blob: 801f4ab83302556e9b48f1ddf139201a05e8c304 [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/smp_lock.h>
65#include <linux/swap.h>
66#include <linux/slab.h>
67#include <linux/loop.h>
David Howells863d5b82006-08-29 19:06:14 +010068#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070070#include <linux/freezer.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>
75#include <linux/gfp.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070076#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020077#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79#include <asm/uaccess.h>
80
Ken Chen73285082007-05-08 00:28:20 -070081static LIST_HEAD(loop_devices);
82static DEFINE_MUTEX(loop_devices_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Laurent Vivier476a4812008-03-26 12:11:53 +010084static int max_part;
85static int part_shift;
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*
88 * Transfer functions
89 */
90static int transfer_none(struct loop_device *lo, int cmd,
91 struct page *raw_page, unsigned raw_off,
92 struct page *loop_page, unsigned loop_off,
93 int size, sector_t real_block)
94{
95 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
96 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
97
98 if (cmd == READ)
99 memcpy(loop_buf, raw_buf, size);
100 else
101 memcpy(raw_buf, loop_buf, size);
102
103 kunmap_atomic(raw_buf, KM_USER0);
104 kunmap_atomic(loop_buf, KM_USER1);
105 cond_resched();
106 return 0;
107}
108
109static int transfer_xor(struct loop_device *lo, int cmd,
110 struct page *raw_page, unsigned raw_off,
111 struct page *loop_page, unsigned loop_off,
112 int size, sector_t real_block)
113{
114 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
115 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
116 char *in, *out, *key;
117 int i, keysize;
118
119 if (cmd == READ) {
120 in = raw_buf;
121 out = loop_buf;
122 } else {
123 in = loop_buf;
124 out = raw_buf;
125 }
126
127 key = lo->lo_encrypt_key;
128 keysize = lo->lo_encrypt_key_size;
129 for (i = 0; i < size; i++)
130 *out++ = *in++ ^ key[(i & 511) % keysize];
131
132 kunmap_atomic(raw_buf, KM_USER0);
133 kunmap_atomic(loop_buf, KM_USER1);
134 cond_resched();
135 return 0;
136}
137
138static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
139{
140 if (unlikely(info->lo_encrypt_key_size <= 0))
141 return -EINVAL;
142 return 0;
143}
144
145static struct loop_func_table none_funcs = {
146 .number = LO_CRYPT_NONE,
147 .transfer = transfer_none,
148};
149
150static struct loop_func_table xor_funcs = {
151 .number = LO_CRYPT_XOR,
152 .transfer = transfer_xor,
153 .init = xor_init
154};
155
156/* xfer_funcs[0] is special - its release function is never called */
157static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
158 &none_funcs,
159 &xor_funcs
160};
161
162static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163{
164 loff_t size, offset, loopsize;
165
166 /* Compute loopsize in bytes */
167 size = i_size_read(file->f_mapping->host);
168 offset = lo->lo_offset;
169 loopsize = size - offset;
170 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
171 loopsize = lo->lo_sizelimit;
172
173 /*
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
176 */
177 return loopsize >> 9;
178}
179
180static int
181figure_loop_size(struct loop_device *lo)
182{
183 loff_t size = get_loop_size(lo, lo->lo_backing_file);
184 sector_t x = (sector_t)size;
185
186 if (unlikely((loff_t)x != size))
187 return -EFBIG;
188
Ken Chen73285082007-05-08 00:28:20 -0700189 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 return 0;
191}
192
193static inline int
194lo_do_transfer(struct loop_device *lo, int cmd,
195 struct page *rpage, unsigned roffs,
196 struct page *lpage, unsigned loffs,
197 int size, sector_t rblock)
198{
199 if (unlikely(!lo->transfer))
200 return 0;
201
202 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
203}
204
205/**
206 * do_lo_send_aops - helper for writing data to a loop device
207 *
208 * This is the fast version for backing filesystems which implement the address
Nick Pigginafddba42007-10-16 01:25:01 -0700209 * space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 */
211static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400212 loff_t pos, struct page *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
215 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 pgoff_t index;
217 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800218 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800220 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 index = pos >> PAGE_CACHE_SHIFT;
222 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
223 bv_offs = bvec->bv_offset;
224 len = bvec->bv_len;
225 while (len > 0) {
226 sector_t IV;
Nick Pigginafddba42007-10-16 01:25:01 -0700227 unsigned size, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 int transfer_result;
Nick Pigginafddba42007-10-16 01:25:01 -0700229 struct page *page;
230 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
233 size = PAGE_CACHE_SIZE - offset;
234 if (size > len)
235 size = len;
Nick Pigginafddba42007-10-16 01:25:01 -0700236
237 ret = pagecache_write_begin(file, mapping, pos, size, 0,
238 &page, &fsdata);
239 if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
243 bvec->bv_page, bv_offs, size, IV);
Nick Pigginafddba42007-10-16 01:25:01 -0700244 copied = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 if (unlikely(transfer_result))
Nick Pigginafddba42007-10-16 01:25:01 -0700246 copied = 0;
247
248 ret = pagecache_write_end(file, mapping, pos, size, copied,
249 page, fsdata);
Dmitry Monakhov8268f5a2007-10-16 01:25:02 -0700250 if (ret < 0 || ret != copied)
Nick Pigginafddba42007-10-16 01:25:01 -0700251 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700252
253 if (unlikely(transfer_result))
254 goto fail;
255
256 bv_offs += copied;
257 len -= copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 offset = 0;
259 index++;
Nick Pigginafddba42007-10-16 01:25:01 -0700260 pos += copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800262 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800264 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266fail:
267 ret = -1;
268 goto out;
269}
270
271/**
272 * __do_lo_send_write - helper for writing data to a loop device
273 *
274 * This helper just factors out common code between do_lo_send_direct_write()
275 * and do_lo_send_write().
276 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800277static int __do_lo_send_write(struct file *file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100278 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 ssize_t bw;
281 mm_segment_t old_fs = get_fs();
282
283 set_fs(get_ds());
284 bw = file->f_op->write(file, buf, len, &pos);
285 set_fs(old_fs);
286 if (likely(bw == len))
287 return 0;
288 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
289 (unsigned long long)pos, len);
290 if (bw >= 0)
291 bw = -EIO;
292 return bw;
293}
294
295/**
296 * do_lo_send_direct_write - helper for writing data to a loop device
297 *
298 * This is the fast, non-transforming version for backing filesystems which do
Nick Pigginafddba42007-10-16 01:25:01 -0700299 * not implement the address space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * It uses the write file operation which should be present on all writeable
301 * filesystems.
302 */
303static int do_lo_send_direct_write(struct loop_device *lo,
Al Viro511de732007-10-08 12:10:13 -0400304 struct bio_vec *bvec, loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100307 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 bvec->bv_len, pos);
309 kunmap(bvec->bv_page);
310 cond_resched();
311 return bw;
312}
313
314/**
315 * do_lo_send_write - helper for writing data to a loop device
316 *
317 * This is the slow, transforming version for filesystems which do not
Nick Pigginafddba42007-10-16 01:25:01 -0700318 * implement the address space operations write_begin and write_end. It
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * uses the write file operation which should be present on all writeable
320 * filesystems.
321 *
322 * Using fops->write is slower than using aops->{prepare,commit}_write in the
323 * transforming case because we need to double buffer the data as we cannot do
324 * the transformations in place as we do not have direct access to the
325 * destination pages of the backing file.
326 */
327static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400328 loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
330 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
331 bvec->bv_offset, bvec->bv_len, pos >> 9);
332 if (likely(!ret))
333 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100334 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 pos);
336 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
337 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
338 if (ret > 0)
339 ret = -EIO;
340 return ret;
341}
342
Al Viro511de732007-10-08 12:10:13 -0400343static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
Al Viro511de732007-10-08 12:10:13 -0400345 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 struct page *page);
347 struct bio_vec *bvec;
348 struct page *page = NULL;
349 int i, ret = 0;
350
351 do_lo_send = do_lo_send_aops;
352 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
353 do_lo_send = do_lo_send_direct_write;
354 if (lo->transfer != transfer_none) {
355 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
356 if (unlikely(!page))
357 goto fail;
358 kmap(page);
359 do_lo_send = do_lo_send_write;
360 }
361 }
362 bio_for_each_segment(bvec, bio, i) {
Al Viro511de732007-10-08 12:10:13 -0400363 ret = do_lo_send(lo, bvec, pos, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 if (ret < 0)
365 break;
366 pos += bvec->bv_len;
367 }
368 if (page) {
369 kunmap(page);
370 __free_page(page);
371 }
372out:
373 return ret;
374fail:
375 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
376 ret = -ENOMEM;
377 goto out;
378}
379
380struct lo_read_data {
381 struct loop_device *lo;
382 struct page *page;
383 unsigned offset;
384 int bsize;
385};
386
387static int
Jens Axboefd582142007-06-12 21:20:37 +0200388lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
389 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Jens Axboefd582142007-06-12 21:20:37 +0200391 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200393 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 sector_t IV;
Roel Kluina3941ec2009-03-05 08:03:53 +0100395 int size, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Jens Axboecac36bb2007-06-14 13:10:48 +0200397 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200398 if (unlikely(ret))
399 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Jens Axboefd582142007-06-12 21:20:37 +0200401 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
402 (buf->offset >> 9);
403 size = sd->len;
404 if (size > p->bsize)
405 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Jens Axboefd582142007-06-12 21:20:37 +0200407 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 printk(KERN_ERR "loop: transfer error block %ld\n",
409 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200410 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
412
413 flush_dcache_page(p->page);
414
Jens Axboefd582142007-06-12 21:20:37 +0200415 if (size > 0)
416 p->offset += size;
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return size;
419}
420
421static int
Jens Axboefd582142007-06-12 21:20:37 +0200422lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
423{
424 return __splice_from_pipe(pipe, sd, lo_splice_actor);
425}
426
427static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428do_lo_receive(struct loop_device *lo,
429 struct bio_vec *bvec, int bsize, loff_t pos)
430{
431 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200432 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200434 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 cookie.lo = lo;
437 cookie.page = bvec->bv_page;
438 cookie.offset = bvec->bv_offset;
439 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200440
441 sd.len = 0;
442 sd.total_len = bvec->bv_len;
443 sd.flags = 0;
444 sd.pos = pos;
445 sd.u.data = &cookie;
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200448 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
449
450 if (retval < 0)
451 return retval;
452
453 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456static int
457lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
458{
459 struct bio_vec *bvec;
460 int i, ret = 0;
461
462 bio_for_each_segment(bvec, bio, i) {
463 ret = do_lo_receive(lo, bvec, bsize, pos);
464 if (ret < 0)
465 break;
466 pos += bvec->bv_len;
467 }
468 return ret;
469}
470
471static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
472{
473 loff_t pos;
474 int ret;
475
476 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100477
478 if (bio_rw(bio) == WRITE) {
479 int barrier = bio_barrier(bio);
480 struct file *file = lo->lo_backing_file;
481
482 if (barrier) {
483 if (unlikely(!file->f_op->fsync)) {
484 ret = -EOPNOTSUPP;
485 goto out;
486 }
487
488 ret = vfs_fsync(file, file->f_path.dentry, 0);
489 if (unlikely(ret)) {
490 ret = -EIO;
491 goto out;
492 }
493 }
494
Al Viro511de732007-10-08 12:10:13 -0400495 ret = lo_send(lo, bio, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100496
497 if (barrier && !ret) {
498 ret = vfs_fsync(file, file->f_path.dentry, 0);
499 if (unlikely(ret))
500 ret = -EIO;
501 }
502 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100504
505out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return ret;
507}
508
509/*
510 * Add bio to back of pending list
511 */
512static void loop_add_bio(struct loop_device *lo, struct bio *bio)
513{
Akinobu Mitae6863072009-04-17 08:41:21 +0200514 bio_list_add(&lo->lo_bio_list, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
517/*
518 * Grab first pending buffer
519 */
520static struct bio *loop_get_bio(struct loop_device *lo)
521{
Akinobu Mitae6863072009-04-17 08:41:21 +0200522 return bio_list_pop(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523}
524
Jens Axboe165125e2007-07-24 09:28:11 +0200525static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
527 struct loop_device *lo = q->queuedata;
528 int rw = bio_rw(old_bio);
529
Nick Piggin35a82d12005-06-23 00:09:06 -0700530 if (rw == READA)
531 rw = READ;
532
533 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 spin_lock_irq(&lo->lo_lock);
536 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700537 goto out;
538 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
539 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700541 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700542 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700546 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200547 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
551/*
552 * kick off io on the underlying address space
553 */
Jens Axboe165125e2007-07-24 09:28:11 +0200554static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 struct loop_device *lo = q->queuedata;
557
Nick Piggin75ad23b2008-04-29 14:48:33 +0200558 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 blk_run_address_space(lo->lo_backing_file->f_mapping);
560}
561
562struct switch_request {
563 struct file *file;
564 struct completion wait;
565};
566
567static void do_loop_switch(struct loop_device *, struct switch_request *);
568
569static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
570{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 if (unlikely(!bio->bi_bdev)) {
572 do_loop_switch(lo, bio->bi_private);
573 bio_put(bio);
574 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700575 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200576 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 }
578}
579
580/*
581 * worker thread that handles reads/writes to file backed loop devices,
582 * to avoid blocking in our make_request_fn. it also does loop decrypting
583 * on reads for block backed loop, as that is too heavy to do from
584 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700585 *
586 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
587 * calling kthread_stop(). Therefore once kthread_should_stop() is
588 * true, make_request will not place any more requests. Therefore
589 * once kthread_should_stop() is true and lo_bio is NULL, we are
590 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 */
592static int loop_thread(void *data)
593{
594 struct loop_device *lo = data;
595 struct bio *bio;
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 set_user_nice(current, -20);
598
Akinobu Mitae6863072009-04-17 08:41:21 +0200599 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700601 wait_event_interruptible(lo->lo_event,
Akinobu Mitae6863072009-04-17 08:41:21 +0200602 !bio_list_empty(&lo->lo_bio_list) ||
603 kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700604
Akinobu Mitae6863072009-04-17 08:41:21 +0200605 if (bio_list_empty(&lo->lo_bio_list))
Nick Piggin35a82d12005-06-23 00:09:06 -0700606 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700607 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700609 spin_unlock_irq(&lo->lo_lock);
610
611 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 }
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 return 0;
616}
617
618/*
619 * loop_switch performs the hard work of switching a backing store.
620 * First it needs to flush existing IO, it does this by sending a magic
621 * BIO down the pipe. The completion of this BIO does the actual switch.
622 */
623static int loop_switch(struct loop_device *lo, struct file *file)
624{
625 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100626 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 if (!bio)
628 return -ENOMEM;
629 init_completion(&w.wait);
630 w.file = file;
631 bio->bi_private = &w;
632 bio->bi_bdev = NULL;
633 loop_make_request(lo->lo_queue, bio);
634 wait_for_completion(&w.wait);
635 return 0;
636}
637
638/*
Milan Broz14f27932008-12-12 14:48:27 +0100639 * Helper to flush the IOs in loop, but keeping loop thread running
640 */
641static int loop_flush(struct loop_device *lo)
642{
643 /* loop not yet configured, no running thread, nothing to flush */
644 if (!lo->lo_thread)
645 return 0;
646
647 return loop_switch(lo, NULL);
648}
649
650/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 * Do the actual switch; called from the BIO completion routine
652 */
653static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
654{
655 struct file *file = p->file;
656 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100657 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Milan Broz14f27932008-12-12 14:48:27 +0100659 /* if no new file, only flush of queued bios requested */
660 if (!file)
661 goto out;
662
663 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
665 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700666 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
667 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 lo->old_gfp_mask = mapping_gfp_mask(mapping);
669 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100670out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 complete(&p->wait);
672}
673
674
675/*
676 * loop_change_fd switched the backing store of a loopback device to
677 * a new file. This is useful for operating system installers to free up
678 * the original file and in High Availability environments to switch to
679 * an alternative location for the content in case of server meltdown.
680 * This can only work if the loop device is used read-only, and if the
681 * new backing store is the same size and type as the old backing store.
682 */
Al Virobb214882008-03-02 09:29:48 -0500683static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
684 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
686 struct file *file, *old_file;
687 struct inode *inode;
688 int error;
689
690 error = -ENXIO;
691 if (lo->lo_state != Lo_bound)
692 goto out;
693
694 /* the loop device has to be read-only */
695 error = -EINVAL;
696 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
697 goto out;
698
699 error = -EBADF;
700 file = fget(arg);
701 if (!file)
702 goto out;
703
704 inode = file->f_mapping->host;
705 old_file = lo->lo_backing_file;
706
707 error = -EINVAL;
708
709 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
710 goto out_putf;
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 /* size of the new backing store needs to be the same */
713 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
714 goto out_putf;
715
716 /* and ... switch */
717 error = loop_switch(lo, file);
718 if (error)
719 goto out_putf;
720
721 fput(old_file);
Laurent Vivier476a4812008-03-26 12:11:53 +0100722 if (max_part > 0)
723 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return 0;
725
726 out_putf:
727 fput(file);
728 out:
729 return error;
730}
731
732static inline int is_loop_device(struct file *file)
733{
734 struct inode *i = file->f_mapping->host;
735
736 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
737}
738
Al Virobb214882008-03-02 09:29:48 -0500739static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 struct block_device *bdev, unsigned int arg)
741{
742 struct file *file, *f;
743 struct inode *inode;
744 struct address_space *mapping;
745 unsigned lo_blocksize;
746 int lo_flags = 0;
747 int error;
748 loff_t size;
749
750 /* This is safe, since we have a reference from open(). */
751 __module_get(THIS_MODULE);
752
753 error = -EBADF;
754 file = fget(arg);
755 if (!file)
756 goto out;
757
758 error = -EBUSY;
759 if (lo->lo_state != Lo_unbound)
760 goto out_putf;
761
762 /* Avoid recursion */
763 f = file;
764 while (is_loop_device(f)) {
765 struct loop_device *l;
766
Al Virobb214882008-03-02 09:29:48 -0500767 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 goto out_putf;
769
770 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
771 if (l->lo_state == Lo_unbound) {
772 error = -EINVAL;
773 goto out_putf;
774 }
775 f = l->lo_backing_file;
776 }
777
778 mapping = file->f_mapping;
779 inode = mapping->host;
780
781 if (!(file->f_mode & FMODE_WRITE))
782 lo_flags |= LO_FLAGS_READ_ONLY;
783
784 error = -EINVAL;
785 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700786 const struct address_space_operations *aops = mapping->a_ops;
Miklos Szeredi68181732009-05-07 15:37:36 +0200787
Nick Piggin4e02ed42008-10-29 14:00:55 -0700788 if (aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 lo_flags |= LO_FLAGS_USE_AOPS;
790 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
791 lo_flags |= LO_FLAGS_READ_ONLY;
792
Theodore Ts'oba52de12006-09-27 01:50:49 -0700793 lo_blocksize = S_ISBLK(inode->i_mode) ?
794 inode->i_bdev->bd_block_size : PAGE_SIZE;
795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 error = 0;
797 } else {
798 goto out_putf;
799 }
800
801 size = get_loop_size(lo, file);
802
803 if ((loff_t)(sector_t)size != size) {
804 error = -EFBIG;
805 goto out_putf;
806 }
807
Al Virobb214882008-03-02 09:29:48 -0500808 if (!(mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 lo_flags |= LO_FLAGS_READ_ONLY;
810
811 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
812
813 lo->lo_blocksize = lo_blocksize;
814 lo->lo_device = bdev;
815 lo->lo_flags = lo_flags;
816 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700817 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 lo->ioctl = NULL;
819 lo->lo_sizelimit = 0;
820 lo->old_gfp_mask = mapping_gfp_mask(mapping);
821 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
822
Akinobu Mitae6863072009-04-17 08:41:21 +0200823 bio_list_init(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 /*
826 * set queue make_request_fn, and add limits based on lower level
827 * device
828 */
829 blk_queue_make_request(lo->lo_queue, loop_make_request);
830 lo->lo_queue->queuedata = lo;
831 lo->lo_queue->unplug_fn = loop_unplug;
832
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100833 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
834 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
835
Ken Chen73285082007-05-08 00:28:20 -0700836 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 bd_set_size(bdev, size << 9);
838
839 set_blocksize(bdev, lo_blocksize);
840
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700841 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
842 lo->lo_number);
843 if (IS_ERR(lo->lo_thread)) {
844 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700845 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700846 }
847 lo->lo_state = Lo_bound;
848 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100849 if (max_part > 0)
850 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 return 0;
852
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700853out_clr:
854 lo->lo_thread = NULL;
855 lo->lo_device = NULL;
856 lo->lo_backing_file = NULL;
857 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700858 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700859 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700860 bd_set_size(bdev, 0);
861 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
862 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 out_putf:
864 fput(file);
865 out:
866 /* This is safe: open() is still holding a reference. */
867 module_put(THIS_MODULE);
868 return error;
869}
870
871static int
872loop_release_xfer(struct loop_device *lo)
873{
874 int err = 0;
875 struct loop_func_table *xfer = lo->lo_encryption;
876
877 if (xfer) {
878 if (xfer->release)
879 err = xfer->release(lo);
880 lo->transfer = NULL;
881 lo->lo_encryption = NULL;
882 module_put(xfer->owner);
883 }
884 return err;
885}
886
887static int
888loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
889 const struct loop_info64 *i)
890{
891 int err = 0;
892
893 if (xfer) {
894 struct module *owner = xfer->owner;
895
896 if (!try_module_get(owner))
897 return -EINVAL;
898 if (xfer->init)
899 err = xfer->init(lo, i);
900 if (err)
901 module_put(owner);
902 else
903 lo->lo_encryption = xfer;
904 }
905 return err;
906}
907
908static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
909{
910 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400911 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 if (lo->lo_state != Lo_bound)
914 return -ENXIO;
915
916 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
917 return -EBUSY;
918
919 if (filp == NULL)
920 return -EINVAL;
921
922 spin_lock_irq(&lo->lo_lock);
923 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 spin_unlock_irq(&lo->lo_lock);
925
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700926 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Milan Broz8ae30b82008-12-12 14:50:49 +0100928 lo->lo_queue->unplug_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 lo->lo_backing_file = NULL;
930
931 loop_release_xfer(lo);
932 lo->transfer = NULL;
933 lo->ioctl = NULL;
934 lo->lo_device = NULL;
935 lo->lo_encryption = NULL;
936 lo->lo_offset = 0;
937 lo->lo_sizelimit = 0;
938 lo->lo_encrypt_key_size = 0;
939 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700940 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
942 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
943 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -0500944 if (bdev)
945 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700946 set_capacity(lo->lo_disk, 0);
Al Virobb214882008-03-02 09:29:48 -0500947 if (bdev)
948 bd_set_size(bdev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 mapping_set_gfp_mask(filp->f_mapping, gfp);
950 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 /* This is safe: open() is still holding a reference. */
952 module_put(THIS_MODULE);
Laurent Vivier476a4812008-03-26 12:11:53 +0100953 if (max_part > 0)
954 ioctl_by_bdev(bdev, BLKRRPART, 0);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +0100955 mutex_unlock(&lo->lo_ctl_mutex);
956 /*
957 * Need not hold lo_ctl_mutex to fput backing file.
958 * Calling fput holding lo_ctl_mutex triggers a circular
959 * lock dependency possibility warning as fput can take
960 * bd_mutex which is usually taken before lo_ctl_mutex.
961 */
962 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 return 0;
964}
965
966static int
967loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
968{
969 int err;
970 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +1100971 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
David Howellsb0fafa82008-11-14 10:38:41 +1100973 if (lo->lo_encrypt_key_size &&
974 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 !capable(CAP_SYS_ADMIN))
976 return -EPERM;
977 if (lo->lo_state != Lo_bound)
978 return -ENXIO;
979 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
980 return -EINVAL;
981
982 err = loop_release_xfer(lo);
983 if (err)
984 return err;
985
986 if (info->lo_encrypt_type) {
987 unsigned int type = info->lo_encrypt_type;
988
989 if (type >= MAX_LO_CRYPT)
990 return -EINVAL;
991 xfer = xfer_funcs[type];
992 if (xfer == NULL)
993 return -EINVAL;
994 } else
995 xfer = NULL;
996
997 err = loop_init_xfer(lo, xfer, info);
998 if (err)
999 return err;
1000
1001 if (lo->lo_offset != info->lo_offset ||
1002 lo->lo_sizelimit != info->lo_sizelimit) {
1003 lo->lo_offset = info->lo_offset;
1004 lo->lo_sizelimit = info->lo_sizelimit;
1005 if (figure_loop_size(lo))
1006 return -EFBIG;
1007 }
1008
1009 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1010 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1011 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1012 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1013
1014 if (!xfer)
1015 xfer = &none_funcs;
1016 lo->transfer = xfer->transfer;
1017 lo->ioctl = xfer->ioctl;
1018
David Woodhouse96c58652008-02-06 01:36:27 -08001019 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1020 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1021 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1024 lo->lo_init[0] = info->lo_init[0];
1025 lo->lo_init[1] = info->lo_init[1];
1026 if (info->lo_encrypt_key_size) {
1027 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1028 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001029 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 }
1031
1032 return 0;
1033}
1034
1035static int
1036loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1037{
1038 struct file *file = lo->lo_backing_file;
1039 struct kstat stat;
1040 int error;
1041
1042 if (lo->lo_state != Lo_bound)
1043 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001044 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 if (error)
1046 return error;
1047 memset(info, 0, sizeof(*info));
1048 info->lo_number = lo->lo_number;
1049 info->lo_device = huge_encode_dev(stat.dev);
1050 info->lo_inode = stat.ino;
1051 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1052 info->lo_offset = lo->lo_offset;
1053 info->lo_sizelimit = lo->lo_sizelimit;
1054 info->lo_flags = lo->lo_flags;
1055 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1056 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1057 info->lo_encrypt_type =
1058 lo->lo_encryption ? lo->lo_encryption->number : 0;
1059 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1060 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1061 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1062 lo->lo_encrypt_key_size);
1063 }
1064 return 0;
1065}
1066
1067static void
1068loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1069{
1070 memset(info64, 0, sizeof(*info64));
1071 info64->lo_number = info->lo_number;
1072 info64->lo_device = info->lo_device;
1073 info64->lo_inode = info->lo_inode;
1074 info64->lo_rdevice = info->lo_rdevice;
1075 info64->lo_offset = info->lo_offset;
1076 info64->lo_sizelimit = 0;
1077 info64->lo_encrypt_type = info->lo_encrypt_type;
1078 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1079 info64->lo_flags = info->lo_flags;
1080 info64->lo_init[0] = info->lo_init[0];
1081 info64->lo_init[1] = info->lo_init[1];
1082 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1083 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1084 else
1085 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1086 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1087}
1088
1089static int
1090loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1091{
1092 memset(info, 0, sizeof(*info));
1093 info->lo_number = info64->lo_number;
1094 info->lo_device = info64->lo_device;
1095 info->lo_inode = info64->lo_inode;
1096 info->lo_rdevice = info64->lo_rdevice;
1097 info->lo_offset = info64->lo_offset;
1098 info->lo_encrypt_type = info64->lo_encrypt_type;
1099 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1100 info->lo_flags = info64->lo_flags;
1101 info->lo_init[0] = info64->lo_init[0];
1102 info->lo_init[1] = info64->lo_init[1];
1103 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1104 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1105 else
1106 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1107 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1108
1109 /* error in case values were truncated */
1110 if (info->lo_device != info64->lo_device ||
1111 info->lo_rdevice != info64->lo_rdevice ||
1112 info->lo_inode != info64->lo_inode ||
1113 info->lo_offset != info64->lo_offset)
1114 return -EOVERFLOW;
1115
1116 return 0;
1117}
1118
1119static int
1120loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1121{
1122 struct loop_info info;
1123 struct loop_info64 info64;
1124
1125 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1126 return -EFAULT;
1127 loop_info64_from_old(&info, &info64);
1128 return loop_set_status(lo, &info64);
1129}
1130
1131static int
1132loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1133{
1134 struct loop_info64 info64;
1135
1136 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1137 return -EFAULT;
1138 return loop_set_status(lo, &info64);
1139}
1140
1141static int
1142loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1143 struct loop_info info;
1144 struct loop_info64 info64;
1145 int err = 0;
1146
1147 if (!arg)
1148 err = -EINVAL;
1149 if (!err)
1150 err = loop_get_status(lo, &info64);
1151 if (!err)
1152 err = loop_info64_to_old(&info64, &info);
1153 if (!err && copy_to_user(arg, &info, sizeof(info)))
1154 err = -EFAULT;
1155
1156 return err;
1157}
1158
1159static int
1160loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1161 struct loop_info64 info64;
1162 int err = 0;
1163
1164 if (!arg)
1165 err = -EINVAL;
1166 if (!err)
1167 err = loop_get_status(lo, &info64);
1168 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1169 err = -EFAULT;
1170
1171 return err;
1172}
1173
J. R. Okajima53d66602009-03-31 15:23:43 -07001174static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1175{
1176 int err;
1177 sector_t sec;
1178 loff_t sz;
1179
1180 err = -ENXIO;
1181 if (unlikely(lo->lo_state != Lo_bound))
1182 goto out;
1183 err = figure_loop_size(lo);
1184 if (unlikely(err))
1185 goto out;
1186 sec = get_capacity(lo->lo_disk);
1187 /* the width of sector_t may be narrow for bit-shift */
1188 sz = sec;
1189 sz <<= 9;
1190 mutex_lock(&bdev->bd_mutex);
1191 bd_set_size(bdev, sz);
1192 mutex_unlock(&bdev->bd_mutex);
1193
1194 out:
1195 return err;
1196}
1197
Al Virobb214882008-03-02 09:29:48 -05001198static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 unsigned int cmd, unsigned long arg)
1200{
Al Virobb214882008-03-02 09:29:48 -05001201 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 int err;
1203
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001204 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 switch (cmd) {
1206 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001207 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 break;
1209 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001210 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 break;
1212 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001213 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Al Virobb214882008-03-02 09:29:48 -05001214 err = loop_clr_fd(lo, bdev);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001215 if (!err)
1216 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 break;
1218 case LOOP_SET_STATUS:
1219 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1220 break;
1221 case LOOP_GET_STATUS:
1222 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1223 break;
1224 case LOOP_SET_STATUS64:
1225 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1226 break;
1227 case LOOP_GET_STATUS64:
1228 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1229 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001230 case LOOP_SET_CAPACITY:
1231 err = -EPERM;
1232 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1233 err = loop_set_capacity(lo, bdev);
1234 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 default:
1236 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1237 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001238 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001239
1240out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return err;
1242}
1243
David Howells863d5b82006-08-29 19:06:14 +01001244#ifdef CONFIG_COMPAT
1245struct compat_loop_info {
1246 compat_int_t lo_number; /* ioctl r/o */
1247 compat_dev_t lo_device; /* ioctl r/o */
1248 compat_ulong_t lo_inode; /* ioctl r/o */
1249 compat_dev_t lo_rdevice; /* ioctl r/o */
1250 compat_int_t lo_offset;
1251 compat_int_t lo_encrypt_type;
1252 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1253 compat_int_t lo_flags; /* ioctl r/o */
1254 char lo_name[LO_NAME_SIZE];
1255 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1256 compat_ulong_t lo_init[2];
1257 char reserved[4];
1258};
1259
1260/*
1261 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1262 * - noinlined to reduce stack space usage in main part of driver
1263 */
1264static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001265loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b82006-08-29 19:06:14 +01001266 struct loop_info64 *info64)
1267{
1268 struct compat_loop_info info;
1269
1270 if (copy_from_user(&info, arg, sizeof(info)))
1271 return -EFAULT;
1272
1273 memset(info64, 0, sizeof(*info64));
1274 info64->lo_number = info.lo_number;
1275 info64->lo_device = info.lo_device;
1276 info64->lo_inode = info.lo_inode;
1277 info64->lo_rdevice = info.lo_rdevice;
1278 info64->lo_offset = info.lo_offset;
1279 info64->lo_sizelimit = 0;
1280 info64->lo_encrypt_type = info.lo_encrypt_type;
1281 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1282 info64->lo_flags = info.lo_flags;
1283 info64->lo_init[0] = info.lo_init[0];
1284 info64->lo_init[1] = info.lo_init[1];
1285 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1286 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1287 else
1288 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1289 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1290 return 0;
1291}
1292
1293/*
1294 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1295 * - noinlined to reduce stack space usage in main part of driver
1296 */
1297static noinline int
1298loop_info64_to_compat(const struct loop_info64 *info64,
1299 struct compat_loop_info __user *arg)
1300{
1301 struct compat_loop_info info;
1302
1303 memset(&info, 0, sizeof(info));
1304 info.lo_number = info64->lo_number;
1305 info.lo_device = info64->lo_device;
1306 info.lo_inode = info64->lo_inode;
1307 info.lo_rdevice = info64->lo_rdevice;
1308 info.lo_offset = info64->lo_offset;
1309 info.lo_encrypt_type = info64->lo_encrypt_type;
1310 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1311 info.lo_flags = info64->lo_flags;
1312 info.lo_init[0] = info64->lo_init[0];
1313 info.lo_init[1] = info64->lo_init[1];
1314 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1315 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1316 else
1317 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1318 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1319
1320 /* error in case values were truncated */
1321 if (info.lo_device != info64->lo_device ||
1322 info.lo_rdevice != info64->lo_rdevice ||
1323 info.lo_inode != info64->lo_inode ||
1324 info.lo_offset != info64->lo_offset ||
1325 info.lo_init[0] != info64->lo_init[0] ||
1326 info.lo_init[1] != info64->lo_init[1])
1327 return -EOVERFLOW;
1328
1329 if (copy_to_user(arg, &info, sizeof(info)))
1330 return -EFAULT;
1331 return 0;
1332}
1333
1334static int
1335loop_set_status_compat(struct loop_device *lo,
1336 const struct compat_loop_info __user *arg)
1337{
1338 struct loop_info64 info64;
1339 int ret;
1340
1341 ret = loop_info64_from_compat(arg, &info64);
1342 if (ret < 0)
1343 return ret;
1344 return loop_set_status(lo, &info64);
1345}
1346
1347static int
1348loop_get_status_compat(struct loop_device *lo,
1349 struct compat_loop_info __user *arg)
1350{
1351 struct loop_info64 info64;
1352 int err = 0;
1353
1354 if (!arg)
1355 err = -EINVAL;
1356 if (!err)
1357 err = loop_get_status(lo, &info64);
1358 if (!err)
1359 err = loop_info64_to_compat(&info64, arg);
1360 return err;
1361}
1362
Al Virobb214882008-03-02 09:29:48 -05001363static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1364 unsigned int cmd, unsigned long arg)
David Howells863d5b82006-08-29 19:06:14 +01001365{
Al Virobb214882008-03-02 09:29:48 -05001366 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b82006-08-29 19:06:14 +01001367 int err;
1368
David Howells863d5b82006-08-29 19:06:14 +01001369 switch(cmd) {
1370 case LOOP_SET_STATUS:
1371 mutex_lock(&lo->lo_ctl_mutex);
1372 err = loop_set_status_compat(
1373 lo, (const struct compat_loop_info __user *) arg);
1374 mutex_unlock(&lo->lo_ctl_mutex);
1375 break;
1376 case LOOP_GET_STATUS:
1377 mutex_lock(&lo->lo_ctl_mutex);
1378 err = loop_get_status_compat(
1379 lo, (struct compat_loop_info __user *) arg);
1380 mutex_unlock(&lo->lo_ctl_mutex);
1381 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001382 case LOOP_SET_CAPACITY:
David Howells863d5b82006-08-29 19:06:14 +01001383 case LOOP_CLR_FD:
1384 case LOOP_GET_STATUS64:
1385 case LOOP_SET_STATUS64:
1386 arg = (unsigned long) compat_ptr(arg);
1387 case LOOP_SET_FD:
1388 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001389 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b82006-08-29 19:06:14 +01001390 break;
1391 default:
1392 err = -ENOIOCTLCMD;
1393 break;
1394 }
David Howells863d5b82006-08-29 19:06:14 +01001395 return err;
1396}
1397#endif
1398
Al Virobb214882008-03-02 09:29:48 -05001399static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Al Virobb214882008-03-02 09:29:48 -05001401 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Ingo Molnarf85221d2006-03-23 03:00:38 -08001403 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001405 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
1407 return 0;
1408}
1409
Al Virobb214882008-03-02 09:29:48 -05001410static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
Al Virobb214882008-03-02 09:29:48 -05001412 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001413 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Ingo Molnarf85221d2006-03-23 03:00:38 -08001415 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001416
Milan Broz14f27932008-12-12 14:48:27 +01001417 if (--lo->lo_refcnt)
1418 goto out;
1419
1420 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1421 /*
1422 * In autoclear mode, stop the loop thread
1423 * and remove configuration after last close.
1424 */
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001425 err = loop_clr_fd(lo, NULL);
1426 if (!err)
1427 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001428 } else {
1429 /*
1430 * Otherwise keep thread (if running) and config,
1431 * but flush possible ongoing bios in thread.
1432 */
1433 loop_flush(lo);
1434 }
David Woodhouse96c58652008-02-06 01:36:27 -08001435
Milan Broz14f27932008-12-12 14:48:27 +01001436out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001437 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001438out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return 0;
1440}
1441
1442static struct block_device_operations lo_fops = {
1443 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001444 .open = lo_open,
1445 .release = lo_release,
1446 .ioctl = lo_ioctl,
David Howells863d5b82006-08-29 19:06:14 +01001447#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001448 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b82006-08-29 19:06:14 +01001449#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450};
1451
1452/*
1453 * And now the modules code and kernel interface.
1454 */
Ken Chen73285082007-05-08 00:28:20 -07001455static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001457MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001458module_param(max_part, int, 0);
1459MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460MODULE_LICENSE("GPL");
1461MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1462
1463int loop_register_transfer(struct loop_func_table *funcs)
1464{
1465 unsigned int n = funcs->number;
1466
1467 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1468 return -EINVAL;
1469 xfer_funcs[n] = funcs;
1470 return 0;
1471}
1472
1473int loop_unregister_transfer(int number)
1474{
1475 unsigned int n = number;
1476 struct loop_device *lo;
1477 struct loop_func_table *xfer;
1478
1479 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1480 return -EINVAL;
1481
1482 xfer_funcs[n] = NULL;
1483
Ken Chen73285082007-05-08 00:28:20 -07001484 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001485 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
1487 if (lo->lo_encryption == xfer)
1488 loop_release_xfer(lo);
1489
Ingo Molnarf85221d2006-03-23 03:00:38 -08001490 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 }
1492
1493 return 0;
1494}
1495
1496EXPORT_SYMBOL(loop_register_transfer);
1497EXPORT_SYMBOL(loop_unregister_transfer);
1498
Ken Chena47653f2007-06-08 13:46:44 -07001499static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001500{
1501 struct loop_device *lo;
1502 struct gendisk *disk;
1503
1504 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1505 if (!lo)
1506 goto out;
1507
1508 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1509 if (!lo->lo_queue)
1510 goto out_free_dev;
1511
Laurent Vivier476a4812008-03-26 12:11:53 +01001512 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001513 if (!disk)
1514 goto out_free_queue;
1515
1516 mutex_init(&lo->lo_ctl_mutex);
1517 lo->lo_number = i;
1518 lo->lo_thread = NULL;
1519 init_waitqueue_head(&lo->lo_event);
1520 spin_lock_init(&lo->lo_lock);
1521 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001522 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001523 disk->fops = &lo_fops;
1524 disk->private_data = lo;
1525 disk->queue = lo->lo_queue;
1526 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001527 return lo;
1528
1529out_free_queue:
1530 blk_cleanup_queue(lo->lo_queue);
1531out_free_dev:
1532 kfree(lo);
1533out:
Al Viro07002e92007-05-12 16:23:15 -04001534 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001535}
1536
Ken Chena47653f2007-06-08 13:46:44 -07001537static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001538{
Ken Chen73285082007-05-08 00:28:20 -07001539 blk_cleanup_queue(lo->lo_queue);
1540 put_disk(lo->lo_disk);
1541 list_del(&lo->lo_list);
1542 kfree(lo);
1543}
1544
Ken Chena47653f2007-06-08 13:46:44 -07001545static struct loop_device *loop_init_one(int i)
1546{
1547 struct loop_device *lo;
1548
1549 list_for_each_entry(lo, &loop_devices, lo_list) {
1550 if (lo->lo_number == i)
1551 return lo;
1552 }
1553
1554 lo = loop_alloc(i);
1555 if (lo) {
1556 add_disk(lo->lo_disk);
1557 list_add_tail(&lo->lo_list, &loop_devices);
1558 }
1559 return lo;
1560}
1561
1562static void loop_del_one(struct loop_device *lo)
1563{
1564 del_gendisk(lo->lo_disk);
1565 loop_free(lo);
1566}
1567
Ken Chen73285082007-05-08 00:28:20 -07001568static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1569{
Al Viro705962c2007-05-13 05:52:32 -04001570 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001571 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001572
Al Viro705962c2007-05-13 05:52:32 -04001573 mutex_lock(&loop_devices_mutex);
1574 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001575 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001576 mutex_unlock(&loop_devices_mutex);
1577
1578 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001579 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001580}
1581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582static int __init loop_init(void)
1583{
Ken Chena47653f2007-06-08 13:46:44 -07001584 int i, nr;
1585 unsigned long range;
1586 struct loop_device *lo, *next;
1587
1588 /*
1589 * loop module now has a feature to instantiate underlying device
1590 * structure on-demand, provided that there is an access dev node.
1591 * However, this will not work well with user space tool that doesn't
1592 * know about such "feature". In order to not break any existing
1593 * tool, we do the following:
1594 *
1595 * (1) if max_loop is specified, create that many upfront, and this
1596 * also becomes a hard limit.
1597 * (2) if max_loop is not specified, create 8 loop device on module
1598 * load, user can further extend loop device by create dev node
1599 * themselves and have kernel automatically instantiate actual
1600 * device on-demand.
1601 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001602
1603 part_shift = 0;
1604 if (max_part > 0)
1605 part_shift = fls(max_part);
1606
1607 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001608 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Ken Chen73285082007-05-08 00:28:20 -07001610 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001611 nr = max_loop;
1612 range = max_loop;
1613 } else {
1614 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001615 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 }
Ken Chena47653f2007-06-08 13:46:44 -07001617
1618 if (register_blkdev(LOOP_MAJOR, "loop"))
1619 return -EIO;
1620
1621 for (i = 0; i < nr; i++) {
1622 lo = loop_alloc(i);
1623 if (!lo)
1624 goto Enomem;
1625 list_add_tail(&lo->lo_list, &loop_devices);
1626 }
1627
1628 /* point of no return */
1629
1630 list_for_each_entry(lo, &loop_devices, lo_list)
1631 add_disk(lo->lo_disk);
1632
1633 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1634 THIS_MODULE, loop_probe, NULL, NULL);
1635
Ken Chen73285082007-05-08 00:28:20 -07001636 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001638
1639Enomem:
1640 printk(KERN_INFO "loop: out of memory\n");
1641
1642 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1643 loop_free(lo);
1644
1645 unregister_blkdev(LOOP_MAJOR, "loop");
1646 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
1648
Ken Chen73285082007-05-08 00:28:20 -07001649static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650{
Ken Chena47653f2007-06-08 13:46:44 -07001651 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001652 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Laurent Vivier476a4812008-03-26 12:11:53 +01001654 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001655
Ken Chen73285082007-05-08 00:28:20 -07001656 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1657 loop_del_one(lo);
1658
Ken Chena47653f2007-06-08 13:46:44 -07001659 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001660 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661}
1662
1663module_init(loop_init);
1664module_exit(loop_exit);
1665
1666#ifndef MODULE
1667static int __init max_loop_setup(char *str)
1668{
1669 max_loop = simple_strtol(str, NULL, 0);
1670 return 1;
1671}
1672
1673__setup("max_loop=", max_loop_setup);
1674#endif