projects/linux·blob v4.2 · refs

1891 lines · 45.8 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/compat.h>
67#include <linux/suspend.h>
68#include <linux/freezer.h>
69#include <linux/mutex.h>
70#include <linux/writeback.h>
71#include <linux/completion.h>
72#include <linux/highmem.h>
73#include <linux/kthread.h>
74#include <linux/splice.h>
75#include <linux/sysfs.h>
76#include <linux/miscdevice.h>
77#include <linux/falloc.h>
78#include <linux/uio.h>
79#include "loop.h"
80
81#include <asm/uaccess.h>
82
83static DEFINE_IDR(loop_index_idr);
84static DEFINE_MUTEX(loop_index_mutex);
85
86static int max_part;
87static int part_shift;
88
89static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
93{
94 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
96 char *in, *out, *key;
97 int i, keysize;
98
99 if (cmd == READ) {
100 in = raw_buf;
101 out = loop_buf;
102 } else {
103 in = loop_buf;
104 out = raw_buf;
105 }
106
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
111
112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
114 cond_resched();
115 return 0;
116}
117
118static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
119{
120 if (unlikely(info->lo_encrypt_key_size <= 0))
121 return -EINVAL;
122 return 0;
123}
124
125static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
127};
128
129static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
132 .init = xor_init
133};
134
135/* xfer_funcs[0] is special - its release function is never called */
136static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
137 &none_funcs,
138 &xor_funcs
139};
140
141static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
142{
143 loff_t loopsize;
144
145 /* Compute loopsize in bytes */
146 loopsize = i_size_read(file->f_mapping->host);
147 if (offset > 0)
148 loopsize -= offset;
149 /* offset is beyond i_size, weird but possible */
150 if (loopsize < 0)
151 return 0;
152
153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
155 /*
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
158 */
159 return loopsize >> 9;
160}
161
162static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163{
164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
165}
166
167static int
168figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
169{
170 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
171 sector_t x = (sector_t)size;
172 struct block_device *bdev = lo->lo_device;
173
174 if (unlikely((loff_t)x != size))
175 return -EFBIG;
176 if (lo->lo_offset != offset)
177 lo->lo_offset = offset;
178 if (lo->lo_sizelimit != sizelimit)
179 lo->lo_sizelimit = sizelimit;
180 set_capacity(lo->lo_disk, x);
181 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
182 /* let user-space know about the new size */
183 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
184 return 0;
185}
186
187static inline int
188lo_do_transfer(struct loop_device *lo, int cmd,
189 struct page *rpage, unsigned roffs,
190 struct page *lpage, unsigned loffs,
191 int size, sector_t rblock)
192{
193 int ret;
194
195 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
196 if (likely(!ret))
197 return 0;
198
199 printk_ratelimited(KERN_ERR
200 "loop: Transfer error at byte offset %llu, length %i.\n",
201 (unsigned long long)rblock << 9, size);
202 return ret;
203}
204
205static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
206{
207 struct iov_iter i;
208 ssize_t bw;
209
210 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
211
212 file_start_write(file);
213 bw = vfs_iter_write(file, &i, ppos);
214 file_end_write(file);
215
216 if (likely(bw == bvec->bv_len))
217 return 0;
218
219 printk_ratelimited(KERN_ERR
220 "loop: Write error at byte offset %llu, length %i.\n",
221 (unsigned long long)*ppos, bvec->bv_len);
222 if (bw >= 0)
223 bw = -EIO;
224 return bw;
225}
226
227static int lo_write_simple(struct loop_device *lo, struct request *rq,
228 loff_t pos)
229{
230 struct bio_vec bvec;
231 struct req_iterator iter;
232 int ret = 0;
233
234 rq_for_each_segment(bvec, rq, iter) {
235 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
236 if (ret < 0)
237 break;
238 cond_resched();
239 }
240
241 return ret;
242}
243
244/*
245 * This is the slow, transforming version that needs to double buffer the
246 * data as it cannot do the transformations in place without having direct
247 * access to the destination pages of the backing file.
248 */
249static int lo_write_transfer(struct loop_device *lo, struct request *rq,
250 loff_t pos)
251{
252 struct bio_vec bvec, b;
253 struct req_iterator iter;
254 struct page *page;
255 int ret = 0;
256
257 page = alloc_page(GFP_NOIO);
258 if (unlikely(!page))
259 return -ENOMEM;
260
261 rq_for_each_segment(bvec, rq, iter) {
262 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
263 bvec.bv_offset, bvec.bv_len, pos >> 9);
264 if (unlikely(ret))
265 break;
266
267 b.bv_page = page;
268 b.bv_offset = 0;
269 b.bv_len = bvec.bv_len;
270 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
271 if (ret < 0)
272 break;
273 }
274
275 __free_page(page);
276 return ret;
277}
278
279static int lo_read_simple(struct loop_device *lo, struct request *rq,
280 loff_t pos)
281{
282 struct bio_vec bvec;
283 struct req_iterator iter;
284 struct iov_iter i;
285 ssize_t len;
286
287 rq_for_each_segment(bvec, rq, iter) {
288 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
289 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
290 if (len < 0)
291 return len;
292
293 flush_dcache_page(bvec.bv_page);
294
295 if (len != bvec.bv_len) {
296 struct bio *bio;
297
298 __rq_for_each_bio(bio, rq)
299 zero_fill_bio(bio);
300 break;
301 }
302 cond_resched();
303 }
304
305 return 0;
306}
307
308static int lo_read_transfer(struct loop_device *lo, struct request *rq,
309 loff_t pos)
310{
311 struct bio_vec bvec, b;
312 struct req_iterator iter;
313 struct iov_iter i;
314 struct page *page;
315 ssize_t len;
316 int ret = 0;
317
318 page = alloc_page(GFP_NOIO);
319 if (unlikely(!page))
320 return -ENOMEM;
321
322 rq_for_each_segment(bvec, rq, iter) {
323 loff_t offset = pos;
324
325 b.bv_page = page;
326 b.bv_offset = 0;
327 b.bv_len = bvec.bv_len;
328
329 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
330 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
331 if (len < 0) {
332 ret = len;
333 goto out_free_page;
334 }
335
336 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
337 bvec.bv_offset, len, offset >> 9);
338 if (ret)
339 goto out_free_page;
340
341 flush_dcache_page(bvec.bv_page);
342
343 if (len != bvec.bv_len) {
344 struct bio *bio;
345
346 __rq_for_each_bio(bio, rq)
347 zero_fill_bio(bio);
348 break;
349 }
350 }
351
352 ret = 0;
353out_free_page:
354 __free_page(page);
355 return ret;
356}
357
358static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
359{
360 /*
361 * We use punch hole to reclaim the free space used by the
362 * image a.k.a. discard. However we do not support discard if
363 * encryption is enabled, because it may give an attacker
364 * useful information.
365 */
366 struct file *file = lo->lo_backing_file;
367 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
368 int ret;
369
370 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
371 ret = -EOPNOTSUPP;
372 goto out;
373 }
374
375 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
376 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
377 ret = -EIO;
378 out:
379 return ret;
380}
381
382static int lo_req_flush(struct loop_device *lo, struct request *rq)
383{
384 struct file *file = lo->lo_backing_file;
385 int ret = vfs_fsync(file, 0);
386 if (unlikely(ret && ret != -EINVAL))
387 ret = -EIO;
388
389 return ret;
390}
391
392static int do_req_filebacked(struct loop_device *lo, struct request *rq)
393{
394 loff_t pos;
395 int ret;
396
397 pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
398
399 if (rq->cmd_flags & REQ_WRITE) {
400 if (rq->cmd_flags & REQ_FLUSH)
401 ret = lo_req_flush(lo, rq);
402 else if (rq->cmd_flags & REQ_DISCARD)
403 ret = lo_discard(lo, rq, pos);
404 else if (lo->transfer)
405 ret = lo_write_transfer(lo, rq, pos);
406 else
407 ret = lo_write_simple(lo, rq, pos);
408
409 } else {
410 if (lo->transfer)
411 ret = lo_read_transfer(lo, rq, pos);
412 else
413 ret = lo_read_simple(lo, rq, pos);
414 }
415
416 return ret;
417}
418
419struct switch_request {
420 struct file *file;
421 struct completion wait;
422};
423
424/*
425 * Do the actual switch; called from the BIO completion routine
426 */
427static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
428{
429 struct file *file = p->file;
430 struct file *old_file = lo->lo_backing_file;
431 struct address_space *mapping;
432
433 /* if no new file, only flush of queued bios requested */
434 if (!file)
435 return;
436
437 mapping = file->f_mapping;
438 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
439 lo->lo_backing_file = file;
440 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
441 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
442 lo->old_gfp_mask = mapping_gfp_mask(mapping);
443 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
444}
445
446/*
447 * loop_switch performs the hard work of switching a backing store.
448 * First it needs to flush existing IO, it does this by sending a magic
449 * BIO down the pipe. The completion of this BIO does the actual switch.
450 */
451static int loop_switch(struct loop_device *lo, struct file *file)
452{
453 struct switch_request w;
454
455 w.file = file;
456
457 /* freeze queue and wait for completion of scheduled requests */
458 blk_mq_freeze_queue(lo->lo_queue);
459
460 /* do the switch action */
461 do_loop_switch(lo, &w);
462
463 /* unfreeze */
464 blk_mq_unfreeze_queue(lo->lo_queue);
465
466 return 0;
467}
468
469/*
470 * Helper to flush the IOs in loop, but keeping loop thread running
471 */
472static int loop_flush(struct loop_device *lo)
473{
474 return loop_switch(lo, NULL);
475}
476
477static void loop_reread_partitions(struct loop_device *lo,
478 struct block_device *bdev)
479{
480 int rc;
481
482 /*
483 * bd_mutex has been held already in release path, so don't
484 * acquire it if this function is called in such case.
485 *
486 * If the reread partition isn't from release path, lo_refcnt
487 * must be at least one and it can only become zero when the
488 * current holder is released.
489 */
490 if (!atomic_read(&lo->lo_refcnt))
491 rc = __blkdev_reread_part(bdev);
492 else
493 rc = blkdev_reread_part(bdev);
494 if (rc)
495 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
496 __func__, lo->lo_number, lo->lo_file_name, rc);
497}
498
499/*
500 * loop_change_fd switched the backing store of a loopback device to
501 * a new file. This is useful for operating system installers to free up
502 * the original file and in High Availability environments to switch to
503 * an alternative location for the content in case of server meltdown.
504 * This can only work if the loop device is used read-only, and if the
505 * new backing store is the same size and type as the old backing store.
506 */
507static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
508 unsigned int arg)
509{
510 struct file *file, *old_file;
511 struct inode *inode;
512 int error;
513
514 error = -ENXIO;
515 if (lo->lo_state != Lo_bound)
516 goto out;
517
518 /* the loop device has to be read-only */
519 error = -EINVAL;
520 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
521 goto out;
522
523 error = -EBADF;
524 file = fget(arg);
525 if (!file)
526 goto out;
527
528 inode = file->f_mapping->host;
529 old_file = lo->lo_backing_file;
530
531 error = -EINVAL;
532
533 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
534 goto out_putf;
535
536 /* size of the new backing store needs to be the same */
537 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
538 goto out_putf;
539
540 /* and ... switch */
541 error = loop_switch(lo, file);
542 if (error)
543 goto out_putf;
544
545 fput(old_file);
546 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
547 loop_reread_partitions(lo, bdev);
548 return 0;
549
550 out_putf:
551 fput(file);
552 out:
553 return error;
554}
555
556static inline int is_loop_device(struct file *file)
557{
558 struct inode *i = file->f_mapping->host;
559
560 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
561}
562
563/* loop sysfs attributes */
564
565static ssize_t loop_attr_show(struct device *dev, char *page,
566 ssize_t (*callback)(struct loop_device *, char *))
567{
568 struct gendisk *disk = dev_to_disk(dev);
569 struct loop_device *lo = disk->private_data;
570
571 return callback(lo, page);
572}
573
574#define LOOP_ATTR_RO(_name) \
575static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
576static ssize_t loop_attr_do_show_##_name(struct device *d, \
577 struct device_attribute *attr, char *b) \
578{ \
579 return loop_attr_show(d, b, loop_attr_##_name##_show); \
580} \
581static struct device_attribute loop_attr_##_name = \
582 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
583
584static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
585{
586 ssize_t ret;
587 char *p = NULL;
588
589 spin_lock_irq(&lo->lo_lock);
590 if (lo->lo_backing_file)
591 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
592 spin_unlock_irq(&lo->lo_lock);
593
594 if (IS_ERR_OR_NULL(p))
595 ret = PTR_ERR(p);
596 else {
597 ret = strlen(p);
598 memmove(buf, p, ret);
599 buf[ret++] = '\n';
600 buf[ret] = 0;
601 }
602
603 return ret;
604}
605
606static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
607{
608 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
609}
610
611static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
612{
613 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
614}
615
616static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
617{
618 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
619
620 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
621}
622
623static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
624{
625 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
626
627 return sprintf(buf, "%s\n", partscan ? "1" : "0");
628}
629
630LOOP_ATTR_RO(backing_file);
631LOOP_ATTR_RO(offset);
632LOOP_ATTR_RO(sizelimit);
633LOOP_ATTR_RO(autoclear);
634LOOP_ATTR_RO(partscan);
635
636static struct attribute *loop_attrs[] = {
637 &loop_attr_backing_file.attr,
638 &loop_attr_offset.attr,
639 &loop_attr_sizelimit.attr,
640 &loop_attr_autoclear.attr,
641 &loop_attr_partscan.attr,
642 NULL,
643};
644
645static struct attribute_group loop_attribute_group = {
646 .name = "loop",
647 .attrs= loop_attrs,
648};
649
650static int loop_sysfs_init(struct loop_device *lo)
651{
652 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
653 &loop_attribute_group);
654}
655
656static void loop_sysfs_exit(struct loop_device *lo)
657{
658 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
659 &loop_attribute_group);
660}
661
662static void loop_config_discard(struct loop_device *lo)
663{
664 struct file *file = lo->lo_backing_file;
665 struct inode *inode = file->f_mapping->host;
666 struct request_queue *q = lo->lo_queue;
667
668 /*
669 * We use punch hole to reclaim the free space used by the
670 * image a.k.a. discard. However we do not support discard if
671 * encryption is enabled, because it may give an attacker
672 * useful information.
673 */
674 if ((!file->f_op->fallocate) ||
675 lo->lo_encrypt_key_size) {
676 q->limits.discard_granularity = 0;
677 q->limits.discard_alignment = 0;
678 q->limits.max_discard_sectors = 0;
679 q->limits.discard_zeroes_data = 0;
680 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
681 return;
682 }
683
684 q->limits.discard_granularity = inode->i_sb->s_blocksize;
685 q->limits.discard_alignment = 0;
686 q->limits.max_discard_sectors = UINT_MAX >> 9;
687 q->limits.discard_zeroes_data = 1;
688 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
689}
690
691static int loop_set_fd(struct loop_device *lo, fmode_t mode,
692 struct block_device *bdev, unsigned int arg)
693{
694 struct file *file, *f;
695 struct inode *inode;
696 struct address_space *mapping;
697 unsigned lo_blocksize;
698 int lo_flags = 0;
699 int error;
700 loff_t size;
701
702 /* This is safe, since we have a reference from open(). */
703 __module_get(THIS_MODULE);
704
705 error = -EBADF;
706 file = fget(arg);
707 if (!file)
708 goto out;
709
710 error = -EBUSY;
711 if (lo->lo_state != Lo_unbound)
712 goto out_putf;
713
714 /* Avoid recursion */
715 f = file;
716 while (is_loop_device(f)) {
717 struct loop_device *l;
718
719 if (f->f_mapping->host->i_bdev == bdev)
720 goto out_putf;
721
722 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
723 if (l->lo_state == Lo_unbound) {
724 error = -EINVAL;
725 goto out_putf;
726 }
727 f = l->lo_backing_file;
728 }
729
730 mapping = file->f_mapping;
731 inode = mapping->host;
732
733 error = -EINVAL;
734 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
735 goto out_putf;
736
737 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
738 !file->f_op->write_iter)
739 lo_flags |= LO_FLAGS_READ_ONLY;
740
741 lo_blocksize = S_ISBLK(inode->i_mode) ?
742 inode->i_bdev->bd_block_size : PAGE_SIZE;
743
744 error = -EFBIG;
745 size = get_loop_size(lo, file);
746 if ((loff_t)(sector_t)size != size)
747 goto out_putf;
748 error = -ENOMEM;
749 lo->wq = alloc_workqueue("kloopd%d",
750 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 16,
751 lo->lo_number);
752 if (!lo->wq)
753 goto out_putf;
754
755 error = 0;
756
757 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
758
759 lo->lo_blocksize = lo_blocksize;
760 lo->lo_device = bdev;
761 lo->lo_flags = lo_flags;
762 lo->lo_backing_file = file;
763 lo->transfer = NULL;
764 lo->ioctl = NULL;
765 lo->lo_sizelimit = 0;
766 lo->old_gfp_mask = mapping_gfp_mask(mapping);
767 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
768
769 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
770 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
771
772 set_capacity(lo->lo_disk, size);
773 bd_set_size(bdev, size << 9);
774 loop_sysfs_init(lo);
775 /* let user-space know about the new size */
776 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
777
778 set_blocksize(bdev, lo_blocksize);
779
780 lo->lo_state = Lo_bound;
781 if (part_shift)
782 lo->lo_flags |= LO_FLAGS_PARTSCAN;
783 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
784 loop_reread_partitions(lo, bdev);
785
786 /* Grab the block_device to prevent its destruction after we
787 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
788 */
789 bdgrab(bdev);
790 return 0;
791
792 out_putf:
793 fput(file);
794 out:
795 /* This is safe: open() is still holding a reference. */
796 module_put(THIS_MODULE);
797 return error;
798}
799
800static int
801loop_release_xfer(struct loop_device *lo)
802{
803 int err = 0;
804 struct loop_func_table *xfer = lo->lo_encryption;
805
806 if (xfer) {
807 if (xfer->release)
808 err = xfer->release(lo);
809 lo->transfer = NULL;
810 lo->lo_encryption = NULL;
811 module_put(xfer->owner);
812 }
813 return err;
814}
815
816static int
817loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
818 const struct loop_info64 *i)
819{
820 int err = 0;
821
822 if (xfer) {
823 struct module *owner = xfer->owner;
824
825 if (!try_module_get(owner))
826 return -EINVAL;
827 if (xfer->init)
828 err = xfer->init(lo, i);
829 if (err)
830 module_put(owner);
831 else
832 lo->lo_encryption = xfer;
833 }
834 return err;
835}
836
837static int loop_clr_fd(struct loop_device *lo)
838{
839 struct file *filp = lo->lo_backing_file;
840 gfp_t gfp = lo->old_gfp_mask;
841 struct block_device *bdev = lo->lo_device;
842
843 if (lo->lo_state != Lo_bound)
844 return -ENXIO;
845
846 /*
847 * If we've explicitly asked to tear down the loop device,
848 * and it has an elevated reference count, set it for auto-teardown when
849 * the last reference goes away. This stops $!~#$@ udev from
850 * preventing teardown because it decided that it needs to run blkid on
851 * the loopback device whenever they appear. xfstests is notorious for
852 * failing tests because blkid via udev races with a losetup
853 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
854 * command to fail with EBUSY.
855 */
856 if (atomic_read(&lo->lo_refcnt) > 1) {
857 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
858 mutex_unlock(&lo->lo_ctl_mutex);
859 return 0;
860 }
861
862 if (filp == NULL)
863 return -EINVAL;
864
865 /* freeze request queue during the transition */
866 blk_mq_freeze_queue(lo->lo_queue);
867
868 spin_lock_irq(&lo->lo_lock);
869 lo->lo_state = Lo_rundown;
870 lo->lo_backing_file = NULL;
871 spin_unlock_irq(&lo->lo_lock);
872
873 loop_release_xfer(lo);
874 lo->transfer = NULL;
875 lo->ioctl = NULL;
876 lo->lo_device = NULL;
877 lo->lo_encryption = NULL;
878 lo->lo_offset = 0;
879 lo->lo_sizelimit = 0;
880 lo->lo_encrypt_key_size = 0;
881 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
882 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
883 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
884 if (bdev) {
885 bdput(bdev);
886 invalidate_bdev(bdev);
887 }
888 set_capacity(lo->lo_disk, 0);
889 loop_sysfs_exit(lo);
890 if (bdev) {
891 bd_set_size(bdev, 0);
892 /* let user-space know about this change */
893 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
894 }
895 mapping_set_gfp_mask(filp->f_mapping, gfp);
896 lo->lo_state = Lo_unbound;
897 /* This is safe: open() is still holding a reference. */
898 module_put(THIS_MODULE);
899 blk_mq_unfreeze_queue(lo->lo_queue);
900
901 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
902 loop_reread_partitions(lo, bdev);
903 lo->lo_flags = 0;
904 if (!part_shift)
905 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
906 destroy_workqueue(lo->wq);
907 lo->wq = NULL;
908 mutex_unlock(&lo->lo_ctl_mutex);
909 /*
910 * Need not hold lo_ctl_mutex to fput backing file.
911 * Calling fput holding lo_ctl_mutex triggers a circular
912 * lock dependency possibility warning as fput can take
913 * bd_mutex which is usually taken before lo_ctl_mutex.
914 */
915 fput(filp);
916 return 0;
917}
918
919static int
920loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
921{
922 int err;
923 struct loop_func_table *xfer;
924 kuid_t uid = current_uid();
925
926 if (lo->lo_encrypt_key_size &&
927 !uid_eq(lo->lo_key_owner, uid) &&
928 !capable(CAP_SYS_ADMIN))
929 return -EPERM;
930 if (lo->lo_state != Lo_bound)
931 return -ENXIO;
932 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
933 return -EINVAL;
934
935 err = loop_release_xfer(lo);
936 if (err)
937 return err;
938
939 if (info->lo_encrypt_type) {
940 unsigned int type = info->lo_encrypt_type;
941
942 if (type >= MAX_LO_CRYPT)
943 return -EINVAL;
944 xfer = xfer_funcs[type];
945 if (xfer == NULL)
946 return -EINVAL;
947 } else
948 xfer = NULL;
949
950 err = loop_init_xfer(lo, xfer, info);
951 if (err)
952 return err;
953
954 if (lo->lo_offset != info->lo_offset ||
955 lo->lo_sizelimit != info->lo_sizelimit)
956 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
957 return -EFBIG;
958
959 loop_config_discard(lo);
960
961 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
962 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
963 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
964 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
965
966 if (!xfer)
967 xfer = &none_funcs;
968 lo->transfer = xfer->transfer;
969 lo->ioctl = xfer->ioctl;
970
971 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
972 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
973 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
974
975 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
976 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
977 lo->lo_flags |= LO_FLAGS_PARTSCAN;
978 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
979 loop_reread_partitions(lo, lo->lo_device);
980 }
981
982 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
983 lo->lo_init[0] = info->lo_init[0];
984 lo->lo_init[1] = info->lo_init[1];
985 if (info->lo_encrypt_key_size) {
986 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
987 info->lo_encrypt_key_size);
988 lo->lo_key_owner = uid;
989 }
990
991 return 0;
992}
993
994static int
995loop_get_status(struct loop_device *lo, struct loop_info64 *info)
996{
997 struct file *file = lo->lo_backing_file;
998 struct kstat stat;
999 int error;
1000
1001 if (lo->lo_state != Lo_bound)
1002 return -ENXIO;
1003 error = vfs_getattr(&file->f_path, &stat);
1004 if (error)
1005 return error;
1006 memset(info, 0, sizeof(*info));
1007 info->lo_number = lo->lo_number;
1008 info->lo_device = huge_encode_dev(stat.dev);
1009 info->lo_inode = stat.ino;
1010 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1011 info->lo_offset = lo->lo_offset;
1012 info->lo_sizelimit = lo->lo_sizelimit;
1013 info->lo_flags = lo->lo_flags;
1014 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1015 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1016 info->lo_encrypt_type =
1017 lo->lo_encryption ? lo->lo_encryption->number : 0;
1018 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1019 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1020 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1021 lo->lo_encrypt_key_size);
1022 }
1023 return 0;
1024}
1025
1026static void
1027loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1028{
1029 memset(info64, 0, sizeof(*info64));
1030 info64->lo_number = info->lo_number;
1031 info64->lo_device = info->lo_device;
1032 info64->lo_inode = info->lo_inode;
1033 info64->lo_rdevice = info->lo_rdevice;
1034 info64->lo_offset = info->lo_offset;
1035 info64->lo_sizelimit = 0;
1036 info64->lo_encrypt_type = info->lo_encrypt_type;
1037 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1038 info64->lo_flags = info->lo_flags;
1039 info64->lo_init[0] = info->lo_init[0];
1040 info64->lo_init[1] = info->lo_init[1];
1041 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1042 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1043 else
1044 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1045 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1046}
1047
1048static int
1049loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1050{
1051 memset(info, 0, sizeof(*info));
1052 info->lo_number = info64->lo_number;
1053 info->lo_device = info64->lo_device;
1054 info->lo_inode = info64->lo_inode;
1055 info->lo_rdevice = info64->lo_rdevice;
1056 info->lo_offset = info64->lo_offset;
1057 info->lo_encrypt_type = info64->lo_encrypt_type;
1058 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1059 info->lo_flags = info64->lo_flags;
1060 info->lo_init[0] = info64->lo_init[0];
1061 info->lo_init[1] = info64->lo_init[1];
1062 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1063 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1064 else
1065 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1066 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1067
1068 /* error in case values were truncated */
1069 if (info->lo_device != info64->lo_device ||
1070 info->lo_rdevice != info64->lo_rdevice ||
1071 info->lo_inode != info64->lo_inode ||
1072 info->lo_offset != info64->lo_offset)
1073 return -EOVERFLOW;
1074
1075 return 0;
1076}
1077
1078static int
1079loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1080{
1081 struct loop_info info;
1082 struct loop_info64 info64;
1083
1084 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1085 return -EFAULT;
1086 loop_info64_from_old(&info, &info64);
1087 return loop_set_status(lo, &info64);
1088}
1089
1090static int
1091loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1092{
1093 struct loop_info64 info64;
1094
1095 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1096 return -EFAULT;
1097 return loop_set_status(lo, &info64);
1098}
1099
1100static int
1101loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1102 struct loop_info info;
1103 struct loop_info64 info64;
1104 int err = 0;
1105
1106 if (!arg)
1107 err = -EINVAL;
1108 if (!err)
1109 err = loop_get_status(lo, &info64);
1110 if (!err)
1111 err = loop_info64_to_old(&info64, &info);
1112 if (!err && copy_to_user(arg, &info, sizeof(info)))
1113 err = -EFAULT;
1114
1115 return err;
1116}
1117
1118static int
1119loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1120 struct loop_info64 info64;
1121 int err = 0;
1122
1123 if (!arg)
1124 err = -EINVAL;
1125 if (!err)
1126 err = loop_get_status(lo, &info64);
1127 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1128 err = -EFAULT;
1129
1130 return err;
1131}
1132
1133static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1134{
1135 if (unlikely(lo->lo_state != Lo_bound))
1136 return -ENXIO;
1137
1138 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1139}
1140
1141static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1142 unsigned int cmd, unsigned long arg)
1143{
1144 struct loop_device *lo = bdev->bd_disk->private_data;
1145 int err;
1146
1147 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1148 switch (cmd) {
1149 case LOOP_SET_FD:
1150 err = loop_set_fd(lo, mode, bdev, arg);
1151 break;
1152 case LOOP_CHANGE_FD:
1153 err = loop_change_fd(lo, bdev, arg);
1154 break;
1155 case LOOP_CLR_FD:
1156 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1157 err = loop_clr_fd(lo);
1158 if (!err)
1159 goto out_unlocked;
1160 break;
1161 case LOOP_SET_STATUS:
1162 err = -EPERM;
1163 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1164 err = loop_set_status_old(lo,
1165 (struct loop_info __user *)arg);
1166 break;
1167 case LOOP_GET_STATUS:
1168 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1169 break;
1170 case LOOP_SET_STATUS64:
1171 err = -EPERM;
1172 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1173 err = loop_set_status64(lo,
1174 (struct loop_info64 __user *) arg);
1175 break;
1176 case LOOP_GET_STATUS64:
1177 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1178 break;
1179 case LOOP_SET_CAPACITY:
1180 err = -EPERM;
1181 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1182 err = loop_set_capacity(lo, bdev);
1183 break;
1184 default:
1185 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1186 }
1187 mutex_unlock(&lo->lo_ctl_mutex);
1188
1189out_unlocked:
1190 return err;
1191}
1192
1193#ifdef CONFIG_COMPAT
1194struct compat_loop_info {
1195 compat_int_t lo_number; /* ioctl r/o */
1196 compat_dev_t lo_device; /* ioctl r/o */
1197 compat_ulong_t lo_inode; /* ioctl r/o */
1198 compat_dev_t lo_rdevice; /* ioctl r/o */
1199 compat_int_t lo_offset;
1200 compat_int_t lo_encrypt_type;
1201 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1202 compat_int_t lo_flags; /* ioctl r/o */
1203 char lo_name[LO_NAME_SIZE];
1204 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1205 compat_ulong_t lo_init[2];
1206 char reserved[4];
1207};
1208
1209/*
1210 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1211 * - noinlined to reduce stack space usage in main part of driver
1212 */
1213static noinline int
1214loop_info64_from_compat(const struct compat_loop_info __user *arg,
1215 struct loop_info64 *info64)
1216{
1217 struct compat_loop_info info;
1218
1219 if (copy_from_user(&info, arg, sizeof(info)))
1220 return -EFAULT;
1221
1222 memset(info64, 0, sizeof(*info64));
1223 info64->lo_number = info.lo_number;
1224 info64->lo_device = info.lo_device;
1225 info64->lo_inode = info.lo_inode;
1226 info64->lo_rdevice = info.lo_rdevice;
1227 info64->lo_offset = info.lo_offset;
1228 info64->lo_sizelimit = 0;
1229 info64->lo_encrypt_type = info.lo_encrypt_type;
1230 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1231 info64->lo_flags = info.lo_flags;
1232 info64->lo_init[0] = info.lo_init[0];
1233 info64->lo_init[1] = info.lo_init[1];
1234 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1235 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1236 else
1237 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1238 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1239 return 0;
1240}
1241
1242/*
1243 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1244 * - noinlined to reduce stack space usage in main part of driver
1245 */
1246static noinline int
1247loop_info64_to_compat(const struct loop_info64 *info64,
1248 struct compat_loop_info __user *arg)
1249{
1250 struct compat_loop_info info;
1251
1252 memset(&info, 0, sizeof(info));
1253 info.lo_number = info64->lo_number;
1254 info.lo_device = info64->lo_device;
1255 info.lo_inode = info64->lo_inode;
1256 info.lo_rdevice = info64->lo_rdevice;
1257 info.lo_offset = info64->lo_offset;
1258 info.lo_encrypt_type = info64->lo_encrypt_type;
1259 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1260 info.lo_flags = info64->lo_flags;
1261 info.lo_init[0] = info64->lo_init[0];
1262 info.lo_init[1] = info64->lo_init[1];
1263 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1264 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1265 else
1266 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1267 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1268
1269 /* error in case values were truncated */
1270 if (info.lo_device != info64->lo_device ||
1271 info.lo_rdevice != info64->lo_rdevice ||
1272 info.lo_inode != info64->lo_inode ||
1273 info.lo_offset != info64->lo_offset ||
1274 info.lo_init[0] != info64->lo_init[0] ||
1275 info.lo_init[1] != info64->lo_init[1])
1276 return -EOVERFLOW;
1277
1278 if (copy_to_user(arg, &info, sizeof(info)))
1279 return -EFAULT;
1280 return 0;
1281}
1282
1283static int
1284loop_set_status_compat(struct loop_device *lo,
1285 const struct compat_loop_info __user *arg)
1286{
1287 struct loop_info64 info64;
1288 int ret;
1289
1290 ret = loop_info64_from_compat(arg, &info64);
1291 if (ret < 0)
1292 return ret;
1293 return loop_set_status(lo, &info64);
1294}
1295
1296static int
1297loop_get_status_compat(struct loop_device *lo,
1298 struct compat_loop_info __user *arg)
1299{
1300 struct loop_info64 info64;
1301 int err = 0;
1302
1303 if (!arg)
1304 err = -EINVAL;
1305 if (!err)
1306 err = loop_get_status(lo, &info64);
1307 if (!err)
1308 err = loop_info64_to_compat(&info64, arg);
1309 return err;
1310}
1311
1312static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1313 unsigned int cmd, unsigned long arg)
1314{
1315 struct loop_device *lo = bdev->bd_disk->private_data;
1316 int err;
1317
1318 switch(cmd) {
1319 case LOOP_SET_STATUS:
1320 mutex_lock(&lo->lo_ctl_mutex);
1321 err = loop_set_status_compat(
1322 lo, (const struct compat_loop_info __user *) arg);
1323 mutex_unlock(&lo->lo_ctl_mutex);
1324 break;
1325 case LOOP_GET_STATUS:
1326 mutex_lock(&lo->lo_ctl_mutex);
1327 err = loop_get_status_compat(
1328 lo, (struct compat_loop_info __user *) arg);
1329 mutex_unlock(&lo->lo_ctl_mutex);
1330 break;
1331 case LOOP_SET_CAPACITY:
1332 case LOOP_CLR_FD:
1333 case LOOP_GET_STATUS64:
1334 case LOOP_SET_STATUS64:
1335 arg = (unsigned long) compat_ptr(arg);
1336 case LOOP_SET_FD:
1337 case LOOP_CHANGE_FD:
1338 err = lo_ioctl(bdev, mode, cmd, arg);
1339 break;
1340 default:
1341 err = -ENOIOCTLCMD;
1342 break;
1343 }
1344 return err;
1345}
1346#endif
1347
1348static int lo_open(struct block_device *bdev, fmode_t mode)
1349{
1350 struct loop_device *lo;
1351 int err = 0;
1352
1353 mutex_lock(&loop_index_mutex);
1354 lo = bdev->bd_disk->private_data;
1355 if (!lo) {
1356 err = -ENXIO;
1357 goto out;
1358 }
1359
1360 atomic_inc(&lo->lo_refcnt);
1361out:
1362 mutex_unlock(&loop_index_mutex);
1363 return err;
1364}
1365
1366static void lo_release(struct gendisk *disk, fmode_t mode)
1367{
1368 struct loop_device *lo = disk->private_data;
1369 int err;
1370
1371 if (atomic_dec_return(&lo->lo_refcnt))
1372 return;
1373
1374 mutex_lock(&lo->lo_ctl_mutex);
1375 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1376 /*
1377 * In autoclear mode, stop the loop thread
1378 * and remove configuration after last close.
1379 */
1380 err = loop_clr_fd(lo);
1381 if (!err)
1382 return;
1383 } else {
1384 /*
1385 * Otherwise keep thread (if running) and config,
1386 * but flush possible ongoing bios in thread.
1387 */
1388 loop_flush(lo);
1389 }
1390
1391 mutex_unlock(&lo->lo_ctl_mutex);
1392}
1393
1394static const struct block_device_operations lo_fops = {
1395 .owner = THIS_MODULE,
1396 .open = lo_open,
1397 .release = lo_release,
1398 .ioctl = lo_ioctl,
1399#ifdef CONFIG_COMPAT
1400 .compat_ioctl = lo_compat_ioctl,
1401#endif
1402};
1403
1404/*
1405 * And now the modules code and kernel interface.
1406 */
1407static int max_loop;
1408module_param(max_loop, int, S_IRUGO);
1409MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1410module_param(max_part, int, S_IRUGO);
1411MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1412MODULE_LICENSE("GPL");
1413MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1414
1415int loop_register_transfer(struct loop_func_table *funcs)
1416{
1417 unsigned int n = funcs->number;
1418
1419 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1420 return -EINVAL;
1421 xfer_funcs[n] = funcs;
1422 return 0;
1423}
1424
1425static int unregister_transfer_cb(int id, void *ptr, void *data)
1426{
1427 struct loop_device *lo = ptr;
1428 struct loop_func_table *xfer = data;
1429
1430 mutex_lock(&lo->lo_ctl_mutex);
1431 if (lo->lo_encryption == xfer)
1432 loop_release_xfer(lo);
1433 mutex_unlock(&lo->lo_ctl_mutex);
1434 return 0;
1435}
1436
1437int loop_unregister_transfer(int number)
1438{
1439 unsigned int n = number;
1440 struct loop_func_table *xfer;
1441
1442 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1443 return -EINVAL;
1444
1445 xfer_funcs[n] = NULL;
1446 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1447 return 0;
1448}
1449
1450EXPORT_SYMBOL(loop_register_transfer);
1451EXPORT_SYMBOL(loop_unregister_transfer);
1452
1453static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1454 const struct blk_mq_queue_data *bd)
1455{
1456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1457 struct loop_device *lo = cmd->rq->q->queuedata;
1458
1459 blk_mq_start_request(bd->rq);
1460
1461 if (lo->lo_state != Lo_bound)
1462 return -EIO;
1463
1464 if (cmd->rq->cmd_flags & REQ_WRITE) {
1465 struct loop_device *lo = cmd->rq->q->queuedata;
1466 bool need_sched = true;
1467
1468 spin_lock_irq(&lo->lo_lock);
1469 if (lo->write_started)
1470 need_sched = false;
1471 else
1472 lo->write_started = true;
1473 list_add_tail(&cmd->list, &lo->write_cmd_head);
1474 spin_unlock_irq(&lo->lo_lock);
1475
1476 if (need_sched)
1477 queue_work(lo->wq, &lo->write_work);
1478 } else {
1479 queue_work(lo->wq, &cmd->read_work);
1480 }
1481
1482 return BLK_MQ_RQ_QUEUE_OK;
1483}
1484
1485static void loop_handle_cmd(struct loop_cmd *cmd)
1486{
1487 const bool write = cmd->rq->cmd_flags & REQ_WRITE;
1488 struct loop_device *lo = cmd->rq->q->queuedata;
1489 int ret = -EIO;
1490
1491 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY))
1492 goto failed;
1493
1494 ret = do_req_filebacked(lo, cmd->rq);
1495
1496 failed:
1497 if (ret)
1498 cmd->rq->errors = -EIO;
1499 blk_mq_complete_request(cmd->rq);
1500}
1501
1502static void loop_queue_write_work(struct work_struct *work)
1503{
1504 struct loop_device *lo =
1505 container_of(work, struct loop_device, write_work);
1506 LIST_HEAD(cmd_list);
1507
1508 spin_lock_irq(&lo->lo_lock);
1509 repeat:
1510 list_splice_init(&lo->write_cmd_head, &cmd_list);
1511 spin_unlock_irq(&lo->lo_lock);
1512
1513 while (!list_empty(&cmd_list)) {
1514 struct loop_cmd *cmd = list_first_entry(&cmd_list,
1515 struct loop_cmd, list);
1516 list_del_init(&cmd->list);
1517 loop_handle_cmd(cmd);
1518 }
1519
1520 spin_lock_irq(&lo->lo_lock);
1521 if (!list_empty(&lo->write_cmd_head))
1522 goto repeat;
1523 lo->write_started = false;
1524 spin_unlock_irq(&lo->lo_lock);
1525}
1526
1527static void loop_queue_read_work(struct work_struct *work)
1528{
1529 struct loop_cmd *cmd =
1530 container_of(work, struct loop_cmd, read_work);
1531
1532 loop_handle_cmd(cmd);
1533}
1534
1535static int loop_init_request(void *data, struct request *rq,
1536 unsigned int hctx_idx, unsigned int request_idx,
1537 unsigned int numa_node)
1538{
1539 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1540
1541 cmd->rq = rq;
1542 INIT_WORK(&cmd->read_work, loop_queue_read_work);
1543
1544 return 0;
1545}
1546
1547static struct blk_mq_ops loop_mq_ops = {
1548 .queue_rq = loop_queue_rq,
1549 .map_queue = blk_mq_map_queue,
1550 .init_request = loop_init_request,
1551};
1552
1553static int loop_add(struct loop_device **l, int i)
1554{
1555 struct loop_device *lo;
1556 struct gendisk *disk;
1557 int err;
1558
1559 err = -ENOMEM;
1560 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1561 if (!lo)
1562 goto out;
1563
1564 lo->lo_state = Lo_unbound;
1565
1566 /* allocate id, if @id >= 0, we're requesting that specific id */
1567 if (i >= 0) {
1568 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1569 if (err == -ENOSPC)
1570 err = -EEXIST;
1571 } else {
1572 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1573 }
1574 if (err < 0)
1575 goto out_free_dev;
1576 i = err;
1577
1578 err = -ENOMEM;
1579 lo->tag_set.ops = &loop_mq_ops;
1580 lo->tag_set.nr_hw_queues = 1;
1581 lo->tag_set.queue_depth = 128;
1582 lo->tag_set.numa_node = NUMA_NO_NODE;
1583 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1584 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1585 lo->tag_set.driver_data = lo;
1586
1587 err = blk_mq_alloc_tag_set(&lo->tag_set);
1588 if (err)
1589 goto out_free_idr;
1590
1591 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1592 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1593 err = PTR_ERR(lo->lo_queue);
1594 goto out_cleanup_tags;
1595 }
1596 lo->lo_queue->queuedata = lo;
1597
1598 INIT_LIST_HEAD(&lo->write_cmd_head);
1599 INIT_WORK(&lo->write_work, loop_queue_write_work);
1600
1601 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1602 if (!disk)
1603 goto out_free_queue;
1604
1605 /*
1606 * Disable partition scanning by default. The in-kernel partition
1607 * scanning can be requested individually per-device during its
1608 * setup. Userspace can always add and remove partitions from all
1609 * devices. The needed partition minors are allocated from the
1610 * extended minor space, the main loop device numbers will continue
1611 * to match the loop minors, regardless of the number of partitions
1612 * used.
1613 *
1614 * If max_part is given, partition scanning is globally enabled for
1615 * all loop devices. The minors for the main loop devices will be
1616 * multiples of max_part.
1617 *
1618 * Note: Global-for-all-devices, set-only-at-init, read-only module
1619 * parameteters like 'max_loop' and 'max_part' make things needlessly
1620 * complicated, are too static, inflexible and may surprise
1621 * userspace tools. Parameters like this in general should be avoided.
1622 */
1623 if (!part_shift)
1624 disk->flags |= GENHD_FL_NO_PART_SCAN;
1625 disk->flags |= GENHD_FL_EXT_DEVT;
1626 mutex_init(&lo->lo_ctl_mutex);
1627 atomic_set(&lo->lo_refcnt, 0);
1628 lo->lo_number = i;
1629 spin_lock_init(&lo->lo_lock);
1630 disk->major = LOOP_MAJOR;
1631 disk->first_minor = i << part_shift;
1632 disk->fops = &lo_fops;
1633 disk->private_data = lo;
1634 disk->queue = lo->lo_queue;
1635 sprintf(disk->disk_name, "loop%d", i);
1636 add_disk(disk);
1637 *l = lo;
1638 return lo->lo_number;
1639
1640out_free_queue:
1641 blk_cleanup_queue(lo->lo_queue);
1642out_cleanup_tags:
1643 blk_mq_free_tag_set(&lo->tag_set);
1644out_free_idr:
1645 idr_remove(&loop_index_idr, i);
1646out_free_dev:
1647 kfree(lo);
1648out:
1649 return err;
1650}
1651
1652static void loop_remove(struct loop_device *lo)
1653{
1654 blk_cleanup_queue(lo->lo_queue);
1655 del_gendisk(lo->lo_disk);
1656 blk_mq_free_tag_set(&lo->tag_set);
1657 put_disk(lo->lo_disk);
1658 kfree(lo);
1659}
1660
1661static int find_free_cb(int id, void *ptr, void *data)
1662{
1663 struct loop_device *lo = ptr;
1664 struct loop_device **l = data;
1665
1666 if (lo->lo_state == Lo_unbound) {
1667 *l = lo;
1668 return 1;
1669 }
1670 return 0;
1671}
1672
1673static int loop_lookup(struct loop_device **l, int i)
1674{
1675 struct loop_device *lo;
1676 int ret = -ENODEV;
1677
1678 if (i < 0) {
1679 int err;
1680
1681 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1682 if (err == 1) {
1683 *l = lo;
1684 ret = lo->lo_number;
1685 }
1686 goto out;
1687 }
1688
1689 /* lookup and return a specific i */
1690 lo = idr_find(&loop_index_idr, i);
1691 if (lo) {
1692 *l = lo;
1693 ret = lo->lo_number;
1694 }
1695out:
1696 return ret;
1697}
1698
1699static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1700{
1701 struct loop_device *lo;
1702 struct kobject *kobj;
1703 int err;
1704
1705 mutex_lock(&loop_index_mutex);
1706 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1707 if (err < 0)
1708 err = loop_add(&lo, MINOR(dev) >> part_shift);
1709 if (err < 0)
1710 kobj = NULL;
1711 else
1712 kobj = get_disk(lo->lo_disk);
1713 mutex_unlock(&loop_index_mutex);
1714
1715 *part = 0;
1716 return kobj;
1717}
1718
1719static long loop_control_ioctl(struct file *file, unsigned int cmd,
1720 unsigned long parm)
1721{
1722 struct loop_device *lo;
1723 int ret = -ENOSYS;
1724
1725 mutex_lock(&loop_index_mutex);
1726 switch (cmd) {
1727 case LOOP_CTL_ADD:
1728 ret = loop_lookup(&lo, parm);
1729 if (ret >= 0) {
1730 ret = -EEXIST;
1731 break;
1732 }
1733 ret = loop_add(&lo, parm);
1734 break;
1735 case LOOP_CTL_REMOVE:
1736 ret = loop_lookup(&lo, parm);
1737 if (ret < 0)
1738 break;
1739 mutex_lock(&lo->lo_ctl_mutex);
1740 if (lo->lo_state != Lo_unbound) {
1741 ret = -EBUSY;
1742 mutex_unlock(&lo->lo_ctl_mutex);
1743 break;
1744 }
1745 if (atomic_read(&lo->lo_refcnt) > 0) {
1746 ret = -EBUSY;
1747 mutex_unlock(&lo->lo_ctl_mutex);
1748 break;
1749 }
1750 lo->lo_disk->private_data = NULL;
1751 mutex_unlock(&lo->lo_ctl_mutex);
1752 idr_remove(&loop_index_idr, lo->lo_number);
1753 loop_remove(lo);
1754 break;
1755 case LOOP_CTL_GET_FREE:
1756 ret = loop_lookup(&lo, -1);
1757 if (ret >= 0)
1758 break;
1759 ret = loop_add(&lo, -1);
1760 }
1761 mutex_unlock(&loop_index_mutex);
1762
1763 return ret;
1764}
1765
1766static const struct file_operations loop_ctl_fops = {
1767 .open = nonseekable_open,
1768 .unlocked_ioctl = loop_control_ioctl,
1769 .compat_ioctl = loop_control_ioctl,
1770 .owner = THIS_MODULE,
1771 .llseek = noop_llseek,
1772};
1773
1774static struct miscdevice loop_misc = {
1775 .minor = LOOP_CTRL_MINOR,
1776 .name = "loop-control",
1777 .fops = &loop_ctl_fops,
1778};
1779
1780MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1781MODULE_ALIAS("devname:loop-control");
1782
1783static int __init loop_init(void)
1784{
1785 int i, nr;
1786 unsigned long range;
1787 struct loop_device *lo;
1788 int err;
1789
1790 err = misc_register(&loop_misc);
1791 if (err < 0)
1792 return err;
1793
1794 part_shift = 0;
1795 if (max_part > 0) {
1796 part_shift = fls(max_part);
1797
1798 /*
1799 * Adjust max_part according to part_shift as it is exported
1800 * to user space so that user can decide correct minor number
1801 * if [s]he want to create more devices.
1802 *
1803 * Note that -1 is required because partition 0 is reserved
1804 * for the whole disk.
1805 */
1806 max_part = (1UL << part_shift) - 1;
1807 }
1808
1809 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1810 err = -EINVAL;
1811 goto misc_out;
1812 }
1813
1814 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1815 err = -EINVAL;
1816 goto misc_out;
1817 }
1818
1819 /*
1820 * If max_loop is specified, create that many devices upfront.
1821 * This also becomes a hard limit. If max_loop is not specified,
1822 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1823 * init time. Loop devices can be requested on-demand with the
1824 * /dev/loop-control interface, or be instantiated by accessing
1825 * a 'dead' device node.
1826 */
1827 if (max_loop) {
1828 nr = max_loop;
1829 range = max_loop << part_shift;
1830 } else {
1831 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1832 range = 1UL << MINORBITS;
1833 }
1834
1835 if (register_blkdev(LOOP_MAJOR, "loop")) {
1836 err = -EIO;
1837 goto misc_out;
1838 }
1839
1840 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1841 THIS_MODULE, loop_probe, NULL, NULL);
1842
1843 /* pre-create number of devices given by config or max_loop */
1844 mutex_lock(&loop_index_mutex);
1845 for (i = 0; i < nr; i++)
1846 loop_add(&lo, i);
1847 mutex_unlock(&loop_index_mutex);
1848
1849 printk(KERN_INFO "loop: module loaded\n");
1850 return 0;
1851
1852misc_out:
1853 misc_deregister(&loop_misc);
1854 return err;
1855}
1856
1857static int loop_exit_cb(int id, void *ptr, void *data)
1858{
1859 struct loop_device *lo = ptr;
1860
1861 loop_remove(lo);
1862 return 0;
1863}
1864
1865static void __exit loop_exit(void)
1866{
1867 unsigned long range;
1868
1869 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1870
1871 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1872 idr_destroy(&loop_index_idr);
1873
1874 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1875 unregister_blkdev(LOOP_MAJOR, "loop");
1876
1877 misc_deregister(&loop_misc);
1878}
1879
1880module_init(loop_init);
1881module_exit(loop_exit);
1882
1883#ifndef MODULE
1884static int __init max_loop_setup(char *str)
1885{
1886 max_loop = simple_strtol(str, NULL, 0);
1887 return 1;
1888}
1889
1890__setup("max_loop=", max_loop_setup);
1891#endif