projects/linux·blob v3.0-rc2 · refs

1775 lines · 43.0 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 LIST_HEAD(loop_devices);
82static DEFINE_MUTEX(loop_devices_mutex);
83
84static int max_part;
85static int part_shift;
86
87/*
88 * Transfer functions
89 */
90static int transfer_none(struct loop_device *lo, int cmd,
91 struct page *raw_page, unsigned raw_off,
92 struct page *loop_page, unsigned loop_off,
93 int size, sector_t real_block)
94{
95 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
96 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
97
98 if (cmd == READ)
99 memcpy(loop_buf, raw_buf, size);
100 else
101 memcpy(raw_buf, loop_buf, size);
102
103 kunmap_atomic(loop_buf, KM_USER1);
104 kunmap_atomic(raw_buf, KM_USER0);
105 cond_resched();
106 return 0;
107}
108
109static int transfer_xor(struct loop_device *lo, int cmd,
110 struct page *raw_page, unsigned raw_off,
111 struct page *loop_page, unsigned loop_off,
112 int size, sector_t real_block)
113{
114 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
115 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
116 char *in, *out, *key;
117 int i, keysize;
118
119 if (cmd == READ) {
120 in = raw_buf;
121 out = loop_buf;
122 } else {
123 in = loop_buf;
124 out = raw_buf;
125 }
126
127 key = lo->lo_encrypt_key;
128 keysize = lo->lo_encrypt_key_size;
129 for (i = 0; i < size; i++)
130 *out++ = *in++ ^ key[(i & 511) % keysize];
131
132 kunmap_atomic(loop_buf, KM_USER1);
133 kunmap_atomic(raw_buf, KM_USER0);
134 cond_resched();
135 return 0;
136}
137
138static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
139{
140 if (unlikely(info->lo_encrypt_key_size <= 0))
141 return -EINVAL;
142 return 0;
143}
144
145static struct loop_func_table none_funcs = {
146 .number = LO_CRYPT_NONE,
147 .transfer = transfer_none,
148};
149
150static struct loop_func_table xor_funcs = {
151 .number = LO_CRYPT_XOR,
152 .transfer = transfer_xor,
153 .init = xor_init
154};
155
156/* xfer_funcs[0] is special - its release function is never called */
157static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
158 &none_funcs,
159 &xor_funcs
160};
161
162static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163{
164 loff_t size, offset, loopsize;
165
166 /* Compute loopsize in bytes */
167 size = i_size_read(file->f_mapping->host);
168 offset = lo->lo_offset;
169 loopsize = size - offset;
170 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
171 loopsize = lo->lo_sizelimit;
172
173 /*
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
176 */
177 return loopsize >> 9;
178}
179
180static int
181figure_loop_size(struct loop_device *lo)
182{
183 loff_t size = get_loop_size(lo, lo->lo_backing_file);
184 sector_t x = (sector_t)size;
185
186 if (unlikely((loff_t)x != size))
187 return -EFBIG;
188
189 set_capacity(lo->lo_disk, x);
190 return 0;
191}
192
193static inline int
194lo_do_transfer(struct loop_device *lo, int cmd,
195 struct page *rpage, unsigned roffs,
196 struct page *lpage, unsigned loffs,
197 int size, sector_t rblock)
198{
199 if (unlikely(!lo->transfer))
200 return 0;
201
202 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
203}
204
205/**
206 * do_lo_send_aops - helper for writing data to a loop device
207 *
208 * This is the fast version for backing filesystems which implement the address
209 * space operations write_begin and write_end.
210 */
211static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
212 loff_t pos, struct page *unused)
213{
214 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
215 struct address_space *mapping = file->f_mapping;
216 pgoff_t index;
217 unsigned offset, bv_offs;
218 int len, ret;
219
220 mutex_lock(&mapping->host->i_mutex);
221 index = pos >> PAGE_CACHE_SHIFT;
222 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
223 bv_offs = bvec->bv_offset;
224 len = bvec->bv_len;
225 while (len > 0) {
226 sector_t IV;
227 unsigned size, copied;
228 int transfer_result;
229 struct page *page;
230 void *fsdata;
231
232 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
233 size = PAGE_CACHE_SIZE - offset;
234 if (size > len)
235 size = len;
236
237 ret = pagecache_write_begin(file, mapping, pos, size, 0,
238 &page, &fsdata);
239 if (ret)
240 goto fail;
241
242 file_update_time(file);
243
244 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
245 bvec->bv_page, bv_offs, size, IV);
246 copied = size;
247 if (unlikely(transfer_result))
248 copied = 0;
249
250 ret = pagecache_write_end(file, mapping, pos, size, copied,
251 page, fsdata);
252 if (ret < 0 || ret != copied)
253 goto fail;
254
255 if (unlikely(transfer_result))
256 goto fail;
257
258 bv_offs += copied;
259 len -= copied;
260 offset = 0;
261 index++;
262 pos += copied;
263 }
264 ret = 0;
265out:
266 mutex_unlock(&mapping->host->i_mutex);
267 return ret;
268fail:
269 ret = -1;
270 goto out;
271}
272
273/**
274 * __do_lo_send_write - helper for writing data to a loop device
275 *
276 * This helper just factors out common code between do_lo_send_direct_write()
277 * and do_lo_send_write().
278 */
279static int __do_lo_send_write(struct file *file,
280 u8 *buf, const int len, loff_t pos)
281{
282 ssize_t bw;
283 mm_segment_t old_fs = get_fs();
284
285 set_fs(get_ds());
286 bw = file->f_op->write(file, buf, len, &pos);
287 set_fs(old_fs);
288 if (likely(bw == len))
289 return 0;
290 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
291 (unsigned long long)pos, len);
292 if (bw >= 0)
293 bw = -EIO;
294 return bw;
295}
296
297/**
298 * do_lo_send_direct_write - helper for writing data to a loop device
299 *
300 * This is the fast, non-transforming version for backing filesystems which do
301 * not implement the address space operations write_begin and write_end.
302 * It uses the write file operation which should be present on all writeable
303 * filesystems.
304 */
305static int do_lo_send_direct_write(struct loop_device *lo,
306 struct bio_vec *bvec, loff_t pos, struct page *page)
307{
308 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
309 kmap(bvec->bv_page) + bvec->bv_offset,
310 bvec->bv_len, pos);
311 kunmap(bvec->bv_page);
312 cond_resched();
313 return bw;
314}
315
316/**
317 * do_lo_send_write - helper for writing data to a loop device
318 *
319 * This is the slow, transforming version for filesystems which do not
320 * implement the address space operations write_begin and write_end. It
321 * uses the write file operation which should be present on all writeable
322 * filesystems.
323 *
324 * Using fops->write is slower than using aops->{prepare,commit}_write in the
325 * transforming case because we need to double buffer the data as we cannot do
326 * the transformations in place as we do not have direct access to the
327 * destination pages of the backing file.
328 */
329static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
330 loff_t pos, struct page *page)
331{
332 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
333 bvec->bv_offset, bvec->bv_len, pos >> 9);
334 if (likely(!ret))
335 return __do_lo_send_write(lo->lo_backing_file,
336 page_address(page), bvec->bv_len,
337 pos);
338 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
339 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
340 if (ret > 0)
341 ret = -EIO;
342 return ret;
343}
344
345static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
346{
347 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
348 struct page *page);
349 struct bio_vec *bvec;
350 struct page *page = NULL;
351 int i, ret = 0;
352
353 do_lo_send = do_lo_send_aops;
354 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
355 do_lo_send = do_lo_send_direct_write;
356 if (lo->transfer != transfer_none) {
357 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
358 if (unlikely(!page))
359 goto fail;
360 kmap(page);
361 do_lo_send = do_lo_send_write;
362 }
363 }
364 bio_for_each_segment(bvec, bio, i) {
365 ret = do_lo_send(lo, bvec, pos, page);
366 if (ret < 0)
367 break;
368 pos += bvec->bv_len;
369 }
370 if (page) {
371 kunmap(page);
372 __free_page(page);
373 }
374out:
375 return ret;
376fail:
377 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
378 ret = -ENOMEM;
379 goto out;
380}
381
382struct lo_read_data {
383 struct loop_device *lo;
384 struct page *page;
385 unsigned offset;
386 int bsize;
387};
388
389static int
390lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
391 struct splice_desc *sd)
392{
393 struct lo_read_data *p = sd->u.data;
394 struct loop_device *lo = p->lo;
395 struct page *page = buf->page;
396 sector_t IV;
397 int size;
398
399 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
400 (buf->offset >> 9);
401 size = sd->len;
402 if (size > p->bsize)
403 size = p->bsize;
404
405 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
406 printk(KERN_ERR "loop: transfer error block %ld\n",
407 page->index);
408 size = -EINVAL;
409 }
410
411 flush_dcache_page(p->page);
412
413 if (size > 0)
414 p->offset += size;
415
416 return size;
417}
418
419static int
420lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
421{
422 return __splice_from_pipe(pipe, sd, lo_splice_actor);
423}
424
425static int
426do_lo_receive(struct loop_device *lo,
427 struct bio_vec *bvec, int bsize, loff_t pos)
428{
429 struct lo_read_data cookie;
430 struct splice_desc sd;
431 struct file *file;
432 long retval;
433
434 cookie.lo = lo;
435 cookie.page = bvec->bv_page;
436 cookie.offset = bvec->bv_offset;
437 cookie.bsize = bsize;
438
439 sd.len = 0;
440 sd.total_len = bvec->bv_len;
441 sd.flags = 0;
442 sd.pos = pos;
443 sd.u.data = &cookie;
444
445 file = lo->lo_backing_file;
446 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
447
448 if (retval < 0)
449 return retval;
450
451 return 0;
452}
453
454static int
455lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
456{
457 struct bio_vec *bvec;
458 int i, ret = 0;
459
460 bio_for_each_segment(bvec, bio, i) {
461 ret = do_lo_receive(lo, bvec, bsize, pos);
462 if (ret < 0)
463 break;
464 pos += bvec->bv_len;
465 }
466 return ret;
467}
468
469static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
470{
471 loff_t pos;
472 int ret;
473
474 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
475
476 if (bio_rw(bio) == WRITE) {
477 struct file *file = lo->lo_backing_file;
478
479 if (bio->bi_rw & REQ_FLUSH) {
480 ret = vfs_fsync(file, 0);
481 if (unlikely(ret && ret != -EINVAL)) {
482 ret = -EIO;
483 goto out;
484 }
485 }
486
487 ret = lo_send(lo, bio, pos);
488
489 if ((bio->bi_rw & REQ_FUA) && !ret) {
490 ret = vfs_fsync(file, 0);
491 if (unlikely(ret && ret != -EINVAL))
492 ret = -EIO;
493 }
494 } else
495 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
496
497out:
498 return ret;
499}
500
501/*
502 * Add bio to back of pending list
503 */
504static void loop_add_bio(struct loop_device *lo, struct bio *bio)
505{
506 bio_list_add(&lo->lo_bio_list, bio);
507}
508
509/*
510 * Grab first pending buffer
511 */
512static struct bio *loop_get_bio(struct loop_device *lo)
513{
514 return bio_list_pop(&lo->lo_bio_list);
515}
516
517static int loop_make_request(struct request_queue *q, struct bio *old_bio)
518{
519 struct loop_device *lo = q->queuedata;
520 int rw = bio_rw(old_bio);
521
522 if (rw == READA)
523 rw = READ;
524
525 BUG_ON(!lo || (rw != READ && rw != WRITE));
526
527 spin_lock_irq(&lo->lo_lock);
528 if (lo->lo_state != Lo_bound)
529 goto out;
530 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
531 goto out;
532 loop_add_bio(lo, old_bio);
533 wake_up(&lo->lo_event);
534 spin_unlock_irq(&lo->lo_lock);
535 return 0;
536
537out:
538 spin_unlock_irq(&lo->lo_lock);
539 bio_io_error(old_bio);
540 return 0;
541}
542
543struct switch_request {
544 struct file *file;
545 struct completion wait;
546};
547
548static void do_loop_switch(struct loop_device *, struct switch_request *);
549
550static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
551{
552 if (unlikely(!bio->bi_bdev)) {
553 do_loop_switch(lo, bio->bi_private);
554 bio_put(bio);
555 } else {
556 int ret = do_bio_filebacked(lo, bio);
557 bio_endio(bio, ret);
558 }
559}
560
561/*
562 * worker thread that handles reads/writes to file backed loop devices,
563 * to avoid blocking in our make_request_fn. it also does loop decrypting
564 * on reads for block backed loop, as that is too heavy to do from
565 * b_end_io context where irqs may be disabled.
566 *
567 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
568 * calling kthread_stop(). Therefore once kthread_should_stop() is
569 * true, make_request will not place any more requests. Therefore
570 * once kthread_should_stop() is true and lo_bio is NULL, we are
571 * done with the loop.
572 */
573static int loop_thread(void *data)
574{
575 struct loop_device *lo = data;
576 struct bio *bio;
577
578 set_user_nice(current, -20);
579
580 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
581
582 wait_event_interruptible(lo->lo_event,
583 !bio_list_empty(&lo->lo_bio_list) ||
584 kthread_should_stop());
585
586 if (bio_list_empty(&lo->lo_bio_list))
587 continue;
588 spin_lock_irq(&lo->lo_lock);
589 bio = loop_get_bio(lo);
590 spin_unlock_irq(&lo->lo_lock);
591
592 BUG_ON(!bio);
593 loop_handle_bio(lo, bio);
594 }
595
596 return 0;
597}
598
599/*
600 * loop_switch performs the hard work of switching a backing store.
601 * First it needs to flush existing IO, it does this by sending a magic
602 * BIO down the pipe. The completion of this BIO does the actual switch.
603 */
604static int loop_switch(struct loop_device *lo, struct file *file)
605{
606 struct switch_request w;
607 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
608 if (!bio)
609 return -ENOMEM;
610 init_completion(&w.wait);
611 w.file = file;
612 bio->bi_private = &w;
613 bio->bi_bdev = NULL;
614 loop_make_request(lo->lo_queue, bio);
615 wait_for_completion(&w.wait);
616 return 0;
617}
618
619/*
620 * Helper to flush the IOs in loop, but keeping loop thread running
621 */
622static int loop_flush(struct loop_device *lo)
623{
624 /* loop not yet configured, no running thread, nothing to flush */
625 if (!lo->lo_thread)
626 return 0;
627
628 return loop_switch(lo, NULL);
629}
630
631/*
632 * Do the actual switch; called from the BIO completion routine
633 */
634static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
635{
636 struct file *file = p->file;
637 struct file *old_file = lo->lo_backing_file;
638 struct address_space *mapping;
639
640 /* if no new file, only flush of queued bios requested */
641 if (!file)
642 goto out;
643
644 mapping = file->f_mapping;
645 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
646 lo->lo_backing_file = file;
647 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
648 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
649 lo->old_gfp_mask = mapping_gfp_mask(mapping);
650 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
651out:
652 complete(&p->wait);
653}
654
655
656/*
657 * loop_change_fd switched the backing store of a loopback device to
658 * a new file. This is useful for operating system installers to free up
659 * the original file and in High Availability environments to switch to
660 * an alternative location for the content in case of server meltdown.
661 * This can only work if the loop device is used read-only, and if the
662 * new backing store is the same size and type as the old backing store.
663 */
664static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
665 unsigned int arg)
666{
667 struct file *file, *old_file;
668 struct inode *inode;
669 int error;
670
671 error = -ENXIO;
672 if (lo->lo_state != Lo_bound)
673 goto out;
674
675 /* the loop device has to be read-only */
676 error = -EINVAL;
677 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
678 goto out;
679
680 error = -EBADF;
681 file = fget(arg);
682 if (!file)
683 goto out;
684
685 inode = file->f_mapping->host;
686 old_file = lo->lo_backing_file;
687
688 error = -EINVAL;
689
690 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
691 goto out_putf;
692
693 /* size of the new backing store needs to be the same */
694 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
695 goto out_putf;
696
697 /* and ... switch */
698 error = loop_switch(lo, file);
699 if (error)
700 goto out_putf;
701
702 fput(old_file);
703 if (max_part > 0)
704 ioctl_by_bdev(bdev, BLKRRPART, 0);
705 return 0;
706
707 out_putf:
708 fput(file);
709 out:
710 return error;
711}
712
713static inline int is_loop_device(struct file *file)
714{
715 struct inode *i = file->f_mapping->host;
716
717 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
718}
719
720/* loop sysfs attributes */
721
722static ssize_t loop_attr_show(struct device *dev, char *page,
723 ssize_t (*callback)(struct loop_device *, char *))
724{
725 struct loop_device *l, *lo = NULL;
726
727 mutex_lock(&loop_devices_mutex);
728 list_for_each_entry(l, &loop_devices, lo_list)
729 if (disk_to_dev(l->lo_disk) == dev) {
730 lo = l;
731 break;
732 }
733 mutex_unlock(&loop_devices_mutex);
734
735 return lo ? callback(lo, page) : -EIO;
736}
737
738#define LOOP_ATTR_RO(_name) \
739static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
740static ssize_t loop_attr_do_show_##_name(struct device *d, \
741 struct device_attribute *attr, char *b) \
742{ \
743 return loop_attr_show(d, b, loop_attr_##_name##_show); \
744} \
745static struct device_attribute loop_attr_##_name = \
746 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
747
748static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
749{
750 ssize_t ret;
751 char *p = NULL;
752
753 mutex_lock(&lo->lo_ctl_mutex);
754 if (lo->lo_backing_file)
755 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
756 mutex_unlock(&lo->lo_ctl_mutex);
757
758 if (IS_ERR_OR_NULL(p))
759 ret = PTR_ERR(p);
760 else {
761 ret = strlen(p);
762 memmove(buf, p, ret);
763 buf[ret++] = '\n';
764 buf[ret] = 0;
765 }
766
767 return ret;
768}
769
770static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
771{
772 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
773}
774
775static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
776{
777 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
778}
779
780static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
781{
782 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
783
784 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
785}
786
787LOOP_ATTR_RO(backing_file);
788LOOP_ATTR_RO(offset);
789LOOP_ATTR_RO(sizelimit);
790LOOP_ATTR_RO(autoclear);
791
792static struct attribute *loop_attrs[] = {
793 &loop_attr_backing_file.attr,
794 &loop_attr_offset.attr,
795 &loop_attr_sizelimit.attr,
796 &loop_attr_autoclear.attr,
797 NULL,
798};
799
800static struct attribute_group loop_attribute_group = {
801 .name = "loop",
802 .attrs= loop_attrs,
803};
804
805static int loop_sysfs_init(struct loop_device *lo)
806{
807 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
808 &loop_attribute_group);
809}
810
811static void loop_sysfs_exit(struct loop_device *lo)
812{
813 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
814 &loop_attribute_group);
815}
816
817static int loop_set_fd(struct loop_device *lo, fmode_t mode,
818 struct block_device *bdev, unsigned int arg)
819{
820 struct file *file, *f;
821 struct inode *inode;
822 struct address_space *mapping;
823 unsigned lo_blocksize;
824 int lo_flags = 0;
825 int error;
826 loff_t size;
827
828 /* This is safe, since we have a reference from open(). */
829 __module_get(THIS_MODULE);
830
831 error = -EBADF;
832 file = fget(arg);
833 if (!file)
834 goto out;
835
836 error = -EBUSY;
837 if (lo->lo_state != Lo_unbound)
838 goto out_putf;
839
840 /* Avoid recursion */
841 f = file;
842 while (is_loop_device(f)) {
843 struct loop_device *l;
844
845 if (f->f_mapping->host->i_bdev == bdev)
846 goto out_putf;
847
848 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
849 if (l->lo_state == Lo_unbound) {
850 error = -EINVAL;
851 goto out_putf;
852 }
853 f = l->lo_backing_file;
854 }
855
856 mapping = file->f_mapping;
857 inode = mapping->host;
858
859 if (!(file->f_mode & FMODE_WRITE))
860 lo_flags |= LO_FLAGS_READ_ONLY;
861
862 error = -EINVAL;
863 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
864 const struct address_space_operations *aops = mapping->a_ops;
865
866 if (aops->write_begin)
867 lo_flags |= LO_FLAGS_USE_AOPS;
868 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
869 lo_flags |= LO_FLAGS_READ_ONLY;
870
871 lo_blocksize = S_ISBLK(inode->i_mode) ?
872 inode->i_bdev->bd_block_size : PAGE_SIZE;
873
874 error = 0;
875 } else {
876 goto out_putf;
877 }
878
879 size = get_loop_size(lo, file);
880
881 if ((loff_t)(sector_t)size != size) {
882 error = -EFBIG;
883 goto out_putf;
884 }
885
886 if (!(mode & FMODE_WRITE))
887 lo_flags |= LO_FLAGS_READ_ONLY;
888
889 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
890
891 lo->lo_blocksize = lo_blocksize;
892 lo->lo_device = bdev;
893 lo->lo_flags = lo_flags;
894 lo->lo_backing_file = file;
895 lo->transfer = transfer_none;
896 lo->ioctl = NULL;
897 lo->lo_sizelimit = 0;
898 lo->old_gfp_mask = mapping_gfp_mask(mapping);
899 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
900
901 bio_list_init(&lo->lo_bio_list);
902
903 /*
904 * set queue make_request_fn, and add limits based on lower level
905 * device
906 */
907 blk_queue_make_request(lo->lo_queue, loop_make_request);
908 lo->lo_queue->queuedata = lo;
909
910 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
911 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
912
913 set_capacity(lo->lo_disk, size);
914 bd_set_size(bdev, size << 9);
915 loop_sysfs_init(lo);
916 /* let user-space know about the new size */
917 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
918
919 set_blocksize(bdev, lo_blocksize);
920
921 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
922 lo->lo_number);
923 if (IS_ERR(lo->lo_thread)) {
924 error = PTR_ERR(lo->lo_thread);
925 goto out_clr;
926 }
927 lo->lo_state = Lo_bound;
928 wake_up_process(lo->lo_thread);
929 if (max_part > 0)
930 ioctl_by_bdev(bdev, BLKRRPART, 0);
931 return 0;
932
933out_clr:
934 loop_sysfs_exit(lo);
935 lo->lo_thread = NULL;
936 lo->lo_device = NULL;
937 lo->lo_backing_file = NULL;
938 lo->lo_flags = 0;
939 set_capacity(lo->lo_disk, 0);
940 invalidate_bdev(bdev);
941 bd_set_size(bdev, 0);
942 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
943 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
944 lo->lo_state = Lo_unbound;
945 out_putf:
946 fput(file);
947 out:
948 /* This is safe: open() is still holding a reference. */
949 module_put(THIS_MODULE);
950 return error;
951}
952
953static int
954loop_release_xfer(struct loop_device *lo)
955{
956 int err = 0;
957 struct loop_func_table *xfer = lo->lo_encryption;
958
959 if (xfer) {
960 if (xfer->release)
961 err = xfer->release(lo);
962 lo->transfer = NULL;
963 lo->lo_encryption = NULL;
964 module_put(xfer->owner);
965 }
966 return err;
967}
968
969static int
970loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
971 const struct loop_info64 *i)
972{
973 int err = 0;
974
975 if (xfer) {
976 struct module *owner = xfer->owner;
977
978 if (!try_module_get(owner))
979 return -EINVAL;
980 if (xfer->init)
981 err = xfer->init(lo, i);
982 if (err)
983 module_put(owner);
984 else
985 lo->lo_encryption = xfer;
986 }
987 return err;
988}
989
990static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
991{
992 struct file *filp = lo->lo_backing_file;
993 gfp_t gfp = lo->old_gfp_mask;
994
995 if (lo->lo_state != Lo_bound)
996 return -ENXIO;
997
998 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
999 return -EBUSY;
1000
1001 if (filp == NULL)
1002 return -EINVAL;
1003
1004 spin_lock_irq(&lo->lo_lock);
1005 lo->lo_state = Lo_rundown;
1006 spin_unlock_irq(&lo->lo_lock);
1007
1008 kthread_stop(lo->lo_thread);
1009
1010 lo->lo_backing_file = NULL;
1011
1012 loop_release_xfer(lo);
1013 lo->transfer = NULL;
1014 lo->ioctl = NULL;
1015 lo->lo_device = NULL;
1016 lo->lo_encryption = NULL;
1017 lo->lo_offset = 0;
1018 lo->lo_sizelimit = 0;
1019 lo->lo_encrypt_key_size = 0;
1020 lo->lo_flags = 0;
1021 lo->lo_thread = NULL;
1022 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1023 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1024 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1025 if (bdev)
1026 invalidate_bdev(bdev);
1027 set_capacity(lo->lo_disk, 0);
1028 loop_sysfs_exit(lo);
1029 if (bdev) {
1030 bd_set_size(bdev, 0);
1031 /* let user-space know about this change */
1032 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1033 }
1034 mapping_set_gfp_mask(filp->f_mapping, gfp);
1035 lo->lo_state = Lo_unbound;
1036 /* This is safe: open() is still holding a reference. */
1037 module_put(THIS_MODULE);
1038 if (max_part > 0 && bdev)
1039 ioctl_by_bdev(bdev, BLKRRPART, 0);
1040 mutex_unlock(&lo->lo_ctl_mutex);
1041 /*
1042 * Need not hold lo_ctl_mutex to fput backing file.
1043 * Calling fput holding lo_ctl_mutex triggers a circular
1044 * lock dependency possibility warning as fput can take
1045 * bd_mutex which is usually taken before lo_ctl_mutex.
1046 */
1047 fput(filp);
1048 return 0;
1049}
1050
1051static int
1052loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1053{
1054 int err;
1055 struct loop_func_table *xfer;
1056 uid_t uid = current_uid();
1057
1058 if (lo->lo_encrypt_key_size &&
1059 lo->lo_key_owner != uid &&
1060 !capable(CAP_SYS_ADMIN))
1061 return -EPERM;
1062 if (lo->lo_state != Lo_bound)
1063 return -ENXIO;
1064 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1065 return -EINVAL;
1066
1067 err = loop_release_xfer(lo);
1068 if (err)
1069 return err;
1070
1071 if (info->lo_encrypt_type) {
1072 unsigned int type = info->lo_encrypt_type;
1073
1074 if (type >= MAX_LO_CRYPT)
1075 return -EINVAL;
1076 xfer = xfer_funcs[type];
1077 if (xfer == NULL)
1078 return -EINVAL;
1079 } else
1080 xfer = NULL;
1081
1082 err = loop_init_xfer(lo, xfer, info);
1083 if (err)
1084 return err;
1085
1086 if (lo->lo_offset != info->lo_offset ||
1087 lo->lo_sizelimit != info->lo_sizelimit) {
1088 lo->lo_offset = info->lo_offset;
1089 lo->lo_sizelimit = info->lo_sizelimit;
1090 if (figure_loop_size(lo))
1091 return -EFBIG;
1092 }
1093
1094 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1095 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1096 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1097 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1098
1099 if (!xfer)
1100 xfer = &none_funcs;
1101 lo->transfer = xfer->transfer;
1102 lo->ioctl = xfer->ioctl;
1103
1104 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1105 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1106 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1107
1108 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1109 lo->lo_init[0] = info->lo_init[0];
1110 lo->lo_init[1] = info->lo_init[1];
1111 if (info->lo_encrypt_key_size) {
1112 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1113 info->lo_encrypt_key_size);
1114 lo->lo_key_owner = uid;
1115 }
1116
1117 return 0;
1118}
1119
1120static int
1121loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1122{
1123 struct file *file = lo->lo_backing_file;
1124 struct kstat stat;
1125 int error;
1126
1127 if (lo->lo_state != Lo_bound)
1128 return -ENXIO;
1129 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1130 if (error)
1131 return error;
1132 memset(info, 0, sizeof(*info));
1133 info->lo_number = lo->lo_number;
1134 info->lo_device = huge_encode_dev(stat.dev);
1135 info->lo_inode = stat.ino;
1136 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1137 info->lo_offset = lo->lo_offset;
1138 info->lo_sizelimit = lo->lo_sizelimit;
1139 info->lo_flags = lo->lo_flags;
1140 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1141 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1142 info->lo_encrypt_type =
1143 lo->lo_encryption ? lo->lo_encryption->number : 0;
1144 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1145 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1146 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1147 lo->lo_encrypt_key_size);
1148 }
1149 return 0;
1150}
1151
1152static void
1153loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1154{
1155 memset(info64, 0, sizeof(*info64));
1156 info64->lo_number = info->lo_number;
1157 info64->lo_device = info->lo_device;
1158 info64->lo_inode = info->lo_inode;
1159 info64->lo_rdevice = info->lo_rdevice;
1160 info64->lo_offset = info->lo_offset;
1161 info64->lo_sizelimit = 0;
1162 info64->lo_encrypt_type = info->lo_encrypt_type;
1163 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1164 info64->lo_flags = info->lo_flags;
1165 info64->lo_init[0] = info->lo_init[0];
1166 info64->lo_init[1] = info->lo_init[1];
1167 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1168 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1169 else
1170 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1171 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1172}
1173
1174static int
1175loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1176{
1177 memset(info, 0, sizeof(*info));
1178 info->lo_number = info64->lo_number;
1179 info->lo_device = info64->lo_device;
1180 info->lo_inode = info64->lo_inode;
1181 info->lo_rdevice = info64->lo_rdevice;
1182 info->lo_offset = info64->lo_offset;
1183 info->lo_encrypt_type = info64->lo_encrypt_type;
1184 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1185 info->lo_flags = info64->lo_flags;
1186 info->lo_init[0] = info64->lo_init[0];
1187 info->lo_init[1] = info64->lo_init[1];
1188 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1189 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1190 else
1191 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1192 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1193
1194 /* error in case values were truncated */
1195 if (info->lo_device != info64->lo_device ||
1196 info->lo_rdevice != info64->lo_rdevice ||
1197 info->lo_inode != info64->lo_inode ||
1198 info->lo_offset != info64->lo_offset)
1199 return -EOVERFLOW;
1200
1201 return 0;
1202}
1203
1204static int
1205loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1206{
1207 struct loop_info info;
1208 struct loop_info64 info64;
1209
1210 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1211 return -EFAULT;
1212 loop_info64_from_old(&info, &info64);
1213 return loop_set_status(lo, &info64);
1214}
1215
1216static int
1217loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1218{
1219 struct loop_info64 info64;
1220
1221 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1222 return -EFAULT;
1223 return loop_set_status(lo, &info64);
1224}
1225
1226static int
1227loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1228 struct loop_info info;
1229 struct loop_info64 info64;
1230 int err = 0;
1231
1232 if (!arg)
1233 err = -EINVAL;
1234 if (!err)
1235 err = loop_get_status(lo, &info64);
1236 if (!err)
1237 err = loop_info64_to_old(&info64, &info);
1238 if (!err && copy_to_user(arg, &info, sizeof(info)))
1239 err = -EFAULT;
1240
1241 return err;
1242}
1243
1244static int
1245loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1246 struct loop_info64 info64;
1247 int err = 0;
1248
1249 if (!arg)
1250 err = -EINVAL;
1251 if (!err)
1252 err = loop_get_status(lo, &info64);
1253 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1254 err = -EFAULT;
1255
1256 return err;
1257}
1258
1259static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1260{
1261 int err;
1262 sector_t sec;
1263 loff_t sz;
1264
1265 err = -ENXIO;
1266 if (unlikely(lo->lo_state != Lo_bound))
1267 goto out;
1268 err = figure_loop_size(lo);
1269 if (unlikely(err))
1270 goto out;
1271 sec = get_capacity(lo->lo_disk);
1272 /* the width of sector_t may be narrow for bit-shift */
1273 sz = sec;
1274 sz <<= 9;
1275 mutex_lock(&bdev->bd_mutex);
1276 bd_set_size(bdev, sz);
1277 /* let user-space know about the new size */
1278 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1279 mutex_unlock(&bdev->bd_mutex);
1280
1281 out:
1282 return err;
1283}
1284
1285static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1286 unsigned int cmd, unsigned long arg)
1287{
1288 struct loop_device *lo = bdev->bd_disk->private_data;
1289 int err;
1290
1291 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1292 switch (cmd) {
1293 case LOOP_SET_FD:
1294 err = loop_set_fd(lo, mode, bdev, arg);
1295 break;
1296 case LOOP_CHANGE_FD:
1297 err = loop_change_fd(lo, bdev, arg);
1298 break;
1299 case LOOP_CLR_FD:
1300 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1301 err = loop_clr_fd(lo, bdev);
1302 if (!err)
1303 goto out_unlocked;
1304 break;
1305 case LOOP_SET_STATUS:
1306 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1307 break;
1308 case LOOP_GET_STATUS:
1309 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1310 break;
1311 case LOOP_SET_STATUS64:
1312 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1313 break;
1314 case LOOP_GET_STATUS64:
1315 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1316 break;
1317 case LOOP_SET_CAPACITY:
1318 err = -EPERM;
1319 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1320 err = loop_set_capacity(lo, bdev);
1321 break;
1322 default:
1323 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1324 }
1325 mutex_unlock(&lo->lo_ctl_mutex);
1326
1327out_unlocked:
1328 return err;
1329}
1330
1331#ifdef CONFIG_COMPAT
1332struct compat_loop_info {
1333 compat_int_t lo_number; /* ioctl r/o */
1334 compat_dev_t lo_device; /* ioctl r/o */
1335 compat_ulong_t lo_inode; /* ioctl r/o */
1336 compat_dev_t lo_rdevice; /* ioctl r/o */
1337 compat_int_t lo_offset;
1338 compat_int_t lo_encrypt_type;
1339 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1340 compat_int_t lo_flags; /* ioctl r/o */
1341 char lo_name[LO_NAME_SIZE];
1342 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1343 compat_ulong_t lo_init[2];
1344 char reserved[4];
1345};
1346
1347/*
1348 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1349 * - noinlined to reduce stack space usage in main part of driver
1350 */
1351static noinline int
1352loop_info64_from_compat(const struct compat_loop_info __user *arg,
1353 struct loop_info64 *info64)
1354{
1355 struct compat_loop_info info;
1356
1357 if (copy_from_user(&info, arg, sizeof(info)))
1358 return -EFAULT;
1359
1360 memset(info64, 0, sizeof(*info64));
1361 info64->lo_number = info.lo_number;
1362 info64->lo_device = info.lo_device;
1363 info64->lo_inode = info.lo_inode;
1364 info64->lo_rdevice = info.lo_rdevice;
1365 info64->lo_offset = info.lo_offset;
1366 info64->lo_sizelimit = 0;
1367 info64->lo_encrypt_type = info.lo_encrypt_type;
1368 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1369 info64->lo_flags = info.lo_flags;
1370 info64->lo_init[0] = info.lo_init[0];
1371 info64->lo_init[1] = info.lo_init[1];
1372 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1373 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1374 else
1375 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1376 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1377 return 0;
1378}
1379
1380/*
1381 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1382 * - noinlined to reduce stack space usage in main part of driver
1383 */
1384static noinline int
1385loop_info64_to_compat(const struct loop_info64 *info64,
1386 struct compat_loop_info __user *arg)
1387{
1388 struct compat_loop_info info;
1389
1390 memset(&info, 0, sizeof(info));
1391 info.lo_number = info64->lo_number;
1392 info.lo_device = info64->lo_device;
1393 info.lo_inode = info64->lo_inode;
1394 info.lo_rdevice = info64->lo_rdevice;
1395 info.lo_offset = info64->lo_offset;
1396 info.lo_encrypt_type = info64->lo_encrypt_type;
1397 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1398 info.lo_flags = info64->lo_flags;
1399 info.lo_init[0] = info64->lo_init[0];
1400 info.lo_init[1] = info64->lo_init[1];
1401 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1402 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1403 else
1404 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1405 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1406
1407 /* error in case values were truncated */
1408 if (info.lo_device != info64->lo_device ||
1409 info.lo_rdevice != info64->lo_rdevice ||
1410 info.lo_inode != info64->lo_inode ||
1411 info.lo_offset != info64->lo_offset ||
1412 info.lo_init[0] != info64->lo_init[0] ||
1413 info.lo_init[1] != info64->lo_init[1])
1414 return -EOVERFLOW;
1415
1416 if (copy_to_user(arg, &info, sizeof(info)))
1417 return -EFAULT;
1418 return 0;
1419}
1420
1421static int
1422loop_set_status_compat(struct loop_device *lo,
1423 const struct compat_loop_info __user *arg)
1424{
1425 struct loop_info64 info64;
1426 int ret;
1427
1428 ret = loop_info64_from_compat(arg, &info64);
1429 if (ret < 0)
1430 return ret;
1431 return loop_set_status(lo, &info64);
1432}
1433
1434static int
1435loop_get_status_compat(struct loop_device *lo,
1436 struct compat_loop_info __user *arg)
1437{
1438 struct loop_info64 info64;
1439 int err = 0;
1440
1441 if (!arg)
1442 err = -EINVAL;
1443 if (!err)
1444 err = loop_get_status(lo, &info64);
1445 if (!err)
1446 err = loop_info64_to_compat(&info64, arg);
1447 return err;
1448}
1449
1450static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1451 unsigned int cmd, unsigned long arg)
1452{
1453 struct loop_device *lo = bdev->bd_disk->private_data;
1454 int err;
1455
1456 switch(cmd) {
1457 case LOOP_SET_STATUS:
1458 mutex_lock(&lo->lo_ctl_mutex);
1459 err = loop_set_status_compat(
1460 lo, (const struct compat_loop_info __user *) arg);
1461 mutex_unlock(&lo->lo_ctl_mutex);
1462 break;
1463 case LOOP_GET_STATUS:
1464 mutex_lock(&lo->lo_ctl_mutex);
1465 err = loop_get_status_compat(
1466 lo, (struct compat_loop_info __user *) arg);
1467 mutex_unlock(&lo->lo_ctl_mutex);
1468 break;
1469 case LOOP_SET_CAPACITY:
1470 case LOOP_CLR_FD:
1471 case LOOP_GET_STATUS64:
1472 case LOOP_SET_STATUS64:
1473 arg = (unsigned long) compat_ptr(arg);
1474 case LOOP_SET_FD:
1475 case LOOP_CHANGE_FD:
1476 err = lo_ioctl(bdev, mode, cmd, arg);
1477 break;
1478 default:
1479 err = -ENOIOCTLCMD;
1480 break;
1481 }
1482 return err;
1483}
1484#endif
1485
1486static int lo_open(struct block_device *bdev, fmode_t mode)
1487{
1488 struct loop_device *lo = bdev->bd_disk->private_data;
1489
1490 mutex_lock(&lo->lo_ctl_mutex);
1491 lo->lo_refcnt++;
1492 mutex_unlock(&lo->lo_ctl_mutex);
1493
1494 return 0;
1495}
1496
1497static int lo_release(struct gendisk *disk, fmode_t mode)
1498{
1499 struct loop_device *lo = disk->private_data;
1500 int err;
1501
1502 mutex_lock(&lo->lo_ctl_mutex);
1503
1504 if (--lo->lo_refcnt)
1505 goto out;
1506
1507 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1508 /*
1509 * In autoclear mode, stop the loop thread
1510 * and remove configuration after last close.
1511 */
1512 err = loop_clr_fd(lo, NULL);
1513 if (!err)
1514 goto out_unlocked;
1515 } else {
1516 /*
1517 * Otherwise keep thread (if running) and config,
1518 * but flush possible ongoing bios in thread.
1519 */
1520 loop_flush(lo);
1521 }
1522
1523out:
1524 mutex_unlock(&lo->lo_ctl_mutex);
1525out_unlocked:
1526 return 0;
1527}
1528
1529static const struct block_device_operations lo_fops = {
1530 .owner = THIS_MODULE,
1531 .open = lo_open,
1532 .release = lo_release,
1533 .ioctl = lo_ioctl,
1534#ifdef CONFIG_COMPAT
1535 .compat_ioctl = lo_compat_ioctl,
1536#endif
1537};
1538
1539/*
1540 * And now the modules code and kernel interface.
1541 */
1542static int max_loop;
1543module_param(max_loop, int, S_IRUGO);
1544MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1545module_param(max_part, int, S_IRUGO);
1546MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1547MODULE_LICENSE("GPL");
1548MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1549
1550int loop_register_transfer(struct loop_func_table *funcs)
1551{
1552 unsigned int n = funcs->number;
1553
1554 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1555 return -EINVAL;
1556 xfer_funcs[n] = funcs;
1557 return 0;
1558}
1559
1560int loop_unregister_transfer(int number)
1561{
1562 unsigned int n = number;
1563 struct loop_device *lo;
1564 struct loop_func_table *xfer;
1565
1566 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1567 return -EINVAL;
1568
1569 xfer_funcs[n] = NULL;
1570
1571 list_for_each_entry(lo, &loop_devices, lo_list) {
1572 mutex_lock(&lo->lo_ctl_mutex);
1573
1574 if (lo->lo_encryption == xfer)
1575 loop_release_xfer(lo);
1576
1577 mutex_unlock(&lo->lo_ctl_mutex);
1578 }
1579
1580 return 0;
1581}
1582
1583EXPORT_SYMBOL(loop_register_transfer);
1584EXPORT_SYMBOL(loop_unregister_transfer);
1585
1586static struct loop_device *loop_alloc(int i)
1587{
1588 struct loop_device *lo;
1589 struct gendisk *disk;
1590
1591 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1592 if (!lo)
1593 goto out;
1594
1595 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1596 if (!lo->lo_queue)
1597 goto out_free_dev;
1598
1599 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1600 if (!disk)
1601 goto out_free_queue;
1602
1603 mutex_init(&lo->lo_ctl_mutex);
1604 lo->lo_number = i;
1605 lo->lo_thread = NULL;
1606 init_waitqueue_head(&lo->lo_event);
1607 spin_lock_init(&lo->lo_lock);
1608 disk->major = LOOP_MAJOR;
1609 disk->first_minor = i << part_shift;
1610 disk->fops = &lo_fops;
1611 disk->private_data = lo;
1612 disk->queue = lo->lo_queue;
1613 sprintf(disk->disk_name, "loop%d", i);
1614 return lo;
1615
1616out_free_queue:
1617 blk_cleanup_queue(lo->lo_queue);
1618out_free_dev:
1619 kfree(lo);
1620out:
1621 return NULL;
1622}
1623
1624static void loop_free(struct loop_device *lo)
1625{
1626 blk_cleanup_queue(lo->lo_queue);
1627 put_disk(lo->lo_disk);
1628 list_del(&lo->lo_list);
1629 kfree(lo);
1630}
1631
1632static struct loop_device *loop_init_one(int i)
1633{
1634 struct loop_device *lo;
1635
1636 list_for_each_entry(lo, &loop_devices, lo_list) {
1637 if (lo->lo_number == i)
1638 return lo;
1639 }
1640
1641 lo = loop_alloc(i);
1642 if (lo) {
1643 add_disk(lo->lo_disk);
1644 list_add_tail(&lo->lo_list, &loop_devices);
1645 }
1646 return lo;
1647}
1648
1649static void loop_del_one(struct loop_device *lo)
1650{
1651 del_gendisk(lo->lo_disk);
1652 loop_free(lo);
1653}
1654
1655static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1656{
1657 struct loop_device *lo;
1658 struct kobject *kobj;
1659
1660 mutex_lock(&loop_devices_mutex);
1661 lo = loop_init_one(MINOR(dev) >> part_shift);
1662 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1663 mutex_unlock(&loop_devices_mutex);
1664
1665 *part = 0;
1666 return kobj;
1667}
1668
1669static int __init loop_init(void)
1670{
1671 int i, nr;
1672 unsigned long range;
1673 struct loop_device *lo, *next;
1674
1675 /*
1676 * loop module now has a feature to instantiate underlying device
1677 * structure on-demand, provided that there is an access dev node.
1678 * However, this will not work well with user space tool that doesn't
1679 * know about such "feature". In order to not break any existing
1680 * tool, we do the following:
1681 *
1682 * (1) if max_loop is specified, create that many upfront, and this
1683 * also becomes a hard limit.
1684 * (2) if max_loop is not specified, create 8 loop device on module
1685 * load, user can further extend loop device by create dev node
1686 * themselves and have kernel automatically instantiate actual
1687 * device on-demand.
1688 */
1689
1690 part_shift = 0;
1691 if (max_part > 0) {
1692 part_shift = fls(max_part);
1693
1694 /*
1695 * Adjust max_part according to part_shift as it is exported
1696 * to user space so that user can decide correct minor number
1697 * if [s]he want to create more devices.
1698 *
1699 * Note that -1 is required because partition 0 is reserved
1700 * for the whole disk.
1701 */
1702 max_part = (1UL << part_shift) - 1;
1703 }
1704
1705 if ((1UL << part_shift) > DISK_MAX_PARTS)
1706 return -EINVAL;
1707
1708 if (max_loop > 1UL << (MINORBITS - part_shift))
1709 return -EINVAL;
1710
1711 if (max_loop) {
1712 nr = max_loop;
1713 range = max_loop << part_shift;
1714 } else {
1715 nr = 8;
1716 range = 1UL << MINORBITS;
1717 }
1718
1719 if (register_blkdev(LOOP_MAJOR, "loop"))
1720 return -EIO;
1721
1722 for (i = 0; i < nr; i++) {
1723 lo = loop_alloc(i);
1724 if (!lo)
1725 goto Enomem;
1726 list_add_tail(&lo->lo_list, &loop_devices);
1727 }
1728
1729 /* point of no return */
1730
1731 list_for_each_entry(lo, &loop_devices, lo_list)
1732 add_disk(lo->lo_disk);
1733
1734 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1735 THIS_MODULE, loop_probe, NULL, NULL);
1736
1737 printk(KERN_INFO "loop: module loaded\n");
1738 return 0;
1739
1740Enomem:
1741 printk(KERN_INFO "loop: out of memory\n");
1742
1743 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1744 loop_free(lo);
1745
1746 unregister_blkdev(LOOP_MAJOR, "loop");
1747 return -ENOMEM;
1748}
1749
1750static void __exit loop_exit(void)
1751{
1752 unsigned long range;
1753 struct loop_device *lo, *next;
1754
1755 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1756
1757 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1758 loop_del_one(lo);
1759
1760 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1761 unregister_blkdev(LOOP_MAJOR, "loop");
1762}
1763
1764module_init(loop_init);
1765module_exit(loop_exit);
1766
1767#ifndef MODULE
1768static int __init max_loop_setup(char *str)
1769{
1770 max_loop = simple_strtol(str, NULL, 0);
1771 return 1;
1772}
1773
1774__setup("max_loop=", max_loop_setup);
1775#endif