projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: Shorten interrupt disabled regions
[linux-2.6-block.git] / drivers / block / loop.c
CommitLineData
1da177e4
LT		 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.
96de0e25 32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
1da177e4
LT		 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
4e02ed4b 43 * operations write_begin is not available on the backing filesystem.
1da177e4
LT		 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
1da177e4
LT		 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>
1da177e4
LT		 64#include <linux/swap.h>
65#include <linux/slab.h>
863d5b82 66#include <linux/compat.h>
1da177e4 67#include <linux/suspend.h>
83144186 68#include <linux/freezer.h>
2a48fc0a 69#include <linux/mutex.h>
1da177e4 70#include <linux/writeback.h>
1da177e4
LT		 71#include <linux/completion.h>
72#include <linux/highmem.h>
6c997918 73#include <linux/kthread.h>
d6b29d7c 74#include <linux/splice.h>
ee862730 75#include <linux/sysfs.h>
770fe30a 76#include <linux/miscdevice.h>
dfaa2ef6 77#include <linux/falloc.h>
283e7e5d 78#include <linux/uio.h>
83a87611 79#include "loop.h"
1da177e4 80
7c0f6ba6 81#include <linux/uaccess.h>
1da177e4 82
34dd82af
KS		 83static DEFINE_IDR(loop_index_idr);
84static DEFINE_MUTEX(loop_index_mutex);
1da177e4 85
476a4813
LV		 86static int max_part;
87static int part_shift;
88
1da177e4
LT		 89static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
93{
cfd8005c
CW		 94 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
1da177e4
LT		 96 char *in, *out, *key;
97 int i, keysize;
98
99 if (cmd == READ) {
100 in = raw_buf;
101 out = loop_buf;
102 } else {
103 in = loop_buf;
104 out = raw_buf;
105 }
106
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
111
cfd8005c
CW		 112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
1da177e4
LT		 114 cond_resched();
115 return 0;
116}
117
118static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
119{
120 if (unlikely(info->lo_encrypt_key_size <= 0))
121 return -EINVAL;
122 return 0;
123}
124
125static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
aa4d8616 127};
1da177e4
LT		 128
129static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
132 .init = xor_init
aa4d8616 133};
1da177e4
LT		 134
135/* xfer_funcs[0] is special - its release function is never called */
136static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
137 &none_funcs,
138 &xor_funcs
139};
140
7035b5df 141static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
1da177e4 142{
b7a1da69 143 loff_t loopsize;
1da177e4
LT		 144
145 /* Compute loopsize in bytes */
b7a1da69
GC		 146 loopsize = i_size_read(file->f_mapping->host);
147 if (offset > 0)
148 loopsize -= offset;
149 /* offset is beyond i_size, weird but possible */
7035b5df
DM		 150 if (loopsize < 0)
151 return 0;
1da177e4 152
7035b5df
DM		 153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
1da177e4
LT		 155 /*
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
158 */
159 return loopsize >> 9;
160}
161
7035b5df
DM		 162static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163{
164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
165}
166
2e5ab5f3
ML		 167static void __loop_update_dio(struct loop_device *lo, bool dio)
168{
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
174 bool use_dio;
175
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
179 }
180
181 /*
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
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
89d790ab 189 * of requests in sane applications should be PAGE_SIZE aligned
2e5ab5f3
ML		 190 */
191 if (dio) {
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
195 !lo->transfer)
196 use_dio = true;
197 else
198 use_dio = false;
199 } else {
200 use_dio = false;
201 }
202
203 if (lo->use_dio == use_dio)
204 return;
205
206 /* flush dirty pages before changing direct IO */
207 vfs_fsync(file, 0);
208
209 /*
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
213 */
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
40326d8a 216 if (use_dio) {
8b904b5b 217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2e5ab5f3 218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
40326d8a 219 } else {
8b904b5b 220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2e5ab5f3 221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
40326d8a 222 }
2e5ab5f3
ML		 223 blk_mq_unfreeze_queue(lo->lo_queue);
224}
225
1da177e4 226static int
1e6ec9ea 227figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
1da177e4 228{
7035b5df 229 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
1da177e4 230 sector_t x = (sector_t)size;
7b0576a3 231 struct block_device *bdev = lo->lo_device;
1da177e4
LT		 232
233 if (unlikely((loff_t)x != size))
234 return -EFBIG;
7035b5df
DM		 235 if (lo->lo_offset != offset)
236 lo->lo_offset = offset;
237 if (lo->lo_sizelimit != sizelimit)
238 lo->lo_sizelimit = sizelimit;
73285082 239 set_capacity(lo->lo_disk, x);
7b0576a3
GC		 240 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
241 /* let user-space know about the new size */
242 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
7035b5df 243 return 0;
1da177e4
LT		 244}
245
246static inline int
247lo_do_transfer(struct loop_device *lo, int cmd,
248 struct page *rpage, unsigned roffs,
249 struct page *lpage, unsigned loffs,
250 int size, sector_t rblock)
251{
aa4d8616
CH		 252 int ret;
253
254 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
255 if (likely(!ret))
1da177e4
LT		 256 return 0;
257
aa4d8616
CH		 258 printk_ratelimited(KERN_ERR
259 "loop: Transfer error at byte offset %llu, length %i.\n",
260 (unsigned long long)rblock << 9, size);
261 return ret;
1da177e4
LT		 262}
263
aa4d8616 264static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
1da177e4 265{
aa4d8616 266 struct iov_iter i;
1da177e4 267 ssize_t bw;
283e7e5d 268
1d037577 269 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
1da177e4 270
03d95eb2 271 file_start_write(file);
abbb6589 272 bw = vfs_iter_write(file, &i, ppos, 0);
03d95eb2 273 file_end_write(file);
aa4d8616
CH		 274
275 if (likely(bw == bvec->bv_len))
1da177e4 276 return 0;
aa4d8616
CH		 277
278 printk_ratelimited(KERN_ERR
279 "loop: Write error at byte offset %llu, length %i.\n",
280 (unsigned long long)*ppos, bvec->bv_len);
1da177e4
LT		 281 if (bw >= 0)
282 bw = -EIO;
283 return bw;
284}
285
aa4d8616
CH		 286static int lo_write_simple(struct loop_device *lo, struct request *rq,
287 loff_t pos)
1da177e4 288{
aa4d8616
CH		 289 struct bio_vec bvec;
290 struct req_iterator iter;
291 int ret = 0;
292
293 rq_for_each_segment(bvec, rq, iter) {
294 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
295 if (ret < 0)
296 break;
297 cond_resched();
298 }
299
300 return ret;
1da177e4
LT		 301}
302
aa4d8616 303/*
456be148
CH		 304 * This is the slow, transforming version that needs to double buffer the
305 * data as it cannot do the transformations in place without having direct
306 * access to the destination pages of the backing file.
1da177e4 307 */
aa4d8616
CH		 308static int lo_write_transfer(struct loop_device *lo, struct request *rq,
309 loff_t pos)
1da177e4 310{
aa4d8616 311 struct bio_vec bvec, b;
30112013 312 struct req_iterator iter;
aa4d8616 313 struct page *page;
7988613b 314 int ret = 0;
1da177e4 315
aa4d8616
CH		 316 page = alloc_page(GFP_NOIO);
317 if (unlikely(!page))
318 return -ENOMEM;
456be148 319
30112013 320 rq_for_each_segment(bvec, rq, iter) {
aa4d8616
CH		 321 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
322 bvec.bv_offset, bvec.bv_len, pos >> 9);
323 if (unlikely(ret))
324 break;
325
326 b.bv_page = page;
327 b.bv_offset = 0;
328 b.bv_len = bvec.bv_len;
329 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
1da177e4
LT		 330 if (ret < 0)
331 break;
1da177e4 332 }
aa4d8616
CH		 333
334 __free_page(page);
1da177e4 335 return ret;
1da177e4
LT		 336}
337
aa4d8616
CH		 338static int lo_read_simple(struct loop_device *lo, struct request *rq,
339 loff_t pos)
1da177e4 340{
aa4d8616
CH		 341 struct bio_vec bvec;
342 struct req_iterator iter;
343 struct iov_iter i;
344 ssize_t len;
1da177e4 345
aa4d8616
CH		 346 rq_for_each_segment(bvec, rq, iter) {
347 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
18e9710e 348 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
aa4d8616
CH		 349 if (len < 0)
350 return len;
1da177e4 351
aa4d8616 352 flush_dcache_page(bvec.bv_page);
fd582140 353
aa4d8616
CH		 354 if (len != bvec.bv_len) {
355 struct bio *bio;
1da177e4 356
aa4d8616
CH		 357 __rq_for_each_bio(bio, rq)
358 zero_fill_bio(bio);
359 break;
360 }
361 cond_resched();
362 }
363
364 return 0;
fd582140
JA		 365}
366
aa4d8616
CH		 367static int lo_read_transfer(struct loop_device *lo, struct request *rq,
368 loff_t pos)
1da177e4 369{
aa4d8616
CH		 370 struct bio_vec bvec, b;
371 struct req_iterator iter;
372 struct iov_iter i;
373 struct page *page;
374 ssize_t len;
375 int ret = 0;
1da177e4 376
aa4d8616
CH		 377 page = alloc_page(GFP_NOIO);
378 if (unlikely(!page))
379 return -ENOMEM;
fd582140 380
aa4d8616
CH		 381 rq_for_each_segment(bvec, rq, iter) {
382 loff_t offset = pos;
fd582140 383
aa4d8616
CH		 384 b.bv_page = page;
385 b.bv_offset = 0;
386 b.bv_len = bvec.bv_len;
fd582140 387
aa4d8616 388 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
18e9710e 389 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
aa4d8616
CH		 390 if (len < 0) {
391 ret = len;
392 goto out_free_page;
393 }
1da177e4 394
aa4d8616
CH		 395 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
396 bvec.bv_offset, len, offset >> 9);
397 if (ret)
398 goto out_free_page;
1da177e4 399
aa4d8616 400 flush_dcache_page(bvec.bv_page);
306df071 401
aa4d8616 402 if (len != bvec.bv_len) {
30112013
ML		 403 struct bio *bio;
404
405 __rq_for_each_bio(bio, rq)
406 zero_fill_bio(bio);
1da177e4 407 break;
306df071 408 }
1da177e4 409 }
aa4d8616
CH		 410
411 ret = 0;
412out_free_page:
413 __free_page(page);
414 return ret;
1da177e4
LT		 415}
416
cf655d95
ML		 417static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
418{
419 /*
420 * We use punch hole to reclaim the free space used by the
421 * image a.k.a. discard. However we do not support discard if
422 * encryption is enabled, because it may give an attacker
423 * useful information.
424 */
425 struct file *file = lo->lo_backing_file;
426 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
427 int ret;
428
429 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
430 ret = -EOPNOTSUPP;
431 goto out;
432 }
433
434 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
435 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
436 ret = -EIO;
437 out:
438 return ret;
439}
440
441static int lo_req_flush(struct loop_device *lo, struct request *rq)
442{
443 struct file *file = lo->lo_backing_file;
444 int ret = vfs_fsync(file, 0);
445 if (unlikely(ret && ret != -EINVAL))
446 ret = -EIO;
447
448 return ret;
449}
450
fe2cb290 451static void lo_complete_rq(struct request *rq)
bc07c10a 452{
fe2cb290 453 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
f9de14bc 454 blk_status_t ret = BLK_STS_OK;
bc07c10a 455
f9de14bc
JA		 456 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
457 req_op(rq) != REQ_OP_READ) {
458 if (cmd->ret < 0)
459 ret = BLK_STS_IOERR;
460 goto end_io;
bc07c10a 461 }
fe2cb290 462
f9de14bc
JA		 463 /*
464 * Short READ - if we got some data, advance our request and
465 * retry it. If we got no data, end the rest with EIO.
466 */
467 if (cmd->ret) {
468 blk_update_request(rq, BLK_STS_OK, cmd->ret);
469 cmd->ret = 0;
470 blk_mq_requeue_request(rq, true);
471 } else {
472 if (cmd->use_aio) {
473 struct bio *bio = rq->bio;
474
475 while (bio) {
476 zero_fill_bio(bio);
477 bio = bio->bi_next;
478 }
479 }
480 ret = BLK_STS_IOERR;
481end_io:
482 blk_mq_end_request(rq, ret);
483 }
bc07c10a
ML		 484}
485
92d77332
SL		 486static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
487{
1894e916
JA		 488 struct request *rq = blk_mq_rq_from_pdu(cmd);
489
92d77332
SL		 490 if (!atomic_dec_and_test(&cmd->ref))
491 return;
492 kfree(cmd->bvec);
493 cmd->bvec = NULL;
1894e916 494 blk_mq_complete_request(rq);
92d77332
SL		 495}
496
bc07c10a
ML		 497static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
498{
499 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
bc07c10a 500
d4478e92
SL		 501 if (cmd->css)
502 css_put(cmd->css);
fe2cb290 503 cmd->ret = ret;
92d77332 504 lo_rw_aio_do_completion(cmd);
bc07c10a
ML		 505}
506
507static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
508 loff_t pos, bool rw)
509{
510 struct iov_iter iter;
511 struct bio_vec *bvec;
1894e916 512 struct request *rq = blk_mq_rq_from_pdu(cmd);
40326d8a 513 struct bio *bio = rq->bio;
bc07c10a 514 struct file *file = lo->lo_backing_file;
40326d8a
SL		 515 unsigned int offset;
516 int segments = 0;
bc07c10a
ML		 517 int ret;
518
40326d8a
SL		 519 if (rq->bio != rq->biotail) {
520 struct req_iterator iter;
521 struct bio_vec tmp;
522
523 __rq_for_each_bio(bio, rq)
524 segments += bio_segments(bio);
525 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
526 if (!bvec)
527 return -EIO;
528 cmd->bvec = bvec;
529
530 /*
531 * The bios of the request may be started from the middle of
532 * the 'bvec' because of bio splitting, so we can't directly
533 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
534 * API will take care of all details for us.
535 */
536 rq_for_each_segment(tmp, rq, iter) {
537 *bvec = tmp;
538 bvec++;
539 }
540 bvec = cmd->bvec;
541 offset = 0;
542 } else {
543 /*
544 * Same here, this bio may be started from the middle of the
545 * 'bvec' because of bio splitting, so offset from the bvec
546 * must be passed to iov iterator
547 */
548 offset = bio->bi_iter.bi_bvec_done;
549 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
550 segments = bio_segments(bio);
551 }
92d77332 552 atomic_set(&cmd->ref, 2);
bc07c10a 553
bc07c10a 554 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
40326d8a
SL		 555 segments, blk_rq_bytes(rq));
556 iter.iov_offset = offset;
bc07c10a
ML		 557
558 cmd->iocb.ki_pos = pos;
559 cmd->iocb.ki_filp = file;
560 cmd->iocb.ki_complete = lo_rw_aio_complete;
561 cmd->iocb.ki_flags = IOCB_DIRECT;
d4478e92
SL		 562 if (cmd->css)
563 kthread_associate_blkcg(cmd->css);
bc07c10a
ML		 564
565 if (rw == WRITE)
bb7462b6 566 ret = call_write_iter(file, &cmd->iocb, &iter);
bc07c10a 567 else
bb7462b6 568 ret = call_read_iter(file, &cmd->iocb, &iter);
bc07c10a 569
92d77332 570 lo_rw_aio_do_completion(cmd);
d4478e92 571 kthread_associate_blkcg(NULL);
92d77332 572
bc07c10a
ML		 573 if (ret != -EIOCBQUEUED)
574 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
575 return 0;
576}
577
c1c87c2b 578static int do_req_filebacked(struct loop_device *lo, struct request *rq)
bc07c10a
ML		 579{
580 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
c1c87c2b 581 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
bc07c10a
ML		 582
583 /*
584 * lo_write_simple and lo_read_simple should have been covered
585 * by io submit style function like lo_rw_aio(), one blocker
586 * is that lo_read_simple() need to call flush_dcache_page after
587 * the page is written from kernel, and it isn't easy to handle
588 * this in io submit style function which submits all segments
589 * of the req at one time. And direct read IO doesn't need to
590 * run flush_dcache_page().
591 */
c1c87c2b
CH		 592 switch (req_op(rq)) {
593 case REQ_OP_FLUSH:
594 return lo_req_flush(lo, rq);
595 case REQ_OP_DISCARD:
19372e27 596 case REQ_OP_WRITE_ZEROES:
c1c87c2b
CH		 597 return lo_discard(lo, rq, pos);
598 case REQ_OP_WRITE:
599 if (lo->transfer)
600 return lo_write_transfer(lo, rq, pos);
601 else if (cmd->use_aio)
602 return lo_rw_aio(lo, cmd, pos, WRITE);
af65aa8e 603 else
c1c87c2b
CH		 604 return lo_write_simple(lo, rq, pos);
605 case REQ_OP_READ:
aa4d8616 606 if (lo->transfer)
c1c87c2b
CH		 607 return lo_read_transfer(lo, rq, pos);
608 else if (cmd->use_aio)
609 return lo_rw_aio(lo, cmd, pos, READ);
aa4d8616 610 else
c1c87c2b
CH		 611 return lo_read_simple(lo, rq, pos);
612 default:
613 WARN_ON_ONCE(1);
614 return -EIO;
615 break;
aa4d8616 616 }
1da177e4
LT		 617}
618
2e5ab5f3
ML		 619static inline void loop_update_dio(struct loop_device *lo)
620{
621 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
622 lo->use_dio);
623}
624
06f0e9e6
ML		 625static void loop_reread_partitions(struct loop_device *lo,
626 struct block_device *bdev)
627{
628 int rc;
629
630 /*
631 * bd_mutex has been held already in release path, so don't
632 * acquire it if this function is called in such case.
633 *
634 * If the reread partition isn't from release path, lo_refcnt
635 * must be at least one and it can only become zero when the
636 * current holder is released.
637 */
638 if (!atomic_read(&lo->lo_refcnt))
639 rc = __blkdev_reread_part(bdev);
640 else
641 rc = blkdev_reread_part(bdev);
642 if (rc)
643 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
644 __func__, lo->lo_number, lo->lo_file_name, rc);
645}
646
1da177e4
LT		 647/*
648 * loop_change_fd switched the backing store of a loopback device to
649 * a new file. This is useful for operating system installers to free up
650 * the original file and in High Availability environments to switch to
651 * an alternative location for the content in case of server meltdown.
652 * This can only work if the loop device is used read-only, and if the
653 * new backing store is the same size and type as the old backing store.
654 */
bb214884
AV		 655static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
656 unsigned int arg)
1da177e4
LT		 657{
658 struct file *file, *old_file;
659 struct inode *inode;
660 int error;
661
662 error = -ENXIO;
663 if (lo->lo_state != Lo_bound)
664 goto out;
665
666 /* the loop device has to be read-only */
667 error = -EINVAL;
668 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
669 goto out;
670
671 error = -EBADF;
672 file = fget(arg);
673 if (!file)
674 goto out;
675
676 inode = file->f_mapping->host;
677 old_file = lo->lo_backing_file;
678
679 error = -EINVAL;
680
681 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 goto out_putf;
683
1da177e4
LT		 684 /* size of the new backing store needs to be the same */
685 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
686 goto out_putf;
687
688 /* and ... switch */
43cade80
OS		 689 blk_mq_freeze_queue(lo->lo_queue);
690 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
691 lo->lo_backing_file = file;
692 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
693 mapping_set_gfp_mask(file->f_mapping,
694 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
695 loop_update_dio(lo);
696 blk_mq_unfreeze_queue(lo->lo_queue);
1da177e4
LT		 697
698 fput(old_file);
e03c8dd1 699 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
06f0e9e6 700 loop_reread_partitions(lo, bdev);
1da177e4
LT		 701 return 0;
702
703 out_putf:
704 fput(file);
705 out:
706 return error;
707}
708
709static inline int is_loop_device(struct file *file)
710{
711 struct inode *i = file->f_mapping->host;
712
713 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
714}
715
ee862730
MB		 716/* loop sysfs attributes */
717
718static ssize_t loop_attr_show(struct device *dev, char *page,
719 ssize_t (*callback)(struct loop_device *, char *))
720{
34dd82af
KS		 721 struct gendisk *disk = dev_to_disk(dev);
722 struct loop_device *lo = disk->private_data;
ee862730 723
34dd82af 724 return callback(lo, page);
ee862730
MB		 725}
726
727#define LOOP_ATTR_RO(_name) \
728static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
729static ssize_t loop_attr_do_show_##_name(struct device *d, \
730 struct device_attribute *attr, char *b) \
731{ \
732 return loop_attr_show(d, b, loop_attr_##_name##_show); \
733} \
734static struct device_attribute loop_attr_##_name = \
735 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
736
737static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
738{
739 ssize_t ret;
740 char *p = NULL;
741
05eb0f25 742 spin_lock_irq(&lo->lo_lock);
ee862730 743 if (lo->lo_backing_file)
9bf39ab2 744 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
05eb0f25 745 spin_unlock_irq(&lo->lo_lock);
ee862730
MB		 746
747 if (IS_ERR_OR_NULL(p))
748 ret = PTR_ERR(p);
749 else {
750 ret = strlen(p);
751 memmove(buf, p, ret);
752 buf[ret++] = '\n';
753 buf[ret] = 0;
754 }
755
756 return ret;
757}
758
759static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
760{
761 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
762}
763
764static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
765{
766 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
767}
768
769static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
770{
771 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
772
773 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
774}
775
e03c8dd1
KS		 776static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
777{
778 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
779
780 return sprintf(buf, "%s\n", partscan ? "1" : "0");
781}
782
2e5ab5f3
ML		 783static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
784{
785 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
786
787 return sprintf(buf, "%s\n", dio ? "1" : "0");
788}
789
ee862730
MB		 790LOOP_ATTR_RO(backing_file);
791LOOP_ATTR_RO(offset);
792LOOP_ATTR_RO(sizelimit);
793LOOP_ATTR_RO(autoclear);
e03c8dd1 794LOOP_ATTR_RO(partscan);
2e5ab5f3 795LOOP_ATTR_RO(dio);
ee862730
MB		 796
797static struct attribute *loop_attrs[] = {
798 &loop_attr_backing_file.attr,
799 &loop_attr_offset.attr,
800 &loop_attr_sizelimit.attr,
801 &loop_attr_autoclear.attr,
e03c8dd1 802 &loop_attr_partscan.attr,
2e5ab5f3 803 &loop_attr_dio.attr,
ee862730
MB		 804 NULL,
805};
806
807static struct attribute_group loop_attribute_group = {
808 .name = "loop",
809 .attrs= loop_attrs,
810};
811
812static int loop_sysfs_init(struct loop_device *lo)
813{
814 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
815 &loop_attribute_group);
816}
817
818static void loop_sysfs_exit(struct loop_device *lo)
819{
820 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
821 &loop_attribute_group);
822}
823
dfaa2ef6
LC		 824static void loop_config_discard(struct loop_device *lo)
825{
826 struct file *file = lo->lo_backing_file;
827 struct inode *inode = file->f_mapping->host;
828 struct request_queue *q = lo->lo_queue;
829
830 /*
831 * We use punch hole to reclaim the free space used by the
12a64d2f 832 * image a.k.a. discard. However we do not support discard if
dfaa2ef6
LC		 833 * encryption is enabled, because it may give an attacker
834 * useful information.
835 */
836 if ((!file->f_op->fallocate) ||
837 lo->lo_encrypt_key_size) {
838 q->limits.discard_granularity = 0;
839 q->limits.discard_alignment = 0;
2bb4cd5c 840 blk_queue_max_discard_sectors(q, 0);
19372e27 841 blk_queue_max_write_zeroes_sectors(q, 0);
8b904b5b 842 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
dfaa2ef6
LC		 843 return;
844 }
845
846 q->limits.discard_granularity = inode->i_sb->s_blocksize;
dfaf3c03 847 q->limits.discard_alignment = 0;
f2c6df7d 848
1e6ec9ea
OS		 849 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
850 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
8b904b5b 851 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
dfaa2ef6
LC		 852}
853
e03a3d7a
ML		 854static void loop_unprepare_queue(struct loop_device *lo)
855{
3989144f 856 kthread_flush_worker(&lo->worker);
e03a3d7a
ML		 857 kthread_stop(lo->worker_task);
858}
859
b2ee7d46
N		 860static int loop_kthread_worker_fn(void *worker_ptr)
861{
862 current->flags |= PF_LESS_THROTTLE;
863 return kthread_worker_fn(worker_ptr);
864}
865
e03a3d7a
ML		 866static int loop_prepare_queue(struct loop_device *lo)
867{
3989144f 868 kthread_init_worker(&lo->worker);
b2ee7d46 869 lo->worker_task = kthread_run(loop_kthread_worker_fn,
e03a3d7a
ML		 870 &lo->worker, "loop%d", lo->lo_number);
871 if (IS_ERR(lo->worker_task))
872 return -ENOMEM;
873 set_user_nice(lo->worker_task, MIN_NICE);
874 return 0;
875}
876
bb214884 877static int loop_set_fd(struct loop_device *lo, fmode_t mode,
1da177e4
LT		 878 struct block_device *bdev, unsigned int arg)
879{
880 struct file *file, *f;
881 struct inode *inode;
882 struct address_space *mapping;
1da177e4
LT		 883 int lo_flags = 0;
884 int error;
885 loff_t size;
886
887 /* This is safe, since we have a reference from open(). */
888 __module_get(THIS_MODULE);
889
890 error = -EBADF;
891 file = fget(arg);
892 if (!file)
893 goto out;
894
895 error = -EBUSY;
896 if (lo->lo_state != Lo_unbound)
897 goto out_putf;
898
899 /* Avoid recursion */
900 f = file;
901 while (is_loop_device(f)) {
902 struct loop_device *l;
903
bb214884 904 if (f->f_mapping->host->i_bdev == bdev)
1da177e4
LT		 905 goto out_putf;
906
907 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
908 if (l->lo_state == Lo_unbound) {
909 error = -EINVAL;
910 goto out_putf;
911 }
912 f = l->lo_backing_file;
913 }
914
915 mapping = file->f_mapping;
916 inode = mapping->host;
917
1da177e4 918 error = -EINVAL;
456be148
CH		 919 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
920 goto out_putf;
1da177e4 921
456be148 922 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
283e7e5d 923 !file->f_op->write_iter)
456be148 924 lo_flags |= LO_FLAGS_READ_ONLY;
ba52de12 925
456be148 926 error = -EFBIG;
1da177e4 927 size = get_loop_size(lo, file);
456be148 928 if ((loff_t)(sector_t)size != size)
1da177e4 929 goto out_putf;
e03a3d7a
ML		 930 error = loop_prepare_queue(lo);
931 if (error)
f4aa4c7b 932 goto out_putf;
1da177e4 933
456be148 934 error = 0;
1da177e4
LT		 935
936 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
937
2e5ab5f3 938 lo->use_dio = false;
1da177e4
LT		 939 lo->lo_device = bdev;
940 lo->lo_flags = lo_flags;
941 lo->lo_backing_file = file;
aa4d8616 942 lo->transfer = NULL;
1da177e4
LT		 943 lo->ioctl = NULL;
944 lo->lo_sizelimit = 0;
945 lo->old_gfp_mask = mapping_gfp_mask(mapping);
946 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
947
68db1961 948 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
21d0727f 949 blk_queue_write_cache(lo->lo_queue, true, false);
68db1961 950
2e5ab5f3 951 loop_update_dio(lo);
73285082 952 set_capacity(lo->lo_disk, size);
1da177e4 953 bd_set_size(bdev, size << 9);
ee862730 954 loop_sysfs_init(lo);
c3473c63
DZ		 955 /* let user-space know about the new size */
956 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1da177e4 957
8a0740c4
OS		 958 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
959 block_size(inode->i_bdev) : PAGE_SIZE);
1da177e4 960
6c997918 961 lo->lo_state = Lo_bound;
e03c8dd1
KS		 962 if (part_shift)
963 lo->lo_flags |= LO_FLAGS_PARTSCAN;
964 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
06f0e9e6 965 loop_reread_partitions(lo, bdev);
c1681bf8
AP		 966
967 /* Grab the block_device to prevent its destruction after we
968 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
969 */
970 bdgrab(bdev);
1da177e4
LT		 971 return 0;
972
973 out_putf:
974 fput(file);
975 out:
976 /* This is safe: open() is still holding a reference. */
977 module_put(THIS_MODULE);
978 return error;
979}
980
981static int
982loop_release_xfer(struct loop_device *lo)
983{
984 int err = 0;
985 struct loop_func_table *xfer = lo->lo_encryption;
986
987 if (xfer) {
988 if (xfer->release)
989 err = xfer->release(lo);
990 lo->transfer = NULL;
991 lo->lo_encryption = NULL;
992 module_put(xfer->owner);
993 }
994 return err;
995}
996
997static int
998loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
999 const struct loop_info64 *i)
1000{
1001 int err = 0;
1002
1003 if (xfer) {
1004 struct module *owner = xfer->owner;
1005
1006 if (!try_module_get(owner))
1007 return -EINVAL;
1008 if (xfer->init)
1009 err = xfer->init(lo, i);
1010 if (err)
1011 module_put(owner);
1012 else
1013 lo->lo_encryption = xfer;
1014 }
1015 return err;
1016}
1017
4c823cc3 1018static int loop_clr_fd(struct loop_device *lo)
1da177e4
LT		 1019{
1020 struct file *filp = lo->lo_backing_file;
b4e3ca1a 1021 gfp_t gfp = lo->old_gfp_mask;
4c823cc3 1022 struct block_device *bdev = lo->lo_device;
1da177e4
LT		 1023
1024 if (lo->lo_state != Lo_bound)
1025 return -ENXIO;
1026
a1ecac3b
DC		 1027 /*
1028 * If we've explicitly asked to tear down the loop device,
1029 * and it has an elevated reference count, set it for auto-teardown when
1030 * the last reference goes away. This stops $!~#$@ udev from
1031 * preventing teardown because it decided that it needs to run blkid on
1032 * the loopback device whenever they appear. xfstests is notorious for
1033 * failing tests because blkid via udev races with a losetup
1034 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1035 * command to fail with EBUSY.
1036 */
f8933667 1037 if (atomic_read(&lo->lo_refcnt) > 1) {
a1ecac3b
DC		 1038 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1039 mutex_unlock(&lo->lo_ctl_mutex);
1040 return 0;
1041 }
1da177e4
LT		 1042
1043 if (filp == NULL)
1044 return -EINVAL;
1045
f8933667
ML		 1046 /* freeze request queue during the transition */
1047 blk_mq_freeze_queue(lo->lo_queue);
1048
1da177e4
LT		 1049 spin_lock_irq(&lo->lo_lock);
1050 lo->lo_state = Lo_rundown;
1da177e4 1051 lo->lo_backing_file = NULL;
05eb0f25 1052 spin_unlock_irq(&lo->lo_lock);
1da177e4
LT		 1053
1054 loop_release_xfer(lo);
1055 lo->transfer = NULL;
1056 lo->ioctl = NULL;
1057 lo->lo_device = NULL;
1058 lo->lo_encryption = NULL;
1059 lo->lo_offset = 0;
1060 lo->lo_sizelimit = 0;
1061 lo->lo_encrypt_key_size = 0;
1da177e4
LT		 1062 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1063 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1064 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
89e4fdec 1065 blk_queue_logical_block_size(lo->lo_queue, 512);
bf093753
OS		 1066 blk_queue_physical_block_size(lo->lo_queue, 512);
1067 blk_queue_io_min(lo->lo_queue, 512);
c1681bf8
AP		 1068 if (bdev) {
1069 bdput(bdev);
bb214884 1070 invalidate_bdev(bdev);
c1681bf8 1071 }
73285082 1072 set_capacity(lo->lo_disk, 0);
51a0bb0c 1073 loop_sysfs_exit(lo);
c3473c63 1074 if (bdev) {
bb214884 1075 bd_set_size(bdev, 0);
c3473c63
DZ		 1076 /* let user-space know about this change */
1077 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1078 }
1da177e4
LT		 1079 mapping_set_gfp_mask(filp->f_mapping, gfp);
1080 lo->lo_state = Lo_unbound;
1da177e4
LT		 1081 /* This is safe: open() is still holding a reference. */
1082 module_put(THIS_MODULE);
f8933667
ML		 1083 blk_mq_unfreeze_queue(lo->lo_queue);
1084
c2fccc1c 1085 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
06f0e9e6 1086 loop_reread_partitions(lo, bdev);
e03c8dd1
KS		 1087 lo->lo_flags = 0;
1088 if (!part_shift)
1089 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
e03a3d7a 1090 loop_unprepare_queue(lo);
f028f3b2
NK		 1091 mutex_unlock(&lo->lo_ctl_mutex);
1092 /*
1093 * Need not hold lo_ctl_mutex to fput backing file.
1094 * Calling fput holding lo_ctl_mutex triggers a circular
1095 * lock dependency possibility warning as fput can take
1096 * bd_mutex which is usually taken before lo_ctl_mutex.
1097 */
1098 fput(filp);
1da177e4
LT		 1099 return 0;
1100}
1101
1102static int
1103loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1104{
1105 int err;
1106 struct loop_func_table *xfer;
e4849737 1107 kuid_t uid = current_uid();
1da177e4 1108
b0fafa81 1109 if (lo->lo_encrypt_key_size &&
e4849737 1110 !uid_eq(lo->lo_key_owner, uid) &&
1da177e4
LT		 1111 !capable(CAP_SYS_ADMIN))
1112 return -EPERM;
1113 if (lo->lo_state != Lo_bound)
1114 return -ENXIO;
1115 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1116 return -EINVAL;
1117
ecdd0959
ML		 1118 /* I/O need to be drained during transfer transition */
1119 blk_mq_freeze_queue(lo->lo_queue);
1120
1da177e4
LT		 1121 err = loop_release_xfer(lo);
1122 if (err)
ecdd0959 1123 goto exit;
1da177e4
LT		 1124
1125 if (info->lo_encrypt_type) {
1126 unsigned int type = info->lo_encrypt_type;
1127
1e047eaa
TH		 1128 if (type >= MAX_LO_CRYPT) {
1129 err = -EINVAL;
1130 goto exit;
1131 }
1da177e4 1132 xfer = xfer_funcs[type];
1e047eaa
TH		 1133 if (xfer == NULL) {
1134 err = -EINVAL;
1135 goto exit;
1136 }
1da177e4
LT		 1137 } else
1138 xfer = NULL;
1139
1140 err = loop_init_xfer(lo, xfer, info);
1141 if (err)
ecdd0959 1142 goto exit;
1da177e4
LT		 1143
1144 if (lo->lo_offset != info->lo_offset ||
1e6ec9ea
OS		 1145 lo->lo_sizelimit != info->lo_sizelimit) {
1146 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
ecdd0959
ML		 1147 err = -EFBIG;
1148 goto exit;
1149 }
b040ad9c 1150 }
541c742a 1151
dfaa2ef6 1152 loop_config_discard(lo);
1da177e4
LT		 1153
1154 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1155 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1156 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1157 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1158
1159 if (!xfer)
1160 xfer = &none_funcs;
1161 lo->transfer = xfer->transfer;
1162 lo->ioctl = xfer->ioctl;
1163
96c58655
DW		 1164 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1165 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1166 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1167
1da177e4
LT		 1168 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1169 lo->lo_init[0] = info->lo_init[0];
1170 lo->lo_init[1] = info->lo_init[1];
1171 if (info->lo_encrypt_key_size) {
1172 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1173 info->lo_encrypt_key_size);
b0fafa81 1174 lo->lo_key_owner = uid;
aa4d8616 1175 }
1da177e4 1176
2e5ab5f3
ML		 1177 /* update dio if lo_offset or transfer is changed */
1178 __loop_update_dio(lo, lo->use_dio);
1179
ecdd0959
ML		 1180 exit:
1181 blk_mq_unfreeze_queue(lo->lo_queue);
e02898b4
OS		 1182
1183 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1184 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1185 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1186 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1187 loop_reread_partitions(lo, lo->lo_device);
1188 }
1189
ecdd0959 1190 return err;
1da177e4
LT		 1191}
1192
1193static int
1194loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1195{
2d1d4c1e 1196 struct file *file;
1da177e4 1197 struct kstat stat;
2d1d4c1e 1198 int ret;
1da177e4 1199
2d1d4c1e
OS		 1200 if (lo->lo_state != Lo_bound) {
1201 mutex_unlock(&lo->lo_ctl_mutex);
1da177e4 1202 return -ENXIO;
2d1d4c1e
OS		 1203 }
1204
1da177e4
LT		 1205 memset(info, 0, sizeof(*info));
1206 info->lo_number = lo->lo_number;
1da177e4
LT		 1207 info->lo_offset = lo->lo_offset;
1208 info->lo_sizelimit = lo->lo_sizelimit;
1209 info->lo_flags = lo->lo_flags;
1210 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1211 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1212 info->lo_encrypt_type =
1213 lo->lo_encryption ? lo->lo_encryption->number : 0;
1214 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1215 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1216 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1217 lo->lo_encrypt_key_size);
1218 }
2d1d4c1e
OS		 1219
1220 /* Drop lo_ctl_mutex while we call into the filesystem. */
1221 file = get_file(lo->lo_backing_file);
1222 mutex_unlock(&lo->lo_ctl_mutex);
1223 ret = vfs_getattr(&file->f_path, &stat, STATX_INO,
1224 AT_STATX_SYNC_AS_STAT);
1225 if (!ret) {
1226 info->lo_device = huge_encode_dev(stat.dev);
1227 info->lo_inode = stat.ino;
1228 info->lo_rdevice = huge_encode_dev(stat.rdev);
1229 }
1230 fput(file);
1231 return ret;
1da177e4
LT		 1232}
1233
1234static void
1235loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1236{
1237 memset(info64, 0, sizeof(*info64));
1238 info64->lo_number = info->lo_number;
1239 info64->lo_device = info->lo_device;
1240 info64->lo_inode = info->lo_inode;
1241 info64->lo_rdevice = info->lo_rdevice;
1242 info64->lo_offset = info->lo_offset;
1243 info64->lo_sizelimit = 0;
1244 info64->lo_encrypt_type = info->lo_encrypt_type;
1245 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1246 info64->lo_flags = info->lo_flags;
1247 info64->lo_init[0] = info->lo_init[0];
1248 info64->lo_init[1] = info->lo_init[1];
1249 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1250 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1251 else
1252 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1253 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1254}
1255
1256static int
1257loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1258{
1259 memset(info, 0, sizeof(*info));
1260 info->lo_number = info64->lo_number;
1261 info->lo_device = info64->lo_device;
1262 info->lo_inode = info64->lo_inode;
1263 info->lo_rdevice = info64->lo_rdevice;
1264 info->lo_offset = info64->lo_offset;
1265 info->lo_encrypt_type = info64->lo_encrypt_type;
1266 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1267 info->lo_flags = info64->lo_flags;
1268 info->lo_init[0] = info64->lo_init[0];
1269 info->lo_init[1] = info64->lo_init[1];
1270 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1271 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1272 else
1273 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1274 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1275
1276 /* error in case values were truncated */
1277 if (info->lo_device != info64->lo_device ||
1278 info->lo_rdevice != info64->lo_rdevice ||
1279 info->lo_inode != info64->lo_inode ||
1280 info->lo_offset != info64->lo_offset)
1281 return -EOVERFLOW;
1282
1283 return 0;
1284}
1285
1286static int
1287loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1288{
1289 struct loop_info info;
1290 struct loop_info64 info64;
1291
1292 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1293 return -EFAULT;
1294 loop_info64_from_old(&info, &info64);
1295 return loop_set_status(lo, &info64);
1296}
1297
1298static int
1299loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1300{
1301 struct loop_info64 info64;
1302
1303 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1304 return -EFAULT;
1305 return loop_set_status(lo, &info64);
1306}
1307
1308static int
1309loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1310 struct loop_info info;
1311 struct loop_info64 info64;
bdac616d 1312 int err;
1da177e4 1313
bdac616d
OS		 1314 if (!arg) {
1315 mutex_unlock(&lo->lo_ctl_mutex);
1316 return -EINVAL;
1317 }
1318 err = loop_get_status(lo, &info64);
1da177e4
LT		 1319 if (!err)
1320 err = loop_info64_to_old(&info64, &info);
1321 if (!err && copy_to_user(arg, &info, sizeof(info)))
1322 err = -EFAULT;
1323
1324 return err;
1325}
1326
1327static int
1328loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1329 struct loop_info64 info64;
bdac616d 1330 int err;
1da177e4 1331
bdac616d
OS		 1332 if (!arg) {
1333 mutex_unlock(&lo->lo_ctl_mutex);
1334 return -EINVAL;
1335 }
1336 err = loop_get_status(lo, &info64);
1da177e4
LT		 1337 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1338 err = -EFAULT;
1339
1340 return err;
1341}
1342
51001b7d 1343static int loop_set_capacity(struct loop_device *lo)
53d66608 1344{
53d66608 1345 if (unlikely(lo->lo_state != Lo_bound))
7b0576a3 1346 return -ENXIO;
53d66608 1347
1e6ec9ea 1348 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
53d66608
O		 1349}
1350
ab1cb278
ML		 1351static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1352{
1353 int error = -ENXIO;
1354 if (lo->lo_state != Lo_bound)
1355 goto out;
1356
1357 __loop_update_dio(lo, !!arg);
1358 if (lo->use_dio == !!arg)
1359 return 0;
1360 error = -EINVAL;
1361 out:
1362 return error;
1363}
1364
89e4fdec
OS		 1365static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1366{
1367 if (lo->lo_state != Lo_bound)
1368 return -ENXIO;
1369
1370 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1371 return -EINVAL;
1372
1373 blk_mq_freeze_queue(lo->lo_queue);
1374
1375 blk_queue_logical_block_size(lo->lo_queue, arg);
bf093753
OS		 1376 blk_queue_physical_block_size(lo->lo_queue, arg);
1377 blk_queue_io_min(lo->lo_queue, arg);
89e4fdec
OS		 1378 loop_update_dio(lo);
1379
1380 blk_mq_unfreeze_queue(lo->lo_queue);
1381
1382 return 0;
1383}
1384
bb214884 1385static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 1386 unsigned int cmd, unsigned long arg)
1387{
bb214884 1388 struct loop_device *lo = bdev->bd_disk->private_data;
1da177e4
LT		 1389 int err;
1390
3148ffbd
OS		 1391 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1);
1392 if (err)
1393 goto out_unlocked;
1394
1da177e4
LT		 1395 switch (cmd) {
1396 case LOOP_SET_FD:
bb214884 1397 err = loop_set_fd(lo, mode, bdev, arg);
1da177e4
LT		 1398 break;
1399 case LOOP_CHANGE_FD:
bb214884 1400 err = loop_change_fd(lo, bdev, arg);
1da177e4
LT		 1401 break;
1402 case LOOP_CLR_FD:
f028f3b2 1403 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
4c823cc3 1404 err = loop_clr_fd(lo);
f028f3b2
NK		 1405 if (!err)
1406 goto out_unlocked;
1da177e4
LT		 1407 break;
1408 case LOOP_SET_STATUS:
7035b5df
DM		 1409 err = -EPERM;
1410 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1411 err = loop_set_status_old(lo,
1412 (struct loop_info __user *)arg);
1da177e4
LT		 1413 break;
1414 case LOOP_GET_STATUS:
1415 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
2d1d4c1e
OS		 1416 /* loop_get_status() unlocks lo_ctl_mutex */
1417 goto out_unlocked;
1da177e4 1418 case LOOP_SET_STATUS64:
7035b5df
DM		 1419 err = -EPERM;
1420 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1421 err = loop_set_status64(lo,
1422 (struct loop_info64 __user *) arg);
1da177e4
LT		 1423 break;
1424 case LOOP_GET_STATUS64:
1425 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
2d1d4c1e
OS		 1426 /* loop_get_status() unlocks lo_ctl_mutex */
1427 goto out_unlocked;
53d66608
O		 1428 case LOOP_SET_CAPACITY:
1429 err = -EPERM;
1430 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
51001b7d 1431 err = loop_set_capacity(lo);
53d66608 1432 break;
ab1cb278
ML		 1433 case LOOP_SET_DIRECT_IO:
1434 err = -EPERM;
1435 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1436 err = loop_set_dio(lo, arg);
1437 break;
89e4fdec
OS		 1438 case LOOP_SET_BLOCK_SIZE:
1439 err = -EPERM;
1440 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1441 err = loop_set_block_size(lo, arg);
1442 break;
1da177e4
LT		 1443 default:
1444 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1445 }
f85221dd 1446 mutex_unlock(&lo->lo_ctl_mutex);
f028f3b2
NK		 1447
1448out_unlocked:
1da177e4
LT		 1449 return err;
1450}
1451
863d5b82
DH		 1452#ifdef CONFIG_COMPAT
1453struct compat_loop_info {
1454 compat_int_t lo_number; /* ioctl r/o */
1455 compat_dev_t lo_device; /* ioctl r/o */
1456 compat_ulong_t lo_inode; /* ioctl r/o */
1457 compat_dev_t lo_rdevice; /* ioctl r/o */
1458 compat_int_t lo_offset;
1459 compat_int_t lo_encrypt_type;
1460 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1461 compat_int_t lo_flags; /* ioctl r/o */
1462 char lo_name[LO_NAME_SIZE];
1463 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1464 compat_ulong_t lo_init[2];
1465 char reserved[4];
1466};
1467
1468/*
1469 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1470 * - noinlined to reduce stack space usage in main part of driver
1471 */
1472static noinline int
ba674cfc 1473loop_info64_from_compat(const struct compat_loop_info __user *arg,
863d5b82
DH		 1474 struct loop_info64 *info64)
1475{
1476 struct compat_loop_info info;
1477
1478 if (copy_from_user(&info, arg, sizeof(info)))
1479 return -EFAULT;
1480
1481 memset(info64, 0, sizeof(*info64));
1482 info64->lo_number = info.lo_number;
1483 info64->lo_device = info.lo_device;
1484 info64->lo_inode = info.lo_inode;
1485 info64->lo_rdevice = info.lo_rdevice;
1486 info64->lo_offset = info.lo_offset;
1487 info64->lo_sizelimit = 0;
1488 info64->lo_encrypt_type = info.lo_encrypt_type;
1489 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1490 info64->lo_flags = info.lo_flags;
1491 info64->lo_init[0] = info.lo_init[0];
1492 info64->lo_init[1] = info.lo_init[1];
1493 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1494 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1495 else
1496 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1497 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1498 return 0;
1499}
1500
1501/*
1502 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1503 * - noinlined to reduce stack space usage in main part of driver
1504 */
1505static noinline int
1506loop_info64_to_compat(const struct loop_info64 *info64,
1507 struct compat_loop_info __user *arg)
1508{
1509 struct compat_loop_info info;
1510
1511 memset(&info, 0, sizeof(info));
1512 info.lo_number = info64->lo_number;
1513 info.lo_device = info64->lo_device;
1514 info.lo_inode = info64->lo_inode;
1515 info.lo_rdevice = info64->lo_rdevice;
1516 info.lo_offset = info64->lo_offset;
1517 info.lo_encrypt_type = info64->lo_encrypt_type;
1518 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1519 info.lo_flags = info64->lo_flags;
1520 info.lo_init[0] = info64->lo_init[0];
1521 info.lo_init[1] = info64->lo_init[1];
1522 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1523 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1524 else
1525 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1526 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1527
1528 /* error in case values were truncated */
1529 if (info.lo_device != info64->lo_device ||
1530 info.lo_rdevice != info64->lo_rdevice ||
1531 info.lo_inode != info64->lo_inode ||
1532 info.lo_offset != info64->lo_offset ||
1533 info.lo_init[0] != info64->lo_init[0] ||
1534 info.lo_init[1] != info64->lo_init[1])
1535 return -EOVERFLOW;
1536
1537 if (copy_to_user(arg, &info, sizeof(info)))
1538 return -EFAULT;
1539 return 0;
1540}
1541
1542static int
1543loop_set_status_compat(struct loop_device *lo,
1544 const struct compat_loop_info __user *arg)
1545{
1546 struct loop_info64 info64;
1547 int ret;
1548
1549 ret = loop_info64_from_compat(arg, &info64);
1550 if (ret < 0)
1551 return ret;
1552 return loop_set_status(lo, &info64);
1553}
1554
1555static int
1556loop_get_status_compat(struct loop_device *lo,
1557 struct compat_loop_info __user *arg)
1558{
1559 struct loop_info64 info64;
bdac616d 1560 int err;
863d5b82 1561
bdac616d
OS		 1562 if (!arg) {
1563 mutex_unlock(&lo->lo_ctl_mutex);
1564 return -EINVAL;
1565 }
1566 err = loop_get_status(lo, &info64);
863d5b82
DH		 1567 if (!err)
1568 err = loop_info64_to_compat(&info64, arg);
1569 return err;
1570}
1571
bb214884
AV		 1572static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1573 unsigned int cmd, unsigned long arg)
863d5b82 1574{
bb214884 1575 struct loop_device *lo = bdev->bd_disk->private_data;
863d5b82
DH		 1576 int err;
1577
863d5b82
DH		 1578 switch(cmd) {
1579 case LOOP_SET_STATUS:
3148ffbd
OS		 1580 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1581 if (!err) {
1582 err = loop_set_status_compat(lo,
1583 (const struct compat_loop_info __user *)arg);
1584 mutex_unlock(&lo->lo_ctl_mutex);
1585 }
863d5b82
DH		 1586 break;
1587 case LOOP_GET_STATUS:
3148ffbd
OS		 1588 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1589 if (!err) {
1590 err = loop_get_status_compat(lo,
1591 (struct compat_loop_info __user *)arg);
1592 /* loop_get_status() unlocks lo_ctl_mutex */
1593 }
863d5b82 1594 break;
53d66608 1595 case LOOP_SET_CAPACITY:
863d5b82
DH		 1596 case LOOP_CLR_FD:
1597 case LOOP_GET_STATUS64:
1598 case LOOP_SET_STATUS64:
1599 arg = (unsigned long) compat_ptr(arg);
1600 case LOOP_SET_FD:
1601 case LOOP_CHANGE_FD:
bb214884 1602 err = lo_ioctl(bdev, mode, cmd, arg);
863d5b82
DH		 1603 break;
1604 default:
1605 err = -ENOIOCTLCMD;
1606 break;
1607 }
863d5b82
DH		 1608 return err;
1609}
1610#endif
1611
bb214884 1612static int lo_open(struct block_device *bdev, fmode_t mode)
1da177e4 1613{
770fe30a
KS		 1614 struct loop_device *lo;
1615 int err = 0;
1616
1617 mutex_lock(&loop_index_mutex);
1618 lo = bdev->bd_disk->private_data;
1619 if (!lo) {
1620 err = -ENXIO;
1621 goto out;
1622 }
1da177e4 1623
f8933667 1624 atomic_inc(&lo->lo_refcnt);
770fe30a
KS		 1625out:
1626 mutex_unlock(&loop_index_mutex);
1627 return err;
1da177e4
LT		 1628}
1629
ae665016 1630static void __lo_release(struct loop_device *lo)
1da177e4 1631{
ffcd7dca 1632 int err;
1da177e4 1633
f8933667
ML		 1634 if (atomic_dec_return(&lo->lo_refcnt))
1635 return;
14f27939 1636
f8933667 1637 mutex_lock(&lo->lo_ctl_mutex);
14f27939
MB		 1638 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1639 /*
1640 * In autoclear mode, stop the loop thread
1641 * and remove configuration after last close.
1642 */
4c823cc3 1643 err = loop_clr_fd(lo);
ffcd7dca 1644 if (!err)
db2a144b 1645 return;
43cade80 1646 } else if (lo->lo_state == Lo_bound) {
14f27939
MB		 1647 /*
1648 * Otherwise keep thread (if running) and config,
1649 * but flush possible ongoing bios in thread.
1650 */
43cade80
OS		 1651 blk_mq_freeze_queue(lo->lo_queue);
1652 blk_mq_unfreeze_queue(lo->lo_queue);
14f27939 1653 }
96c58655 1654
f85221dd 1655 mutex_unlock(&lo->lo_ctl_mutex);
1da177e4
LT		 1656}
1657
ae665016
LT		 1658static void lo_release(struct gendisk *disk, fmode_t mode)
1659{
1660 mutex_lock(&loop_index_mutex);
1661 __lo_release(disk->private_data);
1662 mutex_unlock(&loop_index_mutex);
1663}
1664
83d5cde4 1665static const struct block_device_operations lo_fops = {
1da177e4 1666 .owner = THIS_MODULE,
bb214884
AV		 1667 .open = lo_open,
1668 .release = lo_release,
1669 .ioctl = lo_ioctl,
863d5b82 1670#ifdef CONFIG_COMPAT
bb214884 1671 .compat_ioctl = lo_compat_ioctl,
863d5b82 1672#endif
1da177e4
LT		 1673};
1674
1675/*
1676 * And now the modules code and kernel interface.
1677 */
73285082 1678static int max_loop;
ac04fee0 1679module_param(max_loop, int, S_IRUGO);
a47653fc 1680MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
ac04fee0 1681module_param(max_part, int, S_IRUGO);
476a4813 1682MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1da177e4
LT		 1683MODULE_LICENSE("GPL");
1684MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1685
1686int loop_register_transfer(struct loop_func_table *funcs)
1687{
1688 unsigned int n = funcs->number;
1689
1690 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1691 return -EINVAL;
1692 xfer_funcs[n] = funcs;
1693 return 0;
1694}
1695
34dd82af
KS		 1696static int unregister_transfer_cb(int id, void *ptr, void *data)
1697{
1698 struct loop_device *lo = ptr;
1699 struct loop_func_table *xfer = data;
1700
1701 mutex_lock(&lo->lo_ctl_mutex);
1702 if (lo->lo_encryption == xfer)
1703 loop_release_xfer(lo);
1704 mutex_unlock(&lo->lo_ctl_mutex);
1705 return 0;
1706}
1707
1da177e4
LT		 1708int loop_unregister_transfer(int number)
1709{
1710 unsigned int n = number;
1da177e4
LT		 1711 struct loop_func_table *xfer;
1712
1713 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1714 return -EINVAL;
1715
1716 xfer_funcs[n] = NULL;
34dd82af 1717 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1da177e4
LT		 1718 return 0;
1719}
1720
1721EXPORT_SYMBOL(loop_register_transfer);
1722EXPORT_SYMBOL(loop_unregister_transfer);
1723
fc17b653 1724static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
b5dd2f60
ML		 1725 const struct blk_mq_queue_data *bd)
1726{
1894e916
JA		 1727 struct request *rq = bd->rq;
1728 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1729 struct loop_device *lo = rq->q->queuedata;
b5dd2f60 1730
1894e916 1731 blk_mq_start_request(rq);
b5dd2f60 1732
f4aa4c7b 1733 if (lo->lo_state != Lo_bound)
fc17b653 1734 return BLK_STS_IOERR;
f4aa4c7b 1735
1894e916 1736 switch (req_op(rq)) {
f0225cac
CH		 1737 case REQ_OP_FLUSH:
1738 case REQ_OP_DISCARD:
19372e27 1739 case REQ_OP_WRITE_ZEROES:
bc07c10a 1740 cmd->use_aio = false;
f0225cac
CH		 1741 break;
1742 default:
1743 cmd->use_aio = lo->use_dio;
1744 break;
1745 }
bc07c10a 1746
d4478e92 1747 /* always use the first bio's css */
0b508bc9 1748#ifdef CONFIG_BLK_CGROUP
1894e916
JA		 1749 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1750 cmd->css = rq->bio->bi_css;
d4478e92
SL		 1751 css_get(cmd->css);
1752 } else
1753#endif
1754 cmd->css = NULL;
3989144f 1755 kthread_queue_work(&lo->worker, &cmd->work);
b5dd2f60 1756
fc17b653 1757 return BLK_STS_OK;
b5dd2f60
ML		 1758}
1759
1760static void loop_handle_cmd(struct loop_cmd *cmd)
1761{
1894e916
JA		 1762 struct request *rq = blk_mq_rq_from_pdu(cmd);
1763 const bool write = op_is_write(req_op(rq));
1764 struct loop_device *lo = rq->q->queuedata;
f4829a9b 1765 int ret = 0;
b5dd2f60 1766
f4829a9b
CH		 1767 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1768 ret = -EIO;
b5dd2f60 1769 goto failed;
f4829a9b 1770 }
b5dd2f60 1771
1894e916 1772 ret = do_req_filebacked(lo, rq);
b5dd2f60 1773 failed:
bc07c10a 1774 /* complete non-aio request */
fe2cb290
CH		 1775 if (!cmd->use_aio || ret) {
1776 cmd->ret = ret ? -EIO : 0;
1894e916 1777 blk_mq_complete_request(rq);
fe2cb290 1778 }
b5dd2f60
ML		 1779}
1780
e03a3d7a 1781static void loop_queue_work(struct kthread_work *work)
b5dd2f60
ML		 1782{
1783 struct loop_cmd *cmd =
e03a3d7a 1784 container_of(work, struct loop_cmd, work);
b5dd2f60
ML		 1785
1786 loop_handle_cmd(cmd);
1787}
1788
d6296d39
CH		 1789static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1790 unsigned int hctx_idx, unsigned int numa_node)
b5dd2f60
ML		 1791{
1792 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1793
3989144f 1794 kthread_init_work(&cmd->work, loop_queue_work);
b5dd2f60
ML		 1795 return 0;
1796}
1797
f363b089 1798static const struct blk_mq_ops loop_mq_ops = {
b5dd2f60 1799 .queue_rq = loop_queue_rq,
b5dd2f60 1800 .init_request = loop_init_request,
fe2cb290 1801 .complete = lo_complete_rq,
b5dd2f60
ML		 1802};
1803
34dd82af 1804static int loop_add(struct loop_device **l, int i)
73285082
KC		 1805{
1806 struct loop_device *lo;
1807 struct gendisk *disk;
34dd82af 1808 int err;
73285082 1809
68d740d7 1810 err = -ENOMEM;
73285082 1811 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
68d740d7 1812 if (!lo)
73285082 1813 goto out;
34dd82af 1814
ef7e7c82
MP		 1815 lo->lo_state = Lo_unbound;
1816
c718aa65 1817 /* allocate id, if @id >= 0, we're requesting that specific id */
34dd82af 1818 if (i >= 0) {
c718aa65
TH		 1819 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1820 if (err == -ENOSPC)
34dd82af 1821 err = -EEXIST;
34dd82af 1822 } else {
c718aa65 1823 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
34dd82af
KS		 1824 }
1825 if (err < 0)
1826 goto out_free_dev;
c718aa65 1827 i = err;
73285082 1828
183cfb57 1829 err = -ENOMEM;
b5dd2f60
ML		 1830 lo->tag_set.ops = &loop_mq_ops;
1831 lo->tag_set.nr_hw_queues = 1;
1832 lo->tag_set.queue_depth = 128;
1833 lo->tag_set.numa_node = NUMA_NO_NODE;
1834 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1835 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1836 lo->tag_set.driver_data = lo;
1837
1838 err = blk_mq_alloc_tag_set(&lo->tag_set);
1839 if (err)
3ec981e3 1840 goto out_free_idr;
73285082 1841
b5dd2f60
ML		 1842 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1843 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1844 err = PTR_ERR(lo->lo_queue);
1845 goto out_cleanup_tags;
1846 }
ef7e7c82
MP		 1847 lo->lo_queue->queuedata = lo;
1848
54bb0ade 1849 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
40326d8a 1850
5b5e20f4 1851 /*
40326d8a
SL		 1852 * By default, we do buffer IO, so it doesn't make sense to enable
1853 * merge because the I/O submitted to backing file is handled page by
1854 * page. For directio mode, merge does help to dispatch bigger request
1855 * to underlayer disk. We will enable merge once directio is enabled.
5b5e20f4 1856 */
8b904b5b 1857 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
5b5e20f4 1858
7a649737 1859 err = -ENOMEM;
476a4813 1860 disk = lo->lo_disk = alloc_disk(1 << part_shift);
73285082
KC		 1861 if (!disk)
1862 goto out_free_queue;
1863
e03c8dd1
KS		 1864 /*
1865 * Disable partition scanning by default. The in-kernel partition
1866 * scanning can be requested individually per-device during its
1867 * setup. Userspace can always add and remove partitions from all
1868 * devices. The needed partition minors are allocated from the
1869 * extended minor space, the main loop device numbers will continue
1870 * to match the loop minors, regardless of the number of partitions
1871 * used.
1872 *
1873 * If max_part is given, partition scanning is globally enabled for
1874 * all loop devices. The minors for the main loop devices will be
1875 * multiples of max_part.
1876 *
1877 * Note: Global-for-all-devices, set-only-at-init, read-only module
1878 * parameteters like 'max_loop' and 'max_part' make things needlessly
1879 * complicated, are too static, inflexible and may surprise
1880 * userspace tools. Parameters like this in general should be avoided.
1881 */
1882 if (!part_shift)
1883 disk->flags |= GENHD_FL_NO_PART_SCAN;
1884 disk->flags |= GENHD_FL_EXT_DEVT;
73285082 1885 mutex_init(&lo->lo_ctl_mutex);
f8933667 1886 atomic_set(&lo->lo_refcnt, 0);
73285082 1887 lo->lo_number = i;
73285082
KC		 1888 spin_lock_init(&lo->lo_lock);
1889 disk->major = LOOP_MAJOR;
476a4813 1890 disk->first_minor = i << part_shift;
73285082
KC		 1891 disk->fops = &lo_fops;
1892 disk->private_data = lo;
1893 disk->queue = lo->lo_queue;
1894 sprintf(disk->disk_name, "loop%d", i);
34dd82af
KS		 1895 add_disk(disk);
1896 *l = lo;
1897 return lo->lo_number;
73285082
KC		 1898
1899out_free_queue:
1900 blk_cleanup_queue(lo->lo_queue);
b5dd2f60
ML		 1901out_cleanup_tags:
1902 blk_mq_free_tag_set(&lo->tag_set);
3ec981e3
MP		 1903out_free_idr:
1904 idr_remove(&loop_index_idr, i);
73285082
KC		 1905out_free_dev:
1906 kfree(lo);
1907out:
34dd82af 1908 return err;
73285082
KC		 1909}
1910
34dd82af 1911static void loop_remove(struct loop_device *lo)
1da177e4 1912{
6cd18e71 1913 del_gendisk(lo->lo_disk);
0fa8ebdd 1914 blk_cleanup_queue(lo->lo_queue);
b5dd2f60 1915 blk_mq_free_tag_set(&lo->tag_set);
73285082 1916 put_disk(lo->lo_disk);
73285082
KC		 1917 kfree(lo);
1918}
1da177e4 1919
770fe30a
KS		 1920static int find_free_cb(int id, void *ptr, void *data)
1921{
1922 struct loop_device *lo = ptr;
1923 struct loop_device **l = data;
1924
1925 if (lo->lo_state == Lo_unbound) {
1926 *l = lo;
1927 return 1;
1928 }
1929 return 0;
1930}
1931
34dd82af 1932static int loop_lookup(struct loop_device **l, int i)
a47653fc
KC		 1933{
1934 struct loop_device *lo;
34dd82af 1935 int ret = -ENODEV;
a47653fc 1936
770fe30a
KS		 1937 if (i < 0) {
1938 int err;
1939
1940 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1941 if (err == 1) {
1942 *l = lo;
1943 ret = lo->lo_number;
1944 }
1945 goto out;
a47653fc
KC		 1946 }
1947
770fe30a 1948 /* lookup and return a specific i */
34dd82af 1949 lo = idr_find(&loop_index_idr, i);
a47653fc 1950 if (lo) {
34dd82af
KS		 1951 *l = lo;
1952 ret = lo->lo_number;
a47653fc 1953 }
770fe30a 1954out:
34dd82af 1955 return ret;
a47653fc
KC		 1956}
1957
73285082
KC		 1958static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1959{
705962cc 1960 struct loop_device *lo;
07002e99 1961 struct kobject *kobj;
34dd82af 1962 int err;
73285082 1963
34dd82af
KS		 1964 mutex_lock(&loop_index_mutex);
1965 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1966 if (err < 0)
1967 err = loop_add(&lo, MINOR(dev) >> part_shift);
1968 if (err < 0)
a207f593 1969 kobj = NULL;
34dd82af 1970 else
3079c22e 1971 kobj = get_disk_and_module(lo->lo_disk);
34dd82af 1972 mutex_unlock(&loop_index_mutex);
73285082
KC		 1973
1974 *part = 0;
07002e99 1975 return kobj;
73285082
KC		 1976}
1977
770fe30a
KS		 1978static long loop_control_ioctl(struct file *file, unsigned int cmd,
1979 unsigned long parm)
1980{
1981 struct loop_device *lo;
1982 int ret = -ENOSYS;
1983
1984 mutex_lock(&loop_index_mutex);
1985 switch (cmd) {
1986 case LOOP_CTL_ADD:
1987 ret = loop_lookup(&lo, parm);
1988 if (ret >= 0) {
1989 ret = -EEXIST;
1990 break;
1991 }
1992 ret = loop_add(&lo, parm);
1993 break;
1994 case LOOP_CTL_REMOVE:
1995 ret = loop_lookup(&lo, parm);
1996 if (ret < 0)
1997 break;
3148ffbd
OS		 1998 ret = mutex_lock_killable(&lo->lo_ctl_mutex);
1999 if (ret)
2000 break;
770fe30a
KS		 2001 if (lo->lo_state != Lo_unbound) {
2002 ret = -EBUSY;
2003 mutex_unlock(&lo->lo_ctl_mutex);
2004 break;
2005 }
f8933667 2006 if (atomic_read(&lo->lo_refcnt) > 0) {
770fe30a
KS		 2007 ret = -EBUSY;
2008 mutex_unlock(&lo->lo_ctl_mutex);
2009 break;
2010 }
2011 lo->lo_disk->private_data = NULL;
2012 mutex_unlock(&lo->lo_ctl_mutex);
2013 idr_remove(&loop_index_idr, lo->lo_number);
2014 loop_remove(lo);
2015 break;
2016 case LOOP_CTL_GET_FREE:
2017 ret = loop_lookup(&lo, -1);
2018 if (ret >= 0)
2019 break;
2020 ret = loop_add(&lo, -1);
2021 }
2022 mutex_unlock(&loop_index_mutex);
2023
2024 return ret;
2025}
2026
2027static const struct file_operations loop_ctl_fops = {
2028 .open = nonseekable_open,
2029 .unlocked_ioctl = loop_control_ioctl,
2030 .compat_ioctl = loop_control_ioctl,
2031 .owner = THIS_MODULE,
2032 .llseek = noop_llseek,
2033};
2034
2035static struct miscdevice loop_misc = {
2036 .minor = LOOP_CTRL_MINOR,
2037 .name = "loop-control",
2038 .fops = &loop_ctl_fops,
2039};
2040
2041MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2042MODULE_ALIAS("devname:loop-control");
2043
73285082
KC		 2044static int __init loop_init(void)
2045{
a47653fc
KC		 2046 int i, nr;
2047 unsigned long range;
34dd82af 2048 struct loop_device *lo;
770fe30a 2049 int err;
a47653fc 2050
476a4813 2051 part_shift = 0;
ac04fee0 2052 if (max_part > 0) {
476a4813
LV		 2053 part_shift = fls(max_part);
2054
ac04fee0
NK		 2055 /*
2056 * Adjust max_part according to part_shift as it is exported
2057 * to user space so that user can decide correct minor number
2058 * if [s]he want to create more devices.
2059 *
2060 * Note that -1 is required because partition 0 is reserved
2061 * for the whole disk.
2062 */
2063 max_part = (1UL << part_shift) - 1;
2064 }
2065
b1a66504
GC		 2066 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2067 err = -EINVAL;
a8c1d064 2068 goto err_out;
b1a66504 2069 }
78f4bb36 2070
b1a66504
GC		 2071 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2072 err = -EINVAL;
a8c1d064 2073 goto err_out;
b1a66504 2074 }
1da177e4 2075
d134b00b
KS		 2076 /*
2077 * If max_loop is specified, create that many devices upfront.
2078 * This also becomes a hard limit. If max_loop is not specified,
2079 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2080 * init time. Loop devices can be requested on-demand with the
2081 * /dev/loop-control interface, or be instantiated by accessing
2082 * a 'dead' device node.
2083 */
73285082 2084 if (max_loop) {
a47653fc 2085 nr = max_loop;
a1c15c59 2086 range = max_loop << part_shift;
a47653fc 2087 } else {
d134b00b 2088 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
a1c15c59 2089 range = 1UL << MINORBITS;
a47653fc
KC		 2090 }
2091
a8c1d064
AV		 2092 err = misc_register(&loop_misc);
2093 if (err < 0)
2094 goto err_out;
2095
2096
b1a66504
GC		 2097 if (register_blkdev(LOOP_MAJOR, "loop")) {
2098 err = -EIO;
2099 goto misc_out;
2100 }
1da177e4 2101
a47653fc
KC		 2102 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2103 THIS_MODULE, loop_probe, NULL, NULL);
2104
d134b00b 2105 /* pre-create number of devices given by config or max_loop */
34dd82af
KS		 2106 mutex_lock(&loop_index_mutex);
2107 for (i = 0; i < nr; i++)
2108 loop_add(&lo, i);
2109 mutex_unlock(&loop_index_mutex);
2110
73285082 2111 printk(KERN_INFO "loop: module loaded\n");
1da177e4 2112 return 0;
b1a66504
GC		 2113
2114misc_out:
2115 misc_deregister(&loop_misc);
a8c1d064 2116err_out:
b1a66504 2117 return err;
34dd82af 2118}
a47653fc 2119
34dd82af
KS		 2120static int loop_exit_cb(int id, void *ptr, void *data)
2121{
2122 struct loop_device *lo = ptr;
a47653fc 2123
34dd82af
KS		 2124 loop_remove(lo);
2125 return 0;
1da177e4
LT		 2126}
2127
73285082 2128static void __exit loop_exit(void)
1da177e4 2129{
a47653fc 2130 unsigned long range;
1da177e4 2131
a1c15c59 2132 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
a47653fc 2133
34dd82af 2134 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
34dd82af 2135 idr_destroy(&loop_index_idr);
73285082 2136
a47653fc 2137 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
00d59405 2138 unregister_blkdev(LOOP_MAJOR, "loop");
770fe30a
KS		 2139
2140 misc_deregister(&loop_misc);
1da177e4
LT		 2141}
2142
2143module_init(loop_init);
2144module_exit(loop_exit);
2145
2146#ifndef MODULE
2147static int __init max_loop_setup(char *str)
2148{
2149 max_loop = simple_strtol(str, NULL, 0);
2150 return 1;
2151}
2152
2153__setup("max_loop=", max_loop_setup);
2154#endif
Linux 6.x block layer and io_uring tree(s)