?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / block / loop.c
blob728681a20b7f4df082468f2b83336e195f824872
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 - Vincent.Renardias@waw.com Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 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, rhp@draper.net
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 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, <eriki@himolde.no>, 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 <axboe@suse.de>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, 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
81 #include "loop.h"
82
83 #include <linux/uaccess.h>
84
85 static DEFINE_IDR(loop_index_idr);
86 static DEFINE_MUTEX(loop_ctl_mutex);
87
88 static int max_part;
89 static int part_shift;
90
91 static int transfer_xor(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95 {
96 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
98 char *in, *out, *key;
99 int i, keysize;
100
101 if (cmd == READ) {
102 in = raw_buf;
103 out = loop_buf;
104 } else {
105 in = loop_buf;
106 out = raw_buf;
107 }
108
109 key = lo->lo_encrypt_key;
110 keysize = lo->lo_encrypt_key_size;
111 for (i = 0; i < size; i++)
112 *out++ = *in++ ^ key[(i & 511) % keysize];
113
114 kunmap_atomic(loop_buf);
115 kunmap_atomic(raw_buf);
116 cond_resched();
117 return 0;
118 }
119
120 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
121 {
122 if (unlikely(info->lo_encrypt_key_size <= 0))
123 return -EINVAL;
124 return 0;
125 }
126
127 static struct loop_func_table none_funcs = {
128 .number = LO_CRYPT_NONE,
129 };
130
131 static struct loop_func_table xor_funcs = {
132 .number = LO_CRYPT_XOR,
133 .transfer = transfer_xor,
134 .init = xor_init
135 };
136
137 /* xfer_funcs[0] is special - its release function is never called */
138 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
139 &none_funcs,
140 &xor_funcs
141 };
142
143 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
144 {
145 loff_t loopsize;
146
147 /* Compute loopsize in bytes */
148 loopsize = i_size_read(file->f_mapping->host);
149 if (offset > 0)
150 loopsize -= offset;
151 /* offset is beyond i_size, weird but possible */
152 if (loopsize < 0)
153 return 0;
154
155 if (sizelimit > 0 && sizelimit < loopsize)
156 loopsize = sizelimit;
157 /*
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
160 */
161 return loopsize >> 9;
162 }
163
164 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
165 {
166 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167 }
168
169 static void __loop_update_dio(struct loop_device *lo, bool dio)
170 {
171 struct file *file = lo->lo_backing_file;
172 struct address_space *mapping = file->f_mapping;
173 struct inode *inode = mapping->host;
174 unsigned short sb_bsize = 0;
175 unsigned dio_align = 0;
176 bool use_dio;
177
178 if (inode->i_sb->s_bdev) {
179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
180 dio_align = sb_bsize - 1;
181 }
182
183 /*
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
188 *
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
191 * of requests in sane applications should be PAGE_SIZE aligned
192 */
193 if (dio) {
194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
195 !(lo->lo_offset & dio_align) &&
196 mapping->a_ops->direct_IO &&
197 !lo->transfer)
198 use_dio = true;
199 else
200 use_dio = false;
201 } else {
202 use_dio = false;
203 }
204
205 if (lo->use_dio == use_dio)
206 return;
207
208 /* flush dirty pages before changing direct IO */
209 vfs_fsync(file, 0);
210
211 /*
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
215 */
216 blk_mq_freeze_queue(lo->lo_queue);
217 lo->use_dio = use_dio;
218 if (use_dio) {
219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
220 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
221 } else {
222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
224 }
225 blk_mq_unfreeze_queue(lo->lo_queue);
226 }
227
228 static int
229 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
230 {
231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
232 sector_t x = (sector_t)size;
233 struct block_device *bdev = lo->lo_device;
234
235 if (unlikely((loff_t)x != size))
236 return -EFBIG;
237 if (lo->lo_offset != offset)
238 lo->lo_offset = offset;
239 if (lo->lo_sizelimit != sizelimit)
240 lo->lo_sizelimit = sizelimit;
241 set_capacity(lo->lo_disk, x);
242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
245 return 0;
246 }
247
248 static inline int
249 lo_do_transfer(struct loop_device *lo, int cmd,
250 struct page *rpage, unsigned roffs,
251 struct page *lpage, unsigned loffs,
252 int size, sector_t rblock)
253 {
254 int ret;
255
256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
257 if (likely(!ret))
258 return 0;
259
260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock << 9, size);
263 return ret;
264 }
265
266 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
267 {
268 struct iov_iter i;
269 ssize_t bw;
270
271 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
272
273 file_start_write(file);
274 bw = vfs_iter_write(file, &i, ppos, 0);
275 file_end_write(file);
276
277 if (likely(bw == bvec->bv_len))
278 return 0;
279
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos, bvec->bv_len);
283 if (bw >= 0)
284 bw = -EIO;
285 return bw;
286 }
287
288 static int lo_write_simple(struct loop_device *lo, struct request *rq,
289 loff_t pos)
290 {
291 struct bio_vec bvec;
292 struct req_iterator iter;
293 int ret = 0;
294
295 rq_for_each_segment(bvec, rq, iter) {
296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
297 if (ret < 0)
298 break;
299 cond_resched();
300 }
301
302 return ret;
303 }
304
305 /*
306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
309 */
310 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
311 loff_t pos)
312 {
313 struct bio_vec bvec, b;
314 struct req_iterator iter;
315 struct page *page;
316 int ret = 0;
317
318 page = alloc_page(GFP_NOIO);
319 if (unlikely(!page))
320 return -ENOMEM;
321
322 rq_for_each_segment(bvec, rq, iter) {
323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
324 bvec.bv_offset, bvec.bv_len, pos >> 9);
325 if (unlikely(ret))
326 break;
327
328 b.bv_page = page;
329 b.bv_offset = 0;
330 b.bv_len = bvec.bv_len;
331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
332 if (ret < 0)
333 break;
334 }
335
336 __free_page(page);
337 return ret;
338 }
339
340 static int lo_read_simple(struct loop_device *lo, struct request *rq,
341 loff_t pos)
342 {
343 struct bio_vec bvec;
344 struct req_iterator iter;
345 struct iov_iter i;
346 ssize_t len;
347
348 rq_for_each_segment(bvec, rq, iter) {
349 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
351 if (len < 0)
352 return len;
353
354 flush_dcache_page(bvec.bv_page);
355
356 if (len != bvec.bv_len) {
357 struct bio *bio;
358
359 __rq_for_each_bio(bio, rq)
360 zero_fill_bio(bio);
361 break;
362 }
363 cond_resched();
364 }
365
366 return 0;
367 }
368
369 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
370 loff_t pos)
371 {
372 struct bio_vec bvec, b;
373 struct req_iterator iter;
374 struct iov_iter i;
375 struct page *page;
376 ssize_t len;
377 int ret = 0;
378
379 page = alloc_page(GFP_NOIO);
380 if (unlikely(!page))
381 return -ENOMEM;
382
383 rq_for_each_segment(bvec, rq, iter) {
384 loff_t offset = pos;
385
386 b.bv_page = page;
387 b.bv_offset = 0;
388 b.bv_len = bvec.bv_len;
389
390 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
392 if (len < 0) {
393 ret = len;
394 goto out_free_page;
395 }
396
397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
398 bvec.bv_offset, len, offset >> 9);
399 if (ret)
400 goto out_free_page;
401
402 flush_dcache_page(bvec.bv_page);
403
404 if (len != bvec.bv_len) {
405 struct bio *bio;
406
407 __rq_for_each_bio(bio, rq)
408 zero_fill_bio(bio);
409 break;
410 }
411 }
412
413 ret = 0;
414 out_free_page:
415 __free_page(page);
416 return ret;
417 }
418
419 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
420 int mode)
421 {
422 /*
423 * We use fallocate to manipulate the space mappings used by the image
424 * a.k.a. discard/zerorange. However we do not support this if
425 * encryption is enabled, because it may give an attacker useful
426 * information.
427 */
428 struct file *file = lo->lo_backing_file;
429 struct request_queue *q = lo->lo_queue;
430 int ret;
431
432 mode |= FALLOC_FL_KEEP_SIZE;
433
434 if (!blk_queue_discard(q)) {
435 ret = -EOPNOTSUPP;
436 goto out;
437 }
438
439 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
440 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
441 ret = -EIO;
442 out:
443 return ret;
444 }
445
446 static int lo_req_flush(struct loop_device *lo, struct request *rq)
447 {
448 struct file *file = lo->lo_backing_file;
449 int ret = vfs_fsync(file, 0);
450 if (unlikely(ret && ret != -EINVAL))
451 ret = -EIO;
452
453 return ret;
454 }
455
456 static void lo_complete_rq(struct request *rq)
457 {
458 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
459 blk_status_t ret = BLK_STS_OK;
460
461 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
462 req_op(rq) != REQ_OP_READ) {
463 if (cmd->ret < 0)
464 ret = BLK_STS_IOERR;
465 goto end_io;
466 }
467
468 /*
469 * Short READ - if we got some data, advance our request and
470 * retry it. If we got no data, end the rest with EIO.
471 */
472 if (cmd->ret) {
473 blk_update_request(rq, BLK_STS_OK, cmd->ret);
474 cmd->ret = 0;
475 blk_mq_requeue_request(rq, true);
476 } else {
477 if (cmd->use_aio) {
478 struct bio *bio = rq->bio;
479
480 while (bio) {
481 zero_fill_bio(bio);
482 bio = bio->bi_next;
483 }
484 }
485 ret = BLK_STS_IOERR;
486 end_io:
487 blk_mq_end_request(rq, ret);
488 }
489 }
490
491 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
492 {
493 struct request *rq = blk_mq_rq_from_pdu(cmd);
494
495 if (!atomic_dec_and_test(&cmd->ref))
496 return;
497 kfree(cmd->bvec);
498 cmd->bvec = NULL;
499 blk_mq_complete_request(rq);
500 }
501
502 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
503 {
504 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
505
506 if (cmd->css)
507 css_put(cmd->css);
508 cmd->ret = ret;
509 lo_rw_aio_do_completion(cmd);
510 }
511
512 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
513 loff_t pos, bool rw)
514 {
515 struct iov_iter iter;
516 struct bio_vec *bvec;
517 struct request *rq = blk_mq_rq_from_pdu(cmd);
518 struct bio *bio = rq->bio;
519 struct file *file = lo->lo_backing_file;
520 unsigned int offset;
521 int segments = 0;
522 int ret;
523
524 if (rq->bio != rq->biotail) {
525 struct req_iterator iter;
526 struct bio_vec tmp;
527
528 __rq_for_each_bio(bio, rq)
529 segments += bio_segments(bio);
530 bvec = kmalloc_array(segments, sizeof(struct bio_vec),
531 GFP_NOIO);
532 if (!bvec)
533 return -EIO;
534 cmd->bvec = bvec;
535
536 /*
537 * The bios of the request may be started from the middle of
538 * the 'bvec' because of bio splitting, so we can't directly
539 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
540 * API will take care of all details for us.
541 */
542 rq_for_each_segment(tmp, rq, iter) {
543 *bvec = tmp;
544 bvec++;
545 }
546 bvec = cmd->bvec;
547 offset = 0;
548 } else {
549 /*
550 * Same here, this bio may be started from the middle of the
551 * 'bvec' because of bio splitting, so offset from the bvec
552 * must be passed to iov iterator
553 */
554 offset = bio->bi_iter.bi_bvec_done;
555 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
556 segments = bio_segments(bio);
557 }
558 atomic_set(&cmd->ref, 2);
559
560 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
561 segments, blk_rq_bytes(rq));
562 iter.iov_offset = offset;
563
564 cmd->iocb.ki_pos = pos;
565 cmd->iocb.ki_filp = file;
566 cmd->iocb.ki_complete = lo_rw_aio_complete;
567 cmd->iocb.ki_flags = IOCB_DIRECT;
568 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
569 if (cmd->css)
570 kthread_associate_blkcg(cmd->css);
571
572 if (rw == WRITE)
573 ret = call_write_iter(file, &cmd->iocb, &iter);
574 else
575 ret = call_read_iter(file, &cmd->iocb, &iter);
576
577 lo_rw_aio_do_completion(cmd);
578 kthread_associate_blkcg(NULL);
579
580 if (ret != -EIOCBQUEUED)
581 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
582 return 0;
583 }
584
585 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
586 {
587 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
588 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
589
590 /*
591 * lo_write_simple and lo_read_simple should have been covered
592 * by io submit style function like lo_rw_aio(), one blocker
593 * is that lo_read_simple() need to call flush_dcache_page after
594 * the page is written from kernel, and it isn't easy to handle
595 * this in io submit style function which submits all segments
596 * of the req at one time. And direct read IO doesn't need to
597 * run flush_dcache_page().
598 */
599 switch (req_op(rq)) {
600 case REQ_OP_FLUSH:
601 return lo_req_flush(lo, rq);
602 case REQ_OP_WRITE_ZEROES:
603 /*
604 * If the caller doesn't want deallocation, call zeroout to
605 * write zeroes the range. Otherwise, punch them out.
606 */
607 return lo_fallocate(lo, rq, pos,
608 (rq->cmd_flags & REQ_NOUNMAP) ?
609 FALLOC_FL_ZERO_RANGE :
610 FALLOC_FL_PUNCH_HOLE);
611 case REQ_OP_DISCARD:
612 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
613 case REQ_OP_WRITE:
614 if (lo->transfer)
615 return lo_write_transfer(lo, rq, pos);
616 else if (cmd->use_aio)
617 return lo_rw_aio(lo, cmd, pos, WRITE);
618 else
619 return lo_write_simple(lo, rq, pos);
620 case REQ_OP_READ:
621 if (lo->transfer)
622 return lo_read_transfer(lo, rq, pos);
623 else if (cmd->use_aio)
624 return lo_rw_aio(lo, cmd, pos, READ);
625 else
626 return lo_read_simple(lo, rq, pos);
627 default:
628 WARN_ON_ONCE(1);
629 return -EIO;
630 break;
631 }
632 }
633
634 static inline void loop_update_dio(struct loop_device *lo)
635 {
636 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
637 lo->use_dio);
638 }
639
640 static void loop_reread_partitions(struct loop_device *lo,
641 struct block_device *bdev)
642 {
643 int rc;
644
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 static inline int is_loop_device(struct file *file)
652 {
653 struct inode *i = file->f_mapping->host;
654
655 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
656 }
657
658 static int loop_validate_file(struct file *file, struct block_device *bdev)
659 {
660 struct inode *inode = file->f_mapping->host;
661 struct file *f = file;
662
663 /* Avoid recursion */
664 while (is_loop_device(f)) {
665 struct loop_device *l;
666
667 if (f->f_mapping->host->i_bdev == bdev)
668 return -EBADF;
669
670 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
671 if (l->lo_state != Lo_bound) {
672 return -EINVAL;
673 }
674 f = l->lo_backing_file;
675 }
676 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
677 return -EINVAL;
678 return 0;
679 }
680
681 /*
682 * loop_change_fd switched the backing store of a loopback device to
683 * a new file. This is useful for operating system installers to free up
684 * the original file and in High Availability environments to switch to
685 * an alternative location for the content in case of server meltdown.
686 * This can only work if the loop device is used read-only, and if the
687 * new backing store is the same size and type as the old backing store.
688 */
689 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
690 unsigned int arg)
691 {
692 struct file *file = NULL, *old_file;
693 int error;
694 bool partscan;
695
696 error = mutex_lock_killable(&loop_ctl_mutex);
697 if (error)
698 return error;
699 error = -ENXIO;
700 if (lo->lo_state != Lo_bound)
701 goto out_err;
702
703 /* the loop device has to be read-only */
704 error = -EINVAL;
705 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
706 goto out_err;
707
708 error = -EBADF;
709 file = fget(arg);
710 if (!file)
711 goto out_err;
712
713 error = loop_validate_file(file, bdev);
714 if (error)
715 goto out_err;
716
717 old_file = lo->lo_backing_file;
718
719 error = -EINVAL;
720
721 /* size of the new backing store needs to be the same */
722 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
723 goto out_err;
724
725 /* and ... switch */
726 blk_mq_freeze_queue(lo->lo_queue);
727 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
728 lo->lo_backing_file = file;
729 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
730 mapping_set_gfp_mask(file->f_mapping,
731 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
732 loop_update_dio(lo);
733 blk_mq_unfreeze_queue(lo->lo_queue);
734 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
735 mutex_unlock(&loop_ctl_mutex);
736 /*
737 * We must drop file reference outside of loop_ctl_mutex as dropping
738 * the file ref can take bd_mutex which creates circular locking
739 * dependency.
740 */
741 fput(old_file);
742 if (partscan)
743 loop_reread_partitions(lo, bdev);
744 return 0;
745
746 out_err:
747 mutex_unlock(&loop_ctl_mutex);
748 if (file)
749 fput(file);
750 return error;
751 }
752
753 /* loop sysfs attributes */
754
755 static ssize_t loop_attr_show(struct device *dev, char *page,
756 ssize_t (*callback)(struct loop_device *, char *))
757 {
758 struct gendisk *disk = dev_to_disk(dev);
759 struct loop_device *lo = disk->private_data;
760
761 return callback(lo, page);
762 }
763
764 #define LOOP_ATTR_RO(_name) \
765 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
766 static ssize_t loop_attr_do_show_##_name(struct device *d, \
767 struct device_attribute *attr, char *b) \
768 { \
769 return loop_attr_show(d, b, loop_attr_##_name##_show); \
770 } \
771 static struct device_attribute loop_attr_##_name = \
772 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
773
774 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
775 {
776 ssize_t ret;
777 char *p = NULL;
778
779 spin_lock_irq(&lo->lo_lock);
780 if (lo->lo_backing_file)
781 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
782 spin_unlock_irq(&lo->lo_lock);
783
784 if (IS_ERR_OR_NULL(p))
785 ret = PTR_ERR(p);
786 else {
787 ret = strlen(p);
788 memmove(buf, p, ret);
789 buf[ret++] = '\n';
790 buf[ret] = 0;
791 }
792
793 return ret;
794 }
795
796 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
797 {
798 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
799 }
800
801 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
802 {
803 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
804 }
805
806 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
807 {
808 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
809
810 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
811 }
812
813 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
814 {
815 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
816
817 return sprintf(buf, "%s\n", partscan ? "1" : "0");
818 }
819
820 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
821 {
822 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
823
824 return sprintf(buf, "%s\n", dio ? "1" : "0");
825 }
826
827 LOOP_ATTR_RO(backing_file);
828 LOOP_ATTR_RO(offset);
829 LOOP_ATTR_RO(sizelimit);
830 LOOP_ATTR_RO(autoclear);
831 LOOP_ATTR_RO(partscan);
832 LOOP_ATTR_RO(dio);
833
834 static struct attribute *loop_attrs[] = {
835 &loop_attr_backing_file.attr,
836 &loop_attr_offset.attr,
837 &loop_attr_sizelimit.attr,
838 &loop_attr_autoclear.attr,
839 &loop_attr_partscan.attr,
840 &loop_attr_dio.attr,
841 NULL,
842 };
843
844 static struct attribute_group loop_attribute_group = {
845 .name = "loop",
846 .attrs= loop_attrs,
847 };
848
849 static void loop_sysfs_init(struct loop_device *lo)
850 {
851 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
852 &loop_attribute_group);
853 }
854
855 static void loop_sysfs_exit(struct loop_device *lo)
856 {
857 if (lo->sysfs_inited)
858 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
859 &loop_attribute_group);
860 }
861
862 static void loop_config_discard(struct loop_device *lo)
863 {
864 struct file *file = lo->lo_backing_file;
865 struct inode *inode = file->f_mapping->host;
866 struct request_queue *q = lo->lo_queue;
867
868 /*
869 * If the backing device is a block device, mirror its zeroing
870 * capability. Set the discard sectors to the block device's zeroing
871 * capabilities because loop discards result in blkdev_issue_zeroout(),
872 * not blkdev_issue_discard(). This maintains consistent behavior with
873 * file-backed loop devices: discarded regions read back as zero.
874 */
875 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
876 struct request_queue *backingq;
877
878 backingq = bdev_get_queue(inode->i_bdev);
879 blk_queue_max_discard_sectors(q,
880 backingq->limits.max_write_zeroes_sectors);
881
882 blk_queue_max_write_zeroes_sectors(q,
883 backingq->limits.max_write_zeroes_sectors);
884
885 /*
886 * We use punch hole to reclaim the free space used by the
887 * image a.k.a. discard. However we do not support discard if
888 * encryption is enabled, because it may give an attacker
889 * useful information.
890 */
891 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
892 q->limits.discard_granularity = 0;
893 q->limits.discard_alignment = 0;
894 blk_queue_max_discard_sectors(q, 0);
895 blk_queue_max_write_zeroes_sectors(q, 0);
896
897 } else {
898 q->limits.discard_granularity = inode->i_sb->s_blocksize;
899 q->limits.discard_alignment = 0;
900
901 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
902 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
903 }
904
905 if (q->limits.max_write_zeroes_sectors)
906 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
907 else
908 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
909 }
910
911 static void loop_unprepare_queue(struct loop_device *lo)
912 {
913 kthread_flush_worker(&lo->worker);
914 kthread_stop(lo->worker_task);
915 }
916
917 static int loop_kthread_worker_fn(void *worker_ptr)
918 {
919 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
920 return kthread_worker_fn(worker_ptr);
921 }
922
923 static int loop_prepare_queue(struct loop_device *lo)
924 {
925 kthread_init_worker(&lo->worker);
926 lo->worker_task = kthread_run(loop_kthread_worker_fn,
927 &lo->worker, "loop%d", lo->lo_number);
928 if (IS_ERR(lo->worker_task))
929 return -ENOMEM;
930 set_user_nice(lo->worker_task, MIN_NICE);
931 return 0;
932 }
933
934 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
935 struct block_device *bdev, unsigned int arg)
936 {
937 struct file *file;
938 struct inode *inode;
939 struct address_space *mapping;
940 int lo_flags = 0;
941 int error;
942 loff_t size;
943 bool partscan;
944
945 /* This is safe, since we have a reference from open(). */
946 __module_get(THIS_MODULE);
947
948 error = -EBADF;
949 file = fget(arg);
950 if (!file)
951 goto out;
952
953 error = mutex_lock_killable(&loop_ctl_mutex);
954 if (error)
955 goto out_putf;
956
957 error = -EBUSY;
958 if (lo->lo_state != Lo_unbound)
959 goto out_unlock;
960
961 error = loop_validate_file(file, bdev);
962 if (error)
963 goto out_unlock;
964
965 mapping = file->f_mapping;
966 inode = mapping->host;
967
968 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
969 !file->f_op->write_iter)
970 lo_flags |= LO_FLAGS_READ_ONLY;
971
972 error = -EFBIG;
973 size = get_loop_size(lo, file);
974 if ((loff_t)(sector_t)size != size)
975 goto out_unlock;
976 error = loop_prepare_queue(lo);
977 if (error)
978 goto out_unlock;
979
980 error = 0;
981
982 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
983
984 lo->use_dio = false;
985 lo->lo_device = bdev;
986 lo->lo_flags = lo_flags;
987 lo->lo_backing_file = file;
988 lo->transfer = NULL;
989 lo->ioctl = NULL;
990 lo->lo_sizelimit = 0;
991 lo->old_gfp_mask = mapping_gfp_mask(mapping);
992 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
993
994 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
995 blk_queue_write_cache(lo->lo_queue, true, false);
996
997 loop_update_dio(lo);
998 set_capacity(lo->lo_disk, size);
999 bd_set_size(bdev, size << 9);
1000 loop_sysfs_init(lo);
1001 /* let user-space know about the new size */
1002 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1003
1004 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1005 block_size(inode->i_bdev) : PAGE_SIZE);
1006
1007 lo->lo_state = Lo_bound;
1008 if (part_shift)
1009 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1010 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1011
1012 /* Grab the block_device to prevent its destruction after we
1013 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1014 */
1015 bdgrab(bdev);
1016 mutex_unlock(&loop_ctl_mutex);
1017 if (partscan)
1018 loop_reread_partitions(lo, bdev);
1019 return 0;
1020
1021 out_unlock:
1022 mutex_unlock(&loop_ctl_mutex);
1023 out_putf:
1024 fput(file);
1025 out:
1026 /* This is safe: open() is still holding a reference. */
1027 module_put(THIS_MODULE);
1028 return error;
1029 }
1030
1031 static int
1032 loop_release_xfer(struct loop_device *lo)
1033 {
1034 int err = 0;
1035 struct loop_func_table *xfer = lo->lo_encryption;
1036
1037 if (xfer) {
1038 if (xfer->release)
1039 err = xfer->release(lo);
1040 lo->transfer = NULL;
1041 lo->lo_encryption = NULL;
1042 module_put(xfer->owner);
1043 }
1044 return err;
1045 }
1046
1047 static int
1048 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1049 const struct loop_info64 *i)
1050 {
1051 int err = 0;
1052
1053 if (xfer) {
1054 struct module *owner = xfer->owner;
1055
1056 if (!try_module_get(owner))
1057 return -EINVAL;
1058 if (xfer->init)
1059 err = xfer->init(lo, i);
1060 if (err)
1061 module_put(owner);
1062 else
1063 lo->lo_encryption = xfer;
1064 }
1065 return err;
1066 }
1067
1068 static int __loop_clr_fd(struct loop_device *lo, bool release)
1069 {
1070 struct file *filp = NULL;
1071 gfp_t gfp = lo->old_gfp_mask;
1072 struct block_device *bdev = lo->lo_device;
1073 int err = 0;
1074 bool partscan = false;
1075 int lo_number;
1076
1077 mutex_lock(&loop_ctl_mutex);
1078 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1079 err = -ENXIO;
1080 goto out_unlock;
1081 }
1082
1083 filp = lo->lo_backing_file;
1084 if (filp == NULL) {
1085 err = -EINVAL;
1086 goto out_unlock;
1087 }
1088
1089 /* freeze request queue during the transition */
1090 blk_mq_freeze_queue(lo->lo_queue);
1091
1092 spin_lock_irq(&lo->lo_lock);
1093 lo->lo_backing_file = NULL;
1094 spin_unlock_irq(&lo->lo_lock);
1095
1096 loop_release_xfer(lo);
1097 lo->transfer = NULL;
1098 lo->ioctl = NULL;
1099 lo->lo_device = NULL;
1100 lo->lo_encryption = NULL;
1101 lo->lo_offset = 0;
1102 lo->lo_sizelimit = 0;
1103 lo->lo_encrypt_key_size = 0;
1104 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1105 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1106 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1107 blk_queue_logical_block_size(lo->lo_queue, 512);
1108 blk_queue_physical_block_size(lo->lo_queue, 512);
1109 blk_queue_io_min(lo->lo_queue, 512);
1110 if (bdev) {
1111 bdput(bdev);
1112 invalidate_bdev(bdev);
1113 bdev->bd_inode->i_mapping->wb_err = 0;
1114 }
1115 set_capacity(lo->lo_disk, 0);
1116 loop_sysfs_exit(lo);
1117 if (bdev) {
1118 bd_set_size(bdev, 0);
1119 /* let user-space know about this change */
1120 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1121 }
1122 mapping_set_gfp_mask(filp->f_mapping, gfp);
1123 /* This is safe: open() is still holding a reference. */
1124 module_put(THIS_MODULE);
1125 blk_mq_unfreeze_queue(lo->lo_queue);
1126
1127 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1128 lo_number = lo->lo_number;
1129 loop_unprepare_queue(lo);
1130 out_unlock:
1131 mutex_unlock(&loop_ctl_mutex);
1132 if (partscan) {
1133 /*
1134 * bd_mutex has been held already in release path, so don't
1135 * acquire it if this function is called in such case.
1136 *
1137 * If the reread partition isn't from release path, lo_refcnt
1138 * must be at least one and it can only become zero when the
1139 * current holder is released.
1140 */
1141 if (release)
1142 err = __blkdev_reread_part(bdev);
1143 else
1144 err = blkdev_reread_part(bdev);
1145 if (err)
1146 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1147 __func__, lo_number, err);
1148 /* Device is gone, no point in returning error */
1149 err = 0;
1150 }
1151
1152 /*
1153 * lo->lo_state is set to Lo_unbound here after above partscan has
1154 * finished.
1155 *
1156 * There cannot be anybody else entering __loop_clr_fd() as
1157 * lo->lo_backing_file is already cleared and Lo_rundown state
1158 * protects us from all the other places trying to change the 'lo'
1159 * device.
1160 */
1161 mutex_lock(&loop_ctl_mutex);
1162 lo->lo_flags = 0;
1163 if (!part_shift)
1164 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1165 lo->lo_state = Lo_unbound;
1166 mutex_unlock(&loop_ctl_mutex);
1167
1168 /*
1169 * Need not hold loop_ctl_mutex to fput backing file.
1170 * Calling fput holding loop_ctl_mutex triggers a circular
1171 * lock dependency possibility warning as fput can take
1172 * bd_mutex which is usually taken before loop_ctl_mutex.
1173 */
1174 if (filp)
1175 fput(filp);
1176 return err;
1177 }
1178
1179 static int loop_clr_fd(struct loop_device *lo)
1180 {
1181 int err;
1182
1183 err = mutex_lock_killable(&loop_ctl_mutex);
1184 if (err)
1185 return err;
1186 if (lo->lo_state != Lo_bound) {
1187 mutex_unlock(&loop_ctl_mutex);
1188 return -ENXIO;
1189 }
1190 /*
1191 * If we've explicitly asked to tear down the loop device,
1192 * and it has an elevated reference count, set it for auto-teardown when
1193 * the last reference goes away. This stops $!~#$@ udev from
1194 * preventing teardown because it decided that it needs to run blkid on
1195 * the loopback device whenever they appear. xfstests is notorious for
1196 * failing tests because blkid via udev races with a losetup
1197 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1198 * command to fail with EBUSY.
1199 */
1200 if (atomic_read(&lo->lo_refcnt) > 1) {
1201 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1202 mutex_unlock(&loop_ctl_mutex);
1203 return 0;
1204 }
1205 lo->lo_state = Lo_rundown;
1206 mutex_unlock(&loop_ctl_mutex);
1207
1208 return __loop_clr_fd(lo, false);
1209 }
1210
1211 static int
1212 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1213 {
1214 int err;
1215 struct loop_func_table *xfer;
1216 kuid_t uid = current_uid();
1217 struct block_device *bdev;
1218 bool partscan = false;
1219
1220 err = mutex_lock_killable(&loop_ctl_mutex);
1221 if (err)
1222 return err;
1223 if (lo->lo_encrypt_key_size &&
1224 !uid_eq(lo->lo_key_owner, uid) &&
1225 !capable(CAP_SYS_ADMIN)) {
1226 err = -EPERM;
1227 goto out_unlock;
1228 }
1229 if (lo->lo_state != Lo_bound) {
1230 err = -ENXIO;
1231 goto out_unlock;
1232 }
1233 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1234 err = -EINVAL;
1235 goto out_unlock;
1236 }
1237
1238 if (lo->lo_offset != info->lo_offset ||
1239 lo->lo_sizelimit != info->lo_sizelimit) {
1240 sync_blockdev(lo->lo_device);
1241 invalidate_bdev(lo->lo_device);
1242 }
1243
1244 /* I/O need to be drained during transfer transition */
1245 blk_mq_freeze_queue(lo->lo_queue);
1246
1247 err = loop_release_xfer(lo);
1248 if (err)
1249 goto out_unfreeze;
1250
1251 if (info->lo_encrypt_type) {
1252 unsigned int type = info->lo_encrypt_type;
1253
1254 if (type >= MAX_LO_CRYPT) {
1255 err = -EINVAL;
1256 goto out_unfreeze;
1257 }
1258 xfer = xfer_funcs[type];
1259 if (xfer == NULL) {
1260 err = -EINVAL;
1261 goto out_unfreeze;
1262 }
1263 } else
1264 xfer = NULL;
1265
1266 err = loop_init_xfer(lo, xfer, info);
1267 if (err)
1268 goto out_unfreeze;
1269
1270 if (lo->lo_offset != info->lo_offset ||
1271 lo->lo_sizelimit != info->lo_sizelimit) {
1272 /* kill_bdev should have truncated all the pages */
1273 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1274 err = -EAGAIN;
1275 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1276 __func__, lo->lo_number, lo->lo_file_name,
1277 lo->lo_device->bd_inode->i_mapping->nrpages);
1278 goto out_unfreeze;
1279 }
1280 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1281 err = -EFBIG;
1282 goto out_unfreeze;
1283 }
1284 }
1285
1286 loop_config_discard(lo);
1287
1288 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1289 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1290 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1291 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1292
1293 if (!xfer)
1294 xfer = &none_funcs;
1295 lo->transfer = xfer->transfer;
1296 lo->ioctl = xfer->ioctl;
1297
1298 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1299 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1300 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1301
1302 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1303 lo->lo_init[0] = info->lo_init[0];
1304 lo->lo_init[1] = info->lo_init[1];
1305 if (info->lo_encrypt_key_size) {
1306 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1307 info->lo_encrypt_key_size);
1308 lo->lo_key_owner = uid;
1309 }
1310
1311 /* update dio if lo_offset or transfer is changed */
1312 __loop_update_dio(lo, lo->use_dio);
1313
1314 out_unfreeze:
1315 blk_mq_unfreeze_queue(lo->lo_queue);
1316
1317 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1318 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1319 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1320 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1321 bdev = lo->lo_device;
1322 partscan = true;
1323 }
1324 out_unlock:
1325 mutex_unlock(&loop_ctl_mutex);
1326 if (partscan)
1327 loop_reread_partitions(lo, bdev);
1328
1329 return err;
1330 }
1331
1332 static int
1333 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1334 {
1335 struct path path;
1336 struct kstat stat;
1337 int ret;
1338
1339 ret = mutex_lock_killable(&loop_ctl_mutex);
1340 if (ret)
1341 return ret;
1342 if (lo->lo_state != Lo_bound) {
1343 mutex_unlock(&loop_ctl_mutex);
1344 return -ENXIO;
1345 }
1346
1347 memset(info, 0, sizeof(*info));
1348 info->lo_number = lo->lo_number;
1349 info->lo_offset = lo->lo_offset;
1350 info->lo_sizelimit = lo->lo_sizelimit;
1351 info->lo_flags = lo->lo_flags;
1352 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1353 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1354 info->lo_encrypt_type =
1355 lo->lo_encryption ? lo->lo_encryption->number : 0;
1356 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1357 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1358 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1359 lo->lo_encrypt_key_size);
1360 }
1361
1362 /* Drop loop_ctl_mutex while we call into the filesystem. */
1363 path = lo->lo_backing_file->f_path;
1364 path_get(&path);
1365 mutex_unlock(&loop_ctl_mutex);
1366 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1367 if (!ret) {
1368 info->lo_device = huge_encode_dev(stat.dev);
1369 info->lo_inode = stat.ino;
1370 info->lo_rdevice = huge_encode_dev(stat.rdev);
1371 }
1372 path_put(&path);
1373 return ret;
1374 }
1375
1376 static void
1377 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1378 {
1379 memset(info64, 0, sizeof(*info64));
1380 info64->lo_number = info->lo_number;
1381 info64->lo_device = info->lo_device;
1382 info64->lo_inode = info->lo_inode;
1383 info64->lo_rdevice = info->lo_rdevice;
1384 info64->lo_offset = info->lo_offset;
1385 info64->lo_sizelimit = 0;
1386 info64->lo_encrypt_type = info->lo_encrypt_type;
1387 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1388 info64->lo_flags = info->lo_flags;
1389 info64->lo_init[0] = info->lo_init[0];
1390 info64->lo_init[1] = info->lo_init[1];
1391 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1392 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1393 else
1394 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1395 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1396 }
1397
1398 static int
1399 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1400 {
1401 memset(info, 0, sizeof(*info));
1402 info->lo_number = info64->lo_number;
1403 info->lo_device = info64->lo_device;
1404 info->lo_inode = info64->lo_inode;
1405 info->lo_rdevice = info64->lo_rdevice;
1406 info->lo_offset = info64->lo_offset;
1407 info->lo_encrypt_type = info64->lo_encrypt_type;
1408 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1409 info->lo_flags = info64->lo_flags;
1410 info->lo_init[0] = info64->lo_init[0];
1411 info->lo_init[1] = info64->lo_init[1];
1412 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1413 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1414 else
1415 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1416 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1417
1418 /* error in case values were truncated */
1419 if (info->lo_device != info64->lo_device ||
1420 info->lo_rdevice != info64->lo_rdevice ||
1421 info->lo_inode != info64->lo_inode ||
1422 info->lo_offset != info64->lo_offset)
1423 return -EOVERFLOW;
1424
1425 return 0;
1426 }
1427
1428 static int
1429 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1430 {
1431 struct loop_info info;
1432 struct loop_info64 info64;
1433
1434 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1435 return -EFAULT;
1436 loop_info64_from_old(&info, &info64);
1437 return loop_set_status(lo, &info64);
1438 }
1439
1440 static int
1441 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1442 {
1443 struct loop_info64 info64;
1444
1445 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1446 return -EFAULT;
1447 return loop_set_status(lo, &info64);
1448 }
1449
1450 static int
1451 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1452 struct loop_info info;
1453 struct loop_info64 info64;
1454 int err;
1455
1456 if (!arg)
1457 return -EINVAL;
1458 err = loop_get_status(lo, &info64);
1459 if (!err)
1460 err = loop_info64_to_old(&info64, &info);
1461 if (!err && copy_to_user(arg, &info, sizeof(info)))
1462 err = -EFAULT;
1463
1464 return err;
1465 }
1466
1467 static int
1468 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1469 struct loop_info64 info64;
1470 int err;
1471
1472 if (!arg)
1473 return -EINVAL;
1474 err = loop_get_status(lo, &info64);
1475 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1476 err = -EFAULT;
1477
1478 return err;
1479 }
1480
1481 static int loop_set_capacity(struct loop_device *lo)
1482 {
1483 if (unlikely(lo->lo_state != Lo_bound))
1484 return -ENXIO;
1485
1486 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1487 }
1488
1489 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1490 {
1491 int error = -ENXIO;
1492 if (lo->lo_state != Lo_bound)
1493 goto out;
1494
1495 __loop_update_dio(lo, !!arg);
1496 if (lo->use_dio == !!arg)
1497 return 0;
1498 error = -EINVAL;
1499 out:
1500 return error;
1501 }
1502
1503 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1504 {
1505 int err = 0;
1506
1507 if (lo->lo_state != Lo_bound)
1508 return -ENXIO;
1509
1510 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1511 return -EINVAL;
1512
1513 if (lo->lo_queue->limits.logical_block_size != arg) {
1514 sync_blockdev(lo->lo_device);
1515 invalidate_bdev(lo->lo_device);
1516 }
1517
1518 blk_mq_freeze_queue(lo->lo_queue);
1519
1520 /* invalidate_bdev should have truncated all the pages */
1521 if (lo->lo_queue->limits.logical_block_size != arg &&
1522 lo->lo_device->bd_inode->i_mapping->nrpages) {
1523 err = -EAGAIN;
1524 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1525 __func__, lo->lo_number, lo->lo_file_name,
1526 lo->lo_device->bd_inode->i_mapping->nrpages);
1527 goto out_unfreeze;
1528 }
1529
1530 blk_queue_logical_block_size(lo->lo_queue, arg);
1531 blk_queue_physical_block_size(lo->lo_queue, arg);
1532 blk_queue_io_min(lo->lo_queue, arg);
1533 loop_update_dio(lo);
1534 out_unfreeze:
1535 blk_mq_unfreeze_queue(lo->lo_queue);
1536
1537 return err;
1538 }
1539
1540 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1541 unsigned long arg)
1542 {
1543 int err;
1544
1545 err = mutex_lock_killable(&loop_ctl_mutex);
1546 if (err)
1547 return err;
1548 switch (cmd) {
1549 case LOOP_SET_CAPACITY:
1550 err = loop_set_capacity(lo);
1551 break;
1552 case LOOP_SET_DIRECT_IO:
1553 err = loop_set_dio(lo, arg);
1554 break;
1555 case LOOP_SET_BLOCK_SIZE:
1556 err = loop_set_block_size(lo, arg);
1557 break;
1558 default:
1559 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1560 }
1561 mutex_unlock(&loop_ctl_mutex);
1562 return err;
1563 }
1564
1565 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1566 unsigned int cmd, unsigned long arg)
1567 {
1568 struct loop_device *lo = bdev->bd_disk->private_data;
1569 int err;
1570
1571 switch (cmd) {
1572 case LOOP_SET_FD:
1573 return loop_set_fd(lo, mode, bdev, arg);
1574 case LOOP_CHANGE_FD:
1575 return loop_change_fd(lo, bdev, arg);
1576 case LOOP_CLR_FD:
1577 return loop_clr_fd(lo);
1578 case LOOP_SET_STATUS:
1579 err = -EPERM;
1580 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1581 err = loop_set_status_old(lo,
1582 (struct loop_info __user *)arg);
1583 }
1584 break;
1585 case LOOP_GET_STATUS:
1586 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1587 case LOOP_SET_STATUS64:
1588 err = -EPERM;
1589 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1590 err = loop_set_status64(lo,
1591 (struct loop_info64 __user *) arg);
1592 }
1593 break;
1594 case LOOP_GET_STATUS64:
1595 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1596 case LOOP_SET_CAPACITY:
1597 case LOOP_SET_DIRECT_IO:
1598 case LOOP_SET_BLOCK_SIZE:
1599 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1600 return -EPERM;
1601 /* Fall through */
1602 default:
1603 err = lo_simple_ioctl(lo, cmd, arg);
1604 break;
1605 }
1606
1607 return err;
1608 }
1609
1610 #ifdef CONFIG_COMPAT
1611 struct compat_loop_info {
1612 compat_int_t lo_number; /* ioctl r/o */
1613 compat_dev_t lo_device; /* ioctl r/o */
1614 compat_ulong_t lo_inode; /* ioctl r/o */
1615 compat_dev_t lo_rdevice; /* ioctl r/o */
1616 compat_int_t lo_offset;
1617 compat_int_t lo_encrypt_type;
1618 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1619 compat_int_t lo_flags; /* ioctl r/o */
1620 char lo_name[LO_NAME_SIZE];
1621 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1622 compat_ulong_t lo_init[2];
1623 char reserved[4];
1624 };
1625
1626 /*
1627 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1628 * - noinlined to reduce stack space usage in main part of driver
1629 */
1630 static noinline int
1631 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1632 struct loop_info64 *info64)
1633 {
1634 struct compat_loop_info info;
1635
1636 if (copy_from_user(&info, arg, sizeof(info)))
1637 return -EFAULT;
1638
1639 memset(info64, 0, sizeof(*info64));
1640 info64->lo_number = info.lo_number;
1641 info64->lo_device = info.lo_device;
1642 info64->lo_inode = info.lo_inode;
1643 info64->lo_rdevice = info.lo_rdevice;
1644 info64->lo_offset = info.lo_offset;
1645 info64->lo_sizelimit = 0;
1646 info64->lo_encrypt_type = info.lo_encrypt_type;
1647 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1648 info64->lo_flags = info.lo_flags;
1649 info64->lo_init[0] = info.lo_init[0];
1650 info64->lo_init[1] = info.lo_init[1];
1651 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1652 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1653 else
1654 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1655 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1656 return 0;
1657 }
1658
1659 /*
1660 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1661 * - noinlined to reduce stack space usage in main part of driver
1662 */
1663 static noinline int
1664 loop_info64_to_compat(const struct loop_info64 *info64,
1665 struct compat_loop_info __user *arg)
1666 {
1667 struct compat_loop_info info;
1668
1669 memset(&info, 0, sizeof(info));
1670 info.lo_number = info64->lo_number;
1671 info.lo_device = info64->lo_device;
1672 info.lo_inode = info64->lo_inode;
1673 info.lo_rdevice = info64->lo_rdevice;
1674 info.lo_offset = info64->lo_offset;
1675 info.lo_encrypt_type = info64->lo_encrypt_type;
1676 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1677 info.lo_flags = info64->lo_flags;
1678 info.lo_init[0] = info64->lo_init[0];
1679 info.lo_init[1] = info64->lo_init[1];
1680 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1681 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1682 else
1683 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1684 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1685
1686 /* error in case values were truncated */
1687 if (info.lo_device != info64->lo_device ||
1688 info.lo_rdevice != info64->lo_rdevice ||
1689 info.lo_inode != info64->lo_inode ||
1690 info.lo_offset != info64->lo_offset ||
1691 info.lo_init[0] != info64->lo_init[0] ||
1692 info.lo_init[1] != info64->lo_init[1])
1693 return -EOVERFLOW;
1694
1695 if (copy_to_user(arg, &info, sizeof(info)))
1696 return -EFAULT;
1697 return 0;
1698 }
1699
1700 static int
1701 loop_set_status_compat(struct loop_device *lo,
1702 const struct compat_loop_info __user *arg)
1703 {
1704 struct loop_info64 info64;
1705 int ret;
1706
1707 ret = loop_info64_from_compat(arg, &info64);
1708 if (ret < 0)
1709 return ret;
1710 return loop_set_status(lo, &info64);
1711 }
1712
1713 static int
1714 loop_get_status_compat(struct loop_device *lo,
1715 struct compat_loop_info __user *arg)
1716 {
1717 struct loop_info64 info64;
1718 int err;
1719
1720 if (!arg)
1721 return -EINVAL;
1722 err = loop_get_status(lo, &info64);
1723 if (!err)
1724 err = loop_info64_to_compat(&info64, arg);
1725 return err;
1726 }
1727
1728 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1729 unsigned int cmd, unsigned long arg)
1730 {
1731 struct loop_device *lo = bdev->bd_disk->private_data;
1732 int err;
1733
1734 switch(cmd) {
1735 case LOOP_SET_STATUS:
1736 err = loop_set_status_compat(lo,
1737 (const struct compat_loop_info __user *)arg);
1738 break;
1739 case LOOP_GET_STATUS:
1740 err = loop_get_status_compat(lo,
1741 (struct compat_loop_info __user *)arg);
1742 break;
1743 case LOOP_SET_CAPACITY:
1744 case LOOP_CLR_FD:
1745 case LOOP_GET_STATUS64:
1746 case LOOP_SET_STATUS64:
1747 arg = (unsigned long) compat_ptr(arg);
1748 /* fall through */
1749 case LOOP_SET_FD:
1750 case LOOP_CHANGE_FD:
1751 case LOOP_SET_BLOCK_SIZE:
1752 case LOOP_SET_DIRECT_IO:
1753 err = lo_ioctl(bdev, mode, cmd, arg);
1754 break;
1755 default:
1756 err = -ENOIOCTLCMD;
1757 break;
1758 }
1759 return err;
1760 }
1761 #endif
1762
1763 static int lo_open(struct block_device *bdev, fmode_t mode)
1764 {
1765 struct loop_device *lo;
1766 int err;
1767
1768 err = mutex_lock_killable(&loop_ctl_mutex);
1769 if (err)
1770 return err;
1771 lo = bdev->bd_disk->private_data;
1772 if (!lo) {
1773 err = -ENXIO;
1774 goto out;
1775 }
1776
1777 atomic_inc(&lo->lo_refcnt);
1778 out:
1779 mutex_unlock(&loop_ctl_mutex);
1780 return err;
1781 }
1782
1783 static void lo_release(struct gendisk *disk, fmode_t mode)
1784 {
1785 struct loop_device *lo;
1786
1787 mutex_lock(&loop_ctl_mutex);
1788 lo = disk->private_data;
1789 if (atomic_dec_return(&lo->lo_refcnt))
1790 goto out_unlock;
1791
1792 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1793 if (lo->lo_state != Lo_bound)
1794 goto out_unlock;
1795 lo->lo_state = Lo_rundown;
1796 mutex_unlock(&loop_ctl_mutex);
1797 /*
1798 * In autoclear mode, stop the loop thread
1799 * and remove configuration after last close.
1800 */
1801 __loop_clr_fd(lo, true);
1802 return;
1803 } else if (lo->lo_state == Lo_bound) {
1804 /*
1805 * Otherwise keep thread (if running) and config,
1806 * but flush possible ongoing bios in thread.
1807 */
1808 blk_mq_freeze_queue(lo->lo_queue);
1809 blk_mq_unfreeze_queue(lo->lo_queue);
1810 }
1811
1812 out_unlock:
1813 mutex_unlock(&loop_ctl_mutex);
1814 }
1815
1816 static const struct block_device_operations lo_fops = {
1817 .owner = THIS_MODULE,
1818 .open = lo_open,
1819 .release = lo_release,
1820 .ioctl = lo_ioctl,
1821 #ifdef CONFIG_COMPAT
1822 .compat_ioctl = lo_compat_ioctl,
1823 #endif
1824 };
1825
1826 /*
1827 * And now the modules code and kernel interface.
1828 */
1829 static int max_loop;
1830 module_param(max_loop, int, 0444);
1831 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1832 module_param(max_part, int, 0444);
1833 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1834 MODULE_LICENSE("GPL");
1835 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1836
1837 int loop_register_transfer(struct loop_func_table *funcs)
1838 {
1839 unsigned int n = funcs->number;
1840
1841 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1842 return -EINVAL;
1843 xfer_funcs[n] = funcs;
1844 return 0;
1845 }
1846
1847 static int unregister_transfer_cb(int id, void *ptr, void *data)
1848 {
1849 struct loop_device *lo = ptr;
1850 struct loop_func_table *xfer = data;
1851
1852 mutex_lock(&loop_ctl_mutex);
1853 if (lo->lo_encryption == xfer)
1854 loop_release_xfer(lo);
1855 mutex_unlock(&loop_ctl_mutex);
1856 return 0;
1857 }
1858
1859 int loop_unregister_transfer(int number)
1860 {
1861 unsigned int n = number;
1862 struct loop_func_table *xfer;
1863
1864 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1865 return -EINVAL;
1866
1867 xfer_funcs[n] = NULL;
1868 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1869 return 0;
1870 }
1871
1872 EXPORT_SYMBOL(loop_register_transfer);
1873 EXPORT_SYMBOL(loop_unregister_transfer);
1874
1875 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1876 const struct blk_mq_queue_data *bd)
1877 {
1878 struct request *rq = bd->rq;
1879 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1880 struct loop_device *lo = rq->q->queuedata;
1881
1882 blk_mq_start_request(rq);
1883
1884 if (lo->lo_state != Lo_bound)
1885 return BLK_STS_IOERR;
1886
1887 switch (req_op(rq)) {
1888 case REQ_OP_FLUSH:
1889 case REQ_OP_DISCARD:
1890 case REQ_OP_WRITE_ZEROES:
1891 cmd->use_aio = false;
1892 break;
1893 default:
1894 cmd->use_aio = lo->use_dio;
1895 break;
1896 }
1897
1898 /* always use the first bio's css */
1899 #ifdef CONFIG_BLK_CGROUP
1900 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1901 cmd->css = rq->bio->bi_css;
1902 css_get(cmd->css);
1903 } else
1904 #endif
1905 cmd->css = NULL;
1906 kthread_queue_work(&lo->worker, &cmd->work);
1907
1908 return BLK_STS_OK;
1909 }
1910
1911 static void loop_handle_cmd(struct loop_cmd *cmd)
1912 {
1913 struct request *rq = blk_mq_rq_from_pdu(cmd);
1914 const bool write = op_is_write(req_op(rq));
1915 struct loop_device *lo = rq->q->queuedata;
1916 int ret = 0;
1917
1918 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1919 ret = -EIO;
1920 goto failed;
1921 }
1922
1923 ret = do_req_filebacked(lo, rq);
1924 failed:
1925 /* complete non-aio request */
1926 if (!cmd->use_aio || ret) {
1927 cmd->ret = ret ? -EIO : 0;
1928 blk_mq_complete_request(rq);
1929 }
1930 }
1931
1932 static void loop_queue_work(struct kthread_work *work)
1933 {
1934 struct loop_cmd *cmd =
1935 container_of(work, struct loop_cmd, work);
1936
1937 loop_handle_cmd(cmd);
1938 }
1939
1940 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1941 unsigned int hctx_idx, unsigned int numa_node)
1942 {
1943 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1944
1945 kthread_init_work(&cmd->work, loop_queue_work);
1946 return 0;
1947 }
1948
1949 static const struct blk_mq_ops loop_mq_ops = {
1950 .queue_rq = loop_queue_rq,
1951 .init_request = loop_init_request,
1952 .complete = lo_complete_rq,
1953 };
1954
1955 static int loop_add(struct loop_device **l, int i)
1956 {
1957 struct loop_device *lo;
1958 struct gendisk *disk;
1959 int err;
1960
1961 err = -ENOMEM;
1962 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1963 if (!lo)
1964 goto out;
1965
1966 lo->lo_state = Lo_unbound;
1967
1968 /* allocate id, if @id >= 0, we're requesting that specific id */
1969 if (i >= 0) {
1970 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1971 if (err == -ENOSPC)
1972 err = -EEXIST;
1973 } else {
1974 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1975 }
1976 if (err < 0)
1977 goto out_free_dev;
1978 i = err;
1979
1980 err = -ENOMEM;
1981 lo->tag_set.ops = &loop_mq_ops;
1982 lo->tag_set.nr_hw_queues = 1;
1983 lo->tag_set.queue_depth = 128;
1984 lo->tag_set.numa_node = NUMA_NO_NODE;
1985 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1986 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1987 lo->tag_set.driver_data = lo;
1988
1989 err = blk_mq_alloc_tag_set(&lo->tag_set);
1990 if (err)
1991 goto out_free_idr;
1992
1993 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1994 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1995 err = PTR_ERR(lo->lo_queue);
1996 goto out_cleanup_tags;
1997 }
1998 lo->lo_queue->queuedata = lo;
1999
2000 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2001
2002 /*
2003 * By default, we do buffer IO, so it doesn't make sense to enable
2004 * merge because the I/O submitted to backing file is handled page by
2005 * page. For directio mode, merge does help to dispatch bigger request
2006 * to underlayer disk. We will enable merge once directio is enabled.
2007 */
2008 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2009
2010 err = -ENOMEM;
2011 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2012 if (!disk)
2013 goto out_free_queue;
2014
2015 /*
2016 * Disable partition scanning by default. The in-kernel partition
2017 * scanning can be requested individually per-device during its
2018 * setup. Userspace can always add and remove partitions from all
2019 * devices. The needed partition minors are allocated from the
2020 * extended minor space, the main loop device numbers will continue
2021 * to match the loop minors, regardless of the number of partitions
2022 * used.
2023 *
2024 * If max_part is given, partition scanning is globally enabled for
2025 * all loop devices. The minors for the main loop devices will be
2026 * multiples of max_part.
2027 *
2028 * Note: Global-for-all-devices, set-only-at-init, read-only module
2029 * parameteters like 'max_loop' and 'max_part' make things needlessly
2030 * complicated, are too static, inflexible and may surprise
2031 * userspace tools. Parameters like this in general should be avoided.
2032 */
2033 if (!part_shift)
2034 disk->flags |= GENHD_FL_NO_PART_SCAN;
2035 disk->flags |= GENHD_FL_EXT_DEVT;
2036 atomic_set(&lo->lo_refcnt, 0);
2037 lo->lo_number = i;
2038 spin_lock_init(&lo->lo_lock);
2039 disk->major = LOOP_MAJOR;
2040 disk->first_minor = i << part_shift;
2041 disk->fops = &lo_fops;
2042 disk->private_data = lo;
2043 disk->queue = lo->lo_queue;
2044 sprintf(disk->disk_name, "loop%d", i);
2045 add_disk(disk);
2046 *l = lo;
2047 return lo->lo_number;
2048
2049 out_free_queue:
2050 blk_cleanup_queue(lo->lo_queue);
2051 out_cleanup_tags:
2052 blk_mq_free_tag_set(&lo->tag_set);
2053 out_free_idr:
2054 idr_remove(&loop_index_idr, i);
2055 out_free_dev:
2056 kfree(lo);
2057 out:
2058 return err;
2059 }
2060
2061 static void loop_remove(struct loop_device *lo)
2062 {
2063 del_gendisk(lo->lo_disk);
2064 blk_cleanup_queue(lo->lo_queue);
2065 blk_mq_free_tag_set(&lo->tag_set);
2066 put_disk(lo->lo_disk);
2067 kfree(lo);
2068 }
2069
2070 static int find_free_cb(int id, void *ptr, void *data)
2071 {
2072 struct loop_device *lo = ptr;
2073 struct loop_device **l = data;
2074
2075 if (lo->lo_state == Lo_unbound) {
2076 *l = lo;
2077 return 1;
2078 }
2079 return 0;
2080 }
2081
2082 static int loop_lookup(struct loop_device **l, int i)
2083 {
2084 struct loop_device *lo;
2085 int ret = -ENODEV;
2086
2087 if (i < 0) {
2088 int err;
2089
2090 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2091 if (err == 1) {
2092 *l = lo;
2093 ret = lo->lo_number;
2094 }
2095 goto out;
2096 }
2097
2098 /* lookup and return a specific i */
2099 lo = idr_find(&loop_index_idr, i);
2100 if (lo) {
2101 *l = lo;
2102 ret = lo->lo_number;
2103 }
2104 out:
2105 return ret;
2106 }
2107
2108 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2109 {
2110 struct loop_device *lo;
2111 struct kobject *kobj;
2112 int err;
2113
2114 mutex_lock(&loop_ctl_mutex);
2115 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2116 if (err < 0)
2117 err = loop_add(&lo, MINOR(dev) >> part_shift);
2118 if (err < 0)
2119 kobj = NULL;
2120 else
2121 kobj = get_disk_and_module(lo->lo_disk);
2122 mutex_unlock(&loop_ctl_mutex);
2123
2124 *part = 0;
2125 return kobj;
2126 }
2127
2128 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2129 unsigned long parm)
2130 {
2131 struct loop_device *lo;
2132 int ret;
2133
2134 ret = mutex_lock_killable(&loop_ctl_mutex);
2135 if (ret)
2136 return ret;
2137
2138 ret = -ENOSYS;
2139 switch (cmd) {
2140 case LOOP_CTL_ADD:
2141 ret = loop_lookup(&lo, parm);
2142 if (ret >= 0) {
2143 ret = -EEXIST;
2144 break;
2145 }
2146 ret = loop_add(&lo, parm);
2147 break;
2148 case LOOP_CTL_REMOVE:
2149 ret = loop_lookup(&lo, parm);
2150 if (ret < 0)
2151 break;
2152 if (lo->lo_state != Lo_unbound) {
2153 ret = -EBUSY;
2154 break;
2155 }
2156 if (atomic_read(&lo->lo_refcnt) > 0) {
2157 ret = -EBUSY;
2158 break;
2159 }
2160 lo->lo_disk->private_data = NULL;
2161 idr_remove(&loop_index_idr, lo->lo_number);
2162 loop_remove(lo);
2163 break;
2164 case LOOP_CTL_GET_FREE:
2165 ret = loop_lookup(&lo, -1);
2166 if (ret >= 0)
2167 break;
2168 ret = loop_add(&lo, -1);
2169 }
2170 mutex_unlock(&loop_ctl_mutex);
2171
2172 return ret;
2173 }
2174
2175 static const struct file_operations loop_ctl_fops = {
2176 .open = nonseekable_open,
2177 .unlocked_ioctl = loop_control_ioctl,
2178 .compat_ioctl = loop_control_ioctl,
2179 .owner = THIS_MODULE,
2180 .llseek = noop_llseek,
2181 };
2182
2183 static struct miscdevice loop_misc = {
2184 .minor = LOOP_CTRL_MINOR,
2185 .name = "loop-control",
2186 .fops = &loop_ctl_fops,
2187 };
2188
2189 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2190 MODULE_ALIAS("devname:loop-control");
2191
2192 static int __init loop_init(void)
2193 {
2194 int i, nr;
2195 unsigned long range;
2196 struct loop_device *lo;
2197 int err;
2198
2199 part_shift = 0;
2200 if (max_part > 0) {
2201 part_shift = fls(max_part);
2202
2203 /*
2204 * Adjust max_part according to part_shift as it is exported
2205 * to user space so that user can decide correct minor number
2206 * if [s]he want to create more devices.
2207 *
2208 * Note that -1 is required because partition 0 is reserved
2209 * for the whole disk.
2210 */
2211 max_part = (1UL << part_shift) - 1;
2212 }
2213
2214 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2215 err = -EINVAL;
2216 goto err_out;
2217 }
2218
2219 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2220 err = -EINVAL;
2221 goto err_out;
2222 }
2223
2224 /*
2225 * If max_loop is specified, create that many devices upfront.
2226 * This also becomes a hard limit. If max_loop is not specified,
2227 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2228 * init time. Loop devices can be requested on-demand with the
2229 * /dev/loop-control interface, or be instantiated by accessing
2230 * a 'dead' device node.
2231 */
2232 if (max_loop) {
2233 nr = max_loop;
2234 range = max_loop << part_shift;
2235 } else {
2236 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2237 range = 1UL << MINORBITS;
2238 }
2239
2240 err = misc_register(&loop_misc);
2241 if (err < 0)
2242 goto err_out;
2243
2244
2245 if (register_blkdev(LOOP_MAJOR, "loop")) {
2246 err = -EIO;
2247 goto misc_out;
2248 }
2249
2250 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2251 THIS_MODULE, loop_probe, NULL, NULL);
2252
2253 /* pre-create number of devices given by config or max_loop */
2254 mutex_lock(&loop_ctl_mutex);
2255 for (i = 0; i < nr; i++)
2256 loop_add(&lo, i);
2257 mutex_unlock(&loop_ctl_mutex);
2258
2259 printk(KERN_INFO "loop: module loaded\n");
2260 return 0;
2261
2262 misc_out:
2263 misc_deregister(&loop_misc);
2264 err_out:
2265 return err;
2266 }
2267
2268 static int loop_exit_cb(int id, void *ptr, void *data)
2269 {
2270 struct loop_device *lo = ptr;
2271
2272 loop_remove(lo);
2273 return 0;
2274 }
2275
2276 static void __exit loop_exit(void)
2277 {
2278 unsigned long range;
2279
2280 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2281
2282 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2283 idr_destroy(&loop_index_idr);
2284
2285 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2286 unregister_blkdev(LOOP_MAJOR, "loop");
2287
2288 misc_deregister(&loop_misc);
2289 }
2290
2291 module_init(loop_init);
2292 module_exit(loop_exit);
2293
2294 #ifndef MODULE
2295 static int __init max_loop_setup(char *str)
2296 {
2297 max_loop = simple_strtol(str, NULL, 0);
2298 return 1;
2299 }
2300
2301 __setup("max_loop=", max_loop_setup);
2302 #endif
Linux with added FPC-III support