projects/linux·blob v6.13 · refs

2314 lines · 57.3 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 1993 by Theodore Ts'o.
4 */
5#include <linux/module.h>
6#include <linux/moduleparam.h>
7#include <linux/sched.h>
8#include <linux/fs.h>
9#include <linux/pagemap.h>
10#include <linux/file.h>
11#include <linux/stat.h>
12#include <linux/errno.h>
13#include <linux/major.h>
14#include <linux/wait.h>
15#include <linux/blkpg.h>
16#include <linux/init.h>
17#include <linux/swap.h>
18#include <linux/slab.h>
19#include <linux/compat.h>
20#include <linux/suspend.h>
21#include <linux/freezer.h>
22#include <linux/mutex.h>
23#include <linux/writeback.h>
24#include <linux/completion.h>
25#include <linux/highmem.h>
26#include <linux/splice.h>
27#include <linux/sysfs.h>
28#include <linux/miscdevice.h>
29#include <linux/falloc.h>
30#include <linux/uio.h>
31#include <linux/ioprio.h>
32#include <linux/blk-cgroup.h>
33#include <linux/sched/mm.h>
34#include <linux/statfs.h>
35#include <linux/uaccess.h>
36#include <linux/blk-mq.h>
37#include <linux/spinlock.h>
38#include <uapi/linux/loop.h>
39
40/* Possible states of device */
41enum {
42 Lo_unbound,
43 Lo_bound,
44 Lo_rundown,
45 Lo_deleting,
46};
47
48struct loop_func_table;
49
50struct loop_device {
51 int lo_number;
52 loff_t lo_offset;
53 loff_t lo_sizelimit;
54 int lo_flags;
55 char lo_file_name[LO_NAME_SIZE];
56
57 struct file * lo_backing_file;
58 struct block_device *lo_device;
59
60 gfp_t old_gfp_mask;
61
62 spinlock_t lo_lock;
63 int lo_state;
64 spinlock_t lo_work_lock;
65 struct workqueue_struct *workqueue;
66 struct work_struct rootcg_work;
67 struct list_head rootcg_cmd_list;
68 struct list_head idle_worker_list;
69 struct rb_root worker_tree;
70 struct timer_list timer;
71 bool use_dio;
72 bool sysfs_inited;
73
74 struct request_queue *lo_queue;
75 struct blk_mq_tag_set tag_set;
76 struct gendisk *lo_disk;
77 struct mutex lo_mutex;
78 bool idr_visible;
79};
80
81struct loop_cmd {
82 struct list_head list_entry;
83 bool use_aio; /* use AIO interface to handle I/O */
84 atomic_t ref; /* only for aio */
85 long ret;
86 struct kiocb iocb;
87 struct bio_vec *bvec;
88 struct cgroup_subsys_state *blkcg_css;
89 struct cgroup_subsys_state *memcg_css;
90};
91
92#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
93#define LOOP_DEFAULT_HW_Q_DEPTH 128
94
95static DEFINE_IDR(loop_index_idr);
96static DEFINE_MUTEX(loop_ctl_mutex);
97static DEFINE_MUTEX(loop_validate_mutex);
98
99/**
100 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
101 *
102 * @lo: struct loop_device
103 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
104 *
105 * Returns 0 on success, -EINTR otherwise.
106 *
107 * Since loop_validate_file() traverses on other "struct loop_device" if
108 * is_loop_device() is true, we need a global lock for serializing concurrent
109 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
110 */
111static int loop_global_lock_killable(struct loop_device *lo, bool global)
112{
113 int err;
114
115 if (global) {
116 err = mutex_lock_killable(&loop_validate_mutex);
117 if (err)
118 return err;
119 }
120 err = mutex_lock_killable(&lo->lo_mutex);
121 if (err && global)
122 mutex_unlock(&loop_validate_mutex);
123 return err;
124}
125
126/**
127 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
128 *
129 * @lo: struct loop_device
130 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
131 */
132static void loop_global_unlock(struct loop_device *lo, bool global)
133{
134 mutex_unlock(&lo->lo_mutex);
135 if (global)
136 mutex_unlock(&loop_validate_mutex);
137}
138
139static int max_part;
140static int part_shift;
141
142static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
143{
144 loff_t loopsize;
145
146 /* Compute loopsize in bytes */
147 loopsize = i_size_read(file->f_mapping->host);
148 if (offset > 0)
149 loopsize -= offset;
150 /* offset is beyond i_size, weird but possible */
151 if (loopsize < 0)
152 return 0;
153
154 if (sizelimit > 0 && sizelimit < loopsize)
155 loopsize = sizelimit;
156 /*
157 * Unfortunately, if we want to do I/O on the device,
158 * the number of 512-byte sectors has to fit into a sector_t.
159 */
160 return loopsize >> 9;
161}
162
163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164{
165 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
166}
167
168/*
169 * We support direct I/O only if lo_offset is aligned with the logical I/O size
170 * of backing device, and the logical block size of loop is bigger than that of
171 * the backing device.
172 */
173static bool lo_bdev_can_use_dio(struct loop_device *lo,
174 struct block_device *backing_bdev)
175{
176 unsigned int sb_bsize = bdev_logical_block_size(backing_bdev);
177
178 if (queue_logical_block_size(lo->lo_queue) < sb_bsize)
179 return false;
180 if (lo->lo_offset & (sb_bsize - 1))
181 return false;
182 return true;
183}
184
185static void __loop_update_dio(struct loop_device *lo, bool dio)
186{
187 struct file *file = lo->lo_backing_file;
188 struct inode *inode = file->f_mapping->host;
189 struct block_device *backing_bdev = NULL;
190 bool use_dio;
191
192 if (S_ISBLK(inode->i_mode))
193 backing_bdev = I_BDEV(inode);
194 else if (inode->i_sb->s_bdev)
195 backing_bdev = inode->i_sb->s_bdev;
196
197 use_dio = dio && (file->f_mode & FMODE_CAN_ODIRECT) &&
198 (!backing_bdev || lo_bdev_can_use_dio(lo, backing_bdev));
199
200 if (lo->use_dio == use_dio)
201 return;
202
203 /* flush dirty pages before changing direct IO */
204 vfs_fsync(file, 0);
205
206 /*
207 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
208 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
209 * will get updated by ioctl(LOOP_GET_STATUS)
210 */
211 if (lo->lo_state == Lo_bound)
212 blk_mq_freeze_queue(lo->lo_queue);
213 lo->use_dio = use_dio;
214 if (use_dio)
215 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
216 else
217 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
218 if (lo->lo_state == Lo_bound)
219 blk_mq_unfreeze_queue(lo->lo_queue);
220}
221
222/**
223 * loop_set_size() - sets device size and notifies userspace
224 * @lo: struct loop_device to set the size for
225 * @size: new size of the loop device
226 *
227 * Callers must validate that the size passed into this function fits into
228 * a sector_t, eg using loop_validate_size()
229 */
230static void loop_set_size(struct loop_device *lo, loff_t size)
231{
232 if (!set_capacity_and_notify(lo->lo_disk, size))
233 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
234}
235
236static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
237{
238 struct iov_iter i;
239 ssize_t bw;
240
241 iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len);
242
243 bw = vfs_iter_write(file, &i, ppos, 0);
244
245 if (likely(bw == bvec->bv_len))
246 return 0;
247
248 printk_ratelimited(KERN_ERR
249 "loop: Write error at byte offset %llu, length %i.\n",
250 (unsigned long long)*ppos, bvec->bv_len);
251 if (bw >= 0)
252 bw = -EIO;
253 return bw;
254}
255
256static int lo_write_simple(struct loop_device *lo, struct request *rq,
257 loff_t pos)
258{
259 struct bio_vec bvec;
260 struct req_iterator iter;
261 int ret = 0;
262
263 rq_for_each_segment(bvec, rq, iter) {
264 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
265 if (ret < 0)
266 break;
267 cond_resched();
268 }
269
270 return ret;
271}
272
273static int lo_read_simple(struct loop_device *lo, struct request *rq,
274 loff_t pos)
275{
276 struct bio_vec bvec;
277 struct req_iterator iter;
278 struct iov_iter i;
279 ssize_t len;
280
281 rq_for_each_segment(bvec, rq, iter) {
282 iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len);
283 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
284 if (len < 0)
285 return len;
286
287 flush_dcache_page(bvec.bv_page);
288
289 if (len != bvec.bv_len) {
290 struct bio *bio;
291
292 __rq_for_each_bio(bio, rq)
293 zero_fill_bio(bio);
294 break;
295 }
296 cond_resched();
297 }
298
299 return 0;
300}
301
302static void loop_clear_limits(struct loop_device *lo, int mode)
303{
304 struct queue_limits lim = queue_limits_start_update(lo->lo_queue);
305
306 if (mode & FALLOC_FL_ZERO_RANGE)
307 lim.max_write_zeroes_sectors = 0;
308
309 if (mode & FALLOC_FL_PUNCH_HOLE) {
310 lim.max_hw_discard_sectors = 0;
311 lim.discard_granularity = 0;
312 }
313
314 queue_limits_commit_update(lo->lo_queue, &lim);
315}
316
317static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
318 int mode)
319{
320 /*
321 * We use fallocate to manipulate the space mappings used by the image
322 * a.k.a. discard/zerorange.
323 */
324 struct file *file = lo->lo_backing_file;
325 int ret;
326
327 mode |= FALLOC_FL_KEEP_SIZE;
328
329 if (!bdev_max_discard_sectors(lo->lo_device))
330 return -EOPNOTSUPP;
331
332 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
333 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
334 return -EIO;
335
336 /*
337 * We initially configure the limits in a hope that fallocate is
338 * supported and clear them here if that turns out not to be true.
339 */
340 if (unlikely(ret == -EOPNOTSUPP))
341 loop_clear_limits(lo, mode);
342
343 return ret;
344}
345
346static int lo_req_flush(struct loop_device *lo, struct request *rq)
347{
348 int ret = vfs_fsync(lo->lo_backing_file, 0);
349 if (unlikely(ret && ret != -EINVAL))
350 ret = -EIO;
351
352 return ret;
353}
354
355static void lo_complete_rq(struct request *rq)
356{
357 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
358 blk_status_t ret = BLK_STS_OK;
359
360 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
361 req_op(rq) != REQ_OP_READ) {
362 if (cmd->ret < 0)
363 ret = errno_to_blk_status(cmd->ret);
364 goto end_io;
365 }
366
367 /*
368 * Short READ - if we got some data, advance our request and
369 * retry it. If we got no data, end the rest with EIO.
370 */
371 if (cmd->ret) {
372 blk_update_request(rq, BLK_STS_OK, cmd->ret);
373 cmd->ret = 0;
374 blk_mq_requeue_request(rq, true);
375 } else {
376 if (cmd->use_aio) {
377 struct bio *bio = rq->bio;
378
379 while (bio) {
380 zero_fill_bio(bio);
381 bio = bio->bi_next;
382 }
383 }
384 ret = BLK_STS_IOERR;
385end_io:
386 blk_mq_end_request(rq, ret);
387 }
388}
389
390static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
391{
392 struct request *rq = blk_mq_rq_from_pdu(cmd);
393
394 if (!atomic_dec_and_test(&cmd->ref))
395 return;
396 kfree(cmd->bvec);
397 cmd->bvec = NULL;
398 if (likely(!blk_should_fake_timeout(rq->q)))
399 blk_mq_complete_request(rq);
400}
401
402static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
403{
404 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
405
406 cmd->ret = ret;
407 lo_rw_aio_do_completion(cmd);
408}
409
410static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
411 loff_t pos, int rw)
412{
413 struct iov_iter iter;
414 struct req_iterator rq_iter;
415 struct bio_vec *bvec;
416 struct request *rq = blk_mq_rq_from_pdu(cmd);
417 struct bio *bio = rq->bio;
418 struct file *file = lo->lo_backing_file;
419 struct bio_vec tmp;
420 unsigned int offset;
421 int nr_bvec = 0;
422 int ret;
423
424 rq_for_each_bvec(tmp, rq, rq_iter)
425 nr_bvec++;
426
427 if (rq->bio != rq->biotail) {
428
429 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
430 GFP_NOIO);
431 if (!bvec)
432 return -EIO;
433 cmd->bvec = bvec;
434
435 /*
436 * The bios of the request may be started from the middle of
437 * the 'bvec' because of bio splitting, so we can't directly
438 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
439 * API will take care of all details for us.
440 */
441 rq_for_each_bvec(tmp, rq, rq_iter) {
442 *bvec = tmp;
443 bvec++;
444 }
445 bvec = cmd->bvec;
446 offset = 0;
447 } else {
448 /*
449 * Same here, this bio may be started from the middle of the
450 * 'bvec' because of bio splitting, so offset from the bvec
451 * must be passed to iov iterator
452 */
453 offset = bio->bi_iter.bi_bvec_done;
454 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
455 }
456 atomic_set(&cmd->ref, 2);
457
458 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
459 iter.iov_offset = offset;
460
461 cmd->iocb.ki_pos = pos;
462 cmd->iocb.ki_filp = file;
463 cmd->iocb.ki_complete = lo_rw_aio_complete;
464 cmd->iocb.ki_flags = IOCB_DIRECT;
465 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
466
467 if (rw == ITER_SOURCE)
468 ret = file->f_op->write_iter(&cmd->iocb, &iter);
469 else
470 ret = file->f_op->read_iter(&cmd->iocb, &iter);
471
472 lo_rw_aio_do_completion(cmd);
473
474 if (ret != -EIOCBQUEUED)
475 lo_rw_aio_complete(&cmd->iocb, ret);
476 return 0;
477}
478
479static int do_req_filebacked(struct loop_device *lo, struct request *rq)
480{
481 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
482 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
483
484 /*
485 * lo_write_simple and lo_read_simple should have been covered
486 * by io submit style function like lo_rw_aio(), one blocker
487 * is that lo_read_simple() need to call flush_dcache_page after
488 * the page is written from kernel, and it isn't easy to handle
489 * this in io submit style function which submits all segments
490 * of the req at one time. And direct read IO doesn't need to
491 * run flush_dcache_page().
492 */
493 switch (req_op(rq)) {
494 case REQ_OP_FLUSH:
495 return lo_req_flush(lo, rq);
496 case REQ_OP_WRITE_ZEROES:
497 /*
498 * If the caller doesn't want deallocation, call zeroout to
499 * write zeroes the range. Otherwise, punch them out.
500 */
501 return lo_fallocate(lo, rq, pos,
502 (rq->cmd_flags & REQ_NOUNMAP) ?
503 FALLOC_FL_ZERO_RANGE :
504 FALLOC_FL_PUNCH_HOLE);
505 case REQ_OP_DISCARD:
506 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
507 case REQ_OP_WRITE:
508 if (cmd->use_aio)
509 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
510 else
511 return lo_write_simple(lo, rq, pos);
512 case REQ_OP_READ:
513 if (cmd->use_aio)
514 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
515 else
516 return lo_read_simple(lo, rq, pos);
517 default:
518 WARN_ON_ONCE(1);
519 return -EIO;
520 }
521}
522
523static inline void loop_update_dio(struct loop_device *lo)
524{
525 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
526 lo->use_dio);
527}
528
529static void loop_reread_partitions(struct loop_device *lo)
530{
531 int rc;
532
533 mutex_lock(&lo->lo_disk->open_mutex);
534 rc = bdev_disk_changed(lo->lo_disk, false);
535 mutex_unlock(&lo->lo_disk->open_mutex);
536 if (rc)
537 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
538 __func__, lo->lo_number, lo->lo_file_name, rc);
539}
540
541static inline int is_loop_device(struct file *file)
542{
543 struct inode *i = file->f_mapping->host;
544
545 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
546}
547
548static int loop_validate_file(struct file *file, struct block_device *bdev)
549{
550 struct inode *inode = file->f_mapping->host;
551 struct file *f = file;
552
553 /* Avoid recursion */
554 while (is_loop_device(f)) {
555 struct loop_device *l;
556
557 lockdep_assert_held(&loop_validate_mutex);
558 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
559 return -EBADF;
560
561 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
562 if (l->lo_state != Lo_bound)
563 return -EINVAL;
564 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
565 rmb();
566 f = l->lo_backing_file;
567 }
568 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
569 return -EINVAL;
570 return 0;
571}
572
573/*
574 * loop_change_fd switched the backing store of a loopback device to
575 * a new file. This is useful for operating system installers to free up
576 * the original file and in High Availability environments to switch to
577 * an alternative location for the content in case of server meltdown.
578 * This can only work if the loop device is used read-only, and if the
579 * new backing store is the same size and type as the old backing store.
580 */
581static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
582 unsigned int arg)
583{
584 struct file *file = fget(arg);
585 struct file *old_file;
586 int error;
587 bool partscan;
588 bool is_loop;
589
590 if (!file)
591 return -EBADF;
592
593 /* suppress uevents while reconfiguring the device */
594 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
595
596 is_loop = is_loop_device(file);
597 error = loop_global_lock_killable(lo, is_loop);
598 if (error)
599 goto out_putf;
600 error = -ENXIO;
601 if (lo->lo_state != Lo_bound)
602 goto out_err;
603
604 /* the loop device has to be read-only */
605 error = -EINVAL;
606 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
607 goto out_err;
608
609 error = loop_validate_file(file, bdev);
610 if (error)
611 goto out_err;
612
613 old_file = lo->lo_backing_file;
614
615 error = -EINVAL;
616
617 /* size of the new backing store needs to be the same */
618 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
619 goto out_err;
620
621 /* and ... switch */
622 disk_force_media_change(lo->lo_disk);
623 blk_mq_freeze_queue(lo->lo_queue);
624 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
625 lo->lo_backing_file = file;
626 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
627 mapping_set_gfp_mask(file->f_mapping,
628 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
629 loop_update_dio(lo);
630 blk_mq_unfreeze_queue(lo->lo_queue);
631 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
632 loop_global_unlock(lo, is_loop);
633
634 /*
635 * Flush loop_validate_file() before fput(), for l->lo_backing_file
636 * might be pointing at old_file which might be the last reference.
637 */
638 if (!is_loop) {
639 mutex_lock(&loop_validate_mutex);
640 mutex_unlock(&loop_validate_mutex);
641 }
642 /*
643 * We must drop file reference outside of lo_mutex as dropping
644 * the file ref can take open_mutex which creates circular locking
645 * dependency.
646 */
647 fput(old_file);
648 if (partscan)
649 loop_reread_partitions(lo);
650
651 error = 0;
652done:
653 /* enable and uncork uevent now that we are done */
654 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
655 return error;
656
657out_err:
658 loop_global_unlock(lo, is_loop);
659out_putf:
660 fput(file);
661 goto done;
662}
663
664/* loop sysfs attributes */
665
666static ssize_t loop_attr_show(struct device *dev, char *page,
667 ssize_t (*callback)(struct loop_device *, char *))
668{
669 struct gendisk *disk = dev_to_disk(dev);
670 struct loop_device *lo = disk->private_data;
671
672 return callback(lo, page);
673}
674
675#define LOOP_ATTR_RO(_name) \
676static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
677static ssize_t loop_attr_do_show_##_name(struct device *d, \
678 struct device_attribute *attr, char *b) \
679{ \
680 return loop_attr_show(d, b, loop_attr_##_name##_show); \
681} \
682static struct device_attribute loop_attr_##_name = \
683 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
684
685static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
686{
687 ssize_t ret;
688 char *p = NULL;
689
690 spin_lock_irq(&lo->lo_lock);
691 if (lo->lo_backing_file)
692 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
693 spin_unlock_irq(&lo->lo_lock);
694
695 if (IS_ERR_OR_NULL(p))
696 ret = PTR_ERR(p);
697 else {
698 ret = strlen(p);
699 memmove(buf, p, ret);
700 buf[ret++] = '\n';
701 buf[ret] = 0;
702 }
703
704 return ret;
705}
706
707static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
708{
709 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
710}
711
712static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
713{
714 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
715}
716
717static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
718{
719 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
720
721 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
722}
723
724static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
725{
726 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
727
728 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
729}
730
731static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
732{
733 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
734
735 return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
736}
737
738LOOP_ATTR_RO(backing_file);
739LOOP_ATTR_RO(offset);
740LOOP_ATTR_RO(sizelimit);
741LOOP_ATTR_RO(autoclear);
742LOOP_ATTR_RO(partscan);
743LOOP_ATTR_RO(dio);
744
745static struct attribute *loop_attrs[] = {
746 &loop_attr_backing_file.attr,
747 &loop_attr_offset.attr,
748 &loop_attr_sizelimit.attr,
749 &loop_attr_autoclear.attr,
750 &loop_attr_partscan.attr,
751 &loop_attr_dio.attr,
752 NULL,
753};
754
755static struct attribute_group loop_attribute_group = {
756 .name = "loop",
757 .attrs= loop_attrs,
758};
759
760static void loop_sysfs_init(struct loop_device *lo)
761{
762 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
763 &loop_attribute_group);
764}
765
766static void loop_sysfs_exit(struct loop_device *lo)
767{
768 if (lo->sysfs_inited)
769 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
770 &loop_attribute_group);
771}
772
773static void loop_get_discard_config(struct loop_device *lo,
774 u32 *granularity, u32 *max_discard_sectors)
775{
776 struct file *file = lo->lo_backing_file;
777 struct inode *inode = file->f_mapping->host;
778 struct kstatfs sbuf;
779
780 /*
781 * If the backing device is a block device, mirror its zeroing
782 * capability. Set the discard sectors to the block device's zeroing
783 * capabilities because loop discards result in blkdev_issue_zeroout(),
784 * not blkdev_issue_discard(). This maintains consistent behavior with
785 * file-backed loop devices: discarded regions read back as zero.
786 */
787 if (S_ISBLK(inode->i_mode)) {
788 struct block_device *bdev = I_BDEV(inode);
789
790 *max_discard_sectors = bdev_write_zeroes_sectors(bdev);
791 *granularity = bdev_discard_granularity(bdev);
792
793 /*
794 * We use punch hole to reclaim the free space used by the
795 * image a.k.a. discard.
796 */
797 } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
798 *max_discard_sectors = UINT_MAX >> 9;
799 *granularity = sbuf.f_bsize;
800 }
801}
802
803struct loop_worker {
804 struct rb_node rb_node;
805 struct work_struct work;
806 struct list_head cmd_list;
807 struct list_head idle_list;
808 struct loop_device *lo;
809 struct cgroup_subsys_state *blkcg_css;
810 unsigned long last_ran_at;
811};
812
813static void loop_workfn(struct work_struct *work);
814
815#ifdef CONFIG_BLK_CGROUP
816static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
817{
818 return !css || css == blkcg_root_css;
819}
820#else
821static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
822{
823 return !css;
824}
825#endif
826
827static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
828{
829 struct rb_node **node, *parent = NULL;
830 struct loop_worker *cur_worker, *worker = NULL;
831 struct work_struct *work;
832 struct list_head *cmd_list;
833
834 spin_lock_irq(&lo->lo_work_lock);
835
836 if (queue_on_root_worker(cmd->blkcg_css))
837 goto queue_work;
838
839 node = &lo->worker_tree.rb_node;
840
841 while (*node) {
842 parent = *node;
843 cur_worker = container_of(*node, struct loop_worker, rb_node);
844 if (cur_worker->blkcg_css == cmd->blkcg_css) {
845 worker = cur_worker;
846 break;
847 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
848 node = &(*node)->rb_left;
849 } else {
850 node = &(*node)->rb_right;
851 }
852 }
853 if (worker)
854 goto queue_work;
855
856 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
857 /*
858 * In the event we cannot allocate a worker, just queue on the
859 * rootcg worker and issue the I/O as the rootcg
860 */
861 if (!worker) {
862 cmd->blkcg_css = NULL;
863 if (cmd->memcg_css)
864 css_put(cmd->memcg_css);
865 cmd->memcg_css = NULL;
866 goto queue_work;
867 }
868
869 worker->blkcg_css = cmd->blkcg_css;
870 css_get(worker->blkcg_css);
871 INIT_WORK(&worker->work, loop_workfn);
872 INIT_LIST_HEAD(&worker->cmd_list);
873 INIT_LIST_HEAD(&worker->idle_list);
874 worker->lo = lo;
875 rb_link_node(&worker->rb_node, parent, node);
876 rb_insert_color(&worker->rb_node, &lo->worker_tree);
877queue_work:
878 if (worker) {
879 /*
880 * We need to remove from the idle list here while
881 * holding the lock so that the idle timer doesn't
882 * free the worker
883 */
884 if (!list_empty(&worker->idle_list))
885 list_del_init(&worker->idle_list);
886 work = &worker->work;
887 cmd_list = &worker->cmd_list;
888 } else {
889 work = &lo->rootcg_work;
890 cmd_list = &lo->rootcg_cmd_list;
891 }
892 list_add_tail(&cmd->list_entry, cmd_list);
893 queue_work(lo->workqueue, work);
894 spin_unlock_irq(&lo->lo_work_lock);
895}
896
897static void loop_set_timer(struct loop_device *lo)
898{
899 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
900}
901
902static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
903{
904 struct loop_worker *pos, *worker;
905
906 spin_lock_irq(&lo->lo_work_lock);
907 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
908 idle_list) {
909 if (!delete_all &&
910 time_is_after_jiffies(worker->last_ran_at +
911 LOOP_IDLE_WORKER_TIMEOUT))
912 break;
913 list_del(&worker->idle_list);
914 rb_erase(&worker->rb_node, &lo->worker_tree);
915 css_put(worker->blkcg_css);
916 kfree(worker);
917 }
918 if (!list_empty(&lo->idle_worker_list))
919 loop_set_timer(lo);
920 spin_unlock_irq(&lo->lo_work_lock);
921}
922
923static void loop_free_idle_workers_timer(struct timer_list *timer)
924{
925 struct loop_device *lo = container_of(timer, struct loop_device, timer);
926
927 return loop_free_idle_workers(lo, false);
928}
929
930/**
931 * loop_set_status_from_info - configure device from loop_info
932 * @lo: struct loop_device to configure
933 * @info: struct loop_info64 to configure the device with
934 *
935 * Configures the loop device parameters according to the passed
936 * in loop_info64 configuration.
937 */
938static int
939loop_set_status_from_info(struct loop_device *lo,
940 const struct loop_info64 *info)
941{
942 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
943 return -EINVAL;
944
945 switch (info->lo_encrypt_type) {
946 case LO_CRYPT_NONE:
947 break;
948 case LO_CRYPT_XOR:
949 pr_warn("support for the xor transformation has been removed.\n");
950 return -EINVAL;
951 case LO_CRYPT_CRYPTOAPI:
952 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
953 return -EINVAL;
954 default:
955 return -EINVAL;
956 }
957
958 /* Avoid assigning overflow values */
959 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
960 return -EOVERFLOW;
961
962 lo->lo_offset = info->lo_offset;
963 lo->lo_sizelimit = info->lo_sizelimit;
964
965 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
966 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
967 lo->lo_flags = info->lo_flags;
968 return 0;
969}
970
971static unsigned int loop_default_blocksize(struct loop_device *lo,
972 struct block_device *backing_bdev)
973{
974 /* In case of direct I/O, match underlying block size */
975 if ((lo->lo_backing_file->f_flags & O_DIRECT) && backing_bdev)
976 return bdev_logical_block_size(backing_bdev);
977 return SECTOR_SIZE;
978}
979
980static int loop_reconfigure_limits(struct loop_device *lo, unsigned int bsize)
981{
982 struct file *file = lo->lo_backing_file;
983 struct inode *inode = file->f_mapping->host;
984 struct block_device *backing_bdev = NULL;
985 struct queue_limits lim;
986 u32 granularity = 0, max_discard_sectors = 0;
987
988 if (S_ISBLK(inode->i_mode))
989 backing_bdev = I_BDEV(inode);
990 else if (inode->i_sb->s_bdev)
991 backing_bdev = inode->i_sb->s_bdev;
992
993 if (!bsize)
994 bsize = loop_default_blocksize(lo, backing_bdev);
995
996 loop_get_discard_config(lo, &granularity, &max_discard_sectors);
997
998 lim = queue_limits_start_update(lo->lo_queue);
999 lim.logical_block_size = bsize;
1000 lim.physical_block_size = bsize;
1001 lim.io_min = bsize;
1002 lim.features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL);
1003 if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY))
1004 lim.features |= BLK_FEAT_WRITE_CACHE;
1005 if (backing_bdev && !bdev_nonrot(backing_bdev))
1006 lim.features |= BLK_FEAT_ROTATIONAL;
1007 lim.max_hw_discard_sectors = max_discard_sectors;
1008 lim.max_write_zeroes_sectors = max_discard_sectors;
1009 if (max_discard_sectors)
1010 lim.discard_granularity = granularity;
1011 else
1012 lim.discard_granularity = 0;
1013 return queue_limits_commit_update(lo->lo_queue, &lim);
1014}
1015
1016static int loop_configure(struct loop_device *lo, blk_mode_t mode,
1017 struct block_device *bdev,
1018 const struct loop_config *config)
1019{
1020 struct file *file = fget(config->fd);
1021 struct address_space *mapping;
1022 int error;
1023 loff_t size;
1024 bool partscan;
1025 bool is_loop;
1026
1027 if (!file)
1028 return -EBADF;
1029 is_loop = is_loop_device(file);
1030
1031 /* This is safe, since we have a reference from open(). */
1032 __module_get(THIS_MODULE);
1033
1034 /*
1035 * If we don't hold exclusive handle for the device, upgrade to it
1036 * here to avoid changing device under exclusive owner.
1037 */
1038 if (!(mode & BLK_OPEN_EXCL)) {
1039 error = bd_prepare_to_claim(bdev, loop_configure, NULL);
1040 if (error)
1041 goto out_putf;
1042 }
1043
1044 error = loop_global_lock_killable(lo, is_loop);
1045 if (error)
1046 goto out_bdev;
1047
1048 error = -EBUSY;
1049 if (lo->lo_state != Lo_unbound)
1050 goto out_unlock;
1051
1052 error = loop_validate_file(file, bdev);
1053 if (error)
1054 goto out_unlock;
1055
1056 mapping = file->f_mapping;
1057
1058 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1059 error = -EINVAL;
1060 goto out_unlock;
1061 }
1062
1063 error = loop_set_status_from_info(lo, &config->info);
1064 if (error)
1065 goto out_unlock;
1066
1067 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1068 !file->f_op->write_iter)
1069 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1070
1071 if (!lo->workqueue) {
1072 lo->workqueue = alloc_workqueue("loop%d",
1073 WQ_UNBOUND | WQ_FREEZABLE,
1074 0, lo->lo_number);
1075 if (!lo->workqueue) {
1076 error = -ENOMEM;
1077 goto out_unlock;
1078 }
1079 }
1080
1081 /* suppress uevents while reconfiguring the device */
1082 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1083
1084 disk_force_media_change(lo->lo_disk);
1085 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1086
1087 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1088 lo->lo_device = bdev;
1089 lo->lo_backing_file = file;
1090 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1091 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1092
1093 error = loop_reconfigure_limits(lo, config->block_size);
1094 if (error)
1095 goto out_unlock;
1096
1097 loop_update_dio(lo);
1098 loop_sysfs_init(lo);
1099
1100 size = get_loop_size(lo, file);
1101 loop_set_size(lo, size);
1102
1103 /* Order wrt reading lo_state in loop_validate_file(). */
1104 wmb();
1105
1106 lo->lo_state = Lo_bound;
1107 if (part_shift)
1108 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1109 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1110 if (partscan)
1111 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1112
1113 /* enable and uncork uevent now that we are done */
1114 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1115
1116 loop_global_unlock(lo, is_loop);
1117 if (partscan)
1118 loop_reread_partitions(lo);
1119
1120 if (!(mode & BLK_OPEN_EXCL))
1121 bd_abort_claiming(bdev, loop_configure);
1122
1123 return 0;
1124
1125out_unlock:
1126 loop_global_unlock(lo, is_loop);
1127out_bdev:
1128 if (!(mode & BLK_OPEN_EXCL))
1129 bd_abort_claiming(bdev, loop_configure);
1130out_putf:
1131 fput(file);
1132 /* This is safe: open() is still holding a reference. */
1133 module_put(THIS_MODULE);
1134 return error;
1135}
1136
1137static void __loop_clr_fd(struct loop_device *lo)
1138{
1139 struct queue_limits lim;
1140 struct file *filp;
1141 gfp_t gfp = lo->old_gfp_mask;
1142
1143 spin_lock_irq(&lo->lo_lock);
1144 filp = lo->lo_backing_file;
1145 lo->lo_backing_file = NULL;
1146 spin_unlock_irq(&lo->lo_lock);
1147
1148 lo->lo_device = NULL;
1149 lo->lo_offset = 0;
1150 lo->lo_sizelimit = 0;
1151 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1152
1153 /* reset the block size to the default */
1154 lim = queue_limits_start_update(lo->lo_queue);
1155 lim.logical_block_size = SECTOR_SIZE;
1156 lim.physical_block_size = SECTOR_SIZE;
1157 lim.io_min = SECTOR_SIZE;
1158 queue_limits_commit_update(lo->lo_queue, &lim);
1159
1160 invalidate_disk(lo->lo_disk);
1161 loop_sysfs_exit(lo);
1162 /* let user-space know about this change */
1163 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1164 mapping_set_gfp_mask(filp->f_mapping, gfp);
1165 /* This is safe: open() is still holding a reference. */
1166 module_put(THIS_MODULE);
1167
1168 disk_force_media_change(lo->lo_disk);
1169
1170 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1171 int err;
1172
1173 /*
1174 * open_mutex has been held already in release path, so don't
1175 * acquire it if this function is called in such case.
1176 *
1177 * If the reread partition isn't from release path, lo_refcnt
1178 * must be at least one and it can only become zero when the
1179 * current holder is released.
1180 */
1181 err = bdev_disk_changed(lo->lo_disk, false);
1182 if (err)
1183 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1184 __func__, lo->lo_number, err);
1185 /* Device is gone, no point in returning error */
1186 }
1187
1188 /*
1189 * lo->lo_state is set to Lo_unbound here after above partscan has
1190 * finished. There cannot be anybody else entering __loop_clr_fd() as
1191 * Lo_rundown state protects us from all the other places trying to
1192 * change the 'lo' device.
1193 */
1194 lo->lo_flags = 0;
1195 if (!part_shift)
1196 set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1197 mutex_lock(&lo->lo_mutex);
1198 lo->lo_state = Lo_unbound;
1199 mutex_unlock(&lo->lo_mutex);
1200
1201 /*
1202 * Need not hold lo_mutex to fput backing file. Calling fput holding
1203 * lo_mutex triggers a circular lock dependency possibility warning as
1204 * fput can take open_mutex which is usually taken before lo_mutex.
1205 */
1206 fput(filp);
1207}
1208
1209static int loop_clr_fd(struct loop_device *lo)
1210{
1211 int err;
1212
1213 /*
1214 * Since lo_ioctl() is called without locks held, it is possible that
1215 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1216 *
1217 * Therefore, use global lock when setting Lo_rundown state in order to
1218 * make sure that loop_validate_file() will fail if the "struct file"
1219 * which loop_configure()/loop_change_fd() found via fget() was this
1220 * loop device.
1221 */
1222 err = loop_global_lock_killable(lo, true);
1223 if (err)
1224 return err;
1225 if (lo->lo_state != Lo_bound) {
1226 loop_global_unlock(lo, true);
1227 return -ENXIO;
1228 }
1229 /*
1230 * Mark the device for removing the backing device on last close.
1231 * If we are the only opener, also switch the state to roundown here to
1232 * prevent new openers from coming in.
1233 */
1234
1235 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1236 if (disk_openers(lo->lo_disk) == 1)
1237 lo->lo_state = Lo_rundown;
1238 loop_global_unlock(lo, true);
1239
1240 return 0;
1241}
1242
1243static int
1244loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1245{
1246 int err;
1247 int prev_lo_flags;
1248 bool partscan = false;
1249 bool size_changed = false;
1250
1251 err = mutex_lock_killable(&lo->lo_mutex);
1252 if (err)
1253 return err;
1254 if (lo->lo_state != Lo_bound) {
1255 err = -ENXIO;
1256 goto out_unlock;
1257 }
1258
1259 if (lo->lo_offset != info->lo_offset ||
1260 lo->lo_sizelimit != info->lo_sizelimit) {
1261 size_changed = true;
1262 sync_blockdev(lo->lo_device);
1263 invalidate_bdev(lo->lo_device);
1264 }
1265
1266 /* I/O need to be drained during transfer transition */
1267 blk_mq_freeze_queue(lo->lo_queue);
1268
1269 prev_lo_flags = lo->lo_flags;
1270
1271 err = loop_set_status_from_info(lo, info);
1272 if (err)
1273 goto out_unfreeze;
1274
1275 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1276 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1277 /* For those flags, use the previous values instead */
1278 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1279 /* For flags that can't be cleared, use previous values too */
1280 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1281
1282 if (size_changed) {
1283 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1284 lo->lo_backing_file);
1285 loop_set_size(lo, new_size);
1286 }
1287
1288 /* update dio if lo_offset or transfer is changed */
1289 __loop_update_dio(lo, lo->use_dio);
1290
1291out_unfreeze:
1292 blk_mq_unfreeze_queue(lo->lo_queue);
1293
1294 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1295 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1296 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1297 partscan = true;
1298 }
1299out_unlock:
1300 mutex_unlock(&lo->lo_mutex);
1301 if (partscan)
1302 loop_reread_partitions(lo);
1303
1304 return err;
1305}
1306
1307static int
1308loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1309{
1310 struct path path;
1311 struct kstat stat;
1312 int ret;
1313
1314 ret = mutex_lock_killable(&lo->lo_mutex);
1315 if (ret)
1316 return ret;
1317 if (lo->lo_state != Lo_bound) {
1318 mutex_unlock(&lo->lo_mutex);
1319 return -ENXIO;
1320 }
1321
1322 memset(info, 0, sizeof(*info));
1323 info->lo_number = lo->lo_number;
1324 info->lo_offset = lo->lo_offset;
1325 info->lo_sizelimit = lo->lo_sizelimit;
1326 info->lo_flags = lo->lo_flags;
1327 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1328
1329 /* Drop lo_mutex while we call into the filesystem. */
1330 path = lo->lo_backing_file->f_path;
1331 path_get(&path);
1332 mutex_unlock(&lo->lo_mutex);
1333 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1334 if (!ret) {
1335 info->lo_device = huge_encode_dev(stat.dev);
1336 info->lo_inode = stat.ino;
1337 info->lo_rdevice = huge_encode_dev(stat.rdev);
1338 }
1339 path_put(&path);
1340 return ret;
1341}
1342
1343static void
1344loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1345{
1346 memset(info64, 0, sizeof(*info64));
1347 info64->lo_number = info->lo_number;
1348 info64->lo_device = info->lo_device;
1349 info64->lo_inode = info->lo_inode;
1350 info64->lo_rdevice = info->lo_rdevice;
1351 info64->lo_offset = info->lo_offset;
1352 info64->lo_sizelimit = 0;
1353 info64->lo_flags = info->lo_flags;
1354 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1355}
1356
1357static int
1358loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1359{
1360 memset(info, 0, sizeof(*info));
1361 info->lo_number = info64->lo_number;
1362 info->lo_device = info64->lo_device;
1363 info->lo_inode = info64->lo_inode;
1364 info->lo_rdevice = info64->lo_rdevice;
1365 info->lo_offset = info64->lo_offset;
1366 info->lo_flags = info64->lo_flags;
1367 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1368
1369 /* error in case values were truncated */
1370 if (info->lo_device != info64->lo_device ||
1371 info->lo_rdevice != info64->lo_rdevice ||
1372 info->lo_inode != info64->lo_inode ||
1373 info->lo_offset != info64->lo_offset)
1374 return -EOVERFLOW;
1375
1376 return 0;
1377}
1378
1379static int
1380loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1381{
1382 struct loop_info info;
1383 struct loop_info64 info64;
1384
1385 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1386 return -EFAULT;
1387 loop_info64_from_old(&info, &info64);
1388 return loop_set_status(lo, &info64);
1389}
1390
1391static int
1392loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1393{
1394 struct loop_info64 info64;
1395
1396 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1397 return -EFAULT;
1398 return loop_set_status(lo, &info64);
1399}
1400
1401static int
1402loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1403 struct loop_info info;
1404 struct loop_info64 info64;
1405 int err;
1406
1407 if (!arg)
1408 return -EINVAL;
1409 err = loop_get_status(lo, &info64);
1410 if (!err)
1411 err = loop_info64_to_old(&info64, &info);
1412 if (!err && copy_to_user(arg, &info, sizeof(info)))
1413 err = -EFAULT;
1414
1415 return err;
1416}
1417
1418static int
1419loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1420 struct loop_info64 info64;
1421 int err;
1422
1423 if (!arg)
1424 return -EINVAL;
1425 err = loop_get_status(lo, &info64);
1426 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1427 err = -EFAULT;
1428
1429 return err;
1430}
1431
1432static int loop_set_capacity(struct loop_device *lo)
1433{
1434 loff_t size;
1435
1436 if (unlikely(lo->lo_state != Lo_bound))
1437 return -ENXIO;
1438
1439 size = get_loop_size(lo, lo->lo_backing_file);
1440 loop_set_size(lo, size);
1441
1442 return 0;
1443}
1444
1445static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1446{
1447 int error = -ENXIO;
1448 if (lo->lo_state != Lo_bound)
1449 goto out;
1450
1451 __loop_update_dio(lo, !!arg);
1452 if (lo->use_dio == !!arg)
1453 return 0;
1454 error = -EINVAL;
1455 out:
1456 return error;
1457}
1458
1459static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1460{
1461 int err = 0;
1462
1463 if (lo->lo_state != Lo_bound)
1464 return -ENXIO;
1465
1466 if (lo->lo_queue->limits.logical_block_size == arg)
1467 return 0;
1468
1469 sync_blockdev(lo->lo_device);
1470 invalidate_bdev(lo->lo_device);
1471
1472 blk_mq_freeze_queue(lo->lo_queue);
1473 err = loop_reconfigure_limits(lo, arg);
1474 loop_update_dio(lo);
1475 blk_mq_unfreeze_queue(lo->lo_queue);
1476
1477 return err;
1478}
1479
1480static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1481 unsigned long arg)
1482{
1483 int err;
1484
1485 err = mutex_lock_killable(&lo->lo_mutex);
1486 if (err)
1487 return err;
1488 switch (cmd) {
1489 case LOOP_SET_CAPACITY:
1490 err = loop_set_capacity(lo);
1491 break;
1492 case LOOP_SET_DIRECT_IO:
1493 err = loop_set_dio(lo, arg);
1494 break;
1495 case LOOP_SET_BLOCK_SIZE:
1496 err = loop_set_block_size(lo, arg);
1497 break;
1498 default:
1499 err = -EINVAL;
1500 }
1501 mutex_unlock(&lo->lo_mutex);
1502 return err;
1503}
1504
1505static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1506 unsigned int cmd, unsigned long arg)
1507{
1508 struct loop_device *lo = bdev->bd_disk->private_data;
1509 void __user *argp = (void __user *) arg;
1510 int err;
1511
1512 switch (cmd) {
1513 case LOOP_SET_FD: {
1514 /*
1515 * Legacy case - pass in a zeroed out struct loop_config with
1516 * only the file descriptor set , which corresponds with the
1517 * default parameters we'd have used otherwise.
1518 */
1519 struct loop_config config;
1520
1521 memset(&config, 0, sizeof(config));
1522 config.fd = arg;
1523
1524 return loop_configure(lo, mode, bdev, &config);
1525 }
1526 case LOOP_CONFIGURE: {
1527 struct loop_config config;
1528
1529 if (copy_from_user(&config, argp, sizeof(config)))
1530 return -EFAULT;
1531
1532 return loop_configure(lo, mode, bdev, &config);
1533 }
1534 case LOOP_CHANGE_FD:
1535 return loop_change_fd(lo, bdev, arg);
1536 case LOOP_CLR_FD:
1537 return loop_clr_fd(lo);
1538 case LOOP_SET_STATUS:
1539 err = -EPERM;
1540 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1541 err = loop_set_status_old(lo, argp);
1542 break;
1543 case LOOP_GET_STATUS:
1544 return loop_get_status_old(lo, argp);
1545 case LOOP_SET_STATUS64:
1546 err = -EPERM;
1547 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1548 err = loop_set_status64(lo, argp);
1549 break;
1550 case LOOP_GET_STATUS64:
1551 return loop_get_status64(lo, argp);
1552 case LOOP_SET_CAPACITY:
1553 case LOOP_SET_DIRECT_IO:
1554 case LOOP_SET_BLOCK_SIZE:
1555 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1556 return -EPERM;
1557 fallthrough;
1558 default:
1559 err = lo_simple_ioctl(lo, cmd, arg);
1560 break;
1561 }
1562
1563 return err;
1564}
1565
1566#ifdef CONFIG_COMPAT
1567struct compat_loop_info {
1568 compat_int_t lo_number; /* ioctl r/o */
1569 compat_dev_t lo_device; /* ioctl r/o */
1570 compat_ulong_t lo_inode; /* ioctl r/o */
1571 compat_dev_t lo_rdevice; /* ioctl r/o */
1572 compat_int_t lo_offset;
1573 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1574 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1575 compat_int_t lo_flags; /* ioctl r/o */
1576 char lo_name[LO_NAME_SIZE];
1577 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1578 compat_ulong_t lo_init[2];
1579 char reserved[4];
1580};
1581
1582/*
1583 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1584 * - noinlined to reduce stack space usage in main part of driver
1585 */
1586static noinline int
1587loop_info64_from_compat(const struct compat_loop_info __user *arg,
1588 struct loop_info64 *info64)
1589{
1590 struct compat_loop_info info;
1591
1592 if (copy_from_user(&info, arg, sizeof(info)))
1593 return -EFAULT;
1594
1595 memset(info64, 0, sizeof(*info64));
1596 info64->lo_number = info.lo_number;
1597 info64->lo_device = info.lo_device;
1598 info64->lo_inode = info.lo_inode;
1599 info64->lo_rdevice = info.lo_rdevice;
1600 info64->lo_offset = info.lo_offset;
1601 info64->lo_sizelimit = 0;
1602 info64->lo_flags = info.lo_flags;
1603 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1604 return 0;
1605}
1606
1607/*
1608 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1609 * - noinlined to reduce stack space usage in main part of driver
1610 */
1611static noinline int
1612loop_info64_to_compat(const struct loop_info64 *info64,
1613 struct compat_loop_info __user *arg)
1614{
1615 struct compat_loop_info info;
1616
1617 memset(&info, 0, sizeof(info));
1618 info.lo_number = info64->lo_number;
1619 info.lo_device = info64->lo_device;
1620 info.lo_inode = info64->lo_inode;
1621 info.lo_rdevice = info64->lo_rdevice;
1622 info.lo_offset = info64->lo_offset;
1623 info.lo_flags = info64->lo_flags;
1624 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1625
1626 /* error in case values were truncated */
1627 if (info.lo_device != info64->lo_device ||
1628 info.lo_rdevice != info64->lo_rdevice ||
1629 info.lo_inode != info64->lo_inode ||
1630 info.lo_offset != info64->lo_offset)
1631 return -EOVERFLOW;
1632
1633 if (copy_to_user(arg, &info, sizeof(info)))
1634 return -EFAULT;
1635 return 0;
1636}
1637
1638static int
1639loop_set_status_compat(struct loop_device *lo,
1640 const struct compat_loop_info __user *arg)
1641{
1642 struct loop_info64 info64;
1643 int ret;
1644
1645 ret = loop_info64_from_compat(arg, &info64);
1646 if (ret < 0)
1647 return ret;
1648 return loop_set_status(lo, &info64);
1649}
1650
1651static int
1652loop_get_status_compat(struct loop_device *lo,
1653 struct compat_loop_info __user *arg)
1654{
1655 struct loop_info64 info64;
1656 int err;
1657
1658 if (!arg)
1659 return -EINVAL;
1660 err = loop_get_status(lo, &info64);
1661 if (!err)
1662 err = loop_info64_to_compat(&info64, arg);
1663 return err;
1664}
1665
1666static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1667 unsigned int cmd, unsigned long arg)
1668{
1669 struct loop_device *lo = bdev->bd_disk->private_data;
1670 int err;
1671
1672 switch(cmd) {
1673 case LOOP_SET_STATUS:
1674 err = loop_set_status_compat(lo,
1675 (const struct compat_loop_info __user *)arg);
1676 break;
1677 case LOOP_GET_STATUS:
1678 err = loop_get_status_compat(lo,
1679 (struct compat_loop_info __user *)arg);
1680 break;
1681 case LOOP_SET_CAPACITY:
1682 case LOOP_CLR_FD:
1683 case LOOP_GET_STATUS64:
1684 case LOOP_SET_STATUS64:
1685 case LOOP_CONFIGURE:
1686 arg = (unsigned long) compat_ptr(arg);
1687 fallthrough;
1688 case LOOP_SET_FD:
1689 case LOOP_CHANGE_FD:
1690 case LOOP_SET_BLOCK_SIZE:
1691 case LOOP_SET_DIRECT_IO:
1692 err = lo_ioctl(bdev, mode, cmd, arg);
1693 break;
1694 default:
1695 err = -ENOIOCTLCMD;
1696 break;
1697 }
1698 return err;
1699}
1700#endif
1701
1702static int lo_open(struct gendisk *disk, blk_mode_t mode)
1703{
1704 struct loop_device *lo = disk->private_data;
1705 int err;
1706
1707 err = mutex_lock_killable(&lo->lo_mutex);
1708 if (err)
1709 return err;
1710
1711 if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown)
1712 err = -ENXIO;
1713 mutex_unlock(&lo->lo_mutex);
1714 return err;
1715}
1716
1717static void lo_release(struct gendisk *disk)
1718{
1719 struct loop_device *lo = disk->private_data;
1720 bool need_clear = false;
1721
1722 if (disk_openers(disk) > 0)
1723 return;
1724 /*
1725 * Clear the backing device information if this is the last close of
1726 * a device that's been marked for auto clear, or on which LOOP_CLR_FD
1727 * has been called.
1728 */
1729
1730 mutex_lock(&lo->lo_mutex);
1731 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR))
1732 lo->lo_state = Lo_rundown;
1733
1734 need_clear = (lo->lo_state == Lo_rundown);
1735 mutex_unlock(&lo->lo_mutex);
1736
1737 if (need_clear)
1738 __loop_clr_fd(lo);
1739}
1740
1741static void lo_free_disk(struct gendisk *disk)
1742{
1743 struct loop_device *lo = disk->private_data;
1744
1745 if (lo->workqueue)
1746 destroy_workqueue(lo->workqueue);
1747 loop_free_idle_workers(lo, true);
1748 timer_shutdown_sync(&lo->timer);
1749 mutex_destroy(&lo->lo_mutex);
1750 kfree(lo);
1751}
1752
1753static const struct block_device_operations lo_fops = {
1754 .owner = THIS_MODULE,
1755 .open = lo_open,
1756 .release = lo_release,
1757 .ioctl = lo_ioctl,
1758#ifdef CONFIG_COMPAT
1759 .compat_ioctl = lo_compat_ioctl,
1760#endif
1761 .free_disk = lo_free_disk,
1762};
1763
1764/*
1765 * And now the modules code and kernel interface.
1766 */
1767
1768/*
1769 * If max_loop is specified, create that many devices upfront.
1770 * This also becomes a hard limit. If max_loop is not specified,
1771 * the default isn't a hard limit (as before commit 85c50197716c
1772 * changed the default value from 0 for max_loop=0 reasons), just
1773 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1774 * init time. Loop devices can be requested on-demand with the
1775 * /dev/loop-control interface, or be instantiated by accessing
1776 * a 'dead' device node.
1777 */
1778static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1779
1780#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1781static bool max_loop_specified;
1782
1783static int max_loop_param_set_int(const char *val,
1784 const struct kernel_param *kp)
1785{
1786 int ret;
1787
1788 ret = param_set_int(val, kp);
1789 if (ret < 0)
1790 return ret;
1791
1792 max_loop_specified = true;
1793 return 0;
1794}
1795
1796static const struct kernel_param_ops max_loop_param_ops = {
1797 .set = max_loop_param_set_int,
1798 .get = param_get_int,
1799};
1800
1801module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1802MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1803#else
1804module_param(max_loop, int, 0444);
1805MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1806#endif
1807
1808module_param(max_part, int, 0444);
1809MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1810
1811static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1812
1813static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1814{
1815 int qd, ret;
1816
1817 ret = kstrtoint(s, 0, &qd);
1818 if (ret < 0)
1819 return ret;
1820 if (qd < 1)
1821 return -EINVAL;
1822 hw_queue_depth = qd;
1823 return 0;
1824}
1825
1826static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1827 .set = loop_set_hw_queue_depth,
1828 .get = param_get_int,
1829};
1830
1831device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1832MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1833
1834MODULE_DESCRIPTION("Loopback device support");
1835MODULE_LICENSE("GPL");
1836MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1837
1838static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1839 const struct blk_mq_queue_data *bd)
1840{
1841 struct request *rq = bd->rq;
1842 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1843 struct loop_device *lo = rq->q->queuedata;
1844
1845 blk_mq_start_request(rq);
1846
1847 if (lo->lo_state != Lo_bound)
1848 return BLK_STS_IOERR;
1849
1850 switch (req_op(rq)) {
1851 case REQ_OP_FLUSH:
1852 case REQ_OP_DISCARD:
1853 case REQ_OP_WRITE_ZEROES:
1854 cmd->use_aio = false;
1855 break;
1856 default:
1857 cmd->use_aio = lo->use_dio;
1858 break;
1859 }
1860
1861 /* always use the first bio's css */
1862 cmd->blkcg_css = NULL;
1863 cmd->memcg_css = NULL;
1864#ifdef CONFIG_BLK_CGROUP
1865 if (rq->bio) {
1866 cmd->blkcg_css = bio_blkcg_css(rq->bio);
1867#ifdef CONFIG_MEMCG
1868 if (cmd->blkcg_css) {
1869 cmd->memcg_css =
1870 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1871 &memory_cgrp_subsys);
1872 }
1873#endif
1874 }
1875#endif
1876 loop_queue_work(lo, cmd);
1877
1878 return BLK_STS_OK;
1879}
1880
1881static void loop_handle_cmd(struct loop_cmd *cmd)
1882{
1883 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1884 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1885 struct request *rq = blk_mq_rq_from_pdu(cmd);
1886 const bool write = op_is_write(req_op(rq));
1887 struct loop_device *lo = rq->q->queuedata;
1888 int ret = 0;
1889 struct mem_cgroup *old_memcg = NULL;
1890 const bool use_aio = cmd->use_aio;
1891
1892 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1893 ret = -EIO;
1894 goto failed;
1895 }
1896
1897 if (cmd_blkcg_css)
1898 kthread_associate_blkcg(cmd_blkcg_css);
1899 if (cmd_memcg_css)
1900 old_memcg = set_active_memcg(
1901 mem_cgroup_from_css(cmd_memcg_css));
1902
1903 /*
1904 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1905 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1906 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1907 * not yet been completed.
1908 */
1909 ret = do_req_filebacked(lo, rq);
1910
1911 if (cmd_blkcg_css)
1912 kthread_associate_blkcg(NULL);
1913
1914 if (cmd_memcg_css) {
1915 set_active_memcg(old_memcg);
1916 css_put(cmd_memcg_css);
1917 }
1918 failed:
1919 /* complete non-aio request */
1920 if (!use_aio || ret) {
1921 if (ret == -EOPNOTSUPP)
1922 cmd->ret = ret;
1923 else
1924 cmd->ret = ret ? -EIO : 0;
1925 if (likely(!blk_should_fake_timeout(rq->q)))
1926 blk_mq_complete_request(rq);
1927 }
1928}
1929
1930static void loop_process_work(struct loop_worker *worker,
1931 struct list_head *cmd_list, struct loop_device *lo)
1932{
1933 int orig_flags = current->flags;
1934 struct loop_cmd *cmd;
1935
1936 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1937 spin_lock_irq(&lo->lo_work_lock);
1938 while (!list_empty(cmd_list)) {
1939 cmd = container_of(
1940 cmd_list->next, struct loop_cmd, list_entry);
1941 list_del(cmd_list->next);
1942 spin_unlock_irq(&lo->lo_work_lock);
1943
1944 loop_handle_cmd(cmd);
1945 cond_resched();
1946
1947 spin_lock_irq(&lo->lo_work_lock);
1948 }
1949
1950 /*
1951 * We only add to the idle list if there are no pending cmds
1952 * *and* the worker will not run again which ensures that it
1953 * is safe to free any worker on the idle list
1954 */
1955 if (worker && !work_pending(&worker->work)) {
1956 worker->last_ran_at = jiffies;
1957 list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1958 loop_set_timer(lo);
1959 }
1960 spin_unlock_irq(&lo->lo_work_lock);
1961 current->flags = orig_flags;
1962}
1963
1964static void loop_workfn(struct work_struct *work)
1965{
1966 struct loop_worker *worker =
1967 container_of(work, struct loop_worker, work);
1968 loop_process_work(worker, &worker->cmd_list, worker->lo);
1969}
1970
1971static void loop_rootcg_workfn(struct work_struct *work)
1972{
1973 struct loop_device *lo =
1974 container_of(work, struct loop_device, rootcg_work);
1975 loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1976}
1977
1978static const struct blk_mq_ops loop_mq_ops = {
1979 .queue_rq = loop_queue_rq,
1980 .complete = lo_complete_rq,
1981};
1982
1983static int loop_add(int i)
1984{
1985 struct queue_limits lim = {
1986 /*
1987 * Random number picked from the historic block max_sectors cap.
1988 */
1989 .max_hw_sectors = 2560u,
1990 };
1991 struct loop_device *lo;
1992 struct gendisk *disk;
1993 int err;
1994
1995 err = -ENOMEM;
1996 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1997 if (!lo)
1998 goto out;
1999 lo->worker_tree = RB_ROOT;
2000 INIT_LIST_HEAD(&lo->idle_worker_list);
2001 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
2002 lo->lo_state = Lo_unbound;
2003
2004 err = mutex_lock_killable(&loop_ctl_mutex);
2005 if (err)
2006 goto out_free_dev;
2007
2008 /* allocate id, if @id >= 0, we're requesting that specific id */
2009 if (i >= 0) {
2010 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2011 if (err == -ENOSPC)
2012 err = -EEXIST;
2013 } else {
2014 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2015 }
2016 mutex_unlock(&loop_ctl_mutex);
2017 if (err < 0)
2018 goto out_free_dev;
2019 i = err;
2020
2021 lo->tag_set.ops = &loop_mq_ops;
2022 lo->tag_set.nr_hw_queues = 1;
2023 lo->tag_set.queue_depth = hw_queue_depth;
2024 lo->tag_set.numa_node = NUMA_NO_NODE;
2025 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2026 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
2027 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2028 lo->tag_set.driver_data = lo;
2029
2030 err = blk_mq_alloc_tag_set(&lo->tag_set);
2031 if (err)
2032 goto out_free_idr;
2033
2034 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2035 if (IS_ERR(disk)) {
2036 err = PTR_ERR(disk);
2037 goto out_cleanup_tags;
2038 }
2039 lo->lo_queue = lo->lo_disk->queue;
2040
2041 /*
2042 * Disable partition scanning by default. The in-kernel partition
2043 * scanning can be requested individually per-device during its
2044 * setup. Userspace can always add and remove partitions from all
2045 * devices. The needed partition minors are allocated from the
2046 * extended minor space, the main loop device numbers will continue
2047 * to match the loop minors, regardless of the number of partitions
2048 * used.
2049 *
2050 * If max_part is given, partition scanning is globally enabled for
2051 * all loop devices. The minors for the main loop devices will be
2052 * multiples of max_part.
2053 *
2054 * Note: Global-for-all-devices, set-only-at-init, read-only module
2055 * parameteters like 'max_loop' and 'max_part' make things needlessly
2056 * complicated, are too static, inflexible and may surprise
2057 * userspace tools. Parameters like this in general should be avoided.
2058 */
2059 if (!part_shift)
2060 set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2061 mutex_init(&lo->lo_mutex);
2062 lo->lo_number = i;
2063 spin_lock_init(&lo->lo_lock);
2064 spin_lock_init(&lo->lo_work_lock);
2065 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2066 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2067 disk->major = LOOP_MAJOR;
2068 disk->first_minor = i << part_shift;
2069 disk->minors = 1 << part_shift;
2070 disk->fops = &lo_fops;
2071 disk->private_data = lo;
2072 disk->queue = lo->lo_queue;
2073 disk->events = DISK_EVENT_MEDIA_CHANGE;
2074 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2075 sprintf(disk->disk_name, "loop%d", i);
2076 /* Make this loop device reachable from pathname. */
2077 err = add_disk(disk);
2078 if (err)
2079 goto out_cleanup_disk;
2080
2081 /* Show this loop device. */
2082 mutex_lock(&loop_ctl_mutex);
2083 lo->idr_visible = true;
2084 mutex_unlock(&loop_ctl_mutex);
2085
2086 return i;
2087
2088out_cleanup_disk:
2089 put_disk(disk);
2090out_cleanup_tags:
2091 blk_mq_free_tag_set(&lo->tag_set);
2092out_free_idr:
2093 mutex_lock(&loop_ctl_mutex);
2094 idr_remove(&loop_index_idr, i);
2095 mutex_unlock(&loop_ctl_mutex);
2096out_free_dev:
2097 kfree(lo);
2098out:
2099 return err;
2100}
2101
2102static void loop_remove(struct loop_device *lo)
2103{
2104 /* Make this loop device unreachable from pathname. */
2105 del_gendisk(lo->lo_disk);
2106 blk_mq_free_tag_set(&lo->tag_set);
2107
2108 mutex_lock(&loop_ctl_mutex);
2109 idr_remove(&loop_index_idr, lo->lo_number);
2110 mutex_unlock(&loop_ctl_mutex);
2111
2112 put_disk(lo->lo_disk);
2113}
2114
2115#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2116static void loop_probe(dev_t dev)
2117{
2118 int idx = MINOR(dev) >> part_shift;
2119
2120 if (max_loop_specified && max_loop && idx >= max_loop)
2121 return;
2122 loop_add(idx);
2123}
2124#else
2125#define loop_probe NULL
2126#endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2127
2128static int loop_control_remove(int idx)
2129{
2130 struct loop_device *lo;
2131 int ret;
2132
2133 if (idx < 0) {
2134 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2135 return -EINVAL;
2136 }
2137
2138 /* Hide this loop device for serialization. */
2139 ret = mutex_lock_killable(&loop_ctl_mutex);
2140 if (ret)
2141 return ret;
2142 lo = idr_find(&loop_index_idr, idx);
2143 if (!lo || !lo->idr_visible)
2144 ret = -ENODEV;
2145 else
2146 lo->idr_visible = false;
2147 mutex_unlock(&loop_ctl_mutex);
2148 if (ret)
2149 return ret;
2150
2151 /* Check whether this loop device can be removed. */
2152 ret = mutex_lock_killable(&lo->lo_mutex);
2153 if (ret)
2154 goto mark_visible;
2155 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2156 mutex_unlock(&lo->lo_mutex);
2157 ret = -EBUSY;
2158 goto mark_visible;
2159 }
2160 /* Mark this loop device as no more bound, but not quite unbound yet */
2161 lo->lo_state = Lo_deleting;
2162 mutex_unlock(&lo->lo_mutex);
2163
2164 loop_remove(lo);
2165 return 0;
2166
2167mark_visible:
2168 /* Show this loop device again. */
2169 mutex_lock(&loop_ctl_mutex);
2170 lo->idr_visible = true;
2171 mutex_unlock(&loop_ctl_mutex);
2172 return ret;
2173}
2174
2175static int loop_control_get_free(int idx)
2176{
2177 struct loop_device *lo;
2178 int id, ret;
2179
2180 ret = mutex_lock_killable(&loop_ctl_mutex);
2181 if (ret)
2182 return ret;
2183 idr_for_each_entry(&loop_index_idr, lo, id) {
2184 /* Hitting a race results in creating a new loop device which is harmless. */
2185 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2186 goto found;
2187 }
2188 mutex_unlock(&loop_ctl_mutex);
2189 return loop_add(-1);
2190found:
2191 mutex_unlock(&loop_ctl_mutex);
2192 return id;
2193}
2194
2195static long loop_control_ioctl(struct file *file, unsigned int cmd,
2196 unsigned long parm)
2197{
2198 switch (cmd) {
2199 case LOOP_CTL_ADD:
2200 return loop_add(parm);
2201 case LOOP_CTL_REMOVE:
2202 return loop_control_remove(parm);
2203 case LOOP_CTL_GET_FREE:
2204 return loop_control_get_free(parm);
2205 default:
2206 return -ENOSYS;
2207 }
2208}
2209
2210static const struct file_operations loop_ctl_fops = {
2211 .open = nonseekable_open,
2212 .unlocked_ioctl = loop_control_ioctl,
2213 .compat_ioctl = loop_control_ioctl,
2214 .owner = THIS_MODULE,
2215 .llseek = noop_llseek,
2216};
2217
2218static struct miscdevice loop_misc = {
2219 .minor = LOOP_CTRL_MINOR,
2220 .name = "loop-control",
2221 .fops = &loop_ctl_fops,
2222};
2223
2224MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2225MODULE_ALIAS("devname:loop-control");
2226
2227static int __init loop_init(void)
2228{
2229 int i;
2230 int err;
2231
2232 part_shift = 0;
2233 if (max_part > 0) {
2234 part_shift = fls(max_part);
2235
2236 /*
2237 * Adjust max_part according to part_shift as it is exported
2238 * to user space so that user can decide correct minor number
2239 * if [s]he want to create more devices.
2240 *
2241 * Note that -1 is required because partition 0 is reserved
2242 * for the whole disk.
2243 */
2244 max_part = (1UL << part_shift) - 1;
2245 }
2246
2247 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2248 err = -EINVAL;
2249 goto err_out;
2250 }
2251
2252 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2253 err = -EINVAL;
2254 goto err_out;
2255 }
2256
2257 err = misc_register(&loop_misc);
2258 if (err < 0)
2259 goto err_out;
2260
2261
2262 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2263 err = -EIO;
2264 goto misc_out;
2265 }
2266
2267 /* pre-create number of devices given by config or max_loop */
2268 for (i = 0; i < max_loop; i++)
2269 loop_add(i);
2270
2271 printk(KERN_INFO "loop: module loaded\n");
2272 return 0;
2273
2274misc_out:
2275 misc_deregister(&loop_misc);
2276err_out:
2277 return err;
2278}
2279
2280static void __exit loop_exit(void)
2281{
2282 struct loop_device *lo;
2283 int id;
2284
2285 unregister_blkdev(LOOP_MAJOR, "loop");
2286 misc_deregister(&loop_misc);
2287
2288 /*
2289 * There is no need to use loop_ctl_mutex here, for nobody else can
2290 * access loop_index_idr when this module is unloading (unless forced
2291 * module unloading is requested). If this is not a clean unloading,
2292 * we have no means to avoid kernel crash.
2293 */
2294 idr_for_each_entry(&loop_index_idr, lo, id)
2295 loop_remove(lo);
2296
2297 idr_destroy(&loop_index_idr);
2298}
2299
2300module_init(loop_init);
2301module_exit(loop_exit);
2302
2303#ifndef MODULE
2304static int __init max_loop_setup(char *str)
2305{
2306 max_loop = simple_strtol(str, NULL, 0);
2307#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2308 max_loop_specified = true;
2309#endif
2310 return 1;
2311}
2312
2313__setup("max_loop=", max_loop_setup);
2314#endif