blob: 5f6d1a5cce11fc3bebdf1f0ab29c875db7850e09 [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
53#include <linux/config.h>
54#include <linux/module.h>
55#include <linux/moduleparam.h>
56#include <linux/sched.h>
57#include <linux/fs.h>
58#include <linux/file.h>
59#include <linux/stat.h>
60#include <linux/errno.h>
61#include <linux/major.h>
62#include <linux/wait.h>
63#include <linux/blkdev.h>
64#include <linux/blkpg.h>
65#include <linux/init.h>
66#include <linux/devfs_fs_kernel.h>
67#include <linux/smp_lock.h>
68#include <linux/swap.h>
69#include <linux/slab.h>
70#include <linux/loop.h>
71#include <linux/suspend.h>
72#include <linux/writeback.h>
73#include <linux/buffer_head.h> /* for invalidate_bdev() */
74#include <linux/completion.h>
75#include <linux/highmem.h>
76#include <linux/gfp.h>
77
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;
213 struct address_space_operations *aops = mapping->a_ops;
214 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,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 u8 __user *buf, const int len, loff_t pos)
299{
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,
327 (u8 __user *)kmap(bvec->bv_page) + bvec->bv_offset,
328 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,
354 (u8 __user *)page_address(page), bvec->bv_len,
355 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;
527 lo->lo_pending++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 loop_add_bio(lo, old_bio);
Nick Piggin35a82d12005-06-23 00:09:06 -0700529 spin_unlock_irq(&lo->lo_lock);
Ingo Molnar11b751a2006-01-09 15:59:27 -0800530 complete(&lo->lo_bh_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700534 if (lo->lo_pending == 0)
Ingo Molnar11b751a2006-01-09 15:59:27 -0800535 complete(&lo->lo_bh_done);
Nick Piggin35a82d12005-06-23 00:09:06 -0700536 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 bio_io_error(old_bio, old_bio->bi_size);
538 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
541/*
542 * kick off io on the underlying address space
543 */
544static void loop_unplug(request_queue_t *q)
545{
546 struct loop_device *lo = q->queuedata;
547
548 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
549 blk_run_address_space(lo->lo_backing_file->f_mapping);
550}
551
552struct switch_request {
553 struct file *file;
554 struct completion wait;
555};
556
557static void do_loop_switch(struct loop_device *, struct switch_request *);
558
559static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
560{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 if (unlikely(!bio->bi_bdev)) {
562 do_loop_switch(lo, bio->bi_private);
563 bio_put(bio);
564 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700565 int ret = do_bio_filebacked(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 bio_endio(bio, bio->bi_size, ret);
567 }
568}
569
570/*
571 * worker thread that handles reads/writes to file backed loop devices,
572 * to avoid blocking in our make_request_fn. it also does loop decrypting
573 * on reads for block backed loop, as that is too heavy to do from
574 * b_end_io context where irqs may be disabled.
575 */
576static int loop_thread(void *data)
577{
578 struct loop_device *lo = data;
579 struct bio *bio;
580
581 daemonize("loop%d", lo->lo_number);
582
583 /*
584 * loop can be used in an encrypted device,
585 * hence, it mustn't be stopped at all
586 * because it could be indirectly used during suspension
587 */
588 current->flags |= PF_NOFREEZE;
589
590 set_user_nice(current, -20);
591
592 lo->lo_state = Lo_bound;
Nick Piggin35a82d12005-06-23 00:09:06 -0700593 lo->lo_pending = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 /*
Ingo Molnar11b751a2006-01-09 15:59:27 -0800596 * complete it, we are running
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 */
Ingo Molnar11b751a2006-01-09 15:59:27 -0800598 complete(&lo->lo_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 for (;;) {
Nick Piggin35a82d12005-06-23 00:09:06 -0700601 int pending;
602
Ingo Molnar11b751a2006-01-09 15:59:27 -0800603 if (wait_for_completion_interruptible(&lo->lo_bh_done))
Nick Piggin35a82d12005-06-23 00:09:06 -0700604 continue;
605
606 spin_lock_irq(&lo->lo_lock);
607
608 /*
Ingo Molnar11b751a2006-01-09 15:59:27 -0800609 * could be completed because of tear-down, not pending work
Nick Piggin35a82d12005-06-23 00:09:06 -0700610 */
611 if (unlikely(!lo->lo_pending)) {
612 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 break;
Nick Piggin35a82d12005-06-23 00:09:06 -0700614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700617 lo->lo_pending--;
618 pending = lo->lo_pending;
619 spin_unlock_irq(&lo->lo_lock);
620
621 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 loop_handle_bio(lo, bio);
623
624 /*
625 * upped both for pending work and tear-down, lo_pending
626 * will hit zero then
627 */
Nick Piggin35a82d12005-06-23 00:09:06 -0700628 if (unlikely(!pending))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 break;
630 }
631
Ingo Molnar11b751a2006-01-09 15:59:27 -0800632 complete(&lo->lo_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return 0;
634}
635
636/*
637 * loop_switch performs the hard work of switching a backing store.
638 * First it needs to flush existing IO, it does this by sending a magic
639 * BIO down the pipe. The completion of this BIO does the actual switch.
640 */
641static int loop_switch(struct loop_device *lo, struct file *file)
642{
643 struct switch_request w;
644 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
645 if (!bio)
646 return -ENOMEM;
647 init_completion(&w.wait);
648 w.file = file;
649 bio->bi_private = &w;
650 bio->bi_bdev = NULL;
651 loop_make_request(lo->lo_queue, bio);
652 wait_for_completion(&w.wait);
653 return 0;
654}
655
656/*
657 * Do the actual switch; called from the BIO completion routine
658 */
659static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
660{
661 struct file *file = p->file;
662 struct file *old_file = lo->lo_backing_file;
663 struct address_space *mapping = file->f_mapping;
664
665 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
666 lo->lo_backing_file = file;
667 lo->lo_blocksize = mapping->host->i_blksize;
668 lo->old_gfp_mask = mapping_gfp_mask(mapping);
669 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
670 complete(&p->wait);
671}
672
673
674/*
675 * loop_change_fd switched the backing store of a loopback device to
676 * a new file. This is useful for operating system installers to free up
677 * the original file and in High Availability environments to switch to
678 * an alternative location for the content in case of server meltdown.
679 * This can only work if the loop device is used read-only, and if the
680 * new backing store is the same size and type as the old backing store.
681 */
682static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
683 struct block_device *bdev, unsigned int arg)
684{
685 struct file *file, *old_file;
686 struct inode *inode;
687 int error;
688
689 error = -ENXIO;
690 if (lo->lo_state != Lo_bound)
691 goto out;
692
693 /* the loop device has to be read-only */
694 error = -EINVAL;
695 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
696 goto out;
697
698 error = -EBADF;
699 file = fget(arg);
700 if (!file)
701 goto out;
702
703 inode = file->f_mapping->host;
704 old_file = lo->lo_backing_file;
705
706 error = -EINVAL;
707
708 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
709 goto out_putf;
710
711 /* new backing store needs to support loop (eg sendfile) */
712 if (!inode->i_fop->sendfile)
713 goto out_putf;
714
715 /* size of the new backing store needs to be the same */
716 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
717 goto out_putf;
718
719 /* and ... switch */
720 error = loop_switch(lo, file);
721 if (error)
722 goto out_putf;
723
724 fput(old_file);
725 return 0;
726
727 out_putf:
728 fput(file);
729 out:
730 return error;
731}
732
733static inline int is_loop_device(struct file *file)
734{
735 struct inode *i = file->f_mapping->host;
736
737 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
738}
739
740static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
741 struct block_device *bdev, unsigned int arg)
742{
743 struct file *file, *f;
744 struct inode *inode;
745 struct address_space *mapping;
746 unsigned lo_blocksize;
747 int lo_flags = 0;
748 int error;
749 loff_t size;
750
751 /* This is safe, since we have a reference from open(). */
752 __module_get(THIS_MODULE);
753
754 error = -EBADF;
755 file = fget(arg);
756 if (!file)
757 goto out;
758
759 error = -EBUSY;
760 if (lo->lo_state != Lo_unbound)
761 goto out_putf;
762
763 /* Avoid recursion */
764 f = file;
765 while (is_loop_device(f)) {
766 struct loop_device *l;
767
768 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
769 goto out_putf;
770
771 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
772 if (l->lo_state == Lo_unbound) {
773 error = -EINVAL;
774 goto out_putf;
775 }
776 f = l->lo_backing_file;
777 }
778
779 mapping = file->f_mapping;
780 inode = mapping->host;
781
782 if (!(file->f_mode & FMODE_WRITE))
783 lo_flags |= LO_FLAGS_READ_ONLY;
784
785 error = -EINVAL;
786 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
787 struct address_space_operations *aops = mapping->a_ops;
788 /*
789 * If we can't read - sorry. If we only can't write - well,
790 * it's going to be read-only.
791 */
792 if (!file->f_op->sendfile)
793 goto out_putf;
794 if (aops->prepare_write && aops->commit_write)
795 lo_flags |= LO_FLAGS_USE_AOPS;
796 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
797 lo_flags |= LO_FLAGS_READ_ONLY;
798
799 lo_blocksize = inode->i_blksize;
800 error = 0;
801 } else {
802 goto out_putf;
803 }
804
805 size = get_loop_size(lo, file);
806
807 if ((loff_t)(sector_t)size != size) {
808 error = -EFBIG;
809 goto out_putf;
810 }
811
812 if (!(lo_file->f_mode & FMODE_WRITE))
813 lo_flags |= LO_FLAGS_READ_ONLY;
814
815 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
816
817 lo->lo_blocksize = lo_blocksize;
818 lo->lo_device = bdev;
819 lo->lo_flags = lo_flags;
820 lo->lo_backing_file = file;
821 lo->transfer = NULL;
822 lo->ioctl = NULL;
823 lo->lo_sizelimit = 0;
824 lo->old_gfp_mask = mapping_gfp_mask(mapping);
825 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
826
827 lo->lo_bio = lo->lo_biotail = NULL;
828
829 /*
830 * set queue make_request_fn, and add limits based on lower level
831 * device
832 */
833 blk_queue_make_request(lo->lo_queue, loop_make_request);
834 lo->lo_queue->queuedata = lo;
835 lo->lo_queue->unplug_fn = loop_unplug;
836
837 set_capacity(disks[lo->lo_number], size);
838 bd_set_size(bdev, size << 9);
839
840 set_blocksize(bdev, lo_blocksize);
841
842 kernel_thread(loop_thread, lo, CLONE_KERNEL);
Ingo Molnar11b751a2006-01-09 15:59:27 -0800843 wait_for_completion(&lo->lo_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 return 0;
845
846 out_putf:
847 fput(file);
848 out:
849 /* This is safe: open() is still holding a reference. */
850 module_put(THIS_MODULE);
851 return error;
852}
853
854static int
855loop_release_xfer(struct loop_device *lo)
856{
857 int err = 0;
858 struct loop_func_table *xfer = lo->lo_encryption;
859
860 if (xfer) {
861 if (xfer->release)
862 err = xfer->release(lo);
863 lo->transfer = NULL;
864 lo->lo_encryption = NULL;
865 module_put(xfer->owner);
866 }
867 return err;
868}
869
870static int
871loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
872 const struct loop_info64 *i)
873{
874 int err = 0;
875
876 if (xfer) {
877 struct module *owner = xfer->owner;
878
879 if (!try_module_get(owner))
880 return -EINVAL;
881 if (xfer->init)
882 err = xfer->init(lo, i);
883 if (err)
884 module_put(owner);
885 else
886 lo->lo_encryption = xfer;
887 }
888 return err;
889}
890
891static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
892{
893 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400894 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
896 if (lo->lo_state != Lo_bound)
897 return -ENXIO;
898
899 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
900 return -EBUSY;
901
902 if (filp == NULL)
903 return -EINVAL;
904
905 spin_lock_irq(&lo->lo_lock);
906 lo->lo_state = Lo_rundown;
Nick Piggin35a82d12005-06-23 00:09:06 -0700907 lo->lo_pending--;
908 if (!lo->lo_pending)
Ingo Molnar11b751a2006-01-09 15:59:27 -0800909 complete(&lo->lo_bh_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 spin_unlock_irq(&lo->lo_lock);
911
Ingo Molnar11b751a2006-01-09 15:59:27 -0800912 wait_for_completion(&lo->lo_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
914 lo->lo_backing_file = NULL;
915
916 loop_release_xfer(lo);
917 lo->transfer = NULL;
918 lo->ioctl = NULL;
919 lo->lo_device = NULL;
920 lo->lo_encryption = NULL;
921 lo->lo_offset = 0;
922 lo->lo_sizelimit = 0;
923 lo->lo_encrypt_key_size = 0;
924 lo->lo_flags = 0;
925 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
926 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
927 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
928 invalidate_bdev(bdev, 0);
929 set_capacity(disks[lo->lo_number], 0);
930 bd_set_size(bdev, 0);
931 mapping_set_gfp_mask(filp->f_mapping, gfp);
932 lo->lo_state = Lo_unbound;
933 fput(filp);
934 /* This is safe: open() is still holding a reference. */
935 module_put(THIS_MODULE);
936 return 0;
937}
938
939static int
940loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
941{
942 int err;
943 struct loop_func_table *xfer;
944
945 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
946 !capable(CAP_SYS_ADMIN))
947 return -EPERM;
948 if (lo->lo_state != Lo_bound)
949 return -ENXIO;
950 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
951 return -EINVAL;
952
953 err = loop_release_xfer(lo);
954 if (err)
955 return err;
956
957 if (info->lo_encrypt_type) {
958 unsigned int type = info->lo_encrypt_type;
959
960 if (type >= MAX_LO_CRYPT)
961 return -EINVAL;
962 xfer = xfer_funcs[type];
963 if (xfer == NULL)
964 return -EINVAL;
965 } else
966 xfer = NULL;
967
968 err = loop_init_xfer(lo, xfer, info);
969 if (err)
970 return err;
971
972 if (lo->lo_offset != info->lo_offset ||
973 lo->lo_sizelimit != info->lo_sizelimit) {
974 lo->lo_offset = info->lo_offset;
975 lo->lo_sizelimit = info->lo_sizelimit;
976 if (figure_loop_size(lo))
977 return -EFBIG;
978 }
979
980 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
981 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
982 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
983 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
984
985 if (!xfer)
986 xfer = &none_funcs;
987 lo->transfer = xfer->transfer;
988 lo->ioctl = xfer->ioctl;
989
990 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
991 lo->lo_init[0] = info->lo_init[0];
992 lo->lo_init[1] = info->lo_init[1];
993 if (info->lo_encrypt_key_size) {
994 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
995 info->lo_encrypt_key_size);
996 lo->lo_key_owner = current->uid;
997 }
998
999 return 0;
1000}
1001
1002static int
1003loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1004{
1005 struct file *file = lo->lo_backing_file;
1006 struct kstat stat;
1007 int error;
1008
1009 if (lo->lo_state != Lo_bound)
1010 return -ENXIO;
1011 error = vfs_getattr(file->f_vfsmnt, file->f_dentry, &stat);
1012 if (error)
1013 return error;
1014 memset(info, 0, sizeof(*info));
1015 info->lo_number = lo->lo_number;
1016 info->lo_device = huge_encode_dev(stat.dev);
1017 info->lo_inode = stat.ino;
1018 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1019 info->lo_offset = lo->lo_offset;
1020 info->lo_sizelimit = lo->lo_sizelimit;
1021 info->lo_flags = lo->lo_flags;
1022 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1023 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1024 info->lo_encrypt_type =
1025 lo->lo_encryption ? lo->lo_encryption->number : 0;
1026 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1027 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1028 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1029 lo->lo_encrypt_key_size);
1030 }
1031 return 0;
1032}
1033
1034static void
1035loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1036{
1037 memset(info64, 0, sizeof(*info64));
1038 info64->lo_number = info->lo_number;
1039 info64->lo_device = info->lo_device;
1040 info64->lo_inode = info->lo_inode;
1041 info64->lo_rdevice = info->lo_rdevice;
1042 info64->lo_offset = info->lo_offset;
1043 info64->lo_sizelimit = 0;
1044 info64->lo_encrypt_type = info->lo_encrypt_type;
1045 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1046 info64->lo_flags = info->lo_flags;
1047 info64->lo_init[0] = info->lo_init[0];
1048 info64->lo_init[1] = info->lo_init[1];
1049 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1050 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1051 else
1052 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1053 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1054}
1055
1056static int
1057loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1058{
1059 memset(info, 0, sizeof(*info));
1060 info->lo_number = info64->lo_number;
1061 info->lo_device = info64->lo_device;
1062 info->lo_inode = info64->lo_inode;
1063 info->lo_rdevice = info64->lo_rdevice;
1064 info->lo_offset = info64->lo_offset;
1065 info->lo_encrypt_type = info64->lo_encrypt_type;
1066 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1067 info->lo_flags = info64->lo_flags;
1068 info->lo_init[0] = info64->lo_init[0];
1069 info->lo_init[1] = info64->lo_init[1];
1070 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1071 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1072 else
1073 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1074 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1075
1076 /* error in case values were truncated */
1077 if (info->lo_device != info64->lo_device ||
1078 info->lo_rdevice != info64->lo_rdevice ||
1079 info->lo_inode != info64->lo_inode ||
1080 info->lo_offset != info64->lo_offset)
1081 return -EOVERFLOW;
1082
1083 return 0;
1084}
1085
1086static int
1087loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1088{
1089 struct loop_info info;
1090 struct loop_info64 info64;
1091
1092 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1093 return -EFAULT;
1094 loop_info64_from_old(&info, &info64);
1095 return loop_set_status(lo, &info64);
1096}
1097
1098static int
1099loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1100{
1101 struct loop_info64 info64;
1102
1103 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1104 return -EFAULT;
1105 return loop_set_status(lo, &info64);
1106}
1107
1108static int
1109loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1110 struct loop_info info;
1111 struct loop_info64 info64;
1112 int err = 0;
1113
1114 if (!arg)
1115 err = -EINVAL;
1116 if (!err)
1117 err = loop_get_status(lo, &info64);
1118 if (!err)
1119 err = loop_info64_to_old(&info64, &info);
1120 if (!err && copy_to_user(arg, &info, sizeof(info)))
1121 err = -EFAULT;
1122
1123 return err;
1124}
1125
1126static int
1127loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1128 struct loop_info64 info64;
1129 int err = 0;
1130
1131 if (!arg)
1132 err = -EINVAL;
1133 if (!err)
1134 err = loop_get_status(lo, &info64);
1135 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1136 err = -EFAULT;
1137
1138 return err;
1139}
1140
1141static int lo_ioctl(struct inode * inode, struct file * file,
1142 unsigned int cmd, unsigned long arg)
1143{
1144 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1145 int err;
1146
1147 down(&lo->lo_ctl_mutex);
1148 switch (cmd) {
1149 case LOOP_SET_FD:
1150 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1151 break;
1152 case LOOP_CHANGE_FD:
1153 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1154 break;
1155 case LOOP_CLR_FD:
1156 err = loop_clr_fd(lo, inode->i_bdev);
1157 break;
1158 case LOOP_SET_STATUS:
1159 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1160 break;
1161 case LOOP_GET_STATUS:
1162 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1163 break;
1164 case LOOP_SET_STATUS64:
1165 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1166 break;
1167 case LOOP_GET_STATUS64:
1168 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1169 break;
1170 default:
1171 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1172 }
1173 up(&lo->lo_ctl_mutex);
1174 return err;
1175}
1176
1177static int lo_open(struct inode *inode, struct file *file)
1178{
1179 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1180
1181 down(&lo->lo_ctl_mutex);
1182 lo->lo_refcnt++;
1183 up(&lo->lo_ctl_mutex);
1184
1185 return 0;
1186}
1187
1188static int lo_release(struct inode *inode, struct file *file)
1189{
1190 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1191
1192 down(&lo->lo_ctl_mutex);
1193 --lo->lo_refcnt;
1194 up(&lo->lo_ctl_mutex);
1195
1196 return 0;
1197}
1198
1199static struct block_device_operations lo_fops = {
1200 .owner = THIS_MODULE,
1201 .open = lo_open,
1202 .release = lo_release,
1203 .ioctl = lo_ioctl,
1204};
1205
1206/*
1207 * And now the modules code and kernel interface.
1208 */
1209module_param(max_loop, int, 0);
1210MODULE_PARM_DESC(max_loop, "Maximum number of loop devices (1-256)");
1211MODULE_LICENSE("GPL");
1212MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1213
1214int loop_register_transfer(struct loop_func_table *funcs)
1215{
1216 unsigned int n = funcs->number;
1217
1218 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1219 return -EINVAL;
1220 xfer_funcs[n] = funcs;
1221 return 0;
1222}
1223
1224int loop_unregister_transfer(int number)
1225{
1226 unsigned int n = number;
1227 struct loop_device *lo;
1228 struct loop_func_table *xfer;
1229
1230 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1231 return -EINVAL;
1232
1233 xfer_funcs[n] = NULL;
1234
1235 for (lo = &loop_dev[0]; lo < &loop_dev[max_loop]; lo++) {
1236 down(&lo->lo_ctl_mutex);
1237
1238 if (lo->lo_encryption == xfer)
1239 loop_release_xfer(lo);
1240
1241 up(&lo->lo_ctl_mutex);
1242 }
1243
1244 return 0;
1245}
1246
1247EXPORT_SYMBOL(loop_register_transfer);
1248EXPORT_SYMBOL(loop_unregister_transfer);
1249
1250static int __init loop_init(void)
1251{
1252 int i;
1253
1254 if (max_loop < 1 || max_loop > 256) {
1255 printk(KERN_WARNING "loop: invalid max_loop (must be between"
1256 " 1 and 256), using default (8)\n");
1257 max_loop = 8;
1258 }
1259
1260 if (register_blkdev(LOOP_MAJOR, "loop"))
1261 return -EIO;
1262
1263 loop_dev = kmalloc(max_loop * sizeof(struct loop_device), GFP_KERNEL);
1264 if (!loop_dev)
1265 goto out_mem1;
1266 memset(loop_dev, 0, max_loop * sizeof(struct loop_device));
1267
1268 disks = kmalloc(max_loop * sizeof(struct gendisk *), GFP_KERNEL);
1269 if (!disks)
1270 goto out_mem2;
1271
1272 for (i = 0; i < max_loop; i++) {
1273 disks[i] = alloc_disk(1);
1274 if (!disks[i])
1275 goto out_mem3;
1276 }
1277
1278 devfs_mk_dir("loop");
1279
1280 for (i = 0; i < max_loop; i++) {
1281 struct loop_device *lo = &loop_dev[i];
1282 struct gendisk *disk = disks[i];
1283
1284 memset(lo, 0, sizeof(*lo));
1285 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1286 if (!lo->lo_queue)
1287 goto out_mem4;
1288 init_MUTEX(&lo->lo_ctl_mutex);
Ingo Molnar11b751a2006-01-09 15:59:27 -08001289 init_completion(&lo->lo_done);
1290 init_completion(&lo->lo_bh_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 lo->lo_number = i;
1292 spin_lock_init(&lo->lo_lock);
1293 disk->major = LOOP_MAJOR;
1294 disk->first_minor = i;
1295 disk->fops = &lo_fops;
1296 sprintf(disk->disk_name, "loop%d", i);
1297 sprintf(disk->devfs_name, "loop/%d", i);
1298 disk->private_data = lo;
1299 disk->queue = lo->lo_queue;
1300 }
1301
1302 /* We cannot fail after we call this, so another loop!*/
1303 for (i = 0; i < max_loop; i++)
1304 add_disk(disks[i]);
1305 printk(KERN_INFO "loop: loaded (max %d devices)\n", max_loop);
1306 return 0;
1307
1308out_mem4:
1309 while (i--)
1310 blk_put_queue(loop_dev[i].lo_queue);
1311 devfs_remove("loop");
1312 i = max_loop;
1313out_mem3:
1314 while (i--)
1315 put_disk(disks[i]);
1316 kfree(disks);
1317out_mem2:
1318 kfree(loop_dev);
1319out_mem1:
1320 unregister_blkdev(LOOP_MAJOR, "loop");
1321 printk(KERN_ERR "loop: ran out of memory\n");
1322 return -ENOMEM;
1323}
1324
1325static void loop_exit(void)
1326{
1327 int i;
1328
1329 for (i = 0; i < max_loop; i++) {
1330 del_gendisk(disks[i]);
1331 blk_put_queue(loop_dev[i].lo_queue);
1332 put_disk(disks[i]);
1333 }
1334 devfs_remove("loop");
1335 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1336 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1337
1338 kfree(disks);
1339 kfree(loop_dev);
1340}
1341
1342module_init(loop_init);
1343module_exit(loop_exit);
1344
1345#ifndef MODULE
1346static int __init max_loop_setup(char *str)
1347{
1348 max_loop = simple_strtol(str, NULL, 0);
1349 return 1;
1350}
1351
1352__setup("max_loop=", max_loop_setup);
1353#endif