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(lock: &loop_validate_mutex);
115 if (err)
116 return err;
117 }
118 err = mutex_lock_killable(lock: &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 lo_calculate_size(struct loop_device *lo, struct file *file)
141{
142 loff_t loopsize;
143 int ret;
144
145 if (S_ISBLK(file_inode(file)->i_mode)) {
146 loopsize = i_size_read(inode: file->f_mapping->host);
147 } else {
148 struct kstat stat;
149
150 /*
151 * Get the accurate file size. This provides better results than
152 * cached inode data, particularly for network filesystems where
153 * metadata may be stale.
154 */
155 ret = vfs_getattr_nosec(&file->f_path, &stat, STATX_SIZE, 0);
156 if (ret)
157 return 0;
158
159 loopsize = stat.size;
160 }
161
162 if (lo->lo_offset > 0)
163 loopsize -= lo->lo_offset;
164 /* offset is beyond i_size, weird but possible */
165 if (loopsize < 0)
166 return 0;
167 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
168 loopsize = lo->lo_sizelimit;
169 /*
170 * Unfortunately, if we want to do I/O on the device,
171 * the number of 512-byte sectors has to fit into a sector_t.
172 */
173 return loopsize >> 9;
174}
175
176/*
177 * We support direct I/O only if lo_offset is aligned with the logical I/O size
178 * of backing device, and the logical block size of loop is bigger than that of
179 * the backing device.
180 */
181static bool lo_can_use_dio(struct loop_device *lo)
182{
183 if (!(lo->lo_backing_file->f_mode & FMODE_CAN_ODIRECT))
184 return false;
185 if (queue_logical_block_size(q: lo->lo_queue) < lo->lo_min_dio_size)
186 return false;
187 if (lo->lo_offset & (lo->lo_min_dio_size - 1))
188 return false;
189 return true;
190}
191
192/*
193 * Direct I/O can be enabled either by using an O_DIRECT file descriptor, or by
194 * passing in the LO_FLAGS_DIRECT_IO flag from userspace. It will be silently
195 * disabled when the device block size is too small or the offset is unaligned.
196 *
197 * loop_get_status will always report the effective LO_FLAGS_DIRECT_IO flag and
198 * not the originally passed in one.
199 */
200static inline void loop_update_dio(struct loop_device *lo)
201{
202 lockdep_assert_held(&lo->lo_mutex);
203 WARN_ON_ONCE(lo->lo_state == Lo_bound &&
204 lo->lo_queue->mq_freeze_depth == 0);
205
206 if ((lo->lo_flags & LO_FLAGS_DIRECT_IO) && !lo_can_use_dio(lo))
207 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
208}
209
210/**
211 * loop_set_size() - sets device size and notifies userspace
212 * @lo: struct loop_device to set the size for
213 * @size: new size of the loop device
214 *
215 * Callers must validate that the size passed into this function fits into
216 * a sector_t, eg using loop_validate_size()
217 */
218static void loop_set_size(struct loop_device *lo, loff_t size)
219{
220 if (!set_capacity_and_notify(disk: lo->lo_disk, size))
221 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
222}
223
224static void loop_clear_limits(struct loop_device *lo, int mode)
225{
226 struct queue_limits lim = queue_limits_start_update(q: lo->lo_queue);
227
228 if (mode & FALLOC_FL_ZERO_RANGE)
229 lim.max_write_zeroes_sectors = 0;
230
231 if (mode & FALLOC_FL_PUNCH_HOLE) {
232 lim.max_hw_discard_sectors = 0;
233 lim.discard_granularity = 0;
234 }
235
236 /*
237 * XXX: this updates the queue limits without freezing the queue, which
238 * is against the locking protocol and dangerous. But we can't just
239 * freeze the queue as we're inside the ->queue_rq method here. So this
240 * should move out into a workqueue unless we get the file operations to
241 * advertise if they support specific fallocate operations.
242 */
243 queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
244}
245
246static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
247 int mode)
248{
249 /*
250 * We use fallocate to manipulate the space mappings used by the image
251 * a.k.a. discard/zerorange.
252 */
253 struct file *file = lo->lo_backing_file;
254 int ret;
255
256 mode |= FALLOC_FL_KEEP_SIZE;
257
258 if (!bdev_max_discard_sectors(bdev: lo->lo_device))
259 return -EOPNOTSUPP;
260
261 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
262 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
263 return -EIO;
264
265 /*
266 * We initially configure the limits in a hope that fallocate is
267 * supported and clear them here if that turns out not to be true.
268 */
269 if (unlikely(ret == -EOPNOTSUPP))
270 loop_clear_limits(lo, mode);
271
272 return ret;
273}
274
275static int lo_req_flush(struct loop_device *lo, struct request *rq)
276{
277 int ret = vfs_fsync(file: lo->lo_backing_file, datasync: 0);
278 if (unlikely(ret && ret != -EINVAL))
279 ret = -EIO;
280
281 return ret;
282}
283
284static void lo_complete_rq(struct request *rq)
285{
286 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
287 blk_status_t ret = BLK_STS_OK;
288
289 if (cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
290 req_op(req: rq) != REQ_OP_READ) {
291 if (cmd->ret < 0)
292 ret = errno_to_blk_status(errno: cmd->ret);
293 goto end_io;
294 }
295
296 /*
297 * Short READ - if we got some data, advance our request and
298 * retry it. If we got no data, end the rest with EIO.
299 */
300 if (cmd->ret) {
301 blk_update_request(rq, BLK_STS_OK, nr_bytes: cmd->ret);
302 cmd->ret = 0;
303 blk_mq_requeue_request(rq, kick_requeue_list: true);
304 } else {
305 struct bio *bio = rq->bio;
306
307 while (bio) {
308 zero_fill_bio(bio);
309 bio = bio->bi_next;
310 }
311
312 ret = BLK_STS_IOERR;
313end_io:
314 blk_mq_end_request(rq, error: ret);
315 }
316}
317
318static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
319{
320 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
321
322 if (!atomic_dec_and_test(v: &cmd->ref))
323 return;
324 kfree(objp: cmd->bvec);
325 cmd->bvec = NULL;
326 if (req_op(req: rq) == REQ_OP_WRITE)
327 kiocb_end_write(iocb: &cmd->iocb);
328 if (likely(!blk_should_fake_timeout(rq->q)))
329 blk_mq_complete_request(rq);
330}
331
332static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
333{
334 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
335
336 cmd->ret = ret;
337 lo_rw_aio_do_completion(cmd);
338}
339
340static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
341 loff_t pos, int rw)
342{
343 struct iov_iter iter;
344 struct req_iterator rq_iter;
345 struct bio_vec *bvec;
346 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
347 struct bio *bio = rq->bio;
348 struct file *file = lo->lo_backing_file;
349 struct bio_vec tmp;
350 unsigned int offset;
351 int nr_bvec = 0;
352 int ret;
353
354 rq_for_each_bvec(tmp, rq, rq_iter)
355 nr_bvec++;
356
357 if (rq->bio != rq->biotail) {
358
359 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
360 GFP_NOIO);
361 if (!bvec)
362 return -EIO;
363 cmd->bvec = bvec;
364
365 /*
366 * The bios of the request may be started from the middle of
367 * the 'bvec' because of bio splitting, so we can't directly
368 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
369 * API will take care of all details for us.
370 */
371 rq_for_each_bvec(tmp, rq, rq_iter) {
372 *bvec = tmp;
373 bvec++;
374 }
375 bvec = cmd->bvec;
376 offset = 0;
377 } else {
378 /*
379 * Same here, this bio may be started from the middle of the
380 * 'bvec' because of bio splitting, so offset from the bvec
381 * must be passed to iov iterator
382 */
383 offset = bio->bi_iter.bi_bvec_done;
384 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
385 }
386 atomic_set(v: &cmd->ref, i: 2);
387
388 iov_iter_bvec(i: &iter, direction: rw, bvec, nr_segs: nr_bvec, count: blk_rq_bytes(rq));
389 iter.iov_offset = offset;
390
391 cmd->iocb.ki_pos = pos;
392 cmd->iocb.ki_filp = file;
393 cmd->iocb.ki_ioprio = req_get_ioprio(req: rq);
394 if (cmd->use_aio) {
395 cmd->iocb.ki_complete = lo_rw_aio_complete;
396 cmd->iocb.ki_flags = IOCB_DIRECT;
397 } else {
398 cmd->iocb.ki_complete = NULL;
399 cmd->iocb.ki_flags = 0;
400 }
401
402 if (rw == ITER_SOURCE) {
403 kiocb_start_write(iocb: &cmd->iocb);
404 ret = file->f_op->write_iter(&cmd->iocb, &iter);
405 } else
406 ret = file->f_op->read_iter(&cmd->iocb, &iter);
407
408 lo_rw_aio_do_completion(cmd);
409
410 if (ret != -EIOCBQUEUED)
411 lo_rw_aio_complete(iocb: &cmd->iocb, ret);
412 return -EIOCBQUEUED;
413}
414
415static int do_req_filebacked(struct loop_device *lo, struct request *rq)
416{
417 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
418 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
419
420 switch (req_op(req: rq)) {
421 case REQ_OP_FLUSH:
422 return lo_req_flush(lo, rq);
423 case REQ_OP_WRITE_ZEROES:
424 /*
425 * If the caller doesn't want deallocation, call zeroout to
426 * write zeroes the range. Otherwise, punch them out.
427 */
428 return lo_fallocate(lo, rq, pos,
429 mode: (rq->cmd_flags & REQ_NOUNMAP) ?
430 FALLOC_FL_ZERO_RANGE :
431 FALLOC_FL_PUNCH_HOLE);
432 case REQ_OP_DISCARD:
433 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
434 case REQ_OP_WRITE:
435 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
436 case REQ_OP_READ:
437 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
438 default:
439 WARN_ON_ONCE(1);
440 return -EIO;
441 }
442}
443
444static void loop_reread_partitions(struct loop_device *lo)
445{
446 int rc;
447
448 mutex_lock(lock: &lo->lo_disk->open_mutex);
449 rc = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
450 mutex_unlock(lock: &lo->lo_disk->open_mutex);
451 if (rc)
452 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
453 __func__, lo->lo_number, lo->lo_file_name, rc);
454}
455
456static unsigned int loop_query_min_dio_size(struct loop_device *lo)
457{
458 struct file *file = lo->lo_backing_file;
459 struct block_device *sb_bdev = file->f_mapping->host->i_sb->s_bdev;
460 struct kstat st;
461
462 /*
463 * Use the minimal dio alignment of the file system if provided.
464 */
465 if (!vfs_getattr(&file->f_path, &st, STATX_DIOALIGN, 0) &&
466 (st.result_mask & STATX_DIOALIGN))
467 return st.dio_offset_align;
468
469 /*
470 * In a perfect world this wouldn't be needed, but as of Linux 6.13 only
471 * a handful of file systems support the STATX_DIOALIGN flag.
472 */
473 if (sb_bdev)
474 return bdev_logical_block_size(bdev: sb_bdev);
475 return SECTOR_SIZE;
476}
477
478static inline int is_loop_device(struct file *file)
479{
480 struct inode *i = file->f_mapping->host;
481
482 return i && S_ISBLK(i->i_mode) && imajor(inode: i) == LOOP_MAJOR;
483}
484
485static int loop_validate_file(struct file *file, struct block_device *bdev)
486{
487 struct inode *inode = file->f_mapping->host;
488 struct file *f = file;
489
490 /* Avoid recursion */
491 while (is_loop_device(file: f)) {
492 struct loop_device *l;
493
494 lockdep_assert_held(&loop_validate_mutex);
495 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
496 return -EBADF;
497
498 l = I_BDEV(inode: f->f_mapping->host)->bd_disk->private_data;
499 if (l->lo_state != Lo_bound)
500 return -EINVAL;
501 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
502 rmb();
503 f = l->lo_backing_file;
504 }
505 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
506 return -EINVAL;
507 return 0;
508}
509
510static void loop_assign_backing_file(struct loop_device *lo, struct file *file)
511{
512 lo->lo_backing_file = file;
513 lo->old_gfp_mask = mapping_gfp_mask(mapping: file->f_mapping);
514 mapping_set_gfp_mask(m: file->f_mapping,
515 mask: lo->old_gfp_mask & ~(__GFP_IO | __GFP_FS));
516 if (lo->lo_backing_file->f_flags & O_DIRECT)
517 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
518 lo->lo_min_dio_size = loop_query_min_dio_size(lo);
519}
520
521static int loop_check_backing_file(struct file *file)
522{
523 if (!file->f_op->read_iter)
524 return -EINVAL;
525
526 if ((file->f_mode & FMODE_WRITE) && !file->f_op->write_iter)
527 return -EINVAL;
528
529 return 0;
530}
531
532/*
533 * loop_change_fd switched the backing store of a loopback device to
534 * a new file. This is useful for operating system installers to free up
535 * the original file and in High Availability environments to switch to
536 * an alternative location for the content in case of server meltdown.
537 * This can only work if the loop device is used read-only, and if the
538 * new backing store is the same size and type as the old backing store.
539 */
540static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
541 unsigned int arg)
542{
543 struct file *file = fget(fd: arg);
544 struct file *old_file;
545 unsigned int memflags;
546 int error;
547 bool partscan;
548 bool is_loop;
549
550 if (!file)
551 return -EBADF;
552
553 error = loop_check_backing_file(file);
554 if (error) {
555 fput(file);
556 return error;
557 }
558
559 /* suppress uevents while reconfiguring the device */
560 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
561
562 is_loop = is_loop_device(file);
563 error = loop_global_lock_killable(lo, global: is_loop);
564 if (error)
565 goto out_putf;
566 error = -ENXIO;
567 if (lo->lo_state != Lo_bound)
568 goto out_err;
569
570 /* the loop device has to be read-only */
571 error = -EINVAL;
572 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
573 goto out_err;
574
575 error = loop_validate_file(file, bdev);
576 if (error)
577 goto out_err;
578
579 old_file = lo->lo_backing_file;
580
581 error = -EINVAL;
582
583 /* size of the new backing store needs to be the same */
584 if (lo_calculate_size(lo, file) != lo_calculate_size(lo, file: old_file))
585 goto out_err;
586
587 /*
588 * We might switch to direct I/O mode for the loop device, write back
589 * all dirty data the page cache now that so that the individual I/O
590 * operations don't have to do that.
591 */
592 vfs_fsync(file, datasync: 0);
593
594 /* and ... switch */
595 disk_force_media_change(disk: lo->lo_disk);
596 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
597 mapping_set_gfp_mask(m: old_file->f_mapping, mask: lo->old_gfp_mask);
598 loop_assign_backing_file(lo, file);
599 loop_update_dio(lo);
600 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
601 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
602 loop_global_unlock(lo, global: is_loop);
603
604 /*
605 * Flush loop_validate_file() before fput(), for l->lo_backing_file
606 * might be pointing at old_file which might be the last reference.
607 */
608 if (!is_loop) {
609 mutex_lock(lock: &loop_validate_mutex);
610 mutex_unlock(lock: &loop_validate_mutex);
611 }
612 /*
613 * We must drop file reference outside of lo_mutex as dropping
614 * the file ref can take open_mutex which creates circular locking
615 * dependency.
616 */
617 fput(old_file);
618 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
619 if (partscan)
620 loop_reread_partitions(lo);
621
622 error = 0;
623done:
624 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
625 return error;
626
627out_err:
628 loop_global_unlock(lo, global: is_loop);
629out_putf:
630 fput(file);
631 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
632 goto done;
633}
634
635/* loop sysfs attributes */
636
637static ssize_t loop_attr_show(struct device *dev, char *page,
638 ssize_t (*callback)(struct loop_device *, char *))
639{
640 struct gendisk *disk = dev_to_disk(dev);
641 struct loop_device *lo = disk->private_data;
642
643 return callback(lo, page);
644}
645
646#define LOOP_ATTR_RO(_name) \
647static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
648static ssize_t loop_attr_do_show_##_name(struct device *d, \
649 struct device_attribute *attr, char *b) \
650{ \
651 return loop_attr_show(d, b, loop_attr_##_name##_show); \
652} \
653static struct device_attribute loop_attr_##_name = \
654 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
655
656static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
657{
658 ssize_t ret;
659 char *p = NULL;
660
661 spin_lock_irq(lock: &lo->lo_lock);
662 if (lo->lo_backing_file)
663 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
664 spin_unlock_irq(lock: &lo->lo_lock);
665
666 if (IS_ERR_OR_NULL(ptr: p))
667 ret = PTR_ERR(ptr: p);
668 else {
669 ret = strlen(p);
670 memmove(dest: buf, src: p, count: ret);
671 buf[ret++] = '\n';
672 buf[ret] = 0;
673 }
674
675 return ret;
676}
677
678static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
679{
680 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_offset);
681}
682
683static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
684{
685 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_sizelimit);
686}
687
688static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
689{
690 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
691
692 return sysfs_emit(buf, fmt: "%s\n", autoclear ? "1" : "0");
693}
694
695static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
696{
697 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
698
699 return sysfs_emit(buf, fmt: "%s\n", partscan ? "1" : "0");
700}
701
702static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
703{
704 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
705
706 return sysfs_emit(buf, fmt: "%s\n", dio ? "1" : "0");
707}
708
709LOOP_ATTR_RO(backing_file);
710LOOP_ATTR_RO(offset);
711LOOP_ATTR_RO(sizelimit);
712LOOP_ATTR_RO(autoclear);
713LOOP_ATTR_RO(partscan);
714LOOP_ATTR_RO(dio);
715
716static struct attribute *loop_attrs[] = {
717 &loop_attr_backing_file.attr,
718 &loop_attr_offset.attr,
719 &loop_attr_sizelimit.attr,
720 &loop_attr_autoclear.attr,
721 &loop_attr_partscan.attr,
722 &loop_attr_dio.attr,
723 NULL,
724};
725
726static struct attribute_group loop_attribute_group = {
727 .name = "loop",
728 .attrs= loop_attrs,
729};
730
731static void loop_sysfs_init(struct loop_device *lo)
732{
733 lo->sysfs_inited = !sysfs_create_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
734 grp: &loop_attribute_group);
735}
736
737static void loop_sysfs_exit(struct loop_device *lo)
738{
739 if (lo->sysfs_inited)
740 sysfs_remove_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
741 grp: &loop_attribute_group);
742}
743
744static void loop_get_discard_config(struct loop_device *lo,
745 u32 *granularity, u32 *max_discard_sectors)
746{
747 struct file *file = lo->lo_backing_file;
748 struct inode *inode = file->f_mapping->host;
749 struct kstatfs sbuf;
750
751 /*
752 * If the backing device is a block device, mirror its zeroing
753 * capability. Set the discard sectors to the block device's zeroing
754 * capabilities because loop discards result in blkdev_issue_zeroout(),
755 * not blkdev_issue_discard(). This maintains consistent behavior with
756 * file-backed loop devices: discarded regions read back as zero.
757 */
758 if (S_ISBLK(inode->i_mode)) {
759 struct block_device *bdev = I_BDEV(inode);
760
761 *max_discard_sectors = bdev_write_zeroes_sectors(bdev);
762 *granularity = bdev_discard_granularity(bdev);
763
764 /*
765 * We use punch hole to reclaim the free space used by the
766 * image a.k.a. discard.
767 */
768 } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
769 *max_discard_sectors = UINT_MAX >> 9;
770 *granularity = sbuf.f_bsize;
771 }
772}
773
774struct loop_worker {
775 struct rb_node rb_node;
776 struct work_struct work;
777 struct list_head cmd_list;
778 struct list_head idle_list;
779 struct loop_device *lo;
780 struct cgroup_subsys_state *blkcg_css;
781 unsigned long last_ran_at;
782};
783
784static void loop_workfn(struct work_struct *work);
785
786#ifdef CONFIG_BLK_CGROUP
787static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
788{
789 return !css || css == blkcg_root_css;
790}
791#else
792static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
793{
794 return !css;
795}
796#endif
797
798static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
799{
800 struct rb_node **node, *parent = NULL;
801 struct loop_worker *cur_worker, *worker = NULL;
802 struct work_struct *work;
803 struct list_head *cmd_list;
804
805 spin_lock_irq(lock: &lo->lo_work_lock);
806
807 if (queue_on_root_worker(css: cmd->blkcg_css))
808 goto queue_work;
809
810 node = &lo->worker_tree.rb_node;
811
812 while (*node) {
813 parent = *node;
814 cur_worker = container_of(*node, struct loop_worker, rb_node);
815 if (cur_worker->blkcg_css == cmd->blkcg_css) {
816 worker = cur_worker;
817 break;
818 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
819 node = &(*node)->rb_left;
820 } else {
821 node = &(*node)->rb_right;
822 }
823 }
824 if (worker)
825 goto queue_work;
826
827 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT);
828 /*
829 * In the event we cannot allocate a worker, just queue on the
830 * rootcg worker and issue the I/O as the rootcg
831 */
832 if (!worker) {
833 cmd->blkcg_css = NULL;
834 if (cmd->memcg_css)
835 css_put(css: cmd->memcg_css);
836 cmd->memcg_css = NULL;
837 goto queue_work;
838 }
839
840 worker->blkcg_css = cmd->blkcg_css;
841 css_get(css: worker->blkcg_css);
842 INIT_WORK(&worker->work, loop_workfn);
843 INIT_LIST_HEAD(list: &worker->cmd_list);
844 INIT_LIST_HEAD(list: &worker->idle_list);
845 worker->lo = lo;
846 rb_link_node(node: &worker->rb_node, parent, rb_link: node);
847 rb_insert_color(&worker->rb_node, &lo->worker_tree);
848queue_work:
849 if (worker) {
850 /*
851 * We need to remove from the idle list here while
852 * holding the lock so that the idle timer doesn't
853 * free the worker
854 */
855 if (!list_empty(head: &worker->idle_list))
856 list_del_init(entry: &worker->idle_list);
857 work = &worker->work;
858 cmd_list = &worker->cmd_list;
859 } else {
860 work = &lo->rootcg_work;
861 cmd_list = &lo->rootcg_cmd_list;
862 }
863 list_add_tail(new: &cmd->list_entry, head: cmd_list);
864 queue_work(wq: lo->workqueue, work);
865 spin_unlock_irq(lock: &lo->lo_work_lock);
866}
867
868static void loop_set_timer(struct loop_device *lo)
869{
870 timer_reduce(timer: &lo->timer, expires: jiffies + LOOP_IDLE_WORKER_TIMEOUT);
871}
872
873static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
874{
875 struct loop_worker *pos, *worker;
876
877 spin_lock_irq(lock: &lo->lo_work_lock);
878 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
879 idle_list) {
880 if (!delete_all &&
881 time_is_after_jiffies(worker->last_ran_at +
882 LOOP_IDLE_WORKER_TIMEOUT))
883 break;
884 list_del(entry: &worker->idle_list);
885 rb_erase(&worker->rb_node, &lo->worker_tree);
886 css_put(css: worker->blkcg_css);
887 kfree(objp: worker);
888 }
889 if (!list_empty(head: &lo->idle_worker_list))
890 loop_set_timer(lo);
891 spin_unlock_irq(lock: &lo->lo_work_lock);
892}
893
894static void loop_free_idle_workers_timer(struct timer_list *timer)
895{
896 struct loop_device *lo = container_of(timer, struct loop_device, timer);
897
898 return loop_free_idle_workers(lo, delete_all: false);
899}
900
901/**
902 * loop_set_status_from_info - configure device from loop_info
903 * @lo: struct loop_device to configure
904 * @info: struct loop_info64 to configure the device with
905 *
906 * Configures the loop device parameters according to the passed
907 * in loop_info64 configuration.
908 */
909static int
910loop_set_status_from_info(struct loop_device *lo,
911 const struct loop_info64 *info)
912{
913 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
914 return -EINVAL;
915
916 switch (info->lo_encrypt_type) {
917 case LO_CRYPT_NONE:
918 break;
919 case LO_CRYPT_XOR:
920 pr_warn("support for the xor transformation has been removed.\n");
921 return -EINVAL;
922 case LO_CRYPT_CRYPTOAPI:
923 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
924 return -EINVAL;
925 default:
926 return -EINVAL;
927 }
928
929 /* Avoid assigning overflow values */
930 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
931 return -EOVERFLOW;
932
933 lo->lo_offset = info->lo_offset;
934 lo->lo_sizelimit = info->lo_sizelimit;
935
936 memcpy(to: lo->lo_file_name, from: info->lo_file_name, LO_NAME_SIZE);
937 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
938 return 0;
939}
940
941static unsigned int loop_default_blocksize(struct loop_device *lo)
942{
943 /* In case of direct I/O, match underlying minimum I/O size */
944 if (lo->lo_flags & LO_FLAGS_DIRECT_IO)
945 return lo->lo_min_dio_size;
946 return SECTOR_SIZE;
947}
948
949static void loop_update_limits(struct loop_device *lo, struct queue_limits *lim,
950 unsigned int bsize)
951{
952 struct file *file = lo->lo_backing_file;
953 struct inode *inode = file->f_mapping->host;
954 struct block_device *backing_bdev = NULL;
955 u32 granularity = 0, max_discard_sectors = 0;
956
957 if (S_ISBLK(inode->i_mode))
958 backing_bdev = I_BDEV(inode);
959 else if (inode->i_sb->s_bdev)
960 backing_bdev = inode->i_sb->s_bdev;
961
962 if (!bsize)
963 bsize = loop_default_blocksize(lo);
964
965 loop_get_discard_config(lo, granularity: &granularity, max_discard_sectors: &max_discard_sectors);
966
967 lim->logical_block_size = bsize;
968 lim->physical_block_size = bsize;
969 lim->io_min = bsize;
970 lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL);
971 if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY))
972 lim->features |= BLK_FEAT_WRITE_CACHE;
973 if (backing_bdev && !bdev_nonrot(bdev: backing_bdev))
974 lim->features |= BLK_FEAT_ROTATIONAL;
975 lim->max_hw_discard_sectors = max_discard_sectors;
976 lim->max_write_zeroes_sectors = max_discard_sectors;
977 if (max_discard_sectors)
978 lim->discard_granularity = granularity;
979 else
980 lim->discard_granularity = 0;
981}
982
983static int loop_configure(struct loop_device *lo, blk_mode_t mode,
984 struct block_device *bdev,
985 const struct loop_config *config)
986{
987 struct file *file = fget(fd: config->fd);
988 struct queue_limits lim;
989 int error;
990 loff_t size;
991 bool partscan;
992 bool is_loop;
993
994 if (!file)
995 return -EBADF;
996
997 error = loop_check_backing_file(file);
998 if (error) {
999 fput(file);
1000 return error;
1001 }
1002
1003 is_loop = is_loop_device(file);
1004
1005 /* This is safe, since we have a reference from open(). */
1006 __module_get(THIS_MODULE);
1007
1008 /*
1009 * If we don't hold exclusive handle for the device, upgrade to it
1010 * here to avoid changing device under exclusive owner.
1011 */
1012 if (!(mode & BLK_OPEN_EXCL)) {
1013 error = bd_prepare_to_claim(bdev, holder: loop_configure, NULL);
1014 if (error)
1015 goto out_putf;
1016 }
1017
1018 error = loop_global_lock_killable(lo, global: is_loop);
1019 if (error)
1020 goto out_bdev;
1021
1022 error = -EBUSY;
1023 if (lo->lo_state != Lo_unbound)
1024 goto out_unlock;
1025
1026 error = loop_validate_file(file, bdev);
1027 if (error)
1028 goto out_unlock;
1029
1030 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1031 error = -EINVAL;
1032 goto out_unlock;
1033 }
1034
1035 error = loop_set_status_from_info(lo, info: &config->info);
1036 if (error)
1037 goto out_unlock;
1038 lo->lo_flags = config->info.lo_flags;
1039
1040 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1041 !file->f_op->write_iter)
1042 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1043
1044 if (!lo->workqueue) {
1045 lo->workqueue = alloc_workqueue("loop%d",
1046 WQ_UNBOUND | WQ_FREEZABLE,
1047 0, lo->lo_number);
1048 if (!lo->workqueue) {
1049 error = -ENOMEM;
1050 goto out_unlock;
1051 }
1052 }
1053
1054 /* suppress uevents while reconfiguring the device */
1055 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
1056
1057 disk_force_media_change(disk: lo->lo_disk);
1058 set_disk_ro(disk: lo->lo_disk, read_only: (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1059
1060 lo->lo_device = bdev;
1061 loop_assign_backing_file(lo, file);
1062
1063 lim = queue_limits_start_update(q: lo->lo_queue);
1064 loop_update_limits(lo, lim: &lim, bsize: config->block_size);
1065 /* No need to freeze the queue as the device isn't bound yet. */
1066 error = queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1067 if (error)
1068 goto out_unlock;
1069
1070 /*
1071 * We might switch to direct I/O mode for the loop device, write back
1072 * all dirty data the page cache now that so that the individual I/O
1073 * operations don't have to do that.
1074 */
1075 vfs_fsync(file, datasync: 0);
1076
1077 loop_update_dio(lo);
1078 loop_sysfs_init(lo);
1079
1080 size = lo_calculate_size(lo, file);
1081 loop_set_size(lo, size);
1082
1083 /* Order wrt reading lo_state in loop_validate_file(). */
1084 wmb();
1085
1086 lo->lo_state = Lo_bound;
1087 if (part_shift)
1088 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1089 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1090 if (partscan)
1091 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1092
1093 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
1094 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
1095
1096 loop_global_unlock(lo, global: is_loop);
1097 if (partscan)
1098 loop_reread_partitions(lo);
1099
1100 if (!(mode & BLK_OPEN_EXCL))
1101 bd_abort_claiming(bdev, holder: loop_configure);
1102
1103 return 0;
1104
1105out_unlock:
1106 loop_global_unlock(lo, global: is_loop);
1107out_bdev:
1108 if (!(mode & BLK_OPEN_EXCL))
1109 bd_abort_claiming(bdev, holder: loop_configure);
1110out_putf:
1111 fput(file);
1112 /* This is safe: open() is still holding a reference. */
1113 module_put(THIS_MODULE);
1114 return error;
1115}
1116
1117static void __loop_clr_fd(struct loop_device *lo)
1118{
1119 struct queue_limits lim;
1120 struct file *filp;
1121 gfp_t gfp = lo->old_gfp_mask;
1122
1123 spin_lock_irq(lock: &lo->lo_lock);
1124 filp = lo->lo_backing_file;
1125 lo->lo_backing_file = NULL;
1126 spin_unlock_irq(lock: &lo->lo_lock);
1127
1128 lo->lo_device = NULL;
1129 lo->lo_offset = 0;
1130 lo->lo_sizelimit = 0;
1131 memset(s: lo->lo_file_name, c: 0, LO_NAME_SIZE);
1132
1133 /*
1134 * Reset the block size to the default.
1135 *
1136 * No queue freezing needed because this is called from the final
1137 * ->release call only, so there can't be any outstanding I/O.
1138 */
1139 lim = queue_limits_start_update(q: lo->lo_queue);
1140 lim.logical_block_size = SECTOR_SIZE;
1141 lim.physical_block_size = SECTOR_SIZE;
1142 lim.io_min = SECTOR_SIZE;
1143 queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1144
1145 invalidate_disk(disk: lo->lo_disk);
1146 loop_sysfs_exit(lo);
1147 /* let user-space know about this change */
1148 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
1149 mapping_set_gfp_mask(m: filp->f_mapping, mask: gfp);
1150 /* This is safe: open() is still holding a reference. */
1151 module_put(THIS_MODULE);
1152
1153 disk_force_media_change(disk: lo->lo_disk);
1154
1155 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1156 int err;
1157
1158 /*
1159 * open_mutex has been held already in release path, so don't
1160 * acquire it if this function is called in such case.
1161 *
1162 * If the reread partition isn't from release path, lo_refcnt
1163 * must be at least one and it can only become zero when the
1164 * current holder is released.
1165 */
1166 err = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
1167 if (err)
1168 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1169 __func__, lo->lo_number, err);
1170 /* Device is gone, no point in returning error */
1171 }
1172
1173 /*
1174 * lo->lo_state is set to Lo_unbound here after above partscan has
1175 * finished. There cannot be anybody else entering __loop_clr_fd() as
1176 * Lo_rundown state protects us from all the other places trying to
1177 * change the 'lo' device.
1178 */
1179 lo->lo_flags = 0;
1180 if (!part_shift)
1181 set_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1182 mutex_lock(lock: &lo->lo_mutex);
1183 lo->lo_state = Lo_unbound;
1184 mutex_unlock(lock: &lo->lo_mutex);
1185
1186 /*
1187 * Need not hold lo_mutex to fput backing file. Calling fput holding
1188 * lo_mutex triggers a circular lock dependency possibility warning as
1189 * fput can take open_mutex which is usually taken before lo_mutex.
1190 */
1191 fput(filp);
1192}
1193
1194static int loop_clr_fd(struct loop_device *lo)
1195{
1196 int err;
1197
1198 /*
1199 * Since lo_ioctl() is called without locks held, it is possible that
1200 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1201 *
1202 * Therefore, use global lock when setting Lo_rundown state in order to
1203 * make sure that loop_validate_file() will fail if the "struct file"
1204 * which loop_configure()/loop_change_fd() found via fget() was this
1205 * loop device.
1206 */
1207 err = loop_global_lock_killable(lo, global: true);
1208 if (err)
1209 return err;
1210 if (lo->lo_state != Lo_bound) {
1211 loop_global_unlock(lo, global: true);
1212 return -ENXIO;
1213 }
1214 /*
1215 * Mark the device for removing the backing device on last close.
1216 * If we are the only opener, also switch the state to roundown here to
1217 * prevent new openers from coming in.
1218 */
1219
1220 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1221 if (disk_openers(disk: lo->lo_disk) == 1)
1222 lo->lo_state = Lo_rundown;
1223 loop_global_unlock(lo, global: true);
1224
1225 return 0;
1226}
1227
1228static int
1229loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1230{
1231 int err;
1232 bool partscan = false;
1233 bool size_changed = false;
1234 unsigned int memflags;
1235
1236 err = mutex_lock_killable(lock: &lo->lo_mutex);
1237 if (err)
1238 return err;
1239 if (lo->lo_state != Lo_bound) {
1240 err = -ENXIO;
1241 goto out_unlock;
1242 }
1243
1244 if (lo->lo_offset != info->lo_offset ||
1245 lo->lo_sizelimit != info->lo_sizelimit) {
1246 size_changed = true;
1247 sync_blockdev(bdev: lo->lo_device);
1248 invalidate_bdev(bdev: lo->lo_device);
1249 }
1250
1251 /* I/O needs to be drained before changing lo_offset or lo_sizelimit */
1252 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1253
1254 err = loop_set_status_from_info(lo, info);
1255 if (err)
1256 goto out_unfreeze;
1257
1258 partscan = !(lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1259 (info->lo_flags & LO_FLAGS_PARTSCAN);
1260
1261 lo->lo_flags &= ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1262 lo->lo_flags |= (info->lo_flags & LOOP_SET_STATUS_SETTABLE_FLAGS);
1263
1264 /* update the direct I/O flag if lo_offset changed */
1265 loop_update_dio(lo);
1266
1267out_unfreeze:
1268 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1269 if (partscan)
1270 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1271 if (!err && size_changed) {
1272 loff_t new_size = lo_calculate_size(lo, file: lo->lo_backing_file);
1273 loop_set_size(lo, size: new_size);
1274 }
1275out_unlock:
1276 mutex_unlock(lock: &lo->lo_mutex);
1277 if (partscan)
1278 loop_reread_partitions(lo);
1279
1280 return err;
1281}
1282
1283static int
1284loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1285{
1286 struct path path;
1287 struct kstat stat;
1288 int ret;
1289
1290 ret = mutex_lock_killable(lock: &lo->lo_mutex);
1291 if (ret)
1292 return ret;
1293 if (lo->lo_state != Lo_bound) {
1294 mutex_unlock(lock: &lo->lo_mutex);
1295 return -ENXIO;
1296 }
1297
1298 memset(s: info, c: 0, n: sizeof(*info));
1299 info->lo_number = lo->lo_number;
1300 info->lo_offset = lo->lo_offset;
1301 info->lo_sizelimit = lo->lo_sizelimit;
1302 info->lo_flags = lo->lo_flags;
1303 memcpy(to: info->lo_file_name, from: lo->lo_file_name, LO_NAME_SIZE);
1304
1305 /* Drop lo_mutex while we call into the filesystem. */
1306 path = lo->lo_backing_file->f_path;
1307 path_get(&path);
1308 mutex_unlock(lock: &lo->lo_mutex);
1309 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1310 if (!ret) {
1311 info->lo_device = huge_encode_dev(dev: stat.dev);
1312 info->lo_inode = stat.ino;
1313 info->lo_rdevice = huge_encode_dev(dev: stat.rdev);
1314 }
1315 path_put(&path);
1316 return ret;
1317}
1318
1319static void
1320loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1321{
1322 memset(s: info64, c: 0, n: sizeof(*info64));
1323 info64->lo_number = info->lo_number;
1324 info64->lo_device = info->lo_device;
1325 info64->lo_inode = info->lo_inode;
1326 info64->lo_rdevice = info->lo_rdevice;
1327 info64->lo_offset = info->lo_offset;
1328 info64->lo_sizelimit = 0;
1329 info64->lo_flags = info->lo_flags;
1330 memcpy(to: info64->lo_file_name, from: info->lo_name, LO_NAME_SIZE);
1331}
1332
1333static int
1334loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1335{
1336 memset(s: info, c: 0, n: sizeof(*info));
1337 info->lo_number = info64->lo_number;
1338 info->lo_device = info64->lo_device;
1339 info->lo_inode = info64->lo_inode;
1340 info->lo_rdevice = info64->lo_rdevice;
1341 info->lo_offset = info64->lo_offset;
1342 info->lo_flags = info64->lo_flags;
1343 memcpy(to: info->lo_name, from: info64->lo_file_name, LO_NAME_SIZE);
1344
1345 /* error in case values were truncated */
1346 if (info->lo_device != info64->lo_device ||
1347 info->lo_rdevice != info64->lo_rdevice ||
1348 info->lo_inode != info64->lo_inode ||
1349 info->lo_offset != info64->lo_offset)
1350 return -EOVERFLOW;
1351
1352 return 0;
1353}
1354
1355static int
1356loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1357{
1358 struct loop_info info;
1359 struct loop_info64 info64;
1360
1361 if (copy_from_user(to: &info, from: arg, n: sizeof (struct loop_info)))
1362 return -EFAULT;
1363 loop_info64_from_old(info: &info, info64: &info64);
1364 return loop_set_status(lo, info: &info64);
1365}
1366
1367static int
1368loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1369{
1370 struct loop_info64 info64;
1371
1372 if (copy_from_user(to: &info64, from: arg, n: sizeof (struct loop_info64)))
1373 return -EFAULT;
1374 return loop_set_status(lo, info: &info64);
1375}
1376
1377static int
1378loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1379 struct loop_info info;
1380 struct loop_info64 info64;
1381 int err;
1382
1383 if (!arg)
1384 return -EINVAL;
1385 err = loop_get_status(lo, info: &info64);
1386 if (!err)
1387 err = loop_info64_to_old(info64: &info64, info: &info);
1388 if (!err && copy_to_user(to: arg, from: &info, n: sizeof(info)))
1389 err = -EFAULT;
1390
1391 return err;
1392}
1393
1394static int
1395loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1396 struct loop_info64 info64;
1397 int err;
1398
1399 if (!arg)
1400 return -EINVAL;
1401 err = loop_get_status(lo, info: &info64);
1402 if (!err && copy_to_user(to: arg, from: &info64, n: sizeof(info64)))
1403 err = -EFAULT;
1404
1405 return err;
1406}
1407
1408static int loop_set_capacity(struct loop_device *lo)
1409{
1410 loff_t size;
1411
1412 if (unlikely(lo->lo_state != Lo_bound))
1413 return -ENXIO;
1414
1415 size = lo_calculate_size(lo, file: lo->lo_backing_file);
1416 loop_set_size(lo, size);
1417
1418 return 0;
1419}
1420
1421static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1422{
1423 bool use_dio = !!arg;
1424 unsigned int memflags;
1425
1426 if (lo->lo_state != Lo_bound)
1427 return -ENXIO;
1428 if (use_dio == !!(lo->lo_flags & LO_FLAGS_DIRECT_IO))
1429 return 0;
1430
1431 if (use_dio) {
1432 if (!lo_can_use_dio(lo))
1433 return -EINVAL;
1434 /* flush dirty pages before starting to use direct I/O */
1435 vfs_fsync(file: lo->lo_backing_file, datasync: 0);
1436 }
1437
1438 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1439 if (use_dio)
1440 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
1441 else
1442 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
1443 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1444 return 0;
1445}
1446
1447static int loop_set_block_size(struct loop_device *lo, blk_mode_t mode,
1448 struct block_device *bdev, unsigned long arg)
1449{
1450 struct queue_limits lim;
1451 unsigned int memflags;
1452 int err = 0;
1453
1454 /*
1455 * If we don't hold exclusive handle for the device, upgrade to it
1456 * here to avoid changing device under exclusive owner.
1457 */
1458 if (!(mode & BLK_OPEN_EXCL)) {
1459 err = bd_prepare_to_claim(bdev, holder: loop_set_block_size, NULL);
1460 if (err)
1461 return err;
1462 }
1463
1464 err = mutex_lock_killable(lock: &lo->lo_mutex);
1465 if (err)
1466 goto abort_claim;
1467
1468 if (lo->lo_state != Lo_bound) {
1469 err = -ENXIO;
1470 goto unlock;
1471 }
1472
1473 if (lo->lo_queue->limits.logical_block_size == arg)
1474 goto unlock;
1475
1476 sync_blockdev(bdev: lo->lo_device);
1477 invalidate_bdev(bdev: lo->lo_device);
1478
1479 lim = queue_limits_start_update(q: lo->lo_queue);
1480 loop_update_limits(lo, lim: &lim, bsize: arg);
1481
1482 memflags = blk_mq_freeze_queue(q: lo->lo_queue);
1483 err = queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
1484 loop_update_dio(lo);
1485 blk_mq_unfreeze_queue(q: lo->lo_queue, memflags);
1486
1487unlock:
1488 mutex_unlock(lock: &lo->lo_mutex);
1489abort_claim:
1490 if (!(mode & BLK_OPEN_EXCL))
1491 bd_abort_claiming(bdev, holder: loop_set_block_size);
1492 return err;
1493}
1494
1495static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1496 unsigned long arg)
1497{
1498 int err;
1499
1500 err = mutex_lock_killable(lock: &lo->lo_mutex);
1501 if (err)
1502 return err;
1503 switch (cmd) {
1504 case LOOP_SET_CAPACITY:
1505 err = loop_set_capacity(lo);
1506 break;
1507 case LOOP_SET_DIRECT_IO:
1508 err = loop_set_dio(lo, arg);
1509 break;
1510 default:
1511 err = -EINVAL;
1512 }
1513 mutex_unlock(lock: &lo->lo_mutex);
1514 return err;
1515}
1516
1517static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1518 unsigned int cmd, unsigned long arg)
1519{
1520 struct loop_device *lo = bdev->bd_disk->private_data;
1521 void __user *argp = (void __user *) arg;
1522 int err;
1523
1524 switch (cmd) {
1525 case LOOP_SET_FD: {
1526 /*
1527 * Legacy case - pass in a zeroed out struct loop_config with
1528 * only the file descriptor set , which corresponds with the
1529 * default parameters we'd have used otherwise.
1530 */
1531 struct loop_config config;
1532
1533 memset(s: &config, c: 0, n: sizeof(config));
1534 config.fd = arg;
1535
1536 return loop_configure(lo, mode, bdev, config: &config);
1537 }
1538 case LOOP_CONFIGURE: {
1539 struct loop_config config;
1540
1541 if (copy_from_user(to: &config, from: argp, n: sizeof(config)))
1542 return -EFAULT;
1543
1544 return loop_configure(lo, mode, bdev, config: &config);
1545 }
1546 case LOOP_CHANGE_FD:
1547 return loop_change_fd(lo, bdev, arg);
1548 case LOOP_CLR_FD:
1549 return loop_clr_fd(lo);
1550 case LOOP_SET_STATUS:
1551 err = -EPERM;
1552 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1553 err = loop_set_status_old(lo, arg: argp);
1554 break;
1555 case LOOP_GET_STATUS:
1556 return loop_get_status_old(lo, arg: argp);
1557 case LOOP_SET_STATUS64:
1558 err = -EPERM;
1559 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1560 err = loop_set_status64(lo, arg: argp);
1561 break;
1562 case LOOP_GET_STATUS64:
1563 return loop_get_status64(lo, arg: argp);
1564 case LOOP_SET_BLOCK_SIZE:
1565 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1566 return -EPERM;
1567 return loop_set_block_size(lo, mode, bdev, arg);
1568 case LOOP_SET_CAPACITY:
1569 case LOOP_SET_DIRECT_IO:
1570 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1571 return -EPERM;
1572 fallthrough;
1573 default:
1574 err = lo_simple_ioctl(lo, cmd, arg);
1575 break;
1576 }
1577
1578 return err;
1579}
1580
1581#ifdef CONFIG_COMPAT
1582struct compat_loop_info {
1583 compat_int_t lo_number; /* ioctl r/o */
1584 compat_dev_t lo_device; /* ioctl r/o */
1585 compat_ulong_t lo_inode; /* ioctl r/o */
1586 compat_dev_t lo_rdevice; /* ioctl r/o */
1587 compat_int_t lo_offset;
1588 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1589 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1590 compat_int_t lo_flags; /* ioctl r/o */
1591 char lo_name[LO_NAME_SIZE];
1592 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1593 compat_ulong_t lo_init[2];
1594 char reserved[4];
1595};
1596
1597/*
1598 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1599 * - noinlined to reduce stack space usage in main part of driver
1600 */
1601static noinline int
1602loop_info64_from_compat(const struct compat_loop_info __user *arg,
1603 struct loop_info64 *info64)
1604{
1605 struct compat_loop_info info;
1606
1607 if (copy_from_user(to: &info, from: arg, n: sizeof(info)))
1608 return -EFAULT;
1609
1610 memset(s: info64, c: 0, n: sizeof(*info64));
1611 info64->lo_number = info.lo_number;
1612 info64->lo_device = info.lo_device;
1613 info64->lo_inode = info.lo_inode;
1614 info64->lo_rdevice = info.lo_rdevice;
1615 info64->lo_offset = info.lo_offset;
1616 info64->lo_sizelimit = 0;
1617 info64->lo_flags = info.lo_flags;
1618 memcpy(to: info64->lo_file_name, from: info.lo_name, LO_NAME_SIZE);
1619 return 0;
1620}
1621
1622/*
1623 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1624 * - noinlined to reduce stack space usage in main part of driver
1625 */
1626static noinline int
1627loop_info64_to_compat(const struct loop_info64 *info64,
1628 struct compat_loop_info __user *arg)
1629{
1630 struct compat_loop_info info;
1631
1632 memset(s: &info, c: 0, n: sizeof(info));
1633 info.lo_number = info64->lo_number;
1634 info.lo_device = info64->lo_device;
1635 info.lo_inode = info64->lo_inode;
1636 info.lo_rdevice = info64->lo_rdevice;
1637 info.lo_offset = info64->lo_offset;
1638 info.lo_flags = info64->lo_flags;
1639 memcpy(to: info.lo_name, from: info64->lo_file_name, LO_NAME_SIZE);
1640
1641 /* error in case values were truncated */
1642 if (info.lo_device != info64->lo_device ||
1643 info.lo_rdevice != info64->lo_rdevice ||
1644 info.lo_inode != info64->lo_inode ||
1645 info.lo_offset != info64->lo_offset)
1646 return -EOVERFLOW;
1647
1648 if (copy_to_user(to: arg, from: &info, n: sizeof(info)))
1649 return -EFAULT;
1650 return 0;
1651}
1652
1653static int
1654loop_set_status_compat(struct loop_device *lo,
1655 const struct compat_loop_info __user *arg)
1656{
1657 struct loop_info64 info64;
1658 int ret;
1659
1660 ret = loop_info64_from_compat(arg, info64: &info64);
1661 if (ret < 0)
1662 return ret;
1663 return loop_set_status(lo, info: &info64);
1664}
1665
1666static int
1667loop_get_status_compat(struct loop_device *lo,
1668 struct compat_loop_info __user *arg)
1669{
1670 struct loop_info64 info64;
1671 int err;
1672
1673 if (!arg)
1674 return -EINVAL;
1675 err = loop_get_status(lo, info: &info64);
1676 if (!err)
1677 err = loop_info64_to_compat(info64: &info64, arg);
1678 return err;
1679}
1680
1681static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1682 unsigned int cmd, unsigned long arg)
1683{
1684 struct loop_device *lo = bdev->bd_disk->private_data;
1685 int err;
1686
1687 switch(cmd) {
1688 case LOOP_SET_STATUS:
1689 err = loop_set_status_compat(lo,
1690 arg: (const struct compat_loop_info __user *)arg);
1691 break;
1692 case LOOP_GET_STATUS:
1693 err = loop_get_status_compat(lo,
1694 arg: (struct compat_loop_info __user *)arg);
1695 break;
1696 case LOOP_SET_CAPACITY:
1697 case LOOP_CLR_FD:
1698 case LOOP_GET_STATUS64:
1699 case LOOP_SET_STATUS64:
1700 case LOOP_CONFIGURE:
1701 arg = (unsigned long) compat_ptr(uptr: arg);
1702 fallthrough;
1703 case LOOP_SET_FD:
1704 case LOOP_CHANGE_FD:
1705 case LOOP_SET_BLOCK_SIZE:
1706 case LOOP_SET_DIRECT_IO:
1707 err = lo_ioctl(bdev, mode, cmd, arg);
1708 break;
1709 default:
1710 err = -ENOIOCTLCMD;
1711 break;
1712 }
1713 return err;
1714}
1715#endif
1716
1717static int lo_open(struct gendisk *disk, blk_mode_t mode)
1718{
1719 struct loop_device *lo = disk->private_data;
1720 int err;
1721
1722 err = mutex_lock_killable(lock: &lo->lo_mutex);
1723 if (err)
1724 return err;
1725
1726 if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown)
1727 err = -ENXIO;
1728 mutex_unlock(lock: &lo->lo_mutex);
1729 return err;
1730}
1731
1732static void lo_release(struct gendisk *disk)
1733{
1734 struct loop_device *lo = disk->private_data;
1735 bool need_clear = false;
1736
1737 if (disk_openers(disk) > 0)
1738 return;
1739 /*
1740 * Clear the backing device information if this is the last close of
1741 * a device that's been marked for auto clear, or on which LOOP_CLR_FD
1742 * has been called.
1743 */
1744
1745 mutex_lock(lock: &lo->lo_mutex);
1746 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR))
1747 lo->lo_state = Lo_rundown;
1748
1749 need_clear = (lo->lo_state == Lo_rundown);
1750 mutex_unlock(lock: &lo->lo_mutex);
1751
1752 if (need_clear)
1753 __loop_clr_fd(lo);
1754}
1755
1756static void lo_free_disk(struct gendisk *disk)
1757{
1758 struct loop_device *lo = disk->private_data;
1759
1760 if (lo->workqueue)
1761 destroy_workqueue(wq: lo->workqueue);
1762 loop_free_idle_workers(lo, delete_all: true);
1763 timer_shutdown_sync(timer: &lo->timer);
1764 mutex_destroy(lock: &lo->lo_mutex);
1765 kfree(objp: lo);
1766}
1767
1768static const struct block_device_operations lo_fops = {
1769 .owner = THIS_MODULE,
1770 .open = lo_open,
1771 .release = lo_release,
1772 .ioctl = lo_ioctl,
1773#ifdef CONFIG_COMPAT
1774 .compat_ioctl = lo_compat_ioctl,
1775#endif
1776 .free_disk = lo_free_disk,
1777};
1778
1779/*
1780 * And now the modules code and kernel interface.
1781 */
1782
1783/*
1784 * If max_loop is specified, create that many devices upfront.
1785 * This also becomes a hard limit. If max_loop is not specified,
1786 * the default isn't a hard limit (as before commit 85c50197716c
1787 * changed the default value from 0 for max_loop=0 reasons), just
1788 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1789 * init time. Loop devices can be requested on-demand with the
1790 * /dev/loop-control interface, or be instantiated by accessing
1791 * a 'dead' device node.
1792 */
1793static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1794
1795#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1796static bool max_loop_specified;
1797
1798static int max_loop_param_set_int(const char *val,
1799 const struct kernel_param *kp)
1800{
1801 int ret;
1802
1803 ret = param_set_int(val, kp);
1804 if (ret < 0)
1805 return ret;
1806
1807 max_loop_specified = true;
1808 return 0;
1809}
1810
1811static const struct kernel_param_ops max_loop_param_ops = {
1812 .set = max_loop_param_set_int,
1813 .get = param_get_int,
1814};
1815
1816module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1817MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1818#else
1819module_param(max_loop, int, 0444);
1820MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1821#endif
1822
1823module_param(max_part, int, 0444);
1824MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1825
1826static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1827
1828static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1829{
1830 int qd, ret;
1831
1832 ret = kstrtoint(s, base: 0, res: &qd);
1833 if (ret < 0)
1834 return ret;
1835 if (qd < 1)
1836 return -EINVAL;
1837 hw_queue_depth = qd;
1838 return 0;
1839}
1840
1841static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1842 .set = loop_set_hw_queue_depth,
1843 .get = param_get_int,
1844};
1845
1846device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1847MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1848
1849MODULE_DESCRIPTION("Loopback device support");
1850MODULE_LICENSE("GPL");
1851MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1852
1853static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1854 const struct blk_mq_queue_data *bd)
1855{
1856 struct request *rq = bd->rq;
1857 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1858 struct loop_device *lo = rq->q->queuedata;
1859
1860 blk_mq_start_request(rq);
1861
1862 if (lo->lo_state != Lo_bound)
1863 return BLK_STS_IOERR;
1864
1865 switch (req_op(req: rq)) {
1866 case REQ_OP_FLUSH:
1867 case REQ_OP_DISCARD:
1868 case REQ_OP_WRITE_ZEROES:
1869 cmd->use_aio = false;
1870 break;
1871 default:
1872 cmd->use_aio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1873 break;
1874 }
1875
1876 /* always use the first bio's css */
1877 cmd->blkcg_css = NULL;
1878 cmd->memcg_css = NULL;
1879#ifdef CONFIG_BLK_CGROUP
1880 if (rq->bio) {
1881 cmd->blkcg_css = bio_blkcg_css(bio: rq->bio);
1882#ifdef CONFIG_MEMCG
1883 if (cmd->blkcg_css) {
1884 cmd->memcg_css =
1885 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1886 &memory_cgrp_subsys);
1887 }
1888#endif
1889 }
1890#endif
1891 loop_queue_work(lo, cmd);
1892
1893 return BLK_STS_OK;
1894}
1895
1896static void loop_handle_cmd(struct loop_cmd *cmd)
1897{
1898 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1899 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1900 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
1901 const bool write = op_is_write(op: req_op(req: rq));
1902 struct loop_device *lo = rq->q->queuedata;
1903 int ret = 0;
1904 struct mem_cgroup *old_memcg = NULL;
1905
1906 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1907 ret = -EIO;
1908 goto failed;
1909 }
1910
1911 if (cmd_blkcg_css)
1912 kthread_associate_blkcg(css: cmd_blkcg_css);
1913 if (cmd_memcg_css)
1914 old_memcg = set_active_memcg(
1915 mem_cgroup_from_css(css: cmd_memcg_css));
1916
1917 /*
1918 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1919 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1920 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1921 * not yet been completed.
1922 */
1923 ret = do_req_filebacked(lo, rq);
1924
1925 if (cmd_blkcg_css)
1926 kthread_associate_blkcg(NULL);
1927
1928 if (cmd_memcg_css) {
1929 set_active_memcg(old_memcg);
1930 css_put(css: cmd_memcg_css);
1931 }
1932 failed:
1933 /* complete non-aio request */
1934 if (ret != -EIOCBQUEUED) {
1935 if (ret == -EOPNOTSUPP)
1936 cmd->ret = ret;
1937 else
1938 cmd->ret = ret ? -EIO : 0;
1939 if (likely(!blk_should_fake_timeout(rq->q)))
1940 blk_mq_complete_request(rq);
1941 }
1942}
1943
1944static void loop_process_work(struct loop_worker *worker,
1945 struct list_head *cmd_list, struct loop_device *lo)
1946{
1947 int orig_flags = current->flags;
1948 struct loop_cmd *cmd;
1949
1950 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1951 spin_lock_irq(lock: &lo->lo_work_lock);
1952 while (!list_empty(head: cmd_list)) {
1953 cmd = container_of(
1954 cmd_list->next, struct loop_cmd, list_entry);
1955 list_del(entry: cmd_list->next);
1956 spin_unlock_irq(lock: &lo->lo_work_lock);
1957
1958 loop_handle_cmd(cmd);
1959 cond_resched();
1960
1961 spin_lock_irq(lock: &lo->lo_work_lock);
1962 }
1963
1964 /*
1965 * We only add to the idle list if there are no pending cmds
1966 * *and* the worker will not run again which ensures that it
1967 * is safe to free any worker on the idle list
1968 */
1969 if (worker && !work_pending(&worker->work)) {
1970 worker->last_ran_at = jiffies;
1971 list_add_tail(new: &worker->idle_list, head: &lo->idle_worker_list);
1972 loop_set_timer(lo);
1973 }
1974 spin_unlock_irq(lock: &lo->lo_work_lock);
1975 current->flags = orig_flags;
1976}
1977
1978static void loop_workfn(struct work_struct *work)
1979{
1980 struct loop_worker *worker =
1981 container_of(work, struct loop_worker, work);
1982 loop_process_work(worker, cmd_list: &worker->cmd_list, lo: worker->lo);
1983}
1984
1985static void loop_rootcg_workfn(struct work_struct *work)
1986{
1987 struct loop_device *lo =
1988 container_of(work, struct loop_device, rootcg_work);
1989 loop_process_work(NULL, cmd_list: &lo->rootcg_cmd_list, lo);
1990}
1991
1992static const struct blk_mq_ops loop_mq_ops = {
1993 .queue_rq = loop_queue_rq,
1994 .complete = lo_complete_rq,
1995};
1996
1997static int loop_add(int i)
1998{
1999 struct queue_limits lim = {
2000 /*
2001 * Random number picked from the historic block max_sectors cap.
2002 */
2003 .max_hw_sectors = 2560u,
2004 };
2005 struct loop_device *lo;
2006 struct gendisk *disk;
2007 int err;
2008
2009 err = -ENOMEM;
2010 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2011 if (!lo)
2012 goto out;
2013 lo->worker_tree = RB_ROOT;
2014 INIT_LIST_HEAD(list: &lo->idle_worker_list);
2015 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
2016 lo->lo_state = Lo_unbound;
2017
2018 err = mutex_lock_killable(lock: &loop_ctl_mutex);
2019 if (err)
2020 goto out_free_dev;
2021
2022 /* allocate id, if @id >= 0, we're requesting that specific id */
2023 if (i >= 0) {
2024 err = idr_alloc(&loop_index_idr, ptr: lo, start: i, end: i + 1, GFP_KERNEL);
2025 if (err == -ENOSPC)
2026 err = -EEXIST;
2027 } else {
2028 err = idr_alloc(&loop_index_idr, ptr: lo, start: 0, end: 0, GFP_KERNEL);
2029 }
2030 mutex_unlock(lock: &loop_ctl_mutex);
2031 if (err < 0)
2032 goto out_free_dev;
2033 i = err;
2034
2035 lo->tag_set.ops = &loop_mq_ops;
2036 lo->tag_set.nr_hw_queues = 1;
2037 lo->tag_set.queue_depth = hw_queue_depth;
2038 lo->tag_set.numa_node = NUMA_NO_NODE;
2039 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2040 lo->tag_set.flags = BLK_MQ_F_STACKING | BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2041 lo->tag_set.driver_data = lo;
2042
2043 err = blk_mq_alloc_tag_set(set: &lo->tag_set);
2044 if (err)
2045 goto out_free_idr;
2046
2047 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2048 if (IS_ERR(ptr: disk)) {
2049 err = PTR_ERR(ptr: disk);
2050 goto out_cleanup_tags;
2051 }
2052 lo->lo_queue = lo->lo_disk->queue;
2053
2054 /*
2055 * Disable partition scanning by default. The in-kernel partition
2056 * scanning can be requested individually per-device during its
2057 * setup. Userspace can always add and remove partitions from all
2058 * devices. The needed partition minors are allocated from the
2059 * extended minor space, the main loop device numbers will continue
2060 * to match the loop minors, regardless of the number of partitions
2061 * used.
2062 *
2063 * If max_part is given, partition scanning is globally enabled for
2064 * all loop devices. The minors for the main loop devices will be
2065 * multiples of max_part.
2066 *
2067 * Note: Global-for-all-devices, set-only-at-init, read-only module
2068 * parameteters like 'max_loop' and 'max_part' make things needlessly
2069 * complicated, are too static, inflexible and may surprise
2070 * userspace tools. Parameters like this in general should be avoided.
2071 */
2072 if (!part_shift)
2073 set_bit(GD_SUPPRESS_PART_SCAN, addr: &disk->state);
2074 mutex_init(&lo->lo_mutex);
2075 lo->lo_number = i;
2076 spin_lock_init(&lo->lo_lock);
2077 spin_lock_init(&lo->lo_work_lock);
2078 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2079 INIT_LIST_HEAD(list: &lo->rootcg_cmd_list);
2080 disk->major = LOOP_MAJOR;
2081 disk->first_minor = i << part_shift;
2082 disk->minors = 1 << part_shift;
2083 disk->fops = &lo_fops;
2084 disk->private_data = lo;
2085 disk->queue = lo->lo_queue;
2086 disk->events = DISK_EVENT_MEDIA_CHANGE;
2087 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2088 sprintf(buf: disk->disk_name, fmt: "loop%d", i);
2089 /* Make this loop device reachable from pathname. */
2090 err = add_disk(disk);
2091 if (err)
2092 goto out_cleanup_disk;
2093
2094 /* Show this loop device. */
2095 mutex_lock(lock: &loop_ctl_mutex);
2096 lo->idr_visible = true;
2097 mutex_unlock(lock: &loop_ctl_mutex);
2098
2099 return i;
2100
2101out_cleanup_disk:
2102 put_disk(disk);
2103out_cleanup_tags:
2104 blk_mq_free_tag_set(set: &lo->tag_set);
2105out_free_idr:
2106 mutex_lock(lock: &loop_ctl_mutex);
2107 idr_remove(&loop_index_idr, id: i);
2108 mutex_unlock(lock: &loop_ctl_mutex);
2109out_free_dev:
2110 kfree(objp: lo);
2111out:
2112 return err;
2113}
2114
2115static void loop_remove(struct loop_device *lo)
2116{
2117 /* Make this loop device unreachable from pathname. */
2118 del_gendisk(gp: lo->lo_disk);
2119 blk_mq_free_tag_set(set: &lo->tag_set);
2120
2121 mutex_lock(lock: &loop_ctl_mutex);
2122 idr_remove(&loop_index_idr, id: lo->lo_number);
2123 mutex_unlock(lock: &loop_ctl_mutex);
2124
2125 put_disk(disk: lo->lo_disk);
2126}
2127
2128#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2129static void loop_probe(dev_t dev)
2130{
2131 int idx = MINOR(dev) >> part_shift;
2132
2133 if (max_loop_specified && max_loop && idx >= max_loop)
2134 return;
2135 loop_add(i: idx);
2136}
2137#else
2138#define loop_probe NULL
2139#endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2140
2141static int loop_control_remove(int idx)
2142{
2143 struct loop_device *lo;
2144 int ret;
2145
2146 if (idx < 0) {
2147 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2148 return -EINVAL;
2149 }
2150
2151 /* Hide this loop device for serialization. */
2152 ret = mutex_lock_killable(lock: &loop_ctl_mutex);
2153 if (ret)
2154 return ret;
2155 lo = idr_find(&loop_index_idr, id: idx);
2156 if (!lo || !lo->idr_visible)
2157 ret = -ENODEV;
2158 else
2159 lo->idr_visible = false;
2160 mutex_unlock(lock: &loop_ctl_mutex);
2161 if (ret)
2162 return ret;
2163
2164 /* Check whether this loop device can be removed. */
2165 ret = mutex_lock_killable(lock: &lo->lo_mutex);
2166 if (ret)
2167 goto mark_visible;
2168 if (lo->lo_state != Lo_unbound || disk_openers(disk: lo->lo_disk) > 0) {
2169 mutex_unlock(lock: &lo->lo_mutex);
2170 ret = -EBUSY;
2171 goto mark_visible;
2172 }
2173 /* Mark this loop device as no more bound, but not quite unbound yet */
2174 lo->lo_state = Lo_deleting;
2175 mutex_unlock(lock: &lo->lo_mutex);
2176
2177 loop_remove(lo);
2178 return 0;
2179
2180mark_visible:
2181 /* Show this loop device again. */
2182 mutex_lock(lock: &loop_ctl_mutex);
2183 lo->idr_visible = true;
2184 mutex_unlock(lock: &loop_ctl_mutex);
2185 return ret;
2186}
2187
2188static int loop_control_get_free(int idx)
2189{
2190 struct loop_device *lo;
2191 int id, ret;
2192
2193 ret = mutex_lock_killable(lock: &loop_ctl_mutex);
2194 if (ret)
2195 return ret;
2196 idr_for_each_entry(&loop_index_idr, lo, id) {
2197 /* Hitting a race results in creating a new loop device which is harmless. */
2198 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2199 goto found;
2200 }
2201 mutex_unlock(lock: &loop_ctl_mutex);
2202 return loop_add(i: -1);
2203found:
2204 mutex_unlock(lock: &loop_ctl_mutex);
2205 return id;
2206}
2207
2208static long loop_control_ioctl(struct file *file, unsigned int cmd,
2209 unsigned long parm)
2210{
2211 switch (cmd) {
2212 case LOOP_CTL_ADD:
2213 return loop_add(i: parm);
2214 case LOOP_CTL_REMOVE:
2215 return loop_control_remove(idx: parm);
2216 case LOOP_CTL_GET_FREE:
2217 return loop_control_get_free(idx: parm);
2218 default:
2219 return -ENOSYS;
2220 }
2221}
2222
2223static const struct file_operations loop_ctl_fops = {
2224 .open = nonseekable_open,
2225 .unlocked_ioctl = loop_control_ioctl,
2226 .compat_ioctl = loop_control_ioctl,
2227 .owner = THIS_MODULE,
2228 .llseek = noop_llseek,
2229};
2230
2231static struct miscdevice loop_misc = {
2232 .minor = LOOP_CTRL_MINOR,
2233 .name = "loop-control",
2234 .fops = &loop_ctl_fops,
2235};
2236
2237MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2238MODULE_ALIAS("devname:loop-control");
2239
2240static int __init loop_init(void)
2241{
2242 int i;
2243 int err;
2244
2245 part_shift = 0;
2246 if (max_part > 0) {
2247 part_shift = fls(x: max_part);
2248
2249 /*
2250 * Adjust max_part according to part_shift as it is exported
2251 * to user space so that user can decide correct minor number
2252 * if [s]he want to create more devices.
2253 *
2254 * Note that -1 is required because partition 0 is reserved
2255 * for the whole disk.
2256 */
2257 max_part = (1UL << part_shift) - 1;
2258 }
2259
2260 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2261 err = -EINVAL;
2262 goto err_out;
2263 }
2264
2265 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2266 err = -EINVAL;
2267 goto err_out;
2268 }
2269
2270 err = misc_register(misc: &loop_misc);
2271 if (err < 0)
2272 goto err_out;
2273
2274
2275 if (__register_blkdev(LOOP_MAJOR, name: "loop", probe: loop_probe)) {
2276 err = -EIO;
2277 goto misc_out;
2278 }
2279
2280 /* pre-create number of devices given by config or max_loop */
2281 for (i = 0; i < max_loop; i++)
2282 loop_add(i);
2283
2284 printk(KERN_INFO "loop: module loaded\n");
2285 return 0;
2286
2287misc_out:
2288 misc_deregister(misc: &loop_misc);
2289err_out:
2290 return err;
2291}
2292
2293static void __exit loop_exit(void)
2294{
2295 struct loop_device *lo;
2296 int id;
2297
2298 unregister_blkdev(LOOP_MAJOR, name: "loop");
2299 misc_deregister(misc: &loop_misc);
2300
2301 /*
2302 * There is no need to use loop_ctl_mutex here, for nobody else can
2303 * access loop_index_idr when this module is unloading (unless forced
2304 * module unloading is requested). If this is not a clean unloading,
2305 * we have no means to avoid kernel crash.
2306 */
2307 idr_for_each_entry(&loop_index_idr, lo, id)
2308 loop_remove(lo);
2309
2310 idr_destroy(&loop_index_idr);
2311}
2312
2313module_init(loop_init);
2314module_exit(loop_exit);
2315
2316#ifndef MODULE
2317static int __init max_loop_setup(char *str)
2318{
2319 max_loop = simple_strtol(str, NULL, 0);
2320#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2321 max_loop_specified = true;
2322#endif
2323 return 1;
2324}
2325
2326__setup("max_loop=", max_loop_setup);
2327#endif
2328