blob: 40e715531aa65f0e63babd13757ad6dfb104d22a [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 Viro83a87612013-05-12 10:14:07 -040078#include "loop.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80#include <asm/uaccess.h>
81
Kay Sievers34dd82a2011-07-31 22:08:04 +020082static DEFINE_IDR(loop_index_idr);
83static DEFINE_MUTEX(loop_index_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Laurent Vivier476a4812008-03-26 12:11:53 +010085static int max_part;
86static int part_shift;
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
89 * Transfer functions
90 */
91static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
Cong Wangcfd80052011-11-25 23:14:18 +080096 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99 if (cmd == READ)
100 memcpy(loop_buf, raw_buf, size);
101 else
102 memcpy(raw_buf, loop_buf, size);
103
Cong Wangcfd80052011-11-25 23:14:18 +0800104 kunmap_atomic(loop_buf);
105 kunmap_atomic(raw_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 cond_resched();
107 return 0;
108}
109
110static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
114{
Cong Wangcfd80052011-11-25 23:14:18 +0800115 char *raw_buf = kmap_atomic(raw_page) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page) + loop_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 char *in, *out, *key;
118 int i, keysize;
119
120 if (cmd == READ) {
121 in = raw_buf;
122 out = loop_buf;
123 } else {
124 in = loop_buf;
125 out = raw_buf;
126 }
127
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
132
Cong Wangcfd80052011-11-25 23:14:18 +0800133 kunmap_atomic(loop_buf);
134 kunmap_atomic(raw_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 cond_resched();
136 return 0;
137}
138
139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140{
141 if (unlikely(info->lo_encrypt_key_size <= 0))
142 return -EINVAL;
143 return 0;
144}
145
146static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
149};
150
151static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
154 .init = xor_init
155};
156
157/* xfer_funcs[0] is special - its release function is never called */
158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 &none_funcs,
160 &xor_funcs
161};
162
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100163static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Guo Chaob7a1da62013-02-21 15:16:50 -0800165 loff_t loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 /* Compute loopsize in bytes */
Guo Chaob7a1da62013-02-21 15:16:50 -0800168 loopsize = i_size_read(file->f_mapping->host);
169 if (offset > 0)
170 loopsize -= offset;
171 /* offset is beyond i_size, weird but possible */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100172 if (loopsize < 0)
173 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100175 if (sizelimit > 0 && sizelimit < loopsize)
176 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 /*
178 * Unfortunately, if we want to do I/O on the device,
179 * the number of 512-byte sectors has to fit into a sector_t.
180 */
181 return loopsize >> 9;
182}
183
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100184static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100186 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
187}
188
189static int
190figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
191{
192 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 sector_t x = (sector_t)size;
Guo Chao7b0576a2013-02-21 15:16:47 -0800194 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 if (unlikely((loff_t)x != size))
197 return -EFBIG;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100198 if (lo->lo_offset != offset)
199 lo->lo_offset = offset;
200 if (lo->lo_sizelimit != sizelimit)
201 lo->lo_sizelimit = sizelimit;
Ken Chen73285082007-05-08 00:28:20 -0700202 set_capacity(lo->lo_disk, x);
Guo Chao7b0576a2013-02-21 15:16:47 -0800203 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
204 /* let user-space know about the new size */
205 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static inline int
210lo_do_transfer(struct loop_device *lo, int cmd,
211 struct page *rpage, unsigned roffs,
212 struct page *lpage, unsigned loffs,
213 int size, sector_t rblock)
214{
215 if (unlikely(!lo->transfer))
216 return 0;
217
218 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
219}
220
221/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 * __do_lo_send_write - helper for writing data to a loop device
223 *
224 * This helper just factors out common code between do_lo_send_direct_write()
225 * and do_lo_send_write().
226 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800227static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100228 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 ssize_t bw;
231 mm_segment_t old_fs = get_fs();
232
Al Viro03d95eb2013-03-20 13:04:20 -0400233 file_start_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 set_fs(get_ds());
235 bw = file->f_op->write(file, buf, len, &pos);
236 set_fs(old_fs);
Al Viro03d95eb2013-03-20 13:04:20 -0400237 file_end_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (likely(bw == len))
239 return 0;
240 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
241 (unsigned long long)pos, len);
242 if (bw >= 0)
243 bw = -EIO;
244 return bw;
245}
246
247/**
248 * do_lo_send_direct_write - helper for writing data to a loop device
249 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200250 * This is the fast, non-transforming version that does not need double
251 * buffering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 */
253static int do_lo_send_direct_write(struct loop_device *lo,
Al Viro511de732007-10-08 12:10:13 -0400254 struct bio_vec *bvec, loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100257 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 bvec->bv_len, pos);
259 kunmap(bvec->bv_page);
260 cond_resched();
261 return bw;
262}
263
264/**
265 * do_lo_send_write - helper for writing data to a loop device
266 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200267 * This is the slow, transforming version that needs to double buffer the
268 * data as it cannot do the transformations in place without having direct
269 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 */
271static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400272 loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
275 bvec->bv_offset, bvec->bv_len, pos >> 9);
276 if (likely(!ret))
277 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100278 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 pos);
280 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
281 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
282 if (ret > 0)
283 ret = -EIO;
284 return ret;
285}
286
Al Viro511de732007-10-08 12:10:13 -0400287static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
Al Viro511de732007-10-08 12:10:13 -0400289 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 struct page *page);
291 struct bio_vec *bvec;
292 struct page *page = NULL;
293 int i, ret = 0;
294
Christoph Hellwig456be142011-10-17 12:57:20 +0200295 if (lo->transfer != transfer_none) {
296 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
297 if (unlikely(!page))
298 goto fail;
299 kmap(page);
300 do_lo_send = do_lo_send_write;
301 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 do_lo_send = do_lo_send_direct_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 }
Christoph Hellwig456be142011-10-17 12:57:20 +0200304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 bio_for_each_segment(bvec, bio, i) {
Al Viro511de732007-10-08 12:10:13 -0400306 ret = do_lo_send(lo, bvec, pos, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 if (ret < 0)
308 break;
309 pos += bvec->bv_len;
310 }
311 if (page) {
312 kunmap(page);
313 __free_page(page);
314 }
315out:
316 return ret;
317fail:
318 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
319 ret = -ENOMEM;
320 goto out;
321}
322
323struct lo_read_data {
324 struct loop_device *lo;
325 struct page *page;
326 unsigned offset;
327 int bsize;
328};
329
330static int
Jens Axboefd582142007-06-12 21:20:37 +0200331lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
332 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Jens Axboefd582142007-06-12 21:20:37 +0200334 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200336 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 sector_t IV;
Jens Axboe3603b8e2010-12-20 09:15:19 +0100338 int size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Jens Axboefd582142007-06-12 21:20:37 +0200340 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
341 (buf->offset >> 9);
342 size = sd->len;
343 if (size > p->bsize)
344 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Jens Axboefd582142007-06-12 21:20:37 +0200346 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 printk(KERN_ERR "loop: transfer error block %ld\n",
348 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200349 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351
352 flush_dcache_page(p->page);
353
Jens Axboefd582142007-06-12 21:20:37 +0200354 if (size > 0)
355 p->offset += size;
356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return size;
358}
359
360static int
Jens Axboefd582142007-06-12 21:20:37 +0200361lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
362{
363 return __splice_from_pipe(pipe, sd, lo_splice_actor);
364}
365
Dave Young306df072012-02-08 22:07:19 +0100366static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367do_lo_receive(struct loop_device *lo,
368 struct bio_vec *bvec, int bsize, loff_t pos)
369{
370 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200371 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 struct file *file;
Dave Young306df072012-02-08 22:07:19 +0100373 ssize_t retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 cookie.lo = lo;
376 cookie.page = bvec->bv_page;
377 cookie.offset = bvec->bv_offset;
378 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200379
380 sd.len = 0;
381 sd.total_len = bvec->bv_len;
382 sd.flags = 0;
383 sd.pos = pos;
384 sd.u.data = &cookie;
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200387 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
388
Dave Young306df072012-02-08 22:07:19 +0100389 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
392static int
393lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
394{
395 struct bio_vec *bvec;
Dave Young306df072012-02-08 22:07:19 +0100396 ssize_t s;
397 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 bio_for_each_segment(bvec, bio, i) {
Dave Young306df072012-02-08 22:07:19 +0100400 s = do_lo_receive(lo, bvec, bsize, pos);
401 if (s < 0)
402 return s;
403
404 if (s != bvec->bv_len) {
405 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 break;
Dave Young306df072012-02-08 22:07:19 +0100407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 pos += bvec->bv_len;
409 }
Dave Young306df072012-02-08 22:07:19 +0100410 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
412
413static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
414{
415 loff_t pos;
416 int ret;
417
418 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100419
420 if (bio_rw(bio) == WRITE) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100421 struct file *file = lo->lo_backing_file;
422
Tejun Heo6259f282010-09-03 11:56:17 +0200423 if (bio->bi_rw & REQ_FLUSH) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100424 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200425 if (unlikely(ret && ret != -EINVAL)) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100426 ret = -EIO;
427 goto out;
428 }
429 }
430
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200431 /*
432 * We use punch hole to reclaim the free space used by the
Dave Youngae957572011-11-25 09:41:25 +0100433 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200434 * encryption is enabled, because it may give an attacker
435 * useful information.
436 */
437 if (bio->bi_rw & REQ_DISCARD) {
438 struct file *file = lo->lo_backing_file;
439 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
440
441 if ((!file->f_op->fallocate) ||
442 lo->lo_encrypt_key_size) {
443 ret = -EOPNOTSUPP;
444 goto out;
445 }
446 ret = file->f_op->fallocate(file, mode, pos,
447 bio->bi_size);
448 if (unlikely(ret && ret != -EINVAL &&
449 ret != -EOPNOTSUPP))
450 ret = -EIO;
451 goto out;
452 }
453
Al Viro511de732007-10-08 12:10:13 -0400454 ret = lo_send(lo, bio, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100455
Tejun Heo6259f282010-09-03 11:56:17 +0200456 if ((bio->bi_rw & REQ_FUA) && !ret) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100457 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200458 if (unlikely(ret && ret != -EINVAL))
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100459 ret = -EIO;
460 }
461 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100463
464out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 return ret;
466}
467
468/*
469 * Add bio to back of pending list
470 */
471static void loop_add_bio(struct loop_device *lo, struct bio *bio)
472{
Lukas Czerner7b5a3522012-11-30 11:42:41 +0100473 lo->lo_bio_count++;
Akinobu Mitae6863072009-04-17 08:41:21 +0200474 bio_list_add(&lo->lo_bio_list, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477/*
478 * Grab first pending buffer
479 */
480static struct bio *loop_get_bio(struct loop_device *lo)
481{
Lukas Czerner7b5a3522012-11-30 11:42:41 +0100482 lo->lo_bio_count--;
Akinobu Mitae6863072009-04-17 08:41:21 +0200483 return bio_list_pop(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200486static void loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
488 struct loop_device *lo = q->queuedata;
489 int rw = bio_rw(old_bio);
490
Nick Piggin35a82d12005-06-23 00:09:06 -0700491 if (rw == READA)
492 rw = READ;
493
494 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 spin_lock_irq(&lo->lo_lock);
497 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700498 goto out;
499 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
500 goto out;
Lukas Czerner7b5a3522012-11-30 11:42:41 +0100501 if (lo->lo_bio_count >= q->nr_congestion_on)
502 wait_event_lock_irq(lo->lo_req_wait,
503 lo->lo_bio_count < q->nr_congestion_off,
504 lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700506 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700507 spin_unlock_irq(&lo->lo_lock);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200508 return;
Nick Piggin35a82d12005-06-23 00:09:06 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700511 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200512 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515struct switch_request {
516 struct file *file;
517 struct completion wait;
518};
519
520static void do_loop_switch(struct loop_device *, struct switch_request *);
521
522static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
523{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (unlikely(!bio->bi_bdev)) {
525 do_loop_switch(lo, bio->bi_private);
526 bio_put(bio);
527 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700528 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200529 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 }
531}
532
533/*
534 * worker thread that handles reads/writes to file backed loop devices,
535 * to avoid blocking in our make_request_fn. it also does loop decrypting
536 * on reads for block backed loop, as that is too heavy to do from
537 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700538 *
539 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
540 * calling kthread_stop(). Therefore once kthread_should_stop() is
541 * true, make_request will not place any more requests. Therefore
542 * once kthread_should_stop() is true and lo_bio is NULL, we are
543 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
545static int loop_thread(void *data)
546{
547 struct loop_device *lo = data;
548 struct bio *bio;
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 set_user_nice(current, -20);
551
Akinobu Mitae6863072009-04-17 08:41:21 +0200552 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700554 wait_event_interruptible(lo->lo_event,
Akinobu Mitae6863072009-04-17 08:41:21 +0200555 !bio_list_empty(&lo->lo_bio_list) ||
556 kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700557
Akinobu Mitae6863072009-04-17 08:41:21 +0200558 if (bio_list_empty(&lo->lo_bio_list))
Nick Piggin35a82d12005-06-23 00:09:06 -0700559 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700560 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 bio = loop_get_bio(lo);
Lukas Czerner7b5a3522012-11-30 11:42:41 +0100562 if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off)
563 wake_up(&lo->lo_req_wait);
Nick Piggin35a82d12005-06-23 00:09:06 -0700564 spin_unlock_irq(&lo->lo_lock);
565
566 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return 0;
571}
572
573/*
574 * loop_switch performs the hard work of switching a backing store.
575 * First it needs to flush existing IO, it does this by sending a magic
576 * BIO down the pipe. The completion of this BIO does the actual switch.
577 */
578static int loop_switch(struct loop_device *lo, struct file *file)
579{
580 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100581 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (!bio)
583 return -ENOMEM;
584 init_completion(&w.wait);
585 w.file = file;
586 bio->bi_private = &w;
587 bio->bi_bdev = NULL;
588 loop_make_request(lo->lo_queue, bio);
589 wait_for_completion(&w.wait);
590 return 0;
591}
592
593/*
Milan Broz14f27932008-12-12 14:48:27 +0100594 * Helper to flush the IOs in loop, but keeping loop thread running
595 */
596static int loop_flush(struct loop_device *lo)
597{
598 /* loop not yet configured, no running thread, nothing to flush */
599 if (!lo->lo_thread)
600 return 0;
601
602 return loop_switch(lo, NULL);
603}
604
605/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * Do the actual switch; called from the BIO completion routine
607 */
608static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
609{
610 struct file *file = p->file;
611 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100612 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Milan Broz14f27932008-12-12 14:48:27 +0100614 /* if no new file, only flush of queued bios requested */
615 if (!file)
616 goto out;
617
618 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
620 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700621 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
622 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 lo->old_gfp_mask = mapping_gfp_mask(mapping);
624 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100625out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 complete(&p->wait);
627}
628
629
630/*
631 * loop_change_fd switched the backing store of a loopback device to
632 * a new file. This is useful for operating system installers to free up
633 * the original file and in High Availability environments to switch to
634 * an alternative location for the content in case of server meltdown.
635 * This can only work if the loop device is used read-only, and if the
636 * new backing store is the same size and type as the old backing store.
637 */
Al Virobb214882008-03-02 09:29:48 -0500638static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
639 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 struct file *file, *old_file;
642 struct inode *inode;
643 int error;
644
645 error = -ENXIO;
646 if (lo->lo_state != Lo_bound)
647 goto out;
648
649 /* the loop device has to be read-only */
650 error = -EINVAL;
651 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
652 goto out;
653
654 error = -EBADF;
655 file = fget(arg);
656 if (!file)
657 goto out;
658
659 inode = file->f_mapping->host;
660 old_file = lo->lo_backing_file;
661
662 error = -EINVAL;
663
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
665 goto out_putf;
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 /* size of the new backing store needs to be the same */
668 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
669 goto out_putf;
670
671 /* and ... switch */
672 error = loop_switch(lo, file);
673 if (error)
674 goto out_putf;
675
676 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200677 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100678 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return 0;
680
681 out_putf:
682 fput(file);
683 out:
684 return error;
685}
686
687static inline int is_loop_device(struct file *file)
688{
689 struct inode *i = file->f_mapping->host;
690
691 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
692}
693
Milan Brozee862732010-08-23 15:16:00 +0200694/* loop sysfs attributes */
695
696static ssize_t loop_attr_show(struct device *dev, char *page,
697 ssize_t (*callback)(struct loop_device *, char *))
698{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200699 struct gendisk *disk = dev_to_disk(dev);
700 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200701
Kay Sievers34dd82a2011-07-31 22:08:04 +0200702 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200703}
704
705#define LOOP_ATTR_RO(_name) \
706static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
707static ssize_t loop_attr_do_show_##_name(struct device *d, \
708 struct device_attribute *attr, char *b) \
709{ \
710 return loop_attr_show(d, b, loop_attr_##_name##_show); \
711} \
712static struct device_attribute loop_attr_##_name = \
713 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
714
715static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
716{
717 ssize_t ret;
718 char *p = NULL;
719
Kay Sievers05eb0f22011-07-31 22:21:35 +0200720 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200721 if (lo->lo_backing_file)
722 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200723 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200724
725 if (IS_ERR_OR_NULL(p))
726 ret = PTR_ERR(p);
727 else {
728 ret = strlen(p);
729 memmove(buf, p, ret);
730 buf[ret++] = '\n';
731 buf[ret] = 0;
732 }
733
734 return ret;
735}
736
737static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
738{
739 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
740}
741
742static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
743{
744 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
745}
746
747static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
748{
749 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
750
751 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
752}
753
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200754static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
755{
756 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
757
758 return sprintf(buf, "%s\n", partscan ? "1" : "0");
759}
760
Milan Brozee862732010-08-23 15:16:00 +0200761LOOP_ATTR_RO(backing_file);
762LOOP_ATTR_RO(offset);
763LOOP_ATTR_RO(sizelimit);
764LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200765LOOP_ATTR_RO(partscan);
Milan Brozee862732010-08-23 15:16:00 +0200766
767static struct attribute *loop_attrs[] = {
768 &loop_attr_backing_file.attr,
769 &loop_attr_offset.attr,
770 &loop_attr_sizelimit.attr,
771 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200772 &loop_attr_partscan.attr,
Milan Brozee862732010-08-23 15:16:00 +0200773 NULL,
774};
775
776static struct attribute_group loop_attribute_group = {
777 .name = "loop",
778 .attrs= loop_attrs,
779};
780
781static int loop_sysfs_init(struct loop_device *lo)
782{
783 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
784 &loop_attribute_group);
785}
786
787static void loop_sysfs_exit(struct loop_device *lo)
788{
789 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
790 &loop_attribute_group);
791}
792
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200793static void loop_config_discard(struct loop_device *lo)
794{
795 struct file *file = lo->lo_backing_file;
796 struct inode *inode = file->f_mapping->host;
797 struct request_queue *q = lo->lo_queue;
798
799 /*
800 * We use punch hole to reclaim the free space used by the
801 * image a.k.a. discard. However we do support discard if
802 * encryption is enabled, because it may give an attacker
803 * useful information.
804 */
805 if ((!file->f_op->fallocate) ||
806 lo->lo_encrypt_key_size) {
807 q->limits.discard_granularity = 0;
808 q->limits.discard_alignment = 0;
809 q->limits.max_discard_sectors = 0;
810 q->limits.discard_zeroes_data = 0;
811 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
812 return;
813 }
814
815 q->limits.discard_granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaf3c02011-12-02 14:47:03 +0100816 q->limits.discard_alignment = 0;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200817 q->limits.max_discard_sectors = UINT_MAX >> 9;
818 q->limits.discard_zeroes_data = 1;
819 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
820}
821
Al Virobb214882008-03-02 09:29:48 -0500822static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 struct block_device *bdev, unsigned int arg)
824{
825 struct file *file, *f;
826 struct inode *inode;
827 struct address_space *mapping;
828 unsigned lo_blocksize;
829 int lo_flags = 0;
830 int error;
831 loff_t size;
832
833 /* This is safe, since we have a reference from open(). */
834 __module_get(THIS_MODULE);
835
836 error = -EBADF;
837 file = fget(arg);
838 if (!file)
839 goto out;
840
841 error = -EBUSY;
842 if (lo->lo_state != Lo_unbound)
843 goto out_putf;
844
845 /* Avoid recursion */
846 f = file;
847 while (is_loop_device(f)) {
848 struct loop_device *l;
849
Al Virobb214882008-03-02 09:29:48 -0500850 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 goto out_putf;
852
853 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
854 if (l->lo_state == Lo_unbound) {
855 error = -EINVAL;
856 goto out_putf;
857 }
858 f = l->lo_backing_file;
859 }
860
861 mapping = file->f_mapping;
862 inode = mapping->host;
863
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 error = -EINVAL;
Christoph Hellwig456be142011-10-17 12:57:20 +0200865 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Christoph Hellwig456be142011-10-17 12:57:20 +0200868 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
869 !file->f_op->write)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 lo_flags |= LO_FLAGS_READ_ONLY;
871
Christoph Hellwig456be142011-10-17 12:57:20 +0200872 lo_blocksize = S_ISBLK(inode->i_mode) ?
873 inode->i_bdev->bd_block_size : PAGE_SIZE;
874
875 error = -EFBIG;
876 size = get_loop_size(lo, file);
877 if ((loff_t)(sector_t)size != size)
878 goto out_putf;
879
880 error = 0;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
883
884 lo->lo_blocksize = lo_blocksize;
885 lo->lo_device = bdev;
886 lo->lo_flags = lo_flags;
887 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700888 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 lo->ioctl = NULL;
890 lo->lo_sizelimit = 0;
Lukas Czerner7b5a3522012-11-30 11:42:41 +0100891 lo->lo_bio_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 lo->old_gfp_mask = mapping_gfp_mask(mapping);
893 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
894
Akinobu Mitae6863072009-04-17 08:41:21 +0200895 bio_list_init(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 /*
898 * set queue make_request_fn, and add limits based on lower level
899 * device
900 */
901 blk_queue_make_request(lo->lo_queue, loop_make_request);
902 lo->lo_queue->queuedata = lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100904 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Tejun Heo4913efe2010-09-03 11:56:16 +0200905 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100906
Ken Chen73285082007-05-08 00:28:20 -0700907 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200909 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200910 /* let user-space know about the new size */
911 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 set_blocksize(bdev, lo_blocksize);
914
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700915 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
916 lo->lo_number);
917 if (IS_ERR(lo->lo_thread)) {
918 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700919 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700920 }
921 lo->lo_state = Lo_bound;
922 wake_up_process(lo->lo_thread);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200923 if (part_shift)
924 lo->lo_flags |= LO_FLAGS_PARTSCAN;
925 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100926 ioctl_by_bdev(bdev, BLKRRPART, 0);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700927
928 /* Grab the block_device to prevent its destruction after we
929 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
930 */
931 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 return 0;
933
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700934out_clr:
Milan Brozee862732010-08-23 15:16:00 +0200935 loop_sysfs_exit(lo);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700936 lo->lo_thread = NULL;
937 lo->lo_device = NULL;
938 lo->lo_backing_file = NULL;
939 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700940 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700941 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700942 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200943 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700944 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
945 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 out_putf:
947 fput(file);
948 out:
949 /* This is safe: open() is still holding a reference. */
950 module_put(THIS_MODULE);
951 return error;
952}
953
954static int
955loop_release_xfer(struct loop_device *lo)
956{
957 int err = 0;
958 struct loop_func_table *xfer = lo->lo_encryption;
959
960 if (xfer) {
961 if (xfer->release)
962 err = xfer->release(lo);
963 lo->transfer = NULL;
964 lo->lo_encryption = NULL;
965 module_put(xfer->owner);
966 }
967 return err;
968}
969
970static int
971loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
972 const struct loop_info64 *i)
973{
974 int err = 0;
975
976 if (xfer) {
977 struct module *owner = xfer->owner;
978
979 if (!try_module_get(owner))
980 return -EINVAL;
981 if (xfer->init)
982 err = xfer->init(lo, i);
983 if (err)
984 module_put(owner);
985 else
986 lo->lo_encryption = xfer;
987 }
988 return err;
989}
990
Ayan George4c823cc2011-09-21 10:02:13 +0200991static int loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
993 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400994 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +0200995 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 if (lo->lo_state != Lo_bound)
998 return -ENXIO;
999
Dave Chinnera1ecac32012-09-28 10:42:23 +02001000 /*
1001 * If we've explicitly asked to tear down the loop device,
1002 * and it has an elevated reference count, set it for auto-teardown when
1003 * the last reference goes away. This stops $!~#$@ udev from
1004 * preventing teardown because it decided that it needs to run blkid on
1005 * the loopback device whenever they appear. xfstests is notorious for
1006 * failing tests because blkid via udev races with a losetup
1007 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1008 * command to fail with EBUSY.
1009 */
1010 if (lo->lo_refcnt > 1) {
1011 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1012 mutex_unlock(&lo->lo_ctl_mutex);
1013 return 0;
1014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 if (filp == NULL)
1017 return -EINVAL;
1018
1019 spin_lock_irq(&lo->lo_lock);
1020 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 spin_unlock_irq(&lo->lo_lock);
1022
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001023 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Kay Sievers05eb0f22011-07-31 22:21:35 +02001025 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001027 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 loop_release_xfer(lo);
1030 lo->transfer = NULL;
1031 lo->ioctl = NULL;
1032 lo->lo_device = NULL;
1033 lo->lo_encryption = NULL;
1034 lo->lo_offset = 0;
1035 lo->lo_sizelimit = 0;
1036 lo->lo_encrypt_key_size = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001037 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1039 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1040 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001041 if (bdev) {
1042 bdput(bdev);
Al Virobb214882008-03-02 09:29:48 -05001043 invalidate_bdev(bdev);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001044 }
Ken Chen73285082007-05-08 00:28:20 -07001045 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001046 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001047 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001048 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001049 /* let user-space know about this change */
1050 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1051 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 mapping_set_gfp_mask(filp->f_mapping, gfp);
1053 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 /* This is safe: open() is still holding a reference. */
1055 module_put(THIS_MODULE);
Jens Axboec2fccc12013-04-08 10:12:11 +02001056 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1057 ioctl_by_bdev(bdev, BLKRRPART, 0);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001058 lo->lo_flags = 0;
1059 if (!part_shift)
1060 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001061 mutex_unlock(&lo->lo_ctl_mutex);
1062 /*
1063 * Need not hold lo_ctl_mutex to fput backing file.
1064 * Calling fput holding lo_ctl_mutex triggers a circular
1065 * lock dependency possibility warning as fput can take
1066 * bd_mutex which is usually taken before lo_ctl_mutex.
1067 */
1068 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return 0;
1070}
1071
1072static int
1073loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1074{
1075 int err;
1076 struct loop_func_table *xfer;
Eric W. Biedermane4849732012-02-11 11:23:51 -08001077 kuid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
David Howellsb0fafa82008-11-14 10:38:41 +11001079 if (lo->lo_encrypt_key_size &&
Eric W. Biedermane4849732012-02-11 11:23:51 -08001080 !uid_eq(lo->lo_key_owner, uid) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 !capable(CAP_SYS_ADMIN))
1082 return -EPERM;
1083 if (lo->lo_state != Lo_bound)
1084 return -ENXIO;
1085 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1086 return -EINVAL;
1087
1088 err = loop_release_xfer(lo);
1089 if (err)
1090 return err;
1091
1092 if (info->lo_encrypt_type) {
1093 unsigned int type = info->lo_encrypt_type;
1094
1095 if (type >= MAX_LO_CRYPT)
1096 return -EINVAL;
1097 xfer = xfer_funcs[type];
1098 if (xfer == NULL)
1099 return -EINVAL;
1100 } else
1101 xfer = NULL;
1102
1103 err = loop_init_xfer(lo, xfer, info);
1104 if (err)
1105 return err;
1106
1107 if (lo->lo_offset != info->lo_offset ||
Guo Chao7b0576a2013-02-21 15:16:47 -08001108 lo->lo_sizelimit != info->lo_sizelimit)
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001109 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 return -EFBIG;
Guo Chao541c7422013-02-21 15:16:46 -08001111
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001112 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1115 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1116 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1117 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1118
1119 if (!xfer)
1120 xfer = &none_funcs;
1121 lo->transfer = xfer->transfer;
1122 lo->ioctl = xfer->ioctl;
1123
David Woodhouse96c58652008-02-06 01:36:27 -08001124 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1125 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1126 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1127
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001128 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1129 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1130 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1131 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1132 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1133 }
1134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1136 lo->lo_init[0] = info->lo_init[0];
1137 lo->lo_init[1] = info->lo_init[1];
1138 if (info->lo_encrypt_key_size) {
1139 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1140 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001141 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
1143
1144 return 0;
1145}
1146
1147static int
1148loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1149{
1150 struct file *file = lo->lo_backing_file;
1151 struct kstat stat;
1152 int error;
1153
1154 if (lo->lo_state != Lo_bound)
1155 return -ENXIO;
Al Viro3dadecc2013-01-24 02:18:08 -05001156 error = vfs_getattr(&file->f_path, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (error)
1158 return error;
1159 memset(info, 0, sizeof(*info));
1160 info->lo_number = lo->lo_number;
1161 info->lo_device = huge_encode_dev(stat.dev);
1162 info->lo_inode = stat.ino;
1163 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1164 info->lo_offset = lo->lo_offset;
1165 info->lo_sizelimit = lo->lo_sizelimit;
1166 info->lo_flags = lo->lo_flags;
1167 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1168 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1169 info->lo_encrypt_type =
1170 lo->lo_encryption ? lo->lo_encryption->number : 0;
1171 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1172 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1173 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1174 lo->lo_encrypt_key_size);
1175 }
1176 return 0;
1177}
1178
1179static void
1180loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1181{
1182 memset(info64, 0, sizeof(*info64));
1183 info64->lo_number = info->lo_number;
1184 info64->lo_device = info->lo_device;
1185 info64->lo_inode = info->lo_inode;
1186 info64->lo_rdevice = info->lo_rdevice;
1187 info64->lo_offset = info->lo_offset;
1188 info64->lo_sizelimit = 0;
1189 info64->lo_encrypt_type = info->lo_encrypt_type;
1190 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1191 info64->lo_flags = info->lo_flags;
1192 info64->lo_init[0] = info->lo_init[0];
1193 info64->lo_init[1] = info->lo_init[1];
1194 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1195 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1196 else
1197 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1198 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1199}
1200
1201static int
1202loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1203{
1204 memset(info, 0, sizeof(*info));
1205 info->lo_number = info64->lo_number;
1206 info->lo_device = info64->lo_device;
1207 info->lo_inode = info64->lo_inode;
1208 info->lo_rdevice = info64->lo_rdevice;
1209 info->lo_offset = info64->lo_offset;
1210 info->lo_encrypt_type = info64->lo_encrypt_type;
1211 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1212 info->lo_flags = info64->lo_flags;
1213 info->lo_init[0] = info64->lo_init[0];
1214 info->lo_init[1] = info64->lo_init[1];
1215 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1216 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1217 else
1218 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1219 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1220
1221 /* error in case values were truncated */
1222 if (info->lo_device != info64->lo_device ||
1223 info->lo_rdevice != info64->lo_rdevice ||
1224 info->lo_inode != info64->lo_inode ||
1225 info->lo_offset != info64->lo_offset)
1226 return -EOVERFLOW;
1227
1228 return 0;
1229}
1230
1231static int
1232loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1233{
1234 struct loop_info info;
1235 struct loop_info64 info64;
1236
1237 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1238 return -EFAULT;
1239 loop_info64_from_old(&info, &info64);
1240 return loop_set_status(lo, &info64);
1241}
1242
1243static int
1244loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1245{
1246 struct loop_info64 info64;
1247
1248 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1249 return -EFAULT;
1250 return loop_set_status(lo, &info64);
1251}
1252
1253static int
1254loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1255 struct loop_info info;
1256 struct loop_info64 info64;
1257 int err = 0;
1258
1259 if (!arg)
1260 err = -EINVAL;
1261 if (!err)
1262 err = loop_get_status(lo, &info64);
1263 if (!err)
1264 err = loop_info64_to_old(&info64, &info);
1265 if (!err && copy_to_user(arg, &info, sizeof(info)))
1266 err = -EFAULT;
1267
1268 return err;
1269}
1270
1271static int
1272loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1273 struct loop_info64 info64;
1274 int err = 0;
1275
1276 if (!arg)
1277 err = -EINVAL;
1278 if (!err)
1279 err = loop_get_status(lo, &info64);
1280 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1281 err = -EFAULT;
1282
1283 return err;
1284}
1285
J. R. Okajima53d66602009-03-31 15:23:43 -07001286static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1287{
J. R. Okajima53d66602009-03-31 15:23:43 -07001288 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001289 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001290
Guo Chao7b0576a2013-02-21 15:16:47 -08001291 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001292}
1293
Al Virobb214882008-03-02 09:29:48 -05001294static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 unsigned int cmd, unsigned long arg)
1296{
Al Virobb214882008-03-02 09:29:48 -05001297 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 int err;
1299
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001300 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 switch (cmd) {
1302 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001303 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 break;
1305 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001306 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 break;
1308 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001309 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Ayan George4c823cc2011-09-21 10:02:13 +02001310 err = loop_clr_fd(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001311 if (!err)
1312 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 break;
1314 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001315 err = -EPERM;
1316 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1317 err = loop_set_status_old(lo,
1318 (struct loop_info __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 break;
1320 case LOOP_GET_STATUS:
1321 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1322 break;
1323 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001324 err = -EPERM;
1325 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1326 err = loop_set_status64(lo,
1327 (struct loop_info64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 break;
1329 case LOOP_GET_STATUS64:
1330 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1331 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001332 case LOOP_SET_CAPACITY:
1333 err = -EPERM;
1334 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1335 err = loop_set_capacity(lo, bdev);
1336 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 default:
1338 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1339 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001340 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001341
1342out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 return err;
1344}
1345
David Howells863d5b822006-08-29 19:06:14 +01001346#ifdef CONFIG_COMPAT
1347struct compat_loop_info {
1348 compat_int_t lo_number; /* ioctl r/o */
1349 compat_dev_t lo_device; /* ioctl r/o */
1350 compat_ulong_t lo_inode; /* ioctl r/o */
1351 compat_dev_t lo_rdevice; /* ioctl r/o */
1352 compat_int_t lo_offset;
1353 compat_int_t lo_encrypt_type;
1354 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1355 compat_int_t lo_flags; /* ioctl r/o */
1356 char lo_name[LO_NAME_SIZE];
1357 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1358 compat_ulong_t lo_init[2];
1359 char reserved[4];
1360};
1361
1362/*
1363 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1364 * - noinlined to reduce stack space usage in main part of driver
1365 */
1366static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001367loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001368 struct loop_info64 *info64)
1369{
1370 struct compat_loop_info info;
1371
1372 if (copy_from_user(&info, arg, sizeof(info)))
1373 return -EFAULT;
1374
1375 memset(info64, 0, sizeof(*info64));
1376 info64->lo_number = info.lo_number;
1377 info64->lo_device = info.lo_device;
1378 info64->lo_inode = info.lo_inode;
1379 info64->lo_rdevice = info.lo_rdevice;
1380 info64->lo_offset = info.lo_offset;
1381 info64->lo_sizelimit = 0;
1382 info64->lo_encrypt_type = info.lo_encrypt_type;
1383 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1384 info64->lo_flags = info.lo_flags;
1385 info64->lo_init[0] = info.lo_init[0];
1386 info64->lo_init[1] = info.lo_init[1];
1387 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1388 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1389 else
1390 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1391 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1392 return 0;
1393}
1394
1395/*
1396 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1397 * - noinlined to reduce stack space usage in main part of driver
1398 */
1399static noinline int
1400loop_info64_to_compat(const struct loop_info64 *info64,
1401 struct compat_loop_info __user *arg)
1402{
1403 struct compat_loop_info info;
1404
1405 memset(&info, 0, sizeof(info));
1406 info.lo_number = info64->lo_number;
1407 info.lo_device = info64->lo_device;
1408 info.lo_inode = info64->lo_inode;
1409 info.lo_rdevice = info64->lo_rdevice;
1410 info.lo_offset = info64->lo_offset;
1411 info.lo_encrypt_type = info64->lo_encrypt_type;
1412 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1413 info.lo_flags = info64->lo_flags;
1414 info.lo_init[0] = info64->lo_init[0];
1415 info.lo_init[1] = info64->lo_init[1];
1416 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1417 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1418 else
1419 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1420 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1421
1422 /* error in case values were truncated */
1423 if (info.lo_device != info64->lo_device ||
1424 info.lo_rdevice != info64->lo_rdevice ||
1425 info.lo_inode != info64->lo_inode ||
1426 info.lo_offset != info64->lo_offset ||
1427 info.lo_init[0] != info64->lo_init[0] ||
1428 info.lo_init[1] != info64->lo_init[1])
1429 return -EOVERFLOW;
1430
1431 if (copy_to_user(arg, &info, sizeof(info)))
1432 return -EFAULT;
1433 return 0;
1434}
1435
1436static int
1437loop_set_status_compat(struct loop_device *lo,
1438 const struct compat_loop_info __user *arg)
1439{
1440 struct loop_info64 info64;
1441 int ret;
1442
1443 ret = loop_info64_from_compat(arg, &info64);
1444 if (ret < 0)
1445 return ret;
1446 return loop_set_status(lo, &info64);
1447}
1448
1449static int
1450loop_get_status_compat(struct loop_device *lo,
1451 struct compat_loop_info __user *arg)
1452{
1453 struct loop_info64 info64;
1454 int err = 0;
1455
1456 if (!arg)
1457 err = -EINVAL;
1458 if (!err)
1459 err = loop_get_status(lo, &info64);
1460 if (!err)
1461 err = loop_info64_to_compat(&info64, arg);
1462 return err;
1463}
1464
Al Virobb214882008-03-02 09:29:48 -05001465static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1466 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001467{
Al Virobb214882008-03-02 09:29:48 -05001468 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001469 int err;
1470
David Howells863d5b822006-08-29 19:06:14 +01001471 switch(cmd) {
1472 case LOOP_SET_STATUS:
1473 mutex_lock(&lo->lo_ctl_mutex);
1474 err = loop_set_status_compat(
1475 lo, (const struct compat_loop_info __user *) arg);
1476 mutex_unlock(&lo->lo_ctl_mutex);
1477 break;
1478 case LOOP_GET_STATUS:
1479 mutex_lock(&lo->lo_ctl_mutex);
1480 err = loop_get_status_compat(
1481 lo, (struct compat_loop_info __user *) arg);
1482 mutex_unlock(&lo->lo_ctl_mutex);
1483 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001484 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001485 case LOOP_CLR_FD:
1486 case LOOP_GET_STATUS64:
1487 case LOOP_SET_STATUS64:
1488 arg = (unsigned long) compat_ptr(arg);
1489 case LOOP_SET_FD:
1490 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001491 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001492 break;
1493 default:
1494 err = -ENOIOCTLCMD;
1495 break;
1496 }
David Howells863d5b822006-08-29 19:06:14 +01001497 return err;
1498}
1499#endif
1500
Al Virobb214882008-03-02 09:29:48 -05001501static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Kay Sievers770fe302011-07-31 22:08:04 +02001503 struct loop_device *lo;
1504 int err = 0;
1505
1506 mutex_lock(&loop_index_mutex);
1507 lo = bdev->bd_disk->private_data;
1508 if (!lo) {
1509 err = -ENXIO;
1510 goto out;
1511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Ingo Molnarf85221d2006-03-23 03:00:38 -08001513 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001515 mutex_unlock(&lo->lo_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001516out:
1517 mutex_unlock(&loop_index_mutex);
1518 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
1520
Al Virodb2a1442013-05-05 21:52:57 -04001521static void lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
Al Virobb214882008-03-02 09:29:48 -05001523 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001524 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Ingo Molnarf85221d2006-03-23 03:00:38 -08001526 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001527
Milan Broz14f27932008-12-12 14:48:27 +01001528 if (--lo->lo_refcnt)
1529 goto out;
1530
1531 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1532 /*
1533 * In autoclear mode, stop the loop thread
1534 * and remove configuration after last close.
1535 */
Ayan George4c823cc2011-09-21 10:02:13 +02001536 err = loop_clr_fd(lo);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001537 if (!err)
Al Virodb2a1442013-05-05 21:52:57 -04001538 return;
Milan Broz14f27932008-12-12 14:48:27 +01001539 } else {
1540 /*
1541 * Otherwise keep thread (if running) and config,
1542 * but flush possible ongoing bios in thread.
1543 */
1544 loop_flush(lo);
1545 }
David Woodhouse96c58652008-02-06 01:36:27 -08001546
Milan Broz14f27932008-12-12 14:48:27 +01001547out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001548 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001551static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001553 .open = lo_open,
1554 .release = lo_release,
1555 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001556#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001557 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001558#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559};
1560
1561/*
1562 * And now the modules code and kernel interface.
1563 */
Ken Chen73285082007-05-08 00:28:20 -07001564static int max_loop;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001565module_param(max_loop, int, S_IRUGO);
Ken Chena47653f2007-06-08 13:46:44 -07001566MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Namhyung Kimac04fee2011-05-27 07:59:25 +02001567module_param(max_part, int, S_IRUGO);
Laurent Vivier476a4812008-03-26 12:11:53 +01001568MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569MODULE_LICENSE("GPL");
1570MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1571
1572int loop_register_transfer(struct loop_func_table *funcs)
1573{
1574 unsigned int n = funcs->number;
1575
1576 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1577 return -EINVAL;
1578 xfer_funcs[n] = funcs;
1579 return 0;
1580}
1581
Kay Sievers34dd82a2011-07-31 22:08:04 +02001582static int unregister_transfer_cb(int id, void *ptr, void *data)
1583{
1584 struct loop_device *lo = ptr;
1585 struct loop_func_table *xfer = data;
1586
1587 mutex_lock(&lo->lo_ctl_mutex);
1588 if (lo->lo_encryption == xfer)
1589 loop_release_xfer(lo);
1590 mutex_unlock(&lo->lo_ctl_mutex);
1591 return 0;
1592}
1593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594int loop_unregister_transfer(int number)
1595{
1596 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 struct loop_func_table *xfer;
1598
1599 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1600 return -EINVAL;
1601
1602 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001603 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 return 0;
1605}
1606
1607EXPORT_SYMBOL(loop_register_transfer);
1608EXPORT_SYMBOL(loop_unregister_transfer);
1609
Kay Sievers34dd82a2011-07-31 22:08:04 +02001610static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001611{
1612 struct loop_device *lo;
1613 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001614 int err;
Ken Chen73285082007-05-08 00:28:20 -07001615
Silva Paulo68d740d2012-07-14 15:39:58 -07001616 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001617 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07001618 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07001619 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001620
Tejun Heoc718aa62013-02-27 17:03:58 -08001621 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001622 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08001623 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1624 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02001625 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001626 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08001627 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001628 }
1629 if (err < 0)
1630 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08001631 i = err;
Ken Chen73285082007-05-08 00:28:20 -07001632
Wei Yongjun183cfb52013-03-22 08:59:19 -06001633 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001634 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1635 if (!lo->lo_queue)
1636 goto out_free_dev;
1637
Laurent Vivier476a4812008-03-26 12:11:53 +01001638 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001639 if (!disk)
1640 goto out_free_queue;
1641
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001642 /*
1643 * Disable partition scanning by default. The in-kernel partition
1644 * scanning can be requested individually per-device during its
1645 * setup. Userspace can always add and remove partitions from all
1646 * devices. The needed partition minors are allocated from the
1647 * extended minor space, the main loop device numbers will continue
1648 * to match the loop minors, regardless of the number of partitions
1649 * used.
1650 *
1651 * If max_part is given, partition scanning is globally enabled for
1652 * all loop devices. The minors for the main loop devices will be
1653 * multiples of max_part.
1654 *
1655 * Note: Global-for-all-devices, set-only-at-init, read-only module
1656 * parameteters like 'max_loop' and 'max_part' make things needlessly
1657 * complicated, are too static, inflexible and may surprise
1658 * userspace tools. Parameters like this in general should be avoided.
1659 */
1660 if (!part_shift)
1661 disk->flags |= GENHD_FL_NO_PART_SCAN;
1662 disk->flags |= GENHD_FL_EXT_DEVT;
Ken Chen73285082007-05-08 00:28:20 -07001663 mutex_init(&lo->lo_ctl_mutex);
1664 lo->lo_number = i;
1665 lo->lo_thread = NULL;
1666 init_waitqueue_head(&lo->lo_event);
Lukas Czerner7b5a3522012-11-30 11:42:41 +01001667 init_waitqueue_head(&lo->lo_req_wait);
Ken Chen73285082007-05-08 00:28:20 -07001668 spin_lock_init(&lo->lo_lock);
1669 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001670 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001671 disk->fops = &lo_fops;
1672 disk->private_data = lo;
1673 disk->queue = lo->lo_queue;
1674 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001675 add_disk(disk);
1676 *l = lo;
1677 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001678
1679out_free_queue:
1680 blk_cleanup_queue(lo->lo_queue);
1681out_free_dev:
1682 kfree(lo);
1683out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001684 return err;
Ken Chen73285082007-05-08 00:28:20 -07001685}
1686
Kay Sievers34dd82a2011-07-31 22:08:04 +02001687static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001688{
Kay Sievers34dd82a2011-07-31 22:08:04 +02001689 del_gendisk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001690 blk_cleanup_queue(lo->lo_queue);
1691 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001692 kfree(lo);
1693}
1694
Kay Sievers770fe302011-07-31 22:08:04 +02001695static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001696{
Kay Sievers770fe302011-07-31 22:08:04 +02001697 struct loop_device *lo = ptr;
1698 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001699
Kay Sievers770fe302011-07-31 22:08:04 +02001700 if (lo->lo_state == Lo_unbound) {
1701 *l = lo;
1702 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001703 }
Kay Sievers770fe302011-07-31 22:08:04 +02001704 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001705}
1706
Kay Sievers34dd82a2011-07-31 22:08:04 +02001707static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001708{
Ken Chena47653f2007-06-08 13:46:44 -07001709 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001710 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001711
Kay Sievers770fe302011-07-31 22:08:04 +02001712 if (i < 0) {
1713 int err;
1714
1715 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1716 if (err == 1) {
1717 *l = lo;
1718 ret = lo->lo_number;
1719 }
1720 goto out;
1721 }
1722
1723 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001724 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07001725 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02001726 *l = lo;
1727 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07001728 }
Kay Sievers770fe302011-07-31 22:08:04 +02001729out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001730 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07001731}
1732
Ken Chen73285082007-05-08 00:28:20 -07001733static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1734{
Al Viro705962c2007-05-13 05:52:32 -04001735 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001736 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001737 int err;
Ken Chen73285082007-05-08 00:28:20 -07001738
Kay Sievers34dd82a2011-07-31 22:08:04 +02001739 mutex_lock(&loop_index_mutex);
1740 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1741 if (err < 0)
1742 err = loop_add(&lo, MINOR(dev) >> part_shift);
1743 if (err < 0)
1744 kobj = ERR_PTR(err);
1745 else
1746 kobj = get_disk(lo->lo_disk);
1747 mutex_unlock(&loop_index_mutex);
Ken Chen73285082007-05-08 00:28:20 -07001748
1749 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001750 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001751}
1752
Kay Sievers770fe302011-07-31 22:08:04 +02001753static long loop_control_ioctl(struct file *file, unsigned int cmd,
1754 unsigned long parm)
1755{
1756 struct loop_device *lo;
1757 int ret = -ENOSYS;
1758
1759 mutex_lock(&loop_index_mutex);
1760 switch (cmd) {
1761 case LOOP_CTL_ADD:
1762 ret = loop_lookup(&lo, parm);
1763 if (ret >= 0) {
1764 ret = -EEXIST;
1765 break;
1766 }
1767 ret = loop_add(&lo, parm);
1768 break;
1769 case LOOP_CTL_REMOVE:
1770 ret = loop_lookup(&lo, parm);
1771 if (ret < 0)
1772 break;
1773 mutex_lock(&lo->lo_ctl_mutex);
1774 if (lo->lo_state != Lo_unbound) {
1775 ret = -EBUSY;
1776 mutex_unlock(&lo->lo_ctl_mutex);
1777 break;
1778 }
1779 if (lo->lo_refcnt > 0) {
1780 ret = -EBUSY;
1781 mutex_unlock(&lo->lo_ctl_mutex);
1782 break;
1783 }
1784 lo->lo_disk->private_data = NULL;
1785 mutex_unlock(&lo->lo_ctl_mutex);
1786 idr_remove(&loop_index_idr, lo->lo_number);
1787 loop_remove(lo);
1788 break;
1789 case LOOP_CTL_GET_FREE:
1790 ret = loop_lookup(&lo, -1);
1791 if (ret >= 0)
1792 break;
1793 ret = loop_add(&lo, -1);
1794 }
1795 mutex_unlock(&loop_index_mutex);
1796
1797 return ret;
1798}
1799
1800static const struct file_operations loop_ctl_fops = {
1801 .open = nonseekable_open,
1802 .unlocked_ioctl = loop_control_ioctl,
1803 .compat_ioctl = loop_control_ioctl,
1804 .owner = THIS_MODULE,
1805 .llseek = noop_llseek,
1806};
1807
1808static struct miscdevice loop_misc = {
1809 .minor = LOOP_CTRL_MINOR,
1810 .name = "loop-control",
1811 .fops = &loop_ctl_fops,
1812};
1813
1814MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1815MODULE_ALIAS("devname:loop-control");
1816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817static int __init loop_init(void)
1818{
Ken Chena47653f2007-06-08 13:46:44 -07001819 int i, nr;
1820 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001821 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02001822 int err;
Ken Chena47653f2007-06-08 13:46:44 -07001823
Kay Sievers770fe302011-07-31 22:08:04 +02001824 err = misc_register(&loop_misc);
1825 if (err < 0)
1826 return err;
Laurent Vivier476a4812008-03-26 12:11:53 +01001827
1828 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001829 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01001830 part_shift = fls(max_part);
1831
Namhyung Kimac04fee2011-05-27 07:59:25 +02001832 /*
1833 * Adjust max_part according to part_shift as it is exported
1834 * to user space so that user can decide correct minor number
1835 * if [s]he want to create more devices.
1836 *
1837 * Note that -1 is required because partition 0 is reserved
1838 * for the whole disk.
1839 */
1840 max_part = (1UL << part_shift) - 1;
1841 }
1842
Guo Chaob1a66502013-02-21 15:16:49 -08001843 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1844 err = -EINVAL;
1845 goto misc_out;
1846 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02001847
Guo Chaob1a66502013-02-21 15:16:49 -08001848 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1849 err = -EINVAL;
1850 goto misc_out;
1851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Kay Sieversd134b002011-07-31 22:08:04 +02001853 /*
1854 * If max_loop is specified, create that many devices upfront.
1855 * This also becomes a hard limit. If max_loop is not specified,
1856 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1857 * init time. Loop devices can be requested on-demand with the
1858 * /dev/loop-control interface, or be instantiated by accessing
1859 * a 'dead' device node.
1860 */
Ken Chen73285082007-05-08 00:28:20 -07001861 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001862 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001863 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07001864 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02001865 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001866 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
Ken Chena47653f2007-06-08 13:46:44 -07001868
Guo Chaob1a66502013-02-21 15:16:49 -08001869 if (register_blkdev(LOOP_MAJOR, "loop")) {
1870 err = -EIO;
1871 goto misc_out;
1872 }
Ken Chena47653f2007-06-08 13:46:44 -07001873
Ken Chena47653f2007-06-08 13:46:44 -07001874 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1875 THIS_MODULE, loop_probe, NULL, NULL);
1876
Kay Sieversd134b002011-07-31 22:08:04 +02001877 /* pre-create number of devices given by config or max_loop */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001878 mutex_lock(&loop_index_mutex);
1879 for (i = 0; i < nr; i++)
1880 loop_add(&lo, i);
1881 mutex_unlock(&loop_index_mutex);
1882
Ken Chen73285082007-05-08 00:28:20 -07001883 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08001885
1886misc_out:
1887 misc_deregister(&loop_misc);
1888 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001889}
Ken Chena47653f2007-06-08 13:46:44 -07001890
Kay Sievers34dd82a2011-07-31 22:08:04 +02001891static int loop_exit_cb(int id, void *ptr, void *data)
1892{
1893 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07001894
Kay Sievers34dd82a2011-07-31 22:08:04 +02001895 loop_remove(lo);
1896 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897}
1898
Ken Chen73285082007-05-08 00:28:20 -07001899static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900{
Ken Chena47653f2007-06-08 13:46:44 -07001901 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Namhyung Kima1c15c52011-05-24 16:48:55 +02001903 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07001904
Kay Sievers34dd82a2011-07-31 22:08:04 +02001905 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001906 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07001907
Ken Chena47653f2007-06-08 13:46:44 -07001908 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001909 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02001910
1911 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912}
1913
1914module_init(loop_init);
1915module_exit(loop_exit);
1916
1917#ifndef MODULE
1918static int __init max_loop_setup(char *str)
1919{
1920 max_loop = simple_strtol(str, NULL, 0);
1921 return 1;
1922}
1923
1924__setup("max_loop=", max_loop_setup);
1925#endif