1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
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 <linux/ioprio.h>
80#include <linux/blk-cgroup.h>
81
82#include "loop.h"
83
84#include <linux/uaccess.h>
85
86static DEFINE_IDR(loop_index_idr);
87static DEFINE_MUTEX(loop_ctl_mutex);
88
89static int max_part;
90static int part_shift;
91
92static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
96{
97 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
99 char *in, *out, *key;
100 int i, keysize;
101
102 if (cmd == READ) {
103 in = raw_buf;
104 out = loop_buf;
105 } else {
106 in = loop_buf;
107 out = raw_buf;
108 }
109
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
114
115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
117 cond_resched();
118 return 0;
119}
120
121static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122{
123 if (unlikely(info->lo_encrypt_key_size <= 0))
124 return -EINVAL;
125 return 0;
126}
127
128static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
130};
131
132static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
135 .init = xor_init
136};
137
138
139static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140 &none_funcs,
141 &xor_funcs
142};
143
144static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145{
146 loff_t loopsize;
147
148
149 loopsize = i_size_read(file->f_mapping->host);
150 if (offset > 0)
151 loopsize -= offset;
152
153 if (loopsize < 0)
154 return 0;
155
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
158
159
160
161
162 return loopsize >> 9;
163}
164
165static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166{
167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
168}
169
170static void __loop_update_dio(struct loop_device *lo, bool dio)
171{
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
177 bool use_dio;
178
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
182 }
183
184
185
186
187
188
189
190
191
192
193
194 if (dio) {
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
198 !lo->transfer)
199 use_dio = true;
200 else
201 use_dio = false;
202 } else {
203 use_dio = false;
204 }
205
206 if (lo->use_dio == use_dio)
207 return;
208
209
210 vfs_fsync(file, 0);
211
212
213
214
215
216
217 blk_mq_freeze_queue(lo->lo_queue);
218 lo->use_dio = use_dio;
219 if (use_dio) {
220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
222 } else {
223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
225 }
226 blk_mq_unfreeze_queue(lo->lo_queue);
227}
228
229static int
230figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
231{
232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
233 sector_t x = (sector_t)size;
234 struct block_device *bdev = lo->lo_device;
235
236 if (unlikely((loff_t)x != size))
237 return -EFBIG;
238 if (lo->lo_offset != offset)
239 lo->lo_offset = offset;
240 if (lo->lo_sizelimit != sizelimit)
241 lo->lo_sizelimit = sizelimit;
242 set_capacity(lo->lo_disk, x);
243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
244
245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
246 return 0;
247}
248
249static inline int
250lo_do_transfer(struct loop_device *lo, int cmd,
251 struct page *rpage, unsigned roffs,
252 struct page *lpage, unsigned loffs,
253 int size, sector_t rblock)
254{
255 int ret;
256
257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
258 if (likely(!ret))
259 return 0;
260
261 printk_ratelimited(KERN_ERR
262 "loop: Transfer error at byte offset %llu, length %i.\n",
263 (unsigned long long)rblock << 9, size);
264 return ret;
265}
266
267static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
268{
269 struct iov_iter i;
270 ssize_t bw;
271
272 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
273
274 file_start_write(file);
275 bw = vfs_iter_write(file, &i, ppos, 0);
276 file_end_write(file);
277
278 if (likely(bw == bvec->bv_len))
279 return 0;
280
281 printk_ratelimited(KERN_ERR
282 "loop: Write error at byte offset %llu, length %i.\n",
283 (unsigned long long)*ppos, bvec->bv_len);
284 if (bw >= 0)
285 bw = -EIO;
286 return bw;
287}
288
289static int lo_write_simple(struct loop_device *lo, struct request *rq,
290 loff_t pos)
291{
292 struct bio_vec bvec;
293 struct req_iterator iter;
294 int ret = 0;
295
296 rq_for_each_segment(bvec, rq, iter) {
297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
298 if (ret < 0)
299 break;
300 cond_resched();
301 }
302
303 return ret;
304}
305
306
307
308
309
310
311static int lo_write_transfer(struct loop_device *lo, struct request *rq,
312 loff_t pos)
313{
314 struct bio_vec bvec, b;
315 struct req_iterator iter;
316 struct page *page;
317 int ret = 0;
318
319 page = alloc_page(GFP_NOIO);
320 if (unlikely(!page))
321 return -ENOMEM;
322
323 rq_for_each_segment(bvec, rq, iter) {
324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
325 bvec.bv_offset, bvec.bv_len, pos >> 9);
326 if (unlikely(ret))
327 break;
328
329 b.bv_page = page;
330 b.bv_offset = 0;
331 b.bv_len = bvec.bv_len;
332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
333 if (ret < 0)
334 break;
335 }
336
337 __free_page(page);
338 return ret;
339}
340
341static int lo_read_simple(struct loop_device *lo, struct request *rq,
342 loff_t pos)
343{
344 struct bio_vec bvec;
345 struct req_iterator iter;
346 struct iov_iter i;
347 ssize_t len;
348
349 rq_for_each_segment(bvec, rq, iter) {
350 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
352 if (len < 0)
353 return len;
354
355 flush_dcache_page(bvec.bv_page);
356
357 if (len != bvec.bv_len) {
358 struct bio *bio;
359
360 __rq_for_each_bio(bio, rq)
361 zero_fill_bio(bio);
362 break;
363 }
364 cond_resched();
365 }
366
367 return 0;
368}
369
370static int lo_read_transfer(struct loop_device *lo, struct request *rq,
371 loff_t pos)
372{
373 struct bio_vec bvec, b;
374 struct req_iterator iter;
375 struct iov_iter i;
376 struct page *page;
377 ssize_t len;
378 int ret = 0;
379
380 page = alloc_page(GFP_NOIO);
381 if (unlikely(!page))
382 return -ENOMEM;
383
384 rq_for_each_segment(bvec, rq, iter) {
385 loff_t offset = pos;
386
387 b.bv_page = page;
388 b.bv_offset = 0;
389 b.bv_len = bvec.bv_len;
390
391 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
393 if (len < 0) {
394 ret = len;
395 goto out_free_page;
396 }
397
398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
399 bvec.bv_offset, len, offset >> 9);
400 if (ret)
401 goto out_free_page;
402
403 flush_dcache_page(bvec.bv_page);
404
405 if (len != bvec.bv_len) {
406 struct bio *bio;
407
408 __rq_for_each_bio(bio, rq)
409 zero_fill_bio(bio);
410 break;
411 }
412 }
413
414 ret = 0;
415out_free_page:
416 __free_page(page);
417 return ret;
418}
419
420static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
421{
422
423
424
425
426
427
428 struct file *file = lo->lo_backing_file;
429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
430 int ret;
431
432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
433 ret = -EOPNOTSUPP;
434 goto out;
435 }
436
437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
439 ret = -EIO;
440 out:
441 return ret;
442}
443
444static int lo_req_flush(struct loop_device *lo, struct request *rq)
445{
446 struct file *file = lo->lo_backing_file;
447 int ret = vfs_fsync(file, 0);
448 if (unlikely(ret && ret != -EINVAL))
449 ret = -EIO;
450
451 return ret;
452}
453
454static void lo_complete_rq(struct request *rq)
455{
456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
457 blk_status_t ret = BLK_STS_OK;
458
459 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
460 req_op(rq) != REQ_OP_READ) {
461 if (cmd->ret < 0)
462 ret = BLK_STS_IOERR;
463 goto end_io;
464 }
465
466
467
468
469
470 if (cmd->ret) {
471 blk_update_request(rq, BLK_STS_OK, cmd->ret);
472 cmd->ret = 0;
473 blk_mq_requeue_request(rq, true);
474 } else {
475 if (cmd->use_aio) {
476 struct bio *bio = rq->bio;
477
478 while (bio) {
479 zero_fill_bio(bio);
480 bio = bio->bi_next;
481 }
482 }
483 ret = BLK_STS_IOERR;
484end_io:
485 blk_mq_end_request(rq, ret);
486 }
487}
488
489static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
490{
491 struct request *rq = blk_mq_rq_from_pdu(cmd);
492
493 if (!atomic_dec_and_test(&cmd->ref))
494 return;
495 kfree(cmd->bvec);
496 cmd->bvec = NULL;
497 blk_mq_complete_request(rq);
498}
499
500static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
501{
502 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
503
504 if (cmd->css)
505 css_put(cmd->css);
506 cmd->ret = ret;
507 lo_rw_aio_do_completion(cmd);
508}
509
510static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
511 loff_t pos, bool rw)
512{
513 struct iov_iter iter;
514 struct req_iterator rq_iter;
515 struct bio_vec *bvec;
516 struct request *rq = blk_mq_rq_from_pdu(cmd);
517 struct bio *bio = rq->bio;
518 struct file *file = lo->lo_backing_file;
519 struct bio_vec tmp;
520 unsigned int offset;
521 int nr_bvec = 0;
522 int ret;
523
524 rq_for_each_bvec(tmp, rq, rq_iter)
525 nr_bvec++;
526
527 if (rq->bio != rq->biotail) {
528
529 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
530 GFP_NOIO);
531 if (!bvec)
532 return -EIO;
533 cmd->bvec = bvec;
534
535
536
537
538
539
540
541 rq_for_each_bvec(tmp, rq, rq_iter) {
542 *bvec = tmp;
543 bvec++;
544 }
545 bvec = cmd->bvec;
546 offset = 0;
547 } else {
548
549
550
551
552
553 offset = bio->bi_iter.bi_bvec_done;
554 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
555 }
556 atomic_set(&cmd->ref, 2);
557
558 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
559 iter.iov_offset = offset;
560
561 cmd->iocb.ki_pos = pos;
562 cmd->iocb.ki_filp = file;
563 cmd->iocb.ki_complete = lo_rw_aio_complete;
564 cmd->iocb.ki_flags = IOCB_DIRECT;
565 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
566 if (cmd->css)
567 kthread_associate_blkcg(cmd->css);
568
569 if (rw == WRITE)
570 ret = call_write_iter(file, &cmd->iocb, &iter);
571 else
572 ret = call_read_iter(file, &cmd->iocb, &iter);
573
574 lo_rw_aio_do_completion(cmd);
575 kthread_associate_blkcg(NULL);
576
577 if (ret != -EIOCBQUEUED)
578 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
579 return 0;
580}
581
582static int do_req_filebacked(struct loop_device *lo, struct request *rq)
583{
584 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
585 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
586
587
588
589
590
591
592
593
594
595
596 switch (req_op(rq)) {
597 case REQ_OP_FLUSH:
598 return lo_req_flush(lo, rq);
599 case REQ_OP_DISCARD:
600 case REQ_OP_WRITE_ZEROES:
601 return lo_discard(lo, rq, pos);
602 case REQ_OP_WRITE:
603 if (lo->transfer)
604 return lo_write_transfer(lo, rq, pos);
605 else if (cmd->use_aio)
606 return lo_rw_aio(lo, cmd, pos, WRITE);
607 else
608 return lo_write_simple(lo, rq, pos);
609 case REQ_OP_READ:
610 if (lo->transfer)
611 return lo_read_transfer(lo, rq, pos);
612 else if (cmd->use_aio)
613 return lo_rw_aio(lo, cmd, pos, READ);
614 else
615 return lo_read_simple(lo, rq, pos);
616 default:
617 WARN_ON_ONCE(1);
618 return -EIO;
619 }
620}
621
622static inline void loop_update_dio(struct loop_device *lo)
623{
624 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
625 lo->use_dio);
626}
627
628static void loop_reread_partitions(struct loop_device *lo,
629 struct block_device *bdev)
630{
631 int rc;
632
633 rc = blkdev_reread_part(bdev);
634 if (rc)
635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
636 __func__, lo->lo_number, lo->lo_file_name, rc);
637}
638
639static inline int is_loop_device(struct file *file)
640{
641 struct inode *i = file->f_mapping->host;
642
643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
644}
645
646static int loop_validate_file(struct file *file, struct block_device *bdev)
647{
648 struct inode *inode = file->f_mapping->host;
649 struct file *f = file;
650
651
652 while (is_loop_device(f)) {
653 struct loop_device *l;
654
655 if (f->f_mapping->host->i_bdev == bdev)
656 return -EBADF;
657
658 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
659 if (l->lo_state != Lo_bound) {
660 return -EINVAL;
661 }
662 f = l->lo_backing_file;
663 }
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
665 return -EINVAL;
666 return 0;
667}
668
669
670
671
672
673
674
675
676
677static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
678 unsigned int arg)
679{
680 struct file *file = NULL, *old_file;
681 int error;
682 bool partscan;
683
684 error = mutex_lock_killable(&loop_ctl_mutex);
685 if (error)
686 return error;
687 error = -ENXIO;
688 if (lo->lo_state != Lo_bound)
689 goto out_err;
690
691
692 error = -EINVAL;
693 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
694 goto out_err;
695
696 error = -EBADF;
697 file = fget(arg);
698 if (!file)
699 goto out_err;
700
701 error = loop_validate_file(file, bdev);
702 if (error)
703 goto out_err;
704
705 old_file = lo->lo_backing_file;
706
707 error = -EINVAL;
708
709
710 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
711 goto out_err;
712
713
714 blk_mq_freeze_queue(lo->lo_queue);
715 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
716 lo->lo_backing_file = file;
717 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
718 mapping_set_gfp_mask(file->f_mapping,
719 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
720 loop_update_dio(lo);
721 blk_mq_unfreeze_queue(lo->lo_queue);
722 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
723 mutex_unlock(&loop_ctl_mutex);
724
725
726
727
728
729 fput(old_file);
730 if (partscan)
731 loop_reread_partitions(lo, bdev);
732 return 0;
733
734out_err:
735 mutex_unlock(&loop_ctl_mutex);
736 if (file)
737 fput(file);
738 return error;
739}
740
741
742
743static ssize_t loop_attr_show(struct device *dev, char *page,
744 ssize_t (*callback)(struct loop_device *, char *))
745{
746 struct gendisk *disk = dev_to_disk(dev);
747 struct loop_device *lo = disk->private_data;
748
749 return callback(lo, page);
750}
751
752#define LOOP_ATTR_RO(_name) \
753static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
754static ssize_t loop_attr_do_show_##_name(struct device *d, \
755 struct device_attribute *attr, char *b) \
756{ \
757 return loop_attr_show(d, b, loop_attr_##_name##_show); \
758} \
759static struct device_attribute loop_attr_##_name = \
760 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
761
762static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
763{
764 ssize_t ret;
765 char *p = NULL;
766
767 spin_lock_irq(&lo->lo_lock);
768 if (lo->lo_backing_file)
769 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
770 spin_unlock_irq(&lo->lo_lock);
771
772 if (IS_ERR_OR_NULL(p))
773 ret = PTR_ERR(p);
774 else {
775 ret = strlen(p);
776 memmove(buf, p, ret);
777 buf[ret++] = '\n';
778 buf[ret] = 0;
779 }
780
781 return ret;
782}
783
784static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
785{
786 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
787}
788
789static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
790{
791 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
792}
793
794static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
795{
796 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
797
798 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
799}
800
801static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
802{
803 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
804
805 return sprintf(buf, "%s\n", partscan ? "1" : "0");
806}
807
808static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
809{
810 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
811
812 return sprintf(buf, "%s\n", dio ? "1" : "0");
813}
814
815LOOP_ATTR_RO(backing_file);
816LOOP_ATTR_RO(offset);
817LOOP_ATTR_RO(sizelimit);
818LOOP_ATTR_RO(autoclear);
819LOOP_ATTR_RO(partscan);
820LOOP_ATTR_RO(dio);
821
822static struct attribute *loop_attrs[] = {
823 &loop_attr_backing_file.attr,
824 &loop_attr_offset.attr,
825 &loop_attr_sizelimit.attr,
826 &loop_attr_autoclear.attr,
827 &loop_attr_partscan.attr,
828 &loop_attr_dio.attr,
829 NULL,
830};
831
832static struct attribute_group loop_attribute_group = {
833 .name = "loop",
834 .attrs= loop_attrs,
835};
836
837static void loop_sysfs_init(struct loop_device *lo)
838{
839 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
840 &loop_attribute_group);
841}
842
843static void loop_sysfs_exit(struct loop_device *lo)
844{
845 if (lo->sysfs_inited)
846 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
847 &loop_attribute_group);
848}
849
850static void loop_config_discard(struct loop_device *lo)
851{
852 struct file *file = lo->lo_backing_file;
853 struct inode *inode = file->f_mapping->host;
854 struct request_queue *q = lo->lo_queue;
855
856
857
858
859
860
861
862 if ((!file->f_op->fallocate) ||
863 lo->lo_encrypt_key_size) {
864 q->limits.discard_granularity = 0;
865 q->limits.discard_alignment = 0;
866 blk_queue_max_discard_sectors(q, 0);
867 blk_queue_max_write_zeroes_sectors(q, 0);
868 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
869 return;
870 }
871
872 q->limits.discard_granularity = inode->i_sb->s_blocksize;
873 q->limits.discard_alignment = 0;
874
875 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
876 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
877 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
878}
879
880static void loop_unprepare_queue(struct loop_device *lo)
881{
882 kthread_flush_worker(&lo->worker);
883 kthread_stop(lo->worker_task);
884}
885
886static int loop_kthread_worker_fn(void *worker_ptr)
887{
888 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
889 return kthread_worker_fn(worker_ptr);
890}
891
892static int loop_prepare_queue(struct loop_device *lo)
893{
894 kthread_init_worker(&lo->worker);
895 lo->worker_task = kthread_run(loop_kthread_worker_fn,
896 &lo->worker, "loop%d", lo->lo_number);
897 if (IS_ERR(lo->worker_task))
898 return -ENOMEM;
899 set_user_nice(lo->worker_task, MIN_NICE);
900 return 0;
901}
902
903static void loop_update_rotational(struct loop_device *lo)
904{
905 struct file *file = lo->lo_backing_file;
906 struct inode *file_inode = file->f_mapping->host;
907 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
908 struct request_queue *q = lo->lo_queue;
909 bool nonrot = true;
910
911
912 if (file_bdev)
913 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
914
915 if (nonrot)
916 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
917 else
918 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
919}
920
921static int loop_set_fd(struct loop_device *lo, fmode_t mode,
922 struct block_device *bdev, unsigned int arg)
923{
924 struct file *file;
925 struct inode *inode;
926 struct address_space *mapping;
927 struct block_device *claimed_bdev = NULL;
928 int lo_flags = 0;
929 int error;
930 loff_t size;
931 bool partscan;
932
933
934 __module_get(THIS_MODULE);
935
936 error = -EBADF;
937 file = fget(arg);
938 if (!file)
939 goto out;
940
941
942
943
944
945 if (!(mode & FMODE_EXCL)) {
946 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
947 if (IS_ERR(claimed_bdev)) {
948 error = PTR_ERR(claimed_bdev);
949 goto out_putf;
950 }
951 }
952
953 error = mutex_lock_killable(&loop_ctl_mutex);
954 if (error)
955 goto out_bdev;
956
957 error = -EBUSY;
958 if (lo->lo_state != Lo_unbound)
959 goto out_unlock;
960
961 error = loop_validate_file(file, bdev);
962 if (error)
963 goto out_unlock;
964
965 mapping = file->f_mapping;
966 inode = mapping->host;
967
968 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
969 !file->f_op->write_iter)
970 lo_flags |= LO_FLAGS_READ_ONLY;
971
972 error = -EFBIG;
973 size = get_loop_size(lo, file);
974 if ((loff_t)(sector_t)size != size)
975 goto out_unlock;
976 error = loop_prepare_queue(lo);
977 if (error)
978 goto out_unlock;
979
980 error = 0;
981
982 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
983
984 lo->use_dio = false;
985 lo->lo_device = bdev;
986 lo->lo_flags = lo_flags;
987 lo->lo_backing_file = file;
988 lo->transfer = NULL;
989 lo->ioctl = NULL;
990 lo->lo_sizelimit = 0;
991 lo->old_gfp_mask = mapping_gfp_mask(mapping);
992 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
993
994 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
995 blk_queue_write_cache(lo->lo_queue, true, false);
996
997 loop_update_rotational(lo);
998 loop_update_dio(lo);
999 set_capacity(lo->lo_disk, size);
1000 bd_set_size(bdev, size << 9);
1001 loop_sysfs_init(lo);
1002
1003 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1004
1005 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1006 block_size(inode->i_bdev) : PAGE_SIZE);
1007
1008 lo->lo_state = Lo_bound;
1009 if (part_shift)
1010 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1011 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1012
1013
1014
1015
1016 bdgrab(bdev);
1017 mutex_unlock(&loop_ctl_mutex);
1018 if (partscan)
1019 loop_reread_partitions(lo, bdev);
1020 if (claimed_bdev)
1021 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1022 return 0;
1023
1024out_unlock:
1025 mutex_unlock(&loop_ctl_mutex);
1026out_bdev:
1027 if (claimed_bdev)
1028 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1029out_putf:
1030 fput(file);
1031out:
1032
1033 module_put(THIS_MODULE);
1034 return error;
1035}
1036
1037static int
1038loop_release_xfer(struct loop_device *lo)
1039{
1040 int err = 0;
1041 struct loop_func_table *xfer = lo->lo_encryption;
1042
1043 if (xfer) {
1044 if (xfer->release)
1045 err = xfer->release(lo);
1046 lo->transfer = NULL;
1047 lo->lo_encryption = NULL;
1048 module_put(xfer->owner);
1049 }
1050 return err;
1051}
1052
1053static int
1054loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1055 const struct loop_info64 *i)
1056{
1057 int err = 0;
1058
1059 if (xfer) {
1060 struct module *owner = xfer->owner;
1061
1062 if (!try_module_get(owner))
1063 return -EINVAL;
1064 if (xfer->init)
1065 err = xfer->init(lo, i);
1066 if (err)
1067 module_put(owner);
1068 else
1069 lo->lo_encryption = xfer;
1070 }
1071 return err;
1072}
1073
1074static int __loop_clr_fd(struct loop_device *lo, bool release)
1075{
1076 struct file *filp = NULL;
1077 gfp_t gfp = lo->old_gfp_mask;
1078 struct block_device *bdev = lo->lo_device;
1079 int err = 0;
1080 bool partscan = false;
1081 int lo_number;
1082
1083 mutex_lock(&loop_ctl_mutex);
1084 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1085 err = -ENXIO;
1086 goto out_unlock;
1087 }
1088
1089 filp = lo->lo_backing_file;
1090 if (filp == NULL) {
1091 err = -EINVAL;
1092 goto out_unlock;
1093 }
1094
1095
1096 blk_mq_freeze_queue(lo->lo_queue);
1097
1098 spin_lock_irq(&lo->lo_lock);
1099 lo->lo_backing_file = NULL;
1100 spin_unlock_irq(&lo->lo_lock);
1101
1102 loop_release_xfer(lo);
1103 lo->transfer = NULL;
1104 lo->ioctl = NULL;
1105 lo->lo_device = NULL;
1106 lo->lo_encryption = NULL;
1107 lo->lo_offset = 0;
1108 lo->lo_sizelimit = 0;
1109 lo->lo_encrypt_key_size = 0;
1110 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1111 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1112 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1113 blk_queue_logical_block_size(lo->lo_queue, 512);
1114 blk_queue_physical_block_size(lo->lo_queue, 512);
1115 blk_queue_io_min(lo->lo_queue, 512);
1116 if (bdev) {
1117 bdput(bdev);
1118 invalidate_bdev(bdev);
1119 bdev->bd_inode->i_mapping->wb_err = 0;
1120 }
1121 set_capacity(lo->lo_disk, 0);
1122 loop_sysfs_exit(lo);
1123 if (bdev) {
1124 bd_set_size(bdev, 0);
1125
1126 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1127 }
1128 mapping_set_gfp_mask(filp->f_mapping, gfp);
1129
1130 module_put(THIS_MODULE);
1131 blk_mq_unfreeze_queue(lo->lo_queue);
1132
1133 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1134 lo_number = lo->lo_number;
1135 loop_unprepare_queue(lo);
1136out_unlock:
1137 mutex_unlock(&loop_ctl_mutex);
1138 if (partscan) {
1139
1140
1141
1142
1143
1144
1145
1146
1147 if (release)
1148 err = __blkdev_reread_part(bdev);
1149 else
1150 err = blkdev_reread_part(bdev);
1151 if (err)
1152 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1153 __func__, lo_number, err);
1154
1155 err = 0;
1156 }
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167 mutex_lock(&loop_ctl_mutex);
1168 lo->lo_flags = 0;
1169 if (!part_shift)
1170 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1171 lo->lo_state = Lo_unbound;
1172 mutex_unlock(&loop_ctl_mutex);
1173
1174
1175
1176
1177
1178
1179
1180 if (filp)
1181 fput(filp);
1182 return err;
1183}
1184
1185static int loop_clr_fd(struct loop_device *lo)
1186{
1187 int err;
1188
1189 err = mutex_lock_killable(&loop_ctl_mutex);
1190 if (err)
1191 return err;
1192 if (lo->lo_state != Lo_bound) {
1193 mutex_unlock(&loop_ctl_mutex);
1194 return -ENXIO;
1195 }
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206 if (atomic_read(&lo->lo_refcnt) > 1) {
1207 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1208 mutex_unlock(&loop_ctl_mutex);
1209 return 0;
1210 }
1211 lo->lo_state = Lo_rundown;
1212 mutex_unlock(&loop_ctl_mutex);
1213
1214 return __loop_clr_fd(lo, false);
1215}
1216
1217static int
1218loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1219{
1220 int err;
1221 struct loop_func_table *xfer;
1222 kuid_t uid = current_uid();
1223 struct block_device *bdev;
1224 bool partscan = false;
1225
1226 err = mutex_lock_killable(&loop_ctl_mutex);
1227 if (err)
1228 return err;
1229 if (lo->lo_encrypt_key_size &&
1230 !uid_eq(lo->lo_key_owner, uid) &&
1231 !capable(CAP_SYS_ADMIN)) {
1232 err = -EPERM;
1233 goto out_unlock;
1234 }
1235 if (lo->lo_state != Lo_bound) {
1236 err = -ENXIO;
1237 goto out_unlock;
1238 }
1239 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1240 err = -EINVAL;
1241 goto out_unlock;
1242 }
1243
1244 if (lo->lo_offset != info->lo_offset ||
1245 lo->lo_sizelimit != info->lo_sizelimit) {
1246 sync_blockdev(lo->lo_device);
1247 kill_bdev(lo->lo_device);
1248 }
1249
1250
1251 blk_mq_freeze_queue(lo->lo_queue);
1252
1253 err = loop_release_xfer(lo);
1254 if (err)
1255 goto out_unfreeze;
1256
1257 if (info->lo_encrypt_type) {
1258 unsigned int type = info->lo_encrypt_type;
1259
1260 if (type >= MAX_LO_CRYPT) {
1261 err = -EINVAL;
1262 goto out_unfreeze;
1263 }
1264 xfer = xfer_funcs[type];
1265 if (xfer == NULL) {
1266 err = -EINVAL;
1267 goto out_unfreeze;
1268 }
1269 } else
1270 xfer = NULL;
1271
1272 err = loop_init_xfer(lo, xfer, info);
1273 if (err)
1274 goto out_unfreeze;
1275
1276 if (lo->lo_offset != info->lo_offset ||
1277 lo->lo_sizelimit != info->lo_sizelimit) {
1278
1279 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1280 err = -EAGAIN;
1281 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1282 __func__, lo->lo_number, lo->lo_file_name,
1283 lo->lo_device->bd_inode->i_mapping->nrpages);
1284 goto out_unfreeze;
1285 }
1286 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1287 err = -EFBIG;
1288 goto out_unfreeze;
1289 }
1290 }
1291
1292 loop_config_discard(lo);
1293
1294 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1295 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1296 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1297 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1298
1299 if (!xfer)
1300 xfer = &none_funcs;
1301 lo->transfer = xfer->transfer;
1302 lo->ioctl = xfer->ioctl;
1303
1304 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1305 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1306 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1307
1308 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1309 lo->lo_init[0] = info->lo_init[0];
1310 lo->lo_init[1] = info->lo_init[1];
1311 if (info->lo_encrypt_key_size) {
1312 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1313 info->lo_encrypt_key_size);
1314 lo->lo_key_owner = uid;
1315 }
1316
1317
1318 __loop_update_dio(lo, lo->use_dio);
1319
1320out_unfreeze:
1321 blk_mq_unfreeze_queue(lo->lo_queue);
1322
1323 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1324 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1325 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1326 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1327 bdev = lo->lo_device;
1328 partscan = true;
1329 }
1330out_unlock:
1331 mutex_unlock(&loop_ctl_mutex);
1332 if (partscan)
1333 loop_reread_partitions(lo, bdev);
1334
1335 return err;
1336}
1337
1338static int
1339loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1340{
1341 struct path path;
1342 struct kstat stat;
1343 int ret;
1344
1345 ret = mutex_lock_killable(&loop_ctl_mutex);
1346 if (ret)
1347 return ret;
1348 if (lo->lo_state != Lo_bound) {
1349 mutex_unlock(&loop_ctl_mutex);
1350 return -ENXIO;
1351 }
1352
1353 memset(info, 0, sizeof(*info));
1354 info->lo_number = lo->lo_number;
1355 info->lo_offset = lo->lo_offset;
1356 info->lo_sizelimit = lo->lo_sizelimit;
1357 info->lo_flags = lo->lo_flags;
1358 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1359 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1360 info->lo_encrypt_type =
1361 lo->lo_encryption ? lo->lo_encryption->number : 0;
1362 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1363 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1364 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1365 lo->lo_encrypt_key_size);
1366 }
1367
1368
1369 path = lo->lo_backing_file->f_path;
1370 path_get(&path);
1371 mutex_unlock(&loop_ctl_mutex);
1372 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1373 if (!ret) {
1374 info->lo_device = huge_encode_dev(stat.dev);
1375 info->lo_inode = stat.ino;
1376 info->lo_rdevice = huge_encode_dev(stat.rdev);
1377 }
1378 path_put(&path);
1379 return ret;
1380}
1381
1382static void
1383loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1384{
1385 memset(info64, 0, sizeof(*info64));
1386 info64->lo_number = info->lo_number;
1387 info64->lo_device = info->lo_device;
1388 info64->lo_inode = info->lo_inode;
1389 info64->lo_rdevice = info->lo_rdevice;
1390 info64->lo_offset = info->lo_offset;
1391 info64->lo_sizelimit = 0;
1392 info64->lo_encrypt_type = info->lo_encrypt_type;
1393 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1394 info64->lo_flags = info->lo_flags;
1395 info64->lo_init[0] = info->lo_init[0];
1396 info64->lo_init[1] = info->lo_init[1];
1397 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1398 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1399 else
1400 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1401 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1402}
1403
1404static int
1405loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1406{
1407 memset(info, 0, sizeof(*info));
1408 info->lo_number = info64->lo_number;
1409 info->lo_device = info64->lo_device;
1410 info->lo_inode = info64->lo_inode;
1411 info->lo_rdevice = info64->lo_rdevice;
1412 info->lo_offset = info64->lo_offset;
1413 info->lo_encrypt_type = info64->lo_encrypt_type;
1414 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1415 info->lo_flags = info64->lo_flags;
1416 info->lo_init[0] = info64->lo_init[0];
1417 info->lo_init[1] = info64->lo_init[1];
1418 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1419 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1420 else
1421 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1422 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1423
1424
1425 if (info->lo_device != info64->lo_device ||
1426 info->lo_rdevice != info64->lo_rdevice ||
1427 info->lo_inode != info64->lo_inode ||
1428 info->lo_offset != info64->lo_offset)
1429 return -EOVERFLOW;
1430
1431 return 0;
1432}
1433
1434static int
1435loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1436{
1437 struct loop_info info;
1438 struct loop_info64 info64;
1439
1440 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1441 return -EFAULT;
1442 loop_info64_from_old(&info, &info64);
1443 return loop_set_status(lo, &info64);
1444}
1445
1446static int
1447loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1448{
1449 struct loop_info64 info64;
1450
1451 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1452 return -EFAULT;
1453 return loop_set_status(lo, &info64);
1454}
1455
1456static int
1457loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1458 struct loop_info info;
1459 struct loop_info64 info64;
1460 int err;
1461
1462 if (!arg)
1463 return -EINVAL;
1464 err = loop_get_status(lo, &info64);
1465 if (!err)
1466 err = loop_info64_to_old(&info64, &info);
1467 if (!err && copy_to_user(arg, &info, sizeof(info)))
1468 err = -EFAULT;
1469
1470 return err;
1471}
1472
1473static int
1474loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1475 struct loop_info64 info64;
1476 int err;
1477
1478 if (!arg)
1479 return -EINVAL;
1480 err = loop_get_status(lo, &info64);
1481 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1482 err = -EFAULT;
1483
1484 return err;
1485}
1486
1487static int loop_set_capacity(struct loop_device *lo)
1488{
1489 if (unlikely(lo->lo_state != Lo_bound))
1490 return -ENXIO;
1491
1492 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1493}
1494
1495static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1496{
1497 int error = -ENXIO;
1498 if (lo->lo_state != Lo_bound)
1499 goto out;
1500
1501 __loop_update_dio(lo, !!arg);
1502 if (lo->use_dio == !!arg)
1503 return 0;
1504 error = -EINVAL;
1505 out:
1506 return error;
1507}
1508
1509static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1510{
1511 int err = 0;
1512
1513 if (lo->lo_state != Lo_bound)
1514 return -ENXIO;
1515
1516 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1517 return -EINVAL;
1518
1519 if (lo->lo_queue->limits.logical_block_size != arg) {
1520 sync_blockdev(lo->lo_device);
1521 kill_bdev(lo->lo_device);
1522 }
1523
1524 blk_mq_freeze_queue(lo->lo_queue);
1525
1526
1527 if (lo->lo_queue->limits.logical_block_size != arg &&
1528 lo->lo_device->bd_inode->i_mapping->nrpages) {
1529 err = -EAGAIN;
1530 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1531 __func__, lo->lo_number, lo->lo_file_name,
1532 lo->lo_device->bd_inode->i_mapping->nrpages);
1533 goto out_unfreeze;
1534 }
1535
1536 blk_queue_logical_block_size(lo->lo_queue, arg);
1537 blk_queue_physical_block_size(lo->lo_queue, arg);
1538 blk_queue_io_min(lo->lo_queue, arg);
1539 loop_update_dio(lo);
1540out_unfreeze:
1541 blk_mq_unfreeze_queue(lo->lo_queue);
1542
1543 return err;
1544}
1545
1546static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1547 unsigned long arg)
1548{
1549 int err;
1550
1551 err = mutex_lock_killable(&loop_ctl_mutex);
1552 if (err)
1553 return err;
1554 switch (cmd) {
1555 case LOOP_SET_CAPACITY:
1556 err = loop_set_capacity(lo);
1557 break;
1558 case LOOP_SET_DIRECT_IO:
1559 err = loop_set_dio(lo, arg);
1560 break;
1561 case LOOP_SET_BLOCK_SIZE:
1562 err = loop_set_block_size(lo, arg);
1563 break;
1564 default:
1565 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1566 }
1567 mutex_unlock(&loop_ctl_mutex);
1568 return err;
1569}
1570
1571static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1572 unsigned int cmd, unsigned long arg)
1573{
1574 struct loop_device *lo = bdev->bd_disk->private_data;
1575 int err;
1576
1577 switch (cmd) {
1578 case LOOP_SET_FD:
1579 return loop_set_fd(lo, mode, bdev, arg);
1580 case LOOP_CHANGE_FD:
1581 return loop_change_fd(lo, bdev, arg);
1582 case LOOP_CLR_FD:
1583 return loop_clr_fd(lo);
1584 case LOOP_SET_STATUS:
1585 err = -EPERM;
1586 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1587 err = loop_set_status_old(lo,
1588 (struct loop_info __user *)arg);
1589 }
1590 break;
1591 case LOOP_GET_STATUS:
1592 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1593 case LOOP_SET_STATUS64:
1594 err = -EPERM;
1595 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1596 err = loop_set_status64(lo,
1597 (struct loop_info64 __user *) arg);
1598 }
1599 break;
1600 case LOOP_GET_STATUS64:
1601 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1602 case LOOP_SET_CAPACITY:
1603 case LOOP_SET_DIRECT_IO:
1604 case LOOP_SET_BLOCK_SIZE:
1605 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1606 return -EPERM;
1607
1608 default:
1609 err = lo_simple_ioctl(lo, cmd, arg);
1610 break;
1611 }
1612
1613 return err;
1614}
1615
1616#ifdef CONFIG_COMPAT
1617struct compat_loop_info {
1618 compat_int_t lo_number;
1619 compat_dev_t lo_device;
1620 compat_ulong_t lo_inode;
1621 compat_dev_t lo_rdevice;
1622 compat_int_t lo_offset;
1623 compat_int_t lo_encrypt_type;
1624 compat_int_t lo_encrypt_key_size;
1625 compat_int_t lo_flags;
1626 char lo_name[LO_NAME_SIZE];
1627 unsigned char lo_encrypt_key[LO_KEY_SIZE];
1628 compat_ulong_t lo_init[2];
1629 char reserved[4];
1630};
1631
1632
1633
1634
1635
1636static noinline int
1637loop_info64_from_compat(const struct compat_loop_info __user *arg,
1638 struct loop_info64 *info64)
1639{
1640 struct compat_loop_info info;
1641
1642 if (copy_from_user(&info, arg, sizeof(info)))
1643 return -EFAULT;
1644
1645 memset(info64, 0, sizeof(*info64));
1646 info64->lo_number = info.lo_number;
1647 info64->lo_device = info.lo_device;
1648 info64->lo_inode = info.lo_inode;
1649 info64->lo_rdevice = info.lo_rdevice;
1650 info64->lo_offset = info.lo_offset;
1651 info64->lo_sizelimit = 0;
1652 info64->lo_encrypt_type = info.lo_encrypt_type;
1653 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1654 info64->lo_flags = info.lo_flags;
1655 info64->lo_init[0] = info.lo_init[0];
1656 info64->lo_init[1] = info.lo_init[1];
1657 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1658 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1659 else
1660 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1661 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1662 return 0;
1663}
1664
1665
1666
1667
1668
1669static noinline int
1670loop_info64_to_compat(const struct loop_info64 *info64,
1671 struct compat_loop_info __user *arg)
1672{
1673 struct compat_loop_info info;
1674
1675 memset(&info, 0, sizeof(info));
1676 info.lo_number = info64->lo_number;
1677 info.lo_device = info64->lo_device;
1678 info.lo_inode = info64->lo_inode;
1679 info.lo_rdevice = info64->lo_rdevice;
1680 info.lo_offset = info64->lo_offset;
1681 info.lo_encrypt_type = info64->lo_encrypt_type;
1682 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1683 info.lo_flags = info64->lo_flags;
1684 info.lo_init[0] = info64->lo_init[0];
1685 info.lo_init[1] = info64->lo_init[1];
1686 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1687 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1688 else
1689 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1690 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1691
1692
1693 if (info.lo_device != info64->lo_device ||
1694 info.lo_rdevice != info64->lo_rdevice ||
1695 info.lo_inode != info64->lo_inode ||
1696 info.lo_offset != info64->lo_offset ||
1697 info.lo_init[0] != info64->lo_init[0] ||
1698 info.lo_init[1] != info64->lo_init[1])
1699 return -EOVERFLOW;
1700
1701 if (copy_to_user(arg, &info, sizeof(info)))
1702 return -EFAULT;
1703 return 0;
1704}
1705
1706static int
1707loop_set_status_compat(struct loop_device *lo,
1708 const struct compat_loop_info __user *arg)
1709{
1710 struct loop_info64 info64;
1711 int ret;
1712
1713 ret = loop_info64_from_compat(arg, &info64);
1714 if (ret < 0)
1715 return ret;
1716 return loop_set_status(lo, &info64);
1717}
1718
1719static int
1720loop_get_status_compat(struct loop_device *lo,
1721 struct compat_loop_info __user *arg)
1722{
1723 struct loop_info64 info64;
1724 int err;
1725
1726 if (!arg)
1727 return -EINVAL;
1728 err = loop_get_status(lo, &info64);
1729 if (!err)
1730 err = loop_info64_to_compat(&info64, arg);
1731 return err;
1732}
1733
1734static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1735 unsigned int cmd, unsigned long arg)
1736{
1737 struct loop_device *lo = bdev->bd_disk->private_data;
1738 int err;
1739
1740 switch(cmd) {
1741 case LOOP_SET_STATUS:
1742 err = loop_set_status_compat(lo,
1743 (const struct compat_loop_info __user *)arg);
1744 break;
1745 case LOOP_GET_STATUS:
1746 err = loop_get_status_compat(lo,
1747 (struct compat_loop_info __user *)arg);
1748 break;
1749 case LOOP_SET_CAPACITY:
1750 case LOOP_CLR_FD:
1751 case LOOP_GET_STATUS64:
1752 case LOOP_SET_STATUS64:
1753 arg = (unsigned long) compat_ptr(arg);
1754
1755 case LOOP_SET_FD:
1756 case LOOP_CHANGE_FD:
1757 case LOOP_SET_BLOCK_SIZE:
1758 err = lo_ioctl(bdev, mode, cmd, arg);
1759 break;
1760 default:
1761 err = -ENOIOCTLCMD;
1762 break;
1763 }
1764 return err;
1765}
1766#endif
1767
1768static int lo_open(struct block_device *bdev, fmode_t mode)
1769{
1770 struct loop_device *lo;
1771 int err;
1772
1773 err = mutex_lock_killable(&loop_ctl_mutex);
1774 if (err)
1775 return err;
1776 lo = bdev->bd_disk->private_data;
1777 if (!lo) {
1778 err = -ENXIO;
1779 goto out;
1780 }
1781
1782 atomic_inc(&lo->lo_refcnt);
1783out:
1784 mutex_unlock(&loop_ctl_mutex);
1785 return err;
1786}
1787
1788static void lo_release(struct gendisk *disk, fmode_t mode)
1789{
1790 struct loop_device *lo;
1791
1792 mutex_lock(&loop_ctl_mutex);
1793 lo = disk->private_data;
1794 if (atomic_dec_return(&lo->lo_refcnt))
1795 goto out_unlock;
1796
1797 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1798 if (lo->lo_state != Lo_bound)
1799 goto out_unlock;
1800 lo->lo_state = Lo_rundown;
1801 mutex_unlock(&loop_ctl_mutex);
1802
1803
1804
1805
1806 __loop_clr_fd(lo, true);
1807 return;
1808 } else if (lo->lo_state == Lo_bound) {
1809
1810
1811
1812
1813 blk_mq_freeze_queue(lo->lo_queue);
1814 blk_mq_unfreeze_queue(lo->lo_queue);
1815 }
1816
1817out_unlock:
1818 mutex_unlock(&loop_ctl_mutex);
1819}
1820
1821static const struct block_device_operations lo_fops = {
1822 .owner = THIS_MODULE,
1823 .open = lo_open,
1824 .release = lo_release,
1825 .ioctl = lo_ioctl,
1826#ifdef CONFIG_COMPAT
1827 .compat_ioctl = lo_compat_ioctl,
1828#endif
1829};
1830
1831
1832
1833
1834static int max_loop;
1835module_param(max_loop, int, 0444);
1836MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1837module_param(max_part, int, 0444);
1838MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1839MODULE_LICENSE("GPL");
1840MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1841
1842int loop_register_transfer(struct loop_func_table *funcs)
1843{
1844 unsigned int n = funcs->number;
1845
1846 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1847 return -EINVAL;
1848 xfer_funcs[n] = funcs;
1849 return 0;
1850}
1851
1852static int unregister_transfer_cb(int id, void *ptr, void *data)
1853{
1854 struct loop_device *lo = ptr;
1855 struct loop_func_table *xfer = data;
1856
1857 mutex_lock(&loop_ctl_mutex);
1858 if (lo->lo_encryption == xfer)
1859 loop_release_xfer(lo);
1860 mutex_unlock(&loop_ctl_mutex);
1861 return 0;
1862}
1863
1864int loop_unregister_transfer(int number)
1865{
1866 unsigned int n = number;
1867 struct loop_func_table *xfer;
1868
1869 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1870 return -EINVAL;
1871
1872 xfer_funcs[n] = NULL;
1873 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1874 return 0;
1875}
1876
1877EXPORT_SYMBOL(loop_register_transfer);
1878EXPORT_SYMBOL(loop_unregister_transfer);
1879
1880static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1881 const struct blk_mq_queue_data *bd)
1882{
1883 struct request *rq = bd->rq;
1884 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1885 struct loop_device *lo = rq->q->queuedata;
1886
1887 blk_mq_start_request(rq);
1888
1889 if (lo->lo_state != Lo_bound)
1890 return BLK_STS_IOERR;
1891
1892 switch (req_op(rq)) {
1893 case REQ_OP_FLUSH:
1894 case REQ_OP_DISCARD:
1895 case REQ_OP_WRITE_ZEROES:
1896 cmd->use_aio = false;
1897 break;
1898 default:
1899 cmd->use_aio = lo->use_dio;
1900 break;
1901 }
1902
1903
1904#ifdef CONFIG_BLK_CGROUP
1905 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1906 cmd->css = &bio_blkcg(rq->bio)->css;
1907 css_get(cmd->css);
1908 } else
1909#endif
1910 cmd->css = NULL;
1911 kthread_queue_work(&lo->worker, &cmd->work);
1912
1913 return BLK_STS_OK;
1914}
1915
1916static void loop_handle_cmd(struct loop_cmd *cmd)
1917{
1918 struct request *rq = blk_mq_rq_from_pdu(cmd);
1919 const bool write = op_is_write(req_op(rq));
1920 struct loop_device *lo = rq->q->queuedata;
1921 int ret = 0;
1922
1923 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1924 ret = -EIO;
1925 goto failed;
1926 }
1927
1928 ret = do_req_filebacked(lo, rq);
1929 failed:
1930
1931 if (!cmd->use_aio || ret) {
1932 cmd->ret = ret ? -EIO : 0;
1933 blk_mq_complete_request(rq);
1934 }
1935}
1936
1937static void loop_queue_work(struct kthread_work *work)
1938{
1939 struct loop_cmd *cmd =
1940 container_of(work, struct loop_cmd, work);
1941
1942 loop_handle_cmd(cmd);
1943}
1944
1945static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1946 unsigned int hctx_idx, unsigned int numa_node)
1947{
1948 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1949
1950 kthread_init_work(&cmd->work, loop_queue_work);
1951 return 0;
1952}
1953
1954static const struct blk_mq_ops loop_mq_ops = {
1955 .queue_rq = loop_queue_rq,
1956 .init_request = loop_init_request,
1957 .complete = lo_complete_rq,
1958};
1959
1960static int loop_add(struct loop_device **l, int i)
1961{
1962 struct loop_device *lo;
1963 struct gendisk *disk;
1964 int err;
1965
1966 err = -ENOMEM;
1967 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1968 if (!lo)
1969 goto out;
1970
1971 lo->lo_state = Lo_unbound;
1972
1973
1974 if (i >= 0) {
1975 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1976 if (err == -ENOSPC)
1977 err = -EEXIST;
1978 } else {
1979 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1980 }
1981 if (err < 0)
1982 goto out_free_dev;
1983 i = err;
1984
1985 err = -ENOMEM;
1986 lo->tag_set.ops = &loop_mq_ops;
1987 lo->tag_set.nr_hw_queues = 1;
1988 lo->tag_set.queue_depth = 128;
1989 lo->tag_set.numa_node = NUMA_NO_NODE;
1990 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1991 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1992 lo->tag_set.driver_data = lo;
1993
1994 err = blk_mq_alloc_tag_set(&lo->tag_set);
1995 if (err)
1996 goto out_free_idr;
1997
1998 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1999 if (IS_ERR(lo->lo_queue)) {
2000 err = PTR_ERR(lo->lo_queue);
2001 goto out_cleanup_tags;
2002 }
2003 lo->lo_queue->queuedata = lo;
2004
2005 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2006
2007
2008
2009
2010
2011
2012
2013 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2014
2015 err = -ENOMEM;
2016 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2017 if (!disk)
2018 goto out_free_queue;
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038 if (!part_shift)
2039 disk->flags |= GENHD_FL_NO_PART_SCAN;
2040 disk->flags |= GENHD_FL_EXT_DEVT;
2041 atomic_set(&lo->lo_refcnt, 0);
2042 lo->lo_number = i;
2043 spin_lock_init(&lo->lo_lock);
2044 disk->major = LOOP_MAJOR;
2045 disk->first_minor = i << part_shift;
2046 disk->fops = &lo_fops;
2047 disk->private_data = lo;
2048 disk->queue = lo->lo_queue;
2049 sprintf(disk->disk_name, "loop%d", i);
2050 add_disk(disk);
2051 *l = lo;
2052 return lo->lo_number;
2053
2054out_free_queue:
2055 blk_cleanup_queue(lo->lo_queue);
2056out_cleanup_tags:
2057 blk_mq_free_tag_set(&lo->tag_set);
2058out_free_idr:
2059 idr_remove(&loop_index_idr, i);
2060out_free_dev:
2061 kfree(lo);
2062out:
2063 return err;
2064}
2065
2066static void loop_remove(struct loop_device *lo)
2067{
2068 del_gendisk(lo->lo_disk);
2069 blk_cleanup_queue(lo->lo_queue);
2070 blk_mq_free_tag_set(&lo->tag_set);
2071 put_disk(lo->lo_disk);
2072 kfree(lo);
2073}
2074
2075static int find_free_cb(int id, void *ptr, void *data)
2076{
2077 struct loop_device *lo = ptr;
2078 struct loop_device **l = data;
2079
2080 if (lo->lo_state == Lo_unbound) {
2081 *l = lo;
2082 return 1;
2083 }
2084 return 0;
2085}
2086
2087static int loop_lookup(struct loop_device **l, int i)
2088{
2089 struct loop_device *lo;
2090 int ret = -ENODEV;
2091
2092 if (i < 0) {
2093 int err;
2094
2095 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2096 if (err == 1) {
2097 *l = lo;
2098 ret = lo->lo_number;
2099 }
2100 goto out;
2101 }
2102
2103
2104 lo = idr_find(&loop_index_idr, i);
2105 if (lo) {
2106 *l = lo;
2107 ret = lo->lo_number;
2108 }
2109out:
2110 return ret;
2111}
2112
2113static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2114{
2115 struct loop_device *lo;
2116 struct kobject *kobj;
2117 int err;
2118
2119 mutex_lock(&loop_ctl_mutex);
2120 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2121 if (err < 0)
2122 err = loop_add(&lo, MINOR(dev) >> part_shift);
2123 if (err < 0)
2124 kobj = NULL;
2125 else
2126 kobj = get_disk_and_module(lo->lo_disk);
2127 mutex_unlock(&loop_ctl_mutex);
2128
2129 *part = 0;
2130 return kobj;
2131}
2132
2133static long loop_control_ioctl(struct file *file, unsigned int cmd,
2134 unsigned long parm)
2135{
2136 struct loop_device *lo;
2137 int ret;
2138
2139 ret = mutex_lock_killable(&loop_ctl_mutex);
2140 if (ret)
2141 return ret;
2142
2143 ret = -ENOSYS;
2144 switch (cmd) {
2145 case LOOP_CTL_ADD:
2146 ret = loop_lookup(&lo, parm);
2147 if (ret >= 0) {
2148 ret = -EEXIST;
2149 break;
2150 }
2151 ret = loop_add(&lo, parm);
2152 break;
2153 case LOOP_CTL_REMOVE:
2154 ret = loop_lookup(&lo, parm);
2155 if (ret < 0)
2156 break;
2157 if (lo->lo_state != Lo_unbound) {
2158 ret = -EBUSY;
2159 break;
2160 }
2161 if (atomic_read(&lo->lo_refcnt) > 0) {
2162 ret = -EBUSY;
2163 break;
2164 }
2165 lo->lo_disk->private_data = NULL;
2166 idr_remove(&loop_index_idr, lo->lo_number);
2167 loop_remove(lo);
2168 break;
2169 case LOOP_CTL_GET_FREE:
2170 ret = loop_lookup(&lo, -1);
2171 if (ret >= 0)
2172 break;
2173 ret = loop_add(&lo, -1);
2174 }
2175 mutex_unlock(&loop_ctl_mutex);
2176
2177 return ret;
2178}
2179
2180static const struct file_operations loop_ctl_fops = {
2181 .open = nonseekable_open,
2182 .unlocked_ioctl = loop_control_ioctl,
2183 .compat_ioctl = loop_control_ioctl,
2184 .owner = THIS_MODULE,
2185 .llseek = noop_llseek,
2186};
2187
2188static struct miscdevice loop_misc = {
2189 .minor = LOOP_CTRL_MINOR,
2190 .name = "loop-control",
2191 .fops = &loop_ctl_fops,
2192};
2193
2194MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2195MODULE_ALIAS("devname:loop-control");
2196
2197static int __init loop_init(void)
2198{
2199 int i, nr;
2200 unsigned long range;
2201 struct loop_device *lo;
2202 int err;
2203
2204 part_shift = 0;
2205 if (max_part > 0) {
2206 part_shift = fls(max_part);
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216 max_part = (1UL << part_shift) - 1;
2217 }
2218
2219 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2220 err = -EINVAL;
2221 goto err_out;
2222 }
2223
2224 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2225 err = -EINVAL;
2226 goto err_out;
2227 }
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237 if (max_loop) {
2238 nr = max_loop;
2239 range = max_loop << part_shift;
2240 } else {
2241 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2242 range = 1UL << MINORBITS;
2243 }
2244
2245 err = misc_register(&loop_misc);
2246 if (err < 0)
2247 goto err_out;
2248
2249
2250 if (register_blkdev(LOOP_MAJOR, "loop")) {
2251 err = -EIO;
2252 goto misc_out;
2253 }
2254
2255 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2256 THIS_MODULE, loop_probe, NULL, NULL);
2257
2258
2259 mutex_lock(&loop_ctl_mutex);
2260 for (i = 0; i < nr; i++)
2261 loop_add(&lo, i);
2262 mutex_unlock(&loop_ctl_mutex);
2263
2264 printk(KERN_INFO "loop: module loaded\n");
2265 return 0;
2266
2267misc_out:
2268 misc_deregister(&loop_misc);
2269err_out:
2270 return err;
2271}
2272
2273static int loop_exit_cb(int id, void *ptr, void *data)
2274{
2275 struct loop_device *lo = ptr;
2276
2277 loop_remove(lo);
2278 return 0;
2279}
2280
2281static void __exit loop_exit(void)
2282{
2283 unsigned long range;
2284
2285 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2286
2287 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2288 idr_destroy(&loop_index_idr);
2289
2290 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2291 unregister_blkdev(LOOP_MAJOR, "loop");
2292
2293 misc_deregister(&loop_misc);
2294}
2295
2296module_init(loop_init);
2297module_exit(loop_exit);
2298
2299#ifndef MODULE
2300static int __init max_loop_setup(char *str)
2301{
2302 max_loop = simple_strtol(str, NULL, 0);
2303 return 1;
2304}
2305
2306__setup("max_loop=", max_loop_setup);
2307#endif
2308