blob: b35e08876dd4b77cb1b661571c204d51e610fcd9 [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;
216 int len, ret = 0;
217
218 down(&mapping->host->i_sem);
219 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;
235 if (unlikely(aops->prepare_write(file, page, offset,
236 offset + size)))
237 goto unlock;
238 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
239 bvec->bv_page, bv_offs, size, IV);
240 if (unlikely(transfer_result)) {
241 char *kaddr;
242
243 /*
244 * The transfer failed, but we still write the data to
245 * keep prepare/commit calls balanced.
246 */
247 printk(KERN_ERR "loop: transfer error block %llu\n",
248 (unsigned long long)index);
249 kaddr = kmap_atomic(page, KM_USER0);
250 memset(kaddr + offset, 0, size);
251 kunmap_atomic(kaddr, KM_USER0);
252 }
253 flush_dcache_page(page);
254 if (unlikely(aops->commit_write(file, page, offset,
255 offset + size)))
256 goto unlock;
257 if (unlikely(transfer_result))
258 goto unlock;
259 bv_offs += size;
260 len -= size;
261 offset = 0;
262 index++;
263 pos += size;
264 unlock_page(page);
265 page_cache_release(page);
266 }
267out:
268 up(&mapping->host->i_sem);
269 return ret;
270unlock:
271 unlock_page(page);
272 page_cache_release(page);
273fail:
274 ret = -1;
275 goto out;
276}
277
278/**
279 * __do_lo_send_write - helper for writing data to a loop device
280 *
281 * This helper just factors out common code between do_lo_send_direct_write()
282 * and do_lo_send_write().
283 */
284static inline int __do_lo_send_write(struct file *file,
285 u8 __user *buf, const int len, loff_t pos)
286{
287 ssize_t bw;
288 mm_segment_t old_fs = get_fs();
289
290 set_fs(get_ds());
291 bw = file->f_op->write(file, buf, len, &pos);
292 set_fs(old_fs);
293 if (likely(bw == len))
294 return 0;
295 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
296 (unsigned long long)pos, len);
297 if (bw >= 0)
298 bw = -EIO;
299 return bw;
300}
301
302/**
303 * do_lo_send_direct_write - helper for writing data to a loop device
304 *
305 * This is the fast, non-transforming version for backing filesystems which do
306 * not implement the address space operations prepare_write and commit_write.
307 * It uses the write file operation which should be present on all writeable
308 * filesystems.
309 */
310static int do_lo_send_direct_write(struct loop_device *lo,
311 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
312{
313 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
314 (u8 __user *)kmap(bvec->bv_page) + bvec->bv_offset,
315 bvec->bv_len, pos);
316 kunmap(bvec->bv_page);
317 cond_resched();
318 return bw;
319}
320
321/**
322 * do_lo_send_write - helper for writing data to a loop device
323 *
324 * This is the slow, transforming version for filesystems which do not
325 * implement the address space operations prepare_write and commit_write. It
326 * uses the write file operation which should be present on all writeable
327 * filesystems.
328 *
329 * Using fops->write is slower than using aops->{prepare,commit}_write in the
330 * transforming case because we need to double buffer the data as we cannot do
331 * the transformations in place as we do not have direct access to the
332 * destination pages of the backing file.
333 */
334static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
335 int bsize, loff_t pos, struct page *page)
336{
337 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
338 bvec->bv_offset, bvec->bv_len, pos >> 9);
339 if (likely(!ret))
340 return __do_lo_send_write(lo->lo_backing_file,
341 (u8 __user *)page_address(page), bvec->bv_len,
342 pos);
343 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
344 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
345 if (ret > 0)
346 ret = -EIO;
347 return ret;
348}
349
350static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
351 loff_t pos)
352{
353 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
354 struct page *page);
355 struct bio_vec *bvec;
356 struct page *page = NULL;
357 int i, ret = 0;
358
359 do_lo_send = do_lo_send_aops;
360 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
361 do_lo_send = do_lo_send_direct_write;
362 if (lo->transfer != transfer_none) {
363 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
364 if (unlikely(!page))
365 goto fail;
366 kmap(page);
367 do_lo_send = do_lo_send_write;
368 }
369 }
370 bio_for_each_segment(bvec, bio, i) {
371 ret = do_lo_send(lo, bvec, bsize, pos, page);
372 if (ret < 0)
373 break;
374 pos += bvec->bv_len;
375 }
376 if (page) {
377 kunmap(page);
378 __free_page(page);
379 }
380out:
381 return ret;
382fail:
383 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
384 ret = -ENOMEM;
385 goto out;
386}
387
388struct lo_read_data {
389 struct loop_device *lo;
390 struct page *page;
391 unsigned offset;
392 int bsize;
393};
394
395static int
396lo_read_actor(read_descriptor_t *desc, struct page *page,
397 unsigned long offset, unsigned long size)
398{
399 unsigned long count = desc->count;
400 struct lo_read_data *p = desc->arg.data;
401 struct loop_device *lo = p->lo;
402 sector_t IV;
403
404 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
405
406 if (size > count)
407 size = count;
408
409 if (lo_do_transfer(lo, READ, page, offset, p->page, p->offset, size, IV)) {
410 size = 0;
411 printk(KERN_ERR "loop: transfer error block %ld\n",
412 page->index);
413 desc->error = -EINVAL;
414 }
415
416 flush_dcache_page(p->page);
417
418 desc->count = count - size;
419 desc->written += size;
420 p->offset += size;
421 return size;
422}
423
424static int
425do_lo_receive(struct loop_device *lo,
426 struct bio_vec *bvec, int bsize, loff_t pos)
427{
428 struct lo_read_data cookie;
429 struct file *file;
430 int retval;
431
432 cookie.lo = lo;
433 cookie.page = bvec->bv_page;
434 cookie.offset = bvec->bv_offset;
435 cookie.bsize = bsize;
436 file = lo->lo_backing_file;
437 retval = file->f_op->sendfile(file, &pos, bvec->bv_len,
438 lo_read_actor, &cookie);
439 return (retval < 0)? retval: 0;
440}
441
442static int
443lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
444{
445 struct bio_vec *bvec;
446 int i, ret = 0;
447
448 bio_for_each_segment(bvec, bio, i) {
449 ret = do_lo_receive(lo, bvec, bsize, pos);
450 if (ret < 0)
451 break;
452 pos += bvec->bv_len;
453 }
454 return ret;
455}
456
457static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
458{
459 loff_t pos;
460 int ret;
461
462 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
463 if (bio_rw(bio) == WRITE)
464 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
465 else
466 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
467 return ret;
468}
469
470/*
471 * Add bio to back of pending list
472 */
473static void loop_add_bio(struct loop_device *lo, struct bio *bio)
474{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 if (lo->lo_biotail) {
476 lo->lo_biotail->bi_next = bio;
477 lo->lo_biotail = bio;
478 } else
479 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
482/*
483 * Grab first pending buffer
484 */
485static struct bio *loop_get_bio(struct loop_device *lo)
486{
487 struct bio *bio;
488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if ((bio = lo->lo_bio)) {
490 if (bio == lo->lo_biotail)
491 lo->lo_biotail = NULL;
492 lo->lo_bio = bio->bi_next;
493 bio->bi_next = NULL;
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 return bio;
497}
498
499static int loop_make_request(request_queue_t *q, struct bio *old_bio)
500{
501 struct loop_device *lo = q->queuedata;
502 int rw = bio_rw(old_bio);
503
Nick Piggin35a82d12005-06-23 00:09:06 -0700504 if (rw == READA)
505 rw = READ;
506
507 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 spin_lock_irq(&lo->lo_lock);
510 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700511 goto out;
512 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
513 goto out;
514 lo->lo_pending++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 loop_add_bio(lo, old_bio);
Nick Piggin35a82d12005-06-23 00:09:06 -0700516 spin_unlock_irq(&lo->lo_lock);
517 up(&lo->lo_bh_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700521 if (lo->lo_pending == 0)
522 up(&lo->lo_bh_mutex);
523 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 bio_io_error(old_bio, old_bio->bi_size);
525 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
528/*
529 * kick off io on the underlying address space
530 */
531static void loop_unplug(request_queue_t *q)
532{
533 struct loop_device *lo = q->queuedata;
534
535 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
536 blk_run_address_space(lo->lo_backing_file->f_mapping);
537}
538
539struct switch_request {
540 struct file *file;
541 struct completion wait;
542};
543
544static void do_loop_switch(struct loop_device *, struct switch_request *);
545
546static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
547{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 if (unlikely(!bio->bi_bdev)) {
549 do_loop_switch(lo, bio->bi_private);
550 bio_put(bio);
551 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700552 int ret = do_bio_filebacked(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 bio_endio(bio, bio->bi_size, ret);
554 }
555}
556
557/*
558 * worker thread that handles reads/writes to file backed loop devices,
559 * to avoid blocking in our make_request_fn. it also does loop decrypting
560 * on reads for block backed loop, as that is too heavy to do from
561 * b_end_io context where irqs may be disabled.
562 */
563static int loop_thread(void *data)
564{
565 struct loop_device *lo = data;
566 struct bio *bio;
567
568 daemonize("loop%d", lo->lo_number);
569
570 /*
571 * loop can be used in an encrypted device,
572 * hence, it mustn't be stopped at all
573 * because it could be indirectly used during suspension
574 */
575 current->flags |= PF_NOFREEZE;
576
577 set_user_nice(current, -20);
578
579 lo->lo_state = Lo_bound;
Nick Piggin35a82d12005-06-23 00:09:06 -0700580 lo->lo_pending = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582 /*
583 * up sem, we are running
584 */
585 up(&lo->lo_sem);
586
587 for (;;) {
Nick Piggin35a82d12005-06-23 00:09:06 -0700588 int pending;
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /*
Nick Piggin35a82d12005-06-23 00:09:06 -0700591 * interruptible just to not contribute to load avg
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 */
Nick Piggin35a82d12005-06-23 00:09:06 -0700593 if (down_interruptible(&lo->lo_bh_mutex))
594 continue;
595
596 spin_lock_irq(&lo->lo_lock);
597
598 /*
599 * could be upped because of tear-down, not pending work
600 */
601 if (unlikely(!lo->lo_pending)) {
602 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 break;
Nick Piggin35a82d12005-06-23 00:09:06 -0700604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700607 lo->lo_pending--;
608 pending = lo->lo_pending;
609 spin_unlock_irq(&lo->lo_lock);
610
611 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 loop_handle_bio(lo, bio);
613
614 /*
615 * upped both for pending work and tear-down, lo_pending
616 * will hit zero then
617 */
Nick Piggin35a82d12005-06-23 00:09:06 -0700618 if (unlikely(!pending))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 break;
620 }
621
622 up(&lo->lo_sem);
623 return 0;
624}
625
626/*
627 * loop_switch performs the hard work of switching a backing store.
628 * First it needs to flush existing IO, it does this by sending a magic
629 * BIO down the pipe. The completion of this BIO does the actual switch.
630 */
631static int loop_switch(struct loop_device *lo, struct file *file)
632{
633 struct switch_request w;
634 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
635 if (!bio)
636 return -ENOMEM;
637 init_completion(&w.wait);
638 w.file = file;
639 bio->bi_private = &w;
640 bio->bi_bdev = NULL;
641 loop_make_request(lo->lo_queue, bio);
642 wait_for_completion(&w.wait);
643 return 0;
644}
645
646/*
647 * Do the actual switch; called from the BIO completion routine
648 */
649static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
650{
651 struct file *file = p->file;
652 struct file *old_file = lo->lo_backing_file;
653 struct address_space *mapping = file->f_mapping;
654
655 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
656 lo->lo_backing_file = file;
657 lo->lo_blocksize = mapping->host->i_blksize;
658 lo->old_gfp_mask = mapping_gfp_mask(mapping);
659 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
660 complete(&p->wait);
661}
662
663
664/*
665 * loop_change_fd switched the backing store of a loopback device to
666 * a new file. This is useful for operating system installers to free up
667 * the original file and in High Availability environments to switch to
668 * an alternative location for the content in case of server meltdown.
669 * This can only work if the loop device is used read-only, and if the
670 * new backing store is the same size and type as the old backing store.
671 */
672static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
673 struct block_device *bdev, unsigned int arg)
674{
675 struct file *file, *old_file;
676 struct inode *inode;
677 int error;
678
679 error = -ENXIO;
680 if (lo->lo_state != Lo_bound)
681 goto out;
682
683 /* the loop device has to be read-only */
684 error = -EINVAL;
685 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
686 goto out;
687
688 error = -EBADF;
689 file = fget(arg);
690 if (!file)
691 goto out;
692
693 inode = file->f_mapping->host;
694 old_file = lo->lo_backing_file;
695
696 error = -EINVAL;
697
698 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
699 goto out_putf;
700
701 /* new backing store needs to support loop (eg sendfile) */
702 if (!inode->i_fop->sendfile)
703 goto out_putf;
704
705 /* size of the new backing store needs to be the same */
706 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
707 goto out_putf;
708
709 /* and ... switch */
710 error = loop_switch(lo, file);
711 if (error)
712 goto out_putf;
713
714 fput(old_file);
715 return 0;
716
717 out_putf:
718 fput(file);
719 out:
720 return error;
721}
722
723static inline int is_loop_device(struct file *file)
724{
725 struct inode *i = file->f_mapping->host;
726
727 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
728}
729
730static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
731 struct block_device *bdev, unsigned int arg)
732{
733 struct file *file, *f;
734 struct inode *inode;
735 struct address_space *mapping;
736 unsigned lo_blocksize;
737 int lo_flags = 0;
738 int error;
739 loff_t size;
740
741 /* This is safe, since we have a reference from open(). */
742 __module_get(THIS_MODULE);
743
744 error = -EBADF;
745 file = fget(arg);
746 if (!file)
747 goto out;
748
749 error = -EBUSY;
750 if (lo->lo_state != Lo_unbound)
751 goto out_putf;
752
753 /* Avoid recursion */
754 f = file;
755 while (is_loop_device(f)) {
756 struct loop_device *l;
757
758 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
759 goto out_putf;
760
761 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
762 if (l->lo_state == Lo_unbound) {
763 error = -EINVAL;
764 goto out_putf;
765 }
766 f = l->lo_backing_file;
767 }
768
769 mapping = file->f_mapping;
770 inode = mapping->host;
771
772 if (!(file->f_mode & FMODE_WRITE))
773 lo_flags |= LO_FLAGS_READ_ONLY;
774
775 error = -EINVAL;
776 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
777 struct address_space_operations *aops = mapping->a_ops;
778 /*
779 * If we can't read - sorry. If we only can't write - well,
780 * it's going to be read-only.
781 */
782 if (!file->f_op->sendfile)
783 goto out_putf;
784 if (aops->prepare_write && aops->commit_write)
785 lo_flags |= LO_FLAGS_USE_AOPS;
786 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
787 lo_flags |= LO_FLAGS_READ_ONLY;
788
789 lo_blocksize = inode->i_blksize;
790 error = 0;
791 } else {
792 goto out_putf;
793 }
794
795 size = get_loop_size(lo, file);
796
797 if ((loff_t)(sector_t)size != size) {
798 error = -EFBIG;
799 goto out_putf;
800 }
801
802 if (!(lo_file->f_mode & FMODE_WRITE))
803 lo_flags |= LO_FLAGS_READ_ONLY;
804
805 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
806
807 lo->lo_blocksize = lo_blocksize;
808 lo->lo_device = bdev;
809 lo->lo_flags = lo_flags;
810 lo->lo_backing_file = file;
811 lo->transfer = NULL;
812 lo->ioctl = NULL;
813 lo->lo_sizelimit = 0;
814 lo->old_gfp_mask = mapping_gfp_mask(mapping);
815 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
816
817 lo->lo_bio = lo->lo_biotail = NULL;
818
819 /*
820 * set queue make_request_fn, and add limits based on lower level
821 * device
822 */
823 blk_queue_make_request(lo->lo_queue, loop_make_request);
824 lo->lo_queue->queuedata = lo;
825 lo->lo_queue->unplug_fn = loop_unplug;
826
827 set_capacity(disks[lo->lo_number], size);
828 bd_set_size(bdev, size << 9);
829
830 set_blocksize(bdev, lo_blocksize);
831
832 kernel_thread(loop_thread, lo, CLONE_KERNEL);
833 down(&lo->lo_sem);
834 return 0;
835
836 out_putf:
837 fput(file);
838 out:
839 /* This is safe: open() is still holding a reference. */
840 module_put(THIS_MODULE);
841 return error;
842}
843
844static int
845loop_release_xfer(struct loop_device *lo)
846{
847 int err = 0;
848 struct loop_func_table *xfer = lo->lo_encryption;
849
850 if (xfer) {
851 if (xfer->release)
852 err = xfer->release(lo);
853 lo->transfer = NULL;
854 lo->lo_encryption = NULL;
855 module_put(xfer->owner);
856 }
857 return err;
858}
859
860static int
861loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
862 const struct loop_info64 *i)
863{
864 int err = 0;
865
866 if (xfer) {
867 struct module *owner = xfer->owner;
868
869 if (!try_module_get(owner))
870 return -EINVAL;
871 if (xfer->init)
872 err = xfer->init(lo, i);
873 if (err)
874 module_put(owner);
875 else
876 lo->lo_encryption = xfer;
877 }
878 return err;
879}
880
881static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
882{
883 struct file *filp = lo->lo_backing_file;
884 int gfp = lo->old_gfp_mask;
885
886 if (lo->lo_state != Lo_bound)
887 return -ENXIO;
888
889 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
890 return -EBUSY;
891
892 if (filp == NULL)
893 return -EINVAL;
894
895 spin_lock_irq(&lo->lo_lock);
896 lo->lo_state = Lo_rundown;
Nick Piggin35a82d12005-06-23 00:09:06 -0700897 lo->lo_pending--;
898 if (!lo->lo_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 up(&lo->lo_bh_mutex);
900 spin_unlock_irq(&lo->lo_lock);
901
902 down(&lo->lo_sem);
903
904 lo->lo_backing_file = NULL;
905
906 loop_release_xfer(lo);
907 lo->transfer = NULL;
908 lo->ioctl = NULL;
909 lo->lo_device = NULL;
910 lo->lo_encryption = NULL;
911 lo->lo_offset = 0;
912 lo->lo_sizelimit = 0;
913 lo->lo_encrypt_key_size = 0;
914 lo->lo_flags = 0;
915 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
916 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
917 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
918 invalidate_bdev(bdev, 0);
919 set_capacity(disks[lo->lo_number], 0);
920 bd_set_size(bdev, 0);
921 mapping_set_gfp_mask(filp->f_mapping, gfp);
922 lo->lo_state = Lo_unbound;
923 fput(filp);
924 /* This is safe: open() is still holding a reference. */
925 module_put(THIS_MODULE);
926 return 0;
927}
928
929static int
930loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
931{
932 int err;
933 struct loop_func_table *xfer;
934
935 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
936 !capable(CAP_SYS_ADMIN))
937 return -EPERM;
938 if (lo->lo_state != Lo_bound)
939 return -ENXIO;
940 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
941 return -EINVAL;
942
943 err = loop_release_xfer(lo);
944 if (err)
945 return err;
946
947 if (info->lo_encrypt_type) {
948 unsigned int type = info->lo_encrypt_type;
949
950 if (type >= MAX_LO_CRYPT)
951 return -EINVAL;
952 xfer = xfer_funcs[type];
953 if (xfer == NULL)
954 return -EINVAL;
955 } else
956 xfer = NULL;
957
958 err = loop_init_xfer(lo, xfer, info);
959 if (err)
960 return err;
961
962 if (lo->lo_offset != info->lo_offset ||
963 lo->lo_sizelimit != info->lo_sizelimit) {
964 lo->lo_offset = info->lo_offset;
965 lo->lo_sizelimit = info->lo_sizelimit;
966 if (figure_loop_size(lo))
967 return -EFBIG;
968 }
969
970 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
971 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
972 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
973 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
974
975 if (!xfer)
976 xfer = &none_funcs;
977 lo->transfer = xfer->transfer;
978 lo->ioctl = xfer->ioctl;
979
980 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
981 lo->lo_init[0] = info->lo_init[0];
982 lo->lo_init[1] = info->lo_init[1];
983 if (info->lo_encrypt_key_size) {
984 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
985 info->lo_encrypt_key_size);
986 lo->lo_key_owner = current->uid;
987 }
988
989 return 0;
990}
991
992static int
993loop_get_status(struct loop_device *lo, struct loop_info64 *info)
994{
995 struct file *file = lo->lo_backing_file;
996 struct kstat stat;
997 int error;
998
999 if (lo->lo_state != Lo_bound)
1000 return -ENXIO;
1001 error = vfs_getattr(file->f_vfsmnt, file->f_dentry, &stat);
1002 if (error)
1003 return error;
1004 memset(info, 0, sizeof(*info));
1005 info->lo_number = lo->lo_number;
1006 info->lo_device = huge_encode_dev(stat.dev);
1007 info->lo_inode = stat.ino;
1008 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1009 info->lo_offset = lo->lo_offset;
1010 info->lo_sizelimit = lo->lo_sizelimit;
1011 info->lo_flags = lo->lo_flags;
1012 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1013 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1014 info->lo_encrypt_type =
1015 lo->lo_encryption ? lo->lo_encryption->number : 0;
1016 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1017 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1018 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1019 lo->lo_encrypt_key_size);
1020 }
1021 return 0;
1022}
1023
1024static void
1025loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1026{
1027 memset(info64, 0, sizeof(*info64));
1028 info64->lo_number = info->lo_number;
1029 info64->lo_device = info->lo_device;
1030 info64->lo_inode = info->lo_inode;
1031 info64->lo_rdevice = info->lo_rdevice;
1032 info64->lo_offset = info->lo_offset;
1033 info64->lo_sizelimit = 0;
1034 info64->lo_encrypt_type = info->lo_encrypt_type;
1035 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1036 info64->lo_flags = info->lo_flags;
1037 info64->lo_init[0] = info->lo_init[0];
1038 info64->lo_init[1] = info->lo_init[1];
1039 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1040 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1041 else
1042 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1043 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1044}
1045
1046static int
1047loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1048{
1049 memset(info, 0, sizeof(*info));
1050 info->lo_number = info64->lo_number;
1051 info->lo_device = info64->lo_device;
1052 info->lo_inode = info64->lo_inode;
1053 info->lo_rdevice = info64->lo_rdevice;
1054 info->lo_offset = info64->lo_offset;
1055 info->lo_encrypt_type = info64->lo_encrypt_type;
1056 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1057 info->lo_flags = info64->lo_flags;
1058 info->lo_init[0] = info64->lo_init[0];
1059 info->lo_init[1] = info64->lo_init[1];
1060 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1061 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1062 else
1063 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1064 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1065
1066 /* error in case values were truncated */
1067 if (info->lo_device != info64->lo_device ||
1068 info->lo_rdevice != info64->lo_rdevice ||
1069 info->lo_inode != info64->lo_inode ||
1070 info->lo_offset != info64->lo_offset)
1071 return -EOVERFLOW;
1072
1073 return 0;
1074}
1075
1076static int
1077loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1078{
1079 struct loop_info info;
1080 struct loop_info64 info64;
1081
1082 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1083 return -EFAULT;
1084 loop_info64_from_old(&info, &info64);
1085 return loop_set_status(lo, &info64);
1086}
1087
1088static int
1089loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1090{
1091 struct loop_info64 info64;
1092
1093 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1094 return -EFAULT;
1095 return loop_set_status(lo, &info64);
1096}
1097
1098static int
1099loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1100 struct loop_info info;
1101 struct loop_info64 info64;
1102 int err = 0;
1103
1104 if (!arg)
1105 err = -EINVAL;
1106 if (!err)
1107 err = loop_get_status(lo, &info64);
1108 if (!err)
1109 err = loop_info64_to_old(&info64, &info);
1110 if (!err && copy_to_user(arg, &info, sizeof(info)))
1111 err = -EFAULT;
1112
1113 return err;
1114}
1115
1116static int
1117loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1118 struct loop_info64 info64;
1119 int err = 0;
1120
1121 if (!arg)
1122 err = -EINVAL;
1123 if (!err)
1124 err = loop_get_status(lo, &info64);
1125 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1126 err = -EFAULT;
1127
1128 return err;
1129}
1130
1131static int lo_ioctl(struct inode * inode, struct file * file,
1132 unsigned int cmd, unsigned long arg)
1133{
1134 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1135 int err;
1136
1137 down(&lo->lo_ctl_mutex);
1138 switch (cmd) {
1139 case LOOP_SET_FD:
1140 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1141 break;
1142 case LOOP_CHANGE_FD:
1143 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1144 break;
1145 case LOOP_CLR_FD:
1146 err = loop_clr_fd(lo, inode->i_bdev);
1147 break;
1148 case LOOP_SET_STATUS:
1149 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1150 break;
1151 case LOOP_GET_STATUS:
1152 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1153 break;
1154 case LOOP_SET_STATUS64:
1155 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1156 break;
1157 case LOOP_GET_STATUS64:
1158 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1159 break;
1160 default:
1161 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1162 }
1163 up(&lo->lo_ctl_mutex);
1164 return err;
1165}
1166
1167static int lo_open(struct inode *inode, struct file *file)
1168{
1169 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1170
1171 down(&lo->lo_ctl_mutex);
1172 lo->lo_refcnt++;
1173 up(&lo->lo_ctl_mutex);
1174
1175 return 0;
1176}
1177
1178static int lo_release(struct inode *inode, struct file *file)
1179{
1180 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1181
1182 down(&lo->lo_ctl_mutex);
1183 --lo->lo_refcnt;
1184 up(&lo->lo_ctl_mutex);
1185
1186 return 0;
1187}
1188
1189static struct block_device_operations lo_fops = {
1190 .owner = THIS_MODULE,
1191 .open = lo_open,
1192 .release = lo_release,
1193 .ioctl = lo_ioctl,
1194};
1195
1196/*
1197 * And now the modules code and kernel interface.
1198 */
1199module_param(max_loop, int, 0);
1200MODULE_PARM_DESC(max_loop, "Maximum number of loop devices (1-256)");
1201MODULE_LICENSE("GPL");
1202MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1203
1204int loop_register_transfer(struct loop_func_table *funcs)
1205{
1206 unsigned int n = funcs->number;
1207
1208 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1209 return -EINVAL;
1210 xfer_funcs[n] = funcs;
1211 return 0;
1212}
1213
1214int loop_unregister_transfer(int number)
1215{
1216 unsigned int n = number;
1217 struct loop_device *lo;
1218 struct loop_func_table *xfer;
1219
1220 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1221 return -EINVAL;
1222
1223 xfer_funcs[n] = NULL;
1224
1225 for (lo = &loop_dev[0]; lo < &loop_dev[max_loop]; lo++) {
1226 down(&lo->lo_ctl_mutex);
1227
1228 if (lo->lo_encryption == xfer)
1229 loop_release_xfer(lo);
1230
1231 up(&lo->lo_ctl_mutex);
1232 }
1233
1234 return 0;
1235}
1236
1237EXPORT_SYMBOL(loop_register_transfer);
1238EXPORT_SYMBOL(loop_unregister_transfer);
1239
1240static int __init loop_init(void)
1241{
1242 int i;
1243
1244 if (max_loop < 1 || max_loop > 256) {
1245 printk(KERN_WARNING "loop: invalid max_loop (must be between"
1246 " 1 and 256), using default (8)\n");
1247 max_loop = 8;
1248 }
1249
1250 if (register_blkdev(LOOP_MAJOR, "loop"))
1251 return -EIO;
1252
1253 loop_dev = kmalloc(max_loop * sizeof(struct loop_device), GFP_KERNEL);
1254 if (!loop_dev)
1255 goto out_mem1;
1256 memset(loop_dev, 0, max_loop * sizeof(struct loop_device));
1257
1258 disks = kmalloc(max_loop * sizeof(struct gendisk *), GFP_KERNEL);
1259 if (!disks)
1260 goto out_mem2;
1261
1262 for (i = 0; i < max_loop; i++) {
1263 disks[i] = alloc_disk(1);
1264 if (!disks[i])
1265 goto out_mem3;
1266 }
1267
1268 devfs_mk_dir("loop");
1269
1270 for (i = 0; i < max_loop; i++) {
1271 struct loop_device *lo = &loop_dev[i];
1272 struct gendisk *disk = disks[i];
1273
1274 memset(lo, 0, sizeof(*lo));
1275 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1276 if (!lo->lo_queue)
1277 goto out_mem4;
1278 init_MUTEX(&lo->lo_ctl_mutex);
1279 init_MUTEX_LOCKED(&lo->lo_sem);
1280 init_MUTEX_LOCKED(&lo->lo_bh_mutex);
1281 lo->lo_number = i;
1282 spin_lock_init(&lo->lo_lock);
1283 disk->major = LOOP_MAJOR;
1284 disk->first_minor = i;
1285 disk->fops = &lo_fops;
1286 sprintf(disk->disk_name, "loop%d", i);
1287 sprintf(disk->devfs_name, "loop/%d", i);
1288 disk->private_data = lo;
1289 disk->queue = lo->lo_queue;
1290 }
1291
1292 /* We cannot fail after we call this, so another loop!*/
1293 for (i = 0; i < max_loop; i++)
1294 add_disk(disks[i]);
1295 printk(KERN_INFO "loop: loaded (max %d devices)\n", max_loop);
1296 return 0;
1297
1298out_mem4:
1299 while (i--)
1300 blk_put_queue(loop_dev[i].lo_queue);
1301 devfs_remove("loop");
1302 i = max_loop;
1303out_mem3:
1304 while (i--)
1305 put_disk(disks[i]);
1306 kfree(disks);
1307out_mem2:
1308 kfree(loop_dev);
1309out_mem1:
1310 unregister_blkdev(LOOP_MAJOR, "loop");
1311 printk(KERN_ERR "loop: ran out of memory\n");
1312 return -ENOMEM;
1313}
1314
1315static void loop_exit(void)
1316{
1317 int i;
1318
1319 for (i = 0; i < max_loop; i++) {
1320 del_gendisk(disks[i]);
1321 blk_put_queue(loop_dev[i].lo_queue);
1322 put_disk(disks[i]);
1323 }
1324 devfs_remove("loop");
1325 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1326 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1327
1328 kfree(disks);
1329 kfree(loop_dev);
1330}
1331
1332module_init(loop_init);
1333module_exit(loop_exit);
1334
1335#ifndef MODULE
1336static int __init max_loop_setup(char *str)
1337{
1338 max_loop = simple_strtol(str, NULL, 0);
1339 return 1;
1340}
1341
1342__setup("max_loop=", max_loop_setup);
1343#endif