blob: cb1191d6e945f27bd1c5582321cd6ef5496727bf [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
Martijn Coenen5795b6f2020-05-13 15:38:37 +0200231/**
Martijn Coenen34489142020-05-13 15:38:45 +0200232 * loop_validate_block_size() - validates the passed in block size
233 * @bsize: size to validate
234 */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100235static int
Martijn Coenen34489142020-05-13 15:38:45 +0200236loop_validate_block_size(unsigned short bsize)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100237{
Martijn Coenen34489142020-05-13 15:38:45 +0200238 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
239 return -EINVAL;
240
241 return 0;
242}
243
244/**
Martijn Coenen5795b6f2020-05-13 15:38:37 +0200245 * loop_set_size() - sets device size and notifies userspace
246 * @lo: struct loop_device to set the size for
247 * @size: new size of the loop device
248 *
249 * Callers must validate that the size passed into this function fits into
250 * a sector_t, eg using loop_validate_size()
251 */
252static void loop_set_size(struct loop_device *lo, loff_t size)
253{
Guo Chao7b0576a2013-02-21 15:16:47 -0800254 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Christoph Hellwig611bee52020-08-23 11:10:41 +0200256 bd_set_nr_sectors(bdev, size);
Martijn Coenen716ad092020-05-13 15:38:38 +0200257
258 set_capacity_revalidate_and_notify(lo->lo_disk, size, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261static inline int
262lo_do_transfer(struct loop_device *lo, int cmd,
263 struct page *rpage, unsigned roffs,
264 struct page *lpage, unsigned loffs,
265 int size, sector_t rblock)
266{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200267 int ret;
268
269 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
270 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return 0;
272
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200273 printk_ratelimited(KERN_ERR
274 "loop: Transfer error at byte offset %llu, length %i.\n",
275 (unsigned long long)rblock << 9, size);
276 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200279static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200281 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 ssize_t bw;
Al Viro283e7e52015-04-03 15:21:59 -0400283
Christoph Hellwigb6207432019-06-26 15:49:28 +0200284 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Al Viro03d95eb2013-03-20 13:04:20 -0400286 file_start_write(file);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300287 bw = vfs_iter_write(file, &i, ppos, 0);
Al Viro03d95eb2013-03-20 13:04:20 -0400288 file_end_write(file);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200289
290 if (likely(bw == bvec->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return 0;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200292
293 printk_ratelimited(KERN_ERR
294 "loop: Write error at byte offset %llu, length %i.\n",
295 (unsigned long long)*ppos, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (bw >= 0)
297 bw = -EIO;
298 return bw;
299}
300
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200301static int lo_write_simple(struct loop_device *lo, struct request *rq,
302 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200304 struct bio_vec bvec;
305 struct req_iterator iter;
306 int ret = 0;
307
308 rq_for_each_segment(bvec, rq, iter) {
309 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
310 if (ret < 0)
311 break;
312 cond_resched();
313 }
314
315 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200318/*
Christoph Hellwig456be142011-10-17 12:57:20 +0200319 * This is the slow, transforming version that needs to double buffer the
320 * data as it cannot do the transformations in place without having direct
321 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 */
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200323static int lo_write_transfer(struct loop_device *lo, struct request *rq,
324 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200326 struct bio_vec bvec, b;
Ming Lei30112012014-12-31 13:22:58 +0000327 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200328 struct page *page;
Kent Overstreet79886132013-11-23 17:19:00 -0800329 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200331 page = alloc_page(GFP_NOIO);
332 if (unlikely(!page))
333 return -ENOMEM;
Christoph Hellwig456be142011-10-17 12:57:20 +0200334
Ming Lei30112012014-12-31 13:22:58 +0000335 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200336 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
337 bvec.bv_offset, bvec.bv_len, pos >> 9);
338 if (unlikely(ret))
339 break;
340
341 b.bv_page = page;
342 b.bv_offset = 0;
343 b.bv_len = bvec.bv_len;
344 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (ret < 0)
346 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200348
349 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200353static int lo_read_simple(struct loop_device *lo, struct request *rq,
354 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
Kent Overstreet79886132013-11-23 17:19:00 -0800356 struct bio_vec bvec;
Ming Lei30112012014-12-31 13:22:58 +0000357 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200358 struct iov_iter i;
359 ssize_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
Ming Lei30112012014-12-31 13:22:58 +0000361 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigb6207432019-06-26 15:49:28 +0200362 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300363 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200364 if (len < 0)
365 return len;
Dave Young306df072012-02-08 22:07:19 +0100366
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200367 flush_dcache_page(bvec.bv_page);
368
369 if (len != bvec.bv_len) {
Ming Lei30112012014-12-31 13:22:58 +0000370 struct bio *bio;
371
372 __rq_for_each_bio(bio, rq)
373 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 break;
Dave Young306df072012-02-08 22:07:19 +0100375 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200376 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200378
Dave Young306df072012-02-08 22:07:19 +0100379 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
381
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200382static int lo_read_transfer(struct loop_device *lo, struct request *rq,
383 loff_t pos)
384{
385 struct bio_vec bvec, b;
386 struct req_iterator iter;
387 struct iov_iter i;
388 struct page *page;
389 ssize_t len;
390 int ret = 0;
391
392 page = alloc_page(GFP_NOIO);
393 if (unlikely(!page))
394 return -ENOMEM;
395
396 rq_for_each_segment(bvec, rq, iter) {
397 loff_t offset = pos;
398
399 b.bv_page = page;
400 b.bv_offset = 0;
401 b.bv_len = bvec.bv_len;
402
Christoph Hellwigb6207432019-06-26 15:49:28 +0200403 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300404 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200405 if (len < 0) {
406 ret = len;
407 goto out_free_page;
408 }
409
410 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
411 bvec.bv_offset, len, offset >> 9);
412 if (ret)
413 goto out_free_page;
414
415 flush_dcache_page(bvec.bv_page);
416
417 if (len != bvec.bv_len) {
418 struct bio *bio;
419
420 __rq_for_each_bio(bio, rq)
421 zero_fill_bio(bio);
422 break;
423 }
424 }
425
426 ret = 0;
427out_free_page:
428 __free_page(page);
429 return ret;
430}
431
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700432static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
433 int mode)
Ming Leicf655d92014-12-31 13:22:59 +0000434{
435 /*
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700436 * We use fallocate to manipulate the space mappings used by the image
437 * a.k.a. discard/zerorange. However we do not support this if
438 * encryption is enabled, because it may give an attacker useful
439 * information.
Ming Leicf655d92014-12-31 13:22:59 +0000440 */
441 struct file *file = lo->lo_backing_file;
Evan Greenc52abf52020-04-03 16:43:04 +0200442 struct request_queue *q = lo->lo_queue;
Ming Leicf655d92014-12-31 13:22:59 +0000443 int ret;
444
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700445 mode |= FALLOC_FL_KEEP_SIZE;
446
Evan Greenc52abf52020-04-03 16:43:04 +0200447 if (!blk_queue_discard(q)) {
Ming Leicf655d92014-12-31 13:22:59 +0000448 ret = -EOPNOTSUPP;
449 goto out;
450 }
451
452 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
453 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
454 ret = -EIO;
455 out:
456 return ret;
457}
458
459static int lo_req_flush(struct loop_device *lo, struct request *rq)
460{
461 struct file *file = lo->lo_backing_file;
462 int ret = vfs_fsync(file, 0);
463 if (unlikely(ret && ret != -EINVAL))
464 ret = -EIO;
465
466 return ret;
467}
468
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200469static void lo_complete_rq(struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800470{
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200471 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Jens Axboef9de14b2018-04-13 16:25:57 -0600472 blk_status_t ret = BLK_STS_OK;
Ming Leibc07c102015-08-17 10:31:51 +0800473
Jens Axboef9de14b2018-04-13 16:25:57 -0600474 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
475 req_op(rq) != REQ_OP_READ) {
476 if (cmd->ret < 0)
Evan Green8cd55082020-04-03 16:43:03 +0200477 ret = errno_to_blk_status(cmd->ret);
Jens Axboef9de14b2018-04-13 16:25:57 -0600478 goto end_io;
Ming Leibc07c102015-08-17 10:31:51 +0800479 }
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200480
Jens Axboef9de14b2018-04-13 16:25:57 -0600481 /*
482 * Short READ - if we got some data, advance our request and
483 * retry it. If we got no data, end the rest with EIO.
484 */
485 if (cmd->ret) {
486 blk_update_request(rq, BLK_STS_OK, cmd->ret);
487 cmd->ret = 0;
488 blk_mq_requeue_request(rq, true);
489 } else {
490 if (cmd->use_aio) {
491 struct bio *bio = rq->bio;
492
493 while (bio) {
494 zero_fill_bio(bio);
495 bio = bio->bi_next;
496 }
497 }
498 ret = BLK_STS_IOERR;
499end_io:
500 blk_mq_end_request(rq, ret);
501 }
Ming Leibc07c102015-08-17 10:31:51 +0800502}
503
Shaohua Li92d77332017-09-01 11:15:17 -0700504static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
505{
Jens Axboe1894e9162018-04-13 16:24:29 -0600506 struct request *rq = blk_mq_rq_from_pdu(cmd);
507
Shaohua Li92d77332017-09-01 11:15:17 -0700508 if (!atomic_dec_and_test(&cmd->ref))
509 return;
510 kfree(cmd->bvec);
511 cmd->bvec = NULL;
Christoph Hellwig15f73f52020-06-11 08:44:47 +0200512 if (likely(!blk_should_fake_timeout(rq->q)))
513 blk_mq_complete_request(rq);
Shaohua Li92d77332017-09-01 11:15:17 -0700514}
515
Ming Leibc07c102015-08-17 10:31:51 +0800516static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
517{
518 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
Ming Leibc07c102015-08-17 10:31:51 +0800519
Shaohua Lid4478e92017-09-25 13:07:22 -0600520 if (cmd->css)
521 css_put(cmd->css);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200522 cmd->ret = ret;
Shaohua Li92d77332017-09-01 11:15:17 -0700523 lo_rw_aio_do_completion(cmd);
Ming Leibc07c102015-08-17 10:31:51 +0800524}
525
526static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
527 loff_t pos, bool rw)
528{
529 struct iov_iter iter;
Ming Lei86af5952019-02-15 19:13:17 +0800530 struct req_iterator rq_iter;
Ming Leibc07c102015-08-17 10:31:51 +0800531 struct bio_vec *bvec;
Jens Axboe1894e9162018-04-13 16:24:29 -0600532 struct request *rq = blk_mq_rq_from_pdu(cmd);
Shaohua Li40326d82017-08-31 22:09:46 -0700533 struct bio *bio = rq->bio;
Ming Leibc07c102015-08-17 10:31:51 +0800534 struct file *file = lo->lo_backing_file;
Ming Lei86af5952019-02-15 19:13:17 +0800535 struct bio_vec tmp;
Shaohua Li40326d82017-08-31 22:09:46 -0700536 unsigned int offset;
Ming Lei86af5952019-02-15 19:13:17 +0800537 int nr_bvec = 0;
Ming Leibc07c102015-08-17 10:31:51 +0800538 int ret;
539
Ming Lei86af5952019-02-15 19:13:17 +0800540 rq_for_each_bvec(tmp, rq, rq_iter)
541 nr_bvec++;
Ming Leibc07c102015-08-17 10:31:51 +0800542
Ming Lei86af5952019-02-15 19:13:17 +0800543 if (rq->bio != rq->biotail) {
544
545 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
Kees Cook6da2ec52018-06-12 13:55:00 -0700546 GFP_NOIO);
Shaohua Li40326d82017-08-31 22:09:46 -0700547 if (!bvec)
548 return -EIO;
549 cmd->bvec = bvec;
550
551 /*
552 * The bios of the request may be started from the middle of
553 * the 'bvec' because of bio splitting, so we can't directly
Ming Lei86af5952019-02-15 19:13:17 +0800554 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
Shaohua Li40326d82017-08-31 22:09:46 -0700555 * API will take care of all details for us.
556 */
Ming Lei86af5952019-02-15 19:13:17 +0800557 rq_for_each_bvec(tmp, rq, rq_iter) {
Shaohua Li40326d82017-08-31 22:09:46 -0700558 *bvec = tmp;
559 bvec++;
560 }
561 bvec = cmd->bvec;
562 offset = 0;
563 } else {
564 /*
565 * Same here, this bio may be started from the middle of the
566 * 'bvec' because of bio splitting, so offset from the bvec
567 * must be passed to iov iterator
568 */
569 offset = bio->bi_iter.bi_bvec_done;
570 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
Shaohua Li40326d82017-08-31 22:09:46 -0700571 }
Shaohua Li92d77332017-09-01 11:15:17 -0700572 atomic_set(&cmd->ref, 2);
Shaohua Li40326d82017-08-31 22:09:46 -0700573
Christoph Hellwigb6207432019-06-26 15:49:28 +0200574 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
Shaohua Li40326d82017-08-31 22:09:46 -0700575 iter.iov_offset = offset;
Ming Leibc07c102015-08-17 10:31:51 +0800576
577 cmd->iocb.ki_pos = pos;
578 cmd->iocb.ki_filp = file;
579 cmd->iocb.ki_complete = lo_rw_aio_complete;
580 cmd->iocb.ki_flags = IOCB_DIRECT;
Adam Manzanaresd9a08a92018-05-22 10:52:19 -0700581 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
Shaohua Lid4478e92017-09-25 13:07:22 -0600582 if (cmd->css)
583 kthread_associate_blkcg(cmd->css);
Ming Leibc07c102015-08-17 10:31:51 +0800584
585 if (rw == WRITE)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100586 ret = call_write_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800587 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100588 ret = call_read_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800589
Shaohua Li92d77332017-09-01 11:15:17 -0700590 lo_rw_aio_do_completion(cmd);
Shaohua Lid4478e92017-09-25 13:07:22 -0600591 kthread_associate_blkcg(NULL);
Shaohua Li92d77332017-09-01 11:15:17 -0700592
Ming Leibc07c102015-08-17 10:31:51 +0800593 if (ret != -EIOCBQUEUED)
594 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
595 return 0;
596}
597
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200598static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800599{
600 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200601 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800602
603 /*
604 * lo_write_simple and lo_read_simple should have been covered
605 * by io submit style function like lo_rw_aio(), one blocker
606 * is that lo_read_simple() need to call flush_dcache_page after
607 * the page is written from kernel, and it isn't easy to handle
608 * this in io submit style function which submits all segments
609 * of the req at one time. And direct read IO doesn't need to
610 * run flush_dcache_page().
611 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200612 switch (req_op(rq)) {
613 case REQ_OP_FLUSH:
614 return lo_req_flush(lo, rq);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200615 case REQ_OP_WRITE_ZEROES:
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700616 /*
617 * If the caller doesn't want deallocation, call zeroout to
618 * write zeroes the range. Otherwise, punch them out.
619 */
620 return lo_fallocate(lo, rq, pos,
621 (rq->cmd_flags & REQ_NOUNMAP) ?
622 FALLOC_FL_ZERO_RANGE :
623 FALLOC_FL_PUNCH_HOLE);
624 case REQ_OP_DISCARD:
625 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200626 case REQ_OP_WRITE:
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200627 if (lo->transfer)
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200628 return lo_write_transfer(lo, rq, pos);
629 else if (cmd->use_aio)
630 return lo_rw_aio(lo, cmd, pos, WRITE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200631 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200632 return lo_write_simple(lo, rq, pos);
633 case REQ_OP_READ:
634 if (lo->transfer)
635 return lo_read_transfer(lo, rq, pos);
636 else if (cmd->use_aio)
637 return lo_rw_aio(lo, cmd, pos, READ);
638 else
639 return lo_read_simple(lo, rq, pos);
640 default:
641 WARN_ON_ONCE(1);
642 return -EIO;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800646static inline void loop_update_dio(struct loop_device *lo)
647{
Ira Weinyefbe3c22020-04-30 07:41:33 -0700648 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
649 lo->use_dio);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800650}
651
Ming Lei06f0e9e2015-05-06 12:26:24 +0800652static void loop_reread_partitions(struct loop_device *lo,
653 struct block_device *bdev)
654{
655 int rc;
656
Christoph Hellwigf0b870d2019-11-14 15:34:36 +0100657 mutex_lock(&bdev->bd_mutex);
658 rc = bdev_disk_changed(bdev, false);
659 mutex_unlock(&bdev->bd_mutex);
Ming Lei06f0e9e2015-05-06 12:26:24 +0800660 if (rc)
661 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
662 __func__, lo->lo_number, lo->lo_file_name, rc);
663}
664
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400665static inline int is_loop_device(struct file *file)
666{
667 struct inode *i = file->f_mapping->host;
668
669 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
670}
671
672static int loop_validate_file(struct file *file, struct block_device *bdev)
673{
674 struct inode *inode = file->f_mapping->host;
675 struct file *f = file;
676
677 /* Avoid recursion */
678 while (is_loop_device(f)) {
679 struct loop_device *l;
680
681 if (f->f_mapping->host->i_bdev == bdev)
682 return -EBADF;
683
684 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
Dongli Zhangf7c8a412019-03-18 20:23:17 +0800685 if (l->lo_state != Lo_bound) {
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400686 return -EINVAL;
687 }
688 f = l->lo_backing_file;
689 }
690 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
691 return -EINVAL;
692 return 0;
693}
694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695/*
696 * loop_change_fd switched the backing store of a loopback device to
697 * a new file. This is useful for operating system installers to free up
698 * the original file and in High Availability environments to switch to
699 * an alternative location for the content in case of server meltdown.
700 * This can only work if the loop device is used read-only, and if the
701 * new backing store is the same size and type as the old backing store.
702 */
Al Virobb214882008-03-02 09:29:48 -0500703static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
704 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Jan Kara1dded9a2018-11-08 14:01:15 +0100706 struct file *file = NULL, *old_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 int error;
Jan Kara85b0a542018-11-08 14:01:13 +0100708 bool partscan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Jan Karac28445f2018-11-08 14:01:16 +0100710 error = mutex_lock_killable(&loop_ctl_mutex);
Jan Karac3710772018-11-08 14:01:11 +0100711 if (error)
712 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 error = -ENXIO;
714 if (lo->lo_state != Lo_bound)
Jan Kara1dded9a2018-11-08 14:01:15 +0100715 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
717 /* the loop device has to be read-only */
718 error = -EINVAL;
719 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
Jan Kara1dded9a2018-11-08 14:01:15 +0100720 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 error = -EBADF;
723 file = fget(arg);
724 if (!file)
Jan Kara1dded9a2018-11-08 14:01:15 +0100725 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400727 error = loop_validate_file(file, bdev);
728 if (error)
Jan Kara1dded9a2018-11-08 14:01:15 +0100729 goto out_err;
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 old_file = lo->lo_backing_file;
732
733 error = -EINVAL;
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 /* size of the new backing store needs to be the same */
736 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
Jan Kara1dded9a2018-11-08 14:01:15 +0100737 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 /* and ... switch */
Omar Sandoval43cade802017-08-24 00:03:44 -0700740 blk_mq_freeze_queue(lo->lo_queue);
741 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
742 lo->lo_backing_file = file;
743 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
744 mapping_set_gfp_mask(file->f_mapping,
745 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
746 loop_update_dio(lo);
747 blk_mq_unfreeze_queue(lo->lo_queue);
Jan Kara85b0a542018-11-08 14:01:13 +0100748 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Jan Karac3710772018-11-08 14:01:11 +0100749 mutex_unlock(&loop_ctl_mutex);
Jan Kara1dded9a2018-11-08 14:01:15 +0100750 /*
751 * We must drop file reference outside of loop_ctl_mutex as dropping
752 * the file ref can take bd_mutex which creates circular locking
753 * dependency.
754 */
755 fput(old_file);
Jan Kara85b0a542018-11-08 14:01:13 +0100756 if (partscan)
757 loop_reread_partitions(lo, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return 0;
759
Jan Kara1dded9a2018-11-08 14:01:15 +0100760out_err:
Jan Karac3710772018-11-08 14:01:11 +0100761 mutex_unlock(&loop_ctl_mutex);
Jan Kara1dded9a2018-11-08 14:01:15 +0100762 if (file)
763 fput(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return error;
765}
766
Milan Brozee862732010-08-23 15:16:00 +0200767/* loop sysfs attributes */
768
769static ssize_t loop_attr_show(struct device *dev, char *page,
770 ssize_t (*callback)(struct loop_device *, char *))
771{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200772 struct gendisk *disk = dev_to_disk(dev);
773 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200774
Kay Sievers34dd82a2011-07-31 22:08:04 +0200775 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200776}
777
778#define LOOP_ATTR_RO(_name) \
779static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
780static ssize_t loop_attr_do_show_##_name(struct device *d, \
781 struct device_attribute *attr, char *b) \
782{ \
783 return loop_attr_show(d, b, loop_attr_##_name##_show); \
784} \
785static struct device_attribute loop_attr_##_name = \
Joe Perches5657a812018-05-24 13:38:59 -0600786 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
Milan Brozee862732010-08-23 15:16:00 +0200787
788static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
789{
790 ssize_t ret;
791 char *p = NULL;
792
Kay Sievers05eb0f22011-07-31 22:21:35 +0200793 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200794 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200795 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200796 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200797
798 if (IS_ERR_OR_NULL(p))
799 ret = PTR_ERR(p);
800 else {
801 ret = strlen(p);
802 memmove(buf, p, ret);
803 buf[ret++] = '\n';
804 buf[ret] = 0;
805 }
806
807 return ret;
808}
809
810static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
811{
812 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
813}
814
815static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
816{
817 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
818}
819
820static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
821{
822 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
823
824 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
825}
826
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200827static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
828{
829 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
830
831 return sprintf(buf, "%s\n", partscan ? "1" : "0");
832}
833
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800834static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
835{
836 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
837
838 return sprintf(buf, "%s\n", dio ? "1" : "0");
839}
840
Milan Brozee862732010-08-23 15:16:00 +0200841LOOP_ATTR_RO(backing_file);
842LOOP_ATTR_RO(offset);
843LOOP_ATTR_RO(sizelimit);
844LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200845LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800846LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200847
848static struct attribute *loop_attrs[] = {
849 &loop_attr_backing_file.attr,
850 &loop_attr_offset.attr,
851 &loop_attr_sizelimit.attr,
852 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200853 &loop_attr_partscan.attr,
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800854 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200855 NULL,
856};
857
858static struct attribute_group loop_attribute_group = {
859 .name = "loop",
860 .attrs= loop_attrs,
861};
862
Tetsuo Handad3349b62018-05-04 10:58:09 -0600863static void loop_sysfs_init(struct loop_device *lo)
Milan Brozee862732010-08-23 15:16:00 +0200864{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600865 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
866 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200867}
868
869static void loop_sysfs_exit(struct loop_device *lo)
870{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600871 if (lo->sysfs_inited)
872 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
873 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200874}
875
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200876static void loop_config_discard(struct loop_device *lo)
877{
878 struct file *file = lo->lo_backing_file;
879 struct inode *inode = file->f_mapping->host;
880 struct request_queue *q = lo->lo_queue;
Ming Leibcb21c82020-08-17 18:01:30 +0800881 u32 granularity, max_discard_sectors;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200882
883 /*
Evan Greenc52abf52020-04-03 16:43:04 +0200884 * If the backing device is a block device, mirror its zeroing
885 * capability. Set the discard sectors to the block device's zeroing
886 * capabilities because loop discards result in blkdev_issue_zeroout(),
887 * not blkdev_issue_discard(). This maintains consistent behavior with
888 * file-backed loop devices: discarded regions read back as zero.
889 */
890 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
891 struct request_queue *backingq;
892
893 backingq = bdev_get_queue(inode->i_bdev);
Evan Greenc52abf52020-04-03 16:43:04 +0200894
Ming Leibcb21c82020-08-17 18:01:30 +0800895 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
896 granularity = backingq->limits.discard_granularity ?:
897 queue_physical_block_size(backingq);
Evan Greenc52abf52020-04-03 16:43:04 +0200898
899 /*
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200900 * We use punch hole to reclaim the free space used by the
Olaf Hering12a64d22014-01-21 14:39:24 -0800901 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200902 * encryption is enabled, because it may give an attacker
903 * useful information.
904 */
Evan Greenc52abf52020-04-03 16:43:04 +0200905 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
Ming Leibcb21c82020-08-17 18:01:30 +0800906 max_discard_sectors = 0;
907 granularity = 0;
Evan Greenc52abf52020-04-03 16:43:04 +0200908
909 } else {
Ming Leibcb21c82020-08-17 18:01:30 +0800910 max_discard_sectors = UINT_MAX >> 9;
911 granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200912 }
913
Ming Leibcb21c82020-08-17 18:01:30 +0800914 if (max_discard_sectors) {
915 q->limits.discard_granularity = granularity;
916 blk_queue_max_discard_sectors(q, max_discard_sectors);
917 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
Evan Greenc52abf52020-04-03 16:43:04 +0200918 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
Ming Leibcb21c82020-08-17 18:01:30 +0800919 } else {
920 q->limits.discard_granularity = 0;
921 blk_queue_max_discard_sectors(q, 0);
922 blk_queue_max_write_zeroes_sectors(q, 0);
Evan Greenc52abf52020-04-03 16:43:04 +0200923 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
Ming Leibcb21c82020-08-17 18:01:30 +0800924 }
925 q->limits.discard_alignment = 0;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200926}
927
Ming Leie03a3d72015-08-17 10:31:48 +0800928static void loop_unprepare_queue(struct loop_device *lo)
929{
Petr Mladek39891442016-10-11 13:55:20 -0700930 kthread_flush_worker(&lo->worker);
Ming Leie03a3d72015-08-17 10:31:48 +0800931 kthread_stop(lo->worker_task);
932}
933
NeilBrownb2ee7d42017-06-16 15:02:09 +1000934static int loop_kthread_worker_fn(void *worker_ptr)
935{
NeilBrowna37b0712020-06-01 21:48:18 -0700936 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
NeilBrownb2ee7d42017-06-16 15:02:09 +1000937 return kthread_worker_fn(worker_ptr);
938}
939
Ming Leie03a3d72015-08-17 10:31:48 +0800940static int loop_prepare_queue(struct loop_device *lo)
941{
Petr Mladek39891442016-10-11 13:55:20 -0700942 kthread_init_worker(&lo->worker);
NeilBrownb2ee7d42017-06-16 15:02:09 +1000943 lo->worker_task = kthread_run(loop_kthread_worker_fn,
Ming Leie03a3d72015-08-17 10:31:48 +0800944 &lo->worker, "loop%d", lo->lo_number);
945 if (IS_ERR(lo->worker_task))
946 return -ENOMEM;
947 set_user_nice(lo->worker_task, MIN_NICE);
948 return 0;
949}
950
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -0700951static void loop_update_rotational(struct loop_device *lo)
952{
953 struct file *file = lo->lo_backing_file;
954 struct inode *file_inode = file->f_mapping->host;
955 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
956 struct request_queue *q = lo->lo_queue;
957 bool nonrot = true;
958
959 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
960 if (file_bdev)
961 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
962
963 if (nonrot)
964 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
965 else
966 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
967}
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969static int
970loop_release_xfer(struct loop_device *lo)
971{
972 int err = 0;
973 struct loop_func_table *xfer = lo->lo_encryption;
974
975 if (xfer) {
976 if (xfer->release)
977 err = xfer->release(lo);
978 lo->transfer = NULL;
979 lo->lo_encryption = NULL;
980 module_put(xfer->owner);
981 }
982 return err;
983}
984
985static int
986loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
987 const struct loop_info64 *i)
988{
989 int err = 0;
990
991 if (xfer) {
992 struct module *owner = xfer->owner;
993
994 if (!try_module_get(owner))
995 return -EINVAL;
996 if (xfer->init)
997 err = xfer->init(lo, i);
998 if (err)
999 module_put(owner);
1000 else
1001 lo->lo_encryption = xfer;
1002 }
1003 return err;
1004}
1005
Martijn Coenen62ab4662020-05-13 15:38:42 +02001006/**
1007 * loop_set_status_from_info - configure device from loop_info
1008 * @lo: struct loop_device to configure
1009 * @info: struct loop_info64 to configure the device with
1010 *
1011 * Configures the loop device parameters according to the passed
1012 * in loop_info64 configuration.
1013 */
1014static int
1015loop_set_status_from_info(struct loop_device *lo,
1016 const struct loop_info64 *info)
1017{
1018 int err;
1019 struct loop_func_table *xfer;
1020 kuid_t uid = current_uid();
1021
1022 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1023 return -EINVAL;
1024
1025 err = loop_release_xfer(lo);
1026 if (err)
1027 return err;
1028
1029 if (info->lo_encrypt_type) {
1030 unsigned int type = info->lo_encrypt_type;
1031
1032 if (type >= MAX_LO_CRYPT)
1033 return -EINVAL;
1034 xfer = xfer_funcs[type];
1035 if (xfer == NULL)
1036 return -EINVAL;
1037 } else
1038 xfer = NULL;
1039
1040 err = loop_init_xfer(lo, xfer, info);
1041 if (err)
1042 return err;
1043
1044 lo->lo_offset = info->lo_offset;
1045 lo->lo_sizelimit = info->lo_sizelimit;
1046 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1047 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1048 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1049 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1050
1051 if (!xfer)
1052 xfer = &none_funcs;
1053 lo->transfer = xfer->transfer;
1054 lo->ioctl = xfer->ioctl;
1055
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001056 lo->lo_flags = info->lo_flags;
Martijn Coenen62ab4662020-05-13 15:38:42 +02001057
1058 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1059 lo->lo_init[0] = info->lo_init[0];
1060 lo->lo_init[1] = info->lo_init[1];
1061 if (info->lo_encrypt_key_size) {
1062 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1063 info->lo_encrypt_key_size);
1064 lo->lo_key_owner = uid;
1065 }
1066
1067 return 0;
1068}
1069
Martijn Coenen34489142020-05-13 15:38:45 +02001070static int loop_configure(struct loop_device *lo, fmode_t mode,
1071 struct block_device *bdev,
1072 const struct loop_config *config)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 struct file *file;
1075 struct inode *inode;
1076 struct address_space *mapping;
1077 struct block_device *claimed_bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 int error;
1079 loff_t size;
1080 bool partscan;
Martijn Coenen34489142020-05-13 15:38:45 +02001081 unsigned short bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
1083 /* This is safe, since we have a reference from open(). */
1084 __module_get(THIS_MODULE);
1085
1086 error = -EBADF;
Martijn Coenen34489142020-05-13 15:38:45 +02001087 file = fget(config->fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if (!file)
1089 goto out;
1090
1091 /*
1092 * If we don't hold exclusive handle for the device, upgrade to it
1093 * here to avoid changing device under exclusive owner.
1094 */
1095 if (!(mode & FMODE_EXCL)) {
Christoph Hellwigecbe6bc2020-07-16 16:33:09 +02001096 claimed_bdev = bdev->bd_contains;
1097 error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
1098 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
1101
1102 error = mutex_lock_killable(&loop_ctl_mutex);
1103 if (error)
1104 goto out_bdev;
1105
1106 error = -EBUSY;
1107 if (lo->lo_state != Lo_unbound)
1108 goto out_unlock;
1109
1110 error = loop_validate_file(file, bdev);
1111 if (error)
1112 goto out_unlock;
1113
1114 mapping = file->f_mapping;
1115 inode = mapping->host;
1116
Martijn Coenen34489142020-05-13 15:38:45 +02001117 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1118 error = -EINVAL;
1119 goto out_unlock;
1120 }
1121
1122 if (config->block_size) {
1123 error = loop_validate_block_size(config->block_size);
1124 if (error)
1125 goto out_unlock;
1126 }
1127
1128 error = loop_set_status_from_info(lo, &config->info);
1129 if (error)
1130 goto out_unlock;
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1133 !file->f_op->write_iter)
Martijn Coenen34489142020-05-13 15:38:45 +02001134 lo->lo_flags |= LO_FLAGS_READ_ONLY;
Martijn Coenen083a6a52020-05-13 15:38:36 +02001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 error = loop_prepare_queue(lo);
1137 if (error)
1138 goto out_unlock;
1139
Martijn Coenen34489142020-05-13 15:38:45 +02001140 set_device_ro(bdev, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Martijn Coenen34489142020-05-13 15:38:45 +02001142 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 lo->lo_device = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 lo->lo_backing_file = file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1146 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1147
Martijn Coenen34489142020-05-13 15:38:45 +02001148 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 blk_queue_write_cache(lo->lo_queue, true, false);
1150
Martijn Coenen34489142020-05-13 15:38:45 +02001151 if (config->block_size)
1152 bsize = config->block_size;
Linus Torvalds96ed3202020-06-02 19:45:12 -07001153 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 /* In case of direct I/O, match underlying block size */
Martijn Coenen34489142020-05-13 15:38:45 +02001155 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1156 else
1157 bsize = 512;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Martijn Coenen34489142020-05-13 15:38:45 +02001159 blk_queue_logical_block_size(lo->lo_queue, bsize);
1160 blk_queue_physical_block_size(lo->lo_queue, bsize);
1161 blk_queue_io_min(lo->lo_queue, bsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
1163 loop_update_rotational(lo);
1164 loop_update_dio(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 loop_sysfs_init(lo);
Martijn Coenen79e5dc52020-08-25 09:18:29 +02001166
1167 size = get_loop_size(lo, file);
Martijn Coenen5795b6f2020-05-13 15:38:37 +02001168 loop_set_size(lo, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1171 block_size(inode->i_bdev) : PAGE_SIZE);
1172
1173 lo->lo_state = Lo_bound;
1174 if (part_shift)
1175 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1176 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Lennart Poetteringfe6a8fc2020-08-10 19:16:32 +02001177 if (partscan)
1178 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /* Grab the block_device to prevent its destruction after we
1181 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1182 */
1183 bdgrab(bdev);
1184 mutex_unlock(&loop_ctl_mutex);
1185 if (partscan)
1186 loop_reread_partitions(lo, bdev);
1187 if (claimed_bdev)
Martijn Coenen34489142020-05-13 15:38:45 +02001188 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return 0;
1190
1191out_unlock:
1192 mutex_unlock(&loop_ctl_mutex);
1193out_bdev:
1194 if (claimed_bdev)
Martijn Coenen34489142020-05-13 15:38:45 +02001195 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196out_putf:
1197 fput(file);
1198out:
1199 /* This is safe: open() is still holding a reference. */
1200 module_put(THIS_MODULE);
1201 return error;
1202}
1203
Jan Kara0da03ca2018-11-08 14:01:14 +01001204static int __loop_clr_fd(struct loop_device *lo, bool release)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Jan Kara7ccd0792018-11-08 14:01:07 +01001206 struct file *filp = NULL;
Al Virob4e3ca12005-10-21 03:22:34 -04001207 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +02001208 struct block_device *bdev = lo->lo_device;
Jan Kara7ccd0792018-11-08 14:01:07 +01001209 int err = 0;
Jan Kara0da03ca2018-11-08 14:01:14 +01001210 bool partscan = false;
1211 int lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Jan Kara7ccd0792018-11-08 14:01:07 +01001213 mutex_lock(&loop_ctl_mutex);
1214 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1215 err = -ENXIO;
1216 goto out_unlock;
1217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Jan Kara7ccd0792018-11-08 14:01:07 +01001219 filp = lo->lo_backing_file;
1220 if (filp == NULL) {
1221 err = -EINVAL;
1222 goto out_unlock;
1223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Ming Leif8933662015-05-06 12:26:23 +08001225 /* freeze request queue during the transition */
1226 blk_mq_freeze_queue(lo->lo_queue);
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001230 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 loop_release_xfer(lo);
1233 lo->transfer = NULL;
1234 lo->ioctl = NULL;
1235 lo->lo_device = NULL;
1236 lo->lo_encryption = NULL;
1237 lo->lo_offset = 0;
1238 lo->lo_sizelimit = 0;
1239 lo->lo_encrypt_key_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1241 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1242 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001243 blk_queue_logical_block_size(lo->lo_queue, 512);
Omar Sandovalbf093752017-09-05 14:24:47 -07001244 blk_queue_physical_block_size(lo->lo_queue, 512);
1245 blk_queue_io_min(lo->lo_queue, 512);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001246 if (bdev) {
1247 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -05001248 invalidate_bdev(bdev);
Jeff Laytoneedffa22018-05-21 14:35:03 -04001249 bdev->bd_inode->i_mapping->wb_err = 0;
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001250 }
Ken Chen73285082007-05-08 00:28:20 -07001251 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001252 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001253 if (bdev) {
Christoph Hellwig611bee52020-08-23 11:10:41 +02001254 bd_set_nr_sectors(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001255 /* let user-space know about this change */
1256 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 mapping_set_gfp_mask(filp->f_mapping, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 /* This is safe: open() is still holding a reference. */
1260 module_put(THIS_MODULE);
Ming Leif8933662015-05-06 12:26:23 +08001261 blk_mq_unfreeze_queue(lo->lo_queue);
1262
Jan Kara0da03ca2018-11-08 14:01:14 +01001263 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1264 lo_number = lo->lo_number;
Jan Kara0da03ca2018-11-08 14:01:14 +01001265 loop_unprepare_queue(lo);
1266out_unlock:
1267 mutex_unlock(&loop_ctl_mutex);
1268 if (partscan) {
Jan Karad57f3372018-11-08 14:01:12 +01001269 /*
1270 * bd_mutex has been held already in release path, so don't
1271 * acquire it if this function is called in such case.
1272 *
1273 * If the reread partition isn't from release path, lo_refcnt
1274 * must be at least one and it can only become zero when the
1275 * current holder is released.
1276 */
Christoph Hellwigf0b870d2019-11-14 15:34:36 +01001277 if (!release)
1278 mutex_lock(&bdev->bd_mutex);
1279 err = bdev_disk_changed(bdev, false);
1280 if (!release)
1281 mutex_unlock(&bdev->bd_mutex);
Dongli Zhang40853d62019-02-22 22:10:19 +08001282 if (err)
1283 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1284 __func__, lo_number, err);
Jan Karad57f3372018-11-08 14:01:12 +01001285 /* Device is gone, no point in returning error */
1286 err = 0;
1287 }
Dongli Zhang758a58d2019-02-22 22:10:20 +08001288
1289 /*
1290 * lo->lo_state is set to Lo_unbound here after above partscan has
1291 * finished.
1292 *
1293 * There cannot be anybody else entering __loop_clr_fd() as
1294 * lo->lo_backing_file is already cleared and Lo_rundown state
1295 * protects us from all the other places trying to change the 'lo'
1296 * device.
1297 */
1298 mutex_lock(&loop_ctl_mutex);
1299 lo->lo_flags = 0;
1300 if (!part_shift)
1301 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1302 lo->lo_state = Lo_unbound;
1303 mutex_unlock(&loop_ctl_mutex);
1304
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001305 /*
Tetsuo Handa310ca162018-11-08 14:01:02 +01001306 * Need not hold loop_ctl_mutex to fput backing file.
1307 * Calling fput holding loop_ctl_mutex triggers a circular
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001308 * lock dependency possibility warning as fput can take
Tetsuo Handa310ca162018-11-08 14:01:02 +01001309 * bd_mutex which is usually taken before loop_ctl_mutex.
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001310 */
Jan Kara7ccd0792018-11-08 14:01:07 +01001311 if (filp)
1312 fput(filp);
1313 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
Jan Karaa2505b72018-11-08 14:01:06 +01001316static int loop_clr_fd(struct loop_device *lo)
1317{
Jan Kara7ccd0792018-11-08 14:01:07 +01001318 int err;
1319
Jan Karac28445f2018-11-08 14:01:16 +01001320 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara7ccd0792018-11-08 14:01:07 +01001321 if (err)
1322 return err;
1323 if (lo->lo_state != Lo_bound) {
1324 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001325 return -ENXIO;
Jan Kara7ccd0792018-11-08 14:01:07 +01001326 }
Jan Karaa2505b72018-11-08 14:01:06 +01001327 /*
1328 * If we've explicitly asked to tear down the loop device,
1329 * and it has an elevated reference count, set it for auto-teardown when
1330 * the last reference goes away. This stops $!~#$@ udev from
1331 * preventing teardown because it decided that it needs to run blkid on
1332 * the loopback device whenever they appear. xfstests is notorious for
1333 * failing tests because blkid via udev races with a losetup
1334 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1335 * command to fail with EBUSY.
1336 */
1337 if (atomic_read(&lo->lo_refcnt) > 1) {
1338 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1339 mutex_unlock(&loop_ctl_mutex);
1340 return 0;
1341 }
1342 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001343 mutex_unlock(&loop_ctl_mutex);
Jan Karaa2505b72018-11-08 14:01:06 +01001344
Jan Kara0da03ca2018-11-08 14:01:14 +01001345 return __loop_clr_fd(lo, false);
Jan Karaa2505b72018-11-08 14:01:06 +01001346}
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348static int
1349loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1350{
1351 int err;
Jan Kara85b0a542018-11-08 14:01:13 +01001352 struct block_device *bdev;
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001353 kuid_t uid = current_uid();
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001354 int prev_lo_flags;
Jan Kara85b0a542018-11-08 14:01:13 +01001355 bool partscan = false;
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001356 bool size_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Jan Karac28445f2018-11-08 14:01:16 +01001358 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara550df5fd2018-11-08 14:01:09 +01001359 if (err)
1360 return err;
David Howellsb0fafa82008-11-14 10:38:41 +11001361 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001362 !uid_eq(lo->lo_key_owner, uid) &&
Jan Kara550df5fd2018-11-08 14:01:09 +01001363 !capable(CAP_SYS_ADMIN)) {
1364 err = -EPERM;
1365 goto out_unlock;
1366 }
1367 if (lo->lo_state != Lo_bound) {
1368 err = -ENXIO;
1369 goto out_unlock;
1370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001372 if (lo->lo_offset != info->lo_offset ||
1373 lo->lo_sizelimit != info->lo_sizelimit) {
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001374 size_changed = true;
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001375 sync_blockdev(lo->lo_device);
Zheng Binf4bd34b2020-06-18 12:21:37 +08001376 invalidate_bdev(lo->lo_device);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001377 }
1378
Ming Leiecdd0952017-02-11 11:40:45 +08001379 /* I/O need to be drained during transfer transition */
1380 blk_mq_freeze_queue(lo->lo_queue);
1381
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001382 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
Zheng Binf4bd34b2020-06-18 12:21:37 +08001383 /* If any pages were dirtied after invalidate_bdev(), try again */
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001384 err = -EAGAIN;
1385 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1386 __func__, lo->lo_number, lo->lo_file_name,
1387 lo->lo_device->bd_inode->i_mapping->nrpages);
1388 goto out_unfreeze;
1389 }
1390
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001391 prev_lo_flags = lo->lo_flags;
1392
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001393 err = loop_set_status_from_info(lo, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001395 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001397 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
Martijn Coenen6ac92fb2020-06-04 22:25:20 +02001398 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001399 /* For those flags, use the previous values instead */
1400 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1401 /* For flags that can't be cleared, use previous values too */
1402 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Martijn Coenenb0bd1582020-05-13 15:38:39 +02001404 if (size_changed) {
1405 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1406 lo->lo_backing_file);
1407 loop_set_size(lo, new_size);
Arnd Bergmannb040ad92017-06-09 12:19:18 +02001408 }
Guo Chao541c7422013-02-21 15:16:46 -08001409
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001410 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001412 /* update dio if lo_offset or transfer is changed */
1413 __loop_update_dio(lo, lo->use_dio);
1414
Jan Kara550df5fd2018-11-08 14:01:09 +01001415out_unfreeze:
Ming Leiecdd0952017-02-11 11:40:45 +08001416 blk_mq_unfreeze_queue(lo->lo_queue);
Omar Sandovale02898b42017-03-01 10:42:38 -08001417
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001418 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1419 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
Omar Sandovale02898b42017-03-01 10:42:38 -08001420 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
Jan Kara85b0a542018-11-08 14:01:13 +01001421 bdev = lo->lo_device;
1422 partscan = true;
Omar Sandovale02898b42017-03-01 10:42:38 -08001423 }
Jan Kara550df5fd2018-11-08 14:01:09 +01001424out_unlock:
1425 mutex_unlock(&loop_ctl_mutex);
Jan Kara85b0a542018-11-08 14:01:13 +01001426 if (partscan)
1427 loop_reread_partitions(lo, bdev);
Omar Sandovale02898b42017-03-01 10:42:38 -08001428
Ming Leiecdd0952017-02-11 11:40:45 +08001429 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430}
1431
1432static int
1433loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1434{
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001435 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 struct kstat stat;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001437 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Jan Karac28445f2018-11-08 14:01:16 +01001439 ret = mutex_lock_killable(&loop_ctl_mutex);
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001440 if (ret)
1441 return ret;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001442 if (lo->lo_state != Lo_bound) {
Tetsuo Handa310ca162018-11-08 14:01:02 +01001443 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return -ENXIO;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001445 }
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 memset(info, 0, sizeof(*info));
1448 info->lo_number = lo->lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 info->lo_offset = lo->lo_offset;
1450 info->lo_sizelimit = lo->lo_sizelimit;
1451 info->lo_flags = lo->lo_flags;
1452 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1453 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1454 info->lo_encrypt_type =
1455 lo->lo_encryption ? lo->lo_encryption->number : 0;
1456 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1457 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1458 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1459 lo->lo_encrypt_key_size);
1460 }
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001461
Tetsuo Handa310ca162018-11-08 14:01:02 +01001462 /* Drop loop_ctl_mutex while we call into the filesystem. */
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001463 path = lo->lo_backing_file->f_path;
1464 path_get(&path);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001465 mutex_unlock(&loop_ctl_mutex);
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001466 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001467 if (!ret) {
1468 info->lo_device = huge_encode_dev(stat.dev);
1469 info->lo_inode = stat.ino;
1470 info->lo_rdevice = huge_encode_dev(stat.rdev);
1471 }
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001472 path_put(&path);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001473 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
1476static void
1477loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1478{
1479 memset(info64, 0, sizeof(*info64));
1480 info64->lo_number = info->lo_number;
1481 info64->lo_device = info->lo_device;
1482 info64->lo_inode = info->lo_inode;
1483 info64->lo_rdevice = info->lo_rdevice;
1484 info64->lo_offset = info->lo_offset;
1485 info64->lo_sizelimit = 0;
1486 info64->lo_encrypt_type = info->lo_encrypt_type;
1487 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1488 info64->lo_flags = info->lo_flags;
1489 info64->lo_init[0] = info->lo_init[0];
1490 info64->lo_init[1] = info->lo_init[1];
1491 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1492 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1493 else
1494 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1495 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1496}
1497
1498static int
1499loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1500{
1501 memset(info, 0, sizeof(*info));
1502 info->lo_number = info64->lo_number;
1503 info->lo_device = info64->lo_device;
1504 info->lo_inode = info64->lo_inode;
1505 info->lo_rdevice = info64->lo_rdevice;
1506 info->lo_offset = info64->lo_offset;
1507 info->lo_encrypt_type = info64->lo_encrypt_type;
1508 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1509 info->lo_flags = info64->lo_flags;
1510 info->lo_init[0] = info64->lo_init[0];
1511 info->lo_init[1] = info64->lo_init[1];
1512 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1513 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1514 else
1515 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1516 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1517
1518 /* error in case values were truncated */
1519 if (info->lo_device != info64->lo_device ||
1520 info->lo_rdevice != info64->lo_rdevice ||
1521 info->lo_inode != info64->lo_inode ||
1522 info->lo_offset != info64->lo_offset)
1523 return -EOVERFLOW;
1524
1525 return 0;
1526}
1527
1528static int
1529loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1530{
1531 struct loop_info info;
1532 struct loop_info64 info64;
1533
1534 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1535 return -EFAULT;
1536 loop_info64_from_old(&info, &info64);
1537 return loop_set_status(lo, &info64);
1538}
1539
1540static int
1541loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1542{
1543 struct loop_info64 info64;
1544
1545 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1546 return -EFAULT;
1547 return loop_set_status(lo, &info64);
1548}
1549
1550static int
1551loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1552 struct loop_info info;
1553 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001554 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001556 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001557 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001558 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 if (!err)
1560 err = loop_info64_to_old(&info64, &info);
1561 if (!err && copy_to_user(arg, &info, sizeof(info)))
1562 err = -EFAULT;
1563
1564 return err;
1565}
1566
1567static int
1568loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1569 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001570 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001572 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001573 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001574 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1576 err = -EFAULT;
1577
1578 return err;
1579}
1580
Hannes Reinecke51001b72017-06-08 13:46:44 +02001581static int loop_set_capacity(struct loop_device *lo)
J. R. Okajima53d66602009-03-31 15:23:43 -07001582{
Martijn Coenen0a6ed1b2020-05-13 15:38:40 +02001583 loff_t size;
1584
J. R. Okajima53d66602009-03-31 15:23:43 -07001585 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001586 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001587
Martijn Coenen0a6ed1b2020-05-13 15:38:40 +02001588 size = get_loop_size(lo, lo->lo_backing_file);
1589 loop_set_size(lo, size);
Martijn Coenen083a6a52020-05-13 15:38:36 +02001590
1591 return 0;
J. R. Okajima53d66602009-03-31 15:23:43 -07001592}
1593
Ming Leiab1cb272015-08-17 10:31:50 +08001594static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1595{
1596 int error = -ENXIO;
1597 if (lo->lo_state != Lo_bound)
1598 goto out;
1599
1600 __loop_update_dio(lo, !!arg);
1601 if (lo->use_dio == !!arg)
1602 return 0;
1603 error = -EINVAL;
1604 out:
1605 return error;
1606}
1607
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001608static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1609{
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001610 int err = 0;
1611
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001612 if (lo->lo_state != Lo_bound)
1613 return -ENXIO;
1614
Martijn Coenen34489142020-05-13 15:38:45 +02001615 err = loop_validate_block_size(arg);
1616 if (err)
1617 return err;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001618
Martijn Coenen7e81f992020-03-10 14:12:30 +01001619 if (lo->lo_queue->limits.logical_block_size == arg)
1620 return 0;
1621
1622 sync_blockdev(lo->lo_device);
Zheng Binf4bd34b2020-06-18 12:21:37 +08001623 invalidate_bdev(lo->lo_device);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001624
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001625 blk_mq_freeze_queue(lo->lo_queue);
1626
Zheng Binf4bd34b2020-06-18 12:21:37 +08001627 /* invalidate_bdev should have truncated all the pages */
Martijn Coenen7e81f992020-03-10 14:12:30 +01001628 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001629 err = -EAGAIN;
1630 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1631 __func__, lo->lo_number, lo->lo_file_name,
1632 lo->lo_device->bd_inode->i_mapping->nrpages);
1633 goto out_unfreeze;
1634 }
1635
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001636 blk_queue_logical_block_size(lo->lo_queue, arg);
Omar Sandovalbf093752017-09-05 14:24:47 -07001637 blk_queue_physical_block_size(lo->lo_queue, arg);
1638 blk_queue_io_min(lo->lo_queue, arg);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001639 loop_update_dio(lo);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001640out_unfreeze:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001641 blk_mq_unfreeze_queue(lo->lo_queue);
1642
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001643 return err;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001644}
1645
Jan Karaa1316542018-11-08 14:01:05 +01001646static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1647 unsigned long arg)
1648{
1649 int err;
1650
Jan Karac28445f2018-11-08 14:01:16 +01001651 err = mutex_lock_killable(&loop_ctl_mutex);
Jan Karaa1316542018-11-08 14:01:05 +01001652 if (err)
1653 return err;
1654 switch (cmd) {
1655 case LOOP_SET_CAPACITY:
1656 err = loop_set_capacity(lo);
1657 break;
1658 case LOOP_SET_DIRECT_IO:
1659 err = loop_set_dio(lo, arg);
1660 break;
1661 case LOOP_SET_BLOCK_SIZE:
1662 err = loop_set_block_size(lo, arg);
1663 break;
1664 default:
1665 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1666 }
1667 mutex_unlock(&loop_ctl_mutex);
1668 return err;
1669}
1670
Al Virobb214882008-03-02 09:29:48 -05001671static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 unsigned int cmd, unsigned long arg)
1673{
Al Virobb214882008-03-02 09:29:48 -05001674 struct loop_device *lo = bdev->bd_disk->private_data;
Martijn Coenen571fae62020-05-13 15:38:43 +02001675 void __user *argp = (void __user *) arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 int err;
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 switch (cmd) {
Martijn Coenen34489142020-05-13 15:38:45 +02001679 case LOOP_SET_FD: {
1680 /*
1681 * Legacy case - pass in a zeroed out struct loop_config with
1682 * only the file descriptor set , which corresponds with the
1683 * default parameters we'd have used otherwise.
1684 */
1685 struct loop_config config;
1686
1687 memset(&config, 0, sizeof(config));
1688 config.fd = arg;
1689
1690 return loop_configure(lo, mode, bdev, &config);
1691 }
1692 case LOOP_CONFIGURE: {
1693 struct loop_config config;
1694
1695 if (copy_from_user(&config, argp, sizeof(config)))
1696 return -EFAULT;
1697
1698 return loop_configure(lo, mode, bdev, &config);
1699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 case LOOP_CHANGE_FD:
Jan Karac3710772018-11-08 14:01:11 +01001701 return loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 case LOOP_CLR_FD:
Jan Kara7ccd0792018-11-08 14:01:07 +01001703 return loop_clr_fd(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001705 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001706 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Martijn Coenen571fae62020-05-13 15:38:43 +02001707 err = loop_set_status_old(lo, argp);
Jan Karaa1316542018-11-08 14:01:05 +01001708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 break;
1710 case LOOP_GET_STATUS:
Martijn Coenen571fae62020-05-13 15:38:43 +02001711 return loop_get_status_old(lo, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001713 err = -EPERM;
Jan Karaa1316542018-11-08 14:01:05 +01001714 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
Martijn Coenen571fae62020-05-13 15:38:43 +02001715 err = loop_set_status64(lo, argp);
Jan Karaa1316542018-11-08 14:01:05 +01001716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 break;
1718 case LOOP_GET_STATUS64:
Martijn Coenen571fae62020-05-13 15:38:43 +02001719 return loop_get_status64(lo, argp);
J. R. Okajima53d66602009-03-31 15:23:43 -07001720 case LOOP_SET_CAPACITY:
Ming Leiab1cb272015-08-17 10:31:50 +08001721 case LOOP_SET_DIRECT_IO:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001722 case LOOP_SET_BLOCK_SIZE:
Jan Karaa1316542018-11-08 14:01:05 +01001723 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1724 return -EPERM;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001725 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 default:
Jan Karaa1316542018-11-08 14:01:05 +01001727 err = lo_simple_ioctl(lo, cmd, arg);
1728 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 return err;
1732}
1733
David Howells863d5b822006-08-29 19:06:14 +01001734#ifdef CONFIG_COMPAT
1735struct compat_loop_info {
1736 compat_int_t lo_number; /* ioctl r/o */
1737 compat_dev_t lo_device; /* ioctl r/o */
1738 compat_ulong_t lo_inode; /* ioctl r/o */
1739 compat_dev_t lo_rdevice; /* ioctl r/o */
1740 compat_int_t lo_offset;
1741 compat_int_t lo_encrypt_type;
1742 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1743 compat_int_t lo_flags; /* ioctl r/o */
1744 char lo_name[LO_NAME_SIZE];
1745 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1746 compat_ulong_t lo_init[2];
1747 char reserved[4];
1748};
1749
1750/*
1751 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1752 * - noinlined to reduce stack space usage in main part of driver
1753 */
1754static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001755loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001756 struct loop_info64 *info64)
1757{
1758 struct compat_loop_info info;
1759
1760 if (copy_from_user(&info, arg, sizeof(info)))
1761 return -EFAULT;
1762
1763 memset(info64, 0, sizeof(*info64));
1764 info64->lo_number = info.lo_number;
1765 info64->lo_device = info.lo_device;
1766 info64->lo_inode = info.lo_inode;
1767 info64->lo_rdevice = info.lo_rdevice;
1768 info64->lo_offset = info.lo_offset;
1769 info64->lo_sizelimit = 0;
1770 info64->lo_encrypt_type = info.lo_encrypt_type;
1771 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1772 info64->lo_flags = info.lo_flags;
1773 info64->lo_init[0] = info.lo_init[0];
1774 info64->lo_init[1] = info.lo_init[1];
1775 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1776 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1777 else
1778 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1779 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1780 return 0;
1781}
1782
1783/*
1784 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1785 * - noinlined to reduce stack space usage in main part of driver
1786 */
1787static noinline int
1788loop_info64_to_compat(const struct loop_info64 *info64,
1789 struct compat_loop_info __user *arg)
1790{
1791 struct compat_loop_info info;
1792
1793 memset(&info, 0, sizeof(info));
1794 info.lo_number = info64->lo_number;
1795 info.lo_device = info64->lo_device;
1796 info.lo_inode = info64->lo_inode;
1797 info.lo_rdevice = info64->lo_rdevice;
1798 info.lo_offset = info64->lo_offset;
1799 info.lo_encrypt_type = info64->lo_encrypt_type;
1800 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1801 info.lo_flags = info64->lo_flags;
1802 info.lo_init[0] = info64->lo_init[0];
1803 info.lo_init[1] = info64->lo_init[1];
1804 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1805 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1806 else
1807 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1808 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1809
1810 /* error in case values were truncated */
1811 if (info.lo_device != info64->lo_device ||
1812 info.lo_rdevice != info64->lo_rdevice ||
1813 info.lo_inode != info64->lo_inode ||
1814 info.lo_offset != info64->lo_offset ||
1815 info.lo_init[0] != info64->lo_init[0] ||
1816 info.lo_init[1] != info64->lo_init[1])
1817 return -EOVERFLOW;
1818
1819 if (copy_to_user(arg, &info, sizeof(info)))
1820 return -EFAULT;
1821 return 0;
1822}
1823
1824static int
1825loop_set_status_compat(struct loop_device *lo,
1826 const struct compat_loop_info __user *arg)
1827{
1828 struct loop_info64 info64;
1829 int ret;
1830
1831 ret = loop_info64_from_compat(arg, &info64);
1832 if (ret < 0)
1833 return ret;
1834 return loop_set_status(lo, &info64);
1835}
1836
1837static int
1838loop_get_status_compat(struct loop_device *lo,
1839 struct compat_loop_info __user *arg)
1840{
1841 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001842 int err;
David Howells863d5b822006-08-29 19:06:14 +01001843
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001844 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001845 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001846 err = loop_get_status(lo, &info64);
David Howells863d5b822006-08-29 19:06:14 +01001847 if (!err)
1848 err = loop_info64_to_compat(&info64, arg);
1849 return err;
1850}
1851
Al Virobb214882008-03-02 09:29:48 -05001852static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1853 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001854{
Al Virobb214882008-03-02 09:29:48 -05001855 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001856 int err;
1857
David Howells863d5b822006-08-29 19:06:14 +01001858 switch(cmd) {
1859 case LOOP_SET_STATUS:
Jan Kara550df5fd2018-11-08 14:01:09 +01001860 err = loop_set_status_compat(lo,
1861 (const struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001862 break;
1863 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001864 err = loop_get_status_compat(lo,
1865 (struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001866 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001867 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001868 case LOOP_CLR_FD:
1869 case LOOP_GET_STATUS64:
1870 case LOOP_SET_STATUS64:
Martijn Coenen34489142020-05-13 15:38:45 +02001871 case LOOP_CONFIGURE:
David Howells863d5b822006-08-29 19:06:14 +01001872 arg = (unsigned long) compat_ptr(arg);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001873 fallthrough;
David Howells863d5b822006-08-29 19:06:14 +01001874 case LOOP_SET_FD:
1875 case LOOP_CHANGE_FD:
Evan Green9fea4b32018-07-02 16:03:46 -07001876 case LOOP_SET_BLOCK_SIZE:
Alessio Balsinifdbe4ee2019-08-07 01:48:28 +01001877 case LOOP_SET_DIRECT_IO:
Al Virobb214882008-03-02 09:29:48 -05001878 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001879 break;
1880 default:
1881 err = -ENOIOCTLCMD;
1882 break;
1883 }
David Howells863d5b822006-08-29 19:06:14 +01001884 return err;
1885}
1886#endif
1887
Al Virobb214882008-03-02 09:29:48 -05001888static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Kay Sievers770fe302011-07-31 22:08:04 +02001890 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01001891 int err;
Kay Sievers770fe302011-07-31 22:08:04 +02001892
Jan Kara0a42e992018-11-08 14:01:04 +01001893 err = mutex_lock_killable(&loop_ctl_mutex);
1894 if (err)
1895 return err;
Kay Sievers770fe302011-07-31 22:08:04 +02001896 lo = bdev->bd_disk->private_data;
1897 if (!lo) {
1898 err = -ENXIO;
1899 goto out;
1900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Ming Leif8933662015-05-06 12:26:23 +08001902 atomic_inc(&lo->lo_refcnt);
Kay Sievers770fe302011-07-31 22:08:04 +02001903out:
Jan Kara0a42e992018-11-08 14:01:04 +01001904 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001905 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906}
1907
Jan Kara967d1dc2018-11-08 14:01:03 +01001908static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909{
Jan Kara967d1dc2018-11-08 14:01:03 +01001910 struct loop_device *lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Jan Kara0a42e992018-11-08 14:01:04 +01001912 mutex_lock(&loop_ctl_mutex);
Jan Kara967d1dc2018-11-08 14:01:03 +01001913 lo = disk->private_data;
Ming Leif8933662015-05-06 12:26:23 +08001914 if (atomic_dec_return(&lo->lo_refcnt))
Jan Kara0a42e992018-11-08 14:01:04 +01001915 goto out_unlock;
Ming Leif8933662015-05-06 12:26:23 +08001916
Milan Broz14f27932008-12-12 14:48:27 +01001917 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
Jan Karaa2505b72018-11-08 14:01:06 +01001918 if (lo->lo_state != Lo_bound)
1919 goto out_unlock;
1920 lo->lo_state = Lo_rundown;
Jan Kara7ccd0792018-11-08 14:01:07 +01001921 mutex_unlock(&loop_ctl_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001922 /*
1923 * In autoclear mode, stop the loop thread
1924 * and remove configuration after last close.
1925 */
Jan Kara0da03ca2018-11-08 14:01:14 +01001926 __loop_clr_fd(lo, true);
Jan Kara7ccd0792018-11-08 14:01:07 +01001927 return;
Omar Sandoval43cade802017-08-24 00:03:44 -07001928 } else if (lo->lo_state == Lo_bound) {
Milan Broz14f27932008-12-12 14:48:27 +01001929 /*
1930 * Otherwise keep thread (if running) and config,
1931 * but flush possible ongoing bios in thread.
1932 */
Omar Sandoval43cade802017-08-24 00:03:44 -07001933 blk_mq_freeze_queue(lo->lo_queue);
1934 blk_mq_unfreeze_queue(lo->lo_queue);
Milan Broz14f27932008-12-12 14:48:27 +01001935 }
David Woodhouse96c58652008-02-06 01:36:27 -08001936
Jan Kara0a42e992018-11-08 14:01:04 +01001937out_unlock:
Tetsuo Handa310ca162018-11-08 14:01:02 +01001938 mutex_unlock(&loop_ctl_mutex);
Linus Torvaldsae665012018-01-05 16:26:00 -08001939}
1940
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001941static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001943 .open = lo_open,
1944 .release = lo_release,
1945 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001946#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001947 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001948#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949};
1950
1951/*
1952 * And now the modules code and kernel interface.
1953 */
Ken Chen73285082007-05-08 00:28:20 -07001954static int max_loop;
Joe Perches5657a812018-05-24 13:38:59 -06001955module_param(max_loop, int, 0444);
Ken Chena47653f2007-06-08 13:46:44 -07001956MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Joe Perches5657a812018-05-24 13:38:59 -06001957module_param(max_part, int, 0444);
Laurent Vivier476a4812008-03-26 12:11:53 +01001958MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959MODULE_LICENSE("GPL");
1960MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1961
1962int loop_register_transfer(struct loop_func_table *funcs)
1963{
1964 unsigned int n = funcs->number;
1965
1966 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1967 return -EINVAL;
1968 xfer_funcs[n] = funcs;
1969 return 0;
1970}
1971
Kay Sievers34dd82a2011-07-31 22:08:04 +02001972static int unregister_transfer_cb(int id, void *ptr, void *data)
1973{
1974 struct loop_device *lo = ptr;
1975 struct loop_func_table *xfer = data;
1976
Tetsuo Handa310ca162018-11-08 14:01:02 +01001977 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001978 if (lo->lo_encryption == xfer)
1979 loop_release_xfer(lo);
Tetsuo Handa310ca162018-11-08 14:01:02 +01001980 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001981 return 0;
1982}
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984int loop_unregister_transfer(int number)
1985{
1986 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 struct loop_func_table *xfer;
1988
1989 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1990 return -EINVAL;
1991
1992 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001993 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return 0;
1995}
1996
1997EXPORT_SYMBOL(loop_register_transfer);
1998EXPORT_SYMBOL(loop_unregister_transfer);
1999
Christoph Hellwigfc17b652017-06-03 09:38:05 +02002000static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
Ming Leib5dd2f62014-12-31 13:22:57 +00002001 const struct blk_mq_queue_data *bd)
2002{
Jens Axboe1894e9162018-04-13 16:24:29 -06002003 struct request *rq = bd->rq;
2004 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2005 struct loop_device *lo = rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00002006
Jens Axboe1894e9162018-04-13 16:24:29 -06002007 blk_mq_start_request(rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00002008
Ming Leif4aa4c72015-05-05 19:49:54 +08002009 if (lo->lo_state != Lo_bound)
Christoph Hellwigfc17b652017-06-03 09:38:05 +02002010 return BLK_STS_IOERR;
Ming Leif4aa4c72015-05-05 19:49:54 +08002011
Jens Axboe1894e9162018-04-13 16:24:29 -06002012 switch (req_op(rq)) {
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02002013 case REQ_OP_FLUSH:
2014 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +02002015 case REQ_OP_WRITE_ZEROES:
Ming Leibc07c102015-08-17 10:31:51 +08002016 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02002017 break;
2018 default:
2019 cmd->use_aio = lo->use_dio;
2020 break;
2021 }
Ming Leibc07c102015-08-17 10:31:51 +08002022
Shaohua Lid4478e92017-09-25 13:07:22 -06002023 /* always use the first bio's css */
Shaohua Li0b508bc2017-09-26 11:02:12 -07002024#ifdef CONFIG_BLK_CGROUP
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002025 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2026 cmd->css = &bio_blkcg(rq->bio)->css;
Shaohua Lid4478e92017-09-25 13:07:22 -06002027 css_get(cmd->css);
2028 } else
2029#endif
2030 cmd->css = NULL;
Petr Mladek39891442016-10-11 13:55:20 -07002031 kthread_queue_work(&lo->worker, &cmd->work);
Ming Leib5dd2f62014-12-31 13:22:57 +00002032
Christoph Hellwigfc17b652017-06-03 09:38:05 +02002033 return BLK_STS_OK;
Ming Leib5dd2f62014-12-31 13:22:57 +00002034}
2035
2036static void loop_handle_cmd(struct loop_cmd *cmd)
2037{
Jens Axboe1894e9162018-04-13 16:24:29 -06002038 struct request *rq = blk_mq_rq_from_pdu(cmd);
2039 const bool write = op_is_write(req_op(rq));
2040 struct loop_device *lo = rq->q->queuedata;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02002041 int ret = 0;
Ming Leib5dd2f62014-12-31 13:22:57 +00002042
Christoph Hellwigf4829a92015-09-27 21:01:50 +02002043 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2044 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00002045 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02002046 }
Ming Leib5dd2f62014-12-31 13:22:57 +00002047
Jens Axboe1894e9162018-04-13 16:24:29 -06002048 ret = do_req_filebacked(lo, rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00002049 failed:
Ming Leibc07c102015-08-17 10:31:51 +08002050 /* complete non-aio request */
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02002051 if (!cmd->use_aio || ret) {
Evan Green8cd55082020-04-03 16:43:03 +02002052 if (ret == -EOPNOTSUPP)
2053 cmd->ret = ret;
2054 else
2055 cmd->ret = ret ? -EIO : 0;
Christoph Hellwig15f73f52020-06-11 08:44:47 +02002056 if (likely(!blk_should_fake_timeout(rq->q)))
2057 blk_mq_complete_request(rq);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02002058 }
Ming Leib5dd2f62014-12-31 13:22:57 +00002059}
2060
Ming Leie03a3d72015-08-17 10:31:48 +08002061static void loop_queue_work(struct kthread_work *work)
Ming Leib5dd2f62014-12-31 13:22:57 +00002062{
2063 struct loop_cmd *cmd =
Ming Leie03a3d72015-08-17 10:31:48 +08002064 container_of(work, struct loop_cmd, work);
Ming Leib5dd2f62014-12-31 13:22:57 +00002065
2066 loop_handle_cmd(cmd);
2067}
2068
Christoph Hellwigd6296d392017-05-01 10:19:08 -06002069static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2070 unsigned int hctx_idx, unsigned int numa_node)
Ming Leib5dd2f62014-12-31 13:22:57 +00002071{
2072 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2073
Petr Mladek39891442016-10-11 13:55:20 -07002074 kthread_init_work(&cmd->work, loop_queue_work);
Ming Leib5dd2f62014-12-31 13:22:57 +00002075 return 0;
2076}
2077
Eric Biggersf363b082017-03-30 13:39:16 -07002078static const struct blk_mq_ops loop_mq_ops = {
Ming Leib5dd2f62014-12-31 13:22:57 +00002079 .queue_rq = loop_queue_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00002080 .init_request = loop_init_request,
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02002081 .complete = lo_complete_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00002082};
2083
Kay Sievers34dd82a2011-07-31 22:08:04 +02002084static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07002085{
2086 struct loop_device *lo;
2087 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002088 int err;
Ken Chen73285082007-05-08 00:28:20 -07002089
Silva Paulo68d740d2012-07-14 15:39:58 -07002090 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07002091 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07002092 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07002093 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002094
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06002095 lo->lo_state = Lo_unbound;
2096
Tejun Heoc718aa62013-02-27 17:03:58 -08002097 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002098 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08002099 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2100 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02002101 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002102 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08002103 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002104 }
2105 if (err < 0)
2106 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08002107 i = err;
Ken Chen73285082007-05-08 00:28:20 -07002108
Wei Yongjun183cfb52013-03-22 08:59:19 -06002109 err = -ENOMEM;
Ming Leib5dd2f62014-12-31 13:22:57 +00002110 lo->tag_set.ops = &loop_mq_ops;
2111 lo->tag_set.nr_hw_queues = 1;
2112 lo->tag_set.queue_depth = 128;
2113 lo->tag_set.numa_node = NUMA_NO_NODE;
2114 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
Ming Leibf0beec2020-05-29 15:53:15 +02002115 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
Ming Leib5dd2f62014-12-31 13:22:57 +00002116 lo->tag_set.driver_data = lo;
2117
2118 err = blk_mq_alloc_tag_set(&lo->tag_set);
2119 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002120 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07002121
Ming Leib5dd2f62014-12-31 13:22:57 +00002122 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
Chengguang Xu38a34992018-12-16 17:35:00 +08002123 if (IS_ERR(lo->lo_queue)) {
Ming Leib5dd2f62014-12-31 13:22:57 +00002124 err = PTR_ERR(lo->lo_queue);
2125 goto out_cleanup_tags;
2126 }
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06002127 lo->lo_queue->queuedata = lo;
2128
Shaohua Li54bb0ad2017-08-31 22:09:45 -07002129 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
Shaohua Li40326d82017-08-31 22:09:46 -07002130
Ming Lei5b5e20f2015-08-17 10:31:47 +08002131 /*
Shaohua Li40326d82017-08-31 22:09:46 -07002132 * By default, we do buffer IO, so it doesn't make sense to enable
2133 * merge because the I/O submitted to backing file is handled page by
2134 * page. For directio mode, merge does help to dispatch bigger request
2135 * to underlayer disk. We will enable merge once directio is enabled.
Ming Lei5b5e20f2015-08-17 10:31:47 +08002136 */
Bart Van Assche8b904b52018-03-07 17:10:10 -08002137 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei5b5e20f2015-08-17 10:31:47 +08002138
Minfei Huang7a649732016-06-07 10:05:15 +08002139 err = -ENOMEM;
Laurent Vivier476a4812008-03-26 12:11:53 +01002140 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07002141 if (!disk)
2142 goto out_free_queue;
2143
Kay Sieverse03c8dd2011-08-23 20:12:04 +02002144 /*
2145 * Disable partition scanning by default. The in-kernel partition
2146 * scanning can be requested individually per-device during its
2147 * setup. Userspace can always add and remove partitions from all
2148 * devices. The needed partition minors are allocated from the
2149 * extended minor space, the main loop device numbers will continue
2150 * to match the loop minors, regardless of the number of partitions
2151 * used.
2152 *
2153 * If max_part is given, partition scanning is globally enabled for
2154 * all loop devices. The minors for the main loop devices will be
2155 * multiples of max_part.
2156 *
2157 * Note: Global-for-all-devices, set-only-at-init, read-only module
2158 * parameteters like 'max_loop' and 'max_part' make things needlessly
2159 * complicated, are too static, inflexible and may surprise
2160 * userspace tools. Parameters like this in general should be avoided.
2161 */
2162 if (!part_shift)
2163 disk->flags |= GENHD_FL_NO_PART_SCAN;
2164 disk->flags |= GENHD_FL_EXT_DEVT;
Ming Leif8933662015-05-06 12:26:23 +08002165 atomic_set(&lo->lo_refcnt, 0);
Ken Chen73285082007-05-08 00:28:20 -07002166 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07002167 spin_lock_init(&lo->lo_lock);
2168 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01002169 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07002170 disk->fops = &lo_fops;
2171 disk->private_data = lo;
2172 disk->queue = lo->lo_queue;
2173 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002174 add_disk(disk);
2175 *l = lo;
2176 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07002177
2178out_free_queue:
2179 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00002180out_cleanup_tags:
2181 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002182out_free_idr:
2183 idr_remove(&loop_index_idr, i);
Ken Chen73285082007-05-08 00:28:20 -07002184out_free_dev:
2185 kfree(lo);
2186out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002187 return err;
Ken Chen73285082007-05-08 00:28:20 -07002188}
2189
Kay Sievers34dd82a2011-07-31 22:08:04 +02002190static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07002191{
NeilBrown6cd18e72015-04-27 14:12:22 +10002192 del_gendisk(lo->lo_disk);
Bart Van Assche0fa8ebd2018-02-28 10:15:28 -08002193 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00002194 blk_mq_free_tag_set(&lo->tag_set);
Ken Chen73285082007-05-08 00:28:20 -07002195 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07002196 kfree(lo);
2197}
2198
Kay Sievers770fe302011-07-31 22:08:04 +02002199static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07002200{
Kay Sievers770fe302011-07-31 22:08:04 +02002201 struct loop_device *lo = ptr;
2202 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07002203
Kay Sievers770fe302011-07-31 22:08:04 +02002204 if (lo->lo_state == Lo_unbound) {
2205 *l = lo;
2206 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07002207 }
Kay Sievers770fe302011-07-31 22:08:04 +02002208 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07002209}
2210
Kay Sievers34dd82a2011-07-31 22:08:04 +02002211static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07002212{
Ken Chena47653f2007-06-08 13:46:44 -07002213 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002214 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07002215
Kay Sievers770fe302011-07-31 22:08:04 +02002216 if (i < 0) {
2217 int err;
2218
2219 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2220 if (err == 1) {
2221 *l = lo;
2222 ret = lo->lo_number;
2223 }
2224 goto out;
2225 }
2226
2227 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002228 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07002229 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02002230 *l = lo;
2231 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07002232 }
Kay Sievers770fe302011-07-31 22:08:04 +02002233out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002234 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07002235}
2236
Ken Chen73285082007-05-08 00:28:20 -07002237static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2238{
Al Viro705962c2007-05-13 05:52:32 -04002239 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04002240 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002241 int err;
Ken Chen73285082007-05-08 00:28:20 -07002242
Jan Kara0a42e992018-11-08 14:01:04 +01002243 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002244 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2245 if (err < 0)
2246 err = loop_add(&lo, MINOR(dev) >> part_shift);
2247 if (err < 0)
Mikulas Patockaa207f592013-10-14 12:13:24 -04002248 kobj = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002249 else
Jan Kara3079c222018-02-26 13:01:38 +01002250 kobj = get_disk_and_module(lo->lo_disk);
Jan Kara0a42e992018-11-08 14:01:04 +01002251 mutex_unlock(&loop_ctl_mutex);
Ken Chen73285082007-05-08 00:28:20 -07002252
2253 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04002254 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07002255}
2256
Kay Sievers770fe302011-07-31 22:08:04 +02002257static long loop_control_ioctl(struct file *file, unsigned int cmd,
2258 unsigned long parm)
2259{
2260 struct loop_device *lo;
Jan Kara0a42e992018-11-08 14:01:04 +01002261 int ret;
Kay Sievers770fe302011-07-31 22:08:04 +02002262
Jan Kara0a42e992018-11-08 14:01:04 +01002263 ret = mutex_lock_killable(&loop_ctl_mutex);
2264 if (ret)
2265 return ret;
2266
2267 ret = -ENOSYS;
Kay Sievers770fe302011-07-31 22:08:04 +02002268 switch (cmd) {
2269 case LOOP_CTL_ADD:
2270 ret = loop_lookup(&lo, parm);
2271 if (ret >= 0) {
2272 ret = -EEXIST;
2273 break;
2274 }
2275 ret = loop_add(&lo, parm);
2276 break;
2277 case LOOP_CTL_REMOVE:
2278 ret = loop_lookup(&lo, parm);
2279 if (ret < 0)
2280 break;
Kay Sievers770fe302011-07-31 22:08:04 +02002281 if (lo->lo_state != Lo_unbound) {
2282 ret = -EBUSY;
Kay Sievers770fe302011-07-31 22:08:04 +02002283 break;
2284 }
Ming Leif8933662015-05-06 12:26:23 +08002285 if (atomic_read(&lo->lo_refcnt) > 0) {
Kay Sievers770fe302011-07-31 22:08:04 +02002286 ret = -EBUSY;
Kay Sievers770fe302011-07-31 22:08:04 +02002287 break;
2288 }
2289 lo->lo_disk->private_data = NULL;
Kay Sievers770fe302011-07-31 22:08:04 +02002290 idr_remove(&loop_index_idr, lo->lo_number);
2291 loop_remove(lo);
2292 break;
2293 case LOOP_CTL_GET_FREE:
2294 ret = loop_lookup(&lo, -1);
2295 if (ret >= 0)
2296 break;
2297 ret = loop_add(&lo, -1);
2298 }
Jan Kara0a42e992018-11-08 14:01:04 +01002299 mutex_unlock(&loop_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02002300
2301 return ret;
2302}
2303
2304static const struct file_operations loop_ctl_fops = {
2305 .open = nonseekable_open,
2306 .unlocked_ioctl = loop_control_ioctl,
2307 .compat_ioctl = loop_control_ioctl,
2308 .owner = THIS_MODULE,
2309 .llseek = noop_llseek,
2310};
2311
2312static struct miscdevice loop_misc = {
2313 .minor = LOOP_CTRL_MINOR,
2314 .name = "loop-control",
2315 .fops = &loop_ctl_fops,
2316};
2317
2318MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2319MODULE_ALIAS("devname:loop-control");
2320
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321static int __init loop_init(void)
2322{
Ken Chena47653f2007-06-08 13:46:44 -07002323 int i, nr;
2324 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002325 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02002326 int err;
Ken Chena47653f2007-06-08 13:46:44 -07002327
Laurent Vivier476a4812008-03-26 12:11:53 +01002328 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02002329 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01002330 part_shift = fls(max_part);
2331
Namhyung Kimac04fee2011-05-27 07:59:25 +02002332 /*
2333 * Adjust max_part according to part_shift as it is exported
2334 * to user space so that user can decide correct minor number
2335 * if [s]he want to create more devices.
2336 *
2337 * Note that -1 is required because partition 0 is reserved
2338 * for the whole disk.
2339 */
2340 max_part = (1UL << part_shift) - 1;
2341 }
2342
Guo Chaob1a66502013-02-21 15:16:49 -08002343 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2344 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002345 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002346 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02002347
Guo Chaob1a66502013-02-21 15:16:49 -08002348 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2349 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002350 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Kay Sieversd134b002011-07-31 22:08:04 +02002353 /*
2354 * If max_loop is specified, create that many devices upfront.
2355 * This also becomes a hard limit. If max_loop is not specified,
2356 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2357 * init time. Loop devices can be requested on-demand with the
2358 * /dev/loop-control interface, or be instantiated by accessing
2359 * a 'dead' device node.
2360 */
Ken Chen73285082007-05-08 00:28:20 -07002361 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07002362 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002363 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07002364 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02002365 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02002366 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 }
Ken Chena47653f2007-06-08 13:46:44 -07002368
Anton Volkova8c1d062017-08-07 15:37:50 +03002369 err = misc_register(&loop_misc);
2370 if (err < 0)
2371 goto err_out;
2372
2373
Guo Chaob1a66502013-02-21 15:16:49 -08002374 if (register_blkdev(LOOP_MAJOR, "loop")) {
2375 err = -EIO;
2376 goto misc_out;
2377 }
Ken Chena47653f2007-06-08 13:46:44 -07002378
Ken Chena47653f2007-06-08 13:46:44 -07002379 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2380 THIS_MODULE, loop_probe, NULL, NULL);
2381
Kay Sieversd134b002011-07-31 22:08:04 +02002382 /* pre-create number of devices given by config or max_loop */
Jan Kara0a42e992018-11-08 14:01:04 +01002383 mutex_lock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002384 for (i = 0; i < nr; i++)
2385 loop_add(&lo, i);
Jan Kara0a42e992018-11-08 14:01:04 +01002386 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002387
Ken Chen73285082007-05-08 00:28:20 -07002388 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08002390
2391misc_out:
2392 misc_deregister(&loop_misc);
Anton Volkova8c1d062017-08-07 15:37:50 +03002393err_out:
Guo Chaob1a66502013-02-21 15:16:49 -08002394 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002395}
Ken Chena47653f2007-06-08 13:46:44 -07002396
Kay Sievers34dd82a2011-07-31 22:08:04 +02002397static int loop_exit_cb(int id, void *ptr, void *data)
2398{
2399 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07002400
Kay Sievers34dd82a2011-07-31 22:08:04 +02002401 loop_remove(lo);
2402 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403}
2404
Ken Chen73285082007-05-08 00:28:20 -07002405static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
Ken Chena47653f2007-06-08 13:46:44 -07002407 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
Namhyung Kima1c15c52011-05-24 16:48:55 +02002409 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07002410
Luis Chamberlain200f9332020-06-19 20:47:27 +00002411 mutex_lock(&loop_ctl_mutex);
2412
Kay Sievers34dd82a2011-07-31 22:08:04 +02002413 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002414 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07002415
Ken Chena47653f2007-06-08 13:46:44 -07002416 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07002417 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02002418
2419 misc_deregister(&loop_misc);
Luis Chamberlain200f9332020-06-19 20:47:27 +00002420
2421 mutex_unlock(&loop_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423
2424module_init(loop_init);
2425module_exit(loop_exit);
2426
2427#ifndef MODULE
2428static int __init max_loop_setup(char *str)
2429{
2430 max_loop = simple_strtol(str, NULL, 0);
2431 return 1;
2432}
2433
2434__setup("max_loop=", max_loop_setup);
2435#endif