1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 1993 by Theodore Ts'o.
4 */
5#include <linux/module.h>
6#include <linux/moduleparam.h>
7#include <linux/sched.h>
8#include <linux/fs.h>
9#include <linux/pagemap.h>
10#include <linux/file.h>
11#include <linux/stat.h>
12#include <linux/errno.h>
13#include <linux/major.h>
14#include <linux/wait.h>
15#include <linux/blkpg.h>
16#include <linux/init.h>
17#include <linux/swap.h>
18#include <linux/slab.h>
19#include <linux/compat.h>
20#include <linux/suspend.h>
21#include <linux/freezer.h>
22#include <linux/mutex.h>
23#include <linux/writeback.h>
24#include <linux/completion.h>
25#include <linux/highmem.h>
26#include <linux/splice.h>
27#include <linux/sysfs.h>
28#include <linux/miscdevice.h>
29#include <linux/falloc.h>
30#include <linux/uio.h>
31#include <linux/ioprio.h>
32#include <linux/blk-cgroup.h>
33#include <linux/sched/mm.h>
34#include <linux/statfs.h>
35#include <linux/uaccess.h>
36#include <linux/blk-mq.h>
37#include <linux/spinlock.h>
38#include <uapi/linux/loop.h>
39
40/* Possible states of device */
41enum {
42 Lo_unbound,
43 Lo_bound,
44 Lo_rundown,
45 Lo_deleting,
46};
47
48struct loop_func_table;
49
50struct loop_device {
51 int lo_number;
52 loff_t lo_offset;
53 loff_t lo_sizelimit;
54 int lo_flags;
55 char lo_file_name[LO_NAME_SIZE];
56
57 struct file * lo_backing_file;
58 struct block_device *lo_device;
59
60 gfp_t old_gfp_mask;
61
62 spinlock_t lo_lock;
63 int lo_state;
64 spinlock_t lo_work_lock;
65 struct workqueue_struct *workqueue;
66 struct work_struct rootcg_work;
67 struct list_head rootcg_cmd_list;
68 struct list_head idle_worker_list;
69 struct rb_root worker_tree;
70 struct timer_list timer;
71 bool use_dio;
72 bool sysfs_inited;
73
74 struct request_queue *lo_queue;
75 struct blk_mq_tag_set tag_set;
76 struct gendisk *lo_disk;
77 struct mutex lo_mutex;
78 bool idr_visible;
79};
80
81struct loop_cmd {
82 struct list_head list_entry;
83 bool use_aio; /* use AIO interface to handle I/O */
84 atomic_t ref; /* only for aio */
85 long ret;
86 struct kiocb iocb;
87 struct bio_vec *bvec;
88 struct cgroup_subsys_state *blkcg_css;
89 struct cgroup_subsys_state *memcg_css;
90};
91
92#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
93#define LOOP_DEFAULT_HW_Q_DEPTH 128
94
95static DEFINE_IDR(loop_index_idr);
96static DEFINE_MUTEX(loop_ctl_mutex);
97static DEFINE_MUTEX(loop_validate_mutex);
98
99/**
100 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
101 *
102 * @lo: struct loop_device
103 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
104 *
105 * Returns 0 on success, -EINTR otherwise.
106 *
107 * Since loop_validate_file() traverses on other "struct loop_device" if
108 * is_loop_device() is true, we need a global lock for serializing concurrent
109 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
110 */
111static int loop_global_lock_killable(struct loop_device *lo, bool global)
112{
113 int err;
114
115 if (global) {
116 err = mutex_lock_killable(&loop_validate_mutex);
117 if (err)
118 return err;
119 }
120 err = mutex_lock_killable(&lo->lo_mutex);
121 if (err && global)
122 mutex_unlock(lock: &loop_validate_mutex);
123 return err;
124}
125
126/**
127 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
128 *
129 * @lo: struct loop_device
130 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
131 */
132static void loop_global_unlock(struct loop_device *lo, bool global)
133{
134 mutex_unlock(lock: &lo->lo_mutex);
135 if (global)
136 mutex_unlock(lock: &loop_validate_mutex);
137}
138
139static int max_part;
140static int part_shift;
141
142static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
143{
144 loff_t loopsize;
145
146 /* Compute loopsize in bytes */
147 loopsize = i_size_read(inode: file->f_mapping->host);
148 if (offset > 0)
149 loopsize -= offset;
150 /* offset is beyond i_size, weird but possible */
151 if (loopsize < 0)
152 return 0;
153
154 if (sizelimit > 0 && sizelimit < loopsize)
155 loopsize = sizelimit;
156 /*
157 * Unfortunately, if we want to do I/O on the device,
158 * the number of 512-byte sectors has to fit into a sector_t.
159 */
160 return loopsize >> 9;
161}
162
163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164{
165 return get_size(offset: lo->lo_offset, sizelimit: lo->lo_sizelimit, file);
166}
167
168/*
169 * We support direct I/O only if lo_offset is aligned with the logical I/O size
170 * of backing device, and the logical block size of loop is bigger than that of
171 * the backing device.
172 */
173static bool lo_bdev_can_use_dio(struct loop_device *lo,
174 struct block_device *backing_bdev)
175{
176 unsigned short sb_bsize = bdev_logical_block_size(bdev: backing_bdev);
177
178 if (queue_logical_block_size(q: lo->lo_queue) < sb_bsize)
179 return false;
180 if (lo->lo_offset & (sb_bsize - 1))
181 return false;
182 return true;
183}
184
185static void __loop_update_dio(struct loop_device *lo, bool dio)
186{
187 struct file *file = lo->lo_backing_file;
188 struct inode *inode = file->f_mapping->host;
189 struct block_device *backing_bdev = NULL;
190 bool use_dio;
191
192 if (S_ISBLK(inode->i_mode))
193 backing_bdev = I_BDEV(inode);
194 else if (inode->i_sb->s_bdev)
195 backing_bdev = inode->i_sb->s_bdev;
196
197 use_dio = dio && (file->f_mode & FMODE_CAN_ODIRECT) &&
198 (!backing_bdev || lo_bdev_can_use_dio(lo, backing_bdev));
199
200 if (lo->use_dio == use_dio)
201 return;
202
203 /* flush dirty pages before changing direct IO */
204 vfs_fsync(file, datasync: 0);
205
206 /*
207 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
208 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
209 * will get updated by ioctl(LOOP_GET_STATUS)
210 */
211 if (lo->lo_state == Lo_bound)
212 blk_mq_freeze_queue(q: lo->lo_queue);
213 lo->use_dio = use_dio;
214 if (use_dio) {
215 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q: lo->lo_queue);
216 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
217 } else {
218 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q: lo->lo_queue);
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 }
221 if (lo->lo_state == Lo_bound)
222 blk_mq_unfreeze_queue(q: lo->lo_queue);
223}
224
225/**
226 * loop_set_size() - sets device size and notifies userspace
227 * @lo: struct loop_device to set the size for
228 * @size: new size of the loop device
229 *
230 * Callers must validate that the size passed into this function fits into
231 * a sector_t, eg using loop_validate_size()
232 */
233static void loop_set_size(struct loop_device *lo, loff_t size)
234{
235 if (!set_capacity_and_notify(disk: lo->lo_disk, size))
236 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
237}
238
239static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
240{
241 struct iov_iter i;
242 ssize_t bw;
243
244 iov_iter_bvec(i: &i, ITER_SOURCE, bvec, nr_segs: 1, count: bvec->bv_len);
245
246 bw = vfs_iter_write(file, iter: &i, ppos, flags: 0);
247
248 if (likely(bw == bvec->bv_len))
249 return 0;
250
251 printk_ratelimited(KERN_ERR
252 "loop: Write error at byte offset %llu, length %i.\n",
253 (unsigned long long)*ppos, bvec->bv_len);
254 if (bw >= 0)
255 bw = -EIO;
256 return bw;
257}
258
259static int lo_write_simple(struct loop_device *lo, struct request *rq,
260 loff_t pos)
261{
262 struct bio_vec bvec;
263 struct req_iterator iter;
264 int ret = 0;
265
266 rq_for_each_segment(bvec, rq, iter) {
267 ret = lo_write_bvec(file: lo->lo_backing_file, bvec: &bvec, ppos: &pos);
268 if (ret < 0)
269 break;
270 cond_resched();
271 }
272
273 return ret;
274}
275
276static int lo_read_simple(struct loop_device *lo, struct request *rq,
277 loff_t pos)
278{
279 struct bio_vec bvec;
280 struct req_iterator iter;
281 struct iov_iter i;
282 ssize_t len;
283
284 rq_for_each_segment(bvec, rq, iter) {
285 iov_iter_bvec(i: &i, ITER_DEST, bvec: &bvec, nr_segs: 1, count: bvec.bv_len);
286 len = vfs_iter_read(file: lo->lo_backing_file, iter: &i, ppos: &pos, flags: 0);
287 if (len < 0)
288 return len;
289
290 flush_dcache_page(page: bvec.bv_page);
291
292 if (len != bvec.bv_len) {
293 struct bio *bio;
294
295 __rq_for_each_bio(bio, rq)
296 zero_fill_bio(bio);
297 break;
298 }
299 cond_resched();
300 }
301
302 return 0;
303}
304
305static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
306 int mode)
307{
308 /*
309 * We use fallocate to manipulate the space mappings used by the image
310 * a.k.a. discard/zerorange.
311 */
312 struct file *file = lo->lo_backing_file;
313 int ret;
314
315 mode |= FALLOC_FL_KEEP_SIZE;
316
317 if (!bdev_max_discard_sectors(bdev: lo->lo_device))
318 return -EOPNOTSUPP;
319
320 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
321 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
322 return -EIO;
323 return ret;
324}
325
326static int lo_req_flush(struct loop_device *lo, struct request *rq)
327{
328 int ret = vfs_fsync(file: lo->lo_backing_file, datasync: 0);
329 if (unlikely(ret && ret != -EINVAL))
330 ret = -EIO;
331
332 return ret;
333}
334
335static void lo_complete_rq(struct request *rq)
336{
337 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
338 blk_status_t ret = BLK_STS_OK;
339
340 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
341 req_op(req: rq) != REQ_OP_READ) {
342 if (cmd->ret < 0)
343 ret = errno_to_blk_status(errno: cmd->ret);
344 goto end_io;
345 }
346
347 /*
348 * Short READ - if we got some data, advance our request and
349 * retry it. If we got no data, end the rest with EIO.
350 */
351 if (cmd->ret) {
352 blk_update_request(rq, BLK_STS_OK, nr_bytes: cmd->ret);
353 cmd->ret = 0;
354 blk_mq_requeue_request(rq, kick_requeue_list: true);
355 } else {
356 if (cmd->use_aio) {
357 struct bio *bio = rq->bio;
358
359 while (bio) {
360 zero_fill_bio(bio);
361 bio = bio->bi_next;
362 }
363 }
364 ret = BLK_STS_IOERR;
365end_io:
366 blk_mq_end_request(rq, error: ret);
367 }
368}
369
370static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
371{
372 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
373
374 if (!atomic_dec_and_test(v: &cmd->ref))
375 return;
376 kfree(objp: cmd->bvec);
377 cmd->bvec = NULL;
378 if (likely(!blk_should_fake_timeout(rq->q)))
379 blk_mq_complete_request(rq);
380}
381
382static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
383{
384 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
385
386 cmd->ret = ret;
387 lo_rw_aio_do_completion(cmd);
388}
389
390static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
391 loff_t pos, int rw)
392{
393 struct iov_iter iter;
394 struct req_iterator rq_iter;
395 struct bio_vec *bvec;
396 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
397 struct bio *bio = rq->bio;
398 struct file *file = lo->lo_backing_file;
399 struct bio_vec tmp;
400 unsigned int offset;
401 int nr_bvec = 0;
402 int ret;
403
404 rq_for_each_bvec(tmp, rq, rq_iter)
405 nr_bvec++;
406
407 if (rq->bio != rq->biotail) {
408
409 bvec = kmalloc_array(n: nr_bvec, size: sizeof(struct bio_vec),
410 GFP_NOIO);
411 if (!bvec)
412 return -EIO;
413 cmd->bvec = bvec;
414
415 /*
416 * The bios of the request may be started from the middle of
417 * the 'bvec' because of bio splitting, so we can't directly
418 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
419 * API will take care of all details for us.
420 */
421 rq_for_each_bvec(tmp, rq, rq_iter) {
422 *bvec = tmp;
423 bvec++;
424 }
425 bvec = cmd->bvec;
426 offset = 0;
427 } else {
428 /*
429 * Same here, this bio may be started from the middle of the
430 * 'bvec' because of bio splitting, so offset from the bvec
431 * must be passed to iov iterator
432 */
433 offset = bio->bi_iter.bi_bvec_done;
434 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
435 }
436 atomic_set(v: &cmd->ref, i: 2);
437
438 iov_iter_bvec(i: &iter, direction: rw, bvec, nr_segs: nr_bvec, count: blk_rq_bytes(rq));
439 iter.iov_offset = offset;
440
441 cmd->iocb.ki_pos = pos;
442 cmd->iocb.ki_filp = file;
443 cmd->iocb.ki_complete = lo_rw_aio_complete;
444 cmd->iocb.ki_flags = IOCB_DIRECT;
445 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
446
447 if (rw == ITER_SOURCE)
448 ret = call_write_iter(file, kio: &cmd->iocb, iter: &iter);
449 else
450 ret = call_read_iter(file, kio: &cmd->iocb, iter: &iter);
451
452 lo_rw_aio_do_completion(cmd);
453
454 if (ret != -EIOCBQUEUED)
455 lo_rw_aio_complete(iocb: &cmd->iocb, ret);
456 return 0;
457}
458
459static int do_req_filebacked(struct loop_device *lo, struct request *rq)
460{
461 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
462 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
463
464 /*
465 * lo_write_simple and lo_read_simple should have been covered
466 * by io submit style function like lo_rw_aio(), one blocker
467 * is that lo_read_simple() need to call flush_dcache_page after
468 * the page is written from kernel, and it isn't easy to handle
469 * this in io submit style function which submits all segments
470 * of the req at one time. And direct read IO doesn't need to
471 * run flush_dcache_page().
472 */
473 switch (req_op(req: rq)) {
474 case REQ_OP_FLUSH:
475 return lo_req_flush(lo, rq);
476 case REQ_OP_WRITE_ZEROES:
477 /*
478 * If the caller doesn't want deallocation, call zeroout to
479 * write zeroes the range. Otherwise, punch them out.
480 */
481 return lo_fallocate(lo, rq, pos,
482 mode: (rq->cmd_flags & REQ_NOUNMAP) ?
483 FALLOC_FL_ZERO_RANGE :
484 FALLOC_FL_PUNCH_HOLE);
485 case REQ_OP_DISCARD:
486 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
487 case REQ_OP_WRITE:
488 if (cmd->use_aio)
489 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
490 else
491 return lo_write_simple(lo, rq, pos);
492 case REQ_OP_READ:
493 if (cmd->use_aio)
494 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
495 else
496 return lo_read_simple(lo, rq, pos);
497 default:
498 WARN_ON_ONCE(1);
499 return -EIO;
500 }
501}
502
503static inline void loop_update_dio(struct loop_device *lo)
504{
505 __loop_update_dio(lo, dio: (lo->lo_backing_file->f_flags & O_DIRECT) |
506 lo->use_dio);
507}
508
509static void loop_reread_partitions(struct loop_device *lo)
510{
511 int rc;
512
513 mutex_lock(&lo->lo_disk->open_mutex);
514 rc = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
515 mutex_unlock(lock: &lo->lo_disk->open_mutex);
516 if (rc)
517 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
518 __func__, lo->lo_number, lo->lo_file_name, rc);
519}
520
521static inline int is_loop_device(struct file *file)
522{
523 struct inode *i = file->f_mapping->host;
524
525 return i && S_ISBLK(i->i_mode) && imajor(inode: i) == LOOP_MAJOR;
526}
527
528static int loop_validate_file(struct file *file, struct block_device *bdev)
529{
530 struct inode *inode = file->f_mapping->host;
531 struct file *f = file;
532
533 /* Avoid recursion */
534 while (is_loop_device(file: f)) {
535 struct loop_device *l;
536
537 lockdep_assert_held(&loop_validate_mutex);
538 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
539 return -EBADF;
540
541 l = I_BDEV(inode: f->f_mapping->host)->bd_disk->private_data;
542 if (l->lo_state != Lo_bound)
543 return -EINVAL;
544 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
545 rmb();
546 f = l->lo_backing_file;
547 }
548 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
549 return -EINVAL;
550 return 0;
551}
552
553/*
554 * loop_change_fd switched the backing store of a loopback device to
555 * a new file. This is useful for operating system installers to free up
556 * the original file and in High Availability environments to switch to
557 * an alternative location for the content in case of server meltdown.
558 * This can only work if the loop device is used read-only, and if the
559 * new backing store is the same size and type as the old backing store.
560 */
561static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
562 unsigned int arg)
563{
564 struct file *file = fget(fd: arg);
565 struct file *old_file;
566 int error;
567 bool partscan;
568 bool is_loop;
569
570 if (!file)
571 return -EBADF;
572
573 /* suppress uevents while reconfiguring the device */
574 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
575
576 is_loop = is_loop_device(file);
577 error = loop_global_lock_killable(lo, global: is_loop);
578 if (error)
579 goto out_putf;
580 error = -ENXIO;
581 if (lo->lo_state != Lo_bound)
582 goto out_err;
583
584 /* the loop device has to be read-only */
585 error = -EINVAL;
586 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
587 goto out_err;
588
589 error = loop_validate_file(file, bdev);
590 if (error)
591 goto out_err;
592
593 old_file = lo->lo_backing_file;
594
595 error = -EINVAL;
596
597 /* size of the new backing store needs to be the same */
598 if (get_loop_size(lo, file) != get_loop_size(lo, file: old_file))
599 goto out_err;
600
601 /* and ... switch */
602 disk_force_media_change(disk: lo->lo_disk);
603 blk_mq_freeze_queue(q: lo->lo_queue);
604 mapping_set_gfp_mask(m: old_file->f_mapping, mask: lo->old_gfp_mask);
605 lo->lo_backing_file = file;
606 lo->old_gfp_mask = mapping_gfp_mask(mapping: file->f_mapping);
607 mapping_set_gfp_mask(m: file->f_mapping,
608 mask: lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
609 loop_update_dio(lo);
610 blk_mq_unfreeze_queue(q: lo->lo_queue);
611 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
612 loop_global_unlock(lo, global: is_loop);
613
614 /*
615 * Flush loop_validate_file() before fput(), for l->lo_backing_file
616 * might be pointing at old_file which might be the last reference.
617 */
618 if (!is_loop) {
619 mutex_lock(&loop_validate_mutex);
620 mutex_unlock(lock: &loop_validate_mutex);
621 }
622 /*
623 * We must drop file reference outside of lo_mutex as dropping
624 * the file ref can take open_mutex which creates circular locking
625 * dependency.
626 */
627 fput(old_file);
628 if (partscan)
629 loop_reread_partitions(lo);
630
631 error = 0;
632done:
633 /* enable and uncork uevent now that we are done */
634 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
635 return error;
636
637out_err:
638 loop_global_unlock(lo, global: is_loop);
639out_putf:
640 fput(file);
641 goto done;
642}
643
644/* loop sysfs attributes */
645
646static ssize_t loop_attr_show(struct device *dev, char *page,
647 ssize_t (*callback)(struct loop_device *, char *))
648{
649 struct gendisk *disk = dev_to_disk(dev);
650 struct loop_device *lo = disk->private_data;
651
652 return callback(lo, page);
653}
654
655#define LOOP_ATTR_RO(_name) \
656static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
657static ssize_t loop_attr_do_show_##_name(struct device *d, \
658 struct device_attribute *attr, char *b) \
659{ \
660 return loop_attr_show(d, b, loop_attr_##_name##_show); \
661} \
662static struct device_attribute loop_attr_##_name = \
663 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
664
665static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
666{
667 ssize_t ret;
668 char *p = NULL;
669
670 spin_lock_irq(lock: &lo->lo_lock);
671 if (lo->lo_backing_file)
672 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
673 spin_unlock_irq(lock: &lo->lo_lock);
674
675 if (IS_ERR_OR_NULL(ptr: p))
676 ret = PTR_ERR(ptr: p);
677 else {
678 ret = strlen(p);
679 memmove(buf, p, ret);
680 buf[ret++] = '\n';
681 buf[ret] = 0;
682 }
683
684 return ret;
685}
686
687static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
688{
689 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_offset);
690}
691
692static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
693{
694 return sysfs_emit(buf, fmt: "%llu\n", (unsigned long long)lo->lo_sizelimit);
695}
696
697static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
698{
699 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
700
701 return sysfs_emit(buf, fmt: "%s\n", autoclear ? "1" : "0");
702}
703
704static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
705{
706 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
707
708 return sysfs_emit(buf, fmt: "%s\n", partscan ? "1" : "0");
709}
710
711static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
712{
713 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
714
715 return sysfs_emit(buf, fmt: "%s\n", dio ? "1" : "0");
716}
717
718LOOP_ATTR_RO(backing_file);
719LOOP_ATTR_RO(offset);
720LOOP_ATTR_RO(sizelimit);
721LOOP_ATTR_RO(autoclear);
722LOOP_ATTR_RO(partscan);
723LOOP_ATTR_RO(dio);
724
725static struct attribute *loop_attrs[] = {
726 &loop_attr_backing_file.attr,
727 &loop_attr_offset.attr,
728 &loop_attr_sizelimit.attr,
729 &loop_attr_autoclear.attr,
730 &loop_attr_partscan.attr,
731 &loop_attr_dio.attr,
732 NULL,
733};
734
735static struct attribute_group loop_attribute_group = {
736 .name = "loop",
737 .attrs= loop_attrs,
738};
739
740static void loop_sysfs_init(struct loop_device *lo)
741{
742 lo->sysfs_inited = !sysfs_create_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
743 grp: &loop_attribute_group);
744}
745
746static void loop_sysfs_exit(struct loop_device *lo)
747{
748 if (lo->sysfs_inited)
749 sysfs_remove_group(kobj: &disk_to_dev(lo->lo_disk)->kobj,
750 grp: &loop_attribute_group);
751}
752
753static void loop_config_discard(struct loop_device *lo,
754 struct queue_limits *lim)
755{
756 struct file *file = lo->lo_backing_file;
757 struct inode *inode = file->f_mapping->host;
758 u32 granularity = 0, max_discard_sectors = 0;
759 struct kstatfs sbuf;
760
761 /*
762 * If the backing device is a block device, mirror its zeroing
763 * capability. Set the discard sectors to the block device's zeroing
764 * capabilities because loop discards result in blkdev_issue_zeroout(),
765 * not blkdev_issue_discard(). This maintains consistent behavior with
766 * file-backed loop devices: discarded regions read back as zero.
767 */
768 if (S_ISBLK(inode->i_mode)) {
769 struct request_queue *backingq = bdev_get_queue(bdev: I_BDEV(inode));
770
771 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
772 granularity = bdev_discard_granularity(bdev: I_BDEV(inode)) ?:
773 queue_physical_block_size(q: backingq);
774
775 /*
776 * We use punch hole to reclaim the free space used by the
777 * image a.k.a. discard.
778 */
779 } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
780 max_discard_sectors = UINT_MAX >> 9;
781 granularity = sbuf.f_bsize;
782 }
783
784 lim->max_hw_discard_sectors = max_discard_sectors;
785 lim->max_write_zeroes_sectors = max_discard_sectors;
786 if (max_discard_sectors)
787 lim->discard_granularity = granularity;
788 else
789 lim->discard_granularity = 0;
790}
791
792struct loop_worker {
793 struct rb_node rb_node;
794 struct work_struct work;
795 struct list_head cmd_list;
796 struct list_head idle_list;
797 struct loop_device *lo;
798 struct cgroup_subsys_state *blkcg_css;
799 unsigned long last_ran_at;
800};
801
802static void loop_workfn(struct work_struct *work);
803
804#ifdef CONFIG_BLK_CGROUP
805static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
806{
807 return !css || css == blkcg_root_css;
808}
809#else
810static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
811{
812 return !css;
813}
814#endif
815
816static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
817{
818 struct rb_node **node, *parent = NULL;
819 struct loop_worker *cur_worker, *worker = NULL;
820 struct work_struct *work;
821 struct list_head *cmd_list;
822
823 spin_lock_irq(lock: &lo->lo_work_lock);
824
825 if (queue_on_root_worker(css: cmd->blkcg_css))
826 goto queue_work;
827
828 node = &lo->worker_tree.rb_node;
829
830 while (*node) {
831 parent = *node;
832 cur_worker = container_of(*node, struct loop_worker, rb_node);
833 if (cur_worker->blkcg_css == cmd->blkcg_css) {
834 worker = cur_worker;
835 break;
836 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
837 node = &(*node)->rb_left;
838 } else {
839 node = &(*node)->rb_right;
840 }
841 }
842 if (worker)
843 goto queue_work;
844
845 worker = kzalloc(size: sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
846 /*
847 * In the event we cannot allocate a worker, just queue on the
848 * rootcg worker and issue the I/O as the rootcg
849 */
850 if (!worker) {
851 cmd->blkcg_css = NULL;
852 if (cmd->memcg_css)
853 css_put(css: cmd->memcg_css);
854 cmd->memcg_css = NULL;
855 goto queue_work;
856 }
857
858 worker->blkcg_css = cmd->blkcg_css;
859 css_get(css: worker->blkcg_css);
860 INIT_WORK(&worker->work, loop_workfn);
861 INIT_LIST_HEAD(list: &worker->cmd_list);
862 INIT_LIST_HEAD(list: &worker->idle_list);
863 worker->lo = lo;
864 rb_link_node(node: &worker->rb_node, parent, rb_link: node);
865 rb_insert_color(&worker->rb_node, &lo->worker_tree);
866queue_work:
867 if (worker) {
868 /*
869 * We need to remove from the idle list here while
870 * holding the lock so that the idle timer doesn't
871 * free the worker
872 */
873 if (!list_empty(head: &worker->idle_list))
874 list_del_init(entry: &worker->idle_list);
875 work = &worker->work;
876 cmd_list = &worker->cmd_list;
877 } else {
878 work = &lo->rootcg_work;
879 cmd_list = &lo->rootcg_cmd_list;
880 }
881 list_add_tail(new: &cmd->list_entry, head: cmd_list);
882 queue_work(wq: lo->workqueue, work);
883 spin_unlock_irq(lock: &lo->lo_work_lock);
884}
885
886static void loop_set_timer(struct loop_device *lo)
887{
888 timer_reduce(timer: &lo->timer, expires: jiffies + LOOP_IDLE_WORKER_TIMEOUT);
889}
890
891static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
892{
893 struct loop_worker *pos, *worker;
894
895 spin_lock_irq(lock: &lo->lo_work_lock);
896 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
897 idle_list) {
898 if (!delete_all &&
899 time_is_after_jiffies(worker->last_ran_at +
900 LOOP_IDLE_WORKER_TIMEOUT))
901 break;
902 list_del(entry: &worker->idle_list);
903 rb_erase(&worker->rb_node, &lo->worker_tree);
904 css_put(css: worker->blkcg_css);
905 kfree(objp: worker);
906 }
907 if (!list_empty(head: &lo->idle_worker_list))
908 loop_set_timer(lo);
909 spin_unlock_irq(lock: &lo->lo_work_lock);
910}
911
912static void loop_free_idle_workers_timer(struct timer_list *timer)
913{
914 struct loop_device *lo = container_of(timer, struct loop_device, timer);
915
916 return loop_free_idle_workers(lo, delete_all: false);
917}
918
919static void loop_update_rotational(struct loop_device *lo)
920{
921 struct file *file = lo->lo_backing_file;
922 struct inode *file_inode = file->f_mapping->host;
923 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
924 struct request_queue *q = lo->lo_queue;
925 bool nonrot = true;
926
927 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
928 if (file_bdev)
929 nonrot = bdev_nonrot(bdev: file_bdev);
930
931 if (nonrot)
932 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
933 else
934 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
935}
936
937/**
938 * loop_set_status_from_info - configure device from loop_info
939 * @lo: struct loop_device to configure
940 * @info: struct loop_info64 to configure the device with
941 *
942 * Configures the loop device parameters according to the passed
943 * in loop_info64 configuration.
944 */
945static int
946loop_set_status_from_info(struct loop_device *lo,
947 const struct loop_info64 *info)
948{
949 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
950 return -EINVAL;
951
952 switch (info->lo_encrypt_type) {
953 case LO_CRYPT_NONE:
954 break;
955 case LO_CRYPT_XOR:
956 pr_warn("support for the xor transformation has been removed.\n");
957 return -EINVAL;
958 case LO_CRYPT_CRYPTOAPI:
959 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
960 return -EINVAL;
961 default:
962 return -EINVAL;
963 }
964
965 /* Avoid assigning overflow values */
966 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
967 return -EOVERFLOW;
968
969 lo->lo_offset = info->lo_offset;
970 lo->lo_sizelimit = info->lo_sizelimit;
971
972 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
973 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
974 lo->lo_flags = info->lo_flags;
975 return 0;
976}
977
978static int loop_reconfigure_limits(struct loop_device *lo, unsigned short bsize,
979 bool update_discard_settings)
980{
981 struct queue_limits lim;
982
983 lim = queue_limits_start_update(q: lo->lo_queue);
984 lim.logical_block_size = bsize;
985 lim.physical_block_size = bsize;
986 lim.io_min = bsize;
987 if (update_discard_settings)
988 loop_config_discard(lo, lim: &lim);
989 return queue_limits_commit_update(q: lo->lo_queue, lim: &lim);
990}
991
992static int loop_configure(struct loop_device *lo, blk_mode_t mode,
993 struct block_device *bdev,
994 const struct loop_config *config)
995{
996 struct file *file = fget(fd: config->fd);
997 struct inode *inode;
998 struct address_space *mapping;
999 int error;
1000 loff_t size;
1001 bool partscan;
1002 unsigned short bsize;
1003 bool is_loop;
1004
1005 if (!file)
1006 return -EBADF;
1007 is_loop = is_loop_device(file);
1008
1009 /* This is safe, since we have a reference from open(). */
1010 __module_get(THIS_MODULE);
1011
1012 /*
1013 * If we don't hold exclusive handle for the device, upgrade to it
1014 * here to avoid changing device under exclusive owner.
1015 */
1016 if (!(mode & BLK_OPEN_EXCL)) {
1017 error = bd_prepare_to_claim(bdev, holder: loop_configure, NULL);
1018 if (error)
1019 goto out_putf;
1020 }
1021
1022 error = loop_global_lock_killable(lo, global: is_loop);
1023 if (error)
1024 goto out_bdev;
1025
1026 error = -EBUSY;
1027 if (lo->lo_state != Lo_unbound)
1028 goto out_unlock;
1029
1030 error = loop_validate_file(file, bdev);
1031 if (error)
1032 goto out_unlock;
1033
1034 mapping = file->f_mapping;
1035 inode = mapping->host;
1036
1037 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1038 error = -EINVAL;
1039 goto out_unlock;
1040 }
1041
1042 if (config->block_size) {
1043 error = blk_validate_block_size(bsize: config->block_size);
1044 if (error)
1045 goto out_unlock;
1046 }
1047
1048 error = loop_set_status_from_info(lo, info: &config->info);
1049 if (error)
1050 goto out_unlock;
1051
1052 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1053 !file->f_op->write_iter)
1054 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1055
1056 if (!lo->workqueue) {
1057 lo->workqueue = alloc_workqueue(fmt: "loop%d",
1058 flags: WQ_UNBOUND | WQ_FREEZABLE,
1059 max_active: 0, lo->lo_number);
1060 if (!lo->workqueue) {
1061 error = -ENOMEM;
1062 goto out_unlock;
1063 }
1064 }
1065
1066 /* suppress uevents while reconfiguring the device */
1067 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 1);
1068
1069 disk_force_media_change(disk: lo->lo_disk);
1070 set_disk_ro(disk: lo->lo_disk, read_only: (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1071
1072 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1073 lo->lo_device = bdev;
1074 lo->lo_backing_file = file;
1075 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1076 mapping_set_gfp_mask(m: mapping, mask: lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1077
1078 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1079 blk_queue_write_cache(q: lo->lo_queue, enabled: true, fua: false);
1080
1081 if (config->block_size)
1082 bsize = config->block_size;
1083 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1084 /* In case of direct I/O, match underlying block size */
1085 bsize = bdev_logical_block_size(bdev: inode->i_sb->s_bdev);
1086 else
1087 bsize = 512;
1088
1089 error = loop_reconfigure_limits(lo, bsize, update_discard_settings: true);
1090 if (WARN_ON_ONCE(error))
1091 goto out_unlock;
1092
1093 loop_update_rotational(lo);
1094 loop_update_dio(lo);
1095 loop_sysfs_init(lo);
1096
1097 size = get_loop_size(lo, file);
1098 loop_set_size(lo, size);
1099
1100 /* Order wrt reading lo_state in loop_validate_file(). */
1101 wmb();
1102
1103 lo->lo_state = Lo_bound;
1104 if (part_shift)
1105 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1106 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1107 if (partscan)
1108 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1109
1110 /* enable and uncork uevent now that we are done */
1111 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), val: 0);
1112
1113 loop_global_unlock(lo, global: is_loop);
1114 if (partscan)
1115 loop_reread_partitions(lo);
1116
1117 if (!(mode & BLK_OPEN_EXCL))
1118 bd_abort_claiming(bdev, holder: loop_configure);
1119
1120 return 0;
1121
1122out_unlock:
1123 loop_global_unlock(lo, global: is_loop);
1124out_bdev:
1125 if (!(mode & BLK_OPEN_EXCL))
1126 bd_abort_claiming(bdev, holder: loop_configure);
1127out_putf:
1128 fput(file);
1129 /* This is safe: open() is still holding a reference. */
1130 module_put(THIS_MODULE);
1131 return error;
1132}
1133
1134static void __loop_clr_fd(struct loop_device *lo, bool release)
1135{
1136 struct file *filp;
1137 gfp_t gfp = lo->old_gfp_mask;
1138
1139 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1140 blk_queue_write_cache(q: lo->lo_queue, enabled: false, fua: false);
1141
1142 /*
1143 * Freeze the request queue when unbinding on a live file descriptor and
1144 * thus an open device. When called from ->release we are guaranteed
1145 * that there is no I/O in progress already.
1146 */
1147 if (!release)
1148 blk_mq_freeze_queue(q: lo->lo_queue);
1149
1150 spin_lock_irq(lock: &lo->lo_lock);
1151 filp = lo->lo_backing_file;
1152 lo->lo_backing_file = NULL;
1153 spin_unlock_irq(lock: &lo->lo_lock);
1154
1155 lo->lo_device = NULL;
1156 lo->lo_offset = 0;
1157 lo->lo_sizelimit = 0;
1158 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1159 loop_reconfigure_limits(lo, bsize: 512, update_discard_settings: false);
1160 invalidate_disk(disk: lo->lo_disk);
1161 loop_sysfs_exit(lo);
1162 /* let user-space know about this change */
1163 kobject_uevent(kobj: &disk_to_dev(lo->lo_disk)->kobj, action: KOBJ_CHANGE);
1164 mapping_set_gfp_mask(m: filp->f_mapping, mask: gfp);
1165 /* This is safe: open() is still holding a reference. */
1166 module_put(THIS_MODULE);
1167 if (!release)
1168 blk_mq_unfreeze_queue(q: lo->lo_queue);
1169
1170 disk_force_media_change(disk: lo->lo_disk);
1171
1172 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1173 int err;
1174
1175 /*
1176 * open_mutex has been held already in release path, so don't
1177 * acquire it if this function is called in such case.
1178 *
1179 * If the reread partition isn't from release path, lo_refcnt
1180 * must be at least one and it can only become zero when the
1181 * current holder is released.
1182 */
1183 if (!release)
1184 mutex_lock(&lo->lo_disk->open_mutex);
1185 err = bdev_disk_changed(disk: lo->lo_disk, invalidate: false);
1186 if (!release)
1187 mutex_unlock(lock: &lo->lo_disk->open_mutex);
1188 if (err)
1189 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1190 __func__, lo->lo_number, err);
1191 /* Device is gone, no point in returning error */
1192 }
1193
1194 /*
1195 * lo->lo_state is set to Lo_unbound here after above partscan has
1196 * finished. There cannot be anybody else entering __loop_clr_fd() as
1197 * Lo_rundown state protects us from all the other places trying to
1198 * change the 'lo' device.
1199 */
1200 lo->lo_flags = 0;
1201 if (!part_shift)
1202 set_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1203 mutex_lock(&lo->lo_mutex);
1204 lo->lo_state = Lo_unbound;
1205 mutex_unlock(lock: &lo->lo_mutex);
1206
1207 /*
1208 * Need not hold lo_mutex to fput backing file. Calling fput holding
1209 * lo_mutex triggers a circular lock dependency possibility warning as
1210 * fput can take open_mutex which is usually taken before lo_mutex.
1211 */
1212 fput(filp);
1213}
1214
1215static int loop_clr_fd(struct loop_device *lo)
1216{
1217 int err;
1218
1219 /*
1220 * Since lo_ioctl() is called without locks held, it is possible that
1221 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1222 *
1223 * Therefore, use global lock when setting Lo_rundown state in order to
1224 * make sure that loop_validate_file() will fail if the "struct file"
1225 * which loop_configure()/loop_change_fd() found via fget() was this
1226 * loop device.
1227 */
1228 err = loop_global_lock_killable(lo, global: true);
1229 if (err)
1230 return err;
1231 if (lo->lo_state != Lo_bound) {
1232 loop_global_unlock(lo, global: true);
1233 return -ENXIO;
1234 }
1235 /*
1236 * If we've explicitly asked to tear down the loop device,
1237 * and it has an elevated reference count, set it for auto-teardown when
1238 * the last reference goes away. This stops $!~#$@ udev from
1239 * preventing teardown because it decided that it needs to run blkid on
1240 * the loopback device whenever they appear. xfstests is notorious for
1241 * failing tests because blkid via udev races with a losetup
1242 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1243 * command to fail with EBUSY.
1244 */
1245 if (disk_openers(disk: lo->lo_disk) > 1) {
1246 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1247 loop_global_unlock(lo, global: true);
1248 return 0;
1249 }
1250 lo->lo_state = Lo_rundown;
1251 loop_global_unlock(lo, global: true);
1252
1253 __loop_clr_fd(lo, release: false);
1254 return 0;
1255}
1256
1257static int
1258loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1259{
1260 int err;
1261 int prev_lo_flags;
1262 bool partscan = false;
1263 bool size_changed = false;
1264
1265 err = mutex_lock_killable(&lo->lo_mutex);
1266 if (err)
1267 return err;
1268 if (lo->lo_state != Lo_bound) {
1269 err = -ENXIO;
1270 goto out_unlock;
1271 }
1272
1273 if (lo->lo_offset != info->lo_offset ||
1274 lo->lo_sizelimit != info->lo_sizelimit) {
1275 size_changed = true;
1276 sync_blockdev(bdev: lo->lo_device);
1277 invalidate_bdev(bdev: lo->lo_device);
1278 }
1279
1280 /* I/O need to be drained during transfer transition */
1281 blk_mq_freeze_queue(q: lo->lo_queue);
1282
1283 prev_lo_flags = lo->lo_flags;
1284
1285 err = loop_set_status_from_info(lo, info);
1286 if (err)
1287 goto out_unfreeze;
1288
1289 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1290 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1291 /* For those flags, use the previous values instead */
1292 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1293 /* For flags that can't be cleared, use previous values too */
1294 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1295
1296 if (size_changed) {
1297 loff_t new_size = get_size(offset: lo->lo_offset, sizelimit: lo->lo_sizelimit,
1298 file: lo->lo_backing_file);
1299 loop_set_size(lo, size: new_size);
1300 }
1301
1302 /* update dio if lo_offset or transfer is changed */
1303 __loop_update_dio(lo, dio: lo->use_dio);
1304
1305out_unfreeze:
1306 blk_mq_unfreeze_queue(q: lo->lo_queue);
1307
1308 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1309 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1310 clear_bit(GD_SUPPRESS_PART_SCAN, addr: &lo->lo_disk->state);
1311 partscan = true;
1312 }
1313out_unlock:
1314 mutex_unlock(lock: &lo->lo_mutex);
1315 if (partscan)
1316 loop_reread_partitions(lo);
1317
1318 return err;
1319}
1320
1321static int
1322loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1323{
1324 struct path path;
1325 struct kstat stat;
1326 int ret;
1327
1328 ret = mutex_lock_killable(&lo->lo_mutex);
1329 if (ret)
1330 return ret;
1331 if (lo->lo_state != Lo_bound) {
1332 mutex_unlock(lock: &lo->lo_mutex);
1333 return -ENXIO;
1334 }
1335
1336 memset(info, 0, sizeof(*info));
1337 info->lo_number = lo->lo_number;
1338 info->lo_offset = lo->lo_offset;
1339 info->lo_sizelimit = lo->lo_sizelimit;
1340 info->lo_flags = lo->lo_flags;
1341 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1342
1343 /* Drop lo_mutex while we call into the filesystem. */
1344 path = lo->lo_backing_file->f_path;
1345 path_get(&path);
1346 mutex_unlock(lock: &lo->lo_mutex);
1347 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1348 if (!ret) {
1349 info->lo_device = huge_encode_dev(dev: stat.dev);
1350 info->lo_inode = stat.ino;
1351 info->lo_rdevice = huge_encode_dev(dev: stat.rdev);
1352 }
1353 path_put(&path);
1354 return ret;
1355}
1356
1357static void
1358loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1359{
1360 memset(info64, 0, sizeof(*info64));
1361 info64->lo_number = info->lo_number;
1362 info64->lo_device = info->lo_device;
1363 info64->lo_inode = info->lo_inode;
1364 info64->lo_rdevice = info->lo_rdevice;
1365 info64->lo_offset = info->lo_offset;
1366 info64->lo_sizelimit = 0;
1367 info64->lo_flags = info->lo_flags;
1368 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1369}
1370
1371static int
1372loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1373{
1374 memset(info, 0, sizeof(*info));
1375 info->lo_number = info64->lo_number;
1376 info->lo_device = info64->lo_device;
1377 info->lo_inode = info64->lo_inode;
1378 info->lo_rdevice = info64->lo_rdevice;
1379 info->lo_offset = info64->lo_offset;
1380 info->lo_flags = info64->lo_flags;
1381 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1382
1383 /* error in case values were truncated */
1384 if (info->lo_device != info64->lo_device ||
1385 info->lo_rdevice != info64->lo_rdevice ||
1386 info->lo_inode != info64->lo_inode ||
1387 info->lo_offset != info64->lo_offset)
1388 return -EOVERFLOW;
1389
1390 return 0;
1391}
1392
1393static int
1394loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1395{
1396 struct loop_info info;
1397 struct loop_info64 info64;
1398
1399 if (copy_from_user(to: &info, from: arg, n: sizeof (struct loop_info)))
1400 return -EFAULT;
1401 loop_info64_from_old(info: &info, info64: &info64);
1402 return loop_set_status(lo, info: &info64);
1403}
1404
1405static int
1406loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1407{
1408 struct loop_info64 info64;
1409
1410 if (copy_from_user(to: &info64, from: arg, n: sizeof (struct loop_info64)))
1411 return -EFAULT;
1412 return loop_set_status(lo, info: &info64);
1413}
1414
1415static int
1416loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1417 struct loop_info info;
1418 struct loop_info64 info64;
1419 int err;
1420
1421 if (!arg)
1422 return -EINVAL;
1423 err = loop_get_status(lo, info: &info64);
1424 if (!err)
1425 err = loop_info64_to_old(info64: &info64, info: &info);
1426 if (!err && copy_to_user(to: arg, from: &info, n: sizeof(info)))
1427 err = -EFAULT;
1428
1429 return err;
1430}
1431
1432static int
1433loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1434 struct loop_info64 info64;
1435 int err;
1436
1437 if (!arg)
1438 return -EINVAL;
1439 err = loop_get_status(lo, info: &info64);
1440 if (!err && copy_to_user(to: arg, from: &info64, n: sizeof(info64)))
1441 err = -EFAULT;
1442
1443 return err;
1444}
1445
1446static int loop_set_capacity(struct loop_device *lo)
1447{
1448 loff_t size;
1449
1450 if (unlikely(lo->lo_state != Lo_bound))
1451 return -ENXIO;
1452
1453 size = get_loop_size(lo, file: lo->lo_backing_file);
1454 loop_set_size(lo, size);
1455
1456 return 0;
1457}
1458
1459static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1460{
1461 int error = -ENXIO;
1462 if (lo->lo_state != Lo_bound)
1463 goto out;
1464
1465 __loop_update_dio(lo, dio: !!arg);
1466 if (lo->use_dio == !!arg)
1467 return 0;
1468 error = -EINVAL;
1469 out:
1470 return error;
1471}
1472
1473static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1474{
1475 int err = 0;
1476
1477 if (lo->lo_state != Lo_bound)
1478 return -ENXIO;
1479
1480 err = blk_validate_block_size(bsize: arg);
1481 if (err)
1482 return err;
1483
1484 if (lo->lo_queue->limits.logical_block_size == arg)
1485 return 0;
1486
1487 sync_blockdev(bdev: lo->lo_device);
1488 invalidate_bdev(bdev: lo->lo_device);
1489
1490 blk_mq_freeze_queue(q: lo->lo_queue);
1491 err = loop_reconfigure_limits(lo, bsize: arg, update_discard_settings: false);
1492 loop_update_dio(lo);
1493 blk_mq_unfreeze_queue(q: lo->lo_queue);
1494
1495 return err;
1496}
1497
1498static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1499 unsigned long arg)
1500{
1501 int err;
1502
1503 err = mutex_lock_killable(&lo->lo_mutex);
1504 if (err)
1505 return err;
1506 switch (cmd) {
1507 case LOOP_SET_CAPACITY:
1508 err = loop_set_capacity(lo);
1509 break;
1510 case LOOP_SET_DIRECT_IO:
1511 err = loop_set_dio(lo, arg);
1512 break;
1513 case LOOP_SET_BLOCK_SIZE:
1514 err = loop_set_block_size(lo, arg);
1515 break;
1516 default:
1517 err = -EINVAL;
1518 }
1519 mutex_unlock(lock: &lo->lo_mutex);
1520 return err;
1521}
1522
1523static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1524 unsigned int cmd, unsigned long arg)
1525{
1526 struct loop_device *lo = bdev->bd_disk->private_data;
1527 void __user *argp = (void __user *) arg;
1528 int err;
1529
1530 switch (cmd) {
1531 case LOOP_SET_FD: {
1532 /*
1533 * Legacy case - pass in a zeroed out struct loop_config with
1534 * only the file descriptor set , which corresponds with the
1535 * default parameters we'd have used otherwise.
1536 */
1537 struct loop_config config;
1538
1539 memset(&config, 0, sizeof(config));
1540 config.fd = arg;
1541
1542 return loop_configure(lo, mode, bdev, config: &config);
1543 }
1544 case LOOP_CONFIGURE: {
1545 struct loop_config config;
1546
1547 if (copy_from_user(to: &config, from: argp, n: sizeof(config)))
1548 return -EFAULT;
1549
1550 return loop_configure(lo, mode, bdev, config: &config);
1551 }
1552 case LOOP_CHANGE_FD:
1553 return loop_change_fd(lo, bdev, arg);
1554 case LOOP_CLR_FD:
1555 return loop_clr_fd(lo);
1556 case LOOP_SET_STATUS:
1557 err = -EPERM;
1558 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1559 err = loop_set_status_old(lo, arg: argp);
1560 break;
1561 case LOOP_GET_STATUS:
1562 return loop_get_status_old(lo, arg: argp);
1563 case LOOP_SET_STATUS64:
1564 err = -EPERM;
1565 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1566 err = loop_set_status64(lo, arg: argp);
1567 break;
1568 case LOOP_GET_STATUS64:
1569 return loop_get_status64(lo, arg: argp);
1570 case LOOP_SET_CAPACITY:
1571 case LOOP_SET_DIRECT_IO:
1572 case LOOP_SET_BLOCK_SIZE:
1573 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1574 return -EPERM;
1575 fallthrough;
1576 default:
1577 err = lo_simple_ioctl(lo, cmd, arg);
1578 break;
1579 }
1580
1581 return err;
1582}
1583
1584#ifdef CONFIG_COMPAT
1585struct compat_loop_info {
1586 compat_int_t lo_number; /* ioctl r/o */
1587 compat_dev_t lo_device; /* ioctl r/o */
1588 compat_ulong_t lo_inode; /* ioctl r/o */
1589 compat_dev_t lo_rdevice; /* ioctl r/o */
1590 compat_int_t lo_offset;
1591 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1592 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1593 compat_int_t lo_flags; /* ioctl r/o */
1594 char lo_name[LO_NAME_SIZE];
1595 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1596 compat_ulong_t lo_init[2];
1597 char reserved[4];
1598};
1599
1600/*
1601 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1602 * - noinlined to reduce stack space usage in main part of driver
1603 */
1604static noinline int
1605loop_info64_from_compat(const struct compat_loop_info __user *arg,
1606 struct loop_info64 *info64)
1607{
1608 struct compat_loop_info info;
1609
1610 if (copy_from_user(to: &info, from: arg, n: sizeof(info)))
1611 return -EFAULT;
1612
1613 memset(info64, 0, sizeof(*info64));
1614 info64->lo_number = info.lo_number;
1615 info64->lo_device = info.lo_device;
1616 info64->lo_inode = info.lo_inode;
1617 info64->lo_rdevice = info.lo_rdevice;
1618 info64->lo_offset = info.lo_offset;
1619 info64->lo_sizelimit = 0;
1620 info64->lo_flags = info.lo_flags;
1621 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1622 return 0;
1623}
1624
1625/*
1626 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1627 * - noinlined to reduce stack space usage in main part of driver
1628 */
1629static noinline int
1630loop_info64_to_compat(const struct loop_info64 *info64,
1631 struct compat_loop_info __user *arg)
1632{
1633 struct compat_loop_info info;
1634
1635 memset(&info, 0, sizeof(info));
1636 info.lo_number = info64->lo_number;
1637 info.lo_device = info64->lo_device;
1638 info.lo_inode = info64->lo_inode;
1639 info.lo_rdevice = info64->lo_rdevice;
1640 info.lo_offset = info64->lo_offset;
1641 info.lo_flags = info64->lo_flags;
1642 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1643
1644 /* error in case values were truncated */
1645 if (info.lo_device != info64->lo_device ||
1646 info.lo_rdevice != info64->lo_rdevice ||
1647 info.lo_inode != info64->lo_inode ||
1648 info.lo_offset != info64->lo_offset)
1649 return -EOVERFLOW;
1650
1651 if (copy_to_user(to: arg, from: &info, n: sizeof(info)))
1652 return -EFAULT;
1653 return 0;
1654}
1655
1656static int
1657loop_set_status_compat(struct loop_device *lo,
1658 const struct compat_loop_info __user *arg)
1659{
1660 struct loop_info64 info64;
1661 int ret;
1662
1663 ret = loop_info64_from_compat(arg, info64: &info64);
1664 if (ret < 0)
1665 return ret;
1666 return loop_set_status(lo, info: &info64);
1667}
1668
1669static int
1670loop_get_status_compat(struct loop_device *lo,
1671 struct compat_loop_info __user *arg)
1672{
1673 struct loop_info64 info64;
1674 int err;
1675
1676 if (!arg)
1677 return -EINVAL;
1678 err = loop_get_status(lo, info: &info64);
1679 if (!err)
1680 err = loop_info64_to_compat(info64: &info64, arg);
1681 return err;
1682}
1683
1684static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1685 unsigned int cmd, unsigned long arg)
1686{
1687 struct loop_device *lo = bdev->bd_disk->private_data;
1688 int err;
1689
1690 switch(cmd) {
1691 case LOOP_SET_STATUS:
1692 err = loop_set_status_compat(lo,
1693 arg: (const struct compat_loop_info __user *)arg);
1694 break;
1695 case LOOP_GET_STATUS:
1696 err = loop_get_status_compat(lo,
1697 arg: (struct compat_loop_info __user *)arg);
1698 break;
1699 case LOOP_SET_CAPACITY:
1700 case LOOP_CLR_FD:
1701 case LOOP_GET_STATUS64:
1702 case LOOP_SET_STATUS64:
1703 case LOOP_CONFIGURE:
1704 arg = (unsigned long) compat_ptr(uptr: arg);
1705 fallthrough;
1706 case LOOP_SET_FD:
1707 case LOOP_CHANGE_FD:
1708 case LOOP_SET_BLOCK_SIZE:
1709 case LOOP_SET_DIRECT_IO:
1710 err = lo_ioctl(bdev, mode, cmd, arg);
1711 break;
1712 default:
1713 err = -ENOIOCTLCMD;
1714 break;
1715 }
1716 return err;
1717}
1718#endif
1719
1720static void lo_release(struct gendisk *disk)
1721{
1722 struct loop_device *lo = disk->private_data;
1723
1724 if (disk_openers(disk) > 0)
1725 return;
1726
1727 mutex_lock(&lo->lo_mutex);
1728 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
1729 lo->lo_state = Lo_rundown;
1730 mutex_unlock(lock: &lo->lo_mutex);
1731 /*
1732 * In autoclear mode, stop the loop thread
1733 * and remove configuration after last close.
1734 */
1735 __loop_clr_fd(lo, release: true);
1736 return;
1737 }
1738 mutex_unlock(lock: &lo->lo_mutex);
1739}
1740
1741static void lo_free_disk(struct gendisk *disk)
1742{
1743 struct loop_device *lo = disk->private_data;
1744
1745 if (lo->workqueue)
1746 destroy_workqueue(wq: lo->workqueue);
1747 loop_free_idle_workers(lo, delete_all: true);
1748 timer_shutdown_sync(timer: &lo->timer);
1749 mutex_destroy(lock: &lo->lo_mutex);
1750 kfree(objp: lo);
1751}
1752
1753static const struct block_device_operations lo_fops = {
1754 .owner = THIS_MODULE,
1755 .release = lo_release,
1756 .ioctl = lo_ioctl,
1757#ifdef CONFIG_COMPAT
1758 .compat_ioctl = lo_compat_ioctl,
1759#endif
1760 .free_disk = lo_free_disk,
1761};
1762
1763/*
1764 * And now the modules code and kernel interface.
1765 */
1766
1767/*
1768 * If max_loop is specified, create that many devices upfront.
1769 * This also becomes a hard limit. If max_loop is not specified,
1770 * the default isn't a hard limit (as before commit 85c50197716c
1771 * changed the default value from 0 for max_loop=0 reasons), just
1772 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1773 * init time. Loop devices can be requested on-demand with the
1774 * /dev/loop-control interface, or be instantiated by accessing
1775 * a 'dead' device node.
1776 */
1777static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1778
1779#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1780static bool max_loop_specified;
1781
1782static int max_loop_param_set_int(const char *val,
1783 const struct kernel_param *kp)
1784{
1785 int ret;
1786
1787 ret = param_set_int(val, kp);
1788 if (ret < 0)
1789 return ret;
1790
1791 max_loop_specified = true;
1792 return 0;
1793}
1794
1795static const struct kernel_param_ops max_loop_param_ops = {
1796 .set = max_loop_param_set_int,
1797 .get = param_get_int,
1798};
1799
1800module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1801MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1802#else
1803module_param(max_loop, int, 0444);
1804MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1805#endif
1806
1807module_param(max_part, int, 0444);
1808MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1809
1810static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1811
1812static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1813{
1814 int qd, ret;
1815
1816 ret = kstrtoint(s, base: 0, res: &qd);
1817 if (ret < 0)
1818 return ret;
1819 if (qd < 1)
1820 return -EINVAL;
1821 hw_queue_depth = qd;
1822 return 0;
1823}
1824
1825static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1826 .set = loop_set_hw_queue_depth,
1827 .get = param_get_int,
1828};
1829
1830device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1831MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1832
1833MODULE_LICENSE("GPL");
1834MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1835
1836static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1837 const struct blk_mq_queue_data *bd)
1838{
1839 struct request *rq = bd->rq;
1840 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1841 struct loop_device *lo = rq->q->queuedata;
1842
1843 blk_mq_start_request(rq);
1844
1845 if (lo->lo_state != Lo_bound)
1846 return BLK_STS_IOERR;
1847
1848 switch (req_op(req: rq)) {
1849 case REQ_OP_FLUSH:
1850 case REQ_OP_DISCARD:
1851 case REQ_OP_WRITE_ZEROES:
1852 cmd->use_aio = false;
1853 break;
1854 default:
1855 cmd->use_aio = lo->use_dio;
1856 break;
1857 }
1858
1859 /* always use the first bio's css */
1860 cmd->blkcg_css = NULL;
1861 cmd->memcg_css = NULL;
1862#ifdef CONFIG_BLK_CGROUP
1863 if (rq->bio) {
1864 cmd->blkcg_css = bio_blkcg_css(bio: rq->bio);
1865#ifdef CONFIG_MEMCG
1866 if (cmd->blkcg_css) {
1867 cmd->memcg_css =
1868 cgroup_get_e_css(cgroup: cmd->blkcg_css->cgroup,
1869 ss: &memory_cgrp_subsys);
1870 }
1871#endif
1872 }
1873#endif
1874 loop_queue_work(lo, cmd);
1875
1876 return BLK_STS_OK;
1877}
1878
1879static void loop_handle_cmd(struct loop_cmd *cmd)
1880{
1881 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1882 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1883 struct request *rq = blk_mq_rq_from_pdu(pdu: cmd);
1884 const bool write = op_is_write(op: req_op(req: rq));
1885 struct loop_device *lo = rq->q->queuedata;
1886 int ret = 0;
1887 struct mem_cgroup *old_memcg = NULL;
1888 const bool use_aio = cmd->use_aio;
1889
1890 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1891 ret = -EIO;
1892 goto failed;
1893 }
1894
1895 if (cmd_blkcg_css)
1896 kthread_associate_blkcg(css: cmd_blkcg_css);
1897 if (cmd_memcg_css)
1898 old_memcg = set_active_memcg(
1899 mem_cgroup_from_css(css: cmd_memcg_css));
1900
1901 /*
1902 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1903 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1904 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1905 * not yet been completed.
1906 */
1907 ret = do_req_filebacked(lo, rq);
1908
1909 if (cmd_blkcg_css)
1910 kthread_associate_blkcg(NULL);
1911
1912 if (cmd_memcg_css) {
1913 set_active_memcg(old_memcg);
1914 css_put(css: cmd_memcg_css);
1915 }
1916 failed:
1917 /* complete non-aio request */
1918 if (!use_aio || ret) {
1919 if (ret == -EOPNOTSUPP)
1920 cmd->ret = ret;
1921 else
1922 cmd->ret = ret ? -EIO : 0;
1923 if (likely(!blk_should_fake_timeout(rq->q)))
1924 blk_mq_complete_request(rq);
1925 }
1926}
1927
1928static void loop_process_work(struct loop_worker *worker,
1929 struct list_head *cmd_list, struct loop_device *lo)
1930{
1931 int orig_flags = current->flags;
1932 struct loop_cmd *cmd;
1933
1934 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1935 spin_lock_irq(lock: &lo->lo_work_lock);
1936 while (!list_empty(head: cmd_list)) {
1937 cmd = container_of(
1938 cmd_list->next, struct loop_cmd, list_entry);
1939 list_del(entry: cmd_list->next);
1940 spin_unlock_irq(lock: &lo->lo_work_lock);
1941
1942 loop_handle_cmd(cmd);
1943 cond_resched();
1944
1945 spin_lock_irq(lock: &lo->lo_work_lock);
1946 }
1947
1948 /*
1949 * We only add to the idle list if there are no pending cmds
1950 * *and* the worker will not run again which ensures that it
1951 * is safe to free any worker on the idle list
1952 */
1953 if (worker && !work_pending(&worker->work)) {
1954 worker->last_ran_at = jiffies;
1955 list_add_tail(new: &worker->idle_list, head: &lo->idle_worker_list);
1956 loop_set_timer(lo);
1957 }
1958 spin_unlock_irq(lock: &lo->lo_work_lock);
1959 current->flags = orig_flags;
1960}
1961
1962static void loop_workfn(struct work_struct *work)
1963{
1964 struct loop_worker *worker =
1965 container_of(work, struct loop_worker, work);
1966 loop_process_work(worker, cmd_list: &worker->cmd_list, lo: worker->lo);
1967}
1968
1969static void loop_rootcg_workfn(struct work_struct *work)
1970{
1971 struct loop_device *lo =
1972 container_of(work, struct loop_device, rootcg_work);
1973 loop_process_work(NULL, cmd_list: &lo->rootcg_cmd_list, lo);
1974}
1975
1976static const struct blk_mq_ops loop_mq_ops = {
1977 .queue_rq = loop_queue_rq,
1978 .complete = lo_complete_rq,
1979};
1980
1981static int loop_add(int i)
1982{
1983 struct queue_limits lim = {
1984 /*
1985 * Random number picked from the historic block max_sectors cap.
1986 */
1987 .max_hw_sectors = 2560u,
1988 };
1989 struct loop_device *lo;
1990 struct gendisk *disk;
1991 int err;
1992
1993 err = -ENOMEM;
1994 lo = kzalloc(size: sizeof(*lo), GFP_KERNEL);
1995 if (!lo)
1996 goto out;
1997 lo->worker_tree = RB_ROOT;
1998 INIT_LIST_HEAD(list: &lo->idle_worker_list);
1999 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
2000 lo->lo_state = Lo_unbound;
2001
2002 err = mutex_lock_killable(&loop_ctl_mutex);
2003 if (err)
2004 goto out_free_dev;
2005
2006 /* allocate id, if @id >= 0, we're requesting that specific id */
2007 if (i >= 0) {
2008 err = idr_alloc(&loop_index_idr, ptr: lo, start: i, end: i + 1, GFP_KERNEL);
2009 if (err == -ENOSPC)
2010 err = -EEXIST;
2011 } else {
2012 err = idr_alloc(&loop_index_idr, ptr: lo, start: 0, end: 0, GFP_KERNEL);
2013 }
2014 mutex_unlock(lock: &loop_ctl_mutex);
2015 if (err < 0)
2016 goto out_free_dev;
2017 i = err;
2018
2019 lo->tag_set.ops = &loop_mq_ops;
2020 lo->tag_set.nr_hw_queues = 1;
2021 lo->tag_set.queue_depth = hw_queue_depth;
2022 lo->tag_set.numa_node = NUMA_NO_NODE;
2023 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2024 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
2025 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2026 lo->tag_set.driver_data = lo;
2027
2028 err = blk_mq_alloc_tag_set(set: &lo->tag_set);
2029 if (err)
2030 goto out_free_idr;
2031
2032 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2033 if (IS_ERR(ptr: disk)) {
2034 err = PTR_ERR(ptr: disk);
2035 goto out_cleanup_tags;
2036 }
2037 lo->lo_queue = lo->lo_disk->queue;
2038
2039 /*
2040 * By default, we do buffer IO, so it doesn't make sense to enable
2041 * merge because the I/O submitted to backing file is handled page by
2042 * page. For directio mode, merge does help to dispatch bigger request
2043 * to underlayer disk. We will enable merge once directio is enabled.
2044 */
2045 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q: lo->lo_queue);
2046
2047 /*
2048 * Disable partition scanning by default. The in-kernel partition
2049 * scanning can be requested individually per-device during its
2050 * setup. Userspace can always add and remove partitions from all
2051 * devices. The needed partition minors are allocated from the
2052 * extended minor space, the main loop device numbers will continue
2053 * to match the loop minors, regardless of the number of partitions
2054 * used.
2055 *
2056 * If max_part is given, partition scanning is globally enabled for
2057 * all loop devices. The minors for the main loop devices will be
2058 * multiples of max_part.
2059 *
2060 * Note: Global-for-all-devices, set-only-at-init, read-only module
2061 * parameteters like 'max_loop' and 'max_part' make things needlessly
2062 * complicated, are too static, inflexible and may surprise
2063 * userspace tools. Parameters like this in general should be avoided.
2064 */
2065 if (!part_shift)
2066 set_bit(GD_SUPPRESS_PART_SCAN, addr: &disk->state);
2067 mutex_init(&lo->lo_mutex);
2068 lo->lo_number = i;
2069 spin_lock_init(&lo->lo_lock);
2070 spin_lock_init(&lo->lo_work_lock);
2071 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2072 INIT_LIST_HEAD(list: &lo->rootcg_cmd_list);
2073 disk->major = LOOP_MAJOR;
2074 disk->first_minor = i << part_shift;
2075 disk->minors = 1 << part_shift;
2076 disk->fops = &lo_fops;
2077 disk->private_data = lo;
2078 disk->queue = lo->lo_queue;
2079 disk->events = DISK_EVENT_MEDIA_CHANGE;
2080 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2081 sprintf(buf: disk->disk_name, fmt: "loop%d", i);
2082 /* Make this loop device reachable from pathname. */
2083 err = add_disk(disk);
2084 if (err)
2085 goto out_cleanup_disk;
2086
2087 /* Show this loop device. */
2088 mutex_lock(&loop_ctl_mutex);
2089 lo->idr_visible = true;
2090 mutex_unlock(lock: &loop_ctl_mutex);
2091
2092 return i;
2093
2094out_cleanup_disk:
2095 put_disk(disk);
2096out_cleanup_tags:
2097 blk_mq_free_tag_set(set: &lo->tag_set);
2098out_free_idr:
2099 mutex_lock(&loop_ctl_mutex);
2100 idr_remove(&loop_index_idr, id: i);
2101 mutex_unlock(lock: &loop_ctl_mutex);
2102out_free_dev:
2103 kfree(objp: lo);
2104out:
2105 return err;
2106}
2107
2108static void loop_remove(struct loop_device *lo)
2109{
2110 /* Make this loop device unreachable from pathname. */
2111 del_gendisk(gp: lo->lo_disk);
2112 blk_mq_free_tag_set(set: &lo->tag_set);
2113
2114 mutex_lock(&loop_ctl_mutex);
2115 idr_remove(&loop_index_idr, id: lo->lo_number);
2116 mutex_unlock(lock: &loop_ctl_mutex);
2117
2118 put_disk(disk: lo->lo_disk);
2119}
2120
2121#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2122static void loop_probe(dev_t dev)
2123{
2124 int idx = MINOR(dev) >> part_shift;
2125
2126 if (max_loop_specified && max_loop && idx >= max_loop)
2127 return;
2128 loop_add(i: idx);
2129}
2130#else
2131#define loop_probe NULL
2132#endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2133
2134static int loop_control_remove(int idx)
2135{
2136 struct loop_device *lo;
2137 int ret;
2138
2139 if (idx < 0) {
2140 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2141 return -EINVAL;
2142 }
2143
2144 /* Hide this loop device for serialization. */
2145 ret = mutex_lock_killable(&loop_ctl_mutex);
2146 if (ret)
2147 return ret;
2148 lo = idr_find(&loop_index_idr, id: idx);
2149 if (!lo || !lo->idr_visible)
2150 ret = -ENODEV;
2151 else
2152 lo->idr_visible = false;
2153 mutex_unlock(lock: &loop_ctl_mutex);
2154 if (ret)
2155 return ret;
2156
2157 /* Check whether this loop device can be removed. */
2158 ret = mutex_lock_killable(&lo->lo_mutex);
2159 if (ret)
2160 goto mark_visible;
2161 if (lo->lo_state != Lo_unbound || disk_openers(disk: lo->lo_disk) > 0) {
2162 mutex_unlock(lock: &lo->lo_mutex);
2163 ret = -EBUSY;
2164 goto mark_visible;
2165 }
2166 /* Mark this loop device as no more bound, but not quite unbound yet */
2167 lo->lo_state = Lo_deleting;
2168 mutex_unlock(lock: &lo->lo_mutex);
2169
2170 loop_remove(lo);
2171 return 0;
2172
2173mark_visible:
2174 /* Show this loop device again. */
2175 mutex_lock(&loop_ctl_mutex);
2176 lo->idr_visible = true;
2177 mutex_unlock(lock: &loop_ctl_mutex);
2178 return ret;
2179}
2180
2181static int loop_control_get_free(int idx)
2182{
2183 struct loop_device *lo;
2184 int id, ret;
2185
2186 ret = mutex_lock_killable(&loop_ctl_mutex);
2187 if (ret)
2188 return ret;
2189 idr_for_each_entry(&loop_index_idr, lo, id) {
2190 /* Hitting a race results in creating a new loop device which is harmless. */
2191 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2192 goto found;
2193 }
2194 mutex_unlock(lock: &loop_ctl_mutex);
2195 return loop_add(i: -1);
2196found:
2197 mutex_unlock(lock: &loop_ctl_mutex);
2198 return id;
2199}
2200
2201static long loop_control_ioctl(struct file *file, unsigned int cmd,
2202 unsigned long parm)
2203{
2204 switch (cmd) {
2205 case LOOP_CTL_ADD:
2206 return loop_add(i: parm);
2207 case LOOP_CTL_REMOVE:
2208 return loop_control_remove(idx: parm);
2209 case LOOP_CTL_GET_FREE:
2210 return loop_control_get_free(idx: parm);
2211 default:
2212 return -ENOSYS;
2213 }
2214}
2215
2216static const struct file_operations loop_ctl_fops = {
2217 .open = nonseekable_open,
2218 .unlocked_ioctl = loop_control_ioctl,
2219 .compat_ioctl = loop_control_ioctl,
2220 .owner = THIS_MODULE,
2221 .llseek = noop_llseek,
2222};
2223
2224static struct miscdevice loop_misc = {
2225 .minor = LOOP_CTRL_MINOR,
2226 .name = "loop-control",
2227 .fops = &loop_ctl_fops,
2228};
2229
2230MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2231MODULE_ALIAS("devname:loop-control");
2232
2233static int __init loop_init(void)
2234{
2235 int i;
2236 int err;
2237
2238 part_shift = 0;
2239 if (max_part > 0) {
2240 part_shift = fls(x: max_part);
2241
2242 /*
2243 * Adjust max_part according to part_shift as it is exported
2244 * to user space so that user can decide correct minor number
2245 * if [s]he want to create more devices.
2246 *
2247 * Note that -1 is required because partition 0 is reserved
2248 * for the whole disk.
2249 */
2250 max_part = (1UL << part_shift) - 1;
2251 }
2252
2253 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2254 err = -EINVAL;
2255 goto err_out;
2256 }
2257
2258 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2259 err = -EINVAL;
2260 goto err_out;
2261 }
2262
2263 err = misc_register(misc: &loop_misc);
2264 if (err < 0)
2265 goto err_out;
2266
2267
2268 if (__register_blkdev(LOOP_MAJOR, name: "loop", probe: loop_probe)) {
2269 err = -EIO;
2270 goto misc_out;
2271 }
2272
2273 /* pre-create number of devices given by config or max_loop */
2274 for (i = 0; i < max_loop; i++)
2275 loop_add(i);
2276
2277 printk(KERN_INFO "loop: module loaded\n");
2278 return 0;
2279
2280misc_out:
2281 misc_deregister(misc: &loop_misc);
2282err_out:
2283 return err;
2284}
2285
2286static void __exit loop_exit(void)
2287{
2288 struct loop_device *lo;
2289 int id;
2290
2291 unregister_blkdev(LOOP_MAJOR, name: "loop");
2292 misc_deregister(misc: &loop_misc);
2293
2294 /*
2295 * There is no need to use loop_ctl_mutex here, for nobody else can
2296 * access loop_index_idr when this module is unloading (unless forced
2297 * module unloading is requested). If this is not a clean unloading,
2298 * we have no means to avoid kernel crash.
2299 */
2300 idr_for_each_entry(&loop_index_idr, lo, id)
2301 loop_remove(lo);
2302
2303 idr_destroy(&loop_index_idr);
2304}
2305
2306module_init(loop_init);
2307module_exit(loop_exit);
2308
2309#ifndef MODULE
2310static int __init max_loop_setup(char *str)
2311{
2312 max_loop = simple_strtol(str, NULL, 0);
2313#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2314 max_loop_specified = true;
2315#endif
2316 return 1;
2317}
2318
2319__setup("max_loop=", max_loop_setup);
2320#endif
2321

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