Loading...
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 - Vincent.Renardias@waw.com Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
33 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <axboe@suse.de>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations 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/completion.h>
73#include <linux/highmem.h>
74#include <linux/kthread.h>
75#include <linux/splice.h>
76#include <linux/sysfs.h>
77#include <linux/miscdevice.h>
78#include <linux/falloc.h>
79
80#include <asm/uaccess.h>
81
82static DEFINE_IDR(loop_index_idr);
83static DEFINE_MUTEX(loop_index_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) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + 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);
105 kunmap_atomic(raw_buf);
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) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page) + 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);
134 kunmap_atomic(raw_buf);
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_size(loff_t offset, loff_t sizelimit, struct file *file)
164{
165 loff_t size, loopsize;
166
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 loopsize = size - offset;
170 /* offset is beyond i_size, wierd but possible */
171 if (loopsize < 0)
172 return 0;
173
174 if (sizelimit > 0 && sizelimit < loopsize)
175 loopsize = sizelimit;
176 /*
177 * Unfortunately, if we want to do I/O on the device,
178 * the number of 512-byte sectors has to fit into a sector_t.
179 */
180 return loopsize >> 9;
181}
182
183static loff_t get_loop_size(struct loop_device *lo, struct file *file)
184{
185 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
186}
187
188static int
189figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
190{
191 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
192 sector_t x = (sector_t)size;
193
194 if (unlikely((loff_t)x != size))
195 return -EFBIG;
196 if (lo->lo_offset != offset)
197 lo->lo_offset = offset;
198 if (lo->lo_sizelimit != sizelimit)
199 lo->lo_sizelimit = sizelimit;
200 set_capacity(lo->lo_disk, x);
201 return 0;
202}
203
204static inline int
205lo_do_transfer(struct loop_device *lo, int cmd,
206 struct page *rpage, unsigned roffs,
207 struct page *lpage, unsigned loffs,
208 int size, sector_t rblock)
209{
210 if (unlikely(!lo->transfer))
211 return 0;
212
213 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
214}
215
216/**
217 * __do_lo_send_write - helper for writing data to a loop device
218 *
219 * This helper just factors out common code between do_lo_send_direct_write()
220 * and do_lo_send_write().
221 */
222static int __do_lo_send_write(struct file *file,
223 u8 *buf, const int len, loff_t pos)
224{
225 ssize_t bw;
226 mm_segment_t old_fs = get_fs();
227
228 set_fs(get_ds());
229 bw = file->f_op->write(file, buf, len, &pos);
230 set_fs(old_fs);
231 if (likely(bw == len))
232 return 0;
233 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
234 (unsigned long long)pos, len);
235 if (bw >= 0)
236 bw = -EIO;
237 return bw;
238}
239
240/**
241 * do_lo_send_direct_write - helper for writing data to a loop device
242 *
243 * This is the fast, non-transforming version that does not need double
244 * buffering.
245 */
246static int do_lo_send_direct_write(struct loop_device *lo,
247 struct bio_vec *bvec, loff_t pos, struct page *page)
248{
249 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
250 kmap(bvec->bv_page) + bvec->bv_offset,
251 bvec->bv_len, pos);
252 kunmap(bvec->bv_page);
253 cond_resched();
254 return bw;
255}
256
257/**
258 * do_lo_send_write - helper for writing data to a loop device
259 *
260 * This is the slow, transforming version that needs to double buffer the
261 * data as it cannot do the transformations in place without having direct
262 * access to the destination pages of the backing file.
263 */
264static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
265 loff_t pos, struct page *page)
266{
267 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
268 bvec->bv_offset, bvec->bv_len, pos >> 9);
269 if (likely(!ret))
270 return __do_lo_send_write(lo->lo_backing_file,
271 page_address(page), bvec->bv_len,
272 pos);
273 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
274 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
275 if (ret > 0)
276 ret = -EIO;
277 return ret;
278}
279
280static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
281{
282 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
283 struct page *page);
284 struct bio_vec *bvec;
285 struct page *page = NULL;
286 int i, ret = 0;
287
288 if (lo->transfer != transfer_none) {
289 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
290 if (unlikely(!page))
291 goto fail;
292 kmap(page);
293 do_lo_send = do_lo_send_write;
294 } else {
295 do_lo_send = do_lo_send_direct_write;
296 }
297
298 bio_for_each_segment(bvec, bio, i) {
299 ret = do_lo_send(lo, bvec, pos, page);
300 if (ret < 0)
301 break;
302 pos += bvec->bv_len;
303 }
304 if (page) {
305 kunmap(page);
306 __free_page(page);
307 }
308out:
309 return ret;
310fail:
311 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
312 ret = -ENOMEM;
313 goto out;
314}
315
316struct lo_read_data {
317 struct loop_device *lo;
318 struct page *page;
319 unsigned offset;
320 int bsize;
321};
322
323static int
324lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
325 struct splice_desc *sd)
326{
327 struct lo_read_data *p = sd->u.data;
328 struct loop_device *lo = p->lo;
329 struct page *page = buf->page;
330 sector_t IV;
331 int size;
332
333 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
334 (buf->offset >> 9);
335 size = sd->len;
336 if (size > p->bsize)
337 size = p->bsize;
338
339 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
340 printk(KERN_ERR "loop: transfer error block %ld\n",
341 page->index);
342 size = -EINVAL;
343 }
344
345 flush_dcache_page(p->page);
346
347 if (size > 0)
348 p->offset += size;
349
350 return size;
351}
352
353static int
354lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
355{
356 return __splice_from_pipe(pipe, sd, lo_splice_actor);
357}
358
359static ssize_t
360do_lo_receive(struct loop_device *lo,
361 struct bio_vec *bvec, int bsize, loff_t pos)
362{
363 struct lo_read_data cookie;
364 struct splice_desc sd;
365 struct file *file;
366 ssize_t retval;
367
368 cookie.lo = lo;
369 cookie.page = bvec->bv_page;
370 cookie.offset = bvec->bv_offset;
371 cookie.bsize = bsize;
372
373 sd.len = 0;
374 sd.total_len = bvec->bv_len;
375 sd.flags = 0;
376 sd.pos = pos;
377 sd.u.data = &cookie;
378
379 file = lo->lo_backing_file;
380 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
381
382 return retval;
383}
384
385static int
386lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
387{
388 struct bio_vec *bvec;
389 ssize_t s;
390 int i;
391
392 bio_for_each_segment(bvec, bio, i) {
393 s = do_lo_receive(lo, bvec, bsize, pos);
394 if (s < 0)
395 return s;
396
397 if (s != bvec->bv_len) {
398 zero_fill_bio(bio);
399 break;
400 }
401 pos += bvec->bv_len;
402 }
403 return 0;
404}
405
406static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
407{
408 loff_t pos;
409 int ret;
410
411 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
412
413 if (bio_rw(bio) == WRITE) {
414 struct file *file = lo->lo_backing_file;
415
416 if (bio->bi_rw & REQ_FLUSH) {
417 ret = vfs_fsync(file, 0);
418 if (unlikely(ret && ret != -EINVAL)) {
419 ret = -EIO;
420 goto out;
421 }
422 }
423
424 /*
425 * We use punch hole to reclaim the free space used by the
426 * image a.k.a. discard. However we do not support discard if
427 * encryption is enabled, because it may give an attacker
428 * useful information.
429 */
430 if (bio->bi_rw & REQ_DISCARD) {
431 struct file *file = lo->lo_backing_file;
432 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
433
434 if ((!file->f_op->fallocate) ||
435 lo->lo_encrypt_key_size) {
436 ret = -EOPNOTSUPP;
437 goto out;
438 }
439 ret = file->f_op->fallocate(file, mode, pos,
440 bio->bi_size);
441 if (unlikely(ret && ret != -EINVAL &&
442 ret != -EOPNOTSUPP))
443 ret = -EIO;
444 goto out;
445 }
446
447 ret = lo_send(lo, bio, pos);
448
449 if ((bio->bi_rw & REQ_FUA) && !ret) {
450 ret = vfs_fsync(file, 0);
451 if (unlikely(ret && ret != -EINVAL))
452 ret = -EIO;
453 }
454 } else
455 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
456
457out:
458 return ret;
459}
460
461/*
462 * Add bio to back of pending list
463 */
464static void loop_add_bio(struct loop_device *lo, struct bio *bio)
465{
466 bio_list_add(&lo->lo_bio_list, bio);
467}
468
469/*
470 * Grab first pending buffer
471 */
472static struct bio *loop_get_bio(struct loop_device *lo)
473{
474 return bio_list_pop(&lo->lo_bio_list);
475}
476
477static void loop_make_request(struct request_queue *q, struct bio *old_bio)
478{
479 struct loop_device *lo = q->queuedata;
480 int rw = bio_rw(old_bio);
481
482 if (rw == READA)
483 rw = READ;
484
485 BUG_ON(!lo || (rw != READ && rw != WRITE));
486
487 spin_lock_irq(&lo->lo_lock);
488 if (lo->lo_state != Lo_bound)
489 goto out;
490 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
491 goto out;
492 loop_add_bio(lo, old_bio);
493 wake_up(&lo->lo_event);
494 spin_unlock_irq(&lo->lo_lock);
495 return;
496
497out:
498 spin_unlock_irq(&lo->lo_lock);
499 bio_io_error(old_bio);
500}
501
502struct switch_request {
503 struct file *file;
504 struct completion wait;
505};
506
507static void do_loop_switch(struct loop_device *, struct switch_request *);
508
509static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
510{
511 if (unlikely(!bio->bi_bdev)) {
512 do_loop_switch(lo, bio->bi_private);
513 bio_put(bio);
514 } else {
515 int ret = do_bio_filebacked(lo, bio);
516 bio_endio(bio, ret);
517 }
518}
519
520/*
521 * worker thread that handles reads/writes to file backed loop devices,
522 * to avoid blocking in our make_request_fn. it also does loop decrypting
523 * on reads for block backed loop, as that is too heavy to do from
524 * b_end_io context where irqs may be disabled.
525 *
526 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
527 * calling kthread_stop(). Therefore once kthread_should_stop() is
528 * true, make_request will not place any more requests. Therefore
529 * once kthread_should_stop() is true and lo_bio is NULL, we are
530 * done with the loop.
531 */
532static int loop_thread(void *data)
533{
534 struct loop_device *lo = data;
535 struct bio *bio;
536
537 set_user_nice(current, -20);
538
539 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
540
541 wait_event_interruptible(lo->lo_event,
542 !bio_list_empty(&lo->lo_bio_list) ||
543 kthread_should_stop());
544
545 if (bio_list_empty(&lo->lo_bio_list))
546 continue;
547 spin_lock_irq(&lo->lo_lock);
548 bio = loop_get_bio(lo);
549 spin_unlock_irq(&lo->lo_lock);
550
551 BUG_ON(!bio);
552 loop_handle_bio(lo, bio);
553 }
554
555 return 0;
556}
557
558/*
559 * loop_switch performs the hard work of switching a backing store.
560 * First it needs to flush existing IO, it does this by sending a magic
561 * BIO down the pipe. The completion of this BIO does the actual switch.
562 */
563static int loop_switch(struct loop_device *lo, struct file *file)
564{
565 struct switch_request w;
566 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
567 if (!bio)
568 return -ENOMEM;
569 init_completion(&w.wait);
570 w.file = file;
571 bio->bi_private = &w;
572 bio->bi_bdev = NULL;
573 loop_make_request(lo->lo_queue, bio);
574 wait_for_completion(&w.wait);
575 return 0;
576}
577
578/*
579 * Helper to flush the IOs in loop, but keeping loop thread running
580 */
581static int loop_flush(struct loop_device *lo)
582{
583 /* loop not yet configured, no running thread, nothing to flush */
584 if (!lo->lo_thread)
585 return 0;
586
587 return loop_switch(lo, NULL);
588}
589
590/*
591 * Do the actual switch; called from the BIO completion routine
592 */
593static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
594{
595 struct file *file = p->file;
596 struct file *old_file = lo->lo_backing_file;
597 struct address_space *mapping;
598
599 /* if no new file, only flush of queued bios requested */
600 if (!file)
601 goto out;
602
603 mapping = file->f_mapping;
604 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
605 lo->lo_backing_file = file;
606 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
607 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
608 lo->old_gfp_mask = mapping_gfp_mask(mapping);
609 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
610out:
611 complete(&p->wait);
612}
613
614
615/*
616 * loop_change_fd switched the backing store of a loopback device to
617 * a new file. This is useful for operating system installers to free up
618 * the original file and in High Availability environments to switch to
619 * an alternative location for the content in case of server meltdown.
620 * This can only work if the loop device is used read-only, and if the
621 * new backing store is the same size and type as the old backing store.
622 */
623static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
624 unsigned int arg)
625{
626 struct file *file, *old_file;
627 struct inode *inode;
628 int error;
629
630 error = -ENXIO;
631 if (lo->lo_state != Lo_bound)
632 goto out;
633
634 /* the loop device has to be read-only */
635 error = -EINVAL;
636 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
637 goto out;
638
639 error = -EBADF;
640 file = fget(arg);
641 if (!file)
642 goto out;
643
644 inode = file->f_mapping->host;
645 old_file = lo->lo_backing_file;
646
647 error = -EINVAL;
648
649 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
650 goto out_putf;
651
652 /* size of the new backing store needs to be the same */
653 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
654 goto out_putf;
655
656 /* and ... switch */
657 error = loop_switch(lo, file);
658 if (error)
659 goto out_putf;
660
661 fput(old_file);
662 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
663 ioctl_by_bdev(bdev, BLKRRPART, 0);
664 return 0;
665
666 out_putf:
667 fput(file);
668 out:
669 return error;
670}
671
672static inline int is_loop_device(struct file *file)
673{
674 struct inode *i = file->f_mapping->host;
675
676 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
677}
678
679/* loop sysfs attributes */
680
681static ssize_t loop_attr_show(struct device *dev, char *page,
682 ssize_t (*callback)(struct loop_device *, char *))
683{
684 struct gendisk *disk = dev_to_disk(dev);
685 struct loop_device *lo = disk->private_data;
686
687 return callback(lo, page);
688}
689
690#define LOOP_ATTR_RO(_name) \
691static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
692static ssize_t loop_attr_do_show_##_name(struct device *d, \
693 struct device_attribute *attr, char *b) \
694{ \
695 return loop_attr_show(d, b, loop_attr_##_name##_show); \
696} \
697static struct device_attribute loop_attr_##_name = \
698 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
699
700static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
701{
702 ssize_t ret;
703 char *p = NULL;
704
705 spin_lock_irq(&lo->lo_lock);
706 if (lo->lo_backing_file)
707 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
708 spin_unlock_irq(&lo->lo_lock);
709
710 if (IS_ERR_OR_NULL(p))
711 ret = PTR_ERR(p);
712 else {
713 ret = strlen(p);
714 memmove(buf, p, ret);
715 buf[ret++] = '\n';
716 buf[ret] = 0;
717 }
718
719 return ret;
720}
721
722static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
723{
724 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
725}
726
727static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
728{
729 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
730}
731
732static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
733{
734 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
735
736 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
737}
738
739static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
740{
741 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
742
743 return sprintf(buf, "%s\n", partscan ? "1" : "0");
744}
745
746LOOP_ATTR_RO(backing_file);
747LOOP_ATTR_RO(offset);
748LOOP_ATTR_RO(sizelimit);
749LOOP_ATTR_RO(autoclear);
750LOOP_ATTR_RO(partscan);
751
752static struct attribute *loop_attrs[] = {
753 &loop_attr_backing_file.attr,
754 &loop_attr_offset.attr,
755 &loop_attr_sizelimit.attr,
756 &loop_attr_autoclear.attr,
757 &loop_attr_partscan.attr,
758 NULL,
759};
760
761static struct attribute_group loop_attribute_group = {
762 .name = "loop",
763 .attrs= loop_attrs,
764};
765
766static int loop_sysfs_init(struct loop_device *lo)
767{
768 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
769 &loop_attribute_group);
770}
771
772static void loop_sysfs_exit(struct loop_device *lo)
773{
774 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
775 &loop_attribute_group);
776}
777
778static void loop_config_discard(struct loop_device *lo)
779{
780 struct file *file = lo->lo_backing_file;
781 struct inode *inode = file->f_mapping->host;
782 struct request_queue *q = lo->lo_queue;
783
784 /*
785 * We use punch hole to reclaim the free space used by the
786 * image a.k.a. discard. However we do support discard if
787 * encryption is enabled, because it may give an attacker
788 * useful information.
789 */
790 if ((!file->f_op->fallocate) ||
791 lo->lo_encrypt_key_size) {
792 q->limits.discard_granularity = 0;
793 q->limits.discard_alignment = 0;
794 q->limits.max_discard_sectors = 0;
795 q->limits.discard_zeroes_data = 0;
796 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
797 return;
798 }
799
800 q->limits.discard_granularity = inode->i_sb->s_blocksize;
801 q->limits.discard_alignment = 0;
802 q->limits.max_discard_sectors = UINT_MAX >> 9;
803 q->limits.discard_zeroes_data = 1;
804 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
805}
806
807static int loop_set_fd(struct loop_device *lo, fmode_t mode,
808 struct block_device *bdev, unsigned int arg)
809{
810 struct file *file, *f;
811 struct inode *inode;
812 struct address_space *mapping;
813 unsigned lo_blocksize;
814 int lo_flags = 0;
815 int error;
816 loff_t size;
817
818 /* This is safe, since we have a reference from open(). */
819 __module_get(THIS_MODULE);
820
821 error = -EBADF;
822 file = fget(arg);
823 if (!file)
824 goto out;
825
826 error = -EBUSY;
827 if (lo->lo_state != Lo_unbound)
828 goto out_putf;
829
830 /* Avoid recursion */
831 f = file;
832 while (is_loop_device(f)) {
833 struct loop_device *l;
834
835 if (f->f_mapping->host->i_bdev == bdev)
836 goto out_putf;
837
838 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
839 if (l->lo_state == Lo_unbound) {
840 error = -EINVAL;
841 goto out_putf;
842 }
843 f = l->lo_backing_file;
844 }
845
846 mapping = file->f_mapping;
847 inode = mapping->host;
848
849 error = -EINVAL;
850 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
851 goto out_putf;
852
853 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
854 !file->f_op->write)
855 lo_flags |= LO_FLAGS_READ_ONLY;
856
857 lo_blocksize = S_ISBLK(inode->i_mode) ?
858 inode->i_bdev->bd_block_size : PAGE_SIZE;
859
860 error = -EFBIG;
861 size = get_loop_size(lo, file);
862 if ((loff_t)(sector_t)size != size)
863 goto out_putf;
864
865 error = 0;
866
867 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
868
869 lo->lo_blocksize = lo_blocksize;
870 lo->lo_device = bdev;
871 lo->lo_flags = lo_flags;
872 lo->lo_backing_file = file;
873 lo->transfer = transfer_none;
874 lo->ioctl = NULL;
875 lo->lo_sizelimit = 0;
876 lo->old_gfp_mask = mapping_gfp_mask(mapping);
877 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
878
879 bio_list_init(&lo->lo_bio_list);
880
881 /*
882 * set queue make_request_fn, and add limits based on lower level
883 * device
884 */
885 blk_queue_make_request(lo->lo_queue, loop_make_request);
886 lo->lo_queue->queuedata = lo;
887
888 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
889 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
890
891 set_capacity(lo->lo_disk, size);
892 bd_set_size(bdev, size << 9);
893 loop_sysfs_init(lo);
894 /* let user-space know about the new size */
895 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
896
897 set_blocksize(bdev, lo_blocksize);
898
899 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
900 lo->lo_number);
901 if (IS_ERR(lo->lo_thread)) {
902 error = PTR_ERR(lo->lo_thread);
903 goto out_clr;
904 }
905 lo->lo_state = Lo_bound;
906 wake_up_process(lo->lo_thread);
907 if (part_shift)
908 lo->lo_flags |= LO_FLAGS_PARTSCAN;
909 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
910 ioctl_by_bdev(bdev, BLKRRPART, 0);
911 return 0;
912
913out_clr:
914 loop_sysfs_exit(lo);
915 lo->lo_thread = NULL;
916 lo->lo_device = NULL;
917 lo->lo_backing_file = NULL;
918 lo->lo_flags = 0;
919 set_capacity(lo->lo_disk, 0);
920 invalidate_bdev(bdev);
921 bd_set_size(bdev, 0);
922 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
923 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
924 lo->lo_state = Lo_unbound;
925 out_putf:
926 fput(file);
927 out:
928 /* This is safe: open() is still holding a reference. */
929 module_put(THIS_MODULE);
930 return error;
931}
932
933static int
934loop_release_xfer(struct loop_device *lo)
935{
936 int err = 0;
937 struct loop_func_table *xfer = lo->lo_encryption;
938
939 if (xfer) {
940 if (xfer->release)
941 err = xfer->release(lo);
942 lo->transfer = NULL;
943 lo->lo_encryption = NULL;
944 module_put(xfer->owner);
945 }
946 return err;
947}
948
949static int
950loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
951 const struct loop_info64 *i)
952{
953 int err = 0;
954
955 if (xfer) {
956 struct module *owner = xfer->owner;
957
958 if (!try_module_get(owner))
959 return -EINVAL;
960 if (xfer->init)
961 err = xfer->init(lo, i);
962 if (err)
963 module_put(owner);
964 else
965 lo->lo_encryption = xfer;
966 }
967 return err;
968}
969
970static int loop_clr_fd(struct loop_device *lo)
971{
972 struct file *filp = lo->lo_backing_file;
973 gfp_t gfp = lo->old_gfp_mask;
974 struct block_device *bdev = lo->lo_device;
975
976 if (lo->lo_state != Lo_bound)
977 return -ENXIO;
978
979 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
980 return -EBUSY;
981
982 if (filp == NULL)
983 return -EINVAL;
984
985 spin_lock_irq(&lo->lo_lock);
986 lo->lo_state = Lo_rundown;
987 spin_unlock_irq(&lo->lo_lock);
988
989 kthread_stop(lo->lo_thread);
990
991 spin_lock_irq(&lo->lo_lock);
992 lo->lo_backing_file = NULL;
993 spin_unlock_irq(&lo->lo_lock);
994
995 loop_release_xfer(lo);
996 lo->transfer = NULL;
997 lo->ioctl = NULL;
998 lo->lo_device = NULL;
999 lo->lo_encryption = NULL;
1000 lo->lo_offset = 0;
1001 lo->lo_sizelimit = 0;
1002 lo->lo_encrypt_key_size = 0;
1003 lo->lo_thread = NULL;
1004 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1005 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1006 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1007 if (bdev)
1008 invalidate_bdev(bdev);
1009 set_capacity(lo->lo_disk, 0);
1010 loop_sysfs_exit(lo);
1011 if (bdev) {
1012 bd_set_size(bdev, 0);
1013 /* let user-space know about this change */
1014 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1015 }
1016 mapping_set_gfp_mask(filp->f_mapping, gfp);
1017 lo->lo_state = Lo_unbound;
1018 /* This is safe: open() is still holding a reference. */
1019 module_put(THIS_MODULE);
1020 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1021 ioctl_by_bdev(bdev, BLKRRPART, 0);
1022 lo->lo_flags = 0;
1023 if (!part_shift)
1024 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1025 mutex_unlock(&lo->lo_ctl_mutex);
1026 /*
1027 * Need not hold lo_ctl_mutex to fput backing file.
1028 * Calling fput holding lo_ctl_mutex triggers a circular
1029 * lock dependency possibility warning as fput can take
1030 * bd_mutex which is usually taken before lo_ctl_mutex.
1031 */
1032 fput(filp);
1033 return 0;
1034}
1035
1036static int
1037loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1038{
1039 int err;
1040 struct loop_func_table *xfer;
1041 uid_t uid = current_uid();
1042
1043 if (lo->lo_encrypt_key_size &&
1044 lo->lo_key_owner != uid &&
1045 !capable(CAP_SYS_ADMIN))
1046 return -EPERM;
1047 if (lo->lo_state != Lo_bound)
1048 return -ENXIO;
1049 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1050 return -EINVAL;
1051
1052 err = loop_release_xfer(lo);
1053 if (err)
1054 return err;
1055
1056 if (info->lo_encrypt_type) {
1057 unsigned int type = info->lo_encrypt_type;
1058
1059 if (type >= MAX_LO_CRYPT)
1060 return -EINVAL;
1061 xfer = xfer_funcs[type];
1062 if (xfer == NULL)
1063 return -EINVAL;
1064 } else
1065 xfer = NULL;
1066
1067 err = loop_init_xfer(lo, xfer, info);
1068 if (err)
1069 return err;
1070
1071 if (lo->lo_offset != info->lo_offset ||
1072 lo->lo_sizelimit != info->lo_sizelimit) {
1073 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1074 return -EFBIG;
1075 }
1076 loop_config_discard(lo);
1077
1078 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1079 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1080 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1081 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1082
1083 if (!xfer)
1084 xfer = &none_funcs;
1085 lo->transfer = xfer->transfer;
1086 lo->ioctl = xfer->ioctl;
1087
1088 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1089 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1090 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1091
1092 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1093 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1094 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1095 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1096 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1097 }
1098
1099 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1100 lo->lo_init[0] = info->lo_init[0];
1101 lo->lo_init[1] = info->lo_init[1];
1102 if (info->lo_encrypt_key_size) {
1103 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1104 info->lo_encrypt_key_size);
1105 lo->lo_key_owner = uid;
1106 }
1107
1108 return 0;
1109}
1110
1111static int
1112loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1113{
1114 struct file *file = lo->lo_backing_file;
1115 struct kstat stat;
1116 int error;
1117
1118 if (lo->lo_state != Lo_bound)
1119 return -ENXIO;
1120 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1121 if (error)
1122 return error;
1123 memset(info, 0, sizeof(*info));
1124 info->lo_number = lo->lo_number;
1125 info->lo_device = huge_encode_dev(stat.dev);
1126 info->lo_inode = stat.ino;
1127 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1128 info->lo_offset = lo->lo_offset;
1129 info->lo_sizelimit = lo->lo_sizelimit;
1130 info->lo_flags = lo->lo_flags;
1131 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1132 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1133 info->lo_encrypt_type =
1134 lo->lo_encryption ? lo->lo_encryption->number : 0;
1135 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1136 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1137 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1138 lo->lo_encrypt_key_size);
1139 }
1140 return 0;
1141}
1142
1143static void
1144loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1145{
1146 memset(info64, 0, sizeof(*info64));
1147 info64->lo_number = info->lo_number;
1148 info64->lo_device = info->lo_device;
1149 info64->lo_inode = info->lo_inode;
1150 info64->lo_rdevice = info->lo_rdevice;
1151 info64->lo_offset = info->lo_offset;
1152 info64->lo_sizelimit = 0;
1153 info64->lo_encrypt_type = info->lo_encrypt_type;
1154 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1155 info64->lo_flags = info->lo_flags;
1156 info64->lo_init[0] = info->lo_init[0];
1157 info64->lo_init[1] = info->lo_init[1];
1158 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1159 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1160 else
1161 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1162 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1163}
1164
1165static int
1166loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1167{
1168 memset(info, 0, sizeof(*info));
1169 info->lo_number = info64->lo_number;
1170 info->lo_device = info64->lo_device;
1171 info->lo_inode = info64->lo_inode;
1172 info->lo_rdevice = info64->lo_rdevice;
1173 info->lo_offset = info64->lo_offset;
1174 info->lo_encrypt_type = info64->lo_encrypt_type;
1175 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1176 info->lo_flags = info64->lo_flags;
1177 info->lo_init[0] = info64->lo_init[0];
1178 info->lo_init[1] = info64->lo_init[1];
1179 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1180 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1181 else
1182 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1183 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1184
1185 /* error in case values were truncated */
1186 if (info->lo_device != info64->lo_device ||
1187 info->lo_rdevice != info64->lo_rdevice ||
1188 info->lo_inode != info64->lo_inode ||
1189 info->lo_offset != info64->lo_offset)
1190 return -EOVERFLOW;
1191
1192 return 0;
1193}
1194
1195static int
1196loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1197{
1198 struct loop_info info;
1199 struct loop_info64 info64;
1200
1201 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1202 return -EFAULT;
1203 loop_info64_from_old(&info, &info64);
1204 return loop_set_status(lo, &info64);
1205}
1206
1207static int
1208loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1209{
1210 struct loop_info64 info64;
1211
1212 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1213 return -EFAULT;
1214 return loop_set_status(lo, &info64);
1215}
1216
1217static int
1218loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1219 struct loop_info info;
1220 struct loop_info64 info64;
1221 int err = 0;
1222
1223 if (!arg)
1224 err = -EINVAL;
1225 if (!err)
1226 err = loop_get_status(lo, &info64);
1227 if (!err)
1228 err = loop_info64_to_old(&info64, &info);
1229 if (!err && copy_to_user(arg, &info, sizeof(info)))
1230 err = -EFAULT;
1231
1232 return err;
1233}
1234
1235static int
1236loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1237 struct loop_info64 info64;
1238 int err = 0;
1239
1240 if (!arg)
1241 err = -EINVAL;
1242 if (!err)
1243 err = loop_get_status(lo, &info64);
1244 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1245 err = -EFAULT;
1246
1247 return err;
1248}
1249
1250static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1251{
1252 int err;
1253 sector_t sec;
1254 loff_t sz;
1255
1256 err = -ENXIO;
1257 if (unlikely(lo->lo_state != Lo_bound))
1258 goto out;
1259 err = figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1260 if (unlikely(err))
1261 goto out;
1262 sec = get_capacity(lo->lo_disk);
1263 /* the width of sector_t may be narrow for bit-shift */
1264 sz = sec;
1265 sz <<= 9;
1266 mutex_lock(&bdev->bd_mutex);
1267 bd_set_size(bdev, sz);
1268 /* let user-space know about the new size */
1269 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1270 mutex_unlock(&bdev->bd_mutex);
1271
1272 out:
1273 return err;
1274}
1275
1276static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1277 unsigned int cmd, unsigned long arg)
1278{
1279 struct loop_device *lo = bdev->bd_disk->private_data;
1280 int err;
1281
1282 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1283 switch (cmd) {
1284 case LOOP_SET_FD:
1285 err = loop_set_fd(lo, mode, bdev, arg);
1286 break;
1287 case LOOP_CHANGE_FD:
1288 err = loop_change_fd(lo, bdev, arg);
1289 break;
1290 case LOOP_CLR_FD:
1291 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1292 err = loop_clr_fd(lo);
1293 if (!err)
1294 goto out_unlocked;
1295 break;
1296 case LOOP_SET_STATUS:
1297 err = -EPERM;
1298 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1299 err = loop_set_status_old(lo,
1300 (struct loop_info __user *)arg);
1301 break;
1302 case LOOP_GET_STATUS:
1303 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1304 break;
1305 case LOOP_SET_STATUS64:
1306 err = -EPERM;
1307 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1308 err = loop_set_status64(lo,
1309 (struct loop_info64 __user *) arg);
1310 break;
1311 case LOOP_GET_STATUS64:
1312 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1313 break;
1314 case LOOP_SET_CAPACITY:
1315 err = -EPERM;
1316 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1317 err = loop_set_capacity(lo, bdev);
1318 break;
1319 default:
1320 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1321 }
1322 mutex_unlock(&lo->lo_ctl_mutex);
1323
1324out_unlocked:
1325 return err;
1326}
1327
1328#ifdef CONFIG_COMPAT
1329struct compat_loop_info {
1330 compat_int_t lo_number; /* ioctl r/o */
1331 compat_dev_t lo_device; /* ioctl r/o */
1332 compat_ulong_t lo_inode; /* ioctl r/o */
1333 compat_dev_t lo_rdevice; /* ioctl r/o */
1334 compat_int_t lo_offset;
1335 compat_int_t lo_encrypt_type;
1336 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1337 compat_int_t lo_flags; /* ioctl r/o */
1338 char lo_name[LO_NAME_SIZE];
1339 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1340 compat_ulong_t lo_init[2];
1341 char reserved[4];
1342};
1343
1344/*
1345 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1346 * - noinlined to reduce stack space usage in main part of driver
1347 */
1348static noinline int
1349loop_info64_from_compat(const struct compat_loop_info __user *arg,
1350 struct loop_info64 *info64)
1351{
1352 struct compat_loop_info info;
1353
1354 if (copy_from_user(&info, arg, sizeof(info)))
1355 return -EFAULT;
1356
1357 memset(info64, 0, sizeof(*info64));
1358 info64->lo_number = info.lo_number;
1359 info64->lo_device = info.lo_device;
1360 info64->lo_inode = info.lo_inode;
1361 info64->lo_rdevice = info.lo_rdevice;
1362 info64->lo_offset = info.lo_offset;
1363 info64->lo_sizelimit = 0;
1364 info64->lo_encrypt_type = info.lo_encrypt_type;
1365 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1366 info64->lo_flags = info.lo_flags;
1367 info64->lo_init[0] = info.lo_init[0];
1368 info64->lo_init[1] = info.lo_init[1];
1369 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1370 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1371 else
1372 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1373 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1374 return 0;
1375}
1376
1377/*
1378 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1379 * - noinlined to reduce stack space usage in main part of driver
1380 */
1381static noinline int
1382loop_info64_to_compat(const struct loop_info64 *info64,
1383 struct compat_loop_info __user *arg)
1384{
1385 struct compat_loop_info info;
1386
1387 memset(&info, 0, sizeof(info));
1388 info.lo_number = info64->lo_number;
1389 info.lo_device = info64->lo_device;
1390 info.lo_inode = info64->lo_inode;
1391 info.lo_rdevice = info64->lo_rdevice;
1392 info.lo_offset = info64->lo_offset;
1393 info.lo_encrypt_type = info64->lo_encrypt_type;
1394 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1395 info.lo_flags = info64->lo_flags;
1396 info.lo_init[0] = info64->lo_init[0];
1397 info.lo_init[1] = info64->lo_init[1];
1398 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1399 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1400 else
1401 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1402 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1403
1404 /* error in case values were truncated */
1405 if (info.lo_device != info64->lo_device ||
1406 info.lo_rdevice != info64->lo_rdevice ||
1407 info.lo_inode != info64->lo_inode ||
1408 info.lo_offset != info64->lo_offset ||
1409 info.lo_init[0] != info64->lo_init[0] ||
1410 info.lo_init[1] != info64->lo_init[1])
1411 return -EOVERFLOW;
1412
1413 if (copy_to_user(arg, &info, sizeof(info)))
1414 return -EFAULT;
1415 return 0;
1416}
1417
1418static int
1419loop_set_status_compat(struct loop_device *lo,
1420 const struct compat_loop_info __user *arg)
1421{
1422 struct loop_info64 info64;
1423 int ret;
1424
1425 ret = loop_info64_from_compat(arg, &info64);
1426 if (ret < 0)
1427 return ret;
1428 return loop_set_status(lo, &info64);
1429}
1430
1431static int
1432loop_get_status_compat(struct loop_device *lo,
1433 struct compat_loop_info __user *arg)
1434{
1435 struct loop_info64 info64;
1436 int err = 0;
1437
1438 if (!arg)
1439 err = -EINVAL;
1440 if (!err)
1441 err = loop_get_status(lo, &info64);
1442 if (!err)
1443 err = loop_info64_to_compat(&info64, arg);
1444 return err;
1445}
1446
1447static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1448 unsigned int cmd, unsigned long arg)
1449{
1450 struct loop_device *lo = bdev->bd_disk->private_data;
1451 int err;
1452
1453 switch(cmd) {
1454 case LOOP_SET_STATUS:
1455 mutex_lock(&lo->lo_ctl_mutex);
1456 err = loop_set_status_compat(
1457 lo, (const struct compat_loop_info __user *) arg);
1458 mutex_unlock(&lo->lo_ctl_mutex);
1459 break;
1460 case LOOP_GET_STATUS:
1461 mutex_lock(&lo->lo_ctl_mutex);
1462 err = loop_get_status_compat(
1463 lo, (struct compat_loop_info __user *) arg);
1464 mutex_unlock(&lo->lo_ctl_mutex);
1465 break;
1466 case LOOP_SET_CAPACITY:
1467 case LOOP_CLR_FD:
1468 case LOOP_GET_STATUS64:
1469 case LOOP_SET_STATUS64:
1470 arg = (unsigned long) compat_ptr(arg);
1471 case LOOP_SET_FD:
1472 case LOOP_CHANGE_FD:
1473 err = lo_ioctl(bdev, mode, cmd, arg);
1474 break;
1475 default:
1476 err = -ENOIOCTLCMD;
1477 break;
1478 }
1479 return err;
1480}
1481#endif
1482
1483static int lo_open(struct block_device *bdev, fmode_t mode)
1484{
1485 struct loop_device *lo;
1486 int err = 0;
1487
1488 mutex_lock(&loop_index_mutex);
1489 lo = bdev->bd_disk->private_data;
1490 if (!lo) {
1491 err = -ENXIO;
1492 goto out;
1493 }
1494
1495 mutex_lock(&lo->lo_ctl_mutex);
1496 lo->lo_refcnt++;
1497 mutex_unlock(&lo->lo_ctl_mutex);
1498out:
1499 mutex_unlock(&loop_index_mutex);
1500 return err;
1501}
1502
1503static int lo_release(struct gendisk *disk, fmode_t mode)
1504{
1505 struct loop_device *lo = disk->private_data;
1506 int err;
1507
1508 mutex_lock(&lo->lo_ctl_mutex);
1509
1510 if (--lo->lo_refcnt)
1511 goto out;
1512
1513 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1514 /*
1515 * In autoclear mode, stop the loop thread
1516 * and remove configuration after last close.
1517 */
1518 err = loop_clr_fd(lo);
1519 if (!err)
1520 goto out_unlocked;
1521 } else {
1522 /*
1523 * Otherwise keep thread (if running) and config,
1524 * but flush possible ongoing bios in thread.
1525 */
1526 loop_flush(lo);
1527 }
1528
1529out:
1530 mutex_unlock(&lo->lo_ctl_mutex);
1531out_unlocked:
1532 return 0;
1533}
1534
1535static const struct block_device_operations lo_fops = {
1536 .owner = THIS_MODULE,
1537 .open = lo_open,
1538 .release = lo_release,
1539 .ioctl = lo_ioctl,
1540#ifdef CONFIG_COMPAT
1541 .compat_ioctl = lo_compat_ioctl,
1542#endif
1543};
1544
1545/*
1546 * And now the modules code and kernel interface.
1547 */
1548static int max_loop;
1549module_param(max_loop, int, S_IRUGO);
1550MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1551module_param(max_part, int, S_IRUGO);
1552MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1553MODULE_LICENSE("GPL");
1554MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1555
1556int loop_register_transfer(struct loop_func_table *funcs)
1557{
1558 unsigned int n = funcs->number;
1559
1560 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1561 return -EINVAL;
1562 xfer_funcs[n] = funcs;
1563 return 0;
1564}
1565
1566static int unregister_transfer_cb(int id, void *ptr, void *data)
1567{
1568 struct loop_device *lo = ptr;
1569 struct loop_func_table *xfer = data;
1570
1571 mutex_lock(&lo->lo_ctl_mutex);
1572 if (lo->lo_encryption == xfer)
1573 loop_release_xfer(lo);
1574 mutex_unlock(&lo->lo_ctl_mutex);
1575 return 0;
1576}
1577
1578int loop_unregister_transfer(int number)
1579{
1580 unsigned int n = number;
1581 struct loop_func_table *xfer;
1582
1583 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1584 return -EINVAL;
1585
1586 xfer_funcs[n] = NULL;
1587 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1588 return 0;
1589}
1590
1591EXPORT_SYMBOL(loop_register_transfer);
1592EXPORT_SYMBOL(loop_unregister_transfer);
1593
1594static int loop_add(struct loop_device **l, int i)
1595{
1596 struct loop_device *lo;
1597 struct gendisk *disk;
1598 int err;
1599
1600 err = -ENOMEM;
1601 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1602 if (!lo)
1603 goto out;
1604
1605 if (!idr_pre_get(&loop_index_idr, GFP_KERNEL))
1606 goto out_free_dev;
1607
1608 if (i >= 0) {
1609 int m;
1610
1611 /* create specific i in the index */
1612 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1613 if (err >= 0 && i != m) {
1614 idr_remove(&loop_index_idr, m);
1615 err = -EEXIST;
1616 }
1617 } else if (i == -1) {
1618 int m;
1619
1620 /* get next free nr */
1621 err = idr_get_new(&loop_index_idr, lo, &m);
1622 if (err >= 0)
1623 i = m;
1624 } else {
1625 err = -EINVAL;
1626 }
1627 if (err < 0)
1628 goto out_free_dev;
1629
1630 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1631 if (!lo->lo_queue)
1632 goto out_free_dev;
1633
1634 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1635 if (!disk)
1636 goto out_free_queue;
1637
1638 /*
1639 * Disable partition scanning by default. The in-kernel partition
1640 * scanning can be requested individually per-device during its
1641 * setup. Userspace can always add and remove partitions from all
1642 * devices. The needed partition minors are allocated from the
1643 * extended minor space, the main loop device numbers will continue
1644 * to match the loop minors, regardless of the number of partitions
1645 * used.
1646 *
1647 * If max_part is given, partition scanning is globally enabled for
1648 * all loop devices. The minors for the main loop devices will be
1649 * multiples of max_part.
1650 *
1651 * Note: Global-for-all-devices, set-only-at-init, read-only module
1652 * parameteters like 'max_loop' and 'max_part' make things needlessly
1653 * complicated, are too static, inflexible and may surprise
1654 * userspace tools. Parameters like this in general should be avoided.
1655 */
1656 if (!part_shift)
1657 disk->flags |= GENHD_FL_NO_PART_SCAN;
1658 disk->flags |= GENHD_FL_EXT_DEVT;
1659 mutex_init(&lo->lo_ctl_mutex);
1660 lo->lo_number = i;
1661 lo->lo_thread = NULL;
1662 init_waitqueue_head(&lo->lo_event);
1663 spin_lock_init(&lo->lo_lock);
1664 disk->major = LOOP_MAJOR;
1665 disk->first_minor = i << part_shift;
1666 disk->fops = &lo_fops;
1667 disk->private_data = lo;
1668 disk->queue = lo->lo_queue;
1669 sprintf(disk->disk_name, "loop%d", i);
1670 add_disk(disk);
1671 *l = lo;
1672 return lo->lo_number;
1673
1674out_free_queue:
1675 blk_cleanup_queue(lo->lo_queue);
1676out_free_dev:
1677 kfree(lo);
1678out:
1679 return err;
1680}
1681
1682static void loop_remove(struct loop_device *lo)
1683{
1684 del_gendisk(lo->lo_disk);
1685 blk_cleanup_queue(lo->lo_queue);
1686 put_disk(lo->lo_disk);
1687 kfree(lo);
1688}
1689
1690static int find_free_cb(int id, void *ptr, void *data)
1691{
1692 struct loop_device *lo = ptr;
1693 struct loop_device **l = data;
1694
1695 if (lo->lo_state == Lo_unbound) {
1696 *l = lo;
1697 return 1;
1698 }
1699 return 0;
1700}
1701
1702static int loop_lookup(struct loop_device **l, int i)
1703{
1704 struct loop_device *lo;
1705 int ret = -ENODEV;
1706
1707 if (i < 0) {
1708 int err;
1709
1710 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1711 if (err == 1) {
1712 *l = lo;
1713 ret = lo->lo_number;
1714 }
1715 goto out;
1716 }
1717
1718 /* lookup and return a specific i */
1719 lo = idr_find(&loop_index_idr, i);
1720 if (lo) {
1721 *l = lo;
1722 ret = lo->lo_number;
1723 }
1724out:
1725 return ret;
1726}
1727
1728static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1729{
1730 struct loop_device *lo;
1731 struct kobject *kobj;
1732 int err;
1733
1734 mutex_lock(&loop_index_mutex);
1735 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1736 if (err < 0)
1737 err = loop_add(&lo, MINOR(dev) >> part_shift);
1738 if (err < 0)
1739 kobj = ERR_PTR(err);
1740 else
1741 kobj = get_disk(lo->lo_disk);
1742 mutex_unlock(&loop_index_mutex);
1743
1744 *part = 0;
1745 return kobj;
1746}
1747
1748static long loop_control_ioctl(struct file *file, unsigned int cmd,
1749 unsigned long parm)
1750{
1751 struct loop_device *lo;
1752 int ret = -ENOSYS;
1753
1754 mutex_lock(&loop_index_mutex);
1755 switch (cmd) {
1756 case LOOP_CTL_ADD:
1757 ret = loop_lookup(&lo, parm);
1758 if (ret >= 0) {
1759 ret = -EEXIST;
1760 break;
1761 }
1762 ret = loop_add(&lo, parm);
1763 break;
1764 case LOOP_CTL_REMOVE:
1765 ret = loop_lookup(&lo, parm);
1766 if (ret < 0)
1767 break;
1768 mutex_lock(&lo->lo_ctl_mutex);
1769 if (lo->lo_state != Lo_unbound) {
1770 ret = -EBUSY;
1771 mutex_unlock(&lo->lo_ctl_mutex);
1772 break;
1773 }
1774 if (lo->lo_refcnt > 0) {
1775 ret = -EBUSY;
1776 mutex_unlock(&lo->lo_ctl_mutex);
1777 break;
1778 }
1779 lo->lo_disk->private_data = NULL;
1780 mutex_unlock(&lo->lo_ctl_mutex);
1781 idr_remove(&loop_index_idr, lo->lo_number);
1782 loop_remove(lo);
1783 break;
1784 case LOOP_CTL_GET_FREE:
1785 ret = loop_lookup(&lo, -1);
1786 if (ret >= 0)
1787 break;
1788 ret = loop_add(&lo, -1);
1789 }
1790 mutex_unlock(&loop_index_mutex);
1791
1792 return ret;
1793}
1794
1795static const struct file_operations loop_ctl_fops = {
1796 .open = nonseekable_open,
1797 .unlocked_ioctl = loop_control_ioctl,
1798 .compat_ioctl = loop_control_ioctl,
1799 .owner = THIS_MODULE,
1800 .llseek = noop_llseek,
1801};
1802
1803static struct miscdevice loop_misc = {
1804 .minor = LOOP_CTRL_MINOR,
1805 .name = "loop-control",
1806 .fops = &loop_ctl_fops,
1807};
1808
1809MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1810MODULE_ALIAS("devname:loop-control");
1811
1812static int __init loop_init(void)
1813{
1814 int i, nr;
1815 unsigned long range;
1816 struct loop_device *lo;
1817 int err;
1818
1819 err = misc_register(&loop_misc);
1820 if (err < 0)
1821 return err;
1822
1823 part_shift = 0;
1824 if (max_part > 0) {
1825 part_shift = fls(max_part);
1826
1827 /*
1828 * Adjust max_part according to part_shift as it is exported
1829 * to user space so that user can decide correct minor number
1830 * if [s]he want to create more devices.
1831 *
1832 * Note that -1 is required because partition 0 is reserved
1833 * for the whole disk.
1834 */
1835 max_part = (1UL << part_shift) - 1;
1836 }
1837
1838 if ((1UL << part_shift) > DISK_MAX_PARTS)
1839 return -EINVAL;
1840
1841 if (max_loop > 1UL << (MINORBITS - part_shift))
1842 return -EINVAL;
1843
1844 /*
1845 * If max_loop is specified, create that many devices upfront.
1846 * This also becomes a hard limit. If max_loop is not specified,
1847 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1848 * init time. Loop devices can be requested on-demand with the
1849 * /dev/loop-control interface, or be instantiated by accessing
1850 * a 'dead' device node.
1851 */
1852 if (max_loop) {
1853 nr = max_loop;
1854 range = max_loop << part_shift;
1855 } else {
1856 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1857 range = 1UL << MINORBITS;
1858 }
1859
1860 if (register_blkdev(LOOP_MAJOR, "loop"))
1861 return -EIO;
1862
1863 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1864 THIS_MODULE, loop_probe, NULL, NULL);
1865
1866 /* pre-create number of devices given by config or max_loop */
1867 mutex_lock(&loop_index_mutex);
1868 for (i = 0; i < nr; i++)
1869 loop_add(&lo, i);
1870 mutex_unlock(&loop_index_mutex);
1871
1872 printk(KERN_INFO "loop: module loaded\n");
1873 return 0;
1874}
1875
1876static int loop_exit_cb(int id, void *ptr, void *data)
1877{
1878 struct loop_device *lo = ptr;
1879
1880 loop_remove(lo);
1881 return 0;
1882}
1883
1884static void __exit loop_exit(void)
1885{
1886 unsigned long range;
1887
1888 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1889
1890 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1891 idr_remove_all(&loop_index_idr);
1892 idr_destroy(&loop_index_idr);
1893
1894 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1895 unregister_blkdev(LOOP_MAJOR, "loop");
1896
1897 misc_deregister(&loop_misc);
1898}
1899
1900module_init(loop_init);
1901module_exit(loop_exit);
1902
1903#ifndef MODULE
1904static int __init max_loop_setup(char *str)
1905{
1906 max_loop = simple_strtol(str, NULL, 0);
1907 return 1;
1908}
1909
1910__setup("max_loop=", max_loop_setup);
1911#endif