projects/linux·blob v4.7 · refs

2058 lines · 50.1 KiB

1/*
2 * linux/drivers/block/loop.c
3 *
4 * Written by Theodore Ts'o, 3/29/93
5 *
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
8 *
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 *
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 *
15 * Fixed do_loop_request() re-entrancy - [email protected] Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <[email protected]> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, [email protected]
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <[email protected]> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <[email protected]>, Oct 31, 1999
33 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <[email protected]>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <[email protected]>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
52#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/file.h>
57#include <linux/stat.h>
58#include <linux/errno.h>
59#include <linux/major.h>
60#include <linux/wait.h>
61#include <linux/blkdev.h>
62#include <linux/blkpg.h>
63#include <linux/init.h>
64#include <linux/swap.h>
65#include <linux/slab.h>
66#include <linux/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 void __loop_update_dio(struct loop_device *lo, bool dio)
168{
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
174 bool use_dio;
175
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
179 }
180
181 /*
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
186 *
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane appplications should be PAGE_SIZE algined
190 */
191 if (dio) {
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
195 !lo->transfer)
196 use_dio = true;
197 else
198 use_dio = false;
199 } else {
200 use_dio = false;
201 }
202
203 if (lo->use_dio == use_dio)
204 return;
205
206 /* flush dirty pages before changing direct IO */
207 vfs_fsync(file, 0);
208
209 /*
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
213 */
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
216 if (use_dio)
217 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
218 else
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 blk_mq_unfreeze_queue(lo->lo_queue);
221}
222
223static int
224figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
225{
226 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
227 sector_t x = (sector_t)size;
228 struct block_device *bdev = lo->lo_device;
229
230 if (unlikely((loff_t)x != size))
231 return -EFBIG;
232 if (lo->lo_offset != offset)
233 lo->lo_offset = offset;
234 if (lo->lo_sizelimit != sizelimit)
235 lo->lo_sizelimit = sizelimit;
236 set_capacity(lo->lo_disk, x);
237 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
238 /* let user-space know about the new size */
239 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
240 return 0;
241}
242
243static inline int
244lo_do_transfer(struct loop_device *lo, int cmd,
245 struct page *rpage, unsigned roffs,
246 struct page *lpage, unsigned loffs,
247 int size, sector_t rblock)
248{
249 int ret;
250
251 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
252 if (likely(!ret))
253 return 0;
254
255 printk_ratelimited(KERN_ERR
256 "loop: Transfer error at byte offset %llu, length %i.\n",
257 (unsigned long long)rblock << 9, size);
258 return ret;
259}
260
261static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
262{
263 struct iov_iter i;
264 ssize_t bw;
265
266 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
267
268 file_start_write(file);
269 bw = vfs_iter_write(file, &i, ppos);
270 file_end_write(file);
271
272 if (likely(bw == bvec->bv_len))
273 return 0;
274
275 printk_ratelimited(KERN_ERR
276 "loop: Write error at byte offset %llu, length %i.\n",
277 (unsigned long long)*ppos, bvec->bv_len);
278 if (bw >= 0)
279 bw = -EIO;
280 return bw;
281}
282
283static int lo_write_simple(struct loop_device *lo, struct request *rq,
284 loff_t pos)
285{
286 struct bio_vec bvec;
287 struct req_iterator iter;
288 int ret = 0;
289
290 rq_for_each_segment(bvec, rq, iter) {
291 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
292 if (ret < 0)
293 break;
294 cond_resched();
295 }
296
297 return ret;
298}
299
300/*
301 * This is the slow, transforming version that needs to double buffer the
302 * data as it cannot do the transformations in place without having direct
303 * access to the destination pages of the backing file.
304 */
305static int lo_write_transfer(struct loop_device *lo, struct request *rq,
306 loff_t pos)
307{
308 struct bio_vec bvec, b;
309 struct req_iterator iter;
310 struct page *page;
311 int ret = 0;
312
313 page = alloc_page(GFP_NOIO);
314 if (unlikely(!page))
315 return -ENOMEM;
316
317 rq_for_each_segment(bvec, rq, iter) {
318 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
319 bvec.bv_offset, bvec.bv_len, pos >> 9);
320 if (unlikely(ret))
321 break;
322
323 b.bv_page = page;
324 b.bv_offset = 0;
325 b.bv_len = bvec.bv_len;
326 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
327 if (ret < 0)
328 break;
329 }
330
331 __free_page(page);
332 return ret;
333}
334
335static int lo_read_simple(struct loop_device *lo, struct request *rq,
336 loff_t pos)
337{
338 struct bio_vec bvec;
339 struct req_iterator iter;
340 struct iov_iter i;
341 ssize_t len;
342
343 rq_for_each_segment(bvec, rq, iter) {
344 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
346 if (len < 0)
347 return len;
348
349 flush_dcache_page(bvec.bv_page);
350
351 if (len != bvec.bv_len) {
352 struct bio *bio;
353
354 __rq_for_each_bio(bio, rq)
355 zero_fill_bio(bio);
356 break;
357 }
358 cond_resched();
359 }
360
361 return 0;
362}
363
364static int lo_read_transfer(struct loop_device *lo, struct request *rq,
365 loff_t pos)
366{
367 struct bio_vec bvec, b;
368 struct req_iterator iter;
369 struct iov_iter i;
370 struct page *page;
371 ssize_t len;
372 int ret = 0;
373
374 page = alloc_page(GFP_NOIO);
375 if (unlikely(!page))
376 return -ENOMEM;
377
378 rq_for_each_segment(bvec, rq, iter) {
379 loff_t offset = pos;
380
381 b.bv_page = page;
382 b.bv_offset = 0;
383 b.bv_len = bvec.bv_len;
384
385 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos);
387 if (len < 0) {
388 ret = len;
389 goto out_free_page;
390 }
391
392 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
393 bvec.bv_offset, len, offset >> 9);
394 if (ret)
395 goto out_free_page;
396
397 flush_dcache_page(bvec.bv_page);
398
399 if (len != bvec.bv_len) {
400 struct bio *bio;
401
402 __rq_for_each_bio(bio, rq)
403 zero_fill_bio(bio);
404 break;
405 }
406 }
407
408 ret = 0;
409out_free_page:
410 __free_page(page);
411 return ret;
412}
413
414static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
415{
416 /*
417 * We use punch hole to reclaim the free space used by the
418 * image a.k.a. discard. However we do not support discard if
419 * encryption is enabled, because it may give an attacker
420 * useful information.
421 */
422 struct file *file = lo->lo_backing_file;
423 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
424 int ret;
425
426 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
427 ret = -EOPNOTSUPP;
428 goto out;
429 }
430
431 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
432 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
433 ret = -EIO;
434 out:
435 return ret;
436}
437
438static int lo_req_flush(struct loop_device *lo, struct request *rq)
439{
440 struct file *file = lo->lo_backing_file;
441 int ret = vfs_fsync(file, 0);
442 if (unlikely(ret && ret != -EINVAL))
443 ret = -EIO;
444
445 return ret;
446}
447
448static inline void handle_partial_read(struct loop_cmd *cmd, long bytes)
449{
450 if (bytes < 0 || (cmd->rq->cmd_flags & REQ_WRITE))
451 return;
452
453 if (unlikely(bytes < blk_rq_bytes(cmd->rq))) {
454 struct bio *bio = cmd->rq->bio;
455
456 bio_advance(bio, bytes);
457 zero_fill_bio(bio);
458 }
459}
460
461static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
462{
463 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
464 struct request *rq = cmd->rq;
465
466 handle_partial_read(cmd, ret);
467
468 if (ret > 0)
469 ret = 0;
470 else if (ret < 0)
471 ret = -EIO;
472
473 blk_mq_complete_request(rq, ret);
474}
475
476static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
477 loff_t pos, bool rw)
478{
479 struct iov_iter iter;
480 struct bio_vec *bvec;
481 struct bio *bio = cmd->rq->bio;
482 struct file *file = lo->lo_backing_file;
483 int ret;
484
485 /* nomerge for loop request queue */
486 WARN_ON(cmd->rq->bio != cmd->rq->biotail);
487
488 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
489 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
490 bio_segments(bio), blk_rq_bytes(cmd->rq));
491 /*
492 * This bio may be started from the middle of the 'bvec'
493 * because of bio splitting, so offset from the bvec must
494 * be passed to iov iterator
495 */
496 iter.iov_offset = bio->bi_iter.bi_bvec_done;
497
498 cmd->iocb.ki_pos = pos;
499 cmd->iocb.ki_filp = file;
500 cmd->iocb.ki_complete = lo_rw_aio_complete;
501 cmd->iocb.ki_flags = IOCB_DIRECT;
502
503 if (rw == WRITE)
504 ret = file->f_op->write_iter(&cmd->iocb, &iter);
505 else
506 ret = file->f_op->read_iter(&cmd->iocb, &iter);
507
508 if (ret != -EIOCBQUEUED)
509 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
510 return 0;
511}
512
513
514static inline int lo_rw_simple(struct loop_device *lo,
515 struct request *rq, loff_t pos, bool rw)
516{
517 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
518
519 if (cmd->use_aio)
520 return lo_rw_aio(lo, cmd, pos, rw);
521
522 /*
523 * lo_write_simple and lo_read_simple should have been covered
524 * by io submit style function like lo_rw_aio(), one blocker
525 * is that lo_read_simple() need to call flush_dcache_page after
526 * the page is written from kernel, and it isn't easy to handle
527 * this in io submit style function which submits all segments
528 * of the req at one time. And direct read IO doesn't need to
529 * run flush_dcache_page().
530 */
531 if (rw == WRITE)
532 return lo_write_simple(lo, rq, pos);
533 else
534 return lo_read_simple(lo, rq, pos);
535}
536
537static int do_req_filebacked(struct loop_device *lo, struct request *rq)
538{
539 loff_t pos;
540 int ret;
541
542 pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
543
544 if (rq->cmd_flags & REQ_WRITE) {
545 if (rq->cmd_flags & REQ_FLUSH)
546 ret = lo_req_flush(lo, rq);
547 else if (rq->cmd_flags & REQ_DISCARD)
548 ret = lo_discard(lo, rq, pos);
549 else if (lo->transfer)
550 ret = lo_write_transfer(lo, rq, pos);
551 else
552 ret = lo_rw_simple(lo, rq, pos, WRITE);
553
554 } else {
555 if (lo->transfer)
556 ret = lo_read_transfer(lo, rq, pos);
557 else
558 ret = lo_rw_simple(lo, rq, pos, READ);
559 }
560
561 return ret;
562}
563
564struct switch_request {
565 struct file *file;
566 struct completion wait;
567};
568
569static inline void loop_update_dio(struct loop_device *lo)
570{
571 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
572 lo->use_dio);
573}
574
575/*
576 * Do the actual switch; called from the BIO completion routine
577 */
578static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
579{
580 struct file *file = p->file;
581 struct file *old_file = lo->lo_backing_file;
582 struct address_space *mapping;
583
584 /* if no new file, only flush of queued bios requested */
585 if (!file)
586 return;
587
588 mapping = file->f_mapping;
589 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
590 lo->lo_backing_file = file;
591 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
592 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
593 lo->old_gfp_mask = mapping_gfp_mask(mapping);
594 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
595 loop_update_dio(lo);
596}
597
598/*
599 * loop_switch performs the hard work of switching a backing store.
600 * First it needs to flush existing IO, it does this by sending a magic
601 * BIO down the pipe. The completion of this BIO does the actual switch.
602 */
603static int loop_switch(struct loop_device *lo, struct file *file)
604{
605 struct switch_request w;
606
607 w.file = file;
608
609 /* freeze queue and wait for completion of scheduled requests */
610 blk_mq_freeze_queue(lo->lo_queue);
611
612 /* do the switch action */
613 do_loop_switch(lo, &w);
614
615 /* unfreeze */
616 blk_mq_unfreeze_queue(lo->lo_queue);
617
618 return 0;
619}
620
621/*
622 * Helper to flush the IOs in loop, but keeping loop thread running
623 */
624static int loop_flush(struct loop_device *lo)
625{
626 return loop_switch(lo, NULL);
627}
628
629static void loop_reread_partitions(struct loop_device *lo,
630 struct block_device *bdev)
631{
632 int rc;
633
634 /*
635 * bd_mutex has been held already in release path, so don't
636 * acquire it if this function is called in such case.
637 *
638 * If the reread partition isn't from release path, lo_refcnt
639 * must be at least one and it can only become zero when the
640 * current holder is released.
641 */
642 if (!atomic_read(&lo->lo_refcnt))
643 rc = __blkdev_reread_part(bdev);
644 else
645 rc = blkdev_reread_part(bdev);
646 if (rc)
647 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
648 __func__, lo->lo_number, lo->lo_file_name, rc);
649}
650
651/*
652 * loop_change_fd switched the backing store of a loopback device to
653 * a new file. This is useful for operating system installers to free up
654 * the original file and in High Availability environments to switch to
655 * an alternative location for the content in case of server meltdown.
656 * This can only work if the loop device is used read-only, and if the
657 * new backing store is the same size and type as the old backing store.
658 */
659static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
660 unsigned int arg)
661{
662 struct file *file, *old_file;
663 struct inode *inode;
664 int error;
665
666 error = -ENXIO;
667 if (lo->lo_state != Lo_bound)
668 goto out;
669
670 /* the loop device has to be read-only */
671 error = -EINVAL;
672 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
673 goto out;
674
675 error = -EBADF;
676 file = fget(arg);
677 if (!file)
678 goto out;
679
680 inode = file->f_mapping->host;
681 old_file = lo->lo_backing_file;
682
683 error = -EINVAL;
684
685 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
686 goto out_putf;
687
688 /* size of the new backing store needs to be the same */
689 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
690 goto out_putf;
691
692 /* and ... switch */
693 error = loop_switch(lo, file);
694 if (error)
695 goto out_putf;
696
697 fput(old_file);
698 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
699 loop_reread_partitions(lo, bdev);
700 return 0;
701
702 out_putf:
703 fput(file);
704 out:
705 return error;
706}
707
708static inline int is_loop_device(struct file *file)
709{
710 struct inode *i = file->f_mapping->host;
711
712 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
713}
714
715/* loop sysfs attributes */
716
717static ssize_t loop_attr_show(struct device *dev, char *page,
718 ssize_t (*callback)(struct loop_device *, char *))
719{
720 struct gendisk *disk = dev_to_disk(dev);
721 struct loop_device *lo = disk->private_data;
722
723 return callback(lo, page);
724}
725
726#define LOOP_ATTR_RO(_name) \
727static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
728static ssize_t loop_attr_do_show_##_name(struct device *d, \
729 struct device_attribute *attr, char *b) \
730{ \
731 return loop_attr_show(d, b, loop_attr_##_name##_show); \
732} \
733static struct device_attribute loop_attr_##_name = \
734 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
735
736static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
737{
738 ssize_t ret;
739 char *p = NULL;
740
741 spin_lock_irq(&lo->lo_lock);
742 if (lo->lo_backing_file)
743 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
744 spin_unlock_irq(&lo->lo_lock);
745
746 if (IS_ERR_OR_NULL(p))
747 ret = PTR_ERR(p);
748 else {
749 ret = strlen(p);
750 memmove(buf, p, ret);
751 buf[ret++] = '\n';
752 buf[ret] = 0;
753 }
754
755 return ret;
756}
757
758static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
759{
760 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
761}
762
763static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
764{
765 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
766}
767
768static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
769{
770 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
771
772 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
773}
774
775static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
776{
777 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
778
779 return sprintf(buf, "%s\n", partscan ? "1" : "0");
780}
781
782static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
783{
784 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
785
786 return sprintf(buf, "%s\n", dio ? "1" : "0");
787}
788
789LOOP_ATTR_RO(backing_file);
790LOOP_ATTR_RO(offset);
791LOOP_ATTR_RO(sizelimit);
792LOOP_ATTR_RO(autoclear);
793LOOP_ATTR_RO(partscan);
794LOOP_ATTR_RO(dio);
795
796static struct attribute *loop_attrs[] = {
797 &loop_attr_backing_file.attr,
798 &loop_attr_offset.attr,
799 &loop_attr_sizelimit.attr,
800 &loop_attr_autoclear.attr,
801 &loop_attr_partscan.attr,
802 &loop_attr_dio.attr,
803 NULL,
804};
805
806static struct attribute_group loop_attribute_group = {
807 .name = "loop",
808 .attrs= loop_attrs,
809};
810
811static int loop_sysfs_init(struct loop_device *lo)
812{
813 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
814 &loop_attribute_group);
815}
816
817static void loop_sysfs_exit(struct loop_device *lo)
818{
819 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
820 &loop_attribute_group);
821}
822
823static void loop_config_discard(struct loop_device *lo)
824{
825 struct file *file = lo->lo_backing_file;
826 struct inode *inode = file->f_mapping->host;
827 struct request_queue *q = lo->lo_queue;
828
829 /*
830 * We use punch hole to reclaim the free space used by the
831 * image a.k.a. discard. However we do not support discard if
832 * encryption is enabled, because it may give an attacker
833 * useful information.
834 */
835 if ((!file->f_op->fallocate) ||
836 lo->lo_encrypt_key_size) {
837 q->limits.discard_granularity = 0;
838 q->limits.discard_alignment = 0;
839 blk_queue_max_discard_sectors(q, 0);
840 q->limits.discard_zeroes_data = 0;
841 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
842 return;
843 }
844
845 q->limits.discard_granularity = inode->i_sb->s_blocksize;
846 q->limits.discard_alignment = 0;
847 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
848 q->limits.discard_zeroes_data = 1;
849 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
850}
851
852static void loop_unprepare_queue(struct loop_device *lo)
853{
854 flush_kthread_worker(&lo->worker);
855 kthread_stop(lo->worker_task);
856}
857
858static int loop_prepare_queue(struct loop_device *lo)
859{
860 init_kthread_worker(&lo->worker);
861 lo->worker_task = kthread_run(kthread_worker_fn,
862 &lo->worker, "loop%d", lo->lo_number);
863 if (IS_ERR(lo->worker_task))
864 return -ENOMEM;
865 set_user_nice(lo->worker_task, MIN_NICE);
866 return 0;
867}
868
869static int loop_set_fd(struct loop_device *lo, fmode_t mode,
870 struct block_device *bdev, unsigned int arg)
871{
872 struct file *file, *f;
873 struct inode *inode;
874 struct address_space *mapping;
875 unsigned lo_blocksize;
876 int lo_flags = 0;
877 int error;
878 loff_t size;
879
880 /* This is safe, since we have a reference from open(). */
881 __module_get(THIS_MODULE);
882
883 error = -EBADF;
884 file = fget(arg);
885 if (!file)
886 goto out;
887
888 error = -EBUSY;
889 if (lo->lo_state != Lo_unbound)
890 goto out_putf;
891
892 /* Avoid recursion */
893 f = file;
894 while (is_loop_device(f)) {
895 struct loop_device *l;
896
897 if (f->f_mapping->host->i_bdev == bdev)
898 goto out_putf;
899
900 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
901 if (l->lo_state == Lo_unbound) {
902 error = -EINVAL;
903 goto out_putf;
904 }
905 f = l->lo_backing_file;
906 }
907
908 mapping = file->f_mapping;
909 inode = mapping->host;
910
911 error = -EINVAL;
912 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
913 goto out_putf;
914
915 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
916 !file->f_op->write_iter)
917 lo_flags |= LO_FLAGS_READ_ONLY;
918
919 lo_blocksize = S_ISBLK(inode->i_mode) ?
920 inode->i_bdev->bd_block_size : PAGE_SIZE;
921
922 error = -EFBIG;
923 size = get_loop_size(lo, file);
924 if ((loff_t)(sector_t)size != size)
925 goto out_putf;
926 error = loop_prepare_queue(lo);
927 if (error)
928 goto out_putf;
929
930 error = 0;
931
932 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
933
934 lo->use_dio = false;
935 lo->lo_blocksize = lo_blocksize;
936 lo->lo_device = bdev;
937 lo->lo_flags = lo_flags;
938 lo->lo_backing_file = file;
939 lo->transfer = NULL;
940 lo->ioctl = NULL;
941 lo->lo_sizelimit = 0;
942 lo->old_gfp_mask = mapping_gfp_mask(mapping);
943 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
944
945 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
946 blk_queue_write_cache(lo->lo_queue, true, false);
947
948 loop_update_dio(lo);
949 set_capacity(lo->lo_disk, size);
950 bd_set_size(bdev, size << 9);
951 loop_sysfs_init(lo);
952 /* let user-space know about the new size */
953 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
954
955 set_blocksize(bdev, lo_blocksize);
956
957 lo->lo_state = Lo_bound;
958 if (part_shift)
959 lo->lo_flags |= LO_FLAGS_PARTSCAN;
960 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
961 loop_reread_partitions(lo, bdev);
962
963 /* Grab the block_device to prevent its destruction after we
964 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
965 */
966 bdgrab(bdev);
967 return 0;
968
969 out_putf:
970 fput(file);
971 out:
972 /* This is safe: open() is still holding a reference. */
973 module_put(THIS_MODULE);
974 return error;
975}
976
977static int
978loop_release_xfer(struct loop_device *lo)
979{
980 int err = 0;
981 struct loop_func_table *xfer = lo->lo_encryption;
982
983 if (xfer) {
984 if (xfer->release)
985 err = xfer->release(lo);
986 lo->transfer = NULL;
987 lo->lo_encryption = NULL;
988 module_put(xfer->owner);
989 }
990 return err;
991}
992
993static int
994loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
995 const struct loop_info64 *i)
996{
997 int err = 0;
998
999 if (xfer) {
1000 struct module *owner = xfer->owner;
1001
1002 if (!try_module_get(owner))
1003 return -EINVAL;
1004 if (xfer->init)
1005 err = xfer->init(lo, i);
1006 if (err)
1007 module_put(owner);
1008 else
1009 lo->lo_encryption = xfer;
1010 }
1011 return err;
1012}
1013
1014static int loop_clr_fd(struct loop_device *lo)
1015{
1016 struct file *filp = lo->lo_backing_file;
1017 gfp_t gfp = lo->old_gfp_mask;
1018 struct block_device *bdev = lo->lo_device;
1019
1020 if (lo->lo_state != Lo_bound)
1021 return -ENXIO;
1022
1023 /*
1024 * If we've explicitly asked to tear down the loop device,
1025 * and it has an elevated reference count, set it for auto-teardown when
1026 * the last reference goes away. This stops $!~#$@ udev from
1027 * preventing teardown because it decided that it needs to run blkid on
1028 * the loopback device whenever they appear. xfstests is notorious for
1029 * failing tests because blkid via udev races with a losetup
1030 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1031 * command to fail with EBUSY.
1032 */
1033 if (atomic_read(&lo->lo_refcnt) > 1) {
1034 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1035 mutex_unlock(&lo->lo_ctl_mutex);
1036 return 0;
1037 }
1038
1039 if (filp == NULL)
1040 return -EINVAL;
1041
1042 /* freeze request queue during the transition */
1043 blk_mq_freeze_queue(lo->lo_queue);
1044
1045 spin_lock_irq(&lo->lo_lock);
1046 lo->lo_state = Lo_rundown;
1047 lo->lo_backing_file = NULL;
1048 spin_unlock_irq(&lo->lo_lock);
1049
1050 loop_release_xfer(lo);
1051 lo->transfer = NULL;
1052 lo->ioctl = NULL;
1053 lo->lo_device = NULL;
1054 lo->lo_encryption = NULL;
1055 lo->lo_offset = 0;
1056 lo->lo_sizelimit = 0;
1057 lo->lo_encrypt_key_size = 0;
1058 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1059 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1060 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1061 if (bdev) {
1062 bdput(bdev);
1063 invalidate_bdev(bdev);
1064 }
1065 set_capacity(lo->lo_disk, 0);
1066 loop_sysfs_exit(lo);
1067 if (bdev) {
1068 bd_set_size(bdev, 0);
1069 /* let user-space know about this change */
1070 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1071 }
1072 mapping_set_gfp_mask(filp->f_mapping, gfp);
1073 lo->lo_state = Lo_unbound;
1074 /* This is safe: open() is still holding a reference. */
1075 module_put(THIS_MODULE);
1076 blk_mq_unfreeze_queue(lo->lo_queue);
1077
1078 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1079 loop_reread_partitions(lo, bdev);
1080 lo->lo_flags = 0;
1081 if (!part_shift)
1082 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1083 loop_unprepare_queue(lo);
1084 mutex_unlock(&lo->lo_ctl_mutex);
1085 /*
1086 * Need not hold lo_ctl_mutex to fput backing file.
1087 * Calling fput holding lo_ctl_mutex triggers a circular
1088 * lock dependency possibility warning as fput can take
1089 * bd_mutex which is usually taken before lo_ctl_mutex.
1090 */
1091 fput(filp);
1092 return 0;
1093}
1094
1095static int
1096loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1097{
1098 int err;
1099 struct loop_func_table *xfer;
1100 kuid_t uid = current_uid();
1101
1102 if (lo->lo_encrypt_key_size &&
1103 !uid_eq(lo->lo_key_owner, uid) &&
1104 !capable(CAP_SYS_ADMIN))
1105 return -EPERM;
1106 if (lo->lo_state != Lo_bound)
1107 return -ENXIO;
1108 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1109 return -EINVAL;
1110
1111 err = loop_release_xfer(lo);
1112 if (err)
1113 return err;
1114
1115 if (info->lo_encrypt_type) {
1116 unsigned int type = info->lo_encrypt_type;
1117
1118 if (type >= MAX_LO_CRYPT)
1119 return -EINVAL;
1120 xfer = xfer_funcs[type];
1121 if (xfer == NULL)
1122 return -EINVAL;
1123 } else
1124 xfer = NULL;
1125
1126 err = loop_init_xfer(lo, xfer, info);
1127 if (err)
1128 return err;
1129
1130 if (lo->lo_offset != info->lo_offset ||
1131 lo->lo_sizelimit != info->lo_sizelimit)
1132 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1133 return -EFBIG;
1134
1135 loop_config_discard(lo);
1136
1137 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1138 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1139 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1140 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1141
1142 if (!xfer)
1143 xfer = &none_funcs;
1144 lo->transfer = xfer->transfer;
1145 lo->ioctl = xfer->ioctl;
1146
1147 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1148 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1149 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1150
1151 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1152 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1153 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1154 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1155 loop_reread_partitions(lo, lo->lo_device);
1156 }
1157
1158 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1159 lo->lo_init[0] = info->lo_init[0];
1160 lo->lo_init[1] = info->lo_init[1];
1161 if (info->lo_encrypt_key_size) {
1162 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1163 info->lo_encrypt_key_size);
1164 lo->lo_key_owner = uid;
1165 }
1166
1167 /* update dio if lo_offset or transfer is changed */
1168 __loop_update_dio(lo, lo->use_dio);
1169
1170 return 0;
1171}
1172
1173static int
1174loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1175{
1176 struct file *file = lo->lo_backing_file;
1177 struct kstat stat;
1178 int error;
1179
1180 if (lo->lo_state != Lo_bound)
1181 return -ENXIO;
1182 error = vfs_getattr(&file->f_path, &stat);
1183 if (error)
1184 return error;
1185 memset(info, 0, sizeof(*info));
1186 info->lo_number = lo->lo_number;
1187 info->lo_device = huge_encode_dev(stat.dev);
1188 info->lo_inode = stat.ino;
1189 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1190 info->lo_offset = lo->lo_offset;
1191 info->lo_sizelimit = lo->lo_sizelimit;
1192 info->lo_flags = lo->lo_flags;
1193 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1194 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1195 info->lo_encrypt_type =
1196 lo->lo_encryption ? lo->lo_encryption->number : 0;
1197 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1198 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1199 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1200 lo->lo_encrypt_key_size);
1201 }
1202 return 0;
1203}
1204
1205static void
1206loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1207{
1208 memset(info64, 0, sizeof(*info64));
1209 info64->lo_number = info->lo_number;
1210 info64->lo_device = info->lo_device;
1211 info64->lo_inode = info->lo_inode;
1212 info64->lo_rdevice = info->lo_rdevice;
1213 info64->lo_offset = info->lo_offset;
1214 info64->lo_sizelimit = 0;
1215 info64->lo_encrypt_type = info->lo_encrypt_type;
1216 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1217 info64->lo_flags = info->lo_flags;
1218 info64->lo_init[0] = info->lo_init[0];
1219 info64->lo_init[1] = info->lo_init[1];
1220 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1221 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1222 else
1223 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1224 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1225}
1226
1227static int
1228loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1229{
1230 memset(info, 0, sizeof(*info));
1231 info->lo_number = info64->lo_number;
1232 info->lo_device = info64->lo_device;
1233 info->lo_inode = info64->lo_inode;
1234 info->lo_rdevice = info64->lo_rdevice;
1235 info->lo_offset = info64->lo_offset;
1236 info->lo_encrypt_type = info64->lo_encrypt_type;
1237 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1238 info->lo_flags = info64->lo_flags;
1239 info->lo_init[0] = info64->lo_init[0];
1240 info->lo_init[1] = info64->lo_init[1];
1241 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1242 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1243 else
1244 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1245 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1246
1247 /* error in case values were truncated */
1248 if (info->lo_device != info64->lo_device ||
1249 info->lo_rdevice != info64->lo_rdevice ||
1250 info->lo_inode != info64->lo_inode ||
1251 info->lo_offset != info64->lo_offset)
1252 return -EOVERFLOW;
1253
1254 return 0;
1255}
1256
1257static int
1258loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1259{
1260 struct loop_info info;
1261 struct loop_info64 info64;
1262
1263 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1264 return -EFAULT;
1265 loop_info64_from_old(&info, &info64);
1266 return loop_set_status(lo, &info64);
1267}
1268
1269static int
1270loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1271{
1272 struct loop_info64 info64;
1273
1274 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1275 return -EFAULT;
1276 return loop_set_status(lo, &info64);
1277}
1278
1279static int
1280loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1281 struct loop_info info;
1282 struct loop_info64 info64;
1283 int err = 0;
1284
1285 if (!arg)
1286 err = -EINVAL;
1287 if (!err)
1288 err = loop_get_status(lo, &info64);
1289 if (!err)
1290 err = loop_info64_to_old(&info64, &info);
1291 if (!err && copy_to_user(arg, &info, sizeof(info)))
1292 err = -EFAULT;
1293
1294 return err;
1295}
1296
1297static int
1298loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1299 struct loop_info64 info64;
1300 int err = 0;
1301
1302 if (!arg)
1303 err = -EINVAL;
1304 if (!err)
1305 err = loop_get_status(lo, &info64);
1306 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1307 err = -EFAULT;
1308
1309 return err;
1310}
1311
1312static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1313{
1314 if (unlikely(lo->lo_state != Lo_bound))
1315 return -ENXIO;
1316
1317 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1318}
1319
1320static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1321{
1322 int error = -ENXIO;
1323 if (lo->lo_state != Lo_bound)
1324 goto out;
1325
1326 __loop_update_dio(lo, !!arg);
1327 if (lo->use_dio == !!arg)
1328 return 0;
1329 error = -EINVAL;
1330 out:
1331 return error;
1332}
1333
1334static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1335 unsigned int cmd, unsigned long arg)
1336{
1337 struct loop_device *lo = bdev->bd_disk->private_data;
1338 int err;
1339
1340 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1341 switch (cmd) {
1342 case LOOP_SET_FD:
1343 err = loop_set_fd(lo, mode, bdev, arg);
1344 break;
1345 case LOOP_CHANGE_FD:
1346 err = loop_change_fd(lo, bdev, arg);
1347 break;
1348 case LOOP_CLR_FD:
1349 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1350 err = loop_clr_fd(lo);
1351 if (!err)
1352 goto out_unlocked;
1353 break;
1354 case LOOP_SET_STATUS:
1355 err = -EPERM;
1356 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1357 err = loop_set_status_old(lo,
1358 (struct loop_info __user *)arg);
1359 break;
1360 case LOOP_GET_STATUS:
1361 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1362 break;
1363 case LOOP_SET_STATUS64:
1364 err = -EPERM;
1365 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1366 err = loop_set_status64(lo,
1367 (struct loop_info64 __user *) arg);
1368 break;
1369 case LOOP_GET_STATUS64:
1370 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1371 break;
1372 case LOOP_SET_CAPACITY:
1373 err = -EPERM;
1374 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1375 err = loop_set_capacity(lo, bdev);
1376 break;
1377 case LOOP_SET_DIRECT_IO:
1378 err = -EPERM;
1379 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1380 err = loop_set_dio(lo, arg);
1381 break;
1382 default:
1383 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1384 }
1385 mutex_unlock(&lo->lo_ctl_mutex);
1386
1387out_unlocked:
1388 return err;
1389}
1390
1391#ifdef CONFIG_COMPAT
1392struct compat_loop_info {
1393 compat_int_t lo_number; /* ioctl r/o */
1394 compat_dev_t lo_device; /* ioctl r/o */
1395 compat_ulong_t lo_inode; /* ioctl r/o */
1396 compat_dev_t lo_rdevice; /* ioctl r/o */
1397 compat_int_t lo_offset;
1398 compat_int_t lo_encrypt_type;
1399 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1400 compat_int_t lo_flags; /* ioctl r/o */
1401 char lo_name[LO_NAME_SIZE];
1402 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1403 compat_ulong_t lo_init[2];
1404 char reserved[4];
1405};
1406
1407/*
1408 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1409 * - noinlined to reduce stack space usage in main part of driver
1410 */
1411static noinline int
1412loop_info64_from_compat(const struct compat_loop_info __user *arg,
1413 struct loop_info64 *info64)
1414{
1415 struct compat_loop_info info;
1416
1417 if (copy_from_user(&info, arg, sizeof(info)))
1418 return -EFAULT;
1419
1420 memset(info64, 0, sizeof(*info64));
1421 info64->lo_number = info.lo_number;
1422 info64->lo_device = info.lo_device;
1423 info64->lo_inode = info.lo_inode;
1424 info64->lo_rdevice = info.lo_rdevice;
1425 info64->lo_offset = info.lo_offset;
1426 info64->lo_sizelimit = 0;
1427 info64->lo_encrypt_type = info.lo_encrypt_type;
1428 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1429 info64->lo_flags = info.lo_flags;
1430 info64->lo_init[0] = info.lo_init[0];
1431 info64->lo_init[1] = info.lo_init[1];
1432 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1433 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1434 else
1435 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1436 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1437 return 0;
1438}
1439
1440/*
1441 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1442 * - noinlined to reduce stack space usage in main part of driver
1443 */
1444static noinline int
1445loop_info64_to_compat(const struct loop_info64 *info64,
1446 struct compat_loop_info __user *arg)
1447{
1448 struct compat_loop_info info;
1449
1450 memset(&info, 0, sizeof(info));
1451 info.lo_number = info64->lo_number;
1452 info.lo_device = info64->lo_device;
1453 info.lo_inode = info64->lo_inode;
1454 info.lo_rdevice = info64->lo_rdevice;
1455 info.lo_offset = info64->lo_offset;
1456 info.lo_encrypt_type = info64->lo_encrypt_type;
1457 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1458 info.lo_flags = info64->lo_flags;
1459 info.lo_init[0] = info64->lo_init[0];
1460 info.lo_init[1] = info64->lo_init[1];
1461 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1462 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1463 else
1464 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1465 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1466
1467 /* error in case values were truncated */
1468 if (info.lo_device != info64->lo_device ||
1469 info.lo_rdevice != info64->lo_rdevice ||
1470 info.lo_inode != info64->lo_inode ||
1471 info.lo_offset != info64->lo_offset ||
1472 info.lo_init[0] != info64->lo_init[0] ||
1473 info.lo_init[1] != info64->lo_init[1])
1474 return -EOVERFLOW;
1475
1476 if (copy_to_user(arg, &info, sizeof(info)))
1477 return -EFAULT;
1478 return 0;
1479}
1480
1481static int
1482loop_set_status_compat(struct loop_device *lo,
1483 const struct compat_loop_info __user *arg)
1484{
1485 struct loop_info64 info64;
1486 int ret;
1487
1488 ret = loop_info64_from_compat(arg, &info64);
1489 if (ret < 0)
1490 return ret;
1491 return loop_set_status(lo, &info64);
1492}
1493
1494static int
1495loop_get_status_compat(struct loop_device *lo,
1496 struct compat_loop_info __user *arg)
1497{
1498 struct loop_info64 info64;
1499 int err = 0;
1500
1501 if (!arg)
1502 err = -EINVAL;
1503 if (!err)
1504 err = loop_get_status(lo, &info64);
1505 if (!err)
1506 err = loop_info64_to_compat(&info64, arg);
1507 return err;
1508}
1509
1510static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1511 unsigned int cmd, unsigned long arg)
1512{
1513 struct loop_device *lo = bdev->bd_disk->private_data;
1514 int err;
1515
1516 switch(cmd) {
1517 case LOOP_SET_STATUS:
1518 mutex_lock(&lo->lo_ctl_mutex);
1519 err = loop_set_status_compat(
1520 lo, (const struct compat_loop_info __user *) arg);
1521 mutex_unlock(&lo->lo_ctl_mutex);
1522 break;
1523 case LOOP_GET_STATUS:
1524 mutex_lock(&lo->lo_ctl_mutex);
1525 err = loop_get_status_compat(
1526 lo, (struct compat_loop_info __user *) arg);
1527 mutex_unlock(&lo->lo_ctl_mutex);
1528 break;
1529 case LOOP_SET_CAPACITY:
1530 case LOOP_CLR_FD:
1531 case LOOP_GET_STATUS64:
1532 case LOOP_SET_STATUS64:
1533 arg = (unsigned long) compat_ptr(arg);
1534 case LOOP_SET_FD:
1535 case LOOP_CHANGE_FD:
1536 err = lo_ioctl(bdev, mode, cmd, arg);
1537 break;
1538 default:
1539 err = -ENOIOCTLCMD;
1540 break;
1541 }
1542 return err;
1543}
1544#endif
1545
1546static int lo_open(struct block_device *bdev, fmode_t mode)
1547{
1548 struct loop_device *lo;
1549 int err = 0;
1550
1551 mutex_lock(&loop_index_mutex);
1552 lo = bdev->bd_disk->private_data;
1553 if (!lo) {
1554 err = -ENXIO;
1555 goto out;
1556 }
1557
1558 atomic_inc(&lo->lo_refcnt);
1559out:
1560 mutex_unlock(&loop_index_mutex);
1561 return err;
1562}
1563
1564static void lo_release(struct gendisk *disk, fmode_t mode)
1565{
1566 struct loop_device *lo = disk->private_data;
1567 int err;
1568
1569 if (atomic_dec_return(&lo->lo_refcnt))
1570 return;
1571
1572 mutex_lock(&lo->lo_ctl_mutex);
1573 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1574 /*
1575 * In autoclear mode, stop the loop thread
1576 * and remove configuration after last close.
1577 */
1578 err = loop_clr_fd(lo);
1579 if (!err)
1580 return;
1581 } else {
1582 /*
1583 * Otherwise keep thread (if running) and config,
1584 * but flush possible ongoing bios in thread.
1585 */
1586 loop_flush(lo);
1587 }
1588
1589 mutex_unlock(&lo->lo_ctl_mutex);
1590}
1591
1592static const struct block_device_operations lo_fops = {
1593 .owner = THIS_MODULE,
1594 .open = lo_open,
1595 .release = lo_release,
1596 .ioctl = lo_ioctl,
1597#ifdef CONFIG_COMPAT
1598 .compat_ioctl = lo_compat_ioctl,
1599#endif
1600};
1601
1602/*
1603 * And now the modules code and kernel interface.
1604 */
1605static int max_loop;
1606module_param(max_loop, int, S_IRUGO);
1607MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1608module_param(max_part, int, S_IRUGO);
1609MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1610MODULE_LICENSE("GPL");
1611MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1612
1613int loop_register_transfer(struct loop_func_table *funcs)
1614{
1615 unsigned int n = funcs->number;
1616
1617 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1618 return -EINVAL;
1619 xfer_funcs[n] = funcs;
1620 return 0;
1621}
1622
1623static int unregister_transfer_cb(int id, void *ptr, void *data)
1624{
1625 struct loop_device *lo = ptr;
1626 struct loop_func_table *xfer = data;
1627
1628 mutex_lock(&lo->lo_ctl_mutex);
1629 if (lo->lo_encryption == xfer)
1630 loop_release_xfer(lo);
1631 mutex_unlock(&lo->lo_ctl_mutex);
1632 return 0;
1633}
1634
1635int loop_unregister_transfer(int number)
1636{
1637 unsigned int n = number;
1638 struct loop_func_table *xfer;
1639
1640 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1641 return -EINVAL;
1642
1643 xfer_funcs[n] = NULL;
1644 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1645 return 0;
1646}
1647
1648EXPORT_SYMBOL(loop_register_transfer);
1649EXPORT_SYMBOL(loop_unregister_transfer);
1650
1651static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1652 const struct blk_mq_queue_data *bd)
1653{
1654 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1655 struct loop_device *lo = cmd->rq->q->queuedata;
1656
1657 blk_mq_start_request(bd->rq);
1658
1659 if (lo->lo_state != Lo_bound)
1660 return -EIO;
1661
1662 if (lo->use_dio && !(cmd->rq->cmd_flags & (REQ_FLUSH |
1663 REQ_DISCARD)))
1664 cmd->use_aio = true;
1665 else
1666 cmd->use_aio = false;
1667
1668 queue_kthread_work(&lo->worker, &cmd->work);
1669
1670 return BLK_MQ_RQ_QUEUE_OK;
1671}
1672
1673static void loop_handle_cmd(struct loop_cmd *cmd)
1674{
1675 const bool write = cmd->rq->cmd_flags & REQ_WRITE;
1676 struct loop_device *lo = cmd->rq->q->queuedata;
1677 int ret = 0;
1678
1679 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1680 ret = -EIO;
1681 goto failed;
1682 }
1683
1684 ret = do_req_filebacked(lo, cmd->rq);
1685 failed:
1686 /* complete non-aio request */
1687 if (!cmd->use_aio || ret)
1688 blk_mq_complete_request(cmd->rq, ret ? -EIO : 0);
1689}
1690
1691static void loop_queue_work(struct kthread_work *work)
1692{
1693 struct loop_cmd *cmd =
1694 container_of(work, struct loop_cmd, work);
1695
1696 loop_handle_cmd(cmd);
1697}
1698
1699static int loop_init_request(void *data, struct request *rq,
1700 unsigned int hctx_idx, unsigned int request_idx,
1701 unsigned int numa_node)
1702{
1703 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1704
1705 cmd->rq = rq;
1706 init_kthread_work(&cmd->work, loop_queue_work);
1707
1708 return 0;
1709}
1710
1711static struct blk_mq_ops loop_mq_ops = {
1712 .queue_rq = loop_queue_rq,
1713 .map_queue = blk_mq_map_queue,
1714 .init_request = loop_init_request,
1715};
1716
1717static int loop_add(struct loop_device **l, int i)
1718{
1719 struct loop_device *lo;
1720 struct gendisk *disk;
1721 int err;
1722
1723 err = -ENOMEM;
1724 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1725 if (!lo)
1726 goto out;
1727
1728 lo->lo_state = Lo_unbound;
1729
1730 /* allocate id, if @id >= 0, we're requesting that specific id */
1731 if (i >= 0) {
1732 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1733 if (err == -ENOSPC)
1734 err = -EEXIST;
1735 } else {
1736 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1737 }
1738 if (err < 0)
1739 goto out_free_dev;
1740 i = err;
1741
1742 err = -ENOMEM;
1743 lo->tag_set.ops = &loop_mq_ops;
1744 lo->tag_set.nr_hw_queues = 1;
1745 lo->tag_set.queue_depth = 128;
1746 lo->tag_set.numa_node = NUMA_NO_NODE;
1747 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1748 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1749 lo->tag_set.driver_data = lo;
1750
1751 err = blk_mq_alloc_tag_set(&lo->tag_set);
1752 if (err)
1753 goto out_free_idr;
1754
1755 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1756 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1757 err = PTR_ERR(lo->lo_queue);
1758 goto out_cleanup_tags;
1759 }
1760 lo->lo_queue->queuedata = lo;
1761
1762 /*
1763 * It doesn't make sense to enable merge because the I/O
1764 * submitted to backing file is handled page by page.
1765 */
1766 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1767
1768 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1769 if (!disk)
1770 goto out_free_queue;
1771
1772 /*
1773 * Disable partition scanning by default. The in-kernel partition
1774 * scanning can be requested individually per-device during its
1775 * setup. Userspace can always add and remove partitions from all
1776 * devices. The needed partition minors are allocated from the
1777 * extended minor space, the main loop device numbers will continue
1778 * to match the loop minors, regardless of the number of partitions
1779 * used.
1780 *
1781 * If max_part is given, partition scanning is globally enabled for
1782 * all loop devices. The minors for the main loop devices will be
1783 * multiples of max_part.
1784 *
1785 * Note: Global-for-all-devices, set-only-at-init, read-only module
1786 * parameteters like 'max_loop' and 'max_part' make things needlessly
1787 * complicated, are too static, inflexible and may surprise
1788 * userspace tools. Parameters like this in general should be avoided.
1789 */
1790 if (!part_shift)
1791 disk->flags |= GENHD_FL_NO_PART_SCAN;
1792 disk->flags |= GENHD_FL_EXT_DEVT;
1793 mutex_init(&lo->lo_ctl_mutex);
1794 atomic_set(&lo->lo_refcnt, 0);
1795 lo->lo_number = i;
1796 spin_lock_init(&lo->lo_lock);
1797 disk->major = LOOP_MAJOR;
1798 disk->first_minor = i << part_shift;
1799 disk->fops = &lo_fops;
1800 disk->private_data = lo;
1801 disk->queue = lo->lo_queue;
1802 sprintf(disk->disk_name, "loop%d", i);
1803 add_disk(disk);
1804 *l = lo;
1805 return lo->lo_number;
1806
1807out_free_queue:
1808 blk_cleanup_queue(lo->lo_queue);
1809out_cleanup_tags:
1810 blk_mq_free_tag_set(&lo->tag_set);
1811out_free_idr:
1812 idr_remove(&loop_index_idr, i);
1813out_free_dev:
1814 kfree(lo);
1815out:
1816 return err;
1817}
1818
1819static void loop_remove(struct loop_device *lo)
1820{
1821 blk_cleanup_queue(lo->lo_queue);
1822 del_gendisk(lo->lo_disk);
1823 blk_mq_free_tag_set(&lo->tag_set);
1824 put_disk(lo->lo_disk);
1825 kfree(lo);
1826}
1827
1828static int find_free_cb(int id, void *ptr, void *data)
1829{
1830 struct loop_device *lo = ptr;
1831 struct loop_device **l = data;
1832
1833 if (lo->lo_state == Lo_unbound) {
1834 *l = lo;
1835 return 1;
1836 }
1837 return 0;
1838}
1839
1840static int loop_lookup(struct loop_device **l, int i)
1841{
1842 struct loop_device *lo;
1843 int ret = -ENODEV;
1844
1845 if (i < 0) {
1846 int err;
1847
1848 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1849 if (err == 1) {
1850 *l = lo;
1851 ret = lo->lo_number;
1852 }
1853 goto out;
1854 }
1855
1856 /* lookup and return a specific i */
1857 lo = idr_find(&loop_index_idr, i);
1858 if (lo) {
1859 *l = lo;
1860 ret = lo->lo_number;
1861 }
1862out:
1863 return ret;
1864}
1865
1866static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1867{
1868 struct loop_device *lo;
1869 struct kobject *kobj;
1870 int err;
1871
1872 mutex_lock(&loop_index_mutex);
1873 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1874 if (err < 0)
1875 err = loop_add(&lo, MINOR(dev) >> part_shift);
1876 if (err < 0)
1877 kobj = NULL;
1878 else
1879 kobj = get_disk(lo->lo_disk);
1880 mutex_unlock(&loop_index_mutex);
1881
1882 *part = 0;
1883 return kobj;
1884}
1885
1886static long loop_control_ioctl(struct file *file, unsigned int cmd,
1887 unsigned long parm)
1888{
1889 struct loop_device *lo;
1890 int ret = -ENOSYS;
1891
1892 mutex_lock(&loop_index_mutex);
1893 switch (cmd) {
1894 case LOOP_CTL_ADD:
1895 ret = loop_lookup(&lo, parm);
1896 if (ret >= 0) {
1897 ret = -EEXIST;
1898 break;
1899 }
1900 ret = loop_add(&lo, parm);
1901 break;
1902 case LOOP_CTL_REMOVE:
1903 ret = loop_lookup(&lo, parm);
1904 if (ret < 0)
1905 break;
1906 mutex_lock(&lo->lo_ctl_mutex);
1907 if (lo->lo_state != Lo_unbound) {
1908 ret = -EBUSY;
1909 mutex_unlock(&lo->lo_ctl_mutex);
1910 break;
1911 }
1912 if (atomic_read(&lo->lo_refcnt) > 0) {
1913 ret = -EBUSY;
1914 mutex_unlock(&lo->lo_ctl_mutex);
1915 break;
1916 }
1917 lo->lo_disk->private_data = NULL;
1918 mutex_unlock(&lo->lo_ctl_mutex);
1919 idr_remove(&loop_index_idr, lo->lo_number);
1920 loop_remove(lo);
1921 break;
1922 case LOOP_CTL_GET_FREE:
1923 ret = loop_lookup(&lo, -1);
1924 if (ret >= 0)
1925 break;
1926 ret = loop_add(&lo, -1);
1927 }
1928 mutex_unlock(&loop_index_mutex);
1929
1930 return ret;
1931}
1932
1933static const struct file_operations loop_ctl_fops = {
1934 .open = nonseekable_open,
1935 .unlocked_ioctl = loop_control_ioctl,
1936 .compat_ioctl = loop_control_ioctl,
1937 .owner = THIS_MODULE,
1938 .llseek = noop_llseek,
1939};
1940
1941static struct miscdevice loop_misc = {
1942 .minor = LOOP_CTRL_MINOR,
1943 .name = "loop-control",
1944 .fops = &loop_ctl_fops,
1945};
1946
1947MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1948MODULE_ALIAS("devname:loop-control");
1949
1950static int __init loop_init(void)
1951{
1952 int i, nr;
1953 unsigned long range;
1954 struct loop_device *lo;
1955 int err;
1956
1957 err = misc_register(&loop_misc);
1958 if (err < 0)
1959 return err;
1960
1961 part_shift = 0;
1962 if (max_part > 0) {
1963 part_shift = fls(max_part);
1964
1965 /*
1966 * Adjust max_part according to part_shift as it is exported
1967 * to user space so that user can decide correct minor number
1968 * if [s]he want to create more devices.
1969 *
1970 * Note that -1 is required because partition 0 is reserved
1971 * for the whole disk.
1972 */
1973 max_part = (1UL << part_shift) - 1;
1974 }
1975
1976 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1977 err = -EINVAL;
1978 goto misc_out;
1979 }
1980
1981 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1982 err = -EINVAL;
1983 goto misc_out;
1984 }
1985
1986 /*
1987 * If max_loop is specified, create that many devices upfront.
1988 * This also becomes a hard limit. If max_loop is not specified,
1989 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1990 * init time. Loop devices can be requested on-demand with the
1991 * /dev/loop-control interface, or be instantiated by accessing
1992 * a 'dead' device node.
1993 */
1994 if (max_loop) {
1995 nr = max_loop;
1996 range = max_loop << part_shift;
1997 } else {
1998 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1999 range = 1UL << MINORBITS;
2000 }
2001
2002 if (register_blkdev(LOOP_MAJOR, "loop")) {
2003 err = -EIO;
2004 goto misc_out;
2005 }
2006
2007 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2008 THIS_MODULE, loop_probe, NULL, NULL);
2009
2010 /* pre-create number of devices given by config or max_loop */
2011 mutex_lock(&loop_index_mutex);
2012 for (i = 0; i < nr; i++)
2013 loop_add(&lo, i);
2014 mutex_unlock(&loop_index_mutex);
2015
2016 printk(KERN_INFO "loop: module loaded\n");
2017 return 0;
2018
2019misc_out:
2020 misc_deregister(&loop_misc);
2021 return err;
2022}
2023
2024static int loop_exit_cb(int id, void *ptr, void *data)
2025{
2026 struct loop_device *lo = ptr;
2027
2028 loop_remove(lo);
2029 return 0;
2030}
2031
2032static void __exit loop_exit(void)
2033{
2034 unsigned long range;
2035
2036 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2037
2038 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2039 idr_destroy(&loop_index_idr);
2040
2041 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2042 unregister_blkdev(LOOP_MAJOR, "loop");
2043
2044 misc_deregister(&loop_misc);
2045}
2046
2047module_init(loop_init);
2048module_exit(loop_exit);
2049
2050#ifndef MODULE
2051static int __init max_loop_setup(char *str)
2052{
2053 max_loop = simple_strtol(str, NULL, 0);
2054 return 1;
2055}
2056
2057__setup("max_loop=", max_loop_setup);
2058#endif