]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - drivers/block/loop.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xen/events: Fix race in set_evtchn_to_irq
[mirror_ubuntu-hirsute-kernel.git] / drivers / block / loop.c
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 #include <linux/blk-cgroup.h>
81
82 #include "loop.h"
83
84 #include <linux/uaccess.h>
85
86 static DEFINE_IDR(loop_index_idr);
87 static DEFINE_MUTEX(loop_ctl_mutex);
88
89 static int max_part;
90 static int part_shift;
91
92 static 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
121 static 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
128 static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
130 };
131
132 static 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 */
139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140 &none_funcs,
141 &xor_funcs
142 };
143
144 static 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
165 static 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
170 static 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 if (lo->lo_state == Lo_bound)
218 blk_mq_freeze_queue(lo->lo_queue);
219 lo->use_dio = use_dio;
220 if (use_dio) {
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
223 } else {
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
226 }
227 if (lo->lo_state == Lo_bound)
228 blk_mq_unfreeze_queue(lo->lo_queue);
229 }
230
231 /**
232 * loop_validate_block_size() - validates the passed in block size
233 * @bsize: size to validate
234 */
235 static int
236 loop_validate_block_size(unsigned short bsize)
237 {
238 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
239 return -EINVAL;
240
241 return 0;
242 }
243
244 /**
245 * loop_set_size() - sets device size and notifies userspace
246 * @lo: struct loop_device to set the size for
247 * @size: new size of the loop device
248 *
249 * Callers must validate that the size passed into this function fits into
250 * a sector_t, eg using loop_validate_size()
251 */
252 static void loop_set_size(struct loop_device *lo, loff_t size)
253 {
254 if (!set_capacity_and_notify(lo->lo_disk, size))
255 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
256 }
257
258 static inline int
259 lo_do_transfer(struct loop_device *lo, int cmd,
260 struct page *rpage, unsigned roffs,
261 struct page *lpage, unsigned loffs,
262 int size, sector_t rblock)
263 {
264 int ret;
265
266 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
267 if (likely(!ret))
268 return 0;
269
270 printk_ratelimited(KERN_ERR
271 "loop: Transfer error at byte offset %llu, length %i.\n",
272 (unsigned long long)rblock << 9, size);
273 return ret;
274 }
275
276 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
277 {
278 struct iov_iter i;
279 ssize_t bw;
280
281 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
282
283 file_start_write(file);
284 bw = vfs_iter_write(file, &i, ppos, 0);
285 file_end_write(file);
286
287 if (likely(bw == bvec->bv_len))
288 return 0;
289
290 printk_ratelimited(KERN_ERR
291 "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)*ppos, bvec->bv_len);
293 if (bw >= 0)
294 bw = -EIO;
295 return bw;
296 }
297
298 static int lo_write_simple(struct loop_device *lo, struct request *rq,
299 loff_t pos)
300 {
301 struct bio_vec bvec;
302 struct req_iterator iter;
303 int ret = 0;
304
305 rq_for_each_segment(bvec, rq, iter) {
306 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
307 if (ret < 0)
308 break;
309 cond_resched();
310 }
311
312 return ret;
313 }
314
315 /*
316 * This is the slow, transforming version that needs to double buffer the
317 * data as it cannot do the transformations in place without having direct
318 * access to the destination pages of the backing file.
319 */
320 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
321 loff_t pos)
322 {
323 struct bio_vec bvec, b;
324 struct req_iterator iter;
325 struct page *page;
326 int ret = 0;
327
328 page = alloc_page(GFP_NOIO);
329 if (unlikely(!page))
330 return -ENOMEM;
331
332 rq_for_each_segment(bvec, rq, iter) {
333 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
334 bvec.bv_offset, bvec.bv_len, pos >> 9);
335 if (unlikely(ret))
336 break;
337
338 b.bv_page = page;
339 b.bv_offset = 0;
340 b.bv_len = bvec.bv_len;
341 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
342 if (ret < 0)
343 break;
344 }
345
346 __free_page(page);
347 return ret;
348 }
349
350 static int lo_read_simple(struct loop_device *lo, struct request *rq,
351 loff_t pos)
352 {
353 struct bio_vec bvec;
354 struct req_iterator iter;
355 struct iov_iter i;
356 ssize_t len;
357
358 rq_for_each_segment(bvec, rq, iter) {
359 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
360 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
361 if (len < 0)
362 return len;
363
364 flush_dcache_page(bvec.bv_page);
365
366 if (len != bvec.bv_len) {
367 struct bio *bio;
368
369 __rq_for_each_bio(bio, rq)
370 zero_fill_bio(bio);
371 break;
372 }
373 cond_resched();
374 }
375
376 return 0;
377 }
378
379 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
380 loff_t pos)
381 {
382 struct bio_vec bvec, b;
383 struct req_iterator iter;
384 struct iov_iter i;
385 struct page *page;
386 ssize_t len;
387 int ret = 0;
388
389 page = alloc_page(GFP_NOIO);
390 if (unlikely(!page))
391 return -ENOMEM;
392
393 rq_for_each_segment(bvec, rq, iter) {
394 loff_t offset = pos;
395
396 b.bv_page = page;
397 b.bv_offset = 0;
398 b.bv_len = bvec.bv_len;
399
400 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
401 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
402 if (len < 0) {
403 ret = len;
404 goto out_free_page;
405 }
406
407 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
408 bvec.bv_offset, len, offset >> 9);
409 if (ret)
410 goto out_free_page;
411
412 flush_dcache_page(bvec.bv_page);
413
414 if (len != bvec.bv_len) {
415 struct bio *bio;
416
417 __rq_for_each_bio(bio, rq)
418 zero_fill_bio(bio);
419 break;
420 }
421 }
422
423 ret = 0;
424 out_free_page:
425 __free_page(page);
426 return ret;
427 }
428
429 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
430 int mode)
431 {
432 /*
433 * We use fallocate to manipulate the space mappings used by the image
434 * a.k.a. discard/zerorange. However we do not support this if
435 * encryption is enabled, because it may give an attacker useful
436 * information.
437 */
438 struct file *file = lo->lo_backing_file;
439 struct request_queue *q = lo->lo_queue;
440 int ret;
441
442 mode |= FALLOC_FL_KEEP_SIZE;
443
444 if (!blk_queue_discard(q)) {
445 ret = -EOPNOTSUPP;
446 goto out;
447 }
448
449 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
450 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
451 ret = -EIO;
452 out:
453 return ret;
454 }
455
456 static int lo_req_flush(struct loop_device *lo, struct request *rq)
457 {
458 struct file *file = lo->lo_backing_file;
459 int ret = vfs_fsync(file, 0);
460 if (unlikely(ret && ret != -EINVAL))
461 ret = -EIO;
462
463 return ret;
464 }
465
466 static void lo_complete_rq(struct request *rq)
467 {
468 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
469 blk_status_t ret = BLK_STS_OK;
470
471 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
472 req_op(rq) != REQ_OP_READ) {
473 if (cmd->ret < 0)
474 ret = errno_to_blk_status(cmd->ret);
475 goto end_io;
476 }
477
478 /*
479 * Short READ - if we got some data, advance our request and
480 * retry it. If we got no data, end the rest with EIO.
481 */
482 if (cmd->ret) {
483 blk_update_request(rq, BLK_STS_OK, cmd->ret);
484 cmd->ret = 0;
485 blk_mq_requeue_request(rq, true);
486 } else {
487 if (cmd->use_aio) {
488 struct bio *bio = rq->bio;
489
490 while (bio) {
491 zero_fill_bio(bio);
492 bio = bio->bi_next;
493 }
494 }
495 ret = BLK_STS_IOERR;
496 end_io:
497 blk_mq_end_request(rq, ret);
498 }
499 }
500
501 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
502 {
503 struct request *rq = blk_mq_rq_from_pdu(cmd);
504
505 if (!atomic_dec_and_test(&cmd->ref))
506 return;
507 kfree(cmd->bvec);
508 cmd->bvec = NULL;
509 if (likely(!blk_should_fake_timeout(rq->q)))
510 blk_mq_complete_request(rq);
511 }
512
513 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
514 {
515 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
516
517 if (cmd->css)
518 css_put(cmd->css);
519 cmd->ret = ret;
520 lo_rw_aio_do_completion(cmd);
521 }
522
523 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
524 loff_t pos, bool rw)
525 {
526 struct iov_iter iter;
527 struct req_iterator rq_iter;
528 struct bio_vec *bvec;
529 struct request *rq = blk_mq_rq_from_pdu(cmd);
530 struct bio *bio = rq->bio;
531 struct file *file = lo->lo_backing_file;
532 struct bio_vec tmp;
533 unsigned int offset;
534 int nr_bvec = 0;
535 int ret;
536
537 rq_for_each_bvec(tmp, rq, rq_iter)
538 nr_bvec++;
539
540 if (rq->bio != rq->biotail) {
541
542 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
543 GFP_NOIO);
544 if (!bvec)
545 return -EIO;
546 cmd->bvec = bvec;
547
548 /*
549 * The bios of the request may be started from the middle of
550 * the 'bvec' because of bio splitting, so we can't directly
551 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
552 * API will take care of all details for us.
553 */
554 rq_for_each_bvec(tmp, rq, rq_iter) {
555 *bvec = tmp;
556 bvec++;
557 }
558 bvec = cmd->bvec;
559 offset = 0;
560 } else {
561 /*
562 * Same here, this bio may be started from the middle of the
563 * 'bvec' because of bio splitting, so offset from the bvec
564 * must be passed to iov iterator
565 */
566 offset = bio->bi_iter.bi_bvec_done;
567 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
568 }
569 atomic_set(&cmd->ref, 2);
570
571 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
572 iter.iov_offset = offset;
573
574 cmd->iocb.ki_pos = pos;
575 cmd->iocb.ki_filp = file;
576 cmd->iocb.ki_complete = lo_rw_aio_complete;
577 cmd->iocb.ki_flags = IOCB_DIRECT;
578 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
579 if (cmd->css)
580 kthread_associate_blkcg(cmd->css);
581
582 if (rw == WRITE)
583 ret = call_write_iter(file, &cmd->iocb, &iter);
584 else
585 ret = call_read_iter(file, &cmd->iocb, &iter);
586
587 lo_rw_aio_do_completion(cmd);
588 kthread_associate_blkcg(NULL);
589
590 if (ret != -EIOCBQUEUED)
591 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
592 return 0;
593 }
594
595 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
596 {
597 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
598 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
599
600 /*
601 * lo_write_simple and lo_read_simple should have been covered
602 * by io submit style function like lo_rw_aio(), one blocker
603 * is that lo_read_simple() need to call flush_dcache_page after
604 * the page is written from kernel, and it isn't easy to handle
605 * this in io submit style function which submits all segments
606 * of the req at one time. And direct read IO doesn't need to
607 * run flush_dcache_page().
608 */
609 switch (req_op(rq)) {
610 case REQ_OP_FLUSH:
611 return lo_req_flush(lo, rq);
612 case REQ_OP_WRITE_ZEROES:
613 /*
614 * If the caller doesn't want deallocation, call zeroout to
615 * write zeroes the range. Otherwise, punch them out.
616 */
617 return lo_fallocate(lo, rq, pos,
618 (rq->cmd_flags & REQ_NOUNMAP) ?
619 FALLOC_FL_ZERO_RANGE :
620 FALLOC_FL_PUNCH_HOLE);
621 case REQ_OP_DISCARD:
622 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
623 case REQ_OP_WRITE:
624 if (lo->transfer)
625 return lo_write_transfer(lo, rq, pos);
626 else if (cmd->use_aio)
627 return lo_rw_aio(lo, cmd, pos, WRITE);
628 else
629 return lo_write_simple(lo, rq, pos);
630 case REQ_OP_READ:
631 if (lo->transfer)
632 return lo_read_transfer(lo, rq, pos);
633 else if (cmd->use_aio)
634 return lo_rw_aio(lo, cmd, pos, READ);
635 else
636 return lo_read_simple(lo, rq, pos);
637 default:
638 WARN_ON_ONCE(1);
639 return -EIO;
640 }
641 }
642
643 static inline void loop_update_dio(struct loop_device *lo)
644 {
645 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
646 lo->use_dio);
647 }
648
649 static struct file *loop_real_file(struct file *file)
650 {
651 struct file *f = NULL;
652
653 if (file->f_path.dentry->d_sb->s_op->real_loop)
654 f = file->f_path.dentry->d_sb->s_op->real_loop(file);
655 return f;
656 }
657
658 static void loop_reread_partitions(struct loop_device *lo,
659 struct block_device *bdev)
660 {
661 int rc;
662
663 mutex_lock(&bdev->bd_mutex);
664 rc = bdev_disk_changed(bdev, false);
665 mutex_unlock(&bdev->bd_mutex);
666 if (rc)
667 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
668 __func__, lo->lo_number, lo->lo_file_name, rc);
669 }
670
671 static inline int is_loop_device(struct file *file)
672 {
673 struct inode *i = file->f_mapping->host;
674
675 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
676 }
677
678 static int loop_validate_file(struct file *file, struct block_device *bdev)
679 {
680 struct inode *inode = file->f_mapping->host;
681 struct file *f = file;
682
683 /* Avoid recursion */
684 while (is_loop_device(f)) {
685 struct loop_device *l;
686
687 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
688 return -EBADF;
689
690 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
691 if (l->lo_state != Lo_bound) {
692 return -EINVAL;
693 }
694 f = l->lo_backing_file;
695 }
696 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
697 return -EINVAL;
698 return 0;
699 }
700
701 /*
702 * loop_change_fd switched the backing store of a loopback device to
703 * a new file. This is useful for operating system installers to free up
704 * the original file and in High Availability environments to switch to
705 * an alternative location for the content in case of server meltdown.
706 * This can only work if the loop device is used read-only, and if the
707 * new backing store is the same size and type as the old backing store.
708 */
709 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
710 unsigned int arg)
711 {
712 struct file *file = NULL, *old_file;
713 struct file *f, *virt_file = NULL, *old_virt_file;
714 int error;
715 bool partscan;
716
717 error = mutex_lock_killable(&loop_ctl_mutex);
718 if (error)
719 return error;
720 error = -ENXIO;
721 if (lo->lo_state != Lo_bound)
722 goto out_err;
723
724 /* the loop device has to be read-only */
725 error = -EINVAL;
726 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
727 goto out_err;
728
729 error = -EBADF;
730 file = fget(arg);
731 if (!file)
732 goto out_err;
733 f = loop_real_file(file);
734 if (f) {
735 virt_file = file;
736 file = f;
737 get_file(file);
738 }
739
740 error = loop_validate_file(file, bdev);
741 if (error)
742 goto out_err;
743
744 old_file = lo->lo_backing_file;
745 old_virt_file = lo->lo_backing_virt_file;
746
747 error = -EINVAL;
748
749 /* size of the new backing store needs to be the same */
750 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
751 goto out_err;
752
753 /* and ... switch */
754 blk_mq_freeze_queue(lo->lo_queue);
755 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
756 lo->lo_backing_file = file;
757 lo->lo_backing_virt_file = virt_file;
758 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
759 mapping_set_gfp_mask(file->f_mapping,
760 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
761 loop_update_dio(lo);
762 blk_mq_unfreeze_queue(lo->lo_queue);
763 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
764 mutex_unlock(&loop_ctl_mutex);
765 /*
766 * We must drop file reference outside of loop_ctl_mutex as dropping
767 * the file ref can take bd_mutex which creates circular locking
768 * dependency.
769 */
770 fput(old_file);
771 if (old_virt_file)
772 fput(old_virt_file);
773 if (partscan)
774 loop_reread_partitions(lo, bdev);
775 return 0;
776
777 out_err:
778 mutex_unlock(&loop_ctl_mutex);
779 if (file)
780 fput(file);
781 if (virt_file)
782 fput(virt_file);
783 return error;
784 }
785
786 /*
787 * for AUFS
788 * no get/put for file.
789 */
790 struct file *loop_backing_file(struct super_block *sb)
791 {
792 struct file *ret;
793 struct loop_device *l;
794
795 ret = NULL;
796 if (MAJOR(sb->s_dev) == LOOP_MAJOR) {
797 l = sb->s_bdev->bd_disk->private_data;
798 ret = l->lo_backing_file;
799 }
800 return ret;
801 }
802 EXPORT_SYMBOL_GPL(loop_backing_file);
803
804 /* loop sysfs attributes */
805
806 static ssize_t loop_attr_show(struct device *dev, char *page,
807 ssize_t (*callback)(struct loop_device *, char *))
808 {
809 struct gendisk *disk = dev_to_disk(dev);
810 struct loop_device *lo = disk->private_data;
811
812 return callback(lo, page);
813 }
814
815 #define LOOP_ATTR_RO(_name) \
816 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
817 static ssize_t loop_attr_do_show_##_name(struct device *d, \
818 struct device_attribute *attr, char *b) \
819 { \
820 return loop_attr_show(d, b, loop_attr_##_name##_show); \
821 } \
822 static struct device_attribute loop_attr_##_name = \
823 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
824
825 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
826 {
827 ssize_t ret;
828 char *p = NULL;
829
830 spin_lock_irq(&lo->lo_lock);
831 if (lo->lo_backing_file)
832 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
833 spin_unlock_irq(&lo->lo_lock);
834
835 if (IS_ERR_OR_NULL(p))
836 ret = PTR_ERR(p);
837 else {
838 ret = strlen(p);
839 memmove(buf, p, ret);
840 buf[ret++] = '\n';
841 buf[ret] = 0;
842 }
843
844 return ret;
845 }
846
847 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
848 {
849 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
850 }
851
852 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
853 {
854 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
855 }
856
857 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
858 {
859 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
860
861 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
862 }
863
864 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
865 {
866 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
867
868 return sprintf(buf, "%s\n", partscan ? "1" : "0");
869 }
870
871 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
872 {
873 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
874
875 return sprintf(buf, "%s\n", dio ? "1" : "0");
876 }
877
878 LOOP_ATTR_RO(backing_file);
879 LOOP_ATTR_RO(offset);
880 LOOP_ATTR_RO(sizelimit);
881 LOOP_ATTR_RO(autoclear);
882 LOOP_ATTR_RO(partscan);
883 LOOP_ATTR_RO(dio);
884
885 static struct attribute *loop_attrs[] = {
886 &loop_attr_backing_file.attr,
887 &loop_attr_offset.attr,
888 &loop_attr_sizelimit.attr,
889 &loop_attr_autoclear.attr,
890 &loop_attr_partscan.attr,
891 &loop_attr_dio.attr,
892 NULL,
893 };
894
895 static struct attribute_group loop_attribute_group = {
896 .name = "loop",
897 .attrs= loop_attrs,
898 };
899
900 static void loop_sysfs_init(struct loop_device *lo)
901 {
902 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
903 &loop_attribute_group);
904 }
905
906 static void loop_sysfs_exit(struct loop_device *lo)
907 {
908 if (lo->sysfs_inited)
909 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
910 &loop_attribute_group);
911 }
912
913 static void loop_config_discard(struct loop_device *lo)
914 {
915 struct file *file = lo->lo_backing_file;
916 struct inode *inode = file->f_mapping->host;
917 struct request_queue *q = lo->lo_queue;
918 u32 granularity, max_discard_sectors;
919
920 /*
921 * If the backing device is a block device, mirror its zeroing
922 * capability. Set the discard sectors to the block device's zeroing
923 * capabilities because loop discards result in blkdev_issue_zeroout(),
924 * not blkdev_issue_discard(). This maintains consistent behavior with
925 * file-backed loop devices: discarded regions read back as zero.
926 */
927 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
928 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
929
930 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
931 granularity = backingq->limits.discard_granularity ?:
932 queue_physical_block_size(backingq);
933
934 /*
935 * We use punch hole to reclaim the free space used by the
936 * image a.k.a. discard. However we do not support discard if
937 * encryption is enabled, because it may give an attacker
938 * useful information.
939 */
940 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
941 max_discard_sectors = 0;
942 granularity = 0;
943
944 } else {
945 max_discard_sectors = UINT_MAX >> 9;
946 granularity = inode->i_sb->s_blocksize;
947 }
948
949 if (max_discard_sectors) {
950 q->limits.discard_granularity = granularity;
951 blk_queue_max_discard_sectors(q, max_discard_sectors);
952 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
953 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
954 } else {
955 q->limits.discard_granularity = 0;
956 blk_queue_max_discard_sectors(q, 0);
957 blk_queue_max_write_zeroes_sectors(q, 0);
958 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
959 }
960 q->limits.discard_alignment = 0;
961 }
962
963 static void loop_unprepare_queue(struct loop_device *lo)
964 {
965 kthread_flush_worker(&lo->worker);
966 kthread_stop(lo->worker_task);
967 }
968
969 static int loop_kthread_worker_fn(void *worker_ptr)
970 {
971 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
972 return kthread_worker_fn(worker_ptr);
973 }
974
975 static int loop_prepare_queue(struct loop_device *lo)
976 {
977 kthread_init_worker(&lo->worker);
978 lo->worker_task = kthread_run(loop_kthread_worker_fn,
979 &lo->worker, "loop%d", lo->lo_number);
980 if (IS_ERR(lo->worker_task))
981 return -ENOMEM;
982 set_user_nice(lo->worker_task, MIN_NICE);
983 return 0;
984 }
985
986 static void loop_update_rotational(struct loop_device *lo)
987 {
988 struct file *file = lo->lo_backing_file;
989 struct inode *file_inode = file->f_mapping->host;
990 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
991 struct request_queue *q = lo->lo_queue;
992 bool nonrot = true;
993
994 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
995 if (file_bdev)
996 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
997
998 if (nonrot)
999 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1000 else
1001 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1002 }
1003
1004 static int
1005 loop_release_xfer(struct loop_device *lo)
1006 {
1007 int err = 0;
1008 struct loop_func_table *xfer = lo->lo_encryption;
1009
1010 if (xfer) {
1011 if (xfer->release)
1012 err = xfer->release(lo);
1013 lo->transfer = NULL;
1014 lo->lo_encryption = NULL;
1015 module_put(xfer->owner);
1016 }
1017 return err;
1018 }
1019
1020 static int
1021 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1022 const struct loop_info64 *i)
1023 {
1024 int err = 0;
1025
1026 if (xfer) {
1027 struct module *owner = xfer->owner;
1028
1029 if (!try_module_get(owner))
1030 return -EINVAL;
1031 if (xfer->init)
1032 err = xfer->init(lo, i);
1033 if (err)
1034 module_put(owner);
1035 else
1036 lo->lo_encryption = xfer;
1037 }
1038 return err;
1039 }
1040
1041 /**
1042 * loop_set_status_from_info - configure device from loop_info
1043 * @lo: struct loop_device to configure
1044 * @info: struct loop_info64 to configure the device with
1045 *
1046 * Configures the loop device parameters according to the passed
1047 * in loop_info64 configuration.
1048 */
1049 static int
1050 loop_set_status_from_info(struct loop_device *lo,
1051 const struct loop_info64 *info)
1052 {
1053 int err;
1054 struct loop_func_table *xfer;
1055 kuid_t uid = current_uid();
1056
1057 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1058 return -EINVAL;
1059
1060 err = loop_release_xfer(lo);
1061 if (err)
1062 return err;
1063
1064 if (info->lo_encrypt_type) {
1065 unsigned int type = info->lo_encrypt_type;
1066
1067 if (type >= MAX_LO_CRYPT)
1068 return -EINVAL;
1069 xfer = xfer_funcs[type];
1070 if (xfer == NULL)
1071 return -EINVAL;
1072 } else
1073 xfer = NULL;
1074
1075 err = loop_init_xfer(lo, xfer, info);
1076 if (err)
1077 return err;
1078
1079 lo->lo_offset = info->lo_offset;
1080 lo->lo_sizelimit = info->lo_sizelimit;
1081 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1082 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1083 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1084 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1085
1086 if (!xfer)
1087 xfer = &none_funcs;
1088 lo->transfer = xfer->transfer;
1089 lo->ioctl = xfer->ioctl;
1090
1091 lo->lo_flags = info->lo_flags;
1092
1093 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1094 lo->lo_init[0] = info->lo_init[0];
1095 lo->lo_init[1] = info->lo_init[1];
1096 if (info->lo_encrypt_key_size) {
1097 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1098 info->lo_encrypt_key_size);
1099 lo->lo_key_owner = uid;
1100 }
1101
1102 return 0;
1103 }
1104
1105 static int loop_configure(struct loop_device *lo, fmode_t mode,
1106 struct block_device *bdev,
1107 const struct loop_config *config)
1108 {
1109 struct file *file, *f, *virt_file = NULL;
1110 struct inode *inode;
1111 struct address_space *mapping;
1112 int error;
1113 loff_t size;
1114 bool partscan;
1115 unsigned short bsize;
1116
1117 /* This is safe, since we have a reference from open(). */
1118 __module_get(THIS_MODULE);
1119
1120 error = -EBADF;
1121 file = fget(config->fd);
1122 if (!file)
1123 goto out;
1124 f = loop_real_file(file);
1125 if (f) {
1126 virt_file = file;
1127 file = f;
1128 get_file(file);
1129 }
1130
1131 /*
1132 * If we don't hold exclusive handle for the device, upgrade to it
1133 * here to avoid changing device under exclusive owner.
1134 */
1135 if (!(mode & FMODE_EXCL)) {
1136 error = bd_prepare_to_claim(bdev, loop_configure);
1137 if (error)
1138 goto out_putf;
1139 }
1140
1141 error = mutex_lock_killable(&loop_ctl_mutex);
1142 if (error)
1143 goto out_bdev;
1144
1145 error = -EBUSY;
1146 if (lo->lo_state != Lo_unbound)
1147 goto out_unlock;
1148
1149 error = loop_validate_file(file, bdev);
1150 if (error)
1151 goto out_unlock;
1152
1153 mapping = file->f_mapping;
1154 inode = mapping->host;
1155
1156 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1157 error = -EINVAL;
1158 goto out_unlock;
1159 }
1160
1161 if (config->block_size) {
1162 error = loop_validate_block_size(config->block_size);
1163 if (error)
1164 goto out_unlock;
1165 }
1166
1167 error = loop_set_status_from_info(lo, &config->info);
1168 if (error)
1169 goto out_unlock;
1170
1171 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1172 !file->f_op->write_iter)
1173 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1174
1175 error = loop_prepare_queue(lo);
1176 if (error)
1177 goto out_unlock;
1178
1179 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1180
1181 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1182 lo->lo_device = bdev;
1183 lo->lo_backing_file = file;
1184 lo->lo_backing_virt_file = virt_file;
1185 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1186 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1187
1188 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1189 blk_queue_write_cache(lo->lo_queue, true, false);
1190
1191 if (config->block_size)
1192 bsize = config->block_size;
1193 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1194 /* In case of direct I/O, match underlying block size */
1195 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1196 else
1197 bsize = 512;
1198
1199 blk_queue_logical_block_size(lo->lo_queue, bsize);
1200 blk_queue_physical_block_size(lo->lo_queue, bsize);
1201 blk_queue_io_min(lo->lo_queue, bsize);
1202
1203 loop_config_discard(lo);
1204 loop_update_rotational(lo);
1205 loop_update_dio(lo);
1206 loop_sysfs_init(lo);
1207
1208 size = get_loop_size(lo, file);
1209 loop_set_size(lo, size);
1210
1211 lo->lo_state = Lo_bound;
1212 if (part_shift)
1213 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1214 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1215 if (partscan)
1216 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1217
1218 /* Grab the block_device to prevent its destruction after we
1219 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1220 */
1221 bdgrab(bdev);
1222 mutex_unlock(&loop_ctl_mutex);
1223 if (partscan)
1224 loop_reread_partitions(lo, bdev);
1225 if (!(mode & FMODE_EXCL))
1226 bd_abort_claiming(bdev, loop_configure);
1227 return 0;
1228
1229 out_unlock:
1230 mutex_unlock(&loop_ctl_mutex);
1231 out_bdev:
1232 if (!(mode & FMODE_EXCL))
1233 bd_abort_claiming(bdev, loop_configure);
1234 out_putf:
1235 fput(file);
1236 if (virt_file)
1237 fput(virt_file);
1238 out:
1239 /* This is safe: open() is still holding a reference. */
1240 module_put(THIS_MODULE);
1241 return error;
1242 }
1243
1244 static int __loop_clr_fd(struct loop_device *lo, bool release)
1245 {
1246 struct file *filp = NULL;
1247 struct file *virt_filp = lo->lo_backing_virt_file;
1248 gfp_t gfp = lo->old_gfp_mask;
1249 struct block_device *bdev = lo->lo_device;
1250 int err = 0;
1251 bool partscan = false;
1252 int lo_number;
1253
1254 mutex_lock(&loop_ctl_mutex);
1255 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1256 err = -ENXIO;
1257 goto out_unlock;
1258 }
1259
1260 filp = lo->lo_backing_file;
1261 if (filp == NULL) {
1262 err = -EINVAL;
1263 goto out_unlock;
1264 }
1265
1266 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1267 blk_queue_write_cache(lo->lo_queue, false, false);
1268
1269 /* freeze request queue during the transition */
1270 blk_mq_freeze_queue(lo->lo_queue);
1271
1272 spin_lock_irq(&lo->lo_lock);
1273 lo->lo_backing_file = NULL;
1274 lo->lo_backing_virt_file = NULL;
1275 spin_unlock_irq(&lo->lo_lock);
1276
1277 loop_release_xfer(lo);
1278 lo->transfer = NULL;
1279 lo->ioctl = NULL;
1280 lo->lo_device = NULL;
1281 lo->lo_encryption = NULL;
1282 lo->lo_offset = 0;
1283 lo->lo_sizelimit = 0;
1284 lo->lo_encrypt_key_size = 0;
1285 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1286 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1287 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1288 blk_queue_logical_block_size(lo->lo_queue, 512);
1289 blk_queue_physical_block_size(lo->lo_queue, 512);
1290 blk_queue_io_min(lo->lo_queue, 512);
1291 if (bdev) {
1292 bdput(bdev);
1293 invalidate_bdev(bdev);
1294 bdev->bd_inode->i_mapping->wb_err = 0;
1295 }
1296 set_capacity(lo->lo_disk, 0);
1297 loop_sysfs_exit(lo);
1298 if (bdev) {
1299 /* let user-space know about this change */
1300 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1301 }
1302 mapping_set_gfp_mask(filp->f_mapping, gfp);
1303 /* This is safe: open() is still holding a reference. */
1304 module_put(THIS_MODULE);
1305 blk_mq_unfreeze_queue(lo->lo_queue);
1306
1307 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1308 lo_number = lo->lo_number;
1309 loop_unprepare_queue(lo);
1310 out_unlock:
1311 mutex_unlock(&loop_ctl_mutex);
1312 if (partscan) {
1313 /*
1314 * bd_mutex has been held already in release path, so don't
1315 * acquire it if this function is called in such case.
1316 *
1317 * If the reread partition isn't from release path, lo_refcnt
1318 * must be at least one and it can only become zero when the
1319 * current holder is released.
1320 */
1321 if (!release)
1322 mutex_lock(&bdev->bd_mutex);
1323 err = bdev_disk_changed(bdev, false);
1324 if (!release)
1325 mutex_unlock(&bdev->bd_mutex);
1326 if (err)
1327 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1328 __func__, lo_number, err);
1329 /* Device is gone, no point in returning error */
1330 err = 0;
1331 }
1332
1333 /*
1334 * lo->lo_state is set to Lo_unbound here after above partscan has
1335 * finished.
1336 *
1337 * There cannot be anybody else entering __loop_clr_fd() as
1338 * lo->lo_backing_file is already cleared and Lo_rundown state
1339 * protects us from all the other places trying to change the 'lo'
1340 * device.
1341 */
1342 mutex_lock(&loop_ctl_mutex);
1343 lo->lo_flags = 0;
1344 if (!part_shift)
1345 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1346 lo->lo_state = Lo_unbound;
1347 mutex_unlock(&loop_ctl_mutex);
1348
1349 /*
1350 * Need not hold loop_ctl_mutex to fput backing file.
1351 * Calling fput holding loop_ctl_mutex triggers a circular
1352 * lock dependency possibility warning as fput can take
1353 * bd_mutex which is usually taken before loop_ctl_mutex.
1354 */
1355 if (filp)
1356 fput(filp);
1357 if (virt_filp)
1358 fput(virt_filp);
1359 return err;
1360 }
1361
1362 static int loop_clr_fd(struct loop_device *lo)
1363 {
1364 int err;
1365
1366 err = mutex_lock_killable(&loop_ctl_mutex);
1367 if (err)
1368 return err;
1369 if (lo->lo_state != Lo_bound) {
1370 mutex_unlock(&loop_ctl_mutex);
1371 return -ENXIO;
1372 }
1373 /*
1374 * If we've explicitly asked to tear down the loop device,
1375 * and it has an elevated reference count, set it for auto-teardown when
1376 * the last reference goes away. This stops $!~#$@ udev from
1377 * preventing teardown because it decided that it needs to run blkid on
1378 * the loopback device whenever they appear. xfstests is notorious for
1379 * failing tests because blkid via udev races with a losetup
1380 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1381 * command to fail with EBUSY.
1382 */
1383 if (atomic_read(&lo->lo_refcnt) > 1) {
1384 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1385 mutex_unlock(&loop_ctl_mutex);
1386 return 0;
1387 }
1388 lo->lo_state = Lo_rundown;
1389 mutex_unlock(&loop_ctl_mutex);
1390
1391 return __loop_clr_fd(lo, false);
1392 }
1393
1394 static int
1395 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1396 {
1397 int err;
1398 struct block_device *bdev;
1399 kuid_t uid = current_uid();
1400 int prev_lo_flags;
1401 bool partscan = false;
1402 bool size_changed = false;
1403
1404 err = mutex_lock_killable(&loop_ctl_mutex);
1405 if (err)
1406 return err;
1407 if (lo->lo_encrypt_key_size &&
1408 !uid_eq(lo->lo_key_owner, uid) &&
1409 !capable(CAP_SYS_ADMIN)) {
1410 err = -EPERM;
1411 goto out_unlock;
1412 }
1413 if (lo->lo_state != Lo_bound) {
1414 err = -ENXIO;
1415 goto out_unlock;
1416 }
1417
1418 if (lo->lo_offset != info->lo_offset ||
1419 lo->lo_sizelimit != info->lo_sizelimit) {
1420 size_changed = true;
1421 sync_blockdev(lo->lo_device);
1422 invalidate_bdev(lo->lo_device);
1423 }
1424
1425 /* I/O need to be drained during transfer transition */
1426 blk_mq_freeze_queue(lo->lo_queue);
1427
1428 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1429 /* If any pages were dirtied after invalidate_bdev(), try again */
1430 err = -EAGAIN;
1431 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1432 __func__, lo->lo_number, lo->lo_file_name,
1433 lo->lo_device->bd_inode->i_mapping->nrpages);
1434 goto out_unfreeze;
1435 }
1436
1437 prev_lo_flags = lo->lo_flags;
1438
1439 err = loop_set_status_from_info(lo, info);
1440 if (err)
1441 goto out_unfreeze;
1442
1443 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1444 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1445 /* For those flags, use the previous values instead */
1446 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1447 /* For flags that can't be cleared, use previous values too */
1448 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1449
1450 if (size_changed) {
1451 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1452 lo->lo_backing_file);
1453 loop_set_size(lo, new_size);
1454 }
1455
1456 loop_config_discard(lo);
1457
1458 /* update dio if lo_offset or transfer is changed */
1459 __loop_update_dio(lo, lo->use_dio);
1460
1461 out_unfreeze:
1462 blk_mq_unfreeze_queue(lo->lo_queue);
1463
1464 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1465 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1466 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1467 bdev = lo->lo_device;
1468 partscan = true;
1469 }
1470 out_unlock:
1471 mutex_unlock(&loop_ctl_mutex);
1472 if (partscan)
1473 loop_reread_partitions(lo, bdev);
1474
1475 return err;
1476 }
1477
1478 static int
1479 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1480 {
1481 struct path path;
1482 struct kstat stat;
1483 int ret;
1484
1485 ret = mutex_lock_killable(&loop_ctl_mutex);
1486 if (ret)
1487 return ret;
1488 if (lo->lo_state != Lo_bound) {
1489 mutex_unlock(&loop_ctl_mutex);
1490 return -ENXIO;
1491 }
1492
1493 memset(info, 0, sizeof(*info));
1494 info->lo_number = lo->lo_number;
1495 info->lo_offset = lo->lo_offset;
1496 info->lo_sizelimit = lo->lo_sizelimit;
1497 info->lo_flags = lo->lo_flags;
1498 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1499 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1500 info->lo_encrypt_type =
1501 lo->lo_encryption ? lo->lo_encryption->number : 0;
1502 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1503 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1504 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1505 lo->lo_encrypt_key_size);
1506 }
1507
1508 /* Drop loop_ctl_mutex while we call into the filesystem. */
1509 path = lo->lo_backing_file->f_path;
1510 path_get(&path);
1511 mutex_unlock(&loop_ctl_mutex);
1512 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1513 if (!ret) {
1514 info->lo_device = huge_encode_dev(stat.dev);
1515 info->lo_inode = stat.ino;
1516 info->lo_rdevice = huge_encode_dev(stat.rdev);
1517 }
1518 path_put(&path);
1519 return ret;
1520 }
1521
1522 static void
1523 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1524 {
1525 memset(info64, 0, sizeof(*info64));
1526 info64->lo_number = info->lo_number;
1527 info64->lo_device = info->lo_device;
1528 info64->lo_inode = info->lo_inode;
1529 info64->lo_rdevice = info->lo_rdevice;
1530 info64->lo_offset = info->lo_offset;
1531 info64->lo_sizelimit = 0;
1532 info64->lo_encrypt_type = info->lo_encrypt_type;
1533 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1534 info64->lo_flags = info->lo_flags;
1535 info64->lo_init[0] = info->lo_init[0];
1536 info64->lo_init[1] = info->lo_init[1];
1537 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1538 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1539 else
1540 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1541 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1542 }
1543
1544 static int
1545 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1546 {
1547 memset(info, 0, sizeof(*info));
1548 info->lo_number = info64->lo_number;
1549 info->lo_device = info64->lo_device;
1550 info->lo_inode = info64->lo_inode;
1551 info->lo_rdevice = info64->lo_rdevice;
1552 info->lo_offset = info64->lo_offset;
1553 info->lo_encrypt_type = info64->lo_encrypt_type;
1554 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1555 info->lo_flags = info64->lo_flags;
1556 info->lo_init[0] = info64->lo_init[0];
1557 info->lo_init[1] = info64->lo_init[1];
1558 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1559 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1560 else
1561 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1562 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1563
1564 /* error in case values were truncated */
1565 if (info->lo_device != info64->lo_device ||
1566 info->lo_rdevice != info64->lo_rdevice ||
1567 info->lo_inode != info64->lo_inode ||
1568 info->lo_offset != info64->lo_offset)
1569 return -EOVERFLOW;
1570
1571 return 0;
1572 }
1573
1574 static int
1575 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1576 {
1577 struct loop_info info;
1578 struct loop_info64 info64;
1579
1580 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1581 return -EFAULT;
1582 loop_info64_from_old(&info, &info64);
1583 return loop_set_status(lo, &info64);
1584 }
1585
1586 static int
1587 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1588 {
1589 struct loop_info64 info64;
1590
1591 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1592 return -EFAULT;
1593 return loop_set_status(lo, &info64);
1594 }
1595
1596 static int
1597 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1598 struct loop_info info;
1599 struct loop_info64 info64;
1600 int err;
1601
1602 if (!arg)
1603 return -EINVAL;
1604 err = loop_get_status(lo, &info64);
1605 if (!err)
1606 err = loop_info64_to_old(&info64, &info);
1607 if (!err && copy_to_user(arg, &info, sizeof(info)))
1608 err = -EFAULT;
1609
1610 return err;
1611 }
1612
1613 static int
1614 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1615 struct loop_info64 info64;
1616 int err;
1617
1618 if (!arg)
1619 return -EINVAL;
1620 err = loop_get_status(lo, &info64);
1621 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1622 err = -EFAULT;
1623
1624 return err;
1625 }
1626
1627 static int loop_set_capacity(struct loop_device *lo)
1628 {
1629 loff_t size;
1630
1631 if (unlikely(lo->lo_state != Lo_bound))
1632 return -ENXIO;
1633
1634 size = get_loop_size(lo, lo->lo_backing_file);
1635 loop_set_size(lo, size);
1636
1637 return 0;
1638 }
1639
1640 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1641 {
1642 int error = -ENXIO;
1643 if (lo->lo_state != Lo_bound)
1644 goto out;
1645
1646 __loop_update_dio(lo, !!arg);
1647 if (lo->use_dio == !!arg)
1648 return 0;
1649 error = -EINVAL;
1650 out:
1651 return error;
1652 }
1653
1654 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1655 {
1656 int err = 0;
1657
1658 if (lo->lo_state != Lo_bound)
1659 return -ENXIO;
1660
1661 err = loop_validate_block_size(arg);
1662 if (err)
1663 return err;
1664
1665 if (lo->lo_queue->limits.logical_block_size == arg)
1666 return 0;
1667
1668 sync_blockdev(lo->lo_device);
1669 invalidate_bdev(lo->lo_device);
1670
1671 blk_mq_freeze_queue(lo->lo_queue);
1672
1673 /* invalidate_bdev should have truncated all the pages */
1674 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1675 err = -EAGAIN;
1676 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1677 __func__, lo->lo_number, lo->lo_file_name,
1678 lo->lo_device->bd_inode->i_mapping->nrpages);
1679 goto out_unfreeze;
1680 }
1681
1682 blk_queue_logical_block_size(lo->lo_queue, arg);
1683 blk_queue_physical_block_size(lo->lo_queue, arg);
1684 blk_queue_io_min(lo->lo_queue, arg);
1685 loop_update_dio(lo);
1686 out_unfreeze:
1687 blk_mq_unfreeze_queue(lo->lo_queue);
1688
1689 return err;
1690 }
1691
1692 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1693 unsigned long arg)
1694 {
1695 int err;
1696
1697 err = mutex_lock_killable(&loop_ctl_mutex);
1698 if (err)
1699 return err;
1700 switch (cmd) {
1701 case LOOP_SET_CAPACITY:
1702 err = loop_set_capacity(lo);
1703 break;
1704 case LOOP_SET_DIRECT_IO:
1705 err = loop_set_dio(lo, arg);
1706 break;
1707 case LOOP_SET_BLOCK_SIZE:
1708 err = loop_set_block_size(lo, arg);
1709 break;
1710 default:
1711 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1712 }
1713 mutex_unlock(&loop_ctl_mutex);
1714 return err;
1715 }
1716
1717 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1718 unsigned int cmd, unsigned long arg)
1719 {
1720 struct loop_device *lo = bdev->bd_disk->private_data;
1721 void __user *argp = (void __user *) arg;
1722 int err;
1723
1724 switch (cmd) {
1725 case LOOP_SET_FD: {
1726 /*
1727 * Legacy case - pass in a zeroed out struct loop_config with
1728 * only the file descriptor set , which corresponds with the
1729 * default parameters we'd have used otherwise.
1730 */
1731 struct loop_config config;
1732
1733 memset(&config, 0, sizeof(config));
1734 config.fd = arg;
1735
1736 return loop_configure(lo, mode, bdev, &config);
1737 }
1738 case LOOP_CONFIGURE: {
1739 struct loop_config config;
1740
1741 if (copy_from_user(&config, argp, sizeof(config)))
1742 return -EFAULT;
1743
1744 return loop_configure(lo, mode, bdev, &config);
1745 }
1746 case LOOP_CHANGE_FD:
1747 return loop_change_fd(lo, bdev, arg);
1748 case LOOP_CLR_FD:
1749 return loop_clr_fd(lo);
1750 case LOOP_SET_STATUS:
1751 err = -EPERM;
1752 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1753 err = loop_set_status_old(lo, argp);
1754 }
1755 break;
1756 case LOOP_GET_STATUS:
1757 return loop_get_status_old(lo, argp);
1758 case LOOP_SET_STATUS64:
1759 err = -EPERM;
1760 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1761 err = loop_set_status64(lo, argp);
1762 }
1763 break;
1764 case LOOP_GET_STATUS64:
1765 return loop_get_status64(lo, argp);
1766 case LOOP_SET_CAPACITY:
1767 case LOOP_SET_DIRECT_IO:
1768 case LOOP_SET_BLOCK_SIZE:
1769 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1770 return -EPERM;
1771 fallthrough;
1772 default:
1773 err = lo_simple_ioctl(lo, cmd, arg);
1774 break;
1775 }
1776
1777 return err;
1778 }
1779
1780 #ifdef CONFIG_COMPAT
1781 struct compat_loop_info {
1782 compat_int_t lo_number; /* ioctl r/o */
1783 compat_dev_t lo_device; /* ioctl r/o */
1784 compat_ulong_t lo_inode; /* ioctl r/o */
1785 compat_dev_t lo_rdevice; /* ioctl r/o */
1786 compat_int_t lo_offset;
1787 compat_int_t lo_encrypt_type;
1788 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1789 compat_int_t lo_flags; /* ioctl r/o */
1790 char lo_name[LO_NAME_SIZE];
1791 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1792 compat_ulong_t lo_init[2];
1793 char reserved[4];
1794 };
1795
1796 /*
1797 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1798 * - noinlined to reduce stack space usage in main part of driver
1799 */
1800 static noinline int
1801 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1802 struct loop_info64 *info64)
1803 {
1804 struct compat_loop_info info;
1805
1806 if (copy_from_user(&info, arg, sizeof(info)))
1807 return -EFAULT;
1808
1809 memset(info64, 0, sizeof(*info64));
1810 info64->lo_number = info.lo_number;
1811 info64->lo_device = info.lo_device;
1812 info64->lo_inode = info.lo_inode;
1813 info64->lo_rdevice = info.lo_rdevice;
1814 info64->lo_offset = info.lo_offset;
1815 info64->lo_sizelimit = 0;
1816 info64->lo_encrypt_type = info.lo_encrypt_type;
1817 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1818 info64->lo_flags = info.lo_flags;
1819 info64->lo_init[0] = info.lo_init[0];
1820 info64->lo_init[1] = info.lo_init[1];
1821 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1822 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1823 else
1824 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1825 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1826 return 0;
1827 }
1828
1829 /*
1830 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1831 * - noinlined to reduce stack space usage in main part of driver
1832 */
1833 static noinline int
1834 loop_info64_to_compat(const struct loop_info64 *info64,
1835 struct compat_loop_info __user *arg)
1836 {
1837 struct compat_loop_info info;
1838
1839 memset(&info, 0, sizeof(info));
1840 info.lo_number = info64->lo_number;
1841 info.lo_device = info64->lo_device;
1842 info.lo_inode = info64->lo_inode;
1843 info.lo_rdevice = info64->lo_rdevice;
1844 info.lo_offset = info64->lo_offset;
1845 info.lo_encrypt_type = info64->lo_encrypt_type;
1846 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1847 info.lo_flags = info64->lo_flags;
1848 info.lo_init[0] = info64->lo_init[0];
1849 info.lo_init[1] = info64->lo_init[1];
1850 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1851 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1852 else
1853 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1854 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1855
1856 /* error in case values were truncated */
1857 if (info.lo_device != info64->lo_device ||
1858 info.lo_rdevice != info64->lo_rdevice ||
1859 info.lo_inode != info64->lo_inode ||
1860 info.lo_offset != info64->lo_offset ||
1861 info.lo_init[0] != info64->lo_init[0] ||
1862 info.lo_init[1] != info64->lo_init[1])
1863 return -EOVERFLOW;
1864
1865 if (copy_to_user(arg, &info, sizeof(info)))
1866 return -EFAULT;
1867 return 0;
1868 }
1869
1870 static int
1871 loop_set_status_compat(struct loop_device *lo,
1872 const struct compat_loop_info __user *arg)
1873 {
1874 struct loop_info64 info64;
1875 int ret;
1876
1877 ret = loop_info64_from_compat(arg, &info64);
1878 if (ret < 0)
1879 return ret;
1880 return loop_set_status(lo, &info64);
1881 }
1882
1883 static int
1884 loop_get_status_compat(struct loop_device *lo,
1885 struct compat_loop_info __user *arg)
1886 {
1887 struct loop_info64 info64;
1888 int err;
1889
1890 if (!arg)
1891 return -EINVAL;
1892 err = loop_get_status(lo, &info64);
1893 if (!err)
1894 err = loop_info64_to_compat(&info64, arg);
1895 return err;
1896 }
1897
1898 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1899 unsigned int cmd, unsigned long arg)
1900 {
1901 struct loop_device *lo = bdev->bd_disk->private_data;
1902 int err;
1903
1904 switch(cmd) {
1905 case LOOP_SET_STATUS:
1906 err = loop_set_status_compat(lo,
1907 (const struct compat_loop_info __user *)arg);
1908 break;
1909 case LOOP_GET_STATUS:
1910 err = loop_get_status_compat(lo,
1911 (struct compat_loop_info __user *)arg);
1912 break;
1913 case LOOP_SET_CAPACITY:
1914 case LOOP_CLR_FD:
1915 case LOOP_GET_STATUS64:
1916 case LOOP_SET_STATUS64:
1917 case LOOP_CONFIGURE:
1918 arg = (unsigned long) compat_ptr(arg);
1919 fallthrough;
1920 case LOOP_SET_FD:
1921 case LOOP_CHANGE_FD:
1922 case LOOP_SET_BLOCK_SIZE:
1923 case LOOP_SET_DIRECT_IO:
1924 err = lo_ioctl(bdev, mode, cmd, arg);
1925 break;
1926 default:
1927 err = -ENOIOCTLCMD;
1928 break;
1929 }
1930 return err;
1931 }
1932 #endif
1933
1934 static int lo_open(struct block_device *bdev, fmode_t mode)
1935 {
1936 struct loop_device *lo;
1937 int err;
1938
1939 err = mutex_lock_killable(&loop_ctl_mutex);
1940 if (err)
1941 return err;
1942 lo = bdev->bd_disk->private_data;
1943 if (!lo) {
1944 err = -ENXIO;
1945 goto out;
1946 }
1947
1948 atomic_inc(&lo->lo_refcnt);
1949 out:
1950 mutex_unlock(&loop_ctl_mutex);
1951 return err;
1952 }
1953
1954 static void lo_release(struct gendisk *disk, fmode_t mode)
1955 {
1956 struct loop_device *lo;
1957
1958 mutex_lock(&loop_ctl_mutex);
1959 lo = disk->private_data;
1960 if (atomic_dec_return(&lo->lo_refcnt))
1961 goto out_unlock;
1962
1963 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1964 if (lo->lo_state != Lo_bound)
1965 goto out_unlock;
1966 lo->lo_state = Lo_rundown;
1967 mutex_unlock(&loop_ctl_mutex);
1968 /*
1969 * In autoclear mode, stop the loop thread
1970 * and remove configuration after last close.
1971 */
1972 __loop_clr_fd(lo, true);
1973 return;
1974 } else if (lo->lo_state == Lo_bound) {
1975 /*
1976 * Otherwise keep thread (if running) and config,
1977 * but flush possible ongoing bios in thread.
1978 */
1979 blk_mq_freeze_queue(lo->lo_queue);
1980 blk_mq_unfreeze_queue(lo->lo_queue);
1981 }
1982
1983 out_unlock:
1984 mutex_unlock(&loop_ctl_mutex);
1985 }
1986
1987 static const struct block_device_operations lo_fops = {
1988 .owner = THIS_MODULE,
1989 .open = lo_open,
1990 .release = lo_release,
1991 .ioctl = lo_ioctl,
1992 #ifdef CONFIG_COMPAT
1993 .compat_ioctl = lo_compat_ioctl,
1994 #endif
1995 };
1996
1997 /*
1998 * And now the modules code and kernel interface.
1999 */
2000 static int max_loop;
2001 module_param(max_loop, int, 0444);
2002 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
2003 module_param(max_part, int, 0444);
2004 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
2005 MODULE_LICENSE("GPL");
2006 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
2007
2008 int loop_register_transfer(struct loop_func_table *funcs)
2009 {
2010 unsigned int n = funcs->number;
2011
2012 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
2013 return -EINVAL;
2014 xfer_funcs[n] = funcs;
2015 return 0;
2016 }
2017
2018 static int unregister_transfer_cb(int id, void *ptr, void *data)
2019 {
2020 struct loop_device *lo = ptr;
2021 struct loop_func_table *xfer = data;
2022
2023 mutex_lock(&loop_ctl_mutex);
2024 if (lo->lo_encryption == xfer)
2025 loop_release_xfer(lo);
2026 mutex_unlock(&loop_ctl_mutex);
2027 return 0;
2028 }
2029
2030 int loop_unregister_transfer(int number)
2031 {
2032 unsigned int n = number;
2033 struct loop_func_table *xfer;
2034
2035 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
2036 return -EINVAL;
2037
2038 xfer_funcs[n] = NULL;
2039 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
2040 return 0;
2041 }
2042
2043 EXPORT_SYMBOL(loop_register_transfer);
2044 EXPORT_SYMBOL(loop_unregister_transfer);
2045
2046 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
2047 const struct blk_mq_queue_data *bd)
2048 {
2049 struct request *rq = bd->rq;
2050 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2051 struct loop_device *lo = rq->q->queuedata;
2052
2053 blk_mq_start_request(rq);
2054
2055 if (lo->lo_state != Lo_bound)
2056 return BLK_STS_IOERR;
2057
2058 switch (req_op(rq)) {
2059 case REQ_OP_FLUSH:
2060 case REQ_OP_DISCARD:
2061 case REQ_OP_WRITE_ZEROES:
2062 cmd->use_aio = false;
2063 break;
2064 default:
2065 cmd->use_aio = lo->use_dio;
2066 break;
2067 }
2068
2069 /* always use the first bio's css */
2070 #ifdef CONFIG_BLK_CGROUP
2071 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2072 cmd->css = &bio_blkcg(rq->bio)->css;
2073 css_get(cmd->css);
2074 } else
2075 #endif
2076 cmd->css = NULL;
2077 kthread_queue_work(&lo->worker, &cmd->work);
2078
2079 return BLK_STS_OK;
2080 }
2081
2082 static void loop_handle_cmd(struct loop_cmd *cmd)
2083 {
2084 struct request *rq = blk_mq_rq_from_pdu(cmd);
2085 const bool write = op_is_write(req_op(rq));
2086 struct loop_device *lo = rq->q->queuedata;
2087 int ret = 0;
2088
2089 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2090 ret = -EIO;
2091 goto failed;
2092 }
2093
2094 ret = do_req_filebacked(lo, rq);
2095 failed:
2096 /* complete non-aio request */
2097 if (!cmd->use_aio || ret) {
2098 if (ret == -EOPNOTSUPP)
2099 cmd->ret = ret;
2100 else
2101 cmd->ret = ret ? -EIO : 0;
2102 if (likely(!blk_should_fake_timeout(rq->q)))
2103 blk_mq_complete_request(rq);
2104 }
2105 }
2106
2107 static void loop_queue_work(struct kthread_work *work)
2108 {
2109 struct loop_cmd *cmd =
2110 container_of(work, struct loop_cmd, work);
2111
2112 loop_handle_cmd(cmd);
2113 }
2114
2115 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2116 unsigned int hctx_idx, unsigned int numa_node)
2117 {
2118 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2119
2120 kthread_init_work(&cmd->work, loop_queue_work);
2121 return 0;
2122 }
2123
2124 static const struct blk_mq_ops loop_mq_ops = {
2125 .queue_rq = loop_queue_rq,
2126 .init_request = loop_init_request,
2127 .complete = lo_complete_rq,
2128 };
2129
2130 static int loop_add(struct loop_device **l, int i)
2131 {
2132 struct loop_device *lo;
2133 struct gendisk *disk;
2134 int err;
2135
2136 err = -ENOMEM;
2137 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2138 if (!lo)
2139 goto out;
2140
2141 lo->lo_state = Lo_unbound;
2142
2143 /* allocate id, if @id >= 0, we're requesting that specific id */
2144 if (i >= 0) {
2145 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2146 if (err == -ENOSPC)
2147 err = -EEXIST;
2148 } else {
2149 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2150 }
2151 if (err < 0)
2152 goto out_free_dev;
2153 i = err;
2154
2155 err = -ENOMEM;
2156 lo->tag_set.ops = &loop_mq_ops;
2157 lo->tag_set.nr_hw_queues = 1;
2158 lo->tag_set.queue_depth = 128;
2159 lo->tag_set.numa_node = NUMA_NO_NODE;
2160 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2161 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2162 lo->tag_set.driver_data = lo;
2163
2164 err = blk_mq_alloc_tag_set(&lo->tag_set);
2165 if (err)
2166 goto out_free_idr;
2167
2168 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2169 if (IS_ERR(lo->lo_queue)) {
2170 err = PTR_ERR(lo->lo_queue);
2171 goto out_cleanup_tags;
2172 }
2173 lo->lo_queue->queuedata = lo;
2174
2175 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2176
2177 /*
2178 * By default, we do buffer IO, so it doesn't make sense to enable
2179 * merge because the I/O submitted to backing file is handled page by
2180 * page. For directio mode, merge does help to dispatch bigger request
2181 * to underlayer disk. We will enable merge once directio is enabled.
2182 */
2183 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2184
2185 err = -ENOMEM;
2186 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2187 if (!disk)
2188 goto out_free_queue;
2189
2190 /*
2191 * Disable partition scanning by default. The in-kernel partition
2192 * scanning can be requested individually per-device during its
2193 * setup. Userspace can always add and remove partitions from all
2194 * devices. The needed partition minors are allocated from the
2195 * extended minor space, the main loop device numbers will continue
2196 * to match the loop minors, regardless of the number of partitions
2197 * used.
2198 *
2199 * If max_part is given, partition scanning is globally enabled for
2200 * all loop devices. The minors for the main loop devices will be
2201 * multiples of max_part.
2202 *
2203 * Note: Global-for-all-devices, set-only-at-init, read-only module
2204 * parameteters like 'max_loop' and 'max_part' make things needlessly
2205 * complicated, are too static, inflexible and may surprise
2206 * userspace tools. Parameters like this in general should be avoided.
2207 */
2208 if (!part_shift)
2209 disk->flags |= GENHD_FL_NO_PART_SCAN;
2210 disk->flags |= GENHD_FL_EXT_DEVT;
2211 atomic_set(&lo->lo_refcnt, 0);
2212 lo->lo_number = i;
2213 spin_lock_init(&lo->lo_lock);
2214 disk->major = LOOP_MAJOR;
2215 disk->first_minor = i << part_shift;
2216 disk->fops = &lo_fops;
2217 disk->private_data = lo;
2218 disk->queue = lo->lo_queue;
2219 sprintf(disk->disk_name, "loop%d", i);
2220 add_disk(disk);
2221 *l = lo;
2222 return lo->lo_number;
2223
2224 out_free_queue:
2225 blk_cleanup_queue(lo->lo_queue);
2226 out_cleanup_tags:
2227 blk_mq_free_tag_set(&lo->tag_set);
2228 out_free_idr:
2229 idr_remove(&loop_index_idr, i);
2230 out_free_dev:
2231 kfree(lo);
2232 out:
2233 return err;
2234 }
2235
2236 static void loop_remove(struct loop_device *lo)
2237 {
2238 del_gendisk(lo->lo_disk);
2239 blk_cleanup_queue(lo->lo_queue);
2240 blk_mq_free_tag_set(&lo->tag_set);
2241 put_disk(lo->lo_disk);
2242 kfree(lo);
2243 }
2244
2245 static int find_free_cb(int id, void *ptr, void *data)
2246 {
2247 struct loop_device *lo = ptr;
2248 struct loop_device **l = data;
2249
2250 if (lo->lo_state == Lo_unbound) {
2251 *l = lo;
2252 return 1;
2253 }
2254 return 0;
2255 }
2256
2257 static int loop_lookup(struct loop_device **l, int i)
2258 {
2259 struct loop_device *lo;
2260 int ret = -ENODEV;
2261
2262 if (i < 0) {
2263 int err;
2264
2265 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2266 if (err == 1) {
2267 *l = lo;
2268 ret = lo->lo_number;
2269 }
2270 goto out;
2271 }
2272
2273 /* lookup and return a specific i */
2274 lo = idr_find(&loop_index_idr, i);
2275 if (lo) {
2276 *l = lo;
2277 ret = lo->lo_number;
2278 }
2279 out:
2280 return ret;
2281 }
2282
2283 static void loop_probe(dev_t dev)
2284 {
2285 int idx = MINOR(dev) >> part_shift;
2286 struct loop_device *lo;
2287
2288 if (max_loop && idx >= max_loop)
2289 return;
2290
2291 mutex_lock(&loop_ctl_mutex);
2292 if (loop_lookup(&lo, idx) < 0)
2293 loop_add(&lo, idx);
2294 mutex_unlock(&loop_ctl_mutex);
2295 }
2296
2297 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2298 unsigned long parm)
2299 {
2300 struct loop_device *lo;
2301 int ret;
2302
2303 ret = mutex_lock_killable(&loop_ctl_mutex);
2304 if (ret)
2305 return ret;
2306
2307 ret = -ENOSYS;
2308 switch (cmd) {
2309 case LOOP_CTL_ADD:
2310 ret = loop_lookup(&lo, parm);
2311 if (ret >= 0) {
2312 ret = -EEXIST;
2313 break;
2314 }
2315 ret = loop_add(&lo, parm);
2316 break;
2317 case LOOP_CTL_REMOVE:
2318 ret = loop_lookup(&lo, parm);
2319 if (ret < 0)
2320 break;
2321 if (lo->lo_state != Lo_unbound) {
2322 ret = -EBUSY;
2323 break;
2324 }
2325 if (atomic_read(&lo->lo_refcnt) > 0) {
2326 ret = -EBUSY;
2327 break;
2328 }
2329 lo->lo_disk->private_data = NULL;
2330 idr_remove(&loop_index_idr, lo->lo_number);
2331 loop_remove(lo);
2332 break;
2333 case LOOP_CTL_GET_FREE:
2334 ret = loop_lookup(&lo, -1);
2335 if (ret >= 0)
2336 break;
2337 ret = loop_add(&lo, -1);
2338 }
2339 mutex_unlock(&loop_ctl_mutex);
2340
2341 return ret;
2342 }
2343
2344 static const struct file_operations loop_ctl_fops = {
2345 .open = nonseekable_open,
2346 .unlocked_ioctl = loop_control_ioctl,
2347 .compat_ioctl = loop_control_ioctl,
2348 .owner = THIS_MODULE,
2349 .llseek = noop_llseek,
2350 };
2351
2352 static struct miscdevice loop_misc = {
2353 .minor = LOOP_CTRL_MINOR,
2354 .name = "loop-control",
2355 .fops = &loop_ctl_fops,
2356 };
2357
2358 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2359 MODULE_ALIAS("devname:loop-control");
2360
2361 static int __init loop_init(void)
2362 {
2363 int i, nr;
2364 struct loop_device *lo;
2365 int err;
2366
2367 part_shift = 0;
2368 if (max_part > 0) {
2369 part_shift = fls(max_part);
2370
2371 /*
2372 * Adjust max_part according to part_shift as it is exported
2373 * to user space so that user can decide correct minor number
2374 * if [s]he want to create more devices.
2375 *
2376 * Note that -1 is required because partition 0 is reserved
2377 * for the whole disk.
2378 */
2379 max_part = (1UL << part_shift) - 1;
2380 }
2381
2382 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2383 err = -EINVAL;
2384 goto err_out;
2385 }
2386
2387 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2388 err = -EINVAL;
2389 goto err_out;
2390 }
2391
2392 /*
2393 * If max_loop is specified, create that many devices upfront.
2394 * This also becomes a hard limit. If max_loop is not specified,
2395 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2396 * init time. Loop devices can be requested on-demand with the
2397 * /dev/loop-control interface, or be instantiated by accessing
2398 * a 'dead' device node.
2399 */
2400 if (max_loop)
2401 nr = max_loop;
2402 else
2403 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2404
2405 err = misc_register(&loop_misc);
2406 if (err < 0)
2407 goto err_out;
2408
2409
2410 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2411 err = -EIO;
2412 goto misc_out;
2413 }
2414
2415 /* pre-create number of devices given by config or max_loop */
2416 mutex_lock(&loop_ctl_mutex);
2417 for (i = 0; i < nr; i++)
2418 loop_add(&lo, i);
2419 mutex_unlock(&loop_ctl_mutex);
2420
2421 printk(KERN_INFO "loop: module loaded\n");
2422 return 0;
2423
2424 misc_out:
2425 misc_deregister(&loop_misc);
2426 err_out:
2427 return err;
2428 }
2429
2430 static int loop_exit_cb(int id, void *ptr, void *data)
2431 {
2432 struct loop_device *lo = ptr;
2433
2434 loop_remove(lo);
2435 return 0;
2436 }
2437
2438 static void __exit loop_exit(void)
2439 {
2440 mutex_lock(&loop_ctl_mutex);
2441
2442 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2443 idr_destroy(&loop_index_idr);
2444
2445 unregister_blkdev(LOOP_MAJOR, "loop");
2446
2447 misc_deregister(&loop_misc);
2448
2449 mutex_unlock(&loop_ctl_mutex);
2450 }
2451
2452 module_init(loop_init);
2453 module_exit(loop_exit);
2454
2455 #ifndef MODULE
2456 static int __init max_loop_setup(char *str)
2457 {
2458 max_loop = simple_strtol(str, NULL, 0);
2459 return 1;
2460 }
2461
2462 __setup("max_loop=", max_loop_setup);
2463 #endif
Ubuntu Hirsute Linux kernel mirror