blob: 3d806820280e3bc4aaa5e81d6bea411f6597e400 [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>
72#include <linux/buffer_head.h> /* for invalidate_bdev() */
73#include <linux/completion.h>
74#include <linux/highmem.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070075#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020076#include <linux/splice.h>
Milan Brozee862732010-08-23 15:16:00 +020077#include <linux/sysfs.h>
Kay Sievers770fe302011-07-31 22:08:04 +020078#include <linux/miscdevice.h>
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +020079#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81#include <asm/uaccess.h>
82
Kay Sievers34dd82a2011-07-31 22:08:04 +020083static DEFINE_IDR(loop_index_idr);
84static DEFINE_MUTEX(loop_index_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Laurent Vivier476a4812008-03-26 12:11:53 +010086static int max_part;
87static int part_shift;
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/*
90 * Transfer functions
91 */
92static int transfer_none(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
96{
97 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
99
100 if (cmd == READ)
101 memcpy(loop_buf, raw_buf, size);
102 else
103 memcpy(raw_buf, loop_buf, size);
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 kunmap_atomic(loop_buf, KM_USER1);
Peter Zijlstra61ecdb802010-10-26 14:21:47 -0700106 kunmap_atomic(raw_buf, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 cond_resched();
108 return 0;
109}
110
111static int transfer_xor(struct loop_device *lo, int cmd,
112 struct page *raw_page, unsigned raw_off,
113 struct page *loop_page, unsigned loop_off,
114 int size, sector_t real_block)
115{
116 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
117 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
118 char *in, *out, *key;
119 int i, keysize;
120
121 if (cmd == READ) {
122 in = raw_buf;
123 out = loop_buf;
124 } else {
125 in = loop_buf;
126 out = raw_buf;
127 }
128
129 key = lo->lo_encrypt_key;
130 keysize = lo->lo_encrypt_key_size;
131 for (i = 0; i < size; i++)
132 *out++ = *in++ ^ key[(i & 511) % keysize];
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 kunmap_atomic(loop_buf, KM_USER1);
Peter Zijlstra61ecdb802010-10-26 14:21:47 -0700135 kunmap_atomic(raw_buf, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 cond_resched();
137 return 0;
138}
139
140static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
141{
142 if (unlikely(info->lo_encrypt_key_size <= 0))
143 return -EINVAL;
144 return 0;
145}
146
147static struct loop_func_table none_funcs = {
148 .number = LO_CRYPT_NONE,
149 .transfer = transfer_none,
150};
151
152static struct loop_func_table xor_funcs = {
153 .number = LO_CRYPT_XOR,
154 .transfer = transfer_xor,
155 .init = xor_init
156};
157
158/* xfer_funcs[0] is special - its release function is never called */
159static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
160 &none_funcs,
161 &xor_funcs
162};
163
164static loff_t get_loop_size(struct loop_device *lo, struct file *file)
165{
166 loff_t size, offset, loopsize;
167
168 /* Compute loopsize in bytes */
169 size = i_size_read(file->f_mapping->host);
170 offset = lo->lo_offset;
171 loopsize = size - offset;
172 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
173 loopsize = lo->lo_sizelimit;
174
175 /*
176 * Unfortunately, if we want to do I/O on the device,
177 * the number of 512-byte sectors has to fit into a sector_t.
178 */
179 return loopsize >> 9;
180}
181
182static int
183figure_loop_size(struct loop_device *lo)
184{
185 loff_t size = get_loop_size(lo, lo->lo_backing_file);
186 sector_t x = (sector_t)size;
187
188 if (unlikely((loff_t)x != size))
189 return -EFBIG;
190
Ken Chen73285082007-05-08 00:28:20 -0700191 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return 0;
193}
194
195static inline int
196lo_do_transfer(struct loop_device *lo, int cmd,
197 struct page *rpage, unsigned roffs,
198 struct page *lpage, unsigned loffs,
199 int size, sector_t rblock)
200{
201 if (unlikely(!lo->transfer))
202 return 0;
203
204 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
205}
206
207/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 * __do_lo_send_write - helper for writing data to a loop device
209 *
210 * This helper just factors out common code between do_lo_send_direct_write()
211 * and do_lo_send_write().
212 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800213static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100214 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 ssize_t bw;
217 mm_segment_t old_fs = get_fs();
218
219 set_fs(get_ds());
220 bw = file->f_op->write(file, buf, len, &pos);
221 set_fs(old_fs);
222 if (likely(bw == len))
223 return 0;
224 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
225 (unsigned long long)pos, len);
226 if (bw >= 0)
227 bw = -EIO;
228 return bw;
229}
230
231/**
232 * do_lo_send_direct_write - helper for writing data to a loop device
233 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200234 * This is the fast, non-transforming version that does not need double
235 * buffering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 */
237static int do_lo_send_direct_write(struct loop_device *lo,
Al Viro511de732007-10-08 12:10:13 -0400238 struct bio_vec *bvec, loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100241 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 bvec->bv_len, pos);
243 kunmap(bvec->bv_page);
244 cond_resched();
245 return bw;
246}
247
248/**
249 * do_lo_send_write - helper for writing data to a loop device
250 *
Christoph Hellwig456be142011-10-17 12:57:20 +0200251 * This is the slow, transforming version that needs to double buffer the
252 * data as it cannot do the transformations in place without having direct
253 * access to the destination pages of the backing file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 */
255static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400256 loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
258 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
259 bvec->bv_offset, bvec->bv_len, pos >> 9);
260 if (likely(!ret))
261 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100262 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 pos);
264 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
265 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
266 if (ret > 0)
267 ret = -EIO;
268 return ret;
269}
270
Al Viro511de732007-10-08 12:10:13 -0400271static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Al Viro511de732007-10-08 12:10:13 -0400273 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 struct page *page);
275 struct bio_vec *bvec;
276 struct page *page = NULL;
277 int i, ret = 0;
278
Christoph Hellwig456be142011-10-17 12:57:20 +0200279 if (lo->transfer != transfer_none) {
280 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
281 if (unlikely(!page))
282 goto fail;
283 kmap(page);
284 do_lo_send = do_lo_send_write;
285 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 do_lo_send = do_lo_send_direct_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 }
Christoph Hellwig456be142011-10-17 12:57:20 +0200288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 bio_for_each_segment(bvec, bio, i) {
Al Viro511de732007-10-08 12:10:13 -0400290 ret = do_lo_send(lo, bvec, pos, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 if (ret < 0)
292 break;
293 pos += bvec->bv_len;
294 }
295 if (page) {
296 kunmap(page);
297 __free_page(page);
298 }
299out:
300 return ret;
301fail:
302 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
303 ret = -ENOMEM;
304 goto out;
305}
306
307struct lo_read_data {
308 struct loop_device *lo;
309 struct page *page;
310 unsigned offset;
311 int bsize;
312};
313
314static int
Jens Axboefd582142007-06-12 21:20:37 +0200315lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
316 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
Jens Axboefd582142007-06-12 21:20:37 +0200318 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200320 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 sector_t IV;
Jens Axboe3603b8e2010-12-20 09:15:19 +0100322 int size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Jens Axboefd582142007-06-12 21:20:37 +0200324 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
325 (buf->offset >> 9);
326 size = sd->len;
327 if (size > p->bsize)
328 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Jens Axboefd582142007-06-12 21:20:37 +0200330 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 printk(KERN_ERR "loop: transfer error block %ld\n",
332 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200333 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
335
336 flush_dcache_page(p->page);
337
Jens Axboefd582142007-06-12 21:20:37 +0200338 if (size > 0)
339 p->offset += size;
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return size;
342}
343
344static int
Jens Axboefd582142007-06-12 21:20:37 +0200345lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
346{
347 return __splice_from_pipe(pipe, sd, lo_splice_actor);
348}
349
350static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351do_lo_receive(struct loop_device *lo,
352 struct bio_vec *bvec, int bsize, loff_t pos)
353{
354 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200355 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200357 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 cookie.lo = lo;
360 cookie.page = bvec->bv_page;
361 cookie.offset = bvec->bv_offset;
362 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200363
364 sd.len = 0;
365 sd.total_len = bvec->bv_len;
366 sd.flags = 0;
367 sd.pos = pos;
368 sd.u.data = &cookie;
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200371 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
372
373 if (retval < 0)
374 return retval;
375
376 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
379static int
380lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
381{
382 struct bio_vec *bvec;
383 int i, ret = 0;
384
385 bio_for_each_segment(bvec, bio, i) {
386 ret = do_lo_receive(lo, bvec, bsize, pos);
387 if (ret < 0)
388 break;
389 pos += bvec->bv_len;
390 }
391 return ret;
392}
393
394static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
395{
396 loff_t pos;
397 int ret;
398
399 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100400
401 if (bio_rw(bio) == WRITE) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100402 struct file *file = lo->lo_backing_file;
403
Tejun Heo6259f282010-09-03 11:56:17 +0200404 if (bio->bi_rw & REQ_FLUSH) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100405 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200406 if (unlikely(ret && ret != -EINVAL)) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100407 ret = -EIO;
408 goto out;
409 }
410 }
411
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200412 /*
413 * We use punch hole to reclaim the free space used by the
414 * image a.k.a. discard. However we do support discard if
415 * encryption is enabled, because it may give an attacker
416 * useful information.
417 */
418 if (bio->bi_rw & REQ_DISCARD) {
419 struct file *file = lo->lo_backing_file;
420 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
421
422 if ((!file->f_op->fallocate) ||
423 lo->lo_encrypt_key_size) {
424 ret = -EOPNOTSUPP;
425 goto out;
426 }
427 ret = file->f_op->fallocate(file, mode, pos,
428 bio->bi_size);
429 if (unlikely(ret && ret != -EINVAL &&
430 ret != -EOPNOTSUPP))
431 ret = -EIO;
432 goto out;
433 }
434
Al Viro511de732007-10-08 12:10:13 -0400435 ret = lo_send(lo, bio, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100436
Tejun Heo6259f282010-09-03 11:56:17 +0200437 if ((bio->bi_rw & REQ_FUA) && !ret) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100438 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200439 if (unlikely(ret && ret != -EINVAL))
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100440 ret = -EIO;
441 }
442 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100444
445out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 return ret;
447}
448
449/*
450 * Add bio to back of pending list
451 */
452static void loop_add_bio(struct loop_device *lo, struct bio *bio)
453{
Akinobu Mitae6863072009-04-17 08:41:21 +0200454 bio_list_add(&lo->lo_bio_list, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457/*
458 * Grab first pending buffer
459 */
460static struct bio *loop_get_bio(struct loop_device *lo)
461{
Akinobu Mitae6863072009-04-17 08:41:21 +0200462 return bio_list_pop(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200465static void loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
467 struct loop_device *lo = q->queuedata;
468 int rw = bio_rw(old_bio);
469
Nick Piggin35a82d12005-06-23 00:09:06 -0700470 if (rw == READA)
471 rw = READ;
472
473 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 spin_lock_irq(&lo->lo_lock);
476 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700477 goto out;
478 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
479 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700481 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700482 spin_unlock_irq(&lo->lo_lock);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200483 return;
Nick Piggin35a82d12005-06-23 00:09:06 -0700484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700486 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200487 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490struct switch_request {
491 struct file *file;
492 struct completion wait;
493};
494
495static void do_loop_switch(struct loop_device *, struct switch_request *);
496
497static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
498{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (unlikely(!bio->bi_bdev)) {
500 do_loop_switch(lo, bio->bi_private);
501 bio_put(bio);
502 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700503 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200504 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
506}
507
508/*
509 * worker thread that handles reads/writes to file backed loop devices,
510 * to avoid blocking in our make_request_fn. it also does loop decrypting
511 * on reads for block backed loop, as that is too heavy to do from
512 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700513 *
514 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
515 * calling kthread_stop(). Therefore once kthread_should_stop() is
516 * true, make_request will not place any more requests. Therefore
517 * once kthread_should_stop() is true and lo_bio is NULL, we are
518 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 */
520static int loop_thread(void *data)
521{
522 struct loop_device *lo = data;
523 struct bio *bio;
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 set_user_nice(current, -20);
526
Akinobu Mitae6863072009-04-17 08:41:21 +0200527 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700529 wait_event_interruptible(lo->lo_event,
Akinobu Mitae6863072009-04-17 08:41:21 +0200530 !bio_list_empty(&lo->lo_bio_list) ||
531 kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700532
Akinobu Mitae6863072009-04-17 08:41:21 +0200533 if (bio_list_empty(&lo->lo_bio_list))
Nick Piggin35a82d12005-06-23 00:09:06 -0700534 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700535 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700537 spin_unlock_irq(&lo->lo_lock);
538
539 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 return 0;
544}
545
546/*
547 * loop_switch performs the hard work of switching a backing store.
548 * First it needs to flush existing IO, it does this by sending a magic
549 * BIO down the pipe. The completion of this BIO does the actual switch.
550 */
551static int loop_switch(struct loop_device *lo, struct file *file)
552{
553 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100554 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (!bio)
556 return -ENOMEM;
557 init_completion(&w.wait);
558 w.file = file;
559 bio->bi_private = &w;
560 bio->bi_bdev = NULL;
561 loop_make_request(lo->lo_queue, bio);
562 wait_for_completion(&w.wait);
563 return 0;
564}
565
566/*
Milan Broz14f27932008-12-12 14:48:27 +0100567 * Helper to flush the IOs in loop, but keeping loop thread running
568 */
569static int loop_flush(struct loop_device *lo)
570{
571 /* loop not yet configured, no running thread, nothing to flush */
572 if (!lo->lo_thread)
573 return 0;
574
575 return loop_switch(lo, NULL);
576}
577
578/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * Do the actual switch; called from the BIO completion routine
580 */
581static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
582{
583 struct file *file = p->file;
584 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100585 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Milan Broz14f27932008-12-12 14:48:27 +0100587 /* if no new file, only flush of queued bios requested */
588 if (!file)
589 goto out;
590
591 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
593 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700594 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
595 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 lo->old_gfp_mask = mapping_gfp_mask(mapping);
597 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100598out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 complete(&p->wait);
600}
601
602
603/*
604 * loop_change_fd switched the backing store of a loopback device to
605 * a new file. This is useful for operating system installers to free up
606 * the original file and in High Availability environments to switch to
607 * an alternative location for the content in case of server meltdown.
608 * This can only work if the loop device is used read-only, and if the
609 * new backing store is the same size and type as the old backing store.
610 */
Al Virobb214882008-03-02 09:29:48 -0500611static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
612 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
614 struct file *file, *old_file;
615 struct inode *inode;
616 int error;
617
618 error = -ENXIO;
619 if (lo->lo_state != Lo_bound)
620 goto out;
621
622 /* the loop device has to be read-only */
623 error = -EINVAL;
624 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
625 goto out;
626
627 error = -EBADF;
628 file = fget(arg);
629 if (!file)
630 goto out;
631
632 inode = file->f_mapping->host;
633 old_file = lo->lo_backing_file;
634
635 error = -EINVAL;
636
637 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
638 goto out_putf;
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 /* size of the new backing store needs to be the same */
641 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
642 goto out_putf;
643
644 /* and ... switch */
645 error = loop_switch(lo, file);
646 if (error)
647 goto out_putf;
648
649 fput(old_file);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200650 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100651 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 return 0;
653
654 out_putf:
655 fput(file);
656 out:
657 return error;
658}
659
660static inline int is_loop_device(struct file *file)
661{
662 struct inode *i = file->f_mapping->host;
663
664 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
665}
666
Milan Brozee862732010-08-23 15:16:00 +0200667/* loop sysfs attributes */
668
669static ssize_t loop_attr_show(struct device *dev, char *page,
670 ssize_t (*callback)(struct loop_device *, char *))
671{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200672 struct gendisk *disk = dev_to_disk(dev);
673 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200674
Kay Sievers34dd82a2011-07-31 22:08:04 +0200675 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200676}
677
678#define LOOP_ATTR_RO(_name) \
679static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
680static ssize_t loop_attr_do_show_##_name(struct device *d, \
681 struct device_attribute *attr, char *b) \
682{ \
683 return loop_attr_show(d, b, loop_attr_##_name##_show); \
684} \
685static struct device_attribute loop_attr_##_name = \
686 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
687
688static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
689{
690 ssize_t ret;
691 char *p = NULL;
692
Kay Sievers05eb0f22011-07-31 22:21:35 +0200693 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200694 if (lo->lo_backing_file)
695 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200696 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200697
698 if (IS_ERR_OR_NULL(p))
699 ret = PTR_ERR(p);
700 else {
701 ret = strlen(p);
702 memmove(buf, p, ret);
703 buf[ret++] = '\n';
704 buf[ret] = 0;
705 }
706
707 return ret;
708}
709
710static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
711{
712 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
713}
714
715static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
716{
717 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
718}
719
720static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
721{
722 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
723
724 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
725}
726
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200727static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
728{
729 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
730
731 return sprintf(buf, "%s\n", partscan ? "1" : "0");
732}
733
Milan Brozee862732010-08-23 15:16:00 +0200734LOOP_ATTR_RO(backing_file);
735LOOP_ATTR_RO(offset);
736LOOP_ATTR_RO(sizelimit);
737LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200738LOOP_ATTR_RO(partscan);
Milan Brozee862732010-08-23 15:16:00 +0200739
740static struct attribute *loop_attrs[] = {
741 &loop_attr_backing_file.attr,
742 &loop_attr_offset.attr,
743 &loop_attr_sizelimit.attr,
744 &loop_attr_autoclear.attr,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200745 &loop_attr_partscan.attr,
Milan Brozee862732010-08-23 15:16:00 +0200746 NULL,
747};
748
749static struct attribute_group loop_attribute_group = {
750 .name = "loop",
751 .attrs= loop_attrs,
752};
753
754static int loop_sysfs_init(struct loop_device *lo)
755{
756 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
757 &loop_attribute_group);
758}
759
760static void loop_sysfs_exit(struct loop_device *lo)
761{
762 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
763 &loop_attribute_group);
764}
765
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200766static void loop_config_discard(struct loop_device *lo)
767{
768 struct file *file = lo->lo_backing_file;
769 struct inode *inode = file->f_mapping->host;
770 struct request_queue *q = lo->lo_queue;
771
772 /*
773 * We use punch hole to reclaim the free space used by the
774 * image a.k.a. discard. However we do support discard if
775 * encryption is enabled, because it may give an attacker
776 * useful information.
777 */
778 if ((!file->f_op->fallocate) ||
779 lo->lo_encrypt_key_size) {
780 q->limits.discard_granularity = 0;
781 q->limits.discard_alignment = 0;
782 q->limits.max_discard_sectors = 0;
783 q->limits.discard_zeroes_data = 0;
784 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
785 return;
786 }
787
788 q->limits.discard_granularity = inode->i_sb->s_blocksize;
789 q->limits.discard_alignment = inode->i_sb->s_blocksize;
790 q->limits.max_discard_sectors = UINT_MAX >> 9;
791 q->limits.discard_zeroes_data = 1;
792 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
793}
794
Al Virobb214882008-03-02 09:29:48 -0500795static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 struct block_device *bdev, unsigned int arg)
797{
798 struct file *file, *f;
799 struct inode *inode;
800 struct address_space *mapping;
801 unsigned lo_blocksize;
802 int lo_flags = 0;
803 int error;
804 loff_t size;
805
806 /* This is safe, since we have a reference from open(). */
807 __module_get(THIS_MODULE);
808
809 error = -EBADF;
810 file = fget(arg);
811 if (!file)
812 goto out;
813
814 error = -EBUSY;
815 if (lo->lo_state != Lo_unbound)
816 goto out_putf;
817
818 /* Avoid recursion */
819 f = file;
820 while (is_loop_device(f)) {
821 struct loop_device *l;
822
Al Virobb214882008-03-02 09:29:48 -0500823 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 goto out_putf;
825
826 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
827 if (l->lo_state == Lo_unbound) {
828 error = -EINVAL;
829 goto out_putf;
830 }
831 f = l->lo_backing_file;
832 }
833
834 mapping = file->f_mapping;
835 inode = mapping->host;
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 error = -EINVAL;
Christoph Hellwig456be142011-10-17 12:57:20 +0200838 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Christoph Hellwig456be142011-10-17 12:57:20 +0200841 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
842 !file->f_op->write)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 lo_flags |= LO_FLAGS_READ_ONLY;
844
Christoph Hellwig456be142011-10-17 12:57:20 +0200845 lo_blocksize = S_ISBLK(inode->i_mode) ?
846 inode->i_bdev->bd_block_size : PAGE_SIZE;
847
848 error = -EFBIG;
849 size = get_loop_size(lo, file);
850 if ((loff_t)(sector_t)size != size)
851 goto out_putf;
852
853 error = 0;
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
856
857 lo->lo_blocksize = lo_blocksize;
858 lo->lo_device = bdev;
859 lo->lo_flags = lo_flags;
860 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700861 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 lo->ioctl = NULL;
863 lo->lo_sizelimit = 0;
864 lo->old_gfp_mask = mapping_gfp_mask(mapping);
865 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
866
Akinobu Mitae6863072009-04-17 08:41:21 +0200867 bio_list_init(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 /*
870 * set queue make_request_fn, and add limits based on lower level
871 * device
872 */
873 blk_queue_make_request(lo->lo_queue, loop_make_request);
874 lo->lo_queue->queuedata = lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100876 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Tejun Heo4913efe2010-09-03 11:56:16 +0200877 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100878
Ken Chen73285082007-05-08 00:28:20 -0700879 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200881 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200882 /* let user-space know about the new size */
883 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 set_blocksize(bdev, lo_blocksize);
886
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700887 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
888 lo->lo_number);
889 if (IS_ERR(lo->lo_thread)) {
890 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700891 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700892 }
893 lo->lo_state = Lo_bound;
894 wake_up_process(lo->lo_thread);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200895 if (part_shift)
896 lo->lo_flags |= LO_FLAGS_PARTSCAN;
897 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
Laurent Vivier476a4812008-03-26 12:11:53 +0100898 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 return 0;
900
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700901out_clr:
Milan Brozee862732010-08-23 15:16:00 +0200902 loop_sysfs_exit(lo);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700903 lo->lo_thread = NULL;
904 lo->lo_device = NULL;
905 lo->lo_backing_file = NULL;
906 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700907 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700908 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700909 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200910 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700911 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
912 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 out_putf:
914 fput(file);
915 out:
916 /* This is safe: open() is still holding a reference. */
917 module_put(THIS_MODULE);
918 return error;
919}
920
921static int
922loop_release_xfer(struct loop_device *lo)
923{
924 int err = 0;
925 struct loop_func_table *xfer = lo->lo_encryption;
926
927 if (xfer) {
928 if (xfer->release)
929 err = xfer->release(lo);
930 lo->transfer = NULL;
931 lo->lo_encryption = NULL;
932 module_put(xfer->owner);
933 }
934 return err;
935}
936
937static int
938loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
939 const struct loop_info64 *i)
940{
941 int err = 0;
942
943 if (xfer) {
944 struct module *owner = xfer->owner;
945
946 if (!try_module_get(owner))
947 return -EINVAL;
948 if (xfer->init)
949 err = xfer->init(lo, i);
950 if (err)
951 module_put(owner);
952 else
953 lo->lo_encryption = xfer;
954 }
955 return err;
956}
957
Ayan George4c823cc2011-09-21 10:02:13 +0200958static int loop_clr_fd(struct loop_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
960 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400961 gfp_t gfp = lo->old_gfp_mask;
Ayan George4c823cc2011-09-21 10:02:13 +0200962 struct block_device *bdev = lo->lo_device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964 if (lo->lo_state != Lo_bound)
965 return -ENXIO;
966
967 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
968 return -EBUSY;
969
970 if (filp == NULL)
971 return -EINVAL;
972
973 spin_lock_irq(&lo->lo_lock);
974 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 spin_unlock_irq(&lo->lo_lock);
976
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700977 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
Kay Sievers05eb0f22011-07-31 22:21:35 +0200979 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +0200981 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 loop_release_xfer(lo);
984 lo->transfer = NULL;
985 lo->ioctl = NULL;
986 lo->lo_device = NULL;
987 lo->lo_encryption = NULL;
988 lo->lo_offset = 0;
989 lo->lo_sizelimit = 0;
990 lo->lo_encrypt_key_size = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700991 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
993 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
994 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -0500995 if (bdev)
996 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700997 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -0600998 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200999 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001000 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001001 /* let user-space know about this change */
1002 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 mapping_set_gfp_mask(filp->f_mapping, gfp);
1005 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 /* This is safe: open() is still holding a reference. */
1007 module_put(THIS_MODULE);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001008 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
Laurent Vivier476a4812008-03-26 12:11:53 +01001009 ioctl_by_bdev(bdev, BLKRRPART, 0);
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001010 lo->lo_flags = 0;
1011 if (!part_shift)
1012 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001013 mutex_unlock(&lo->lo_ctl_mutex);
1014 /*
1015 * Need not hold lo_ctl_mutex to fput backing file.
1016 * Calling fput holding lo_ctl_mutex triggers a circular
1017 * lock dependency possibility warning as fput can take
1018 * bd_mutex which is usually taken before lo_ctl_mutex.
1019 */
1020 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return 0;
1022}
1023
1024static int
1025loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1026{
1027 int err;
1028 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +11001029 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
David Howellsb0fafa82008-11-14 10:38:41 +11001031 if (lo->lo_encrypt_key_size &&
1032 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 !capable(CAP_SYS_ADMIN))
1034 return -EPERM;
1035 if (lo->lo_state != Lo_bound)
1036 return -ENXIO;
1037 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1038 return -EINVAL;
1039
1040 err = loop_release_xfer(lo);
1041 if (err)
1042 return err;
1043
1044 if (info->lo_encrypt_type) {
1045 unsigned int type = info->lo_encrypt_type;
1046
1047 if (type >= MAX_LO_CRYPT)
1048 return -EINVAL;
1049 xfer = xfer_funcs[type];
1050 if (xfer == NULL)
1051 return -EINVAL;
1052 } else
1053 xfer = NULL;
1054
1055 err = loop_init_xfer(lo, xfer, info);
1056 if (err)
1057 return err;
1058
1059 if (lo->lo_offset != info->lo_offset ||
1060 lo->lo_sizelimit != info->lo_sizelimit) {
1061 lo->lo_offset = info->lo_offset;
1062 lo->lo_sizelimit = info->lo_sizelimit;
1063 if (figure_loop_size(lo))
1064 return -EFBIG;
1065 }
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001066 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1069 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1070 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1071 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1072
1073 if (!xfer)
1074 xfer = &none_funcs;
1075 lo->transfer = xfer->transfer;
1076 lo->ioctl = xfer->ioctl;
1077
David Woodhouse96c58652008-02-06 01:36:27 -08001078 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1079 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1080 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1081
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001082 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1083 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1084 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1085 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1086 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1087 }
1088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1090 lo->lo_init[0] = info->lo_init[0];
1091 lo->lo_init[1] = info->lo_init[1];
1092 if (info->lo_encrypt_key_size) {
1093 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1094 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001095 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097
1098 return 0;
1099}
1100
1101static int
1102loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1103{
1104 struct file *file = lo->lo_backing_file;
1105 struct kstat stat;
1106 int error;
1107
1108 if (lo->lo_state != Lo_bound)
1109 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001110 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 if (error)
1112 return error;
1113 memset(info, 0, sizeof(*info));
1114 info->lo_number = lo->lo_number;
1115 info->lo_device = huge_encode_dev(stat.dev);
1116 info->lo_inode = stat.ino;
1117 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1118 info->lo_offset = lo->lo_offset;
1119 info->lo_sizelimit = lo->lo_sizelimit;
1120 info->lo_flags = lo->lo_flags;
1121 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1122 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1123 info->lo_encrypt_type =
1124 lo->lo_encryption ? lo->lo_encryption->number : 0;
1125 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1126 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1127 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1128 lo->lo_encrypt_key_size);
1129 }
1130 return 0;
1131}
1132
1133static void
1134loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1135{
1136 memset(info64, 0, sizeof(*info64));
1137 info64->lo_number = info->lo_number;
1138 info64->lo_device = info->lo_device;
1139 info64->lo_inode = info->lo_inode;
1140 info64->lo_rdevice = info->lo_rdevice;
1141 info64->lo_offset = info->lo_offset;
1142 info64->lo_sizelimit = 0;
1143 info64->lo_encrypt_type = info->lo_encrypt_type;
1144 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1145 info64->lo_flags = info->lo_flags;
1146 info64->lo_init[0] = info->lo_init[0];
1147 info64->lo_init[1] = info->lo_init[1];
1148 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1149 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1150 else
1151 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1152 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1153}
1154
1155static int
1156loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1157{
1158 memset(info, 0, sizeof(*info));
1159 info->lo_number = info64->lo_number;
1160 info->lo_device = info64->lo_device;
1161 info->lo_inode = info64->lo_inode;
1162 info->lo_rdevice = info64->lo_rdevice;
1163 info->lo_offset = info64->lo_offset;
1164 info->lo_encrypt_type = info64->lo_encrypt_type;
1165 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1166 info->lo_flags = info64->lo_flags;
1167 info->lo_init[0] = info64->lo_init[0];
1168 info->lo_init[1] = info64->lo_init[1];
1169 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1170 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1171 else
1172 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1173 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1174
1175 /* error in case values were truncated */
1176 if (info->lo_device != info64->lo_device ||
1177 info->lo_rdevice != info64->lo_rdevice ||
1178 info->lo_inode != info64->lo_inode ||
1179 info->lo_offset != info64->lo_offset)
1180 return -EOVERFLOW;
1181
1182 return 0;
1183}
1184
1185static int
1186loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1187{
1188 struct loop_info info;
1189 struct loop_info64 info64;
1190
1191 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1192 return -EFAULT;
1193 loop_info64_from_old(&info, &info64);
1194 return loop_set_status(lo, &info64);
1195}
1196
1197static int
1198loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1199{
1200 struct loop_info64 info64;
1201
1202 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1203 return -EFAULT;
1204 return loop_set_status(lo, &info64);
1205}
1206
1207static int
1208loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1209 struct loop_info info;
1210 struct loop_info64 info64;
1211 int err = 0;
1212
1213 if (!arg)
1214 err = -EINVAL;
1215 if (!err)
1216 err = loop_get_status(lo, &info64);
1217 if (!err)
1218 err = loop_info64_to_old(&info64, &info);
1219 if (!err && copy_to_user(arg, &info, sizeof(info)))
1220 err = -EFAULT;
1221
1222 return err;
1223}
1224
1225static int
1226loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1227 struct loop_info64 info64;
1228 int err = 0;
1229
1230 if (!arg)
1231 err = -EINVAL;
1232 if (!err)
1233 err = loop_get_status(lo, &info64);
1234 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1235 err = -EFAULT;
1236
1237 return err;
1238}
1239
J. R. Okajima53d66602009-03-31 15:23:43 -07001240static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1241{
1242 int err;
1243 sector_t sec;
1244 loff_t sz;
1245
1246 err = -ENXIO;
1247 if (unlikely(lo->lo_state != Lo_bound))
1248 goto out;
1249 err = figure_loop_size(lo);
1250 if (unlikely(err))
1251 goto out;
1252 sec = get_capacity(lo->lo_disk);
1253 /* the width of sector_t may be narrow for bit-shift */
1254 sz = sec;
1255 sz <<= 9;
1256 mutex_lock(&bdev->bd_mutex);
1257 bd_set_size(bdev, sz);
David Zeuthenc3473c62010-05-03 14:08:59 +02001258 /* let user-space know about the new size */
1259 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
J. R. Okajima53d66602009-03-31 15:23:43 -07001260 mutex_unlock(&bdev->bd_mutex);
1261
1262 out:
1263 return err;
1264}
1265
Al Virobb214882008-03-02 09:29:48 -05001266static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 unsigned int cmd, unsigned long arg)
1268{
Al Virobb214882008-03-02 09:29:48 -05001269 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 int err;
1271
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001272 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 switch (cmd) {
1274 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001275 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 break;
1277 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001278 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 break;
1280 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001281 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Ayan George4c823cc2011-09-21 10:02:13 +02001282 err = loop_clr_fd(lo);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001283 if (!err)
1284 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 break;
1286 case LOOP_SET_STATUS:
1287 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1288 break;
1289 case LOOP_GET_STATUS:
1290 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1291 break;
1292 case LOOP_SET_STATUS64:
1293 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1294 break;
1295 case LOOP_GET_STATUS64:
1296 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1297 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001298 case LOOP_SET_CAPACITY:
1299 err = -EPERM;
1300 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1301 err = loop_set_capacity(lo, bdev);
1302 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 default:
1304 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1305 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001306 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001307
1308out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 return err;
1310}
1311
David Howells863d5b822006-08-29 19:06:14 +01001312#ifdef CONFIG_COMPAT
1313struct compat_loop_info {
1314 compat_int_t lo_number; /* ioctl r/o */
1315 compat_dev_t lo_device; /* ioctl r/o */
1316 compat_ulong_t lo_inode; /* ioctl r/o */
1317 compat_dev_t lo_rdevice; /* ioctl r/o */
1318 compat_int_t lo_offset;
1319 compat_int_t lo_encrypt_type;
1320 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1321 compat_int_t lo_flags; /* ioctl r/o */
1322 char lo_name[LO_NAME_SIZE];
1323 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1324 compat_ulong_t lo_init[2];
1325 char reserved[4];
1326};
1327
1328/*
1329 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1330 * - noinlined to reduce stack space usage in main part of driver
1331 */
1332static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001333loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001334 struct loop_info64 *info64)
1335{
1336 struct compat_loop_info info;
1337
1338 if (copy_from_user(&info, arg, sizeof(info)))
1339 return -EFAULT;
1340
1341 memset(info64, 0, sizeof(*info64));
1342 info64->lo_number = info.lo_number;
1343 info64->lo_device = info.lo_device;
1344 info64->lo_inode = info.lo_inode;
1345 info64->lo_rdevice = info.lo_rdevice;
1346 info64->lo_offset = info.lo_offset;
1347 info64->lo_sizelimit = 0;
1348 info64->lo_encrypt_type = info.lo_encrypt_type;
1349 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1350 info64->lo_flags = info.lo_flags;
1351 info64->lo_init[0] = info.lo_init[0];
1352 info64->lo_init[1] = info.lo_init[1];
1353 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1354 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1355 else
1356 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1357 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1358 return 0;
1359}
1360
1361/*
1362 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1363 * - noinlined to reduce stack space usage in main part of driver
1364 */
1365static noinline int
1366loop_info64_to_compat(const struct loop_info64 *info64,
1367 struct compat_loop_info __user *arg)
1368{
1369 struct compat_loop_info info;
1370
1371 memset(&info, 0, sizeof(info));
1372 info.lo_number = info64->lo_number;
1373 info.lo_device = info64->lo_device;
1374 info.lo_inode = info64->lo_inode;
1375 info.lo_rdevice = info64->lo_rdevice;
1376 info.lo_offset = info64->lo_offset;
1377 info.lo_encrypt_type = info64->lo_encrypt_type;
1378 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1379 info.lo_flags = info64->lo_flags;
1380 info.lo_init[0] = info64->lo_init[0];
1381 info.lo_init[1] = info64->lo_init[1];
1382 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1383 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1384 else
1385 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1386 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1387
1388 /* error in case values were truncated */
1389 if (info.lo_device != info64->lo_device ||
1390 info.lo_rdevice != info64->lo_rdevice ||
1391 info.lo_inode != info64->lo_inode ||
1392 info.lo_offset != info64->lo_offset ||
1393 info.lo_init[0] != info64->lo_init[0] ||
1394 info.lo_init[1] != info64->lo_init[1])
1395 return -EOVERFLOW;
1396
1397 if (copy_to_user(arg, &info, sizeof(info)))
1398 return -EFAULT;
1399 return 0;
1400}
1401
1402static int
1403loop_set_status_compat(struct loop_device *lo,
1404 const struct compat_loop_info __user *arg)
1405{
1406 struct loop_info64 info64;
1407 int ret;
1408
1409 ret = loop_info64_from_compat(arg, &info64);
1410 if (ret < 0)
1411 return ret;
1412 return loop_set_status(lo, &info64);
1413}
1414
1415static int
1416loop_get_status_compat(struct loop_device *lo,
1417 struct compat_loop_info __user *arg)
1418{
1419 struct loop_info64 info64;
1420 int err = 0;
1421
1422 if (!arg)
1423 err = -EINVAL;
1424 if (!err)
1425 err = loop_get_status(lo, &info64);
1426 if (!err)
1427 err = loop_info64_to_compat(&info64, arg);
1428 return err;
1429}
1430
Al Virobb214882008-03-02 09:29:48 -05001431static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1432 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001433{
Al Virobb214882008-03-02 09:29:48 -05001434 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001435 int err;
1436
David Howells863d5b822006-08-29 19:06:14 +01001437 switch(cmd) {
1438 case LOOP_SET_STATUS:
1439 mutex_lock(&lo->lo_ctl_mutex);
1440 err = loop_set_status_compat(
1441 lo, (const struct compat_loop_info __user *) arg);
1442 mutex_unlock(&lo->lo_ctl_mutex);
1443 break;
1444 case LOOP_GET_STATUS:
1445 mutex_lock(&lo->lo_ctl_mutex);
1446 err = loop_get_status_compat(
1447 lo, (struct compat_loop_info __user *) arg);
1448 mutex_unlock(&lo->lo_ctl_mutex);
1449 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001450 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001451 case LOOP_CLR_FD:
1452 case LOOP_GET_STATUS64:
1453 case LOOP_SET_STATUS64:
1454 arg = (unsigned long) compat_ptr(arg);
1455 case LOOP_SET_FD:
1456 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001457 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001458 break;
1459 default:
1460 err = -ENOIOCTLCMD;
1461 break;
1462 }
David Howells863d5b822006-08-29 19:06:14 +01001463 return err;
1464}
1465#endif
1466
Al Virobb214882008-03-02 09:29:48 -05001467static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
Kay Sievers770fe302011-07-31 22:08:04 +02001469 struct loop_device *lo;
1470 int err = 0;
1471
1472 mutex_lock(&loop_index_mutex);
1473 lo = bdev->bd_disk->private_data;
1474 if (!lo) {
1475 err = -ENXIO;
1476 goto out;
1477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Ingo Molnarf85221d2006-03-23 03:00:38 -08001479 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001481 mutex_unlock(&lo->lo_ctl_mutex);
Kay Sievers770fe302011-07-31 22:08:04 +02001482out:
1483 mutex_unlock(&loop_index_mutex);
1484 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485}
1486
Al Virobb214882008-03-02 09:29:48 -05001487static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
Al Virobb214882008-03-02 09:29:48 -05001489 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001490 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
Ingo Molnarf85221d2006-03-23 03:00:38 -08001492 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001493
Milan Broz14f27932008-12-12 14:48:27 +01001494 if (--lo->lo_refcnt)
1495 goto out;
1496
1497 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1498 /*
1499 * In autoclear mode, stop the loop thread
1500 * and remove configuration after last close.
1501 */
Ayan George4c823cc2011-09-21 10:02:13 +02001502 err = loop_clr_fd(lo);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001503 if (!err)
1504 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001505 } else {
1506 /*
1507 * Otherwise keep thread (if running) and config,
1508 * but flush possible ongoing bios in thread.
1509 */
1510 loop_flush(lo);
1511 }
David Woodhouse96c58652008-02-06 01:36:27 -08001512
Milan Broz14f27932008-12-12 14:48:27 +01001513out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001514 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001515out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return 0;
1517}
1518
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001519static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001521 .open = lo_open,
1522 .release = lo_release,
1523 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001524#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001525 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001526#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527};
1528
1529/*
1530 * And now the modules code and kernel interface.
1531 */
Ken Chen73285082007-05-08 00:28:20 -07001532static int max_loop;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001533module_param(max_loop, int, S_IRUGO);
Ken Chena47653f2007-06-08 13:46:44 -07001534MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Namhyung Kimac04fee2011-05-27 07:59:25 +02001535module_param(max_part, int, S_IRUGO);
Laurent Vivier476a4812008-03-26 12:11:53 +01001536MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537MODULE_LICENSE("GPL");
1538MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1539
1540int loop_register_transfer(struct loop_func_table *funcs)
1541{
1542 unsigned int n = funcs->number;
1543
1544 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1545 return -EINVAL;
1546 xfer_funcs[n] = funcs;
1547 return 0;
1548}
1549
Kay Sievers34dd82a2011-07-31 22:08:04 +02001550static int unregister_transfer_cb(int id, void *ptr, void *data)
1551{
1552 struct loop_device *lo = ptr;
1553 struct loop_func_table *xfer = data;
1554
1555 mutex_lock(&lo->lo_ctl_mutex);
1556 if (lo->lo_encryption == xfer)
1557 loop_release_xfer(lo);
1558 mutex_unlock(&lo->lo_ctl_mutex);
1559 return 0;
1560}
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562int loop_unregister_transfer(int number)
1563{
1564 unsigned int n = number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct loop_func_table *xfer;
1566
1567 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1568 return -EINVAL;
1569
1570 xfer_funcs[n] = NULL;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001571 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 return 0;
1573}
1574
1575EXPORT_SYMBOL(loop_register_transfer);
1576EXPORT_SYMBOL(loop_unregister_transfer);
1577
Kay Sievers34dd82a2011-07-31 22:08:04 +02001578static int loop_add(struct loop_device **l, int i)
Ken Chen73285082007-05-08 00:28:20 -07001579{
1580 struct loop_device *lo;
1581 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001582 int err;
Ken Chen73285082007-05-08 00:28:20 -07001583
1584 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001585 if (!lo) {
1586 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001587 goto out;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001588 }
1589
1590 err = idr_pre_get(&loop_index_idr, GFP_KERNEL);
1591 if (err < 0)
1592 goto out_free_dev;
1593
1594 if (i >= 0) {
1595 int m;
1596
1597 /* create specific i in the index */
1598 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1599 if (err >= 0 && i != m) {
1600 idr_remove(&loop_index_idr, m);
1601 err = -EEXIST;
1602 }
Kay Sievers770fe302011-07-31 22:08:04 +02001603 } else if (i == -1) {
1604 int m;
1605
1606 /* get next free nr */
1607 err = idr_get_new(&loop_index_idr, lo, &m);
1608 if (err >= 0)
1609 i = m;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001610 } else {
1611 err = -EINVAL;
1612 }
1613 if (err < 0)
1614 goto out_free_dev;
Ken Chen73285082007-05-08 00:28:20 -07001615
1616 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1617 if (!lo->lo_queue)
1618 goto out_free_dev;
1619
Laurent Vivier476a4812008-03-26 12:11:53 +01001620 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001621 if (!disk)
1622 goto out_free_queue;
1623
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001624 /*
1625 * Disable partition scanning by default. The in-kernel partition
1626 * scanning can be requested individually per-device during its
1627 * setup. Userspace can always add and remove partitions from all
1628 * devices. The needed partition minors are allocated from the
1629 * extended minor space, the main loop device numbers will continue
1630 * to match the loop minors, regardless of the number of partitions
1631 * used.
1632 *
1633 * If max_part is given, partition scanning is globally enabled for
1634 * all loop devices. The minors for the main loop devices will be
1635 * multiples of max_part.
1636 *
1637 * Note: Global-for-all-devices, set-only-at-init, read-only module
1638 * parameteters like 'max_loop' and 'max_part' make things needlessly
1639 * complicated, are too static, inflexible and may surprise
1640 * userspace tools. Parameters like this in general should be avoided.
1641 */
1642 if (!part_shift)
1643 disk->flags |= GENHD_FL_NO_PART_SCAN;
1644 disk->flags |= GENHD_FL_EXT_DEVT;
Ken Chen73285082007-05-08 00:28:20 -07001645 mutex_init(&lo->lo_ctl_mutex);
1646 lo->lo_number = i;
1647 lo->lo_thread = NULL;
1648 init_waitqueue_head(&lo->lo_event);
1649 spin_lock_init(&lo->lo_lock);
1650 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001651 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001652 disk->fops = &lo_fops;
1653 disk->private_data = lo;
1654 disk->queue = lo->lo_queue;
1655 sprintf(disk->disk_name, "loop%d", i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001656 add_disk(disk);
1657 *l = lo;
1658 return lo->lo_number;
Ken Chen73285082007-05-08 00:28:20 -07001659
1660out_free_queue:
1661 blk_cleanup_queue(lo->lo_queue);
1662out_free_dev:
1663 kfree(lo);
1664out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001665 return err;
Ken Chen73285082007-05-08 00:28:20 -07001666}
1667
Kay Sievers34dd82a2011-07-31 22:08:04 +02001668static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001669{
Kay Sievers34dd82a2011-07-31 22:08:04 +02001670 del_gendisk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001671 blk_cleanup_queue(lo->lo_queue);
1672 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07001673 kfree(lo);
1674}
1675
Kay Sievers770fe302011-07-31 22:08:04 +02001676static int find_free_cb(int id, void *ptr, void *data)
Ken Chena47653f2007-06-08 13:46:44 -07001677{
Kay Sievers770fe302011-07-31 22:08:04 +02001678 struct loop_device *lo = ptr;
1679 struct loop_device **l = data;
Ken Chena47653f2007-06-08 13:46:44 -07001680
Kay Sievers770fe302011-07-31 22:08:04 +02001681 if (lo->lo_state == Lo_unbound) {
1682 *l = lo;
1683 return 1;
Ken Chena47653f2007-06-08 13:46:44 -07001684 }
Kay Sievers770fe302011-07-31 22:08:04 +02001685 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001686}
1687
Kay Sievers34dd82a2011-07-31 22:08:04 +02001688static int loop_lookup(struct loop_device **l, int i)
Ken Chena47653f2007-06-08 13:46:44 -07001689{
Ken Chena47653f2007-06-08 13:46:44 -07001690 struct loop_device *lo;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001691 int ret = -ENODEV;
Ken Chena47653f2007-06-08 13:46:44 -07001692
Kay Sievers770fe302011-07-31 22:08:04 +02001693 if (i < 0) {
1694 int err;
1695
1696 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1697 if (err == 1) {
1698 *l = lo;
1699 ret = lo->lo_number;
1700 }
1701 goto out;
1702 }
1703
1704 /* lookup and return a specific i */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001705 lo = idr_find(&loop_index_idr, i);
Ken Chena47653f2007-06-08 13:46:44 -07001706 if (lo) {
Kay Sievers34dd82a2011-07-31 22:08:04 +02001707 *l = lo;
1708 ret = lo->lo_number;
Ken Chena47653f2007-06-08 13:46:44 -07001709 }
Kay Sievers770fe302011-07-31 22:08:04 +02001710out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02001711 return ret;
Ken Chena47653f2007-06-08 13:46:44 -07001712}
1713
Ken Chen73285082007-05-08 00:28:20 -07001714static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1715{
Al Viro705962c2007-05-13 05:52:32 -04001716 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001717 struct kobject *kobj;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001718 int err;
Ken Chen73285082007-05-08 00:28:20 -07001719
Kay Sievers34dd82a2011-07-31 22:08:04 +02001720 mutex_lock(&loop_index_mutex);
1721 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1722 if (err < 0)
1723 err = loop_add(&lo, MINOR(dev) >> part_shift);
1724 if (err < 0)
1725 kobj = ERR_PTR(err);
1726 else
1727 kobj = get_disk(lo->lo_disk);
1728 mutex_unlock(&loop_index_mutex);
Ken Chen73285082007-05-08 00:28:20 -07001729
1730 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001731 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001732}
1733
Kay Sievers770fe302011-07-31 22:08:04 +02001734static long loop_control_ioctl(struct file *file, unsigned int cmd,
1735 unsigned long parm)
1736{
1737 struct loop_device *lo;
1738 int ret = -ENOSYS;
1739
1740 mutex_lock(&loop_index_mutex);
1741 switch (cmd) {
1742 case LOOP_CTL_ADD:
1743 ret = loop_lookup(&lo, parm);
1744 if (ret >= 0) {
1745 ret = -EEXIST;
1746 break;
1747 }
1748 ret = loop_add(&lo, parm);
1749 break;
1750 case LOOP_CTL_REMOVE:
1751 ret = loop_lookup(&lo, parm);
1752 if (ret < 0)
1753 break;
1754 mutex_lock(&lo->lo_ctl_mutex);
1755 if (lo->lo_state != Lo_unbound) {
1756 ret = -EBUSY;
1757 mutex_unlock(&lo->lo_ctl_mutex);
1758 break;
1759 }
1760 if (lo->lo_refcnt > 0) {
1761 ret = -EBUSY;
1762 mutex_unlock(&lo->lo_ctl_mutex);
1763 break;
1764 }
1765 lo->lo_disk->private_data = NULL;
1766 mutex_unlock(&lo->lo_ctl_mutex);
1767 idr_remove(&loop_index_idr, lo->lo_number);
1768 loop_remove(lo);
1769 break;
1770 case LOOP_CTL_GET_FREE:
1771 ret = loop_lookup(&lo, -1);
1772 if (ret >= 0)
1773 break;
1774 ret = loop_add(&lo, -1);
1775 }
1776 mutex_unlock(&loop_index_mutex);
1777
1778 return ret;
1779}
1780
1781static const struct file_operations loop_ctl_fops = {
1782 .open = nonseekable_open,
1783 .unlocked_ioctl = loop_control_ioctl,
1784 .compat_ioctl = loop_control_ioctl,
1785 .owner = THIS_MODULE,
1786 .llseek = noop_llseek,
1787};
1788
1789static struct miscdevice loop_misc = {
1790 .minor = LOOP_CTRL_MINOR,
1791 .name = "loop-control",
1792 .fops = &loop_ctl_fops,
1793};
1794
1795MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1796MODULE_ALIAS("devname:loop-control");
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798static int __init loop_init(void)
1799{
Ken Chena47653f2007-06-08 13:46:44 -07001800 int i, nr;
1801 unsigned long range;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001802 struct loop_device *lo;
Kay Sievers770fe302011-07-31 22:08:04 +02001803 int err;
Ken Chena47653f2007-06-08 13:46:44 -07001804
Kay Sievers770fe302011-07-31 22:08:04 +02001805 err = misc_register(&loop_misc);
1806 if (err < 0)
1807 return err;
Laurent Vivier476a4812008-03-26 12:11:53 +01001808
1809 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02001810 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01001811 part_shift = fls(max_part);
1812
Namhyung Kimac04fee2011-05-27 07:59:25 +02001813 /*
1814 * Adjust max_part according to part_shift as it is exported
1815 * to user space so that user can decide correct minor number
1816 * if [s]he want to create more devices.
1817 *
1818 * Note that -1 is required because partition 0 is reserved
1819 * for the whole disk.
1820 */
1821 max_part = (1UL << part_shift) - 1;
1822 }
1823
Namhyung Kim78f4bb32011-05-24 16:48:54 +02001824 if ((1UL << part_shift) > DISK_MAX_PARTS)
1825 return -EINVAL;
1826
Laurent Vivier476a4812008-03-26 12:11:53 +01001827 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001828 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Kay Sieversd134b002011-07-31 22:08:04 +02001830 /*
1831 * If max_loop is specified, create that many devices upfront.
1832 * This also becomes a hard limit. If max_loop is not specified,
1833 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1834 * init time. Loop devices can be requested on-demand with the
1835 * /dev/loop-control interface, or be instantiated by accessing
1836 * a 'dead' device node.
1837 */
Ken Chen73285082007-05-08 00:28:20 -07001838 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001839 nr = max_loop;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001840 range = max_loop << part_shift;
Ken Chena47653f2007-06-08 13:46:44 -07001841 } else {
Kay Sieversd134b002011-07-31 22:08:04 +02001842 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Namhyung Kima1c15c52011-05-24 16:48:55 +02001843 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 }
Ken Chena47653f2007-06-08 13:46:44 -07001845
1846 if (register_blkdev(LOOP_MAJOR, "loop"))
1847 return -EIO;
1848
Ken Chena47653f2007-06-08 13:46:44 -07001849 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1850 THIS_MODULE, loop_probe, NULL, NULL);
1851
Kay Sieversd134b002011-07-31 22:08:04 +02001852 /* pre-create number of devices given by config or max_loop */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001853 mutex_lock(&loop_index_mutex);
1854 for (i = 0; i < nr; i++)
1855 loop_add(&lo, i);
1856 mutex_unlock(&loop_index_mutex);
1857
Ken Chen73285082007-05-08 00:28:20 -07001858 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 return 0;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001860}
Ken Chena47653f2007-06-08 13:46:44 -07001861
Kay Sievers34dd82a2011-07-31 22:08:04 +02001862static int loop_exit_cb(int id, void *ptr, void *data)
1863{
1864 struct loop_device *lo = ptr;
Ken Chena47653f2007-06-08 13:46:44 -07001865
Kay Sievers34dd82a2011-07-31 22:08:04 +02001866 loop_remove(lo);
1867 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868}
1869
Ken Chen73285082007-05-08 00:28:20 -07001870static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
Ken Chena47653f2007-06-08 13:46:44 -07001872 unsigned long range;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Namhyung Kima1c15c52011-05-24 16:48:55 +02001874 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
Ken Chena47653f2007-06-08 13:46:44 -07001875
Kay Sievers34dd82a2011-07-31 22:08:04 +02001876 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1877 idr_remove_all(&loop_index_idr);
1878 idr_destroy(&loop_index_idr);
Ken Chen73285082007-05-08 00:28:20 -07001879
Ken Chena47653f2007-06-08 13:46:44 -07001880 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001881 unregister_blkdev(LOOP_MAJOR, "loop");
Kay Sievers770fe302011-07-31 22:08:04 +02001882
1883 misc_deregister(&loop_misc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884}
1885
1886module_init(loop_init);
1887module_exit(loop_exit);
1888
1889#ifndef MODULE
1890static int __init max_loop_setup(char *str)
1891{
1892 max_loop = simple_strtol(str, NULL, 0);
1893 return 1;
1894}
1895
1896__setup("max_loop=", max_loop_setup);
1897#endif