blob: beab6d2643cb0fb27419bed3aa16b87034d2de0f [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78#include <asm/uaccess.h>
79
80static int max_loop = 8;
81static struct loop_device *loop_dev;
82static struct gendisk **disks;
83
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
186 set_capacity(disks[lo->lo_number], x);
187 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)) {
247 char *kaddr;
248
249 /*
250 * The transfer failed, but we still write the data to
251 * keep prepare/commit calls balanced.
252 */
253 printk(KERN_ERR "loop: transfer error block %llu\n",
254 (unsigned long long)index);
255 kaddr = kmap_atomic(page, KM_USER0);
256 memset(kaddr + offset, 0, size);
257 kunmap_atomic(kaddr, KM_USER0);
258 }
259 flush_dcache_page(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800260 ret = aops->commit_write(file, page, offset,
261 offset + size);
262 if (unlikely(ret)) {
263 if (ret == AOP_TRUNCATED_PAGE) {
264 page_cache_release(page);
265 continue;
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 goto unlock;
Zach Brown994fc28c2005-12-15 14:28:17 -0800268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 if (unlikely(transfer_result))
270 goto unlock;
271 bv_offs += size;
272 len -= size;
273 offset = 0;
274 index++;
275 pos += size;
276 unlock_page(page);
277 page_cache_release(page);
278 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800279 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800281 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 return ret;
283unlock:
284 unlock_page(page);
285 page_cache_release(page);
286fail:
287 ret = -1;
288 goto out;
289}
290
291/**
292 * __do_lo_send_write - helper for writing data to a loop device
293 *
294 * This helper just factors out common code between do_lo_send_direct_write()
295 * and do_lo_send_write().
296 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800297static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100298 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 ssize_t bw;
301 mm_segment_t old_fs = get_fs();
302
303 set_fs(get_ds());
304 bw = file->f_op->write(file, buf, len, &pos);
305 set_fs(old_fs);
306 if (likely(bw == len))
307 return 0;
308 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
309 (unsigned long long)pos, len);
310 if (bw >= 0)
311 bw = -EIO;
312 return bw;
313}
314
315/**
316 * do_lo_send_direct_write - helper for writing data to a loop device
317 *
318 * This is the fast, non-transforming version for backing filesystems which do
319 * not implement the address space operations prepare_write and commit_write.
320 * It uses the write file operation which should be present on all writeable
321 * filesystems.
322 */
323static int do_lo_send_direct_write(struct loop_device *lo,
324 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
325{
326 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100327 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 bvec->bv_len, pos);
329 kunmap(bvec->bv_page);
330 cond_resched();
331 return bw;
332}
333
334/**
335 * do_lo_send_write - helper for writing data to a loop device
336 *
337 * This is the slow, transforming version for filesystems which do not
338 * implement the address space operations prepare_write and commit_write. It
339 * uses the write file operation which should be present on all writeable
340 * filesystems.
341 *
342 * Using fops->write is slower than using aops->{prepare,commit}_write in the
343 * transforming case because we need to double buffer the data as we cannot do
344 * the transformations in place as we do not have direct access to the
345 * destination pages of the backing file.
346 */
347static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
348 int bsize, loff_t pos, struct page *page)
349{
350 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
351 bvec->bv_offset, bvec->bv_len, pos >> 9);
352 if (likely(!ret))
353 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100354 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 pos);
356 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
357 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
358 if (ret > 0)
359 ret = -EIO;
360 return ret;
361}
362
363static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
364 loff_t pos)
365{
366 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
367 struct page *page);
368 struct bio_vec *bvec;
369 struct page *page = NULL;
370 int i, ret = 0;
371
372 do_lo_send = do_lo_send_aops;
373 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
374 do_lo_send = do_lo_send_direct_write;
375 if (lo->transfer != transfer_none) {
376 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
377 if (unlikely(!page))
378 goto fail;
379 kmap(page);
380 do_lo_send = do_lo_send_write;
381 }
382 }
383 bio_for_each_segment(bvec, bio, i) {
384 ret = do_lo_send(lo, bvec, bsize, pos, page);
385 if (ret < 0)
386 break;
387 pos += bvec->bv_len;
388 }
389 if (page) {
390 kunmap(page);
391 __free_page(page);
392 }
393out:
394 return ret;
395fail:
396 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
397 ret = -ENOMEM;
398 goto out;
399}
400
401struct lo_read_data {
402 struct loop_device *lo;
403 struct page *page;
404 unsigned offset;
405 int bsize;
406};
407
408static int
409lo_read_actor(read_descriptor_t *desc, struct page *page,
410 unsigned long offset, unsigned long size)
411{
412 unsigned long count = desc->count;
413 struct lo_read_data *p = desc->arg.data;
414 struct loop_device *lo = p->lo;
415 sector_t IV;
416
417 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
418
419 if (size > count)
420 size = count;
421
422 if (lo_do_transfer(lo, READ, page, offset, p->page, p->offset, size, IV)) {
423 size = 0;
424 printk(KERN_ERR "loop: transfer error block %ld\n",
425 page->index);
426 desc->error = -EINVAL;
427 }
428
429 flush_dcache_page(p->page);
430
431 desc->count = count - size;
432 desc->written += size;
433 p->offset += size;
434 return size;
435}
436
437static int
438do_lo_receive(struct loop_device *lo,
439 struct bio_vec *bvec, int bsize, loff_t pos)
440{
441 struct lo_read_data cookie;
442 struct file *file;
443 int retval;
444
445 cookie.lo = lo;
446 cookie.page = bvec->bv_page;
447 cookie.offset = bvec->bv_offset;
448 cookie.bsize = bsize;
449 file = lo->lo_backing_file;
450 retval = file->f_op->sendfile(file, &pos, bvec->bv_len,
451 lo_read_actor, &cookie);
452 return (retval < 0)? retval: 0;
453}
454
455static int
456lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
457{
458 struct bio_vec *bvec;
459 int i, ret = 0;
460
461 bio_for_each_segment(bvec, bio, i) {
462 ret = do_lo_receive(lo, bvec, bsize, pos);
463 if (ret < 0)
464 break;
465 pos += bvec->bv_len;
466 }
467 return ret;
468}
469
470static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
471{
472 loff_t pos;
473 int ret;
474
475 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
476 if (bio_rw(bio) == WRITE)
477 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
478 else
479 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
480 return ret;
481}
482
483/*
484 * Add bio to back of pending list
485 */
486static void loop_add_bio(struct loop_device *lo, struct bio *bio)
487{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 if (lo->lo_biotail) {
489 lo->lo_biotail->bi_next = bio;
490 lo->lo_biotail = bio;
491 } else
492 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
494
495/*
496 * Grab first pending buffer
497 */
498static struct bio *loop_get_bio(struct loop_device *lo)
499{
500 struct bio *bio;
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 if ((bio = lo->lo_bio)) {
503 if (bio == lo->lo_biotail)
504 lo->lo_biotail = NULL;
505 lo->lo_bio = bio->bi_next;
506 bio->bi_next = NULL;
507 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 return bio;
510}
511
512static int loop_make_request(request_queue_t *q, struct bio *old_bio)
513{
514 struct loop_device *lo = q->queuedata;
515 int rw = bio_rw(old_bio);
516
Nick Piggin35a82d12005-06-23 00:09:06 -0700517 if (rw == READA)
518 rw = READ;
519
520 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
522 spin_lock_irq(&lo->lo_lock);
523 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700524 goto out;
525 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
526 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700528 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700529 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700533 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 bio_io_error(old_bio, old_bio->bi_size);
535 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
538/*
539 * kick off io on the underlying address space
540 */
541static void loop_unplug(request_queue_t *q)
542{
543 struct loop_device *lo = q->queuedata;
544
545 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
546 blk_run_address_space(lo->lo_backing_file->f_mapping);
547}
548
549struct switch_request {
550 struct file *file;
551 struct completion wait;
552};
553
554static void do_loop_switch(struct loop_device *, struct switch_request *);
555
556static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
557{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (unlikely(!bio->bi_bdev)) {
559 do_loop_switch(lo, bio->bi_private);
560 bio_put(bio);
561 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700562 int ret = do_bio_filebacked(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 bio_endio(bio, bio->bi_size, ret);
564 }
565}
566
567/*
568 * worker thread that handles reads/writes to file backed loop devices,
569 * to avoid blocking in our make_request_fn. it also does loop decrypting
570 * on reads for block backed loop, as that is too heavy to do from
571 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700572 *
573 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
574 * calling kthread_stop(). Therefore once kthread_should_stop() is
575 * true, make_request will not place any more requests. Therefore
576 * once kthread_should_stop() is true and lo_bio is NULL, we are
577 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 */
579static int loop_thread(void *data)
580{
581 struct loop_device *lo = data;
582 struct bio *bio;
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 /*
585 * loop can be used in an encrypted device,
586 * hence, it mustn't be stopped at all
587 * because it could be indirectly used during suspension
588 */
589 current->flags |= PF_NOFREEZE;
590
591 set_user_nice(current, -20);
592
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700593 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700595 wait_event_interruptible(lo->lo_event,
596 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700597
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700598 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700599 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700600 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700602 spin_unlock_irq(&lo->lo_lock);
603
604 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return 0;
609}
610
611/*
612 * loop_switch performs the hard work of switching a backing store.
613 * First it needs to flush existing IO, it does this by sending a magic
614 * BIO down the pipe. The completion of this BIO does the actual switch.
615 */
616static int loop_switch(struct loop_device *lo, struct file *file)
617{
618 struct switch_request w;
619 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
620 if (!bio)
621 return -ENOMEM;
622 init_completion(&w.wait);
623 w.file = file;
624 bio->bi_private = &w;
625 bio->bi_bdev = NULL;
626 loop_make_request(lo->lo_queue, bio);
627 wait_for_completion(&w.wait);
628 return 0;
629}
630
631/*
632 * Do the actual switch; called from the BIO completion routine
633 */
634static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
635{
636 struct file *file = p->file;
637 struct file *old_file = lo->lo_backing_file;
638 struct address_space *mapping = file->f_mapping;
639
640 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
641 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700642 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
643 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 lo->old_gfp_mask = mapping_gfp_mask(mapping);
645 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
646 complete(&p->wait);
647}
648
649
650/*
651 * loop_change_fd switched the backing store of a loopback device to
652 * a new file. This is useful for operating system installers to free up
653 * the original file and in High Availability environments to switch to
654 * an alternative location for the content in case of server meltdown.
655 * This can only work if the loop device is used read-only, and if the
656 * new backing store is the same size and type as the old backing store.
657 */
658static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
659 struct block_device *bdev, unsigned int arg)
660{
661 struct file *file, *old_file;
662 struct inode *inode;
663 int error;
664
665 error = -ENXIO;
666 if (lo->lo_state != Lo_bound)
667 goto out;
668
669 /* the loop device has to be read-only */
670 error = -EINVAL;
671 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
672 goto out;
673
674 error = -EBADF;
675 file = fget(arg);
676 if (!file)
677 goto out;
678
679 inode = file->f_mapping->host;
680 old_file = lo->lo_backing_file;
681
682 error = -EINVAL;
683
684 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
685 goto out_putf;
686
687 /* new backing store needs to support loop (eg sendfile) */
688 if (!inode->i_fop->sendfile)
689 goto out_putf;
690
691 /* size of the new backing store needs to be the same */
692 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
693 goto out_putf;
694
695 /* and ... switch */
696 error = loop_switch(lo, file);
697 if (error)
698 goto out_putf;
699
700 fput(old_file);
701 return 0;
702
703 out_putf:
704 fput(file);
705 out:
706 return error;
707}
708
709static inline int is_loop_device(struct file *file)
710{
711 struct inode *i = file->f_mapping->host;
712
713 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
714}
715
716static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
717 struct block_device *bdev, unsigned int arg)
718{
719 struct file *file, *f;
720 struct inode *inode;
721 struct address_space *mapping;
722 unsigned lo_blocksize;
723 int lo_flags = 0;
724 int error;
725 loff_t size;
726
727 /* This is safe, since we have a reference from open(). */
728 __module_get(THIS_MODULE);
729
730 error = -EBADF;
731 file = fget(arg);
732 if (!file)
733 goto out;
734
735 error = -EBUSY;
736 if (lo->lo_state != Lo_unbound)
737 goto out_putf;
738
739 /* Avoid recursion */
740 f = file;
741 while (is_loop_device(f)) {
742 struct loop_device *l;
743
744 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
745 goto out_putf;
746
747 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
748 if (l->lo_state == Lo_unbound) {
749 error = -EINVAL;
750 goto out_putf;
751 }
752 f = l->lo_backing_file;
753 }
754
755 mapping = file->f_mapping;
756 inode = mapping->host;
757
758 if (!(file->f_mode & FMODE_WRITE))
759 lo_flags |= LO_FLAGS_READ_ONLY;
760
761 error = -EINVAL;
762 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700763 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 /*
765 * If we can't read - sorry. If we only can't write - well,
766 * it's going to be read-only.
767 */
768 if (!file->f_op->sendfile)
769 goto out_putf;
770 if (aops->prepare_write && aops->commit_write)
771 lo_flags |= LO_FLAGS_USE_AOPS;
772 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
773 lo_flags |= LO_FLAGS_READ_ONLY;
774
Theodore Ts'oba52de12006-09-27 01:50:49 -0700775 lo_blocksize = S_ISBLK(inode->i_mode) ?
776 inode->i_bdev->bd_block_size : PAGE_SIZE;
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 error = 0;
779 } else {
780 goto out_putf;
781 }
782
783 size = get_loop_size(lo, file);
784
785 if ((loff_t)(sector_t)size != size) {
786 error = -EFBIG;
787 goto out_putf;
788 }
789
790 if (!(lo_file->f_mode & FMODE_WRITE))
791 lo_flags |= LO_FLAGS_READ_ONLY;
792
793 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
794
795 lo->lo_blocksize = lo_blocksize;
796 lo->lo_device = bdev;
797 lo->lo_flags = lo_flags;
798 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700799 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 lo->ioctl = NULL;
801 lo->lo_sizelimit = 0;
802 lo->old_gfp_mask = mapping_gfp_mask(mapping);
803 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
804
805 lo->lo_bio = lo->lo_biotail = NULL;
806
807 /*
808 * set queue make_request_fn, and add limits based on lower level
809 * device
810 */
811 blk_queue_make_request(lo->lo_queue, loop_make_request);
812 lo->lo_queue->queuedata = lo;
813 lo->lo_queue->unplug_fn = loop_unplug;
814
815 set_capacity(disks[lo->lo_number], size);
816 bd_set_size(bdev, size << 9);
817
818 set_blocksize(bdev, lo_blocksize);
819
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700820 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
821 lo->lo_number);
822 if (IS_ERR(lo->lo_thread)) {
823 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700824 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700825 }
826 lo->lo_state = Lo_bound;
827 wake_up_process(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return 0;
829
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700830out_clr:
831 lo->lo_thread = NULL;
832 lo->lo_device = NULL;
833 lo->lo_backing_file = NULL;
834 lo->lo_flags = 0;
835 set_capacity(disks[lo->lo_number], 0);
836 invalidate_bdev(bdev, 0);
837 bd_set_size(bdev, 0);
838 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
839 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 out_putf:
841 fput(file);
842 out:
843 /* This is safe: open() is still holding a reference. */
844 module_put(THIS_MODULE);
845 return error;
846}
847
848static int
849loop_release_xfer(struct loop_device *lo)
850{
851 int err = 0;
852 struct loop_func_table *xfer = lo->lo_encryption;
853
854 if (xfer) {
855 if (xfer->release)
856 err = xfer->release(lo);
857 lo->transfer = NULL;
858 lo->lo_encryption = NULL;
859 module_put(xfer->owner);
860 }
861 return err;
862}
863
864static int
865loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
866 const struct loop_info64 *i)
867{
868 int err = 0;
869
870 if (xfer) {
871 struct module *owner = xfer->owner;
872
873 if (!try_module_get(owner))
874 return -EINVAL;
875 if (xfer->init)
876 err = xfer->init(lo, i);
877 if (err)
878 module_put(owner);
879 else
880 lo->lo_encryption = xfer;
881 }
882 return err;
883}
884
885static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
886{
887 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400888 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 if (lo->lo_state != Lo_bound)
891 return -ENXIO;
892
893 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
894 return -EBUSY;
895
896 if (filp == NULL)
897 return -EINVAL;
898
899 spin_lock_irq(&lo->lo_lock);
900 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 spin_unlock_irq(&lo->lo_lock);
902
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700903 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 lo->lo_backing_file = NULL;
906
907 loop_release_xfer(lo);
908 lo->transfer = NULL;
909 lo->ioctl = NULL;
910 lo->lo_device = NULL;
911 lo->lo_encryption = NULL;
912 lo->lo_offset = 0;
913 lo->lo_sizelimit = 0;
914 lo->lo_encrypt_key_size = 0;
915 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700916 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
918 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
919 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
920 invalidate_bdev(bdev, 0);
921 set_capacity(disks[lo->lo_number], 0);
922 bd_set_size(bdev, 0);
923 mapping_set_gfp_mask(filp->f_mapping, gfp);
924 lo->lo_state = Lo_unbound;
925 fput(filp);
926 /* This is safe: open() is still holding a reference. */
927 module_put(THIS_MODULE);
928 return 0;
929}
930
931static int
932loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
933{
934 int err;
935 struct loop_func_table *xfer;
936
937 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
938 !capable(CAP_SYS_ADMIN))
939 return -EPERM;
940 if (lo->lo_state != Lo_bound)
941 return -ENXIO;
942 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
943 return -EINVAL;
944
945 err = loop_release_xfer(lo);
946 if (err)
947 return err;
948
949 if (info->lo_encrypt_type) {
950 unsigned int type = info->lo_encrypt_type;
951
952 if (type >= MAX_LO_CRYPT)
953 return -EINVAL;
954 xfer = xfer_funcs[type];
955 if (xfer == NULL)
956 return -EINVAL;
957 } else
958 xfer = NULL;
959
960 err = loop_init_xfer(lo, xfer, info);
961 if (err)
962 return err;
963
964 if (lo->lo_offset != info->lo_offset ||
965 lo->lo_sizelimit != info->lo_sizelimit) {
966 lo->lo_offset = info->lo_offset;
967 lo->lo_sizelimit = info->lo_sizelimit;
968 if (figure_loop_size(lo))
969 return -EFBIG;
970 }
971
972 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
973 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
974 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
975 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
976
977 if (!xfer)
978 xfer = &none_funcs;
979 lo->transfer = xfer->transfer;
980 lo->ioctl = xfer->ioctl;
981
982 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
983 lo->lo_init[0] = info->lo_init[0];
984 lo->lo_init[1] = info->lo_init[1];
985 if (info->lo_encrypt_key_size) {
986 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
987 info->lo_encrypt_key_size);
988 lo->lo_key_owner = current->uid;
989 }
990
991 return 0;
992}
993
994static int
995loop_get_status(struct loop_device *lo, struct loop_info64 *info)
996{
997 struct file *file = lo->lo_backing_file;
998 struct kstat stat;
999 int error;
1000
1001 if (lo->lo_state != Lo_bound)
1002 return -ENXIO;
1003 error = vfs_getattr(file->f_vfsmnt, file->f_dentry, &stat);
1004 if (error)
1005 return error;
1006 memset(info, 0, sizeof(*info));
1007 info->lo_number = lo->lo_number;
1008 info->lo_device = huge_encode_dev(stat.dev);
1009 info->lo_inode = stat.ino;
1010 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1011 info->lo_offset = lo->lo_offset;
1012 info->lo_sizelimit = lo->lo_sizelimit;
1013 info->lo_flags = lo->lo_flags;
1014 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1015 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1016 info->lo_encrypt_type =
1017 lo->lo_encryption ? lo->lo_encryption->number : 0;
1018 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1019 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1020 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1021 lo->lo_encrypt_key_size);
1022 }
1023 return 0;
1024}
1025
1026static void
1027loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1028{
1029 memset(info64, 0, sizeof(*info64));
1030 info64->lo_number = info->lo_number;
1031 info64->lo_device = info->lo_device;
1032 info64->lo_inode = info->lo_inode;
1033 info64->lo_rdevice = info->lo_rdevice;
1034 info64->lo_offset = info->lo_offset;
1035 info64->lo_sizelimit = 0;
1036 info64->lo_encrypt_type = info->lo_encrypt_type;
1037 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1038 info64->lo_flags = info->lo_flags;
1039 info64->lo_init[0] = info->lo_init[0];
1040 info64->lo_init[1] = info->lo_init[1];
1041 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1042 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1043 else
1044 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1045 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1046}
1047
1048static int
1049loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1050{
1051 memset(info, 0, sizeof(*info));
1052 info->lo_number = info64->lo_number;
1053 info->lo_device = info64->lo_device;
1054 info->lo_inode = info64->lo_inode;
1055 info->lo_rdevice = info64->lo_rdevice;
1056 info->lo_offset = info64->lo_offset;
1057 info->lo_encrypt_type = info64->lo_encrypt_type;
1058 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1059 info->lo_flags = info64->lo_flags;
1060 info->lo_init[0] = info64->lo_init[0];
1061 info->lo_init[1] = info64->lo_init[1];
1062 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1063 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1064 else
1065 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1066 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1067
1068 /* error in case values were truncated */
1069 if (info->lo_device != info64->lo_device ||
1070 info->lo_rdevice != info64->lo_rdevice ||
1071 info->lo_inode != info64->lo_inode ||
1072 info->lo_offset != info64->lo_offset)
1073 return -EOVERFLOW;
1074
1075 return 0;
1076}
1077
1078static int
1079loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1080{
1081 struct loop_info info;
1082 struct loop_info64 info64;
1083
1084 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1085 return -EFAULT;
1086 loop_info64_from_old(&info, &info64);
1087 return loop_set_status(lo, &info64);
1088}
1089
1090static int
1091loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1092{
1093 struct loop_info64 info64;
1094
1095 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1096 return -EFAULT;
1097 return loop_set_status(lo, &info64);
1098}
1099
1100static int
1101loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1102 struct loop_info info;
1103 struct loop_info64 info64;
1104 int err = 0;
1105
1106 if (!arg)
1107 err = -EINVAL;
1108 if (!err)
1109 err = loop_get_status(lo, &info64);
1110 if (!err)
1111 err = loop_info64_to_old(&info64, &info);
1112 if (!err && copy_to_user(arg, &info, sizeof(info)))
1113 err = -EFAULT;
1114
1115 return err;
1116}
1117
1118static int
1119loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1120 struct loop_info64 info64;
1121 int err = 0;
1122
1123 if (!arg)
1124 err = -EINVAL;
1125 if (!err)
1126 err = loop_get_status(lo, &info64);
1127 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1128 err = -EFAULT;
1129
1130 return err;
1131}
1132
1133static int lo_ioctl(struct inode * inode, struct file * file,
1134 unsigned int cmd, unsigned long arg)
1135{
1136 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1137 int err;
1138
Ingo Molnarf85221d2006-03-23 03:00:38 -08001139 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 switch (cmd) {
1141 case LOOP_SET_FD:
1142 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1143 break;
1144 case LOOP_CHANGE_FD:
1145 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1146 break;
1147 case LOOP_CLR_FD:
1148 err = loop_clr_fd(lo, inode->i_bdev);
1149 break;
1150 case LOOP_SET_STATUS:
1151 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1152 break;
1153 case LOOP_GET_STATUS:
1154 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1155 break;
1156 case LOOP_SET_STATUS64:
1157 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1158 break;
1159 case LOOP_GET_STATUS64:
1160 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1161 break;
1162 default:
1163 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1164 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001165 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 return err;
1167}
1168
David Howells863d5b822006-08-29 19:06:14 +01001169#ifdef CONFIG_COMPAT
1170struct compat_loop_info {
1171 compat_int_t lo_number; /* ioctl r/o */
1172 compat_dev_t lo_device; /* ioctl r/o */
1173 compat_ulong_t lo_inode; /* ioctl r/o */
1174 compat_dev_t lo_rdevice; /* ioctl r/o */
1175 compat_int_t lo_offset;
1176 compat_int_t lo_encrypt_type;
1177 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1178 compat_int_t lo_flags; /* ioctl r/o */
1179 char lo_name[LO_NAME_SIZE];
1180 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1181 compat_ulong_t lo_init[2];
1182 char reserved[4];
1183};
1184
1185/*
1186 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1187 * - noinlined to reduce stack space usage in main part of driver
1188 */
1189static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001190loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001191 struct loop_info64 *info64)
1192{
1193 struct compat_loop_info info;
1194
1195 if (copy_from_user(&info, arg, sizeof(info)))
1196 return -EFAULT;
1197
1198 memset(info64, 0, sizeof(*info64));
1199 info64->lo_number = info.lo_number;
1200 info64->lo_device = info.lo_device;
1201 info64->lo_inode = info.lo_inode;
1202 info64->lo_rdevice = info.lo_rdevice;
1203 info64->lo_offset = info.lo_offset;
1204 info64->lo_sizelimit = 0;
1205 info64->lo_encrypt_type = info.lo_encrypt_type;
1206 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1207 info64->lo_flags = info.lo_flags;
1208 info64->lo_init[0] = info.lo_init[0];
1209 info64->lo_init[1] = info.lo_init[1];
1210 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1211 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1212 else
1213 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1214 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1215 return 0;
1216}
1217
1218/*
1219 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1220 * - noinlined to reduce stack space usage in main part of driver
1221 */
1222static noinline int
1223loop_info64_to_compat(const struct loop_info64 *info64,
1224 struct compat_loop_info __user *arg)
1225{
1226 struct compat_loop_info info;
1227
1228 memset(&info, 0, sizeof(info));
1229 info.lo_number = info64->lo_number;
1230 info.lo_device = info64->lo_device;
1231 info.lo_inode = info64->lo_inode;
1232 info.lo_rdevice = info64->lo_rdevice;
1233 info.lo_offset = info64->lo_offset;
1234 info.lo_encrypt_type = info64->lo_encrypt_type;
1235 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1236 info.lo_flags = info64->lo_flags;
1237 info.lo_init[0] = info64->lo_init[0];
1238 info.lo_init[1] = info64->lo_init[1];
1239 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1240 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1241 else
1242 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1243 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1244
1245 /* error in case values were truncated */
1246 if (info.lo_device != info64->lo_device ||
1247 info.lo_rdevice != info64->lo_rdevice ||
1248 info.lo_inode != info64->lo_inode ||
1249 info.lo_offset != info64->lo_offset ||
1250 info.lo_init[0] != info64->lo_init[0] ||
1251 info.lo_init[1] != info64->lo_init[1])
1252 return -EOVERFLOW;
1253
1254 if (copy_to_user(arg, &info, sizeof(info)))
1255 return -EFAULT;
1256 return 0;
1257}
1258
1259static int
1260loop_set_status_compat(struct loop_device *lo,
1261 const struct compat_loop_info __user *arg)
1262{
1263 struct loop_info64 info64;
1264 int ret;
1265
1266 ret = loop_info64_from_compat(arg, &info64);
1267 if (ret < 0)
1268 return ret;
1269 return loop_set_status(lo, &info64);
1270}
1271
1272static int
1273loop_get_status_compat(struct loop_device *lo,
1274 struct compat_loop_info __user *arg)
1275{
1276 struct loop_info64 info64;
1277 int err = 0;
1278
1279 if (!arg)
1280 err = -EINVAL;
1281 if (!err)
1282 err = loop_get_status(lo, &info64);
1283 if (!err)
1284 err = loop_info64_to_compat(&info64, arg);
1285 return err;
1286}
1287
1288static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1289{
1290 struct inode *inode = file->f_dentry->d_inode;
1291 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1292 int err;
1293
1294 lock_kernel();
1295 switch(cmd) {
1296 case LOOP_SET_STATUS:
1297 mutex_lock(&lo->lo_ctl_mutex);
1298 err = loop_set_status_compat(
1299 lo, (const struct compat_loop_info __user *) arg);
1300 mutex_unlock(&lo->lo_ctl_mutex);
1301 break;
1302 case LOOP_GET_STATUS:
1303 mutex_lock(&lo->lo_ctl_mutex);
1304 err = loop_get_status_compat(
1305 lo, (struct compat_loop_info __user *) arg);
1306 mutex_unlock(&lo->lo_ctl_mutex);
1307 break;
1308 case LOOP_CLR_FD:
1309 case LOOP_GET_STATUS64:
1310 case LOOP_SET_STATUS64:
1311 arg = (unsigned long) compat_ptr(arg);
1312 case LOOP_SET_FD:
1313 case LOOP_CHANGE_FD:
1314 err = lo_ioctl(inode, file, cmd, arg);
1315 break;
1316 default:
1317 err = -ENOIOCTLCMD;
1318 break;
1319 }
1320 unlock_kernel();
1321 return err;
1322}
1323#endif
1324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325static int lo_open(struct inode *inode, struct file *file)
1326{
1327 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1328
Ingo Molnarf85221d2006-03-23 03:00:38 -08001329 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001331 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 return 0;
1334}
1335
1336static int lo_release(struct inode *inode, struct file *file)
1337{
1338 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1339
Ingo Molnarf85221d2006-03-23 03:00:38 -08001340 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 --lo->lo_refcnt;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001342 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
1344 return 0;
1345}
1346
1347static struct block_device_operations lo_fops = {
1348 .owner = THIS_MODULE,
1349 .open = lo_open,
1350 .release = lo_release,
1351 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001352#ifdef CONFIG_COMPAT
1353 .compat_ioctl = lo_compat_ioctl,
1354#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355};
1356
1357/*
1358 * And now the modules code and kernel interface.
1359 */
1360module_param(max_loop, int, 0);
1361MODULE_PARM_DESC(max_loop, "Maximum number of loop devices (1-256)");
1362MODULE_LICENSE("GPL");
1363MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1364
1365int loop_register_transfer(struct loop_func_table *funcs)
1366{
1367 unsigned int n = funcs->number;
1368
1369 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1370 return -EINVAL;
1371 xfer_funcs[n] = funcs;
1372 return 0;
1373}
1374
1375int loop_unregister_transfer(int number)
1376{
1377 unsigned int n = number;
1378 struct loop_device *lo;
1379 struct loop_func_table *xfer;
1380
1381 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1382 return -EINVAL;
1383
1384 xfer_funcs[n] = NULL;
1385
1386 for (lo = &loop_dev[0]; lo < &loop_dev[max_loop]; lo++) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001387 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 if (lo->lo_encryption == xfer)
1390 loop_release_xfer(lo);
1391
Ingo Molnarf85221d2006-03-23 03:00:38 -08001392 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
1394
1395 return 0;
1396}
1397
1398EXPORT_SYMBOL(loop_register_transfer);
1399EXPORT_SYMBOL(loop_unregister_transfer);
1400
1401static int __init loop_init(void)
1402{
1403 int i;
1404
1405 if (max_loop < 1 || max_loop > 256) {
1406 printk(KERN_WARNING "loop: invalid max_loop (must be between"
1407 " 1 and 256), using default (8)\n");
1408 max_loop = 8;
1409 }
1410
1411 if (register_blkdev(LOOP_MAJOR, "loop"))
1412 return -EIO;
1413
1414 loop_dev = kmalloc(max_loop * sizeof(struct loop_device), GFP_KERNEL);
1415 if (!loop_dev)
1416 goto out_mem1;
1417 memset(loop_dev, 0, max_loop * sizeof(struct loop_device));
1418
1419 disks = kmalloc(max_loop * sizeof(struct gendisk *), GFP_KERNEL);
1420 if (!disks)
1421 goto out_mem2;
1422
1423 for (i = 0; i < max_loop; i++) {
1424 disks[i] = alloc_disk(1);
1425 if (!disks[i])
1426 goto out_mem3;
1427 }
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 for (i = 0; i < max_loop; i++) {
1430 struct loop_device *lo = &loop_dev[i];
1431 struct gendisk *disk = disks[i];
1432
1433 memset(lo, 0, sizeof(*lo));
1434 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1435 if (!lo->lo_queue)
1436 goto out_mem4;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001437 mutex_init(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 lo->lo_number = i;
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001439 lo->lo_thread = NULL;
1440 init_waitqueue_head(&lo->lo_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 spin_lock_init(&lo->lo_lock);
1442 disk->major = LOOP_MAJOR;
1443 disk->first_minor = i;
1444 disk->fops = &lo_fops;
1445 sprintf(disk->disk_name, "loop%d", i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 disk->private_data = lo;
1447 disk->queue = lo->lo_queue;
1448 }
1449
1450 /* We cannot fail after we call this, so another loop!*/
1451 for (i = 0; i < max_loop; i++)
1452 add_disk(disks[i]);
1453 printk(KERN_INFO "loop: loaded (max %d devices)\n", max_loop);
1454 return 0;
1455
1456out_mem4:
1457 while (i--)
Al Viro1312f402006-03-12 11:02:03 -05001458 blk_cleanup_queue(loop_dev[i].lo_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 i = max_loop;
1460out_mem3:
1461 while (i--)
1462 put_disk(disks[i]);
1463 kfree(disks);
1464out_mem2:
1465 kfree(loop_dev);
1466out_mem1:
1467 unregister_blkdev(LOOP_MAJOR, "loop");
1468 printk(KERN_ERR "loop: ran out of memory\n");
1469 return -ENOMEM;
1470}
1471
1472static void loop_exit(void)
1473{
1474 int i;
1475
1476 for (i = 0; i < max_loop; i++) {
1477 del_gendisk(disks[i]);
Al Viro1312f402006-03-12 11:02:03 -05001478 blk_cleanup_queue(loop_dev[i].lo_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 put_disk(disks[i]);
1480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1482 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1483
1484 kfree(disks);
1485 kfree(loop_dev);
1486}
1487
1488module_init(loop_init);
1489module_exit(loop_exit);
1490
1491#ifndef MODULE
1492static int __init max_loop_setup(char *str)
1493{
1494 max_loop = simple_strtol(str, NULL, 0);
1495 return 1;
1496}
1497
1498__setup("max_loop=", max_loop_setup);
1499#endif