blob: f6c204f62b1e36c5d22e43d255553712f67f8d72 [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>
Al Viro83a87612013-05-12 10:14:07 -040079#include "loop.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080081#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Kay Sievers34dd82a2011-07-31 22:08:04 +020083static DEFINE_IDR(loop_index_idr);
84static DEFINE_MUTEX(loop_index_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Laurent Vivier476a4812008-03-26 12:11:53 +010086static int max_part;
87static int part_shift;
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
93{
Cong Wangcfd80052011-11-25 23:14:18 +080094 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 char *in, *out, *key;
97 int i, keysize;
98
99 if (cmd == READ) {
100 in = raw_buf;
101 out = loop_buf;
102 } else {
103 in = loop_buf;
104 out = raw_buf;
105 }
106
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
111
Cong Wangcfd80052011-11-25 23:14:18 +0800112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 cond_resched();
115 return 0;
116}
117
118static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
119{
120 if (unlikely(info->lo_encrypt_key_size <= 0))
121 return -EINVAL;
122 return 0;
123}
124
125static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200127};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
132 .init = xor_init
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200133};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135/* xfer_funcs[0] is special - its release function is never called */
136static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
137 &none_funcs,
138 &xor_funcs
139};
140
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100141static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
Guo Chaob7a1da62013-02-21 15:16:50 -0800143 loff_t loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 /* Compute loopsize in bytes */
Guo Chaob7a1da62013-02-21 15:16:50 -0800146 loopsize = i_size_read(file->f_mapping->host);
147 if (offset > 0)
148 loopsize -= offset;
149 /* offset is beyond i_size, weird but possible */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100150 if (loopsize < 0)
151 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 /*
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
158 */
159 return loopsize >> 9;
160}
161
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100162static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
165}
166
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800167static void __loop_update_dio(struct loop_device *lo, bool dio)
168{
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
174 bool use_dio;
175
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
179 }
180
181 /*
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
186 *
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
Masahiro Yamada89d790a2017-02-27 14:29:01 -0800189 * of requests in sane applications should be PAGE_SIZE aligned
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800190 */
191 if (dio) {
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
195 !lo->transfer)
196 use_dio = true;
197 else
198 use_dio = false;
199 } else {
200 use_dio = false;
201 }
202
203 if (lo->use_dio == use_dio)
204 return;
205
206 /* flush dirty pages before changing direct IO */
207 vfs_fsync(file, 0);
208
209 /*
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
213 */
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
216 if (use_dio)
217 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
218 else
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 blk_mq_unfreeze_queue(lo->lo_queue);
221}
222
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100223static int
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -0700224figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100225{
226 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 sector_t x = (sector_t)size;
Guo Chao7b0576a2013-02-21 15:16:47 -0800228 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 if (unlikely((loff_t)x != size))
231 return -EFBIG;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100232 if (lo->lo_offset != offset)
233 lo->lo_offset = offset;
234 if (lo->lo_sizelimit != sizelimit)
235 lo->lo_sizelimit = sizelimit;
Ken Chen73285082007-05-08 00:28:20 -0700236 set_capacity(lo->lo_disk, x);
Guo Chao7b0576a2013-02-21 15:16:47 -0800237 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
238 /* let user-space know about the new size */
239 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100240 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243static inline int
244lo_do_transfer(struct loop_device *lo, int cmd,
245 struct page *rpage, unsigned roffs,
246 struct page *lpage, unsigned loffs,
247 int size, sector_t rblock)
248{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200249 int ret;
250
251 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
252 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return 0;
254
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200255 printk_ratelimited(KERN_ERR
256 "loop: Transfer error at byte offset %llu, length %i.\n",
257 (unsigned long long)rblock << 9, size);
258 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200261static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200263 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 ssize_t bw;
Al Viro283e7e52015-04-03 15:21:59 -0400265
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200266 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Al Viro03d95eb2013-03-20 13:04:20 -0400268 file_start_write(file);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300269 bw = vfs_iter_write(file, &i, ppos, 0);
Al Viro03d95eb2013-03-20 13:04:20 -0400270 file_end_write(file);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200271
272 if (likely(bw == bvec->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return 0;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200274
275 printk_ratelimited(KERN_ERR
276 "loop: Write error at byte offset %llu, length %i.\n",
277 (unsigned long long)*ppos, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (bw >= 0)
279 bw = -EIO;
280 return bw;
281}
282
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200283static int lo_write_simple(struct loop_device *lo, struct request *rq,
284 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200286 struct bio_vec bvec;
287 struct req_iterator iter;
288 int ret = 0;
289
290 rq_for_each_segment(bvec, rq, iter) {
291 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
292 if (ret < 0)
293 break;
294 cond_resched();
295 }
296
297 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
299
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200300/*
Christoph Hellwig456be142011-10-17 12:57:20 +0200301 * This is the slow, transforming version that needs to double buffer the
302 * data as it cannot do the transformations in place without having direct
303 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 */
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200305static int lo_write_transfer(struct loop_device *lo, struct request *rq,
306 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200308 struct bio_vec bvec, b;
Ming Lei30112012014-12-31 13:22:58 +0000309 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200310 struct page *page;
Kent Overstreet79886132013-11-23 17:19:00 -0800311 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200313 page = alloc_page(GFP_NOIO);
314 if (unlikely(!page))
315 return -ENOMEM;
Christoph Hellwig456be142011-10-17 12:57:20 +0200316
Ming Lei30112012014-12-31 13:22:58 +0000317 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200318 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
319 bvec.bv_offset, bvec.bv_len, pos >> 9);
320 if (unlikely(ret))
321 break;
322
323 b.bv_page = page;
324 b.bv_offset = 0;
325 b.bv_len = bvec.bv_len;
326 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (ret < 0)
328 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200330
331 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200335static int lo_read_simple(struct loop_device *lo, struct request *rq,
336 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Kent Overstreet79886132013-11-23 17:19:00 -0800338 struct bio_vec bvec;
Ming Lei30112012014-12-31 13:22:58 +0000339 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200340 struct iov_iter i;
341 ssize_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Ming Lei30112012014-12-31 13:22:58 +0000343 rq_for_each_segment(bvec, rq, iter) {
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200344 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200346 if (len < 0)
347 return len;
Dave Young306df072012-02-08 22:07:19 +0100348
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200349 flush_dcache_page(bvec.bv_page);
350
351 if (len != bvec.bv_len) {
Ming Lei30112012014-12-31 13:22:58 +0000352 struct bio *bio;
353
354 __rq_for_each_bio(bio, rq)
355 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 break;
Dave Young306df072012-02-08 22:07:19 +0100357 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200358 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200360
Dave Young306df072012-02-08 22:07:19 +0100361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200364static int lo_read_transfer(struct loop_device *lo, struct request *rq,
365 loff_t pos)
366{
367 struct bio_vec bvec, b;
368 struct req_iterator iter;
369 struct iov_iter i;
370 struct page *page;
371 ssize_t len;
372 int ret = 0;
373
374 page = alloc_page(GFP_NOIO);
375 if (unlikely(!page))
376 return -ENOMEM;
377
378 rq_for_each_segment(bvec, rq, iter) {
379 loff_t offset = pos;
380
381 b.bv_page = page;
382 b.bv_offset = 0;
383 b.bv_len = bvec.bv_len;
384
385 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200387 if (len < 0) {
388 ret = len;
389 goto out_free_page;
390 }
391
392 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
393 bvec.bv_offset, len, offset >> 9);
394 if (ret)
395 goto out_free_page;
396
397 flush_dcache_page(bvec.bv_page);
398
399 if (len != bvec.bv_len) {
400 struct bio *bio;
401
402 __rq_for_each_bio(bio, rq)
403 zero_fill_bio(bio);
404 break;
405 }
406 }
407
408 ret = 0;
409out_free_page:
410 __free_page(page);
411 return ret;
412}
413
Ming Leicf655d92014-12-31 13:22:59 +0000414static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
415{
416 /*
417 * We use punch hole to reclaim the free space used by the
418 * image a.k.a. discard. However we do not support discard if
419 * encryption is enabled, because it may give an attacker
420 * useful information.
421 */
422 struct file *file = lo->lo_backing_file;
423 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
424 int ret;
425
426 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
427 ret = -EOPNOTSUPP;
428 goto out;
429 }
430
431 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
432 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
433 ret = -EIO;
434 out:
435 return ret;
436}
437
438static int lo_req_flush(struct loop_device *lo, struct request *rq)
439{
440 struct file *file = lo->lo_backing_file;
441 int ret = vfs_fsync(file, 0);
442 if (unlikely(ret && ret != -EINVAL))
443 ret = -EIO;
444
445 return ret;
446}
447
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200448static void lo_complete_rq(struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800449{
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200450 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Ming Leibc07c102015-08-17 10:31:51 +0800451
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200452 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio &&
453 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) {
Ming Leibc07c102015-08-17 10:31:51 +0800454 struct bio *bio = cmd->rq->bio;
455
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200456 bio_advance(bio, cmd->ret);
Ming Leibc07c102015-08-17 10:31:51 +0800457 zero_fill_bio(bio);
458 }
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200459
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200460 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
Ming Leibc07c102015-08-17 10:31:51 +0800461}
462
463static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
464{
465 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
Ming Leibc07c102015-08-17 10:31:51 +0800466
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200467 cmd->ret = ret;
Christoph Hellwig08e00292017-04-20 16:03:09 +0200468 blk_mq_complete_request(cmd->rq);
Ming Leibc07c102015-08-17 10:31:51 +0800469}
470
471static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
472 loff_t pos, bool rw)
473{
474 struct iov_iter iter;
475 struct bio_vec *bvec;
476 struct bio *bio = cmd->rq->bio;
477 struct file *file = lo->lo_backing_file;
478 int ret;
479
480 /* nomerge for loop request queue */
481 WARN_ON(cmd->rq->bio != cmd->rq->biotail);
482
483 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
484 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
485 bio_segments(bio), blk_rq_bytes(cmd->rq));
Ming Leia7297a62016-04-15 18:51:28 +0800486 /*
487 * This bio may be started from the middle of the 'bvec'
488 * because of bio splitting, so offset from the bvec must
489 * be passed to iov iterator
490 */
491 iter.iov_offset = bio->bi_iter.bi_bvec_done;
Ming Leibc07c102015-08-17 10:31:51 +0800492
493 cmd->iocb.ki_pos = pos;
494 cmd->iocb.ki_filp = file;
495 cmd->iocb.ki_complete = lo_rw_aio_complete;
496 cmd->iocb.ki_flags = IOCB_DIRECT;
497
498 if (rw == WRITE)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100499 ret = call_write_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800500 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100501 ret = call_read_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800502
503 if (ret != -EIOCBQUEUED)
504 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
505 return 0;
506}
507
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200508static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800509{
510 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200511 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800512
513 /*
514 * lo_write_simple and lo_read_simple should have been covered
515 * by io submit style function like lo_rw_aio(), one blocker
516 * is that lo_read_simple() need to call flush_dcache_page after
517 * the page is written from kernel, and it isn't easy to handle
518 * this in io submit style function which submits all segments
519 * of the req at one time. And direct read IO doesn't need to
520 * run flush_dcache_page().
521 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200522 switch (req_op(rq)) {
523 case REQ_OP_FLUSH:
524 return lo_req_flush(lo, rq);
525 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +0200526 case REQ_OP_WRITE_ZEROES:
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200527 return lo_discard(lo, rq, pos);
528 case REQ_OP_WRITE:
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200529 if (lo->transfer)
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200530 return lo_write_transfer(lo, rq, pos);
531 else if (cmd->use_aio)
532 return lo_rw_aio(lo, cmd, pos, WRITE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200533 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200534 return lo_write_simple(lo, rq, pos);
535 case REQ_OP_READ:
536 if (lo->transfer)
537 return lo_read_transfer(lo, rq, pos);
538 else if (cmd->use_aio)
539 return lo_rw_aio(lo, cmd, pos, READ);
540 else
541 return lo_read_simple(lo, rq, pos);
542 default:
543 WARN_ON_ONCE(1);
544 return -EIO;
545 break;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800549static inline void loop_update_dio(struct loop_device *lo)
550{
551 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
552 lo->use_dio);
553}
554
Ming Lei06f0e9e2015-05-06 12:26:24 +0800555static void loop_reread_partitions(struct loop_device *lo,
556 struct block_device *bdev)
557{
558 int rc;
559
560 /*
561 * bd_mutex has been held already in release path, so don't
562 * acquire it if this function is called in such case.
563 *
564 * If the reread partition isn't from release path, lo_refcnt
565 * must be at least one and it can only become zero when the
566 * current holder is released.
567 */
568 if (!atomic_read(&lo->lo_refcnt))
569 rc = __blkdev_reread_part(bdev);
570 else
571 rc = blkdev_reread_part(bdev);
572 if (rc)
573 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
574 __func__, lo->lo_number, lo->lo_file_name, rc);
575}
576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577/*
578 * loop_change_fd switched the backing store of a loopback device to
579 * a new file. This is useful for operating system installers to free up
580 * the original file and in High Availability environments to switch to
581 * an alternative location for the content in case of server meltdown.
582 * This can only work if the loop device is used read-only, and if the
583 * new backing store is the same size and type as the old backing store.
584 */
Al Virobb214882008-03-02 09:29:48 -0500585static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
586 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
588 struct file *file, *old_file;
589 struct inode *inode;
590 int error;
591
592 error = -ENXIO;
593 if (lo->lo_state != Lo_bound)
594 goto out;
595
596 /* the loop device has to be read-only */
597 error = -EINVAL;
598 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
599 goto out;
600
601 error = -EBADF;
602 file = fget(arg);
603 if (!file)
604 goto out;
605
606 inode = file->f_mapping->host;
607 old_file = lo->lo_backing_file;
608
609 error = -EINVAL;
610
611 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
612 goto out_putf;
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 /* size of the new backing store needs to be the same */
615 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
616 goto out_putf;
617
618 /* and ... switch */
Omar Sandoval43cade802017-08-24 00:03:44 -0700619 blk_mq_freeze_queue(lo->lo_queue);
620 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
621 lo->lo_backing_file = file;
622 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
623 mapping_set_gfp_mask(file->f_mapping,
624 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
625 loop_update_dio(lo);
626 blk_mq_unfreeze_queue(lo->lo_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200629 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800630 loop_reread_partitions(lo, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 return 0;
632
633 out_putf:
634 fput(file);
635 out:
636 return error;
637}
638
639static inline int is_loop_device(struct file *file)
640{
641 struct inode *i = file->f_mapping->host;
642
643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
644}
645
Milan Brozee862732010-08-23 15:16:00 +0200646/* loop sysfs attributes */
647
648static ssize_t loop_attr_show(struct device *dev, char *page,
649 ssize_t (*callback)(struct loop_device *, char *))
650{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200651 struct gendisk *disk = dev_to_disk(dev);
652 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200653
Kay Sievers34dd82a2011-07-31 22:08:04 +0200654 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200655}
656
657#define LOOP_ATTR_RO(_name) \
658static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
659static ssize_t loop_attr_do_show_##_name(struct device *d, \
660 struct device_attribute *attr, char *b) \
661{ \
662 return loop_attr_show(d, b, loop_attr_##_name##_show); \
663} \
664static struct device_attribute loop_attr_##_name = \
665 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
666
667static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
668{
669 ssize_t ret;
670 char *p = NULL;
671
Kay Sievers05eb0f22011-07-31 22:21:35 +0200672 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200673 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200674 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200675 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200676
677 if (IS_ERR_OR_NULL(p))
678 ret = PTR_ERR(p);
679 else {
680 ret = strlen(p);
681 memmove(buf, p, ret);
682 buf[ret++] = '\n';
683 buf[ret] = 0;
684 }
685
686 return ret;
687}
688
689static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
690{
691 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
692}
693
694static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
695{
696 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
697}
698
699static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
700{
701 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
702
703 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
704}
705
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200706static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
707{
708 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
709
710 return sprintf(buf, "%s\n", partscan ? "1" : "0");
711}
712
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800713static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
714{
715 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
716
717 return sprintf(buf, "%s\n", dio ? "1" : "0");
718}
719
Milan Brozee862732010-08-23 15:16:00 +0200720LOOP_ATTR_RO(backing_file);
721LOOP_ATTR_RO(offset);
722LOOP_ATTR_RO(sizelimit);
723LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200724LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800725LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200726
727static struct attribute *loop_attrs[] = {
728 &loop_attr_backing_file.attr,
729 &loop_attr_offset.attr,
730 &loop_attr_sizelimit.attr,
731 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200732 &loop_attr_partscan.attr,
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800733 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200734 NULL,
735};
736
737static struct attribute_group loop_attribute_group = {
738 .name = "loop",
739 .attrs= loop_attrs,
740};
741
742static int loop_sysfs_init(struct loop_device *lo)
743{
744 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
745 &loop_attribute_group);
746}
747
748static void loop_sysfs_exit(struct loop_device *lo)
749{
750 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
751 &loop_attribute_group);
752}
753
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200754static void loop_config_discard(struct loop_device *lo)
755{
756 struct file *file = lo->lo_backing_file;
757 struct inode *inode = file->f_mapping->host;
758 struct request_queue *q = lo->lo_queue;
759
760 /*
761 * We use punch hole to reclaim the free space used by the
Olaf Hering12a64d22014-01-21 14:39:24 -0800762 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200763 * encryption is enabled, because it may give an attacker
764 * useful information.
765 */
766 if ((!file->f_op->fallocate) ||
767 lo->lo_encrypt_key_size) {
768 q->limits.discard_granularity = 0;
769 q->limits.discard_alignment = 0;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600770 blk_queue_max_discard_sectors(q, 0);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200771 blk_queue_max_write_zeroes_sectors(q, 0);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200772 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
773 return;
774 }
775
776 q->limits.discard_granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaf3c02011-12-02 14:47:03 +0100777 q->limits.discard_alignment = 0;
Hannes Reineckef2c6df72017-06-08 13:46:45 +0200778
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -0700779 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
780 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200781 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
782}
783
Ming Leie03a3d72015-08-17 10:31:48 +0800784static void loop_unprepare_queue(struct loop_device *lo)
785{
Petr Mladek39891442016-10-11 13:55:20 -0700786 kthread_flush_worker(&lo->worker);
Ming Leie03a3d72015-08-17 10:31:48 +0800787 kthread_stop(lo->worker_task);
788}
789
NeilBrownb2ee7d42017-06-16 15:02:09 +1000790static int loop_kthread_worker_fn(void *worker_ptr)
791{
792 current->flags |= PF_LESS_THROTTLE;
793 return kthread_worker_fn(worker_ptr);
794}
795
Ming Leie03a3d72015-08-17 10:31:48 +0800796static int loop_prepare_queue(struct loop_device *lo)
797{
Petr Mladek39891442016-10-11 13:55:20 -0700798 kthread_init_worker(&lo->worker);
NeilBrownb2ee7d42017-06-16 15:02:09 +1000799 lo->worker_task = kthread_run(loop_kthread_worker_fn,
Ming Leie03a3d72015-08-17 10:31:48 +0800800 &lo->worker, "loop%d", lo->lo_number);
801 if (IS_ERR(lo->worker_task))
802 return -ENOMEM;
803 set_user_nice(lo->worker_task, MIN_NICE);
804 return 0;
805}
806
Al Virobb214882008-03-02 09:29:48 -0500807static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct block_device *bdev, unsigned int arg)
809{
810 struct file *file, *f;
811 struct inode *inode;
812 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 int lo_flags = 0;
814 int error;
815 loff_t size;
816
817 /* This is safe, since we have a reference from open(). */
818 __module_get(THIS_MODULE);
819
820 error = -EBADF;
821 file = fget(arg);
822 if (!file)
823 goto out;
824
825 error = -EBUSY;
826 if (lo->lo_state != Lo_unbound)
827 goto out_putf;
828
829 /* Avoid recursion */
830 f = file;
831 while (is_loop_device(f)) {
832 struct loop_device *l;
833
Al Virobb214882008-03-02 09:29:48 -0500834 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 goto out_putf;
836
837 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
838 if (l->lo_state == Lo_unbound) {
839 error = -EINVAL;
840 goto out_putf;
841 }
842 f = l->lo_backing_file;
843 }
844
845 mapping = file->f_mapping;
846 inode = mapping->host;
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 error = -EINVAL;
Christoph Hellwig456be142011-10-17 12:57:20 +0200849 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Christoph Hellwig456be142011-10-17 12:57:20 +0200852 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
Al Viro283e7e52015-04-03 15:21:59 -0400853 !file->f_op->write_iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 lo_flags |= LO_FLAGS_READ_ONLY;
855
Christoph Hellwig456be142011-10-17 12:57:20 +0200856 error = -EFBIG;
857 size = get_loop_size(lo, file);
858 if ((loff_t)(sector_t)size != size)
859 goto out_putf;
Ming Leie03a3d72015-08-17 10:31:48 +0800860 error = loop_prepare_queue(lo);
861 if (error)
Ming Leif4aa4c72015-05-05 19:49:54 +0800862 goto out_putf;
Christoph Hellwig456be142011-10-17 12:57:20 +0200863
864 error = 0;
865
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
867
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800868 lo->use_dio = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 lo->lo_device = bdev;
870 lo->lo_flags = lo_flags;
871 lo->lo_backing_file = file;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200872 lo->transfer = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 lo->ioctl = NULL;
874 lo->lo_sizelimit = 0;
875 lo->old_gfp_mask = mapping_gfp_mask(mapping);
876 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
877
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100878 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Jens Axboe21d07272016-03-30 10:09:35 -0600879 blk_queue_write_cache(lo->lo_queue, true, false);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100880
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800881 loop_update_dio(lo);
Ken Chen73285082007-05-08 00:28:20 -0700882 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200884 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200885 /* let user-space know about the new size */
886 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Omar Sandoval8a0740c2017-08-24 00:03:41 -0700888 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
889 block_size(inode->i_bdev) : PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700891 lo->lo_state = Lo_bound;
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200892 if (part_shift)
893 lo->lo_flags |= LO_FLAGS_PARTSCAN;
894 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800895 loop_reread_partitions(lo, bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700896
897 /* Grab the block_device to prevent its destruction after we
898 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
899 */
900 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return 0;
902
903 out_putf:
904 fput(file);
905 out:
906 /* This is safe: open() is still holding a reference. */
907 module_put(THIS_MODULE);
908 return error;
909}
910
911static int
912loop_release_xfer(struct loop_device *lo)
913{
914 int err = 0;
915 struct loop_func_table *xfer = lo->lo_encryption;
916
917 if (xfer) {
918 if (xfer->release)
919 err = xfer->release(lo);
920 lo->transfer = NULL;
921 lo->lo_encryption = NULL;
922 module_put(xfer->owner);
923 }
924 return err;
925}
926
927static int
928loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
929 const struct loop_info64 *i)
930{
931 int err = 0;
932
933 if (xfer) {
934 struct module *owner = xfer->owner;
935
936 if (!try_module_get(owner))
937 return -EINVAL;
938 if (xfer->init)
939 err = xfer->init(lo, i);
940 if (err)
941 module_put(owner);
942 else
943 lo->lo_encryption = xfer;
944 }
945 return err;
946}
947
Ayan George4c823cc2011-09-21 10:02:13 +0200948static int loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
950 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400951 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +0200952 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 if (lo->lo_state != Lo_bound)
955 return -ENXIO;
956
Dave Chinnera1ecac32012-09-28 10:42:23 +0200957 /*
958 * If we've explicitly asked to tear down the loop device,
959 * and it has an elevated reference count, set it for auto-teardown when
960 * the last reference goes away. This stops $!~#$@ udev from
961 * preventing teardown because it decided that it needs to run blkid on
962 * the loopback device whenever they appear. xfstests is notorious for
963 * failing tests because blkid via udev races with a losetup
964 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
965 * command to fail with EBUSY.
966 */
Ming Leif8933662015-05-06 12:26:23 +0800967 if (atomic_read(&lo->lo_refcnt) > 1) {
Dave Chinnera1ecac32012-09-28 10:42:23 +0200968 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
969 mutex_unlock(&lo->lo_ctl_mutex);
970 return 0;
971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 if (filp == NULL)
974 return -EINVAL;
975
Ming Leif8933662015-05-06 12:26:23 +0800976 /* freeze request queue during the transition */
977 blk_mq_freeze_queue(lo->lo_queue);
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 spin_lock_irq(&lo->lo_lock);
980 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +0200982 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 loop_release_xfer(lo);
985 lo->transfer = NULL;
986 lo->ioctl = NULL;
987 lo->lo_device = NULL;
988 lo->lo_encryption = NULL;
989 lo->lo_offset = 0;
990 lo->lo_sizelimit = 0;
991 lo->lo_encrypt_key_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
993 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
994 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Omar Sandoval89e4fde2017-08-24 00:03:43 -0700995 blk_queue_logical_block_size(lo->lo_queue, 512);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700996 if (bdev) {
997 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -0500998 invalidate_bdev(bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700999 }
Ken Chen73285082007-05-08 00:28:20 -07001000 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001001 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001002 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001003 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001004 /* let user-space know about this change */
1005 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 mapping_set_gfp_mask(filp->f_mapping, gfp);
1008 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 /* This is safe: open() is still holding a reference. */
1010 module_put(THIS_MODULE);
Ming Leif8933662015-05-06 12:26:23 +08001011 blk_mq_unfreeze_queue(lo->lo_queue);
1012
Jens Axboec2fccc12013-04-08 10:12:11 +02001013 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
Ming Lei06f0e9e2015-05-06 12:26:24 +08001014 loop_reread_partitions(lo, bdev);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001015 lo->lo_flags = 0;
1016 if (!part_shift)
1017 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Ming Leie03a3d72015-08-17 10:31:48 +08001018 loop_unprepare_queue(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001019 mutex_unlock(&lo->lo_ctl_mutex);
1020 /*
1021 * Need not hold lo_ctl_mutex to fput backing file.
1022 * Calling fput holding lo_ctl_mutex triggers a circular
1023 * lock dependency possibility warning as fput can take
1024 * bd_mutex which is usually taken before lo_ctl_mutex.
1025 */
1026 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 return 0;
1028}
1029
1030static int
1031loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1032{
1033 int err;
1034 struct loop_func_table *xfer;
Eric W. Biedermane4849732012-02-11 11:23:51 -08001035 kuid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
David Howellsb0fafa82008-11-14 10:38:41 +11001037 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001038 !uid_eq(lo->lo_key_owner, uid) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 !capable(CAP_SYS_ADMIN))
1040 return -EPERM;
1041 if (lo->lo_state != Lo_bound)
1042 return -ENXIO;
1043 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1044 return -EINVAL;
1045
Ming Leiecdd0952017-02-11 11:40:45 +08001046 /* I/O need to be drained during transfer transition */
1047 blk_mq_freeze_queue(lo->lo_queue);
1048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 err = loop_release_xfer(lo);
1050 if (err)
Ming Leiecdd0952017-02-11 11:40:45 +08001051 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 if (info->lo_encrypt_type) {
1054 unsigned int type = info->lo_encrypt_type;
1055
1056 if (type >= MAX_LO_CRYPT)
1057 return -EINVAL;
1058 xfer = xfer_funcs[type];
1059 if (xfer == NULL)
1060 return -EINVAL;
1061 } else
1062 xfer = NULL;
1063
1064 err = loop_init_xfer(lo, xfer, info);
1065 if (err)
Ming Leiecdd0952017-02-11 11:40:45 +08001066 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 if (lo->lo_offset != info->lo_offset ||
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001069 lo->lo_sizelimit != info->lo_sizelimit) {
1070 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
Ming Leiecdd0952017-02-11 11:40:45 +08001071 err = -EFBIG;
1072 goto exit;
1073 }
Arnd Bergmannb040ad92017-06-09 12:19:18 +02001074 }
Guo Chao541c7422013-02-21 15:16:46 -08001075
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001076 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1079 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1080 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1081 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1082
1083 if (!xfer)
1084 xfer = &none_funcs;
1085 lo->transfer = xfer->transfer;
1086 lo->ioctl = xfer->ioctl;
1087
David Woodhouse96c58652008-02-06 01:36:27 -08001088 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1089 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1090 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1093 lo->lo_init[0] = info->lo_init[0];
1094 lo->lo_init[1] = info->lo_init[1];
1095 if (info->lo_encrypt_key_size) {
1096 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1097 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001098 lo->lo_key_owner = uid;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +02001099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Ming Lei2e5ab5f2015-08-17 10:31:49 +08001101 /* update dio if lo_offset or transfer is changed */
1102 __loop_update_dio(lo, lo->use_dio);
1103
Ming Leiecdd0952017-02-11 11:40:45 +08001104 exit:
1105 blk_mq_unfreeze_queue(lo->lo_queue);
Omar Sandovale02898b42017-03-01 10:42:38 -08001106
1107 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1108 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1109 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1110 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1111 loop_reread_partitions(lo, lo->lo_device);
1112 }
1113
Ming Leiecdd0952017-02-11 11:40:45 +08001114 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
1117static int
1118loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1119{
1120 struct file *file = lo->lo_backing_file;
1121 struct kstat stat;
1122 int error;
1123
1124 if (lo->lo_state != Lo_bound)
1125 return -ENXIO;
David Howellsa528d352017-01-31 16:46:22 +00001126 error = vfs_getattr(&file->f_path, &stat,
1127 STATX_INO, AT_STATX_SYNC_AS_STAT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (error)
1129 return error;
1130 memset(info, 0, sizeof(*info));
1131 info->lo_number = lo->lo_number;
1132 info->lo_device = huge_encode_dev(stat.dev);
1133 info->lo_inode = stat.ino;
1134 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1135 info->lo_offset = lo->lo_offset;
1136 info->lo_sizelimit = lo->lo_sizelimit;
1137 info->lo_flags = lo->lo_flags;
1138 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1139 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1140 info->lo_encrypt_type =
1141 lo->lo_encryption ? lo->lo_encryption->number : 0;
1142 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1143 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1144 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1145 lo->lo_encrypt_key_size);
1146 }
1147 return 0;
1148}
1149
1150static void
1151loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1152{
1153 memset(info64, 0, sizeof(*info64));
1154 info64->lo_number = info->lo_number;
1155 info64->lo_device = info->lo_device;
1156 info64->lo_inode = info->lo_inode;
1157 info64->lo_rdevice = info->lo_rdevice;
1158 info64->lo_offset = info->lo_offset;
1159 info64->lo_sizelimit = 0;
1160 info64->lo_encrypt_type = info->lo_encrypt_type;
1161 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1162 info64->lo_flags = info->lo_flags;
1163 info64->lo_init[0] = info->lo_init[0];
1164 info64->lo_init[1] = info->lo_init[1];
1165 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1166 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1167 else
1168 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1169 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1170}
1171
1172static int
1173loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1174{
1175 memset(info, 0, sizeof(*info));
1176 info->lo_number = info64->lo_number;
1177 info->lo_device = info64->lo_device;
1178 info->lo_inode = info64->lo_inode;
1179 info->lo_rdevice = info64->lo_rdevice;
1180 info->lo_offset = info64->lo_offset;
1181 info->lo_encrypt_type = info64->lo_encrypt_type;
1182 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1183 info->lo_flags = info64->lo_flags;
1184 info->lo_init[0] = info64->lo_init[0];
1185 info->lo_init[1] = info64->lo_init[1];
1186 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1187 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1188 else
1189 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1190 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1191
1192 /* error in case values were truncated */
1193 if (info->lo_device != info64->lo_device ||
1194 info->lo_rdevice != info64->lo_rdevice ||
1195 info->lo_inode != info64->lo_inode ||
1196 info->lo_offset != info64->lo_offset)
1197 return -EOVERFLOW;
1198
1199 return 0;
1200}
1201
1202static int
1203loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1204{
1205 struct loop_info info;
1206 struct loop_info64 info64;
1207
1208 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1209 return -EFAULT;
1210 loop_info64_from_old(&info, &info64);
1211 return loop_set_status(lo, &info64);
1212}
1213
1214static int
1215loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1216{
1217 struct loop_info64 info64;
1218
1219 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1220 return -EFAULT;
1221 return loop_set_status(lo, &info64);
1222}
1223
1224static int
1225loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1226 struct loop_info info;
1227 struct loop_info64 info64;
1228 int err = 0;
1229
1230 if (!arg)
1231 err = -EINVAL;
1232 if (!err)
1233 err = loop_get_status(lo, &info64);
1234 if (!err)
1235 err = loop_info64_to_old(&info64, &info);
1236 if (!err && copy_to_user(arg, &info, sizeof(info)))
1237 err = -EFAULT;
1238
1239 return err;
1240}
1241
1242static int
1243loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1244 struct loop_info64 info64;
1245 int err = 0;
1246
1247 if (!arg)
1248 err = -EINVAL;
1249 if (!err)
1250 err = loop_get_status(lo, &info64);
1251 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1252 err = -EFAULT;
1253
1254 return err;
1255}
1256
Hannes Reinecke51001b72017-06-08 13:46:44 +02001257static int loop_set_capacity(struct loop_device *lo)
J. R. Okajima53d66602009-03-31 15:23:43 -07001258{
J. R. Okajima53d66602009-03-31 15:23:43 -07001259 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001260 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001261
Omar Sandoval1e6ec9e2017-08-23 14:54:59 -07001262 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001263}
1264
Ming Leiab1cb272015-08-17 10:31:50 +08001265static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1266{
1267 int error = -ENXIO;
1268 if (lo->lo_state != Lo_bound)
1269 goto out;
1270
1271 __loop_update_dio(lo, !!arg);
1272 if (lo->use_dio == !!arg)
1273 return 0;
1274 error = -EINVAL;
1275 out:
1276 return error;
1277}
1278
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001279static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1280{
1281 if (lo->lo_state != Lo_bound)
1282 return -ENXIO;
1283
1284 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1285 return -EINVAL;
1286
1287 blk_mq_freeze_queue(lo->lo_queue);
1288
1289 blk_queue_logical_block_size(lo->lo_queue, arg);
1290 loop_update_dio(lo);
1291
1292 blk_mq_unfreeze_queue(lo->lo_queue);
1293
1294 return 0;
1295}
1296
Al Virobb214882008-03-02 09:29:48 -05001297static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 unsigned int cmd, unsigned long arg)
1299{
Al Virobb214882008-03-02 09:29:48 -05001300 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 int err;
1302
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001303 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 switch (cmd) {
1305 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001306 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 break;
1308 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001309 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 break;
1311 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001312 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Ayan George4c823cc2011-09-21 10:02:13 +02001313 err = loop_clr_fd(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001314 if (!err)
1315 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 break;
1317 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001318 err = -EPERM;
1319 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1320 err = loop_set_status_old(lo,
1321 (struct loop_info __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 break;
1323 case LOOP_GET_STATUS:
1324 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1325 break;
1326 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001327 err = -EPERM;
1328 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1329 err = loop_set_status64(lo,
1330 (struct loop_info64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 break;
1332 case LOOP_GET_STATUS64:
1333 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1334 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001335 case LOOP_SET_CAPACITY:
1336 err = -EPERM;
1337 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
Hannes Reinecke51001b72017-06-08 13:46:44 +02001338 err = loop_set_capacity(lo);
J. R. Okajima53d66602009-03-31 15:23:43 -07001339 break;
Ming Leiab1cb272015-08-17 10:31:50 +08001340 case LOOP_SET_DIRECT_IO:
1341 err = -EPERM;
1342 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1343 err = loop_set_dio(lo, arg);
1344 break;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001345 case LOOP_SET_BLOCK_SIZE:
1346 err = -EPERM;
1347 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1348 err = loop_set_block_size(lo, arg);
1349 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 default:
1351 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1352 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001353 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001354
1355out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return err;
1357}
1358
David Howells863d5b822006-08-29 19:06:14 +01001359#ifdef CONFIG_COMPAT
1360struct compat_loop_info {
1361 compat_int_t lo_number; /* ioctl r/o */
1362 compat_dev_t lo_device; /* ioctl r/o */
1363 compat_ulong_t lo_inode; /* ioctl r/o */
1364 compat_dev_t lo_rdevice; /* ioctl r/o */
1365 compat_int_t lo_offset;
1366 compat_int_t lo_encrypt_type;
1367 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1368 compat_int_t lo_flags; /* ioctl r/o */
1369 char lo_name[LO_NAME_SIZE];
1370 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1371 compat_ulong_t lo_init[2];
1372 char reserved[4];
1373};
1374
1375/*
1376 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1377 * - noinlined to reduce stack space usage in main part of driver
1378 */
1379static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001380loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001381 struct loop_info64 *info64)
1382{
1383 struct compat_loop_info info;
1384
1385 if (copy_from_user(&info, arg, sizeof(info)))
1386 return -EFAULT;
1387
1388 memset(info64, 0, sizeof(*info64));
1389 info64->lo_number = info.lo_number;
1390 info64->lo_device = info.lo_device;
1391 info64->lo_inode = info.lo_inode;
1392 info64->lo_rdevice = info.lo_rdevice;
1393 info64->lo_offset = info.lo_offset;
1394 info64->lo_sizelimit = 0;
1395 info64->lo_encrypt_type = info.lo_encrypt_type;
1396 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1397 info64->lo_flags = info.lo_flags;
1398 info64->lo_init[0] = info.lo_init[0];
1399 info64->lo_init[1] = info.lo_init[1];
1400 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1401 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1402 else
1403 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1404 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1405 return 0;
1406}
1407
1408/*
1409 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1410 * - noinlined to reduce stack space usage in main part of driver
1411 */
1412static noinline int
1413loop_info64_to_compat(const struct loop_info64 *info64,
1414 struct compat_loop_info __user *arg)
1415{
1416 struct compat_loop_info info;
1417
1418 memset(&info, 0, sizeof(info));
1419 info.lo_number = info64->lo_number;
1420 info.lo_device = info64->lo_device;
1421 info.lo_inode = info64->lo_inode;
1422 info.lo_rdevice = info64->lo_rdevice;
1423 info.lo_offset = info64->lo_offset;
1424 info.lo_encrypt_type = info64->lo_encrypt_type;
1425 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1426 info.lo_flags = info64->lo_flags;
1427 info.lo_init[0] = info64->lo_init[0];
1428 info.lo_init[1] = info64->lo_init[1];
1429 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1430 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1431 else
1432 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1433 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1434
1435 /* error in case values were truncated */
1436 if (info.lo_device != info64->lo_device ||
1437 info.lo_rdevice != info64->lo_rdevice ||
1438 info.lo_inode != info64->lo_inode ||
1439 info.lo_offset != info64->lo_offset ||
1440 info.lo_init[0] != info64->lo_init[0] ||
1441 info.lo_init[1] != info64->lo_init[1])
1442 return -EOVERFLOW;
1443
1444 if (copy_to_user(arg, &info, sizeof(info)))
1445 return -EFAULT;
1446 return 0;
1447}
1448
1449static int
1450loop_set_status_compat(struct loop_device *lo,
1451 const struct compat_loop_info __user *arg)
1452{
1453 struct loop_info64 info64;
1454 int ret;
1455
1456 ret = loop_info64_from_compat(arg, &info64);
1457 if (ret < 0)
1458 return ret;
1459 return loop_set_status(lo, &info64);
1460}
1461
1462static int
1463loop_get_status_compat(struct loop_device *lo,
1464 struct compat_loop_info __user *arg)
1465{
1466 struct loop_info64 info64;
1467 int err = 0;
1468
1469 if (!arg)
1470 err = -EINVAL;
1471 if (!err)
1472 err = loop_get_status(lo, &info64);
1473 if (!err)
1474 err = loop_info64_to_compat(&info64, arg);
1475 return err;
1476}
1477
Al Virobb214882008-03-02 09:29:48 -05001478static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1479 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001480{
Al Virobb214882008-03-02 09:29:48 -05001481 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001482 int err;
1483
David Howells863d5b822006-08-29 19:06:14 +01001484 switch(cmd) {
1485 case LOOP_SET_STATUS:
1486 mutex_lock(&lo->lo_ctl_mutex);
1487 err = loop_set_status_compat(
1488 lo, (const struct compat_loop_info __user *) arg);
1489 mutex_unlock(&lo->lo_ctl_mutex);
1490 break;
1491 case LOOP_GET_STATUS:
1492 mutex_lock(&lo->lo_ctl_mutex);
1493 err = loop_get_status_compat(
1494 lo, (struct compat_loop_info __user *) arg);
1495 mutex_unlock(&lo->lo_ctl_mutex);
1496 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001497 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001498 case LOOP_CLR_FD:
1499 case LOOP_GET_STATUS64:
1500 case LOOP_SET_STATUS64:
1501 arg = (unsigned long) compat_ptr(arg);
1502 case LOOP_SET_FD:
1503 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001504 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001505 break;
1506 default:
1507 err = -ENOIOCTLCMD;
1508 break;
1509 }
David Howells863d5b822006-08-29 19:06:14 +01001510 return err;
1511}
1512#endif
1513
Al Virobb214882008-03-02 09:29:48 -05001514static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
Kay Sievers770fe302011-07-31 22:08:04 +02001516 struct loop_device *lo;
1517 int err = 0;
1518
1519 mutex_lock(&loop_index_mutex);
1520 lo = bdev->bd_disk->private_data;
1521 if (!lo) {
1522 err = -ENXIO;
1523 goto out;
1524 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Ming Leif8933662015-05-06 12:26:23 +08001526 atomic_inc(&lo->lo_refcnt);
Kay Sievers770fe302011-07-31 22:08:04 +02001527out:
1528 mutex_unlock(&loop_index_mutex);
1529 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
Al Virodb2a1442013-05-05 21:52:57 -04001532static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
Al Virobb214882008-03-02 09:29:48 -05001534 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001535 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
Ming Leif8933662015-05-06 12:26:23 +08001537 if (atomic_dec_return(&lo->lo_refcnt))
1538 return;
1539
Ingo Molnarf85221d2006-03-23 03:00:38 -08001540 mutex_lock(&lo->lo_ctl_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001541 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1542 /*
1543 * In autoclear mode, stop the loop thread
1544 * and remove configuration after last close.
1545 */
Ayan George4c823cc2011-09-21 10:02:13 +02001546 err = loop_clr_fd(lo);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001547 if (!err)
Al Virodb2a1442013-05-05 21:52:57 -04001548 return;
Omar Sandoval43cade802017-08-24 00:03:44 -07001549 } else if (lo->lo_state == Lo_bound) {
Milan Broz14f27932008-12-12 14:48:27 +01001550 /*
1551 * Otherwise keep thread (if running) and config,
1552 * but flush possible ongoing bios in thread.
1553 */
Omar Sandoval43cade802017-08-24 00:03:44 -07001554 blk_mq_freeze_queue(lo->lo_queue);
1555 blk_mq_unfreeze_queue(lo->lo_queue);
Milan Broz14f27932008-12-12 14:48:27 +01001556 }
David Woodhouse96c58652008-02-06 01:36:27 -08001557
Ingo Molnarf85221d2006-03-23 03:00:38 -08001558 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001561static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001563 .open = lo_open,
1564 .release = lo_release,
1565 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001566#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001567 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001568#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569};
1570
1571/*
1572 * And now the modules code and kernel interface.
1573 */
Ken Chen73285082007-05-08 00:28:20 -07001574static int max_loop;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001575module_param(max_loop, int, S_IRUGO);
Ken Chena47653f2007-06-08 13:46:44 -07001576MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Namhyung Kimac04fee2011-05-27 07:59:25 +02001577module_param(max_part, int, S_IRUGO);
Laurent Vivier476a4812008-03-26 12:11:53 +01001578MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579MODULE_LICENSE("GPL");
1580MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1581
1582int loop_register_transfer(struct loop_func_table *funcs)
1583{
1584 unsigned int n = funcs->number;
1585
1586 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1587 return -EINVAL;
1588 xfer_funcs[n] = funcs;
1589 return 0;
1590}
1591
Kay Sievers34dd82a2011-07-31 22:08:04 +02001592static int unregister_transfer_cb(int id, void *ptr, void *data)
1593{
1594 struct loop_device *lo = ptr;
1595 struct loop_func_table *xfer = data;
1596
1597 mutex_lock(&lo->lo_ctl_mutex);
1598 if (lo->lo_encryption == xfer)
1599 loop_release_xfer(lo);
1600 mutex_unlock(&lo->lo_ctl_mutex);
1601 return 0;
1602}
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604int loop_unregister_transfer(int number)
1605{
1606 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 struct loop_func_table *xfer;
1608
1609 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1610 return -EINVAL;
1611
1612 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001613 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 return 0;
1615}
1616
1617EXPORT_SYMBOL(loop_register_transfer);
1618EXPORT_SYMBOL(loop_unregister_transfer);
1619
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001620static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
Ming Leib5dd2f62014-12-31 13:22:57 +00001621 const struct blk_mq_queue_data *bd)
1622{
1623 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
Ming Leif4aa4c72015-05-05 19:49:54 +08001624 struct loop_device *lo = cmd->rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00001625
1626 blk_mq_start_request(bd->rq);
1627
Ming Leif4aa4c72015-05-05 19:49:54 +08001628 if (lo->lo_state != Lo_bound)
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001629 return BLK_STS_IOERR;
Ming Leif4aa4c72015-05-05 19:49:54 +08001630
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001631 switch (req_op(cmd->rq)) {
1632 case REQ_OP_FLUSH:
1633 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +02001634 case REQ_OP_WRITE_ZEROES:
Ming Leibc07c102015-08-17 10:31:51 +08001635 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001636 break;
1637 default:
1638 cmd->use_aio = lo->use_dio;
1639 break;
1640 }
Ming Leibc07c102015-08-17 10:31:51 +08001641
Petr Mladek39891442016-10-11 13:55:20 -07001642 kthread_queue_work(&lo->worker, &cmd->work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001643
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001644 return BLK_STS_OK;
Ming Leib5dd2f62014-12-31 13:22:57 +00001645}
1646
1647static void loop_handle_cmd(struct loop_cmd *cmd)
1648{
Mike Christiea8ebb052016-06-05 14:31:45 -05001649 const bool write = op_is_write(req_op(cmd->rq));
Ming Leib5dd2f62014-12-31 13:22:57 +00001650 struct loop_device *lo = cmd->rq->q->queuedata;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001651 int ret = 0;
Ming Leib5dd2f62014-12-31 13:22:57 +00001652
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001653 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1654 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00001655 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001656 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001657
Ming Lei30112012014-12-31 13:22:58 +00001658 ret = do_req_filebacked(lo, cmd->rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001659 failed:
Ming Leibc07c102015-08-17 10:31:51 +08001660 /* complete non-aio request */
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001661 if (!cmd->use_aio || ret) {
1662 cmd->ret = ret ? -EIO : 0;
Christoph Hellwig08e00292017-04-20 16:03:09 +02001663 blk_mq_complete_request(cmd->rq);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001664 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001665}
1666
Ming Leie03a3d72015-08-17 10:31:48 +08001667static void loop_queue_work(struct kthread_work *work)
Ming Leib5dd2f62014-12-31 13:22:57 +00001668{
1669 struct loop_cmd *cmd =
Ming Leie03a3d72015-08-17 10:31:48 +08001670 container_of(work, struct loop_cmd, work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001671
1672 loop_handle_cmd(cmd);
1673}
1674
Christoph Hellwigd6296d32017-05-01 10:19:08 -06001675static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1676 unsigned int hctx_idx, unsigned int numa_node)
Ming Leib5dd2f62014-12-31 13:22:57 +00001677{
1678 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1679
1680 cmd->rq = rq;
Petr Mladek39891442016-10-11 13:55:20 -07001681 kthread_init_work(&cmd->work, loop_queue_work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001682
1683 return 0;
1684}
1685
Eric Biggersf363b082017-03-30 13:39:16 -07001686static const struct blk_mq_ops loop_mq_ops = {
Ming Leib5dd2f62014-12-31 13:22:57 +00001687 .queue_rq = loop_queue_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00001688 .init_request = loop_init_request,
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001689 .complete = lo_complete_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00001690};
1691
Kay Sievers34dd82a2011-07-31 22:08:04 +02001692static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001693{
1694 struct loop_device *lo;
1695 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001696 int err;
Ken Chen73285082007-05-08 00:28:20 -07001697
Silva Paulo68d740d2012-07-14 15:39:58 -07001698 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001699 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07001700 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07001701 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001702
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001703 lo->lo_state = Lo_unbound;
1704
Tejun Heoc718aa62013-02-27 17:03:58 -08001705 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001706 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08001707 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1708 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02001709 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001710 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08001711 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001712 }
1713 if (err < 0)
1714 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08001715 i = err;
Ken Chen73285082007-05-08 00:28:20 -07001716
Wei Yongjun183cfb52013-03-22 08:59:19 -06001717 err = -ENOMEM;
Ming Leib5dd2f62014-12-31 13:22:57 +00001718 lo->tag_set.ops = &loop_mq_ops;
1719 lo->tag_set.nr_hw_queues = 1;
1720 lo->tag_set.queue_depth = 128;
1721 lo->tag_set.numa_node = NUMA_NO_NODE;
1722 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1723 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1724 lo->tag_set.driver_data = lo;
1725
1726 err = blk_mq_alloc_tag_set(&lo->tag_set);
1727 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001728 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07001729
Ming Leib5dd2f62014-12-31 13:22:57 +00001730 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1731 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1732 err = PTR_ERR(lo->lo_queue);
1733 goto out_cleanup_tags;
1734 }
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001735 lo->lo_queue->queuedata = lo;
1736
Omar Sandoval6c6b6f22017-08-24 00:03:42 -07001737 blk_queue_physical_block_size(lo->lo_queue, PAGE_SIZE);
1738
Ming Lei5b5e20f2015-08-17 10:31:47 +08001739 /*
1740 * It doesn't make sense to enable merge because the I/O
1741 * submitted to backing file is handled page by page.
1742 */
1743 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1744
Minfei Huang7a649732016-06-07 10:05:15 +08001745 err = -ENOMEM;
Laurent Vivier476a4812008-03-26 12:11:53 +01001746 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001747 if (!disk)
1748 goto out_free_queue;
1749
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001750 /*
1751 * Disable partition scanning by default. The in-kernel partition
1752 * scanning can be requested individually per-device during its
1753 * setup. Userspace can always add and remove partitions from all
1754 * devices. The needed partition minors are allocated from the
1755 * extended minor space, the main loop device numbers will continue
1756 * to match the loop minors, regardless of the number of partitions
1757 * used.
1758 *
1759 * If max_part is given, partition scanning is globally enabled for
1760 * all loop devices. The minors for the main loop devices will be
1761 * multiples of max_part.
1762 *
1763 * Note: Global-for-all-devices, set-only-at-init, read-only module
1764 * parameteters like 'max_loop' and 'max_part' make things needlessly
1765 * complicated, are too static, inflexible and may surprise
1766 * userspace tools. Parameters like this in general should be avoided.
1767 */
1768 if (!part_shift)
1769 disk->flags |= GENHD_FL_NO_PART_SCAN;
1770 disk->flags |= GENHD_FL_EXT_DEVT;
Ken Chen73285082007-05-08 00:28:20 -07001771 mutex_init(&lo->lo_ctl_mutex);
Ming Leif8933662015-05-06 12:26:23 +08001772 atomic_set(&lo->lo_refcnt, 0);
Ken Chen73285082007-05-08 00:28:20 -07001773 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07001774 spin_lock_init(&lo->lo_lock);
1775 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001776 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001777 disk->fops = &lo_fops;
1778 disk->private_data = lo;
1779 disk->queue = lo->lo_queue;
1780 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001781 add_disk(disk);
1782 *l = lo;
1783 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001784
1785out_free_queue:
1786 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00001787out_cleanup_tags:
1788 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001789out_free_idr:
1790 idr_remove(&loop_index_idr, i);
Ken Chen73285082007-05-08 00:28:20 -07001791out_free_dev:
1792 kfree(lo);
1793out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001794 return err;
Ken Chen73285082007-05-08 00:28:20 -07001795}
1796
Kay Sievers34dd82a2011-07-31 22:08:04 +02001797static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001798{
Ken Chen73285082007-05-08 00:28:20 -07001799 blk_cleanup_queue(lo->lo_queue);
NeilBrown6cd18e72015-04-27 14:12:22 +10001800 del_gendisk(lo->lo_disk);
Ming Leib5dd2f62014-12-31 13:22:57 +00001801 blk_mq_free_tag_set(&lo->tag_set);
Ken Chen73285082007-05-08 00:28:20 -07001802 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001803 kfree(lo);
1804}
1805
Kay Sievers770fe302011-07-31 22:08:04 +02001806static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001807{
Kay Sievers770fe302011-07-31 22:08:04 +02001808 struct loop_device *lo = ptr;
1809 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001810
Kay Sievers770fe302011-07-31 22:08:04 +02001811 if (lo->lo_state == Lo_unbound) {
1812 *l = lo;
1813 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001814 }
Kay Sievers770fe302011-07-31 22:08:04 +02001815 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001816}
1817
Kay Sievers34dd82a2011-07-31 22:08:04 +02001818static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001819{
Ken Chena47653f2007-06-08 13:46:44 -07001820 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001821 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001822
Kay Sievers770fe302011-07-31 22:08:04 +02001823 if (i < 0) {
1824 int err;
1825
1826 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1827 if (err == 1) {
1828 *l = lo;
1829 ret = lo->lo_number;
1830 }
1831 goto out;
1832 }
1833
1834 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001835 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07001836 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02001837 *l = lo;
1838 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07001839 }
Kay Sievers770fe302011-07-31 22:08:04 +02001840out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001841 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07001842}
1843
Ken Chen73285082007-05-08 00:28:20 -07001844static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1845{
Al Viro705962c2007-05-13 05:52:32 -04001846 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001847 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001848 int err;
Ken Chen73285082007-05-08 00:28:20 -07001849
Kay Sievers34dd82a2011-07-31 22:08:04 +02001850 mutex_lock(&loop_index_mutex);
1851 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1852 if (err < 0)
1853 err = loop_add(&lo, MINOR(dev) >> part_shift);
1854 if (err < 0)
Mikulas Patockaa207f592013-10-14 12:13:24 -04001855 kobj = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001856 else
1857 kobj = get_disk(lo->lo_disk);
1858 mutex_unlock(&loop_index_mutex);
Ken Chen73285082007-05-08 00:28:20 -07001859
1860 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001861 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001862}
1863
Kay Sievers770fe302011-07-31 22:08:04 +02001864static long loop_control_ioctl(struct file *file, unsigned int cmd,
1865 unsigned long parm)
1866{
1867 struct loop_device *lo;
1868 int ret = -ENOSYS;
1869
1870 mutex_lock(&loop_index_mutex);
1871 switch (cmd) {
1872 case LOOP_CTL_ADD:
1873 ret = loop_lookup(&lo, parm);
1874 if (ret >= 0) {
1875 ret = -EEXIST;
1876 break;
1877 }
1878 ret = loop_add(&lo, parm);
1879 break;
1880 case LOOP_CTL_REMOVE:
1881 ret = loop_lookup(&lo, parm);
1882 if (ret < 0)
1883 break;
1884 mutex_lock(&lo->lo_ctl_mutex);
1885 if (lo->lo_state != Lo_unbound) {
1886 ret = -EBUSY;
1887 mutex_unlock(&lo->lo_ctl_mutex);
1888 break;
1889 }
Ming Leif8933662015-05-06 12:26:23 +08001890 if (atomic_read(&lo->lo_refcnt) > 0) {
Kay Sievers770fe302011-07-31 22:08:04 +02001891 ret = -EBUSY;
1892 mutex_unlock(&lo->lo_ctl_mutex);
1893 break;
1894 }
1895 lo->lo_disk->private_data = NULL;
1896 mutex_unlock(&lo->lo_ctl_mutex);
1897 idr_remove(&loop_index_idr, lo->lo_number);
1898 loop_remove(lo);
1899 break;
1900 case LOOP_CTL_GET_FREE:
1901 ret = loop_lookup(&lo, -1);
1902 if (ret >= 0)
1903 break;
1904 ret = loop_add(&lo, -1);
1905 }
1906 mutex_unlock(&loop_index_mutex);
1907
1908 return ret;
1909}
1910
1911static const struct file_operations loop_ctl_fops = {
1912 .open = nonseekable_open,
1913 .unlocked_ioctl = loop_control_ioctl,
1914 .compat_ioctl = loop_control_ioctl,
1915 .owner = THIS_MODULE,
1916 .llseek = noop_llseek,
1917};
1918
1919static struct miscdevice loop_misc = {
1920 .minor = LOOP_CTRL_MINOR,
1921 .name = "loop-control",
1922 .fops = &loop_ctl_fops,
1923};
1924
1925MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1926MODULE_ALIAS("devname:loop-control");
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928static int __init loop_init(void)
1929{
Ken Chena47653f2007-06-08 13:46:44 -07001930 int i, nr;
1931 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001932 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02001933 int err;
Ken Chena47653f2007-06-08 13:46:44 -07001934
Laurent Vivier476a4812008-03-26 12:11:53 +01001935 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001936 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01001937 part_shift = fls(max_part);
1938
Namhyung Kimac04fee2011-05-27 07:59:25 +02001939 /*
1940 * Adjust max_part according to part_shift as it is exported
1941 * to user space so that user can decide correct minor number
1942 * if [s]he want to create more devices.
1943 *
1944 * Note that -1 is required because partition 0 is reserved
1945 * for the whole disk.
1946 */
1947 max_part = (1UL << part_shift) - 1;
1948 }
1949
Guo Chaob1a66502013-02-21 15:16:49 -08001950 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1951 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03001952 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08001953 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02001954
Guo Chaob1a66502013-02-21 15:16:49 -08001955 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1956 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03001957 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08001958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Kay Sieversd134b002011-07-31 22:08:04 +02001960 /*
1961 * If max_loop is specified, create that many devices upfront.
1962 * This also becomes a hard limit. If max_loop is not specified,
1963 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1964 * init time. Loop devices can be requested on-demand with the
1965 * /dev/loop-control interface, or be instantiated by accessing
1966 * a 'dead' device node.
1967 */
Ken Chen73285082007-05-08 00:28:20 -07001968 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001969 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001970 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07001971 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02001972 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001973 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 }
Ken Chena47653f2007-06-08 13:46:44 -07001975
Anton Volkova8c1d062017-08-07 15:37:50 +03001976 err = misc_register(&loop_misc);
1977 if (err < 0)
1978 goto err_out;
1979
1980
Guo Chaob1a66502013-02-21 15:16:49 -08001981 if (register_blkdev(LOOP_MAJOR, "loop")) {
1982 err = -EIO;
1983 goto misc_out;
1984 }
Ken Chena47653f2007-06-08 13:46:44 -07001985
Ken Chena47653f2007-06-08 13:46:44 -07001986 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1987 THIS_MODULE, loop_probe, NULL, NULL);
1988
Kay Sieversd134b002011-07-31 22:08:04 +02001989 /* pre-create number of devices given by config or max_loop */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001990 mutex_lock(&loop_index_mutex);
1991 for (i = 0; i < nr; i++)
1992 loop_add(&lo, i);
1993 mutex_unlock(&loop_index_mutex);
1994
Ken Chen73285082007-05-08 00:28:20 -07001995 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08001997
1998misc_out:
1999 misc_deregister(&loop_misc);
Anton Volkova8c1d062017-08-07 15:37:50 +03002000err_out:
Guo Chaob1a66502013-02-21 15:16:49 -08002001 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002002}
Ken Chena47653f2007-06-08 13:46:44 -07002003
Kay Sievers34dd82a2011-07-31 22:08:04 +02002004static int loop_exit_cb(int id, void *ptr, void *data)
2005{
2006 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07002007
Kay Sievers34dd82a2011-07-31 22:08:04 +02002008 loop_remove(lo);
2009 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010}
2011
Ken Chen73285082007-05-08 00:28:20 -07002012static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
Ken Chena47653f2007-06-08 13:46:44 -07002014 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Namhyung Kima1c15c52011-05-24 16:48:55 +02002016 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07002017
Kay Sievers34dd82a2011-07-31 22:08:04 +02002018 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002019 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07002020
Ken Chena47653f2007-06-08 13:46:44 -07002021 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07002022 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02002023
2024 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025}
2026
2027module_init(loop_init);
2028module_exit(loop_exit);
2029
2030#ifndef MODULE
2031static int __init max_loop_setup(char *str)
2032{
2033 max_loop = simple_strtol(str, NULL, 0);
2034 return 1;
2035}
2036
2037__setup("max_loop=", max_loop_setup);
2038#endif