]>
Commit | Line | Data |
---|---|---|
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 |
83 | static DEFINE_IDR(loop_index_idr); |
84 | static DEFINE_MUTEX(loop_index_mutex); | |
1da177e4 | 85 | |
476a4813 LV |
86 | static int max_part; |
87 | static int part_shift; | |
88 | ||
1da177e4 LT |
89 | static 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 | ||
118 | static 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 | ||
125 | static struct loop_func_table none_funcs = { | |
126 | .number = LO_CRYPT_NONE, | |
aa4d8616 | 127 | }; |
1da177e4 LT |
128 | |
129 | static 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 */ | |
136 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
137 | &none_funcs, | |
138 | &xor_funcs | |
139 | }; | |
140 | ||
7035b5df | 141 | static 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 |
162 | static 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 |
167 | static 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; | |
216 | if (use_dio) | |
217 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; | |
218 | else | |
219 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; | |
220 | blk_mq_unfreeze_queue(lo->lo_queue); | |
221 | } | |
222 | ||
1da177e4 | 223 | static int |
1e6ec9ea | 224 | figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) |
1da177e4 | 225 | { |
7035b5df | 226 | loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); |
1da177e4 | 227 | sector_t x = (sector_t)size; |
7b0576a3 | 228 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
229 | |
230 | if (unlikely((loff_t)x != size)) | |
231 | return -EFBIG; | |
7035b5df DM |
232 | if (lo->lo_offset != offset) |
233 | lo->lo_offset = offset; | |
234 | if (lo->lo_sizelimit != sizelimit) | |
235 | lo->lo_sizelimit = sizelimit; | |
73285082 | 236 | set_capacity(lo->lo_disk, x); |
7b0576a3 GC |
237 | bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); |
238 | /* let user-space know about the new size */ | |
239 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
7035b5df | 240 | return 0; |
1da177e4 LT |
241 | } |
242 | ||
243 | static inline int | |
244 | lo_do_transfer(struct loop_device *lo, int cmd, | |
245 | struct page *rpage, unsigned roffs, | |
246 | struct page *lpage, unsigned loffs, | |
247 | int size, sector_t rblock) | |
248 | { | |
aa4d8616 CH |
249 | int ret; |
250 | ||
251 | ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
252 | if (likely(!ret)) | |
1da177e4 LT |
253 | return 0; |
254 | ||
aa4d8616 CH |
255 | printk_ratelimited(KERN_ERR |
256 | "loop: Transfer error at byte offset %llu, length %i.\n", | |
257 | (unsigned long long)rblock << 9, size); | |
258 | return ret; | |
1da177e4 LT |
259 | } |
260 | ||
aa4d8616 | 261 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 262 | { |
aa4d8616 | 263 | struct iov_iter i; |
1da177e4 | 264 | ssize_t bw; |
283e7e5d | 265 | |
aa4d8616 | 266 | iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); |
1da177e4 | 267 | |
03d95eb2 | 268 | file_start_write(file); |
abbb6589 | 269 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 270 | file_end_write(file); |
aa4d8616 CH |
271 | |
272 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 273 | return 0; |
aa4d8616 CH |
274 | |
275 | printk_ratelimited(KERN_ERR | |
276 | "loop: Write error at byte offset %llu, length %i.\n", | |
277 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
278 | if (bw >= 0) |
279 | bw = -EIO; | |
280 | return bw; | |
281 | } | |
282 | ||
aa4d8616 CH |
283 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
284 | loff_t pos) | |
1da177e4 | 285 | { |
aa4d8616 CH |
286 | struct bio_vec bvec; |
287 | struct req_iterator iter; | |
288 | int ret = 0; | |
289 | ||
290 | rq_for_each_segment(bvec, rq, iter) { | |
291 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
292 | if (ret < 0) | |
293 | break; | |
294 | cond_resched(); | |
295 | } | |
296 | ||
297 | return ret; | |
1da177e4 LT |
298 | } |
299 | ||
aa4d8616 | 300 | /* |
456be148 CH |
301 | * This is the slow, transforming version that needs to double buffer the |
302 | * data as it cannot do the transformations in place without having direct | |
303 | * access to the destination pages of the backing file. | |
1da177e4 | 304 | */ |
aa4d8616 CH |
305 | static int lo_write_transfer(struct loop_device *lo, struct request *rq, |
306 | loff_t pos) | |
1da177e4 | 307 | { |
aa4d8616 | 308 | struct bio_vec bvec, b; |
30112013 | 309 | struct req_iterator iter; |
aa4d8616 | 310 | struct page *page; |
7988613b | 311 | int ret = 0; |
1da177e4 | 312 | |
aa4d8616 CH |
313 | page = alloc_page(GFP_NOIO); |
314 | if (unlikely(!page)) | |
315 | return -ENOMEM; | |
456be148 | 316 | |
30112013 | 317 | rq_for_each_segment(bvec, rq, iter) { |
aa4d8616 CH |
318 | ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, |
319 | bvec.bv_offset, bvec.bv_len, pos >> 9); | |
320 | if (unlikely(ret)) | |
321 | break; | |
322 | ||
323 | b.bv_page = page; | |
324 | b.bv_offset = 0; | |
325 | b.bv_len = bvec.bv_len; | |
326 | ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); | |
1da177e4 LT |
327 | if (ret < 0) |
328 | break; | |
1da177e4 | 329 | } |
aa4d8616 CH |
330 | |
331 | __free_page(page); | |
1da177e4 | 332 | return ret; |
1da177e4 LT |
333 | } |
334 | ||
aa4d8616 CH |
335 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
336 | loff_t pos) | |
1da177e4 | 337 | { |
aa4d8616 CH |
338 | struct bio_vec bvec; |
339 | struct req_iterator iter; | |
340 | struct iov_iter i; | |
341 | ssize_t len; | |
1da177e4 | 342 | |
aa4d8616 CH |
343 | rq_for_each_segment(bvec, rq, iter) { |
344 | iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); | |
18e9710e | 345 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
346 | if (len < 0) |
347 | return len; | |
1da177e4 | 348 | |
aa4d8616 | 349 | flush_dcache_page(bvec.bv_page); |
fd582140 | 350 | |
aa4d8616 CH |
351 | if (len != bvec.bv_len) { |
352 | struct bio *bio; | |
1da177e4 | 353 | |
aa4d8616 CH |
354 | __rq_for_each_bio(bio, rq) |
355 | zero_fill_bio(bio); | |
356 | break; | |
357 | } | |
358 | cond_resched(); | |
359 | } | |
360 | ||
361 | return 0; | |
fd582140 JA |
362 | } |
363 | ||
aa4d8616 CH |
364 | static int lo_read_transfer(struct loop_device *lo, struct request *rq, |
365 | loff_t pos) | |
1da177e4 | 366 | { |
aa4d8616 CH |
367 | struct bio_vec bvec, b; |
368 | struct req_iterator iter; | |
369 | struct iov_iter i; | |
370 | struct page *page; | |
371 | ssize_t len; | |
372 | int ret = 0; | |
1da177e4 | 373 | |
aa4d8616 CH |
374 | page = alloc_page(GFP_NOIO); |
375 | if (unlikely(!page)) | |
376 | return -ENOMEM; | |
fd582140 | 377 | |
aa4d8616 CH |
378 | rq_for_each_segment(bvec, rq, iter) { |
379 | loff_t offset = pos; | |
fd582140 | 380 | |
aa4d8616 CH |
381 | b.bv_page = page; |
382 | b.bv_offset = 0; | |
383 | b.bv_len = bvec.bv_len; | |
fd582140 | 384 | |
aa4d8616 | 385 | iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); |
18e9710e | 386 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
387 | if (len < 0) { |
388 | ret = len; | |
389 | goto out_free_page; | |
390 | } | |
1da177e4 | 391 | |
aa4d8616 CH |
392 | ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, |
393 | bvec.bv_offset, len, offset >> 9); | |
394 | if (ret) | |
395 | goto out_free_page; | |
1da177e4 | 396 | |
aa4d8616 | 397 | flush_dcache_page(bvec.bv_page); |
306df071 | 398 | |
aa4d8616 | 399 | if (len != bvec.bv_len) { |
30112013 ML |
400 | struct bio *bio; |
401 | ||
402 | __rq_for_each_bio(bio, rq) | |
403 | zero_fill_bio(bio); | |
1da177e4 | 404 | break; |
306df071 | 405 | } |
1da177e4 | 406 | } |
aa4d8616 CH |
407 | |
408 | ret = 0; | |
409 | out_free_page: | |
410 | __free_page(page); | |
411 | return ret; | |
1da177e4 LT |
412 | } |
413 | ||
cf655d95 ML |
414 | static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) |
415 | { | |
416 | /* | |
417 | * We use punch hole to reclaim the free space used by the | |
418 | * image a.k.a. discard. However we do not support discard if | |
419 | * encryption is enabled, because it may give an attacker | |
420 | * useful information. | |
421 | */ | |
422 | struct file *file = lo->lo_backing_file; | |
423 | int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; | |
424 | int ret; | |
425 | ||
426 | if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { | |
427 | ret = -EOPNOTSUPP; | |
428 | goto out; | |
429 | } | |
430 | ||
431 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
432 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
433 | ret = -EIO; | |
434 | out: | |
435 | return ret; | |
436 | } | |
437 | ||
438 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
439 | { | |
440 | struct file *file = lo->lo_backing_file; | |
441 | int ret = vfs_fsync(file, 0); | |
442 | if (unlikely(ret && ret != -EINVAL)) | |
443 | ret = -EIO; | |
444 | ||
445 | return ret; | |
446 | } | |
447 | ||
fe2cb290 | 448 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 449 | { |
fe2cb290 | 450 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
bc07c10a | 451 | |
fe2cb290 CH |
452 | if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio && |
453 | cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) { | |
bc07c10a ML |
454 | struct bio *bio = cmd->rq->bio; |
455 | ||
fe2cb290 | 456 | bio_advance(bio, cmd->ret); |
bc07c10a ML |
457 | zero_fill_bio(bio); |
458 | } | |
fe2cb290 | 459 | |
2a842aca | 460 | blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK); |
bc07c10a ML |
461 | } |
462 | ||
463 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) | |
464 | { | |
465 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 466 | |
fe2cb290 | 467 | cmd->ret = ret; |
08e0029a | 468 | blk_mq_complete_request(cmd->rq); |
bc07c10a ML |
469 | } |
470 | ||
471 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
472 | loff_t pos, bool rw) | |
473 | { | |
474 | struct iov_iter iter; | |
475 | struct bio_vec *bvec; | |
476 | struct bio *bio = cmd->rq->bio; | |
477 | struct file *file = lo->lo_backing_file; | |
478 | int ret; | |
479 | ||
480 | /* nomerge for loop request queue */ | |
481 | WARN_ON(cmd->rq->bio != cmd->rq->biotail); | |
482 | ||
483 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
484 | iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, | |
485 | bio_segments(bio), blk_rq_bytes(cmd->rq)); | |
a7297a6a ML |
486 | /* |
487 | * This bio may be started from the middle of the 'bvec' | |
488 | * because of bio splitting, so offset from the bvec must | |
489 | * be passed to iov iterator | |
490 | */ | |
491 | iter.iov_offset = bio->bi_iter.bi_bvec_done; | |
bc07c10a ML |
492 | |
493 | cmd->iocb.ki_pos = pos; | |
494 | cmd->iocb.ki_filp = file; | |
495 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
496 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
497 | ||
498 | if (rw == WRITE) | |
bb7462b6 | 499 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 500 | else |
bb7462b6 | 501 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a ML |
502 | |
503 | if (ret != -EIOCBQUEUED) | |
504 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
505 | return 0; | |
506 | } | |
507 | ||
c1c87c2b | 508 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
509 | { |
510 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 511 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
512 | |
513 | /* | |
514 | * lo_write_simple and lo_read_simple should have been covered | |
515 | * by io submit style function like lo_rw_aio(), one blocker | |
516 | * is that lo_read_simple() need to call flush_dcache_page after | |
517 | * the page is written from kernel, and it isn't easy to handle | |
518 | * this in io submit style function which submits all segments | |
519 | * of the req at one time. And direct read IO doesn't need to | |
520 | * run flush_dcache_page(). | |
521 | */ | |
c1c87c2b CH |
522 | switch (req_op(rq)) { |
523 | case REQ_OP_FLUSH: | |
524 | return lo_req_flush(lo, rq); | |
525 | case REQ_OP_DISCARD: | |
19372e27 | 526 | case REQ_OP_WRITE_ZEROES: |
c1c87c2b CH |
527 | return lo_discard(lo, rq, pos); |
528 | case REQ_OP_WRITE: | |
529 | if (lo->transfer) | |
530 | return lo_write_transfer(lo, rq, pos); | |
531 | else if (cmd->use_aio) | |
532 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 533 | else |
c1c87c2b CH |
534 | return lo_write_simple(lo, rq, pos); |
535 | case REQ_OP_READ: | |
aa4d8616 | 536 | if (lo->transfer) |
c1c87c2b CH |
537 | return lo_read_transfer(lo, rq, pos); |
538 | else if (cmd->use_aio) | |
539 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 540 | else |
c1c87c2b CH |
541 | return lo_read_simple(lo, rq, pos); |
542 | default: | |
543 | WARN_ON_ONCE(1); | |
544 | return -EIO; | |
545 | break; | |
aa4d8616 | 546 | } |
1da177e4 LT |
547 | } |
548 | ||
1da177e4 | 549 | struct switch_request { |
b6450630 | 550 | struct file *file, *virt_file; |
1da177e4 LT |
551 | struct completion wait; |
552 | }; | |
553 | ||
2e5ab5f3 ML |
554 | static inline void loop_update_dio(struct loop_device *lo) |
555 | { | |
556 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
557 | lo->use_dio); | |
558 | } | |
559 | ||
1da177e4 | 560 | /* |
b5dd2f60 | 561 | * Do the actual switch; called from the BIO completion routine |
1da177e4 | 562 | */ |
b5dd2f60 | 563 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) |
1da177e4 | 564 | { |
b5dd2f60 ML |
565 | struct file *file = p->file; |
566 | struct file *old_file = lo->lo_backing_file; | |
567 | struct address_space *mapping; | |
35a82d1a | 568 | |
b5dd2f60 ML |
569 | /* if no new file, only flush of queued bios requested */ |
570 | if (!file) | |
571 | return; | |
1da177e4 | 572 | |
b5dd2f60 ML |
573 | mapping = file->f_mapping; |
574 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
575 | lo->lo_backing_file = file; | |
b6450630 | 576 | lo->lo_backing_virt_file = p->virt_file; |
b5dd2f60 ML |
577 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
578 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
579 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
580 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
2e5ab5f3 | 581 | loop_update_dio(lo); |
1da177e4 LT |
582 | } |
583 | ||
584 | /* | |
585 | * loop_switch performs the hard work of switching a backing store. | |
586 | * First it needs to flush existing IO, it does this by sending a magic | |
587 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
588 | */ | |
b6450630 SF |
589 | static int loop_switch(struct loop_device *lo, struct file *file, |
590 | struct file *virt_file) | |
1da177e4 LT |
591 | { |
592 | struct switch_request w; | |
b5dd2f60 | 593 | |
1da177e4 | 594 | w.file = file; |
b6450630 | 595 | w.virt_file = virt_file; |
b5dd2f60 ML |
596 | |
597 | /* freeze queue and wait for completion of scheduled requests */ | |
598 | blk_mq_freeze_queue(lo->lo_queue); | |
599 | ||
600 | /* do the switch action */ | |
601 | do_loop_switch(lo, &w); | |
602 | ||
603 | /* unfreeze */ | |
604 | blk_mq_unfreeze_queue(lo->lo_queue); | |
605 | ||
1da177e4 LT |
606 | return 0; |
607 | } | |
608 | ||
14f27939 MB |
609 | /* |
610 | * Helper to flush the IOs in loop, but keeping loop thread running | |
611 | */ | |
612 | static int loop_flush(struct loop_device *lo) | |
613 | { | |
64604957 JW |
614 | /* loop not yet configured, no running thread, nothing to flush */ |
615 | if (lo->lo_state != Lo_bound) | |
616 | return 0; | |
b6450630 SF |
617 | return loop_switch(lo, NULL, NULL); |
618 | } | |
619 | ||
620 | static struct file *loop_real_file(struct file *file) | |
621 | { | |
622 | struct file *f = NULL; | |
623 | ||
624 | if (file->f_path.dentry->d_sb->s_op->real_loop) | |
625 | f = file->f_path.dentry->d_sb->s_op->real_loop(file); | |
626 | return f; | |
18507e1a SF |
627 | } |
628 | ||
06f0e9e6 ML |
629 | static void loop_reread_partitions(struct loop_device *lo, |
630 | struct block_device *bdev) | |
631 | { | |
632 | int rc; | |
633 | ||
634 | /* | |
635 | * bd_mutex has been held already in release path, so don't | |
636 | * acquire it if this function is called in such case. | |
637 | * | |
638 | * If the reread partition isn't from release path, lo_refcnt | |
639 | * must be at least one and it can only become zero when the | |
640 | * current holder is released. | |
641 | */ | |
642 | if (!atomic_read(&lo->lo_refcnt)) | |
643 | rc = __blkdev_reread_part(bdev); | |
644 | else | |
645 | rc = blkdev_reread_part(bdev); | |
646 | if (rc) | |
647 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
648 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
649 | } | |
650 | ||
1da177e4 LT |
651 | /* |
652 | * loop_change_fd switched the backing store of a loopback device to | |
653 | * a new file. This is useful for operating system installers to free up | |
654 | * the original file and in High Availability environments to switch to | |
655 | * an alternative location for the content in case of server meltdown. | |
656 | * This can only work if the loop device is used read-only, and if the | |
657 | * new backing store is the same size and type as the old backing store. | |
658 | */ | |
bb214884 AV |
659 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
660 | unsigned int arg) | |
1da177e4 LT |
661 | { |
662 | struct file *file, *old_file; | |
b6450630 | 663 | struct file *f, *virt_file = NULL, *old_virt_file; |
1da177e4 LT |
664 | struct inode *inode; |
665 | int error; | |
666 | ||
667 | error = -ENXIO; | |
668 | if (lo->lo_state != Lo_bound) | |
669 | goto out; | |
670 | ||
671 | /* the loop device has to be read-only */ | |
672 | error = -EINVAL; | |
673 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
674 | goto out; | |
675 | ||
676 | error = -EBADF; | |
677 | file = fget(arg); | |
678 | if (!file) | |
679 | goto out; | |
b6450630 SF |
680 | f = loop_real_file(file); |
681 | if (f) { | |
682 | virt_file = file; | |
683 | file = f; | |
684 | get_file(file); | |
685 | } | |
1da177e4 LT |
686 | |
687 | inode = file->f_mapping->host; | |
688 | old_file = lo->lo_backing_file; | |
b6450630 | 689 | old_virt_file = lo->lo_backing_virt_file; |
1da177e4 LT |
690 | |
691 | error = -EINVAL; | |
692 | ||
693 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
694 | goto out_putf; | |
695 | ||
1da177e4 LT |
696 | /* size of the new backing store needs to be the same */ |
697 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
698 | goto out_putf; | |
699 | ||
700 | /* and ... switch */ | |
b6450630 | 701 | error = loop_switch(lo, file, virt_file); |
1da177e4 LT |
702 | if (error) |
703 | goto out_putf; | |
704 | ||
705 | fput(old_file); | |
b6450630 SF |
706 | if (old_virt_file) |
707 | fput(old_virt_file); | |
e03c8dd1 | 708 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
06f0e9e6 | 709 | loop_reread_partitions(lo, bdev); |
1da177e4 LT |
710 | return 0; |
711 | ||
712 | out_putf: | |
713 | fput(file); | |
b6450630 SF |
714 | if (virt_file) |
715 | fput(virt_file); | |
1da177e4 LT |
716 | out: |
717 | return error; | |
718 | } | |
719 | ||
720 | static inline int is_loop_device(struct file *file) | |
721 | { | |
722 | struct inode *i = file->f_mapping->host; | |
723 | ||
724 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
725 | } | |
726 | ||
b6450630 SF |
727 | /* |
728 | * for AUFS | |
729 | * no get/put for file. | |
730 | */ | |
731 | struct file *loop_backing_file(struct super_block *sb) | |
732 | { | |
733 | struct file *ret; | |
734 | struct loop_device *l; | |
735 | ||
736 | ret = NULL; | |
737 | if (MAJOR(sb->s_dev) == LOOP_MAJOR) { | |
738 | l = sb->s_bdev->bd_disk->private_data; | |
739 | ret = l->lo_backing_file; | |
740 | } | |
741 | return ret; | |
742 | } | |
743 | EXPORT_SYMBOL_GPL(loop_backing_file); | |
744 | ||
ee862730 MB |
745 | /* loop sysfs attributes */ |
746 | ||
747 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
748 | ssize_t (*callback)(struct loop_device *, char *)) | |
749 | { | |
34dd82af KS |
750 | struct gendisk *disk = dev_to_disk(dev); |
751 | struct loop_device *lo = disk->private_data; | |
ee862730 | 752 | |
34dd82af | 753 | return callback(lo, page); |
ee862730 MB |
754 | } |
755 | ||
756 | #define LOOP_ATTR_RO(_name) \ | |
757 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
758 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
759 | struct device_attribute *attr, char *b) \ | |
760 | { \ | |
761 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
762 | } \ | |
763 | static struct device_attribute loop_attr_##_name = \ | |
764 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
765 | ||
766 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
767 | { | |
768 | ssize_t ret; | |
769 | char *p = NULL; | |
770 | ||
05eb0f25 | 771 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 772 | if (lo->lo_backing_file) |
9bf39ab2 | 773 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 774 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
775 | |
776 | if (IS_ERR_OR_NULL(p)) | |
777 | ret = PTR_ERR(p); | |
778 | else { | |
779 | ret = strlen(p); | |
780 | memmove(buf, p, ret); | |
781 | buf[ret++] = '\n'; | |
782 | buf[ret] = 0; | |
783 | } | |
784 | ||
785 | return ret; | |
786 | } | |
787 | ||
788 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
789 | { | |
790 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
791 | } | |
792 | ||
793 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
794 | { | |
795 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
796 | } | |
797 | ||
798 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
799 | { | |
800 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
801 | ||
802 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
803 | } | |
804 | ||
e03c8dd1 KS |
805 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
806 | { | |
807 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
808 | ||
809 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
810 | } | |
811 | ||
2e5ab5f3 ML |
812 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
813 | { | |
814 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
815 | ||
816 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
817 | } | |
818 | ||
ee862730 MB |
819 | LOOP_ATTR_RO(backing_file); |
820 | LOOP_ATTR_RO(offset); | |
821 | LOOP_ATTR_RO(sizelimit); | |
822 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 823 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 824 | LOOP_ATTR_RO(dio); |
ee862730 MB |
825 | |
826 | static struct attribute *loop_attrs[] = { | |
827 | &loop_attr_backing_file.attr, | |
828 | &loop_attr_offset.attr, | |
829 | &loop_attr_sizelimit.attr, | |
830 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 831 | &loop_attr_partscan.attr, |
2e5ab5f3 | 832 | &loop_attr_dio.attr, |
ee862730 MB |
833 | NULL, |
834 | }; | |
835 | ||
836 | static struct attribute_group loop_attribute_group = { | |
837 | .name = "loop", | |
838 | .attrs= loop_attrs, | |
839 | }; | |
840 | ||
841 | static int loop_sysfs_init(struct loop_device *lo) | |
842 | { | |
843 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
844 | &loop_attribute_group); | |
845 | } | |
846 | ||
847 | static void loop_sysfs_exit(struct loop_device *lo) | |
848 | { | |
849 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
850 | &loop_attribute_group); | |
851 | } | |
852 | ||
dfaa2ef6 LC |
853 | static void loop_config_discard(struct loop_device *lo) |
854 | { | |
855 | struct file *file = lo->lo_backing_file; | |
856 | struct inode *inode = file->f_mapping->host; | |
857 | struct request_queue *q = lo->lo_queue; | |
858 | ||
859 | /* | |
860 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 861 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
862 | * encryption is enabled, because it may give an attacker |
863 | * useful information. | |
864 | */ | |
865 | if ((!file->f_op->fallocate) || | |
866 | lo->lo_encrypt_key_size) { | |
867 | q->limits.discard_granularity = 0; | |
868 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 869 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 870 | blk_queue_max_write_zeroes_sectors(q, 0); |
dfaa2ef6 LC |
871 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
872 | return; | |
873 | } | |
874 | ||
875 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 876 | q->limits.discard_alignment = 0; |
f2c6df7d | 877 | |
1e6ec9ea OS |
878 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
879 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
dfaa2ef6 LC |
880 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
881 | } | |
882 | ||
e03a3d7a ML |
883 | static void loop_unprepare_queue(struct loop_device *lo) |
884 | { | |
3989144f | 885 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
886 | kthread_stop(lo->worker_task); |
887 | } | |
888 | ||
b2ee7d46 N |
889 | static int loop_kthread_worker_fn(void *worker_ptr) |
890 | { | |
891 | current->flags |= PF_LESS_THROTTLE; | |
892 | return kthread_worker_fn(worker_ptr); | |
893 | } | |
894 | ||
e03a3d7a ML |
895 | static int loop_prepare_queue(struct loop_device *lo) |
896 | { | |
3989144f | 897 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 898 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
899 | &lo->worker, "loop%d", lo->lo_number); |
900 | if (IS_ERR(lo->worker_task)) | |
901 | return -ENOMEM; | |
902 | set_user_nice(lo->worker_task, MIN_NICE); | |
903 | return 0; | |
904 | } | |
905 | ||
bb214884 | 906 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
907 | struct block_device *bdev, unsigned int arg) |
908 | { | |
b6450630 | 909 | struct file *file, *f, *virt_file = NULL; |
1da177e4 LT |
910 | struct inode *inode; |
911 | struct address_space *mapping; | |
912 | unsigned lo_blocksize; | |
913 | int lo_flags = 0; | |
914 | int error; | |
915 | loff_t size; | |
916 | ||
917 | /* This is safe, since we have a reference from open(). */ | |
918 | __module_get(THIS_MODULE); | |
919 | ||
920 | error = -EBADF; | |
921 | file = fget(arg); | |
922 | if (!file) | |
923 | goto out; | |
b6450630 SF |
924 | f = loop_real_file(file); |
925 | if (f) { | |
926 | virt_file = file; | |
927 | file = f; | |
928 | get_file(file); | |
929 | } | |
1da177e4 LT |
930 | |
931 | error = -EBUSY; | |
932 | if (lo->lo_state != Lo_unbound) | |
933 | goto out_putf; | |
934 | ||
935 | /* Avoid recursion */ | |
936 | f = file; | |
937 | while (is_loop_device(f)) { | |
938 | struct loop_device *l; | |
939 | ||
bb214884 | 940 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
941 | goto out_putf; |
942 | ||
943 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
944 | if (l->lo_state == Lo_unbound) { | |
945 | error = -EINVAL; | |
946 | goto out_putf; | |
947 | } | |
948 | f = l->lo_backing_file; | |
949 | } | |
950 | ||
951 | mapping = file->f_mapping; | |
952 | inode = mapping->host; | |
953 | ||
1da177e4 | 954 | error = -EINVAL; |
456be148 CH |
955 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
956 | goto out_putf; | |
1da177e4 | 957 | |
456be148 | 958 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 959 | !file->f_op->write_iter) |
456be148 | 960 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 961 | |
456be148 CH |
962 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
963 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 | 964 | |
456be148 | 965 | error = -EFBIG; |
1da177e4 | 966 | size = get_loop_size(lo, file); |
456be148 | 967 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 968 | goto out_putf; |
e03a3d7a ML |
969 | error = loop_prepare_queue(lo); |
970 | if (error) | |
f4aa4c7b | 971 | goto out_putf; |
1da177e4 | 972 | |
456be148 | 973 | error = 0; |
1da177e4 LT |
974 | |
975 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
976 | ||
2e5ab5f3 | 977 | lo->use_dio = false; |
1da177e4 LT |
978 | lo->lo_blocksize = lo_blocksize; |
979 | lo->lo_device = bdev; | |
980 | lo->lo_flags = lo_flags; | |
981 | lo->lo_backing_file = file; | |
b6450630 | 982 | lo->lo_backing_virt_file = virt_file; |
aa4d8616 | 983 | lo->transfer = NULL; |
1da177e4 LT |
984 | lo->ioctl = NULL; |
985 | lo->lo_sizelimit = 0; | |
986 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
987 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
988 | ||
68db1961 | 989 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 990 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 991 | |
2e5ab5f3 | 992 | loop_update_dio(lo); |
73285082 | 993 | set_capacity(lo->lo_disk, size); |
1da177e4 | 994 | bd_set_size(bdev, size << 9); |
ee862730 | 995 | loop_sysfs_init(lo); |
c3473c63 DZ |
996 | /* let user-space know about the new size */ |
997 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 LT |
998 | |
999 | set_blocksize(bdev, lo_blocksize); | |
1000 | ||
6c997918 | 1001 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1002 | if (part_shift) |
1003 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1004 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
06f0e9e6 | 1005 | loop_reread_partitions(lo, bdev); |
c1681bf8 AP |
1006 | |
1007 | /* Grab the block_device to prevent its destruction after we | |
1008 | * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). | |
1009 | */ | |
1010 | bdgrab(bdev); | |
1da177e4 LT |
1011 | return 0; |
1012 | ||
1013 | out_putf: | |
1014 | fput(file); | |
b6450630 SF |
1015 | if (virt_file) |
1016 | fput(virt_file); | |
1da177e4 LT |
1017 | out: |
1018 | /* This is safe: open() is still holding a reference. */ | |
1019 | module_put(THIS_MODULE); | |
1020 | return error; | |
1021 | } | |
1022 | ||
1023 | static int | |
1024 | loop_release_xfer(struct loop_device *lo) | |
1025 | { | |
1026 | int err = 0; | |
1027 | struct loop_func_table *xfer = lo->lo_encryption; | |
1028 | ||
1029 | if (xfer) { | |
1030 | if (xfer->release) | |
1031 | err = xfer->release(lo); | |
1032 | lo->transfer = NULL; | |
1033 | lo->lo_encryption = NULL; | |
1034 | module_put(xfer->owner); | |
1035 | } | |
1036 | return err; | |
1037 | } | |
1038 | ||
1039 | static int | |
1040 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1041 | const struct loop_info64 *i) | |
1042 | { | |
1043 | int err = 0; | |
1044 | ||
1045 | if (xfer) { | |
1046 | struct module *owner = xfer->owner; | |
1047 | ||
1048 | if (!try_module_get(owner)) | |
1049 | return -EINVAL; | |
1050 | if (xfer->init) | |
1051 | err = xfer->init(lo, i); | |
1052 | if (err) | |
1053 | module_put(owner); | |
1054 | else | |
1055 | lo->lo_encryption = xfer; | |
1056 | } | |
1057 | return err; | |
1058 | } | |
1059 | ||
4c823cc3 | 1060 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
1061 | { |
1062 | struct file *filp = lo->lo_backing_file; | |
b6450630 | 1063 | struct file *virt_filp = lo->lo_backing_virt_file; |
b4e3ca1a | 1064 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1065 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
1066 | |
1067 | if (lo->lo_state != Lo_bound) | |
1068 | return -ENXIO; | |
1069 | ||
a1ecac3b DC |
1070 | /* |
1071 | * If we've explicitly asked to tear down the loop device, | |
1072 | * and it has an elevated reference count, set it for auto-teardown when | |
1073 | * the last reference goes away. This stops $!~#$@ udev from | |
1074 | * preventing teardown because it decided that it needs to run blkid on | |
1075 | * the loopback device whenever they appear. xfstests is notorious for | |
1076 | * failing tests because blkid via udev races with a losetup | |
1077 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1078 | * command to fail with EBUSY. | |
1079 | */ | |
f8933667 | 1080 | if (atomic_read(&lo->lo_refcnt) > 1) { |
a1ecac3b DC |
1081 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
1082 | mutex_unlock(&lo->lo_ctl_mutex); | |
1083 | return 0; | |
1084 | } | |
1da177e4 LT |
1085 | |
1086 | if (filp == NULL) | |
1087 | return -EINVAL; | |
1088 | ||
f8933667 ML |
1089 | /* freeze request queue during the transition */ |
1090 | blk_mq_freeze_queue(lo->lo_queue); | |
1091 | ||
1da177e4 LT |
1092 | spin_lock_irq(&lo->lo_lock); |
1093 | lo->lo_state = Lo_rundown; | |
1da177e4 | 1094 | lo->lo_backing_file = NULL; |
b6450630 | 1095 | lo->lo_backing_virt_file = NULL; |
05eb0f25 | 1096 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1097 | |
1098 | loop_release_xfer(lo); | |
1099 | lo->transfer = NULL; | |
1100 | lo->ioctl = NULL; | |
1101 | lo->lo_device = NULL; | |
1102 | lo->lo_encryption = NULL; | |
1103 | lo->lo_offset = 0; | |
1104 | lo->lo_sizelimit = 0; | |
1105 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1106 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1107 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1108 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
c1681bf8 AP |
1109 | if (bdev) { |
1110 | bdput(bdev); | |
bb214884 | 1111 | invalidate_bdev(bdev); |
c1681bf8 | 1112 | } |
73285082 | 1113 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1114 | loop_sysfs_exit(lo); |
c3473c63 | 1115 | if (bdev) { |
bb214884 | 1116 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1117 | /* let user-space know about this change */ |
1118 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1119 | } | |
1da177e4 LT |
1120 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1121 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1122 | /* This is safe: open() is still holding a reference. */ |
1123 | module_put(THIS_MODULE); | |
f8933667 ML |
1124 | blk_mq_unfreeze_queue(lo->lo_queue); |
1125 | ||
c2fccc1c | 1126 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
06f0e9e6 | 1127 | loop_reread_partitions(lo, bdev); |
e03c8dd1 KS |
1128 | lo->lo_flags = 0; |
1129 | if (!part_shift) | |
1130 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
e03a3d7a | 1131 | loop_unprepare_queue(lo); |
f028f3b2 NK |
1132 | mutex_unlock(&lo->lo_ctl_mutex); |
1133 | /* | |
1134 | * Need not hold lo_ctl_mutex to fput backing file. | |
1135 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1136 | * lock dependency possibility warning as fput can take | |
1137 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1138 | */ | |
1139 | fput(filp); | |
b6450630 SF |
1140 | if (virt_filp) |
1141 | fput(virt_filp); | |
1da177e4 LT |
1142 | return 0; |
1143 | } | |
1144 | ||
1145 | static int | |
1146 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1147 | { | |
1148 | int err; | |
1149 | struct loop_func_table *xfer; | |
e4849737 | 1150 | kuid_t uid = current_uid(); |
1da177e4 | 1151 | |
b0fafa81 | 1152 | if (lo->lo_encrypt_key_size && |
e4849737 | 1153 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1154 | !capable(CAP_SYS_ADMIN)) |
1155 | return -EPERM; | |
1156 | if (lo->lo_state != Lo_bound) | |
1157 | return -ENXIO; | |
1158 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1159 | return -EINVAL; | |
1160 | ||
ecdd0959 ML |
1161 | /* I/O need to be drained during transfer transition */ |
1162 | blk_mq_freeze_queue(lo->lo_queue); | |
1163 | ||
1da177e4 LT |
1164 | err = loop_release_xfer(lo); |
1165 | if (err) | |
ecdd0959 | 1166 | goto exit; |
1da177e4 LT |
1167 | |
1168 | if (info->lo_encrypt_type) { | |
1169 | unsigned int type = info->lo_encrypt_type; | |
1170 | ||
1171 | if (type >= MAX_LO_CRYPT) | |
1172 | return -EINVAL; | |
1173 | xfer = xfer_funcs[type]; | |
1174 | if (xfer == NULL) | |
1175 | return -EINVAL; | |
1176 | } else | |
1177 | xfer = NULL; | |
1178 | ||
1179 | err = loop_init_xfer(lo, xfer, info); | |
1180 | if (err) | |
ecdd0959 | 1181 | goto exit; |
1da177e4 LT |
1182 | |
1183 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea OS |
1184 | lo->lo_sizelimit != info->lo_sizelimit) { |
1185 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { | |
ecdd0959 ML |
1186 | err = -EFBIG; |
1187 | goto exit; | |
1188 | } | |
b040ad9c | 1189 | } |
541c742a | 1190 | |
dfaa2ef6 | 1191 | loop_config_discard(lo); |
1da177e4 LT |
1192 | |
1193 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1194 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1195 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1196 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1197 | ||
1198 | if (!xfer) | |
1199 | xfer = &none_funcs; | |
1200 | lo->transfer = xfer->transfer; | |
1201 | lo->ioctl = xfer->ioctl; | |
1202 | ||
96c58655 DW |
1203 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1204 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1205 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1206 | ||
1da177e4 LT |
1207 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1208 | lo->lo_init[0] = info->lo_init[0]; | |
1209 | lo->lo_init[1] = info->lo_init[1]; | |
1210 | if (info->lo_encrypt_key_size) { | |
1211 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1212 | info->lo_encrypt_key_size); | |
b0fafa81 | 1213 | lo->lo_key_owner = uid; |
aa4d8616 | 1214 | } |
1da177e4 | 1215 | |
2e5ab5f3 ML |
1216 | /* update dio if lo_offset or transfer is changed */ |
1217 | __loop_update_dio(lo, lo->use_dio); | |
1218 | ||
ecdd0959 ML |
1219 | exit: |
1220 | blk_mq_unfreeze_queue(lo->lo_queue); | |
e02898b4 OS |
1221 | |
1222 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1223 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1224 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1225 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1226 | loop_reread_partitions(lo, lo->lo_device); | |
1227 | } | |
1228 | ||
ecdd0959 | 1229 | return err; |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | static int | |
1233 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1234 | { | |
1235 | struct file *file = lo->lo_backing_file; | |
1236 | struct kstat stat; | |
1237 | int error; | |
1238 | ||
1239 | if (lo->lo_state != Lo_bound) | |
1240 | return -ENXIO; | |
a528d35e DH |
1241 | error = vfs_getattr(&file->f_path, &stat, |
1242 | STATX_INO, AT_STATX_SYNC_AS_STAT); | |
1da177e4 LT |
1243 | if (error) |
1244 | return error; | |
1245 | memset(info, 0, sizeof(*info)); | |
1246 | info->lo_number = lo->lo_number; | |
1247 | info->lo_device = huge_encode_dev(stat.dev); | |
1248 | info->lo_inode = stat.ino; | |
1249 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1250 | info->lo_offset = lo->lo_offset; | |
1251 | info->lo_sizelimit = lo->lo_sizelimit; | |
1252 | info->lo_flags = lo->lo_flags; | |
1253 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1254 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1255 | info->lo_encrypt_type = | |
1256 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1257 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1258 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1259 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1260 | lo->lo_encrypt_key_size); | |
1261 | } | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | static void | |
1266 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1267 | { | |
1268 | memset(info64, 0, sizeof(*info64)); | |
1269 | info64->lo_number = info->lo_number; | |
1270 | info64->lo_device = info->lo_device; | |
1271 | info64->lo_inode = info->lo_inode; | |
1272 | info64->lo_rdevice = info->lo_rdevice; | |
1273 | info64->lo_offset = info->lo_offset; | |
1274 | info64->lo_sizelimit = 0; | |
1275 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1276 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1277 | info64->lo_flags = info->lo_flags; | |
1278 | info64->lo_init[0] = info->lo_init[0]; | |
1279 | info64->lo_init[1] = info->lo_init[1]; | |
1280 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1281 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1282 | else | |
1283 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1284 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1285 | } | |
1286 | ||
1287 | static int | |
1288 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1289 | { | |
1290 | memset(info, 0, sizeof(*info)); | |
1291 | info->lo_number = info64->lo_number; | |
1292 | info->lo_device = info64->lo_device; | |
1293 | info->lo_inode = info64->lo_inode; | |
1294 | info->lo_rdevice = info64->lo_rdevice; | |
1295 | info->lo_offset = info64->lo_offset; | |
1296 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1297 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1298 | info->lo_flags = info64->lo_flags; | |
1299 | info->lo_init[0] = info64->lo_init[0]; | |
1300 | info->lo_init[1] = info64->lo_init[1]; | |
1301 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1302 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1303 | else | |
1304 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1305 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1306 | ||
1307 | /* error in case values were truncated */ | |
1308 | if (info->lo_device != info64->lo_device || | |
1309 | info->lo_rdevice != info64->lo_rdevice || | |
1310 | info->lo_inode != info64->lo_inode || | |
1311 | info->lo_offset != info64->lo_offset) | |
1312 | return -EOVERFLOW; | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | static int | |
1318 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1319 | { | |
1320 | struct loop_info info; | |
1321 | struct loop_info64 info64; | |
1322 | ||
1323 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1324 | return -EFAULT; | |
1325 | loop_info64_from_old(&info, &info64); | |
1326 | return loop_set_status(lo, &info64); | |
1327 | } | |
1328 | ||
1329 | static int | |
1330 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1331 | { | |
1332 | struct loop_info64 info64; | |
1333 | ||
1334 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1335 | return -EFAULT; | |
1336 | return loop_set_status(lo, &info64); | |
1337 | } | |
1338 | ||
1339 | static int | |
1340 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1341 | struct loop_info info; | |
1342 | struct loop_info64 info64; | |
1343 | int err = 0; | |
1344 | ||
1345 | if (!arg) | |
1346 | err = -EINVAL; | |
1347 | if (!err) | |
1348 | err = loop_get_status(lo, &info64); | |
1349 | if (!err) | |
1350 | err = loop_info64_to_old(&info64, &info); | |
1351 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1352 | err = -EFAULT; | |
1353 | ||
1354 | return err; | |
1355 | } | |
1356 | ||
1357 | static int | |
1358 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1359 | struct loop_info64 info64; | |
1360 | int err = 0; | |
1361 | ||
1362 | if (!arg) | |
1363 | err = -EINVAL; | |
1364 | if (!err) | |
1365 | err = loop_get_status(lo, &info64); | |
1366 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1367 | err = -EFAULT; | |
1368 | ||
1369 | return err; | |
1370 | } | |
1371 | ||
51001b7d | 1372 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1373 | { |
53d66608 | 1374 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1375 | return -ENXIO; |
53d66608 | 1376 | |
1e6ec9ea | 1377 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1378 | } |
1379 | ||
ab1cb278 ML |
1380 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1381 | { | |
1382 | int error = -ENXIO; | |
1383 | if (lo->lo_state != Lo_bound) | |
1384 | goto out; | |
1385 | ||
1386 | __loop_update_dio(lo, !!arg); | |
1387 | if (lo->use_dio == !!arg) | |
1388 | return 0; | |
1389 | error = -EINVAL; | |
1390 | out: | |
1391 | return error; | |
1392 | } | |
1393 | ||
bb214884 | 1394 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1395 | unsigned int cmd, unsigned long arg) |
1396 | { | |
bb214884 | 1397 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1398 | int err; |
1399 | ||
f028f3b2 | 1400 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1401 | switch (cmd) { |
1402 | case LOOP_SET_FD: | |
bb214884 | 1403 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1404 | break; |
1405 | case LOOP_CHANGE_FD: | |
bb214884 | 1406 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1407 | break; |
1408 | case LOOP_CLR_FD: | |
f028f3b2 | 1409 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1410 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1411 | if (!err) |
1412 | goto out_unlocked; | |
1da177e4 LT |
1413 | break; |
1414 | case LOOP_SET_STATUS: | |
7035b5df DM |
1415 | err = -EPERM; |
1416 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1417 | err = loop_set_status_old(lo, | |
1418 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1419 | break; |
1420 | case LOOP_GET_STATUS: | |
1421 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1422 | break; | |
1423 | case LOOP_SET_STATUS64: | |
7035b5df DM |
1424 | err = -EPERM; |
1425 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1426 | err = loop_set_status64(lo, | |
1427 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1428 | break; |
1429 | case LOOP_GET_STATUS64: | |
1430 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1431 | break; | |
53d66608 O |
1432 | case LOOP_SET_CAPACITY: |
1433 | err = -EPERM; | |
1434 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
51001b7d | 1435 | err = loop_set_capacity(lo); |
53d66608 | 1436 | break; |
ab1cb278 ML |
1437 | case LOOP_SET_DIRECT_IO: |
1438 | err = -EPERM; | |
1439 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1440 | err = loop_set_dio(lo, arg); | |
1441 | break; | |
1da177e4 LT |
1442 | default: |
1443 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1444 | } | |
f85221dd | 1445 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1446 | |
1447 | out_unlocked: | |
1da177e4 LT |
1448 | return err; |
1449 | } | |
1450 | ||
863d5b82 DH |
1451 | #ifdef CONFIG_COMPAT |
1452 | struct compat_loop_info { | |
1453 | compat_int_t lo_number; /* ioctl r/o */ | |
1454 | compat_dev_t lo_device; /* ioctl r/o */ | |
1455 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1456 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1457 | compat_int_t lo_offset; | |
1458 | compat_int_t lo_encrypt_type; | |
1459 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1460 | compat_int_t lo_flags; /* ioctl r/o */ | |
1461 | char lo_name[LO_NAME_SIZE]; | |
1462 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1463 | compat_ulong_t lo_init[2]; | |
1464 | char reserved[4]; | |
1465 | }; | |
1466 | ||
1467 | /* | |
1468 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1469 | * - noinlined to reduce stack space usage in main part of driver | |
1470 | */ | |
1471 | static noinline int | |
ba674cfc | 1472 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1473 | struct loop_info64 *info64) |
1474 | { | |
1475 | struct compat_loop_info info; | |
1476 | ||
1477 | if (copy_from_user(&info, arg, sizeof(info))) | |
1478 | return -EFAULT; | |
1479 | ||
1480 | memset(info64, 0, sizeof(*info64)); | |
1481 | info64->lo_number = info.lo_number; | |
1482 | info64->lo_device = info.lo_device; | |
1483 | info64->lo_inode = info.lo_inode; | |
1484 | info64->lo_rdevice = info.lo_rdevice; | |
1485 | info64->lo_offset = info.lo_offset; | |
1486 | info64->lo_sizelimit = 0; | |
1487 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1488 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1489 | info64->lo_flags = info.lo_flags; | |
1490 | info64->lo_init[0] = info.lo_init[0]; | |
1491 | info64->lo_init[1] = info.lo_init[1]; | |
1492 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1493 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1494 | else | |
1495 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1496 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1497 | return 0; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1502 | * - noinlined to reduce stack space usage in main part of driver | |
1503 | */ | |
1504 | static noinline int | |
1505 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1506 | struct compat_loop_info __user *arg) | |
1507 | { | |
1508 | struct compat_loop_info info; | |
1509 | ||
1510 | memset(&info, 0, sizeof(info)); | |
1511 | info.lo_number = info64->lo_number; | |
1512 | info.lo_device = info64->lo_device; | |
1513 | info.lo_inode = info64->lo_inode; | |
1514 | info.lo_rdevice = info64->lo_rdevice; | |
1515 | info.lo_offset = info64->lo_offset; | |
1516 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1517 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1518 | info.lo_flags = info64->lo_flags; | |
1519 | info.lo_init[0] = info64->lo_init[0]; | |
1520 | info.lo_init[1] = info64->lo_init[1]; | |
1521 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1522 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1523 | else | |
1524 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1525 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1526 | ||
1527 | /* error in case values were truncated */ | |
1528 | if (info.lo_device != info64->lo_device || | |
1529 | info.lo_rdevice != info64->lo_rdevice || | |
1530 | info.lo_inode != info64->lo_inode || | |
1531 | info.lo_offset != info64->lo_offset || | |
1532 | info.lo_init[0] != info64->lo_init[0] || | |
1533 | info.lo_init[1] != info64->lo_init[1]) | |
1534 | return -EOVERFLOW; | |
1535 | ||
1536 | if (copy_to_user(arg, &info, sizeof(info))) | |
1537 | return -EFAULT; | |
1538 | return 0; | |
1539 | } | |
1540 | ||
1541 | static int | |
1542 | loop_set_status_compat(struct loop_device *lo, | |
1543 | const struct compat_loop_info __user *arg) | |
1544 | { | |
1545 | struct loop_info64 info64; | |
1546 | int ret; | |
1547 | ||
1548 | ret = loop_info64_from_compat(arg, &info64); | |
1549 | if (ret < 0) | |
1550 | return ret; | |
1551 | return loop_set_status(lo, &info64); | |
1552 | } | |
1553 | ||
1554 | static int | |
1555 | loop_get_status_compat(struct loop_device *lo, | |
1556 | struct compat_loop_info __user *arg) | |
1557 | { | |
1558 | struct loop_info64 info64; | |
1559 | int err = 0; | |
1560 | ||
1561 | if (!arg) | |
1562 | err = -EINVAL; | |
1563 | if (!err) | |
1564 | err = loop_get_status(lo, &info64); | |
1565 | if (!err) | |
1566 | err = loop_info64_to_compat(&info64, arg); | |
1567 | return err; | |
1568 | } | |
1569 | ||
bb214884 AV |
1570 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1571 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1572 | { |
bb214884 | 1573 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1574 | int err; |
1575 | ||
863d5b82 DH |
1576 | switch(cmd) { |
1577 | case LOOP_SET_STATUS: | |
1578 | mutex_lock(&lo->lo_ctl_mutex); | |
1579 | err = loop_set_status_compat( | |
1580 | lo, (const struct compat_loop_info __user *) arg); | |
1581 | mutex_unlock(&lo->lo_ctl_mutex); | |
1582 | break; | |
1583 | case LOOP_GET_STATUS: | |
1584 | mutex_lock(&lo->lo_ctl_mutex); | |
1585 | err = loop_get_status_compat( | |
1586 | lo, (struct compat_loop_info __user *) arg); | |
1587 | mutex_unlock(&lo->lo_ctl_mutex); | |
1588 | break; | |
53d66608 | 1589 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1590 | case LOOP_CLR_FD: |
1591 | case LOOP_GET_STATUS64: | |
1592 | case LOOP_SET_STATUS64: | |
1593 | arg = (unsigned long) compat_ptr(arg); | |
1594 | case LOOP_SET_FD: | |
1595 | case LOOP_CHANGE_FD: | |
bb214884 | 1596 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1597 | break; |
1598 | default: | |
1599 | err = -ENOIOCTLCMD; | |
1600 | break; | |
1601 | } | |
863d5b82 DH |
1602 | return err; |
1603 | } | |
1604 | #endif | |
1605 | ||
bb214884 | 1606 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1607 | { |
770fe30a KS |
1608 | struct loop_device *lo; |
1609 | int err = 0; | |
1610 | ||
1611 | mutex_lock(&loop_index_mutex); | |
1612 | lo = bdev->bd_disk->private_data; | |
1613 | if (!lo) { | |
1614 | err = -ENXIO; | |
1615 | goto out; | |
1616 | } | |
1da177e4 | 1617 | |
f8933667 | 1618 | atomic_inc(&lo->lo_refcnt); |
770fe30a KS |
1619 | out: |
1620 | mutex_unlock(&loop_index_mutex); | |
1621 | return err; | |
1da177e4 LT |
1622 | } |
1623 | ||
fbbe56c9 | 1624 | static void __lo_release(struct loop_device *lo) |
1da177e4 | 1625 | { |
ffcd7dca | 1626 | int err; |
1da177e4 | 1627 | |
f8933667 ML |
1628 | if (atomic_dec_return(&lo->lo_refcnt)) |
1629 | return; | |
14f27939 | 1630 | |
f8933667 | 1631 | mutex_lock(&lo->lo_ctl_mutex); |
14f27939 MB |
1632 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { |
1633 | /* | |
1634 | * In autoclear mode, stop the loop thread | |
1635 | * and remove configuration after last close. | |
1636 | */ | |
4c823cc3 | 1637 | err = loop_clr_fd(lo); |
ffcd7dca | 1638 | if (!err) |
db2a144b | 1639 | return; |
14f27939 MB |
1640 | } else { |
1641 | /* | |
1642 | * Otherwise keep thread (if running) and config, | |
1643 | * but flush possible ongoing bios in thread. | |
1644 | */ | |
1645 | loop_flush(lo); | |
1646 | } | |
96c58655 | 1647 | |
f85221dd | 1648 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1649 | } |
1650 | ||
fbbe56c9 LT |
1651 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1652 | { | |
1653 | mutex_lock(&loop_index_mutex); | |
1654 | __lo_release(disk->private_data); | |
1655 | mutex_unlock(&loop_index_mutex); | |
1656 | } | |
1657 | ||
83d5cde4 | 1658 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1659 | .owner = THIS_MODULE, |
bb214884 AV |
1660 | .open = lo_open, |
1661 | .release = lo_release, | |
1662 | .ioctl = lo_ioctl, | |
863d5b82 | 1663 | #ifdef CONFIG_COMPAT |
bb214884 | 1664 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1665 | #endif |
1da177e4 LT |
1666 | }; |
1667 | ||
1668 | /* | |
1669 | * And now the modules code and kernel interface. | |
1670 | */ | |
73285082 | 1671 | static int max_loop; |
ac04fee0 | 1672 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1673 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1674 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1675 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1676 | MODULE_LICENSE("GPL"); |
1677 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1678 | ||
1679 | int loop_register_transfer(struct loop_func_table *funcs) | |
1680 | { | |
1681 | unsigned int n = funcs->number; | |
1682 | ||
1683 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1684 | return -EINVAL; | |
1685 | xfer_funcs[n] = funcs; | |
1686 | return 0; | |
1687 | } | |
1688 | ||
34dd82af KS |
1689 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1690 | { | |
1691 | struct loop_device *lo = ptr; | |
1692 | struct loop_func_table *xfer = data; | |
1693 | ||
1694 | mutex_lock(&lo->lo_ctl_mutex); | |
1695 | if (lo->lo_encryption == xfer) | |
1696 | loop_release_xfer(lo); | |
1697 | mutex_unlock(&lo->lo_ctl_mutex); | |
1698 | return 0; | |
1699 | } | |
1700 | ||
1da177e4 LT |
1701 | int loop_unregister_transfer(int number) |
1702 | { | |
1703 | unsigned int n = number; | |
1da177e4 LT |
1704 | struct loop_func_table *xfer; |
1705 | ||
1706 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1707 | return -EINVAL; | |
1708 | ||
1709 | xfer_funcs[n] = NULL; | |
34dd82af | 1710 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1711 | return 0; |
1712 | } | |
1713 | ||
1714 | EXPORT_SYMBOL(loop_register_transfer); | |
1715 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1716 | ||
fc17b653 | 1717 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1718 | const struct blk_mq_queue_data *bd) |
1719 | { | |
1720 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
f4aa4c7b | 1721 | struct loop_device *lo = cmd->rq->q->queuedata; |
b5dd2f60 ML |
1722 | |
1723 | blk_mq_start_request(bd->rq); | |
1724 | ||
f4aa4c7b | 1725 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1726 | return BLK_STS_IOERR; |
f4aa4c7b | 1727 | |
f0225cac CH |
1728 | switch (req_op(cmd->rq)) { |
1729 | case REQ_OP_FLUSH: | |
1730 | case REQ_OP_DISCARD: | |
19372e27 | 1731 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1732 | cmd->use_aio = false; |
f0225cac CH |
1733 | break; |
1734 | default: | |
1735 | cmd->use_aio = lo->use_dio; | |
1736 | break; | |
1737 | } | |
bc07c10a | 1738 | |
3989144f | 1739 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1740 | |
fc17b653 | 1741 | return BLK_STS_OK; |
b5dd2f60 ML |
1742 | } |
1743 | ||
1744 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1745 | { | |
a8ebb056 | 1746 | const bool write = op_is_write(req_op(cmd->rq)); |
b5dd2f60 | 1747 | struct loop_device *lo = cmd->rq->q->queuedata; |
f4829a9b | 1748 | int ret = 0; |
b5dd2f60 | 1749 | |
f4829a9b CH |
1750 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1751 | ret = -EIO; | |
b5dd2f60 | 1752 | goto failed; |
f4829a9b | 1753 | } |
b5dd2f60 | 1754 | |
30112013 | 1755 | ret = do_req_filebacked(lo, cmd->rq); |
b5dd2f60 | 1756 | failed: |
bc07c10a | 1757 | /* complete non-aio request */ |
fe2cb290 CH |
1758 | if (!cmd->use_aio || ret) { |
1759 | cmd->ret = ret ? -EIO : 0; | |
08e0029a | 1760 | blk_mq_complete_request(cmd->rq); |
fe2cb290 | 1761 | } |
b5dd2f60 ML |
1762 | } |
1763 | ||
e03a3d7a | 1764 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1765 | { |
1766 | struct loop_cmd *cmd = | |
e03a3d7a | 1767 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1768 | |
1769 | loop_handle_cmd(cmd); | |
1770 | } | |
1771 | ||
d6296d39 CH |
1772 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1773 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1774 | { |
1775 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1776 | ||
1777 | cmd->rq = rq; | |
3989144f | 1778 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1779 | |
1780 | return 0; | |
1781 | } | |
1782 | ||
f363b089 | 1783 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1784 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 1785 | .init_request = loop_init_request, |
fe2cb290 | 1786 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1787 | }; |
1788 | ||
34dd82af | 1789 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1790 | { |
1791 | struct loop_device *lo; | |
1792 | struct gendisk *disk; | |
34dd82af | 1793 | int err; |
73285082 | 1794 | |
68d740d7 | 1795 | err = -ENOMEM; |
73285082 | 1796 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1797 | if (!lo) |
73285082 | 1798 | goto out; |
34dd82af | 1799 | |
ef7e7c82 MP |
1800 | lo->lo_state = Lo_unbound; |
1801 | ||
c718aa65 | 1802 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1803 | if (i >= 0) { |
c718aa65 TH |
1804 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1805 | if (err == -ENOSPC) | |
34dd82af | 1806 | err = -EEXIST; |
34dd82af | 1807 | } else { |
c718aa65 | 1808 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1809 | } |
1810 | if (err < 0) | |
1811 | goto out_free_dev; | |
c718aa65 | 1812 | i = err; |
73285082 | 1813 | |
183cfb57 | 1814 | err = -ENOMEM; |
b5dd2f60 ML |
1815 | lo->tag_set.ops = &loop_mq_ops; |
1816 | lo->tag_set.nr_hw_queues = 1; | |
1817 | lo->tag_set.queue_depth = 128; | |
1818 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1819 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
1820 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
1821 | lo->tag_set.driver_data = lo; | |
1822 | ||
1823 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1824 | if (err) | |
3ec981e3 | 1825 | goto out_free_idr; |
73285082 | 1826 | |
b5dd2f60 ML |
1827 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
1828 | if (IS_ERR_OR_NULL(lo->lo_queue)) { | |
1829 | err = PTR_ERR(lo->lo_queue); | |
1830 | goto out_cleanup_tags; | |
1831 | } | |
ef7e7c82 MP |
1832 | lo->lo_queue->queuedata = lo; |
1833 | ||
5b5e20f4 ML |
1834 | /* |
1835 | * It doesn't make sense to enable merge because the I/O | |
1836 | * submitted to backing file is handled page by page. | |
1837 | */ | |
1838 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
1839 | ||
7a649737 | 1840 | err = -ENOMEM; |
476a4813 | 1841 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1842 | if (!disk) |
1843 | goto out_free_queue; | |
1844 | ||
e03c8dd1 KS |
1845 | /* |
1846 | * Disable partition scanning by default. The in-kernel partition | |
1847 | * scanning can be requested individually per-device during its | |
1848 | * setup. Userspace can always add and remove partitions from all | |
1849 | * devices. The needed partition minors are allocated from the | |
1850 | * extended minor space, the main loop device numbers will continue | |
1851 | * to match the loop minors, regardless of the number of partitions | |
1852 | * used. | |
1853 | * | |
1854 | * If max_part is given, partition scanning is globally enabled for | |
1855 | * all loop devices. The minors for the main loop devices will be | |
1856 | * multiples of max_part. | |
1857 | * | |
1858 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1859 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1860 | * complicated, are too static, inflexible and may surprise | |
1861 | * userspace tools. Parameters like this in general should be avoided. | |
1862 | */ | |
1863 | if (!part_shift) | |
1864 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1865 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 | 1866 | mutex_init(&lo->lo_ctl_mutex); |
f8933667 | 1867 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 1868 | lo->lo_number = i; |
73285082 KC |
1869 | spin_lock_init(&lo->lo_lock); |
1870 | disk->major = LOOP_MAJOR; | |
476a4813 | 1871 | disk->first_minor = i << part_shift; |
73285082 KC |
1872 | disk->fops = &lo_fops; |
1873 | disk->private_data = lo; | |
1874 | disk->queue = lo->lo_queue; | |
1875 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1876 | add_disk(disk); |
1877 | *l = lo; | |
1878 | return lo->lo_number; | |
73285082 KC |
1879 | |
1880 | out_free_queue: | |
1881 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
1882 | out_cleanup_tags: |
1883 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
1884 | out_free_idr: |
1885 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
1886 | out_free_dev: |
1887 | kfree(lo); | |
1888 | out: | |
34dd82af | 1889 | return err; |
73285082 KC |
1890 | } |
1891 | ||
34dd82af | 1892 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1893 | { |
73285082 | 1894 | blk_cleanup_queue(lo->lo_queue); |
6cd18e71 | 1895 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 1896 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 1897 | put_disk(lo->lo_disk); |
73285082 KC |
1898 | kfree(lo); |
1899 | } | |
1da177e4 | 1900 | |
770fe30a KS |
1901 | static int find_free_cb(int id, void *ptr, void *data) |
1902 | { | |
1903 | struct loop_device *lo = ptr; | |
1904 | struct loop_device **l = data; | |
1905 | ||
1906 | if (lo->lo_state == Lo_unbound) { | |
1907 | *l = lo; | |
1908 | return 1; | |
1909 | } | |
1910 | return 0; | |
1911 | } | |
1912 | ||
34dd82af | 1913 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1914 | { |
1915 | struct loop_device *lo; | |
34dd82af | 1916 | int ret = -ENODEV; |
a47653fc | 1917 | |
770fe30a KS |
1918 | if (i < 0) { |
1919 | int err; | |
1920 | ||
1921 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1922 | if (err == 1) { | |
1923 | *l = lo; | |
1924 | ret = lo->lo_number; | |
1925 | } | |
1926 | goto out; | |
a47653fc KC |
1927 | } |
1928 | ||
770fe30a | 1929 | /* lookup and return a specific i */ |
34dd82af | 1930 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1931 | if (lo) { |
34dd82af KS |
1932 | *l = lo; |
1933 | ret = lo->lo_number; | |
a47653fc | 1934 | } |
770fe30a | 1935 | out: |
34dd82af | 1936 | return ret; |
a47653fc KC |
1937 | } |
1938 | ||
73285082 KC |
1939 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1940 | { | |
705962cc | 1941 | struct loop_device *lo; |
07002e99 | 1942 | struct kobject *kobj; |
34dd82af | 1943 | int err; |
73285082 | 1944 | |
34dd82af KS |
1945 | mutex_lock(&loop_index_mutex); |
1946 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1947 | if (err < 0) | |
1948 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
1949 | if (err < 0) | |
a207f593 | 1950 | kobj = NULL; |
34dd82af KS |
1951 | else |
1952 | kobj = get_disk(lo->lo_disk); | |
1953 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
1954 | |
1955 | *part = 0; | |
07002e99 | 1956 | return kobj; |
73285082 KC |
1957 | } |
1958 | ||
770fe30a KS |
1959 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
1960 | unsigned long parm) | |
1961 | { | |
1962 | struct loop_device *lo; | |
1963 | int ret = -ENOSYS; | |
1964 | ||
1965 | mutex_lock(&loop_index_mutex); | |
1966 | switch (cmd) { | |
1967 | case LOOP_CTL_ADD: | |
1968 | ret = loop_lookup(&lo, parm); | |
1969 | if (ret >= 0) { | |
1970 | ret = -EEXIST; | |
1971 | break; | |
1972 | } | |
1973 | ret = loop_add(&lo, parm); | |
1974 | break; | |
1975 | case LOOP_CTL_REMOVE: | |
1976 | ret = loop_lookup(&lo, parm); | |
1977 | if (ret < 0) | |
1978 | break; | |
1979 | mutex_lock(&lo->lo_ctl_mutex); | |
1980 | if (lo->lo_state != Lo_unbound) { | |
1981 | ret = -EBUSY; | |
1982 | mutex_unlock(&lo->lo_ctl_mutex); | |
1983 | break; | |
1984 | } | |
f8933667 | 1985 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a KS |
1986 | ret = -EBUSY; |
1987 | mutex_unlock(&lo->lo_ctl_mutex); | |
1988 | break; | |
1989 | } | |
1990 | lo->lo_disk->private_data = NULL; | |
1991 | mutex_unlock(&lo->lo_ctl_mutex); | |
1992 | idr_remove(&loop_index_idr, lo->lo_number); | |
1993 | loop_remove(lo); | |
1994 | break; | |
1995 | case LOOP_CTL_GET_FREE: | |
1996 | ret = loop_lookup(&lo, -1); | |
1997 | if (ret >= 0) | |
1998 | break; | |
1999 | ret = loop_add(&lo, -1); | |
2000 | } | |
2001 | mutex_unlock(&loop_index_mutex); | |
2002 | ||
2003 | return ret; | |
2004 | } | |
2005 | ||
2006 | static const struct file_operations loop_ctl_fops = { | |
2007 | .open = nonseekable_open, | |
2008 | .unlocked_ioctl = loop_control_ioctl, | |
2009 | .compat_ioctl = loop_control_ioctl, | |
2010 | .owner = THIS_MODULE, | |
2011 | .llseek = noop_llseek, | |
2012 | }; | |
2013 | ||
2014 | static struct miscdevice loop_misc = { | |
2015 | .minor = LOOP_CTRL_MINOR, | |
2016 | .name = "loop-control", | |
2017 | .fops = &loop_ctl_fops, | |
2018 | }; | |
2019 | ||
2020 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2021 | MODULE_ALIAS("devname:loop-control"); | |
2022 | ||
73285082 KC |
2023 | static int __init loop_init(void) |
2024 | { | |
a47653fc KC |
2025 | int i, nr; |
2026 | unsigned long range; | |
34dd82af | 2027 | struct loop_device *lo; |
770fe30a | 2028 | int err; |
a47653fc | 2029 | |
770fe30a KS |
2030 | err = misc_register(&loop_misc); |
2031 | if (err < 0) | |
2032 | return err; | |
476a4813 LV |
2033 | |
2034 | part_shift = 0; | |
ac04fee0 | 2035 | if (max_part > 0) { |
476a4813 LV |
2036 | part_shift = fls(max_part); |
2037 | ||
ac04fee0 NK |
2038 | /* |
2039 | * Adjust max_part according to part_shift as it is exported | |
2040 | * to user space so that user can decide correct minor number | |
2041 | * if [s]he want to create more devices. | |
2042 | * | |
2043 | * Note that -1 is required because partition 0 is reserved | |
2044 | * for the whole disk. | |
2045 | */ | |
2046 | max_part = (1UL << part_shift) - 1; | |
2047 | } | |
2048 | ||
b1a66504 GC |
2049 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2050 | err = -EINVAL; | |
2051 | goto misc_out; | |
2052 | } | |
78f4bb36 | 2053 | |
b1a66504 GC |
2054 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2055 | err = -EINVAL; | |
2056 | goto misc_out; | |
2057 | } | |
1da177e4 | 2058 | |
d134b00b KS |
2059 | /* |
2060 | * If max_loop is specified, create that many devices upfront. | |
2061 | * This also becomes a hard limit. If max_loop is not specified, | |
2062 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2063 | * init time. Loop devices can be requested on-demand with the | |
2064 | * /dev/loop-control interface, or be instantiated by accessing | |
2065 | * a 'dead' device node. | |
2066 | */ | |
73285082 | 2067 | if (max_loop) { |
a47653fc | 2068 | nr = max_loop; |
a1c15c59 | 2069 | range = max_loop << part_shift; |
a47653fc | 2070 | } else { |
d134b00b | 2071 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2072 | range = 1UL << MINORBITS; |
a47653fc KC |
2073 | } |
2074 | ||
b1a66504 GC |
2075 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2076 | err = -EIO; | |
2077 | goto misc_out; | |
2078 | } | |
1da177e4 | 2079 | |
a47653fc KC |
2080 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2081 | THIS_MODULE, loop_probe, NULL, NULL); | |
2082 | ||
d134b00b | 2083 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
2084 | mutex_lock(&loop_index_mutex); |
2085 | for (i = 0; i < nr; i++) | |
2086 | loop_add(&lo, i); | |
2087 | mutex_unlock(&loop_index_mutex); | |
2088 | ||
73285082 | 2089 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2090 | return 0; |
b1a66504 GC |
2091 | |
2092 | misc_out: | |
2093 | misc_deregister(&loop_misc); | |
2094 | return err; | |
34dd82af | 2095 | } |
a47653fc | 2096 | |
34dd82af KS |
2097 | static int loop_exit_cb(int id, void *ptr, void *data) |
2098 | { | |
2099 | struct loop_device *lo = ptr; | |
a47653fc | 2100 | |
34dd82af KS |
2101 | loop_remove(lo); |
2102 | return 0; | |
1da177e4 LT |
2103 | } |
2104 | ||
73285082 | 2105 | static void __exit loop_exit(void) |
1da177e4 | 2106 | { |
a47653fc | 2107 | unsigned long range; |
1da177e4 | 2108 | |
a1c15c59 | 2109 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2110 | |
34dd82af | 2111 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2112 | idr_destroy(&loop_index_idr); |
73285082 | 2113 | |
a47653fc | 2114 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2115 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2116 | |
2117 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2118 | } |
2119 | ||
2120 | module_init(loop_init); | |
2121 | module_exit(loop_exit); | |
2122 | ||
2123 | #ifndef MODULE | |
2124 | static int __init max_loop_setup(char *str) | |
2125 | { | |
2126 | max_loop = simple_strtol(str, NULL, 0); | |
2127 | return 1; | |
2128 | } | |
2129 | ||
2130 | __setup("max_loop=", max_loop_setup); | |
2131 | #endif |