blob: ddae80825899ae0459bff4a8109d8fc3468868c3 [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 Howells863d5b822006-08-29 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 Viro98ae6ccd2006-10-10 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 Viro98ae6ccd2006-10-10 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 Viro98ae6ccd2006-10-10 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 Axboecac36bb02007-06-14 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{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (lo->lo_biotail) {
515 lo->lo_biotail->bi_next = bio;
516 lo->lo_biotail = bio;
517 } else
518 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
521/*
522 * Grab first pending buffer
523 */
524static struct bio *loop_get_bio(struct loop_device *lo)
525{
526 struct bio *bio;
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if ((bio = lo->lo_bio)) {
529 if (bio == lo->lo_biotail)
530 lo->lo_biotail = NULL;
531 lo->lo_bio = bio->bi_next;
532 bio->bi_next = NULL;
533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 return bio;
536}
537
Jens Axboe165125e2007-07-24 09:28:11 +0200538static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct loop_device *lo = q->queuedata;
541 int rw = bio_rw(old_bio);
542
Nick Piggin35a82d12005-06-23 00:09:06 -0700543 if (rw == READA)
544 rw = READ;
545
546 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 spin_lock_irq(&lo->lo_lock);
549 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700550 goto out;
551 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
552 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700554 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700555 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700559 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200560 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
564/*
565 * kick off io on the underlying address space
566 */
Jens Axboe165125e2007-07-24 09:28:11 +0200567static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 struct loop_device *lo = q->queuedata;
570
Nick Piggin75ad23b2008-04-29 14:48:33 +0200571 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 blk_run_address_space(lo->lo_backing_file->f_mapping);
573}
574
575struct switch_request {
576 struct file *file;
577 struct completion wait;
578};
579
580static void do_loop_switch(struct loop_device *, struct switch_request *);
581
582static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
583{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (unlikely(!bio->bi_bdev)) {
585 do_loop_switch(lo, bio->bi_private);
586 bio_put(bio);
587 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700588 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200589 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
591}
592
593/*
594 * worker thread that handles reads/writes to file backed loop devices,
595 * to avoid blocking in our make_request_fn. it also does loop decrypting
596 * on reads for block backed loop, as that is too heavy to do from
597 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700598 *
599 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
600 * calling kthread_stop(). Therefore once kthread_should_stop() is
601 * true, make_request will not place any more requests. Therefore
602 * once kthread_should_stop() is true and lo_bio is NULL, we are
603 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
605static int loop_thread(void *data)
606{
607 struct loop_device *lo = data;
608 struct bio *bio;
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 set_user_nice(current, -20);
611
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700612 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700614 wait_event_interruptible(lo->lo_event,
615 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700616
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700617 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700618 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700619 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700621 spin_unlock_irq(&lo->lo_lock);
622
623 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return 0;
628}
629
630/*
631 * loop_switch performs the hard work of switching a backing store.
632 * First it needs to flush existing IO, it does this by sending a magic
633 * BIO down the pipe. The completion of this BIO does the actual switch.
634 */
635static int loop_switch(struct loop_device *lo, struct file *file)
636{
637 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100638 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (!bio)
640 return -ENOMEM;
641 init_completion(&w.wait);
642 w.file = file;
643 bio->bi_private = &w;
644 bio->bi_bdev = NULL;
645 loop_make_request(lo->lo_queue, bio);
646 wait_for_completion(&w.wait);
647 return 0;
648}
649
650/*
Milan Broz14f27932008-12-12 14:48:27 +0100651 * Helper to flush the IOs in loop, but keeping loop thread running
652 */
653static int loop_flush(struct loop_device *lo)
654{
655 /* loop not yet configured, no running thread, nothing to flush */
656 if (!lo->lo_thread)
657 return 0;
658
659 return loop_switch(lo, NULL);
660}
661
662/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 * Do the actual switch; called from the BIO completion routine
664 */
665static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
666{
667 struct file *file = p->file;
668 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100669 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Milan Broz14f27932008-12-12 14:48:27 +0100671 /* if no new file, only flush of queued bios requested */
672 if (!file)
673 goto out;
674
675 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
677 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700678 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
679 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 lo->old_gfp_mask = mapping_gfp_mask(mapping);
681 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100682out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 complete(&p->wait);
684}
685
686
687/*
688 * loop_change_fd switched the backing store of a loopback device to
689 * a new file. This is useful for operating system installers to free up
690 * the original file and in High Availability environments to switch to
691 * an alternative location for the content in case of server meltdown.
692 * This can only work if the loop device is used read-only, and if the
693 * new backing store is the same size and type as the old backing store.
694 */
Al Virobb214882008-03-02 09:29:48 -0500695static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
696 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
698 struct file *file, *old_file;
699 struct inode *inode;
700 int error;
701
702 error = -ENXIO;
703 if (lo->lo_state != Lo_bound)
704 goto out;
705
706 /* the loop device has to be read-only */
707 error = -EINVAL;
708 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
709 goto out;
710
711 error = -EBADF;
712 file = fget(arg);
713 if (!file)
714 goto out;
715
716 inode = file->f_mapping->host;
717 old_file = lo->lo_backing_file;
718
719 error = -EINVAL;
720
721 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
722 goto out_putf;
723
Jens Axboefd582142007-06-12 21:20:37 +0200724 /* new backing store needs to support loop (eg splice_read) */
725 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 goto out_putf;
727
728 /* size of the new backing store needs to be the same */
729 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
730 goto out_putf;
731
732 /* and ... switch */
733 error = loop_switch(lo, file);
734 if (error)
735 goto out_putf;
736
737 fput(old_file);
Laurent Vivier476a4812008-03-26 12:11:53 +0100738 if (max_part > 0)
739 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 return 0;
741
742 out_putf:
743 fput(file);
744 out:
745 return error;
746}
747
748static inline int is_loop_device(struct file *file)
749{
750 struct inode *i = file->f_mapping->host;
751
752 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
753}
754
Al Virobb214882008-03-02 09:29:48 -0500755static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 struct block_device *bdev, unsigned int arg)
757{
758 struct file *file, *f;
759 struct inode *inode;
760 struct address_space *mapping;
761 unsigned lo_blocksize;
762 int lo_flags = 0;
763 int error;
764 loff_t size;
765
766 /* This is safe, since we have a reference from open(). */
767 __module_get(THIS_MODULE);
768
769 error = -EBADF;
770 file = fget(arg);
771 if (!file)
772 goto out;
773
774 error = -EBUSY;
775 if (lo->lo_state != Lo_unbound)
776 goto out_putf;
777
778 /* Avoid recursion */
779 f = file;
780 while (is_loop_device(f)) {
781 struct loop_device *l;
782
Al Virobb214882008-03-02 09:29:48 -0500783 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 goto out_putf;
785
786 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
787 if (l->lo_state == Lo_unbound) {
788 error = -EINVAL;
789 goto out_putf;
790 }
791 f = l->lo_backing_file;
792 }
793
794 mapping = file->f_mapping;
795 inode = mapping->host;
796
797 if (!(file->f_mode & FMODE_WRITE))
798 lo_flags |= LO_FLAGS_READ_ONLY;
799
800 error = -EINVAL;
801 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700802 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 /*
804 * If we can't read - sorry. If we only can't write - well,
805 * it's going to be read-only.
806 */
Jens Axboefd582142007-06-12 21:20:37 +0200807 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 goto out_putf;
Nick Piggin4e02ed42008-10-29 14:00:55 -0700809 if (aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 lo_flags |= LO_FLAGS_USE_AOPS;
811 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
812 lo_flags |= LO_FLAGS_READ_ONLY;
813
Theodore Ts'oba52de12006-09-27 01:50:49 -0700814 lo_blocksize = S_ISBLK(inode->i_mode) ?
815 inode->i_bdev->bd_block_size : PAGE_SIZE;
816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 error = 0;
818 } else {
819 goto out_putf;
820 }
821
822 size = get_loop_size(lo, file);
823
824 if ((loff_t)(sector_t)size != size) {
825 error = -EFBIG;
826 goto out_putf;
827 }
828
Al Virobb214882008-03-02 09:29:48 -0500829 if (!(mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 lo_flags |= LO_FLAGS_READ_ONLY;
831
832 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
833
834 lo->lo_blocksize = lo_blocksize;
835 lo->lo_device = bdev;
836 lo->lo_flags = lo_flags;
837 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700838 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 lo->ioctl = NULL;
840 lo->lo_sizelimit = 0;
841 lo->old_gfp_mask = mapping_gfp_mask(mapping);
842 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
843
844 lo->lo_bio = lo->lo_biotail = NULL;
845
846 /*
847 * set queue make_request_fn, and add limits based on lower level
848 * device
849 */
850 blk_queue_make_request(lo->lo_queue, loop_make_request);
851 lo->lo_queue->queuedata = lo;
852 lo->lo_queue->unplug_fn = loop_unplug;
853
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100854 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
855 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
856
Ken Chen73285082007-05-08 00:28:20 -0700857 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 bd_set_size(bdev, size << 9);
859
860 set_blocksize(bdev, lo_blocksize);
861
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700862 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
863 lo->lo_number);
864 if (IS_ERR(lo->lo_thread)) {
865 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700866 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700867 }
868 lo->lo_state = Lo_bound;
869 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100870 if (max_part > 0)
871 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return 0;
873
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700874out_clr:
875 lo->lo_thread = NULL;
876 lo->lo_device = NULL;
877 lo->lo_backing_file = NULL;
878 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700879 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700880 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700881 bd_set_size(bdev, 0);
882 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
883 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 out_putf:
885 fput(file);
886 out:
887 /* This is safe: open() is still holding a reference. */
888 module_put(THIS_MODULE);
889 return error;
890}
891
892static int
893loop_release_xfer(struct loop_device *lo)
894{
895 int err = 0;
896 struct loop_func_table *xfer = lo->lo_encryption;
897
898 if (xfer) {
899 if (xfer->release)
900 err = xfer->release(lo);
901 lo->transfer = NULL;
902 lo->lo_encryption = NULL;
903 module_put(xfer->owner);
904 }
905 return err;
906}
907
908static int
909loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
910 const struct loop_info64 *i)
911{
912 int err = 0;
913
914 if (xfer) {
915 struct module *owner = xfer->owner;
916
917 if (!try_module_get(owner))
918 return -EINVAL;
919 if (xfer->init)
920 err = xfer->init(lo, i);
921 if (err)
922 module_put(owner);
923 else
924 lo->lo_encryption = xfer;
925 }
926 return err;
927}
928
929static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
930{
931 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400932 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 if (lo->lo_state != Lo_bound)
935 return -ENXIO;
936
937 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
938 return -EBUSY;
939
940 if (filp == NULL)
941 return -EINVAL;
942
943 spin_lock_irq(&lo->lo_lock);
944 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 spin_unlock_irq(&lo->lo_lock);
946
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700947 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Milan Broz8ae30b82008-12-12 14:50:49 +0100949 lo->lo_queue->unplug_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 lo->lo_backing_file = NULL;
951
952 loop_release_xfer(lo);
953 lo->transfer = NULL;
954 lo->ioctl = NULL;
955 lo->lo_device = NULL;
956 lo->lo_encryption = NULL;
957 lo->lo_offset = 0;
958 lo->lo_sizelimit = 0;
959 lo->lo_encrypt_key_size = 0;
960 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700961 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
963 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
964 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -0500965 if (bdev)
966 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700967 set_capacity(lo->lo_disk, 0);
Al Virobb214882008-03-02 09:29:48 -0500968 if (bdev)
969 bd_set_size(bdev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 mapping_set_gfp_mask(filp->f_mapping, gfp);
971 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 /* This is safe: open() is still holding a reference. */
973 module_put(THIS_MODULE);
Laurent Vivier476a4812008-03-26 12:11:53 +0100974 if (max_part > 0)
975 ioctl_by_bdev(bdev, BLKRRPART, 0);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +0100976 mutex_unlock(&lo->lo_ctl_mutex);
977 /*
978 * Need not hold lo_ctl_mutex to fput backing file.
979 * Calling fput holding lo_ctl_mutex triggers a circular
980 * lock dependency possibility warning as fput can take
981 * bd_mutex which is usually taken before lo_ctl_mutex.
982 */
983 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 return 0;
985}
986
987static int
988loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
989{
990 int err;
991 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +1100992 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
David Howellsb0fafa82008-11-14 10:38:41 +1100994 if (lo->lo_encrypt_key_size &&
995 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 !capable(CAP_SYS_ADMIN))
997 return -EPERM;
998 if (lo->lo_state != Lo_bound)
999 return -ENXIO;
1000 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1001 return -EINVAL;
1002
1003 err = loop_release_xfer(lo);
1004 if (err)
1005 return err;
1006
1007 if (info->lo_encrypt_type) {
1008 unsigned int type = info->lo_encrypt_type;
1009
1010 if (type >= MAX_LO_CRYPT)
1011 return -EINVAL;
1012 xfer = xfer_funcs[type];
1013 if (xfer == NULL)
1014 return -EINVAL;
1015 } else
1016 xfer = NULL;
1017
1018 err = loop_init_xfer(lo, xfer, info);
1019 if (err)
1020 return err;
1021
1022 if (lo->lo_offset != info->lo_offset ||
1023 lo->lo_sizelimit != info->lo_sizelimit) {
1024 lo->lo_offset = info->lo_offset;
1025 lo->lo_sizelimit = info->lo_sizelimit;
1026 if (figure_loop_size(lo))
1027 return -EFBIG;
1028 }
1029
1030 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1031 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1032 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1033 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1034
1035 if (!xfer)
1036 xfer = &none_funcs;
1037 lo->transfer = xfer->transfer;
1038 lo->ioctl = xfer->ioctl;
1039
David Woodhouse96c58652008-02-06 01:36:27 -08001040 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1041 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1042 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1045 lo->lo_init[0] = info->lo_init[0];
1046 lo->lo_init[1] = info->lo_init[1];
1047 if (info->lo_encrypt_key_size) {
1048 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1049 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001050 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
1052
1053 return 0;
1054}
1055
1056static int
1057loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1058{
1059 struct file *file = lo->lo_backing_file;
1060 struct kstat stat;
1061 int error;
1062
1063 if (lo->lo_state != Lo_bound)
1064 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001065 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (error)
1067 return error;
1068 memset(info, 0, sizeof(*info));
1069 info->lo_number = lo->lo_number;
1070 info->lo_device = huge_encode_dev(stat.dev);
1071 info->lo_inode = stat.ino;
1072 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1073 info->lo_offset = lo->lo_offset;
1074 info->lo_sizelimit = lo->lo_sizelimit;
1075 info->lo_flags = lo->lo_flags;
1076 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1077 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1078 info->lo_encrypt_type =
1079 lo->lo_encryption ? lo->lo_encryption->number : 0;
1080 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1081 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1082 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1083 lo->lo_encrypt_key_size);
1084 }
1085 return 0;
1086}
1087
1088static void
1089loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1090{
1091 memset(info64, 0, sizeof(*info64));
1092 info64->lo_number = info->lo_number;
1093 info64->lo_device = info->lo_device;
1094 info64->lo_inode = info->lo_inode;
1095 info64->lo_rdevice = info->lo_rdevice;
1096 info64->lo_offset = info->lo_offset;
1097 info64->lo_sizelimit = 0;
1098 info64->lo_encrypt_type = info->lo_encrypt_type;
1099 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1100 info64->lo_flags = info->lo_flags;
1101 info64->lo_init[0] = info->lo_init[0];
1102 info64->lo_init[1] = info->lo_init[1];
1103 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1104 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1105 else
1106 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1107 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1108}
1109
1110static int
1111loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1112{
1113 memset(info, 0, sizeof(*info));
1114 info->lo_number = info64->lo_number;
1115 info->lo_device = info64->lo_device;
1116 info->lo_inode = info64->lo_inode;
1117 info->lo_rdevice = info64->lo_rdevice;
1118 info->lo_offset = info64->lo_offset;
1119 info->lo_encrypt_type = info64->lo_encrypt_type;
1120 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1121 info->lo_flags = info64->lo_flags;
1122 info->lo_init[0] = info64->lo_init[0];
1123 info->lo_init[1] = info64->lo_init[1];
1124 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1125 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1126 else
1127 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1128 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1129
1130 /* error in case values were truncated */
1131 if (info->lo_device != info64->lo_device ||
1132 info->lo_rdevice != info64->lo_rdevice ||
1133 info->lo_inode != info64->lo_inode ||
1134 info->lo_offset != info64->lo_offset)
1135 return -EOVERFLOW;
1136
1137 return 0;
1138}
1139
1140static int
1141loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1142{
1143 struct loop_info info;
1144 struct loop_info64 info64;
1145
1146 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1147 return -EFAULT;
1148 loop_info64_from_old(&info, &info64);
1149 return loop_set_status(lo, &info64);
1150}
1151
1152static int
1153loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1154{
1155 struct loop_info64 info64;
1156
1157 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1158 return -EFAULT;
1159 return loop_set_status(lo, &info64);
1160}
1161
1162static int
1163loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1164 struct loop_info info;
1165 struct loop_info64 info64;
1166 int err = 0;
1167
1168 if (!arg)
1169 err = -EINVAL;
1170 if (!err)
1171 err = loop_get_status(lo, &info64);
1172 if (!err)
1173 err = loop_info64_to_old(&info64, &info);
1174 if (!err && copy_to_user(arg, &info, sizeof(info)))
1175 err = -EFAULT;
1176
1177 return err;
1178}
1179
1180static int
1181loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1182 struct loop_info64 info64;
1183 int err = 0;
1184
1185 if (!arg)
1186 err = -EINVAL;
1187 if (!err)
1188 err = loop_get_status(lo, &info64);
1189 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1190 err = -EFAULT;
1191
1192 return err;
1193}
1194
J. R. Okajima53d66602009-03-31 15:23:43 -07001195static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1196{
1197 int err;
1198 sector_t sec;
1199 loff_t sz;
1200
1201 err = -ENXIO;
1202 if (unlikely(lo->lo_state != Lo_bound))
1203 goto out;
1204 err = figure_loop_size(lo);
1205 if (unlikely(err))
1206 goto out;
1207 sec = get_capacity(lo->lo_disk);
1208 /* the width of sector_t may be narrow for bit-shift */
1209 sz = sec;
1210 sz <<= 9;
1211 mutex_lock(&bdev->bd_mutex);
1212 bd_set_size(bdev, sz);
1213 mutex_unlock(&bdev->bd_mutex);
1214
1215 out:
1216 return err;
1217}
1218
Al Virobb214882008-03-02 09:29:48 -05001219static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 unsigned int cmd, unsigned long arg)
1221{
Al Virobb214882008-03-02 09:29:48 -05001222 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 int err;
1224
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001225 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 switch (cmd) {
1227 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001228 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 break;
1230 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001231 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 break;
1233 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001234 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Al Virobb214882008-03-02 09:29:48 -05001235 err = loop_clr_fd(lo, bdev);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001236 if (!err)
1237 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 break;
1239 case LOOP_SET_STATUS:
1240 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1241 break;
1242 case LOOP_GET_STATUS:
1243 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1244 break;
1245 case LOOP_SET_STATUS64:
1246 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1247 break;
1248 case LOOP_GET_STATUS64:
1249 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1250 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001251 case LOOP_SET_CAPACITY:
1252 err = -EPERM;
1253 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1254 err = loop_set_capacity(lo, bdev);
1255 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 default:
1257 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1258 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001259 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001260
1261out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 return err;
1263}
1264
David Howells863d5b822006-08-29 19:06:14 +01001265#ifdef CONFIG_COMPAT
1266struct compat_loop_info {
1267 compat_int_t lo_number; /* ioctl r/o */
1268 compat_dev_t lo_device; /* ioctl r/o */
1269 compat_ulong_t lo_inode; /* ioctl r/o */
1270 compat_dev_t lo_rdevice; /* ioctl r/o */
1271 compat_int_t lo_offset;
1272 compat_int_t lo_encrypt_type;
1273 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1274 compat_int_t lo_flags; /* ioctl r/o */
1275 char lo_name[LO_NAME_SIZE];
1276 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1277 compat_ulong_t lo_init[2];
1278 char reserved[4];
1279};
1280
1281/*
1282 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1283 * - noinlined to reduce stack space usage in main part of driver
1284 */
1285static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001286loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001287 struct loop_info64 *info64)
1288{
1289 struct compat_loop_info info;
1290
1291 if (copy_from_user(&info, arg, sizeof(info)))
1292 return -EFAULT;
1293
1294 memset(info64, 0, sizeof(*info64));
1295 info64->lo_number = info.lo_number;
1296 info64->lo_device = info.lo_device;
1297 info64->lo_inode = info.lo_inode;
1298 info64->lo_rdevice = info.lo_rdevice;
1299 info64->lo_offset = info.lo_offset;
1300 info64->lo_sizelimit = 0;
1301 info64->lo_encrypt_type = info.lo_encrypt_type;
1302 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1303 info64->lo_flags = info.lo_flags;
1304 info64->lo_init[0] = info.lo_init[0];
1305 info64->lo_init[1] = info.lo_init[1];
1306 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1307 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1308 else
1309 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1310 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1311 return 0;
1312}
1313
1314/*
1315 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1316 * - noinlined to reduce stack space usage in main part of driver
1317 */
1318static noinline int
1319loop_info64_to_compat(const struct loop_info64 *info64,
1320 struct compat_loop_info __user *arg)
1321{
1322 struct compat_loop_info info;
1323
1324 memset(&info, 0, sizeof(info));
1325 info.lo_number = info64->lo_number;
1326 info.lo_device = info64->lo_device;
1327 info.lo_inode = info64->lo_inode;
1328 info.lo_rdevice = info64->lo_rdevice;
1329 info.lo_offset = info64->lo_offset;
1330 info.lo_encrypt_type = info64->lo_encrypt_type;
1331 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1332 info.lo_flags = info64->lo_flags;
1333 info.lo_init[0] = info64->lo_init[0];
1334 info.lo_init[1] = info64->lo_init[1];
1335 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1336 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1337 else
1338 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1339 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1340
1341 /* error in case values were truncated */
1342 if (info.lo_device != info64->lo_device ||
1343 info.lo_rdevice != info64->lo_rdevice ||
1344 info.lo_inode != info64->lo_inode ||
1345 info.lo_offset != info64->lo_offset ||
1346 info.lo_init[0] != info64->lo_init[0] ||
1347 info.lo_init[1] != info64->lo_init[1])
1348 return -EOVERFLOW;
1349
1350 if (copy_to_user(arg, &info, sizeof(info)))
1351 return -EFAULT;
1352 return 0;
1353}
1354
1355static int
1356loop_set_status_compat(struct loop_device *lo,
1357 const struct compat_loop_info __user *arg)
1358{
1359 struct loop_info64 info64;
1360 int ret;
1361
1362 ret = loop_info64_from_compat(arg, &info64);
1363 if (ret < 0)
1364 return ret;
1365 return loop_set_status(lo, &info64);
1366}
1367
1368static int
1369loop_get_status_compat(struct loop_device *lo,
1370 struct compat_loop_info __user *arg)
1371{
1372 struct loop_info64 info64;
1373 int err = 0;
1374
1375 if (!arg)
1376 err = -EINVAL;
1377 if (!err)
1378 err = loop_get_status(lo, &info64);
1379 if (!err)
1380 err = loop_info64_to_compat(&info64, arg);
1381 return err;
1382}
1383
Al Virobb214882008-03-02 09:29:48 -05001384static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1385 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001386{
Al Virobb214882008-03-02 09:29:48 -05001387 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001388 int err;
1389
David Howells863d5b822006-08-29 19:06:14 +01001390 switch(cmd) {
1391 case LOOP_SET_STATUS:
1392 mutex_lock(&lo->lo_ctl_mutex);
1393 err = loop_set_status_compat(
1394 lo, (const struct compat_loop_info __user *) arg);
1395 mutex_unlock(&lo->lo_ctl_mutex);
1396 break;
1397 case LOOP_GET_STATUS:
1398 mutex_lock(&lo->lo_ctl_mutex);
1399 err = loop_get_status_compat(
1400 lo, (struct compat_loop_info __user *) arg);
1401 mutex_unlock(&lo->lo_ctl_mutex);
1402 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001403 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001404 case LOOP_CLR_FD:
1405 case LOOP_GET_STATUS64:
1406 case LOOP_SET_STATUS64:
1407 arg = (unsigned long) compat_ptr(arg);
1408 case LOOP_SET_FD:
1409 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001410 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001411 break;
1412 default:
1413 err = -ENOIOCTLCMD;
1414 break;
1415 }
David Howells863d5b822006-08-29 19:06:14 +01001416 return err;
1417}
1418#endif
1419
Al Virobb214882008-03-02 09:29:48 -05001420static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421{
Al Virobb214882008-03-02 09:29:48 -05001422 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Ingo Molnarf85221d2006-03-23 03:00:38 -08001424 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001426 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 return 0;
1429}
1430
Al Virobb214882008-03-02 09:29:48 -05001431static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
Al Virobb214882008-03-02 09:29:48 -05001433 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001434 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Ingo Molnarf85221d2006-03-23 03:00:38 -08001436 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001437
Milan Broz14f27932008-12-12 14:48:27 +01001438 if (--lo->lo_refcnt)
1439 goto out;
1440
1441 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1442 /*
1443 * In autoclear mode, stop the loop thread
1444 * and remove configuration after last close.
1445 */
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001446 err = loop_clr_fd(lo, NULL);
1447 if (!err)
1448 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001449 } else {
1450 /*
1451 * Otherwise keep thread (if running) and config,
1452 * but flush possible ongoing bios in thread.
1453 */
1454 loop_flush(lo);
1455 }
David Woodhouse96c58652008-02-06 01:36:27 -08001456
Milan Broz14f27932008-12-12 14:48:27 +01001457out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001458 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001459out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 return 0;
1461}
1462
1463static struct block_device_operations lo_fops = {
1464 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001465 .open = lo_open,
1466 .release = lo_release,
1467 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001468#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001469 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001470#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471};
1472
1473/*
1474 * And now the modules code and kernel interface.
1475 */
Ken Chen73285082007-05-08 00:28:20 -07001476static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001478MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001479module_param(max_part, int, 0);
1480MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481MODULE_LICENSE("GPL");
1482MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1483
1484int loop_register_transfer(struct loop_func_table *funcs)
1485{
1486 unsigned int n = funcs->number;
1487
1488 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1489 return -EINVAL;
1490 xfer_funcs[n] = funcs;
1491 return 0;
1492}
1493
1494int loop_unregister_transfer(int number)
1495{
1496 unsigned int n = number;
1497 struct loop_device *lo;
1498 struct loop_func_table *xfer;
1499
1500 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1501 return -EINVAL;
1502
1503 xfer_funcs[n] = NULL;
1504
Ken Chen73285082007-05-08 00:28:20 -07001505 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001506 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
1508 if (lo->lo_encryption == xfer)
1509 loop_release_xfer(lo);
1510
Ingo Molnarf85221d2006-03-23 03:00:38 -08001511 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513
1514 return 0;
1515}
1516
1517EXPORT_SYMBOL(loop_register_transfer);
1518EXPORT_SYMBOL(loop_unregister_transfer);
1519
Ken Chena47653f2007-06-08 13:46:44 -07001520static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001521{
1522 struct loop_device *lo;
1523 struct gendisk *disk;
1524
1525 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1526 if (!lo)
1527 goto out;
1528
1529 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1530 if (!lo->lo_queue)
1531 goto out_free_dev;
1532
Laurent Vivier476a4812008-03-26 12:11:53 +01001533 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001534 if (!disk)
1535 goto out_free_queue;
1536
1537 mutex_init(&lo->lo_ctl_mutex);
1538 lo->lo_number = i;
1539 lo->lo_thread = NULL;
1540 init_waitqueue_head(&lo->lo_event);
1541 spin_lock_init(&lo->lo_lock);
1542 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001543 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001544 disk->fops = &lo_fops;
1545 disk->private_data = lo;
1546 disk->queue = lo->lo_queue;
1547 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001548 return lo;
1549
1550out_free_queue:
1551 blk_cleanup_queue(lo->lo_queue);
1552out_free_dev:
1553 kfree(lo);
1554out:
Al Viro07002e92007-05-12 16:23:15 -04001555 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001556}
1557
Ken Chena47653f2007-06-08 13:46:44 -07001558static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001559{
Ken Chen73285082007-05-08 00:28:20 -07001560 blk_cleanup_queue(lo->lo_queue);
1561 put_disk(lo->lo_disk);
1562 list_del(&lo->lo_list);
1563 kfree(lo);
1564}
1565
Ken Chena47653f2007-06-08 13:46:44 -07001566static struct loop_device *loop_init_one(int i)
1567{
1568 struct loop_device *lo;
1569
1570 list_for_each_entry(lo, &loop_devices, lo_list) {
1571 if (lo->lo_number == i)
1572 return lo;
1573 }
1574
1575 lo = loop_alloc(i);
1576 if (lo) {
1577 add_disk(lo->lo_disk);
1578 list_add_tail(&lo->lo_list, &loop_devices);
1579 }
1580 return lo;
1581}
1582
1583static void loop_del_one(struct loop_device *lo)
1584{
1585 del_gendisk(lo->lo_disk);
1586 loop_free(lo);
1587}
1588
Ken Chen73285082007-05-08 00:28:20 -07001589static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1590{
Al Viro705962c2007-05-13 05:52:32 -04001591 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001592 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001593
Al Viro705962c2007-05-13 05:52:32 -04001594 mutex_lock(&loop_devices_mutex);
1595 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001596 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001597 mutex_unlock(&loop_devices_mutex);
1598
1599 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001600 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001601}
1602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603static int __init loop_init(void)
1604{
Ken Chena47653f2007-06-08 13:46:44 -07001605 int i, nr;
1606 unsigned long range;
1607 struct loop_device *lo, *next;
1608
1609 /*
1610 * loop module now has a feature to instantiate underlying device
1611 * structure on-demand, provided that there is an access dev node.
1612 * However, this will not work well with user space tool that doesn't
1613 * know about such "feature". In order to not break any existing
1614 * tool, we do the following:
1615 *
1616 * (1) if max_loop is specified, create that many upfront, and this
1617 * also becomes a hard limit.
1618 * (2) if max_loop is not specified, create 8 loop device on module
1619 * load, user can further extend loop device by create dev node
1620 * themselves and have kernel automatically instantiate actual
1621 * device on-demand.
1622 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001623
1624 part_shift = 0;
1625 if (max_part > 0)
1626 part_shift = fls(max_part);
1627
1628 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001629 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Ken Chen73285082007-05-08 00:28:20 -07001631 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001632 nr = max_loop;
1633 range = max_loop;
1634 } else {
1635 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001636 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
Ken Chena47653f2007-06-08 13:46:44 -07001638
1639 if (register_blkdev(LOOP_MAJOR, "loop"))
1640 return -EIO;
1641
1642 for (i = 0; i < nr; i++) {
1643 lo = loop_alloc(i);
1644 if (!lo)
1645 goto Enomem;
1646 list_add_tail(&lo->lo_list, &loop_devices);
1647 }
1648
1649 /* point of no return */
1650
1651 list_for_each_entry(lo, &loop_devices, lo_list)
1652 add_disk(lo->lo_disk);
1653
1654 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1655 THIS_MODULE, loop_probe, NULL, NULL);
1656
Ken Chen73285082007-05-08 00:28:20 -07001657 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001659
1660Enomem:
1661 printk(KERN_INFO "loop: out of memory\n");
1662
1663 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1664 loop_free(lo);
1665
1666 unregister_blkdev(LOOP_MAJOR, "loop");
1667 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668}
1669
Ken Chen73285082007-05-08 00:28:20 -07001670static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Ken Chena47653f2007-06-08 13:46:44 -07001672 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001673 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Laurent Vivier476a4812008-03-26 12:11:53 +01001675 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001676
Ken Chen73285082007-05-08 00:28:20 -07001677 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1678 loop_del_one(lo);
1679
Ken Chena47653f2007-06-08 13:46:44 -07001680 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001681 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682}
1683
1684module_init(loop_init);
1685module_exit(loop_exit);
1686
1687#ifndef MODULE
1688static int __init max_loop_setup(char *str)
1689{
1690 max_loop = simple_strtol(str, NULL, 0);
1691 return 1;
1692}
1693
1694__setup("max_loop=", max_loop_setup);
1695#endif