projects/linux·blob v3.8 · refs

1934 lines · 47.1 KiB

1/*
2 * linux/drivers/block/loop.c
3 *
4 * Written by Theodore Ts'o, 3/29/93
5 *
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
8 *
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 *
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 *
15 * Fixed do_loop_request() re-entrancy - [email protected] Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <[email protected]> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, [email protected]
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <[email protected]> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <[email protected]>, Oct 31, 1999
33 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <[email protected]>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <[email protected]>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
52#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/file.h>
57#include <linux/stat.h>
58#include <linux/errno.h>
59#include <linux/major.h>
60#include <linux/wait.h>
61#include <linux/blkdev.h>
62#include <linux/blkpg.h>
63#include <linux/init.h>
64#include <linux/swap.h>
65#include <linux/slab.h>
66#include <linux/loop.h>
67#include <linux/compat.h>
68#include <linux/suspend.h>
69#include <linux/freezer.h>
70#include <linux/mutex.h>
71#include <linux/writeback.h>
72#include <linux/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 lo->lo_bio_count++;
467 bio_list_add(&lo->lo_bio_list, bio);
468}
469
470/*
471 * Grab first pending buffer
472 */
473static struct bio *loop_get_bio(struct loop_device *lo)
474{
475 lo->lo_bio_count--;
476 return bio_list_pop(&lo->lo_bio_list);
477}
478
479static void loop_make_request(struct request_queue *q, struct bio *old_bio)
480{
481 struct loop_device *lo = q->queuedata;
482 int rw = bio_rw(old_bio);
483
484 if (rw == READA)
485 rw = READ;
486
487 BUG_ON(!lo || (rw != READ && rw != WRITE));
488
489 spin_lock_irq(&lo->lo_lock);
490 if (lo->lo_state != Lo_bound)
491 goto out;
492 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
493 goto out;
494 if (lo->lo_bio_count >= q->nr_congestion_on)
495 wait_event_lock_irq(lo->lo_req_wait,
496 lo->lo_bio_count < q->nr_congestion_off,
497 lo->lo_lock);
498 loop_add_bio(lo, old_bio);
499 wake_up(&lo->lo_event);
500 spin_unlock_irq(&lo->lo_lock);
501 return;
502
503out:
504 spin_unlock_irq(&lo->lo_lock);
505 bio_io_error(old_bio);
506}
507
508struct switch_request {
509 struct file *file;
510 struct completion wait;
511};
512
513static void do_loop_switch(struct loop_device *, struct switch_request *);
514
515static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
516{
517 if (unlikely(!bio->bi_bdev)) {
518 do_loop_switch(lo, bio->bi_private);
519 bio_put(bio);
520 } else {
521 int ret = do_bio_filebacked(lo, bio);
522 bio_endio(bio, ret);
523 }
524}
525
526/*
527 * worker thread that handles reads/writes to file backed loop devices,
528 * to avoid blocking in our make_request_fn. it also does loop decrypting
529 * on reads for block backed loop, as that is too heavy to do from
530 * b_end_io context where irqs may be disabled.
531 *
532 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
533 * calling kthread_stop(). Therefore once kthread_should_stop() is
534 * true, make_request will not place any more requests. Therefore
535 * once kthread_should_stop() is true and lo_bio is NULL, we are
536 * done with the loop.
537 */
538static int loop_thread(void *data)
539{
540 struct loop_device *lo = data;
541 struct bio *bio;
542
543 set_user_nice(current, -20);
544
545 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
546
547 wait_event_interruptible(lo->lo_event,
548 !bio_list_empty(&lo->lo_bio_list) ||
549 kthread_should_stop());
550
551 if (bio_list_empty(&lo->lo_bio_list))
552 continue;
553 spin_lock_irq(&lo->lo_lock);
554 bio = loop_get_bio(lo);
555 if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off)
556 wake_up(&lo->lo_req_wait);
557 spin_unlock_irq(&lo->lo_lock);
558
559 BUG_ON(!bio);
560 loop_handle_bio(lo, bio);
561 }
562
563 return 0;
564}
565
566/*
567 * loop_switch performs the hard work of switching a backing store.
568 * First it needs to flush existing IO, it does this by sending a magic
569 * BIO down the pipe. The completion of this BIO does the actual switch.
570 */
571static int loop_switch(struct loop_device *lo, struct file *file)
572{
573 struct switch_request w;
574 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
575 if (!bio)
576 return -ENOMEM;
577 init_completion(&w.wait);
578 w.file = file;
579 bio->bi_private = &w;
580 bio->bi_bdev = NULL;
581 loop_make_request(lo->lo_queue, bio);
582 wait_for_completion(&w.wait);
583 return 0;
584}
585
586/*
587 * Helper to flush the IOs in loop, but keeping loop thread running
588 */
589static int loop_flush(struct loop_device *lo)
590{
591 /* loop not yet configured, no running thread, nothing to flush */
592 if (!lo->lo_thread)
593 return 0;
594
595 return loop_switch(lo, NULL);
596}
597
598/*
599 * Do the actual switch; called from the BIO completion routine
600 */
601static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
602{
603 struct file *file = p->file;
604 struct file *old_file = lo->lo_backing_file;
605 struct address_space *mapping;
606
607 /* if no new file, only flush of queued bios requested */
608 if (!file)
609 goto out;
610
611 mapping = file->f_mapping;
612 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
613 lo->lo_backing_file = file;
614 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
615 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
616 lo->old_gfp_mask = mapping_gfp_mask(mapping);
617 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
618out:
619 complete(&p->wait);
620}
621
622
623/*
624 * loop_change_fd switched the backing store of a loopback device to
625 * a new file. This is useful for operating system installers to free up
626 * the original file and in High Availability environments to switch to
627 * an alternative location for the content in case of server meltdown.
628 * This can only work if the loop device is used read-only, and if the
629 * new backing store is the same size and type as the old backing store.
630 */
631static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
632 unsigned int arg)
633{
634 struct file *file, *old_file;
635 struct inode *inode;
636 int error;
637
638 error = -ENXIO;
639 if (lo->lo_state != Lo_bound)
640 goto out;
641
642 /* the loop device has to be read-only */
643 error = -EINVAL;
644 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
645 goto out;
646
647 error = -EBADF;
648 file = fget(arg);
649 if (!file)
650 goto out;
651
652 inode = file->f_mapping->host;
653 old_file = lo->lo_backing_file;
654
655 error = -EINVAL;
656
657 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
658 goto out_putf;
659
660 /* size of the new backing store needs to be the same */
661 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
662 goto out_putf;
663
664 /* and ... switch */
665 error = loop_switch(lo, file);
666 if (error)
667 goto out_putf;
668
669 fput(old_file);
670 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
671 ioctl_by_bdev(bdev, BLKRRPART, 0);
672 return 0;
673
674 out_putf:
675 fput(file);
676 out:
677 return error;
678}
679
680static inline int is_loop_device(struct file *file)
681{
682 struct inode *i = file->f_mapping->host;
683
684 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
685}
686
687/* loop sysfs attributes */
688
689static ssize_t loop_attr_show(struct device *dev, char *page,
690 ssize_t (*callback)(struct loop_device *, char *))
691{
692 struct gendisk *disk = dev_to_disk(dev);
693 struct loop_device *lo = disk->private_data;
694
695 return callback(lo, page);
696}
697
698#define LOOP_ATTR_RO(_name) \
699static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
700static ssize_t loop_attr_do_show_##_name(struct device *d, \
701 struct device_attribute *attr, char *b) \
702{ \
703 return loop_attr_show(d, b, loop_attr_##_name##_show); \
704} \
705static struct device_attribute loop_attr_##_name = \
706 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
707
708static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
709{
710 ssize_t ret;
711 char *p = NULL;
712
713 spin_lock_irq(&lo->lo_lock);
714 if (lo->lo_backing_file)
715 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
716 spin_unlock_irq(&lo->lo_lock);
717
718 if (IS_ERR_OR_NULL(p))
719 ret = PTR_ERR(p);
720 else {
721 ret = strlen(p);
722 memmove(buf, p, ret);
723 buf[ret++] = '\n';
724 buf[ret] = 0;
725 }
726
727 return ret;
728}
729
730static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
731{
732 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
733}
734
735static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
736{
737 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
738}
739
740static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
741{
742 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
743
744 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
745}
746
747static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
748{
749 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
750
751 return sprintf(buf, "%s\n", partscan ? "1" : "0");
752}
753
754LOOP_ATTR_RO(backing_file);
755LOOP_ATTR_RO(offset);
756LOOP_ATTR_RO(sizelimit);
757LOOP_ATTR_RO(autoclear);
758LOOP_ATTR_RO(partscan);
759
760static struct attribute *loop_attrs[] = {
761 &loop_attr_backing_file.attr,
762 &loop_attr_offset.attr,
763 &loop_attr_sizelimit.attr,
764 &loop_attr_autoclear.attr,
765 &loop_attr_partscan.attr,
766 NULL,
767};
768
769static struct attribute_group loop_attribute_group = {
770 .name = "loop",
771 .attrs= loop_attrs,
772};
773
774static int loop_sysfs_init(struct loop_device *lo)
775{
776 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
777 &loop_attribute_group);
778}
779
780static void loop_sysfs_exit(struct loop_device *lo)
781{
782 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
783 &loop_attribute_group);
784}
785
786static void loop_config_discard(struct loop_device *lo)
787{
788 struct file *file = lo->lo_backing_file;
789 struct inode *inode = file->f_mapping->host;
790 struct request_queue *q = lo->lo_queue;
791
792 /*
793 * We use punch hole to reclaim the free space used by the
794 * image a.k.a. discard. However we do support discard if
795 * encryption is enabled, because it may give an attacker
796 * useful information.
797 */
798 if ((!file->f_op->fallocate) ||
799 lo->lo_encrypt_key_size) {
800 q->limits.discard_granularity = 0;
801 q->limits.discard_alignment = 0;
802 q->limits.max_discard_sectors = 0;
803 q->limits.discard_zeroes_data = 0;
804 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
805 return;
806 }
807
808 q->limits.discard_granularity = inode->i_sb->s_blocksize;
809 q->limits.discard_alignment = 0;
810 q->limits.max_discard_sectors = UINT_MAX >> 9;
811 q->limits.discard_zeroes_data = 1;
812 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
813}
814
815static int loop_set_fd(struct loop_device *lo, fmode_t mode,
816 struct block_device *bdev, unsigned int arg)
817{
818 struct file *file, *f;
819 struct inode *inode;
820 struct address_space *mapping;
821 unsigned lo_blocksize;
822 int lo_flags = 0;
823 int error;
824 loff_t size;
825
826 /* This is safe, since we have a reference from open(). */
827 __module_get(THIS_MODULE);
828
829 error = -EBADF;
830 file = fget(arg);
831 if (!file)
832 goto out;
833
834 error = -EBUSY;
835 if (lo->lo_state != Lo_unbound)
836 goto out_putf;
837
838 /* Avoid recursion */
839 f = file;
840 while (is_loop_device(f)) {
841 struct loop_device *l;
842
843 if (f->f_mapping->host->i_bdev == bdev)
844 goto out_putf;
845
846 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
847 if (l->lo_state == Lo_unbound) {
848 error = -EINVAL;
849 goto out_putf;
850 }
851 f = l->lo_backing_file;
852 }
853
854 mapping = file->f_mapping;
855 inode = mapping->host;
856
857 error = -EINVAL;
858 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
859 goto out_putf;
860
861 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
862 !file->f_op->write)
863 lo_flags |= LO_FLAGS_READ_ONLY;
864
865 lo_blocksize = S_ISBLK(inode->i_mode) ?
866 inode->i_bdev->bd_block_size : PAGE_SIZE;
867
868 error = -EFBIG;
869 size = get_loop_size(lo, file);
870 if ((loff_t)(sector_t)size != size)
871 goto out_putf;
872
873 error = 0;
874
875 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
876
877 lo->lo_blocksize = lo_blocksize;
878 lo->lo_device = bdev;
879 lo->lo_flags = lo_flags;
880 lo->lo_backing_file = file;
881 lo->transfer = transfer_none;
882 lo->ioctl = NULL;
883 lo->lo_sizelimit = 0;
884 lo->lo_bio_count = 0;
885 lo->old_gfp_mask = mapping_gfp_mask(mapping);
886 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
887
888 bio_list_init(&lo->lo_bio_list);
889
890 /*
891 * set queue make_request_fn, and add limits based on lower level
892 * device
893 */
894 blk_queue_make_request(lo->lo_queue, loop_make_request);
895 lo->lo_queue->queuedata = lo;
896
897 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
898 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
899
900 set_capacity(lo->lo_disk, size);
901 bd_set_size(bdev, size << 9);
902 loop_sysfs_init(lo);
903 /* let user-space know about the new size */
904 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
905
906 set_blocksize(bdev, lo_blocksize);
907
908 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
909 lo->lo_number);
910 if (IS_ERR(lo->lo_thread)) {
911 error = PTR_ERR(lo->lo_thread);
912 goto out_clr;
913 }
914 lo->lo_state = Lo_bound;
915 wake_up_process(lo->lo_thread);
916 if (part_shift)
917 lo->lo_flags |= LO_FLAGS_PARTSCAN;
918 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
919 ioctl_by_bdev(bdev, BLKRRPART, 0);
920 return 0;
921
922out_clr:
923 loop_sysfs_exit(lo);
924 lo->lo_thread = NULL;
925 lo->lo_device = NULL;
926 lo->lo_backing_file = NULL;
927 lo->lo_flags = 0;
928 set_capacity(lo->lo_disk, 0);
929 invalidate_bdev(bdev);
930 bd_set_size(bdev, 0);
931 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
932 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
933 lo->lo_state = Lo_unbound;
934 out_putf:
935 fput(file);
936 out:
937 /* This is safe: open() is still holding a reference. */
938 module_put(THIS_MODULE);
939 return error;
940}
941
942static int
943loop_release_xfer(struct loop_device *lo)
944{
945 int err = 0;
946 struct loop_func_table *xfer = lo->lo_encryption;
947
948 if (xfer) {
949 if (xfer->release)
950 err = xfer->release(lo);
951 lo->transfer = NULL;
952 lo->lo_encryption = NULL;
953 module_put(xfer->owner);
954 }
955 return err;
956}
957
958static int
959loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
960 const struct loop_info64 *i)
961{
962 int err = 0;
963
964 if (xfer) {
965 struct module *owner = xfer->owner;
966
967 if (!try_module_get(owner))
968 return -EINVAL;
969 if (xfer->init)
970 err = xfer->init(lo, i);
971 if (err)
972 module_put(owner);
973 else
974 lo->lo_encryption = xfer;
975 }
976 return err;
977}
978
979static int loop_clr_fd(struct loop_device *lo)
980{
981 struct file *filp = lo->lo_backing_file;
982 gfp_t gfp = lo->old_gfp_mask;
983 struct block_device *bdev = lo->lo_device;
984
985 if (lo->lo_state != Lo_bound)
986 return -ENXIO;
987
988 /*
989 * If we've explicitly asked to tear down the loop device,
990 * and it has an elevated reference count, set it for auto-teardown when
991 * the last reference goes away. This stops $!~#$@ udev from
992 * preventing teardown because it decided that it needs to run blkid on
993 * the loopback device whenever they appear. xfstests is notorious for
994 * failing tests because blkid via udev races with a losetup
995 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
996 * command to fail with EBUSY.
997 */
998 if (lo->lo_refcnt > 1) {
999 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1000 mutex_unlock(&lo->lo_ctl_mutex);
1001 return 0;
1002 }
1003
1004 if (filp == NULL)
1005 return -EINVAL;
1006
1007 spin_lock_irq(&lo->lo_lock);
1008 lo->lo_state = Lo_rundown;
1009 spin_unlock_irq(&lo->lo_lock);
1010
1011 kthread_stop(lo->lo_thread);
1012
1013 spin_lock_irq(&lo->lo_lock);
1014 lo->lo_backing_file = NULL;
1015 spin_unlock_irq(&lo->lo_lock);
1016
1017 loop_release_xfer(lo);
1018 lo->transfer = NULL;
1019 lo->ioctl = NULL;
1020 lo->lo_device = NULL;
1021 lo->lo_encryption = NULL;
1022 lo->lo_offset = 0;
1023 lo->lo_sizelimit = 0;
1024 lo->lo_encrypt_key_size = 0;
1025 lo->lo_thread = NULL;
1026 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1027 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1028 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1029 if (bdev)
1030 invalidate_bdev(bdev);
1031 set_capacity(lo->lo_disk, 0);
1032 loop_sysfs_exit(lo);
1033 if (bdev) {
1034 bd_set_size(bdev, 0);
1035 /* let user-space know about this change */
1036 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1037 }
1038 mapping_set_gfp_mask(filp->f_mapping, gfp);
1039 lo->lo_state = Lo_unbound;
1040 /* This is safe: open() is still holding a reference. */
1041 module_put(THIS_MODULE);
1042 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1043 ioctl_by_bdev(bdev, BLKRRPART, 0);
1044 lo->lo_flags = 0;
1045 if (!part_shift)
1046 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1047 mutex_unlock(&lo->lo_ctl_mutex);
1048 /*
1049 * Need not hold lo_ctl_mutex to fput backing file.
1050 * Calling fput holding lo_ctl_mutex triggers a circular
1051 * lock dependency possibility warning as fput can take
1052 * bd_mutex which is usually taken before lo_ctl_mutex.
1053 */
1054 fput(filp);
1055 return 0;
1056}
1057
1058static int
1059loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1060{
1061 int err;
1062 struct loop_func_table *xfer;
1063 kuid_t uid = current_uid();
1064
1065 if (lo->lo_encrypt_key_size &&
1066 !uid_eq(lo->lo_key_owner, uid) &&
1067 !capable(CAP_SYS_ADMIN))
1068 return -EPERM;
1069 if (lo->lo_state != Lo_bound)
1070 return -ENXIO;
1071 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1072 return -EINVAL;
1073
1074 err = loop_release_xfer(lo);
1075 if (err)
1076 return err;
1077
1078 if (info->lo_encrypt_type) {
1079 unsigned int type = info->lo_encrypt_type;
1080
1081 if (type >= MAX_LO_CRYPT)
1082 return -EINVAL;
1083 xfer = xfer_funcs[type];
1084 if (xfer == NULL)
1085 return -EINVAL;
1086 } else
1087 xfer = NULL;
1088
1089 err = loop_init_xfer(lo, xfer, info);
1090 if (err)
1091 return err;
1092
1093 if (lo->lo_offset != info->lo_offset ||
1094 lo->lo_sizelimit != info->lo_sizelimit) {
1095 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1096 return -EFBIG;
1097 }
1098 loop_config_discard(lo);
1099
1100 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1101 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1102 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1103 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1104
1105 if (!xfer)
1106 xfer = &none_funcs;
1107 lo->transfer = xfer->transfer;
1108 lo->ioctl = xfer->ioctl;
1109
1110 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1111 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1112 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1113
1114 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1115 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1116 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1117 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1118 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1119 }
1120
1121 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1122 lo->lo_init[0] = info->lo_init[0];
1123 lo->lo_init[1] = info->lo_init[1];
1124 if (info->lo_encrypt_key_size) {
1125 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1126 info->lo_encrypt_key_size);
1127 lo->lo_key_owner = uid;
1128 }
1129
1130 return 0;
1131}
1132
1133static int
1134loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1135{
1136 struct file *file = lo->lo_backing_file;
1137 struct kstat stat;
1138 int error;
1139
1140 if (lo->lo_state != Lo_bound)
1141 return -ENXIO;
1142 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1143 if (error)
1144 return error;
1145 memset(info, 0, sizeof(*info));
1146 info->lo_number = lo->lo_number;
1147 info->lo_device = huge_encode_dev(stat.dev);
1148 info->lo_inode = stat.ino;
1149 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1150 info->lo_offset = lo->lo_offset;
1151 info->lo_sizelimit = lo->lo_sizelimit;
1152 info->lo_flags = lo->lo_flags;
1153 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1154 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1155 info->lo_encrypt_type =
1156 lo->lo_encryption ? lo->lo_encryption->number : 0;
1157 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1158 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1159 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1160 lo->lo_encrypt_key_size);
1161 }
1162 return 0;
1163}
1164
1165static void
1166loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1167{
1168 memset(info64, 0, sizeof(*info64));
1169 info64->lo_number = info->lo_number;
1170 info64->lo_device = info->lo_device;
1171 info64->lo_inode = info->lo_inode;
1172 info64->lo_rdevice = info->lo_rdevice;
1173 info64->lo_offset = info->lo_offset;
1174 info64->lo_sizelimit = 0;
1175 info64->lo_encrypt_type = info->lo_encrypt_type;
1176 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1177 info64->lo_flags = info->lo_flags;
1178 info64->lo_init[0] = info->lo_init[0];
1179 info64->lo_init[1] = info->lo_init[1];
1180 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1181 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1182 else
1183 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1184 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1185}
1186
1187static int
1188loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1189{
1190 memset(info, 0, sizeof(*info));
1191 info->lo_number = info64->lo_number;
1192 info->lo_device = info64->lo_device;
1193 info->lo_inode = info64->lo_inode;
1194 info->lo_rdevice = info64->lo_rdevice;
1195 info->lo_offset = info64->lo_offset;
1196 info->lo_encrypt_type = info64->lo_encrypt_type;
1197 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1198 info->lo_flags = info64->lo_flags;
1199 info->lo_init[0] = info64->lo_init[0];
1200 info->lo_init[1] = info64->lo_init[1];
1201 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1202 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1203 else
1204 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1205 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1206
1207 /* error in case values were truncated */
1208 if (info->lo_device != info64->lo_device ||
1209 info->lo_rdevice != info64->lo_rdevice ||
1210 info->lo_inode != info64->lo_inode ||
1211 info->lo_offset != info64->lo_offset)
1212 return -EOVERFLOW;
1213
1214 return 0;
1215}
1216
1217static int
1218loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1219{
1220 struct loop_info info;
1221 struct loop_info64 info64;
1222
1223 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1224 return -EFAULT;
1225 loop_info64_from_old(&info, &info64);
1226 return loop_set_status(lo, &info64);
1227}
1228
1229static int
1230loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1231{
1232 struct loop_info64 info64;
1233
1234 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1235 return -EFAULT;
1236 return loop_set_status(lo, &info64);
1237}
1238
1239static int
1240loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1241 struct loop_info info;
1242 struct loop_info64 info64;
1243 int err = 0;
1244
1245 if (!arg)
1246 err = -EINVAL;
1247 if (!err)
1248 err = loop_get_status(lo, &info64);
1249 if (!err)
1250 err = loop_info64_to_old(&info64, &info);
1251 if (!err && copy_to_user(arg, &info, sizeof(info)))
1252 err = -EFAULT;
1253
1254 return err;
1255}
1256
1257static int
1258loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1259 struct loop_info64 info64;
1260 int err = 0;
1261
1262 if (!arg)
1263 err = -EINVAL;
1264 if (!err)
1265 err = loop_get_status(lo, &info64);
1266 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1267 err = -EFAULT;
1268
1269 return err;
1270}
1271
1272static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1273{
1274 int err;
1275 sector_t sec;
1276 loff_t sz;
1277
1278 err = -ENXIO;
1279 if (unlikely(lo->lo_state != Lo_bound))
1280 goto out;
1281 err = figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1282 if (unlikely(err))
1283 goto out;
1284 sec = get_capacity(lo->lo_disk);
1285 /* the width of sector_t may be narrow for bit-shift */
1286 sz = sec;
1287 sz <<= 9;
1288 mutex_lock(&bdev->bd_mutex);
1289 bd_set_size(bdev, sz);
1290 /* let user-space know about the new size */
1291 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1292 mutex_unlock(&bdev->bd_mutex);
1293
1294 out:
1295 return err;
1296}
1297
1298static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1299 unsigned int cmd, unsigned long arg)
1300{
1301 struct loop_device *lo = bdev->bd_disk->private_data;
1302 int err;
1303
1304 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1305 switch (cmd) {
1306 case LOOP_SET_FD:
1307 err = loop_set_fd(lo, mode, bdev, arg);
1308 break;
1309 case LOOP_CHANGE_FD:
1310 err = loop_change_fd(lo, bdev, arg);
1311 break;
1312 case LOOP_CLR_FD:
1313 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1314 err = loop_clr_fd(lo);
1315 if (!err)
1316 goto out_unlocked;
1317 break;
1318 case LOOP_SET_STATUS:
1319 err = -EPERM;
1320 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1321 err = loop_set_status_old(lo,
1322 (struct loop_info __user *)arg);
1323 break;
1324 case LOOP_GET_STATUS:
1325 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1326 break;
1327 case LOOP_SET_STATUS64:
1328 err = -EPERM;
1329 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1330 err = loop_set_status64(lo,
1331 (struct loop_info64 __user *) arg);
1332 break;
1333 case LOOP_GET_STATUS64:
1334 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1335 break;
1336 case LOOP_SET_CAPACITY:
1337 err = -EPERM;
1338 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1339 err = loop_set_capacity(lo, bdev);
1340 break;
1341 default:
1342 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1343 }
1344 mutex_unlock(&lo->lo_ctl_mutex);
1345
1346out_unlocked:
1347 return err;
1348}
1349
1350#ifdef CONFIG_COMPAT
1351struct compat_loop_info {
1352 compat_int_t lo_number; /* ioctl r/o */
1353 compat_dev_t lo_device; /* ioctl r/o */
1354 compat_ulong_t lo_inode; /* ioctl r/o */
1355 compat_dev_t lo_rdevice; /* ioctl r/o */
1356 compat_int_t lo_offset;
1357 compat_int_t lo_encrypt_type;
1358 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1359 compat_int_t lo_flags; /* ioctl r/o */
1360 char lo_name[LO_NAME_SIZE];
1361 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1362 compat_ulong_t lo_init[2];
1363 char reserved[4];
1364};
1365
1366/*
1367 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1368 * - noinlined to reduce stack space usage in main part of driver
1369 */
1370static noinline int
1371loop_info64_from_compat(const struct compat_loop_info __user *arg,
1372 struct loop_info64 *info64)
1373{
1374 struct compat_loop_info info;
1375
1376 if (copy_from_user(&info, arg, sizeof(info)))
1377 return -EFAULT;
1378
1379 memset(info64, 0, sizeof(*info64));
1380 info64->lo_number = info.lo_number;
1381 info64->lo_device = info.lo_device;
1382 info64->lo_inode = info.lo_inode;
1383 info64->lo_rdevice = info.lo_rdevice;
1384 info64->lo_offset = info.lo_offset;
1385 info64->lo_sizelimit = 0;
1386 info64->lo_encrypt_type = info.lo_encrypt_type;
1387 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1388 info64->lo_flags = info.lo_flags;
1389 info64->lo_init[0] = info.lo_init[0];
1390 info64->lo_init[1] = info.lo_init[1];
1391 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1392 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1393 else
1394 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1395 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1396 return 0;
1397}
1398
1399/*
1400 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1401 * - noinlined to reduce stack space usage in main part of driver
1402 */
1403static noinline int
1404loop_info64_to_compat(const struct loop_info64 *info64,
1405 struct compat_loop_info __user *arg)
1406{
1407 struct compat_loop_info info;
1408
1409 memset(&info, 0, sizeof(info));
1410 info.lo_number = info64->lo_number;
1411 info.lo_device = info64->lo_device;
1412 info.lo_inode = info64->lo_inode;
1413 info.lo_rdevice = info64->lo_rdevice;
1414 info.lo_offset = info64->lo_offset;
1415 info.lo_encrypt_type = info64->lo_encrypt_type;
1416 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1417 info.lo_flags = info64->lo_flags;
1418 info.lo_init[0] = info64->lo_init[0];
1419 info.lo_init[1] = info64->lo_init[1];
1420 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1421 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1422 else
1423 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1424 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1425
1426 /* error in case values were truncated */
1427 if (info.lo_device != info64->lo_device ||
1428 info.lo_rdevice != info64->lo_rdevice ||
1429 info.lo_inode != info64->lo_inode ||
1430 info.lo_offset != info64->lo_offset ||
1431 info.lo_init[0] != info64->lo_init[0] ||
1432 info.lo_init[1] != info64->lo_init[1])
1433 return -EOVERFLOW;
1434
1435 if (copy_to_user(arg, &info, sizeof(info)))
1436 return -EFAULT;
1437 return 0;
1438}
1439
1440static int
1441loop_set_status_compat(struct loop_device *lo,
1442 const struct compat_loop_info __user *arg)
1443{
1444 struct loop_info64 info64;
1445 int ret;
1446
1447 ret = loop_info64_from_compat(arg, &info64);
1448 if (ret < 0)
1449 return ret;
1450 return loop_set_status(lo, &info64);
1451}
1452
1453static int
1454loop_get_status_compat(struct loop_device *lo,
1455 struct compat_loop_info __user *arg)
1456{
1457 struct loop_info64 info64;
1458 int err = 0;
1459
1460 if (!arg)
1461 err = -EINVAL;
1462 if (!err)
1463 err = loop_get_status(lo, &info64);
1464 if (!err)
1465 err = loop_info64_to_compat(&info64, arg);
1466 return err;
1467}
1468
1469static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1470 unsigned int cmd, unsigned long arg)
1471{
1472 struct loop_device *lo = bdev->bd_disk->private_data;
1473 int err;
1474
1475 switch(cmd) {
1476 case LOOP_SET_STATUS:
1477 mutex_lock(&lo->lo_ctl_mutex);
1478 err = loop_set_status_compat(
1479 lo, (const struct compat_loop_info __user *) arg);
1480 mutex_unlock(&lo->lo_ctl_mutex);
1481 break;
1482 case LOOP_GET_STATUS:
1483 mutex_lock(&lo->lo_ctl_mutex);
1484 err = loop_get_status_compat(
1485 lo, (struct compat_loop_info __user *) arg);
1486 mutex_unlock(&lo->lo_ctl_mutex);
1487 break;
1488 case LOOP_SET_CAPACITY:
1489 case LOOP_CLR_FD:
1490 case LOOP_GET_STATUS64:
1491 case LOOP_SET_STATUS64:
1492 arg = (unsigned long) compat_ptr(arg);
1493 case LOOP_SET_FD:
1494 case LOOP_CHANGE_FD:
1495 err = lo_ioctl(bdev, mode, cmd, arg);
1496 break;
1497 default:
1498 err = -ENOIOCTLCMD;
1499 break;
1500 }
1501 return err;
1502}
1503#endif
1504
1505static int lo_open(struct block_device *bdev, fmode_t mode)
1506{
1507 struct loop_device *lo;
1508 int err = 0;
1509
1510 mutex_lock(&loop_index_mutex);
1511 lo = bdev->bd_disk->private_data;
1512 if (!lo) {
1513 err = -ENXIO;
1514 goto out;
1515 }
1516
1517 mutex_lock(&lo->lo_ctl_mutex);
1518 lo->lo_refcnt++;
1519 mutex_unlock(&lo->lo_ctl_mutex);
1520out:
1521 mutex_unlock(&loop_index_mutex);
1522 return err;
1523}
1524
1525static int lo_release(struct gendisk *disk, fmode_t mode)
1526{
1527 struct loop_device *lo = disk->private_data;
1528 int err;
1529
1530 mutex_lock(&lo->lo_ctl_mutex);
1531
1532 if (--lo->lo_refcnt)
1533 goto out;
1534
1535 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1536 /*
1537 * In autoclear mode, stop the loop thread
1538 * and remove configuration after last close.
1539 */
1540 err = loop_clr_fd(lo);
1541 if (!err)
1542 goto out_unlocked;
1543 } else {
1544 /*
1545 * Otherwise keep thread (if running) and config,
1546 * but flush possible ongoing bios in thread.
1547 */
1548 loop_flush(lo);
1549 }
1550
1551out:
1552 mutex_unlock(&lo->lo_ctl_mutex);
1553out_unlocked:
1554 return 0;
1555}
1556
1557static const struct block_device_operations lo_fops = {
1558 .owner = THIS_MODULE,
1559 .open = lo_open,
1560 .release = lo_release,
1561 .ioctl = lo_ioctl,
1562#ifdef CONFIG_COMPAT
1563 .compat_ioctl = lo_compat_ioctl,
1564#endif
1565};
1566
1567/*
1568 * And now the modules code and kernel interface.
1569 */
1570static int max_loop;
1571module_param(max_loop, int, S_IRUGO);
1572MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1573module_param(max_part, int, S_IRUGO);
1574MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1575MODULE_LICENSE("GPL");
1576MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1577
1578int loop_register_transfer(struct loop_func_table *funcs)
1579{
1580 unsigned int n = funcs->number;
1581
1582 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1583 return -EINVAL;
1584 xfer_funcs[n] = funcs;
1585 return 0;
1586}
1587
1588static int unregister_transfer_cb(int id, void *ptr, void *data)
1589{
1590 struct loop_device *lo = ptr;
1591 struct loop_func_table *xfer = data;
1592
1593 mutex_lock(&lo->lo_ctl_mutex);
1594 if (lo->lo_encryption == xfer)
1595 loop_release_xfer(lo);
1596 mutex_unlock(&lo->lo_ctl_mutex);
1597 return 0;
1598}
1599
1600int loop_unregister_transfer(int number)
1601{
1602 unsigned int n = number;
1603 struct loop_func_table *xfer;
1604
1605 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1606 return -EINVAL;
1607
1608 xfer_funcs[n] = NULL;
1609 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1610 return 0;
1611}
1612
1613EXPORT_SYMBOL(loop_register_transfer);
1614EXPORT_SYMBOL(loop_unregister_transfer);
1615
1616static int loop_add(struct loop_device **l, int i)
1617{
1618 struct loop_device *lo;
1619 struct gendisk *disk;
1620 int err;
1621
1622 err = -ENOMEM;
1623 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1624 if (!lo)
1625 goto out;
1626
1627 if (!idr_pre_get(&loop_index_idr, GFP_KERNEL))
1628 goto out_free_dev;
1629
1630 if (i >= 0) {
1631 int m;
1632
1633 /* create specific i in the index */
1634 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1635 if (err >= 0 && i != m) {
1636 idr_remove(&loop_index_idr, m);
1637 err = -EEXIST;
1638 }
1639 } else if (i == -1) {
1640 int m;
1641
1642 /* get next free nr */
1643 err = idr_get_new(&loop_index_idr, lo, &m);
1644 if (err >= 0)
1645 i = m;
1646 } else {
1647 err = -EINVAL;
1648 }
1649 if (err < 0)
1650 goto out_free_dev;
1651
1652 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1653 if (!lo->lo_queue)
1654 goto out_free_dev;
1655
1656 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1657 if (!disk)
1658 goto out_free_queue;
1659
1660 /*
1661 * Disable partition scanning by default. The in-kernel partition
1662 * scanning can be requested individually per-device during its
1663 * setup. Userspace can always add and remove partitions from all
1664 * devices. The needed partition minors are allocated from the
1665 * extended minor space, the main loop device numbers will continue
1666 * to match the loop minors, regardless of the number of partitions
1667 * used.
1668 *
1669 * If max_part is given, partition scanning is globally enabled for
1670 * all loop devices. The minors for the main loop devices will be
1671 * multiples of max_part.
1672 *
1673 * Note: Global-for-all-devices, set-only-at-init, read-only module
1674 * parameteters like 'max_loop' and 'max_part' make things needlessly
1675 * complicated, are too static, inflexible and may surprise
1676 * userspace tools. Parameters like this in general should be avoided.
1677 */
1678 if (!part_shift)
1679 disk->flags |= GENHD_FL_NO_PART_SCAN;
1680 disk->flags |= GENHD_FL_EXT_DEVT;
1681 mutex_init(&lo->lo_ctl_mutex);
1682 lo->lo_number = i;
1683 lo->lo_thread = NULL;
1684 init_waitqueue_head(&lo->lo_event);
1685 init_waitqueue_head(&lo->lo_req_wait);
1686 spin_lock_init(&lo->lo_lock);
1687 disk->major = LOOP_MAJOR;
1688 disk->first_minor = i << part_shift;
1689 disk->fops = &lo_fops;
1690 disk->private_data = lo;
1691 disk->queue = lo->lo_queue;
1692 sprintf(disk->disk_name, "loop%d", i);
1693 add_disk(disk);
1694 *l = lo;
1695 return lo->lo_number;
1696
1697out_free_queue:
1698 blk_cleanup_queue(lo->lo_queue);
1699out_free_dev:
1700 kfree(lo);
1701out:
1702 return err;
1703}
1704
1705static void loop_remove(struct loop_device *lo)
1706{
1707 del_gendisk(lo->lo_disk);
1708 blk_cleanup_queue(lo->lo_queue);
1709 put_disk(lo->lo_disk);
1710 kfree(lo);
1711}
1712
1713static int find_free_cb(int id, void *ptr, void *data)
1714{
1715 struct loop_device *lo = ptr;
1716 struct loop_device **l = data;
1717
1718 if (lo->lo_state == Lo_unbound) {
1719 *l = lo;
1720 return 1;
1721 }
1722 return 0;
1723}
1724
1725static int loop_lookup(struct loop_device **l, int i)
1726{
1727 struct loop_device *lo;
1728 int ret = -ENODEV;
1729
1730 if (i < 0) {
1731 int err;
1732
1733 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1734 if (err == 1) {
1735 *l = lo;
1736 ret = lo->lo_number;
1737 }
1738 goto out;
1739 }
1740
1741 /* lookup and return a specific i */
1742 lo = idr_find(&loop_index_idr, i);
1743 if (lo) {
1744 *l = lo;
1745 ret = lo->lo_number;
1746 }
1747out:
1748 return ret;
1749}
1750
1751static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1752{
1753 struct loop_device *lo;
1754 struct kobject *kobj;
1755 int err;
1756
1757 mutex_lock(&loop_index_mutex);
1758 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1759 if (err < 0)
1760 err = loop_add(&lo, MINOR(dev) >> part_shift);
1761 if (err < 0)
1762 kobj = ERR_PTR(err);
1763 else
1764 kobj = get_disk(lo->lo_disk);
1765 mutex_unlock(&loop_index_mutex);
1766
1767 *part = 0;
1768 return kobj;
1769}
1770
1771static long loop_control_ioctl(struct file *file, unsigned int cmd,
1772 unsigned long parm)
1773{
1774 struct loop_device *lo;
1775 int ret = -ENOSYS;
1776
1777 mutex_lock(&loop_index_mutex);
1778 switch (cmd) {
1779 case LOOP_CTL_ADD:
1780 ret = loop_lookup(&lo, parm);
1781 if (ret >= 0) {
1782 ret = -EEXIST;
1783 break;
1784 }
1785 ret = loop_add(&lo, parm);
1786 break;
1787 case LOOP_CTL_REMOVE:
1788 ret = loop_lookup(&lo, parm);
1789 if (ret < 0)
1790 break;
1791 mutex_lock(&lo->lo_ctl_mutex);
1792 if (lo->lo_state != Lo_unbound) {
1793 ret = -EBUSY;
1794 mutex_unlock(&lo->lo_ctl_mutex);
1795 break;
1796 }
1797 if (lo->lo_refcnt > 0) {
1798 ret = -EBUSY;
1799 mutex_unlock(&lo->lo_ctl_mutex);
1800 break;
1801 }
1802 lo->lo_disk->private_data = NULL;
1803 mutex_unlock(&lo->lo_ctl_mutex);
1804 idr_remove(&loop_index_idr, lo->lo_number);
1805 loop_remove(lo);
1806 break;
1807 case LOOP_CTL_GET_FREE:
1808 ret = loop_lookup(&lo, -1);
1809 if (ret >= 0)
1810 break;
1811 ret = loop_add(&lo, -1);
1812 }
1813 mutex_unlock(&loop_index_mutex);
1814
1815 return ret;
1816}
1817
1818static const struct file_operations loop_ctl_fops = {
1819 .open = nonseekable_open,
1820 .unlocked_ioctl = loop_control_ioctl,
1821 .compat_ioctl = loop_control_ioctl,
1822 .owner = THIS_MODULE,
1823 .llseek = noop_llseek,
1824};
1825
1826static struct miscdevice loop_misc = {
1827 .minor = LOOP_CTRL_MINOR,
1828 .name = "loop-control",
1829 .fops = &loop_ctl_fops,
1830};
1831
1832MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1833MODULE_ALIAS("devname:loop-control");
1834
1835static int __init loop_init(void)
1836{
1837 int i, nr;
1838 unsigned long range;
1839 struct loop_device *lo;
1840 int err;
1841
1842 err = misc_register(&loop_misc);
1843 if (err < 0)
1844 return err;
1845
1846 part_shift = 0;
1847 if (max_part > 0) {
1848 part_shift = fls(max_part);
1849
1850 /*
1851 * Adjust max_part according to part_shift as it is exported
1852 * to user space so that user can decide correct minor number
1853 * if [s]he want to create more devices.
1854 *
1855 * Note that -1 is required because partition 0 is reserved
1856 * for the whole disk.
1857 */
1858 max_part = (1UL << part_shift) - 1;
1859 }
1860
1861 if ((1UL << part_shift) > DISK_MAX_PARTS)
1862 return -EINVAL;
1863
1864 if (max_loop > 1UL << (MINORBITS - part_shift))
1865 return -EINVAL;
1866
1867 /*
1868 * If max_loop is specified, create that many devices upfront.
1869 * This also becomes a hard limit. If max_loop is not specified,
1870 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1871 * init time. Loop devices can be requested on-demand with the
1872 * /dev/loop-control interface, or be instantiated by accessing
1873 * a 'dead' device node.
1874 */
1875 if (max_loop) {
1876 nr = max_loop;
1877 range = max_loop << part_shift;
1878 } else {
1879 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1880 range = 1UL << MINORBITS;
1881 }
1882
1883 if (register_blkdev(LOOP_MAJOR, "loop"))
1884 return -EIO;
1885
1886 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1887 THIS_MODULE, loop_probe, NULL, NULL);
1888
1889 /* pre-create number of devices given by config or max_loop */
1890 mutex_lock(&loop_index_mutex);
1891 for (i = 0; i < nr; i++)
1892 loop_add(&lo, i);
1893 mutex_unlock(&loop_index_mutex);
1894
1895 printk(KERN_INFO "loop: module loaded\n");
1896 return 0;
1897}
1898
1899static int loop_exit_cb(int id, void *ptr, void *data)
1900{
1901 struct loop_device *lo = ptr;
1902
1903 loop_remove(lo);
1904 return 0;
1905}
1906
1907static void __exit loop_exit(void)
1908{
1909 unsigned long range;
1910
1911 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1912
1913 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1914 idr_remove_all(&loop_index_idr);
1915 idr_destroy(&loop_index_idr);
1916
1917 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1918 unregister_blkdev(LOOP_MAJOR, "loop");
1919
1920 misc_deregister(&loop_misc);
1921}
1922
1923module_init(loop_init);
1924module_exit(loop_exit);
1925
1926#ifndef MODULE
1927static int __init max_loop_setup(char *str)
1928{
1929 max_loop = simple_strtol(str, NULL, 0);
1930 return 1;
1931}
1932
1933__setup("max_loop=", max_loop_setup);
1934#endif