projects/linux·blob v5.2 · refs

2313 lines · 56.4 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 <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/* xfer_funcs[0] is special - its release function is never called */
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 /* Compute loopsize in bytes */
149 loopsize = i_size_read(file->f_mapping->host);
150 if (offset > 0)
151 loopsize -= offset;
152 /* offset is beyond i_size, weird but possible */
153 if (loopsize < 0)
154 return 0;
155
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
158 /*
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
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 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
189 *
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
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 /* flush dirty pages before changing direct IO */
210 vfs_fsync(file, 0);
211
212 /*
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
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 /* let user-space know about the new size */
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 inline void loop_iov_iter_bvec(struct iov_iter *i,
268 unsigned int direction, const struct bio_vec *bvec,
269 unsigned long nr_segs, size_t count)
270{
271 iov_iter_bvec(i, direction, bvec, nr_segs, count);
272 i->type |= ITER_BVEC_FLAG_NO_REF;
273}
274
275static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
276{
277 struct iov_iter i;
278 ssize_t bw;
279
280 loop_iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
281
282 file_start_write(file);
283 bw = vfs_iter_write(file, &i, ppos, 0);
284 file_end_write(file);
285
286 if (likely(bw == bvec->bv_len))
287 return 0;
288
289 printk_ratelimited(KERN_ERR
290 "loop: Write error at byte offset %llu, length %i.\n",
291 (unsigned long long)*ppos, bvec->bv_len);
292 if (bw >= 0)
293 bw = -EIO;
294 return bw;
295}
296
297static int lo_write_simple(struct loop_device *lo, struct request *rq,
298 loff_t pos)
299{
300 struct bio_vec bvec;
301 struct req_iterator iter;
302 int ret = 0;
303
304 rq_for_each_segment(bvec, rq, iter) {
305 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
306 if (ret < 0)
307 break;
308 cond_resched();
309 }
310
311 return ret;
312}
313
314/*
315 * This is the slow, transforming version that needs to double buffer the
316 * data as it cannot do the transformations in place without having direct
317 * access to the destination pages of the backing file.
318 */
319static int lo_write_transfer(struct loop_device *lo, struct request *rq,
320 loff_t pos)
321{
322 struct bio_vec bvec, b;
323 struct req_iterator iter;
324 struct page *page;
325 int ret = 0;
326
327 page = alloc_page(GFP_NOIO);
328 if (unlikely(!page))
329 return -ENOMEM;
330
331 rq_for_each_segment(bvec, rq, iter) {
332 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
333 bvec.bv_offset, bvec.bv_len, pos >> 9);
334 if (unlikely(ret))
335 break;
336
337 b.bv_page = page;
338 b.bv_offset = 0;
339 b.bv_len = bvec.bv_len;
340 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
341 if (ret < 0)
342 break;
343 }
344
345 __free_page(page);
346 return ret;
347}
348
349static int lo_read_simple(struct loop_device *lo, struct request *rq,
350 loff_t pos)
351{
352 struct bio_vec bvec;
353 struct req_iterator iter;
354 struct iov_iter i;
355 ssize_t len;
356
357 rq_for_each_segment(bvec, rq, iter) {
358 loop_iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
359 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
360 if (len < 0)
361 return len;
362
363 flush_dcache_page(bvec.bv_page);
364
365 if (len != bvec.bv_len) {
366 struct bio *bio;
367
368 __rq_for_each_bio(bio, rq)
369 zero_fill_bio(bio);
370 break;
371 }
372 cond_resched();
373 }
374
375 return 0;
376}
377
378static int lo_read_transfer(struct loop_device *lo, struct request *rq,
379 loff_t pos)
380{
381 struct bio_vec bvec, b;
382 struct req_iterator iter;
383 struct iov_iter i;
384 struct page *page;
385 ssize_t len;
386 int ret = 0;
387
388 page = alloc_page(GFP_NOIO);
389 if (unlikely(!page))
390 return -ENOMEM;
391
392 rq_for_each_segment(bvec, rq, iter) {
393 loff_t offset = pos;
394
395 b.bv_page = page;
396 b.bv_offset = 0;
397 b.bv_len = bvec.bv_len;
398
399 loop_iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
400 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
401 if (len < 0) {
402 ret = len;
403 goto out_free_page;
404 }
405
406 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
407 bvec.bv_offset, len, offset >> 9);
408 if (ret)
409 goto out_free_page;
410
411 flush_dcache_page(bvec.bv_page);
412
413 if (len != bvec.bv_len) {
414 struct bio *bio;
415
416 __rq_for_each_bio(bio, rq)
417 zero_fill_bio(bio);
418 break;
419 }
420 }
421
422 ret = 0;
423out_free_page:
424 __free_page(page);
425 return ret;
426}
427
428static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
429{
430 /*
431 * We use punch hole to reclaim the free space used by the
432 * image a.k.a. discard. However we do not support discard if
433 * encryption is enabled, because it may give an attacker
434 * useful information.
435 */
436 struct file *file = lo->lo_backing_file;
437 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
438 int ret;
439
440 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
441 ret = -EOPNOTSUPP;
442 goto out;
443 }
444
445 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
446 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
447 ret = -EIO;
448 out:
449 return ret;
450}
451
452static int lo_req_flush(struct loop_device *lo, struct request *rq)
453{
454 struct file *file = lo->lo_backing_file;
455 int ret = vfs_fsync(file, 0);
456 if (unlikely(ret && ret != -EINVAL))
457 ret = -EIO;
458
459 return ret;
460}
461
462static void lo_complete_rq(struct request *rq)
463{
464 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
465 blk_status_t ret = BLK_STS_OK;
466
467 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
468 req_op(rq) != REQ_OP_READ) {
469 if (cmd->ret < 0)
470 ret = BLK_STS_IOERR;
471 goto end_io;
472 }
473
474 /*
475 * Short READ - if we got some data, advance our request and
476 * retry it. If we got no data, end the rest with EIO.
477 */
478 if (cmd->ret) {
479 blk_update_request(rq, BLK_STS_OK, cmd->ret);
480 cmd->ret = 0;
481 blk_mq_requeue_request(rq, true);
482 } else {
483 if (cmd->use_aio) {
484 struct bio *bio = rq->bio;
485
486 while (bio) {
487 zero_fill_bio(bio);
488 bio = bio->bi_next;
489 }
490 }
491 ret = BLK_STS_IOERR;
492end_io:
493 blk_mq_end_request(rq, ret);
494 }
495}
496
497static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
498{
499 struct request *rq = blk_mq_rq_from_pdu(cmd);
500
501 if (!atomic_dec_and_test(&cmd->ref))
502 return;
503 kfree(cmd->bvec);
504 cmd->bvec = NULL;
505 blk_mq_complete_request(rq);
506}
507
508static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
509{
510 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
511
512 if (cmd->css)
513 css_put(cmd->css);
514 cmd->ret = ret;
515 lo_rw_aio_do_completion(cmd);
516}
517
518static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
519 loff_t pos, bool rw)
520{
521 struct iov_iter iter;
522 struct req_iterator rq_iter;
523 struct bio_vec *bvec;
524 struct request *rq = blk_mq_rq_from_pdu(cmd);
525 struct bio *bio = rq->bio;
526 struct file *file = lo->lo_backing_file;
527 struct bio_vec tmp;
528 unsigned int offset;
529 int nr_bvec = 0;
530 int ret;
531
532 rq_for_each_bvec(tmp, rq, rq_iter)
533 nr_bvec++;
534
535 if (rq->bio != rq->biotail) {
536
537 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
538 GFP_NOIO);
539 if (!bvec)
540 return -EIO;
541 cmd->bvec = bvec;
542
543 /*
544 * The bios of the request may be started from the middle of
545 * the 'bvec' because of bio splitting, so we can't directly
546 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
547 * API will take care of all details for us.
548 */
549 rq_for_each_bvec(tmp, rq, rq_iter) {
550 *bvec = tmp;
551 bvec++;
552 }
553 bvec = cmd->bvec;
554 offset = 0;
555 } else {
556 /*
557 * Same here, this bio may be started from the middle of the
558 * 'bvec' because of bio splitting, so offset from the bvec
559 * must be passed to iov iterator
560 */
561 offset = bio->bi_iter.bi_bvec_done;
562 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
563 }
564 atomic_set(&cmd->ref, 2);
565
566 loop_iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
567 iter.iov_offset = offset;
568
569 cmd->iocb.ki_pos = pos;
570 cmd->iocb.ki_filp = file;
571 cmd->iocb.ki_complete = lo_rw_aio_complete;
572 cmd->iocb.ki_flags = IOCB_DIRECT;
573 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
574 if (cmd->css)
575 kthread_associate_blkcg(cmd->css);
576
577 if (rw == WRITE)
578 ret = call_write_iter(file, &cmd->iocb, &iter);
579 else
580 ret = call_read_iter(file, &cmd->iocb, &iter);
581
582 lo_rw_aio_do_completion(cmd);
583 kthread_associate_blkcg(NULL);
584
585 if (ret != -EIOCBQUEUED)
586 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
587 return 0;
588}
589
590static int do_req_filebacked(struct loop_device *lo, struct request *rq)
591{
592 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
593 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
594
595 /*
596 * lo_write_simple and lo_read_simple should have been covered
597 * by io submit style function like lo_rw_aio(), one blocker
598 * is that lo_read_simple() need to call flush_dcache_page after
599 * the page is written from kernel, and it isn't easy to handle
600 * this in io submit style function which submits all segments
601 * of the req at one time. And direct read IO doesn't need to
602 * run flush_dcache_page().
603 */
604 switch (req_op(rq)) {
605 case REQ_OP_FLUSH:
606 return lo_req_flush(lo, rq);
607 case REQ_OP_DISCARD:
608 case REQ_OP_WRITE_ZEROES:
609 return lo_discard(lo, rq, pos);
610 case REQ_OP_WRITE:
611 if (lo->transfer)
612 return lo_write_transfer(lo, rq, pos);
613 else if (cmd->use_aio)
614 return lo_rw_aio(lo, cmd, pos, WRITE);
615 else
616 return lo_write_simple(lo, rq, pos);
617 case REQ_OP_READ:
618 if (lo->transfer)
619 return lo_read_transfer(lo, rq, pos);
620 else if (cmd->use_aio)
621 return lo_rw_aio(lo, cmd, pos, READ);
622 else
623 return lo_read_simple(lo, rq, pos);
624 default:
625 WARN_ON_ONCE(1);
626 return -EIO;
627 }
628}
629
630static inline void loop_update_dio(struct loop_device *lo)
631{
632 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
633 lo->use_dio);
634}
635
636static void loop_reread_partitions(struct loop_device *lo,
637 struct block_device *bdev)
638{
639 int rc;
640
641 rc = blkdev_reread_part(bdev);
642 if (rc)
643 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
644 __func__, lo->lo_number, lo->lo_file_name, rc);
645}
646
647static inline int is_loop_device(struct file *file)
648{
649 struct inode *i = file->f_mapping->host;
650
651 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
652}
653
654static int loop_validate_file(struct file *file, struct block_device *bdev)
655{
656 struct inode *inode = file->f_mapping->host;
657 struct file *f = file;
658
659 /* Avoid recursion */
660 while (is_loop_device(f)) {
661 struct loop_device *l;
662
663 if (f->f_mapping->host->i_bdev == bdev)
664 return -EBADF;
665
666 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
667 if (l->lo_state != Lo_bound) {
668 return -EINVAL;
669 }
670 f = l->lo_backing_file;
671 }
672 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
673 return -EINVAL;
674 return 0;
675}
676
677/*
678 * loop_change_fd switched the backing store of a loopback device to
679 * a new file. This is useful for operating system installers to free up
680 * the original file and in High Availability environments to switch to
681 * an alternative location for the content in case of server meltdown.
682 * This can only work if the loop device is used read-only, and if the
683 * new backing store is the same size and type as the old backing store.
684 */
685static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
686 unsigned int arg)
687{
688 struct file *file = NULL, *old_file;
689 int error;
690 bool partscan;
691
692 error = mutex_lock_killable(&loop_ctl_mutex);
693 if (error)
694 return error;
695 error = -ENXIO;
696 if (lo->lo_state != Lo_bound)
697 goto out_err;
698
699 /* the loop device has to be read-only */
700 error = -EINVAL;
701 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
702 goto out_err;
703
704 error = -EBADF;
705 file = fget(arg);
706 if (!file)
707 goto out_err;
708
709 error = loop_validate_file(file, bdev);
710 if (error)
711 goto out_err;
712
713 old_file = lo->lo_backing_file;
714
715 error = -EINVAL;
716
717 /* size of the new backing store needs to be the same */
718 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
719 goto out_err;
720
721 /* and ... switch */
722 blk_mq_freeze_queue(lo->lo_queue);
723 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
724 lo->lo_backing_file = file;
725 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
726 mapping_set_gfp_mask(file->f_mapping,
727 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
728 loop_update_dio(lo);
729 blk_mq_unfreeze_queue(lo->lo_queue);
730 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
731 mutex_unlock(&loop_ctl_mutex);
732 /*
733 * We must drop file reference outside of loop_ctl_mutex as dropping
734 * the file ref can take bd_mutex which creates circular locking
735 * dependency.
736 */
737 fput(old_file);
738 if (partscan)
739 loop_reread_partitions(lo, bdev);
740 return 0;
741
742out_err:
743 mutex_unlock(&loop_ctl_mutex);
744 if (file)
745 fput(file);
746 return error;
747}
748
749/* loop sysfs attributes */
750
751static ssize_t loop_attr_show(struct device *dev, char *page,
752 ssize_t (*callback)(struct loop_device *, char *))
753{
754 struct gendisk *disk = dev_to_disk(dev);
755 struct loop_device *lo = disk->private_data;
756
757 return callback(lo, page);
758}
759
760#define LOOP_ATTR_RO(_name) \
761static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
762static ssize_t loop_attr_do_show_##_name(struct device *d, \
763 struct device_attribute *attr, char *b) \
764{ \
765 return loop_attr_show(d, b, loop_attr_##_name##_show); \
766} \
767static struct device_attribute loop_attr_##_name = \
768 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
769
770static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
771{
772 ssize_t ret;
773 char *p = NULL;
774
775 spin_lock_irq(&lo->lo_lock);
776 if (lo->lo_backing_file)
777 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
778 spin_unlock_irq(&lo->lo_lock);
779
780 if (IS_ERR_OR_NULL(p))
781 ret = PTR_ERR(p);
782 else {
783 ret = strlen(p);
784 memmove(buf, p, ret);
785 buf[ret++] = '\n';
786 buf[ret] = 0;
787 }
788
789 return ret;
790}
791
792static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
793{
794 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
795}
796
797static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
798{
799 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
800}
801
802static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
803{
804 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
805
806 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
807}
808
809static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
810{
811 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
812
813 return sprintf(buf, "%s\n", partscan ? "1" : "0");
814}
815
816static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
817{
818 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
819
820 return sprintf(buf, "%s\n", dio ? "1" : "0");
821}
822
823LOOP_ATTR_RO(backing_file);
824LOOP_ATTR_RO(offset);
825LOOP_ATTR_RO(sizelimit);
826LOOP_ATTR_RO(autoclear);
827LOOP_ATTR_RO(partscan);
828LOOP_ATTR_RO(dio);
829
830static struct attribute *loop_attrs[] = {
831 &loop_attr_backing_file.attr,
832 &loop_attr_offset.attr,
833 &loop_attr_sizelimit.attr,
834 &loop_attr_autoclear.attr,
835 &loop_attr_partscan.attr,
836 &loop_attr_dio.attr,
837 NULL,
838};
839
840static struct attribute_group loop_attribute_group = {
841 .name = "loop",
842 .attrs= loop_attrs,
843};
844
845static void loop_sysfs_init(struct loop_device *lo)
846{
847 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
848 &loop_attribute_group);
849}
850
851static void loop_sysfs_exit(struct loop_device *lo)
852{
853 if (lo->sysfs_inited)
854 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
855 &loop_attribute_group);
856}
857
858static void loop_config_discard(struct loop_device *lo)
859{
860 struct file *file = lo->lo_backing_file;
861 struct inode *inode = file->f_mapping->host;
862 struct request_queue *q = lo->lo_queue;
863
864 /*
865 * We use punch hole to reclaim the free space used by the
866 * image a.k.a. discard. However we do not support discard if
867 * encryption is enabled, because it may give an attacker
868 * useful information.
869 */
870 if ((!file->f_op->fallocate) ||
871 lo->lo_encrypt_key_size) {
872 q->limits.discard_granularity = 0;
873 q->limits.discard_alignment = 0;
874 blk_queue_max_discard_sectors(q, 0);
875 blk_queue_max_write_zeroes_sectors(q, 0);
876 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
877 return;
878 }
879
880 q->limits.discard_granularity = inode->i_sb->s_blocksize;
881 q->limits.discard_alignment = 0;
882
883 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
884 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
885 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
886}
887
888static void loop_unprepare_queue(struct loop_device *lo)
889{
890 kthread_flush_worker(&lo->worker);
891 kthread_stop(lo->worker_task);
892}
893
894static int loop_kthread_worker_fn(void *worker_ptr)
895{
896 current->flags |= PF_LESS_THROTTLE;
897 return kthread_worker_fn(worker_ptr);
898}
899
900static int loop_prepare_queue(struct loop_device *lo)
901{
902 kthread_init_worker(&lo->worker);
903 lo->worker_task = kthread_run(loop_kthread_worker_fn,
904 &lo->worker, "loop%d", lo->lo_number);
905 if (IS_ERR(lo->worker_task))
906 return -ENOMEM;
907 set_user_nice(lo->worker_task, MIN_NICE);
908 return 0;
909}
910
911static void loop_update_rotational(struct loop_device *lo)
912{
913 struct file *file = lo->lo_backing_file;
914 struct inode *file_inode = file->f_mapping->host;
915 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
916 struct request_queue *q = lo->lo_queue;
917 bool nonrot = true;
918
919 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
920 if (file_bdev)
921 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
922
923 if (nonrot)
924 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
925 else
926 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
927}
928
929static int loop_set_fd(struct loop_device *lo, fmode_t mode,
930 struct block_device *bdev, unsigned int arg)
931{
932 struct file *file;
933 struct inode *inode;
934 struct address_space *mapping;
935 int lo_flags = 0;
936 int error;
937 loff_t size;
938 bool partscan;
939
940 /* This is safe, since we have a reference from open(). */
941 __module_get(THIS_MODULE);
942
943 error = -EBADF;
944 file = fget(arg);
945 if (!file)
946 goto out;
947
948 /*
949 * If we don't hold exclusive handle for the device, upgrade to it
950 * here to avoid changing device under exclusive owner.
951 */
952 if (!(mode & FMODE_EXCL)) {
953 bdgrab(bdev);
954 error = blkdev_get(bdev, mode | FMODE_EXCL, loop_set_fd);
955 if (error)
956 goto out_putf;
957 }
958
959 error = mutex_lock_killable(&loop_ctl_mutex);
960 if (error)
961 goto out_bdev;
962
963 error = -EBUSY;
964 if (lo->lo_state != Lo_unbound)
965 goto out_unlock;
966
967 error = loop_validate_file(file, bdev);
968 if (error)
969 goto out_unlock;
970
971 mapping = file->f_mapping;
972 inode = mapping->host;
973
974 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
975 !file->f_op->write_iter)
976 lo_flags |= LO_FLAGS_READ_ONLY;
977
978 error = -EFBIG;
979 size = get_loop_size(lo, file);
980 if ((loff_t)(sector_t)size != size)
981 goto out_unlock;
982 error = loop_prepare_queue(lo);
983 if (error)
984 goto out_unlock;
985
986 error = 0;
987
988 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
989
990 lo->use_dio = false;
991 lo->lo_device = bdev;
992 lo->lo_flags = lo_flags;
993 lo->lo_backing_file = file;
994 lo->transfer = NULL;
995 lo->ioctl = NULL;
996 lo->lo_sizelimit = 0;
997 lo->old_gfp_mask = mapping_gfp_mask(mapping);
998 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
999
1000 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1001 blk_queue_write_cache(lo->lo_queue, true, false);
1002
1003 loop_update_rotational(lo);
1004 loop_update_dio(lo);
1005 set_capacity(lo->lo_disk, size);
1006 bd_set_size(bdev, size << 9);
1007 loop_sysfs_init(lo);
1008 /* let user-space know about the new size */
1009 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1010
1011 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1012 block_size(inode->i_bdev) : PAGE_SIZE);
1013
1014 lo->lo_state = Lo_bound;
1015 if (part_shift)
1016 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1017 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1018
1019 /* Grab the block_device to prevent its destruction after we
1020 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1021 */
1022 bdgrab(bdev);
1023 mutex_unlock(&loop_ctl_mutex);
1024 if (partscan)
1025 loop_reread_partitions(lo, bdev);
1026 if (!(mode & FMODE_EXCL))
1027 blkdev_put(bdev, mode | FMODE_EXCL);
1028 return 0;
1029
1030out_unlock:
1031 mutex_unlock(&loop_ctl_mutex);
1032out_bdev:
1033 if (!(mode & FMODE_EXCL))
1034 blkdev_put(bdev, mode | FMODE_EXCL);
1035out_putf:
1036 fput(file);
1037out:
1038 /* This is safe: open() is still holding a reference. */
1039 module_put(THIS_MODULE);
1040 return error;
1041}
1042
1043static int
1044loop_release_xfer(struct loop_device *lo)
1045{
1046 int err = 0;
1047 struct loop_func_table *xfer = lo->lo_encryption;
1048
1049 if (xfer) {
1050 if (xfer->release)
1051 err = xfer->release(lo);
1052 lo->transfer = NULL;
1053 lo->lo_encryption = NULL;
1054 module_put(xfer->owner);
1055 }
1056 return err;
1057}
1058
1059static int
1060loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1061 const struct loop_info64 *i)
1062{
1063 int err = 0;
1064
1065 if (xfer) {
1066 struct module *owner = xfer->owner;
1067
1068 if (!try_module_get(owner))
1069 return -EINVAL;
1070 if (xfer->init)
1071 err = xfer->init(lo, i);
1072 if (err)
1073 module_put(owner);
1074 else
1075 lo->lo_encryption = xfer;
1076 }
1077 return err;
1078}
1079
1080static int __loop_clr_fd(struct loop_device *lo, bool release)
1081{
1082 struct file *filp = NULL;
1083 gfp_t gfp = lo->old_gfp_mask;
1084 struct block_device *bdev = lo->lo_device;
1085 int err = 0;
1086 bool partscan = false;
1087 int lo_number;
1088
1089 mutex_lock(&loop_ctl_mutex);
1090 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1091 err = -ENXIO;
1092 goto out_unlock;
1093 }
1094
1095 filp = lo->lo_backing_file;
1096 if (filp == NULL) {
1097 err = -EINVAL;
1098 goto out_unlock;
1099 }
1100
1101 /* freeze request queue during the transition */
1102 blk_mq_freeze_queue(lo->lo_queue);
1103
1104 spin_lock_irq(&lo->lo_lock);
1105 lo->lo_backing_file = NULL;
1106 spin_unlock_irq(&lo->lo_lock);
1107
1108 loop_release_xfer(lo);
1109 lo->transfer = NULL;
1110 lo->ioctl = NULL;
1111 lo->lo_device = NULL;
1112 lo->lo_encryption = NULL;
1113 lo->lo_offset = 0;
1114 lo->lo_sizelimit = 0;
1115 lo->lo_encrypt_key_size = 0;
1116 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1117 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1118 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1119 blk_queue_logical_block_size(lo->lo_queue, 512);
1120 blk_queue_physical_block_size(lo->lo_queue, 512);
1121 blk_queue_io_min(lo->lo_queue, 512);
1122 if (bdev) {
1123 bdput(bdev);
1124 invalidate_bdev(bdev);
1125 bdev->bd_inode->i_mapping->wb_err = 0;
1126 }
1127 set_capacity(lo->lo_disk, 0);
1128 loop_sysfs_exit(lo);
1129 if (bdev) {
1130 bd_set_size(bdev, 0);
1131 /* let user-space know about this change */
1132 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1133 }
1134 mapping_set_gfp_mask(filp->f_mapping, gfp);
1135 /* This is safe: open() is still holding a reference. */
1136 module_put(THIS_MODULE);
1137 blk_mq_unfreeze_queue(lo->lo_queue);
1138
1139 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1140 lo_number = lo->lo_number;
1141 loop_unprepare_queue(lo);
1142out_unlock:
1143 mutex_unlock(&loop_ctl_mutex);
1144 if (partscan) {
1145 /*
1146 * bd_mutex has been held already in release path, so don't
1147 * acquire it if this function is called in such case.
1148 *
1149 * If the reread partition isn't from release path, lo_refcnt
1150 * must be at least one and it can only become zero when the
1151 * current holder is released.
1152 */
1153 if (release)
1154 err = __blkdev_reread_part(bdev);
1155 else
1156 err = blkdev_reread_part(bdev);
1157 if (err)
1158 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1159 __func__, lo_number, err);
1160 /* Device is gone, no point in returning error */
1161 err = 0;
1162 }
1163
1164 /*
1165 * lo->lo_state is set to Lo_unbound here after above partscan has
1166 * finished.
1167 *
1168 * There cannot be anybody else entering __loop_clr_fd() as
1169 * lo->lo_backing_file is already cleared and Lo_rundown state
1170 * protects us from all the other places trying to change the 'lo'
1171 * device.
1172 */
1173 mutex_lock(&loop_ctl_mutex);
1174 lo->lo_flags = 0;
1175 if (!part_shift)
1176 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1177 lo->lo_state = Lo_unbound;
1178 mutex_unlock(&loop_ctl_mutex);
1179
1180 /*
1181 * Need not hold loop_ctl_mutex to fput backing file.
1182 * Calling fput holding loop_ctl_mutex triggers a circular
1183 * lock dependency possibility warning as fput can take
1184 * bd_mutex which is usually taken before loop_ctl_mutex.
1185 */
1186 if (filp)
1187 fput(filp);
1188 return err;
1189}
1190
1191static int loop_clr_fd(struct loop_device *lo)
1192{
1193 int err;
1194
1195 err = mutex_lock_killable(&loop_ctl_mutex);
1196 if (err)
1197 return err;
1198 if (lo->lo_state != Lo_bound) {
1199 mutex_unlock(&loop_ctl_mutex);
1200 return -ENXIO;
1201 }
1202 /*
1203 * If we've explicitly asked to tear down the loop device,
1204 * and it has an elevated reference count, set it for auto-teardown when
1205 * the last reference goes away. This stops $!~#$@ udev from
1206 * preventing teardown because it decided that it needs to run blkid on
1207 * the loopback device whenever they appear. xfstests is notorious for
1208 * failing tests because blkid via udev races with a losetup
1209 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1210 * command to fail with EBUSY.
1211 */
1212 if (atomic_read(&lo->lo_refcnt) > 1) {
1213 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1214 mutex_unlock(&loop_ctl_mutex);
1215 return 0;
1216 }
1217 lo->lo_state = Lo_rundown;
1218 mutex_unlock(&loop_ctl_mutex);
1219
1220 return __loop_clr_fd(lo, false);
1221}
1222
1223static int
1224loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1225{
1226 int err;
1227 struct loop_func_table *xfer;
1228 kuid_t uid = current_uid();
1229 struct block_device *bdev;
1230 bool partscan = false;
1231
1232 err = mutex_lock_killable(&loop_ctl_mutex);
1233 if (err)
1234 return err;
1235 if (lo->lo_encrypt_key_size &&
1236 !uid_eq(lo->lo_key_owner, uid) &&
1237 !capable(CAP_SYS_ADMIN)) {
1238 err = -EPERM;
1239 goto out_unlock;
1240 }
1241 if (lo->lo_state != Lo_bound) {
1242 err = -ENXIO;
1243 goto out_unlock;
1244 }
1245 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1246 err = -EINVAL;
1247 goto out_unlock;
1248 }
1249
1250 if (lo->lo_offset != info->lo_offset ||
1251 lo->lo_sizelimit != info->lo_sizelimit) {
1252 sync_blockdev(lo->lo_device);
1253 kill_bdev(lo->lo_device);
1254 }
1255
1256 /* I/O need to be drained during transfer transition */
1257 blk_mq_freeze_queue(lo->lo_queue);
1258
1259 err = loop_release_xfer(lo);
1260 if (err)
1261 goto out_unfreeze;
1262
1263 if (info->lo_encrypt_type) {
1264 unsigned int type = info->lo_encrypt_type;
1265
1266 if (type >= MAX_LO_CRYPT) {
1267 err = -EINVAL;
1268 goto out_unfreeze;
1269 }
1270 xfer = xfer_funcs[type];
1271 if (xfer == NULL) {
1272 err = -EINVAL;
1273 goto out_unfreeze;
1274 }
1275 } else
1276 xfer = NULL;
1277
1278 err = loop_init_xfer(lo, xfer, info);
1279 if (err)
1280 goto out_unfreeze;
1281
1282 if (lo->lo_offset != info->lo_offset ||
1283 lo->lo_sizelimit != info->lo_sizelimit) {
1284 /* kill_bdev should have truncated all the pages */
1285 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1286 err = -EAGAIN;
1287 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1288 __func__, lo->lo_number, lo->lo_file_name,
1289 lo->lo_device->bd_inode->i_mapping->nrpages);
1290 goto out_unfreeze;
1291 }
1292 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1293 err = -EFBIG;
1294 goto out_unfreeze;
1295 }
1296 }
1297
1298 loop_config_discard(lo);
1299
1300 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1301 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1302 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1303 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1304
1305 if (!xfer)
1306 xfer = &none_funcs;
1307 lo->transfer = xfer->transfer;
1308 lo->ioctl = xfer->ioctl;
1309
1310 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1311 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1312 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1313
1314 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1315 lo->lo_init[0] = info->lo_init[0];
1316 lo->lo_init[1] = info->lo_init[1];
1317 if (info->lo_encrypt_key_size) {
1318 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1319 info->lo_encrypt_key_size);
1320 lo->lo_key_owner = uid;
1321 }
1322
1323 /* update dio if lo_offset or transfer is changed */
1324 __loop_update_dio(lo, lo->use_dio);
1325
1326out_unfreeze:
1327 blk_mq_unfreeze_queue(lo->lo_queue);
1328
1329 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1330 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1331 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1332 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1333 bdev = lo->lo_device;
1334 partscan = true;
1335 }
1336out_unlock:
1337 mutex_unlock(&loop_ctl_mutex);
1338 if (partscan)
1339 loop_reread_partitions(lo, bdev);
1340
1341 return err;
1342}
1343
1344static int
1345loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1346{
1347 struct path path;
1348 struct kstat stat;
1349 int ret;
1350
1351 ret = mutex_lock_killable(&loop_ctl_mutex);
1352 if (ret)
1353 return ret;
1354 if (lo->lo_state != Lo_bound) {
1355 mutex_unlock(&loop_ctl_mutex);
1356 return -ENXIO;
1357 }
1358
1359 memset(info, 0, sizeof(*info));
1360 info->lo_number = lo->lo_number;
1361 info->lo_offset = lo->lo_offset;
1362 info->lo_sizelimit = lo->lo_sizelimit;
1363 info->lo_flags = lo->lo_flags;
1364 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1365 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1366 info->lo_encrypt_type =
1367 lo->lo_encryption ? lo->lo_encryption->number : 0;
1368 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1369 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1370 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1371 lo->lo_encrypt_key_size);
1372 }
1373
1374 /* Drop loop_ctl_mutex while we call into the filesystem. */
1375 path = lo->lo_backing_file->f_path;
1376 path_get(&path);
1377 mutex_unlock(&loop_ctl_mutex);
1378 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1379 if (!ret) {
1380 info->lo_device = huge_encode_dev(stat.dev);
1381 info->lo_inode = stat.ino;
1382 info->lo_rdevice = huge_encode_dev(stat.rdev);
1383 }
1384 path_put(&path);
1385 return ret;
1386}
1387
1388static void
1389loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1390{
1391 memset(info64, 0, sizeof(*info64));
1392 info64->lo_number = info->lo_number;
1393 info64->lo_device = info->lo_device;
1394 info64->lo_inode = info->lo_inode;
1395 info64->lo_rdevice = info->lo_rdevice;
1396 info64->lo_offset = info->lo_offset;
1397 info64->lo_sizelimit = 0;
1398 info64->lo_encrypt_type = info->lo_encrypt_type;
1399 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1400 info64->lo_flags = info->lo_flags;
1401 info64->lo_init[0] = info->lo_init[0];
1402 info64->lo_init[1] = info->lo_init[1];
1403 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1404 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1405 else
1406 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1407 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1408}
1409
1410static int
1411loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1412{
1413 memset(info, 0, sizeof(*info));
1414 info->lo_number = info64->lo_number;
1415 info->lo_device = info64->lo_device;
1416 info->lo_inode = info64->lo_inode;
1417 info->lo_rdevice = info64->lo_rdevice;
1418 info->lo_offset = info64->lo_offset;
1419 info->lo_encrypt_type = info64->lo_encrypt_type;
1420 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1421 info->lo_flags = info64->lo_flags;
1422 info->lo_init[0] = info64->lo_init[0];
1423 info->lo_init[1] = info64->lo_init[1];
1424 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1425 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1426 else
1427 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1428 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1429
1430 /* error in case values were truncated */
1431 if (info->lo_device != info64->lo_device ||
1432 info->lo_rdevice != info64->lo_rdevice ||
1433 info->lo_inode != info64->lo_inode ||
1434 info->lo_offset != info64->lo_offset)
1435 return -EOVERFLOW;
1436
1437 return 0;
1438}
1439
1440static int
1441loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1442{
1443 struct loop_info info;
1444 struct loop_info64 info64;
1445
1446 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1447 return -EFAULT;
1448 loop_info64_from_old(&info, &info64);
1449 return loop_set_status(lo, &info64);
1450}
1451
1452static int
1453loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1454{
1455 struct loop_info64 info64;
1456
1457 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1458 return -EFAULT;
1459 return loop_set_status(lo, &info64);
1460}
1461
1462static int
1463loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1464 struct loop_info info;
1465 struct loop_info64 info64;
1466 int err;
1467
1468 if (!arg)
1469 return -EINVAL;
1470 err = loop_get_status(lo, &info64);
1471 if (!err)
1472 err = loop_info64_to_old(&info64, &info);
1473 if (!err && copy_to_user(arg, &info, sizeof(info)))
1474 err = -EFAULT;
1475
1476 return err;
1477}
1478
1479static int
1480loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1481 struct loop_info64 info64;
1482 int err;
1483
1484 if (!arg)
1485 return -EINVAL;
1486 err = loop_get_status(lo, &info64);
1487 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1488 err = -EFAULT;
1489
1490 return err;
1491}
1492
1493static int loop_set_capacity(struct loop_device *lo)
1494{
1495 if (unlikely(lo->lo_state != Lo_bound))
1496 return -ENXIO;
1497
1498 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1499}
1500
1501static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1502{
1503 int error = -ENXIO;
1504 if (lo->lo_state != Lo_bound)
1505 goto out;
1506
1507 __loop_update_dio(lo, !!arg);
1508 if (lo->use_dio == !!arg)
1509 return 0;
1510 error = -EINVAL;
1511 out:
1512 return error;
1513}
1514
1515static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1516{
1517 int err = 0;
1518
1519 if (lo->lo_state != Lo_bound)
1520 return -ENXIO;
1521
1522 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1523 return -EINVAL;
1524
1525 if (lo->lo_queue->limits.logical_block_size != arg) {
1526 sync_blockdev(lo->lo_device);
1527 kill_bdev(lo->lo_device);
1528 }
1529
1530 blk_mq_freeze_queue(lo->lo_queue);
1531
1532 /* kill_bdev should have truncated all the pages */
1533 if (lo->lo_queue->limits.logical_block_size != arg &&
1534 lo->lo_device->bd_inode->i_mapping->nrpages) {
1535 err = -EAGAIN;
1536 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1537 __func__, lo->lo_number, lo->lo_file_name,
1538 lo->lo_device->bd_inode->i_mapping->nrpages);
1539 goto out_unfreeze;
1540 }
1541
1542 blk_queue_logical_block_size(lo->lo_queue, arg);
1543 blk_queue_physical_block_size(lo->lo_queue, arg);
1544 blk_queue_io_min(lo->lo_queue, arg);
1545 loop_update_dio(lo);
1546out_unfreeze:
1547 blk_mq_unfreeze_queue(lo->lo_queue);
1548
1549 return err;
1550}
1551
1552static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1553 unsigned long arg)
1554{
1555 int err;
1556
1557 err = mutex_lock_killable(&loop_ctl_mutex);
1558 if (err)
1559 return err;
1560 switch (cmd) {
1561 case LOOP_SET_CAPACITY:
1562 err = loop_set_capacity(lo);
1563 break;
1564 case LOOP_SET_DIRECT_IO:
1565 err = loop_set_dio(lo, arg);
1566 break;
1567 case LOOP_SET_BLOCK_SIZE:
1568 err = loop_set_block_size(lo, arg);
1569 break;
1570 default:
1571 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1572 }
1573 mutex_unlock(&loop_ctl_mutex);
1574 return err;
1575}
1576
1577static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1578 unsigned int cmd, unsigned long arg)
1579{
1580 struct loop_device *lo = bdev->bd_disk->private_data;
1581 int err;
1582
1583 switch (cmd) {
1584 case LOOP_SET_FD:
1585 return loop_set_fd(lo, mode, bdev, arg);
1586 case LOOP_CHANGE_FD:
1587 return loop_change_fd(lo, bdev, arg);
1588 case LOOP_CLR_FD:
1589 return loop_clr_fd(lo);
1590 case LOOP_SET_STATUS:
1591 err = -EPERM;
1592 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1593 err = loop_set_status_old(lo,
1594 (struct loop_info __user *)arg);
1595 }
1596 break;
1597 case LOOP_GET_STATUS:
1598 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1599 case LOOP_SET_STATUS64:
1600 err = -EPERM;
1601 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1602 err = loop_set_status64(lo,
1603 (struct loop_info64 __user *) arg);
1604 }
1605 break;
1606 case LOOP_GET_STATUS64:
1607 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1608 case LOOP_SET_CAPACITY:
1609 case LOOP_SET_DIRECT_IO:
1610 case LOOP_SET_BLOCK_SIZE:
1611 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1612 return -EPERM;
1613 /* Fall through */
1614 default:
1615 err = lo_simple_ioctl(lo, cmd, arg);
1616 break;
1617 }
1618
1619 return err;
1620}
1621
1622#ifdef CONFIG_COMPAT
1623struct compat_loop_info {
1624 compat_int_t lo_number; /* ioctl r/o */
1625 compat_dev_t lo_device; /* ioctl r/o */
1626 compat_ulong_t lo_inode; /* ioctl r/o */
1627 compat_dev_t lo_rdevice; /* ioctl r/o */
1628 compat_int_t lo_offset;
1629 compat_int_t lo_encrypt_type;
1630 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1631 compat_int_t lo_flags; /* ioctl r/o */
1632 char lo_name[LO_NAME_SIZE];
1633 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1634 compat_ulong_t lo_init[2];
1635 char reserved[4];
1636};
1637
1638/*
1639 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1640 * - noinlined to reduce stack space usage in main part of driver
1641 */
1642static noinline int
1643loop_info64_from_compat(const struct compat_loop_info __user *arg,
1644 struct loop_info64 *info64)
1645{
1646 struct compat_loop_info info;
1647
1648 if (copy_from_user(&info, arg, sizeof(info)))
1649 return -EFAULT;
1650
1651 memset(info64, 0, sizeof(*info64));
1652 info64->lo_number = info.lo_number;
1653 info64->lo_device = info.lo_device;
1654 info64->lo_inode = info.lo_inode;
1655 info64->lo_rdevice = info.lo_rdevice;
1656 info64->lo_offset = info.lo_offset;
1657 info64->lo_sizelimit = 0;
1658 info64->lo_encrypt_type = info.lo_encrypt_type;
1659 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1660 info64->lo_flags = info.lo_flags;
1661 info64->lo_init[0] = info.lo_init[0];
1662 info64->lo_init[1] = info.lo_init[1];
1663 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1664 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1665 else
1666 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1667 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1668 return 0;
1669}
1670
1671/*
1672 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1673 * - noinlined to reduce stack space usage in main part of driver
1674 */
1675static noinline int
1676loop_info64_to_compat(const struct loop_info64 *info64,
1677 struct compat_loop_info __user *arg)
1678{
1679 struct compat_loop_info info;
1680
1681 memset(&info, 0, sizeof(info));
1682 info.lo_number = info64->lo_number;
1683 info.lo_device = info64->lo_device;
1684 info.lo_inode = info64->lo_inode;
1685 info.lo_rdevice = info64->lo_rdevice;
1686 info.lo_offset = info64->lo_offset;
1687 info.lo_encrypt_type = info64->lo_encrypt_type;
1688 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1689 info.lo_flags = info64->lo_flags;
1690 info.lo_init[0] = info64->lo_init[0];
1691 info.lo_init[1] = info64->lo_init[1];
1692 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1693 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1694 else
1695 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1696 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1697
1698 /* error in case values were truncated */
1699 if (info.lo_device != info64->lo_device ||
1700 info.lo_rdevice != info64->lo_rdevice ||
1701 info.lo_inode != info64->lo_inode ||
1702 info.lo_offset != info64->lo_offset ||
1703 info.lo_init[0] != info64->lo_init[0] ||
1704 info.lo_init[1] != info64->lo_init[1])
1705 return -EOVERFLOW;
1706
1707 if (copy_to_user(arg, &info, sizeof(info)))
1708 return -EFAULT;
1709 return 0;
1710}
1711
1712static int
1713loop_set_status_compat(struct loop_device *lo,
1714 const struct compat_loop_info __user *arg)
1715{
1716 struct loop_info64 info64;
1717 int ret;
1718
1719 ret = loop_info64_from_compat(arg, &info64);
1720 if (ret < 0)
1721 return ret;
1722 return loop_set_status(lo, &info64);
1723}
1724
1725static int
1726loop_get_status_compat(struct loop_device *lo,
1727 struct compat_loop_info __user *arg)
1728{
1729 struct loop_info64 info64;
1730 int err;
1731
1732 if (!arg)
1733 return -EINVAL;
1734 err = loop_get_status(lo, &info64);
1735 if (!err)
1736 err = loop_info64_to_compat(&info64, arg);
1737 return err;
1738}
1739
1740static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1741 unsigned int cmd, unsigned long arg)
1742{
1743 struct loop_device *lo = bdev->bd_disk->private_data;
1744 int err;
1745
1746 switch(cmd) {
1747 case LOOP_SET_STATUS:
1748 err = loop_set_status_compat(lo,
1749 (const struct compat_loop_info __user *)arg);
1750 break;
1751 case LOOP_GET_STATUS:
1752 err = loop_get_status_compat(lo,
1753 (struct compat_loop_info __user *)arg);
1754 break;
1755 case LOOP_SET_CAPACITY:
1756 case LOOP_CLR_FD:
1757 case LOOP_GET_STATUS64:
1758 case LOOP_SET_STATUS64:
1759 arg = (unsigned long) compat_ptr(arg);
1760 /* fall through */
1761 case LOOP_SET_FD:
1762 case LOOP_CHANGE_FD:
1763 case LOOP_SET_BLOCK_SIZE:
1764 err = lo_ioctl(bdev, mode, cmd, arg);
1765 break;
1766 default:
1767 err = -ENOIOCTLCMD;
1768 break;
1769 }
1770 return err;
1771}
1772#endif
1773
1774static int lo_open(struct block_device *bdev, fmode_t mode)
1775{
1776 struct loop_device *lo;
1777 int err;
1778
1779 err = mutex_lock_killable(&loop_ctl_mutex);
1780 if (err)
1781 return err;
1782 lo = bdev->bd_disk->private_data;
1783 if (!lo) {
1784 err = -ENXIO;
1785 goto out;
1786 }
1787
1788 atomic_inc(&lo->lo_refcnt);
1789out:
1790 mutex_unlock(&loop_ctl_mutex);
1791 return err;
1792}
1793
1794static void lo_release(struct gendisk *disk, fmode_t mode)
1795{
1796 struct loop_device *lo;
1797
1798 mutex_lock(&loop_ctl_mutex);
1799 lo = disk->private_data;
1800 if (atomic_dec_return(&lo->lo_refcnt))
1801 goto out_unlock;
1802
1803 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1804 if (lo->lo_state != Lo_bound)
1805 goto out_unlock;
1806 lo->lo_state = Lo_rundown;
1807 mutex_unlock(&loop_ctl_mutex);
1808 /*
1809 * In autoclear mode, stop the loop thread
1810 * and remove configuration after last close.
1811 */
1812 __loop_clr_fd(lo, true);
1813 return;
1814 } else if (lo->lo_state == Lo_bound) {
1815 /*
1816 * Otherwise keep thread (if running) and config,
1817 * but flush possible ongoing bios in thread.
1818 */
1819 blk_mq_freeze_queue(lo->lo_queue);
1820 blk_mq_unfreeze_queue(lo->lo_queue);
1821 }
1822
1823out_unlock:
1824 mutex_unlock(&loop_ctl_mutex);
1825}
1826
1827static const struct block_device_operations lo_fops = {
1828 .owner = THIS_MODULE,
1829 .open = lo_open,
1830 .release = lo_release,
1831 .ioctl = lo_ioctl,
1832#ifdef CONFIG_COMPAT
1833 .compat_ioctl = lo_compat_ioctl,
1834#endif
1835};
1836
1837/*
1838 * And now the modules code and kernel interface.
1839 */
1840static int max_loop;
1841module_param(max_loop, int, 0444);
1842MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1843module_param(max_part, int, 0444);
1844MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1845MODULE_LICENSE("GPL");
1846MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1847
1848int loop_register_transfer(struct loop_func_table *funcs)
1849{
1850 unsigned int n = funcs->number;
1851
1852 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1853 return -EINVAL;
1854 xfer_funcs[n] = funcs;
1855 return 0;
1856}
1857
1858static int unregister_transfer_cb(int id, void *ptr, void *data)
1859{
1860 struct loop_device *lo = ptr;
1861 struct loop_func_table *xfer = data;
1862
1863 mutex_lock(&loop_ctl_mutex);
1864 if (lo->lo_encryption == xfer)
1865 loop_release_xfer(lo);
1866 mutex_unlock(&loop_ctl_mutex);
1867 return 0;
1868}
1869
1870int loop_unregister_transfer(int number)
1871{
1872 unsigned int n = number;
1873 struct loop_func_table *xfer;
1874
1875 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1876 return -EINVAL;
1877
1878 xfer_funcs[n] = NULL;
1879 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1880 return 0;
1881}
1882
1883EXPORT_SYMBOL(loop_register_transfer);
1884EXPORT_SYMBOL(loop_unregister_transfer);
1885
1886static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1887 const struct blk_mq_queue_data *bd)
1888{
1889 struct request *rq = bd->rq;
1890 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1891 struct loop_device *lo = rq->q->queuedata;
1892
1893 blk_mq_start_request(rq);
1894
1895 if (lo->lo_state != Lo_bound)
1896 return BLK_STS_IOERR;
1897
1898 switch (req_op(rq)) {
1899 case REQ_OP_FLUSH:
1900 case REQ_OP_DISCARD:
1901 case REQ_OP_WRITE_ZEROES:
1902 cmd->use_aio = false;
1903 break;
1904 default:
1905 cmd->use_aio = lo->use_dio;
1906 break;
1907 }
1908
1909 /* always use the first bio's css */
1910#ifdef CONFIG_BLK_CGROUP
1911 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1912 cmd->css = &bio_blkcg(rq->bio)->css;
1913 css_get(cmd->css);
1914 } else
1915#endif
1916 cmd->css = NULL;
1917 kthread_queue_work(&lo->worker, &cmd->work);
1918
1919 return BLK_STS_OK;
1920}
1921
1922static void loop_handle_cmd(struct loop_cmd *cmd)
1923{
1924 struct request *rq = blk_mq_rq_from_pdu(cmd);
1925 const bool write = op_is_write(req_op(rq));
1926 struct loop_device *lo = rq->q->queuedata;
1927 int ret = 0;
1928
1929 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1930 ret = -EIO;
1931 goto failed;
1932 }
1933
1934 ret = do_req_filebacked(lo, rq);
1935 failed:
1936 /* complete non-aio request */
1937 if (!cmd->use_aio || ret) {
1938 cmd->ret = ret ? -EIO : 0;
1939 blk_mq_complete_request(rq);
1940 }
1941}
1942
1943static void loop_queue_work(struct kthread_work *work)
1944{
1945 struct loop_cmd *cmd =
1946 container_of(work, struct loop_cmd, work);
1947
1948 loop_handle_cmd(cmd);
1949}
1950
1951static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1952 unsigned int hctx_idx, unsigned int numa_node)
1953{
1954 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1955
1956 kthread_init_work(&cmd->work, loop_queue_work);
1957 return 0;
1958}
1959
1960static const struct blk_mq_ops loop_mq_ops = {
1961 .queue_rq = loop_queue_rq,
1962 .init_request = loop_init_request,
1963 .complete = lo_complete_rq,
1964};
1965
1966static int loop_add(struct loop_device **l, int i)
1967{
1968 struct loop_device *lo;
1969 struct gendisk *disk;
1970 int err;
1971
1972 err = -ENOMEM;
1973 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1974 if (!lo)
1975 goto out;
1976
1977 lo->lo_state = Lo_unbound;
1978
1979 /* allocate id, if @id >= 0, we're requesting that specific id */
1980 if (i >= 0) {
1981 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1982 if (err == -ENOSPC)
1983 err = -EEXIST;
1984 } else {
1985 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1986 }
1987 if (err < 0)
1988 goto out_free_dev;
1989 i = err;
1990
1991 err = -ENOMEM;
1992 lo->tag_set.ops = &loop_mq_ops;
1993 lo->tag_set.nr_hw_queues = 1;
1994 lo->tag_set.queue_depth = 128;
1995 lo->tag_set.numa_node = NUMA_NO_NODE;
1996 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1997 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1998 lo->tag_set.driver_data = lo;
1999
2000 err = blk_mq_alloc_tag_set(&lo->tag_set);
2001 if (err)
2002 goto out_free_idr;
2003
2004 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2005 if (IS_ERR(lo->lo_queue)) {
2006 err = PTR_ERR(lo->lo_queue);
2007 goto out_cleanup_tags;
2008 }
2009 lo->lo_queue->queuedata = lo;
2010
2011 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2012
2013 /*
2014 * By default, we do buffer IO, so it doesn't make sense to enable
2015 * merge because the I/O submitted to backing file is handled page by
2016 * page. For directio mode, merge does help to dispatch bigger request
2017 * to underlayer disk. We will enable merge once directio is enabled.
2018 */
2019 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2020
2021 err = -ENOMEM;
2022 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2023 if (!disk)
2024 goto out_free_queue;
2025
2026 /*
2027 * Disable partition scanning by default. The in-kernel partition
2028 * scanning can be requested individually per-device during its
2029 * setup. Userspace can always add and remove partitions from all
2030 * devices. The needed partition minors are allocated from the
2031 * extended minor space, the main loop device numbers will continue
2032 * to match the loop minors, regardless of the number of partitions
2033 * used.
2034 *
2035 * If max_part is given, partition scanning is globally enabled for
2036 * all loop devices. The minors for the main loop devices will be
2037 * multiples of max_part.
2038 *
2039 * Note: Global-for-all-devices, set-only-at-init, read-only module
2040 * parameteters like 'max_loop' and 'max_part' make things needlessly
2041 * complicated, are too static, inflexible and may surprise
2042 * userspace tools. Parameters like this in general should be avoided.
2043 */
2044 if (!part_shift)
2045 disk->flags |= GENHD_FL_NO_PART_SCAN;
2046 disk->flags |= GENHD_FL_EXT_DEVT;
2047 atomic_set(&lo->lo_refcnt, 0);
2048 lo->lo_number = i;
2049 spin_lock_init(&lo->lo_lock);
2050 disk->major = LOOP_MAJOR;
2051 disk->first_minor = i << part_shift;
2052 disk->fops = &lo_fops;
2053 disk->private_data = lo;
2054 disk->queue = lo->lo_queue;
2055 sprintf(disk->disk_name, "loop%d", i);
2056 add_disk(disk);
2057 *l = lo;
2058 return lo->lo_number;
2059
2060out_free_queue:
2061 blk_cleanup_queue(lo->lo_queue);
2062out_cleanup_tags:
2063 blk_mq_free_tag_set(&lo->tag_set);
2064out_free_idr:
2065 idr_remove(&loop_index_idr, i);
2066out_free_dev:
2067 kfree(lo);
2068out:
2069 return err;
2070}
2071
2072static void loop_remove(struct loop_device *lo)
2073{
2074 del_gendisk(lo->lo_disk);
2075 blk_cleanup_queue(lo->lo_queue);
2076 blk_mq_free_tag_set(&lo->tag_set);
2077 put_disk(lo->lo_disk);
2078 kfree(lo);
2079}
2080
2081static int find_free_cb(int id, void *ptr, void *data)
2082{
2083 struct loop_device *lo = ptr;
2084 struct loop_device **l = data;
2085
2086 if (lo->lo_state == Lo_unbound) {
2087 *l = lo;
2088 return 1;
2089 }
2090 return 0;
2091}
2092
2093static int loop_lookup(struct loop_device **l, int i)
2094{
2095 struct loop_device *lo;
2096 int ret = -ENODEV;
2097
2098 if (i < 0) {
2099 int err;
2100
2101 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2102 if (err == 1) {
2103 *l = lo;
2104 ret = lo->lo_number;
2105 }
2106 goto out;
2107 }
2108
2109 /* lookup and return a specific i */
2110 lo = idr_find(&loop_index_idr, i);
2111 if (lo) {
2112 *l = lo;
2113 ret = lo->lo_number;
2114 }
2115out:
2116 return ret;
2117}
2118
2119static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2120{
2121 struct loop_device *lo;
2122 struct kobject *kobj;
2123 int err;
2124
2125 mutex_lock(&loop_ctl_mutex);
2126 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2127 if (err < 0)
2128 err = loop_add(&lo, MINOR(dev) >> part_shift);
2129 if (err < 0)
2130 kobj = NULL;
2131 else
2132 kobj = get_disk_and_module(lo->lo_disk);
2133 mutex_unlock(&loop_ctl_mutex);
2134
2135 *part = 0;
2136 return kobj;
2137}
2138
2139static long loop_control_ioctl(struct file *file, unsigned int cmd,
2140 unsigned long parm)
2141{
2142 struct loop_device *lo;
2143 int ret;
2144
2145 ret = mutex_lock_killable(&loop_ctl_mutex);
2146 if (ret)
2147 return ret;
2148
2149 ret = -ENOSYS;
2150 switch (cmd) {
2151 case LOOP_CTL_ADD:
2152 ret = loop_lookup(&lo, parm);
2153 if (ret >= 0) {
2154 ret = -EEXIST;
2155 break;
2156 }
2157 ret = loop_add(&lo, parm);
2158 break;
2159 case LOOP_CTL_REMOVE:
2160 ret = loop_lookup(&lo, parm);
2161 if (ret < 0)
2162 break;
2163 if (lo->lo_state != Lo_unbound) {
2164 ret = -EBUSY;
2165 break;
2166 }
2167 if (atomic_read(&lo->lo_refcnt) > 0) {
2168 ret = -EBUSY;
2169 break;
2170 }
2171 lo->lo_disk->private_data = NULL;
2172 idr_remove(&loop_index_idr, lo->lo_number);
2173 loop_remove(lo);
2174 break;
2175 case LOOP_CTL_GET_FREE:
2176 ret = loop_lookup(&lo, -1);
2177 if (ret >= 0)
2178 break;
2179 ret = loop_add(&lo, -1);
2180 }
2181 mutex_unlock(&loop_ctl_mutex);
2182
2183 return ret;
2184}
2185
2186static const struct file_operations loop_ctl_fops = {
2187 .open = nonseekable_open,
2188 .unlocked_ioctl = loop_control_ioctl,
2189 .compat_ioctl = loop_control_ioctl,
2190 .owner = THIS_MODULE,
2191 .llseek = noop_llseek,
2192};
2193
2194static struct miscdevice loop_misc = {
2195 .minor = LOOP_CTRL_MINOR,
2196 .name = "loop-control",
2197 .fops = &loop_ctl_fops,
2198};
2199
2200MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2201MODULE_ALIAS("devname:loop-control");
2202
2203static int __init loop_init(void)
2204{
2205 int i, nr;
2206 unsigned long range;
2207 struct loop_device *lo;
2208 int err;
2209
2210 part_shift = 0;
2211 if (max_part > 0) {
2212 part_shift = fls(max_part);
2213
2214 /*
2215 * Adjust max_part according to part_shift as it is exported
2216 * to user space so that user can decide correct minor number
2217 * if [s]he want to create more devices.
2218 *
2219 * Note that -1 is required because partition 0 is reserved
2220 * for the whole disk.
2221 */
2222 max_part = (1UL << part_shift) - 1;
2223 }
2224
2225 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2226 err = -EINVAL;
2227 goto err_out;
2228 }
2229
2230 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2231 err = -EINVAL;
2232 goto err_out;
2233 }
2234
2235 /*
2236 * If max_loop is specified, create that many devices upfront.
2237 * This also becomes a hard limit. If max_loop is not specified,
2238 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2239 * init time. Loop devices can be requested on-demand with the
2240 * /dev/loop-control interface, or be instantiated by accessing
2241 * a 'dead' device node.
2242 */
2243 if (max_loop) {
2244 nr = max_loop;
2245 range = max_loop << part_shift;
2246 } else {
2247 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2248 range = 1UL << MINORBITS;
2249 }
2250
2251 err = misc_register(&loop_misc);
2252 if (err < 0)
2253 goto err_out;
2254
2255
2256 if (register_blkdev(LOOP_MAJOR, "loop")) {
2257 err = -EIO;
2258 goto misc_out;
2259 }
2260
2261 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2262 THIS_MODULE, loop_probe, NULL, NULL);
2263
2264 /* pre-create number of devices given by config or max_loop */
2265 mutex_lock(&loop_ctl_mutex);
2266 for (i = 0; i < nr; i++)
2267 loop_add(&lo, i);
2268 mutex_unlock(&loop_ctl_mutex);
2269
2270 printk(KERN_INFO "loop: module loaded\n");
2271 return 0;
2272
2273misc_out:
2274 misc_deregister(&loop_misc);
2275err_out:
2276 return err;
2277}
2278
2279static int loop_exit_cb(int id, void *ptr, void *data)
2280{
2281 struct loop_device *lo = ptr;
2282
2283 loop_remove(lo);
2284 return 0;
2285}
2286
2287static void __exit loop_exit(void)
2288{
2289 unsigned long range;
2290
2291 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2292
2293 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2294 idr_destroy(&loop_index_idr);
2295
2296 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2297 unregister_blkdev(LOOP_MAJOR, "loop");
2298
2299 misc_deregister(&loop_misc);
2300}
2301
2302module_init(loop_init);
2303module_exit(loop_exit);
2304
2305#ifndef MODULE
2306static int __init max_loop_setup(char *str)
2307{
2308 max_loop = simple_strtol(str, NULL, 0);
2309 return 1;
2310}
2311
2312__setup("max_loop=", max_loop_setup);
2313#endif