:45 +08001179 /* I/O need to be drained during transfer transition */
1180 blk_mq_freeze_queue(lo->lo_queue);
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 err = loop_release_xfer(lo);
1183 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001184 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186 if (info->lo_encrypt_type) {
1187 unsigned int type = info->lo_encrypt_type;
1188
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001189 if (type >= MAX_LO_CRYPT) {
1190 err = -EINVAL;
Jan Kara550df5fd2018-11-08 14:01:09 +01001191 goto out_unfreeze;
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 xfer = xfer_funcs[type];
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001194 if (xfer == NULL) {
1195 err = -EINVAL;
Jan Kara550df5fd2018-11-08 14:01:09 +01001196 goto out_unfreeze;
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 } else
1199 xfer = NULL;
1200
1201 err = loop_init_xfer(lo, xfer, info);
1202 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001203 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
1205 if (lo->lo_offset != info->lo_offset ||
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001206 lo->lo_sizelimit != info->lo_sizelimit) {
1207 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
Ming Leiecdd0952017-02-11 11:40:45 +08001208 err = -EFBIG;
Jan Kara550df5fd2018-11-08 14:01:09 +01001209 goto out_unfreeze;
Ming Leiecdd0952017-02-11 11:40:45 +08001210 }
Arnd Bergmannb040ad92017-06-09 12:19:18 +02001211 }
Guo Chao541c7422013-02-21 15:16:46 -08001212
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001213 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
1215 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1216 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1217 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1218 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1219
1220 if (!xfer)
1221 xfer = &none_funcs;
1222 lo->transfer = xfer->transfer;
1223 lo->ioctl = xfer->ioctl;
1224
David Woodhouse96c58652008-02-06 01:36:27 -08001225 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1226 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1227 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1230 lo->lo_init[0] = info->lo_init[0];
1231 lo->lo_init[1] = info->lo_init[1];
1232 if (info->lo_encrypt_key_size) {
1233 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1234 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001235 lo->lo_key_owner = uid;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +02001236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Ming Lei2e5ab5f2015-08-17 10:31:49 +08001238 /* update dio if lo_offset or transfer is changed */
1239 __loop_update_dio(lo, lo->use_dio);
1240
Jan Kara550df5fd2018-11-08 14:01:09 +01001241out_unfreeze:
Ming Leiecdd0952017-02-11 11:40:45 +08001242 blk_mq_unfreeze_queue(lo->lo_queue);
Omar Sandovale02898b42017-03-01 10:42:38 -08001243
1244 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1245 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1246 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1247 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1248 loop_reread_partitions(lo, lo->lo_device);
1249 }
Jan Kara550df5fd2018-11-08 14:01:09 +01001250out_unlock:
1251 mutex_unlock(&loop_ctl_mutex);
Omar Sandovale02898b42017-03-01 10:42:38 -08001252
Ming Leiecdd0952017-02-11 11:40:45 +08001253 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254}
1255
1256static int
1257loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1258{
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001259 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 struct kstat stat;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001261 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001263 ret = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1264 if (ret)
1265 return ret;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001266 if (lo->lo_state != Lo_bound) {
Tetsuo Handa310ca162018-11-08 14:01:02 +01001267 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return -ENXIO;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001269 }
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 memset(info, 0, sizeof(*info));
1272 info->lo_number = lo->lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 info->lo_offset = lo->lo_offset;
1274 info->lo_sizelimit = lo->lo_sizelimit;
1275 info->lo_flags = lo->lo_flags;
1276 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1277 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1278 info->lo_encrypt_type =
1279 lo->lo_encryption ? lo->lo_encryption->number : 0;
1280 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1281 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1282 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1283 lo->lo_encrypt_key_size);
1284 }
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001285
Tetsuo Handa310ca162018-11-08 14:01:02 +01001286 /* Drop loop_ctl_mutex while we call into the filesystem. */
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001287 path = lo->lo_backing_file->f_path;
1288 path_get(&path);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001289 mutex_unlock(&loop_ctl_mutex);
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001290 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001291 if (!ret) {
1292 info->lo_device = huge_encode_dev(stat.dev);
1293 info->lo_inode = stat.ino;
1294 info->lo_rdevice = huge_encode_dev(stat.rdev);
1295 }
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001296 path_put(&path);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001297 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
1299
1300static void
1301loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1302{
1303 memset(info64, 0, sizeof(*info64));
1304 info64->lo_number = info->lo_number;
1305 info64->lo_device = info->lo_device;
1306 info64->lo_inode = info->lo_inode;
1307 info64->lo_rdevice = info->lo_rdevice;
1308 info64->lo_offset = info->lo_offset;
1309 info64->lo_sizelimit = 0;
1310 info64->lo_encrypt_type = info->lo_encrypt_type;
1311 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1312 info64->lo_flags = info->lo_flags;
1313 info64->lo_init[0] = info->lo_init[0];
1314 info64->lo_init[1] = info->lo_init[1];
1315 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1316 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1317 else
1318 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1319 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1320}
1321
1322static int
1323loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1324{
1325 memset(info, 0, sizeof(*info));
1326 info->lo_number = info64->lo_number;
1327 info->lo_device = info64->lo_device;
1328 info->lo_inode = info64->lo_inode;
1329 info->lo_rdevice = info64->lo_rdevice;
1330 info->lo_offset = info64->lo_offset;
1331 info->lo_encrypt_type = info64->lo_encrypt_type;
1332 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1333 info->lo_flags = info64->lo_flags;
1334 info->lo_init[0] = info64->lo_init[0];
1335 info->lo_init[1] = info64->lo_init[1];
1336 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1337 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1338 else
1339 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1340 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1341
1342 /* error in case values were truncated */
1343 if (info->lo_device != info64->lo_device ||
1344 info->lo_rdevice != info64->lo_rdevice ||
1345 info->lo_inode != info64->lo_inode ||
1346 info->lo_offset != info64->lo_offset)
1347 return -EOVERFLOW;
1348
1349 return 0;
1350}
1351
1352static int
1353loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1354{
1355 struct loop_info info;
1356 struct loop_info64 info64;
1357
1358 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1359 return -EFAULT;
1360 loop_info64_from_old(&info, &info64);
1361 return loop_set_status(lo, &info64);
1362}
1363
1364static int
1365loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1366{
1367 struct loop_info64 info64;
1368
1369 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1370 return -EFAULT;
1371 return loop_set_status(lo, &info64);
1372}
1373
1374static int
1375loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1376 struct loop_info info;
1377 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001378 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001380 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001381 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001382 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (!err)
1384 err = loop_info64_to_old(&info64, &info);
1385 if (!err && copy_to_user(arg, &info, sizeof(info)))
1386 err = -EFAULT;
1387
1388 return err;
1389}
1390
1391static int
1392loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1393 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001394 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001396 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001397 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001398 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1400 err = -EFAULT;
1401
1402 return err;
1403}
1404
Hannes Reinecke51001b72017-06-08 13:46:44 +02001405static int loop_set_capacity(struct loop_device *lo)
J. R. Okajima53d66602009-03-31 15:23:43 -07001406{
J. R. Okajima53d66602009-03-31 15:23:43 -07001407 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001408 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001409
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001410 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001411}
1412
Ming Leiab1cb272015-08-17 10:31:50 +08001413static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1414{
1415 int error = -ENXIO;
1416 if (lo->lo_state != Lo_bound)
1417 goto out;
1418
1419 __loop_update_dio(lo, !!arg);
1420 if (lo->use_dio == !!arg)
1421 return 0;
1422 error = -EINVAL;
1423 out:
1424 return error;
1425}
1426
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001427static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1428{
1429 if (lo->lo_state != Lo_bound)
1430 return -ENXIO;
1431
1432 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1433 return -EINVAL;
1434
1435 blk_mq_freeze_queue(lo->lo_queue);
1436
1437 blk_queue_logical_block_size(lo->lo_queue, arg);
Omar Sandovalbf093752017-09-05 14:24:47 -07001438 blk_queue_physical_block_size(lo->lo_queue, arg);
1439 blk_queue_io_min(lo->lo_queue, arg);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001440 loop_update_dio(lo);
1441
1442 blk_mq_unfreeze_queue(lo->lo_queue);
1443
1444 return 0;
1445}
1446
Jan Karaa1316542018-11-08 14:01:05 +01001447static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1448 unsigned long arg)
1449{
1450 int err;
1451
1452 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1453 if (err)
1454 return err;
1455 switch (cmd) {
1456 case LOOP_SET_CAPACITY:
1457 err = loop_set_capacity(lo);
1458 break;
1459 case LOOP_SET_DIRECT_IO:
1460 err = loop_set_dio(lo, arg);
1461 break;
1462 case LOOP_SET_BLOCK_SIZE:
1463 err = loop_set_block_size(lo, arg);
1464 break;
1465 default:
1466 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1467 }
1468 mutex_unlock(&loop_ctl_mutex);
1469 return err;
1470}
1471
Al Virobb214882008-03-02 09:29:48 -05001472static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 unsigned int cmd, unsigned long arg)
1474{
Al Virobb214882008-03-02 09:29:48 -05001475 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 int err;
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 switch (cmd) {
1479 case LOOP_SET_FD:
Jan Kara757ecf42018-11-08 14:01:10 +01001480 return loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 case LOOP_CHANGE_FD:
Jan Karac3710772018-11-08 14:01:11 +01001482 return loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 case LOOP_CLR_FD:
Jan Kara7ccd0792018-11-08 14:01:07 +01001484 return loop_clr_fd(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001486 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001487 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001488 err = loop_set_status_old(lo,
1489 (struct loop_info __user *)arg);
Jan Karaa1316542018-11-08 14:01:05 +01001490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 break;
1492 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001493 return loop_get_status_old(lo, (struct loop_info __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001495 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001496 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001497 err = loop_set_status64(lo,
1498 (struct loop_info64 __user *) arg);
Jan Karaa1316542018-11-08 14:01:05 +01001499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 break;
1501 case LOOP_GET_STATUS64:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001502 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
J. R. Okajima53d66602009-03-31 15:23:43 -07001503 case LOOP_SET_CAPACITY:
Ming Leiab1cb272015-08-17 10:31:50 +08001504 case LOOP_SET_DIRECT_IO:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001505 case LOOP_SET_BLOCK_SIZE:
Jan Karaa1316542018-11-08 14:01:05 +01001506 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1507 return -EPERM;
1508 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 default:
Jan Karaa1316542018-11-08 14:01:05 +01001510 err = lo_simple_ioctl(lo, cmd, arg);
1511 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 return err;
1515}
1516
David Howells863d5b822006-08-29 19:06:14 +01001517#ifdef CONFIG_COMPAT
1518struct compat_loop_info {
1519 compat_int_t lo_number; /* ioctl r/o */
1520 compat_dev_t lo_device; /* ioctl r/o */
1521 compat_ulong_t lo_inode; /* ioctl r/o */
1522 compat_dev_t lo_rdevice; /* ioctl r/o */
1523 compat_int_t lo_offset;
1524 compat_int_t lo_encrypt_type;
1525 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1526 compat_int_t lo_flags; /* ioctl r/o */
1527 char lo_name[LO_NAME_SIZE];
1528 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1529 compat_ulong_t lo_init[2];
1530 char reserved[4];
1531};
1532
1533/*
1534 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1535 * - noinlined to reduce stack space usage in main part of driver
1536 */
1537static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001538loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001539 struct loop_info64 *info64)
1540{
1541 struct compat_loop_info info;
1542
1543 if (copy_from_user(&info, arg, sizeof(info)))
1544 return -EFAULT;
1545
1546 memset(info64, 0, sizeof(*info64));
1547 info64->lo_number = info.lo_number;
1548 info64->lo_device = info.lo_device;
1549 info64->lo_inode = info.lo_inode;
1550 info64->lo_rdevice = info.lo_rdevice;
1551 info64->lo_offset = info.lo_offset;
1552 info64->lo_sizelimit = 0;
1553 info64->lo_encrypt_type = info.lo_encrypt_type;
1554 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1555 info64->lo_flags = info.lo_flags;
1556 info64->lo_init[0] = info.lo_init[0];
1557 info64->lo_init[1] = info.lo_init[1];
1558 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1559 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1560 else
1561 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1562 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1563 return 0;
1564}
1565
1566/*
1567 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1568 * - noinlined to reduce stack space usage in main part of driver
1569 */
1570static noinline int
1571loop_info64_to_compat(const struct loop_info64 *info64,
1572 struct compat_loop_info __user *arg)
1573{
1574 struct compat_loop_info info;
1575
1576 memset(&info, 0, sizeof(info));
1577 info.lo_number = info64->lo_number;
1578 info.lo_device = info64->lo_device;
1579 info.lo_inode = info64->lo_inode;
1580 info.lo_rdevice = info64->lo_rdevice;
1581 info.lo_offset = info64->lo_offset;
1582 info.lo_encrypt_type = info64->lo_encrypt_type;
1583 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1584 info.lo_flags = info64->lo_flags;
1585 info.lo_init[0] = info64->lo_init[0];
1586 info.lo_init[1] = info64->lo_init[1];
1587 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1588 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1589 else
1590 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1591 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1592
1593 /* error in case values were truncated */
1594 if (info.lo_device != info64->lo_device ||
1595 info.lo_rdevice != info64->lo_rdevice ||
1596 info.lo_inode != info64->lo_inode ||
1597 info.lo_offset != info64->lo_offset ||
1598 info.lo_init[0] != info64->lo_init[0] ||
1599 info.lo_init[1] != info64->lo_init[1])
1600 return -EOVERFLOW;
1601
1602 if (copy_to_user(arg, &info, sizeof(info)))
1603 return -EFAULT;
1604 return 0;
1605}
1606
1607static int
1608loop_set_status_compat(struct loop_device *lo,
1609 const struct compat_loop_info __user *arg)
1610{
1611 struct loop_info64 info64;
1612 int ret;
1613
1614 ret = loop_info64_from_compat(arg, &info64);
1615 if (ret < 0)
1616 return ret;
1617 return loop_set_status(lo, &info64);
1618}
1619
1620static int
1621loop_get_status_compat(struct loop_device *lo,
1622 struct compat_loop_info __user *arg)
1623{
1624 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001625 int err;
David Howells863d5b822006-08-29 19:06:14 +01001626
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001627 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001628 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001629 err = loop_get_status(lo, &info64);
David Howells863d5b822006-08-29 19:06:14 +01001630 if (!err)
1631 err = loop_info64_to_compat(&info64, arg);
1632 return err;
1633}
1634
Al Virobb214882008-03-02 09:29:48 -05001635static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1636 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001637{
Al Virobb214882008-03-02 09:29:48 -05001638 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001639 int err;
1640
David Howells863d5b822006-08-29 19:06:14 +01001641 switch(cmd) {
1642 case LOOP_SET_STATUS:
Jan Kara550df5fd2018-11-08 14:01:09 +01001643 err = loop_set_status_compat(lo,
1644 (const struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001645 break;
1646 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001647 err = loop_get_status_compat(lo,
1648 (struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001649 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001650 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001651 case LOOP_CLR_FD:
1652 case LOOP_GET_STATUS64:
1653 case LOOP_SET_STATUS64:
1654 arg = (unsigned long) compat_ptr(arg);
Gustavo A. R. Silvad893ff82018-07-02 12:52:06 -05001655 /* fall through */
David Howells863d5b822006-08-29 19:06:14 +01001656 case LOOP_SET_FD:
1657 case LOOP_CHANGE_FD:
Evan Green9fea4b32018-07-02 16:03:46 -07001658 case LOOP_SET_BLOCK_SIZE:
Al Virobb214882008-03-02 09:29:48 -05001659 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001660 break;
1661 default:
1662 err = -ENOIOCTLCMD;
1663 break;
1664 }
David Howells863d5b822006-08-29 19:06:14 +01001665 return err;
1666}
1667#endif
1668
Al Virobb214882008-03-02 09:29:48 -05001669static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
Kay Sievers770fe302011-07-31 22:08:04 +02001671 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01001672 int err;
Kay Sievers770fe302011-07-31 22:08:04 +02001673
Jan Kara0a42e992018-11-08 14:01:04 +01001674 err = mutex_lock_killable(&loop_ctl_mutex);
1675 if (err)
1676 return err;
Kay Sievers770fe302011-07-31 22:08:04 +02001677 lo = bdev->bd_disk->private_data;
1678 if (!lo) {
1679 err = -ENXIO;
1680 goto out;
1681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Ming Leif8933662015-05-06 12:26:23 +08001683 atomic_inc(&lo->lo_refcnt);
Kay Sievers770fe302011-07-31 22:08:04 +02001684out:
Jan Kara0a42e992018-11-08 14:01:04 +01001685 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001686 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
1688
Jan Kara967d1dc2018-11-08 14:01:03 +01001689static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
Jan Kara967d1dc2018-11-08 14:01:03 +01001691 struct loop_device *lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Jan Kara0a42e992018-11-08 14:01:04 +01001693 mutex_lock(&loop_ctl_mutex);
Jan Kara967d1dc2018-11-08 14:01:03 +01001694 lo = disk->private_data;
Ming Leif8933662015-05-06 12:26:23 +08001695 if (atomic_dec_return(&lo->lo_refcnt))
Jan Kara0a42e992018-11-08 14:01:04 +01001696 goto out_unlock;
Ming Leif8933662015-05-06 12:26:23 +08001697
Milan Broz14f27932008-12-12 14:48:27 +01001698 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
Jan Karaa2505b72018-11-08 14:01:06 +01001699 if (lo->lo_state != Lo_bound)
1700 goto out_unlock;
1701 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001702 mutex_unlock(&loop_ctl_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001703 /*
1704 * In autoclear mode, stop the loop thread
1705 * and remove configuration after last close.
1706 */
Jan Kara7ccd0792018-11-08 14:01:07 +01001707 __loop_clr_fd(lo);
1708 return;
Omar Sandoval43cade802017-08-24 00:03:44 -07001709 } else if (lo->lo_state == Lo_bound) {
Milan Broz14f27932008-12-12 14:48:27 +01001710 /*
1711 * Otherwise keep thread (if running) and config,
1712 * but flush possible ongoing bios in thread.
1713 */
Omar Sandoval43cade802017-08-24 00:03:44 -07001714 blk_mq_freeze_queue(lo->lo_queue);
1715 blk_mq_unfreeze_queue(lo->lo_queue);
Milan Broz14f27932008-12-12 14:48:27 +01001716 }
David Woodhouse96c58652008-02-06 01:36:27 -08001717
Jan Kara0a42e992018-11-08 14:01:04 +01001718out_unlock:
Tetsuo Handa310ca162018-11-08 14:01:02 +01001719 mutex_unlock(&loop_ctl_mutex);
Linus Torvaldsae665012018-01-05 16:26:00 -08001720}
1721
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001722static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001724 .open = lo_open,
1725 .release = lo_release,
1726 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001727#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001728 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001729#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730};
1731
1732/*
1733 * And now the modules code and kernel interface.
1734 */
Ken Chen73285082007-05-08 00:28:20 -07001735static int max_loop;
Joe Perches5657a812018-05-24 13:38:59 -06001736module_param(max_loop, int, 0444);
Ken Chena47653f2007-06-08 13:46:44 -07001737MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Joe Perches5657a812018-05-24 13:38:59 -06001738module_param(max_part, int, 0444);
Laurent Vivier476a4812008-03-26 12:11:53 +01001739MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740MODULE_LICENSE("GPL");
1741MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1742
1743int loop_register_transfer(struct loop_func_table *funcs)
1744{
1745 unsigned int n = funcs->number;
1746
1747 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1748 return -EINVAL;
1749 xfer_funcs[n] = funcs;
1750 return 0;
1751}
1752
Kay Sievers34dd82a2011-07-31 22:08:04 +02001753static int unregister_transfer_cb(int id, void *ptr, void *data)
1754{
1755 struct loop_device *lo = ptr;
1756 struct loop_func_table *xfer = data;
1757
Tetsuo Handa310ca162018-11-08 14:01:02 +01001758 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001759 if (lo->lo_encryption == xfer)
1760 loop_release_xfer(lo);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001761 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001762 return 0;
1763}
1764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765int loop_unregister_transfer(int number)
1766{
1767 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 struct loop_func_table *xfer;
1769
1770 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1771 return -EINVAL;
1772
1773 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001774 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 return 0;
1776}
1777
1778EXPORT_SYMBOL(loop_register_transfer);
1779EXPORT_SYMBOL(loop_unregister_transfer);
1780
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001781static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
Ming Leib5dd2f62014-12-31 13:22:57 +00001782 const struct blk_mq_queue_data *bd)
1783{
Jens Axboe1894e912018-04-13 16:24:29 -06001784 struct request *rq = bd->rq;
1785 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1786 struct loop_device *lo = rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00001787
Jens Axboe1894e912018-04-13 16:24:29 -06001788 blk_mq_start_request(rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001789
Ming Leif4aa4c72015-05-05 19:49:54 +08001790 if (lo->lo_state != Lo_bound)
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001791 return BLK_STS_IOERR;
Ming Leif4aa4c72015-05-05 19:49:54 +08001792
Jens Axboe1894e912018-04-13 16:24:29 -06001793 switch (req_op(rq)) {
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001794 case REQ_OP_FLUSH:
1795 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +02001796 case REQ_OP_WRITE_ZEROES:
Ming Leibc07c102015-08-17 10:31:51 +08001797 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001798 break;
1799 default:
1800 cmd->use_aio = lo->use_dio;
1801 break;
1802 }
Ming Leibc07c102015-08-17 10:31:51 +08001803
Shaohua Lid4478e92017-09-25 13:07:22 -06001804 /* always use the first bio's css */
Shaohua Li0b508bc2017-09-26 11:02:12 -07001805#ifdef CONFIG_BLK_CGROUP
Dennis Zhoub5f29542018-11-01 17:24:10 -04001806 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1807 cmd->css = rq->bio->bi_css;
Shaohua Lid4478e92017-09-25 13:07:22 -06001808 css_get(cmd->css);
1809 } else
1810#endif
1811 cmd->css = NULL;
Petr Mladek39891442016-10-11 13:55:20 -07001812 kthread_queue_work(&lo->worker, &cmd->work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001813
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001814 return BLK_STS_OK;
Ming Leib5dd2f62014-12-31 13:22:57 +00001815}
1816
1817static void loop_handle_cmd(struct loop_cmd *cmd)
1818{
Jens Axboe1894e912018-04-13 16:24:29 -06001819 struct request *rq = blk_mq_rq_from_pdu(cmd);
1820 const bool write = op_is_write(req_op(rq));
1821 struct loop_device *lo = rq->q->queuedata;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001822 int ret = 0;
Ming Leib5dd2f62014-12-31 13:22:57 +00001823
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001824 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1825 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00001826 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001827 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001828
Jens Axboe1894e912018-04-13 16:24:29 -06001829 ret = do_req_filebacked(lo, rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001830 failed:
Ming Leibc07c102015-08-17 10:31:51 +08001831 /* complete non-aio request */
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001832 if (!cmd->use_aio || ret) {
1833 cmd->ret = ret ? -EIO : 0;
Jens Axboe1894e912018-04-13 16:24:29 -06001834 blk_mq_complete_request(rq);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001835 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001836}
1837
Ming Leie03a3d72015-08-17 10:31:48 +08001838static void loop_queue_work(struct kthread_work *work)
Ming Leib5dd2f62014-12-31 13:22:57 +00001839{
1840 struct loop_cmd *cmd =
Ming Leie03a3d72015-08-17 10:31:48 +08001841 container_of(work, struct loop_cmd, work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001842
1843 loop_handle_cmd(cmd);
1844}
1845
Christoph Hellwigd6296d392017-05-01 10:19:08 -06001846static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1847 unsigned int hctx_idx, unsigned int numa_node)
Ming Leib5dd2f62014-12-31 13:22:57 +00001848{
1849 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1850
Petr Mladek39891442016-10-11 13:55:20 -07001851 kthread_init_work(&cmd->work, loop_queue_work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001852 return 0;
1853}
1854
Eric Biggersf363b082017-03-30 13:39:16 -07001855static const struct blk_mq_ops loop_mq_ops = {
Ming Leib5dd2f62014-12-31 13:22:57 +00001856 .queue_rq = loop_queue_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00001857 .init_request = loop_init_request,
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001858 .complete = lo_complete_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00001859};
1860
Kay Sievers34dd82a2011-07-31 22:08:04 +02001861static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001862{
1863 struct loop_device *lo;
1864 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001865 int err;
Ken Chen73285082007-05-08 00:28:20 -07001866
Silva Paulo68d740d2012-07-14 15:39:58 -07001867 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001868 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07001869 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07001870 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001871
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001872 lo->lo_state = Lo_unbound;
1873
Tejun Heoc718aa62013-02-27 17:03:58 -08001874 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001875 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08001876 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1877 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02001878 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001879 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08001880 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001881 }
1882 if (err < 0)
1883 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08001884 i = err;
Ken Chen73285082007-05-08 00:28:20 -07001885
Wei Yongjun183cfb52013-03-22 08:59:19 -06001886 err = -ENOMEM;
Ming Leib5dd2f62014-12-31 13:22:57 +00001887 lo->tag_set.ops = &loop_mq_ops;
1888 lo->tag_set.nr_hw_queues = 1;
1889 lo->tag_set.queue_depth = 128;
1890 lo->tag_set.numa_node = NUMA_NO_NODE;
1891 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1892 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1893 lo->tag_set.driver_data = lo;
1894
1895 err = blk_mq_alloc_tag_set(&lo->tag_set);
1896 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001897 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07001898
Ming Leib5dd2f62014-12-31 13:22:57 +00001899 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1900 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1901 err = PTR_ERR(lo->lo_queue);
1902 goto out_cleanup_tags;
1903 }
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001904 lo->lo_queue->queuedata = lo;
1905
Shaohua Li54bb0ad2017-08-31 22:09:45 -07001906 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
Shaohua Li40326d82017-08-31 22:09:46 -07001907
Ming Lei5b5e20f2015-08-17 10:31:47 +08001908 /*
Shaohua Li40326d82017-08-31 22:09:46 -07001909 * By default, we do buffer IO, so it doesn't make sense to enable
1910 * merge because the I/O submitted to backing file is handled page by
1911 * page. For directio mode, merge does help to dispatch bigger request
1912 * to underlayer disk. We will enable merge once directio is enabled.
Ming Lei5b5e20f2015-08-17 10:31:47 +08001913 */
Bart Van Assche8b904b52018-03-07 17:10:10 -08001914 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei5b5e20f2015-08-17 10:31:47 +08001915
Minfei Huang7a649732016-06-07 10:05:15 +08001916 err = -ENOMEM;
Laurent Vivier476a4812008-03-26 12:11:53 +01001917 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001918 if (!disk)
1919 goto out_free_queue;
1920
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001921 /*
1922 * Disable partition scanning by default. The in-kernel partition
1923 * scanning can be requested individually per-device during its
1924 * setup. Userspace can always add and remove partitions from all
1925 * devices. The needed partition minors are allocated from the
1926 * extended minor space, the main loop device numbers will continue
1927 * to match the loop minors, regardless of the number of partitions
1928 * used.
1929 *
1930 * If max_part is given, partition scanning is globally enabled for
1931 * all loop devices. The minors for the main loop devices will be
1932 * multiples of max_part.
1933 *
1934 * Note: Global-for-all-devices, set-only-at-init, read-only module
1935 * parameteters like 'max_loop' and 'max_part' make things needlessly
1936 * complicated, are too static, inflexible and may surprise
1937 * userspace tools. Parameters like this in general should be avoided.
1938 */
1939 if (!part_shift)
1940 disk->flags |= GENHD_FL_NO_PART_SCAN;
1941 disk->flags |= GENHD_FL_EXT_DEVT;
Ming Leif8933662015-05-06 12:26:23 +08001942 atomic_set(&lo->lo_refcnt, 0);
Ken Chen73285082007-05-08 00:28:20 -07001943 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07001944 spin_lock_init(&lo->lo_lock);
1945 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001946 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001947 disk->fops = &lo_fops;
1948 disk->private_data = lo;
1949 disk->queue = lo->lo_queue;
1950 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001951 add_disk(disk);
1952 *l = lo;
1953 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001954
1955out_free_queue:
1956 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00001957out_cleanup_tags:
1958 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001959out_free_idr:
1960 idr_remove(&loop_index_idr, i);
Ken Chen73285082007-05-08 00:28:20 -07001961out_free_dev:
1962 kfree(lo);
1963out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001964 return err;
Ken Chen73285082007-05-08 00:28:20 -07001965}
1966
Kay Sievers34dd82a2011-07-31 22:08:04 +02001967static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001968{
NeilBrown6cd18e72015-04-27 14:12:22 +10001969 del_gendisk(lo->lo_disk);
Bart Van Assche0fa8ebd2018-02-28 10:15:28 -08001970 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00001971 blk_mq_free_tag_set(&lo->tag_set);
Ken Chen73285082007-05-08 00:28:20 -07001972 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001973 kfree(lo);
1974}
1975
Kay Sievers770fe302011-07-31 22:08:04 +02001976static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001977{
Kay Sievers770fe302011-07-31 22:08:04 +02001978 struct loop_device *lo = ptr;
1979 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001980
Kay Sievers770fe302011-07-31 22:08:04 +02001981 if (lo->lo_state == Lo_unbound) {
1982 *l = lo;
1983 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001984 }
Kay Sievers770fe302011-07-31 22:08:04 +02001985 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001986}
1987
Kay Sievers34dd82a2011-07-31 22:08:04 +02001988static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001989{
Ken Chena47653f2007-06-08 13:46:44 -07001990 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001991 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001992
Kay Sievers770fe302011-07-31 22:08:04 +02001993 if (i < 0) {
1994 int err;
1995
1996 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1997 if (err == 1) {
1998 *l = lo;
1999 ret = lo->lo_number;
2000 }
2001 goto out;
2002 }
2003
2004 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002005 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07002006 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02002007 *l = lo;
2008 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07002009 }
Kay Sievers770fe302011-07-31 22:08:04 +02002010out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002011 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07002012}
2013
Ken Chen73285082007-05-08 00:28:20 -07002014static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2015{
Al Viro705962c2007-05-13 05:52:32 -04002016 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04002017 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002018 int err;
Ken Chen73285082007-05-08 00:28:20 -07002019
Jan Kara0a42e992018-11-08 14:01:04 +01002020 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002021 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2022 if (err < 0)
2023 err = loop_add(&lo, MINOR(dev) >> part_shift);
2024 if (err < 0)
Mikulas Patockaa207f592013-10-14 12:13:24 -04002025 kobj = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002026 else
Jan Kara3079c222018-02-26 13:01:38 +01002027 kobj = get_disk_and_module(lo->lo_disk);
Jan Kara0a42e992018-11-08 14:01:04 +01002028 mutex_unlock(&loop_ctl_mutex);
Ken Chen73285082007-05-08 00:28:20 -07002029
2030 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04002031 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07002032}
2033
Kay Sievers770fe302011-07-31 22:08:04 +02002034static long loop_control_ioctl(struct file *file, unsigned int cmd,
2035 unsigned long parm)
2036{
2037 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01002038 int ret;
Kay Sievers770fe302011-07-31 22:08:04 +02002039
Jan Kara0a42e992018-11-08 14:01:04 +01002040 ret = mutex_lock_killable(&loop_ctl_mutex);
2041 if (ret)
2042 return ret;
2043
2044 ret = -ENOSYS;
Kay Sievers770fe302011-07-31 22:08:04 +02002045 switch (cmd) {
2046 case LOOP_CTL_ADD:
2047 ret = loop_lookup(&lo, parm);
2048 if (ret >= 0) {
2049 ret = -EEXIST;
2050 break;
2051 }
2052 ret = loop_add(&lo, parm);
2053 break;
2054 case LOOP_CTL_REMOVE:
2055 ret = loop_lookup(&lo, parm);
2056 if (ret < 0)
2057 break;
Kay Sievers770fe302011-07-31 22:08:04 +02002058 if (lo->lo_state != Lo_unbound) {
2059 ret = -EBUSY;
Tetsuo Handa310ca162018-11-08 14:01:02 +01002060 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02002061 break;
2062 }
Ming Leif8933662015-05-06 12:26:23 +08002063 if (atomic_read(&lo->lo_refcnt) > 0) {
Kay Sievers770fe302011-07-31 22:08:04 +02002064 ret = -EBUSY;
Tetsuo Handa310ca162018-11-08 14:01:02 +01002065 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02002066 break;
2067 }
2068 lo->lo_disk->private_data = NULL;
Kay Sievers770fe302011-07-31 22:08:04 +02002069 idr_remove(&loop_index_idr, lo->lo_number);
2070 loop_remove(lo);
2071 break;
2072 case LOOP_CTL_GET_FREE:
2073 ret = loop_lookup(&lo, -1);
2074 if (ret >= 0)
2075 break;
2076 ret = loop_add(&lo, -1);
2077 }
Jan Kara0a42e992018-11-08 14:01:04 +01002078 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02002079
2080 return ret;
2081}
2082
2083static const struct file_operations loop_ctl_fops = {
2084 .open = nonseekable_open,
2085 .unlocked_ioctl = loop_control_ioctl,
2086 .compat_ioctl = loop_control_ioctl,
2087 .owner = THIS_MODULE,
2088 .llseek = noop_llseek,
2089};
2090
2091static struct miscdevice loop_misc = {
2092 .minor = LOOP_CTRL_MINOR,
2093 .name = "loop-control",
2094 .fops = &loop_ctl_fops,
2095};
2096
2097MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2098MODULE_ALIAS("devname:loop-control");
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100static int __init loop_init(void)
2101{
Ken Chena47653f2007-06-08 13:46:44 -07002102 int i, nr;
2103 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002104 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02002105 int err;
Ken Chena47653f2007-06-08 13:46:44 -07002106
Laurent Vivier476a4812008-03-26 12:11:53 +01002107 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02002108 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01002109 part_shift = fls(max_part);
2110
Namhyung Kimac04fee2011-05-27 07:59:25 +02002111 /*
2112 * Adjust max_part according to part_shift as it is exported
2113 * to user space so that user can decide correct minor number
2114 * if [s]he want to create more devices.
2115 *
2116 * Note that -1 is required because partition 0 is reserved
2117 * for the whole disk.
2118 */
2119 max_part = (1UL << part_shift) - 1;
2120 }
2121
Guo Chaob1a66502013-02-21 15:16:49 -08002122 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2123 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002124 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002125 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02002126
Guo Chaob1a66502013-02-21 15:16:49 -08002127 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2128 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002129 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Kay Sieversd134b002011-07-31 22:08:04 +02002132 /*
2133 * If max_loop is specified, create that many devices upfront.
2134 * This also becomes a hard limit. If max_loop is not specified,
2135 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2136 * init time. Loop devices can be requested on-demand with the
2137 * /dev/loop-control interface, or be instantiated by accessing
2138 * a 'dead' device node.
2139 */
Ken Chen73285082007-05-08 00:28:20 -07002140 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07002141 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002142 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07002143 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02002144 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002145 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 }
Ken Chena47653f2007-06-08 13:46:44 -07002147
Anton Volkova8c1d062017-08-07 15:37:50 +03002148 err = misc_register(&loop_misc);
2149 if (err < 0)
2150 goto err_out;
2151
2152
Guo Chaob1a66502013-02-21 15:16:49 -08002153 if (register_blkdev(LOOP_MAJOR, "loop")) {
2154 err = -EIO;
2155 goto misc_out;
2156 }
Ken Chena47653f2007-06-08 13:46:44 -07002157
Ken Chena47653f2007-06-08 13:46:44 -07002158 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2159 THIS_MODULE, loop_probe, NULL, NULL);
2160
Kay Sieversd134b002011-07-31 22:08:04 +02002161 /* pre-create number of devices given by config or max_loop */
Jan Kara0a42e992018-11-08 14:01:04 +01002162 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002163 for (i = 0; i < nr; i++)
2164 loop_add(&lo, i);
Jan Kara0a42e992018-11-08 14:01:04 +01002165 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002166
Ken Chen73285082007-05-08 00:28:20 -07002167 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08002169
2170misc_out:
2171 misc_deregister(&loop_misc);
Anton Volkova8c1d062017-08-07 15:37:50 +03002172err_out:
Guo Chaob1a66502013-02-21 15:16:49 -08002173 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002174}
Ken Chena47653f2007-06-08 13:46:44 -07002175
Kay Sievers34dd82a2011-07-31 22:08:04 +02002176static int loop_exit_cb(int id, void *ptr, void *data)
2177{
2178 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07002179
Kay Sievers34dd82a2011-07-31 22:08:04 +02002180 loop_remove(lo);
2181 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182}
2183
Ken Chen73285082007-05-08 00:28:20 -07002184static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185{
Ken Chena47653f2007-06-08 13:46:44 -07002186 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
Namhyung Kima1c15c52011-05-24 16:48:55 +02002188 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07002189
Kay Sievers34dd82a2011-07-31 22:08:04 +02002190 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002191 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07002192
Ken Chena47653f2007-06-08 13:46:44 -07002193 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07002194 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02002195
2196 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197}
2198
2199module_init(loop_init);
2200module_exit(loop_exit);
2201
2202#ifndef MODULE
2203static int __init max_loop_setup(char *str)
2204{
2205 max_loop = simple_strtol(str, NULL, 0);
2206 return 1;
2207}
2208
2209__setup("max_loop=", max_loop_setup);
2210#endif