Line data Source code
1 : /*
2 : * linux/drivers/block/loop.c
3 : *
4 : * Written by Theodore Ts'o, 3/29/93
5 : *
6 : * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 : * permitted under the GNU General Public License.
8 : *
9 : * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 : * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
11 : *
12 : * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 : * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
14 : *
15 : * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
16 : *
17 : * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
18 : *
19 : * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
20 : *
21 : * Loadable modules and other fixes by AK, 1998
22 : *
23 : * Make real block number available to downstream transfer functions, enables
24 : * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 : * Reed H. Petty, rhp@draper.net
26 : *
27 : * Maximum number of loop devices now dynamic via max_loop module parameter.
28 : * Russell Kroll <rkroll@exploits.org> 19990701
29 : *
30 : * Maximum number of loop devices when compiled-in now selectable by passing
31 : * max_loop=<1-255> to the kernel on boot.
32 : * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
33 : *
34 : * Completely rewrite request handling to be make_request_fn style and
35 : * non blocking, pushing work to a helper thread. Lots of fixes from
36 : * Al Viro too.
37 : * Jens Axboe <axboe@suse.de>, Nov 2000
38 : *
39 : * Support up to 256 loop devices
40 : * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
41 : *
42 : * Support for falling back on the write file operation when the address space
43 : * operations write_begin is not available on the backing filesystem.
44 : * Anton Altaparmakov, 16 Feb 2005
45 : *
46 : * Still To Fix:
47 : * - Advisory locking is ignored here.
48 : * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 : *
50 : */
51 :
52 : #include <linux/module.h>
53 : #include <linux/moduleparam.h>
54 : #include <linux/sched.h>
55 : #include <linux/fs.h>
56 : #include <linux/file.h>
57 : #include <linux/stat.h>
58 : #include <linux/errno.h>
59 : #include <linux/major.h>
60 : #include <linux/wait.h>
61 : #include <linux/blkdev.h>
62 : #include <linux/blkpg.h>
63 : #include <linux/init.h>
64 : #include <linux/swap.h>
65 : #include <linux/slab.h>
66 : #include <linux/compat.h>
67 : #include <linux/suspend.h>
68 : #include <linux/freezer.h>
69 : #include <linux/mutex.h>
70 : #include <linux/writeback.h>
71 : #include <linux/completion.h>
72 : #include <linux/highmem.h>
73 : #include <linux/kthread.h>
74 : #include <linux/splice.h>
75 : #include <linux/sysfs.h>
76 : #include <linux/miscdevice.h>
77 : #include <linux/falloc.h>
78 : #include <linux/uio.h>
79 : #include <linux/ioprio.h>
80 : #include <linux/blk-cgroup.h>
81 :
82 : #include "loop.h"
83 :
84 : #include <linux/uaccess.h>
85 :
86 : static DEFINE_IDR(loop_index_idr);
87 : static DEFINE_MUTEX(loop_ctl_mutex);
88 :
89 : static int max_part;
90 : static int part_shift;
91 :
92 0 : static int transfer_xor(struct loop_device *lo, int cmd,
93 : struct page *raw_page, unsigned raw_off,
94 : struct page *loop_page, unsigned loop_off,
95 : int size, sector_t real_block)
96 : {
97 0 : char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 0 : char *loop_buf = kmap_atomic(loop_page) + loop_off;
99 0 : char *in, *out, *key;
100 0 : int i, keysize;
101 :
102 0 : if (cmd == READ) {
103 : in = raw_buf;
104 : out = loop_buf;
105 : } else {
106 0 : in = loop_buf;
107 0 : out = raw_buf;
108 : }
109 :
110 0 : key = lo->lo_encrypt_key;
111 0 : keysize = lo->lo_encrypt_key_size;
112 0 : for (i = 0; i < size; i++)
113 0 : *out++ = *in++ ^ key[(i & 511) % keysize];
114 :
115 0 : kunmap_atomic(loop_buf);
116 0 : kunmap_atomic(raw_buf);
117 0 : cond_resched();
118 0 : return 0;
119 : }
120 :
121 0 : static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122 : {
123 0 : if (unlikely(info->lo_encrypt_key_size <= 0))
124 0 : return -EINVAL;
125 : return 0;
126 : }
127 :
128 : static struct loop_func_table none_funcs = {
129 : .number = LO_CRYPT_NONE,
130 : };
131 :
132 : static struct loop_func_table xor_funcs = {
133 : .number = LO_CRYPT_XOR,
134 : .transfer = transfer_xor,
135 : .init = xor_init
136 : };
137 :
138 : /* xfer_funcs[0] is special - its release function is never called */
139 : static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
140 : &none_funcs,
141 : &xor_funcs
142 : };
143 :
144 0 : static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145 : {
146 0 : loff_t loopsize;
147 :
148 : /* Compute loopsize in bytes */
149 0 : loopsize = i_size_read(file->f_mapping->host);
150 0 : if (offset > 0)
151 0 : loopsize -= offset;
152 : /* offset is beyond i_size, weird but possible */
153 0 : if (loopsize < 0)
154 : return 0;
155 :
156 0 : if (sizelimit > 0 && sizelimit < loopsize)
157 0 : loopsize = sizelimit;
158 : /*
159 : * Unfortunately, if we want to do I/O on the device,
160 : * the number of 512-byte sectors has to fit into a sector_t.
161 : */
162 0 : return loopsize >> 9;
163 : }
164 :
165 0 : static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166 : {
167 0 : return get_size(lo->lo_offset, lo->lo_sizelimit, file);
168 : }
169 :
170 0 : static void __loop_update_dio(struct loop_device *lo, bool dio)
171 : {
172 0 : struct file *file = lo->lo_backing_file;
173 0 : struct address_space *mapping = file->f_mapping;
174 0 : struct inode *inode = mapping->host;
175 0 : unsigned short sb_bsize = 0;
176 0 : unsigned dio_align = 0;
177 0 : bool use_dio;
178 :
179 0 : if (inode->i_sb->s_bdev) {
180 0 : sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 0 : dio_align = sb_bsize - 1;
182 : }
183 :
184 : /*
185 : * We support direct I/O only if lo_offset is aligned with the
186 : * logical I/O size of backing device, and the logical block
187 : * size of loop is bigger than the backing device's and the loop
188 : * needn't transform transfer.
189 : *
190 : * TODO: the above condition may be loosed in the future, and
191 : * direct I/O may be switched runtime at that time because most
192 : * of requests in sane applications should be PAGE_SIZE aligned
193 : */
194 0 : if (dio) {
195 0 : if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 0 : !(lo->lo_offset & dio_align) &&
197 0 : mapping->a_ops->direct_IO &&
198 0 : !lo->transfer)
199 : use_dio = true;
200 : else
201 0 : use_dio = false;
202 : } else {
203 : use_dio = false;
204 : }
205 :
206 0 : if (lo->use_dio == use_dio)
207 : return;
208 :
209 : /* flush dirty pages before changing direct IO */
210 0 : vfs_fsync(file, 0);
211 :
212 : /*
213 : * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 : * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 : * will get updated by ioctl(LOOP_GET_STATUS)
216 : */
217 0 : if (lo->lo_state == Lo_bound)
218 0 : blk_mq_freeze_queue(lo->lo_queue);
219 0 : lo->use_dio = use_dio;
220 0 : if (use_dio) {
221 0 : blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222 0 : lo->lo_flags |= LO_FLAGS_DIRECT_IO;
223 : } else {
224 0 : blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225 0 : lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
226 : }
227 0 : if (lo->lo_state == Lo_bound)
228 0 : blk_mq_unfreeze_queue(lo->lo_queue);
229 : }
230 :
231 : /**
232 : * loop_validate_block_size() - validates the passed in block size
233 : * @bsize: size to validate
234 : */
235 : static int
236 0 : loop_validate_block_size(unsigned short bsize)
237 : {
238 0 : if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
239 0 : return -EINVAL;
240 :
241 : return 0;
242 : }
243 :
244 : /**
245 : * loop_set_size() - sets device size and notifies userspace
246 : * @lo: struct loop_device to set the size for
247 : * @size: new size of the loop device
248 : *
249 : * Callers must validate that the size passed into this function fits into
250 : * a sector_t, eg using loop_validate_size()
251 : */
252 0 : static void loop_set_size(struct loop_device *lo, loff_t size)
253 : {
254 0 : if (!set_capacity_and_notify(lo->lo_disk, size))
255 0 : kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
256 0 : }
257 :
258 : static inline int
259 0 : lo_do_transfer(struct loop_device *lo, int cmd,
260 : struct page *rpage, unsigned roffs,
261 : struct page *lpage, unsigned loffs,
262 : int size, sector_t rblock)
263 : {
264 0 : int ret;
265 :
266 0 : ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
267 0 : if (likely(!ret))
268 : return 0;
269 :
270 0 : printk_ratelimited(KERN_ERR
271 : "loop: Transfer error at byte offset %llu, length %i.\n",
272 : (unsigned long long)rblock << 9, size);
273 : return ret;
274 : }
275 :
276 0 : static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
277 : {
278 0 : struct iov_iter i;
279 0 : ssize_t bw;
280 :
281 0 : iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
282 :
283 0 : file_start_write(file);
284 0 : bw = vfs_iter_write(file, &i, ppos, 0);
285 0 : file_end_write(file);
286 :
287 0 : if (likely(bw == bvec->bv_len))
288 : return 0;
289 :
290 0 : printk_ratelimited(KERN_ERR
291 : "loop: Write error at byte offset %llu, length %i.\n",
292 : (unsigned long long)*ppos, bvec->bv_len);
293 0 : if (bw >= 0)
294 0 : bw = -EIO;
295 0 : return bw;
296 : }
297 :
298 0 : static int lo_write_simple(struct loop_device *lo, struct request *rq,
299 : loff_t pos)
300 : {
301 0 : struct bio_vec bvec;
302 0 : struct req_iterator iter;
303 0 : int ret = 0;
304 :
305 0 : rq_for_each_segment(bvec, rq, iter) {
306 0 : ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
307 0 : if (ret < 0)
308 : break;
309 0 : cond_resched();
310 : }
311 :
312 0 : return ret;
313 : }
314 :
315 : /*
316 : * This is the slow, transforming version that needs to double buffer the
317 : * data as it cannot do the transformations in place without having direct
318 : * access to the destination pages of the backing file.
319 : */
320 0 : static int lo_write_transfer(struct loop_device *lo, struct request *rq,
321 : loff_t pos)
322 : {
323 0 : struct bio_vec bvec, b;
324 0 : struct req_iterator iter;
325 0 : struct page *page;
326 0 : int ret = 0;
327 :
328 0 : page = alloc_page(GFP_NOIO);
329 0 : if (unlikely(!page))
330 : return -ENOMEM;
331 :
332 0 : rq_for_each_segment(bvec, rq, iter) {
333 0 : ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
334 0 : bvec.bv_offset, bvec.bv_len, pos >> 9);
335 0 : if (unlikely(ret))
336 : break;
337 :
338 0 : b.bv_page = page;
339 0 : b.bv_offset = 0;
340 0 : b.bv_len = bvec.bv_len;
341 0 : ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
342 0 : if (ret < 0)
343 : break;
344 : }
345 :
346 0 : __free_page(page);
347 0 : return ret;
348 : }
349 :
350 0 : static int lo_read_simple(struct loop_device *lo, struct request *rq,
351 : loff_t pos)
352 : {
353 0 : struct bio_vec bvec;
354 0 : struct req_iterator iter;
355 0 : struct iov_iter i;
356 0 : ssize_t len;
357 :
358 0 : rq_for_each_segment(bvec, rq, iter) {
359 0 : iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
360 0 : len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
361 0 : if (len < 0)
362 0 : return len;
363 :
364 0 : flush_dcache_page(bvec.bv_page);
365 :
366 0 : if (len != bvec.bv_len) {
367 0 : struct bio *bio;
368 :
369 0 : __rq_for_each_bio(bio, rq)
370 0 : zero_fill_bio(bio);
371 : break;
372 : }
373 0 : cond_resched();
374 : }
375 :
376 : return 0;
377 : }
378 :
379 0 : static int lo_read_transfer(struct loop_device *lo, struct request *rq,
380 : loff_t pos)
381 : {
382 0 : struct bio_vec bvec, b;
383 0 : struct req_iterator iter;
384 0 : struct iov_iter i;
385 0 : struct page *page;
386 0 : ssize_t len;
387 0 : int ret = 0;
388 :
389 0 : page = alloc_page(GFP_NOIO);
390 0 : if (unlikely(!page))
391 : return -ENOMEM;
392 :
393 0 : rq_for_each_segment(bvec, rq, iter) {
394 0 : loff_t offset = pos;
395 :
396 0 : b.bv_page = page;
397 0 : b.bv_offset = 0;
398 0 : b.bv_len = bvec.bv_len;
399 :
400 0 : iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
401 0 : len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
402 0 : if (len < 0) {
403 0 : ret = len;
404 0 : goto out_free_page;
405 : }
406 :
407 0 : ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
408 0 : bvec.bv_offset, len, offset >> 9);
409 0 : if (ret)
410 0 : goto out_free_page;
411 :
412 0 : flush_dcache_page(bvec.bv_page);
413 :
414 0 : if (len != bvec.bv_len) {
415 0 : struct bio *bio;
416 :
417 0 : __rq_for_each_bio(bio, rq)
418 0 : zero_fill_bio(bio);
419 : break;
420 : }
421 : }
422 :
423 : ret = 0;
424 0 : out_free_page:
425 0 : __free_page(page);
426 0 : return ret;
427 : }
428 :
429 0 : static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
430 : int mode)
431 : {
432 : /*
433 : * We use fallocate to manipulate the space mappings used by the image
434 : * a.k.a. discard/zerorange. However we do not support this if
435 : * encryption is enabled, because it may give an attacker useful
436 : * information.
437 : */
438 0 : struct file *file = lo->lo_backing_file;
439 0 : struct request_queue *q = lo->lo_queue;
440 0 : int ret;
441 :
442 0 : mode |= FALLOC_FL_KEEP_SIZE;
443 :
444 0 : if (!blk_queue_discard(q)) {
445 0 : ret = -EOPNOTSUPP;
446 0 : goto out;
447 : }
448 :
449 0 : ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
450 0 : if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
451 0 : ret = -EIO;
452 0 : out:
453 0 : return ret;
454 : }
455 :
456 0 : static int lo_req_flush(struct loop_device *lo, struct request *rq)
457 : {
458 0 : struct file *file = lo->lo_backing_file;
459 0 : int ret = vfs_fsync(file, 0);
460 0 : if (unlikely(ret && ret != -EINVAL))
461 0 : ret = -EIO;
462 :
463 0 : return ret;
464 : }
465 :
466 0 : static void lo_complete_rq(struct request *rq)
467 : {
468 0 : struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
469 0 : blk_status_t ret = BLK_STS_OK;
470 :
471 0 : if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
472 0 : req_op(rq) != REQ_OP_READ) {
473 0 : if (cmd->ret < 0)
474 0 : ret = errno_to_blk_status(cmd->ret);
475 0 : goto end_io;
476 : }
477 :
478 : /*
479 : * Short READ - if we got some data, advance our request and
480 : * retry it. If we got no data, end the rest with EIO.
481 : */
482 0 : if (cmd->ret) {
483 0 : blk_update_request(rq, BLK_STS_OK, cmd->ret);
484 0 : cmd->ret = 0;
485 0 : blk_mq_requeue_request(rq, true);
486 : } else {
487 0 : if (cmd->use_aio) {
488 0 : struct bio *bio = rq->bio;
489 :
490 0 : while (bio) {
491 0 : zero_fill_bio(bio);
492 0 : bio = bio->bi_next;
493 : }
494 : }
495 : ret = BLK_STS_IOERR;
496 0 : end_io:
497 0 : blk_mq_end_request(rq, ret);
498 : }
499 0 : }
500 :
501 0 : static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
502 : {
503 0 : struct request *rq = blk_mq_rq_from_pdu(cmd);
504 :
505 0 : if (!atomic_dec_and_test(&cmd->ref))
506 : return;
507 0 : kfree(cmd->bvec);
508 0 : cmd->bvec = NULL;
509 0 : if (likely(!blk_should_fake_timeout(rq->q)))
510 0 : blk_mq_complete_request(rq);
511 : }
512 :
513 0 : static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
514 : {
515 0 : struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
516 :
517 0 : if (cmd->css)
518 0 : css_put(cmd->css);
519 0 : cmd->ret = ret;
520 0 : lo_rw_aio_do_completion(cmd);
521 0 : }
522 :
523 0 : static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
524 : loff_t pos, bool rw)
525 : {
526 0 : struct iov_iter iter;
527 0 : struct req_iterator rq_iter;
528 0 : struct bio_vec *bvec;
529 0 : struct request *rq = blk_mq_rq_from_pdu(cmd);
530 0 : struct bio *bio = rq->bio;
531 0 : struct file *file = lo->lo_backing_file;
532 0 : struct bio_vec tmp;
533 0 : unsigned int offset;
534 0 : int nr_bvec = 0;
535 0 : int ret;
536 :
537 0 : rq_for_each_bvec(tmp, rq, rq_iter)
538 0 : nr_bvec++;
539 :
540 0 : if (rq->bio != rq->biotail) {
541 :
542 0 : bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
543 : GFP_NOIO);
544 0 : if (!bvec)
545 : return -EIO;
546 0 : cmd->bvec = bvec;
547 :
548 : /*
549 : * The bios of the request may be started from the middle of
550 : * the 'bvec' because of bio splitting, so we can't directly
551 : * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
552 : * API will take care of all details for us.
553 : */
554 0 : rq_for_each_bvec(tmp, rq, rq_iter) {
555 0 : *bvec = tmp;
556 0 : bvec++;
557 : }
558 0 : bvec = cmd->bvec;
559 0 : offset = 0;
560 : } else {
561 : /*
562 : * Same here, this bio may be started from the middle of the
563 : * 'bvec' because of bio splitting, so offset from the bvec
564 : * must be passed to iov iterator
565 : */
566 0 : offset = bio->bi_iter.bi_bvec_done;
567 0 : bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
568 : }
569 0 : atomic_set(&cmd->ref, 2);
570 :
571 0 : iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
572 0 : iter.iov_offset = offset;
573 :
574 0 : cmd->iocb.ki_pos = pos;
575 0 : cmd->iocb.ki_filp = file;
576 0 : cmd->iocb.ki_complete = lo_rw_aio_complete;
577 0 : cmd->iocb.ki_flags = IOCB_DIRECT;
578 0 : cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
579 0 : if (cmd->css)
580 0 : kthread_associate_blkcg(cmd->css);
581 :
582 0 : if (rw == WRITE)
583 0 : ret = call_write_iter(file, &cmd->iocb, &iter);
584 : else
585 0 : ret = call_read_iter(file, &cmd->iocb, &iter);
586 :
587 0 : lo_rw_aio_do_completion(cmd);
588 0 : kthread_associate_blkcg(NULL);
589 :
590 0 : if (ret != -EIOCBQUEUED)
591 0 : cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
592 : return 0;
593 : }
594 :
595 0 : static int do_req_filebacked(struct loop_device *lo, struct request *rq)
596 : {
597 0 : struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
598 0 : loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
599 :
600 : /*
601 : * lo_write_simple and lo_read_simple should have been covered
602 : * by io submit style function like lo_rw_aio(), one blocker
603 : * is that lo_read_simple() need to call flush_dcache_page after
604 : * the page is written from kernel, and it isn't easy to handle
605 : * this in io submit style function which submits all segments
606 : * of the req at one time. And direct read IO doesn't need to
607 : * run flush_dcache_page().
608 : */
609 0 : switch (req_op(rq)) {
610 0 : case REQ_OP_FLUSH:
611 0 : return lo_req_flush(lo, rq);
612 0 : case REQ_OP_WRITE_ZEROES:
613 : /*
614 : * If the caller doesn't want deallocation, call zeroout to
615 : * write zeroes the range. Otherwise, punch them out.
616 : */
617 0 : return lo_fallocate(lo, rq, pos,
618 : (rq->cmd_flags & REQ_NOUNMAP) ?
619 : FALLOC_FL_ZERO_RANGE :
620 : FALLOC_FL_PUNCH_HOLE);
621 0 : case REQ_OP_DISCARD:
622 0 : return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
623 0 : case REQ_OP_WRITE:
624 0 : if (lo->transfer)
625 0 : return lo_write_transfer(lo, rq, pos);
626 0 : else if (cmd->use_aio)
627 0 : return lo_rw_aio(lo, cmd, pos, WRITE);
628 : else
629 0 : return lo_write_simple(lo, rq, pos);
630 0 : case REQ_OP_READ:
631 0 : if (lo->transfer)
632 0 : return lo_read_transfer(lo, rq, pos);
633 0 : else if (cmd->use_aio)
634 0 : return lo_rw_aio(lo, cmd, pos, READ);
635 : else
636 0 : return lo_read_simple(lo, rq, pos);
637 : default:
638 0 : WARN_ON_ONCE(1);
639 0 : return -EIO;
640 : }
641 : }
642 :
643 0 : static inline void loop_update_dio(struct loop_device *lo)
644 : {
645 0 : __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
646 0 : lo->use_dio);
647 0 : }
648 :
649 0 : static void loop_reread_partitions(struct loop_device *lo,
650 : struct block_device *bdev)
651 : {
652 0 : int rc;
653 :
654 0 : mutex_lock(&bdev->bd_mutex);
655 0 : rc = bdev_disk_changed(bdev, false);
656 0 : mutex_unlock(&bdev->bd_mutex);
657 0 : if (rc)
658 0 : pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
659 : __func__, lo->lo_number, lo->lo_file_name, rc);
660 0 : }
661 :
662 0 : static inline int is_loop_device(struct file *file)
663 : {
664 0 : struct inode *i = file->f_mapping->host;
665 :
666 0 : return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
667 : }
668 :
669 0 : static int loop_validate_file(struct file *file, struct block_device *bdev)
670 : {
671 0 : struct inode *inode = file->f_mapping->host;
672 0 : struct file *f = file;
673 :
674 : /* Avoid recursion */
675 0 : while (is_loop_device(f)) {
676 0 : struct loop_device *l;
677 :
678 0 : if (f->f_mapping->host->i_rdev == bdev->bd_dev)
679 : return -EBADF;
680 :
681 0 : l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
682 0 : if (l->lo_state != Lo_bound) {
683 : return -EINVAL;
684 : }
685 0 : f = l->lo_backing_file;
686 : }
687 0 : if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
688 0 : return -EINVAL;
689 : return 0;
690 : }
691 :
692 : /*
693 : * loop_change_fd switched the backing store of a loopback device to
694 : * a new file. This is useful for operating system installers to free up
695 : * the original file and in High Availability environments to switch to
696 : * an alternative location for the content in case of server meltdown.
697 : * This can only work if the loop device is used read-only, and if the
698 : * new backing store is the same size and type as the old backing store.
699 : */
700 0 : static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
701 : unsigned int arg)
702 : {
703 0 : struct file *file = NULL, *old_file;
704 0 : int error;
705 0 : bool partscan;
706 :
707 0 : error = mutex_lock_killable(&lo->lo_mutex);
708 0 : if (error)
709 : return error;
710 0 : error = -ENXIO;
711 0 : if (lo->lo_state != Lo_bound)
712 0 : goto out_err;
713 :
714 : /* the loop device has to be read-only */
715 0 : error = -EINVAL;
716 0 : if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
717 0 : goto out_err;
718 :
719 0 : error = -EBADF;
720 0 : file = fget(arg);
721 0 : if (!file)
722 0 : goto out_err;
723 :
724 0 : error = loop_validate_file(file, bdev);
725 0 : if (error)
726 0 : goto out_err;
727 :
728 0 : old_file = lo->lo_backing_file;
729 :
730 0 : error = -EINVAL;
731 :
732 : /* size of the new backing store needs to be the same */
733 0 : if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
734 0 : goto out_err;
735 :
736 : /* and ... switch */
737 0 : blk_mq_freeze_queue(lo->lo_queue);
738 0 : mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
739 0 : lo->lo_backing_file = file;
740 0 : lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
741 0 : mapping_set_gfp_mask(file->f_mapping,
742 : lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
743 0 : loop_update_dio(lo);
744 0 : blk_mq_unfreeze_queue(lo->lo_queue);
745 0 : partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
746 0 : mutex_unlock(&lo->lo_mutex);
747 : /*
748 : * We must drop file reference outside of lo_mutex as dropping
749 : * the file ref can take bd_mutex which creates circular locking
750 : * dependency.
751 : */
752 0 : fput(old_file);
753 0 : if (partscan)
754 0 : loop_reread_partitions(lo, bdev);
755 : return 0;
756 :
757 0 : out_err:
758 0 : mutex_unlock(&lo->lo_mutex);
759 0 : if (file)
760 0 : fput(file);
761 : return error;
762 : }
763 :
764 : /* loop sysfs attributes */
765 :
766 0 : static ssize_t loop_attr_show(struct device *dev, char *page,
767 : ssize_t (*callback)(struct loop_device *, char *))
768 : {
769 0 : struct gendisk *disk = dev_to_disk(dev);
770 0 : struct loop_device *lo = disk->private_data;
771 :
772 0 : return callback(lo, page);
773 : }
774 :
775 : #define LOOP_ATTR_RO(_name) \
776 : static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
777 : static ssize_t loop_attr_do_show_##_name(struct device *d, \
778 : struct device_attribute *attr, char *b) \
779 : { \
780 : return loop_attr_show(d, b, loop_attr_##_name##_show); \
781 : } \
782 : static struct device_attribute loop_attr_##_name = \
783 : __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
784 :
785 0 : static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
786 : {
787 0 : ssize_t ret;
788 0 : char *p = NULL;
789 :
790 0 : spin_lock_irq(&lo->lo_lock);
791 0 : if (lo->lo_backing_file)
792 0 : p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
793 0 : spin_unlock_irq(&lo->lo_lock);
794 :
795 0 : if (IS_ERR_OR_NULL(p))
796 0 : ret = PTR_ERR(p);
797 : else {
798 0 : ret = strlen(p);
799 0 : memmove(buf, p, ret);
800 0 : buf[ret++] = '\n';
801 0 : buf[ret] = 0;
802 : }
803 :
804 0 : return ret;
805 : }
806 :
807 0 : static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
808 : {
809 0 : return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
810 : }
811 :
812 0 : static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
813 : {
814 0 : return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
815 : }
816 :
817 0 : static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
818 : {
819 0 : int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
820 :
821 0 : return sprintf(buf, "%s\n", autoclear ? "1" : "0");
822 : }
823 :
824 0 : static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
825 : {
826 0 : int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
827 :
828 0 : return sprintf(buf, "%s\n", partscan ? "1" : "0");
829 : }
830 :
831 0 : static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
832 : {
833 0 : int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
834 :
835 0 : return sprintf(buf, "%s\n", dio ? "1" : "0");
836 : }
837 :
838 0 : LOOP_ATTR_RO(backing_file);
839 0 : LOOP_ATTR_RO(offset);
840 0 : LOOP_ATTR_RO(sizelimit);
841 0 : LOOP_ATTR_RO(autoclear);
842 0 : LOOP_ATTR_RO(partscan);
843 0 : LOOP_ATTR_RO(dio);
844 :
845 : static struct attribute *loop_attrs[] = {
846 : &loop_attr_backing_file.attr,
847 : &loop_attr_offset.attr,
848 : &loop_attr_sizelimit.attr,
849 : &loop_attr_autoclear.attr,
850 : &loop_attr_partscan.attr,
851 : &loop_attr_dio.attr,
852 : NULL,
853 : };
854 :
855 : static struct attribute_group loop_attribute_group = {
856 : .name = "loop",
857 : .attrs= loop_attrs,
858 : };
859 :
860 0 : static void loop_sysfs_init(struct loop_device *lo)
861 : {
862 0 : lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
863 : &loop_attribute_group);
864 0 : }
865 :
866 0 : static void loop_sysfs_exit(struct loop_device *lo)
867 : {
868 0 : if (lo->sysfs_inited)
869 0 : sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
870 : &loop_attribute_group);
871 0 : }
872 :
873 0 : static void loop_config_discard(struct loop_device *lo)
874 : {
875 0 : struct file *file = lo->lo_backing_file;
876 0 : struct inode *inode = file->f_mapping->host;
877 0 : struct request_queue *q = lo->lo_queue;
878 0 : u32 granularity, max_discard_sectors;
879 :
880 : /*
881 : * If the backing device is a block device, mirror its zeroing
882 : * capability. Set the discard sectors to the block device's zeroing
883 : * capabilities because loop discards result in blkdev_issue_zeroout(),
884 : * not blkdev_issue_discard(). This maintains consistent behavior with
885 : * file-backed loop devices: discarded regions read back as zero.
886 : */
887 0 : if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
888 0 : struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
889 :
890 0 : max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
891 0 : granularity = backingq->limits.discard_granularity ?:
892 0 : queue_physical_block_size(backingq);
893 :
894 : /*
895 : * We use punch hole to reclaim the free space used by the
896 : * image a.k.a. discard. However we do not support discard if
897 : * encryption is enabled, because it may give an attacker
898 : * useful information.
899 : */
900 0 : } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
901 : max_discard_sectors = 0;
902 : granularity = 0;
903 :
904 : } else {
905 0 : max_discard_sectors = UINT_MAX >> 9;
906 0 : granularity = inode->i_sb->s_blocksize;
907 : }
908 :
909 0 : if (max_discard_sectors) {
910 0 : q->limits.discard_granularity = granularity;
911 0 : blk_queue_max_discard_sectors(q, max_discard_sectors);
912 0 : blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
913 0 : blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
914 : } else {
915 0 : q->limits.discard_granularity = 0;
916 0 : blk_queue_max_discard_sectors(q, 0);
917 0 : blk_queue_max_write_zeroes_sectors(q, 0);
918 0 : blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
919 : }
920 0 : q->limits.discard_alignment = 0;
921 0 : }
922 :
923 0 : static void loop_unprepare_queue(struct loop_device *lo)
924 : {
925 0 : kthread_flush_worker(&lo->worker);
926 0 : kthread_stop(lo->worker_task);
927 0 : }
928 :
929 0 : static int loop_kthread_worker_fn(void *worker_ptr)
930 : {
931 0 : current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
932 0 : return kthread_worker_fn(worker_ptr);
933 : }
934 :
935 0 : static int loop_prepare_queue(struct loop_device *lo)
936 : {
937 0 : kthread_init_worker(&lo->worker);
938 0 : lo->worker_task = kthread_run(loop_kthread_worker_fn,
939 : &lo->worker, "loop%d", lo->lo_number);
940 0 : if (IS_ERR(lo->worker_task))
941 : return -ENOMEM;
942 0 : set_user_nice(lo->worker_task, MIN_NICE);
943 0 : return 0;
944 : }
945 :
946 0 : static void loop_update_rotational(struct loop_device *lo)
947 : {
948 0 : struct file *file = lo->lo_backing_file;
949 0 : struct inode *file_inode = file->f_mapping->host;
950 0 : struct block_device *file_bdev = file_inode->i_sb->s_bdev;
951 0 : struct request_queue *q = lo->lo_queue;
952 0 : bool nonrot = true;
953 :
954 : /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
955 0 : if (file_bdev)
956 0 : nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
957 :
958 0 : if (nonrot)
959 0 : blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
960 : else
961 0 : blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
962 0 : }
963 :
964 : static int
965 0 : loop_release_xfer(struct loop_device *lo)
966 : {
967 0 : int err = 0;
968 0 : struct loop_func_table *xfer = lo->lo_encryption;
969 :
970 0 : if (xfer) {
971 0 : if (xfer->release)
972 0 : err = xfer->release(lo);
973 0 : lo->transfer = NULL;
974 0 : lo->lo_encryption = NULL;
975 0 : module_put(xfer->owner);
976 : }
977 0 : return err;
978 : }
979 :
980 : static int
981 0 : loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
982 : const struct loop_info64 *i)
983 : {
984 0 : int err = 0;
985 :
986 0 : if (xfer) {
987 0 : struct module *owner = xfer->owner;
988 :
989 0 : if (!try_module_get(owner))
990 : return -EINVAL;
991 0 : if (xfer->init)
992 0 : err = xfer->init(lo, i);
993 0 : if (err)
994 0 : module_put(owner);
995 : else
996 0 : lo->lo_encryption = xfer;
997 : }
998 0 : return err;
999 : }
1000 :
1001 : /**
1002 : * loop_set_status_from_info - configure device from loop_info
1003 : * @lo: struct loop_device to configure
1004 : * @info: struct loop_info64 to configure the device with
1005 : *
1006 : * Configures the loop device parameters according to the passed
1007 : * in loop_info64 configuration.
1008 : */
1009 : static int
1010 0 : loop_set_status_from_info(struct loop_device *lo,
1011 : const struct loop_info64 *info)
1012 : {
1013 0 : int err;
1014 0 : struct loop_func_table *xfer;
1015 0 : kuid_t uid = current_uid();
1016 :
1017 0 : if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1018 : return -EINVAL;
1019 :
1020 0 : err = loop_release_xfer(lo);
1021 0 : if (err)
1022 : return err;
1023 :
1024 0 : if (info->lo_encrypt_type) {
1025 0 : unsigned int type = info->lo_encrypt_type;
1026 :
1027 0 : if (type >= MAX_LO_CRYPT)
1028 : return -EINVAL;
1029 0 : xfer = xfer_funcs[type];
1030 0 : if (xfer == NULL)
1031 : return -EINVAL;
1032 : } else
1033 : xfer = NULL;
1034 :
1035 0 : err = loop_init_xfer(lo, xfer, info);
1036 0 : if (err)
1037 : return err;
1038 :
1039 0 : lo->lo_offset = info->lo_offset;
1040 0 : lo->lo_sizelimit = info->lo_sizelimit;
1041 0 : memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1042 0 : memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1043 0 : lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1044 0 : lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1045 :
1046 0 : if (!xfer)
1047 0 : xfer = &none_funcs;
1048 0 : lo->transfer = xfer->transfer;
1049 0 : lo->ioctl = xfer->ioctl;
1050 :
1051 0 : lo->lo_flags = info->lo_flags;
1052 :
1053 0 : lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1054 0 : lo->lo_init[0] = info->lo_init[0];
1055 0 : lo->lo_init[1] = info->lo_init[1];
1056 0 : if (info->lo_encrypt_key_size) {
1057 0 : memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1058 : info->lo_encrypt_key_size);
1059 0 : lo->lo_key_owner = uid;
1060 : }
1061 :
1062 : return 0;
1063 : }
1064 :
1065 0 : static int loop_configure(struct loop_device *lo, fmode_t mode,
1066 : struct block_device *bdev,
1067 : const struct loop_config *config)
1068 : {
1069 0 : struct file *file;
1070 0 : struct inode *inode;
1071 0 : struct address_space *mapping;
1072 0 : int error;
1073 0 : loff_t size;
1074 0 : bool partscan;
1075 0 : unsigned short bsize;
1076 :
1077 : /* This is safe, since we have a reference from open(). */
1078 0 : __module_get(THIS_MODULE);
1079 :
1080 0 : error = -EBADF;
1081 0 : file = fget(config->fd);
1082 0 : if (!file)
1083 0 : goto out;
1084 :
1085 : /*
1086 : * If we don't hold exclusive handle for the device, upgrade to it
1087 : * here to avoid changing device under exclusive owner.
1088 : */
1089 0 : if (!(mode & FMODE_EXCL)) {
1090 0 : error = bd_prepare_to_claim(bdev, loop_configure);
1091 0 : if (error)
1092 0 : goto out_putf;
1093 : }
1094 :
1095 0 : error = mutex_lock_killable(&lo->lo_mutex);
1096 0 : if (error)
1097 0 : goto out_bdev;
1098 :
1099 0 : error = -EBUSY;
1100 0 : if (lo->lo_state != Lo_unbound)
1101 0 : goto out_unlock;
1102 :
1103 0 : error = loop_validate_file(file, bdev);
1104 0 : if (error)
1105 0 : goto out_unlock;
1106 :
1107 0 : mapping = file->f_mapping;
1108 0 : inode = mapping->host;
1109 :
1110 0 : if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1111 0 : error = -EINVAL;
1112 0 : goto out_unlock;
1113 : }
1114 :
1115 0 : if (config->block_size) {
1116 0 : error = loop_validate_block_size(config->block_size);
1117 0 : if (error)
1118 0 : goto out_unlock;
1119 : }
1120 :
1121 0 : error = loop_set_status_from_info(lo, &config->info);
1122 0 : if (error)
1123 0 : goto out_unlock;
1124 :
1125 0 : if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1126 0 : !file->f_op->write_iter)
1127 0 : lo->lo_flags |= LO_FLAGS_READ_ONLY;
1128 :
1129 0 : error = loop_prepare_queue(lo);
1130 0 : if (error)
1131 0 : goto out_unlock;
1132 :
1133 0 : set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1134 :
1135 0 : lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1136 0 : lo->lo_device = bdev;
1137 0 : lo->lo_backing_file = file;
1138 0 : lo->old_gfp_mask = mapping_gfp_mask(mapping);
1139 0 : mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1140 :
1141 0 : if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1142 0 : blk_queue_write_cache(lo->lo_queue, true, false);
1143 :
1144 0 : if (config->block_size)
1145 0 : bsize = config->block_size;
1146 0 : else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1147 : /* In case of direct I/O, match underlying block size */
1148 0 : bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1149 : else
1150 : bsize = 512;
1151 :
1152 0 : blk_queue_logical_block_size(lo->lo_queue, bsize);
1153 0 : blk_queue_physical_block_size(lo->lo_queue, bsize);
1154 0 : blk_queue_io_min(lo->lo_queue, bsize);
1155 :
1156 0 : loop_update_rotational(lo);
1157 0 : loop_update_dio(lo);
1158 0 : loop_sysfs_init(lo);
1159 :
1160 0 : size = get_loop_size(lo, file);
1161 0 : loop_set_size(lo, size);
1162 :
1163 0 : lo->lo_state = Lo_bound;
1164 0 : if (part_shift)
1165 0 : lo->lo_flags |= LO_FLAGS_PARTSCAN;
1166 0 : partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1167 0 : if (partscan)
1168 0 : lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1169 :
1170 : /* Grab the block_device to prevent its destruction after we
1171 : * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1172 : */
1173 0 : bdgrab(bdev);
1174 0 : mutex_unlock(&lo->lo_mutex);
1175 0 : if (partscan)
1176 0 : loop_reread_partitions(lo, bdev);
1177 0 : if (!(mode & FMODE_EXCL))
1178 0 : bd_abort_claiming(bdev, loop_configure);
1179 : return 0;
1180 :
1181 0 : out_unlock:
1182 0 : mutex_unlock(&lo->lo_mutex);
1183 0 : out_bdev:
1184 0 : if (!(mode & FMODE_EXCL))
1185 0 : bd_abort_claiming(bdev, loop_configure);
1186 0 : out_putf:
1187 0 : fput(file);
1188 : out:
1189 : /* This is safe: open() is still holding a reference. */
1190 0 : module_put(THIS_MODULE);
1191 : return error;
1192 : }
1193 :
1194 0 : static int __loop_clr_fd(struct loop_device *lo, bool release)
1195 : {
1196 0 : struct file *filp = NULL;
1197 0 : gfp_t gfp = lo->old_gfp_mask;
1198 0 : struct block_device *bdev = lo->lo_device;
1199 0 : int err = 0;
1200 0 : bool partscan = false;
1201 0 : int lo_number;
1202 :
1203 0 : mutex_lock(&lo->lo_mutex);
1204 0 : if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1205 0 : err = -ENXIO;
1206 0 : goto out_unlock;
1207 : }
1208 :
1209 0 : filp = lo->lo_backing_file;
1210 0 : if (filp == NULL) {
1211 0 : err = -EINVAL;
1212 0 : goto out_unlock;
1213 : }
1214 :
1215 0 : if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1216 0 : blk_queue_write_cache(lo->lo_queue, false, false);
1217 :
1218 : /* freeze request queue during the transition */
1219 0 : blk_mq_freeze_queue(lo->lo_queue);
1220 :
1221 0 : spin_lock_irq(&lo->lo_lock);
1222 0 : lo->lo_backing_file = NULL;
1223 0 : spin_unlock_irq(&lo->lo_lock);
1224 :
1225 0 : loop_release_xfer(lo);
1226 0 : lo->transfer = NULL;
1227 0 : lo->ioctl = NULL;
1228 0 : lo->lo_device = NULL;
1229 0 : lo->lo_encryption = NULL;
1230 0 : lo->lo_offset = 0;
1231 0 : lo->lo_sizelimit = 0;
1232 0 : lo->lo_encrypt_key_size = 0;
1233 0 : memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1234 0 : memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1235 0 : memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1236 0 : blk_queue_logical_block_size(lo->lo_queue, 512);
1237 0 : blk_queue_physical_block_size(lo->lo_queue, 512);
1238 0 : blk_queue_io_min(lo->lo_queue, 512);
1239 0 : if (bdev) {
1240 0 : bdput(bdev);
1241 0 : invalidate_bdev(bdev);
1242 0 : bdev->bd_inode->i_mapping->wb_err = 0;
1243 : }
1244 0 : set_capacity(lo->lo_disk, 0);
1245 0 : loop_sysfs_exit(lo);
1246 0 : if (bdev) {
1247 : /* let user-space know about this change */
1248 0 : kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1249 : }
1250 0 : mapping_set_gfp_mask(filp->f_mapping, gfp);
1251 : /* This is safe: open() is still holding a reference. */
1252 0 : module_put(THIS_MODULE);
1253 0 : blk_mq_unfreeze_queue(lo->lo_queue);
1254 :
1255 0 : partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1256 0 : lo_number = lo->lo_number;
1257 0 : loop_unprepare_queue(lo);
1258 0 : out_unlock:
1259 0 : mutex_unlock(&lo->lo_mutex);
1260 0 : if (partscan) {
1261 : /*
1262 : * bd_mutex has been held already in release path, so don't
1263 : * acquire it if this function is called in such case.
1264 : *
1265 : * If the reread partition isn't from release path, lo_refcnt
1266 : * must be at least one and it can only become zero when the
1267 : * current holder is released.
1268 : */
1269 0 : if (!release)
1270 0 : mutex_lock(&bdev->bd_mutex);
1271 0 : err = bdev_disk_changed(bdev, false);
1272 0 : if (!release)
1273 0 : mutex_unlock(&bdev->bd_mutex);
1274 0 : if (err)
1275 0 : pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1276 : __func__, lo_number, err);
1277 : /* Device is gone, no point in returning error */
1278 : err = 0;
1279 : }
1280 :
1281 : /*
1282 : * lo->lo_state is set to Lo_unbound here after above partscan has
1283 : * finished.
1284 : *
1285 : * There cannot be anybody else entering __loop_clr_fd() as
1286 : * lo->lo_backing_file is already cleared and Lo_rundown state
1287 : * protects us from all the other places trying to change the 'lo'
1288 : * device.
1289 : */
1290 0 : mutex_lock(&lo->lo_mutex);
1291 0 : lo->lo_flags = 0;
1292 0 : if (!part_shift)
1293 0 : lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1294 0 : lo->lo_state = Lo_unbound;
1295 0 : mutex_unlock(&lo->lo_mutex);
1296 :
1297 : /*
1298 : * Need not hold lo_mutex to fput backing file. Calling fput holding
1299 : * lo_mutex triggers a circular lock dependency possibility warning as
1300 : * fput can take bd_mutex which is usually taken before lo_mutex.
1301 : */
1302 0 : if (filp)
1303 0 : fput(filp);
1304 0 : return err;
1305 : }
1306 :
1307 0 : static int loop_clr_fd(struct loop_device *lo)
1308 : {
1309 0 : int err;
1310 :
1311 0 : err = mutex_lock_killable(&lo->lo_mutex);
1312 0 : if (err)
1313 : return err;
1314 0 : if (lo->lo_state != Lo_bound) {
1315 0 : mutex_unlock(&lo->lo_mutex);
1316 0 : return -ENXIO;
1317 : }
1318 : /*
1319 : * If we've explicitly asked to tear down the loop device,
1320 : * and it has an elevated reference count, set it for auto-teardown when
1321 : * the last reference goes away. This stops $!~#$@ udev from
1322 : * preventing teardown because it decided that it needs to run blkid on
1323 : * the loopback device whenever they appear. xfstests is notorious for
1324 : * failing tests because blkid via udev races with a losetup
1325 : * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1326 : * command to fail with EBUSY.
1327 : */
1328 0 : if (atomic_read(&lo->lo_refcnt) > 1) {
1329 0 : lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1330 0 : mutex_unlock(&lo->lo_mutex);
1331 0 : return 0;
1332 : }
1333 0 : lo->lo_state = Lo_rundown;
1334 0 : mutex_unlock(&lo->lo_mutex);
1335 :
1336 0 : return __loop_clr_fd(lo, false);
1337 : }
1338 :
1339 : static int
1340 0 : loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1341 : {
1342 0 : int err;
1343 0 : struct block_device *bdev;
1344 0 : kuid_t uid = current_uid();
1345 0 : int prev_lo_flags;
1346 0 : bool partscan = false;
1347 0 : bool size_changed = false;
1348 :
1349 0 : err = mutex_lock_killable(&lo->lo_mutex);
1350 0 : if (err)
1351 : return err;
1352 0 : if (lo->lo_encrypt_key_size &&
1353 0 : !uid_eq(lo->lo_key_owner, uid) &&
1354 0 : !capable(CAP_SYS_ADMIN)) {
1355 0 : err = -EPERM;
1356 0 : goto out_unlock;
1357 : }
1358 0 : if (lo->lo_state != Lo_bound) {
1359 0 : err = -ENXIO;
1360 0 : goto out_unlock;
1361 : }
1362 :
1363 0 : if (lo->lo_offset != info->lo_offset ||
1364 0 : lo->lo_sizelimit != info->lo_sizelimit) {
1365 0 : size_changed = true;
1366 0 : sync_blockdev(lo->lo_device);
1367 0 : invalidate_bdev(lo->lo_device);
1368 : }
1369 :
1370 : /* I/O need to be drained during transfer transition */
1371 0 : blk_mq_freeze_queue(lo->lo_queue);
1372 :
1373 0 : if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1374 : /* If any pages were dirtied after invalidate_bdev(), try again */
1375 0 : err = -EAGAIN;
1376 0 : pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1377 : __func__, lo->lo_number, lo->lo_file_name,
1378 : lo->lo_device->bd_inode->i_mapping->nrpages);
1379 0 : goto out_unfreeze;
1380 : }
1381 :
1382 0 : prev_lo_flags = lo->lo_flags;
1383 :
1384 0 : err = loop_set_status_from_info(lo, info);
1385 0 : if (err)
1386 0 : goto out_unfreeze;
1387 :
1388 : /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1389 0 : lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1390 : /* For those flags, use the previous values instead */
1391 0 : lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1392 : /* For flags that can't be cleared, use previous values too */
1393 0 : lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1394 :
1395 0 : if (size_changed) {
1396 0 : loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1397 : lo->lo_backing_file);
1398 0 : loop_set_size(lo, new_size);
1399 : }
1400 :
1401 0 : loop_config_discard(lo);
1402 :
1403 : /* update dio if lo_offset or transfer is changed */
1404 0 : __loop_update_dio(lo, lo->use_dio);
1405 :
1406 0 : out_unfreeze:
1407 0 : blk_mq_unfreeze_queue(lo->lo_queue);
1408 :
1409 0 : if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1410 0 : !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1411 0 : lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1412 0 : bdev = lo->lo_device;
1413 0 : partscan = true;
1414 : }
1415 0 : out_unlock:
1416 0 : mutex_unlock(&lo->lo_mutex);
1417 0 : if (partscan)
1418 0 : loop_reread_partitions(lo, bdev);
1419 :
1420 : return err;
1421 : }
1422 :
1423 : static int
1424 0 : loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1425 : {
1426 0 : struct path path;
1427 0 : struct kstat stat;
1428 0 : int ret;
1429 :
1430 0 : ret = mutex_lock_killable(&lo->lo_mutex);
1431 0 : if (ret)
1432 : return ret;
1433 0 : if (lo->lo_state != Lo_bound) {
1434 0 : mutex_unlock(&lo->lo_mutex);
1435 0 : return -ENXIO;
1436 : }
1437 :
1438 0 : memset(info, 0, sizeof(*info));
1439 0 : info->lo_number = lo->lo_number;
1440 0 : info->lo_offset = lo->lo_offset;
1441 0 : info->lo_sizelimit = lo->lo_sizelimit;
1442 0 : info->lo_flags = lo->lo_flags;
1443 0 : memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1444 0 : memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1445 0 : info->lo_encrypt_type =
1446 0 : lo->lo_encryption ? lo->lo_encryption->number : 0;
1447 0 : if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1448 0 : info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1449 0 : memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1450 0 : lo->lo_encrypt_key_size);
1451 : }
1452 :
1453 : /* Drop lo_mutex while we call into the filesystem. */
1454 0 : path = lo->lo_backing_file->f_path;
1455 0 : path_get(&path);
1456 0 : mutex_unlock(&lo->lo_mutex);
1457 0 : ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1458 0 : if (!ret) {
1459 0 : info->lo_device = huge_encode_dev(stat.dev);
1460 0 : info->lo_inode = stat.ino;
1461 0 : info->lo_rdevice = huge_encode_dev(stat.rdev);
1462 : }
1463 0 : path_put(&path);
1464 0 : return ret;
1465 : }
1466 :
1467 : static void
1468 0 : loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1469 : {
1470 0 : memset(info64, 0, sizeof(*info64));
1471 0 : info64->lo_number = info->lo_number;
1472 0 : info64->lo_device = info->lo_device;
1473 0 : info64->lo_inode = info->lo_inode;
1474 0 : info64->lo_rdevice = info->lo_rdevice;
1475 0 : info64->lo_offset = info->lo_offset;
1476 0 : info64->lo_sizelimit = 0;
1477 0 : info64->lo_encrypt_type = info->lo_encrypt_type;
1478 0 : info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1479 0 : info64->lo_flags = info->lo_flags;
1480 0 : info64->lo_init[0] = info->lo_init[0];
1481 0 : info64->lo_init[1] = info->lo_init[1];
1482 0 : if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1483 0 : memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1484 : else
1485 0 : memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1486 0 : memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1487 0 : }
1488 :
1489 : static int
1490 0 : loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1491 : {
1492 0 : memset(info, 0, sizeof(*info));
1493 0 : info->lo_number = info64->lo_number;
1494 0 : info->lo_device = info64->lo_device;
1495 0 : info->lo_inode = info64->lo_inode;
1496 0 : info->lo_rdevice = info64->lo_rdevice;
1497 0 : info->lo_offset = info64->lo_offset;
1498 0 : info->lo_encrypt_type = info64->lo_encrypt_type;
1499 0 : info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1500 0 : info->lo_flags = info64->lo_flags;
1501 0 : info->lo_init[0] = info64->lo_init[0];
1502 0 : info->lo_init[1] = info64->lo_init[1];
1503 0 : if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1504 0 : memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1505 : else
1506 0 : memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1507 0 : memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1508 :
1509 : /* error in case values were truncated */
1510 0 : if (info->lo_device != info64->lo_device ||
1511 0 : info->lo_rdevice != info64->lo_rdevice ||
1512 0 : info->lo_inode != info64->lo_inode ||
1513 0 : info->lo_offset != info64->lo_offset)
1514 0 : return -EOVERFLOW;
1515 :
1516 : return 0;
1517 : }
1518 :
1519 : static int
1520 0 : loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1521 : {
1522 0 : struct loop_info info;
1523 0 : struct loop_info64 info64;
1524 :
1525 0 : if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1526 : return -EFAULT;
1527 0 : loop_info64_from_old(&info, &info64);
1528 0 : return loop_set_status(lo, &info64);
1529 : }
1530 :
1531 : static int
1532 0 : loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1533 : {
1534 0 : struct loop_info64 info64;
1535 :
1536 0 : if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1537 : return -EFAULT;
1538 0 : return loop_set_status(lo, &info64);
1539 : }
1540 :
1541 : static int
1542 0 : loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1543 0 : struct loop_info info;
1544 0 : struct loop_info64 info64;
1545 0 : int err;
1546 :
1547 0 : if (!arg)
1548 : return -EINVAL;
1549 0 : err = loop_get_status(lo, &info64);
1550 0 : if (!err)
1551 0 : err = loop_info64_to_old(&info64, &info);
1552 0 : if (!err && copy_to_user(arg, &info, sizeof(info)))
1553 0 : err = -EFAULT;
1554 :
1555 : return err;
1556 : }
1557 :
1558 : static int
1559 0 : loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1560 0 : struct loop_info64 info64;
1561 0 : int err;
1562 :
1563 0 : if (!arg)
1564 : return -EINVAL;
1565 0 : err = loop_get_status(lo, &info64);
1566 0 : if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1567 0 : err = -EFAULT;
1568 :
1569 : return err;
1570 : }
1571 :
1572 0 : static int loop_set_capacity(struct loop_device *lo)
1573 : {
1574 0 : loff_t size;
1575 :
1576 0 : if (unlikely(lo->lo_state != Lo_bound))
1577 : return -ENXIO;
1578 :
1579 0 : size = get_loop_size(lo, lo->lo_backing_file);
1580 0 : loop_set_size(lo, size);
1581 :
1582 0 : return 0;
1583 : }
1584 :
1585 0 : static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1586 : {
1587 0 : int error = -ENXIO;
1588 0 : if (lo->lo_state != Lo_bound)
1589 0 : goto out;
1590 :
1591 0 : __loop_update_dio(lo, !!arg);
1592 0 : if (lo->use_dio == !!arg)
1593 0 : return 0;
1594 : error = -EINVAL;
1595 : out:
1596 : return error;
1597 : }
1598 :
1599 0 : static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1600 : {
1601 0 : int err = 0;
1602 :
1603 0 : if (lo->lo_state != Lo_bound)
1604 : return -ENXIO;
1605 :
1606 0 : err = loop_validate_block_size(arg);
1607 0 : if (err)
1608 : return err;
1609 :
1610 0 : if (lo->lo_queue->limits.logical_block_size == arg)
1611 : return 0;
1612 :
1613 0 : sync_blockdev(lo->lo_device);
1614 0 : invalidate_bdev(lo->lo_device);
1615 :
1616 0 : blk_mq_freeze_queue(lo->lo_queue);
1617 :
1618 : /* invalidate_bdev should have truncated all the pages */
1619 0 : if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1620 0 : err = -EAGAIN;
1621 0 : pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1622 : __func__, lo->lo_number, lo->lo_file_name,
1623 : lo->lo_device->bd_inode->i_mapping->nrpages);
1624 0 : goto out_unfreeze;
1625 : }
1626 :
1627 0 : blk_queue_logical_block_size(lo->lo_queue, arg);
1628 0 : blk_queue_physical_block_size(lo->lo_queue, arg);
1629 0 : blk_queue_io_min(lo->lo_queue, arg);
1630 0 : loop_update_dio(lo);
1631 0 : out_unfreeze:
1632 0 : blk_mq_unfreeze_queue(lo->lo_queue);
1633 :
1634 0 : return err;
1635 : }
1636 :
1637 0 : static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1638 : unsigned long arg)
1639 : {
1640 0 : int err;
1641 :
1642 0 : err = mutex_lock_killable(&lo->lo_mutex);
1643 0 : if (err)
1644 : return err;
1645 0 : switch (cmd) {
1646 0 : case LOOP_SET_CAPACITY:
1647 0 : err = loop_set_capacity(lo);
1648 0 : break;
1649 0 : case LOOP_SET_DIRECT_IO:
1650 0 : err = loop_set_dio(lo, arg);
1651 0 : break;
1652 0 : case LOOP_SET_BLOCK_SIZE:
1653 0 : err = loop_set_block_size(lo, arg);
1654 0 : break;
1655 0 : default:
1656 0 : err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1657 : }
1658 0 : mutex_unlock(&lo->lo_mutex);
1659 0 : return err;
1660 : }
1661 :
1662 0 : static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1663 : unsigned int cmd, unsigned long arg)
1664 : {
1665 0 : struct loop_device *lo = bdev->bd_disk->private_data;
1666 0 : void __user *argp = (void __user *) arg;
1667 0 : int err;
1668 :
1669 0 : switch (cmd) {
1670 0 : case LOOP_SET_FD: {
1671 : /*
1672 : * Legacy case - pass in a zeroed out struct loop_config with
1673 : * only the file descriptor set , which corresponds with the
1674 : * default parameters we'd have used otherwise.
1675 : */
1676 0 : struct loop_config config;
1677 :
1678 0 : memset(&config, 0, sizeof(config));
1679 0 : config.fd = arg;
1680 :
1681 0 : return loop_configure(lo, mode, bdev, &config);
1682 : }
1683 : case LOOP_CONFIGURE: {
1684 0 : struct loop_config config;
1685 :
1686 0 : if (copy_from_user(&config, argp, sizeof(config)))
1687 : return -EFAULT;
1688 :
1689 0 : return loop_configure(lo, mode, bdev, &config);
1690 : }
1691 0 : case LOOP_CHANGE_FD:
1692 0 : return loop_change_fd(lo, bdev, arg);
1693 0 : case LOOP_CLR_FD:
1694 0 : return loop_clr_fd(lo);
1695 0 : case LOOP_SET_STATUS:
1696 0 : err = -EPERM;
1697 0 : if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1698 0 : err = loop_set_status_old(lo, argp);
1699 : }
1700 : break;
1701 0 : case LOOP_GET_STATUS:
1702 0 : return loop_get_status_old(lo, argp);
1703 0 : case LOOP_SET_STATUS64:
1704 0 : err = -EPERM;
1705 0 : if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1706 0 : err = loop_set_status64(lo, argp);
1707 : }
1708 : break;
1709 0 : case LOOP_GET_STATUS64:
1710 0 : return loop_get_status64(lo, argp);
1711 0 : case LOOP_SET_CAPACITY:
1712 : case LOOP_SET_DIRECT_IO:
1713 : case LOOP_SET_BLOCK_SIZE:
1714 0 : if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1715 : return -EPERM;
1716 0 : fallthrough;
1717 : default:
1718 0 : err = lo_simple_ioctl(lo, cmd, arg);
1719 0 : break;
1720 : }
1721 :
1722 : return err;
1723 : }
1724 :
1725 : #ifdef CONFIG_COMPAT
1726 : struct compat_loop_info {
1727 : compat_int_t lo_number; /* ioctl r/o */
1728 : compat_dev_t lo_device; /* ioctl r/o */
1729 : compat_ulong_t lo_inode; /* ioctl r/o */
1730 : compat_dev_t lo_rdevice; /* ioctl r/o */
1731 : compat_int_t lo_offset;
1732 : compat_int_t lo_encrypt_type;
1733 : compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1734 : compat_int_t lo_flags; /* ioctl r/o */
1735 : char lo_name[LO_NAME_SIZE];
1736 : unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1737 : compat_ulong_t lo_init[2];
1738 : char reserved[4];
1739 : };
1740 :
1741 : /*
1742 : * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1743 : * - noinlined to reduce stack space usage in main part of driver
1744 : */
1745 : static noinline int
1746 0 : loop_info64_from_compat(const struct compat_loop_info __user *arg,
1747 : struct loop_info64 *info64)
1748 : {
1749 0 : struct compat_loop_info info;
1750 :
1751 0 : if (copy_from_user(&info, arg, sizeof(info)))
1752 : return -EFAULT;
1753 :
1754 0 : memset(info64, 0, sizeof(*info64));
1755 0 : info64->lo_number = info.lo_number;
1756 0 : info64->lo_device = info.lo_device;
1757 0 : info64->lo_inode = info.lo_inode;
1758 0 : info64->lo_rdevice = info.lo_rdevice;
1759 0 : info64->lo_offset = info.lo_offset;
1760 0 : info64->lo_sizelimit = 0;
1761 0 : info64->lo_encrypt_type = info.lo_encrypt_type;
1762 0 : info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1763 0 : info64->lo_flags = info.lo_flags;
1764 0 : info64->lo_init[0] = info.lo_init[0];
1765 0 : info64->lo_init[1] = info.lo_init[1];
1766 0 : if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1767 0 : memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1768 : else
1769 0 : memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1770 0 : memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1771 0 : return 0;
1772 : }
1773 :
1774 : /*
1775 : * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1776 : * - noinlined to reduce stack space usage in main part of driver
1777 : */
1778 : static noinline int
1779 0 : loop_info64_to_compat(const struct loop_info64 *info64,
1780 : struct compat_loop_info __user *arg)
1781 : {
1782 0 : struct compat_loop_info info;
1783 :
1784 0 : memset(&info, 0, sizeof(info));
1785 0 : info.lo_number = info64->lo_number;
1786 0 : info.lo_device = info64->lo_device;
1787 0 : info.lo_inode = info64->lo_inode;
1788 0 : info.lo_rdevice = info64->lo_rdevice;
1789 0 : info.lo_offset = info64->lo_offset;
1790 0 : info.lo_encrypt_type = info64->lo_encrypt_type;
1791 0 : info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1792 0 : info.lo_flags = info64->lo_flags;
1793 0 : info.lo_init[0] = info64->lo_init[0];
1794 0 : info.lo_init[1] = info64->lo_init[1];
1795 0 : if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1796 0 : memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1797 : else
1798 0 : memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1799 0 : memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1800 :
1801 : /* error in case values were truncated */
1802 0 : if (info.lo_device != info64->lo_device ||
1803 0 : info.lo_rdevice != info64->lo_rdevice ||
1804 0 : info.lo_inode != info64->lo_inode ||
1805 0 : info.lo_offset != info64->lo_offset ||
1806 0 : info.lo_init[0] != info64->lo_init[0] ||
1807 0 : info.lo_init[1] != info64->lo_init[1])
1808 : return -EOVERFLOW;
1809 :
1810 0 : if (copy_to_user(arg, &info, sizeof(info)))
1811 0 : return -EFAULT;
1812 : return 0;
1813 : }
1814 :
1815 : static int
1816 0 : loop_set_status_compat(struct loop_device *lo,
1817 : const struct compat_loop_info __user *arg)
1818 : {
1819 0 : struct loop_info64 info64;
1820 0 : int ret;
1821 :
1822 0 : ret = loop_info64_from_compat(arg, &info64);
1823 0 : if (ret < 0)
1824 : return ret;
1825 0 : return loop_set_status(lo, &info64);
1826 : }
1827 :
1828 : static int
1829 0 : loop_get_status_compat(struct loop_device *lo,
1830 : struct compat_loop_info __user *arg)
1831 : {
1832 0 : struct loop_info64 info64;
1833 0 : int err;
1834 :
1835 0 : if (!arg)
1836 : return -EINVAL;
1837 0 : err = loop_get_status(lo, &info64);
1838 0 : if (!err)
1839 0 : err = loop_info64_to_compat(&info64, arg);
1840 : return err;
1841 : }
1842 :
1843 0 : static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1844 : unsigned int cmd, unsigned long arg)
1845 : {
1846 0 : struct loop_device *lo = bdev->bd_disk->private_data;
1847 0 : int err;
1848 :
1849 0 : switch(cmd) {
1850 0 : case LOOP_SET_STATUS:
1851 0 : err = loop_set_status_compat(lo,
1852 : (const struct compat_loop_info __user *)arg);
1853 0 : break;
1854 0 : case LOOP_GET_STATUS:
1855 0 : err = loop_get_status_compat(lo,
1856 : (struct compat_loop_info __user *)arg);
1857 0 : break;
1858 0 : case LOOP_SET_CAPACITY:
1859 : case LOOP_CLR_FD:
1860 : case LOOP_GET_STATUS64:
1861 : case LOOP_SET_STATUS64:
1862 : case LOOP_CONFIGURE:
1863 0 : arg = (unsigned long) compat_ptr(arg);
1864 0 : fallthrough;
1865 0 : case LOOP_SET_FD:
1866 : case LOOP_CHANGE_FD:
1867 : case LOOP_SET_BLOCK_SIZE:
1868 : case LOOP_SET_DIRECT_IO:
1869 0 : err = lo_ioctl(bdev, mode, cmd, arg);
1870 0 : break;
1871 : default:
1872 : err = -ENOIOCTLCMD;
1873 : break;
1874 : }
1875 0 : return err;
1876 : }
1877 : #endif
1878 :
1879 16 : static int lo_open(struct block_device *bdev, fmode_t mode)
1880 : {
1881 16 : struct loop_device *lo;
1882 16 : int err;
1883 :
1884 : /*
1885 : * take loop_ctl_mutex to protect lo pointer from race with
1886 : * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
1887 : * release it prior to updating lo->lo_refcnt.
1888 : */
1889 16 : err = mutex_lock_killable(&loop_ctl_mutex);
1890 16 : if (err)
1891 : return err;
1892 16 : lo = bdev->bd_disk->private_data;
1893 16 : if (!lo) {
1894 0 : mutex_unlock(&loop_ctl_mutex);
1895 0 : return -ENXIO;
1896 : }
1897 16 : err = mutex_lock_killable(&lo->lo_mutex);
1898 16 : mutex_unlock(&loop_ctl_mutex);
1899 16 : if (err)
1900 : return err;
1901 16 : atomic_inc(&lo->lo_refcnt);
1902 16 : mutex_unlock(&lo->lo_mutex);
1903 16 : return 0;
1904 : }
1905 :
1906 16 : static void lo_release(struct gendisk *disk, fmode_t mode)
1907 : {
1908 16 : struct loop_device *lo = disk->private_data;
1909 :
1910 16 : mutex_lock(&lo->lo_mutex);
1911 32 : if (atomic_dec_return(&lo->lo_refcnt))
1912 8 : goto out_unlock;
1913 :
1914 8 : if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1915 0 : if (lo->lo_state != Lo_bound)
1916 0 : goto out_unlock;
1917 0 : lo->lo_state = Lo_rundown;
1918 0 : mutex_unlock(&lo->lo_mutex);
1919 : /*
1920 : * In autoclear mode, stop the loop thread
1921 : * and remove configuration after last close.
1922 : */
1923 0 : __loop_clr_fd(lo, true);
1924 0 : return;
1925 8 : } else if (lo->lo_state == Lo_bound) {
1926 : /*
1927 : * Otherwise keep thread (if running) and config,
1928 : * but flush possible ongoing bios in thread.
1929 : */
1930 0 : blk_mq_freeze_queue(lo->lo_queue);
1931 0 : blk_mq_unfreeze_queue(lo->lo_queue);
1932 : }
1933 :
1934 8 : out_unlock:
1935 16 : mutex_unlock(&lo->lo_mutex);
1936 : }
1937 :
1938 : static const struct block_device_operations lo_fops = {
1939 : .owner = THIS_MODULE,
1940 : .open = lo_open,
1941 : .release = lo_release,
1942 : .ioctl = lo_ioctl,
1943 : #ifdef CONFIG_COMPAT
1944 : .compat_ioctl = lo_compat_ioctl,
1945 : #endif
1946 : };
1947 :
1948 : /*
1949 : * And now the modules code and kernel interface.
1950 : */
1951 : static int max_loop;
1952 : module_param(max_loop, int, 0444);
1953 : MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1954 : module_param(max_part, int, 0444);
1955 : MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1956 : MODULE_LICENSE("GPL");
1957 : MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1958 :
1959 0 : int loop_register_transfer(struct loop_func_table *funcs)
1960 : {
1961 0 : unsigned int n = funcs->number;
1962 :
1963 0 : if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1964 : return -EINVAL;
1965 0 : xfer_funcs[n] = funcs;
1966 0 : return 0;
1967 : }
1968 :
1969 0 : static int unregister_transfer_cb(int id, void *ptr, void *data)
1970 : {
1971 0 : struct loop_device *lo = ptr;
1972 0 : struct loop_func_table *xfer = data;
1973 :
1974 0 : mutex_lock(&lo->lo_mutex);
1975 0 : if (lo->lo_encryption == xfer)
1976 0 : loop_release_xfer(lo);
1977 0 : mutex_unlock(&lo->lo_mutex);
1978 0 : return 0;
1979 : }
1980 :
1981 0 : int loop_unregister_transfer(int number)
1982 : {
1983 0 : unsigned int n = number;
1984 0 : struct loop_func_table *xfer;
1985 :
1986 0 : if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1987 : return -EINVAL;
1988 :
1989 0 : xfer_funcs[n] = NULL;
1990 0 : idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1991 0 : return 0;
1992 : }
1993 :
1994 : EXPORT_SYMBOL(loop_register_transfer);
1995 : EXPORT_SYMBOL(loop_unregister_transfer);
1996 :
1997 0 : static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1998 : const struct blk_mq_queue_data *bd)
1999 : {
2000 0 : struct request *rq = bd->rq;
2001 0 : struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2002 0 : struct loop_device *lo = rq->q->queuedata;
2003 :
2004 0 : blk_mq_start_request(rq);
2005 :
2006 0 : if (lo->lo_state != Lo_bound)
2007 : return BLK_STS_IOERR;
2008 :
2009 0 : switch (req_op(rq)) {
2010 0 : case REQ_OP_FLUSH:
2011 : case REQ_OP_DISCARD:
2012 : case REQ_OP_WRITE_ZEROES:
2013 0 : cmd->use_aio = false;
2014 0 : break;
2015 0 : default:
2016 0 : cmd->use_aio = lo->use_dio;
2017 0 : break;
2018 : }
2019 :
2020 : /* always use the first bio's css */
2021 : #ifdef CONFIG_BLK_CGROUP
2022 : if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2023 : cmd->css = &bio_blkcg(rq->bio)->css;
2024 : css_get(cmd->css);
2025 : } else
2026 : #endif
2027 0 : cmd->css = NULL;
2028 0 : kthread_queue_work(&lo->worker, &cmd->work);
2029 :
2030 0 : return BLK_STS_OK;
2031 : }
2032 :
2033 0 : static void loop_handle_cmd(struct loop_cmd *cmd)
2034 : {
2035 0 : struct request *rq = blk_mq_rq_from_pdu(cmd);
2036 0 : const bool write = op_is_write(req_op(rq));
2037 0 : struct loop_device *lo = rq->q->queuedata;
2038 0 : int ret = 0;
2039 :
2040 0 : if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2041 0 : ret = -EIO;
2042 0 : goto failed;
2043 : }
2044 :
2045 0 : ret = do_req_filebacked(lo, rq);
2046 0 : failed:
2047 : /* complete non-aio request */
2048 0 : if (!cmd->use_aio || ret) {
2049 0 : if (ret == -EOPNOTSUPP)
2050 0 : cmd->ret = ret;
2051 : else
2052 0 : cmd->ret = ret ? -EIO : 0;
2053 0 : if (likely(!blk_should_fake_timeout(rq->q)))
2054 0 : blk_mq_complete_request(rq);
2055 : }
2056 0 : }
2057 :
2058 0 : static void loop_queue_work(struct kthread_work *work)
2059 : {
2060 0 : struct loop_cmd *cmd =
2061 0 : container_of(work, struct loop_cmd, work);
2062 :
2063 0 : loop_handle_cmd(cmd);
2064 0 : }
2065 :
2066 1032 : static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2067 : unsigned int hctx_idx, unsigned int numa_node)
2068 : {
2069 1032 : struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2070 :
2071 1032 : kthread_init_work(&cmd->work, loop_queue_work);
2072 1032 : return 0;
2073 : }
2074 :
2075 : static const struct blk_mq_ops loop_mq_ops = {
2076 : .queue_rq = loop_queue_rq,
2077 : .init_request = loop_init_request,
2078 : .complete = lo_complete_rq,
2079 : };
2080 :
2081 8 : static int loop_add(struct loop_device **l, int i)
2082 : {
2083 8 : struct loop_device *lo;
2084 8 : struct gendisk *disk;
2085 8 : int err;
2086 :
2087 8 : err = -ENOMEM;
2088 8 : lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2089 8 : if (!lo)
2090 0 : goto out;
2091 :
2092 8 : lo->lo_state = Lo_unbound;
2093 :
2094 : /* allocate id, if @id >= 0, we're requesting that specific id */
2095 8 : if (i >= 0) {
2096 8 : err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2097 8 : if (err == -ENOSPC)
2098 : err = -EEXIST;
2099 : } else {
2100 0 : err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2101 : }
2102 8 : if (err < 0)
2103 0 : goto out_free_dev;
2104 8 : i = err;
2105 :
2106 8 : err = -ENOMEM;
2107 8 : lo->tag_set.ops = &loop_mq_ops;
2108 8 : lo->tag_set.nr_hw_queues = 1;
2109 8 : lo->tag_set.queue_depth = 128;
2110 8 : lo->tag_set.numa_node = NUMA_NO_NODE;
2111 8 : lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2112 8 : lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2113 8 : lo->tag_set.driver_data = lo;
2114 :
2115 8 : err = blk_mq_alloc_tag_set(&lo->tag_set);
2116 8 : if (err)
2117 0 : goto out_free_idr;
2118 :
2119 8 : lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2120 8 : if (IS_ERR(lo->lo_queue)) {
2121 0 : err = PTR_ERR(lo->lo_queue);
2122 0 : goto out_cleanup_tags;
2123 : }
2124 8 : lo->lo_queue->queuedata = lo;
2125 :
2126 8 : blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2127 :
2128 : /*
2129 : * By default, we do buffer IO, so it doesn't make sense to enable
2130 : * merge because the I/O submitted to backing file is handled page by
2131 : * page. For directio mode, merge does help to dispatch bigger request
2132 : * to underlayer disk. We will enable merge once directio is enabled.
2133 : */
2134 8 : blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2135 :
2136 8 : err = -ENOMEM;
2137 8 : disk = lo->lo_disk = alloc_disk(1 << part_shift);
2138 8 : if (!disk)
2139 0 : goto out_free_queue;
2140 :
2141 : /*
2142 : * Disable partition scanning by default. The in-kernel partition
2143 : * scanning can be requested individually per-device during its
2144 : * setup. Userspace can always add and remove partitions from all
2145 : * devices. The needed partition minors are allocated from the
2146 : * extended minor space, the main loop device numbers will continue
2147 : * to match the loop minors, regardless of the number of partitions
2148 : * used.
2149 : *
2150 : * If max_part is given, partition scanning is globally enabled for
2151 : * all loop devices. The minors for the main loop devices will be
2152 : * multiples of max_part.
2153 : *
2154 : * Note: Global-for-all-devices, set-only-at-init, read-only module
2155 : * parameteters like 'max_loop' and 'max_part' make things needlessly
2156 : * complicated, are too static, inflexible and may surprise
2157 : * userspace tools. Parameters like this in general should be avoided.
2158 : */
2159 8 : if (!part_shift)
2160 8 : disk->flags |= GENHD_FL_NO_PART_SCAN;
2161 8 : disk->flags |= GENHD_FL_EXT_DEVT;
2162 8 : atomic_set(&lo->lo_refcnt, 0);
2163 8 : mutex_init(&lo->lo_mutex);
2164 8 : lo->lo_number = i;
2165 8 : spin_lock_init(&lo->lo_lock);
2166 8 : disk->major = LOOP_MAJOR;
2167 8 : disk->first_minor = i << part_shift;
2168 8 : disk->fops = &lo_fops;
2169 8 : disk->private_data = lo;
2170 8 : disk->queue = lo->lo_queue;
2171 8 : sprintf(disk->disk_name, "loop%d", i);
2172 8 : add_disk(disk);
2173 8 : *l = lo;
2174 8 : return lo->lo_number;
2175 :
2176 0 : out_free_queue:
2177 0 : blk_cleanup_queue(lo->lo_queue);
2178 0 : out_cleanup_tags:
2179 0 : blk_mq_free_tag_set(&lo->tag_set);
2180 0 : out_free_idr:
2181 0 : idr_remove(&loop_index_idr, i);
2182 0 : out_free_dev:
2183 0 : kfree(lo);
2184 : out:
2185 : return err;
2186 : }
2187 :
2188 0 : static void loop_remove(struct loop_device *lo)
2189 : {
2190 0 : del_gendisk(lo->lo_disk);
2191 0 : blk_cleanup_queue(lo->lo_queue);
2192 0 : blk_mq_free_tag_set(&lo->tag_set);
2193 0 : put_disk(lo->lo_disk);
2194 0 : mutex_destroy(&lo->lo_mutex);
2195 0 : kfree(lo);
2196 0 : }
2197 :
2198 0 : static int find_free_cb(int id, void *ptr, void *data)
2199 : {
2200 0 : struct loop_device *lo = ptr;
2201 0 : struct loop_device **l = data;
2202 :
2203 0 : if (lo->lo_state == Lo_unbound) {
2204 0 : *l = lo;
2205 0 : return 1;
2206 : }
2207 : return 0;
2208 : }
2209 :
2210 0 : static int loop_lookup(struct loop_device **l, int i)
2211 : {
2212 0 : struct loop_device *lo;
2213 0 : int ret = -ENODEV;
2214 :
2215 0 : if (i < 0) {
2216 0 : int err;
2217 :
2218 0 : err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2219 0 : if (err == 1) {
2220 0 : *l = lo;
2221 0 : ret = lo->lo_number;
2222 : }
2223 0 : goto out;
2224 : }
2225 :
2226 : /* lookup and return a specific i */
2227 0 : lo = idr_find(&loop_index_idr, i);
2228 0 : if (lo) {
2229 0 : *l = lo;
2230 0 : ret = lo->lo_number;
2231 : }
2232 0 : out:
2233 0 : return ret;
2234 : }
2235 :
2236 0 : static void loop_probe(dev_t dev)
2237 : {
2238 0 : int idx = MINOR(dev) >> part_shift;
2239 0 : struct loop_device *lo;
2240 :
2241 0 : if (max_loop && idx >= max_loop)
2242 0 : return;
2243 :
2244 0 : mutex_lock(&loop_ctl_mutex);
2245 0 : if (loop_lookup(&lo, idx) < 0)
2246 0 : loop_add(&lo, idx);
2247 0 : mutex_unlock(&loop_ctl_mutex);
2248 : }
2249 :
2250 0 : static long loop_control_ioctl(struct file *file, unsigned int cmd,
2251 : unsigned long parm)
2252 : {
2253 0 : struct loop_device *lo;
2254 0 : int ret;
2255 :
2256 0 : ret = mutex_lock_killable(&loop_ctl_mutex);
2257 0 : if (ret)
2258 0 : return ret;
2259 :
2260 0 : ret = -ENOSYS;
2261 0 : switch (cmd) {
2262 0 : case LOOP_CTL_ADD:
2263 0 : ret = loop_lookup(&lo, parm);
2264 0 : if (ret >= 0) {
2265 : ret = -EEXIST;
2266 : break;
2267 : }
2268 0 : ret = loop_add(&lo, parm);
2269 0 : break;
2270 0 : case LOOP_CTL_REMOVE:
2271 0 : ret = loop_lookup(&lo, parm);
2272 0 : if (ret < 0)
2273 : break;
2274 0 : ret = mutex_lock_killable(&lo->lo_mutex);
2275 0 : if (ret)
2276 : break;
2277 0 : if (lo->lo_state != Lo_unbound) {
2278 0 : ret = -EBUSY;
2279 0 : mutex_unlock(&lo->lo_mutex);
2280 0 : break;
2281 : }
2282 0 : if (atomic_read(&lo->lo_refcnt) > 0) {
2283 0 : ret = -EBUSY;
2284 0 : mutex_unlock(&lo->lo_mutex);
2285 0 : break;
2286 : }
2287 0 : lo->lo_disk->private_data = NULL;
2288 0 : mutex_unlock(&lo->lo_mutex);
2289 0 : idr_remove(&loop_index_idr, lo->lo_number);
2290 0 : loop_remove(lo);
2291 0 : break;
2292 0 : case LOOP_CTL_GET_FREE:
2293 0 : ret = loop_lookup(&lo, -1);
2294 0 : if (ret >= 0)
2295 : break;
2296 0 : ret = loop_add(&lo, -1);
2297 : }
2298 0 : mutex_unlock(&loop_ctl_mutex);
2299 :
2300 0 : return ret;
2301 : }
2302 :
2303 : static const struct file_operations loop_ctl_fops = {
2304 : .open = nonseekable_open,
2305 : .unlocked_ioctl = loop_control_ioctl,
2306 : .compat_ioctl = loop_control_ioctl,
2307 : .owner = THIS_MODULE,
2308 : .llseek = noop_llseek,
2309 : };
2310 :
2311 : static struct miscdevice loop_misc = {
2312 : .minor = LOOP_CTRL_MINOR,
2313 : .name = "loop-control",
2314 : .fops = &loop_ctl_fops,
2315 : };
2316 :
2317 : MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2318 : MODULE_ALIAS("devname:loop-control");
2319 :
2320 1 : static int __init loop_init(void)
2321 : {
2322 1 : int i, nr;
2323 1 : struct loop_device *lo;
2324 1 : int err;
2325 :
2326 1 : part_shift = 0;
2327 1 : if (max_part > 0) {
2328 0 : part_shift = fls(max_part);
2329 :
2330 : /*
2331 : * Adjust max_part according to part_shift as it is exported
2332 : * to user space so that user can decide correct minor number
2333 : * if [s]he want to create more devices.
2334 : *
2335 : * Note that -1 is required because partition 0 is reserved
2336 : * for the whole disk.
2337 : */
2338 0 : max_part = (1UL << part_shift) - 1;
2339 : }
2340 :
2341 1 : if ((1UL << part_shift) > DISK_MAX_PARTS) {
2342 0 : err = -EINVAL;
2343 0 : goto err_out;
2344 : }
2345 :
2346 1 : if (max_loop > 1UL << (MINORBITS - part_shift)) {
2347 0 : err = -EINVAL;
2348 0 : goto err_out;
2349 : }
2350 :
2351 : /*
2352 : * If max_loop is specified, create that many devices upfront.
2353 : * This also becomes a hard limit. If max_loop is not specified,
2354 : * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2355 : * init time. Loop devices can be requested on-demand with the
2356 : * /dev/loop-control interface, or be instantiated by accessing
2357 : * a 'dead' device node.
2358 : */
2359 1 : if (max_loop)
2360 0 : nr = max_loop;
2361 : else
2362 : nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2363 :
2364 1 : err = misc_register(&loop_misc);
2365 1 : if (err < 0)
2366 0 : goto err_out;
2367 :
2368 :
2369 1 : if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2370 0 : err = -EIO;
2371 0 : goto misc_out;
2372 : }
2373 :
2374 : /* pre-create number of devices given by config or max_loop */
2375 1 : mutex_lock(&loop_ctl_mutex);
2376 10 : for (i = 0; i < nr; i++)
2377 8 : loop_add(&lo, i);
2378 1 : mutex_unlock(&loop_ctl_mutex);
2379 :
2380 1 : printk(KERN_INFO "loop: module loaded\n");
2381 1 : return 0;
2382 :
2383 0 : misc_out:
2384 0 : misc_deregister(&loop_misc);
2385 : err_out:
2386 : return err;
2387 : }
2388 :
2389 0 : static int loop_exit_cb(int id, void *ptr, void *data)
2390 : {
2391 0 : struct loop_device *lo = ptr;
2392 :
2393 0 : loop_remove(lo);
2394 0 : return 0;
2395 : }
2396 :
2397 0 : static void __exit loop_exit(void)
2398 : {
2399 0 : mutex_lock(&loop_ctl_mutex);
2400 :
2401 0 : idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2402 0 : idr_destroy(&loop_index_idr);
2403 :
2404 0 : unregister_blkdev(LOOP_MAJOR, "loop");
2405 :
2406 0 : misc_deregister(&loop_misc);
2407 :
2408 0 : mutex_unlock(&loop_ctl_mutex);
2409 0 : }
2410 :
2411 : module_init(loop_init);
2412 : module_exit(loop_exit);
2413 :
2414 : #ifndef MODULE
2415 0 : static int __init max_loop_setup(char *str)
2416 : {
2417 0 : max_loop = simple_strtol(str, NULL, 0);
2418 0 : return 1;
2419 : }
2420 :
2421 : __setup("max_loop=", max_loop_setup);
2422 : #endif
|
