blob: 6f011d0d8e97a98c0aa661ae6a2223bb73a72fda [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{
475 unsigned long flags;
476
477 spin_lock_irqsave(&lo->lo_lock, flags);
478 if (lo->lo_biotail) {
479 lo->lo_biotail->bi_next = bio;
480 lo->lo_biotail = bio;
481 } else
482 lo->lo_bio = lo->lo_biotail = bio;
483 spin_unlock_irqrestore(&lo->lo_lock, flags);
484
485 up(&lo->lo_bh_mutex);
486}
487
488/*
489 * Grab first pending buffer
490 */
491static struct bio *loop_get_bio(struct loop_device *lo)
492{
493 struct bio *bio;
494
495 spin_lock_irq(&lo->lo_lock);
496 if ((bio = lo->lo_bio)) {
497 if (bio == lo->lo_biotail)
498 lo->lo_biotail = NULL;
499 lo->lo_bio = bio->bi_next;
500 bio->bi_next = NULL;
501 }
502 spin_unlock_irq(&lo->lo_lock);
503
504 return bio;
505}
506
507static int loop_make_request(request_queue_t *q, struct bio *old_bio)
508{
509 struct loop_device *lo = q->queuedata;
510 int rw = bio_rw(old_bio);
511
512 if (!lo)
513 goto out;
514
515 spin_lock_irq(&lo->lo_lock);
516 if (lo->lo_state != Lo_bound)
517 goto inactive;
518 atomic_inc(&lo->lo_pending);
519 spin_unlock_irq(&lo->lo_lock);
520
521 if (rw == WRITE) {
522 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
523 goto err;
524 } else if (rw == READA) {
525 rw = READ;
526 } else if (rw != READ) {
527 printk(KERN_ERR "loop: unknown command (%x)\n", rw);
528 goto err;
529 }
530 loop_add_bio(lo, old_bio);
531 return 0;
532err:
533 if (atomic_dec_and_test(&lo->lo_pending))
534 up(&lo->lo_bh_mutex);
535out:
536 bio_io_error(old_bio, old_bio->bi_size);
537 return 0;
538inactive:
539 spin_unlock_irq(&lo->lo_lock);
540 goto out;
541}
542
543/*
544 * kick off io on the underlying address space
545 */
546static void loop_unplug(request_queue_t *q)
547{
548 struct loop_device *lo = q->queuedata;
549
550 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
551 blk_run_address_space(lo->lo_backing_file->f_mapping);
552}
553
554struct switch_request {
555 struct file *file;
556 struct completion wait;
557};
558
559static void do_loop_switch(struct loop_device *, struct switch_request *);
560
561static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
562{
563 int ret;
564
565 if (unlikely(!bio->bi_bdev)) {
566 do_loop_switch(lo, bio->bi_private);
567 bio_put(bio);
568 } else {
569 ret = do_bio_filebacked(lo, bio);
570 bio_endio(bio, bio->bi_size, ret);
571 }
572}
573
574/*
575 * worker thread that handles reads/writes to file backed loop devices,
576 * to avoid blocking in our make_request_fn. it also does loop decrypting
577 * on reads for block backed loop, as that is too heavy to do from
578 * b_end_io context where irqs may be disabled.
579 */
580static int loop_thread(void *data)
581{
582 struct loop_device *lo = data;
583 struct bio *bio;
584
585 daemonize("loop%d", lo->lo_number);
586
587 /*
588 * loop can be used in an encrypted device,
589 * hence, it mustn't be stopped at all
590 * because it could be indirectly used during suspension
591 */
592 current->flags |= PF_NOFREEZE;
593
594 set_user_nice(current, -20);
595
596 lo->lo_state = Lo_bound;
597 atomic_inc(&lo->lo_pending);
598
599 /*
600 * up sem, we are running
601 */
602 up(&lo->lo_sem);
603
604 for (;;) {
605 down_interruptible(&lo->lo_bh_mutex);
606 /*
607 * could be upped because of tear-down, not because of
608 * pending work
609 */
610 if (!atomic_read(&lo->lo_pending))
611 break;
612
613 bio = loop_get_bio(lo);
614 if (!bio) {
615 printk("loop: missing bio\n");
616 continue;
617 }
618 loop_handle_bio(lo, bio);
619
620 /*
621 * upped both for pending work and tear-down, lo_pending
622 * will hit zero then
623 */
624 if (atomic_dec_and_test(&lo->lo_pending))
625 break;
626 }
627
628 up(&lo->lo_sem);
629 return 0;
630}
631
632/*
633 * loop_switch performs the hard work of switching a backing store.
634 * First it needs to flush existing IO, it does this by sending a magic
635 * BIO down the pipe. The completion of this BIO does the actual switch.
636 */
637static int loop_switch(struct loop_device *lo, struct file *file)
638{
639 struct switch_request w;
640 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
641 if (!bio)
642 return -ENOMEM;
643 init_completion(&w.wait);
644 w.file = file;
645 bio->bi_private = &w;
646 bio->bi_bdev = NULL;
647 loop_make_request(lo->lo_queue, bio);
648 wait_for_completion(&w.wait);
649 return 0;
650}
651
652/*
653 * Do the actual switch; called from the BIO completion routine
654 */
655static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
656{
657 struct file *file = p->file;
658 struct file *old_file = lo->lo_backing_file;
659 struct address_space *mapping = file->f_mapping;
660
661 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
662 lo->lo_backing_file = file;
663 lo->lo_blocksize = mapping->host->i_blksize;
664 lo->old_gfp_mask = mapping_gfp_mask(mapping);
665 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
666 complete(&p->wait);
667}
668
669
670/*
671 * loop_change_fd switched the backing store of a loopback device to
672 * a new file. This is useful for operating system installers to free up
673 * the original file and in High Availability environments to switch to
674 * an alternative location for the content in case of server meltdown.
675 * This can only work if the loop device is used read-only, and if the
676 * new backing store is the same size and type as the old backing store.
677 */
678static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
679 struct block_device *bdev, unsigned int arg)
680{
681 struct file *file, *old_file;
682 struct inode *inode;
683 int error;
684
685 error = -ENXIO;
686 if (lo->lo_state != Lo_bound)
687 goto out;
688
689 /* the loop device has to be read-only */
690 error = -EINVAL;
691 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
692 goto out;
693
694 error = -EBADF;
695 file = fget(arg);
696 if (!file)
697 goto out;
698
699 inode = file->f_mapping->host;
700 old_file = lo->lo_backing_file;
701
702 error = -EINVAL;
703
704 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
705 goto out_putf;
706
707 /* new backing store needs to support loop (eg sendfile) */
708 if (!inode->i_fop->sendfile)
709 goto out_putf;
710
711 /* size of the new backing store needs to be the same */
712 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
713 goto out_putf;
714
715 /* and ... switch */
716 error = loop_switch(lo, file);
717 if (error)
718 goto out_putf;
719
720 fput(old_file);
721 return 0;
722
723 out_putf:
724 fput(file);
725 out:
726 return error;
727}
728
729static inline int is_loop_device(struct file *file)
730{
731 struct inode *i = file->f_mapping->host;
732
733 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
734}
735
736static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
737 struct block_device *bdev, unsigned int arg)
738{
739 struct file *file, *f;
740 struct inode *inode;
741 struct address_space *mapping;
742 unsigned lo_blocksize;
743 int lo_flags = 0;
744 int error;
745 loff_t size;
746
747 /* This is safe, since we have a reference from open(). */
748 __module_get(THIS_MODULE);
749
750 error = -EBADF;
751 file = fget(arg);
752 if (!file)
753 goto out;
754
755 error = -EBUSY;
756 if (lo->lo_state != Lo_unbound)
757 goto out_putf;
758
759 /* Avoid recursion */
760 f = file;
761 while (is_loop_device(f)) {
762 struct loop_device *l;
763
764 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
765 goto out_putf;
766
767 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
768 if (l->lo_state == Lo_unbound) {
769 error = -EINVAL;
770 goto out_putf;
771 }
772 f = l->lo_backing_file;
773 }
774
775 mapping = file->f_mapping;
776 inode = mapping->host;
777
778 if (!(file->f_mode & FMODE_WRITE))
779 lo_flags |= LO_FLAGS_READ_ONLY;
780
781 error = -EINVAL;
782 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
783 struct address_space_operations *aops = mapping->a_ops;
784 /*
785 * If we can't read - sorry. If we only can't write - well,
786 * it's going to be read-only.
787 */
788 if (!file->f_op->sendfile)
789 goto out_putf;
790 if (aops->prepare_write && aops->commit_write)
791 lo_flags |= LO_FLAGS_USE_AOPS;
792 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
793 lo_flags |= LO_FLAGS_READ_ONLY;
794
795 lo_blocksize = inode->i_blksize;
796 error = 0;
797 } else {
798 goto out_putf;
799 }
800
801 size = get_loop_size(lo, file);
802
803 if ((loff_t)(sector_t)size != size) {
804 error = -EFBIG;
805 goto out_putf;
806 }
807
808 if (!(lo_file->f_mode & FMODE_WRITE))
809 lo_flags |= LO_FLAGS_READ_ONLY;
810
811 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
812
813 lo->lo_blocksize = lo_blocksize;
814 lo->lo_device = bdev;
815 lo->lo_flags = lo_flags;
816 lo->lo_backing_file = file;
817 lo->transfer = NULL;
818 lo->ioctl = NULL;
819 lo->lo_sizelimit = 0;
820 lo->old_gfp_mask = mapping_gfp_mask(mapping);
821 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
822
823 lo->lo_bio = lo->lo_biotail = NULL;
824
825 /*
826 * set queue make_request_fn, and add limits based on lower level
827 * device
828 */
829 blk_queue_make_request(lo->lo_queue, loop_make_request);
830 lo->lo_queue->queuedata = lo;
831 lo->lo_queue->unplug_fn = loop_unplug;
832
833 set_capacity(disks[lo->lo_number], size);
834 bd_set_size(bdev, size << 9);
835
836 set_blocksize(bdev, lo_blocksize);
837
838 kernel_thread(loop_thread, lo, CLONE_KERNEL);
839 down(&lo->lo_sem);
840 return 0;
841
842 out_putf:
843 fput(file);
844 out:
845 /* This is safe: open() is still holding a reference. */
846 module_put(THIS_MODULE);
847 return error;
848}
849
850static int
851loop_release_xfer(struct loop_device *lo)
852{
853 int err = 0;
854 struct loop_func_table *xfer = lo->lo_encryption;
855
856 if (xfer) {
857 if (xfer->release)
858 err = xfer->release(lo);
859 lo->transfer = NULL;
860 lo->lo_encryption = NULL;
861 module_put(xfer->owner);
862 }
863 return err;
864}
865
866static int
867loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
868 const struct loop_info64 *i)
869{
870 int err = 0;
871
872 if (xfer) {
873 struct module *owner = xfer->owner;
874
875 if (!try_module_get(owner))
876 return -EINVAL;
877 if (xfer->init)
878 err = xfer->init(lo, i);
879 if (err)
880 module_put(owner);
881 else
882 lo->lo_encryption = xfer;
883 }
884 return err;
885}
886
887static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
888{
889 struct file *filp = lo->lo_backing_file;
890 int gfp = lo->old_gfp_mask;
891
892 if (lo->lo_state != Lo_bound)
893 return -ENXIO;
894
895 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
896 return -EBUSY;
897
898 if (filp == NULL)
899 return -EINVAL;
900
901 spin_lock_irq(&lo->lo_lock);
902 lo->lo_state = Lo_rundown;
903 if (atomic_dec_and_test(&lo->lo_pending))
904 up(&lo->lo_bh_mutex);
905 spin_unlock_irq(&lo->lo_lock);
906
907 down(&lo->lo_sem);
908
909 lo->lo_backing_file = NULL;
910
911 loop_release_xfer(lo);
912 lo->transfer = NULL;
913 lo->ioctl = NULL;
914 lo->lo_device = NULL;
915 lo->lo_encryption = NULL;
916 lo->lo_offset = 0;
917 lo->lo_sizelimit = 0;
918 lo->lo_encrypt_key_size = 0;
919 lo->lo_flags = 0;
920 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
921 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
922 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
923 invalidate_bdev(bdev, 0);
924 set_capacity(disks[lo->lo_number], 0);
925 bd_set_size(bdev, 0);
926 mapping_set_gfp_mask(filp->f_mapping, gfp);
927 lo->lo_state = Lo_unbound;
928 fput(filp);
929 /* This is safe: open() is still holding a reference. */
930 module_put(THIS_MODULE);
931 return 0;
932}
933
934static int
935loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
936{
937 int err;
938 struct loop_func_table *xfer;
939
940 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
941 !capable(CAP_SYS_ADMIN))
942 return -EPERM;
943 if (lo->lo_state != Lo_bound)
944 return -ENXIO;
945 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
946 return -EINVAL;
947
948 err = loop_release_xfer(lo);
949 if (err)
950 return err;
951
952 if (info->lo_encrypt_type) {
953 unsigned int type = info->lo_encrypt_type;
954
955 if (type >= MAX_LO_CRYPT)
956 return -EINVAL;
957 xfer = xfer_funcs[type];
958 if (xfer == NULL)
959 return -EINVAL;
960 } else
961 xfer = NULL;
962
963 err = loop_init_xfer(lo, xfer, info);
964 if (err)
965 return err;
966
967 if (lo->lo_offset != info->lo_offset ||
968 lo->lo_sizelimit != info->lo_sizelimit) {
969 lo->lo_offset = info->lo_offset;
970 lo->lo_sizelimit = info->lo_sizelimit;
971 if (figure_loop_size(lo))
972 return -EFBIG;
973 }
974
975 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
976 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
977 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
978 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
979
980 if (!xfer)
981 xfer = &none_funcs;
982 lo->transfer = xfer->transfer;
983 lo->ioctl = xfer->ioctl;
984
985 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
986 lo->lo_init[0] = info->lo_init[0];
987 lo->lo_init[1] = info->lo_init[1];
988 if (info->lo_encrypt_key_size) {
989 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
990 info->lo_encrypt_key_size);
991 lo->lo_key_owner = current->uid;
992 }
993
994 return 0;
995}
996
997static int
998loop_get_status(struct loop_device *lo, struct loop_info64 *info)
999{
1000 struct file *file = lo->lo_backing_file;
1001 struct kstat stat;
1002 int error;
1003
1004 if (lo->lo_state != Lo_bound)
1005 return -ENXIO;
1006 error = vfs_getattr(file->f_vfsmnt, file->f_dentry, &stat);
1007 if (error)
1008 return error;
1009 memset(info, 0, sizeof(*info));
1010 info->lo_number = lo->lo_number;
1011 info->lo_device = huge_encode_dev(stat.dev);
1012 info->lo_inode = stat.ino;
1013 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1014 info->lo_offset = lo->lo_offset;
1015 info->lo_sizelimit = lo->lo_sizelimit;
1016 info->lo_flags = lo->lo_flags;
1017 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1018 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1019 info->lo_encrypt_type =
1020 lo->lo_encryption ? lo->lo_encryption->number : 0;
1021 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1022 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1023 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1024 lo->lo_encrypt_key_size);
1025 }
1026 return 0;
1027}
1028
1029static void
1030loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1031{
1032 memset(info64, 0, sizeof(*info64));
1033 info64->lo_number = info->lo_number;
1034 info64->lo_device = info->lo_device;
1035 info64->lo_inode = info->lo_inode;
1036 info64->lo_rdevice = info->lo_rdevice;
1037 info64->lo_offset = info->lo_offset;
1038 info64->lo_sizelimit = 0;
1039 info64->lo_encrypt_type = info->lo_encrypt_type;
1040 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1041 info64->lo_flags = info->lo_flags;
1042 info64->lo_init[0] = info->lo_init[0];
1043 info64->lo_init[1] = info->lo_init[1];
1044 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1045 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1046 else
1047 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1048 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1049}
1050
1051static int
1052loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1053{
1054 memset(info, 0, sizeof(*info));
1055 info->lo_number = info64->lo_number;
1056 info->lo_device = info64->lo_device;
1057 info->lo_inode = info64->lo_inode;
1058 info->lo_rdevice = info64->lo_rdevice;
1059 info->lo_offset = info64->lo_offset;
1060 info->lo_encrypt_type = info64->lo_encrypt_type;
1061 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1062 info->lo_flags = info64->lo_flags;
1063 info->lo_init[0] = info64->lo_init[0];
1064 info->lo_init[1] = info64->lo_init[1];
1065 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1066 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1067 else
1068 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1069 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1070
1071 /* error in case values were truncated */
1072 if (info->lo_device != info64->lo_device ||
1073 info->lo_rdevice != info64->lo_rdevice ||
1074 info->lo_inode != info64->lo_inode ||
1075 info->lo_offset != info64->lo_offset)
1076 return -EOVERFLOW;
1077
1078 return 0;
1079}
1080
1081static int
1082loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1083{
1084 struct loop_info info;
1085 struct loop_info64 info64;
1086
1087 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1088 return -EFAULT;
1089 loop_info64_from_old(&info, &info64);
1090 return loop_set_status(lo, &info64);
1091}
1092
1093static int
1094loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1095{
1096 struct loop_info64 info64;
1097
1098 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1099 return -EFAULT;
1100 return loop_set_status(lo, &info64);
1101}
1102
1103static int
1104loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1105 struct loop_info info;
1106 struct loop_info64 info64;
1107 int err = 0;
1108
1109 if (!arg)
1110 err = -EINVAL;
1111 if (!err)
1112 err = loop_get_status(lo, &info64);
1113 if (!err)
1114 err = loop_info64_to_old(&info64, &info);
1115 if (!err && copy_to_user(arg, &info, sizeof(info)))
1116 err = -EFAULT;
1117
1118 return err;
1119}
1120
1121static int
1122loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1123 struct loop_info64 info64;
1124 int err = 0;
1125
1126 if (!arg)
1127 err = -EINVAL;
1128 if (!err)
1129 err = loop_get_status(lo, &info64);
1130 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1131 err = -EFAULT;
1132
1133 return err;
1134}
1135
1136static int lo_ioctl(struct inode * inode, struct file * file,
1137 unsigned int cmd, unsigned long arg)
1138{
1139 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1140 int err;
1141
1142 down(&lo->lo_ctl_mutex);
1143 switch (cmd) {
1144 case LOOP_SET_FD:
1145 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1146 break;
1147 case LOOP_CHANGE_FD:
1148 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1149 break;
1150 case LOOP_CLR_FD:
1151 err = loop_clr_fd(lo, inode->i_bdev);
1152 break;
1153 case LOOP_SET_STATUS:
1154 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1155 break;
1156 case LOOP_GET_STATUS:
1157 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1158 break;
1159 case LOOP_SET_STATUS64:
1160 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1161 break;
1162 case LOOP_GET_STATUS64:
1163 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1164 break;
1165 default:
1166 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1167 }
1168 up(&lo->lo_ctl_mutex);
1169 return err;
1170}
1171
1172static int lo_open(struct inode *inode, struct file *file)
1173{
1174 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1175
1176 down(&lo->lo_ctl_mutex);
1177 lo->lo_refcnt++;
1178 up(&lo->lo_ctl_mutex);
1179
1180 return 0;
1181}
1182
1183static int lo_release(struct inode *inode, struct file *file)
1184{
1185 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1186
1187 down(&lo->lo_ctl_mutex);
1188 --lo->lo_refcnt;
1189 up(&lo->lo_ctl_mutex);
1190
1191 return 0;
1192}
1193
1194static struct block_device_operations lo_fops = {
1195 .owner = THIS_MODULE,
1196 .open = lo_open,
1197 .release = lo_release,
1198 .ioctl = lo_ioctl,
1199};
1200
1201/*
1202 * And now the modules code and kernel interface.
1203 */
1204module_param(max_loop, int, 0);
1205MODULE_PARM_DESC(max_loop, "Maximum number of loop devices (1-256)");
1206MODULE_LICENSE("GPL");
1207MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1208
1209int loop_register_transfer(struct loop_func_table *funcs)
1210{
1211 unsigned int n = funcs->number;
1212
1213 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1214 return -EINVAL;
1215 xfer_funcs[n] = funcs;
1216 return 0;
1217}
1218
1219int loop_unregister_transfer(int number)
1220{
1221 unsigned int n = number;
1222 struct loop_device *lo;
1223 struct loop_func_table *xfer;
1224
1225 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1226 return -EINVAL;
1227
1228 xfer_funcs[n] = NULL;
1229
1230 for (lo = &loop_dev[0]; lo < &loop_dev[max_loop]; lo++) {
1231 down(&lo->lo_ctl_mutex);
1232
1233 if (lo->lo_encryption == xfer)
1234 loop_release_xfer(lo);
1235
1236 up(&lo->lo_ctl_mutex);
1237 }
1238
1239 return 0;
1240}
1241
1242EXPORT_SYMBOL(loop_register_transfer);
1243EXPORT_SYMBOL(loop_unregister_transfer);
1244
1245static int __init loop_init(void)
1246{
1247 int i;
1248
1249 if (max_loop < 1 || max_loop > 256) {
1250 printk(KERN_WARNING "loop: invalid max_loop (must be between"
1251 " 1 and 256), using default (8)\n");
1252 max_loop = 8;
1253 }
1254
1255 if (register_blkdev(LOOP_MAJOR, "loop"))
1256 return -EIO;
1257
1258 loop_dev = kmalloc(max_loop * sizeof(struct loop_device), GFP_KERNEL);
1259 if (!loop_dev)
1260 goto out_mem1;
1261 memset(loop_dev, 0, max_loop * sizeof(struct loop_device));
1262
1263 disks = kmalloc(max_loop * sizeof(struct gendisk *), GFP_KERNEL);
1264 if (!disks)
1265 goto out_mem2;
1266
1267 for (i = 0; i < max_loop; i++) {
1268 disks[i] = alloc_disk(1);
1269 if (!disks[i])
1270 goto out_mem3;
1271 }
1272
1273 devfs_mk_dir("loop");
1274
1275 for (i = 0; i < max_loop; i++) {
1276 struct loop_device *lo = &loop_dev[i];
1277 struct gendisk *disk = disks[i];
1278
1279 memset(lo, 0, sizeof(*lo));
1280 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1281 if (!lo->lo_queue)
1282 goto out_mem4;
1283 init_MUTEX(&lo->lo_ctl_mutex);
1284 init_MUTEX_LOCKED(&lo->lo_sem);
1285 init_MUTEX_LOCKED(&lo->lo_bh_mutex);
1286 lo->lo_number = i;
1287 spin_lock_init(&lo->lo_lock);
1288 disk->major = LOOP_MAJOR;
1289 disk->first_minor = i;
1290 disk->fops = &lo_fops;
1291 sprintf(disk->disk_name, "loop%d", i);
1292 sprintf(disk->devfs_name, "loop/%d", i);
1293 disk->private_data = lo;
1294 disk->queue = lo->lo_queue;
1295 }
1296
1297 /* We cannot fail after we call this, so another loop!*/
1298 for (i = 0; i < max_loop; i++)
1299 add_disk(disks[i]);
1300 printk(KERN_INFO "loop: loaded (max %d devices)\n", max_loop);
1301 return 0;
1302
1303out_mem4:
1304 while (i--)
1305 blk_put_queue(loop_dev[i].lo_queue);
1306 devfs_remove("loop");
1307 i = max_loop;
1308out_mem3:
1309 while (i--)
1310 put_disk(disks[i]);
1311 kfree(disks);
1312out_mem2:
1313 kfree(loop_dev);
1314out_mem1:
1315 unregister_blkdev(LOOP_MAJOR, "loop");
1316 printk(KERN_ERR "loop: ran out of memory\n");
1317 return -ENOMEM;
1318}
1319
1320static void loop_exit(void)
1321{
1322 int i;
1323
1324 for (i = 0; i < max_loop; i++) {
1325 del_gendisk(disks[i]);
1326 blk_put_queue(loop_dev[i].lo_queue);
1327 put_disk(disks[i]);
1328 }
1329 devfs_remove("loop");
1330 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1331 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1332
1333 kfree(disks);
1334 kfree(loop_dev);
1335}
1336
1337module_init(loop_init);
1338module_exit(loop_exit);
1339
1340#ifndef MODULE
1341static int __init max_loop_setup(char *str)
1342{
1343 max_loop = simple_strtol(str, NULL, 0);
1344 return 1;
1345}
1346
1347__setup("max_loop=", max_loop_setup);
1348#endif