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_device {
49 int lo_number;
50 loff_t lo_offset;
51 loff_t lo_sizelimit;
52 int lo_flags;
53 char lo_file_name[LO_NAME_SIZE];
54
55 struct file *lo_backing_file;
56 unsigned int lo_min_dio_size;
57 struct block_device *lo_device;
58
59 gfp_t old_gfp_mask;
60
61 spinlock_t lo_lock;
62 int lo_state;
63 spinlock_t lo_work_lock;
64 struct workqueue_struct *workqueue;
65 struct work_struct rootcg_work;
66 struct list_head rootcg_cmd_list;
67 struct list_head idle_worker_list;
68 struct rb_root worker_tree;
69 struct timer_list timer;
70 bool sysfs_inited;
71
72 struct request_queue *lo_queue;
73 struct blk_mq_tag_set tag_set;
74 struct gendisk *lo_disk;
75 struct mutex lo_mutex;
76 bool idr_visible;
77};
78
79struct loop_cmd {
80 struct list_head list_entry;
81 bool use_aio; /* use AIO interface to handle I/O */
82 atomic_t ref; /* only for aio */
83 long ret;
84 struct kiocb iocb;
85 struct bio_vec *bvec;
86 struct cgroup_subsys_state *blkcg_css;
87 struct cgroup_subsys_state *memcg_css;
88};
89
90#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
91#define LOOP_DEFAULT_HW_Q_DEPTH 128
92
93static DEFINE_IDR(loop_index_idr);
94static DEFINE_MUTEX(loop_ctl_mutex);
95static DEFINE_MUTEX(loop_validate_mutex);
96
97/**
98 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
99 *
100 * @lo: struct loop_device
101 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
102 *
103 * Returns 0 on success, -EINTR otherwise.
104 *
105 * Since loop_validate_file() traverses on other "struct loop_device" if
106 * is_loop_device() is true, we need a global lock for serializing concurrent
107 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
108 */
109static int loop_global_lock_killable(struct loop_device *lo, bool global)
110{
111 int err;
112
113 if (global) {
114 err = mutex_lock_killable(&loop_validate_mutex);
115 if (err)
116 return err;
117 }
118 err = mutex_lock_killable(&lo->lo_mutex);
119 if (err && global)
120 mutex_unlock(lock: &loop_validate_mutex);
121 return err;
122}
123
124/**
125 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
126 *
127 * @lo: struct loop_device
128 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
129 */
130static void loop_global_unlock(struct loop_device *lo, bool global)
131{
132 mutex_unlock(lock: &lo->lo_mutex);
133 if (global)
134 mutex_unlock(lock: &loop_validate_mutex);
135}
136
137static int max_part;
138static int part_shift;
139
140static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
141{
142 loff_t loopsize;
143
144 /* Compute loopsize in bytes */
145 loopsize = i_size_read(inode: file->f_mapping->host);
146 if (offset > 0)
147 loopsize -= offset;
148 /* offset is beyond i_size, weird but possible */
149 if (loopsize < 0)
150 return 0;
151
152 if (sizelimit > 0 && sizelimit < loopsize)
153 loopsize = sizelimit;
154 /*
155 * Unfortunately, if we want to do I/O on the device,
156 * the number of 512-byte sectors has to fit into a sector_t.
157 */
158 return loopsize >> 9;
159}
160
161static loff_t get_loop_size(struct loop_device *lo, struct file *file)
162{
163 return get_size(offset: lo->lo_offset, sizelimit: lo->lo_sizelimit, file);
164}
165
166/*
167 * We support direct I/O only if lo_offset is aligned with the logical I/O size
168 * of backing device, and the logical block size of loop is bigger than that of
169 * the backing device.
170 */
171static bool lo_can_use_dio(struct loop_device *lo)
172{
173 if (!(lo->lo_backing_file->f_mode & FMODE_CAN_ODIRECT))
174 return false;
175 if (queue_logical_block_size(q: lo->lo_queue) < lo->lo_min_dio_size)
176 return false;
177 if (lo->lo_offset & (lo->lo_min_dio_size - 1))
178 return false;
179 return true;
180}
181
182/*
183 * Direct I/O can be enabled either by using an O_DIRECT file descriptor, or by
184 * passing in the LO_FLAGS_DIRECT_IO flag from userspace. It will be silently
185 * disabled when the device block size is too small or the offset is unaligned.
186 *
187 * loop_get_status will always report the effective LO_FLAGS_DIRECT_IO flag and
188 * not the originally passed in one.
189 */
190static inline void loop_update_dio(struct loop_device *lo)
191{
192 lockdep_assert_held(&lo->lo_mutex);
193 WARN_ON_ONCE(lo->lo_state == Lo_bound &&
194 lo->lo_queue->mq_freeze_depth == 0);
195
196 if ((lo->lo_flags & LO_FLAGS_DIRECT_IO) && !lo_can_use_dio(lo))
197 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
198}
199
200/**
201 * loop_set_size() - sets device size and notifies userspace
202 * @lo: struct loop_device to set the size for
203 * @size: new size of the loop device
204 *
205 * Callers must validate that the size passed into this function fits into
206 * a sector_t, eg using loop_validate_size()
207 */
208static void loop_set_size(struct loop_device *lo, loff_t size)
209{
210 if (!set_capacity_and_notify(disk: lo->lo_disk, size))
211 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
212}
213
214static void loop_clear_limits(struct loop_device *lo, int mode)
215{
216 struct queue_limits lim = queue_limits_start_update(q: lo->lo_queue);
217
218 if (mode & FALLOC_FL_ZERO_RANGE)
219 lim.max_write_zeroes_sectors = 0;
220
221 if (mode & FALLOC_FL_PUNCH_HOLE) {
222 lim.max_hw_discard_sectors = 0;
223 lim.discard_granularity = 0;
224 }
225
226 /*
227 * XXX: this updates the queue limits without freezing the queue, which
228 * is against the locking protocol and dangerous. But we can't just
229 * freeze the queue as we're inside the ->queue_rq method here. So this
230 * should move out into a workqueue unless we get the file operations to
231 * advertise if they support specific fallocate operations.
232 */
233 queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
234}
235
236static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
237 int mode)
238{
239 /*
240 * We use fallocate to manipulate the space mappings used by the image
241 * a.k.a. discard/zerorange.
242 */
243 struct file *file = lo->lo_backing_file;
244 int ret;
245
246 mode |= FALLOC_FL_KEEP_SIZE;
247
248 if (!bdev_max_discard_sectors(bdev: lo->lo_device))
249 return -EOPNOTSUPP;
250
251 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
252 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
253 return -EIO;
254
255 /*
256 * We initially configure the limits in a hope that fallocate is
257 * supported and clear them here if that turns out not to be true.
258 */
259 if (unlikely(ret == -EOPNOTSUPP))
260 loop_clear_limits(lo, mode);
261
262 return ret;
263}
264
265static int lo_req_flush(struct loop_device *lo, struct request *rq)
266{
267 int ret = vfs_fsync(file: lo->lo_backing_file, datasync: 0);
268 if (unlikely(ret && ret != -EINVAL))
269 ret = -EIO;
270
271 return ret;
272}
273
274static void lo_complete_rq(struct request *rq)
275{
276 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
277 blk_status_t ret = BLK_STS_OK;
278
279 if (cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
280 req_op(req: rq) != REQ_OP_READ) {
281 if (cmd->ret < 0)
282 ret = errno_to_blk_status(errno: cmd->ret);
283 goto end_io;
284 }
285
286 /*
287 * Short READ - if we got some data, advance our request and
288 * retry it. If we got no data, end the rest with EIO.
289 */
290 if (cmd->ret) {
291 blk_update_request(rq, BLK_STS_OK, nr_bytes: cmd->ret);
292 cmd->ret = 0;
293 blk_mq_requeue_request(rq, kick_requeue_list: true);
294 } else {
295 struct bio *bio = rq->bio;
296
297 while (bio) {
298 zero_fill_bio(bio);
299 bio = bio->bi_next;
300 }
301
302 ret = BLK_STS_IOERR;
303end_io:
304 blk_mq_end_request(rq, error: ret);
305 }
306}
307
308static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
309{
310 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
311 struct loop_device *lo = rq->q->queuedata;
312
313 if (!atomic_dec_and_test(v: &cmd->ref))
314 return;
315 kfree(objp: cmd->bvec);
316 cmd->bvec = NULL;
317 if (req_op(req: rq) == REQ_OP_WRITE)
318 file_end_write(file: lo->lo_backing_file);
319 if (likely(!blk_should_fake_timeout(rq->q)))
320 blk_mq_complete_request(rq);
321}
322
323static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
324{
325 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
326
327 cmd->ret = ret;
328 lo_rw_aio_do_completion(cmd);
329}
330
331static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
332 loff_t pos, int rw)
333{
334 struct iov_iter iter;
335 struct req_iterator rq_iter;
336 struct bio_vec *bvec;
337 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
338 struct bio *bio = rq->bio;
339 struct file *file = lo->lo_backing_file;
340 struct bio_vec tmp;
341 unsigned int offset;
342 int nr_bvec = 0;
343 int ret;
344
345 rq_for_each_bvec(tmp, rq, rq_iter)
346 nr_bvec++;
347
348 if (rq->bio != rq->biotail) {
349
350 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
351 GFP_NOIO);
352 if (!bvec)
353 return -EIO;
354 cmd->bvec = bvec;
355
356 /*
357 * The bios of the request may be started from the middle of
358 * the 'bvec' because of bio splitting, so we can't directly
359 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
360 * API will take care of all details for us.
361 */
362 rq_for_each_bvec(tmp, rq, rq_iter) {
363 *bvec = tmp;
364 bvec++;
365 }
366 bvec = cmd->bvec;
367 offset = 0;
368 } else {
369 /*
370 * Same here, this bio may be started from the middle of the
371 * 'bvec' because of bio splitting, so offset from the bvec
372 * must be passed to iov iterator
373 */
374 offset = bio->bi_iter.bi_bvec_done;
375 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
376 }
377 atomic_set(v: &cmd->ref, i: 2);
378
379 iov_iter_bvec(i: &iter, direction: rw, bvec, nr_segs: nr_bvec, count: blk_rq_bytes(rq));
380 iter.iov_offset = offset;
381
382 cmd->iocb.ki_pos = pos;
383 cmd->iocb.ki_filp = file;
384 cmd->iocb.ki_ioprio = req_get_ioprio(req: rq);
385 if (cmd->use_aio) {
386 cmd->iocb.ki_complete = lo_rw_aio_complete;
387 cmd->iocb.ki_flags = IOCB_DIRECT;
388 } else {
389 cmd->iocb.ki_complete = NULL;
390 cmd->iocb.ki_flags = 0;
391 }
392
393 if (rw == ITER_SOURCE) {
394 file_start_write(file: lo->lo_backing_file);
395 ret = file->f_op->write_iter(&cmd->iocb, &iter);
396 } else
397 ret = file->f_op->read_iter(&cmd->iocb, &iter);
398
399 lo_rw_aio_do_completion(cmd);
400
401 if (ret != -EIOCBQUEUED)
402 lo_rw_aio_complete(iocb: &cmd->iocb, ret);
403 return -EIOCBQUEUED;
404}
405
406static int do_req_filebacked(struct loop_device *lo, struct request *rq)
407{
408 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
409 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
410
411 switch (req_op(req: rq)) {
412 case REQ_OP_FLUSH:
413 return lo_req_flush(lo, rq);
414 case REQ_OP_WRITE_ZEROES:
415 /*
416 * If the caller doesn't want deallocation, call zeroout to
417 * write zeroes the range. Otherwise, punch them out.
418 */
419 return lo_fallocate(lo, rq, pos,
420 mode: (rq->cmd_flags & REQ_NOUNMAP) ?
421 FALLOC_FL_ZERO_RANGE :
422 FALLOC_FL_PUNCH_HOLE);
423 case REQ_OP_DISCARD:
424 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
425 case REQ_OP_WRITE:
426 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
427 case REQ_OP_READ:
428 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
429 default:
430 WARN_ON_ONCE(1);
431 return -EIO;
432 }
433}
434
435static void loop_reread_partitions(struct loop_device *lo)
436{
437 int rc;
438
439 mutex_lock(&lo->lo_disk->open_mutex);
440 rc = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
441 mutex_unlock(lock: &lo->lo_disk->open_mutex);
442 if (rc)
443 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
444 __func__, lo->lo_number, lo->lo_file_name, rc);
445}
446
447static unsigned int loop_query_min_dio_size(struct loop_device *lo)
448{
449 struct file *file = lo->lo_backing_file;
450 struct block_device *sb_bdev = file->f_mapping->host->i_sb->s_bdev;
451 struct kstat st;
452
453 /*
454 * Use the minimal dio alignment of the file system if provided.
455 */
456 if (!vfs_getattr(&file->f_path, &st, STATX_DIOALIGN, 0) &&
457 (st.result_mask & STATX_DIOALIGN))
458 return st.dio_offset_align;
459
460 /*
461 * In a perfect world this wouldn't be needed, but as of Linux 6.13 only
462 * a handful of file systems support the STATX_DIOALIGN flag.
463 */
464 if (sb_bdev)
465 return bdev_logical_block_size(bdev: sb_bdev);
466 return SECTOR_SIZE;
467}
468
469static inline int is_loop_device(struct file *file)
470{
471 struct inode *i = file->f_mapping->host;
472
473 return i && S_ISBLK(i->i_mode) && imajor(inode: i) == LOOP_MAJOR;
474}
475
476static int loop_validate_file(struct file *file, struct block_device *bdev)
477{
478 struct inode *inode = file->f_mapping->host;
479 struct file *f = file;
480
481 /* Avoid recursion */
482 while (is_loop_device(file: f)) {
483 struct loop_device *l;
484
485 lockdep_assert_held(&loop_validate_mutex);
486 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
487 return -EBADF;
488
489 l = I_BDEV(inode: f->f_mapping->host)->bd_disk->private_data;
490 if (l->lo_state != Lo_bound)
491 return -EINVAL;
492 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
493 rmb();
494 f = l->lo_backing_file;
495 }
496 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
497 return -EINVAL;
498 return 0;
499}
500
501static void loop_assign_backing_file(struct loop_device *lo, struct file *file)
502{
503 lo->lo_backing_file = file;
504 lo->old_gfp_mask = mapping_gfp_mask(mapping: file->f_mapping);
505 mapping_set_gfp_mask(m: file->f_mapping,
506 mask: lo->old_gfp_mask & ~(__GFP_IO | __GFP_FS));
507 if (lo->lo_backing_file->f_flags & O_DIRECT)
508 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
509 lo->lo_min_dio_size = loop_query_min_dio_size(lo);
510}
511
512static int loop_check_backing_file(struct file *file)
513{
514 if (!file->f_op->read_iter)
515 return -EINVAL;
516
517 if ((file->f_mode & FMODE_WRITE) && !file->f_op->write_iter)
518 return -EINVAL;
519
520 return 0;
521}
522
523/*
524 * loop_change_fd switched the backing store of a loopback device to
525 * a new file. This is useful for operating system installers to free up
526 * the original file and in High Availability environments to switch to
527 * an alternative location for the content in case of server meltdown.
528 * This can only work if the loop device is used read-only, and if the
529 * new backing store is the same size and type as the old backing store.
530 */
531static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
532 unsigned int arg)
533{
534 struct file *file = fget(fd: arg);
535 struct file *old_file;
536 unsigned int memflags;
537 int error;
538 bool partscan;
539 bool is_loop;
540
541 if (!file)
542 return -EBADF;
543
544 error = loop_check_backing_file(file);
545 if (error)
546 return error;
547
548 /* suppress uevents while reconfiguring the device */
549 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
550
551 is_loop = is_loop_device(file);
552 error = loop_global_lock_killable(lo, global: is_loop);
553 if (error)
554 goto out_putf;
555 error = -ENXIO;
556 if (lo->lo_state != Lo_bound)
557 goto out_err;
558
559 /* the loop device has to be read-only */
560 error = -EINVAL;
561 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
562 goto out_err;
563
564 error = loop_validate_file(file, bdev);
565 if (error)
566 goto out_err;
567
568 old_file = lo->lo_backing_file;
569
570 error = -EINVAL;
571
572 /* size of the new backing store needs to be the same */
573 if (get_loop_size(lo, file) != get_loop_size(lo, file: old_file))
574 goto out_err;
575
576 /*
577 * We might switch to direct I/O mode for the loop device, write back
578 * all dirty data the page cache now that so that the individual I/O
579 * operations don't have to do that.
580 */
581 vfs_fsync(file, datasync: 0);
582
583 /* and ... switch */
584 disk_force_media_change(disk: lo->lo_disk);
585 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
586 mapping_set_gfp_mask(m: old_file->f_mapping, mask: lo->old_gfp_mask);
587 loop_assign_backing_file(lo, file);
588 loop_update_dio(lo);
589 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
590 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
591 loop_global_unlock(lo, global: is_loop);
592
593 /*
594 * Flush loop_validate_file() before fput(), for l->lo_backing_file
595 * might be pointing at old_file which might be the last reference.
596 */
597 if (!is_loop) {
598 mutex_lock(&loop_validate_mutex);
599 mutex_unlock(lock: &loop_validate_mutex);
600 }
601 /*
602 * We must drop file reference outside of lo_mutex as dropping
603 * the file ref can take open_mutex which creates circular locking
604 * dependency.
605 */
606 fput(old_file);
607 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
608 if (partscan)
609 loop_reread_partitions(lo);
610
611 error = 0;
612done:
613 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
614 return error;
615
616out_err:
617 loop_global_unlock(lo, global: is_loop);
618out_putf:
619 fput(file);
620 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
621 goto done;
622}
623
624/* loop sysfs attributes */
625
626static ssize_t loop_attr_show(struct device *dev, char *page,
627 ssize_t (*callback)(struct loop_device *, char *))
628{
629 struct gendisk *disk = dev_to_disk(dev);
630 struct loop_device *lo = disk->private_data;
631
632 return callback(lo, page);
633}
634
635#define LOOP_ATTR_RO(_name) \
636static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
637static ssize_t loop_attr_do_show_##_name(struct device *d, \
638 struct device_attribute *attr, char *b) \
639{ \
640 return loop_attr_show(d, b, loop_attr_##_name##_show); \
641} \
642static struct device_attribute loop_attr_##_name = \
643 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
644
645static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
646{
647 ssize_t ret;
648 char *p = NULL;
649
650 spin_lock_irq(lock: &lo->lo_lock);
651 if (lo->lo_backing_file)
652 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
653 spin_unlock_irq(lock: &lo->lo_lock);
654
655 if (IS_ERR_OR_NULL(ptr: p))
656 ret = PTR_ERR(ptr: p);
657 else {
658 ret = strlen(p);
659 memmove(buf, p, ret);
660 buf[ret++] = '\n';
661 buf[ret] = 0;
662 }
663
664 return ret;
665}
666
667static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
668{
669 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_offset);
670}
671
672static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
673{
674 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_sizelimit);
675}
676
677static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
678{
679 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
680
681 return sysfs_emit(buf, fmt: "%s\n", autoclear ? "1" : "0");
682}
683
684static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
685{
686 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
687
688 return sysfs_emit(buf, fmt: "%s\n", partscan ? "1" : "0");
689}
690
691static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
692{
693 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
694
695 return sysfs_emit(buf, fmt: "%s\n", dio ? "1" : "0");
696}
697
698LOOP_ATTR_RO(backing_file);
699LOOP_ATTR_RO(offset);
700LOOP_ATTR_RO(sizelimit);
701LOOP_ATTR_RO(autoclear);
702LOOP_ATTR_RO(partscan);
703LOOP_ATTR_RO(dio);
704
705static struct attribute *loop_attrs[] = {
706 &loop_attr_backing_file.attr,
707 &loop_attr_offset.attr,
708 &loop_attr_sizelimit.attr,
709 &loop_attr_autoclear.attr,
710 &loop_attr_partscan.attr,
711 &loop_attr_dio.attr,
712 NULL,
713};
714
715static struct attribute_group loop_attribute_group = {
716 .name = "loop",
717 .attrs= loop_attrs,
718};
719
720static void loop_sysfs_init(struct loop_device *lo)
721{
722 lo->sysfs_inited = !sysfs_create_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
723 grp: &loop_attribute_group);
724}
725
726static void loop_sysfs_exit(struct loop_device *lo)
727{
728 if (lo->sysfs_inited)
729 sysfs_remove_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
730 grp: &loop_attribute_group);
731}
732
733static void loop_get_discard_config(struct loop_device *lo,
734 u32 *granularity, u32 *max_discard_sectors)
735{
736 struct file *file = lo->lo_backing_file;
737 struct inode *inode = file->f_mapping->host;
738 struct kstatfs sbuf;
739
740 /*
741 * If the backing device is a block device, mirror its zeroing
742 * capability. Set the discard sectors to the block device's zeroing
743 * capabilities because loop discards result in blkdev_issue_zeroout(),
744 * not blkdev_issue_discard(). This maintains consistent behavior with
745 * file-backed loop devices: discarded regions read back as zero.
746 */
747 if (S_ISBLK(inode->i_mode)) {
748 struct block_device *bdev = I_BDEV(inode);
749
750 *max_discard_sectors = bdev_write_zeroes_sectors(bdev);
751 *granularity = bdev_discard_granularity(bdev);
752
753 /*
754 * We use punch hole to reclaim the free space used by the
755 * image a.k.a. discard.
756 */
757 } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
758 *max_discard_sectors = UINT_MAX >> 9;
759 *granularity = sbuf.f_bsize;
760 }
761}
762
763struct loop_worker {
764 struct rb_node rb_node;
765 struct work_struct work;
766 struct list_head cmd_list;
767 struct list_head idle_list;
768 struct loop_device *lo;
769 struct cgroup_subsys_state *blkcg_css;
770 unsigned long last_ran_at;
771};
772
773static void loop_workfn(struct work_struct *work);
774
775#ifdef CONFIG_BLK_CGROUP
776static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
777{
778 return !css || css == blkcg_root_css;
779}
780#else
781static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
782{
783 return !css;
784}
785#endif
786
787static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
788{
789 struct rb_node **node, *parent = NULL;
790 struct loop_worker *cur_worker, *worker = NULL;
791 struct work_struct *work;
792 struct list_head *cmd_list;
793
794 spin_lock_irq(lock: &lo->lo_work_lock);
795
796 if (queue_on_root_worker(css: cmd->blkcg_css))
797 goto queue_work;
798
799 node = &lo->worker_tree.rb_node;
800
801 while (*node) {
802 parent = *node;
803 cur_worker = container_of(*node, struct loop_worker, rb_node);
804 if (cur_worker->blkcg_css == cmd->blkcg_css) {
805 worker = cur_worker;
806 break;
807 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
808 node = &(*node)->rb_left;
809 } else {
810 node = &(*node)->rb_right;
811 }
812 }
813 if (worker)
814 goto queue_work;
815
816 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
817 /*
818 * In the event we cannot allocate a worker, just queue on the
819 * rootcg worker and issue the I/O as the rootcg
820 */
821 if (!worker) {
822 cmd->blkcg_css = NULL;
823 if (cmd->memcg_css)
824 css_put(css: cmd->memcg_css);
825 cmd->memcg_css = NULL;
826 goto queue_work;
827 }
828
829 worker->blkcg_css = cmd->blkcg_css;
830 css_get(css: worker->blkcg_css);
831 INIT_WORK(&worker->work, loop_workfn);
832 INIT_LIST_HEAD(list: &worker->cmd_list);
833 INIT_LIST_HEAD(list: &worker->idle_list);
834 worker->lo = lo;
835 rb_link_node(node: &worker->rb_node, parent, rb_link: node);
836 rb_insert_color(&worker->rb_node, &lo->worker_tree);
837queue_work:
838 if (worker) {
839 /*
840 * We need to remove from the idle list here while
841 * holding the lock so that the idle timer doesn't
842 * free the worker
843 */
844 if (!list_empty(head: &worker->idle_list))
845 list_del_init(entry: &worker->idle_list);
846 work = &worker->work;
847 cmd_list = &worker->cmd_list;
848 } else {
849 work = &lo->rootcg_work;
850 cmd_list = &lo->rootcg_cmd_list;
851 }
852 list_add_tail(new: &cmd->list_entry, head: cmd_list);
853 queue_work(wq: lo->workqueue, work);
854 spin_unlock_irq(lock: &lo->lo_work_lock);
855}
856
857static void loop_set_timer(struct loop_device *lo)
858{
859 timer_reduce(timer: &lo->timer, expires: jiffies + LOOP_IDLE_WORKER_TIMEOUT);
860}
861
862static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
863{
864 struct loop_worker *pos, *worker;
865
866 spin_lock_irq(lock: &lo->lo_work_lock);
867 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
868 idle_list) {
869 if (!delete_all &&
870 time_is_after_jiffies(worker->last_ran_at +
871 LOOP_IDLE_WORKER_TIMEOUT))
872 break;
873 list_del(entry: &worker->idle_list);
874 rb_erase(&worker->rb_node, &lo->worker_tree);
875 css_put(css: worker->blkcg_css);
876 kfree(objp: worker);
877 }
878 if (!list_empty(head: &lo->idle_worker_list))
879 loop_set_timer(lo);
880 spin_unlock_irq(lock: &lo->lo_work_lock);
881}
882
883static void loop_free_idle_workers_timer(struct timer_list *timer)
884{
885 struct loop_device *lo = container_of(timer, struct loop_device, timer);
886
887 return loop_free_idle_workers(lo, delete_all: false);
888}
889
890/**
891 * loop_set_status_from_info - configure device from loop_info
892 * @lo: struct loop_device to configure
893 * @info: struct loop_info64 to configure the device with
894 *
895 * Configures the loop device parameters according to the passed
896 * in loop_info64 configuration.
897 */
898static int
899loop_set_status_from_info(struct loop_device *lo,
900 const struct loop_info64 *info)
901{
902 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
903 return -EINVAL;
904
905 switch (info->lo_encrypt_type) {
906 case LO_CRYPT_NONE:
907 break;
908 case LO_CRYPT_XOR:
909 pr_warn("support for the xor transformation has been removed.\n");
910 return -EINVAL;
911 case LO_CRYPT_CRYPTOAPI:
912 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
913 return -EINVAL;
914 default:
915 return -EINVAL;
916 }
917
918 /* Avoid assigning overflow values */
919 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
920 return -EOVERFLOW;
921
922 lo->lo_offset = info->lo_offset;
923 lo->lo_sizelimit = info->lo_sizelimit;
924
925 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
926 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
927 return 0;
928}
929
930static unsigned int loop_default_blocksize(struct loop_device *lo)
931{
932 /* In case of direct I/O, match underlying minimum I/O size */
933 if (lo->lo_flags & LO_FLAGS_DIRECT_IO)
934 return lo->lo_min_dio_size;
935 return SECTOR_SIZE;
936}
937
938static void loop_update_limits(struct loop_device *lo, struct queue_limits *lim,
939 unsigned int bsize)
940{
941 struct file *file = lo->lo_backing_file;
942 struct inode *inode = file->f_mapping->host;
943 struct block_device *backing_bdev = NULL;
944 u32 granularity = 0, max_discard_sectors = 0;
945
946 if (S_ISBLK(inode->i_mode))
947 backing_bdev = I_BDEV(inode);
948 else if (inode->i_sb->s_bdev)
949 backing_bdev = inode->i_sb->s_bdev;
950
951 if (!bsize)
952 bsize = loop_default_blocksize(lo);
953
954 loop_get_discard_config(lo, granularity: &granularity, max_discard_sectors: &max_discard_sectors);
955
956 lim->logical_block_size = bsize;
957 lim->physical_block_size = bsize;
958 lim->io_min = bsize;
959 lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL);
960 if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY))
961 lim->features |= BLK_FEAT_WRITE_CACHE;
962 if (backing_bdev && !bdev_nonrot(bdev: backing_bdev))
963 lim->features |= BLK_FEAT_ROTATIONAL;
964 lim->max_hw_discard_sectors = max_discard_sectors;
965 lim->max_write_zeroes_sectors = max_discard_sectors;
966 if (max_discard_sectors)
967 lim->discard_granularity = granularity;
968 else
969 lim->discard_granularity = 0;
970}
971
972static int loop_configure(struct loop_device *lo, blk_mode_t mode,
973 struct block_device *bdev,
974 const struct loop_config *config)
975{
976 struct file *file = fget(fd: config->fd);
977 struct queue_limits lim;
978 int error;
979 loff_t size;
980 bool partscan;
981 bool is_loop;
982
983 if (!file)
984 return -EBADF;
985
986 error = loop_check_backing_file(file);
987 if (error)
988 return error;
989
990 is_loop = is_loop_device(file);
991
992 /* This is safe, since we have a reference from open(). */
993 __module_get(THIS_MODULE);
994
995 /*
996 * If we don't hold exclusive handle for the device, upgrade to it
997 * here to avoid changing device under exclusive owner.
998 */
999 if (!(mode & BLK_OPEN_EXCL)) {
1000 error = bd_prepare_to_claim(bdev, holder: loop_configure, NULL);
1001 if (error)
1002 goto out_putf;
1003 }
1004
1005 error = loop_global_lock_killable(lo, global: is_loop);
1006 if (error)
1007 goto out_bdev;
1008
1009 error = -EBUSY;
1010 if (lo->lo_state != Lo_unbound)
1011 goto out_unlock;
1012
1013 error = loop_validate_file(file, bdev);
1014 if (error)
1015 goto out_unlock;
1016
1017 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1018 error = -EINVAL;
1019 goto out_unlock;
1020 }
1021
1022 error = loop_set_status_from_info(lo, info: &config->info);
1023 if (error)
1024 goto out_unlock;
1025 lo->lo_flags = config->info.lo_flags;
1026
1027 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1028 !file->f_op->write_iter)
1029 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1030
1031 if (!lo->workqueue) {
1032 lo->workqueue = alloc_workqueue(fmt: "loop%d",
1033 flags: WQ_UNBOUND | WQ_FREEZABLE,
1034 max_active: 0, lo->lo_number);
1035 if (!lo->workqueue) {
1036 error = -ENOMEM;
1037 goto out_unlock;
1038 }
1039 }
1040
1041 /* suppress uevents while reconfiguring the device */
1042 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
1043
1044 disk_force_media_change(disk: lo->lo_disk);
1045 set_disk_ro(disk: lo->lo_disk, read_only: (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1046
1047 lo->lo_device = bdev;
1048 loop_assign_backing_file(lo, file);
1049
1050 lim = queue_limits_start_update(q: lo->lo_queue);
1051 loop_update_limits(lo, lim: &lim, bsize: config->block_size);
1052 /* No need to freeze the queue as the device isn't bound yet. */
1053 error = queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1054 if (error)
1055 goto out_unlock;
1056
1057 /*
1058 * We might switch to direct I/O mode for the loop device, write back
1059 * all dirty data the page cache now that so that the individual I/O
1060 * operations don't have to do that.
1061 */
1062 vfs_fsync(file, datasync: 0);
1063
1064 loop_update_dio(lo);
1065 loop_sysfs_init(lo);
1066
1067 size = get_loop_size(lo, file);
1068 loop_set_size(lo, size);
1069
1070 /* Order wrt reading lo_state in loop_validate_file(). */
1071 wmb();
1072
1073 lo->lo_state = Lo_bound;
1074 if (part_shift)
1075 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1076 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1077 if (partscan)
1078 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1079
1080 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
1081 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
1082
1083 loop_global_unlock(lo, global: is_loop);
1084 if (partscan)
1085 loop_reread_partitions(lo);
1086
1087 if (!(mode & BLK_OPEN_EXCL))
1088 bd_abort_claiming(bdev, holder: loop_configure);
1089
1090 return 0;
1091
1092out_unlock:
1093 loop_global_unlock(lo, global: is_loop);
1094out_bdev:
1095 if (!(mode & BLK_OPEN_EXCL))
1096 bd_abort_claiming(bdev, holder: loop_configure);
1097out_putf:
1098 fput(file);
1099 /* This is safe: open() is still holding a reference. */
1100 module_put(THIS_MODULE);
1101 return error;
1102}
1103
1104static void __loop_clr_fd(struct loop_device *lo)
1105{
1106 struct queue_limits lim;
1107 struct file *filp;
1108 gfp_t gfp = lo->old_gfp_mask;
1109
1110 spin_lock_irq(lock: &lo->lo_lock);
1111 filp = lo->lo_backing_file;
1112 lo->lo_backing_file = NULL;
1113 spin_unlock_irq(lock: &lo->lo_lock);
1114
1115 lo->lo_device = NULL;
1116 lo->lo_offset = 0;
1117 lo->lo_sizelimit = 0;
1118 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1119
1120 /*
1121 * Reset the block size to the default.
1122 *
1123 * No queue freezing needed because this is called from the final
1124 * ->release call only, so there can't be any outstanding I/O.
1125 */
1126 lim = queue_limits_start_update(q: lo->lo_queue);
1127 lim.logical_block_size = SECTOR_SIZE;
1128 lim.physical_block_size = SECTOR_SIZE;
1129 lim.io_min = SECTOR_SIZE;
1130 queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1131
1132 invalidate_disk(disk: lo->lo_disk);
1133 loop_sysfs_exit(lo);
1134 /* let user-space know about this change */
1135 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
1136 mapping_set_gfp_mask(m: filp->f_mapping, mask: gfp);
1137 /* This is safe: open() is still holding a reference. */
1138 module_put(THIS_MODULE);
1139
1140 disk_force_media_change(disk: lo->lo_disk);
1141
1142 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1143 int err;
1144
1145 /*
1146 * open_mutex has been held already in release path, so don't
1147 * acquire it if this function is called in such case.
1148 *
1149 * If the reread partition isn't from release path, lo_refcnt
1150 * must be at least one and it can only become zero when the
1151 * current holder is released.
1152 */
1153 err = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
1154 if (err)
1155 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1156 __func__, lo->lo_number, err);
1157 /* Device is gone, no point in returning error */
1158 }
1159
1160 /*
1161 * lo->lo_state is set to Lo_unbound here after above partscan has
1162 * finished. There cannot be anybody else entering __loop_clr_fd() as
1163 * Lo_rundown state protects us from all the other places trying to
1164 * change the 'lo' device.
1165 */
1166 lo->lo_flags = 0;
1167 if (!part_shift)
1168 set_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1169 mutex_lock(&lo->lo_mutex);
1170 lo->lo_state = Lo_unbound;
1171 mutex_unlock(lock: &lo->lo_mutex);
1172
1173 /*
1174 * Need not hold lo_mutex to fput backing file. Calling fput holding
1175 * lo_mutex triggers a circular lock dependency possibility warning as
1176 * fput can take open_mutex which is usually taken before lo_mutex.
1177 */
1178 fput(filp);
1179}
1180
1181static int loop_clr_fd(struct loop_device *lo)
1182{
1183 int err;
1184
1185 /*
1186 * Since lo_ioctl() is called without locks held, it is possible that
1187 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1188 *
1189 * Therefore, use global lock when setting Lo_rundown state in order to
1190 * make sure that loop_validate_file() will fail if the "struct file"
1191 * which loop_configure()/loop_change_fd() found via fget() was this
1192 * loop device.
1193 */
1194 err = loop_global_lock_killable(lo, global: true);
1195 if (err)
1196 return err;
1197 if (lo->lo_state != Lo_bound) {
1198 loop_global_unlock(lo, global: true);
1199 return -ENXIO;
1200 }
1201 /*
1202 * Mark the device for removing the backing device on last close.
1203 * If we are the only opener, also switch the state to roundown here to
1204 * prevent new openers from coming in.
1205 */
1206
1207 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1208 if (disk_openers(disk: lo->lo_disk) == 1)
1209 lo->lo_state = Lo_rundown;
1210 loop_global_unlock(lo, global: true);
1211
1212 return 0;
1213}
1214
1215static int
1216loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1217{
1218 int err;
1219 bool partscan = false;
1220 bool size_changed = false;
1221 unsigned int memflags;
1222
1223 err = mutex_lock_killable(&lo->lo_mutex);
1224 if (err)
1225 return err;
1226 if (lo->lo_state != Lo_bound) {
1227 err = -ENXIO;
1228 goto out_unlock;
1229 }
1230
1231 if (lo->lo_offset != info->lo_offset ||
1232 lo->lo_sizelimit != info->lo_sizelimit) {
1233 size_changed = true;
1234 sync_blockdev(bdev: lo->lo_device);
1235 invalidate_bdev(bdev: lo->lo_device);
1236 }
1237
1238 /* I/O needs to be drained before changing lo_offset or lo_sizelimit */
1239 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1240
1241 err = loop_set_status_from_info(lo, info);
1242 if (err)
1243 goto out_unfreeze;
1244
1245 partscan = !(lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1246 (info->lo_flags & LO_FLAGS_PARTSCAN);
1247
1248 lo->lo_flags &= ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1249 lo->lo_flags |= (info->lo_flags & LOOP_SET_STATUS_SETTABLE_FLAGS);
1250
1251 if (size_changed) {
1252 loff_t new_size = get_size(offset: lo->lo_offset, sizelimit: lo->lo_sizelimit,
1253 file: lo->lo_backing_file);
1254 loop_set_size(lo, size: new_size);
1255 }
1256
1257 /* update the direct I/O flag if lo_offset changed */
1258 loop_update_dio(lo);
1259
1260out_unfreeze:
1261 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1262 if (partscan)
1263 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1264out_unlock:
1265 mutex_unlock(lock: &lo->lo_mutex);
1266 if (partscan)
1267 loop_reread_partitions(lo);
1268
1269 return err;
1270}
1271
1272static int
1273loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1274{
1275 struct path path;
1276 struct kstat stat;
1277 int ret;
1278
1279 ret = mutex_lock_killable(&lo->lo_mutex);
1280 if (ret)
1281 return ret;
1282 if (lo->lo_state != Lo_bound) {
1283 mutex_unlock(lock: &lo->lo_mutex);
1284 return -ENXIO;
1285 }
1286
1287 memset(info, 0, sizeof(*info));
1288 info->lo_number = lo->lo_number;
1289 info->lo_offset = lo->lo_offset;
1290 info->lo_sizelimit = lo->lo_sizelimit;
1291 info->lo_flags = lo->lo_flags;
1292 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1293
1294 /* Drop lo_mutex while we call into the filesystem. */
1295 path = lo->lo_backing_file->f_path;
1296 path_get(&path);
1297 mutex_unlock(lock: &lo->lo_mutex);
1298 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1299 if (!ret) {
1300 info->lo_device = huge_encode_dev(dev: stat.dev);
1301 info->lo_inode = stat.ino;
1302 info->lo_rdevice = huge_encode_dev(dev: stat.rdev);
1303 }
1304 path_put(&path);
1305 return ret;
1306}
1307
1308static void
1309loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1310{
1311 memset(info64, 0, sizeof(*info64));
1312 info64->lo_number = info->lo_number;
1313 info64->lo_device = info->lo_device;
1314 info64->lo_inode = info->lo_inode;
1315 info64->lo_rdevice = info->lo_rdevice;
1316 info64->lo_offset = info->lo_offset;
1317 info64->lo_sizelimit = 0;
1318 info64->lo_flags = info->lo_flags;
1319 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1320}
1321
1322static int
1323loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1324{
1325 memset(info, 0, sizeof(*info));
1326 info->lo_number = info64->lo_number;
1327 info->lo_device = info64->lo_device;
1328 info->lo_inode = info64->lo_inode;
1329 info->lo_rdevice = info64->lo_rdevice;
1330 info->lo_offset = info64->lo_offset;
1331 info->lo_flags = info64->lo_flags;
1332 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1333
1334 /* error in case values were truncated */
1335 if (info->lo_device != info64->lo_device ||
1336 info->lo_rdevice != info64->lo_rdevice ||
1337 info->lo_inode != info64->lo_inode ||
1338 info->lo_offset != info64->lo_offset)
1339 return -EOVERFLOW;
1340
1341 return 0;
1342}
1343
1344static int
1345loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1346{
1347 struct loop_info info;
1348 struct loop_info64 info64;
1349
1350 if (copy_from_user(to: &info, from: arg, n: sizeof (struct loop_info)))
1351 return -EFAULT;
1352 loop_info64_from_old(info: &info, info64: &info64);
1353 return loop_set_status(lo, info: &info64);
1354}
1355
1356static int
1357loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1358{
1359 struct loop_info64 info64;
1360
1361 if (copy_from_user(to: &info64, from: arg, n: sizeof (struct loop_info64)))
1362 return -EFAULT;
1363 return loop_set_status(lo, info: &info64);
1364}
1365
1366static int
1367loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1368 struct loop_info info;
1369 struct loop_info64 info64;
1370 int err;
1371
1372 if (!arg)
1373 return -EINVAL;
1374 err = loop_get_status(lo, info: &info64);
1375 if (!err)
1376 err = loop_info64_to_old(info64: &info64, info: &info);
1377 if (!err && copy_to_user(to: arg, from: &info, n: sizeof(info)))
1378 err = -EFAULT;
1379
1380 return err;
1381}
1382
1383static int
1384loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1385 struct loop_info64 info64;
1386 int err;
1387
1388 if (!arg)
1389 return -EINVAL;
1390 err = loop_get_status(lo, info: &info64);
1391 if (!err && copy_to_user(to: arg, from: &info64, n: sizeof(info64)))
1392 err = -EFAULT;
1393
1394 return err;
1395}
1396
1397static int loop_set_capacity(struct loop_device *lo)
1398{
1399 loff_t size;
1400
1401 if (unlikely(lo->lo_state != Lo_bound))
1402 return -ENXIO;
1403
1404 size = get_loop_size(lo, file: lo->lo_backing_file);
1405 loop_set_size(lo, size);
1406
1407 return 0;
1408}
1409
1410static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1411{
1412 bool use_dio = !!arg;
1413 unsigned int memflags;
1414
1415 if (lo->lo_state != Lo_bound)
1416 return -ENXIO;
1417 if (use_dio == !!(lo->lo_flags & LO_FLAGS_DIRECT_IO))
1418 return 0;
1419
1420 if (use_dio) {
1421 if (!lo_can_use_dio(lo))
1422 return -EINVAL;
1423 /* flush dirty pages before starting to use direct I/O */
1424 vfs_fsync(file: lo->lo_backing_file, datasync: 0);
1425 }
1426
1427 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1428 if (use_dio)
1429 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
1430 else
1431 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
1432 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1433 return 0;
1434}
1435
1436static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1437{
1438 struct queue_limits lim;
1439 unsigned int memflags;
1440 int err = 0;
1441
1442 if (lo->lo_state != Lo_bound)
1443 return -ENXIO;
1444
1445 if (lo->lo_queue->limits.logical_block_size == arg)
1446 return 0;
1447
1448 sync_blockdev(bdev: lo->lo_device);
1449 invalidate_bdev(bdev: lo->lo_device);
1450
1451 lim = queue_limits_start_update(q: lo->lo_queue);
1452 loop_update_limits(lo, lim: &lim, bsize: arg);
1453
1454 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1455 err = queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1456 loop_update_dio(lo);
1457 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1458
1459 return err;
1460}
1461
1462static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1463 unsigned long arg)
1464{
1465 int err;
1466
1467 err = mutex_lock_killable(&lo->lo_mutex);
1468 if (err)
1469 return err;
1470 switch (cmd) {
1471 case LOOP_SET_CAPACITY:
1472 err = loop_set_capacity(lo);
1473 break;
1474 case LOOP_SET_DIRECT_IO:
1475 err = loop_set_dio(lo, arg);
1476 break;
1477 case LOOP_SET_BLOCK_SIZE:
1478 err = loop_set_block_size(lo, arg);
1479 break;
1480 default:
1481 err = -EINVAL;
1482 }
1483 mutex_unlock(lock: &lo->lo_mutex);
1484 return err;
1485}
1486
1487static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1488 unsigned int cmd, unsigned long arg)
1489{
1490 struct loop_device *lo = bdev->bd_disk->private_data;
1491 void __user *argp = (void __user *) arg;
1492 int err;
1493
1494 switch (cmd) {
1495 case LOOP_SET_FD: {
1496 /*
1497 * Legacy case - pass in a zeroed out struct loop_config with
1498 * only the file descriptor set , which corresponds with the
1499 * default parameters we'd have used otherwise.
1500 */
1501 struct loop_config config;
1502
1503 memset(&config, 0, sizeof(config));
1504 config.fd = arg;
1505
1506 return loop_configure(lo, mode, bdev, config: &config);
1507 }
1508 case LOOP_CONFIGURE: {
1509 struct loop_config config;
1510
1511 if (copy_from_user(to: &config, from: argp, n: sizeof(config)))
1512 return -EFAULT;
1513
1514 return loop_configure(lo, mode, bdev, config: &config);
1515 }
1516 case LOOP_CHANGE_FD:
1517 return loop_change_fd(lo, bdev, arg);
1518 case LOOP_CLR_FD:
1519 return loop_clr_fd(lo);
1520 case LOOP_SET_STATUS:
1521 err = -EPERM;
1522 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1523 err = loop_set_status_old(lo, arg: argp);
1524 break;
1525 case LOOP_GET_STATUS:
1526 return loop_get_status_old(lo, arg: argp);
1527 case LOOP_SET_STATUS64:
1528 err = -EPERM;
1529 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1530 err = loop_set_status64(lo, arg: argp);
1531 break;
1532 case LOOP_GET_STATUS64:
1533 return loop_get_status64(lo, arg: argp);
1534 case LOOP_SET_CAPACITY:
1535 case LOOP_SET_DIRECT_IO:
1536 case LOOP_SET_BLOCK_SIZE:
1537 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1538 return -EPERM;
1539 fallthrough;
1540 default:
1541 err = lo_simple_ioctl(lo, cmd, arg);
1542 break;
1543 }
1544
1545 return err;
1546}
1547
1548#ifdef CONFIG_COMPAT
1549struct compat_loop_info {
1550 compat_int_t lo_number; /* ioctl r/o */
1551 compat_dev_t lo_device; /* ioctl r/o */
1552 compat_ulong_t lo_inode; /* ioctl r/o */
1553 compat_dev_t lo_rdevice; /* ioctl r/o */
1554 compat_int_t lo_offset;
1555 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1556 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1557 compat_int_t lo_flags; /* ioctl r/o */
1558 char lo_name[LO_NAME_SIZE];
1559 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1560 compat_ulong_t lo_init[2];
1561 char reserved[4];
1562};
1563
1564/*
1565 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1566 * - noinlined to reduce stack space usage in main part of driver
1567 */
1568static noinline int
1569loop_info64_from_compat(const struct compat_loop_info __user *arg,
1570 struct loop_info64 *info64)
1571{
1572 struct compat_loop_info info;
1573
1574 if (copy_from_user(to: &info, from: arg, n: sizeof(info)))
1575 return -EFAULT;
1576
1577 memset(info64, 0, sizeof(*info64));
1578 info64->lo_number = info.lo_number;
1579 info64->lo_device = info.lo_device;
1580 info64->lo_inode = info.lo_inode;
1581 info64->lo_rdevice = info.lo_rdevice;
1582 info64->lo_offset = info.lo_offset;
1583 info64->lo_sizelimit = 0;
1584 info64->lo_flags = info.lo_flags;
1585 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1586 return 0;
1587}
1588
1589/*
1590 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1591 * - noinlined to reduce stack space usage in main part of driver
1592 */
1593static noinline int
1594loop_info64_to_compat(const struct loop_info64 *info64,
1595 struct compat_loop_info __user *arg)
1596{
1597 struct compat_loop_info info;
1598
1599 memset(&info, 0, sizeof(info));
1600 info.lo_number = info64->lo_number;
1601 info.lo_device = info64->lo_device;
1602 info.lo_inode = info64->lo_inode;
1603 info.lo_rdevice = info64->lo_rdevice;
1604 info.lo_offset = info64->lo_offset;
1605 info.lo_flags = info64->lo_flags;
1606 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1607
1608 /* error in case values were truncated */
1609 if (info.lo_device != info64->lo_device ||
1610 info.lo_rdevice != info64->lo_rdevice ||
1611 info.lo_inode != info64->lo_inode ||
1612 info.lo_offset != info64->lo_offset)
1613 return -EOVERFLOW;
1614
1615 if (copy_to_user(to: arg, from: &info, n: sizeof(info)))
1616 return -EFAULT;
1617 return 0;
1618}
1619
1620static int
1621loop_set_status_compat(struct loop_device *lo,
1622 const struct compat_loop_info __user *arg)
1623{
1624 struct loop_info64 info64;
1625 int ret;
1626
1627 ret = loop_info64_from_compat(arg, info64: &info64);
1628 if (ret < 0)
1629 return ret;
1630 return loop_set_status(lo, info: &info64);
1631}
1632
1633static int
1634loop_get_status_compat(struct loop_device *lo,
1635 struct compat_loop_info __user *arg)
1636{
1637 struct loop_info64 info64;
1638 int err;
1639
1640 if (!arg)
1641 return -EINVAL;
1642 err = loop_get_status(lo, info: &info64);
1643 if (!err)
1644 err = loop_info64_to_compat(info64: &info64, arg);
1645 return err;
1646}
1647
1648static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1649 unsigned int cmd, unsigned long arg)
1650{
1651 struct loop_device *lo = bdev->bd_disk->private_data;
1652 int err;
1653
1654 switch(cmd) {
1655 case LOOP_SET_STATUS:
1656 err = loop_set_status_compat(lo,
1657 arg: (const struct compat_loop_info __user *)arg);
1658 break;
1659 case LOOP_GET_STATUS:
1660 err = loop_get_status_compat(lo,
1661 arg: (struct compat_loop_info __user *)arg);
1662 break;
1663 case LOOP_SET_CAPACITY:
1664 case LOOP_CLR_FD:
1665 case LOOP_GET_STATUS64:
1666 case LOOP_SET_STATUS64:
1667 case LOOP_CONFIGURE:
1668 arg = (unsigned long) compat_ptr(uptr: arg);
1669 fallthrough;
1670 case LOOP_SET_FD:
1671 case LOOP_CHANGE_FD:
1672 case LOOP_SET_BLOCK_SIZE:
1673 case LOOP_SET_DIRECT_IO:
1674 err = lo_ioctl(bdev, mode, cmd, arg);
1675 break;
1676 default:
1677 err = -ENOIOCTLCMD;
1678 break;
1679 }
1680 return err;
1681}
1682#endif
1683
1684static int lo_open(struct gendisk *disk, blk_mode_t mode)
1685{
1686 struct loop_device *lo = disk->private_data;
1687 int err;
1688
1689 err = mutex_lock_killable(&lo->lo_mutex);
1690 if (err)
1691 return err;
1692
1693 if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown)
1694 err = -ENXIO;
1695 mutex_unlock(lock: &lo->lo_mutex);
1696 return err;
1697}
1698
1699static void lo_release(struct gendisk *disk)
1700{
1701 struct loop_device *lo = disk->private_data;
1702 bool need_clear = false;
1703
1704 if (disk_openers(disk) > 0)
1705 return;
1706 /*
1707 * Clear the backing device information if this is the last close of
1708 * a device that's been marked for auto clear, or on which LOOP_CLR_FD
1709 * has been called.
1710 */
1711
1712 mutex_lock(&lo->lo_mutex);
1713 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR))
1714 lo->lo_state = Lo_rundown;
1715
1716 need_clear = (lo->lo_state == Lo_rundown);
1717 mutex_unlock(lock: &lo->lo_mutex);
1718
1719 if (need_clear)
1720 __loop_clr_fd(lo);
1721}
1722
1723static void lo_free_disk(struct gendisk *disk)
1724{
1725 struct loop_device *lo = disk->private_data;
1726
1727 if (lo->workqueue)
1728 destroy_workqueue(wq: lo->workqueue);
1729 loop_free_idle_workers(lo, delete_all: true);
1730 timer_shutdown_sync(timer: &lo->timer);
1731 mutex_destroy(lock: &lo->lo_mutex);
1732 kfree(objp: lo);
1733}
1734
1735static const struct block_device_operations lo_fops = {
1736 .owner = THIS_MODULE,
1737 .open = lo_open,
1738 .release = lo_release,
1739 .ioctl = lo_ioctl,
1740#ifdef CONFIG_COMPAT
1741 .compat_ioctl = lo_compat_ioctl,
1742#endif
1743 .free_disk = lo_free_disk,
1744};
1745
1746/*
1747 * And now the modules code and kernel interface.
1748 */
1749
1750/*
1751 * If max_loop is specified, create that many devices upfront.
1752 * This also becomes a hard limit. If max_loop is not specified,
1753 * the default isn't a hard limit (as before commit 85c50197716c
1754 * changed the default value from 0 for max_loop=0 reasons), just
1755 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1756 * init time. Loop devices can be requested on-demand with the
1757 * /dev/loop-control interface, or be instantiated by accessing
1758 * a 'dead' device node.
1759 */
1760static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1761
1762#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1763static bool max_loop_specified;
1764
1765static int max_loop_param_set_int(const char *val,
1766 const struct kernel_param *kp)
1767{
1768 int ret;
1769
1770 ret = param_set_int(val, kp);
1771 if (ret < 0)
1772 return ret;
1773
1774 max_loop_specified = true;
1775 return 0;
1776}
1777
1778static const struct kernel_param_ops max_loop_param_ops = {
1779 .set = max_loop_param_set_int,
1780 .get = param_get_int,
1781};
1782
1783module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1784MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1785#else
1786module_param(max_loop, int, 0444);
1787MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1788#endif
1789
1790module_param(max_part, int, 0444);
1791MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1792
1793static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1794
1795static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1796{
1797 int qd, ret;
1798
1799 ret = kstrtoint(s, base: 0, res: &qd);
1800 if (ret < 0)
1801 return ret;
1802 if (qd < 1)
1803 return -EINVAL;
1804 hw_queue_depth = qd;
1805 return 0;
1806}
1807
1808static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1809 .set = loop_set_hw_queue_depth,
1810 .get = param_get_int,
1811};
1812
1813device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1814MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1815
1816MODULE_DESCRIPTION("Loopback device support");
1817MODULE_LICENSE("GPL");
1818MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1819
1820static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1821 const struct blk_mq_queue_data *bd)
1822{
1823 struct request *rq = bd->rq;
1824 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1825 struct loop_device *lo = rq->q->queuedata;
1826
1827 blk_mq_start_request(rq);
1828
1829 if (lo->lo_state != Lo_bound)
1830 return BLK_STS_IOERR;
1831
1832 switch (req_op(req: rq)) {
1833 case REQ_OP_FLUSH:
1834 case REQ_OP_DISCARD:
1835 case REQ_OP_WRITE_ZEROES:
1836 cmd->use_aio = false;
1837 break;
1838 default:
1839 cmd->use_aio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1840 break;
1841 }
1842
1843 /* always use the first bio's css */
1844 cmd->blkcg_css = NULL;
1845 cmd->memcg_css = NULL;
1846#ifdef CONFIG_BLK_CGROUP
1847 if (rq->bio) {
1848 cmd->blkcg_css = bio_blkcg_css(bio: rq->bio);
1849#ifdef CONFIG_MEMCG
1850 if (cmd->blkcg_css) {
1851 cmd->memcg_css =
1852 cgroup_get_e_css(cgroup: cmd->blkcg_css->cgroup,
1853 ss: &memory_cgrp_subsys);
1854 }
1855#endif
1856 }
1857#endif
1858 loop_queue_work(lo, cmd);
1859
1860 return BLK_STS_OK;
1861}
1862
1863static void loop_handle_cmd(struct loop_cmd *cmd)
1864{
1865 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1866 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1867 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
1868 const bool write = op_is_write(op: req_op(req: rq));
1869 struct loop_device *lo = rq->q->queuedata;
1870 int ret = 0;
1871 struct mem_cgroup *old_memcg = NULL;
1872
1873 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1874 ret = -EIO;
1875 goto failed;
1876 }
1877
1878 if (cmd_blkcg_css)
1879 kthread_associate_blkcg(css: cmd_blkcg_css);
1880 if (cmd_memcg_css)
1881 old_memcg = set_active_memcg(
1882 mem_cgroup_from_css(css: cmd_memcg_css));
1883
1884 /*
1885 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1886 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1887 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1888 * not yet been completed.
1889 */
1890 ret = do_req_filebacked(lo, rq);
1891
1892 if (cmd_blkcg_css)
1893 kthread_associate_blkcg(NULL);
1894
1895 if (cmd_memcg_css) {
1896 set_active_memcg(old_memcg);
1897 css_put(css: cmd_memcg_css);
1898 }
1899 failed:
1900 /* complete non-aio request */
1901 if (ret != -EIOCBQUEUED) {
1902 if (ret == -EOPNOTSUPP)
1903 cmd->ret = ret;
1904 else
1905 cmd->ret = ret ? -EIO : 0;
1906 if (likely(!blk_should_fake_timeout(rq->q)))
1907 blk_mq_complete_request(rq);
1908 }
1909}
1910
1911static void loop_process_work(struct loop_worker *worker,
1912 struct list_head *cmd_list, struct loop_device *lo)
1913{
1914 int orig_flags = current->flags;
1915 struct loop_cmd *cmd;
1916
1917 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1918 spin_lock_irq(lock: &lo->lo_work_lock);
1919 while (!list_empty(head: cmd_list)) {
1920 cmd = container_of(
1921 cmd_list->next, struct loop_cmd, list_entry);
1922 list_del(entry: cmd_list->next);
1923 spin_unlock_irq(lock: &lo->lo_work_lock);
1924
1925 loop_handle_cmd(cmd);
1926 cond_resched();
1927
1928 spin_lock_irq(lock: &lo->lo_work_lock);
1929 }
1930
1931 /*
1932 * We only add to the idle list if there are no pending cmds
1933 * *and* the worker will not run again which ensures that it
1934 * is safe to free any worker on the idle list
1935 */
1936 if (worker && !work_pending(&worker->work)) {
1937 worker->last_ran_at = jiffies;
1938 list_add_tail(new: &worker->idle_list, head: &lo->idle_worker_list);
1939 loop_set_timer(lo);
1940 }
1941 spin_unlock_irq(lock: &lo->lo_work_lock);
1942 current->flags = orig_flags;
1943}
1944
1945static void loop_workfn(struct work_struct *work)
1946{
1947 struct loop_worker *worker =
1948 container_of(work, struct loop_worker, work);
1949 loop_process_work(worker, cmd_list: &worker->cmd_list, lo: worker->lo);
1950}
1951
1952static void loop_rootcg_workfn(struct work_struct *work)
1953{
1954 struct loop_device *lo =
1955 container_of(work, struct loop_device, rootcg_work);
1956 loop_process_work(NULL, cmd_list: &lo->rootcg_cmd_list, lo);
1957}
1958
1959static const struct blk_mq_ops loop_mq_ops = {
1960 .queue_rq = loop_queue_rq,
1961 .complete = lo_complete_rq,
1962};
1963
1964static int loop_add(int i)
1965{
1966 struct queue_limits lim = {
1967 /*
1968 * Random number picked from the historic block max_sectors cap.
1969 */
1970 .max_hw_sectors = 2560u,
1971 };
1972 struct loop_device *lo;
1973 struct gendisk *disk;
1974 int err;
1975
1976 err = -ENOMEM;
1977 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1978 if (!lo)
1979 goto out;
1980 lo->worker_tree = RB_ROOT;
1981 INIT_LIST_HEAD(list: &lo->idle_worker_list);
1982 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
1983 lo->lo_state = Lo_unbound;
1984
1985 err = mutex_lock_killable(&loop_ctl_mutex);
1986 if (err)
1987 goto out_free_dev;
1988
1989 /* allocate id, if @id >= 0, we're requesting that specific id */
1990 if (i >= 0) {
1991 err = idr_alloc(&loop_index_idr, ptr: lo, start: i, end: i + 1, GFP_KERNEL);
1992 if (err == -ENOSPC)
1993 err = -EEXIST;
1994 } else {
1995 err = idr_alloc(&loop_index_idr, ptr: lo, start: 0, end: 0, GFP_KERNEL);
1996 }
1997 mutex_unlock(lock: &loop_ctl_mutex);
1998 if (err < 0)
1999 goto out_free_dev;
2000 i = err;
2001
2002 lo->tag_set.ops = &loop_mq_ops;
2003 lo->tag_set.nr_hw_queues = 1;
2004 lo->tag_set.queue_depth = hw_queue_depth;
2005 lo->tag_set.numa_node = NUMA_NO_NODE;
2006 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2007 lo->tag_set.flags = BLK_MQ_F_STACKING | BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2008 lo->tag_set.driver_data = lo;
2009
2010 err = blk_mq_alloc_tag_set(set: &lo->tag_set);
2011 if (err)
2012 goto out_free_idr;
2013
2014 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2015 if (IS_ERR(ptr: disk)) {
2016 err = PTR_ERR(ptr: disk);
2017 goto out_cleanup_tags;
2018 }
2019 lo->lo_queue = lo->lo_disk->queue;
2020
2021 /*
2022 * Disable partition scanning by default. The in-kernel partition
2023 * scanning can be requested individually per-device during its
2024 * setup. Userspace can always add and remove partitions from all
2025 * devices. The needed partition minors are allocated from the
2026 * extended minor space, the main loop device numbers will continue
2027 * to match the loop minors, regardless of the number of partitions
2028 * used.
2029 *
2030 * If max_part is given, partition scanning is globally enabled for
2031 * all loop devices. The minors for the main loop devices will be
2032 * multiples of max_part.
2033 *
2034 * Note: Global-for-all-devices, set-only-at-init, read-only module
2035 * parameteters like 'max_loop' and 'max_part' make things needlessly
2036 * complicated, are too static, inflexible and may surprise
2037 * userspace tools. Parameters like this in general should be avoided.
2038 */
2039 if (!part_shift)
2040 set_bit(GD_SUPPRESS_PART_SCAN, addr: &disk->state);
2041 mutex_init(&lo->lo_mutex);
2042 lo->lo_number = i;
2043 spin_lock_init(&lo->lo_lock);
2044 spin_lock_init(&lo->lo_work_lock);
2045 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2046 INIT_LIST_HEAD(list: &lo->rootcg_cmd_list);
2047 disk->major = LOOP_MAJOR;
2048 disk->first_minor = i << part_shift;
2049 disk->minors = 1 << part_shift;
2050 disk->fops = &lo_fops;
2051 disk->private_data = lo;
2052 disk->queue = lo->lo_queue;
2053 disk->events = DISK_EVENT_MEDIA_CHANGE;
2054 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2055 sprintf(buf: disk->disk_name, fmt: "loop%d", i);
2056 /* Make this loop device reachable from pathname. */
2057 err = add_disk(disk);
2058 if (err)
2059 goto out_cleanup_disk;
2060
2061 /* Show this loop device. */
2062 mutex_lock(&loop_ctl_mutex);
2063 lo->idr_visible = true;
2064 mutex_unlock(lock: &loop_ctl_mutex);
2065
2066 return i;
2067
2068out_cleanup_disk:
2069 put_disk(disk);
2070out_cleanup_tags:
2071 blk_mq_free_tag_set(set: &lo->tag_set);
2072out_free_idr:
2073 mutex_lock(&loop_ctl_mutex);
2074 idr_remove(&loop_index_idr, id: i);
2075 mutex_unlock(lock: &loop_ctl_mutex);
2076out_free_dev:
2077 kfree(objp: lo);
2078out:
2079 return err;
2080}
2081
2082static void loop_remove(struct loop_device *lo)
2083{
2084 /* Make this loop device unreachable from pathname. */
2085 del_gendisk(gp: lo->lo_disk);
2086 blk_mq_free_tag_set(set: &lo->tag_set);
2087
2088 mutex_lock(&loop_ctl_mutex);
2089 idr_remove(&loop_index_idr, id: lo->lo_number);
2090 mutex_unlock(lock: &loop_ctl_mutex);
2091
2092 put_disk(disk: lo->lo_disk);
2093}
2094
2095#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2096static void loop_probe(dev_t dev)
2097{
2098 int idx = MINOR(dev) >> part_shift;
2099
2100 if (max_loop_specified && max_loop && idx >= max_loop)
2101 return;
2102 loop_add(i: idx);
2103}
2104#else
2105#define loop_probe NULL
2106#endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2107
2108static int loop_control_remove(int idx)
2109{
2110 struct loop_device *lo;
2111 int ret;
2112
2113 if (idx < 0) {
2114 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2115 return -EINVAL;
2116 }
2117
2118 /* Hide this loop device for serialization. */
2119 ret = mutex_lock_killable(&loop_ctl_mutex);
2120 if (ret)
2121 return ret;
2122 lo = idr_find(&loop_index_idr, id: idx);
2123 if (!lo || !lo->idr_visible)
2124 ret = -ENODEV;
2125 else
2126 lo->idr_visible = false;
2127 mutex_unlock(lock: &loop_ctl_mutex);
2128 if (ret)
2129 return ret;
2130
2131 /* Check whether this loop device can be removed. */
2132 ret = mutex_lock_killable(&lo->lo_mutex);
2133 if (ret)
2134 goto mark_visible;
2135 if (lo->lo_state != Lo_unbound || disk_openers(disk: lo->lo_disk) > 0) {
2136 mutex_unlock(lock: &lo->lo_mutex);
2137 ret = -EBUSY;
2138 goto mark_visible;
2139 }
2140 /* Mark this loop device as no more bound, but not quite unbound yet */
2141 lo->lo_state = Lo_deleting;
2142 mutex_unlock(lock: &lo->lo_mutex);
2143
2144 loop_remove(lo);
2145 return 0;
2146
2147mark_visible:
2148 /* Show this loop device again. */
2149 mutex_lock(&loop_ctl_mutex);
2150 lo->idr_visible = true;
2151 mutex_unlock(lock: &loop_ctl_mutex);
2152 return ret;
2153}
2154
2155static int loop_control_get_free(int idx)
2156{
2157 struct loop_device *lo;
2158 int id, ret;
2159
2160 ret = mutex_lock_killable(&loop_ctl_mutex);
2161 if (ret)
2162 return ret;
2163 idr_for_each_entry(&loop_index_idr, lo, id) {
2164 /* Hitting a race results in creating a new loop device which is harmless. */
2165 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2166 goto found;
2167 }
2168 mutex_unlock(lock: &loop_ctl_mutex);
2169 return loop_add(i: -1);
2170found:
2171 mutex_unlock(lock: &loop_ctl_mutex);
2172 return id;
2173}
2174
2175static long loop_control_ioctl(struct file *file, unsigned int cmd,
2176 unsigned long parm)
2177{
2178 switch (cmd) {
2179 case LOOP_CTL_ADD:
2180 return loop_add(i: parm);
2181 case LOOP_CTL_REMOVE:
2182 return loop_control_remove(idx: parm);
2183 case LOOP_CTL_GET_FREE:
2184 return loop_control_get_free(idx: parm);
2185 default:
2186 return -ENOSYS;
2187 }
2188}
2189
2190static const struct file_operations loop_ctl_fops = {
2191 .open = nonseekable_open,
2192 .unlocked_ioctl = loop_control_ioctl,
2193 .compat_ioctl = loop_control_ioctl,
2194 .owner = THIS_MODULE,
2195 .llseek = noop_llseek,
2196};
2197
2198static struct miscdevice loop_misc = {
2199 .minor = LOOP_CTRL_MINOR,
2200 .name = "loop-control",
2201 .fops = &loop_ctl_fops,
2202};
2203
2204MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2205MODULE_ALIAS("devname:loop-control");
2206
2207static int __init loop_init(void)
2208{
2209 int i;
2210 int err;
2211
2212 part_shift = 0;
2213 if (max_part > 0) {
2214 part_shift = fls(x: max_part);
2215
2216 /*
2217 * Adjust max_part according to part_shift as it is exported
2218 * to user space so that user can decide correct minor number
2219 * if [s]he want to create more devices.
2220 *
2221 * Note that -1 is required because partition 0 is reserved
2222 * for the whole disk.
2223 */
2224 max_part = (1UL << part_shift) - 1;
2225 }
2226
2227 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2228 err = -EINVAL;
2229 goto err_out;
2230 }
2231
2232 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2233 err = -EINVAL;
2234 goto err_out;
2235 }
2236
2237 err = misc_register(misc: &loop_misc);
2238 if (err < 0)
2239 goto err_out;
2240
2241
2242 if (__register_blkdev(LOOP_MAJOR, name: "loop", probe: loop_probe)) {
2243 err = -EIO;
2244 goto misc_out;
2245 }
2246
2247 /* pre-create number of devices given by config or max_loop */
2248 for (i = 0; i < max_loop; i++)
2249 loop_add(i);
2250
2251 printk(KERN_INFO "loop: module loaded\n");
2252 return 0;
2253
2254misc_out:
2255 misc_deregister(misc: &loop_misc);
2256err_out:
2257 return err;
2258}
2259
2260static void __exit loop_exit(void)
2261{
2262 struct loop_device *lo;
2263 int id;
2264
2265 unregister_blkdev(LOOP_MAJOR, name: "loop");
2266 misc_deregister(misc: &loop_misc);
2267
2268 /*
2269 * There is no need to use loop_ctl_mutex here, for nobody else can
2270 * access loop_index_idr when this module is unloading (unless forced
2271 * module unloading is requested). If this is not a clean unloading,
2272 * we have no means to avoid kernel crash.
2273 */
2274 idr_for_each_entry(&loop_index_idr, lo, id)
2275 loop_remove(lo);
2276
2277 idr_destroy(&loop_index_idr);
2278}
2279
2280module_init(loop_init);
2281module_exit(loop_exit);
2282
2283#ifndef MODULE
2284static int __init max_loop_setup(char *str)
2285{
2286 max_loop = simple_strtol(str, NULL, 0);
2287#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2288 max_loop_specified = true;
2289#endif
2290 return 1;
2291}
2292
2293__setup("max_loop=", max_loop_setup);
2294#endif
2295

source code of linux/drivers/block/loop.c