blob: 6120922f459f3708685cffa46a51e5517c89b18c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/drivers/block/loop.c
3 *
4 * Written by Theodore Ts'o, 3/29/93
5 *
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
8 *
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 *
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 *
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
Jan Engelhardt96de0e22007-10-19 23:21:04 +020032 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <axboe@suse.de>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
Nick Piggin4e02ed42008-10-29 14:00:55 -070043 * operations write_begin is not available on the backing filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/file.h>
57#include <linux/stat.h>
58#include <linux/errno.h>
59#include <linux/major.h>
60#include <linux/wait.h>
61#include <linux/blkdev.h>
62#include <linux/blkpg.h>
63#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/swap.h>
65#include <linux/slab.h>
66#include <linux/loop.h>
David Howells863d5b822006-08-29 19:06:14 +010067#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070069#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/writeback.h>
71#include <linux/buffer_head.h> /* for invalidate_bdev() */
72#include <linux/completion.h>
73#include <linux/highmem.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070074#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020075#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77#include <asm/uaccess.h>
78
Ken Chen73285082007-05-08 00:28:20 -070079static LIST_HEAD(loop_devices);
80static DEFINE_MUTEX(loop_devices_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Laurent Vivier476a4812008-03-26 12:11:53 +010082static int max_part;
83static int part_shift;
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/*
86 * Transfer functions
87 */
88static int transfer_none(struct loop_device *lo, int cmd,
89 struct page *raw_page, unsigned raw_off,
90 struct page *loop_page, unsigned loop_off,
91 int size, sector_t real_block)
92{
93 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
94 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
95
96 if (cmd == READ)
97 memcpy(loop_buf, raw_buf, size);
98 else
99 memcpy(raw_buf, loop_buf, size);
100
101 kunmap_atomic(raw_buf, KM_USER0);
102 kunmap_atomic(loop_buf, KM_USER1);
103 cond_resched();
104 return 0;
105}
106
107static int transfer_xor(struct loop_device *lo, int cmd,
108 struct page *raw_page, unsigned raw_off,
109 struct page *loop_page, unsigned loop_off,
110 int size, sector_t real_block)
111{
112 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
113 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
114 char *in, *out, *key;
115 int i, keysize;
116
117 if (cmd == READ) {
118 in = raw_buf;
119 out = loop_buf;
120 } else {
121 in = loop_buf;
122 out = raw_buf;
123 }
124
125 key = lo->lo_encrypt_key;
126 keysize = lo->lo_encrypt_key_size;
127 for (i = 0; i < size; i++)
128 *out++ = *in++ ^ key[(i & 511) % keysize];
129
130 kunmap_atomic(raw_buf, KM_USER0);
131 kunmap_atomic(loop_buf, KM_USER1);
132 cond_resched();
133 return 0;
134}
135
136static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
137{
138 if (unlikely(info->lo_encrypt_key_size <= 0))
139 return -EINVAL;
140 return 0;
141}
142
143static struct loop_func_table none_funcs = {
144 .number = LO_CRYPT_NONE,
145 .transfer = transfer_none,
146};
147
148static struct loop_func_table xor_funcs = {
149 .number = LO_CRYPT_XOR,
150 .transfer = transfer_xor,
151 .init = xor_init
152};
153
154/* xfer_funcs[0] is special - its release function is never called */
155static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
156 &none_funcs,
157 &xor_funcs
158};
159
160static loff_t get_loop_size(struct loop_device *lo, struct file *file)
161{
162 loff_t size, offset, loopsize;
163
164 /* Compute loopsize in bytes */
165 size = i_size_read(file->f_mapping->host);
166 offset = lo->lo_offset;
167 loopsize = size - offset;
168 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
169 loopsize = lo->lo_sizelimit;
170
171 /*
172 * Unfortunately, if we want to do I/O on the device,
173 * the number of 512-byte sectors has to fit into a sector_t.
174 */
175 return loopsize >> 9;
176}
177
178static int
179figure_loop_size(struct loop_device *lo)
180{
181 loff_t size = get_loop_size(lo, lo->lo_backing_file);
182 sector_t x = (sector_t)size;
183
184 if (unlikely((loff_t)x != size))
185 return -EFBIG;
186
Ken Chen73285082007-05-08 00:28:20 -0700187 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 return 0;
189}
190
191static inline int
192lo_do_transfer(struct loop_device *lo, int cmd,
193 struct page *rpage, unsigned roffs,
194 struct page *lpage, unsigned loffs,
195 int size, sector_t rblock)
196{
197 if (unlikely(!lo->transfer))
198 return 0;
199
200 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
201}
202
203/**
204 * do_lo_send_aops - helper for writing data to a loop device
205 *
206 * This is the fast version for backing filesystems which implement the address
Nick Pigginafddba42007-10-16 01:25:01 -0700207 * space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
209static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400210 loff_t pos, struct page *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
212 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
213 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 pgoff_t index;
215 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800216 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800218 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 index = pos >> PAGE_CACHE_SHIFT;
220 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
221 bv_offs = bvec->bv_offset;
222 len = bvec->bv_len;
223 while (len > 0) {
224 sector_t IV;
Nick Pigginafddba42007-10-16 01:25:01 -0700225 unsigned size, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 int transfer_result;
Nick Pigginafddba42007-10-16 01:25:01 -0700227 struct page *page;
228 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
231 size = PAGE_CACHE_SIZE - offset;
232 if (size > len)
233 size = len;
Nick Pigginafddba42007-10-16 01:25:01 -0700234
235 ret = pagecache_write_begin(file, mapping, pos, size, 0,
236 &page, &fsdata);
237 if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700239
Nikanth Karthikesan02246c42010-04-08 21:39:31 +0200240 file_update_time(file);
241
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) {
Jens Axboe1f98a132009-09-11 14:32:04 +0200479 bool barrier = bio_rw_flagged(bio, BIO_RW_BARRIER);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100480 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
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100488 ret = vfs_fsync(file, 0);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100489 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) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100498 ret = vfs_fsync(file, 0);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100499 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);
David Zeuthenc3473c62010-05-03 14:08:59 +0200838 /* let user-space know about the new size */
839 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 set_blocksize(bdev, lo_blocksize);
842
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700843 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
844 lo->lo_number);
845 if (IS_ERR(lo->lo_thread)) {
846 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700847 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700848 }
849 lo->lo_state = Lo_bound;
850 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100851 if (max_part > 0)
852 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 return 0;
854
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700855out_clr:
856 lo->lo_thread = NULL;
857 lo->lo_device = NULL;
858 lo->lo_backing_file = NULL;
859 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700860 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700861 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700862 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200863 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700864 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
865 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 out_putf:
867 fput(file);
868 out:
869 /* This is safe: open() is still holding a reference. */
870 module_put(THIS_MODULE);
871 return error;
872}
873
874static int
875loop_release_xfer(struct loop_device *lo)
876{
877 int err = 0;
878 struct loop_func_table *xfer = lo->lo_encryption;
879
880 if (xfer) {
881 if (xfer->release)
882 err = xfer->release(lo);
883 lo->transfer = NULL;
884 lo->lo_encryption = NULL;
885 module_put(xfer->owner);
886 }
887 return err;
888}
889
890static int
891loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
892 const struct loop_info64 *i)
893{
894 int err = 0;
895
896 if (xfer) {
897 struct module *owner = xfer->owner;
898
899 if (!try_module_get(owner))
900 return -EINVAL;
901 if (xfer->init)
902 err = xfer->init(lo, i);
903 if (err)
904 module_put(owner);
905 else
906 lo->lo_encryption = xfer;
907 }
908 return err;
909}
910
911static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
912{
913 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400914 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 if (lo->lo_state != Lo_bound)
917 return -ENXIO;
918
919 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
920 return -EBUSY;
921
922 if (filp == NULL)
923 return -EINVAL;
924
925 spin_lock_irq(&lo->lo_lock);
926 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 spin_unlock_irq(&lo->lo_lock);
928
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700929 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Milan Broz8ae30b82008-12-12 14:50:49 +0100931 lo->lo_queue->unplug_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 lo->lo_backing_file = NULL;
933
934 loop_release_xfer(lo);
935 lo->transfer = NULL;
936 lo->ioctl = NULL;
937 lo->lo_device = NULL;
938 lo->lo_encryption = NULL;
939 lo->lo_offset = 0;
940 lo->lo_sizelimit = 0;
941 lo->lo_encrypt_key_size = 0;
942 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700943 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
945 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
946 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -0500947 if (bdev)
948 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700949 set_capacity(lo->lo_disk, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200950 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -0500951 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200952 /* let user-space know about this change */
953 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 mapping_set_gfp_mask(filp->f_mapping, gfp);
956 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 /* This is safe: open() is still holding a reference. */
958 module_put(THIS_MODULE);
Alexey Dobriyancf6e6932009-10-26 16:49:55 -0700959 if (max_part > 0 && bdev)
Laurent Vivier476a4812008-03-26 12:11:53 +0100960 ioctl_by_bdev(bdev, BLKRRPART, 0);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +0100961 mutex_unlock(&lo->lo_ctl_mutex);
962 /*
963 * Need not hold lo_ctl_mutex to fput backing file.
964 * Calling fput holding lo_ctl_mutex triggers a circular
965 * lock dependency possibility warning as fput can take
966 * bd_mutex which is usually taken before lo_ctl_mutex.
967 */
968 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 return 0;
970}
971
972static int
973loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
974{
975 int err;
976 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +1100977 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
David Howellsb0fafa82008-11-14 10:38:41 +1100979 if (lo->lo_encrypt_key_size &&
980 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 !capable(CAP_SYS_ADMIN))
982 return -EPERM;
983 if (lo->lo_state != Lo_bound)
984 return -ENXIO;
985 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
986 return -EINVAL;
987
988 err = loop_release_xfer(lo);
989 if (err)
990 return err;
991
992 if (info->lo_encrypt_type) {
993 unsigned int type = info->lo_encrypt_type;
994
995 if (type >= MAX_LO_CRYPT)
996 return -EINVAL;
997 xfer = xfer_funcs[type];
998 if (xfer == NULL)
999 return -EINVAL;
1000 } else
1001 xfer = NULL;
1002
1003 err = loop_init_xfer(lo, xfer, info);
1004 if (err)
1005 return err;
1006
1007 if (lo->lo_offset != info->lo_offset ||
1008 lo->lo_sizelimit != info->lo_sizelimit) {
1009 lo->lo_offset = info->lo_offset;
1010 lo->lo_sizelimit = info->lo_sizelimit;
1011 if (figure_loop_size(lo))
1012 return -EFBIG;
1013 }
1014
1015 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1016 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1017 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1018 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1019
1020 if (!xfer)
1021 xfer = &none_funcs;
1022 lo->transfer = xfer->transfer;
1023 lo->ioctl = xfer->ioctl;
1024
David Woodhouse96c58652008-02-06 01:36:27 -08001025 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1026 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1027 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1030 lo->lo_init[0] = info->lo_init[0];
1031 lo->lo_init[1] = info->lo_init[1];
1032 if (info->lo_encrypt_key_size) {
1033 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1034 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001035 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037
1038 return 0;
1039}
1040
1041static int
1042loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1043{
1044 struct file *file = lo->lo_backing_file;
1045 struct kstat stat;
1046 int error;
1047
1048 if (lo->lo_state != Lo_bound)
1049 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001050 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (error)
1052 return error;
1053 memset(info, 0, sizeof(*info));
1054 info->lo_number = lo->lo_number;
1055 info->lo_device = huge_encode_dev(stat.dev);
1056 info->lo_inode = stat.ino;
1057 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1058 info->lo_offset = lo->lo_offset;
1059 info->lo_sizelimit = lo->lo_sizelimit;
1060 info->lo_flags = lo->lo_flags;
1061 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1062 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1063 info->lo_encrypt_type =
1064 lo->lo_encryption ? lo->lo_encryption->number : 0;
1065 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1066 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1067 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1068 lo->lo_encrypt_key_size);
1069 }
1070 return 0;
1071}
1072
1073static void
1074loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1075{
1076 memset(info64, 0, sizeof(*info64));
1077 info64->lo_number = info->lo_number;
1078 info64->lo_device = info->lo_device;
1079 info64->lo_inode = info->lo_inode;
1080 info64->lo_rdevice = info->lo_rdevice;
1081 info64->lo_offset = info->lo_offset;
1082 info64->lo_sizelimit = 0;
1083 info64->lo_encrypt_type = info->lo_encrypt_type;
1084 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1085 info64->lo_flags = info->lo_flags;
1086 info64->lo_init[0] = info->lo_init[0];
1087 info64->lo_init[1] = info->lo_init[1];
1088 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1089 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1090 else
1091 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1092 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1093}
1094
1095static int
1096loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1097{
1098 memset(info, 0, sizeof(*info));
1099 info->lo_number = info64->lo_number;
1100 info->lo_device = info64->lo_device;
1101 info->lo_inode = info64->lo_inode;
1102 info->lo_rdevice = info64->lo_rdevice;
1103 info->lo_offset = info64->lo_offset;
1104 info->lo_encrypt_type = info64->lo_encrypt_type;
1105 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1106 info->lo_flags = info64->lo_flags;
1107 info->lo_init[0] = info64->lo_init[0];
1108 info->lo_init[1] = info64->lo_init[1];
1109 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1110 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1111 else
1112 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1113 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1114
1115 /* error in case values were truncated */
1116 if (info->lo_device != info64->lo_device ||
1117 info->lo_rdevice != info64->lo_rdevice ||
1118 info->lo_inode != info64->lo_inode ||
1119 info->lo_offset != info64->lo_offset)
1120 return -EOVERFLOW;
1121
1122 return 0;
1123}
1124
1125static int
1126loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1127{
1128 struct loop_info info;
1129 struct loop_info64 info64;
1130
1131 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1132 return -EFAULT;
1133 loop_info64_from_old(&info, &info64);
1134 return loop_set_status(lo, &info64);
1135}
1136
1137static int
1138loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1139{
1140 struct loop_info64 info64;
1141
1142 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1143 return -EFAULT;
1144 return loop_set_status(lo, &info64);
1145}
1146
1147static int
1148loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1149 struct loop_info info;
1150 struct loop_info64 info64;
1151 int err = 0;
1152
1153 if (!arg)
1154 err = -EINVAL;
1155 if (!err)
1156 err = loop_get_status(lo, &info64);
1157 if (!err)
1158 err = loop_info64_to_old(&info64, &info);
1159 if (!err && copy_to_user(arg, &info, sizeof(info)))
1160 err = -EFAULT;
1161
1162 return err;
1163}
1164
1165static int
1166loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1167 struct loop_info64 info64;
1168 int err = 0;
1169
1170 if (!arg)
1171 err = -EINVAL;
1172 if (!err)
1173 err = loop_get_status(lo, &info64);
1174 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1175 err = -EFAULT;
1176
1177 return err;
1178}
1179
J. R. Okajima53d66602009-03-31 15:23:43 -07001180static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1181{
1182 int err;
1183 sector_t sec;
1184 loff_t sz;
1185
1186 err = -ENXIO;
1187 if (unlikely(lo->lo_state != Lo_bound))
1188 goto out;
1189 err = figure_loop_size(lo);
1190 if (unlikely(err))
1191 goto out;
1192 sec = get_capacity(lo->lo_disk);
1193 /* the width of sector_t may be narrow for bit-shift */
1194 sz = sec;
1195 sz <<= 9;
1196 mutex_lock(&bdev->bd_mutex);
1197 bd_set_size(bdev, sz);
David Zeuthenc3473c62010-05-03 14:08:59 +02001198 /* let user-space know about the new size */
1199 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
J. R. Okajima53d66602009-03-31 15:23:43 -07001200 mutex_unlock(&bdev->bd_mutex);
1201
1202 out:
1203 return err;
1204}
1205
Al Virobb214882008-03-02 09:29:48 -05001206static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 unsigned int cmd, unsigned long arg)
1208{
Al Virobb214882008-03-02 09:29:48 -05001209 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 int err;
1211
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001212 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 switch (cmd) {
1214 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001215 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 break;
1217 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001218 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 break;
1220 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001221 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Al Virobb214882008-03-02 09:29:48 -05001222 err = loop_clr_fd(lo, bdev);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001223 if (!err)
1224 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 break;
1226 case LOOP_SET_STATUS:
1227 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1228 break;
1229 case LOOP_GET_STATUS:
1230 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1231 break;
1232 case LOOP_SET_STATUS64:
1233 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1234 break;
1235 case LOOP_GET_STATUS64:
1236 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1237 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001238 case LOOP_SET_CAPACITY:
1239 err = -EPERM;
1240 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1241 err = loop_set_capacity(lo, bdev);
1242 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 default:
1244 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1245 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001246 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001247
1248out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return err;
1250}
1251
David Howells863d5b822006-08-29 19:06:14 +01001252#ifdef CONFIG_COMPAT
1253struct compat_loop_info {
1254 compat_int_t lo_number; /* ioctl r/o */
1255 compat_dev_t lo_device; /* ioctl r/o */
1256 compat_ulong_t lo_inode; /* ioctl r/o */
1257 compat_dev_t lo_rdevice; /* ioctl r/o */
1258 compat_int_t lo_offset;
1259 compat_int_t lo_encrypt_type;
1260 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1261 compat_int_t lo_flags; /* ioctl r/o */
1262 char lo_name[LO_NAME_SIZE];
1263 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1264 compat_ulong_t lo_init[2];
1265 char reserved[4];
1266};
1267
1268/*
1269 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1270 * - noinlined to reduce stack space usage in main part of driver
1271 */
1272static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001273loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001274 struct loop_info64 *info64)
1275{
1276 struct compat_loop_info info;
1277
1278 if (copy_from_user(&info, arg, sizeof(info)))
1279 return -EFAULT;
1280
1281 memset(info64, 0, sizeof(*info64));
1282 info64->lo_number = info.lo_number;
1283 info64->lo_device = info.lo_device;
1284 info64->lo_inode = info.lo_inode;
1285 info64->lo_rdevice = info.lo_rdevice;
1286 info64->lo_offset = info.lo_offset;
1287 info64->lo_sizelimit = 0;
1288 info64->lo_encrypt_type = info.lo_encrypt_type;
1289 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1290 info64->lo_flags = info.lo_flags;
1291 info64->lo_init[0] = info.lo_init[0];
1292 info64->lo_init[1] = info.lo_init[1];
1293 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1294 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1295 else
1296 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1297 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1298 return 0;
1299}
1300
1301/*
1302 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1303 * - noinlined to reduce stack space usage in main part of driver
1304 */
1305static noinline int
1306loop_info64_to_compat(const struct loop_info64 *info64,
1307 struct compat_loop_info __user *arg)
1308{
1309 struct compat_loop_info info;
1310
1311 memset(&info, 0, sizeof(info));
1312 info.lo_number = info64->lo_number;
1313 info.lo_device = info64->lo_device;
1314 info.lo_inode = info64->lo_inode;
1315 info.lo_rdevice = info64->lo_rdevice;
1316 info.lo_offset = info64->lo_offset;
1317 info.lo_encrypt_type = info64->lo_encrypt_type;
1318 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1319 info.lo_flags = info64->lo_flags;
1320 info.lo_init[0] = info64->lo_init[0];
1321 info.lo_init[1] = info64->lo_init[1];
1322 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1323 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1324 else
1325 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1326 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1327
1328 /* error in case values were truncated */
1329 if (info.lo_device != info64->lo_device ||
1330 info.lo_rdevice != info64->lo_rdevice ||
1331 info.lo_inode != info64->lo_inode ||
1332 info.lo_offset != info64->lo_offset ||
1333 info.lo_init[0] != info64->lo_init[0] ||
1334 info.lo_init[1] != info64->lo_init[1])
1335 return -EOVERFLOW;
1336
1337 if (copy_to_user(arg, &info, sizeof(info)))
1338 return -EFAULT;
1339 return 0;
1340}
1341
1342static int
1343loop_set_status_compat(struct loop_device *lo,
1344 const struct compat_loop_info __user *arg)
1345{
1346 struct loop_info64 info64;
1347 int ret;
1348
1349 ret = loop_info64_from_compat(arg, &info64);
1350 if (ret < 0)
1351 return ret;
1352 return loop_set_status(lo, &info64);
1353}
1354
1355static int
1356loop_get_status_compat(struct loop_device *lo,
1357 struct compat_loop_info __user *arg)
1358{
1359 struct loop_info64 info64;
1360 int err = 0;
1361
1362 if (!arg)
1363 err = -EINVAL;
1364 if (!err)
1365 err = loop_get_status(lo, &info64);
1366 if (!err)
1367 err = loop_info64_to_compat(&info64, arg);
1368 return err;
1369}
1370
Al Virobb214882008-03-02 09:29:48 -05001371static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1372 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001373{
Al Virobb214882008-03-02 09:29:48 -05001374 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001375 int err;
1376
David Howells863d5b822006-08-29 19:06:14 +01001377 switch(cmd) {
1378 case LOOP_SET_STATUS:
1379 mutex_lock(&lo->lo_ctl_mutex);
1380 err = loop_set_status_compat(
1381 lo, (const struct compat_loop_info __user *) arg);
1382 mutex_unlock(&lo->lo_ctl_mutex);
1383 break;
1384 case LOOP_GET_STATUS:
1385 mutex_lock(&lo->lo_ctl_mutex);
1386 err = loop_get_status_compat(
1387 lo, (struct compat_loop_info __user *) arg);
1388 mutex_unlock(&lo->lo_ctl_mutex);
1389 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001390 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001391 case LOOP_CLR_FD:
1392 case LOOP_GET_STATUS64:
1393 case LOOP_SET_STATUS64:
1394 arg = (unsigned long) compat_ptr(arg);
1395 case LOOP_SET_FD:
1396 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001397 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001398 break;
1399 default:
1400 err = -ENOIOCTLCMD;
1401 break;
1402 }
David Howells863d5b822006-08-29 19:06:14 +01001403 return err;
1404}
1405#endif
1406
Al Virobb214882008-03-02 09:29:48 -05001407static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
Al Virobb214882008-03-02 09:29:48 -05001409 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Ingo Molnarf85221d2006-03-23 03:00:38 -08001411 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001413 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 return 0;
1416}
1417
Al Virobb214882008-03-02 09:29:48 -05001418static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
Al Virobb214882008-03-02 09:29:48 -05001420 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001421 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Ingo Molnarf85221d2006-03-23 03:00:38 -08001423 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001424
Milan Broz14f27932008-12-12 14:48:27 +01001425 if (--lo->lo_refcnt)
1426 goto out;
1427
1428 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1429 /*
1430 * In autoclear mode, stop the loop thread
1431 * and remove configuration after last close.
1432 */
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001433 err = loop_clr_fd(lo, NULL);
1434 if (!err)
1435 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001436 } else {
1437 /*
1438 * Otherwise keep thread (if running) and config,
1439 * but flush possible ongoing bios in thread.
1440 */
1441 loop_flush(lo);
1442 }
David Woodhouse96c58652008-02-06 01:36:27 -08001443
Milan Broz14f27932008-12-12 14:48:27 +01001444out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001445 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001446out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 return 0;
1448}
1449
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001450static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001452 .open = lo_open,
1453 .release = lo_release,
1454 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001455#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001456 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001457#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458};
1459
1460/*
1461 * And now the modules code and kernel interface.
1462 */
Ken Chen73285082007-05-08 00:28:20 -07001463static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001465MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001466module_param(max_part, int, 0);
1467MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468MODULE_LICENSE("GPL");
1469MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1470
1471int loop_register_transfer(struct loop_func_table *funcs)
1472{
1473 unsigned int n = funcs->number;
1474
1475 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1476 return -EINVAL;
1477 xfer_funcs[n] = funcs;
1478 return 0;
1479}
1480
1481int loop_unregister_transfer(int number)
1482{
1483 unsigned int n = number;
1484 struct loop_device *lo;
1485 struct loop_func_table *xfer;
1486
1487 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1488 return -EINVAL;
1489
1490 xfer_funcs[n] = NULL;
1491
Ken Chen73285082007-05-08 00:28:20 -07001492 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001493 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 if (lo->lo_encryption == xfer)
1496 loop_release_xfer(lo);
1497
Ingo Molnarf85221d2006-03-23 03:00:38 -08001498 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 }
1500
1501 return 0;
1502}
1503
1504EXPORT_SYMBOL(loop_register_transfer);
1505EXPORT_SYMBOL(loop_unregister_transfer);
1506
Ken Chena47653f2007-06-08 13:46:44 -07001507static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001508{
1509 struct loop_device *lo;
1510 struct gendisk *disk;
1511
1512 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1513 if (!lo)
1514 goto out;
1515
1516 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1517 if (!lo->lo_queue)
1518 goto out_free_dev;
1519
Laurent Vivier476a4812008-03-26 12:11:53 +01001520 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001521 if (!disk)
1522 goto out_free_queue;
1523
1524 mutex_init(&lo->lo_ctl_mutex);
1525 lo->lo_number = i;
1526 lo->lo_thread = NULL;
1527 init_waitqueue_head(&lo->lo_event);
1528 spin_lock_init(&lo->lo_lock);
1529 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001530 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001531 disk->fops = &lo_fops;
1532 disk->private_data = lo;
1533 disk->queue = lo->lo_queue;
1534 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001535 return lo;
1536
1537out_free_queue:
1538 blk_cleanup_queue(lo->lo_queue);
1539out_free_dev:
1540 kfree(lo);
1541out:
Al Viro07002e92007-05-12 16:23:15 -04001542 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001543}
1544
Ken Chena47653f2007-06-08 13:46:44 -07001545static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001546{
Ken Chen73285082007-05-08 00:28:20 -07001547 blk_cleanup_queue(lo->lo_queue);
1548 put_disk(lo->lo_disk);
1549 list_del(&lo->lo_list);
1550 kfree(lo);
1551}
1552
Ken Chena47653f2007-06-08 13:46:44 -07001553static struct loop_device *loop_init_one(int i)
1554{
1555 struct loop_device *lo;
1556
1557 list_for_each_entry(lo, &loop_devices, lo_list) {
1558 if (lo->lo_number == i)
1559 return lo;
1560 }
1561
1562 lo = loop_alloc(i);
1563 if (lo) {
1564 add_disk(lo->lo_disk);
1565 list_add_tail(&lo->lo_list, &loop_devices);
1566 }
1567 return lo;
1568}
1569
1570static void loop_del_one(struct loop_device *lo)
1571{
1572 del_gendisk(lo->lo_disk);
1573 loop_free(lo);
1574}
1575
Ken Chen73285082007-05-08 00:28:20 -07001576static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1577{
Al Viro705962c2007-05-13 05:52:32 -04001578 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001579 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001580
Al Viro705962c2007-05-13 05:52:32 -04001581 mutex_lock(&loop_devices_mutex);
1582 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001583 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001584 mutex_unlock(&loop_devices_mutex);
1585
1586 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001587 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001588}
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590static int __init loop_init(void)
1591{
Ken Chena47653f2007-06-08 13:46:44 -07001592 int i, nr;
1593 unsigned long range;
1594 struct loop_device *lo, *next;
1595
1596 /*
1597 * loop module now has a feature to instantiate underlying device
1598 * structure on-demand, provided that there is an access dev node.
1599 * However, this will not work well with user space tool that doesn't
1600 * know about such "feature". In order to not break any existing
1601 * tool, we do the following:
1602 *
1603 * (1) if max_loop is specified, create that many upfront, and this
1604 * also becomes a hard limit.
1605 * (2) if max_loop is not specified, create 8 loop device on module
1606 * load, user can further extend loop device by create dev node
1607 * themselves and have kernel automatically instantiate actual
1608 * device on-demand.
1609 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001610
1611 part_shift = 0;
1612 if (max_part > 0)
1613 part_shift = fls(max_part);
1614
1615 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001616 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
Ken Chen73285082007-05-08 00:28:20 -07001618 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001619 nr = max_loop;
1620 range = max_loop;
1621 } else {
1622 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001623 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 }
Ken Chena47653f2007-06-08 13:46:44 -07001625
1626 if (register_blkdev(LOOP_MAJOR, "loop"))
1627 return -EIO;
1628
1629 for (i = 0; i < nr; i++) {
1630 lo = loop_alloc(i);
1631 if (!lo)
1632 goto Enomem;
1633 list_add_tail(&lo->lo_list, &loop_devices);
1634 }
1635
1636 /* point of no return */
1637
1638 list_for_each_entry(lo, &loop_devices, lo_list)
1639 add_disk(lo->lo_disk);
1640
1641 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1642 THIS_MODULE, loop_probe, NULL, NULL);
1643
Ken Chen73285082007-05-08 00:28:20 -07001644 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001646
1647Enomem:
1648 printk(KERN_INFO "loop: out of memory\n");
1649
1650 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1651 loop_free(lo);
1652
1653 unregister_blkdev(LOOP_MAJOR, "loop");
1654 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655}
1656
Ken Chen73285082007-05-08 00:28:20 -07001657static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
Ken Chena47653f2007-06-08 13:46:44 -07001659 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001660 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Laurent Vivier476a4812008-03-26 12:11:53 +01001662 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001663
Ken Chen73285082007-05-08 00:28:20 -07001664 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1665 loop_del_one(lo);
1666
Ken Chena47653f2007-06-08 13:46:44 -07001667 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001668 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
1670
1671module_init(loop_init);
1672module_exit(loop_exit);
1673
1674#ifndef MODULE
1675static int __init max_loop_setup(char *str)
1676{
1677 max_loop = simple_strtol(str, NULL, 0);
1678 return 1;
1679}
1680
1681__setup("max_loop=", max_loop_setup);
1682#endif