blob: bbca966f8f66a20b04f62ac01ff41cd5ef18c037 [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>
66#include <linux/loop.h>
David Howells863d5b822006-08-29 19:06:14 +010067#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070069#include <linux/freezer.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020070#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/completion.h>
73#include <linux/highmem.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070074#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020075#include <linux/splice.h>
Milan Brozee862732010-08-23 15:16:00 +020076#include <linux/sysfs.h>
Kay Sievers770fe302011-07-31 22:08:04 +020077#include <linux/miscdevice.h>
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +020078#include <linux/falloc.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{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100165 loff_t size, loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 loopsize = size - offset;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100170 /* offset is beyond i_size, wierd but possible */
171 if (loopsize < 0)
172 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100174 if (sizelimit > 0 && sizelimit < loopsize)
175 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 /*
177 * Unfortunately, if we want to do I/O on the device,
178 * the number of 512-byte sectors has to fit into a sector_t.
179 */
180 return loopsize >> 9;
181}
182
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100183static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100185 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
186}
187
188static int
189figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
190{
191 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 sector_t x = (sector_t)size;
193
194 if (unlikely((loff_t)x != size))
195 return -EFBIG;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100196 if (lo->lo_offset != offset)
197 lo->lo_offset = offset;
198 if (lo->lo_sizelimit != sizelimit)
199 lo->lo_sizelimit = sizelimit;
Ken Chen73285082007-05-08 00:28:20 -0700200 set_capacity(lo->lo_disk, x);
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100201 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203
204static inline int
205lo_do_transfer(struct loop_device *lo, int cmd,
206 struct page *rpage, unsigned roffs,
207 struct page *lpage, unsigned loffs,
208 int size, sector_t rblock)
209{
210 if (unlikely(!lo->transfer))
211 return 0;
212
213 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
214}
215
216/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 * __do_lo_send_write - helper for writing data to a loop device
218 *
219 * This helper just factors out common code between do_lo_send_direct_write()
220 * and do_lo_send_write().
221 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800222static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100223 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
225 ssize_t bw;
226 mm_segment_t old_fs = get_fs();
227
228 set_fs(get_ds());
229 bw = file->f_op->write(file, buf, len, &pos);
230 set_fs(old_fs);
231 if (likely(bw == len))
232 return 0;
233 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
234 (unsigned long long)pos, len);
235 if (bw >= 0)
236 bw = -EIO;
237 return bw;
238}
239
240/**
241 * do_lo_send_direct_write - helper for writing data to a loop device
242 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200243 * This is the fast, non-transforming version that does not need double
244 * buffering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 */
246static int do_lo_send_direct_write(struct loop_device *lo,
Al Viro511de732007-10-08 12:10:13 -0400247 struct bio_vec *bvec, loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100250 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 bvec->bv_len, pos);
252 kunmap(bvec->bv_page);
253 cond_resched();
254 return bw;
255}
256
257/**
258 * do_lo_send_write - helper for writing data to a loop device
259 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200260 * This is the slow, transforming version that needs to double buffer the
261 * data as it cannot do the transformations in place without having direct
262 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
264static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400265 loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
268 bvec->bv_offset, bvec->bv_len, pos >> 9);
269 if (likely(!ret))
270 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100271 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 pos);
273 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
274 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
275 if (ret > 0)
276 ret = -EIO;
277 return ret;
278}
279
Al Viro511de732007-10-08 12:10:13 -0400280static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Al Viro511de732007-10-08 12:10:13 -0400282 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 struct page *page);
284 struct bio_vec *bvec;
285 struct page *page = NULL;
286 int i, ret = 0;
287
Christoph Hellwig456be142011-10-17 12:57:20 +0200288 if (lo->transfer != transfer_none) {
289 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
290 if (unlikely(!page))
291 goto fail;
292 kmap(page);
293 do_lo_send = do_lo_send_write;
294 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 do_lo_send = do_lo_send_direct_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
Christoph Hellwig456be142011-10-17 12:57:20 +0200297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 bio_for_each_segment(bvec, bio, i) {
Al Viro511de732007-10-08 12:10:13 -0400299 ret = do_lo_send(lo, bvec, pos, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 if (ret < 0)
301 break;
302 pos += bvec->bv_len;
303 }
304 if (page) {
305 kunmap(page);
306 __free_page(page);
307 }
308out:
309 return ret;
310fail:
311 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
312 ret = -ENOMEM;
313 goto out;
314}
315
316struct lo_read_data {
317 struct loop_device *lo;
318 struct page *page;
319 unsigned offset;
320 int bsize;
321};
322
323static int
Jens Axboefd582142007-06-12 21:20:37 +0200324lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
325 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
Jens Axboefd582142007-06-12 21:20:37 +0200327 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200329 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 sector_t IV;
Jens Axboe3603b8e2010-12-20 09:15:19 +0100331 int size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Jens Axboefd582142007-06-12 21:20:37 +0200333 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
334 (buf->offset >> 9);
335 size = sd->len;
336 if (size > p->bsize)
337 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Jens Axboefd582142007-06-12 21:20:37 +0200339 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 printk(KERN_ERR "loop: transfer error block %ld\n",
341 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200342 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
344
345 flush_dcache_page(p->page);
346
Jens Axboefd582142007-06-12 21:20:37 +0200347 if (size > 0)
348 p->offset += size;
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return size;
351}
352
353static int
Jens Axboefd582142007-06-12 21:20:37 +0200354lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
355{
356 return __splice_from_pipe(pipe, sd, lo_splice_actor);
357}
358
Dave Young306df072012-02-08 22:07:19 +0100359static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360do_lo_receive(struct loop_device *lo,
361 struct bio_vec *bvec, int bsize, loff_t pos)
362{
363 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200364 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 struct file *file;
Dave Young306df072012-02-08 22:07:19 +0100366 ssize_t retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 cookie.lo = lo;
369 cookie.page = bvec->bv_page;
370 cookie.offset = bvec->bv_offset;
371 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200372
373 sd.len = 0;
374 sd.total_len = bvec->bv_len;
375 sd.flags = 0;
376 sd.pos = pos;
377 sd.u.data = &cookie;
378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200380 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
381
Dave Young306df072012-02-08 22:07:19 +0100382 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385static int
386lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
387{
388 struct bio_vec *bvec;
Dave Young306df072012-02-08 22:07:19 +0100389 ssize_t s;
390 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 bio_for_each_segment(bvec, bio, i) {
Dave Young306df072012-02-08 22:07:19 +0100393 s = do_lo_receive(lo, bvec, bsize, pos);
394 if (s < 0)
395 return s;
396
397 if (s != bvec->bv_len) {
398 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 break;
Dave Young306df072012-02-08 22:07:19 +0100400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 pos += bvec->bv_len;
402 }
Dave Young306df072012-02-08 22:07:19 +0100403 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
406static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
407{
408 loff_t pos;
409 int ret;
410
411 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100412
413 if (bio_rw(bio) == WRITE) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100414 struct file *file = lo->lo_backing_file;
415
Tejun Heo6259f282010-09-03 11:56:17 +0200416 if (bio->bi_rw & REQ_FLUSH) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100417 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200418 if (unlikely(ret && ret != -EINVAL)) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100419 ret = -EIO;
420 goto out;
421 }
422 }
423
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200424 /*
425 * We use punch hole to reclaim the free space used by the
Dave Youngae957572011-11-25 09:41:25 +0100426 * image a.k.a. discard. However we do not support discard if
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200427 * encryption is enabled, because it may give an attacker
428 * useful information.
429 */
430 if (bio->bi_rw & REQ_DISCARD) {
431 struct file *file = lo->lo_backing_file;
432 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
433
434 if ((!file->f_op->fallocate) ||
435 lo->lo_encrypt_key_size) {
436 ret = -EOPNOTSUPP;
437 goto out;
438 }
439 ret = file->f_op->fallocate(file, mode, pos,
440 bio->bi_size);
441 if (unlikely(ret && ret != -EINVAL &&
442 ret != -EOPNOTSUPP))
443 ret = -EIO;
444 goto out;
445 }
446
Al Viro511de732007-10-08 12:10:13 -0400447 ret = lo_send(lo, bio, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100448
Tejun Heo6259f282010-09-03 11:56:17 +0200449 if ((bio->bi_rw & REQ_FUA) && !ret) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100450 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200451 if (unlikely(ret && ret != -EINVAL))
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100452 ret = -EIO;
453 }
454 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100456
457out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 return ret;
459}
460
461/*
462 * Add bio to back of pending list
463 */
464static void loop_add_bio(struct loop_device *lo, struct bio *bio)
465{
Akinobu Mitae6863072009-04-17 08:41:21 +0200466 bio_list_add(&lo->lo_bio_list, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/*
470 * Grab first pending buffer
471 */
472static struct bio *loop_get_bio(struct loop_device *lo)
473{
Akinobu Mitae6863072009-04-17 08:41:21 +0200474 return bio_list_pop(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200477static void loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 struct loop_device *lo = q->queuedata;
480 int rw = bio_rw(old_bio);
481
Nick Piggin35a82d12005-06-23 00:09:06 -0700482 if (rw == READA)
483 rw = READ;
484
485 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 spin_lock_irq(&lo->lo_lock);
488 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700489 goto out;
490 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
491 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700493 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700494 spin_unlock_irq(&lo->lo_lock);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200495 return;
Nick Piggin35a82d12005-06-23 00:09:06 -0700496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700498 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200499 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500}
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502struct switch_request {
503 struct file *file;
504 struct completion wait;
505};
506
507static void do_loop_switch(struct loop_device *, struct switch_request *);
508
509static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
510{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (unlikely(!bio->bi_bdev)) {
512 do_loop_switch(lo, bio->bi_private);
513 bio_put(bio);
514 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700515 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200516 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 }
518}
519
520/*
521 * worker thread that handles reads/writes to file backed loop devices,
522 * to avoid blocking in our make_request_fn. it also does loop decrypting
523 * on reads for block backed loop, as that is too heavy to do from
524 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700525 *
526 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
527 * calling kthread_stop(). Therefore once kthread_should_stop() is
528 * true, make_request will not place any more requests. Therefore
529 * once kthread_should_stop() is true and lo_bio is NULL, we are
530 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 */
532static int loop_thread(void *data)
533{
534 struct loop_device *lo = data;
535 struct bio *bio;
536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 set_user_nice(current, -20);
538
Akinobu Mitae6863072009-04-17 08:41:21 +0200539 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700541 wait_event_interruptible(lo->lo_event,
Akinobu Mitae6863072009-04-17 08:41:21 +0200542 !bio_list_empty(&lo->lo_bio_list) ||
543 kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700544
Akinobu Mitae6863072009-04-17 08:41:21 +0200545 if (bio_list_empty(&lo->lo_bio_list))
Nick Piggin35a82d12005-06-23 00:09:06 -0700546 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700547 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700549 spin_unlock_irq(&lo->lo_lock);
550
551 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 return 0;
556}
557
558/*
559 * loop_switch performs the hard work of switching a backing store.
560 * First it needs to flush existing IO, it does this by sending a magic
561 * BIO down the pipe. The completion of this BIO does the actual switch.
562 */
563static int loop_switch(struct loop_device *lo, struct file *file)
564{
565 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100566 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (!bio)
568 return -ENOMEM;
569 init_completion(&w.wait);
570 w.file = file;
571 bio->bi_private = &w;
572 bio->bi_bdev = NULL;
573 loop_make_request(lo->lo_queue, bio);
574 wait_for_completion(&w.wait);
575 return 0;
576}
577
578/*
Milan Broz14f27932008-12-12 14:48:27 +0100579 * Helper to flush the IOs in loop, but keeping loop thread running
580 */
581static int loop_flush(struct loop_device *lo)
582{
583 /* loop not yet configured, no running thread, nothing to flush */
584 if (!lo->lo_thread)
585 return 0;
586
587 return loop_switch(lo, NULL);
588}
589
590/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 * Do the actual switch; called from the BIO completion routine
592 */
593static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
594{
595 struct file *file = p->file;
596 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100597 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Milan Broz14f27932008-12-12 14:48:27 +0100599 /* if no new file, only flush of queued bios requested */
600 if (!file)
601 goto out;
602
603 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
605 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700606 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
607 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 lo->old_gfp_mask = mapping_gfp_mask(mapping);
609 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100610out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 complete(&p->wait);
612}
613
614
615/*
616 * loop_change_fd switched the backing store of a loopback device to
617 * a new file. This is useful for operating system installers to free up
618 * the original file and in High Availability environments to switch to
619 * an alternative location for the content in case of server meltdown.
620 * This can only work if the loop device is used read-only, and if the
621 * new backing store is the same size and type as the old backing store.
622 */
Al Virobb214882008-03-02 09:29:48 -0500623static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
624 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
626 struct file *file, *old_file;
627 struct inode *inode;
628 int error;
629
630 error = -ENXIO;
631 if (lo->lo_state != Lo_bound)
632 goto out;
633
634 /* the loop device has to be read-only */
635 error = -EINVAL;
636 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
637 goto out;
638
639 error = -EBADF;
640 file = fget(arg);
641 if (!file)
642 goto out;
643
644 inode = file->f_mapping->host;
645 old_file = lo->lo_backing_file;
646
647 error = -EINVAL;
648
649 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
650 goto out_putf;
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 /* size of the new backing store needs to be the same */
653 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
654 goto out_putf;
655
656 /* and ... switch */
657 error = loop_switch(lo, file);
658 if (error)
659 goto out_putf;
660
661 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200662 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100663 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return 0;
665
666 out_putf:
667 fput(file);
668 out:
669 return error;
670}
671
672static inline int is_loop_device(struct file *file)
673{
674 struct inode *i = file->f_mapping->host;
675
676 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
677}
678
Milan Brozee862732010-08-23 15:16:00 +0200679/* loop sysfs attributes */
680
681static ssize_t loop_attr_show(struct device *dev, char *page,
682 ssize_t (*callback)(struct loop_device *, char *))
683{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200684 struct gendisk *disk = dev_to_disk(dev);
685 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200686
Kay Sievers34dd82a2011-07-31 22:08:04 +0200687 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200688}
689
690#define LOOP_ATTR_RO(_name) \
691static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
692static ssize_t loop_attr_do_show_##_name(struct device *d, \
693 struct device_attribute *attr, char *b) \
694{ \
695 return loop_attr_show(d, b, loop_attr_##_name##_show); \
696} \
697static struct device_attribute loop_attr_##_name = \
698 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
699
700static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
701{
702 ssize_t ret;
703 char *p = NULL;
704
Kay Sievers05eb0f22011-07-31 22:21:35 +0200705 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200706 if (lo->lo_backing_file)
707 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200708 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200709
710 if (IS_ERR_OR_NULL(p))
711 ret = PTR_ERR(p);
712 else {
713 ret = strlen(p);
714 memmove(buf, p, ret);
715 buf[ret++] = '\n';
716 buf[ret] = 0;
717 }
718
719 return ret;
720}
721
722static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
723{
724 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
725}
726
727static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
728{
729 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
730}
731
732static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
733{
734 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
735
736 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
737}
738
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200739static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
740{
741 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
742
743 return sprintf(buf, "%s\n", partscan ? "1" : "0");
744}
745
Milan Brozee862732010-08-23 15:16:00 +0200746LOOP_ATTR_RO(backing_file);
747LOOP_ATTR_RO(offset);
748LOOP_ATTR_RO(sizelimit);
749LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200750LOOP_ATTR_RO(partscan);
Milan Brozee862732010-08-23 15:16:00 +0200751
752static struct attribute *loop_attrs[] = {
753 &loop_attr_backing_file.attr,
754 &loop_attr_offset.attr,
755 &loop_attr_sizelimit.attr,
756 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200757 &loop_attr_partscan.attr,
Milan Brozee862732010-08-23 15:16:00 +0200758 NULL,
759};
760
761static struct attribute_group loop_attribute_group = {
762 .name = "loop",
763 .attrs= loop_attrs,
764};
765
766static int loop_sysfs_init(struct loop_device *lo)
767{
768 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
769 &loop_attribute_group);
770}
771
772static void loop_sysfs_exit(struct loop_device *lo)
773{
774 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
775 &loop_attribute_group);
776}
777
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200778static void loop_config_discard(struct loop_device *lo)
779{
780 struct file *file = lo->lo_backing_file;
781 struct inode *inode = file->f_mapping->host;
782 struct request_queue *q = lo->lo_queue;
783
784 /*
785 * We use punch hole to reclaim the free space used by the
786 * image a.k.a. discard. However we do support discard if
787 * encryption is enabled, because it may give an attacker
788 * useful information.
789 */
790 if ((!file->f_op->fallocate) ||
791 lo->lo_encrypt_key_size) {
792 q->limits.discard_granularity = 0;
793 q->limits.discard_alignment = 0;
794 q->limits.max_discard_sectors = 0;
795 q->limits.discard_zeroes_data = 0;
796 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
797 return;
798 }
799
800 q->limits.discard_granularity = inode->i_sb->s_blocksize;
Lukas Czernerdfaf3c02011-12-02 14:47:03 +0100801 q->limits.discard_alignment = 0;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200802 q->limits.max_discard_sectors = UINT_MAX >> 9;
803 q->limits.discard_zeroes_data = 1;
804 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
805}
806
Al Virobb214882008-03-02 09:29:48 -0500807static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct block_device *bdev, unsigned int arg)
809{
810 struct file *file, *f;
811 struct inode *inode;
812 struct address_space *mapping;
813 unsigned lo_blocksize;
814 int lo_flags = 0;
815 int error;
816 loff_t size;
817
818 /* This is safe, since we have a reference from open(). */
819 __module_get(THIS_MODULE);
820
821 error = -EBADF;
822 file = fget(arg);
823 if (!file)
824 goto out;
825
826 error = -EBUSY;
827 if (lo->lo_state != Lo_unbound)
828 goto out_putf;
829
830 /* Avoid recursion */
831 f = file;
832 while (is_loop_device(f)) {
833 struct loop_device *l;
834
Al Virobb214882008-03-02 09:29:48 -0500835 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 goto out_putf;
837
838 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
839 if (l->lo_state == Lo_unbound) {
840 error = -EINVAL;
841 goto out_putf;
842 }
843 f = l->lo_backing_file;
844 }
845
846 mapping = file->f_mapping;
847 inode = mapping->host;
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 error = -EINVAL;
Christoph Hellwig456be142011-10-17 12:57:20 +0200850 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Christoph Hellwig456be142011-10-17 12:57:20 +0200853 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
854 !file->f_op->write)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 lo_flags |= LO_FLAGS_READ_ONLY;
856
Christoph Hellwig456be142011-10-17 12:57:20 +0200857 lo_blocksize = S_ISBLK(inode->i_mode) ?
858 inode->i_bdev->bd_block_size : PAGE_SIZE;
859
860 error = -EFBIG;
861 size = get_loop_size(lo, file);
862 if ((loff_t)(sector_t)size != size)
863 goto out_putf;
864
865 error = 0;
866
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
868
869 lo->lo_blocksize = lo_blocksize;
870 lo->lo_device = bdev;
871 lo->lo_flags = lo_flags;
872 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700873 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 lo->ioctl = NULL;
875 lo->lo_sizelimit = 0;
876 lo->old_gfp_mask = mapping_gfp_mask(mapping);
877 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
878
Akinobu Mitae6863072009-04-17 08:41:21 +0200879 bio_list_init(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 /*
882 * set queue make_request_fn, and add limits based on lower level
883 * device
884 */
885 blk_queue_make_request(lo->lo_queue, loop_make_request);
886 lo->lo_queue->queuedata = lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100888 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Tejun Heo4913efe2010-09-03 11:56:16 +0200889 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100890
Ken Chen73285082007-05-08 00:28:20 -0700891 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200893 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200894 /* let user-space know about the new size */
895 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 set_blocksize(bdev, lo_blocksize);
898
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700899 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
900 lo->lo_number);
901 if (IS_ERR(lo->lo_thread)) {
902 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700903 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700904 }
905 lo->lo_state = Lo_bound;
906 wake_up_process(lo->lo_thread);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200907 if (part_shift)
908 lo->lo_flags |= LO_FLAGS_PARTSCAN;
909 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100910 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 return 0;
912
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700913out_clr:
Milan Brozee862732010-08-23 15:16:00 +0200914 loop_sysfs_exit(lo);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700915 lo->lo_thread = NULL;
916 lo->lo_device = NULL;
917 lo->lo_backing_file = NULL;
918 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700919 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700920 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700921 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200922 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700923 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
924 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 out_putf:
926 fput(file);
927 out:
928 /* This is safe: open() is still holding a reference. */
929 module_put(THIS_MODULE);
930 return error;
931}
932
933static int
934loop_release_xfer(struct loop_device *lo)
935{
936 int err = 0;
937 struct loop_func_table *xfer = lo->lo_encryption;
938
939 if (xfer) {
940 if (xfer->release)
941 err = xfer->release(lo);
942 lo->transfer = NULL;
943 lo->lo_encryption = NULL;
944 module_put(xfer->owner);
945 }
946 return err;
947}
948
949static int
950loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
951 const struct loop_info64 *i)
952{
953 int err = 0;
954
955 if (xfer) {
956 struct module *owner = xfer->owner;
957
958 if (!try_module_get(owner))
959 return -EINVAL;
960 if (xfer->init)
961 err = xfer->init(lo, i);
962 if (err)
963 module_put(owner);
964 else
965 lo->lo_encryption = xfer;
966 }
967 return err;
968}
969
Ayan George4c823cc2011-09-21 10:02:13 +0200970static int loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
972 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400973 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +0200974 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 if (lo->lo_state != Lo_bound)
977 return -ENXIO;
978
979 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
980 return -EBUSY;
981
982 if (filp == NULL)
983 return -EINVAL;
984
985 spin_lock_irq(&lo->lo_lock);
986 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 spin_unlock_irq(&lo->lo_lock);
988
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700989 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Kay Sievers05eb0f22011-07-31 22:21:35 +0200991 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +0200993 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 loop_release_xfer(lo);
996 lo->transfer = NULL;
997 lo->ioctl = NULL;
998 lo->lo_device = NULL;
999 lo->lo_encryption = NULL;
1000 lo->lo_offset = 0;
1001 lo->lo_sizelimit = 0;
1002 lo->lo_encrypt_key_size = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001003 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1005 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1006 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -05001007 if (bdev)
1008 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -07001009 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001010 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001011 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001012 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001013 /* let user-space know about this change */
1014 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 mapping_set_gfp_mask(filp->f_mapping, gfp);
1017 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 /* This is safe: open() is still holding a reference. */
1019 module_put(THIS_MODULE);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001020 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
Laurent Vivier476a4812008-03-26 12:11:53 +01001021 ioctl_by_bdev(bdev, BLKRRPART, 0);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001022 lo->lo_flags = 0;
1023 if (!part_shift)
1024 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001025 mutex_unlock(&lo->lo_ctl_mutex);
1026 /*
1027 * Need not hold lo_ctl_mutex to fput backing file.
1028 * Calling fput holding lo_ctl_mutex triggers a circular
1029 * lock dependency possibility warning as fput can take
1030 * bd_mutex which is usually taken before lo_ctl_mutex.
1031 */
1032 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return 0;
1034}
1035
1036static int
1037loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1038{
1039 int err;
1040 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +11001041 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
David Howellsb0fafa82008-11-14 10:38:41 +11001043 if (lo->lo_encrypt_key_size &&
1044 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 !capable(CAP_SYS_ADMIN))
1046 return -EPERM;
1047 if (lo->lo_state != Lo_bound)
1048 return -ENXIO;
1049 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1050 return -EINVAL;
1051
1052 err = loop_release_xfer(lo);
1053 if (err)
1054 return err;
1055
1056 if (info->lo_encrypt_type) {
1057 unsigned int type = info->lo_encrypt_type;
1058
1059 if (type >= MAX_LO_CRYPT)
1060 return -EINVAL;
1061 xfer = xfer_funcs[type];
1062 if (xfer == NULL)
1063 return -EINVAL;
1064 } else
1065 xfer = NULL;
1066
1067 err = loop_init_xfer(lo, xfer, info);
1068 if (err)
1069 return err;
1070
1071 if (lo->lo_offset != info->lo_offset ||
1072 lo->lo_sizelimit != info->lo_sizelimit) {
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001073 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return -EFBIG;
1075 }
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001076 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1079 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1080 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1081 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1082
1083 if (!xfer)
1084 xfer = &none_funcs;
1085 lo->transfer = xfer->transfer;
1086 lo->ioctl = xfer->ioctl;
1087
David Woodhouse96c58652008-02-06 01:36:27 -08001088 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1089 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1090 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1091
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001092 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1093 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1094 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1095 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1096 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1097 }
1098
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1100 lo->lo_init[0] = info->lo_init[0];
1101 lo->lo_init[1] = info->lo_init[1];
1102 if (info->lo_encrypt_key_size) {
1103 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1104 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001105 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107
1108 return 0;
1109}
1110
1111static int
1112loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1113{
1114 struct file *file = lo->lo_backing_file;
1115 struct kstat stat;
1116 int error;
1117
1118 if (lo->lo_state != Lo_bound)
1119 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001120 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (error)
1122 return error;
1123 memset(info, 0, sizeof(*info));
1124 info->lo_number = lo->lo_number;
1125 info->lo_device = huge_encode_dev(stat.dev);
1126 info->lo_inode = stat.ino;
1127 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1128 info->lo_offset = lo->lo_offset;
1129 info->lo_sizelimit = lo->lo_sizelimit;
1130 info->lo_flags = lo->lo_flags;
1131 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1132 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1133 info->lo_encrypt_type =
1134 lo->lo_encryption ? lo->lo_encryption->number : 0;
1135 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1136 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1137 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1138 lo->lo_encrypt_key_size);
1139 }
1140 return 0;
1141}
1142
1143static void
1144loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1145{
1146 memset(info64, 0, sizeof(*info64));
1147 info64->lo_number = info->lo_number;
1148 info64->lo_device = info->lo_device;
1149 info64->lo_inode = info->lo_inode;
1150 info64->lo_rdevice = info->lo_rdevice;
1151 info64->lo_offset = info->lo_offset;
1152 info64->lo_sizelimit = 0;
1153 info64->lo_encrypt_type = info->lo_encrypt_type;
1154 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1155 info64->lo_flags = info->lo_flags;
1156 info64->lo_init[0] = info->lo_init[0];
1157 info64->lo_init[1] = info->lo_init[1];
1158 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1159 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1160 else
1161 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1162 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1163}
1164
1165static int
1166loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1167{
1168 memset(info, 0, sizeof(*info));
1169 info->lo_number = info64->lo_number;
1170 info->lo_device = info64->lo_device;
1171 info->lo_inode = info64->lo_inode;
1172 info->lo_rdevice = info64->lo_rdevice;
1173 info->lo_offset = info64->lo_offset;
1174 info->lo_encrypt_type = info64->lo_encrypt_type;
1175 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1176 info->lo_flags = info64->lo_flags;
1177 info->lo_init[0] = info64->lo_init[0];
1178 info->lo_init[1] = info64->lo_init[1];
1179 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1180 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1181 else
1182 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1183 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1184
1185 /* error in case values were truncated */
1186 if (info->lo_device != info64->lo_device ||
1187 info->lo_rdevice != info64->lo_rdevice ||
1188 info->lo_inode != info64->lo_inode ||
1189 info->lo_offset != info64->lo_offset)
1190 return -EOVERFLOW;
1191
1192 return 0;
1193}
1194
1195static int
1196loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1197{
1198 struct loop_info info;
1199 struct loop_info64 info64;
1200
1201 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1202 return -EFAULT;
1203 loop_info64_from_old(&info, &info64);
1204 return loop_set_status(lo, &info64);
1205}
1206
1207static int
1208loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1209{
1210 struct loop_info64 info64;
1211
1212 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1213 return -EFAULT;
1214 return loop_set_status(lo, &info64);
1215}
1216
1217static int
1218loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1219 struct loop_info info;
1220 struct loop_info64 info64;
1221 int err = 0;
1222
1223 if (!arg)
1224 err = -EINVAL;
1225 if (!err)
1226 err = loop_get_status(lo, &info64);
1227 if (!err)
1228 err = loop_info64_to_old(&info64, &info);
1229 if (!err && copy_to_user(arg, &info, sizeof(info)))
1230 err = -EFAULT;
1231
1232 return err;
1233}
1234
1235static int
1236loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1237 struct loop_info64 info64;
1238 int err = 0;
1239
1240 if (!arg)
1241 err = -EINVAL;
1242 if (!err)
1243 err = loop_get_status(lo, &info64);
1244 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1245 err = -EFAULT;
1246
1247 return err;
1248}
1249
J. R. Okajima53d66602009-03-31 15:23:43 -07001250static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1251{
1252 int err;
1253 sector_t sec;
1254 loff_t sz;
1255
1256 err = -ENXIO;
1257 if (unlikely(lo->lo_state != Lo_bound))
1258 goto out;
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001259 err = figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
J. R. Okajima53d66602009-03-31 15:23:43 -07001260 if (unlikely(err))
1261 goto out;
1262 sec = get_capacity(lo->lo_disk);
1263 /* the width of sector_t may be narrow for bit-shift */
1264 sz = sec;
1265 sz <<= 9;
1266 mutex_lock(&bdev->bd_mutex);
1267 bd_set_size(bdev, sz);
David Zeuthenc3473c62010-05-03 14:08:59 +02001268 /* let user-space know about the new size */
1269 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
J. R. Okajima53d66602009-03-31 15:23:43 -07001270 mutex_unlock(&bdev->bd_mutex);
1271
1272 out:
1273 return err;
1274}
1275
Al Virobb214882008-03-02 09:29:48 -05001276static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 unsigned int cmd, unsigned long arg)
1278{
Al Virobb214882008-03-02 09:29:48 -05001279 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 int err;
1281
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001282 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 switch (cmd) {
1284 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001285 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 break;
1287 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001288 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 break;
1290 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001291 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Ayan George4c823cc2011-09-21 10:02:13 +02001292 err = loop_clr_fd(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001293 if (!err)
1294 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 break;
1296 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001297 err = -EPERM;
1298 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1299 err = loop_set_status_old(lo,
1300 (struct loop_info __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 break;
1302 case LOOP_GET_STATUS:
1303 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1304 break;
1305 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001306 err = -EPERM;
1307 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1308 err = loop_set_status64(lo,
1309 (struct loop_info64 __user *) arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 break;
1311 case LOOP_GET_STATUS64:
1312 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1313 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001314 case LOOP_SET_CAPACITY:
1315 err = -EPERM;
1316 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1317 err = loop_set_capacity(lo, bdev);
1318 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 default:
1320 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1321 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001322 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001323
1324out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 return err;
1326}
1327
David Howells863d5b822006-08-29 19:06:14 +01001328#ifdef CONFIG_COMPAT
1329struct compat_loop_info {
1330 compat_int_t lo_number; /* ioctl r/o */
1331 compat_dev_t lo_device; /* ioctl r/o */
1332 compat_ulong_t lo_inode; /* ioctl r/o */
1333 compat_dev_t lo_rdevice; /* ioctl r/o */
1334 compat_int_t lo_offset;
1335 compat_int_t lo_encrypt_type;
1336 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1337 compat_int_t lo_flags; /* ioctl r/o */
1338 char lo_name[LO_NAME_SIZE];
1339 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1340 compat_ulong_t lo_init[2];
1341 char reserved[4];
1342};
1343
1344/*
1345 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1346 * - noinlined to reduce stack space usage in main part of driver
1347 */
1348static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001349loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001350 struct loop_info64 *info64)
1351{
1352 struct compat_loop_info info;
1353
1354 if (copy_from_user(&info, arg, sizeof(info)))
1355 return -EFAULT;
1356
1357 memset(info64, 0, sizeof(*info64));
1358 info64->lo_number = info.lo_number;
1359 info64->lo_device = info.lo_device;
1360 info64->lo_inode = info.lo_inode;
1361 info64->lo_rdevice = info.lo_rdevice;
1362 info64->lo_offset = info.lo_offset;
1363 info64->lo_sizelimit = 0;
1364 info64->lo_encrypt_type = info.lo_encrypt_type;
1365 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1366 info64->lo_flags = info.lo_flags;
1367 info64->lo_init[0] = info.lo_init[0];
1368 info64->lo_init[1] = info.lo_init[1];
1369 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1370 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1371 else
1372 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1373 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1374 return 0;
1375}
1376
1377/*
1378 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1379 * - noinlined to reduce stack space usage in main part of driver
1380 */
1381static noinline int
1382loop_info64_to_compat(const struct loop_info64 *info64,
1383 struct compat_loop_info __user *arg)
1384{
1385 struct compat_loop_info info;
1386
1387 memset(&info, 0, sizeof(info));
1388 info.lo_number = info64->lo_number;
1389 info.lo_device = info64->lo_device;
1390 info.lo_inode = info64->lo_inode;
1391 info.lo_rdevice = info64->lo_rdevice;
1392 info.lo_offset = info64->lo_offset;
1393 info.lo_encrypt_type = info64->lo_encrypt_type;
1394 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1395 info.lo_flags = info64->lo_flags;
1396 info.lo_init[0] = info64->lo_init[0];
1397 info.lo_init[1] = info64->lo_init[1];
1398 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1399 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1400 else
1401 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1402 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1403
1404 /* error in case values were truncated */
1405 if (info.lo_device != info64->lo_device ||
1406 info.lo_rdevice != info64->lo_rdevice ||
1407 info.lo_inode != info64->lo_inode ||
1408 info.lo_offset != info64->lo_offset ||
1409 info.lo_init[0] != info64->lo_init[0] ||
1410 info.lo_init[1] != info64->lo_init[1])
1411 return -EOVERFLOW;
1412
1413 if (copy_to_user(arg, &info, sizeof(info)))
1414 return -EFAULT;
1415 return 0;
1416}
1417
1418static int
1419loop_set_status_compat(struct loop_device *lo,
1420 const struct compat_loop_info __user *arg)
1421{
1422 struct loop_info64 info64;
1423 int ret;
1424
1425 ret = loop_info64_from_compat(arg, &info64);
1426 if (ret < 0)
1427 return ret;
1428 return loop_set_status(lo, &info64);
1429}
1430
1431static int
1432loop_get_status_compat(struct loop_device *lo,
1433 struct compat_loop_info __user *arg)
1434{
1435 struct loop_info64 info64;
1436 int err = 0;
1437
1438 if (!arg)
1439 err = -EINVAL;
1440 if (!err)
1441 err = loop_get_status(lo, &info64);
1442 if (!err)
1443 err = loop_info64_to_compat(&info64, arg);
1444 return err;
1445}
1446
Al Virobb214882008-03-02 09:29:48 -05001447static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1448 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001449{
Al Virobb214882008-03-02 09:29:48 -05001450 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001451 int err;
1452
David Howells863d5b822006-08-29 19:06:14 +01001453 switch(cmd) {
1454 case LOOP_SET_STATUS:
1455 mutex_lock(&lo->lo_ctl_mutex);
1456 err = loop_set_status_compat(
1457 lo, (const struct compat_loop_info __user *) arg);
1458 mutex_unlock(&lo->lo_ctl_mutex);
1459 break;
1460 case LOOP_GET_STATUS:
1461 mutex_lock(&lo->lo_ctl_mutex);
1462 err = loop_get_status_compat(
1463 lo, (struct compat_loop_info __user *) arg);
1464 mutex_unlock(&lo->lo_ctl_mutex);
1465 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001466 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001467 case LOOP_CLR_FD:
1468 case LOOP_GET_STATUS64:
1469 case LOOP_SET_STATUS64:
1470 arg = (unsigned long) compat_ptr(arg);
1471 case LOOP_SET_FD:
1472 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001473 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001474 break;
1475 default:
1476 err = -ENOIOCTLCMD;
1477 break;
1478 }
David Howells863d5b822006-08-29 19:06:14 +01001479 return err;
1480}
1481#endif
1482
Al Virobb214882008-03-02 09:29:48 -05001483static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Kay Sievers770fe302011-07-31 22:08:04 +02001485 struct loop_device *lo;
1486 int err = 0;
1487
1488 mutex_lock(&loop_index_mutex);
1489 lo = bdev->bd_disk->private_data;
1490 if (!lo) {
1491 err = -ENXIO;
1492 goto out;
1493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
Ingo Molnarf85221d2006-03-23 03:00:38 -08001495 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001497 mutex_unlock(&lo->lo_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001498out:
1499 mutex_unlock(&loop_index_mutex);
1500 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
1502
Al Virobb214882008-03-02 09:29:48 -05001503static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
Al Virobb214882008-03-02 09:29:48 -05001505 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001506 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Ingo Molnarf85221d2006-03-23 03:00:38 -08001508 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001509
Milan Broz14f27932008-12-12 14:48:27 +01001510 if (--lo->lo_refcnt)
1511 goto out;
1512
1513 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1514 /*
1515 * In autoclear mode, stop the loop thread
1516 * and remove configuration after last close.
1517 */
Ayan George4c823cc2011-09-21 10:02:13 +02001518 err = loop_clr_fd(lo);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001519 if (!err)
1520 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001521 } else {
1522 /*
1523 * Otherwise keep thread (if running) and config,
1524 * but flush possible ongoing bios in thread.
1525 */
1526 loop_flush(lo);
1527 }
David Woodhouse96c58652008-02-06 01:36:27 -08001528
Milan Broz14f27932008-12-12 14:48:27 +01001529out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001530 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001531out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return 0;
1533}
1534
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001535static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001537 .open = lo_open,
1538 .release = lo_release,
1539 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001540#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001541 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001542#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543};
1544
1545/*
1546 * And now the modules code and kernel interface.
1547 */
Ken Chen73285082007-05-08 00:28:20 -07001548static int max_loop;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001549module_param(max_loop, int, S_IRUGO);
Ken Chena47653f2007-06-08 13:46:44 -07001550MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Namhyung Kimac04fee2011-05-27 07:59:25 +02001551module_param(max_part, int, S_IRUGO);
Laurent Vivier476a4812008-03-26 12:11:53 +01001552MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553MODULE_LICENSE("GPL");
1554MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1555
1556int loop_register_transfer(struct loop_func_table *funcs)
1557{
1558 unsigned int n = funcs->number;
1559
1560 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1561 return -EINVAL;
1562 xfer_funcs[n] = funcs;
1563 return 0;
1564}
1565
Kay Sievers34dd82a2011-07-31 22:08:04 +02001566static int unregister_transfer_cb(int id, void *ptr, void *data)
1567{
1568 struct loop_device *lo = ptr;
1569 struct loop_func_table *xfer = data;
1570
1571 mutex_lock(&lo->lo_ctl_mutex);
1572 if (lo->lo_encryption == xfer)
1573 loop_release_xfer(lo);
1574 mutex_unlock(&lo->lo_ctl_mutex);
1575 return 0;
1576}
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578int loop_unregister_transfer(int number)
1579{
1580 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 struct loop_func_table *xfer;
1582
1583 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1584 return -EINVAL;
1585
1586 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001587 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return 0;
1589}
1590
1591EXPORT_SYMBOL(loop_register_transfer);
1592EXPORT_SYMBOL(loop_unregister_transfer);
1593
Kay Sievers34dd82a2011-07-31 22:08:04 +02001594static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001595{
1596 struct loop_device *lo;
1597 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001598 int err;
Ken Chen73285082007-05-08 00:28:20 -07001599
1600 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001601 if (!lo) {
1602 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001603 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001604 }
1605
1606 err = idr_pre_get(&loop_index_idr, GFP_KERNEL);
1607 if (err < 0)
1608 goto out_free_dev;
1609
1610 if (i >= 0) {
1611 int m;
1612
1613 /* create specific i in the index */
1614 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1615 if (err >= 0 && i != m) {
1616 idr_remove(&loop_index_idr, m);
1617 err = -EEXIST;
1618 }
Kay Sievers770fe302011-07-31 22:08:04 +02001619 } else if (i == -1) {
1620 int m;
1621
1622 /* get next free nr */
1623 err = idr_get_new(&loop_index_idr, lo, &m);
1624 if (err >= 0)
1625 i = m;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001626 } else {
1627 err = -EINVAL;
1628 }
1629 if (err < 0)
1630 goto out_free_dev;
Ken Chen73285082007-05-08 00:28:20 -07001631
1632 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1633 if (!lo->lo_queue)
1634 goto out_free_dev;
1635
Laurent Vivier476a4812008-03-26 12:11:53 +01001636 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001637 if (!disk)
1638 goto out_free_queue;
1639
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001640 /*
1641 * Disable partition scanning by default. The in-kernel partition
1642 * scanning can be requested individually per-device during its
1643 * setup. Userspace can always add and remove partitions from all
1644 * devices. The needed partition minors are allocated from the
1645 * extended minor space, the main loop device numbers will continue
1646 * to match the loop minors, regardless of the number of partitions
1647 * used.
1648 *
1649 * If max_part is given, partition scanning is globally enabled for
1650 * all loop devices. The minors for the main loop devices will be
1651 * multiples of max_part.
1652 *
1653 * Note: Global-for-all-devices, set-only-at-init, read-only module
1654 * parameteters like 'max_loop' and 'max_part' make things needlessly
1655 * complicated, are too static, inflexible and may surprise
1656 * userspace tools. Parameters like this in general should be avoided.
1657 */
1658 if (!part_shift)
1659 disk->flags |= GENHD_FL_NO_PART_SCAN;
1660 disk->flags |= GENHD_FL_EXT_DEVT;
Ken Chen73285082007-05-08 00:28:20 -07001661 mutex_init(&lo->lo_ctl_mutex);
1662 lo->lo_number = i;
1663 lo->lo_thread = NULL;
1664 init_waitqueue_head(&lo->lo_event);
1665 spin_lock_init(&lo->lo_lock);
1666 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001667 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001668 disk->fops = &lo_fops;
1669 disk->private_data = lo;
1670 disk->queue = lo->lo_queue;
1671 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001672 add_disk(disk);
1673 *l = lo;
1674 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001675
1676out_free_queue:
1677 blk_cleanup_queue(lo->lo_queue);
1678out_free_dev:
1679 kfree(lo);
1680out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001681 return err;
Ken Chen73285082007-05-08 00:28:20 -07001682}
1683
Kay Sievers34dd82a2011-07-31 22:08:04 +02001684static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001685{
Kay Sievers34dd82a2011-07-31 22:08:04 +02001686 del_gendisk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001687 blk_cleanup_queue(lo->lo_queue);
1688 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001689 kfree(lo);
1690}
1691
Kay Sievers770fe302011-07-31 22:08:04 +02001692static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001693{
Kay Sievers770fe302011-07-31 22:08:04 +02001694 struct loop_device *lo = ptr;
1695 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001696
Kay Sievers770fe302011-07-31 22:08:04 +02001697 if (lo->lo_state == Lo_unbound) {
1698 *l = lo;
1699 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001700 }
Kay Sievers770fe302011-07-31 22:08:04 +02001701 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001702}
1703
Kay Sievers34dd82a2011-07-31 22:08:04 +02001704static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001705{
Ken Chena47653f2007-06-08 13:46:44 -07001706 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001707 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001708
Kay Sievers770fe302011-07-31 22:08:04 +02001709 if (i < 0) {
1710 int err;
1711
1712 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1713 if (err == 1) {
1714 *l = lo;
1715 ret = lo->lo_number;
1716 }
1717 goto out;
1718 }
1719
1720 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001721 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07001722 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02001723 *l = lo;
1724 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07001725 }
Kay Sievers770fe302011-07-31 22:08:04 +02001726out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001727 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07001728}
1729
Ken Chen73285082007-05-08 00:28:20 -07001730static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1731{
Al Viro705962c2007-05-13 05:52:32 -04001732 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001733 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001734 int err;
Ken Chen73285082007-05-08 00:28:20 -07001735
Kay Sievers34dd82a2011-07-31 22:08:04 +02001736 mutex_lock(&loop_index_mutex);
1737 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1738 if (err < 0)
1739 err = loop_add(&lo, MINOR(dev) >> part_shift);
1740 if (err < 0)
1741 kobj = ERR_PTR(err);
1742 else
1743 kobj = get_disk(lo->lo_disk);
1744 mutex_unlock(&loop_index_mutex);
Ken Chen73285082007-05-08 00:28:20 -07001745
1746 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001747 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001748}
1749
Kay Sievers770fe302011-07-31 22:08:04 +02001750static long loop_control_ioctl(struct file *file, unsigned int cmd,
1751 unsigned long parm)
1752{
1753 struct loop_device *lo;
1754 int ret = -ENOSYS;
1755
1756 mutex_lock(&loop_index_mutex);
1757 switch (cmd) {
1758 case LOOP_CTL_ADD:
1759 ret = loop_lookup(&lo, parm);
1760 if (ret >= 0) {
1761 ret = -EEXIST;
1762 break;
1763 }
1764 ret = loop_add(&lo, parm);
1765 break;
1766 case LOOP_CTL_REMOVE:
1767 ret = loop_lookup(&lo, parm);
1768 if (ret < 0)
1769 break;
1770 mutex_lock(&lo->lo_ctl_mutex);
1771 if (lo->lo_state != Lo_unbound) {
1772 ret = -EBUSY;
1773 mutex_unlock(&lo->lo_ctl_mutex);
1774 break;
1775 }
1776 if (lo->lo_refcnt > 0) {
1777 ret = -EBUSY;
1778 mutex_unlock(&lo->lo_ctl_mutex);
1779 break;
1780 }
1781 lo->lo_disk->private_data = NULL;
1782 mutex_unlock(&lo->lo_ctl_mutex);
1783 idr_remove(&loop_index_idr, lo->lo_number);
1784 loop_remove(lo);
1785 break;
1786 case LOOP_CTL_GET_FREE:
1787 ret = loop_lookup(&lo, -1);
1788 if (ret >= 0)
1789 break;
1790 ret = loop_add(&lo, -1);
1791 }
1792 mutex_unlock(&loop_index_mutex);
1793
1794 return ret;
1795}
1796
1797static const struct file_operations loop_ctl_fops = {
1798 .open = nonseekable_open,
1799 .unlocked_ioctl = loop_control_ioctl,
1800 .compat_ioctl = loop_control_ioctl,
1801 .owner = THIS_MODULE,
1802 .llseek = noop_llseek,
1803};
1804
1805static struct miscdevice loop_misc = {
1806 .minor = LOOP_CTRL_MINOR,
1807 .name = "loop-control",
1808 .fops = &loop_ctl_fops,
1809};
1810
1811MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1812MODULE_ALIAS("devname:loop-control");
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814static int __init loop_init(void)
1815{
Ken Chena47653f2007-06-08 13:46:44 -07001816 int i, nr;
1817 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001818 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02001819 int err;
Ken Chena47653f2007-06-08 13:46:44 -07001820
Kay Sievers770fe302011-07-31 22:08:04 +02001821 err = misc_register(&loop_misc);
1822 if (err < 0)
1823 return err;
Laurent Vivier476a4812008-03-26 12:11:53 +01001824
1825 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001826 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01001827 part_shift = fls(max_part);
1828
Namhyung Kimac04fee2011-05-27 07:59:25 +02001829 /*
1830 * Adjust max_part according to part_shift as it is exported
1831 * to user space so that user can decide correct minor number
1832 * if [s]he want to create more devices.
1833 *
1834 * Note that -1 is required because partition 0 is reserved
1835 * for the whole disk.
1836 */
1837 max_part = (1UL << part_shift) - 1;
1838 }
1839
Namhyung Kim78f4bb32011-05-24 16:48:54 +02001840 if ((1UL << part_shift) > DISK_MAX_PARTS)
1841 return -EINVAL;
1842
Laurent Vivier476a4812008-03-26 12:11:53 +01001843 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001844 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
Kay Sieversd134b002011-07-31 22:08:04 +02001846 /*
1847 * If max_loop is specified, create that many devices upfront.
1848 * This also becomes a hard limit. If max_loop is not specified,
1849 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1850 * init time. Loop devices can be requested on-demand with the
1851 * /dev/loop-control interface, or be instantiated by accessing
1852 * a 'dead' device node.
1853 */
Ken Chen73285082007-05-08 00:28:20 -07001854 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001855 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001856 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07001857 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02001858 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001859 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 }
Ken Chena47653f2007-06-08 13:46:44 -07001861
1862 if (register_blkdev(LOOP_MAJOR, "loop"))
1863 return -EIO;
1864
Ken Chena47653f2007-06-08 13:46:44 -07001865 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1866 THIS_MODULE, loop_probe, NULL, NULL);
1867
Kay Sieversd134b002011-07-31 22:08:04 +02001868 /* pre-create number of devices given by config or max_loop */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001869 mutex_lock(&loop_index_mutex);
1870 for (i = 0; i < nr; i++)
1871 loop_add(&lo, i);
1872 mutex_unlock(&loop_index_mutex);
1873
Ken Chen73285082007-05-08 00:28:20 -07001874 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 return 0;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001876}
Ken Chena47653f2007-06-08 13:46:44 -07001877
Kay Sievers34dd82a2011-07-31 22:08:04 +02001878static int loop_exit_cb(int id, void *ptr, void *data)
1879{
1880 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07001881
Kay Sievers34dd82a2011-07-31 22:08:04 +02001882 loop_remove(lo);
1883 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884}
1885
Ken Chen73285082007-05-08 00:28:20 -07001886static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
Ken Chena47653f2007-06-08 13:46:44 -07001888 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Namhyung Kima1c15c52011-05-24 16:48:55 +02001890 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07001891
Kay Sievers34dd82a2011-07-31 22:08:04 +02001892 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1893 idr_remove_all(&loop_index_idr);
1894 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07001895
Ken Chena47653f2007-06-08 13:46:44 -07001896 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001897 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02001898
1899 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900}
1901
1902module_init(loop_init);
1903module_exit(loop_exit);
1904
1905#ifndef MODULE
1906static int __init max_loop_setup(char *str)
1907{
1908 max_loop = simple_strtol(str, NULL, 0);
1909 return 1;
1910}
1911
1912__setup("max_loop=", max_loop_setup);
1913#endif