blob: c9f2107f7095148dc5e4027d7cb8fdc4794aad01 [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
81#include <asm/uaccess.h>
82
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
189 * of requests in sane appplications should be PAGE_SIZE algined
190 */
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
224figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
225{
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 Hellwigaa4d8612015-04-07 18:23:29 +0200269 bw = vfs_iter_write(file, &i, ppos);
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);
345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
346 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);
386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
387 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
Ming Leibc07c102015-08-17 10:31:51 +0800448static inline void handle_partial_read(struct loop_cmd *cmd, long bytes)
449{
Mike Christiea8ebb052016-06-05 14:31:45 -0500450 if (bytes < 0 || op_is_write(req_op(cmd->rq)))
Ming Leibc07c102015-08-17 10:31:51 +0800451 return;
452
453 if (unlikely(bytes < blk_rq_bytes(cmd->rq))) {
454 struct bio *bio = cmd->rq->bio;
455
456 bio_advance(bio, bytes);
457 zero_fill_bio(bio);
458 }
459}
460
461static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
462{
463 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
464 struct request *rq = cmd->rq;
465
466 handle_partial_read(cmd, ret);
467
468 if (ret > 0)
469 ret = 0;
470 else if (ret < 0)
471 ret = -EIO;
472
Jens Axboe54ef2b92015-10-09 10:40:29 -0600473 blk_mq_complete_request(rq, ret);
Ming Leibc07c102015-08-17 10:31:51 +0800474}
475
476static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
477 loff_t pos, bool rw)
478{
479 struct iov_iter iter;
480 struct bio_vec *bvec;
481 struct bio *bio = cmd->rq->bio;
482 struct file *file = lo->lo_backing_file;
483 int ret;
484
485 /* nomerge for loop request queue */
486 WARN_ON(cmd->rq->bio != cmd->rq->biotail);
487
488 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
489 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
490 bio_segments(bio), blk_rq_bytes(cmd->rq));
Ming Leia7297a62016-04-15 18:51:28 +0800491 /*
492 * This bio may be started from the middle of the 'bvec'
493 * because of bio splitting, so offset from the bvec must
494 * be passed to iov iterator
495 */
496 iter.iov_offset = bio->bi_iter.bi_bvec_done;
Ming Leibc07c102015-08-17 10:31:51 +0800497
498 cmd->iocb.ki_pos = pos;
499 cmd->iocb.ki_filp = file;
500 cmd->iocb.ki_complete = lo_rw_aio_complete;
501 cmd->iocb.ki_flags = IOCB_DIRECT;
502
503 if (rw == WRITE)
504 ret = file->f_op->write_iter(&cmd->iocb, &iter);
505 else
506 ret = file->f_op->read_iter(&cmd->iocb, &iter);
507
508 if (ret != -EIOCBQUEUED)
509 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
510 return 0;
511}
512
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200513static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800514{
515 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200516 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800517
518 /*
519 * lo_write_simple and lo_read_simple should have been covered
520 * by io submit style function like lo_rw_aio(), one blocker
521 * is that lo_read_simple() need to call flush_dcache_page after
522 * the page is written from kernel, and it isn't easy to handle
523 * this in io submit style function which submits all segments
524 * of the req at one time. And direct read IO doesn't need to
525 * run flush_dcache_page().
526 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200527 switch (req_op(rq)) {
528 case REQ_OP_FLUSH:
529 return lo_req_flush(lo, rq);
530 case REQ_OP_DISCARD:
531 return lo_discard(lo, rq, pos);
532 case REQ_OP_WRITE:
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200533 if (lo->transfer)
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200534 return lo_write_transfer(lo, rq, pos);
535 else if (cmd->use_aio)
536 return lo_rw_aio(lo, cmd, pos, WRITE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200537 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200538 return lo_write_simple(lo, rq, pos);
539 case REQ_OP_READ:
540 if (lo->transfer)
541 return lo_read_transfer(lo, rq, pos);
542 else if (cmd->use_aio)
543 return lo_rw_aio(lo, cmd, pos, READ);
544 else
545 return lo_read_simple(lo, rq, pos);
546 default:
547 WARN_ON_ONCE(1);
548 return -EIO;
549 break;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551}
552
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553struct switch_request {
554 struct file *file;
555 struct completion wait;
556};
557
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800558static inline void loop_update_dio(struct loop_device *lo)
559{
560 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
561 lo->use_dio);
562}
563
Milan Broz14f27932008-12-12 14:48:27 +0100564/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 * Do the actual switch; called from the BIO completion routine
566 */
567static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
568{
569 struct file *file = p->file;
570 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100571 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Milan Broz14f27932008-12-12 14:48:27 +0100573 /* if no new file, only flush of queued bios requested */
574 if (!file)
Ming Leib5dd2f62014-12-31 13:22:57 +0000575 return;
Milan Broz14f27932008-12-12 14:48:27 +0100576
577 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
579 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700580 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
581 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 lo->old_gfp_mask = mapping_gfp_mask(mapping);
583 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800584 loop_update_dio(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
Ming Leib5dd2f62014-12-31 13:22:57 +0000587/*
588 * loop_switch performs the hard work of switching a backing store.
589 * First it needs to flush existing IO, it does this by sending a magic
590 * BIO down the pipe. The completion of this BIO does the actual switch.
591 */
592static int loop_switch(struct loop_device *lo, struct file *file)
593{
594 struct switch_request w;
595
596 w.file = file;
597
598 /* freeze queue and wait for completion of scheduled requests */
599 blk_mq_freeze_queue(lo->lo_queue);
600
601 /* do the switch action */
602 do_loop_switch(lo, &w);
603
604 /* unfreeze */
605 blk_mq_unfreeze_queue(lo->lo_queue);
606
607 return 0;
608}
609
610/*
611 * Helper to flush the IOs in loop, but keeping loop thread running
612 */
613static int loop_flush(struct loop_device *lo)
614{
615 return loop_switch(lo, NULL);
616}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Ming Lei06f0e9e2015-05-06 12:26:24 +0800618static void loop_reread_partitions(struct loop_device *lo,
619 struct block_device *bdev)
620{
621 int rc;
622
623 /*
624 * bd_mutex has been held already in release path, so don't
625 * acquire it if this function is called in such case.
626 *
627 * If the reread partition isn't from release path, lo_refcnt
628 * must be at least one and it can only become zero when the
629 * current holder is released.
630 */
631 if (!atomic_read(&lo->lo_refcnt))
632 rc = __blkdev_reread_part(bdev);
633 else
634 rc = blkdev_reread_part(bdev);
635 if (rc)
636 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
637 __func__, lo->lo_number, lo->lo_file_name, rc);
638}
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640/*
641 * loop_change_fd switched the backing store of a loopback device to
642 * a new file. This is useful for operating system installers to free up
643 * the original file and in High Availability environments to switch to
644 * an alternative location for the content in case of server meltdown.
645 * This can only work if the loop device is used read-only, and if the
646 * new backing store is the same size and type as the old backing store.
647 */
Al Virobb214882008-03-02 09:29:48 -0500648static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
649 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 struct file *file, *old_file;
652 struct inode *inode;
653 int error;
654
655 error = -ENXIO;
656 if (lo->lo_state != Lo_bound)
657 goto out;
658
659 /* the loop device has to be read-only */
660 error = -EINVAL;
661 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
662 goto out;
663
664 error = -EBADF;
665 file = fget(arg);
666 if (!file)
667 goto out;
668
669 inode = file->f_mapping->host;
670 old_file = lo->lo_backing_file;
671
672 error = -EINVAL;
673
674 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
675 goto out_putf;
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 /* size of the new backing store needs to be the same */
678 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
679 goto out_putf;
680
681 /* and ... switch */
682 error = loop_switch(lo, file);
683 if (error)
684 goto out_putf;
685
686 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200687 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800688 loop_reread_partitions(lo, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return 0;
690
691 out_putf:
692 fput(file);
693 out:
694 return error;
695}
696
697static inline int is_loop_device(struct file *file)
698{
699 struct inode *i = file->f_mapping->host;
700
701 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
702}
703
Milan Brozee862732010-08-23 15:16:00 +0200704/* loop sysfs attributes */
705
706static ssize_t loop_attr_show(struct device *dev, char *page,
707 ssize_t (*callback)(struct loop_device *, char *))
708{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200709 struct gendisk *disk = dev_to_disk(dev);
710 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200711
Kay Sievers34dd82a2011-07-31 22:08:04 +0200712 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200713}
714
715#define LOOP_ATTR_RO(_name) \
716static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
717static ssize_t loop_attr_do_show_##_name(struct device *d, \
718 struct device_attribute *attr, char *b) \
719{ \
720 return loop_attr_show(d, b, loop_attr_##_name##_show); \
721} \
722static struct device_attribute loop_attr_##_name = \
723 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
724
725static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
726{
727 ssize_t ret;
728 char *p = NULL;
729
Kay Sievers05eb0f22011-07-31 22:21:35 +0200730 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200731 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200732 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200733 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200734
735 if (IS_ERR_OR_NULL(p))
736 ret = PTR_ERR(p);
737 else {
738 ret = strlen(p);
739 memmove(buf, p, ret);
740 buf[ret++] = '\n';
741 buf[ret] = 0;
742 }
743
744 return ret;
745}
746
747static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
748{
749 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
750}
751
752static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
753{
754 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
755}
756
757static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
758{
759 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
760
761 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
762}
763
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200764static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
765{
766 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
767
768 return sprintf(buf, "%s\n", partscan ? "1" : "0");
769}
770
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800771static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
772{
773 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
774
775 return sprintf(buf, "%s\n", dio ? "1" : "0");
776}
777
Milan Brozee862732010-08-23 15:16:00 +0200778LOOP_ATTR_RO(backing_file);
779LOOP_ATTR_RO(offset);
780LOOP_ATTR_RO(sizelimit);
781LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200782LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800783LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200784
785static struct attribute *loop_attrs[] = {
786 &loop_attr_backing_file.attr,
787 &loop_attr_offset.attr,
788 &loop_attr_sizelimit.attr,
789 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200790 &loop_attr_partscan.attr,
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800791 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200792 NULL,
793};
794
795static struct attribute_group loop_attribute_group = {
796 .name = "loop",
797 .attrs= loop_attrs,
798};
799
800static int loop_sysfs_init(struct loop_device *lo)
801{
802 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
803 &loop_attribute_group);
804}
805
806static void loop_sysfs_exit(struct loop_device *lo)
807{
808 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
809 &loop_attribute_group);
810}
811
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200812static void loop_config_discard(struct loop_device *lo)
813{
814 struct file *file = lo->lo_backing_file;
815 struct inode *inode = file->f_mapping->host;
816 struct request_queue *q = lo->lo_queue;
817
818 /*
819 * We use punch hole to reclaim the free space used by the
Olaf Hering12a64d22014-01-21 14:39:24 -0800820 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200821 * encryption is enabled, because it may give an attacker
822 * useful information.
823 */
824 if ((!file->f_op->fallocate) ||
825 lo->lo_encrypt_key_size) {
826 q->limits.discard_granularity = 0;
827 q->limits.discard_alignment = 0;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600828 blk_queue_max_discard_sectors(q, 0);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200829 q->limits.discard_zeroes_data = 0;
830 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
831 return;
832 }
833
834 q->limits.discard_granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaf3c02011-12-02 14:47:03 +0100835 q->limits.discard_alignment = 0;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600836 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200837 q->limits.discard_zeroes_data = 1;
838 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
839}
840
Ming Leie03a3d72015-08-17 10:31:48 +0800841static void loop_unprepare_queue(struct loop_device *lo)
842{
843 flush_kthread_worker(&lo->worker);
844 kthread_stop(lo->worker_task);
845}
846
847static int loop_prepare_queue(struct loop_device *lo)
848{
849 init_kthread_worker(&lo->worker);
850 lo->worker_task = kthread_run(kthread_worker_fn,
851 &lo->worker, "loop%d", lo->lo_number);
852 if (IS_ERR(lo->worker_task))
853 return -ENOMEM;
854 set_user_nice(lo->worker_task, MIN_NICE);
855 return 0;
856}
857
Al Virobb214882008-03-02 09:29:48 -0500858static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 struct block_device *bdev, unsigned int arg)
860{
861 struct file *file, *f;
862 struct inode *inode;
863 struct address_space *mapping;
864 unsigned lo_blocksize;
865 int lo_flags = 0;
866 int error;
867 loff_t size;
868
869 /* This is safe, since we have a reference from open(). */
870 __module_get(THIS_MODULE);
871
872 error = -EBADF;
873 file = fget(arg);
874 if (!file)
875 goto out;
876
877 error = -EBUSY;
878 if (lo->lo_state != Lo_unbound)
879 goto out_putf;
880
881 /* Avoid recursion */
882 f = file;
883 while (is_loop_device(f)) {
884 struct loop_device *l;
885
Al Virobb214882008-03-02 09:29:48 -0500886 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 goto out_putf;
888
889 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
890 if (l->lo_state == Lo_unbound) {
891 error = -EINVAL;
892 goto out_putf;
893 }
894 f = l->lo_backing_file;
895 }
896
897 mapping = file->f_mapping;
898 inode = mapping->host;
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 error = -EINVAL;
Christoph Hellwig456be142011-10-17 12:57:20 +0200901 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Christoph Hellwig456be142011-10-17 12:57:20 +0200904 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
Al Viro283e7e52015-04-03 15:21:59 -0400905 !file->f_op->write_iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 lo_flags |= LO_FLAGS_READ_ONLY;
907
Christoph Hellwig456be142011-10-17 12:57:20 +0200908 lo_blocksize = S_ISBLK(inode->i_mode) ?
909 inode->i_bdev->bd_block_size : PAGE_SIZE;
910
911 error = -EFBIG;
912 size = get_loop_size(lo, file);
913 if ((loff_t)(sector_t)size != size)
914 goto out_putf;
Ming Leie03a3d72015-08-17 10:31:48 +0800915 error = loop_prepare_queue(lo);
916 if (error)
Ming Leif4aa4c72015-05-05 19:49:54 +0800917 goto out_putf;
Christoph Hellwig456be142011-10-17 12:57:20 +0200918
919 error = 0;
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
922
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800923 lo->use_dio = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 lo->lo_blocksize = lo_blocksize;
925 lo->lo_device = bdev;
926 lo->lo_flags = lo_flags;
927 lo->lo_backing_file = file;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200928 lo->transfer = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 lo->ioctl = NULL;
930 lo->lo_sizelimit = 0;
931 lo->old_gfp_mask = mapping_gfp_mask(mapping);
932 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
933
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100934 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Jens Axboe21d07272016-03-30 10:09:35 -0600935 blk_queue_write_cache(lo->lo_queue, true, false);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100936
Ming Lei2e5ab5f2015-08-17 10:31:49 +0800937 loop_update_dio(lo);
Ken Chen73285082007-05-08 00:28:20 -0700938 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200940 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200941 /* let user-space know about the new size */
942 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 set_blocksize(bdev, lo_blocksize);
945
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700946 lo->lo_state = Lo_bound;
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200947 if (part_shift)
948 lo->lo_flags |= LO_FLAGS_PARTSCAN;
949 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800950 loop_reread_partitions(lo, bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700951
952 /* Grab the block_device to prevent its destruction after we
953 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
954 */
955 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return 0;
957
958 out_putf:
959 fput(file);
960 out:
961 /* This is safe: open() is still holding a reference. */
962 module_put(THIS_MODULE);
963 return error;
964}
965
966static int
967loop_release_xfer(struct loop_device *lo)
968{
969 int err = 0;
970 struct loop_func_table *xfer = lo->lo_encryption;
971
972 if (xfer) {
973 if (xfer->release)
974 err = xfer->release(lo);
975 lo->transfer = NULL;
976 lo->lo_encryption = NULL;
977 module_put(xfer->owner);
978 }
979 return err;
980}
981
982static int
983loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
984 const struct loop_info64 *i)
985{
986 int err = 0;
987
988 if (xfer) {
989 struct module *owner = xfer->owner;
990
991 if (!try_module_get(owner))
992 return -EINVAL;
993 if (xfer->init)
994 err = xfer->init(lo, i);
995 if (err)
996 module_put(owner);
997 else
998 lo->lo_encryption = xfer;
999 }
1000 return err;
1001}
1002
Ayan George4c823cc2011-09-21 10:02:13 +02001003static int loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -04001006 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +02001007 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 if (lo->lo_state != Lo_bound)
1010 return -ENXIO;
1011
Dave Chinnera1ecac32012-09-28 10:42:23 +02001012 /*
1013 * If we've explicitly asked to tear down the loop device,
1014 * and it has an elevated reference count, set it for auto-teardown when
1015 * the last reference goes away. This stops $!~#$@ udev from
1016 * preventing teardown because it decided that it needs to run blkid on
1017 * the loopback device whenever they appear. xfstests is notorious for
1018 * failing tests because blkid via udev races with a losetup
1019 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1020 * command to fail with EBUSY.
1021 */
Ming Leif8933662015-05-06 12:26:23 +08001022 if (atomic_read(&lo->lo_refcnt) > 1) {
Dave Chinnera1ecac32012-09-28 10:42:23 +02001023 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1024 mutex_unlock(&lo->lo_ctl_mutex);
1025 return 0;
1026 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 if (filp == NULL)
1029 return -EINVAL;
1030
Ming Leif8933662015-05-06 12:26:23 +08001031 /* freeze request queue during the transition */
1032 blk_mq_freeze_queue(lo->lo_queue);
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 spin_lock_irq(&lo->lo_lock);
1035 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001037 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039 loop_release_xfer(lo);
1040 lo->transfer = NULL;
1041 lo->ioctl = NULL;
1042 lo->lo_device = NULL;
1043 lo->lo_encryption = NULL;
1044 lo->lo_offset = 0;
1045 lo->lo_sizelimit = 0;
1046 lo->lo_encrypt_key_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1048 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1049 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001050 if (bdev) {
1051 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -05001052 invalidate_bdev(bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001053 }
Ken Chen73285082007-05-08 00:28:20 -07001054 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001055 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001056 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001057 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001058 /* let user-space know about this change */
1059 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 mapping_set_gfp_mask(filp->f_mapping, gfp);
1062 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 /* This is safe: open() is still holding a reference. */
1064 module_put(THIS_MODULE);
Ming Leif8933662015-05-06 12:26:23 +08001065 blk_mq_unfreeze_queue(lo->lo_queue);
1066
Jens Axboec2fccc12013-04-08 10:12:11 +02001067 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
Ming Lei06f0e9e2015-05-06 12:26:24 +08001068 loop_reread_partitions(lo, bdev);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001069 lo->lo_flags = 0;
1070 if (!part_shift)
1071 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Ming Leie03a3d72015-08-17 10:31:48 +08001072 loop_unprepare_queue(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001073 mutex_unlock(&lo->lo_ctl_mutex);
1074 /*
1075 * Need not hold lo_ctl_mutex to fput backing file.
1076 * Calling fput holding lo_ctl_mutex triggers a circular
1077 * lock dependency possibility warning as fput can take
1078 * bd_mutex which is usually taken before lo_ctl_mutex.
1079 */
1080 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return 0;
1082}
1083
1084static int
1085loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1086{
1087 int err;
1088 struct loop_func_table *xfer;
Eric W. Biedermane4849732012-02-11 11:23:51 -08001089 kuid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
David Howellsb0fafa82008-11-14 10:38:41 +11001091 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001092 !uid_eq(lo->lo_key_owner, uid) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 !capable(CAP_SYS_ADMIN))
1094 return -EPERM;
1095 if (lo->lo_state != Lo_bound)
1096 return -ENXIO;
1097 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1098 return -EINVAL;
1099
1100 err = loop_release_xfer(lo);
1101 if (err)
1102 return err;
1103
1104 if (info->lo_encrypt_type) {
1105 unsigned int type = info->lo_encrypt_type;
1106
1107 if (type >= MAX_LO_CRYPT)
1108 return -EINVAL;
1109 xfer = xfer_funcs[type];
1110 if (xfer == NULL)
1111 return -EINVAL;
1112 } else
1113 xfer = NULL;
1114
1115 err = loop_init_xfer(lo, xfer, info);
1116 if (err)
1117 return err;
1118
1119 if (lo->lo_offset != info->lo_offset ||
Guo Chao7b0576a2013-02-21 15:16:47 -08001120 lo->lo_sizelimit != info->lo_sizelimit)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001121 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 return -EFBIG;
Guo Chao541c7422013-02-21 15:16:46 -08001123
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001124 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1127 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1128 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1129 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1130
1131 if (!xfer)
1132 xfer = &none_funcs;
1133 lo->transfer = xfer->transfer;
1134 lo->ioctl = xfer->ioctl;
1135
David Woodhouse96c58652008-02-06 01:36:27 -08001136 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1137 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1138 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1139
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001140 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1141 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1142 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1143 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
Ming Lei06f0e9e2015-05-06 12:26:24 +08001144 loop_reread_partitions(lo, lo->lo_device);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001145 }
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1148 lo->lo_init[0] = info->lo_init[0];
1149 lo->lo_init[1] = info->lo_init[1];
1150 if (info->lo_encrypt_key_size) {
1151 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1152 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001153 lo->lo_key_owner = uid;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +02001154 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Ming Lei2e5ab5f2015-08-17 10:31:49 +08001156 /* update dio if lo_offset or transfer is changed */
1157 __loop_update_dio(lo, lo->use_dio);
1158
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 return 0;
1160}
1161
1162static int
1163loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1164{
1165 struct file *file = lo->lo_backing_file;
1166 struct kstat stat;
1167 int error;
1168
1169 if (lo->lo_state != Lo_bound)
1170 return -ENXIO;
Al Viro3dadecc2013-01-24 02:18:08 -05001171 error = vfs_getattr(&file->f_path, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (error)
1173 return error;
1174 memset(info, 0, sizeof(*info));
1175 info->lo_number = lo->lo_number;
1176 info->lo_device = huge_encode_dev(stat.dev);
1177 info->lo_inode = stat.ino;
1178 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1179 info->lo_offset = lo->lo_offset;
1180 info->lo_sizelimit = lo->lo_sizelimit;
1181 info->lo_flags = lo->lo_flags;
1182 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1183 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1184 info->lo_encrypt_type =
1185 lo->lo_encryption ? lo->lo_encryption->number : 0;
1186 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1187 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1188 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1189 lo->lo_encrypt_key_size);
1190 }
1191 return 0;
1192}
1193
1194static void
1195loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1196{
1197 memset(info64, 0, sizeof(*info64));
1198 info64->lo_number = info->lo_number;
1199 info64->lo_device = info->lo_device;
1200 info64->lo_inode = info->lo_inode;
1201 info64->lo_rdevice = info->lo_rdevice;
1202 info64->lo_offset = info->lo_offset;
1203 info64->lo_sizelimit = 0;
1204 info64->lo_encrypt_type = info->lo_encrypt_type;
1205 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1206 info64->lo_flags = info->lo_flags;
1207 info64->lo_init[0] = info->lo_init[0];
1208 info64->lo_init[1] = info->lo_init[1];
1209 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1210 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1211 else
1212 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1213 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1214}
1215
1216static int
1217loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1218{
1219 memset(info, 0, sizeof(*info));
1220 info->lo_number = info64->lo_number;
1221 info->lo_device = info64->lo_device;
1222 info->lo_inode = info64->lo_inode;
1223 info->lo_rdevice = info64->lo_rdevice;
1224 info->lo_offset = info64->lo_offset;
1225 info->lo_encrypt_type = info64->lo_encrypt_type;
1226 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1227 info->lo_flags = info64->lo_flags;
1228 info->lo_init[0] = info64->lo_init[0];
1229 info->lo_init[1] = info64->lo_init[1];
1230 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1231 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1232 else
1233 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1234 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1235
1236 /* error in case values were truncated */
1237 if (info->lo_device != info64->lo_device ||
1238 info->lo_rdevice != info64->lo_rdevice ||
1239 info->lo_inode != info64->lo_inode ||
1240 info->lo_offset != info64->lo_offset)
1241 return -EOVERFLOW;
1242
1243 return 0;
1244}
1245
1246static int
1247loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1248{
1249 struct loop_info info;
1250 struct loop_info64 info64;
1251
1252 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1253 return -EFAULT;
1254 loop_info64_from_old(&info, &info64);
1255 return loop_set_status(lo, &info64);
1256}
1257
1258static int
1259loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1260{
1261 struct loop_info64 info64;
1262
1263 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1264 return -EFAULT;
1265 return loop_set_status(lo, &info64);
1266}
1267
1268static int
1269loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1270 struct loop_info info;
1271 struct loop_info64 info64;
1272 int err = 0;
1273
1274 if (!arg)
1275 err = -EINVAL;
1276 if (!err)
1277 err = loop_get_status(lo, &info64);
1278 if (!err)
1279 err = loop_info64_to_old(&info64, &info);
1280 if (!err && copy_to_user(arg, &info, sizeof(info)))
1281 err = -EFAULT;
1282
1283 return err;
1284}
1285
1286static int
1287loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1288 struct loop_info64 info64;
1289 int err = 0;
1290
1291 if (!arg)
1292 err = -EINVAL;
1293 if (!err)
1294 err = loop_get_status(lo, &info64);
1295 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1296 err = -EFAULT;
1297
1298 return err;
1299}
1300
J. R. Okajima53d66602009-03-31 15:23:43 -07001301static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1302{
J. R. Okajima53d66602009-03-31 15:23:43 -07001303 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001304 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001305
Guo Chao7b0576a2013-02-21 15:16:47 -08001306 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001307}
1308
Ming Leiab1cb272015-08-17 10:31:50 +08001309static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1310{
1311 int error = -ENXIO;
1312 if (lo->lo_state != Lo_bound)
1313 goto out;
1314
1315 __loop_update_dio(lo, !!arg);
1316 if (lo->use_dio == !!arg)
1317 return 0;
1318 error = -EINVAL;
1319 out:
1320 return error;
1321}
1322
Al Virobb214882008-03-02 09:29:48 -05001323static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 unsigned int cmd, unsigned long arg)
1325{
Al Virobb214882008-03-02 09:29:48 -05001326 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 int err;
1328
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001329 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 switch (cmd) {
1331 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001332 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 break;
1334 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001335 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 break;
1337 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001338 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Ayan George4c823cc2011-09-21 10:02:13 +02001339 err = loop_clr_fd(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001340 if (!err)
1341 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 break;
1343 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001344 err = -EPERM;
1345 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1346 err = loop_set_status_old(lo,
1347 (struct loop_info __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 break;
1349 case LOOP_GET_STATUS:
1350 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1351 break;
1352 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001353 err = -EPERM;
1354 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1355 err = loop_set_status64(lo,
1356 (struct loop_info64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 break;
1358 case LOOP_GET_STATUS64:
1359 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1360 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001361 case LOOP_SET_CAPACITY:
1362 err = -EPERM;
1363 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1364 err = loop_set_capacity(lo, bdev);
1365 break;
Ming Leiab1cb272015-08-17 10:31:50 +08001366 case LOOP_SET_DIRECT_IO:
1367 err = -EPERM;
1368 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1369 err = loop_set_dio(lo, arg);
1370 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 default:
1372 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1373 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001374 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001375
1376out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return err;
1378}
1379
David Howells863d5b822006-08-29 19:06:14 +01001380#ifdef CONFIG_COMPAT
1381struct compat_loop_info {
1382 compat_int_t lo_number; /* ioctl r/o */
1383 compat_dev_t lo_device; /* ioctl r/o */
1384 compat_ulong_t lo_inode; /* ioctl r/o */
1385 compat_dev_t lo_rdevice; /* ioctl r/o */
1386 compat_int_t lo_offset;
1387 compat_int_t lo_encrypt_type;
1388 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1389 compat_int_t lo_flags; /* ioctl r/o */
1390 char lo_name[LO_NAME_SIZE];
1391 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1392 compat_ulong_t lo_init[2];
1393 char reserved[4];
1394};
1395
1396/*
1397 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1398 * - noinlined to reduce stack space usage in main part of driver
1399 */
1400static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001401loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001402 struct loop_info64 *info64)
1403{
1404 struct compat_loop_info info;
1405
1406 if (copy_from_user(&info, arg, sizeof(info)))
1407 return -EFAULT;
1408
1409 memset(info64, 0, sizeof(*info64));
1410 info64->lo_number = info.lo_number;
1411 info64->lo_device = info.lo_device;
1412 info64->lo_inode = info.lo_inode;
1413 info64->lo_rdevice = info.lo_rdevice;
1414 info64->lo_offset = info.lo_offset;
1415 info64->lo_sizelimit = 0;
1416 info64->lo_encrypt_type = info.lo_encrypt_type;
1417 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1418 info64->lo_flags = info.lo_flags;
1419 info64->lo_init[0] = info.lo_init[0];
1420 info64->lo_init[1] = info.lo_init[1];
1421 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1422 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1423 else
1424 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1425 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1426 return 0;
1427}
1428
1429/*
1430 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1431 * - noinlined to reduce stack space usage in main part of driver
1432 */
1433static noinline int
1434loop_info64_to_compat(const struct loop_info64 *info64,
1435 struct compat_loop_info __user *arg)
1436{
1437 struct compat_loop_info info;
1438
1439 memset(&info, 0, sizeof(info));
1440 info.lo_number = info64->lo_number;
1441 info.lo_device = info64->lo_device;
1442 info.lo_inode = info64->lo_inode;
1443 info.lo_rdevice = info64->lo_rdevice;
1444 info.lo_offset = info64->lo_offset;
1445 info.lo_encrypt_type = info64->lo_encrypt_type;
1446 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1447 info.lo_flags = info64->lo_flags;
1448 info.lo_init[0] = info64->lo_init[0];
1449 info.lo_init[1] = info64->lo_init[1];
1450 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1451 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1452 else
1453 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1454 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1455
1456 /* error in case values were truncated */
1457 if (info.lo_device != info64->lo_device ||
1458 info.lo_rdevice != info64->lo_rdevice ||
1459 info.lo_inode != info64->lo_inode ||
1460 info.lo_offset != info64->lo_offset ||
1461 info.lo_init[0] != info64->lo_init[0] ||
1462 info.lo_init[1] != info64->lo_init[1])
1463 return -EOVERFLOW;
1464
1465 if (copy_to_user(arg, &info, sizeof(info)))
1466 return -EFAULT;
1467 return 0;
1468}
1469
1470static int
1471loop_set_status_compat(struct loop_device *lo,
1472 const struct compat_loop_info __user *arg)
1473{
1474 struct loop_info64 info64;
1475 int ret;
1476
1477 ret = loop_info64_from_compat(arg, &info64);
1478 if (ret < 0)
1479 return ret;
1480 return loop_set_status(lo, &info64);
1481}
1482
1483static int
1484loop_get_status_compat(struct loop_device *lo,
1485 struct compat_loop_info __user *arg)
1486{
1487 struct loop_info64 info64;
1488 int err = 0;
1489
1490 if (!arg)
1491 err = -EINVAL;
1492 if (!err)
1493 err = loop_get_status(lo, &info64);
1494 if (!err)
1495 err = loop_info64_to_compat(&info64, arg);
1496 return err;
1497}
1498
Al Virobb214882008-03-02 09:29:48 -05001499static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1500 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001501{
Al Virobb214882008-03-02 09:29:48 -05001502 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001503 int err;
1504
David Howells863d5b822006-08-29 19:06:14 +01001505 switch(cmd) {
1506 case LOOP_SET_STATUS:
1507 mutex_lock(&lo->lo_ctl_mutex);
1508 err = loop_set_status_compat(
1509 lo, (const struct compat_loop_info __user *) arg);
1510 mutex_unlock(&lo->lo_ctl_mutex);
1511 break;
1512 case LOOP_GET_STATUS:
1513 mutex_lock(&lo->lo_ctl_mutex);
1514 err = loop_get_status_compat(
1515 lo, (struct compat_loop_info __user *) arg);
1516 mutex_unlock(&lo->lo_ctl_mutex);
1517 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001518 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001519 case LOOP_CLR_FD:
1520 case LOOP_GET_STATUS64:
1521 case LOOP_SET_STATUS64:
1522 arg = (unsigned long) compat_ptr(arg);
1523 case LOOP_SET_FD:
1524 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001525 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001526 break;
1527 default:
1528 err = -ENOIOCTLCMD;
1529 break;
1530 }
David Howells863d5b822006-08-29 19:06:14 +01001531 return err;
1532}
1533#endif
1534
Al Virobb214882008-03-02 09:29:48 -05001535static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
Kay Sievers770fe302011-07-31 22:08:04 +02001537 struct loop_device *lo;
1538 int err = 0;
1539
1540 mutex_lock(&loop_index_mutex);
1541 lo = bdev->bd_disk->private_data;
1542 if (!lo) {
1543 err = -ENXIO;
1544 goto out;
1545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Ming Leif8933662015-05-06 12:26:23 +08001547 atomic_inc(&lo->lo_refcnt);
Kay Sievers770fe302011-07-31 22:08:04 +02001548out:
1549 mutex_unlock(&loop_index_mutex);
1550 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
Al Virodb2a1442013-05-05 21:52:57 -04001553static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Al Virobb214882008-03-02 09:29:48 -05001555 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001556 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Ming Leif8933662015-05-06 12:26:23 +08001558 if (atomic_dec_return(&lo->lo_refcnt))
1559 return;
1560
Ingo Molnarf85221d2006-03-23 03:00:38 -08001561 mutex_lock(&lo->lo_ctl_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001562 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1563 /*
1564 * In autoclear mode, stop the loop thread
1565 * and remove configuration after last close.
1566 */
Ayan George4c823cc2011-09-21 10:02:13 +02001567 err = loop_clr_fd(lo);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001568 if (!err)
Al Virodb2a1442013-05-05 21:52:57 -04001569 return;
Milan Broz14f27932008-12-12 14:48:27 +01001570 } else {
1571 /*
1572 * Otherwise keep thread (if running) and config,
1573 * but flush possible ongoing bios in thread.
1574 */
1575 loop_flush(lo);
1576 }
David Woodhouse96c58652008-02-06 01:36:27 -08001577
Ingo Molnarf85221d2006-03-23 03:00:38 -08001578 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
1580
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001581static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001583 .open = lo_open,
1584 .release = lo_release,
1585 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001586#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001587 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001588#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589};
1590
1591/*
1592 * And now the modules code and kernel interface.
1593 */
Ken Chen73285082007-05-08 00:28:20 -07001594static int max_loop;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001595module_param(max_loop, int, S_IRUGO);
Ken Chena47653f2007-06-08 13:46:44 -07001596MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Namhyung Kimac04fee2011-05-27 07:59:25 +02001597module_param(max_part, int, S_IRUGO);
Laurent Vivier476a4812008-03-26 12:11:53 +01001598MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599MODULE_LICENSE("GPL");
1600MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1601
1602int loop_register_transfer(struct loop_func_table *funcs)
1603{
1604 unsigned int n = funcs->number;
1605
1606 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1607 return -EINVAL;
1608 xfer_funcs[n] = funcs;
1609 return 0;
1610}
1611
Kay Sievers34dd82a2011-07-31 22:08:04 +02001612static int unregister_transfer_cb(int id, void *ptr, void *data)
1613{
1614 struct loop_device *lo = ptr;
1615 struct loop_func_table *xfer = data;
1616
1617 mutex_lock(&lo->lo_ctl_mutex);
1618 if (lo->lo_encryption == xfer)
1619 loop_release_xfer(lo);
1620 mutex_unlock(&lo->lo_ctl_mutex);
1621 return 0;
1622}
1623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624int loop_unregister_transfer(int number)
1625{
1626 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 struct loop_func_table *xfer;
1628
1629 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1630 return -EINVAL;
1631
1632 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001633 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 return 0;
1635}
1636
1637EXPORT_SYMBOL(loop_register_transfer);
1638EXPORT_SYMBOL(loop_unregister_transfer);
1639
Ming Leib5dd2f62014-12-31 13:22:57 +00001640static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1641 const struct blk_mq_queue_data *bd)
1642{
1643 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
Ming Leif4aa4c72015-05-05 19:49:54 +08001644 struct loop_device *lo = cmd->rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00001645
1646 blk_mq_start_request(bd->rq);
1647
Ming Leif4aa4c72015-05-05 19:49:54 +08001648 if (lo->lo_state != Lo_bound)
1649 return -EIO;
1650
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001651 switch (req_op(cmd->rq)) {
1652 case REQ_OP_FLUSH:
1653 case REQ_OP_DISCARD:
Ming Leibc07c102015-08-17 10:31:51 +08001654 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001655 break;
1656 default:
1657 cmd->use_aio = lo->use_dio;
1658 break;
1659 }
Ming Leibc07c102015-08-17 10:31:51 +08001660
Ming Leie03a3d72015-08-17 10:31:48 +08001661 queue_kthread_work(&lo->worker, &cmd->work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001662
1663 return BLK_MQ_RQ_QUEUE_OK;
1664}
1665
1666static void loop_handle_cmd(struct loop_cmd *cmd)
1667{
Mike Christiea8ebb052016-06-05 14:31:45 -05001668 const bool write = op_is_write(req_op(cmd->rq));
Ming Leib5dd2f62014-12-31 13:22:57 +00001669 struct loop_device *lo = cmd->rq->q->queuedata;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001670 int ret = 0;
Ming Leib5dd2f62014-12-31 13:22:57 +00001671
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001672 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1673 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00001674 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001675 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001676
Ming Lei30112012014-12-31 13:22:58 +00001677 ret = do_req_filebacked(lo, cmd->rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001678 failed:
Ming Leibc07c102015-08-17 10:31:51 +08001679 /* complete non-aio request */
1680 if (!cmd->use_aio || ret)
Jens Axboe54ef2b92015-10-09 10:40:29 -06001681 blk_mq_complete_request(cmd->rq, ret ? -EIO : 0);
Ming Leib5dd2f62014-12-31 13:22:57 +00001682}
1683
Ming Leie03a3d72015-08-17 10:31:48 +08001684static void loop_queue_work(struct kthread_work *work)
Ming Leib5dd2f62014-12-31 13:22:57 +00001685{
1686 struct loop_cmd *cmd =
Ming Leie03a3d72015-08-17 10:31:48 +08001687 container_of(work, struct loop_cmd, work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001688
1689 loop_handle_cmd(cmd);
1690}
1691
1692static int loop_init_request(void *data, struct request *rq,
1693 unsigned int hctx_idx, unsigned int request_idx,
1694 unsigned int numa_node)
1695{
1696 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1697
1698 cmd->rq = rq;
Ming Leie03a3d72015-08-17 10:31:48 +08001699 init_kthread_work(&cmd->work, loop_queue_work);
Ming Leib5dd2f62014-12-31 13:22:57 +00001700
1701 return 0;
1702}
1703
1704static struct blk_mq_ops loop_mq_ops = {
1705 .queue_rq = loop_queue_rq,
1706 .map_queue = blk_mq_map_queue,
1707 .init_request = loop_init_request,
1708};
1709
Kay Sievers34dd82a2011-07-31 22:08:04 +02001710static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001711{
1712 struct loop_device *lo;
1713 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001714 int err;
Ken Chen73285082007-05-08 00:28:20 -07001715
Silva Paulo68d740d2012-07-14 15:39:58 -07001716 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001717 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07001718 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07001719 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001720
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001721 lo->lo_state = Lo_unbound;
1722
Tejun Heoc718aa62013-02-27 17:03:58 -08001723 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001724 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08001725 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1726 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02001727 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001728 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08001729 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001730 }
1731 if (err < 0)
1732 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08001733 i = err;
Ken Chen73285082007-05-08 00:28:20 -07001734
Wei Yongjun183cfb52013-03-22 08:59:19 -06001735 err = -ENOMEM;
Ming Leib5dd2f62014-12-31 13:22:57 +00001736 lo->tag_set.ops = &loop_mq_ops;
1737 lo->tag_set.nr_hw_queues = 1;
1738 lo->tag_set.queue_depth = 128;
1739 lo->tag_set.numa_node = NUMA_NO_NODE;
1740 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1741 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1742 lo->tag_set.driver_data = lo;
1743
1744 err = blk_mq_alloc_tag_set(&lo->tag_set);
1745 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001746 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07001747
Ming Leib5dd2f62014-12-31 13:22:57 +00001748 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1749 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1750 err = PTR_ERR(lo->lo_queue);
1751 goto out_cleanup_tags;
1752 }
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001753 lo->lo_queue->queuedata = lo;
1754
Ming Lei5b5e20f2015-08-17 10:31:47 +08001755 /*
1756 * It doesn't make sense to enable merge because the I/O
1757 * submitted to backing file is handled page by page.
1758 */
1759 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1760
Minfei Huang7a649732016-06-07 10:05:15 +08001761 err = -ENOMEM;
Laurent Vivier476a4812008-03-26 12:11:53 +01001762 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001763 if (!disk)
1764 goto out_free_queue;
1765
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001766 /*
1767 * Disable partition scanning by default. The in-kernel partition
1768 * scanning can be requested individually per-device during its
1769 * setup. Userspace can always add and remove partitions from all
1770 * devices. The needed partition minors are allocated from the
1771 * extended minor space, the main loop device numbers will continue
1772 * to match the loop minors, regardless of the number of partitions
1773 * used.
1774 *
1775 * If max_part is given, partition scanning is globally enabled for
1776 * all loop devices. The minors for the main loop devices will be
1777 * multiples of max_part.
1778 *
1779 * Note: Global-for-all-devices, set-only-at-init, read-only module
1780 * parameteters like 'max_loop' and 'max_part' make things needlessly
1781 * complicated, are too static, inflexible and may surprise
1782 * userspace tools. Parameters like this in general should be avoided.
1783 */
1784 if (!part_shift)
1785 disk->flags |= GENHD_FL_NO_PART_SCAN;
1786 disk->flags |= GENHD_FL_EXT_DEVT;
Ken Chen73285082007-05-08 00:28:20 -07001787 mutex_init(&lo->lo_ctl_mutex);
Ming Leif8933662015-05-06 12:26:23 +08001788 atomic_set(&lo->lo_refcnt, 0);
Ken Chen73285082007-05-08 00:28:20 -07001789 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07001790 spin_lock_init(&lo->lo_lock);
1791 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001792 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001793 disk->fops = &lo_fops;
1794 disk->private_data = lo;
1795 disk->queue = lo->lo_queue;
1796 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001797 add_disk(disk);
1798 *l = lo;
1799 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001800
1801out_free_queue:
1802 blk_cleanup_queue(lo->lo_queue);
Ming Leib5dd2f62014-12-31 13:22:57 +00001803out_cleanup_tags:
1804 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04001805out_free_idr:
1806 idr_remove(&loop_index_idr, i);
Ken Chen73285082007-05-08 00:28:20 -07001807out_free_dev:
1808 kfree(lo);
1809out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001810 return err;
Ken Chen73285082007-05-08 00:28:20 -07001811}
1812
Kay Sievers34dd82a2011-07-31 22:08:04 +02001813static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001814{
Ken Chen73285082007-05-08 00:28:20 -07001815 blk_cleanup_queue(lo->lo_queue);
NeilBrown6cd18e72015-04-27 14:12:22 +10001816 del_gendisk(lo->lo_disk);
Ming Leib5dd2f62014-12-31 13:22:57 +00001817 blk_mq_free_tag_set(&lo->tag_set);
Ken Chen73285082007-05-08 00:28:20 -07001818 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001819 kfree(lo);
1820}
1821
Kay Sievers770fe302011-07-31 22:08:04 +02001822static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001823{
Kay Sievers770fe302011-07-31 22:08:04 +02001824 struct loop_device *lo = ptr;
1825 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001826
Kay Sievers770fe302011-07-31 22:08:04 +02001827 if (lo->lo_state == Lo_unbound) {
1828 *l = lo;
1829 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001830 }
Kay Sievers770fe302011-07-31 22:08:04 +02001831 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001832}
1833
Kay Sievers34dd82a2011-07-31 22:08:04 +02001834static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001835{
Ken Chena47653f2007-06-08 13:46:44 -07001836 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001837 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001838
Kay Sievers770fe302011-07-31 22:08:04 +02001839 if (i < 0) {
1840 int err;
1841
1842 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1843 if (err == 1) {
1844 *l = lo;
1845 ret = lo->lo_number;
1846 }
1847 goto out;
1848 }
1849
1850 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001851 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07001852 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02001853 *l = lo;
1854 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07001855 }
Kay Sievers770fe302011-07-31 22:08:04 +02001856out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001857 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07001858}
1859
Ken Chen73285082007-05-08 00:28:20 -07001860static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1861{
Al Viro705962c2007-05-13 05:52:32 -04001862 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001863 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001864 int err;
Ken Chen73285082007-05-08 00:28:20 -07001865
Kay Sievers34dd82a2011-07-31 22:08:04 +02001866 mutex_lock(&loop_index_mutex);
1867 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1868 if (err < 0)
1869 err = loop_add(&lo, MINOR(dev) >> part_shift);
1870 if (err < 0)
Mikulas Patockaa207f592013-10-14 12:13:24 -04001871 kobj = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001872 else
1873 kobj = get_disk(lo->lo_disk);
1874 mutex_unlock(&loop_index_mutex);
Ken Chen73285082007-05-08 00:28:20 -07001875
1876 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001877 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001878}
1879
Kay Sievers770fe302011-07-31 22:08:04 +02001880static long loop_control_ioctl(struct file *file, unsigned int cmd,
1881 unsigned long parm)
1882{
1883 struct loop_device *lo;
1884 int ret = -ENOSYS;
1885
1886 mutex_lock(&loop_index_mutex);
1887 switch (cmd) {
1888 case LOOP_CTL_ADD:
1889 ret = loop_lookup(&lo, parm);
1890 if (ret >= 0) {
1891 ret = -EEXIST;
1892 break;
1893 }
1894 ret = loop_add(&lo, parm);
1895 break;
1896 case LOOP_CTL_REMOVE:
1897 ret = loop_lookup(&lo, parm);
1898 if (ret < 0)
1899 break;
1900 mutex_lock(&lo->lo_ctl_mutex);
1901 if (lo->lo_state != Lo_unbound) {
1902 ret = -EBUSY;
1903 mutex_unlock(&lo->lo_ctl_mutex);
1904 break;
1905 }
Ming Leif8933662015-05-06 12:26:23 +08001906 if (atomic_read(&lo->lo_refcnt) > 0) {
Kay Sievers770fe302011-07-31 22:08:04 +02001907 ret = -EBUSY;
1908 mutex_unlock(&lo->lo_ctl_mutex);
1909 break;
1910 }
1911 lo->lo_disk->private_data = NULL;
1912 mutex_unlock(&lo->lo_ctl_mutex);
1913 idr_remove(&loop_index_idr, lo->lo_number);
1914 loop_remove(lo);
1915 break;
1916 case LOOP_CTL_GET_FREE:
1917 ret = loop_lookup(&lo, -1);
1918 if (ret >= 0)
1919 break;
1920 ret = loop_add(&lo, -1);
1921 }
1922 mutex_unlock(&loop_index_mutex);
1923
1924 return ret;
1925}
1926
1927static const struct file_operations loop_ctl_fops = {
1928 .open = nonseekable_open,
1929 .unlocked_ioctl = loop_control_ioctl,
1930 .compat_ioctl = loop_control_ioctl,
1931 .owner = THIS_MODULE,
1932 .llseek = noop_llseek,
1933};
1934
1935static struct miscdevice loop_misc = {
1936 .minor = LOOP_CTRL_MINOR,
1937 .name = "loop-control",
1938 .fops = &loop_ctl_fops,
1939};
1940
1941MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1942MODULE_ALIAS("devname:loop-control");
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944static int __init loop_init(void)
1945{
Ken Chena47653f2007-06-08 13:46:44 -07001946 int i, nr;
1947 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001948 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02001949 int err;
Ken Chena47653f2007-06-08 13:46:44 -07001950
Kay Sievers770fe302011-07-31 22:08:04 +02001951 err = misc_register(&loop_misc);
1952 if (err < 0)
1953 return err;
Laurent Vivier476a4812008-03-26 12:11:53 +01001954
1955 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001956 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01001957 part_shift = fls(max_part);
1958
Namhyung Kimac04fee2011-05-27 07:59:25 +02001959 /*
1960 * Adjust max_part according to part_shift as it is exported
1961 * to user space so that user can decide correct minor number
1962 * if [s]he want to create more devices.
1963 *
1964 * Note that -1 is required because partition 0 is reserved
1965 * for the whole disk.
1966 */
1967 max_part = (1UL << part_shift) - 1;
1968 }
1969
Guo Chaob1a66502013-02-21 15:16:49 -08001970 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1971 err = -EINVAL;
1972 goto misc_out;
1973 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02001974
Guo Chaob1a66502013-02-21 15:16:49 -08001975 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1976 err = -EINVAL;
1977 goto misc_out;
1978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Kay Sieversd134b002011-07-31 22:08:04 +02001980 /*
1981 * If max_loop is specified, create that many devices upfront.
1982 * This also becomes a hard limit. If max_loop is not specified,
1983 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1984 * init time. Loop devices can be requested on-demand with the
1985 * /dev/loop-control interface, or be instantiated by accessing
1986 * a 'dead' device node.
1987 */
Ken Chen73285082007-05-08 00:28:20 -07001988 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001989 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001990 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07001991 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02001992 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001993 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
Ken Chena47653f2007-06-08 13:46:44 -07001995
Guo Chaob1a66502013-02-21 15:16:49 -08001996 if (register_blkdev(LOOP_MAJOR, "loop")) {
1997 err = -EIO;
1998 goto misc_out;
1999 }
Ken Chena47653f2007-06-08 13:46:44 -07002000
Ken Chena47653f2007-06-08 13:46:44 -07002001 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2002 THIS_MODULE, loop_probe, NULL, NULL);
2003
Kay Sieversd134b002011-07-31 22:08:04 +02002004 /* pre-create number of devices given by config or max_loop */
Kay Sievers34dd82a2011-07-31 22:08:04 +02002005 mutex_lock(&loop_index_mutex);
2006 for (i = 0; i < nr; i++)
2007 loop_add(&lo, i);
2008 mutex_unlock(&loop_index_mutex);
2009
Ken Chen73285082007-05-08 00:28:20 -07002010 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08002012
2013misc_out:
2014 misc_deregister(&loop_misc);
2015 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002016}
Ken Chena47653f2007-06-08 13:46:44 -07002017
Kay Sievers34dd82a2011-07-31 22:08:04 +02002018static int loop_exit_cb(int id, void *ptr, void *data)
2019{
2020 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07002021
Kay Sievers34dd82a2011-07-31 22:08:04 +02002022 loop_remove(lo);
2023 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024}
2025
Ken Chen73285082007-05-08 00:28:20 -07002026static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
Ken Chena47653f2007-06-08 13:46:44 -07002028 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Namhyung Kima1c15c52011-05-24 16:48:55 +02002030 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07002031
Kay Sievers34dd82a2011-07-31 22:08:04 +02002032 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002033 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07002034
Ken Chena47653f2007-06-08 13:46:44 -07002035 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07002036 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02002037
2038 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039}
2040
2041module_init(loop_init);
2042module_exit(loop_exit);
2043
2044#ifndef MODULE
2045static int __init max_loop_setup(char *str)
2046{
2047 max_loop = simple_strtol(str, NULL, 0);
2048 return 1;
2049}
2050
2051__setup("max_loop=", max_loop_setup);
2052#endif