blob: 4503290da4078c2916bafdc8af301fac333f5349 [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.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
33 *
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>
71#include <linux/writeback.h>
72#include <linux/buffer_head.h> /* for invalidate_bdev() */
73#include <linux/completion.h>
74#include <linux/highmem.h>
75#include <linux/gfp.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070076#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020077#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79#include <asm/uaccess.h>
80
Ken Chen73285082007-05-08 00:28:20 -070081static LIST_HEAD(loop_devices);
82static DEFINE_MUTEX(loop_devices_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84/*
85 * Transfer functions
86 */
87static int transfer_none(struct loop_device *lo, int cmd,
88 struct page *raw_page, unsigned raw_off,
89 struct page *loop_page, unsigned loop_off,
90 int size, sector_t real_block)
91{
92 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
93 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
94
95 if (cmd == READ)
96 memcpy(loop_buf, raw_buf, size);
97 else
98 memcpy(raw_buf, loop_buf, size);
99
100 kunmap_atomic(raw_buf, KM_USER0);
101 kunmap_atomic(loop_buf, KM_USER1);
102 cond_resched();
103 return 0;
104}
105
106static int transfer_xor(struct loop_device *lo, int cmd,
107 struct page *raw_page, unsigned raw_off,
108 struct page *loop_page, unsigned loop_off,
109 int size, sector_t real_block)
110{
111 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
112 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
113 char *in, *out, *key;
114 int i, keysize;
115
116 if (cmd == READ) {
117 in = raw_buf;
118 out = loop_buf;
119 } else {
120 in = loop_buf;
121 out = raw_buf;
122 }
123
124 key = lo->lo_encrypt_key;
125 keysize = lo->lo_encrypt_key_size;
126 for (i = 0; i < size; i++)
127 *out++ = *in++ ^ key[(i & 511) % keysize];
128
129 kunmap_atomic(raw_buf, KM_USER0);
130 kunmap_atomic(loop_buf, KM_USER1);
131 cond_resched();
132 return 0;
133}
134
135static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
136{
137 if (unlikely(info->lo_encrypt_key_size <= 0))
138 return -EINVAL;
139 return 0;
140}
141
142static struct loop_func_table none_funcs = {
143 .number = LO_CRYPT_NONE,
144 .transfer = transfer_none,
145};
146
147static struct loop_func_table xor_funcs = {
148 .number = LO_CRYPT_XOR,
149 .transfer = transfer_xor,
150 .init = xor_init
151};
152
153/* xfer_funcs[0] is special - its release function is never called */
154static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
155 &none_funcs,
156 &xor_funcs
157};
158
159static loff_t get_loop_size(struct loop_device *lo, struct file *file)
160{
161 loff_t size, offset, loopsize;
162
163 /* Compute loopsize in bytes */
164 size = i_size_read(file->f_mapping->host);
165 offset = lo->lo_offset;
166 loopsize = size - offset;
167 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
168 loopsize = lo->lo_sizelimit;
169
170 /*
171 * Unfortunately, if we want to do I/O on the device,
172 * the number of 512-byte sectors has to fit into a sector_t.
173 */
174 return loopsize >> 9;
175}
176
177static int
178figure_loop_size(struct loop_device *lo)
179{
180 loff_t size = get_loop_size(lo, lo->lo_backing_file);
181 sector_t x = (sector_t)size;
182
183 if (unlikely((loff_t)x != size))
184 return -EFBIG;
185
Ken Chen73285082007-05-08 00:28:20 -0700186 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 return 0;
188}
189
190static inline int
191lo_do_transfer(struct loop_device *lo, int cmd,
192 struct page *rpage, unsigned roffs,
193 struct page *lpage, unsigned loffs,
194 int size, sector_t rblock)
195{
196 if (unlikely(!lo->transfer))
197 return 0;
198
199 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
200}
201
202/**
203 * do_lo_send_aops - helper for writing data to a loop device
204 *
205 * This is the fast version for backing filesystems which implement the address
206 * space operations prepare_write and commit_write.
207 */
208static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
209 int bsize, loff_t pos, struct page *page)
210{
211 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
212 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700213 const struct address_space_operations *aops = mapping->a_ops;
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;
225 unsigned size;
226 int transfer_result;
227
228 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
229 size = PAGE_CACHE_SIZE - offset;
230 if (size > len)
231 size = len;
232 page = grab_cache_page(mapping, index);
233 if (unlikely(!page))
234 goto fail;
Zach Brown994fc28c2005-12-15 14:28:17 -0800235 ret = aops->prepare_write(file, page, offset,
236 offset + size);
237 if (unlikely(ret)) {
238 if (ret == AOP_TRUNCATED_PAGE) {
239 page_cache_release(page);
240 continue;
241 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 goto unlock;
Zach Brown994fc28c2005-12-15 14:28:17 -0800243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
245 bvec->bv_page, bv_offs, size, IV);
246 if (unlikely(transfer_result)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 /*
248 * The transfer failed, but we still write the data to
249 * keep prepare/commit calls balanced.
250 */
251 printk(KERN_ERR "loop: transfer error block %llu\n",
252 (unsigned long long)index);
Nate Diller01f27052007-05-09 02:35:07 -0700253 zero_user_page(page, offset, size, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
255 flush_dcache_page(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800256 ret = aops->commit_write(file, page, offset,
257 offset + size);
258 if (unlikely(ret)) {
259 if (ret == AOP_TRUNCATED_PAGE) {
260 page_cache_release(page);
261 continue;
262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 goto unlock;
Zach Brown994fc28c2005-12-15 14:28:17 -0800264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (unlikely(transfer_result))
266 goto unlock;
267 bv_offs += size;
268 len -= size;
269 offset = 0;
270 index++;
271 pos += size;
272 unlock_page(page);
273 page_cache_release(page);
274 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800275 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800277 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return ret;
279unlock:
280 unlock_page(page);
281 page_cache_release(page);
282fail:
283 ret = -1;
284 goto out;
285}
286
287/**
288 * __do_lo_send_write - helper for writing data to a loop device
289 *
290 * This helper just factors out common code between do_lo_send_direct_write()
291 * and do_lo_send_write().
292 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800293static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100294 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
296 ssize_t bw;
297 mm_segment_t old_fs = get_fs();
298
299 set_fs(get_ds());
300 bw = file->f_op->write(file, buf, len, &pos);
301 set_fs(old_fs);
302 if (likely(bw == len))
303 return 0;
304 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
305 (unsigned long long)pos, len);
306 if (bw >= 0)
307 bw = -EIO;
308 return bw;
309}
310
311/**
312 * do_lo_send_direct_write - helper for writing data to a loop device
313 *
314 * This is the fast, non-transforming version for backing filesystems which do
315 * not implement the address space operations prepare_write and commit_write.
316 * It uses the write file operation which should be present on all writeable
317 * filesystems.
318 */
319static int do_lo_send_direct_write(struct loop_device *lo,
320 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
321{
322 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100323 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 bvec->bv_len, pos);
325 kunmap(bvec->bv_page);
326 cond_resched();
327 return bw;
328}
329
330/**
331 * do_lo_send_write - helper for writing data to a loop device
332 *
333 * This is the slow, transforming version for filesystems which do not
334 * implement the address space operations prepare_write and commit_write. It
335 * uses the write file operation which should be present on all writeable
336 * filesystems.
337 *
338 * Using fops->write is slower than using aops->{prepare,commit}_write in the
339 * transforming case because we need to double buffer the data as we cannot do
340 * the transformations in place as we do not have direct access to the
341 * destination pages of the backing file.
342 */
343static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
344 int bsize, loff_t pos, struct page *page)
345{
346 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
347 bvec->bv_offset, bvec->bv_len, pos >> 9);
348 if (likely(!ret))
349 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100350 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 pos);
352 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
353 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
354 if (ret > 0)
355 ret = -EIO;
356 return ret;
357}
358
359static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
360 loff_t pos)
361{
362 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
363 struct page *page);
364 struct bio_vec *bvec;
365 struct page *page = NULL;
366 int i, ret = 0;
367
368 do_lo_send = do_lo_send_aops;
369 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
370 do_lo_send = do_lo_send_direct_write;
371 if (lo->transfer != transfer_none) {
372 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
373 if (unlikely(!page))
374 goto fail;
375 kmap(page);
376 do_lo_send = do_lo_send_write;
377 }
378 }
379 bio_for_each_segment(bvec, bio, i) {
380 ret = do_lo_send(lo, bvec, bsize, pos, page);
381 if (ret < 0)
382 break;
383 pos += bvec->bv_len;
384 }
385 if (page) {
386 kunmap(page);
387 __free_page(page);
388 }
389out:
390 return ret;
391fail:
392 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
393 ret = -ENOMEM;
394 goto out;
395}
396
397struct lo_read_data {
398 struct loop_device *lo;
399 struct page *page;
400 unsigned offset;
401 int bsize;
402};
403
404static int
Jens Axboefd582142007-06-12 21:20:37 +0200405lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
406 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Jens Axboefd582142007-06-12 21:20:37 +0200408 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200410 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 sector_t IV;
Jens Axboefd582142007-06-12 21:20:37 +0200412 size_t size;
413 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Jens Axboecac36bb02007-06-14 13:10:48 +0200415 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200416 if (unlikely(ret))
417 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Jens Axboefd582142007-06-12 21:20:37 +0200419 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
420 (buf->offset >> 9);
421 size = sd->len;
422 if (size > p->bsize)
423 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Jens Axboefd582142007-06-12 21:20:37 +0200425 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 printk(KERN_ERR "loop: transfer error block %ld\n",
427 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200428 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
430
431 flush_dcache_page(p->page);
432
Jens Axboefd582142007-06-12 21:20:37 +0200433 if (size > 0)
434 p->offset += size;
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return size;
437}
438
439static int
Jens Axboefd582142007-06-12 21:20:37 +0200440lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
441{
442 return __splice_from_pipe(pipe, sd, lo_splice_actor);
443}
444
445static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446do_lo_receive(struct loop_device *lo,
447 struct bio_vec *bvec, int bsize, loff_t pos)
448{
449 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200450 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200452 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 cookie.lo = lo;
455 cookie.page = bvec->bv_page;
456 cookie.offset = bvec->bv_offset;
457 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200458
459 sd.len = 0;
460 sd.total_len = bvec->bv_len;
461 sd.flags = 0;
462 sd.pos = pos;
463 sd.u.data = &cookie;
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200466 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
467
468 if (retval < 0)
469 return retval;
470
471 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472}
473
474static int
475lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
476{
477 struct bio_vec *bvec;
478 int i, ret = 0;
479
480 bio_for_each_segment(bvec, bio, i) {
481 ret = do_lo_receive(lo, bvec, bsize, pos);
482 if (ret < 0)
483 break;
484 pos += bvec->bv_len;
485 }
486 return ret;
487}
488
489static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
490{
491 loff_t pos;
492 int ret;
493
494 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
495 if (bio_rw(bio) == WRITE)
496 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
497 else
498 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
499 return ret;
500}
501
502/*
503 * Add bio to back of pending list
504 */
505static void loop_add_bio(struct loop_device *lo, struct bio *bio)
506{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (lo->lo_biotail) {
508 lo->lo_biotail->bi_next = bio;
509 lo->lo_biotail = bio;
510 } else
511 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
513
514/*
515 * Grab first pending buffer
516 */
517static struct bio *loop_get_bio(struct loop_device *lo)
518{
519 struct bio *bio;
520
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if ((bio = lo->lo_bio)) {
522 if (bio == lo->lo_biotail)
523 lo->lo_biotail = NULL;
524 lo->lo_bio = bio->bi_next;
525 bio->bi_next = NULL;
526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 return bio;
529}
530
531static int loop_make_request(request_queue_t *q, struct bio *old_bio)
532{
533 struct loop_device *lo = q->queuedata;
534 int rw = bio_rw(old_bio);
535
Nick Piggin35a82d12005-06-23 00:09:06 -0700536 if (rw == READA)
537 rw = READ;
538
539 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
541 spin_lock_irq(&lo->lo_lock);
542 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700543 goto out;
544 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
545 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700547 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700548 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700552 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 bio_io_error(old_bio, old_bio->bi_size);
554 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
556
557/*
558 * kick off io on the underlying address space
559 */
560static void loop_unplug(request_queue_t *q)
561{
562 struct loop_device *lo = q->queuedata;
563
564 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
565 blk_run_address_space(lo->lo_backing_file->f_mapping);
566}
567
568struct switch_request {
569 struct file *file;
570 struct completion wait;
571};
572
573static void do_loop_switch(struct loop_device *, struct switch_request *);
574
575static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
576{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 if (unlikely(!bio->bi_bdev)) {
578 do_loop_switch(lo, bio->bi_private);
579 bio_put(bio);
580 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700581 int ret = do_bio_filebacked(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 bio_endio(bio, bio->bi_size, ret);
583 }
584}
585
586/*
587 * worker thread that handles reads/writes to file backed loop devices,
588 * to avoid blocking in our make_request_fn. it also does loop decrypting
589 * on reads for block backed loop, as that is too heavy to do from
590 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700591 *
592 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
593 * calling kthread_stop(). Therefore once kthread_should_stop() is
594 * true, make_request will not place any more requests. Therefore
595 * once kthread_should_stop() is true and lo_bio is NULL, we are
596 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
598static int loop_thread(void *data)
599{
600 struct loop_device *lo = data;
601 struct bio *bio;
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 /*
604 * loop can be used in an encrypted device,
605 * hence, it mustn't be stopped at all
606 * because it could be indirectly used during suspension
607 */
608 current->flags |= PF_NOFREEZE;
609
610 set_user_nice(current, -20);
611
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700612 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700614 wait_event_interruptible(lo->lo_event,
615 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700616
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700617 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700618 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700619 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700621 spin_unlock_irq(&lo->lo_lock);
622
623 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return 0;
628}
629
630/*
631 * loop_switch performs the hard work of switching a backing store.
632 * First it needs to flush existing IO, it does this by sending a magic
633 * BIO down the pipe. The completion of this BIO does the actual switch.
634 */
635static int loop_switch(struct loop_device *lo, struct file *file)
636{
637 struct switch_request w;
638 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
639 if (!bio)
640 return -ENOMEM;
641 init_completion(&w.wait);
642 w.file = file;
643 bio->bi_private = &w;
644 bio->bi_bdev = NULL;
645 loop_make_request(lo->lo_queue, bio);
646 wait_for_completion(&w.wait);
647 return 0;
648}
649
650/*
651 * 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;
657 struct address_space *mapping = file->f_mapping;
658
659 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
660 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700661 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
662 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 lo->old_gfp_mask = mapping_gfp_mask(mapping);
664 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
665 complete(&p->wait);
666}
667
668
669/*
670 * loop_change_fd switched the backing store of a loopback device to
671 * a new file. This is useful for operating system installers to free up
672 * the original file and in High Availability environments to switch to
673 * an alternative location for the content in case of server meltdown.
674 * This can only work if the loop device is used read-only, and if the
675 * new backing store is the same size and type as the old backing store.
676 */
677static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
678 struct block_device *bdev, unsigned int arg)
679{
680 struct file *file, *old_file;
681 struct inode *inode;
682 int error;
683
684 error = -ENXIO;
685 if (lo->lo_state != Lo_bound)
686 goto out;
687
688 /* the loop device has to be read-only */
689 error = -EINVAL;
690 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
691 goto out;
692
693 error = -EBADF;
694 file = fget(arg);
695 if (!file)
696 goto out;
697
698 inode = file->f_mapping->host;
699 old_file = lo->lo_backing_file;
700
701 error = -EINVAL;
702
703 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
704 goto out_putf;
705
Jens Axboefd582142007-06-12 21:20:37 +0200706 /* new backing store needs to support loop (eg splice_read) */
707 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 goto out_putf;
709
710 /* size of the new backing store needs to be the same */
711 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
712 goto out_putf;
713
714 /* and ... switch */
715 error = loop_switch(lo, file);
716 if (error)
717 goto out_putf;
718
719 fput(old_file);
720 return 0;
721
722 out_putf:
723 fput(file);
724 out:
725 return error;
726}
727
728static inline int is_loop_device(struct file *file)
729{
730 struct inode *i = file->f_mapping->host;
731
732 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
733}
734
735static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
736 struct block_device *bdev, unsigned int arg)
737{
738 struct file *file, *f;
739 struct inode *inode;
740 struct address_space *mapping;
741 unsigned lo_blocksize;
742 int lo_flags = 0;
743 int error;
744 loff_t size;
745
746 /* This is safe, since we have a reference from open(). */
747 __module_get(THIS_MODULE);
748
749 error = -EBADF;
750 file = fget(arg);
751 if (!file)
752 goto out;
753
754 error = -EBUSY;
755 if (lo->lo_state != Lo_unbound)
756 goto out_putf;
757
758 /* Avoid recursion */
759 f = file;
760 while (is_loop_device(f)) {
761 struct loop_device *l;
762
763 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
764 goto out_putf;
765
766 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
767 if (l->lo_state == Lo_unbound) {
768 error = -EINVAL;
769 goto out_putf;
770 }
771 f = l->lo_backing_file;
772 }
773
774 mapping = file->f_mapping;
775 inode = mapping->host;
776
777 if (!(file->f_mode & FMODE_WRITE))
778 lo_flags |= LO_FLAGS_READ_ONLY;
779
780 error = -EINVAL;
781 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700782 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 /*
784 * If we can't read - sorry. If we only can't write - well,
785 * it's going to be read-only.
786 */
Jens Axboefd582142007-06-12 21:20:37 +0200787 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 goto out_putf;
789 if (aops->prepare_write && aops->commit_write)
790 lo_flags |= LO_FLAGS_USE_AOPS;
791 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
792 lo_flags |= LO_FLAGS_READ_ONLY;
793
Theodore Ts'oba52de12006-09-27 01:50:49 -0700794 lo_blocksize = S_ISBLK(inode->i_mode) ?
795 inode->i_bdev->bd_block_size : PAGE_SIZE;
796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 error = 0;
798 } else {
799 goto out_putf;
800 }
801
802 size = get_loop_size(lo, file);
803
804 if ((loff_t)(sector_t)size != size) {
805 error = -EFBIG;
806 goto out_putf;
807 }
808
809 if (!(lo_file->f_mode & FMODE_WRITE))
810 lo_flags |= LO_FLAGS_READ_ONLY;
811
812 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
813
814 lo->lo_blocksize = lo_blocksize;
815 lo->lo_device = bdev;
816 lo->lo_flags = lo_flags;
817 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700818 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 lo->ioctl = NULL;
820 lo->lo_sizelimit = 0;
821 lo->old_gfp_mask = mapping_gfp_mask(mapping);
822 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
823
824 lo->lo_bio = lo->lo_biotail = NULL;
825
826 /*
827 * set queue make_request_fn, and add limits based on lower level
828 * device
829 */
830 blk_queue_make_request(lo->lo_queue, loop_make_request);
831 lo->lo_queue->queuedata = lo;
832 lo->lo_queue->unplug_fn = loop_unplug;
833
Ken Chen73285082007-05-08 00:28:20 -0700834 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 bd_set_size(bdev, size << 9);
836
837 set_blocksize(bdev, lo_blocksize);
838
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700839 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
840 lo->lo_number);
841 if (IS_ERR(lo->lo_thread)) {
842 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700843 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700844 }
845 lo->lo_state = Lo_bound;
846 wake_up_process(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 0;
848
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700849out_clr:
850 lo->lo_thread = NULL;
851 lo->lo_device = NULL;
852 lo->lo_backing_file = NULL;
853 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700854 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700855 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700856 bd_set_size(bdev, 0);
857 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
858 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 out_putf:
860 fput(file);
861 out:
862 /* This is safe: open() is still holding a reference. */
863 module_put(THIS_MODULE);
864 return error;
865}
866
867static int
868loop_release_xfer(struct loop_device *lo)
869{
870 int err = 0;
871 struct loop_func_table *xfer = lo->lo_encryption;
872
873 if (xfer) {
874 if (xfer->release)
875 err = xfer->release(lo);
876 lo->transfer = NULL;
877 lo->lo_encryption = NULL;
878 module_put(xfer->owner);
879 }
880 return err;
881}
882
883static int
884loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
885 const struct loop_info64 *i)
886{
887 int err = 0;
888
889 if (xfer) {
890 struct module *owner = xfer->owner;
891
892 if (!try_module_get(owner))
893 return -EINVAL;
894 if (xfer->init)
895 err = xfer->init(lo, i);
896 if (err)
897 module_put(owner);
898 else
899 lo->lo_encryption = xfer;
900 }
901 return err;
902}
903
904static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
905{
906 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400907 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if (lo->lo_state != Lo_bound)
910 return -ENXIO;
911
912 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
913 return -EBUSY;
914
915 if (filp == NULL)
916 return -EINVAL;
917
918 spin_lock_irq(&lo->lo_lock);
919 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 spin_unlock_irq(&lo->lo_lock);
921
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700922 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924 lo->lo_backing_file = NULL;
925
926 loop_release_xfer(lo);
927 lo->transfer = NULL;
928 lo->ioctl = NULL;
929 lo->lo_device = NULL;
930 lo->lo_encryption = NULL;
931 lo->lo_offset = 0;
932 lo->lo_sizelimit = 0;
933 lo->lo_encrypt_key_size = 0;
934 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700935 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
937 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
938 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700939 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700940 set_capacity(lo->lo_disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 bd_set_size(bdev, 0);
942 mapping_set_gfp_mask(filp->f_mapping, gfp);
943 lo->lo_state = Lo_unbound;
944 fput(filp);
945 /* This is safe: open() is still holding a reference. */
946 module_put(THIS_MODULE);
947 return 0;
948}
949
950static int
951loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
952{
953 int err;
954 struct loop_func_table *xfer;
955
956 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
957 !capable(CAP_SYS_ADMIN))
958 return -EPERM;
959 if (lo->lo_state != Lo_bound)
960 return -ENXIO;
961 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
962 return -EINVAL;
963
964 err = loop_release_xfer(lo);
965 if (err)
966 return err;
967
968 if (info->lo_encrypt_type) {
969 unsigned int type = info->lo_encrypt_type;
970
971 if (type >= MAX_LO_CRYPT)
972 return -EINVAL;
973 xfer = xfer_funcs[type];
974 if (xfer == NULL)
975 return -EINVAL;
976 } else
977 xfer = NULL;
978
979 err = loop_init_xfer(lo, xfer, info);
980 if (err)
981 return err;
982
983 if (lo->lo_offset != info->lo_offset ||
984 lo->lo_sizelimit != info->lo_sizelimit) {
985 lo->lo_offset = info->lo_offset;
986 lo->lo_sizelimit = info->lo_sizelimit;
987 if (figure_loop_size(lo))
988 return -EFBIG;
989 }
990
991 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
992 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
993 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
994 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
995
996 if (!xfer)
997 xfer = &none_funcs;
998 lo->transfer = xfer->transfer;
999 lo->ioctl = xfer->ioctl;
1000
1001 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1002 lo->lo_init[0] = info->lo_init[0];
1003 lo->lo_init[1] = info->lo_init[1];
1004 if (info->lo_encrypt_key_size) {
1005 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1006 info->lo_encrypt_key_size);
1007 lo->lo_key_owner = current->uid;
1008 }
1009
1010 return 0;
1011}
1012
1013static int
1014loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1015{
1016 struct file *file = lo->lo_backing_file;
1017 struct kstat stat;
1018 int error;
1019
1020 if (lo->lo_state != Lo_bound)
1021 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001022 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (error)
1024 return error;
1025 memset(info, 0, sizeof(*info));
1026 info->lo_number = lo->lo_number;
1027 info->lo_device = huge_encode_dev(stat.dev);
1028 info->lo_inode = stat.ino;
1029 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1030 info->lo_offset = lo->lo_offset;
1031 info->lo_sizelimit = lo->lo_sizelimit;
1032 info->lo_flags = lo->lo_flags;
1033 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1034 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1035 info->lo_encrypt_type =
1036 lo->lo_encryption ? lo->lo_encryption->number : 0;
1037 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1038 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1039 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1040 lo->lo_encrypt_key_size);
1041 }
1042 return 0;
1043}
1044
1045static void
1046loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1047{
1048 memset(info64, 0, sizeof(*info64));
1049 info64->lo_number = info->lo_number;
1050 info64->lo_device = info->lo_device;
1051 info64->lo_inode = info->lo_inode;
1052 info64->lo_rdevice = info->lo_rdevice;
1053 info64->lo_offset = info->lo_offset;
1054 info64->lo_sizelimit = 0;
1055 info64->lo_encrypt_type = info->lo_encrypt_type;
1056 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1057 info64->lo_flags = info->lo_flags;
1058 info64->lo_init[0] = info->lo_init[0];
1059 info64->lo_init[1] = info->lo_init[1];
1060 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1061 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1062 else
1063 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1064 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1065}
1066
1067static int
1068loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1069{
1070 memset(info, 0, sizeof(*info));
1071 info->lo_number = info64->lo_number;
1072 info->lo_device = info64->lo_device;
1073 info->lo_inode = info64->lo_inode;
1074 info->lo_rdevice = info64->lo_rdevice;
1075 info->lo_offset = info64->lo_offset;
1076 info->lo_encrypt_type = info64->lo_encrypt_type;
1077 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1078 info->lo_flags = info64->lo_flags;
1079 info->lo_init[0] = info64->lo_init[0];
1080 info->lo_init[1] = info64->lo_init[1];
1081 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1082 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1083 else
1084 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1085 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1086
1087 /* error in case values were truncated */
1088 if (info->lo_device != info64->lo_device ||
1089 info->lo_rdevice != info64->lo_rdevice ||
1090 info->lo_inode != info64->lo_inode ||
1091 info->lo_offset != info64->lo_offset)
1092 return -EOVERFLOW;
1093
1094 return 0;
1095}
1096
1097static int
1098loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1099{
1100 struct loop_info info;
1101 struct loop_info64 info64;
1102
1103 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1104 return -EFAULT;
1105 loop_info64_from_old(&info, &info64);
1106 return loop_set_status(lo, &info64);
1107}
1108
1109static int
1110loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1111{
1112 struct loop_info64 info64;
1113
1114 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1115 return -EFAULT;
1116 return loop_set_status(lo, &info64);
1117}
1118
1119static int
1120loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1121 struct loop_info info;
1122 struct loop_info64 info64;
1123 int err = 0;
1124
1125 if (!arg)
1126 err = -EINVAL;
1127 if (!err)
1128 err = loop_get_status(lo, &info64);
1129 if (!err)
1130 err = loop_info64_to_old(&info64, &info);
1131 if (!err && copy_to_user(arg, &info, sizeof(info)))
1132 err = -EFAULT;
1133
1134 return err;
1135}
1136
1137static int
1138loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1139 struct loop_info64 info64;
1140 int err = 0;
1141
1142 if (!arg)
1143 err = -EINVAL;
1144 if (!err)
1145 err = loop_get_status(lo, &info64);
1146 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1147 err = -EFAULT;
1148
1149 return err;
1150}
1151
1152static int lo_ioctl(struct inode * inode, struct file * file,
1153 unsigned int cmd, unsigned long arg)
1154{
1155 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1156 int err;
1157
Ingo Molnarf85221d2006-03-23 03:00:38 -08001158 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 switch (cmd) {
1160 case LOOP_SET_FD:
1161 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1162 break;
1163 case LOOP_CHANGE_FD:
1164 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1165 break;
1166 case LOOP_CLR_FD:
1167 err = loop_clr_fd(lo, inode->i_bdev);
1168 break;
1169 case LOOP_SET_STATUS:
1170 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1171 break;
1172 case LOOP_GET_STATUS:
1173 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1174 break;
1175 case LOOP_SET_STATUS64:
1176 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1177 break;
1178 case LOOP_GET_STATUS64:
1179 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1180 break;
1181 default:
1182 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1183 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001184 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return err;
1186}
1187
David Howells863d5b822006-08-29 19:06:14 +01001188#ifdef CONFIG_COMPAT
1189struct compat_loop_info {
1190 compat_int_t lo_number; /* ioctl r/o */
1191 compat_dev_t lo_device; /* ioctl r/o */
1192 compat_ulong_t lo_inode; /* ioctl r/o */
1193 compat_dev_t lo_rdevice; /* ioctl r/o */
1194 compat_int_t lo_offset;
1195 compat_int_t lo_encrypt_type;
1196 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1197 compat_int_t lo_flags; /* ioctl r/o */
1198 char lo_name[LO_NAME_SIZE];
1199 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1200 compat_ulong_t lo_init[2];
1201 char reserved[4];
1202};
1203
1204/*
1205 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1206 * - noinlined to reduce stack space usage in main part of driver
1207 */
1208static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001209loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001210 struct loop_info64 *info64)
1211{
1212 struct compat_loop_info info;
1213
1214 if (copy_from_user(&info, arg, sizeof(info)))
1215 return -EFAULT;
1216
1217 memset(info64, 0, sizeof(*info64));
1218 info64->lo_number = info.lo_number;
1219 info64->lo_device = info.lo_device;
1220 info64->lo_inode = info.lo_inode;
1221 info64->lo_rdevice = info.lo_rdevice;
1222 info64->lo_offset = info.lo_offset;
1223 info64->lo_sizelimit = 0;
1224 info64->lo_encrypt_type = info.lo_encrypt_type;
1225 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1226 info64->lo_flags = info.lo_flags;
1227 info64->lo_init[0] = info.lo_init[0];
1228 info64->lo_init[1] = info.lo_init[1];
1229 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1230 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1231 else
1232 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1233 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1234 return 0;
1235}
1236
1237/*
1238 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1239 * - noinlined to reduce stack space usage in main part of driver
1240 */
1241static noinline int
1242loop_info64_to_compat(const struct loop_info64 *info64,
1243 struct compat_loop_info __user *arg)
1244{
1245 struct compat_loop_info info;
1246
1247 memset(&info, 0, sizeof(info));
1248 info.lo_number = info64->lo_number;
1249 info.lo_device = info64->lo_device;
1250 info.lo_inode = info64->lo_inode;
1251 info.lo_rdevice = info64->lo_rdevice;
1252 info.lo_offset = info64->lo_offset;
1253 info.lo_encrypt_type = info64->lo_encrypt_type;
1254 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1255 info.lo_flags = info64->lo_flags;
1256 info.lo_init[0] = info64->lo_init[0];
1257 info.lo_init[1] = info64->lo_init[1];
1258 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1259 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1260 else
1261 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1262 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1263
1264 /* error in case values were truncated */
1265 if (info.lo_device != info64->lo_device ||
1266 info.lo_rdevice != info64->lo_rdevice ||
1267 info.lo_inode != info64->lo_inode ||
1268 info.lo_offset != info64->lo_offset ||
1269 info.lo_init[0] != info64->lo_init[0] ||
1270 info.lo_init[1] != info64->lo_init[1])
1271 return -EOVERFLOW;
1272
1273 if (copy_to_user(arg, &info, sizeof(info)))
1274 return -EFAULT;
1275 return 0;
1276}
1277
1278static int
1279loop_set_status_compat(struct loop_device *lo,
1280 const struct compat_loop_info __user *arg)
1281{
1282 struct loop_info64 info64;
1283 int ret;
1284
1285 ret = loop_info64_from_compat(arg, &info64);
1286 if (ret < 0)
1287 return ret;
1288 return loop_set_status(lo, &info64);
1289}
1290
1291static int
1292loop_get_status_compat(struct loop_device *lo,
1293 struct compat_loop_info __user *arg)
1294{
1295 struct loop_info64 info64;
1296 int err = 0;
1297
1298 if (!arg)
1299 err = -EINVAL;
1300 if (!err)
1301 err = loop_get_status(lo, &info64);
1302 if (!err)
1303 err = loop_info64_to_compat(&info64, arg);
1304 return err;
1305}
1306
1307static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1308{
Josef Sipek6c648be2006-12-08 02:36:55 -08001309 struct inode *inode = file->f_path.dentry->d_inode;
David Howells863d5b822006-08-29 19:06:14 +01001310 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1311 int err;
1312
1313 lock_kernel();
1314 switch(cmd) {
1315 case LOOP_SET_STATUS:
1316 mutex_lock(&lo->lo_ctl_mutex);
1317 err = loop_set_status_compat(
1318 lo, (const struct compat_loop_info __user *) arg);
1319 mutex_unlock(&lo->lo_ctl_mutex);
1320 break;
1321 case LOOP_GET_STATUS:
1322 mutex_lock(&lo->lo_ctl_mutex);
1323 err = loop_get_status_compat(
1324 lo, (struct compat_loop_info __user *) arg);
1325 mutex_unlock(&lo->lo_ctl_mutex);
1326 break;
1327 case LOOP_CLR_FD:
1328 case LOOP_GET_STATUS64:
1329 case LOOP_SET_STATUS64:
1330 arg = (unsigned long) compat_ptr(arg);
1331 case LOOP_SET_FD:
1332 case LOOP_CHANGE_FD:
1333 err = lo_ioctl(inode, file, cmd, arg);
1334 break;
1335 default:
1336 err = -ENOIOCTLCMD;
1337 break;
1338 }
1339 unlock_kernel();
1340 return err;
1341}
1342#endif
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344static int lo_open(struct inode *inode, struct file *file)
1345{
1346 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1347
Ingo Molnarf85221d2006-03-23 03:00:38 -08001348 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001350 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 return 0;
1353}
1354
1355static int lo_release(struct inode *inode, struct file *file)
1356{
1357 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1358
Ingo Molnarf85221d2006-03-23 03:00:38 -08001359 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 --lo->lo_refcnt;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001361 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363 return 0;
1364}
1365
1366static struct block_device_operations lo_fops = {
1367 .owner = THIS_MODULE,
1368 .open = lo_open,
1369 .release = lo_release,
1370 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001371#ifdef CONFIG_COMPAT
1372 .compat_ioctl = lo_compat_ioctl,
1373#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374};
1375
1376/*
1377 * And now the modules code and kernel interface.
1378 */
Ken Chen73285082007-05-08 00:28:20 -07001379static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001381MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382MODULE_LICENSE("GPL");
1383MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1384
1385int loop_register_transfer(struct loop_func_table *funcs)
1386{
1387 unsigned int n = funcs->number;
1388
1389 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1390 return -EINVAL;
1391 xfer_funcs[n] = funcs;
1392 return 0;
1393}
1394
1395int loop_unregister_transfer(int number)
1396{
1397 unsigned int n = number;
1398 struct loop_device *lo;
1399 struct loop_func_table *xfer;
1400
1401 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1402 return -EINVAL;
1403
1404 xfer_funcs[n] = NULL;
1405
Ken Chen73285082007-05-08 00:28:20 -07001406 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001407 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 if (lo->lo_encryption == xfer)
1410 loop_release_xfer(lo);
1411
Ingo Molnarf85221d2006-03-23 03:00:38 -08001412 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 }
1414
1415 return 0;
1416}
1417
1418EXPORT_SYMBOL(loop_register_transfer);
1419EXPORT_SYMBOL(loop_unregister_transfer);
1420
Ken Chena47653f2007-06-08 13:46:44 -07001421static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001422{
1423 struct loop_device *lo;
1424 struct gendisk *disk;
1425
1426 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1427 if (!lo)
1428 goto out;
1429
1430 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1431 if (!lo->lo_queue)
1432 goto out_free_dev;
1433
1434 disk = lo->lo_disk = alloc_disk(1);
1435 if (!disk)
1436 goto out_free_queue;
1437
1438 mutex_init(&lo->lo_ctl_mutex);
1439 lo->lo_number = i;
1440 lo->lo_thread = NULL;
1441 init_waitqueue_head(&lo->lo_event);
1442 spin_lock_init(&lo->lo_lock);
1443 disk->major = LOOP_MAJOR;
1444 disk->first_minor = i;
1445 disk->fops = &lo_fops;
1446 disk->private_data = lo;
1447 disk->queue = lo->lo_queue;
1448 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001449 return lo;
1450
1451out_free_queue:
1452 blk_cleanup_queue(lo->lo_queue);
1453out_free_dev:
1454 kfree(lo);
1455out:
Al Viro07002e92007-05-12 16:23:15 -04001456 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001457}
1458
Ken Chena47653f2007-06-08 13:46:44 -07001459static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001460{
Ken Chen73285082007-05-08 00:28:20 -07001461 blk_cleanup_queue(lo->lo_queue);
1462 put_disk(lo->lo_disk);
1463 list_del(&lo->lo_list);
1464 kfree(lo);
1465}
1466
Ken Chena47653f2007-06-08 13:46:44 -07001467static struct loop_device *loop_init_one(int i)
1468{
1469 struct loop_device *lo;
1470
1471 list_for_each_entry(lo, &loop_devices, lo_list) {
1472 if (lo->lo_number == i)
1473 return lo;
1474 }
1475
1476 lo = loop_alloc(i);
1477 if (lo) {
1478 add_disk(lo->lo_disk);
1479 list_add_tail(&lo->lo_list, &loop_devices);
1480 }
1481 return lo;
1482}
1483
1484static void loop_del_one(struct loop_device *lo)
1485{
1486 del_gendisk(lo->lo_disk);
1487 loop_free(lo);
1488}
1489
Ken Chen73285082007-05-08 00:28:20 -07001490static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1491{
Al Viro705962c2007-05-13 05:52:32 -04001492 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001493 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001494
Al Viro705962c2007-05-13 05:52:32 -04001495 mutex_lock(&loop_devices_mutex);
1496 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001497 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001498 mutex_unlock(&loop_devices_mutex);
1499
1500 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001501 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001502}
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504static int __init loop_init(void)
1505{
Ken Chena47653f2007-06-08 13:46:44 -07001506 int i, nr;
1507 unsigned long range;
1508 struct loop_device *lo, *next;
1509
1510 /*
1511 * loop module now has a feature to instantiate underlying device
1512 * structure on-demand, provided that there is an access dev node.
1513 * However, this will not work well with user space tool that doesn't
1514 * know about such "feature". In order to not break any existing
1515 * tool, we do the following:
1516 *
1517 * (1) if max_loop is specified, create that many upfront, and this
1518 * also becomes a hard limit.
1519 * (2) if max_loop is not specified, create 8 loop device on module
1520 * load, user can further extend loop device by create dev node
1521 * themselves and have kernel automatically instantiate actual
1522 * device on-demand.
1523 */
1524 if (max_loop > 1UL << MINORBITS)
1525 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
Ken Chen73285082007-05-08 00:28:20 -07001527 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001528 nr = max_loop;
1529 range = max_loop;
1530 } else {
1531 nr = 8;
1532 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 }
Ken Chena47653f2007-06-08 13:46:44 -07001534
1535 if (register_blkdev(LOOP_MAJOR, "loop"))
1536 return -EIO;
1537
1538 for (i = 0; i < nr; i++) {
1539 lo = loop_alloc(i);
1540 if (!lo)
1541 goto Enomem;
1542 list_add_tail(&lo->lo_list, &loop_devices);
1543 }
1544
1545 /* point of no return */
1546
1547 list_for_each_entry(lo, &loop_devices, lo_list)
1548 add_disk(lo->lo_disk);
1549
1550 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1551 THIS_MODULE, loop_probe, NULL, NULL);
1552
Ken Chen73285082007-05-08 00:28:20 -07001553 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001555
1556Enomem:
1557 printk(KERN_INFO "loop: out of memory\n");
1558
1559 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1560 loop_free(lo);
1561
1562 unregister_blkdev(LOOP_MAJOR, "loop");
1563 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
Ken Chen73285082007-05-08 00:28:20 -07001566static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
Ken Chena47653f2007-06-08 13:46:44 -07001568 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001569 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Ken Chena47653f2007-06-08 13:46:44 -07001571 range = max_loop ? max_loop : 1UL << MINORBITS;
1572
Ken Chen73285082007-05-08 00:28:20 -07001573 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1574 loop_del_one(lo);
1575
Ken Chena47653f2007-06-08 13:46:44 -07001576 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1578 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
1580
1581module_init(loop_init);
1582module_exit(loop_exit);
1583
1584#ifndef MODULE
1585static int __init max_loop_setup(char *str)
1586{
1587 max_loop = simple_strtol(str, NULL, 0);
1588 return 1;
1589}
1590
1591__setup("max_loop=", max_loop_setup);
1592#endif