projects/linux·blob v5.16 · refs

2297 lines · 57.8 KiB

1/*
2 * linux/drivers/block/loop.c
3 *
4 * Written by Theodore Ts'o, 3/29/93
5 *
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
8 *
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 *
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 *
15 * Fixed do_loop_request() re-entrancy - [email protected] Mar 20, 1997
16 *
17 * Added devfs support - Richard Gooch <[email protected]> 16-Jan-1998
18 *
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 *
21 * Loadable modules and other fixes by AK, 1998
22 *
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, [email protected]
26 *
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <[email protected]> 19990701
29 *
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <[email protected]>, Oct 31, 1999
33 *
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
36 * Al Viro too.
37 * Jens Axboe <[email protected]>, Nov 2000
38 *
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <[email protected]>, Feb 2002
41 *
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
52#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/pagemap.h>
57#include <linux/file.h>
58#include <linux/stat.h>
59#include <linux/errno.h>
60#include <linux/major.h>
61#include <linux/wait.h>
62#include <linux/blkdev.h>
63#include <linux/blkpg.h>
64#include <linux/init.h>
65#include <linux/swap.h>
66#include <linux/slab.h>
67#include <linux/compat.h>
68#include <linux/suspend.h>
69#include <linux/freezer.h>
70#include <linux/mutex.h>
71#include <linux/writeback.h>
72#include <linux/completion.h>
73#include <linux/highmem.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#include <linux/sched/mm.h>
82
83#include "loop.h"
84
85#include <linux/uaccess.h>
86
87#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
88
89static DEFINE_IDR(loop_index_idr);
90static DEFINE_MUTEX(loop_ctl_mutex);
91static DEFINE_MUTEX(loop_validate_mutex);
92
93/**
94 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
95 *
96 * @lo: struct loop_device
97 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
98 *
99 * Returns 0 on success, -EINTR otherwise.
100 *
101 * Since loop_validate_file() traverses on other "struct loop_device" if
102 * is_loop_device() is true, we need a global lock for serializing concurrent
103 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
104 */
105static int loop_global_lock_killable(struct loop_device *lo, bool global)
106{
107 int err;
108
109 if (global) {
110 err = mutex_lock_killable(&loop_validate_mutex);
111 if (err)
112 return err;
113 }
114 err = mutex_lock_killable(&lo->lo_mutex);
115 if (err && global)
116 mutex_unlock(&loop_validate_mutex);
117 return err;
118}
119
120/**
121 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
122 *
123 * @lo: struct loop_device
124 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
125 */
126static void loop_global_unlock(struct loop_device *lo, bool global)
127{
128 mutex_unlock(&lo->lo_mutex);
129 if (global)
130 mutex_unlock(&loop_validate_mutex);
131}
132
133static int max_part;
134static int part_shift;
135
136static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
137{
138 loff_t loopsize;
139
140 /* Compute loopsize in bytes */
141 loopsize = i_size_read(file->f_mapping->host);
142 if (offset > 0)
143 loopsize -= offset;
144 /* offset is beyond i_size, weird but possible */
145 if (loopsize < 0)
146 return 0;
147
148 if (sizelimit > 0 && sizelimit < loopsize)
149 loopsize = sizelimit;
150 /*
151 * Unfortunately, if we want to do I/O on the device,
152 * the number of 512-byte sectors has to fit into a sector_t.
153 */
154 return loopsize >> 9;
155}
156
157static loff_t get_loop_size(struct loop_device *lo, struct file *file)
158{
159 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
160}
161
162static void __loop_update_dio(struct loop_device *lo, bool dio)
163{
164 struct file *file = lo->lo_backing_file;
165 struct address_space *mapping = file->f_mapping;
166 struct inode *inode = mapping->host;
167 unsigned short sb_bsize = 0;
168 unsigned dio_align = 0;
169 bool use_dio;
170
171 if (inode->i_sb->s_bdev) {
172 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
173 dio_align = sb_bsize - 1;
174 }
175
176 /*
177 * We support direct I/O only if lo_offset is aligned with the
178 * logical I/O size of backing device, and the logical block
179 * size of loop is bigger than the backing device's.
180 *
181 * TODO: the above condition may be loosed in the future, and
182 * direct I/O may be switched runtime at that time because most
183 * of requests in sane applications should be PAGE_SIZE aligned
184 */
185 if (dio) {
186 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
187 !(lo->lo_offset & dio_align) &&
188 mapping->a_ops->direct_IO)
189 use_dio = true;
190 else
191 use_dio = false;
192 } else {
193 use_dio = false;
194 }
195
196 if (lo->use_dio == use_dio)
197 return;
198
199 /* flush dirty pages before changing direct IO */
200 vfs_fsync(file, 0);
201
202 /*
203 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
204 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
205 * will get updated by ioctl(LOOP_GET_STATUS)
206 */
207 if (lo->lo_state == Lo_bound)
208 blk_mq_freeze_queue(lo->lo_queue);
209 lo->use_dio = use_dio;
210 if (use_dio) {
211 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
212 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
213 } else {
214 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
215 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
216 }
217 if (lo->lo_state == Lo_bound)
218 blk_mq_unfreeze_queue(lo->lo_queue);
219}
220
221/**
222 * loop_set_size() - sets device size and notifies userspace
223 * @lo: struct loop_device to set the size for
224 * @size: new size of the loop device
225 *
226 * Callers must validate that the size passed into this function fits into
227 * a sector_t, eg using loop_validate_size()
228 */
229static void loop_set_size(struct loop_device *lo, loff_t size)
230{
231 if (!set_capacity_and_notify(lo->lo_disk, size))
232 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
233}
234
235static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
236{
237 struct iov_iter i;
238 ssize_t bw;
239
240 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
241
242 file_start_write(file);
243 bw = vfs_iter_write(file, &i, ppos, 0);
244 file_end_write(file);
245
246 if (likely(bw == bvec->bv_len))
247 return 0;
248
249 printk_ratelimited(KERN_ERR
250 "loop: Write error at byte offset %llu, length %i.\n",
251 (unsigned long long)*ppos, bvec->bv_len);
252 if (bw >= 0)
253 bw = -EIO;
254 return bw;
255}
256
257static int lo_write_simple(struct loop_device *lo, struct request *rq,
258 loff_t pos)
259{
260 struct bio_vec bvec;
261 struct req_iterator iter;
262 int ret = 0;
263
264 rq_for_each_segment(bvec, rq, iter) {
265 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
266 if (ret < 0)
267 break;
268 cond_resched();
269 }
270
271 return ret;
272}
273
274static int lo_read_simple(struct loop_device *lo, struct request *rq,
275 loff_t pos)
276{
277 struct bio_vec bvec;
278 struct req_iterator iter;
279 struct iov_iter i;
280 ssize_t len;
281
282 rq_for_each_segment(bvec, rq, iter) {
283 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
284 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
285 if (len < 0)
286 return len;
287
288 flush_dcache_page(bvec.bv_page);
289
290 if (len != bvec.bv_len) {
291 struct bio *bio;
292
293 __rq_for_each_bio(bio, rq)
294 zero_fill_bio(bio);
295 break;
296 }
297 cond_resched();
298 }
299
300 return 0;
301}
302
303static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
304 int mode)
305{
306 /*
307 * We use fallocate to manipulate the space mappings used by the image
308 * a.k.a. discard/zerorange.
309 */
310 struct file *file = lo->lo_backing_file;
311 struct request_queue *q = lo->lo_queue;
312 int ret;
313
314 mode |= FALLOC_FL_KEEP_SIZE;
315
316 if (!blk_queue_discard(q)) {
317 ret = -EOPNOTSUPP;
318 goto out;
319 }
320
321 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
322 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
323 ret = -EIO;
324 out:
325 return ret;
326}
327
328static int lo_req_flush(struct loop_device *lo, struct request *rq)
329{
330 struct file *file = lo->lo_backing_file;
331 int ret = vfs_fsync(file, 0);
332 if (unlikely(ret && ret != -EINVAL))
333 ret = -EIO;
334
335 return ret;
336}
337
338static void lo_complete_rq(struct request *rq)
339{
340 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
341 blk_status_t ret = BLK_STS_OK;
342
343 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
344 req_op(rq) != REQ_OP_READ) {
345 if (cmd->ret < 0)
346 ret = errno_to_blk_status(cmd->ret);
347 goto end_io;
348 }
349
350 /*
351 * Short READ - if we got some data, advance our request and
352 * retry it. If we got no data, end the rest with EIO.
353 */
354 if (cmd->ret) {
355 blk_update_request(rq, BLK_STS_OK, cmd->ret);
356 cmd->ret = 0;
357 blk_mq_requeue_request(rq, true);
358 } else {
359 if (cmd->use_aio) {
360 struct bio *bio = rq->bio;
361
362 while (bio) {
363 zero_fill_bio(bio);
364 bio = bio->bi_next;
365 }
366 }
367 ret = BLK_STS_IOERR;
368end_io:
369 blk_mq_end_request(rq, ret);
370 }
371}
372
373static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
374{
375 struct request *rq = blk_mq_rq_from_pdu(cmd);
376
377 if (!atomic_dec_and_test(&cmd->ref))
378 return;
379 kfree(cmd->bvec);
380 cmd->bvec = NULL;
381 if (likely(!blk_should_fake_timeout(rq->q)))
382 blk_mq_complete_request(rq);
383}
384
385static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
386{
387 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
388
389 cmd->ret = ret;
390 lo_rw_aio_do_completion(cmd);
391}
392
393static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
394 loff_t pos, bool rw)
395{
396 struct iov_iter iter;
397 struct req_iterator rq_iter;
398 struct bio_vec *bvec;
399 struct request *rq = blk_mq_rq_from_pdu(cmd);
400 struct bio *bio = rq->bio;
401 struct file *file = lo->lo_backing_file;
402 struct bio_vec tmp;
403 unsigned int offset;
404 int nr_bvec = 0;
405 int ret;
406
407 rq_for_each_bvec(tmp, rq, rq_iter)
408 nr_bvec++;
409
410 if (rq->bio != rq->biotail) {
411
412 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
413 GFP_NOIO);
414 if (!bvec)
415 return -EIO;
416 cmd->bvec = bvec;
417
418 /*
419 * The bios of the request may be started from the middle of
420 * the 'bvec' because of bio splitting, so we can't directly
421 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
422 * API will take care of all details for us.
423 */
424 rq_for_each_bvec(tmp, rq, rq_iter) {
425 *bvec = tmp;
426 bvec++;
427 }
428 bvec = cmd->bvec;
429 offset = 0;
430 } else {
431 /*
432 * Same here, this bio may be started from the middle of the
433 * 'bvec' because of bio splitting, so offset from the bvec
434 * must be passed to iov iterator
435 */
436 offset = bio->bi_iter.bi_bvec_done;
437 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
438 }
439 atomic_set(&cmd->ref, 2);
440
441 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
442 iter.iov_offset = offset;
443
444 cmd->iocb.ki_pos = pos;
445 cmd->iocb.ki_filp = file;
446 cmd->iocb.ki_complete = lo_rw_aio_complete;
447 cmd->iocb.ki_flags = IOCB_DIRECT;
448 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
449
450 if (rw == WRITE)
451 ret = call_write_iter(file, &cmd->iocb, &iter);
452 else
453 ret = call_read_iter(file, &cmd->iocb, &iter);
454
455 lo_rw_aio_do_completion(cmd);
456
457 if (ret != -EIOCBQUEUED)
458 lo_rw_aio_complete(&cmd->iocb, ret);
459 return 0;
460}
461
462static int do_req_filebacked(struct loop_device *lo, struct request *rq)
463{
464 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
465 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
466
467 /*
468 * lo_write_simple and lo_read_simple should have been covered
469 * by io submit style function like lo_rw_aio(), one blocker
470 * is that lo_read_simple() need to call flush_dcache_page after
471 * the page is written from kernel, and it isn't easy to handle
472 * this in io submit style function which submits all segments
473 * of the req at one time. And direct read IO doesn't need to
474 * run flush_dcache_page().
475 */
476 switch (req_op(rq)) {
477 case REQ_OP_FLUSH:
478 return lo_req_flush(lo, rq);
479 case REQ_OP_WRITE_ZEROES:
480 /*
481 * If the caller doesn't want deallocation, call zeroout to
482 * write zeroes the range. Otherwise, punch them out.
483 */
484 return lo_fallocate(lo, rq, pos,
485 (rq->cmd_flags & REQ_NOUNMAP) ?
486 FALLOC_FL_ZERO_RANGE :
487 FALLOC_FL_PUNCH_HOLE);
488 case REQ_OP_DISCARD:
489 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
490 case REQ_OP_WRITE:
491 if (cmd->use_aio)
492 return lo_rw_aio(lo, cmd, pos, WRITE);
493 else
494 return lo_write_simple(lo, rq, pos);
495 case REQ_OP_READ:
496 if (cmd->use_aio)
497 return lo_rw_aio(lo, cmd, pos, READ);
498 else
499 return lo_read_simple(lo, rq, pos);
500 default:
501 WARN_ON_ONCE(1);
502 return -EIO;
503 }
504}
505
506static inline void loop_update_dio(struct loop_device *lo)
507{
508 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
509 lo->use_dio);
510}
511
512static void loop_reread_partitions(struct loop_device *lo)
513{
514 int rc;
515
516 mutex_lock(&lo->lo_disk->open_mutex);
517 rc = bdev_disk_changed(lo->lo_disk, false);
518 mutex_unlock(&lo->lo_disk->open_mutex);
519 if (rc)
520 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
521 __func__, lo->lo_number, lo->lo_file_name, rc);
522}
523
524static inline int is_loop_device(struct file *file)
525{
526 struct inode *i = file->f_mapping->host;
527
528 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
529}
530
531static int loop_validate_file(struct file *file, struct block_device *bdev)
532{
533 struct inode *inode = file->f_mapping->host;
534 struct file *f = file;
535
536 /* Avoid recursion */
537 while (is_loop_device(f)) {
538 struct loop_device *l;
539
540 lockdep_assert_held(&loop_validate_mutex);
541 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
542 return -EBADF;
543
544 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
545 if (l->lo_state != Lo_bound)
546 return -EINVAL;
547 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
548 rmb();
549 f = l->lo_backing_file;
550 }
551 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
552 return -EINVAL;
553 return 0;
554}
555
556/*
557 * loop_change_fd switched the backing store of a loopback device to
558 * a new file. This is useful for operating system installers to free up
559 * the original file and in High Availability environments to switch to
560 * an alternative location for the content in case of server meltdown.
561 * This can only work if the loop device is used read-only, and if the
562 * new backing store is the same size and type as the old backing store.
563 */
564static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
565 unsigned int arg)
566{
567 struct file *file = fget(arg);
568 struct file *old_file;
569 int error;
570 bool partscan;
571 bool is_loop;
572
573 if (!file)
574 return -EBADF;
575 is_loop = is_loop_device(file);
576 error = loop_global_lock_killable(lo, is_loop);
577 if (error)
578 goto out_putf;
579 error = -ENXIO;
580 if (lo->lo_state != Lo_bound)
581 goto out_err;
582
583 /* the loop device has to be read-only */
584 error = -EINVAL;
585 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
586 goto out_err;
587
588 error = loop_validate_file(file, bdev);
589 if (error)
590 goto out_err;
591
592 old_file = lo->lo_backing_file;
593
594 error = -EINVAL;
595
596 /* size of the new backing store needs to be the same */
597 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
598 goto out_err;
599
600 /* and ... switch */
601 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
602 blk_mq_freeze_queue(lo->lo_queue);
603 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
604 lo->lo_backing_file = file;
605 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
606 mapping_set_gfp_mask(file->f_mapping,
607 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
608 loop_update_dio(lo);
609 blk_mq_unfreeze_queue(lo->lo_queue);
610 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
611 loop_global_unlock(lo, is_loop);
612
613 /*
614 * Flush loop_validate_file() before fput(), for l->lo_backing_file
615 * might be pointing at old_file which might be the last reference.
616 */
617 if (!is_loop) {
618 mutex_lock(&loop_validate_mutex);
619 mutex_unlock(&loop_validate_mutex);
620 }
621 /*
622 * We must drop file reference outside of lo_mutex as dropping
623 * the file ref can take open_mutex which creates circular locking
624 * dependency.
625 */
626 fput(old_file);
627 if (partscan)
628 loop_reread_partitions(lo);
629 return 0;
630
631out_err:
632 loop_global_unlock(lo, is_loop);
633out_putf:
634 fput(file);
635 return error;
636}
637
638/* loop sysfs attributes */
639
640static ssize_t loop_attr_show(struct device *dev, char *page,
641 ssize_t (*callback)(struct loop_device *, char *))
642{
643 struct gendisk *disk = dev_to_disk(dev);
644 struct loop_device *lo = disk->private_data;
645
646 return callback(lo, page);
647}
648
649#define LOOP_ATTR_RO(_name) \
650static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
651static ssize_t loop_attr_do_show_##_name(struct device *d, \
652 struct device_attribute *attr, char *b) \
653{ \
654 return loop_attr_show(d, b, loop_attr_##_name##_show); \
655} \
656static struct device_attribute loop_attr_##_name = \
657 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
658
659static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
660{
661 ssize_t ret;
662 char *p = NULL;
663
664 spin_lock_irq(&lo->lo_lock);
665 if (lo->lo_backing_file)
666 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
667 spin_unlock_irq(&lo->lo_lock);
668
669 if (IS_ERR_OR_NULL(p))
670 ret = PTR_ERR(p);
671 else {
672 ret = strlen(p);
673 memmove(buf, p, ret);
674 buf[ret++] = '\n';
675 buf[ret] = 0;
676 }
677
678 return ret;
679}
680
681static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
682{
683 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
684}
685
686static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
687{
688 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
689}
690
691static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
692{
693 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
694
695 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
696}
697
698static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
699{
700 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
701
702 return sprintf(buf, "%s\n", partscan ? "1" : "0");
703}
704
705static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
706{
707 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
708
709 return sprintf(buf, "%s\n", dio ? "1" : "0");
710}
711
712LOOP_ATTR_RO(backing_file);
713LOOP_ATTR_RO(offset);
714LOOP_ATTR_RO(sizelimit);
715LOOP_ATTR_RO(autoclear);
716LOOP_ATTR_RO(partscan);
717LOOP_ATTR_RO(dio);
718
719static struct attribute *loop_attrs[] = {
720 &loop_attr_backing_file.attr,
721 &loop_attr_offset.attr,
722 &loop_attr_sizelimit.attr,
723 &loop_attr_autoclear.attr,
724 &loop_attr_partscan.attr,
725 &loop_attr_dio.attr,
726 NULL,
727};
728
729static struct attribute_group loop_attribute_group = {
730 .name = "loop",
731 .attrs= loop_attrs,
732};
733
734static void loop_sysfs_init(struct loop_device *lo)
735{
736 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
737 &loop_attribute_group);
738}
739
740static void loop_sysfs_exit(struct loop_device *lo)
741{
742 if (lo->sysfs_inited)
743 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
744 &loop_attribute_group);
745}
746
747static void loop_config_discard(struct loop_device *lo)
748{
749 struct file *file = lo->lo_backing_file;
750 struct inode *inode = file->f_mapping->host;
751 struct request_queue *q = lo->lo_queue;
752 u32 granularity, max_discard_sectors;
753
754 /*
755 * If the backing device is a block device, mirror its zeroing
756 * capability. Set the discard sectors to the block device's zeroing
757 * capabilities because loop discards result in blkdev_issue_zeroout(),
758 * not blkdev_issue_discard(). This maintains consistent behavior with
759 * file-backed loop devices: discarded regions read back as zero.
760 */
761 if (S_ISBLK(inode->i_mode)) {
762 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
763
764 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
765 granularity = backingq->limits.discard_granularity ?:
766 queue_physical_block_size(backingq);
767
768 /*
769 * We use punch hole to reclaim the free space used by the
770 * image a.k.a. discard.
771 */
772 } else if (!file->f_op->fallocate) {
773 max_discard_sectors = 0;
774 granularity = 0;
775
776 } else {
777 max_discard_sectors = UINT_MAX >> 9;
778 granularity = inode->i_sb->s_blocksize;
779 }
780
781 if (max_discard_sectors) {
782 q->limits.discard_granularity = granularity;
783 blk_queue_max_discard_sectors(q, max_discard_sectors);
784 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
785 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
786 } else {
787 q->limits.discard_granularity = 0;
788 blk_queue_max_discard_sectors(q, 0);
789 blk_queue_max_write_zeroes_sectors(q, 0);
790 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
791 }
792 q->limits.discard_alignment = 0;
793}
794
795struct loop_worker {
796 struct rb_node rb_node;
797 struct work_struct work;
798 struct list_head cmd_list;
799 struct list_head idle_list;
800 struct loop_device *lo;
801 struct cgroup_subsys_state *blkcg_css;
802 unsigned long last_ran_at;
803};
804
805static void loop_workfn(struct work_struct *work);
806static void loop_rootcg_workfn(struct work_struct *work);
807static void loop_free_idle_workers(struct timer_list *timer);
808
809#ifdef CONFIG_BLK_CGROUP
810static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
811{
812 return !css || css == blkcg_root_css;
813}
814#else
815static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
816{
817 return !css;
818}
819#endif
820
821static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
822{
823 struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL;
824 struct loop_worker *cur_worker, *worker = NULL;
825 struct work_struct *work;
826 struct list_head *cmd_list;
827
828 spin_lock_irq(&lo->lo_work_lock);
829
830 if (queue_on_root_worker(cmd->blkcg_css))
831 goto queue_work;
832
833 node = &lo->worker_tree.rb_node;
834
835 while (*node) {
836 parent = *node;
837 cur_worker = container_of(*node, struct loop_worker, rb_node);
838 if (cur_worker->blkcg_css == cmd->blkcg_css) {
839 worker = cur_worker;
840 break;
841 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
842 node = &(*node)->rb_left;
843 } else {
844 node = &(*node)->rb_right;
845 }
846 }
847 if (worker)
848 goto queue_work;
849
850 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
851 /*
852 * In the event we cannot allocate a worker, just queue on the
853 * rootcg worker and issue the I/O as the rootcg
854 */
855 if (!worker) {
856 cmd->blkcg_css = NULL;
857 if (cmd->memcg_css)
858 css_put(cmd->memcg_css);
859 cmd->memcg_css = NULL;
860 goto queue_work;
861 }
862
863 worker->blkcg_css = cmd->blkcg_css;
864 css_get(worker->blkcg_css);
865 INIT_WORK(&worker->work, loop_workfn);
866 INIT_LIST_HEAD(&worker->cmd_list);
867 INIT_LIST_HEAD(&worker->idle_list);
868 worker->lo = lo;
869 rb_link_node(&worker->rb_node, parent, node);
870 rb_insert_color(&worker->rb_node, &lo->worker_tree);
871queue_work:
872 if (worker) {
873 /*
874 * We need to remove from the idle list here while
875 * holding the lock so that the idle timer doesn't
876 * free the worker
877 */
878 if (!list_empty(&worker->idle_list))
879 list_del_init(&worker->idle_list);
880 work = &worker->work;
881 cmd_list = &worker->cmd_list;
882 } else {
883 work = &lo->rootcg_work;
884 cmd_list = &lo->rootcg_cmd_list;
885 }
886 list_add_tail(&cmd->list_entry, cmd_list);
887 queue_work(lo->workqueue, work);
888 spin_unlock_irq(&lo->lo_work_lock);
889}
890
891static void loop_update_rotational(struct loop_device *lo)
892{
893 struct file *file = lo->lo_backing_file;
894 struct inode *file_inode = file->f_mapping->host;
895 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
896 struct request_queue *q = lo->lo_queue;
897 bool nonrot = true;
898
899 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
900 if (file_bdev)
901 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
902
903 if (nonrot)
904 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
905 else
906 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
907}
908
909/**
910 * loop_set_status_from_info - configure device from loop_info
911 * @lo: struct loop_device to configure
912 * @info: struct loop_info64 to configure the device with
913 *
914 * Configures the loop device parameters according to the passed
915 * in loop_info64 configuration.
916 */
917static int
918loop_set_status_from_info(struct loop_device *lo,
919 const struct loop_info64 *info)
920{
921 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
922 return -EINVAL;
923
924 switch (info->lo_encrypt_type) {
925 case LO_CRYPT_NONE:
926 break;
927 case LO_CRYPT_XOR:
928 pr_warn("support for the xor transformation has been removed.\n");
929 return -EINVAL;
930 case LO_CRYPT_CRYPTOAPI:
931 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
932 return -EINVAL;
933 default:
934 return -EINVAL;
935 }
936
937 lo->lo_offset = info->lo_offset;
938 lo->lo_sizelimit = info->lo_sizelimit;
939 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
940 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
941 lo->lo_flags = info->lo_flags;
942 return 0;
943}
944
945static int loop_configure(struct loop_device *lo, fmode_t mode,
946 struct block_device *bdev,
947 const struct loop_config *config)
948{
949 struct file *file = fget(config->fd);
950 struct inode *inode;
951 struct address_space *mapping;
952 int error;
953 loff_t size;
954 bool partscan;
955 unsigned short bsize;
956 bool is_loop;
957
958 if (!file)
959 return -EBADF;
960 is_loop = is_loop_device(file);
961
962 /* This is safe, since we have a reference from open(). */
963 __module_get(THIS_MODULE);
964
965 /*
966 * If we don't hold exclusive handle for the device, upgrade to it
967 * here to avoid changing device under exclusive owner.
968 */
969 if (!(mode & FMODE_EXCL)) {
970 error = bd_prepare_to_claim(bdev, loop_configure);
971 if (error)
972 goto out_putf;
973 }
974
975 error = loop_global_lock_killable(lo, is_loop);
976 if (error)
977 goto out_bdev;
978
979 error = -EBUSY;
980 if (lo->lo_state != Lo_unbound)
981 goto out_unlock;
982
983 error = loop_validate_file(file, bdev);
984 if (error)
985 goto out_unlock;
986
987 mapping = file->f_mapping;
988 inode = mapping->host;
989
990 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
991 error = -EINVAL;
992 goto out_unlock;
993 }
994
995 if (config->block_size) {
996 error = blk_validate_block_size(config->block_size);
997 if (error)
998 goto out_unlock;
999 }
1000
1001 error = loop_set_status_from_info(lo, &config->info);
1002 if (error)
1003 goto out_unlock;
1004
1005 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1006 !file->f_op->write_iter)
1007 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1008
1009 lo->workqueue = alloc_workqueue("loop%d",
1010 WQ_UNBOUND | WQ_FREEZABLE,
1011 0,
1012 lo->lo_number);
1013 if (!lo->workqueue) {
1014 error = -ENOMEM;
1015 goto out_unlock;
1016 }
1017
1018 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1019 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1020
1021 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
1022 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
1023 INIT_LIST_HEAD(&lo->idle_worker_list);
1024 lo->worker_tree = RB_ROOT;
1025 timer_setup(&lo->timer, loop_free_idle_workers,
1026 TIMER_DEFERRABLE);
1027 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1028 lo->lo_device = bdev;
1029 lo->lo_backing_file = file;
1030 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1031 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1032
1033 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1034 blk_queue_write_cache(lo->lo_queue, true, false);
1035
1036 if (config->block_size)
1037 bsize = config->block_size;
1038 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1039 /* In case of direct I/O, match underlying block size */
1040 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1041 else
1042 bsize = 512;
1043
1044 blk_queue_logical_block_size(lo->lo_queue, bsize);
1045 blk_queue_physical_block_size(lo->lo_queue, bsize);
1046 blk_queue_io_min(lo->lo_queue, bsize);
1047
1048 loop_config_discard(lo);
1049 loop_update_rotational(lo);
1050 loop_update_dio(lo);
1051 loop_sysfs_init(lo);
1052
1053 size = get_loop_size(lo, file);
1054 loop_set_size(lo, size);
1055
1056 /* Order wrt reading lo_state in loop_validate_file(). */
1057 wmb();
1058
1059 lo->lo_state = Lo_bound;
1060 if (part_shift)
1061 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1062 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1063 if (partscan)
1064 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1065
1066 loop_global_unlock(lo, is_loop);
1067 if (partscan)
1068 loop_reread_partitions(lo);
1069 if (!(mode & FMODE_EXCL))
1070 bd_abort_claiming(bdev, loop_configure);
1071 return 0;
1072
1073out_unlock:
1074 loop_global_unlock(lo, is_loop);
1075out_bdev:
1076 if (!(mode & FMODE_EXCL))
1077 bd_abort_claiming(bdev, loop_configure);
1078out_putf:
1079 fput(file);
1080 /* This is safe: open() is still holding a reference. */
1081 module_put(THIS_MODULE);
1082 return error;
1083}
1084
1085static int __loop_clr_fd(struct loop_device *lo, bool release)
1086{
1087 struct file *filp = NULL;
1088 gfp_t gfp = lo->old_gfp_mask;
1089 int err = 0;
1090 bool partscan = false;
1091 int lo_number;
1092 struct loop_worker *pos, *worker;
1093
1094 /*
1095 * Flush loop_configure() and loop_change_fd(). It is acceptable for
1096 * loop_validate_file() to succeed, for actual clear operation has not
1097 * started yet.
1098 */
1099 mutex_lock(&loop_validate_mutex);
1100 mutex_unlock(&loop_validate_mutex);
1101 /*
1102 * loop_validate_file() now fails because l->lo_state != Lo_bound
1103 * became visible.
1104 */
1105
1106 mutex_lock(&lo->lo_mutex);
1107 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1108 err = -ENXIO;
1109 goto out_unlock;
1110 }
1111
1112 filp = lo->lo_backing_file;
1113 if (filp == NULL) {
1114 err = -EINVAL;
1115 goto out_unlock;
1116 }
1117
1118 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1119 blk_queue_write_cache(lo->lo_queue, false, false);
1120
1121 /* freeze request queue during the transition */
1122 blk_mq_freeze_queue(lo->lo_queue);
1123
1124 destroy_workqueue(lo->workqueue);
1125 spin_lock_irq(&lo->lo_work_lock);
1126 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
1127 idle_list) {
1128 list_del(&worker->idle_list);
1129 rb_erase(&worker->rb_node, &lo->worker_tree);
1130 css_put(worker->blkcg_css);
1131 kfree(worker);
1132 }
1133 spin_unlock_irq(&lo->lo_work_lock);
1134 del_timer_sync(&lo->timer);
1135
1136 spin_lock_irq(&lo->lo_lock);
1137 lo->lo_backing_file = NULL;
1138 spin_unlock_irq(&lo->lo_lock);
1139
1140 lo->lo_device = NULL;
1141 lo->lo_offset = 0;
1142 lo->lo_sizelimit = 0;
1143 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1144 blk_queue_logical_block_size(lo->lo_queue, 512);
1145 blk_queue_physical_block_size(lo->lo_queue, 512);
1146 blk_queue_io_min(lo->lo_queue, 512);
1147 invalidate_disk(lo->lo_disk);
1148 loop_sysfs_exit(lo);
1149 /* let user-space know about this change */
1150 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1151 mapping_set_gfp_mask(filp->f_mapping, gfp);
1152 /* This is safe: open() is still holding a reference. */
1153 module_put(THIS_MODULE);
1154 blk_mq_unfreeze_queue(lo->lo_queue);
1155
1156 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1157 lo_number = lo->lo_number;
1158 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1159out_unlock:
1160 mutex_unlock(&lo->lo_mutex);
1161 if (partscan) {
1162 /*
1163 * open_mutex has been held already in release path, so don't
1164 * acquire it if this function is called in such case.
1165 *
1166 * If the reread partition isn't from release path, lo_refcnt
1167 * must be at least one and it can only become zero when the
1168 * current holder is released.
1169 */
1170 if (!release)
1171 mutex_lock(&lo->lo_disk->open_mutex);
1172 err = bdev_disk_changed(lo->lo_disk, false);
1173 if (!release)
1174 mutex_unlock(&lo->lo_disk->open_mutex);
1175 if (err)
1176 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1177 __func__, lo_number, err);
1178 /* Device is gone, no point in returning error */
1179 err = 0;
1180 }
1181
1182 /*
1183 * lo->lo_state is set to Lo_unbound here after above partscan has
1184 * finished.
1185 *
1186 * There cannot be anybody else entering __loop_clr_fd() as
1187 * lo->lo_backing_file is already cleared and Lo_rundown state
1188 * protects us from all the other places trying to change the 'lo'
1189 * device.
1190 */
1191 mutex_lock(&lo->lo_mutex);
1192 lo->lo_flags = 0;
1193 if (!part_shift)
1194 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1195 lo->lo_state = Lo_unbound;
1196 mutex_unlock(&lo->lo_mutex);
1197
1198 /*
1199 * Need not hold lo_mutex to fput backing file. Calling fput holding
1200 * lo_mutex triggers a circular lock dependency possibility warning as
1201 * fput can take open_mutex which is usually taken before lo_mutex.
1202 */
1203 if (filp)
1204 fput(filp);
1205 return err;
1206}
1207
1208static int loop_clr_fd(struct loop_device *lo)
1209{
1210 int err;
1211
1212 err = mutex_lock_killable(&lo->lo_mutex);
1213 if (err)
1214 return err;
1215 if (lo->lo_state != Lo_bound) {
1216 mutex_unlock(&lo->lo_mutex);
1217 return -ENXIO;
1218 }
1219 /*
1220 * If we've explicitly asked to tear down the loop device,
1221 * and it has an elevated reference count, set it for auto-teardown when
1222 * the last reference goes away. This stops $!~#$@ udev from
1223 * preventing teardown because it decided that it needs to run blkid on
1224 * the loopback device whenever they appear. xfstests is notorious for
1225 * failing tests because blkid via udev races with a losetup
1226 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1227 * command to fail with EBUSY.
1228 */
1229 if (atomic_read(&lo->lo_refcnt) > 1) {
1230 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1231 mutex_unlock(&lo->lo_mutex);
1232 return 0;
1233 }
1234 lo->lo_state = Lo_rundown;
1235 mutex_unlock(&lo->lo_mutex);
1236
1237 return __loop_clr_fd(lo, false);
1238}
1239
1240static int
1241loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1242{
1243 int err;
1244 int prev_lo_flags;
1245 bool partscan = false;
1246 bool size_changed = false;
1247
1248 err = mutex_lock_killable(&lo->lo_mutex);
1249 if (err)
1250 return err;
1251 if (lo->lo_state != Lo_bound) {
1252 err = -ENXIO;
1253 goto out_unlock;
1254 }
1255
1256 if (lo->lo_offset != info->lo_offset ||
1257 lo->lo_sizelimit != info->lo_sizelimit) {
1258 size_changed = true;
1259 sync_blockdev(lo->lo_device);
1260 invalidate_bdev(lo->lo_device);
1261 }
1262
1263 /* I/O need to be drained during transfer transition */
1264 blk_mq_freeze_queue(lo->lo_queue);
1265
1266 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1267 /* If any pages were dirtied after invalidate_bdev(), try again */
1268 err = -EAGAIN;
1269 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1270 __func__, lo->lo_number, lo->lo_file_name,
1271 lo->lo_device->bd_inode->i_mapping->nrpages);
1272 goto out_unfreeze;
1273 }
1274
1275 prev_lo_flags = lo->lo_flags;
1276
1277 err = loop_set_status_from_info(lo, info);
1278 if (err)
1279 goto out_unfreeze;
1280
1281 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1282 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1283 /* For those flags, use the previous values instead */
1284 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1285 /* For flags that can't be cleared, use previous values too */
1286 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1287
1288 if (size_changed) {
1289 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1290 lo->lo_backing_file);
1291 loop_set_size(lo, new_size);
1292 }
1293
1294 loop_config_discard(lo);
1295
1296 /* update dio if lo_offset or transfer is changed */
1297 __loop_update_dio(lo, lo->use_dio);
1298
1299out_unfreeze:
1300 blk_mq_unfreeze_queue(lo->lo_queue);
1301
1302 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1303 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1304 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1305 partscan = true;
1306 }
1307out_unlock:
1308 mutex_unlock(&lo->lo_mutex);
1309 if (partscan)
1310 loop_reread_partitions(lo);
1311
1312 return err;
1313}
1314
1315static int
1316loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1317{
1318 struct path path;
1319 struct kstat stat;
1320 int ret;
1321
1322 ret = mutex_lock_killable(&lo->lo_mutex);
1323 if (ret)
1324 return ret;
1325 if (lo->lo_state != Lo_bound) {
1326 mutex_unlock(&lo->lo_mutex);
1327 return -ENXIO;
1328 }
1329
1330 memset(info, 0, sizeof(*info));
1331 info->lo_number = lo->lo_number;
1332 info->lo_offset = lo->lo_offset;
1333 info->lo_sizelimit = lo->lo_sizelimit;
1334 info->lo_flags = lo->lo_flags;
1335 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1336
1337 /* Drop lo_mutex while we call into the filesystem. */
1338 path = lo->lo_backing_file->f_path;
1339 path_get(&path);
1340 mutex_unlock(&lo->lo_mutex);
1341 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1342 if (!ret) {
1343 info->lo_device = huge_encode_dev(stat.dev);
1344 info->lo_inode = stat.ino;
1345 info->lo_rdevice = huge_encode_dev(stat.rdev);
1346 }
1347 path_put(&path);
1348 return ret;
1349}
1350
1351static void
1352loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1353{
1354 memset(info64, 0, sizeof(*info64));
1355 info64->lo_number = info->lo_number;
1356 info64->lo_device = info->lo_device;
1357 info64->lo_inode = info->lo_inode;
1358 info64->lo_rdevice = info->lo_rdevice;
1359 info64->lo_offset = info->lo_offset;
1360 info64->lo_sizelimit = 0;
1361 info64->lo_flags = info->lo_flags;
1362 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1363}
1364
1365static int
1366loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1367{
1368 memset(info, 0, sizeof(*info));
1369 info->lo_number = info64->lo_number;
1370 info->lo_device = info64->lo_device;
1371 info->lo_inode = info64->lo_inode;
1372 info->lo_rdevice = info64->lo_rdevice;
1373 info->lo_offset = info64->lo_offset;
1374 info->lo_flags = info64->lo_flags;
1375 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1376
1377 /* error in case values were truncated */
1378 if (info->lo_device != info64->lo_device ||
1379 info->lo_rdevice != info64->lo_rdevice ||
1380 info->lo_inode != info64->lo_inode ||
1381 info->lo_offset != info64->lo_offset)
1382 return -EOVERFLOW;
1383
1384 return 0;
1385}
1386
1387static int
1388loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1389{
1390 struct loop_info info;
1391 struct loop_info64 info64;
1392
1393 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1394 return -EFAULT;
1395 loop_info64_from_old(&info, &info64);
1396 return loop_set_status(lo, &info64);
1397}
1398
1399static int
1400loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1401{
1402 struct loop_info64 info64;
1403
1404 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1405 return -EFAULT;
1406 return loop_set_status(lo, &info64);
1407}
1408
1409static int
1410loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1411 struct loop_info info;
1412 struct loop_info64 info64;
1413 int err;
1414
1415 if (!arg)
1416 return -EINVAL;
1417 err = loop_get_status(lo, &info64);
1418 if (!err)
1419 err = loop_info64_to_old(&info64, &info);
1420 if (!err && copy_to_user(arg, &info, sizeof(info)))
1421 err = -EFAULT;
1422
1423 return err;
1424}
1425
1426static int
1427loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1428 struct loop_info64 info64;
1429 int err;
1430
1431 if (!arg)
1432 return -EINVAL;
1433 err = loop_get_status(lo, &info64);
1434 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1435 err = -EFAULT;
1436
1437 return err;
1438}
1439
1440static int loop_set_capacity(struct loop_device *lo)
1441{
1442 loff_t size;
1443
1444 if (unlikely(lo->lo_state != Lo_bound))
1445 return -ENXIO;
1446
1447 size = get_loop_size(lo, lo->lo_backing_file);
1448 loop_set_size(lo, size);
1449
1450 return 0;
1451}
1452
1453static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1454{
1455 int error = -ENXIO;
1456 if (lo->lo_state != Lo_bound)
1457 goto out;
1458
1459 __loop_update_dio(lo, !!arg);
1460 if (lo->use_dio == !!arg)
1461 return 0;
1462 error = -EINVAL;
1463 out:
1464 return error;
1465}
1466
1467static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1468{
1469 int err = 0;
1470
1471 if (lo->lo_state != Lo_bound)
1472 return -ENXIO;
1473
1474 err = blk_validate_block_size(arg);
1475 if (err)
1476 return err;
1477
1478 if (lo->lo_queue->limits.logical_block_size == arg)
1479 return 0;
1480
1481 sync_blockdev(lo->lo_device);
1482 invalidate_bdev(lo->lo_device);
1483
1484 blk_mq_freeze_queue(lo->lo_queue);
1485
1486 /* invalidate_bdev should have truncated all the pages */
1487 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1488 err = -EAGAIN;
1489 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1490 __func__, lo->lo_number, lo->lo_file_name,
1491 lo->lo_device->bd_inode->i_mapping->nrpages);
1492 goto out_unfreeze;
1493 }
1494
1495 blk_queue_logical_block_size(lo->lo_queue, arg);
1496 blk_queue_physical_block_size(lo->lo_queue, arg);
1497 blk_queue_io_min(lo->lo_queue, arg);
1498 loop_update_dio(lo);
1499out_unfreeze:
1500 blk_mq_unfreeze_queue(lo->lo_queue);
1501
1502 return err;
1503}
1504
1505static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1506 unsigned long arg)
1507{
1508 int err;
1509
1510 err = mutex_lock_killable(&lo->lo_mutex);
1511 if (err)
1512 return err;
1513 switch (cmd) {
1514 case LOOP_SET_CAPACITY:
1515 err = loop_set_capacity(lo);
1516 break;
1517 case LOOP_SET_DIRECT_IO:
1518 err = loop_set_dio(lo, arg);
1519 break;
1520 case LOOP_SET_BLOCK_SIZE:
1521 err = loop_set_block_size(lo, arg);
1522 break;
1523 default:
1524 err = -EINVAL;
1525 }
1526 mutex_unlock(&lo->lo_mutex);
1527 return err;
1528}
1529
1530static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1531 unsigned int cmd, unsigned long arg)
1532{
1533 struct loop_device *lo = bdev->bd_disk->private_data;
1534 void __user *argp = (void __user *) arg;
1535 int err;
1536
1537 switch (cmd) {
1538 case LOOP_SET_FD: {
1539 /*
1540 * Legacy case - pass in a zeroed out struct loop_config with
1541 * only the file descriptor set , which corresponds with the
1542 * default parameters we'd have used otherwise.
1543 */
1544 struct loop_config config;
1545
1546 memset(&config, 0, sizeof(config));
1547 config.fd = arg;
1548
1549 return loop_configure(lo, mode, bdev, &config);
1550 }
1551 case LOOP_CONFIGURE: {
1552 struct loop_config config;
1553
1554 if (copy_from_user(&config, argp, sizeof(config)))
1555 return -EFAULT;
1556
1557 return loop_configure(lo, mode, bdev, &config);
1558 }
1559 case LOOP_CHANGE_FD:
1560 return loop_change_fd(lo, bdev, arg);
1561 case LOOP_CLR_FD:
1562 return loop_clr_fd(lo);
1563 case LOOP_SET_STATUS:
1564 err = -EPERM;
1565 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1566 err = loop_set_status_old(lo, argp);
1567 }
1568 break;
1569 case LOOP_GET_STATUS:
1570 return loop_get_status_old(lo, argp);
1571 case LOOP_SET_STATUS64:
1572 err = -EPERM;
1573 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1574 err = loop_set_status64(lo, argp);
1575 }
1576 break;
1577 case LOOP_GET_STATUS64:
1578 return loop_get_status64(lo, argp);
1579 case LOOP_SET_CAPACITY:
1580 case LOOP_SET_DIRECT_IO:
1581 case LOOP_SET_BLOCK_SIZE:
1582 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1583 return -EPERM;
1584 fallthrough;
1585 default:
1586 err = lo_simple_ioctl(lo, cmd, arg);
1587 break;
1588 }
1589
1590 return err;
1591}
1592
1593#ifdef CONFIG_COMPAT
1594struct compat_loop_info {
1595 compat_int_t lo_number; /* ioctl r/o */
1596 compat_dev_t lo_device; /* ioctl r/o */
1597 compat_ulong_t lo_inode; /* ioctl r/o */
1598 compat_dev_t lo_rdevice; /* ioctl r/o */
1599 compat_int_t lo_offset;
1600 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1601 compat_int_t lo_flags; /* ioctl r/o */
1602 char lo_name[LO_NAME_SIZE];
1603 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1604 compat_ulong_t lo_init[2];
1605 char reserved[4];
1606};
1607
1608/*
1609 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1610 * - noinlined to reduce stack space usage in main part of driver
1611 */
1612static noinline int
1613loop_info64_from_compat(const struct compat_loop_info __user *arg,
1614 struct loop_info64 *info64)
1615{
1616 struct compat_loop_info info;
1617
1618 if (copy_from_user(&info, arg, sizeof(info)))
1619 return -EFAULT;
1620
1621 memset(info64, 0, sizeof(*info64));
1622 info64->lo_number = info.lo_number;
1623 info64->lo_device = info.lo_device;
1624 info64->lo_inode = info.lo_inode;
1625 info64->lo_rdevice = info.lo_rdevice;
1626 info64->lo_offset = info.lo_offset;
1627 info64->lo_sizelimit = 0;
1628 info64->lo_flags = info.lo_flags;
1629 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1630 return 0;
1631}
1632
1633/*
1634 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1635 * - noinlined to reduce stack space usage in main part of driver
1636 */
1637static noinline int
1638loop_info64_to_compat(const struct loop_info64 *info64,
1639 struct compat_loop_info __user *arg)
1640{
1641 struct compat_loop_info info;
1642
1643 memset(&info, 0, sizeof(info));
1644 info.lo_number = info64->lo_number;
1645 info.lo_device = info64->lo_device;
1646 info.lo_inode = info64->lo_inode;
1647 info.lo_rdevice = info64->lo_rdevice;
1648 info.lo_offset = info64->lo_offset;
1649 info.lo_flags = info64->lo_flags;
1650 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1651
1652 /* error in case values were truncated */
1653 if (info.lo_device != info64->lo_device ||
1654 info.lo_rdevice != info64->lo_rdevice ||
1655 info.lo_inode != info64->lo_inode ||
1656 info.lo_offset != info64->lo_offset)
1657 return -EOVERFLOW;
1658
1659 if (copy_to_user(arg, &info, sizeof(info)))
1660 return -EFAULT;
1661 return 0;
1662}
1663
1664static int
1665loop_set_status_compat(struct loop_device *lo,
1666 const struct compat_loop_info __user *arg)
1667{
1668 struct loop_info64 info64;
1669 int ret;
1670
1671 ret = loop_info64_from_compat(arg, &info64);
1672 if (ret < 0)
1673 return ret;
1674 return loop_set_status(lo, &info64);
1675}
1676
1677static int
1678loop_get_status_compat(struct loop_device *lo,
1679 struct compat_loop_info __user *arg)
1680{
1681 struct loop_info64 info64;
1682 int err;
1683
1684 if (!arg)
1685 return -EINVAL;
1686 err = loop_get_status(lo, &info64);
1687 if (!err)
1688 err = loop_info64_to_compat(&info64, arg);
1689 return err;
1690}
1691
1692static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1693 unsigned int cmd, unsigned long arg)
1694{
1695 struct loop_device *lo = bdev->bd_disk->private_data;
1696 int err;
1697
1698 switch(cmd) {
1699 case LOOP_SET_STATUS:
1700 err = loop_set_status_compat(lo,
1701 (const struct compat_loop_info __user *)arg);
1702 break;
1703 case LOOP_GET_STATUS:
1704 err = loop_get_status_compat(lo,
1705 (struct compat_loop_info __user *)arg);
1706 break;
1707 case LOOP_SET_CAPACITY:
1708 case LOOP_CLR_FD:
1709 case LOOP_GET_STATUS64:
1710 case LOOP_SET_STATUS64:
1711 case LOOP_CONFIGURE:
1712 arg = (unsigned long) compat_ptr(arg);
1713 fallthrough;
1714 case LOOP_SET_FD:
1715 case LOOP_CHANGE_FD:
1716 case LOOP_SET_BLOCK_SIZE:
1717 case LOOP_SET_DIRECT_IO:
1718 err = lo_ioctl(bdev, mode, cmd, arg);
1719 break;
1720 default:
1721 err = -ENOIOCTLCMD;
1722 break;
1723 }
1724 return err;
1725}
1726#endif
1727
1728static int lo_open(struct block_device *bdev, fmode_t mode)
1729{
1730 struct loop_device *lo = bdev->bd_disk->private_data;
1731 int err;
1732
1733 err = mutex_lock_killable(&lo->lo_mutex);
1734 if (err)
1735 return err;
1736 if (lo->lo_state == Lo_deleting)
1737 err = -ENXIO;
1738 else
1739 atomic_inc(&lo->lo_refcnt);
1740 mutex_unlock(&lo->lo_mutex);
1741 return err;
1742}
1743
1744static void lo_release(struct gendisk *disk, fmode_t mode)
1745{
1746 struct loop_device *lo = disk->private_data;
1747
1748 mutex_lock(&lo->lo_mutex);
1749 if (atomic_dec_return(&lo->lo_refcnt))
1750 goto out_unlock;
1751
1752 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1753 if (lo->lo_state != Lo_bound)
1754 goto out_unlock;
1755 lo->lo_state = Lo_rundown;
1756 mutex_unlock(&lo->lo_mutex);
1757 /*
1758 * In autoclear mode, stop the loop thread
1759 * and remove configuration after last close.
1760 */
1761 __loop_clr_fd(lo, true);
1762 return;
1763 } else if (lo->lo_state == Lo_bound) {
1764 /*
1765 * Otherwise keep thread (if running) and config,
1766 * but flush possible ongoing bios in thread.
1767 */
1768 blk_mq_freeze_queue(lo->lo_queue);
1769 blk_mq_unfreeze_queue(lo->lo_queue);
1770 }
1771
1772out_unlock:
1773 mutex_unlock(&lo->lo_mutex);
1774}
1775
1776static const struct block_device_operations lo_fops = {
1777 .owner = THIS_MODULE,
1778 .open = lo_open,
1779 .release = lo_release,
1780 .ioctl = lo_ioctl,
1781#ifdef CONFIG_COMPAT
1782 .compat_ioctl = lo_compat_ioctl,
1783#endif
1784};
1785
1786/*
1787 * And now the modules code and kernel interface.
1788 */
1789static int max_loop;
1790module_param(max_loop, int, 0444);
1791MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1792module_param(max_part, int, 0444);
1793MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1794MODULE_LICENSE("GPL");
1795MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1796
1797static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1798 const struct blk_mq_queue_data *bd)
1799{
1800 struct request *rq = bd->rq;
1801 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1802 struct loop_device *lo = rq->q->queuedata;
1803
1804 blk_mq_start_request(rq);
1805
1806 if (lo->lo_state != Lo_bound)
1807 return BLK_STS_IOERR;
1808
1809 switch (req_op(rq)) {
1810 case REQ_OP_FLUSH:
1811 case REQ_OP_DISCARD:
1812 case REQ_OP_WRITE_ZEROES:
1813 cmd->use_aio = false;
1814 break;
1815 default:
1816 cmd->use_aio = lo->use_dio;
1817 break;
1818 }
1819
1820 /* always use the first bio's css */
1821 cmd->blkcg_css = NULL;
1822 cmd->memcg_css = NULL;
1823#ifdef CONFIG_BLK_CGROUP
1824 if (rq->bio && rq->bio->bi_blkg) {
1825 cmd->blkcg_css = &bio_blkcg(rq->bio)->css;
1826#ifdef CONFIG_MEMCG
1827 cmd->memcg_css =
1828 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1829 &memory_cgrp_subsys);
1830#endif
1831 }
1832#endif
1833 loop_queue_work(lo, cmd);
1834
1835 return BLK_STS_OK;
1836}
1837
1838static void loop_handle_cmd(struct loop_cmd *cmd)
1839{
1840 struct request *rq = blk_mq_rq_from_pdu(cmd);
1841 const bool write = op_is_write(req_op(rq));
1842 struct loop_device *lo = rq->q->queuedata;
1843 int ret = 0;
1844 struct mem_cgroup *old_memcg = NULL;
1845
1846 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1847 ret = -EIO;
1848 goto failed;
1849 }
1850
1851 if (cmd->blkcg_css)
1852 kthread_associate_blkcg(cmd->blkcg_css);
1853 if (cmd->memcg_css)
1854 old_memcg = set_active_memcg(
1855 mem_cgroup_from_css(cmd->memcg_css));
1856
1857 ret = do_req_filebacked(lo, rq);
1858
1859 if (cmd->blkcg_css)
1860 kthread_associate_blkcg(NULL);
1861
1862 if (cmd->memcg_css) {
1863 set_active_memcg(old_memcg);
1864 css_put(cmd->memcg_css);
1865 }
1866 failed:
1867 /* complete non-aio request */
1868 if (!cmd->use_aio || ret) {
1869 if (ret == -EOPNOTSUPP)
1870 cmd->ret = ret;
1871 else
1872 cmd->ret = ret ? -EIO : 0;
1873 if (likely(!blk_should_fake_timeout(rq->q)))
1874 blk_mq_complete_request(rq);
1875 }
1876}
1877
1878static void loop_set_timer(struct loop_device *lo)
1879{
1880 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
1881}
1882
1883static void loop_process_work(struct loop_worker *worker,
1884 struct list_head *cmd_list, struct loop_device *lo)
1885{
1886 int orig_flags = current->flags;
1887 struct loop_cmd *cmd;
1888
1889 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1890 spin_lock_irq(&lo->lo_work_lock);
1891 while (!list_empty(cmd_list)) {
1892 cmd = container_of(
1893 cmd_list->next, struct loop_cmd, list_entry);
1894 list_del(cmd_list->next);
1895 spin_unlock_irq(&lo->lo_work_lock);
1896
1897 loop_handle_cmd(cmd);
1898 cond_resched();
1899
1900 spin_lock_irq(&lo->lo_work_lock);
1901 }
1902
1903 /*
1904 * We only add to the idle list if there are no pending cmds
1905 * *and* the worker will not run again which ensures that it
1906 * is safe to free any worker on the idle list
1907 */
1908 if (worker && !work_pending(&worker->work)) {
1909 worker->last_ran_at = jiffies;
1910 list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1911 loop_set_timer(lo);
1912 }
1913 spin_unlock_irq(&lo->lo_work_lock);
1914 current->flags = orig_flags;
1915}
1916
1917static void loop_workfn(struct work_struct *work)
1918{
1919 struct loop_worker *worker =
1920 container_of(work, struct loop_worker, work);
1921 loop_process_work(worker, &worker->cmd_list, worker->lo);
1922}
1923
1924static void loop_rootcg_workfn(struct work_struct *work)
1925{
1926 struct loop_device *lo =
1927 container_of(work, struct loop_device, rootcg_work);
1928 loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1929}
1930
1931static void loop_free_idle_workers(struct timer_list *timer)
1932{
1933 struct loop_device *lo = container_of(timer, struct loop_device, timer);
1934 struct loop_worker *pos, *worker;
1935
1936 spin_lock_irq(&lo->lo_work_lock);
1937 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
1938 idle_list) {
1939 if (time_is_after_jiffies(worker->last_ran_at +
1940 LOOP_IDLE_WORKER_TIMEOUT))
1941 break;
1942 list_del(&worker->idle_list);
1943 rb_erase(&worker->rb_node, &lo->worker_tree);
1944 css_put(worker->blkcg_css);
1945 kfree(worker);
1946 }
1947 if (!list_empty(&lo->idle_worker_list))
1948 loop_set_timer(lo);
1949 spin_unlock_irq(&lo->lo_work_lock);
1950}
1951
1952static const struct blk_mq_ops loop_mq_ops = {
1953 .queue_rq = loop_queue_rq,
1954 .complete = lo_complete_rq,
1955};
1956
1957static int loop_add(int i)
1958{
1959 struct loop_device *lo;
1960 struct gendisk *disk;
1961 int err;
1962
1963 err = -ENOMEM;
1964 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1965 if (!lo)
1966 goto out;
1967 lo->lo_state = Lo_unbound;
1968
1969 err = mutex_lock_killable(&loop_ctl_mutex);
1970 if (err)
1971 goto out_free_dev;
1972
1973 /* allocate id, if @id >= 0, we're requesting that specific id */
1974 if (i >= 0) {
1975 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1976 if (err == -ENOSPC)
1977 err = -EEXIST;
1978 } else {
1979 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1980 }
1981 mutex_unlock(&loop_ctl_mutex);
1982 if (err < 0)
1983 goto out_free_dev;
1984 i = err;
1985
1986 lo->tag_set.ops = &loop_mq_ops;
1987 lo->tag_set.nr_hw_queues = 1;
1988 lo->tag_set.queue_depth = 128;
1989 lo->tag_set.numa_node = NUMA_NO_NODE;
1990 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1991 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
1992 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
1993 lo->tag_set.driver_data = lo;
1994
1995 err = blk_mq_alloc_tag_set(&lo->tag_set);
1996 if (err)
1997 goto out_free_idr;
1998
1999 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
2000 if (IS_ERR(disk)) {
2001 err = PTR_ERR(disk);
2002 goto out_cleanup_tags;
2003 }
2004 lo->lo_queue = lo->lo_disk->queue;
2005
2006 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2007
2008 /*
2009 * By default, we do buffer IO, so it doesn't make sense to enable
2010 * merge because the I/O submitted to backing file is handled page by
2011 * page. For directio mode, merge does help to dispatch bigger request
2012 * to underlayer disk. We will enable merge once directio is enabled.
2013 */
2014 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2015
2016 /*
2017 * Disable partition scanning by default. The in-kernel partition
2018 * scanning can be requested individually per-device during its
2019 * setup. Userspace can always add and remove partitions from all
2020 * devices. The needed partition minors are allocated from the
2021 * extended minor space, the main loop device numbers will continue
2022 * to match the loop minors, regardless of the number of partitions
2023 * used.
2024 *
2025 * If max_part is given, partition scanning is globally enabled for
2026 * all loop devices. The minors for the main loop devices will be
2027 * multiples of max_part.
2028 *
2029 * Note: Global-for-all-devices, set-only-at-init, read-only module
2030 * parameteters like 'max_loop' and 'max_part' make things needlessly
2031 * complicated, are too static, inflexible and may surprise
2032 * userspace tools. Parameters like this in general should be avoided.
2033 */
2034 if (!part_shift)
2035 disk->flags |= GENHD_FL_NO_PART_SCAN;
2036 disk->flags |= GENHD_FL_EXT_DEVT;
2037 atomic_set(&lo->lo_refcnt, 0);
2038 mutex_init(&lo->lo_mutex);
2039 lo->lo_number = i;
2040 spin_lock_init(&lo->lo_lock);
2041 spin_lock_init(&lo->lo_work_lock);
2042 disk->major = LOOP_MAJOR;
2043 disk->first_minor = i << part_shift;
2044 disk->minors = 1 << part_shift;
2045 disk->fops = &lo_fops;
2046 disk->private_data = lo;
2047 disk->queue = lo->lo_queue;
2048 disk->events = DISK_EVENT_MEDIA_CHANGE;
2049 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2050 sprintf(disk->disk_name, "loop%d", i);
2051 /* Make this loop device reachable from pathname. */
2052 err = add_disk(disk);
2053 if (err)
2054 goto out_cleanup_disk;
2055
2056 /* Show this loop device. */
2057 mutex_lock(&loop_ctl_mutex);
2058 lo->idr_visible = true;
2059 mutex_unlock(&loop_ctl_mutex);
2060
2061 return i;
2062
2063out_cleanup_disk:
2064 blk_cleanup_disk(disk);
2065out_cleanup_tags:
2066 blk_mq_free_tag_set(&lo->tag_set);
2067out_free_idr:
2068 mutex_lock(&loop_ctl_mutex);
2069 idr_remove(&loop_index_idr, i);
2070 mutex_unlock(&loop_ctl_mutex);
2071out_free_dev:
2072 kfree(lo);
2073out:
2074 return err;
2075}
2076
2077static void loop_remove(struct loop_device *lo)
2078{
2079 /* Make this loop device unreachable from pathname. */
2080 del_gendisk(lo->lo_disk);
2081 blk_cleanup_disk(lo->lo_disk);
2082 blk_mq_free_tag_set(&lo->tag_set);
2083 mutex_lock(&loop_ctl_mutex);
2084 idr_remove(&loop_index_idr, lo->lo_number);
2085 mutex_unlock(&loop_ctl_mutex);
2086 /* There is no route which can find this loop device. */
2087 mutex_destroy(&lo->lo_mutex);
2088 kfree(lo);
2089}
2090
2091static void loop_probe(dev_t dev)
2092{
2093 int idx = MINOR(dev) >> part_shift;
2094
2095 if (max_loop && idx >= max_loop)
2096 return;
2097 loop_add(idx);
2098}
2099
2100static int loop_control_remove(int idx)
2101{
2102 struct loop_device *lo;
2103 int ret;
2104
2105 if (idx < 0) {
2106 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2107 return -EINVAL;
2108 }
2109
2110 /* Hide this loop device for serialization. */
2111 ret = mutex_lock_killable(&loop_ctl_mutex);
2112 if (ret)
2113 return ret;
2114 lo = idr_find(&loop_index_idr, idx);
2115 if (!lo || !lo->idr_visible)
2116 ret = -ENODEV;
2117 else
2118 lo->idr_visible = false;
2119 mutex_unlock(&loop_ctl_mutex);
2120 if (ret)
2121 return ret;
2122
2123 /* Check whether this loop device can be removed. */
2124 ret = mutex_lock_killable(&lo->lo_mutex);
2125 if (ret)
2126 goto mark_visible;
2127 if (lo->lo_state != Lo_unbound ||
2128 atomic_read(&lo->lo_refcnt) > 0) {
2129 mutex_unlock(&lo->lo_mutex);
2130 ret = -EBUSY;
2131 goto mark_visible;
2132 }
2133 /* Mark this loop device no longer open()-able. */
2134 lo->lo_state = Lo_deleting;
2135 mutex_unlock(&lo->lo_mutex);
2136
2137 loop_remove(lo);
2138 return 0;
2139
2140mark_visible:
2141 /* Show this loop device again. */
2142 mutex_lock(&loop_ctl_mutex);
2143 lo->idr_visible = true;
2144 mutex_unlock(&loop_ctl_mutex);
2145 return ret;
2146}
2147
2148static int loop_control_get_free(int idx)
2149{
2150 struct loop_device *lo;
2151 int id, ret;
2152
2153 ret = mutex_lock_killable(&loop_ctl_mutex);
2154 if (ret)
2155 return ret;
2156 idr_for_each_entry(&loop_index_idr, lo, id) {
2157 /* Hitting a race results in creating a new loop device which is harmless. */
2158 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2159 goto found;
2160 }
2161 mutex_unlock(&loop_ctl_mutex);
2162 return loop_add(-1);
2163found:
2164 mutex_unlock(&loop_ctl_mutex);
2165 return id;
2166}
2167
2168static long loop_control_ioctl(struct file *file, unsigned int cmd,
2169 unsigned long parm)
2170{
2171 switch (cmd) {
2172 case LOOP_CTL_ADD:
2173 return loop_add(parm);
2174 case LOOP_CTL_REMOVE:
2175 return loop_control_remove(parm);
2176 case LOOP_CTL_GET_FREE:
2177 return loop_control_get_free(parm);
2178 default:
2179 return -ENOSYS;
2180 }
2181}
2182
2183static const struct file_operations loop_ctl_fops = {
2184 .open = nonseekable_open,
2185 .unlocked_ioctl = loop_control_ioctl,
2186 .compat_ioctl = loop_control_ioctl,
2187 .owner = THIS_MODULE,
2188 .llseek = noop_llseek,
2189};
2190
2191static struct miscdevice loop_misc = {
2192 .minor = LOOP_CTRL_MINOR,
2193 .name = "loop-control",
2194 .fops = &loop_ctl_fops,
2195};
2196
2197MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2198MODULE_ALIAS("devname:loop-control");
2199
2200static int __init loop_init(void)
2201{
2202 int i, nr;
2203 int err;
2204
2205 part_shift = 0;
2206 if (max_part > 0) {
2207 part_shift = fls(max_part);
2208
2209 /*
2210 * Adjust max_part according to part_shift as it is exported
2211 * to user space so that user can decide correct minor number
2212 * if [s]he want to create more devices.
2213 *
2214 * Note that -1 is required because partition 0 is reserved
2215 * for the whole disk.
2216 */
2217 max_part = (1UL << part_shift) - 1;
2218 }
2219
2220 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2221 err = -EINVAL;
2222 goto err_out;
2223 }
2224
2225 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2226 err = -EINVAL;
2227 goto err_out;
2228 }
2229
2230 /*
2231 * If max_loop is specified, create that many devices upfront.
2232 * This also becomes a hard limit. If max_loop is not specified,
2233 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2234 * init time. Loop devices can be requested on-demand with the
2235 * /dev/loop-control interface, or be instantiated by accessing
2236 * a 'dead' device node.
2237 */
2238 if (max_loop)
2239 nr = max_loop;
2240 else
2241 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2242
2243 err = misc_register(&loop_misc);
2244 if (err < 0)
2245 goto err_out;
2246
2247
2248 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2249 err = -EIO;
2250 goto misc_out;
2251 }
2252
2253 /* pre-create number of devices given by config or max_loop */
2254 for (i = 0; i < nr; i++)
2255 loop_add(i);
2256
2257 printk(KERN_INFO "loop: module loaded\n");
2258 return 0;
2259
2260misc_out:
2261 misc_deregister(&loop_misc);
2262err_out:
2263 return err;
2264}
2265
2266static void __exit loop_exit(void)
2267{
2268 struct loop_device *lo;
2269 int id;
2270
2271 unregister_blkdev(LOOP_MAJOR, "loop");
2272 misc_deregister(&loop_misc);
2273
2274 /*
2275 * There is no need to use loop_ctl_mutex here, for nobody else can
2276 * access loop_index_idr when this module is unloading (unless forced
2277 * module unloading is requested). If this is not a clean unloading,
2278 * we have no means to avoid kernel crash.
2279 */
2280 idr_for_each_entry(&loop_index_idr, lo, id)
2281 loop_remove(lo);
2282
2283 idr_destroy(&loop_index_idr);
2284}
2285
2286module_init(loop_init);
2287module_exit(loop_exit);
2288
2289#ifndef MODULE
2290static int __init max_loop_setup(char *str)
2291{
2292 max_loop = simple_strtol(str, NULL, 0);
2293 return 1;
2294}
2295
2296__setup("max_loop=", max_loop_setup);
2297#endif