projects/linux·blob v2.6.37 · refs

1783 lines · 43.1 KiB

1/*
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 - [email protected] Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <[email protected]> 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, [email protected]
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <[email protected]> 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, <[email protected]>, 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 <[email protected]>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <[email protected]>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
52#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/file.h>
57#include <linux/stat.h>
58#include <linux/errno.h>
59#include <linux/major.h>
60#include <linux/wait.h>
61#include <linux/blkdev.h>
62#include <linux/blkpg.h>
63#include <linux/init.h>
64#include <linux/swap.h>
65#include <linux/slab.h>
66#include <linux/loop.h>
67#include <linux/compat.h>
68#include <linux/suspend.h>
69#include <linux/freezer.h>
70#include <linux/mutex.h>
71#include <linux/writeback.h>
72#include <linux/buffer_head.h> /* for invalidate_bdev() */
73#include <linux/completion.h>
74#include <linux/highmem.h>
75#include <linux/kthread.h>
76#include <linux/splice.h>
77#include <linux/sysfs.h>
78
79#include <asm/uaccess.h>
80
81static DEFINE_MUTEX(loop_mutex);
82static LIST_HEAD(loop_devices);
83static DEFINE_MUTEX(loop_devices_mutex);
84
85static int max_part;
86static int part_shift;
87
88/*
89 * Transfer functions
90 */
91static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
96 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
98
99 if (cmd == READ)
100 memcpy(loop_buf, raw_buf, size);
101 else
102 memcpy(raw_buf, loop_buf, size);
103
104 kunmap_atomic(loop_buf, KM_USER1);
105 kunmap_atomic(raw_buf, KM_USER0);
106 cond_resched();
107 return 0;
108}
109
110static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
114{
115 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
117 char *in, *out, *key;
118 int i, keysize;
119
120 if (cmd == READ) {
121 in = raw_buf;
122 out = loop_buf;
123 } else {
124 in = loop_buf;
125 out = raw_buf;
126 }
127
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
132
133 kunmap_atomic(loop_buf, KM_USER1);
134 kunmap_atomic(raw_buf, KM_USER0);
135 cond_resched();
136 return 0;
137}
138
139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140{
141 if (unlikely(info->lo_encrypt_key_size <= 0))
142 return -EINVAL;
143 return 0;
144}
145
146static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
149};
150
151static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
154 .init = xor_init
155};
156
157/* xfer_funcs[0] is special - its release function is never called */
158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 &none_funcs,
160 &xor_funcs
161};
162
163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164{
165 loff_t size, offset, loopsize;
166
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 offset = lo->lo_offset;
170 loopsize = size - offset;
171 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
172 loopsize = lo->lo_sizelimit;
173
174 /*
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
177 */
178 return loopsize >> 9;
179}
180
181static int
182figure_loop_size(struct loop_device *lo)
183{
184 loff_t size = get_loop_size(lo, lo->lo_backing_file);
185 sector_t x = (sector_t)size;
186
187 if (unlikely((loff_t)x != size))
188 return -EFBIG;
189
190 set_capacity(lo->lo_disk, x);
191 return 0;
192}
193
194static inline int
195lo_do_transfer(struct loop_device *lo, int cmd,
196 struct page *rpage, unsigned roffs,
197 struct page *lpage, unsigned loffs,
198 int size, sector_t rblock)
199{
200 if (unlikely(!lo->transfer))
201 return 0;
202
203 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
204}
205
206/**
207 * do_lo_send_aops - helper for writing data to a loop device
208 *
209 * This is the fast version for backing filesystems which implement the address
210 * space operations write_begin and write_end.
211 */
212static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
213 loff_t pos, struct page *unused)
214{
215 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
216 struct address_space *mapping = file->f_mapping;
217 pgoff_t index;
218 unsigned offset, bv_offs;
219 int len, ret;
220
221 mutex_lock(&mapping->host->i_mutex);
222 index = pos >> PAGE_CACHE_SHIFT;
223 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
224 bv_offs = bvec->bv_offset;
225 len = bvec->bv_len;
226 while (len > 0) {
227 sector_t IV;
228 unsigned size, copied;
229 int transfer_result;
230 struct page *page;
231 void *fsdata;
232
233 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
234 size = PAGE_CACHE_SIZE - offset;
235 if (size > len)
236 size = len;
237
238 ret = pagecache_write_begin(file, mapping, pos, size, 0,
239 &page, &fsdata);
240 if (ret)
241 goto fail;
242
243 file_update_time(file);
244
245 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
246 bvec->bv_page, bv_offs, size, IV);
247 copied = size;
248 if (unlikely(transfer_result))
249 copied = 0;
250
251 ret = pagecache_write_end(file, mapping, pos, size, copied,
252 page, fsdata);
253 if (ret < 0 || ret != copied)
254 goto fail;
255
256 if (unlikely(transfer_result))
257 goto fail;
258
259 bv_offs += copied;
260 len -= copied;
261 offset = 0;
262 index++;
263 pos += copied;
264 }
265 ret = 0;
266out:
267 mutex_unlock(&mapping->host->i_mutex);
268 return ret;
269fail:
270 ret = -1;
271 goto out;
272}
273
274/**
275 * __do_lo_send_write - helper for writing data to a loop device
276 *
277 * This helper just factors out common code between do_lo_send_direct_write()
278 * and do_lo_send_write().
279 */
280static int __do_lo_send_write(struct file *file,
281 u8 *buf, const int len, loff_t pos)
282{
283 ssize_t bw;
284 mm_segment_t old_fs = get_fs();
285
286 set_fs(get_ds());
287 bw = file->f_op->write(file, buf, len, &pos);
288 set_fs(old_fs);
289 if (likely(bw == len))
290 return 0;
291 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)pos, len);
293 if (bw >= 0)
294 bw = -EIO;
295 return bw;
296}
297
298/**
299 * do_lo_send_direct_write - helper for writing data to a loop device
300 *
301 * This is the fast, non-transforming version for backing filesystems which do
302 * not implement the address space operations write_begin and write_end.
303 * It uses the write file operation which should be present on all writeable
304 * filesystems.
305 */
306static int do_lo_send_direct_write(struct loop_device *lo,
307 struct bio_vec *bvec, loff_t pos, struct page *page)
308{
309 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
310 kmap(bvec->bv_page) + bvec->bv_offset,
311 bvec->bv_len, pos);
312 kunmap(bvec->bv_page);
313 cond_resched();
314 return bw;
315}
316
317/**
318 * do_lo_send_write - helper for writing data to a loop device
319 *
320 * This is the slow, transforming version for filesystems which do not
321 * implement the address space operations write_begin and write_end. It
322 * uses the write file operation which should be present on all writeable
323 * filesystems.
324 *
325 * Using fops->write is slower than using aops->{prepare,commit}_write in the
326 * transforming case because we need to double buffer the data as we cannot do
327 * the transformations in place as we do not have direct access to the
328 * destination pages of the backing file.
329 */
330static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
331 loff_t pos, struct page *page)
332{
333 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
334 bvec->bv_offset, bvec->bv_len, pos >> 9);
335 if (likely(!ret))
336 return __do_lo_send_write(lo->lo_backing_file,
337 page_address(page), bvec->bv_len,
338 pos);
339 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
340 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
341 if (ret > 0)
342 ret = -EIO;
343 return ret;
344}
345
346static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
347{
348 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
349 struct page *page);
350 struct bio_vec *bvec;
351 struct page *page = NULL;
352 int i, ret = 0;
353
354 do_lo_send = do_lo_send_aops;
355 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
356 do_lo_send = do_lo_send_direct_write;
357 if (lo->transfer != transfer_none) {
358 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
359 if (unlikely(!page))
360 goto fail;
361 kmap(page);
362 do_lo_send = do_lo_send_write;
363 }
364 }
365 bio_for_each_segment(bvec, bio, i) {
366 ret = do_lo_send(lo, bvec, pos, page);
367 if (ret < 0)
368 break;
369 pos += bvec->bv_len;
370 }
371 if (page) {
372 kunmap(page);
373 __free_page(page);
374 }
375out:
376 return ret;
377fail:
378 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
379 ret = -ENOMEM;
380 goto out;
381}
382
383struct lo_read_data {
384 struct loop_device *lo;
385 struct page *page;
386 unsigned offset;
387 int bsize;
388};
389
390static int
391lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
392 struct splice_desc *sd)
393{
394 struct lo_read_data *p = sd->u.data;
395 struct loop_device *lo = p->lo;
396 struct page *page = buf->page;
397 sector_t IV;
398 int size, ret;
399
400 ret = buf->ops->confirm(pipe, buf);
401 if (unlikely(ret))
402 return ret;
403
404 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
405 (buf->offset >> 9);
406 size = sd->len;
407 if (size > p->bsize)
408 size = p->bsize;
409
410 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
411 printk(KERN_ERR "loop: transfer error block %ld\n",
412 page->index);
413 size = -EINVAL;
414 }
415
416 flush_dcache_page(p->page);
417
418 if (size > 0)
419 p->offset += size;
420
421 return size;
422}
423
424static int
425lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
426{
427 return __splice_from_pipe(pipe, sd, lo_splice_actor);
428}
429
430static int
431do_lo_receive(struct loop_device *lo,
432 struct bio_vec *bvec, int bsize, loff_t pos)
433{
434 struct lo_read_data cookie;
435 struct splice_desc sd;
436 struct file *file;
437 long retval;
438
439 cookie.lo = lo;
440 cookie.page = bvec->bv_page;
441 cookie.offset = bvec->bv_offset;
442 cookie.bsize = bsize;
443
444 sd.len = 0;
445 sd.total_len = bvec->bv_len;
446 sd.flags = 0;
447 sd.pos = pos;
448 sd.u.data = &cookie;
449
450 file = lo->lo_backing_file;
451 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
452
453 if (retval < 0)
454 return retval;
455
456 return 0;
457}
458
459static int
460lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
461{
462 struct bio_vec *bvec;
463 int i, ret = 0;
464
465 bio_for_each_segment(bvec, bio, i) {
466 ret = do_lo_receive(lo, bvec, bsize, pos);
467 if (ret < 0)
468 break;
469 pos += bvec->bv_len;
470 }
471 return ret;
472}
473
474static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
475{
476 loff_t pos;
477 int ret;
478
479 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
480
481 if (bio_rw(bio) == WRITE) {
482 struct file *file = lo->lo_backing_file;
483
484 if (bio->bi_rw & REQ_FLUSH) {
485 ret = vfs_fsync(file, 0);
486 if (unlikely(ret && ret != -EINVAL)) {
487 ret = -EIO;
488 goto out;
489 }
490 }
491
492 ret = lo_send(lo, bio, pos);
493
494 if ((bio->bi_rw & REQ_FUA) && !ret) {
495 ret = vfs_fsync(file, 0);
496 if (unlikely(ret && ret != -EINVAL))
497 ret = -EIO;
498 }
499 } else
500 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
501
502out:
503 return ret;
504}
505
506/*
507 * Add bio to back of pending list
508 */
509static void loop_add_bio(struct loop_device *lo, struct bio *bio)
510{
511 bio_list_add(&lo->lo_bio_list, bio);
512}
513
514/*
515 * Grab first pending buffer
516 */
517static struct bio *loop_get_bio(struct loop_device *lo)
518{
519 return bio_list_pop(&lo->lo_bio_list);
520}
521
522static int loop_make_request(struct request_queue *q, struct bio *old_bio)
523{
524 struct loop_device *lo = q->queuedata;
525 int rw = bio_rw(old_bio);
526
527 if (rw == READA)
528 rw = READ;
529
530 BUG_ON(!lo || (rw != READ && rw != WRITE));
531
532 spin_lock_irq(&lo->lo_lock);
533 if (lo->lo_state != Lo_bound)
534 goto out;
535 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
536 goto out;
537 loop_add_bio(lo, old_bio);
538 wake_up(&lo->lo_event);
539 spin_unlock_irq(&lo->lo_lock);
540 return 0;
541
542out:
543 spin_unlock_irq(&lo->lo_lock);
544 bio_io_error(old_bio);
545 return 0;
546}
547
548/*
549 * kick off io on the underlying address space
550 */
551static void loop_unplug(struct request_queue *q)
552{
553 struct loop_device *lo = q->queuedata;
554
555 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
556 blk_run_address_space(lo->lo_backing_file->f_mapping);
557}
558
559struct switch_request {
560 struct file *file;
561 struct completion wait;
562};
563
564static void do_loop_switch(struct loop_device *, struct switch_request *);
565
566static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
567{
568 if (unlikely(!bio->bi_bdev)) {
569 do_loop_switch(lo, bio->bi_private);
570 bio_put(bio);
571 } else {
572 int ret = do_bio_filebacked(lo, bio);
573 bio_endio(bio, ret);
574 }
575}
576
577/*
578 * worker thread that handles reads/writes to file backed loop devices,
579 * to avoid blocking in our make_request_fn. it also does loop decrypting
580 * on reads for block backed loop, as that is too heavy to do from
581 * b_end_io context where irqs may be disabled.
582 *
583 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
584 * calling kthread_stop(). Therefore once kthread_should_stop() is
585 * true, make_request will not place any more requests. Therefore
586 * once kthread_should_stop() is true and lo_bio is NULL, we are
587 * done with the loop.
588 */
589static int loop_thread(void *data)
590{
591 struct loop_device *lo = data;
592 struct bio *bio;
593
594 set_user_nice(current, -20);
595
596 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
597
598 wait_event_interruptible(lo->lo_event,
599 !bio_list_empty(&lo->lo_bio_list) ||
600 kthread_should_stop());
601
602 if (bio_list_empty(&lo->lo_bio_list))
603 continue;
604 spin_lock_irq(&lo->lo_lock);
605 bio = loop_get_bio(lo);
606 spin_unlock_irq(&lo->lo_lock);
607
608 BUG_ON(!bio);
609 loop_handle_bio(lo, bio);
610 }
611
612 return 0;
613}
614
615/*
616 * loop_switch performs the hard work of switching a backing store.
617 * First it needs to flush existing IO, it does this by sending a magic
618 * BIO down the pipe. The completion of this BIO does the actual switch.
619 */
620static int loop_switch(struct loop_device *lo, struct file *file)
621{
622 struct switch_request w;
623 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
624 if (!bio)
625 return -ENOMEM;
626 init_completion(&w.wait);
627 w.file = file;
628 bio->bi_private = &w;
629 bio->bi_bdev = NULL;
630 loop_make_request(lo->lo_queue, bio);
631 wait_for_completion(&w.wait);
632 return 0;
633}
634
635/*
636 * Helper to flush the IOs in loop, but keeping loop thread running
637 */
638static int loop_flush(struct loop_device *lo)
639{
640 /* loop not yet configured, no running thread, nothing to flush */
641 if (!lo->lo_thread)
642 return 0;
643
644 return loop_switch(lo, NULL);
645}
646
647/*
648 * Do the actual switch; called from the BIO completion routine
649 */
650static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
651{
652 struct file *file = p->file;
653 struct file *old_file = lo->lo_backing_file;
654 struct address_space *mapping;
655
656 /* if no new file, only flush of queued bios requested */
657 if (!file)
658 goto out;
659
660 mapping = file->f_mapping;
661 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
662 lo->lo_backing_file = file;
663 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
664 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
665 lo->old_gfp_mask = mapping_gfp_mask(mapping);
666 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
667out:
668 complete(&p->wait);
669}
670
671
672/*
673 * loop_change_fd switched the backing store of a loopback device to
674 * a new file. This is useful for operating system installers to free up
675 * the original file and in High Availability environments to switch to
676 * an alternative location for the content in case of server meltdown.
677 * This can only work if the loop device is used read-only, and if the
678 * new backing store is the same size and type as the old backing store.
679 */
680static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
681 unsigned int arg)
682{
683 struct file *file, *old_file;
684 struct inode *inode;
685 int error;
686
687 error = -ENXIO;
688 if (lo->lo_state != Lo_bound)
689 goto out;
690
691 /* the loop device has to be read-only */
692 error = -EINVAL;
693 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
694 goto out;
695
696 error = -EBADF;
697 file = fget(arg);
698 if (!file)
699 goto out;
700
701 inode = file->f_mapping->host;
702 old_file = lo->lo_backing_file;
703
704 error = -EINVAL;
705
706 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
707 goto out_putf;
708
709 /* size of the new backing store needs to be the same */
710 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
711 goto out_putf;
712
713 /* and ... switch */
714 error = loop_switch(lo, file);
715 if (error)
716 goto out_putf;
717
718 fput(old_file);
719 if (max_part > 0)
720 ioctl_by_bdev(bdev, BLKRRPART, 0);
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
736/* loop sysfs attributes */
737
738static ssize_t loop_attr_show(struct device *dev, char *page,
739 ssize_t (*callback)(struct loop_device *, char *))
740{
741 struct loop_device *l, *lo = NULL;
742
743 mutex_lock(&loop_devices_mutex);
744 list_for_each_entry(l, &loop_devices, lo_list)
745 if (disk_to_dev(l->lo_disk) == dev) {
746 lo = l;
747 break;
748 }
749 mutex_unlock(&loop_devices_mutex);
750
751 return lo ? callback(lo, page) : -EIO;
752}
753
754#define LOOP_ATTR_RO(_name) \
755static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
756static ssize_t loop_attr_do_show_##_name(struct device *d, \
757 struct device_attribute *attr, char *b) \
758{ \
759 return loop_attr_show(d, b, loop_attr_##_name##_show); \
760} \
761static struct device_attribute loop_attr_##_name = \
762 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
763
764static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
765{
766 ssize_t ret;
767 char *p = NULL;
768
769 mutex_lock(&lo->lo_ctl_mutex);
770 if (lo->lo_backing_file)
771 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
772 mutex_unlock(&lo->lo_ctl_mutex);
773
774 if (IS_ERR_OR_NULL(p))
775 ret = PTR_ERR(p);
776 else {
777 ret = strlen(p);
778 memmove(buf, p, ret);
779 buf[ret++] = '\n';
780 buf[ret] = 0;
781 }
782
783 return ret;
784}
785
786static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
787{
788 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
789}
790
791static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
792{
793 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
794}
795
796static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
797{
798 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
799
800 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
801}
802
803LOOP_ATTR_RO(backing_file);
804LOOP_ATTR_RO(offset);
805LOOP_ATTR_RO(sizelimit);
806LOOP_ATTR_RO(autoclear);
807
808static struct attribute *loop_attrs[] = {
809 &loop_attr_backing_file.attr,
810 &loop_attr_offset.attr,
811 &loop_attr_sizelimit.attr,
812 &loop_attr_autoclear.attr,
813 NULL,
814};
815
816static struct attribute_group loop_attribute_group = {
817 .name = "loop",
818 .attrs= loop_attrs,
819};
820
821static int loop_sysfs_init(struct loop_device *lo)
822{
823 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
824 &loop_attribute_group);
825}
826
827static void loop_sysfs_exit(struct loop_device *lo)
828{
829 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
830 &loop_attribute_group);
831}
832
833static int loop_set_fd(struct loop_device *lo, fmode_t mode,
834 struct block_device *bdev, unsigned int arg)
835{
836 struct file *file, *f;
837 struct inode *inode;
838 struct address_space *mapping;
839 unsigned lo_blocksize;
840 int lo_flags = 0;
841 int error;
842 loff_t size;
843
844 /* This is safe, since we have a reference from open(). */
845 __module_get(THIS_MODULE);
846
847 error = -EBADF;
848 file = fget(arg);
849 if (!file)
850 goto out;
851
852 error = -EBUSY;
853 if (lo->lo_state != Lo_unbound)
854 goto out_putf;
855
856 /* Avoid recursion */
857 f = file;
858 while (is_loop_device(f)) {
859 struct loop_device *l;
860
861 if (f->f_mapping->host->i_bdev == bdev)
862 goto out_putf;
863
864 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
865 if (l->lo_state == Lo_unbound) {
866 error = -EINVAL;
867 goto out_putf;
868 }
869 f = l->lo_backing_file;
870 }
871
872 mapping = file->f_mapping;
873 inode = mapping->host;
874
875 if (!(file->f_mode & FMODE_WRITE))
876 lo_flags |= LO_FLAGS_READ_ONLY;
877
878 error = -EINVAL;
879 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
880 const struct address_space_operations *aops = mapping->a_ops;
881
882 if (aops->write_begin)
883 lo_flags |= LO_FLAGS_USE_AOPS;
884 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
885 lo_flags |= LO_FLAGS_READ_ONLY;
886
887 lo_blocksize = S_ISBLK(inode->i_mode) ?
888 inode->i_bdev->bd_block_size : PAGE_SIZE;
889
890 error = 0;
891 } else {
892 goto out_putf;
893 }
894
895 size = get_loop_size(lo, file);
896
897 if ((loff_t)(sector_t)size != size) {
898 error = -EFBIG;
899 goto out_putf;
900 }
901
902 if (!(mode & FMODE_WRITE))
903 lo_flags |= LO_FLAGS_READ_ONLY;
904
905 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
906
907 lo->lo_blocksize = lo_blocksize;
908 lo->lo_device = bdev;
909 lo->lo_flags = lo_flags;
910 lo->lo_backing_file = file;
911 lo->transfer = transfer_none;
912 lo->ioctl = NULL;
913 lo->lo_sizelimit = 0;
914 lo->old_gfp_mask = mapping_gfp_mask(mapping);
915 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
916
917 bio_list_init(&lo->lo_bio_list);
918
919 /*
920 * set queue make_request_fn, and add limits based on lower level
921 * device
922 */
923 blk_queue_make_request(lo->lo_queue, loop_make_request);
924 lo->lo_queue->queuedata = lo;
925 lo->lo_queue->unplug_fn = loop_unplug;
926
927 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
928 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
929
930 set_capacity(lo->lo_disk, size);
931 bd_set_size(bdev, size << 9);
932 loop_sysfs_init(lo);
933 /* let user-space know about the new size */
934 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
935
936 set_blocksize(bdev, lo_blocksize);
937
938 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
939 lo->lo_number);
940 if (IS_ERR(lo->lo_thread)) {
941 error = PTR_ERR(lo->lo_thread);
942 goto out_clr;
943 }
944 lo->lo_state = Lo_bound;
945 wake_up_process(lo->lo_thread);
946 if (max_part > 0)
947 ioctl_by_bdev(bdev, BLKRRPART, 0);
948 return 0;
949
950out_clr:
951 loop_sysfs_exit(lo);
952 lo->lo_thread = NULL;
953 lo->lo_device = NULL;
954 lo->lo_backing_file = NULL;
955 lo->lo_flags = 0;
956 set_capacity(lo->lo_disk, 0);
957 invalidate_bdev(bdev);
958 bd_set_size(bdev, 0);
959 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
960 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
961 lo->lo_state = Lo_unbound;
962 out_putf:
963 fput(file);
964 out:
965 /* This is safe: open() is still holding a reference. */
966 module_put(THIS_MODULE);
967 return error;
968}
969
970static int
971loop_release_xfer(struct loop_device *lo)
972{
973 int err = 0;
974 struct loop_func_table *xfer = lo->lo_encryption;
975
976 if (xfer) {
977 if (xfer->release)
978 err = xfer->release(lo);
979 lo->transfer = NULL;
980 lo->lo_encryption = NULL;
981 module_put(xfer->owner);
982 }
983 return err;
984}
985
986static int
987loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
988 const struct loop_info64 *i)
989{
990 int err = 0;
991
992 if (xfer) {
993 struct module *owner = xfer->owner;
994
995 if (!try_module_get(owner))
996 return -EINVAL;
997 if (xfer->init)
998 err = xfer->init(lo, i);
999 if (err)
1000 module_put(owner);
1001 else
1002 lo->lo_encryption = xfer;
1003 }
1004 return err;
1005}
1006
1007static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
1008{
1009 struct file *filp = lo->lo_backing_file;
1010 gfp_t gfp = lo->old_gfp_mask;
1011
1012 if (lo->lo_state != Lo_bound)
1013 return -ENXIO;
1014
1015 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
1016 return -EBUSY;
1017
1018 if (filp == NULL)
1019 return -EINVAL;
1020
1021 spin_lock_irq(&lo->lo_lock);
1022 lo->lo_state = Lo_rundown;
1023 spin_unlock_irq(&lo->lo_lock);
1024
1025 kthread_stop(lo->lo_thread);
1026
1027 lo->lo_queue->unplug_fn = NULL;
1028 lo->lo_backing_file = NULL;
1029
1030 loop_release_xfer(lo);
1031 lo->transfer = NULL;
1032 lo->ioctl = NULL;
1033 lo->lo_device = NULL;
1034 lo->lo_encryption = NULL;
1035 lo->lo_offset = 0;
1036 lo->lo_sizelimit = 0;
1037 lo->lo_encrypt_key_size = 0;
1038 lo->lo_flags = 0;
1039 lo->lo_thread = NULL;
1040 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1041 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1042 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1043 if (bdev)
1044 invalidate_bdev(bdev);
1045 set_capacity(lo->lo_disk, 0);
1046 loop_sysfs_exit(lo);
1047 if (bdev) {
1048 bd_set_size(bdev, 0);
1049 /* let user-space know about this change */
1050 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1051 }
1052 mapping_set_gfp_mask(filp->f_mapping, gfp);
1053 lo->lo_state = Lo_unbound;
1054 /* This is safe: open() is still holding a reference. */
1055 module_put(THIS_MODULE);
1056 if (max_part > 0 && bdev)
1057 ioctl_by_bdev(bdev, BLKRRPART, 0);
1058 mutex_unlock(&lo->lo_ctl_mutex);
1059 /*
1060 * Need not hold lo_ctl_mutex to fput backing file.
1061 * Calling fput holding lo_ctl_mutex triggers a circular
1062 * lock dependency possibility warning as fput can take
1063 * bd_mutex which is usually taken before lo_ctl_mutex.
1064 */
1065 fput(filp);
1066 return 0;
1067}
1068
1069static int
1070loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1071{
1072 int err;
1073 struct loop_func_table *xfer;
1074 uid_t uid = current_uid();
1075
1076 if (lo->lo_encrypt_key_size &&
1077 lo->lo_key_owner != uid &&
1078 !capable(CAP_SYS_ADMIN))
1079 return -EPERM;
1080 if (lo->lo_state != Lo_bound)
1081 return -ENXIO;
1082 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1083 return -EINVAL;
1084
1085 err = loop_release_xfer(lo);
1086 if (err)
1087 return err;
1088
1089 if (info->lo_encrypt_type) {
1090 unsigned int type = info->lo_encrypt_type;
1091
1092 if (type >= MAX_LO_CRYPT)
1093 return -EINVAL;
1094 xfer = xfer_funcs[type];
1095 if (xfer == NULL)
1096 return -EINVAL;
1097 } else
1098 xfer = NULL;
1099
1100 err = loop_init_xfer(lo, xfer, info);
1101 if (err)
1102 return err;
1103
1104 if (lo->lo_offset != info->lo_offset ||
1105 lo->lo_sizelimit != info->lo_sizelimit) {
1106 lo->lo_offset = info->lo_offset;
1107 lo->lo_sizelimit = info->lo_sizelimit;
1108 if (figure_loop_size(lo))
1109 return -EFBIG;
1110 }
1111
1112 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1113 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1114 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1115 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1116
1117 if (!xfer)
1118 xfer = &none_funcs;
1119 lo->transfer = xfer->transfer;
1120 lo->ioctl = xfer->ioctl;
1121
1122 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1123 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1124 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1125
1126 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1127 lo->lo_init[0] = info->lo_init[0];
1128 lo->lo_init[1] = info->lo_init[1];
1129 if (info->lo_encrypt_key_size) {
1130 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1131 info->lo_encrypt_key_size);
1132 lo->lo_key_owner = uid;
1133 }
1134
1135 return 0;
1136}
1137
1138static int
1139loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1140{
1141 struct file *file = lo->lo_backing_file;
1142 struct kstat stat;
1143 int error;
1144
1145 if (lo->lo_state != Lo_bound)
1146 return -ENXIO;
1147 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1148 if (error)
1149 return error;
1150 memset(info, 0, sizeof(*info));
1151 info->lo_number = lo->lo_number;
1152 info->lo_device = huge_encode_dev(stat.dev);
1153 info->lo_inode = stat.ino;
1154 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1155 info->lo_offset = lo->lo_offset;
1156 info->lo_sizelimit = lo->lo_sizelimit;
1157 info->lo_flags = lo->lo_flags;
1158 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1159 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1160 info->lo_encrypt_type =
1161 lo->lo_encryption ? lo->lo_encryption->number : 0;
1162 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1163 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1164 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1165 lo->lo_encrypt_key_size);
1166 }
1167 return 0;
1168}
1169
1170static void
1171loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1172{
1173 memset(info64, 0, sizeof(*info64));
1174 info64->lo_number = info->lo_number;
1175 info64->lo_device = info->lo_device;
1176 info64->lo_inode = info->lo_inode;
1177 info64->lo_rdevice = info->lo_rdevice;
1178 info64->lo_offset = info->lo_offset;
1179 info64->lo_sizelimit = 0;
1180 info64->lo_encrypt_type = info->lo_encrypt_type;
1181 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1182 info64->lo_flags = info->lo_flags;
1183 info64->lo_init[0] = info->lo_init[0];
1184 info64->lo_init[1] = info->lo_init[1];
1185 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1186 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1187 else
1188 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1189 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1190}
1191
1192static int
1193loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1194{
1195 memset(info, 0, sizeof(*info));
1196 info->lo_number = info64->lo_number;
1197 info->lo_device = info64->lo_device;
1198 info->lo_inode = info64->lo_inode;
1199 info->lo_rdevice = info64->lo_rdevice;
1200 info->lo_offset = info64->lo_offset;
1201 info->lo_encrypt_type = info64->lo_encrypt_type;
1202 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1203 info->lo_flags = info64->lo_flags;
1204 info->lo_init[0] = info64->lo_init[0];
1205 info->lo_init[1] = info64->lo_init[1];
1206 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1207 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1208 else
1209 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1210 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1211
1212 /* error in case values were truncated */
1213 if (info->lo_device != info64->lo_device ||
1214 info->lo_rdevice != info64->lo_rdevice ||
1215 info->lo_inode != info64->lo_inode ||
1216 info->lo_offset != info64->lo_offset)
1217 return -EOVERFLOW;
1218
1219 return 0;
1220}
1221
1222static int
1223loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1224{
1225 struct loop_info info;
1226 struct loop_info64 info64;
1227
1228 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1229 return -EFAULT;
1230 loop_info64_from_old(&info, &info64);
1231 return loop_set_status(lo, &info64);
1232}
1233
1234static int
1235loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1236{
1237 struct loop_info64 info64;
1238
1239 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1240 return -EFAULT;
1241 return loop_set_status(lo, &info64);
1242}
1243
1244static int
1245loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1246 struct loop_info info;
1247 struct loop_info64 info64;
1248 int err = 0;
1249
1250 if (!arg)
1251 err = -EINVAL;
1252 if (!err)
1253 err = loop_get_status(lo, &info64);
1254 if (!err)
1255 err = loop_info64_to_old(&info64, &info);
1256 if (!err && copy_to_user(arg, &info, sizeof(info)))
1257 err = -EFAULT;
1258
1259 return err;
1260}
1261
1262static int
1263loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1264 struct loop_info64 info64;
1265 int err = 0;
1266
1267 if (!arg)
1268 err = -EINVAL;
1269 if (!err)
1270 err = loop_get_status(lo, &info64);
1271 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1272 err = -EFAULT;
1273
1274 return err;
1275}
1276
1277static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1278{
1279 int err;
1280 sector_t sec;
1281 loff_t sz;
1282
1283 err = -ENXIO;
1284 if (unlikely(lo->lo_state != Lo_bound))
1285 goto out;
1286 err = figure_loop_size(lo);
1287 if (unlikely(err))
1288 goto out;
1289 sec = get_capacity(lo->lo_disk);
1290 /* the width of sector_t may be narrow for bit-shift */
1291 sz = sec;
1292 sz <<= 9;
1293 mutex_lock(&bdev->bd_mutex);
1294 bd_set_size(bdev, sz);
1295 /* let user-space know about the new size */
1296 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1297 mutex_unlock(&bdev->bd_mutex);
1298
1299 out:
1300 return err;
1301}
1302
1303static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1304 unsigned int cmd, unsigned long arg)
1305{
1306 struct loop_device *lo = bdev->bd_disk->private_data;
1307 int err;
1308
1309 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1310 switch (cmd) {
1311 case LOOP_SET_FD:
1312 err = loop_set_fd(lo, mode, bdev, arg);
1313 break;
1314 case LOOP_CHANGE_FD:
1315 err = loop_change_fd(lo, bdev, arg);
1316 break;
1317 case LOOP_CLR_FD:
1318 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1319 err = loop_clr_fd(lo, bdev);
1320 if (!err)
1321 goto out_unlocked;
1322 break;
1323 case LOOP_SET_STATUS:
1324 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1325 break;
1326 case LOOP_GET_STATUS:
1327 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1328 break;
1329 case LOOP_SET_STATUS64:
1330 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1331 break;
1332 case LOOP_GET_STATUS64:
1333 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1334 break;
1335 case LOOP_SET_CAPACITY:
1336 err = -EPERM;
1337 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1338 err = loop_set_capacity(lo, bdev);
1339 break;
1340 default:
1341 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1342 }
1343 mutex_unlock(&lo->lo_ctl_mutex);
1344
1345out_unlocked:
1346 return err;
1347}
1348
1349#ifdef CONFIG_COMPAT
1350struct compat_loop_info {
1351 compat_int_t lo_number; /* ioctl r/o */
1352 compat_dev_t lo_device; /* ioctl r/o */
1353 compat_ulong_t lo_inode; /* ioctl r/o */
1354 compat_dev_t lo_rdevice; /* ioctl r/o */
1355 compat_int_t lo_offset;
1356 compat_int_t lo_encrypt_type;
1357 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1358 compat_int_t lo_flags; /* ioctl r/o */
1359 char lo_name[LO_NAME_SIZE];
1360 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1361 compat_ulong_t lo_init[2];
1362 char reserved[4];
1363};
1364
1365/*
1366 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1367 * - noinlined to reduce stack space usage in main part of driver
1368 */
1369static noinline int
1370loop_info64_from_compat(const struct compat_loop_info __user *arg,
1371 struct loop_info64 *info64)
1372{
1373 struct compat_loop_info info;
1374
1375 if (copy_from_user(&info, arg, sizeof(info)))
1376 return -EFAULT;
1377
1378 memset(info64, 0, sizeof(*info64));
1379 info64->lo_number = info.lo_number;
1380 info64->lo_device = info.lo_device;
1381 info64->lo_inode = info.lo_inode;
1382 info64->lo_rdevice = info.lo_rdevice;
1383 info64->lo_offset = info.lo_offset;
1384 info64->lo_sizelimit = 0;
1385 info64->lo_encrypt_type = info.lo_encrypt_type;
1386 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1387 info64->lo_flags = info.lo_flags;
1388 info64->lo_init[0] = info.lo_init[0];
1389 info64->lo_init[1] = info.lo_init[1];
1390 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1391 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1392 else
1393 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1394 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1395 return 0;
1396}
1397
1398/*
1399 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1400 * - noinlined to reduce stack space usage in main part of driver
1401 */
1402static noinline int
1403loop_info64_to_compat(const struct loop_info64 *info64,
1404 struct compat_loop_info __user *arg)
1405{
1406 struct compat_loop_info info;
1407
1408 memset(&info, 0, sizeof(info));
1409 info.lo_number = info64->lo_number;
1410 info.lo_device = info64->lo_device;
1411 info.lo_inode = info64->lo_inode;
1412 info.lo_rdevice = info64->lo_rdevice;
1413 info.lo_offset = info64->lo_offset;
1414 info.lo_encrypt_type = info64->lo_encrypt_type;
1415 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1416 info.lo_flags = info64->lo_flags;
1417 info.lo_init[0] = info64->lo_init[0];
1418 info.lo_init[1] = info64->lo_init[1];
1419 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1420 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1421 else
1422 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1423 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1424
1425 /* error in case values were truncated */
1426 if (info.lo_device != info64->lo_device ||
1427 info.lo_rdevice != info64->lo_rdevice ||
1428 info.lo_inode != info64->lo_inode ||
1429 info.lo_offset != info64->lo_offset ||
1430 info.lo_init[0] != info64->lo_init[0] ||
1431 info.lo_init[1] != info64->lo_init[1])
1432 return -EOVERFLOW;
1433
1434 if (copy_to_user(arg, &info, sizeof(info)))
1435 return -EFAULT;
1436 return 0;
1437}
1438
1439static int
1440loop_set_status_compat(struct loop_device *lo,
1441 const struct compat_loop_info __user *arg)
1442{
1443 struct loop_info64 info64;
1444 int ret;
1445
1446 ret = loop_info64_from_compat(arg, &info64);
1447 if (ret < 0)
1448 return ret;
1449 return loop_set_status(lo, &info64);
1450}
1451
1452static int
1453loop_get_status_compat(struct loop_device *lo,
1454 struct compat_loop_info __user *arg)
1455{
1456 struct loop_info64 info64;
1457 int err = 0;
1458
1459 if (!arg)
1460 err = -EINVAL;
1461 if (!err)
1462 err = loop_get_status(lo, &info64);
1463 if (!err)
1464 err = loop_info64_to_compat(&info64, arg);
1465 return err;
1466}
1467
1468static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1469 unsigned int cmd, unsigned long arg)
1470{
1471 struct loop_device *lo = bdev->bd_disk->private_data;
1472 int err;
1473
1474 switch(cmd) {
1475 case LOOP_SET_STATUS:
1476 mutex_lock(&lo->lo_ctl_mutex);
1477 err = loop_set_status_compat(
1478 lo, (const struct compat_loop_info __user *) arg);
1479 mutex_unlock(&lo->lo_ctl_mutex);
1480 break;
1481 case LOOP_GET_STATUS:
1482 mutex_lock(&lo->lo_ctl_mutex);
1483 err = loop_get_status_compat(
1484 lo, (struct compat_loop_info __user *) arg);
1485 mutex_unlock(&lo->lo_ctl_mutex);
1486 break;
1487 case LOOP_SET_CAPACITY:
1488 case LOOP_CLR_FD:
1489 case LOOP_GET_STATUS64:
1490 case LOOP_SET_STATUS64:
1491 arg = (unsigned long) compat_ptr(arg);
1492 case LOOP_SET_FD:
1493 case LOOP_CHANGE_FD:
1494 err = lo_ioctl(bdev, mode, cmd, arg);
1495 break;
1496 default:
1497 err = -ENOIOCTLCMD;
1498 break;
1499 }
1500 return err;
1501}
1502#endif
1503
1504static int lo_open(struct block_device *bdev, fmode_t mode)
1505{
1506 struct loop_device *lo = bdev->bd_disk->private_data;
1507
1508 mutex_lock(&loop_mutex);
1509 mutex_lock(&lo->lo_ctl_mutex);
1510 lo->lo_refcnt++;
1511 mutex_unlock(&lo->lo_ctl_mutex);
1512 mutex_unlock(&loop_mutex);
1513
1514 return 0;
1515}
1516
1517static int lo_release(struct gendisk *disk, fmode_t mode)
1518{
1519 struct loop_device *lo = disk->private_data;
1520 int err;
1521
1522 mutex_lock(&loop_mutex);
1523 mutex_lock(&lo->lo_ctl_mutex);
1524
1525 if (--lo->lo_refcnt)
1526 goto out;
1527
1528 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1529 /*
1530 * In autoclear mode, stop the loop thread
1531 * and remove configuration after last close.
1532 */
1533 err = loop_clr_fd(lo, NULL);
1534 if (!err)
1535 goto out_unlocked;
1536 } else {
1537 /*
1538 * Otherwise keep thread (if running) and config,
1539 * but flush possible ongoing bios in thread.
1540 */
1541 loop_flush(lo);
1542 }
1543
1544out:
1545 mutex_unlock(&lo->lo_ctl_mutex);
1546out_unlocked:
1547 mutex_unlock(&loop_mutex);
1548 return 0;
1549}
1550
1551static const struct block_device_operations lo_fops = {
1552 .owner = THIS_MODULE,
1553 .open = lo_open,
1554 .release = lo_release,
1555 .ioctl = lo_ioctl,
1556#ifdef CONFIG_COMPAT
1557 .compat_ioctl = lo_compat_ioctl,
1558#endif
1559};
1560
1561/*
1562 * And now the modules code and kernel interface.
1563 */
1564static int max_loop;
1565module_param(max_loop, int, 0);
1566MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1567module_param(max_part, int, 0);
1568MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1569MODULE_LICENSE("GPL");
1570MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1571
1572int loop_register_transfer(struct loop_func_table *funcs)
1573{
1574 unsigned int n = funcs->number;
1575
1576 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1577 return -EINVAL;
1578 xfer_funcs[n] = funcs;
1579 return 0;
1580}
1581
1582int loop_unregister_transfer(int number)
1583{
1584 unsigned int n = number;
1585 struct loop_device *lo;
1586 struct loop_func_table *xfer;
1587
1588 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1589 return -EINVAL;
1590
1591 xfer_funcs[n] = NULL;
1592
1593 list_for_each_entry(lo, &loop_devices, lo_list) {
1594 mutex_lock(&lo->lo_ctl_mutex);
1595
1596 if (lo->lo_encryption == xfer)
1597 loop_release_xfer(lo);
1598
1599 mutex_unlock(&lo->lo_ctl_mutex);
1600 }
1601
1602 return 0;
1603}
1604
1605EXPORT_SYMBOL(loop_register_transfer);
1606EXPORT_SYMBOL(loop_unregister_transfer);
1607
1608static struct loop_device *loop_alloc(int i)
1609{
1610 struct loop_device *lo;
1611 struct gendisk *disk;
1612
1613 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1614 if (!lo)
1615 goto out;
1616
1617 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1618 if (!lo->lo_queue)
1619 goto out_free_dev;
1620
1621 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1622 if (!disk)
1623 goto out_free_queue;
1624
1625 mutex_init(&lo->lo_ctl_mutex);
1626 lo->lo_number = i;
1627 lo->lo_thread = NULL;
1628 init_waitqueue_head(&lo->lo_event);
1629 spin_lock_init(&lo->lo_lock);
1630 disk->major = LOOP_MAJOR;
1631 disk->first_minor = i << part_shift;
1632 disk->fops = &lo_fops;
1633 disk->private_data = lo;
1634 disk->queue = lo->lo_queue;
1635 sprintf(disk->disk_name, "loop%d", i);
1636 return lo;
1637
1638out_free_queue:
1639 blk_cleanup_queue(lo->lo_queue);
1640out_free_dev:
1641 kfree(lo);
1642out:
1643 return NULL;
1644}
1645
1646static void loop_free(struct loop_device *lo)
1647{
1648 blk_cleanup_queue(lo->lo_queue);
1649 put_disk(lo->lo_disk);
1650 list_del(&lo->lo_list);
1651 kfree(lo);
1652}
1653
1654static struct loop_device *loop_init_one(int i)
1655{
1656 struct loop_device *lo;
1657
1658 list_for_each_entry(lo, &loop_devices, lo_list) {
1659 if (lo->lo_number == i)
1660 return lo;
1661 }
1662
1663 lo = loop_alloc(i);
1664 if (lo) {
1665 add_disk(lo->lo_disk);
1666 list_add_tail(&lo->lo_list, &loop_devices);
1667 }
1668 return lo;
1669}
1670
1671static void loop_del_one(struct loop_device *lo)
1672{
1673 del_gendisk(lo->lo_disk);
1674 loop_free(lo);
1675}
1676
1677static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1678{
1679 struct loop_device *lo;
1680 struct kobject *kobj;
1681
1682 mutex_lock(&loop_devices_mutex);
1683 lo = loop_init_one(dev & MINORMASK);
1684 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1685 mutex_unlock(&loop_devices_mutex);
1686
1687 *part = 0;
1688 return kobj;
1689}
1690
1691static int __init loop_init(void)
1692{
1693 int i, nr;
1694 unsigned long range;
1695 struct loop_device *lo, *next;
1696
1697 /*
1698 * loop module now has a feature to instantiate underlying device
1699 * structure on-demand, provided that there is an access dev node.
1700 * However, this will not work well with user space tool that doesn't
1701 * know about such "feature". In order to not break any existing
1702 * tool, we do the following:
1703 *
1704 * (1) if max_loop is specified, create that many upfront, and this
1705 * also becomes a hard limit.
1706 * (2) if max_loop is not specified, create 8 loop device on module
1707 * load, user can further extend loop device by create dev node
1708 * themselves and have kernel automatically instantiate actual
1709 * device on-demand.
1710 */
1711
1712 part_shift = 0;
1713 if (max_part > 0)
1714 part_shift = fls(max_part);
1715
1716 if (max_loop > 1UL << (MINORBITS - part_shift))
1717 return -EINVAL;
1718
1719 if (max_loop) {
1720 nr = max_loop;
1721 range = max_loop;
1722 } else {
1723 nr = 8;
1724 range = 1UL << (MINORBITS - part_shift);
1725 }
1726
1727 if (register_blkdev(LOOP_MAJOR, "loop"))
1728 return -EIO;
1729
1730 for (i = 0; i < nr; i++) {
1731 lo = loop_alloc(i);
1732 if (!lo)
1733 goto Enomem;
1734 list_add_tail(&lo->lo_list, &loop_devices);
1735 }
1736
1737 /* point of no return */
1738
1739 list_for_each_entry(lo, &loop_devices, lo_list)
1740 add_disk(lo->lo_disk);
1741
1742 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1743 THIS_MODULE, loop_probe, NULL, NULL);
1744
1745 printk(KERN_INFO "loop: module loaded\n");
1746 return 0;
1747
1748Enomem:
1749 printk(KERN_INFO "loop: out of memory\n");
1750
1751 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1752 loop_free(lo);
1753
1754 unregister_blkdev(LOOP_MAJOR, "loop");
1755 return -ENOMEM;
1756}
1757
1758static void __exit loop_exit(void)
1759{
1760 unsigned long range;
1761 struct loop_device *lo, *next;
1762
1763 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1764
1765 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1766 loop_del_one(lo);
1767
1768 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1769 unregister_blkdev(LOOP_MAJOR, "loop");
1770}
1771
1772module_init(loop_init);
1773module_exit(loop_exit);
1774
1775#ifndef MODULE
1776static int __init max_loop_setup(char *str)
1777{
1778 max_loop = simple_strtol(str, NULL, 0);
1779 return 1;
1780}
1781
1782__setup("max_loop=", max_loop_setup);
1783#endif