blob: 67d7bc812364588458fe511ba5f28fa8b41e6299 [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
43 * operations prepare_write and/or commit_write are not available on the
44 * backing filesystem.
45 * Anton Altaparmakov, 16 Feb 2005
46 *
47 * Still To Fix:
48 * - Advisory locking is ignored here.
49 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
50 *
51 */
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/module.h>
54#include <linux/moduleparam.h>
55#include <linux/sched.h>
56#include <linux/fs.h>
57#include <linux/file.h>
58#include <linux/stat.h>
59#include <linux/errno.h>
60#include <linux/major.h>
61#include <linux/wait.h>
62#include <linux/blkdev.h>
63#include <linux/blkpg.h>
64#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/smp_lock.h>
66#include <linux/swap.h>
67#include <linux/slab.h>
68#include <linux/loop.h>
David Howells863d5b822006-08-29 19:06:14 +010069#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070071#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/writeback.h>
73#include <linux/buffer_head.h> /* for invalidate_bdev() */
74#include <linux/completion.h>
75#include <linux/highmem.h>
76#include <linux/gfp.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070077#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020078#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80#include <asm/uaccess.h>
81
Ken Chen73285082007-05-08 00:28:20 -070082static LIST_HEAD(loop_devices);
83static DEFINE_MUTEX(loop_devices_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Laurent Vivier476a4812008-03-26 12:11:53 +010085static int max_part;
86static int part_shift;
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
89 * Transfer functions
90 */
91static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
96 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
98
99 if (cmd == READ)
100 memcpy(loop_buf, raw_buf, size);
101 else
102 memcpy(raw_buf, loop_buf, size);
103
104 kunmap_atomic(raw_buf, KM_USER0);
105 kunmap_atomic(loop_buf, KM_USER1);
106 cond_resched();
107 return 0;
108}
109
110static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
114{
115 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
117 char *in, *out, *key;
118 int i, keysize;
119
120 if (cmd == READ) {
121 in = raw_buf;
122 out = loop_buf;
123 } else {
124 in = loop_buf;
125 out = raw_buf;
126 }
127
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
132
133 kunmap_atomic(raw_buf, KM_USER0);
134 kunmap_atomic(loop_buf, KM_USER1);
135 cond_resched();
136 return 0;
137}
138
139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140{
141 if (unlikely(info->lo_encrypt_key_size <= 0))
142 return -EINVAL;
143 return 0;
144}
145
146static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
149};
150
151static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
154 .init = xor_init
155};
156
157/* xfer_funcs[0] is special - its release function is never called */
158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 &none_funcs,
160 &xor_funcs
161};
162
163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164{
165 loff_t size, offset, loopsize;
166
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 offset = lo->lo_offset;
170 loopsize = size - offset;
171 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
172 loopsize = lo->lo_sizelimit;
173
174 /*
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
177 */
178 return loopsize >> 9;
179}
180
181static int
182figure_loop_size(struct loop_device *lo)
183{
184 loff_t size = get_loop_size(lo, lo->lo_backing_file);
185 sector_t x = (sector_t)size;
186
187 if (unlikely((loff_t)x != size))
188 return -EFBIG;
189
Ken Chen73285082007-05-08 00:28:20 -0700190 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return 0;
192}
193
194static inline int
195lo_do_transfer(struct loop_device *lo, int cmd,
196 struct page *rpage, unsigned roffs,
197 struct page *lpage, unsigned loffs,
198 int size, sector_t rblock)
199{
200 if (unlikely(!lo->transfer))
201 return 0;
202
203 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
204}
205
206/**
207 * do_lo_send_aops - helper for writing data to a loop device
208 *
209 * This is the fast version for backing filesystems which implement the address
Nick Pigginafddba42007-10-16 01:25:01 -0700210 * space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 */
212static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
Nick Pigginafddba42007-10-16 01:25:01 -0700213 int bsize, loff_t pos, struct page *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
216 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 pgoff_t index;
218 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800219 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800221 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 index = pos >> PAGE_CACHE_SHIFT;
223 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
224 bv_offs = bvec->bv_offset;
225 len = bvec->bv_len;
226 while (len > 0) {
227 sector_t IV;
Nick Pigginafddba42007-10-16 01:25:01 -0700228 unsigned size, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 int transfer_result;
Nick Pigginafddba42007-10-16 01:25:01 -0700230 struct page *page;
231 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
234 size = PAGE_CACHE_SIZE - offset;
235 if (size > len)
236 size = len;
Nick Pigginafddba42007-10-16 01:25:01 -0700237
238 ret = pagecache_write_begin(file, mapping, pos, size, 0,
239 &page, &fsdata);
240 if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
244 bvec->bv_page, bv_offs, size, IV);
Nick Pigginafddba42007-10-16 01:25:01 -0700245 copied = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (unlikely(transfer_result))
Nick Pigginafddba42007-10-16 01:25:01 -0700247 copied = 0;
248
249 ret = pagecache_write_end(file, mapping, pos, size, copied,
250 page, fsdata);
Dmitry Monakhov8268f5a2007-10-16 01:25:02 -0700251 if (ret < 0 || ret != copied)
Nick Pigginafddba42007-10-16 01:25:01 -0700252 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700253
254 if (unlikely(transfer_result))
255 goto fail;
256
257 bv_offs += copied;
258 len -= copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 offset = 0;
260 index++;
Nick Pigginafddba42007-10-16 01:25:01 -0700261 pos += copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800263 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800265 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267fail:
268 ret = -1;
269 goto out;
270}
271
272/**
273 * __do_lo_send_write - helper for writing data to a loop device
274 *
275 * This helper just factors out common code between do_lo_send_direct_write()
276 * and do_lo_send_write().
277 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800278static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100279 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 ssize_t bw;
282 mm_segment_t old_fs = get_fs();
283
284 set_fs(get_ds());
285 bw = file->f_op->write(file, buf, len, &pos);
286 set_fs(old_fs);
287 if (likely(bw == len))
288 return 0;
289 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
290 (unsigned long long)pos, len);
291 if (bw >= 0)
292 bw = -EIO;
293 return bw;
294}
295
296/**
297 * do_lo_send_direct_write - helper for writing data to a loop device
298 *
299 * This is the fast, non-transforming version for backing filesystems which do
Nick Pigginafddba42007-10-16 01:25:01 -0700300 * not implement the address space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * It uses the write file operation which should be present on all writeable
302 * filesystems.
303 */
304static int do_lo_send_direct_write(struct loop_device *lo,
305 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
306{
307 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100308 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 bvec->bv_len, pos);
310 kunmap(bvec->bv_page);
311 cond_resched();
312 return bw;
313}
314
315/**
316 * do_lo_send_write - helper for writing data to a loop device
317 *
318 * This is the slow, transforming version for filesystems which do not
Nick Pigginafddba42007-10-16 01:25:01 -0700319 * implement the address space operations write_begin and write_end. It
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * uses the write file operation which should be present on all writeable
321 * filesystems.
322 *
323 * Using fops->write is slower than using aops->{prepare,commit}_write in the
324 * transforming case because we need to double buffer the data as we cannot do
325 * the transformations in place as we do not have direct access to the
326 * destination pages of the backing file.
327 */
328static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
329 int bsize, loff_t pos, struct page *page)
330{
331 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
332 bvec->bv_offset, bvec->bv_len, pos >> 9);
333 if (likely(!ret))
334 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100335 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 pos);
337 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
338 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
339 if (ret > 0)
340 ret = -EIO;
341 return ret;
342}
343
344static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
345 loff_t pos)
346{
347 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
348 struct page *page);
349 struct bio_vec *bvec;
350 struct page *page = NULL;
351 int i, ret = 0;
352
353 do_lo_send = do_lo_send_aops;
354 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
355 do_lo_send = do_lo_send_direct_write;
356 if (lo->transfer != transfer_none) {
357 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
358 if (unlikely(!page))
359 goto fail;
360 kmap(page);
361 do_lo_send = do_lo_send_write;
362 }
363 }
364 bio_for_each_segment(bvec, bio, i) {
365 ret = do_lo_send(lo, bvec, bsize, pos, page);
366 if (ret < 0)
367 break;
368 pos += bvec->bv_len;
369 }
370 if (page) {
371 kunmap(page);
372 __free_page(page);
373 }
374out:
375 return ret;
376fail:
377 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
378 ret = -ENOMEM;
379 goto out;
380}
381
382struct lo_read_data {
383 struct loop_device *lo;
384 struct page *page;
385 unsigned offset;
386 int bsize;
387};
388
389static int
Jens Axboefd582142007-06-12 21:20:37 +0200390lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
391 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Jens Axboefd582142007-06-12 21:20:37 +0200393 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200395 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 sector_t IV;
Jens Axboefd582142007-06-12 21:20:37 +0200397 size_t size;
398 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Jens Axboecac36bb02007-06-14 13:10:48 +0200400 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200401 if (unlikely(ret))
402 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Jens Axboefd582142007-06-12 21:20:37 +0200404 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
405 (buf->offset >> 9);
406 size = sd->len;
407 if (size > p->bsize)
408 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Jens Axboefd582142007-06-12 21:20:37 +0200410 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 printk(KERN_ERR "loop: transfer error block %ld\n",
412 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200413 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
415
416 flush_dcache_page(p->page);
417
Jens Axboefd582142007-06-12 21:20:37 +0200418 if (size > 0)
419 p->offset += size;
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return size;
422}
423
424static int
Jens Axboefd582142007-06-12 21:20:37 +0200425lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
426{
427 return __splice_from_pipe(pipe, sd, lo_splice_actor);
428}
429
430static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431do_lo_receive(struct loop_device *lo,
432 struct bio_vec *bvec, int bsize, loff_t pos)
433{
434 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200435 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200437 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 cookie.lo = lo;
440 cookie.page = bvec->bv_page;
441 cookie.offset = bvec->bv_offset;
442 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200443
444 sd.len = 0;
445 sd.total_len = bvec->bv_len;
446 sd.flags = 0;
447 sd.pos = pos;
448 sd.u.data = &cookie;
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200451 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
452
453 if (retval < 0)
454 return retval;
455
456 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
459static int
460lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
461{
462 struct bio_vec *bvec;
463 int i, ret = 0;
464
465 bio_for_each_segment(bvec, bio, i) {
466 ret = do_lo_receive(lo, bvec, bsize, pos);
467 if (ret < 0)
468 break;
469 pos += bvec->bv_len;
470 }
471 return ret;
472}
473
474static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
475{
476 loff_t pos;
477 int ret;
478
479 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
480 if (bio_rw(bio) == WRITE)
481 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
482 else
483 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
484 return ret;
485}
486
487/*
488 * Add bio to back of pending list
489 */
490static void loop_add_bio(struct loop_device *lo, struct bio *bio)
491{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (lo->lo_biotail) {
493 lo->lo_biotail->bi_next = bio;
494 lo->lo_biotail = bio;
495 } else
496 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
499/*
500 * Grab first pending buffer
501 */
502static struct bio *loop_get_bio(struct loop_device *lo)
503{
504 struct bio *bio;
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if ((bio = lo->lo_bio)) {
507 if (bio == lo->lo_biotail)
508 lo->lo_biotail = NULL;
509 lo->lo_bio = bio->bi_next;
510 bio->bi_next = NULL;
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 return bio;
514}
515
Jens Axboe165125e2007-07-24 09:28:11 +0200516static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 struct loop_device *lo = q->queuedata;
519 int rw = bio_rw(old_bio);
520
Nick Piggin35a82d12005-06-23 00:09:06 -0700521 if (rw == READA)
522 rw = READ;
523
524 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 spin_lock_irq(&lo->lo_lock);
527 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700528 goto out;
529 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
530 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700532 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700533 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700537 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200538 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
542/*
543 * kick off io on the underlying address space
544 */
Jens Axboe165125e2007-07-24 09:28:11 +0200545static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct loop_device *lo = q->queuedata;
548
Nick Piggin75ad23b2008-04-29 14:48:33 +0200549 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 blk_run_address_space(lo->lo_backing_file->f_mapping);
551}
552
553struct switch_request {
554 struct file *file;
555 struct completion wait;
556};
557
558static void do_loop_switch(struct loop_device *, struct switch_request *);
559
560static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
561{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (unlikely(!bio->bi_bdev)) {
563 do_loop_switch(lo, bio->bi_private);
564 bio_put(bio);
565 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700566 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200567 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569}
570
571/*
572 * worker thread that handles reads/writes to file backed loop devices,
573 * to avoid blocking in our make_request_fn. it also does loop decrypting
574 * on reads for block backed loop, as that is too heavy to do from
575 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700576 *
577 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
578 * calling kthread_stop(). Therefore once kthread_should_stop() is
579 * true, make_request will not place any more requests. Therefore
580 * once kthread_should_stop() is true and lo_bio is NULL, we are
581 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
583static int loop_thread(void *data)
584{
585 struct loop_device *lo = data;
586 struct bio *bio;
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 set_user_nice(current, -20);
589
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700590 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700592 wait_event_interruptible(lo->lo_event,
593 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700594
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700595 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700596 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700597 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700599 spin_unlock_irq(&lo->lo_lock);
600
601 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return 0;
606}
607
608/*
609 * loop_switch performs the hard work of switching a backing store.
610 * First it needs to flush existing IO, it does this by sending a magic
611 * BIO down the pipe. The completion of this BIO does the actual switch.
612 */
613static int loop_switch(struct loop_device *lo, struct file *file)
614{
615 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100616 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (!bio)
618 return -ENOMEM;
619 init_completion(&w.wait);
620 w.file = file;
621 bio->bi_private = &w;
622 bio->bi_bdev = NULL;
623 loop_make_request(lo->lo_queue, bio);
624 wait_for_completion(&w.wait);
625 return 0;
626}
627
628/*
629 * Do the actual switch; called from the BIO completion routine
630 */
631static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
632{
633 struct file *file = p->file;
634 struct file *old_file = lo->lo_backing_file;
635 struct address_space *mapping = file->f_mapping;
636
637 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
638 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700639 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
640 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 lo->old_gfp_mask = mapping_gfp_mask(mapping);
642 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
643 complete(&p->wait);
644}
645
646
647/*
648 * loop_change_fd switched the backing store of a loopback device to
649 * a new file. This is useful for operating system installers to free up
650 * the original file and in High Availability environments to switch to
651 * an alternative location for the content in case of server meltdown.
652 * This can only work if the loop device is used read-only, and if the
653 * new backing store is the same size and type as the old backing store.
654 */
Al Virobb214882008-03-02 09:29:48 -0500655static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
656 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 struct file *file, *old_file;
659 struct inode *inode;
660 int error;
661
662 error = -ENXIO;
663 if (lo->lo_state != Lo_bound)
664 goto out;
665
666 /* the loop device has to be read-only */
667 error = -EINVAL;
668 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
669 goto out;
670
671 error = -EBADF;
672 file = fget(arg);
673 if (!file)
674 goto out;
675
676 inode = file->f_mapping->host;
677 old_file = lo->lo_backing_file;
678
679 error = -EINVAL;
680
681 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 goto out_putf;
683
Jens Axboefd582142007-06-12 21:20:37 +0200684 /* new backing store needs to support loop (eg splice_read) */
685 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 goto out_putf;
687
688 /* size of the new backing store needs to be the same */
689 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
690 goto out_putf;
691
692 /* and ... switch */
693 error = loop_switch(lo, file);
694 if (error)
695 goto out_putf;
696
697 fput(old_file);
Laurent Vivier476a4812008-03-26 12:11:53 +0100698 if (max_part > 0)
699 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701
702 out_putf:
703 fput(file);
704 out:
705 return error;
706}
707
708static inline int is_loop_device(struct file *file)
709{
710 struct inode *i = file->f_mapping->host;
711
712 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
713}
714
Al Virobb214882008-03-02 09:29:48 -0500715static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 struct block_device *bdev, unsigned int arg)
717{
718 struct file *file, *f;
719 struct inode *inode;
720 struct address_space *mapping;
721 unsigned lo_blocksize;
722 int lo_flags = 0;
723 int error;
724 loff_t size;
725
726 /* This is safe, since we have a reference from open(). */
727 __module_get(THIS_MODULE);
728
729 error = -EBADF;
730 file = fget(arg);
731 if (!file)
732 goto out;
733
734 error = -EBUSY;
735 if (lo->lo_state != Lo_unbound)
736 goto out_putf;
737
738 /* Avoid recursion */
739 f = file;
740 while (is_loop_device(f)) {
741 struct loop_device *l;
742
Al Virobb214882008-03-02 09:29:48 -0500743 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 goto out_putf;
745
746 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
747 if (l->lo_state == Lo_unbound) {
748 error = -EINVAL;
749 goto out_putf;
750 }
751 f = l->lo_backing_file;
752 }
753
754 mapping = file->f_mapping;
755 inode = mapping->host;
756
757 if (!(file->f_mode & FMODE_WRITE))
758 lo_flags |= LO_FLAGS_READ_ONLY;
759
760 error = -EINVAL;
761 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700762 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 /*
764 * If we can't read - sorry. If we only can't write - well,
765 * it's going to be read-only.
766 */
Jens Axboefd582142007-06-12 21:20:37 +0200767 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 goto out_putf;
Nick Pigginafddba42007-10-16 01:25:01 -0700769 if (aops->prepare_write || aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 lo_flags |= LO_FLAGS_USE_AOPS;
771 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
772 lo_flags |= LO_FLAGS_READ_ONLY;
773
Theodore Ts'oba52de12006-09-27 01:50:49 -0700774 lo_blocksize = S_ISBLK(inode->i_mode) ?
775 inode->i_bdev->bd_block_size : PAGE_SIZE;
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 error = 0;
778 } else {
779 goto out_putf;
780 }
781
782 size = get_loop_size(lo, file);
783
784 if ((loff_t)(sector_t)size != size) {
785 error = -EFBIG;
786 goto out_putf;
787 }
788
Al Virobb214882008-03-02 09:29:48 -0500789 if (!(mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 lo_flags |= LO_FLAGS_READ_ONLY;
791
792 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
793
794 lo->lo_blocksize = lo_blocksize;
795 lo->lo_device = bdev;
796 lo->lo_flags = lo_flags;
797 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700798 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 lo->ioctl = NULL;
800 lo->lo_sizelimit = 0;
801 lo->old_gfp_mask = mapping_gfp_mask(mapping);
802 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
803
804 lo->lo_bio = lo->lo_biotail = NULL;
805
806 /*
807 * set queue make_request_fn, and add limits based on lower level
808 * device
809 */
810 blk_queue_make_request(lo->lo_queue, loop_make_request);
811 lo->lo_queue->queuedata = lo;
812 lo->lo_queue->unplug_fn = loop_unplug;
813
Ken Chen73285082007-05-08 00:28:20 -0700814 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 bd_set_size(bdev, size << 9);
816
817 set_blocksize(bdev, lo_blocksize);
818
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700819 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
820 lo->lo_number);
821 if (IS_ERR(lo->lo_thread)) {
822 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700823 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700824 }
825 lo->lo_state = Lo_bound;
826 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100827 if (max_part > 0)
828 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return 0;
830
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700831out_clr:
832 lo->lo_thread = NULL;
833 lo->lo_device = NULL;
834 lo->lo_backing_file = NULL;
835 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700836 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700837 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700838 bd_set_size(bdev, 0);
839 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
840 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 out_putf:
842 fput(file);
843 out:
844 /* This is safe: open() is still holding a reference. */
845 module_put(THIS_MODULE);
846 return error;
847}
848
849static int
850loop_release_xfer(struct loop_device *lo)
851{
852 int err = 0;
853 struct loop_func_table *xfer = lo->lo_encryption;
854
855 if (xfer) {
856 if (xfer->release)
857 err = xfer->release(lo);
858 lo->transfer = NULL;
859 lo->lo_encryption = NULL;
860 module_put(xfer->owner);
861 }
862 return err;
863}
864
865static int
866loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
867 const struct loop_info64 *i)
868{
869 int err = 0;
870
871 if (xfer) {
872 struct module *owner = xfer->owner;
873
874 if (!try_module_get(owner))
875 return -EINVAL;
876 if (xfer->init)
877 err = xfer->init(lo, i);
878 if (err)
879 module_put(owner);
880 else
881 lo->lo_encryption = xfer;
882 }
883 return err;
884}
885
886static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
887{
888 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400889 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 if (lo->lo_state != Lo_bound)
892 return -ENXIO;
893
894 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
895 return -EBUSY;
896
897 if (filp == NULL)
898 return -EINVAL;
899
900 spin_lock_irq(&lo->lo_lock);
901 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 spin_unlock_irq(&lo->lo_lock);
903
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700904 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 lo->lo_backing_file = NULL;
907
908 loop_release_xfer(lo);
909 lo->transfer = NULL;
910 lo->ioctl = NULL;
911 lo->lo_device = NULL;
912 lo->lo_encryption = NULL;
913 lo->lo_offset = 0;
914 lo->lo_sizelimit = 0;
915 lo->lo_encrypt_key_size = 0;
916 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700917 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
919 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
920 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -0500921 if (bdev)
922 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700923 set_capacity(lo->lo_disk, 0);
Al Virobb214882008-03-02 09:29:48 -0500924 if (bdev)
925 bd_set_size(bdev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 mapping_set_gfp_mask(filp->f_mapping, gfp);
927 lo->lo_state = Lo_unbound;
928 fput(filp);
929 /* This is safe: open() is still holding a reference. */
930 module_put(THIS_MODULE);
Laurent Vivier476a4812008-03-26 12:11:53 +0100931 if (max_part > 0)
932 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 return 0;
934}
935
936static int
937loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
938{
939 int err;
940 struct loop_func_table *xfer;
941
942 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
943 !capable(CAP_SYS_ADMIN))
944 return -EPERM;
945 if (lo->lo_state != Lo_bound)
946 return -ENXIO;
947 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
948 return -EINVAL;
949
950 err = loop_release_xfer(lo);
951 if (err)
952 return err;
953
954 if (info->lo_encrypt_type) {
955 unsigned int type = info->lo_encrypt_type;
956
957 if (type >= MAX_LO_CRYPT)
958 return -EINVAL;
959 xfer = xfer_funcs[type];
960 if (xfer == NULL)
961 return -EINVAL;
962 } else
963 xfer = NULL;
964
965 err = loop_init_xfer(lo, xfer, info);
966 if (err)
967 return err;
968
969 if (lo->lo_offset != info->lo_offset ||
970 lo->lo_sizelimit != info->lo_sizelimit) {
971 lo->lo_offset = info->lo_offset;
972 lo->lo_sizelimit = info->lo_sizelimit;
973 if (figure_loop_size(lo))
974 return -EFBIG;
975 }
976
977 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
978 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
979 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
980 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
981
982 if (!xfer)
983 xfer = &none_funcs;
984 lo->transfer = xfer->transfer;
985 lo->ioctl = xfer->ioctl;
986
David Woodhouse96c58652008-02-06 01:36:27 -0800987 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
988 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
989 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
992 lo->lo_init[0] = info->lo_init[0];
993 lo->lo_init[1] = info->lo_init[1];
994 if (info->lo_encrypt_key_size) {
995 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
996 info->lo_encrypt_key_size);
997 lo->lo_key_owner = current->uid;
998 }
999
1000 return 0;
1001}
1002
1003static int
1004loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1005{
1006 struct file *file = lo->lo_backing_file;
1007 struct kstat stat;
1008 int error;
1009
1010 if (lo->lo_state != Lo_bound)
1011 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001012 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (error)
1014 return error;
1015 memset(info, 0, sizeof(*info));
1016 info->lo_number = lo->lo_number;
1017 info->lo_device = huge_encode_dev(stat.dev);
1018 info->lo_inode = stat.ino;
1019 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1020 info->lo_offset = lo->lo_offset;
1021 info->lo_sizelimit = lo->lo_sizelimit;
1022 info->lo_flags = lo->lo_flags;
1023 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1024 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1025 info->lo_encrypt_type =
1026 lo->lo_encryption ? lo->lo_encryption->number : 0;
1027 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1028 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1029 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1030 lo->lo_encrypt_key_size);
1031 }
1032 return 0;
1033}
1034
1035static void
1036loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1037{
1038 memset(info64, 0, sizeof(*info64));
1039 info64->lo_number = info->lo_number;
1040 info64->lo_device = info->lo_device;
1041 info64->lo_inode = info->lo_inode;
1042 info64->lo_rdevice = info->lo_rdevice;
1043 info64->lo_offset = info->lo_offset;
1044 info64->lo_sizelimit = 0;
1045 info64->lo_encrypt_type = info->lo_encrypt_type;
1046 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1047 info64->lo_flags = info->lo_flags;
1048 info64->lo_init[0] = info->lo_init[0];
1049 info64->lo_init[1] = info->lo_init[1];
1050 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1051 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1052 else
1053 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1054 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1055}
1056
1057static int
1058loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1059{
1060 memset(info, 0, sizeof(*info));
1061 info->lo_number = info64->lo_number;
1062 info->lo_device = info64->lo_device;
1063 info->lo_inode = info64->lo_inode;
1064 info->lo_rdevice = info64->lo_rdevice;
1065 info->lo_offset = info64->lo_offset;
1066 info->lo_encrypt_type = info64->lo_encrypt_type;
1067 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1068 info->lo_flags = info64->lo_flags;
1069 info->lo_init[0] = info64->lo_init[0];
1070 info->lo_init[1] = info64->lo_init[1];
1071 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1072 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1073 else
1074 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1075 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1076
1077 /* error in case values were truncated */
1078 if (info->lo_device != info64->lo_device ||
1079 info->lo_rdevice != info64->lo_rdevice ||
1080 info->lo_inode != info64->lo_inode ||
1081 info->lo_offset != info64->lo_offset)
1082 return -EOVERFLOW;
1083
1084 return 0;
1085}
1086
1087static int
1088loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1089{
1090 struct loop_info info;
1091 struct loop_info64 info64;
1092
1093 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1094 return -EFAULT;
1095 loop_info64_from_old(&info, &info64);
1096 return loop_set_status(lo, &info64);
1097}
1098
1099static int
1100loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1101{
1102 struct loop_info64 info64;
1103
1104 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1105 return -EFAULT;
1106 return loop_set_status(lo, &info64);
1107}
1108
1109static int
1110loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1111 struct loop_info info;
1112 struct loop_info64 info64;
1113 int err = 0;
1114
1115 if (!arg)
1116 err = -EINVAL;
1117 if (!err)
1118 err = loop_get_status(lo, &info64);
1119 if (!err)
1120 err = loop_info64_to_old(&info64, &info);
1121 if (!err && copy_to_user(arg, &info, sizeof(info)))
1122 err = -EFAULT;
1123
1124 return err;
1125}
1126
1127static int
1128loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1129 struct loop_info64 info64;
1130 int err = 0;
1131
1132 if (!arg)
1133 err = -EINVAL;
1134 if (!err)
1135 err = loop_get_status(lo, &info64);
1136 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1137 err = -EFAULT;
1138
1139 return err;
1140}
1141
Al Virobb214882008-03-02 09:29:48 -05001142static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 unsigned int cmd, unsigned long arg)
1144{
Al Virobb214882008-03-02 09:29:48 -05001145 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 int err;
1147
Ingo Molnarf85221d2006-03-23 03:00:38 -08001148 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 switch (cmd) {
1150 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001151 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 break;
1153 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001154 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 break;
1156 case LOOP_CLR_FD:
Al Virobb214882008-03-02 09:29:48 -05001157 err = loop_clr_fd(lo, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 break;
1159 case LOOP_SET_STATUS:
1160 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1161 break;
1162 case LOOP_GET_STATUS:
1163 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1164 break;
1165 case LOOP_SET_STATUS64:
1166 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1167 break;
1168 case LOOP_GET_STATUS64:
1169 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1170 break;
1171 default:
1172 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1173 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001174 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 return err;
1176}
1177
David Howells863d5b822006-08-29 19:06:14 +01001178#ifdef CONFIG_COMPAT
1179struct compat_loop_info {
1180 compat_int_t lo_number; /* ioctl r/o */
1181 compat_dev_t lo_device; /* ioctl r/o */
1182 compat_ulong_t lo_inode; /* ioctl r/o */
1183 compat_dev_t lo_rdevice; /* ioctl r/o */
1184 compat_int_t lo_offset;
1185 compat_int_t lo_encrypt_type;
1186 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1187 compat_int_t lo_flags; /* ioctl r/o */
1188 char lo_name[LO_NAME_SIZE];
1189 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1190 compat_ulong_t lo_init[2];
1191 char reserved[4];
1192};
1193
1194/*
1195 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1196 * - noinlined to reduce stack space usage in main part of driver
1197 */
1198static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001199loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001200 struct loop_info64 *info64)
1201{
1202 struct compat_loop_info info;
1203
1204 if (copy_from_user(&info, arg, sizeof(info)))
1205 return -EFAULT;
1206
1207 memset(info64, 0, sizeof(*info64));
1208 info64->lo_number = info.lo_number;
1209 info64->lo_device = info.lo_device;
1210 info64->lo_inode = info.lo_inode;
1211 info64->lo_rdevice = info.lo_rdevice;
1212 info64->lo_offset = info.lo_offset;
1213 info64->lo_sizelimit = 0;
1214 info64->lo_encrypt_type = info.lo_encrypt_type;
1215 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1216 info64->lo_flags = info.lo_flags;
1217 info64->lo_init[0] = info.lo_init[0];
1218 info64->lo_init[1] = info.lo_init[1];
1219 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1220 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1221 else
1222 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1223 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1224 return 0;
1225}
1226
1227/*
1228 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1229 * - noinlined to reduce stack space usage in main part of driver
1230 */
1231static noinline int
1232loop_info64_to_compat(const struct loop_info64 *info64,
1233 struct compat_loop_info __user *arg)
1234{
1235 struct compat_loop_info info;
1236
1237 memset(&info, 0, sizeof(info));
1238 info.lo_number = info64->lo_number;
1239 info.lo_device = info64->lo_device;
1240 info.lo_inode = info64->lo_inode;
1241 info.lo_rdevice = info64->lo_rdevice;
1242 info.lo_offset = info64->lo_offset;
1243 info.lo_encrypt_type = info64->lo_encrypt_type;
1244 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1245 info.lo_flags = info64->lo_flags;
1246 info.lo_init[0] = info64->lo_init[0];
1247 info.lo_init[1] = info64->lo_init[1];
1248 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1249 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1250 else
1251 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1252 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1253
1254 /* error in case values were truncated */
1255 if (info.lo_device != info64->lo_device ||
1256 info.lo_rdevice != info64->lo_rdevice ||
1257 info.lo_inode != info64->lo_inode ||
1258 info.lo_offset != info64->lo_offset ||
1259 info.lo_init[0] != info64->lo_init[0] ||
1260 info.lo_init[1] != info64->lo_init[1])
1261 return -EOVERFLOW;
1262
1263 if (copy_to_user(arg, &info, sizeof(info)))
1264 return -EFAULT;
1265 return 0;
1266}
1267
1268static int
1269loop_set_status_compat(struct loop_device *lo,
1270 const struct compat_loop_info __user *arg)
1271{
1272 struct loop_info64 info64;
1273 int ret;
1274
1275 ret = loop_info64_from_compat(arg, &info64);
1276 if (ret < 0)
1277 return ret;
1278 return loop_set_status(lo, &info64);
1279}
1280
1281static int
1282loop_get_status_compat(struct loop_device *lo,
1283 struct compat_loop_info __user *arg)
1284{
1285 struct loop_info64 info64;
1286 int err = 0;
1287
1288 if (!arg)
1289 err = -EINVAL;
1290 if (!err)
1291 err = loop_get_status(lo, &info64);
1292 if (!err)
1293 err = loop_info64_to_compat(&info64, arg);
1294 return err;
1295}
1296
Al Virobb214882008-03-02 09:29:48 -05001297static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1298 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001299{
Al Virobb214882008-03-02 09:29:48 -05001300 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001301 int err;
1302
David Howells863d5b822006-08-29 19:06:14 +01001303 switch(cmd) {
1304 case LOOP_SET_STATUS:
1305 mutex_lock(&lo->lo_ctl_mutex);
1306 err = loop_set_status_compat(
1307 lo, (const struct compat_loop_info __user *) arg);
1308 mutex_unlock(&lo->lo_ctl_mutex);
1309 break;
1310 case LOOP_GET_STATUS:
1311 mutex_lock(&lo->lo_ctl_mutex);
1312 err = loop_get_status_compat(
1313 lo, (struct compat_loop_info __user *) arg);
1314 mutex_unlock(&lo->lo_ctl_mutex);
1315 break;
1316 case LOOP_CLR_FD:
1317 case LOOP_GET_STATUS64:
1318 case LOOP_SET_STATUS64:
1319 arg = (unsigned long) compat_ptr(arg);
1320 case LOOP_SET_FD:
1321 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001322 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001323 break;
1324 default:
1325 err = -ENOIOCTLCMD;
1326 break;
1327 }
David Howells863d5b822006-08-29 19:06:14 +01001328 return err;
1329}
1330#endif
1331
Al Virobb214882008-03-02 09:29:48 -05001332static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
Al Virobb214882008-03-02 09:29:48 -05001334 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Ingo Molnarf85221d2006-03-23 03:00:38 -08001336 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001338 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 return 0;
1341}
1342
Al Virobb214882008-03-02 09:29:48 -05001343static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
Al Virobb214882008-03-02 09:29:48 -05001345 struct loop_device *lo = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Ingo Molnarf85221d2006-03-23 03:00:38 -08001347 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 --lo->lo_refcnt;
David Woodhouse96c58652008-02-06 01:36:27 -08001349
1350 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt)
Al Virobb214882008-03-02 09:29:48 -05001351 loop_clr_fd(lo, NULL);
David Woodhouse96c58652008-02-06 01:36:27 -08001352
Ingo Molnarf85221d2006-03-23 03:00:38 -08001353 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 return 0;
1356}
1357
1358static struct block_device_operations lo_fops = {
1359 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001360 .open = lo_open,
1361 .release = lo_release,
1362 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001363#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001364 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001365#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366};
1367
1368/*
1369 * And now the modules code and kernel interface.
1370 */
Ken Chen73285082007-05-08 00:28:20 -07001371static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001373MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001374module_param(max_part, int, 0);
1375MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376MODULE_LICENSE("GPL");
1377MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1378
1379int loop_register_transfer(struct loop_func_table *funcs)
1380{
1381 unsigned int n = funcs->number;
1382
1383 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1384 return -EINVAL;
1385 xfer_funcs[n] = funcs;
1386 return 0;
1387}
1388
1389int loop_unregister_transfer(int number)
1390{
1391 unsigned int n = number;
1392 struct loop_device *lo;
1393 struct loop_func_table *xfer;
1394
1395 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1396 return -EINVAL;
1397
1398 xfer_funcs[n] = NULL;
1399
Ken Chen73285082007-05-08 00:28:20 -07001400 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001401 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 if (lo->lo_encryption == xfer)
1404 loop_release_xfer(lo);
1405
Ingo Molnarf85221d2006-03-23 03:00:38 -08001406 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
1408
1409 return 0;
1410}
1411
1412EXPORT_SYMBOL(loop_register_transfer);
1413EXPORT_SYMBOL(loop_unregister_transfer);
1414
Ken Chena47653f2007-06-08 13:46:44 -07001415static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001416{
1417 struct loop_device *lo;
1418 struct gendisk *disk;
1419
1420 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1421 if (!lo)
1422 goto out;
1423
1424 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1425 if (!lo->lo_queue)
1426 goto out_free_dev;
1427
Laurent Vivier476a4812008-03-26 12:11:53 +01001428 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001429 if (!disk)
1430 goto out_free_queue;
1431
1432 mutex_init(&lo->lo_ctl_mutex);
1433 lo->lo_number = i;
1434 lo->lo_thread = NULL;
1435 init_waitqueue_head(&lo->lo_event);
1436 spin_lock_init(&lo->lo_lock);
1437 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001438 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001439 disk->fops = &lo_fops;
1440 disk->private_data = lo;
1441 disk->queue = lo->lo_queue;
1442 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001443 return lo;
1444
1445out_free_queue:
1446 blk_cleanup_queue(lo->lo_queue);
1447out_free_dev:
1448 kfree(lo);
1449out:
Al Viro07002e92007-05-12 16:23:15 -04001450 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001451}
1452
Ken Chena47653f2007-06-08 13:46:44 -07001453static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001454{
Ken Chen73285082007-05-08 00:28:20 -07001455 blk_cleanup_queue(lo->lo_queue);
1456 put_disk(lo->lo_disk);
1457 list_del(&lo->lo_list);
1458 kfree(lo);
1459}
1460
Ken Chena47653f2007-06-08 13:46:44 -07001461static struct loop_device *loop_init_one(int i)
1462{
1463 struct loop_device *lo;
1464
1465 list_for_each_entry(lo, &loop_devices, lo_list) {
1466 if (lo->lo_number == i)
1467 return lo;
1468 }
1469
1470 lo = loop_alloc(i);
1471 if (lo) {
1472 add_disk(lo->lo_disk);
1473 list_add_tail(&lo->lo_list, &loop_devices);
1474 }
1475 return lo;
1476}
1477
1478static void loop_del_one(struct loop_device *lo)
1479{
1480 del_gendisk(lo->lo_disk);
1481 loop_free(lo);
1482}
1483
Ken Chen73285082007-05-08 00:28:20 -07001484static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1485{
Al Viro705962c2007-05-13 05:52:32 -04001486 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001487 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001488
Al Viro705962c2007-05-13 05:52:32 -04001489 mutex_lock(&loop_devices_mutex);
1490 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001491 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001492 mutex_unlock(&loop_devices_mutex);
1493
1494 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001495 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001496}
1497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498static int __init loop_init(void)
1499{
Ken Chena47653f2007-06-08 13:46:44 -07001500 int i, nr;
1501 unsigned long range;
1502 struct loop_device *lo, *next;
1503
1504 /*
1505 * loop module now has a feature to instantiate underlying device
1506 * structure on-demand, provided that there is an access dev node.
1507 * However, this will not work well with user space tool that doesn't
1508 * know about such "feature". In order to not break any existing
1509 * tool, we do the following:
1510 *
1511 * (1) if max_loop is specified, create that many upfront, and this
1512 * also becomes a hard limit.
1513 * (2) if max_loop is not specified, create 8 loop device on module
1514 * load, user can further extend loop device by create dev node
1515 * themselves and have kernel automatically instantiate actual
1516 * device on-demand.
1517 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001518
1519 part_shift = 0;
1520 if (max_part > 0)
1521 part_shift = fls(max_part);
1522
1523 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001524 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Ken Chen73285082007-05-08 00:28:20 -07001526 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001527 nr = max_loop;
1528 range = max_loop;
1529 } else {
1530 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001531 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
Ken Chena47653f2007-06-08 13:46:44 -07001533
1534 if (register_blkdev(LOOP_MAJOR, "loop"))
1535 return -EIO;
1536
1537 for (i = 0; i < nr; i++) {
1538 lo = loop_alloc(i);
1539 if (!lo)
1540 goto Enomem;
1541 list_add_tail(&lo->lo_list, &loop_devices);
1542 }
1543
1544 /* point of no return */
1545
1546 list_for_each_entry(lo, &loop_devices, lo_list)
1547 add_disk(lo->lo_disk);
1548
1549 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1550 THIS_MODULE, loop_probe, NULL, NULL);
1551
Ken Chen73285082007-05-08 00:28:20 -07001552 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001554
1555Enomem:
1556 printk(KERN_INFO "loop: out of memory\n");
1557
1558 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1559 loop_free(lo);
1560
1561 unregister_blkdev(LOOP_MAJOR, "loop");
1562 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
Ken Chen73285082007-05-08 00:28:20 -07001565static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Ken Chena47653f2007-06-08 13:46:44 -07001567 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001568 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Laurent Vivier476a4812008-03-26 12:11:53 +01001570 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001571
Ken Chen73285082007-05-08 00:28:20 -07001572 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1573 loop_del_one(lo);
1574
Ken Chena47653f2007-06-08 13:46:44 -07001575 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001576 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577}
1578
1579module_init(loop_init);
1580module_exit(loop_exit);
1581
1582#ifndef MODULE
1583static int __init max_loop_setup(char *str)
1584{
1585 max_loop = simple_strtol(str, NULL, 0);
1586 return 1;
1587}
1588
1589__setup("max_loop=", max_loop_setup);
1590#endif