projects/linux·blob v6.3 · refs

2290 lines · 57.1 KiB

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