blob: 14372df0f35498bbd0f8adb59c671a92e690c609 [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>
Dennis Zhoudb6638d2018-12-05 12:10:35 -050080#include <linux/blk-cgroup.h>
Adam Manzanaresd9a08a92018-05-22 10:52:19 -070081
Al Viro83a87612013-05-12 10:14:07 -040082#include "loop.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080084#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Kay Sievers34dd82a2011-07-31 22:08:04 +020086static DEFINE_IDR(loop_index_idr);
Tetsuo Handa310ca162018-11-08 14:01:02 +010087static DEFINE_MUTEX(loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Laurent Vivier476a4812008-03-26 12:11:53 +010089static int max_part;
90static int part_shift;
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
96{
Cong Wangcfd80052011-11-25 23:14:18 +080097 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 char *in, *out, *key;
100 int i, keysize;
101
102 if (cmd == READ) {
103 in = raw_buf;
104 out = loop_buf;
105 } else {
106 in = loop_buf;
107 out = raw_buf;
108 }
109
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
114
Cong Wangcfd80052011-11-25 23:14:18 +0800115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 cond_resched();
118 return 0;
119}
120
121static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122{
123 if (unlikely(info->lo_encrypt_key_size <= 0))
124 return -EINVAL;
125 return 0;
126}
127
128static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200130};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
135 .init = xor_init
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200136};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138/* xfer_funcs[0] is special - its release function is never called */
139static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140 &none_funcs,
141 &xor_funcs
142};
143
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100144static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
Guo Chaob7a1da62013-02-21 15:16:50 -0800146 loff_t loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148 /* Compute loopsize in bytes */
Guo Chaob7a1da62013-02-21 15:16:50 -0800149 loopsize = i_size_read(file->f_mapping->host);
150 if (offset > 0)
151 loopsize -= offset;
152 /* offset is beyond i_size, weird but possible */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100153 if (loopsize < 0)
154 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 /*
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
161 */
162 return loopsize >> 9;
163}
164
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100165static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
168}
169
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800170static void __loop_update_dio(struct loop_device *lo, bool dio)
171{
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
177 bool use_dio;
178
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
182 }
183
184 /*
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
189 *
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
Masahiro Yamada89d790a2017-02-27 14:29:01 -0800192 * of requests in sane applications should be PAGE_SIZE aligned
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800193 */
194 if (dio) {
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
198 !lo->transfer)
199 use_dio = true;
200 else
201 use_dio = false;
202 } else {
203 use_dio = false;
204 }
205
206 if (lo->use_dio == use_dio)
207 return;
208
209 /* flush dirty pages before changing direct IO */
210 vfs_fsync(file, 0);
211
212 /*
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
216 */
Martijn Coenen0fbcf5792020-03-10 14:06:54 +0100217 if (lo->lo_state == Lo_bound)
218 blk_mq_freeze_queue(lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800219 lo->use_dio = use_dio;
Shaohua Li40326d82017-08-31 22:09:46 -0700220 if (use_dio) {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800222 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700223 } else {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700226 }
Martijn Coenen0fbcf5792020-03-10 14:06:54 +0100227 if (lo->lo_state == Lo_bound)
228 blk_mq_unfreeze_queue(lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800229}
230
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100231static int
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -0700232figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100233{
234 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 sector_t x = (sector_t)size;
Guo Chao7b0576a2013-02-21 15:16:47 -0800236 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
238 if (unlikely((loff_t)x != size))
239 return -EFBIG;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100240 if (lo->lo_offset != offset)
241 lo->lo_offset = offset;
242 if (lo->lo_sizelimit != sizelimit)
243 lo->lo_sizelimit = sizelimit;
Ken Chen73285082007-05-08 00:28:20 -0700244 set_capacity(lo->lo_disk, x);
Guo Chao7b0576a2013-02-21 15:16:47 -0800245 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
246 /* let user-space know about the new size */
247 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100248 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
251static inline int
252lo_do_transfer(struct loop_device *lo, int cmd,
253 struct page *rpage, unsigned roffs,
254 struct page *lpage, unsigned loffs,
255 int size, sector_t rblock)
256{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200257 int ret;
258
259 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
260 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 return 0;
262
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200263 printk_ratelimited(KERN_ERR
264 "loop: Transfer error at byte offset %llu, length %i.\n",
265 (unsigned long long)rblock << 9, size);
266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200269static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200271 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 ssize_t bw;
Al Viro283e7e52015-04-03 15:21:59 -0400273
Christoph Hellwigb6207432019-06-26 15:49:28 +0200274 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Al Viro03d95eb2013-03-20 13:04:20 -0400276 file_start_write(file);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300277 bw = vfs_iter_write(file, &i, ppos, 0);
Al Viro03d95eb2013-03-20 13:04:20 -0400278 file_end_write(file);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200279
280 if (likely(bw == bvec->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return 0;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200282
283 printk_ratelimited(KERN_ERR
284 "loop: Write error at byte offset %llu, length %i.\n",
285 (unsigned long long)*ppos, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 if (bw >= 0)
287 bw = -EIO;
288 return bw;
289}
290
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200291static int lo_write_simple(struct loop_device *lo, struct request *rq,
292 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200294 struct bio_vec bvec;
295 struct req_iterator iter;
296 int ret = 0;
297
298 rq_for_each_segment(bvec, rq, iter) {
299 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
300 if (ret < 0)
301 break;
302 cond_resched();
303 }
304
305 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200308/*
Christoph Hellwig456be142011-10-17 12:57:20 +0200309 * This is the slow, transforming version that needs to double buffer the
310 * data as it cannot do the transformations in place without having direct
311 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200313static int lo_write_transfer(struct loop_device *lo, struct request *rq,
314 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200316 struct bio_vec bvec, b;
Ming Lei30112012014-12-31 13:22:58 +0000317 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200318 struct page *page;
Kent Overstreet79886132013-11-23 17:19:00 -0800319 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200321 page = alloc_page(GFP_NOIO);
322 if (unlikely(!page))
323 return -ENOMEM;
Christoph Hellwig456be142011-10-17 12:57:20 +0200324
Ming Lei30112012014-12-31 13:22:58 +0000325 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200326 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
327 bvec.bv_offset, bvec.bv_len, pos >> 9);
328 if (unlikely(ret))
329 break;
330
331 b.bv_page = page;
332 b.bv_offset = 0;
333 b.bv_len = bvec.bv_len;
334 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 if (ret < 0)
336 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200338
339 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200343static int lo_read_simple(struct loop_device *lo, struct request *rq,
344 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Kent Overstreet79886132013-11-23 17:19:00 -0800346 struct bio_vec bvec;
Ming Lei30112012014-12-31 13:22:58 +0000347 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200348 struct iov_iter i;
349 ssize_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Ming Lei30112012014-12-31 13:22:58 +0000351 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigb6207432019-06-26 15:49:28 +0200352 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300353 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200354 if (len < 0)
355 return len;
Dave Young306df072012-02-08 22:07:19 +0100356
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200357 flush_dcache_page(bvec.bv_page);
358
359 if (len != bvec.bv_len) {
Ming Lei30112012014-12-31 13:22:58 +0000360 struct bio *bio;
361
362 __rq_for_each_bio(bio, rq)
363 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 break;
Dave Young306df072012-02-08 22:07:19 +0100365 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200366 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200368
Dave Young306df072012-02-08 22:07:19 +0100369 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200372static int lo_read_transfer(struct loop_device *lo, struct request *rq,
373 loff_t pos)
374{
375 struct bio_vec bvec, b;
376 struct req_iterator iter;
377 struct iov_iter i;
378 struct page *page;
379 ssize_t len;
380 int ret = 0;
381
382 page = alloc_page(GFP_NOIO);
383 if (unlikely(!page))
384 return -ENOMEM;
385
386 rq_for_each_segment(bvec, rq, iter) {
387 loff_t offset = pos;
388
389 b.bv_page = page;
390 b.bv_offset = 0;
391 b.bv_len = bvec.bv_len;
392
Christoph Hellwigb6207432019-06-26 15:49:28 +0200393 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300394 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200395 if (len < 0) {
396 ret = len;
397 goto out_free_page;
398 }
399
400 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
401 bvec.bv_offset, len, offset >> 9);
402 if (ret)
403 goto out_free_page;
404
405 flush_dcache_page(bvec.bv_page);
406
407 if (len != bvec.bv_len) {
408 struct bio *bio;
409
410 __rq_for_each_bio(bio, rq)
411 zero_fill_bio(bio);
412 break;
413 }
414 }
415
416 ret = 0;
417out_free_page:
418 __free_page(page);
419 return ret;
420}
421
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700422static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
423 int mode)
Ming Leicf655d92014-12-31 13:22:59 +0000424{
425 /*
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700426 * We use fallocate to manipulate the space mappings used by the image
427 * a.k.a. discard/zerorange. However we do not support this if
428 * encryption is enabled, because it may give an attacker useful
429 * information.
Ming Leicf655d92014-12-31 13:22:59 +0000430 */
431 struct file *file = lo->lo_backing_file;
Evan Greenc52abf52020-04-03 16:43:04 +0200432 struct request_queue *q = lo->lo_queue;
Ming Leicf655d92014-12-31 13:22:59 +0000433 int ret;
434
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700435 mode |= FALLOC_FL_KEEP_SIZE;
436
Evan Greenc52abf52020-04-03 16:43:04 +0200437 if (!blk_queue_discard(q)) {
Ming Leicf655d92014-12-31 13:22:59 +0000438 ret = -EOPNOTSUPP;
439 goto out;
440 }
441
442 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
443 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
444 ret = -EIO;
445 out:
446 return ret;
447}
448
449static int lo_req_flush(struct loop_device *lo, struct request *rq)
450{
451 struct file *file = lo->lo_backing_file;
452 int ret = vfs_fsync(file, 0);
453 if (unlikely(ret && ret != -EINVAL))
454 ret = -EIO;
455
456 return ret;
457}
458
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200459static void lo_complete_rq(struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800460{
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200461 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Jens Axboef9de14b2018-04-13 16:25:57 -0600462 blk_status_t ret = BLK_STS_OK;
Ming Leibc07c102015-08-17 10:31:51 +0800463
Jens Axboef9de14b2018-04-13 16:25:57 -0600464 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
465 req_op(rq) != REQ_OP_READ) {
466 if (cmd->ret < 0)
Evan Green8cd55082020-04-03 16:43:03 +0200467 ret = errno_to_blk_status(cmd->ret);
Jens Axboef9de14b2018-04-13 16:25:57 -0600468 goto end_io;
Ming Leibc07c102015-08-17 10:31:51 +0800469 }
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200470
Jens Axboef9de14b2018-04-13 16:25:57 -0600471 /*
472 * Short READ - if we got some data, advance our request and
473 * retry it. If we got no data, end the rest with EIO.
474 */
475 if (cmd->ret) {
476 blk_update_request(rq, BLK_STS_OK, cmd->ret);
477 cmd->ret = 0;
478 blk_mq_requeue_request(rq, true);
479 } else {
480 if (cmd->use_aio) {
481 struct bio *bio = rq->bio;
482
483 while (bio) {
484 zero_fill_bio(bio);
485 bio = bio->bi_next;
486 }
487 }
488 ret = BLK_STS_IOERR;
489end_io:
490 blk_mq_end_request(rq, ret);
491 }
Ming Leibc07c102015-08-17 10:31:51 +0800492}
493
Shaohua Li92d77332017-09-01 11:15:17 -0700494static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
495{
Jens Axboe1894e9162018-04-13 16:24:29 -0600496 struct request *rq = blk_mq_rq_from_pdu(cmd);
497
Shaohua Li92d77332017-09-01 11:15:17 -0700498 if (!atomic_dec_and_test(&cmd->ref))
499 return;
500 kfree(cmd->bvec);
501 cmd->bvec = NULL;
Jens Axboe1894e9162018-04-13 16:24:29 -0600502 blk_mq_complete_request(rq);
Shaohua Li92d77332017-09-01 11:15:17 -0700503}
504
Ming Leibc07c102015-08-17 10:31:51 +0800505static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
506{
507 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
Ming Leibc07c102015-08-17 10:31:51 +0800508
Shaohua Lid4478e92017-09-25 13:07:22 -0600509 if (cmd->css)
510 css_put(cmd->css);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200511 cmd->ret = ret;
Shaohua Li92d77332017-09-01 11:15:17 -0700512 lo_rw_aio_do_completion(cmd);
Ming Leibc07c102015-08-17 10:31:51 +0800513}
514
515static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
516 loff_t pos, bool rw)
517{
518 struct iov_iter iter;
Ming Lei86af5952019-02-15 19:13:17 +0800519 struct req_iterator rq_iter;
Ming Leibc07c102015-08-17 10:31:51 +0800520 struct bio_vec *bvec;
Jens Axboe1894e9162018-04-13 16:24:29 -0600521 struct request *rq = blk_mq_rq_from_pdu(cmd);
Shaohua Li40326d82017-08-31 22:09:46 -0700522 struct bio *bio = rq->bio;
Ming Leibc07c102015-08-17 10:31:51 +0800523 struct file *file = lo->lo_backing_file;
Ming Lei86af5952019-02-15 19:13:17 +0800524 struct bio_vec tmp;
Shaohua Li40326d82017-08-31 22:09:46 -0700525 unsigned int offset;
Ming Lei86af5952019-02-15 19:13:17 +0800526 int nr_bvec = 0;
Ming Leibc07c102015-08-17 10:31:51 +0800527 int ret;
528
Ming Lei86af5952019-02-15 19:13:17 +0800529 rq_for_each_bvec(tmp, rq, rq_iter)
530 nr_bvec++;
Ming Leibc07c102015-08-17 10:31:51 +0800531
Ming Lei86af5952019-02-15 19:13:17 +0800532 if (rq->bio != rq->biotail) {
533
534 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
Kees Cook6da2ec52018-06-12 13:55:00 -0700535 GFP_NOIO);
Shaohua Li40326d82017-08-31 22:09:46 -0700536 if (!bvec)
537 return -EIO;
538 cmd->bvec = bvec;
539
540 /*
541 * The bios of the request may be started from the middle of
542 * the 'bvec' because of bio splitting, so we can't directly
Ming Lei86af5952019-02-15 19:13:17 +0800543 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
Shaohua Li40326d82017-08-31 22:09:46 -0700544 * API will take care of all details for us.
545 */
Ming Lei86af5952019-02-15 19:13:17 +0800546 rq_for_each_bvec(tmp, rq, rq_iter) {
Shaohua Li40326d82017-08-31 22:09:46 -0700547 *bvec = tmp;
548 bvec++;
549 }
550 bvec = cmd->bvec;
551 offset = 0;
552 } else {
553 /*
554 * Same here, this bio may be started from the middle of the
555 * 'bvec' because of bio splitting, so offset from the bvec
556 * must be passed to iov iterator
557 */
558 offset = bio->bi_iter.bi_bvec_done;
559 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
Shaohua Li40326d82017-08-31 22:09:46 -0700560 }
Shaohua Li92d77332017-09-01 11:15:17 -0700561 atomic_set(&cmd->ref, 2);
Shaohua Li40326d82017-08-31 22:09:46 -0700562
Christoph Hellwigb6207432019-06-26 15:49:28 +0200563 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
Shaohua Li40326d82017-08-31 22:09:46 -0700564 iter.iov_offset = offset;
Ming Leibc07c102015-08-17 10:31:51 +0800565
566 cmd->iocb.ki_pos = pos;
567 cmd->iocb.ki_filp = file;
568 cmd->iocb.ki_complete = lo_rw_aio_complete;
569 cmd->iocb.ki_flags = IOCB_DIRECT;
Adam Manzanaresd9a08a92018-05-22 10:52:19 -0700570 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
Shaohua Lid4478e92017-09-25 13:07:22 -0600571 if (cmd->css)
572 kthread_associate_blkcg(cmd->css);
Ming Leibc07c102015-08-17 10:31:51 +0800573
574 if (rw == WRITE)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100575 ret = call_write_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800576 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100577 ret = call_read_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800578
Shaohua Li92d77332017-09-01 11:15:17 -0700579 lo_rw_aio_do_completion(cmd);
Shaohua Lid4478e92017-09-25 13:07:22 -0600580 kthread_associate_blkcg(NULL);
Shaohua Li92d77332017-09-01 11:15:17 -0700581
Ming Leibc07c102015-08-17 10:31:51 +0800582 if (ret != -EIOCBQUEUED)
583 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
584 return 0;
585}
586
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200587static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800588{
589 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200590 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800591
592 /*
593 * lo_write_simple and lo_read_simple should have been covered
594 * by io submit style function like lo_rw_aio(), one blocker
595 * is that lo_read_simple() need to call flush_dcache_page after
596 * the page is written from kernel, and it isn't easy to handle
597 * this in io submit style function which submits all segments
598 * of the req at one time. And direct read IO doesn't need to
599 * run flush_dcache_page().
600 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200601 switch (req_op(rq)) {
602 case REQ_OP_FLUSH:
603 return lo_req_flush(lo, rq);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200604 case REQ_OP_WRITE_ZEROES:
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700605 /*
606 * If the caller doesn't want deallocation, call zeroout to
607 * write zeroes the range. Otherwise, punch them out.
608 */
609 return lo_fallocate(lo, rq, pos,
610 (rq->cmd_flags & REQ_NOUNMAP) ?
611 FALLOC_FL_ZERO_RANGE :
612 FALLOC_FL_PUNCH_HOLE);
613 case REQ_OP_DISCARD:
614 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200615 case REQ_OP_WRITE:
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200616 if (lo->transfer)
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200617 return lo_write_transfer(lo, rq, pos);
618 else if (cmd->use_aio)
619 return lo_rw_aio(lo, cmd, pos, WRITE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200620 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200621 return lo_write_simple(lo, rq, pos);
622 case REQ_OP_READ:
623 if (lo->transfer)
624 return lo_read_transfer(lo, rq, pos);
625 else if (cmd->use_aio)
626 return lo_rw_aio(lo, cmd, pos, READ);
627 else
628 return lo_read_simple(lo, rq, pos);
629 default:
630 WARN_ON_ONCE(1);
631 return -EIO;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800635static inline void loop_update_dio(struct loop_device *lo)
636{
Ira Weinyefbe3c22020-04-30 07:41:33 -0700637 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
638 lo->use_dio);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800639}
640
Ming Lei06f0e9e2015-05-06 12:26:24 +0800641static void loop_reread_partitions(struct loop_device *lo,
642 struct block_device *bdev)
643{
644 int rc;
645
Christoph Hellwigf0b870d2019-11-14 15:34:36 +0100646 mutex_lock(&bdev->bd_mutex);
647 rc = bdev_disk_changed(bdev, false);
648 mutex_unlock(&bdev->bd_mutex);
Ming Lei06f0e9e2015-05-06 12:26:24 +0800649 if (rc)
650 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
651 __func__, lo->lo_number, lo->lo_file_name, rc);
652}
653
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400654static inline int is_loop_device(struct file *file)
655{
656 struct inode *i = file->f_mapping->host;
657
658 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
659}
660
661static int loop_validate_file(struct file *file, struct block_device *bdev)
662{
663 struct inode *inode = file->f_mapping->host;
664 struct file *f = file;
665
666 /* Avoid recursion */
667 while (is_loop_device(f)) {
668 struct loop_device *l;
669
670 if (f->f_mapping->host->i_bdev == bdev)
671 return -EBADF;
672
673 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
Dongli Zhangf7c8a412019-03-18 20:23:17 +0800674 if (l->lo_state != Lo_bound) {
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400675 return -EINVAL;
676 }
677 f = l->lo_backing_file;
678 }
679 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
680 return -EINVAL;
681 return 0;
682}
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684/*
685 * loop_change_fd switched the backing store of a loopback device to
686 * a new file. This is useful for operating system installers to free up
687 * the original file and in High Availability environments to switch to
688 * an alternative location for the content in case of server meltdown.
689 * This can only work if the loop device is used read-only, and if the
690 * new backing store is the same size and type as the old backing store.
691 */
Al Virobb214882008-03-02 09:29:48 -0500692static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
693 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
Jan Kara1dded9a2018-11-08 14:01:15 +0100695 struct file *file = NULL, *old_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 int error;
Jan Kara85b0a542018-11-08 14:01:13 +0100697 bool partscan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Jan Karac28445f2018-11-08 14:01:16 +0100699 error = mutex_lock_killable(&loop_ctl_mutex);
Jan Karac3710772018-11-08 14:01:11 +0100700 if (error)
701 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 error = -ENXIO;
703 if (lo->lo_state != Lo_bound)
Jan Kara1dded9a2018-11-08 14:01:15 +0100704 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706 /* the loop device has to be read-only */
707 error = -EINVAL;
708 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
Jan Kara1dded9a2018-11-08 14:01:15 +0100709 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
711 error = -EBADF;
712 file = fget(arg);
713 if (!file)
Jan Kara1dded9a2018-11-08 14:01:15 +0100714 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400716 error = loop_validate_file(file, bdev);
717 if (error)
Jan Kara1dded9a2018-11-08 14:01:15 +0100718 goto out_err;
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 old_file = lo->lo_backing_file;
721
722 error = -EINVAL;
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 /* size of the new backing store needs to be the same */
725 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
Jan Kara1dded9a2018-11-08 14:01:15 +0100726 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 /* and ... switch */
Omar Sandoval43cade802017-08-24 00:03:44 -0700729 blk_mq_freeze_queue(lo->lo_queue);
730 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
731 lo->lo_backing_file = file;
732 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
733 mapping_set_gfp_mask(file->f_mapping,
734 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
735 loop_update_dio(lo);
736 blk_mq_unfreeze_queue(lo->lo_queue);
Jan Kara85b0a542018-11-08 14:01:13 +0100737 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Jan Karac3710772018-11-08 14:01:11 +0100738 mutex_unlock(&loop_ctl_mutex);
Jan Kara1dded9a2018-11-08 14:01:15 +0100739 /*
740 * We must drop file reference outside of loop_ctl_mutex as dropping
741 * the file ref can take bd_mutex which creates circular locking
742 * dependency.
743 */
744 fput(old_file);
Jan Kara85b0a542018-11-08 14:01:13 +0100745 if (partscan)
746 loop_reread_partitions(lo, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return 0;
748
Jan Kara1dded9a2018-11-08 14:01:15 +0100749out_err:
Jan Karac3710772018-11-08 14:01:11 +0100750 mutex_unlock(&loop_ctl_mutex);
Jan Kara1dded9a2018-11-08 14:01:15 +0100751 if (file)
752 fput(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return error;
754}
755
Milan Brozee862732010-08-23 15:16:00 +0200756/* loop sysfs attributes */
757
758static ssize_t loop_attr_show(struct device *dev, char *page,
759 ssize_t (*callback)(struct loop_device *, char *))
760{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200761 struct gendisk *disk = dev_to_disk(dev);
762 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200763
Kay Sievers34dd82a2011-07-31 22:08:04 +0200764 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200765}
766
767#define LOOP_ATTR_RO(_name) \
768static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
769static ssize_t loop_attr_do_show_##_name(struct device *d, \
770 struct device_attribute *attr, char *b) \
771{ \
772 return loop_attr_show(d, b, loop_attr_##_name##_show); \
773} \
774static struct device_attribute loop_attr_##_name = \
Joe Perches5657a812018-05-24 13:38:59 -0600775 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
Milan Brozee862732010-08-23 15:16:00 +0200776
777static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
778{
779 ssize_t ret;
780 char *p = NULL;
781
Kay Sievers05eb0f22011-07-31 22:21:35 +0200782 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200783 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200784 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200785 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200786
787 if (IS_ERR_OR_NULL(p))
788 ret = PTR_ERR(p);
789 else {
790 ret = strlen(p);
791 memmove(buf, p, ret);
792 buf[ret++] = '\n';
793 buf[ret] = 0;
794 }
795
796 return ret;
797}
798
799static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
800{
801 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
802}
803
804static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
805{
806 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
807}
808
809static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
810{
811 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
812
813 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
814}
815
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200816static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
817{
818 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
819
820 return sprintf(buf, "%s\n", partscan ? "1" : "0");
821}
822
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800823static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
824{
825 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
826
827 return sprintf(buf, "%s\n", dio ? "1" : "0");
828}
829
Milan Brozee862732010-08-23 15:16:00 +0200830LOOP_ATTR_RO(backing_file);
831LOOP_ATTR_RO(offset);
832LOOP_ATTR_RO(sizelimit);
833LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200834LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800835LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200836
837static struct attribute *loop_attrs[] = {
838 &loop_attr_backing_file.attr,
839 &loop_attr_offset.attr,
840 &loop_attr_sizelimit.attr,
841 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200842 &loop_attr_partscan.attr,
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800843 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200844 NULL,
845};
846
847static struct attribute_group loop_attribute_group = {
848 .name = "loop",
849 .attrs= loop_attrs,
850};
851
Tetsuo Handad3349b62018-05-04 10:58:09 -0600852static void loop_sysfs_init(struct loop_device *lo)
Milan Brozee862732010-08-23 15:16:00 +0200853{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600854 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
855 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200856}
857
858static void loop_sysfs_exit(struct loop_device *lo)
859{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600860 if (lo->sysfs_inited)
861 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
862 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200863}
864
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200865static void loop_config_discard(struct loop_device *lo)
866{
867 struct file *file = lo->lo_backing_file;
868 struct inode *inode = file->f_mapping->host;
869 struct request_queue *q = lo->lo_queue;
870
871 /*
Evan Greenc52abf52020-04-03 16:43:04 +0200872 * If the backing device is a block device, mirror its zeroing
873 * capability. Set the discard sectors to the block device's zeroing
874 * capabilities because loop discards result in blkdev_issue_zeroout(),
875 * not blkdev_issue_discard(). This maintains consistent behavior with
876 * file-backed loop devices: discarded regions read back as zero.
877 */
878 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
879 struct request_queue *backingq;
880
881 backingq = bdev_get_queue(inode->i_bdev);
882 blk_queue_max_discard_sectors(q,
883 backingq->limits.max_write_zeroes_sectors);
884
885 blk_queue_max_write_zeroes_sectors(q,
886 backingq->limits.max_write_zeroes_sectors);
887
888 /*
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200889 * We use punch hole to reclaim the free space used by the
Olaf Hering12a64d22014-01-21 14:39:24 -0800890 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200891 * encryption is enabled, because it may give an attacker
892 * useful information.
893 */
Evan Greenc52abf52020-04-03 16:43:04 +0200894 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200895 q->limits.discard_granularity = 0;
896 q->limits.discard_alignment = 0;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600897 blk_queue_max_discard_sectors(q, 0);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200898 blk_queue_max_write_zeroes_sectors(q, 0);
Evan Greenc52abf52020-04-03 16:43:04 +0200899
900 } else {
901 q->limits.discard_granularity = inode->i_sb->s_blocksize;
902 q->limits.discard_alignment = 0;
903
904 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
905 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200906 }
907
Evan Greenc52abf52020-04-03 16:43:04 +0200908 if (q->limits.max_write_zeroes_sectors)
909 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
910 else
911 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200912}
913
Ming Leie03a3d72015-08-17 10:31:48 +0800914static void loop_unprepare_queue(struct loop_device *lo)
915{
Petr Mladek39891442016-10-11 13:55:20 -0700916 kthread_flush_worker(&lo->worker);
Ming Leie03a3d72015-08-17 10:31:48 +0800917 kthread_stop(lo->worker_task);
918}
919
NeilBrownb2ee7d42017-06-16 15:02:09 +1000920static int loop_kthread_worker_fn(void *worker_ptr)
921{
Mikulas Patockad0a255e2019-08-08 11:17:01 -0400922 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
NeilBrownb2ee7d42017-06-16 15:02:09 +1000923 return kthread_worker_fn(worker_ptr);
924}
925
Ming Leie03a3d72015-08-17 10:31:48 +0800926static int loop_prepare_queue(struct loop_device *lo)
927{
Petr Mladek39891442016-10-11 13:55:20 -0700928 kthread_init_worker(&lo->worker);
NeilBrownb2ee7d42017-06-16 15:02:09 +1000929 lo->worker_task = kthread_run(loop_kthread_worker_fn,
Ming Leie03a3d72015-08-17 10:31:48 +0800930 &lo->worker, "loop%d", lo->lo_number);
931 if (IS_ERR(lo->worker_task))
932 return -ENOMEM;
933 set_user_nice(lo->worker_task, MIN_NICE);
934 return 0;
935}
936
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -0700937static void loop_update_rotational(struct loop_device *lo)
938{
939 struct file *file = lo->lo_backing_file;
940 struct inode *file_inode = file->f_mapping->host;
941 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
942 struct request_queue *q = lo->lo_queue;
943 bool nonrot = true;
944
945 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
946 if (file_bdev)
947 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
948
949 if (nonrot)
950 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
951 else
952 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
953}
954
Al Virobb214882008-03-02 09:29:48 -0500955static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 struct block_device *bdev, unsigned int arg)
957{
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400958 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 struct inode *inode;
960 struct address_space *mapping;
Jan Kara89e524c02019-07-30 13:10:14 +0200961 struct block_device *claimed_bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 int lo_flags = 0;
963 int error;
964 loff_t size;
Jan Kara85b0a542018-11-08 14:01:13 +0100965 bool partscan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /* This is safe, since we have a reference from open(). */
968 __module_get(THIS_MODULE);
969
970 error = -EBADF;
971 file = fget(arg);
972 if (!file)
973 goto out;
974
Jan Kara33ec3e532019-05-16 16:01:27 +0200975 /*
976 * If we don't hold exclusive handle for the device, upgrade to it
977 * here to avoid changing device under exclusive owner.
978 */
979 if (!(mode & FMODE_EXCL)) {
Jan Kara89e524c02019-07-30 13:10:14 +0200980 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
981 if (IS_ERR(claimed_bdev)) {
982 error = PTR_ERR(claimed_bdev);
Jan Kara33ec3e532019-05-16 16:01:27 +0200983 goto out_putf;
Jan Kara89e524c02019-07-30 13:10:14 +0200984 }
Jan Kara33ec3e532019-05-16 16:01:27 +0200985 }
986
Jan Karac28445f2018-11-08 14:01:16 +0100987 error = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara757ecf42018-11-08 14:01:10 +0100988 if (error)
Jan Kara33ec3e532019-05-16 16:01:27 +0200989 goto out_bdev;
Jan Kara757ecf42018-11-08 14:01:10 +0100990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 error = -EBUSY;
992 if (lo->lo_state != Lo_unbound)
Jan Kara757ecf42018-11-08 14:01:10 +0100993 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400995 error = loop_validate_file(file, bdev);
996 if (error)
Jan Kara757ecf42018-11-08 14:01:10 +0100997 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
999 mapping = file->f_mapping;
1000 inode = mapping->host;
1001
Christoph Hellwig456be142011-10-17 12:57:20 +02001002 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
Al Viro283e7e52015-04-03 15:21:59 -04001003 !file->f_op->write_iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 lo_flags |= LO_FLAGS_READ_ONLY;
1005
Christoph Hellwig456be142011-10-17 12:57:20 +02001006 error = -EFBIG;
1007 size = get_loop_size(lo, file);
1008 if ((loff_t)(sector_t)size != size)
Jan Kara757ecf42018-11-08 14:01:10 +01001009 goto out_unlock;
Ming Leie03a3d72015-08-17 10:31:48 +08001010 error = loop_prepare_queue(lo);
1011 if (error)
Jan Kara757ecf42018-11-08 14:01:10 +01001012 goto out_unlock;
Christoph Hellwig456be142011-10-17 12:57:20 +02001013
1014 error = 0;
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
1017
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001018 lo->use_dio = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 lo->lo_device = bdev;
1020 lo->lo_flags = lo_flags;
1021 lo->lo_backing_file = file;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +02001022 lo->transfer = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 lo->ioctl = NULL;
1024 lo->lo_sizelimit = 0;
1025 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1026 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1027
Nikanth Karthikesan68db1962009-03-24 12:29:54 +01001028 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Jens Axboe21d07272016-03-30 10:09:35 -06001029 blk_queue_write_cache(lo->lo_queue, true, false);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +01001030
Ira Weinyefbe3c22020-04-30 07:41:33 -07001031 if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) {
Martijn Coenen85560112019-09-04 21:49:01 +02001032 /* In case of direct I/O, match underlying block size */
1033 unsigned short bsize = bdev_logical_block_size(
1034 inode->i_sb->s_bdev);
1035
1036 blk_queue_logical_block_size(lo->lo_queue, bsize);
1037 blk_queue_physical_block_size(lo->lo_queue, bsize);
1038 blk_queue_io_min(lo->lo_queue, bsize);
1039 }
1040
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -07001041 loop_update_rotational(lo);
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001042 loop_update_dio(lo);
Ken Chen73285082007-05-08 00:28:20 -07001043 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +02001045 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001046 /* let user-space know about the new size */
1047 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Omar Sandoval8a0740c2017-08-24 00:03:41 -07001049 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1050 block_size(inode->i_bdev) : PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001052 lo->lo_state = Lo_bound;
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001053 if (part_shift)
1054 lo->lo_flags |= LO_FLAGS_PARTSCAN;
Jan Kara85b0a542018-11-08 14:01:13 +01001055 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001056
1057 /* Grab the block_device to prevent its destruction after we
Jan Karaa2505b72018-11-08 14:01:06 +01001058 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001059 */
1060 bdgrab(bdev);
Jan Kara757ecf42018-11-08 14:01:10 +01001061 mutex_unlock(&loop_ctl_mutex);
Jan Kara85b0a542018-11-08 14:01:13 +01001062 if (partscan)
1063 loop_reread_partitions(lo, bdev);
Jan Kara89e524c02019-07-30 13:10:14 +02001064 if (claimed_bdev)
1065 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 return 0;
1067
Jan Kara757ecf42018-11-08 14:01:10 +01001068out_unlock:
1069 mutex_unlock(&loop_ctl_mutex);
Jan Kara33ec3e532019-05-16 16:01:27 +02001070out_bdev:
Jan Kara89e524c02019-07-30 13:10:14 +02001071 if (claimed_bdev)
1072 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
Jan Kara757ecf42018-11-08 14:01:10 +01001073out_putf:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 fput(file);
Jan Kara757ecf42018-11-08 14:01:10 +01001075out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 /* This is safe: open() is still holding a reference. */
1077 module_put(THIS_MODULE);
1078 return error;
1079}
1080
1081static int
1082loop_release_xfer(struct loop_device *lo)
1083{
1084 int err = 0;
1085 struct loop_func_table *xfer = lo->lo_encryption;
1086
1087 if (xfer) {
1088 if (xfer->release)
1089 err = xfer->release(lo);
1090 lo->transfer = NULL;
1091 lo->lo_encryption = NULL;
1092 module_put(xfer->owner);
1093 }
1094 return err;
1095}
1096
1097static int
1098loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1099 const struct loop_info64 *i)
1100{
1101 int err = 0;
1102
1103 if (xfer) {
1104 struct module *owner = xfer->owner;
1105
1106 if (!try_module_get(owner))
1107 return -EINVAL;
1108 if (xfer->init)
1109 err = xfer->init(lo, i);
1110 if (err)
1111 module_put(owner);
1112 else
1113 lo->lo_encryption = xfer;
1114 }
1115 return err;
1116}
1117
Jan Kara0da03ca2018-11-08 14:01:14 +01001118static int __loop_clr_fd(struct loop_device *lo, bool release)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
Jan Kara7ccd0792018-11-08 14:01:07 +01001120 struct file *filp = NULL;
Al Virob4e3ca12005-10-21 03:22:34 -04001121 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +02001122 struct block_device *bdev = lo->lo_device;
Jan Kara7ccd0792018-11-08 14:01:07 +01001123 int err = 0;
Jan Kara0da03ca2018-11-08 14:01:14 +01001124 bool partscan = false;
1125 int lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Jan Kara7ccd0792018-11-08 14:01:07 +01001127 mutex_lock(&loop_ctl_mutex);
1128 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1129 err = -ENXIO;
1130 goto out_unlock;
1131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Jan Kara7ccd0792018-11-08 14:01:07 +01001133 filp = lo->lo_backing_file;
1134 if (filp == NULL) {
1135 err = -EINVAL;
1136 goto out_unlock;
1137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Ming Leif8933662015-05-06 12:26:23 +08001139 /* freeze request queue during the transition */
1140 blk_mq_freeze_queue(lo->lo_queue);
1141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001144 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 loop_release_xfer(lo);
1147 lo->transfer = NULL;
1148 lo->ioctl = NULL;
1149 lo->lo_device = NULL;
1150 lo->lo_encryption = NULL;
1151 lo->lo_offset = 0;
1152 lo->lo_sizelimit = 0;
1153 lo->lo_encrypt_key_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1155 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1156 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001157 blk_queue_logical_block_size(lo->lo_queue, 512);
Omar Sandovalbf093752017-09-05 14:24:47 -07001158 blk_queue_physical_block_size(lo->lo_queue, 512);
1159 blk_queue_io_min(lo->lo_queue, 512);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001160 if (bdev) {
1161 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -05001162 invalidate_bdev(bdev);
Jeff Laytoneedffa22018-05-21 14:35:03 -04001163 bdev->bd_inode->i_mapping->wb_err = 0;
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001164 }
Ken Chen73285082007-05-08 00:28:20 -07001165 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001166 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001167 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001168 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001169 /* let user-space know about this change */
1170 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 mapping_set_gfp_mask(filp->f_mapping, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 /* This is safe: open() is still holding a reference. */
1174 module_put(THIS_MODULE);
Ming Leif8933662015-05-06 12:26:23 +08001175 blk_mq_unfreeze_queue(lo->lo_queue);
1176
Jan Kara0da03ca2018-11-08 14:01:14 +01001177 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1178 lo_number = lo->lo_number;
Jan Kara0da03ca2018-11-08 14:01:14 +01001179 loop_unprepare_queue(lo);
1180out_unlock:
1181 mutex_unlock(&loop_ctl_mutex);
1182 if (partscan) {
Jan Karad57f3372018-11-08 14:01:12 +01001183 /*
1184 * bd_mutex has been held already in release path, so don't
1185 * acquire it if this function is called in such case.
1186 *
1187 * If the reread partition isn't from release path, lo_refcnt
1188 * must be at least one and it can only become zero when the
1189 * current holder is released.
1190 */
Christoph Hellwigf0b870d2019-11-14 15:34:36 +01001191 if (!release)
1192 mutex_lock(&bdev->bd_mutex);
1193 err = bdev_disk_changed(bdev, false);
1194 if (!release)
1195 mutex_unlock(&bdev->bd_mutex);
Dongli Zhang40853d62019-02-22 22:10:19 +08001196 if (err)
1197 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1198 __func__, lo_number, err);
Jan Karad57f3372018-11-08 14:01:12 +01001199 /* Device is gone, no point in returning error */
1200 err = 0;
1201 }
Dongli Zhang758a58d2019-02-22 22:10:20 +08001202
1203 /*
1204 * lo->lo_state is set to Lo_unbound here after above partscan has
1205 * finished.
1206 *
1207 * There cannot be anybody else entering __loop_clr_fd() as
1208 * lo->lo_backing_file is already cleared and Lo_rundown state
1209 * protects us from all the other places trying to change the 'lo'
1210 * device.
1211 */
1212 mutex_lock(&loop_ctl_mutex);
1213 lo->lo_flags = 0;
1214 if (!part_shift)
1215 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1216 lo->lo_state = Lo_unbound;
1217 mutex_unlock(&loop_ctl_mutex);
1218
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001219 /*
Tetsuo Handa310ca162018-11-08 14:01:02 +01001220 * Need not hold loop_ctl_mutex to fput backing file.
1221 * Calling fput holding loop_ctl_mutex triggers a circular
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001222 * lock dependency possibility warning as fput can take
Tetsuo Handa310ca162018-11-08 14:01:02 +01001223 * bd_mutex which is usually taken before loop_ctl_mutex.
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001224 */
Jan Kara7ccd0792018-11-08 14:01:07 +01001225 if (filp)
1226 fput(filp);
1227 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
1229
Jan Karaa2505b72018-11-08 14:01:06 +01001230static int loop_clr_fd(struct loop_device *lo)
1231{
Jan Kara7ccd0792018-11-08 14:01:07 +01001232 int err;
1233
Jan Karac28445f2018-11-08 14:01:16 +01001234 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara7ccd0792018-11-08 14:01:07 +01001235 if (err)
1236 return err;
1237 if (lo->lo_state != Lo_bound) {
1238 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001239 return -ENXIO;
Jan Kara7ccd0792018-11-08 14:01:07 +01001240 }
Jan Karaa2505b72018-11-08 14:01:06 +01001241 /*
1242 * If we've explicitly asked to tear down the loop device,
1243 * and it has an elevated reference count, set it for auto-teardown when
1244 * the last reference goes away. This stops $!~#$@ udev from
1245 * preventing teardown because it decided that it needs to run blkid on
1246 * the loopback device whenever they appear. xfstests is notorious for
1247 * failing tests because blkid via udev races with a losetup
1248 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1249 * command to fail with EBUSY.
1250 */
1251 if (atomic_read(&lo->lo_refcnt) > 1) {
1252 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1253 mutex_unlock(&loop_ctl_mutex);
1254 return 0;
1255 }
1256 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001257 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001258
Jan Kara0da03ca2018-11-08 14:01:14 +01001259 return __loop_clr_fd(lo, false);
Jan Karaa2505b72018-11-08 14:01:06 +01001260}
1261
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262static int
1263loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1264{
1265 int err;
1266 struct loop_func_table *xfer;
Eric W. Biedermane4849732012-02-11 11:23:51 -08001267 kuid_t uid = current_uid();
Jan Kara85b0a542018-11-08 14:01:13 +01001268 struct block_device *bdev;
1269 bool partscan = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Jan Karac28445f2018-11-08 14:01:16 +01001271 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara550df5fd2018-11-08 14:01:09 +01001272 if (err)
1273 return err;
David Howellsb0fafa82008-11-14 10:38:41 +11001274 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001275 !uid_eq(lo->lo_key_owner, uid) &&
Jan Kara550df5fd2018-11-08 14:01:09 +01001276 !capable(CAP_SYS_ADMIN)) {
1277 err = -EPERM;
1278 goto out_unlock;
1279 }
1280 if (lo->lo_state != Lo_bound) {
1281 err = -ENXIO;
1282 goto out_unlock;
1283 }
1284 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1285 err = -EINVAL;
1286 goto out_unlock;
1287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001289 if (lo->lo_offset != info->lo_offset ||
1290 lo->lo_sizelimit != info->lo_sizelimit) {
1291 sync_blockdev(lo->lo_device);
1292 kill_bdev(lo->lo_device);
1293 }
1294
Ming Leiecdd0952017-02-11 11:40:45 +08001295 /* I/O need to be drained during transfer transition */
1296 blk_mq_freeze_queue(lo->lo_queue);
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 err = loop_release_xfer(lo);
1299 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001300 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
1302 if (info->lo_encrypt_type) {
1303 unsigned int type = info->lo_encrypt_type;
1304
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001305 if (type >= MAX_LO_CRYPT) {
1306 err = -EINVAL;
Jan Kara550df5fd2018-11-08 14:01:09 +01001307 goto out_unfreeze;
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 xfer = xfer_funcs[type];
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001310 if (xfer == NULL) {
1311 err = -EINVAL;
Jan Kara550df5fd2018-11-08 14:01:09 +01001312 goto out_unfreeze;
Tetsuo Handa1e047ea2018-04-06 10:03:17 +09001313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 } else
1315 xfer = NULL;
1316
1317 err = loop_init_xfer(lo, xfer, info);
1318 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001319 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321 if (lo->lo_offset != info->lo_offset ||
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001322 lo->lo_sizelimit != info->lo_sizelimit) {
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001323 /* kill_bdev should have truncated all the pages */
1324 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1325 err = -EAGAIN;
1326 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1327 __func__, lo->lo_number, lo->lo_file_name,
1328 lo->lo_device->bd_inode->i_mapping->nrpages);
1329 goto out_unfreeze;
1330 }
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001331 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
Ming Leiecdd0952017-02-11 11:40:45 +08001332 err = -EFBIG;
Jan Kara550df5fd2018-11-08 14:01:09 +01001333 goto out_unfreeze;
Ming Leiecdd0952017-02-11 11:40:45 +08001334 }
Arnd Bergmannb040ad92017-06-09 12:19:18 +02001335 }
Guo Chao541c7422013-02-21 15:16:46 -08001336
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001337 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1340 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1341 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1342 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1343
1344 if (!xfer)
1345 xfer = &none_funcs;
1346 lo->transfer = xfer->transfer;
1347 lo->ioctl = xfer->ioctl;
1348
David Woodhouse96c58652008-02-06 01:36:27 -08001349 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1350 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1351 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1354 lo->lo_init[0] = info->lo_init[0];
1355 lo->lo_init[1] = info->lo_init[1];
1356 if (info->lo_encrypt_key_size) {
1357 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1358 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001359 lo->lo_key_owner = uid;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +02001360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001362 /* update dio if lo_offset or transfer is changed */
1363 __loop_update_dio(lo, lo->use_dio);
1364
Jan Kara550df5fd2018-11-08 14:01:09 +01001365out_unfreeze:
Ming Leiecdd0952017-02-11 11:40:45 +08001366 blk_mq_unfreeze_queue(lo->lo_queue);
Omar Sandovale02898b42017-03-01 10:42:38 -08001367
1368 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1369 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1370 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1371 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
Jan Kara85b0a542018-11-08 14:01:13 +01001372 bdev = lo->lo_device;
1373 partscan = true;
Omar Sandovale02898b42017-03-01 10:42:38 -08001374 }
Jan Kara550df5fd2018-11-08 14:01:09 +01001375out_unlock:
1376 mutex_unlock(&loop_ctl_mutex);
Jan Kara85b0a542018-11-08 14:01:13 +01001377 if (partscan)
1378 loop_reread_partitions(lo, bdev);
Omar Sandovale02898b42017-03-01 10:42:38 -08001379
Ming Leiecdd0952017-02-11 11:40:45 +08001380 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381}
1382
1383static int
1384loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1385{
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001386 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 struct kstat stat;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001388 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Jan Karac28445f2018-11-08 14:01:16 +01001390 ret = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001391 if (ret)
1392 return ret;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001393 if (lo->lo_state != Lo_bound) {
Tetsuo Handa310ca162018-11-08 14:01:02 +01001394 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return -ENXIO;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001396 }
1397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 memset(info, 0, sizeof(*info));
1399 info->lo_number = lo->lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 info->lo_offset = lo->lo_offset;
1401 info->lo_sizelimit = lo->lo_sizelimit;
1402 info->lo_flags = lo->lo_flags;
1403 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1404 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1405 info->lo_encrypt_type =
1406 lo->lo_encryption ? lo->lo_encryption->number : 0;
1407 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1408 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1409 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1410 lo->lo_encrypt_key_size);
1411 }
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001412
Tetsuo Handa310ca162018-11-08 14:01:02 +01001413 /* Drop loop_ctl_mutex while we call into the filesystem. */
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001414 path = lo->lo_backing_file->f_path;
1415 path_get(&path);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001416 mutex_unlock(&loop_ctl_mutex);
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001417 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001418 if (!ret) {
1419 info->lo_device = huge_encode_dev(stat.dev);
1420 info->lo_inode = stat.ino;
1421 info->lo_rdevice = huge_encode_dev(stat.rdev);
1422 }
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001423 path_put(&path);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001424 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425}
1426
1427static void
1428loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1429{
1430 memset(info64, 0, sizeof(*info64));
1431 info64->lo_number = info->lo_number;
1432 info64->lo_device = info->lo_device;
1433 info64->lo_inode = info->lo_inode;
1434 info64->lo_rdevice = info->lo_rdevice;
1435 info64->lo_offset = info->lo_offset;
1436 info64->lo_sizelimit = 0;
1437 info64->lo_encrypt_type = info->lo_encrypt_type;
1438 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1439 info64->lo_flags = info->lo_flags;
1440 info64->lo_init[0] = info->lo_init[0];
1441 info64->lo_init[1] = info->lo_init[1];
1442 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1443 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1444 else
1445 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1446 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1447}
1448
1449static int
1450loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1451{
1452 memset(info, 0, sizeof(*info));
1453 info->lo_number = info64->lo_number;
1454 info->lo_device = info64->lo_device;
1455 info->lo_inode = info64->lo_inode;
1456 info->lo_rdevice = info64->lo_rdevice;
1457 info->lo_offset = info64->lo_offset;
1458 info->lo_encrypt_type = info64->lo_encrypt_type;
1459 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1460 info->lo_flags = info64->lo_flags;
1461 info->lo_init[0] = info64->lo_init[0];
1462 info->lo_init[1] = info64->lo_init[1];
1463 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1464 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1465 else
1466 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1467 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1468
1469 /* error in case values were truncated */
1470 if (info->lo_device != info64->lo_device ||
1471 info->lo_rdevice != info64->lo_rdevice ||
1472 info->lo_inode != info64->lo_inode ||
1473 info->lo_offset != info64->lo_offset)
1474 return -EOVERFLOW;
1475
1476 return 0;
1477}
1478
1479static int
1480loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1481{
1482 struct loop_info info;
1483 struct loop_info64 info64;
1484
1485 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1486 return -EFAULT;
1487 loop_info64_from_old(&info, &info64);
1488 return loop_set_status(lo, &info64);
1489}
1490
1491static int
1492loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1493{
1494 struct loop_info64 info64;
1495
1496 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1497 return -EFAULT;
1498 return loop_set_status(lo, &info64);
1499}
1500
1501static int
1502loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1503 struct loop_info info;
1504 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001505 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001507 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001508 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001509 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 if (!err)
1511 err = loop_info64_to_old(&info64, &info);
1512 if (!err && copy_to_user(arg, &info, sizeof(info)))
1513 err = -EFAULT;
1514
1515 return err;
1516}
1517
1518static int
1519loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1520 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001521 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001523 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001524 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001525 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1527 err = -EFAULT;
1528
1529 return err;
1530}
1531
Hannes Reinecke51001b72017-06-08 13:46:44 +02001532static int loop_set_capacity(struct loop_device *lo)
J. R. Okajima53d66602009-03-31 15:23:43 -07001533{
J. R. Okajima53d66602009-03-31 15:23:43 -07001534 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001535 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001536
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001537 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001538}
1539
Ming Leiab1cb272015-08-17 10:31:50 +08001540static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1541{
1542 int error = -ENXIO;
1543 if (lo->lo_state != Lo_bound)
1544 goto out;
1545
1546 __loop_update_dio(lo, !!arg);
1547 if (lo->use_dio == !!arg)
1548 return 0;
1549 error = -EINVAL;
1550 out:
1551 return error;
1552}
1553
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001554static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1555{
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001556 int err = 0;
1557
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001558 if (lo->lo_state != Lo_bound)
1559 return -ENXIO;
1560
1561 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1562 return -EINVAL;
1563
Martijn Coenen7e81f992020-03-10 14:12:30 +01001564 if (lo->lo_queue->limits.logical_block_size == arg)
1565 return 0;
1566
1567 sync_blockdev(lo->lo_device);
1568 kill_bdev(lo->lo_device);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001569
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001570 blk_mq_freeze_queue(lo->lo_queue);
1571
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001572 /* kill_bdev should have truncated all the pages */
Martijn Coenen7e81f992020-03-10 14:12:30 +01001573 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001574 err = -EAGAIN;
1575 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1576 __func__, lo->lo_number, lo->lo_file_name,
1577 lo->lo_device->bd_inode->i_mapping->nrpages);
1578 goto out_unfreeze;
1579 }
1580
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001581 blk_queue_logical_block_size(lo->lo_queue, arg);
Omar Sandovalbf093752017-09-05 14:24:47 -07001582 blk_queue_physical_block_size(lo->lo_queue, arg);
1583 blk_queue_io_min(lo->lo_queue, arg);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001584 loop_update_dio(lo);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001585out_unfreeze:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001586 blk_mq_unfreeze_queue(lo->lo_queue);
1587
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001588 return err;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001589}
1590
Jan Karaa1316542018-11-08 14:01:05 +01001591static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1592 unsigned long arg)
1593{
1594 int err;
1595
Jan Karac28445f2018-11-08 14:01:16 +01001596 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Karaa1316542018-11-08 14:01:05 +01001597 if (err)
1598 return err;
1599 switch (cmd) {
1600 case LOOP_SET_CAPACITY:
1601 err = loop_set_capacity(lo);
1602 break;
1603 case LOOP_SET_DIRECT_IO:
1604 err = loop_set_dio(lo, arg);
1605 break;
1606 case LOOP_SET_BLOCK_SIZE:
1607 err = loop_set_block_size(lo, arg);
1608 break;
1609 default:
1610 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1611 }
1612 mutex_unlock(&loop_ctl_mutex);
1613 return err;
1614}
1615
Al Virobb214882008-03-02 09:29:48 -05001616static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 unsigned int cmd, unsigned long arg)
1618{
Al Virobb214882008-03-02 09:29:48 -05001619 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 int err;
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 switch (cmd) {
1623 case LOOP_SET_FD:
Jan Kara757ecf42018-11-08 14:01:10 +01001624 return loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 case LOOP_CHANGE_FD:
Jan Karac3710772018-11-08 14:01:11 +01001626 return loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 case LOOP_CLR_FD:
Jan Kara7ccd0792018-11-08 14:01:07 +01001628 return loop_clr_fd(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001630 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001631 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001632 err = loop_set_status_old(lo,
1633 (struct loop_info __user *)arg);
Jan Karaa1316542018-11-08 14:01:05 +01001634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 break;
1636 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001637 return loop_get_status_old(lo, (struct loop_info __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001639 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001640 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001641 err = loop_set_status64(lo,
1642 (struct loop_info64 __user *) arg);
Jan Karaa1316542018-11-08 14:01:05 +01001643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 break;
1645 case LOOP_GET_STATUS64:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001646 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
J. R. Okajima53d66602009-03-31 15:23:43 -07001647 case LOOP_SET_CAPACITY:
Ming Leiab1cb272015-08-17 10:31:50 +08001648 case LOOP_SET_DIRECT_IO:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001649 case LOOP_SET_BLOCK_SIZE:
Jan Karaa1316542018-11-08 14:01:05 +01001650 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1651 return -EPERM;
1652 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 default:
Jan Karaa1316542018-11-08 14:01:05 +01001654 err = lo_simple_ioctl(lo, cmd, arg);
1655 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 }
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return err;
1659}
1660
David Howells863d5b822006-08-29 19:06:14 +01001661#ifdef CONFIG_COMPAT
1662struct compat_loop_info {
1663 compat_int_t lo_number; /* ioctl r/o */
1664 compat_dev_t lo_device; /* ioctl r/o */
1665 compat_ulong_t lo_inode; /* ioctl r/o */
1666 compat_dev_t lo_rdevice; /* ioctl r/o */
1667 compat_int_t lo_offset;
1668 compat_int_t lo_encrypt_type;
1669 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1670 compat_int_t lo_flags; /* ioctl r/o */
1671 char lo_name[LO_NAME_SIZE];
1672 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1673 compat_ulong_t lo_init[2];
1674 char reserved[4];
1675};
1676
1677/*
1678 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1679 * - noinlined to reduce stack space usage in main part of driver
1680 */
1681static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001682loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001683 struct loop_info64 *info64)
1684{
1685 struct compat_loop_info info;
1686
1687 if (copy_from_user(&info, arg, sizeof(info)))
1688 return -EFAULT;
1689
1690 memset(info64, 0, sizeof(*info64));
1691 info64->lo_number = info.lo_number;
1692 info64->lo_device = info.lo_device;
1693 info64->lo_inode = info.lo_inode;
1694 info64->lo_rdevice = info.lo_rdevice;
1695 info64->lo_offset = info.lo_offset;
1696 info64->lo_sizelimit = 0;
1697 info64->lo_encrypt_type = info.lo_encrypt_type;
1698 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1699 info64->lo_flags = info.lo_flags;
1700 info64->lo_init[0] = info.lo_init[0];
1701 info64->lo_init[1] = info.lo_init[1];
1702 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1703 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1704 else
1705 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1706 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1707 return 0;
1708}
1709
1710/*
1711 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1712 * - noinlined to reduce stack space usage in main part of driver
1713 */
1714static noinline int
1715loop_info64_to_compat(const struct loop_info64 *info64,
1716 struct compat_loop_info __user *arg)
1717{
1718 struct compat_loop_info info;
1719
1720 memset(&info, 0, sizeof(info));
1721 info.lo_number = info64->lo_number;
1722 info.lo_device = info64->lo_device;
1723 info.lo_inode = info64->lo_inode;
1724 info.lo_rdevice = info64->lo_rdevice;
1725 info.lo_offset = info64->lo_offset;
1726 info.lo_encrypt_type = info64->lo_encrypt_type;
1727 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1728 info.lo_flags = info64->lo_flags;
1729 info.lo_init[0] = info64->lo_init[0];
1730 info.lo_init[1] = info64->lo_init[1];
1731 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1732 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1733 else
1734 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1735 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1736
1737 /* error in case values were truncated */
1738 if (info.lo_device != info64->lo_device ||
1739 info.lo_rdevice != info64->lo_rdevice ||
1740 info.lo_inode != info64->lo_inode ||
1741 info.lo_offset != info64->lo_offset ||
1742 info.lo_init[0] != info64->lo_init[0] ||
1743 info.lo_init[1] != info64->lo_init[1])
1744 return -EOVERFLOW;
1745
1746 if (copy_to_user(arg, &info, sizeof(info)))
1747 return -EFAULT;
1748 return 0;
1749}
1750
1751static int
1752loop_set_status_compat(struct loop_device *lo,
1753 const struct compat_loop_info __user *arg)
1754{
1755 struct loop_info64 info64;
1756 int ret;
1757
1758 ret = loop_info64_from_compat(arg, &info64);
1759 if (ret < 0)
1760 return ret;
1761 return loop_set_status(lo, &info64);
1762}
1763
1764static int
1765loop_get_status_compat(struct loop_device *lo,
1766 struct compat_loop_info __user *arg)
1767{
1768 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001769 int err;
David Howells863d5b822006-08-29 19:06:14 +01001770
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001771 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001772 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001773 err = loop_get_status(lo, &info64);
David Howells863d5b822006-08-29 19:06:14 +01001774 if (!err)
1775 err = loop_info64_to_compat(&info64, arg);
1776 return err;
1777}
1778
Al Virobb214882008-03-02 09:29:48 -05001779static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1780 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001781{
Al Virobb214882008-03-02 09:29:48 -05001782 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001783 int err;
1784
David Howells863d5b822006-08-29 19:06:14 +01001785 switch(cmd) {
1786 case LOOP_SET_STATUS:
Jan Kara550df5fd2018-11-08 14:01:09 +01001787 err = loop_set_status_compat(lo,
1788 (const struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001789 break;
1790 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001791 err = loop_get_status_compat(lo,
1792 (struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001793 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001794 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001795 case LOOP_CLR_FD:
1796 case LOOP_GET_STATUS64:
1797 case LOOP_SET_STATUS64:
1798 arg = (unsigned long) compat_ptr(arg);
Gustavo A. R. Silvad893ff82018-07-02 12:52:06 -05001799 /* fall through */
David Howells863d5b822006-08-29 19:06:14 +01001800 case LOOP_SET_FD:
1801 case LOOP_CHANGE_FD:
Evan Green9fea4b32018-07-02 16:03:46 -07001802 case LOOP_SET_BLOCK_SIZE:
Alessio Balsinifdbe4ee2019-08-07 01:48:28 +01001803 case LOOP_SET_DIRECT_IO:
Al Virobb214882008-03-02 09:29:48 -05001804 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001805 break;
1806 default:
1807 err = -ENOIOCTLCMD;
1808 break;
1809 }
David Howells863d5b822006-08-29 19:06:14 +01001810 return err;
1811}
1812#endif
1813
Al Virobb214882008-03-02 09:29:48 -05001814static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
Kay Sievers770fe302011-07-31 22:08:04 +02001816 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01001817 int err;
Kay Sievers770fe302011-07-31 22:08:04 +02001818
Jan Kara0a42e992018-11-08 14:01:04 +01001819 err = mutex_lock_killable(&loop_ctl_mutex);
1820 if (err)
1821 return err;
Kay Sievers770fe302011-07-31 22:08:04 +02001822 lo = bdev->bd_disk->private_data;
1823 if (!lo) {
1824 err = -ENXIO;
1825 goto out;
1826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Ming Leif8933662015-05-06 12:26:23 +08001828 atomic_inc(&lo->lo_refcnt);
Kay Sievers770fe302011-07-31 22:08:04 +02001829out:
Jan Kara0a42e992018-11-08 14:01:04 +01001830 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001831 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832}
1833
Jan Kara967d1dc2018-11-08 14:01:03 +01001834static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
Jan Kara967d1dc2018-11-08 14:01:03 +01001836 struct loop_device *lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Jan Kara0a42e992018-11-08 14:01:04 +01001838 mutex_lock(&loop_ctl_mutex);
Jan Kara967d1dc2018-11-08 14:01:03 +01001839 lo = disk->private_data;
Ming Leif8933662015-05-06 12:26:23 +08001840 if (atomic_dec_return(&lo->lo_refcnt))
Jan Kara0a42e992018-11-08 14:01:04 +01001841 goto out_unlock;
Ming Leif8933662015-05-06 12:26:23 +08001842
Milan Broz14f27932008-12-12 14:48:27 +01001843 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
Jan Karaa2505b72018-11-08 14:01:06 +01001844 if (lo->lo_state != Lo_bound)
1845 goto out_unlock;
1846 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001847 mutex_unlock(&loop_ctl_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001848 /*
1849 * In autoclear mode, stop the loop thread
1850 * and remove configuration after last close.
1851 */
Jan Kara0da03ca2018-11-08 14:01:14 +01001852 __loop_clr_fd(lo, true);
Jan Kara7ccd0792018-11-08 14:01:07 +01001853 return;
Omar Sandoval43cade802017-08-24 00:03:44 -07001854 } else if (lo->lo_state == Lo_bound) {
Milan Broz14f27932008-12-12 14:48:27 +01001855 /*
1856 * Otherwise keep thread (if running) and config,
1857 * but flush possible ongoing bios in thread.
1858 */
Omar Sandoval43cade802017-08-24 00:03:44 -07001859 blk_mq_freeze_queue(lo->lo_queue);
1860 blk_mq_unfreeze_queue(lo->lo_queue);
Milan Broz14f27932008-12-12 14:48:27 +01001861 }
David Woodhouse96c58652008-02-06 01:36:27 -08001862
Jan Kara0a42e992018-11-08 14:01:04 +01001863out_unlock:
Tetsuo Handa310ca162018-11-08 14:01:02 +01001864 mutex_unlock(&loop_ctl_mutex);
Linus Torvaldsae665012018-01-05 16:26:00 -08001865}
1866
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001867static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001869 .open = lo_open,
1870 .release = lo_release,
1871 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001872#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001873 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001874#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875};
1876
1877/*
1878 * And now the modules code and kernel interface.
1879 */
Ken Chen73285082007-05-08 00:28:20 -07001880static int max_loop;
Joe Perches5657a812018-05-24 13:38:59 -06001881module_param(max_loop, int, 0444);
Ken Chena47653f2007-06-08 13:46:44 -07001882MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Joe Perches5657a812018-05-24 13:38:59 -06001883module_param(max_part, int, 0444);
Laurent Vivier476a4812008-03-26 12:11:53 +01001884MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885MODULE_LICENSE("GPL");
1886MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1887
1888int loop_register_transfer(struct loop_func_table *funcs)
1889{
1890 unsigned int n = funcs->number;
1891
1892 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1893 return -EINVAL;
1894 xfer_funcs[n] = funcs;
1895 return 0;
1896}
1897
Kay Sievers34dd82a2011-07-31 22:08:04 +02001898static int unregister_transfer_cb(int id, void *ptr, void *data)
1899{
1900 struct loop_device *lo = ptr;
1901 struct loop_func_table *xfer = data;
1902
Tetsuo Handa310ca162018-11-08 14:01:02 +01001903 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001904 if (lo->lo_encryption == xfer)
1905 loop_release_xfer(lo);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001906 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001907 return 0;
1908}
1909
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910int loop_unregister_transfer(int number)
1911{
1912 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 struct loop_func_table *xfer;
1914
1915 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1916 return -EINVAL;
1917
1918 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001919 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 return 0;
1921}
1922
1923EXPORT_SYMBOL(loop_register_transfer);
1924EXPORT_SYMBOL(loop_unregister_transfer);
1925
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001926static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
Ming Leib5dd2f62014-12-31 13:22:57 +00001927 const struct blk_mq_queue_data *bd)
1928{
Jens Axboe1894e9162018-04-13 16:24:29 -06001929 struct request *rq = bd->rq;
1930 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1931 struct loop_device *lo = rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00001932
Jens Axboe1894e9162018-04-13 16:24:29 -06001933 blk_mq_start_request(rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001934
Ming Leif4aa4c72015-05-05 19:49:54 +08001935 if (lo->lo_state != Lo_bound)
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001936 return BLK_STS_IOERR;
Ming Leif4aa4c72015-05-05 19:49:54 +08001937
Jens Axboe1894e9162018-04-13 16:24:29 -06001938 switch (req_op(rq)) {
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001939 case REQ_OP_FLUSH:
1940 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +02001941 case REQ_OP_WRITE_ZEROES:
Ming Leibc07c102015-08-17 10:31:51 +08001942 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001943 break;
1944 default:
1945 cmd->use_aio = lo->use_dio;
1946 break;
1947 }
Ming Leibc07c102015-08-17 10:31:51 +08001948
Shaohua Lid4478e92017-09-25 13:07:22 -06001949 /* always use the first bio's css */
Shaohua Li0b508bc2017-09-26 11:02:12 -07001950#ifdef CONFIG_BLK_CGROUP
Dennis Zhoudb6638d2018-12-05 12:10:35 -05001951 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1952 cmd->css = &bio_blkcg(rq->bio)->css;
Shaohua Lid4478e92017-09-25 13:07:22 -06001953 css_get(cmd->css);
1954 } else
1955#endif
1956 cmd->css = NULL;
Petr Mladek39891442016-10-11 13:55:20 -07001957 kthread_queue_work(&lo->worker, &cmd->work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001958
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001959 return BLK_STS_OK;
Ming Leib5dd2f62014-12-31 13:22:57 +00001960}
1961
1962static void loop_handle_cmd(struct loop_cmd *cmd)
1963{
Jens Axboe1894e9162018-04-13 16:24:29 -06001964 struct request *rq = blk_mq_rq_from_pdu(cmd);
1965 const bool write = op_is_write(req_op(rq));
1966 struct loop_device *lo = rq->q->queuedata;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001967 int ret = 0;
Ming Leib5dd2f62014-12-31 13:22:57 +00001968
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001969 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1970 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00001971 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001972 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001973
Jens Axboe1894e9162018-04-13 16:24:29 -06001974 ret = do_req_filebacked(lo, rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001975 failed:
Ming Leibc07c102015-08-17 10:31:51 +08001976 /* complete non-aio request */
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001977 if (!cmd->use_aio || ret) {
Evan Green8cd55082020-04-03 16:43:03 +02001978 if (ret == -EOPNOTSUPP)
1979 cmd->ret = ret;
1980 else
1981 cmd->ret = ret ? -EIO : 0;
Jens Axboe1894e9162018-04-13 16:24:29 -06001982 blk_mq_complete_request(rq);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001983 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001984}
1985
Ming Leie03a3d72015-08-17 10:31:48 +08001986static void loop_queue_work(struct kthread_work *work)
Ming Leib5dd2f62014-12-31 13:22:57 +00001987{
1988 struct loop_cmd *cmd =
Ming Leie03a3d72015-08-17 10:31:48 +08001989 container_of(work, struct loop_cmd, work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001990
1991 loop_handle_cmd(cmd);
1992}
1993
Christoph Hellwigd6296d392017-05-01 10:19:08 -06001994static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1995 unsigned int hctx_idx, unsigned int numa_node)
Ming Leib5dd2f62014-12-31 13:22:57 +00001996{
1997 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1998
Petr Mladek39891442016-10-11 13:55:20 -07001999 kthread_init_work(&cmd->work, loop_queue_work);
Ming Leib5dd2f62014-12-31 13:22:57 +00002000 return 0;
2001}
2002
Eric Biggersf363b082017-03-30 13:39:16 -07002003static const struct blk_mq_ops loop_mq_ops = {
Ming Leib5dd2f62014-12-31 13:22:57 +00002004 .queue_rq = loop_queue_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00002005 .init_request = loop_init_request,
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02002006 .complete = lo_complete_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00002007};
2008
Kay Sievers34dd82a2011-07-31 22:08:04 +02002009static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07002010{
2011 struct loop_device *lo;
2012 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002013 int err;
Ken Chen73285082007-05-08 00:28:20 -07002014
Silva Paulo68d740d2012-07-14 15:39:58 -07002015 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07002016 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07002017 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07002018 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002019
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06002020 lo->lo_state = Lo_unbound;
2021
Tejun Heoc718aa62013-02-27 17:03:58 -08002022 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002023 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08002024 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2025 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02002026 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002027 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08002028 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002029 }
2030 if (err < 0)
2031 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08002032 i = err;
Ken Chen73285082007-05-08 00:28:20 -07002033
Wei Yongjun183cfb52013-03-22 08:59:19 -06002034 err = -ENOMEM;
Ming Leib5dd2f62014-12-31 13:22:57 +00002035 lo->tag_set.ops = &loop_mq_ops;
2036 lo->tag_set.nr_hw_queues = 1;
2037 lo->tag_set.queue_depth = 128;
2038 lo->tag_set.numa_node = NUMA_NO_NODE;
2039 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
Ming Lei56d18f62019-02-15 19:13:24 +08002040 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
Ming Leib5dd2f62014-12-31 13:22:57 +00002041 lo->tag_set.driver_data = lo;
2042
2043 err = blk_mq_alloc_tag_set(&lo->tag_set);
2044 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002045 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07002046
Ming Leib5dd2f62014-12-31 13:22:57 +00002047 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
Chengguang Xu38a34992018-12-16 17:35:00 +08002048 if (IS_ERR(lo->lo_queue)) {
Ming Leib5dd2f62014-12-31 13:22:57 +00002049 err = PTR_ERR(lo->lo_queue);
2050 goto out_cleanup_tags;
2051 }
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06002052 lo->lo_queue->queuedata = lo;
2053
Shaohua Li54bb0ad2017-08-31 22:09:45 -07002054 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
Shaohua Li40326d82017-08-31 22:09:46 -07002055
Ming Lei5b5e20f2015-08-17 10:31:47 +08002056 /*
Shaohua Li40326d82017-08-31 22:09:46 -07002057 * By default, we do buffer IO, so it doesn't make sense to enable
2058 * merge because the I/O submitted to backing file is handled page by
2059 * page. For directio mode, merge does help to dispatch bigger request
2060 * to underlayer disk. We will enable merge once directio is enabled.
Ming Lei5b5e20f2015-08-17 10:31:47 +08002061 */
Bart Van Assche8b904b52018-03-07 17:10:10 -08002062 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei5b5e20f2015-08-17 10:31:47 +08002063
Minfei Huang7a649732016-06-07 10:05:15 +08002064 err = -ENOMEM;
Laurent Vivier476a4812008-03-26 12:11:53 +01002065 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07002066 if (!disk)
2067 goto out_free_queue;
2068
Kay Sieverse03c8dd2011-08-23 20:12:04 +02002069 /*
2070 * Disable partition scanning by default. The in-kernel partition
2071 * scanning can be requested individually per-device during its
2072 * setup. Userspace can always add and remove partitions from all
2073 * devices. The needed partition minors are allocated from the
2074 * extended minor space, the main loop device numbers will continue
2075 * to match the loop minors, regardless of the number of partitions
2076 * used.
2077 *
2078 * If max_part is given, partition scanning is globally enabled for
2079 * all loop devices. The minors for the main loop devices will be
2080 * multiples of max_part.
2081 *
2082 * Note: Global-for-all-devices, set-only-at-init, read-only module
2083 * parameteters like 'max_loop' and 'max_part' make things needlessly
2084 * complicated, are too static, inflexible and may surprise
2085 * userspace tools. Parameters like this in general should be avoided.
2086 */
2087 if (!part_shift)
2088 disk->flags |= GENHD_FL_NO_PART_SCAN;
2089 disk->flags |= GENHD_FL_EXT_DEVT;
Ming Leif8933662015-05-06 12:26:23 +08002090 atomic_set(&lo->lo_refcnt, 0);
Ken Chen73285082007-05-08 00:28:20 -07002091 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07002092 spin_lock_init(&lo->lo_lock);
2093 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01002094 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07002095 disk->fops = &lo_fops;
2096 disk->private_data = lo;
2097 disk->queue = lo->lo_queue;
2098 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002099 add_disk(disk);
2100 *l = lo;
2101 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07002102
2103out_free_queue:
2104 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00002105out_cleanup_tags:
2106 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002107out_free_idr:
2108 idr_remove(&loop_index_idr, i);
Ken Chen73285082007-05-08 00:28:20 -07002109out_free_dev:
2110 kfree(lo);
2111out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002112 return err;
Ken Chen73285082007-05-08 00:28:20 -07002113}
2114
Kay Sievers34dd82a2011-07-31 22:08:04 +02002115static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07002116{
NeilBrown6cd18e72015-04-27 14:12:22 +10002117 del_gendisk(lo->lo_disk);
Bart Van Assche0fa8ebd2018-02-28 10:15:28 -08002118 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00002119 blk_mq_free_tag_set(&lo->tag_set);
Ken Chen73285082007-05-08 00:28:20 -07002120 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07002121 kfree(lo);
2122}
2123
Kay Sievers770fe302011-07-31 22:08:04 +02002124static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07002125{
Kay Sievers770fe302011-07-31 22:08:04 +02002126 struct loop_device *lo = ptr;
2127 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07002128
Kay Sievers770fe302011-07-31 22:08:04 +02002129 if (lo->lo_state == Lo_unbound) {
2130 *l = lo;
2131 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07002132 }
Kay Sievers770fe302011-07-31 22:08:04 +02002133 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07002134}
2135
Kay Sievers34dd82a2011-07-31 22:08:04 +02002136static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07002137{
Ken Chena47653f2007-06-08 13:46:44 -07002138 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002139 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07002140
Kay Sievers770fe302011-07-31 22:08:04 +02002141 if (i < 0) {
2142 int err;
2143
2144 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2145 if (err == 1) {
2146 *l = lo;
2147 ret = lo->lo_number;
2148 }
2149 goto out;
2150 }
2151
2152 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002153 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07002154 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02002155 *l = lo;
2156 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07002157 }
Kay Sievers770fe302011-07-31 22:08:04 +02002158out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002159 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07002160}
2161
Ken Chen73285082007-05-08 00:28:20 -07002162static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2163{
Al Viro705962c2007-05-13 05:52:32 -04002164 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04002165 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002166 int err;
Ken Chen73285082007-05-08 00:28:20 -07002167
Jan Kara0a42e992018-11-08 14:01:04 +01002168 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002169 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2170 if (err < 0)
2171 err = loop_add(&lo, MINOR(dev) >> part_shift);
2172 if (err < 0)
Mikulas Patockaa207f592013-10-14 12:13:24 -04002173 kobj = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002174 else
Jan Kara3079c222018-02-26 13:01:38 +01002175 kobj = get_disk_and_module(lo->lo_disk);
Jan Kara0a42e992018-11-08 14:01:04 +01002176 mutex_unlock(&loop_ctl_mutex);
Ken Chen73285082007-05-08 00:28:20 -07002177
2178 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04002179 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07002180}
2181
Kay Sievers770fe302011-07-31 22:08:04 +02002182static long loop_control_ioctl(struct file *file, unsigned int cmd,
2183 unsigned long parm)
2184{
2185 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01002186 int ret;
Kay Sievers770fe302011-07-31 22:08:04 +02002187
Jan Kara0a42e992018-11-08 14:01:04 +01002188 ret = mutex_lock_killable(&loop_ctl_mutex);
2189 if (ret)
2190 return ret;
2191
2192 ret = -ENOSYS;
Kay Sievers770fe302011-07-31 22:08:04 +02002193 switch (cmd) {
2194 case LOOP_CTL_ADD:
2195 ret = loop_lookup(&lo, parm);
2196 if (ret >= 0) {
2197 ret = -EEXIST;
2198 break;
2199 }
2200 ret = loop_add(&lo, parm);
2201 break;
2202 case LOOP_CTL_REMOVE:
2203 ret = loop_lookup(&lo, parm);
2204 if (ret < 0)
2205 break;
Kay Sievers770fe302011-07-31 22:08:04 +02002206 if (lo->lo_state != Lo_unbound) {
2207 ret = -EBUSY;
Kay Sievers770fe302011-07-31 22:08:04 +02002208 break;
2209 }
Ming Leif8933662015-05-06 12:26:23 +08002210 if (atomic_read(&lo->lo_refcnt) > 0) {
Kay Sievers770fe302011-07-31 22:08:04 +02002211 ret = -EBUSY;
Kay Sievers770fe302011-07-31 22:08:04 +02002212 break;
2213 }
2214 lo->lo_disk->private_data = NULL;
Kay Sievers770fe302011-07-31 22:08:04 +02002215 idr_remove(&loop_index_idr, lo->lo_number);
2216 loop_remove(lo);
2217 break;
2218 case LOOP_CTL_GET_FREE:
2219 ret = loop_lookup(&lo, -1);
2220 if (ret >= 0)
2221 break;
2222 ret = loop_add(&lo, -1);
2223 }
Jan Kara0a42e992018-11-08 14:01:04 +01002224 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02002225
2226 return ret;
2227}
2228
2229static const struct file_operations loop_ctl_fops = {
2230 .open = nonseekable_open,
2231 .unlocked_ioctl = loop_control_ioctl,
2232 .compat_ioctl = loop_control_ioctl,
2233 .owner = THIS_MODULE,
2234 .llseek = noop_llseek,
2235};
2236
2237static struct miscdevice loop_misc = {
2238 .minor = LOOP_CTRL_MINOR,
2239 .name = "loop-control",
2240 .fops = &loop_ctl_fops,
2241};
2242
2243MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2244MODULE_ALIAS("devname:loop-control");
2245
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246static int __init loop_init(void)
2247{
Ken Chena47653f2007-06-08 13:46:44 -07002248 int i, nr;
2249 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002250 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02002251 int err;
Ken Chena47653f2007-06-08 13:46:44 -07002252
Laurent Vivier476a4812008-03-26 12:11:53 +01002253 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02002254 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01002255 part_shift = fls(max_part);
2256
Namhyung Kimac04fee2011-05-27 07:59:25 +02002257 /*
2258 * Adjust max_part according to part_shift as it is exported
2259 * to user space so that user can decide correct minor number
2260 * if [s]he want to create more devices.
2261 *
2262 * Note that -1 is required because partition 0 is reserved
2263 * for the whole disk.
2264 */
2265 max_part = (1UL << part_shift) - 1;
2266 }
2267
Guo Chaob1a66502013-02-21 15:16:49 -08002268 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2269 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002270 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002271 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02002272
Guo Chaob1a66502013-02-21 15:16:49 -08002273 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2274 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002275 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Kay Sieversd134b002011-07-31 22:08:04 +02002278 /*
2279 * If max_loop is specified, create that many devices upfront.
2280 * This also becomes a hard limit. If max_loop is not specified,
2281 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2282 * init time. Loop devices can be requested on-demand with the
2283 * /dev/loop-control interface, or be instantiated by accessing
2284 * a 'dead' device node.
2285 */
Ken Chen73285082007-05-08 00:28:20 -07002286 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07002287 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002288 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07002289 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02002290 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002291 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 }
Ken Chena47653f2007-06-08 13:46:44 -07002293
Anton Volkova8c1d062017-08-07 15:37:50 +03002294 err = misc_register(&loop_misc);
2295 if (err < 0)
2296 goto err_out;
2297
2298
Guo Chaob1a66502013-02-21 15:16:49 -08002299 if (register_blkdev(LOOP_MAJOR, "loop")) {
2300 err = -EIO;
2301 goto misc_out;
2302 }
Ken Chena47653f2007-06-08 13:46:44 -07002303
Ken Chena47653f2007-06-08 13:46:44 -07002304 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2305 THIS_MODULE, loop_probe, NULL, NULL);
2306
Kay Sieversd134b002011-07-31 22:08:04 +02002307 /* pre-create number of devices given by config or max_loop */
Jan Kara0a42e992018-11-08 14:01:04 +01002308 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002309 for (i = 0; i < nr; i++)
2310 loop_add(&lo, i);
Jan Kara0a42e992018-11-08 14:01:04 +01002311 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002312
Ken Chen73285082007-05-08 00:28:20 -07002313 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08002315
2316misc_out:
2317 misc_deregister(&loop_misc);
Anton Volkova8c1d062017-08-07 15:37:50 +03002318err_out:
Guo Chaob1a66502013-02-21 15:16:49 -08002319 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002320}
Ken Chena47653f2007-06-08 13:46:44 -07002321
Kay Sievers34dd82a2011-07-31 22:08:04 +02002322static int loop_exit_cb(int id, void *ptr, void *data)
2323{
2324 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07002325
Kay Sievers34dd82a2011-07-31 22:08:04 +02002326 loop_remove(lo);
2327 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328}
2329
Ken Chen73285082007-05-08 00:28:20 -07002330static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Ken Chena47653f2007-06-08 13:46:44 -07002332 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Namhyung Kima1c15c52011-05-24 16:48:55 +02002334 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07002335
Kay Sievers34dd82a2011-07-31 22:08:04 +02002336 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002337 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07002338
Ken Chena47653f2007-06-08 13:46:44 -07002339 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07002340 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02002341
2342 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343}
2344
2345module_init(loop_init);
2346module_exit(loop_exit);
2347
2348#ifndef MODULE
2349static int __init max_loop_setup(char *str)
2350{
2351 max_loop = simple_strtol(str, NULL, 0);
2352 return 1;
2353}
2354
2355__setup("max_loop=", max_loop_setup);
2356#endif