]>
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; | |
40326d8a SL |
216 | if (use_dio) { |
217 | queue_flag_clear_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
2e5ab5f3 | 218 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
40326d8a SL |
219 | } else { |
220 | queue_flag_set_unlocked(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 | 226 | static int |
1e6ec9ea | 227 | figure_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 | ||
246 | static inline int | |
247 | lo_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 | 264 | static 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 | |
aa4d8616 | 269 | iov_iter_bvec(&i, ITER_BVEC, 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 |
286 | static 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 |
308 | static 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 |
338 | static 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 |
367 | static 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; | |
412 | out_free_page: | |
413 | __free_page(page); | |
414 | return ret; | |
1da177e4 LT |
415 | } |
416 | ||
cf655d95 ML |
417 | static 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 | ||
441 | static 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 | 451 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 452 | { |
fe2cb290 | 453 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
bc07c10a | 454 | |
fe2cb290 CH |
455 | if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio && |
456 | cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) { | |
bc07c10a ML |
457 | struct bio *bio = cmd->rq->bio; |
458 | ||
fe2cb290 | 459 | bio_advance(bio, cmd->ret); |
bc07c10a ML |
460 | zero_fill_bio(bio); |
461 | } | |
fe2cb290 | 462 | |
2a842aca | 463 | blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK); |
bc07c10a ML |
464 | } |
465 | ||
92d77332 SL |
466 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
467 | { | |
468 | if (!atomic_dec_and_test(&cmd->ref)) | |
469 | return; | |
470 | kfree(cmd->bvec); | |
471 | cmd->bvec = NULL; | |
472 | blk_mq_complete_request(cmd->rq); | |
473 | } | |
474 | ||
bc07c10a ML |
475 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) |
476 | { | |
477 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 478 | |
fe2cb290 | 479 | cmd->ret = ret; |
92d77332 | 480 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
481 | } |
482 | ||
483 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
484 | loff_t pos, bool rw) | |
485 | { | |
486 | struct iov_iter iter; | |
487 | struct bio_vec *bvec; | |
40326d8a SL |
488 | struct request *rq = cmd->rq; |
489 | struct bio *bio = rq->bio; | |
bc07c10a | 490 | struct file *file = lo->lo_backing_file; |
40326d8a SL |
491 | unsigned int offset; |
492 | int segments = 0; | |
bc07c10a ML |
493 | int ret; |
494 | ||
40326d8a SL |
495 | if (rq->bio != rq->biotail) { |
496 | struct req_iterator iter; | |
497 | struct bio_vec tmp; | |
498 | ||
499 | __rq_for_each_bio(bio, rq) | |
500 | segments += bio_segments(bio); | |
501 | bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO); | |
502 | if (!bvec) | |
503 | return -EIO; | |
504 | cmd->bvec = bvec; | |
505 | ||
506 | /* | |
507 | * The bios of the request may be started from the middle of | |
508 | * the 'bvec' because of bio splitting, so we can't directly | |
509 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment | |
510 | * API will take care of all details for us. | |
511 | */ | |
512 | rq_for_each_segment(tmp, rq, iter) { | |
513 | *bvec = tmp; | |
514 | bvec++; | |
515 | } | |
516 | bvec = cmd->bvec; | |
517 | offset = 0; | |
518 | } else { | |
519 | /* | |
520 | * Same here, this bio may be started from the middle of the | |
521 | * 'bvec' because of bio splitting, so offset from the bvec | |
522 | * must be passed to iov iterator | |
523 | */ | |
524 | offset = bio->bi_iter.bi_bvec_done; | |
525 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
526 | segments = bio_segments(bio); | |
527 | } | |
92d77332 | 528 | atomic_set(&cmd->ref, 2); |
bc07c10a | 529 | |
bc07c10a | 530 | iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, |
40326d8a SL |
531 | segments, blk_rq_bytes(rq)); |
532 | iter.iov_offset = offset; | |
bc07c10a ML |
533 | |
534 | cmd->iocb.ki_pos = pos; | |
535 | cmd->iocb.ki_filp = file; | |
536 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
537 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
538 | ||
539 | if (rw == WRITE) | |
bb7462b6 | 540 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 541 | else |
bb7462b6 | 542 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 543 | |
92d77332 SL |
544 | lo_rw_aio_do_completion(cmd); |
545 | ||
bc07c10a ML |
546 | if (ret != -EIOCBQUEUED) |
547 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
548 | return 0; | |
549 | } | |
550 | ||
c1c87c2b | 551 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
552 | { |
553 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 554 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
555 | |
556 | /* | |
557 | * lo_write_simple and lo_read_simple should have been covered | |
558 | * by io submit style function like lo_rw_aio(), one blocker | |
559 | * is that lo_read_simple() need to call flush_dcache_page after | |
560 | * the page is written from kernel, and it isn't easy to handle | |
561 | * this in io submit style function which submits all segments | |
562 | * of the req at one time. And direct read IO doesn't need to | |
563 | * run flush_dcache_page(). | |
564 | */ | |
c1c87c2b CH |
565 | switch (req_op(rq)) { |
566 | case REQ_OP_FLUSH: | |
567 | return lo_req_flush(lo, rq); | |
568 | case REQ_OP_DISCARD: | |
19372e27 | 569 | case REQ_OP_WRITE_ZEROES: |
c1c87c2b CH |
570 | return lo_discard(lo, rq, pos); |
571 | case REQ_OP_WRITE: | |
572 | if (lo->transfer) | |
573 | return lo_write_transfer(lo, rq, pos); | |
574 | else if (cmd->use_aio) | |
575 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 576 | else |
c1c87c2b CH |
577 | return lo_write_simple(lo, rq, pos); |
578 | case REQ_OP_READ: | |
aa4d8616 | 579 | if (lo->transfer) |
c1c87c2b CH |
580 | return lo_read_transfer(lo, rq, pos); |
581 | else if (cmd->use_aio) | |
582 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 583 | else |
c1c87c2b CH |
584 | return lo_read_simple(lo, rq, pos); |
585 | default: | |
586 | WARN_ON_ONCE(1); | |
587 | return -EIO; | |
588 | break; | |
aa4d8616 | 589 | } |
1da177e4 LT |
590 | } |
591 | ||
2e5ab5f3 ML |
592 | static inline void loop_update_dio(struct loop_device *lo) |
593 | { | |
594 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
595 | lo->use_dio); | |
596 | } | |
597 | ||
06f0e9e6 ML |
598 | static void loop_reread_partitions(struct loop_device *lo, |
599 | struct block_device *bdev) | |
600 | { | |
601 | int rc; | |
602 | ||
603 | /* | |
604 | * bd_mutex has been held already in release path, so don't | |
605 | * acquire it if this function is called in such case. | |
606 | * | |
607 | * If the reread partition isn't from release path, lo_refcnt | |
608 | * must be at least one and it can only become zero when the | |
609 | * current holder is released. | |
610 | */ | |
611 | if (!atomic_read(&lo->lo_refcnt)) | |
612 | rc = __blkdev_reread_part(bdev); | |
613 | else | |
614 | rc = blkdev_reread_part(bdev); | |
615 | if (rc) | |
616 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
617 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
618 | } | |
619 | ||
1da177e4 LT |
620 | /* |
621 | * loop_change_fd switched the backing store of a loopback device to | |
622 | * a new file. This is useful for operating system installers to free up | |
623 | * the original file and in High Availability environments to switch to | |
624 | * an alternative location for the content in case of server meltdown. | |
625 | * This can only work if the loop device is used read-only, and if the | |
626 | * new backing store is the same size and type as the old backing store. | |
627 | */ | |
bb214884 AV |
628 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
629 | unsigned int arg) | |
1da177e4 LT |
630 | { |
631 | struct file *file, *old_file; | |
632 | struct inode *inode; | |
633 | int error; | |
634 | ||
635 | error = -ENXIO; | |
636 | if (lo->lo_state != Lo_bound) | |
637 | goto out; | |
638 | ||
639 | /* the loop device has to be read-only */ | |
640 | error = -EINVAL; | |
641 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
642 | goto out; | |
643 | ||
644 | error = -EBADF; | |
645 | file = fget(arg); | |
646 | if (!file) | |
647 | goto out; | |
648 | ||
649 | inode = file->f_mapping->host; | |
650 | old_file = lo->lo_backing_file; | |
651 | ||
652 | error = -EINVAL; | |
653 | ||
654 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
655 | goto out_putf; | |
656 | ||
1da177e4 LT |
657 | /* size of the new backing store needs to be the same */ |
658 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
659 | goto out_putf; | |
660 | ||
661 | /* and ... switch */ | |
43cade80 OS |
662 | blk_mq_freeze_queue(lo->lo_queue); |
663 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
664 | lo->lo_backing_file = file; | |
665 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
666 | mapping_set_gfp_mask(file->f_mapping, | |
667 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
668 | loop_update_dio(lo); | |
669 | blk_mq_unfreeze_queue(lo->lo_queue); | |
1da177e4 LT |
670 | |
671 | fput(old_file); | |
e03c8dd1 | 672 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
06f0e9e6 | 673 | loop_reread_partitions(lo, bdev); |
1da177e4 LT |
674 | return 0; |
675 | ||
676 | out_putf: | |
677 | fput(file); | |
678 | out: | |
679 | return error; | |
680 | } | |
681 | ||
682 | static inline int is_loop_device(struct file *file) | |
683 | { | |
684 | struct inode *i = file->f_mapping->host; | |
685 | ||
686 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
687 | } | |
688 | ||
ee862730 MB |
689 | /* loop sysfs attributes */ |
690 | ||
691 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
692 | ssize_t (*callback)(struct loop_device *, char *)) | |
693 | { | |
34dd82af KS |
694 | struct gendisk *disk = dev_to_disk(dev); |
695 | struct loop_device *lo = disk->private_data; | |
ee862730 | 696 | |
34dd82af | 697 | return callback(lo, page); |
ee862730 MB |
698 | } |
699 | ||
700 | #define LOOP_ATTR_RO(_name) \ | |
701 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
702 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
703 | struct device_attribute *attr, char *b) \ | |
704 | { \ | |
705 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
706 | } \ | |
707 | static struct device_attribute loop_attr_##_name = \ | |
708 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
709 | ||
710 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
711 | { | |
712 | ssize_t ret; | |
713 | char *p = NULL; | |
714 | ||
05eb0f25 | 715 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 716 | if (lo->lo_backing_file) |
9bf39ab2 | 717 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 718 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
719 | |
720 | if (IS_ERR_OR_NULL(p)) | |
721 | ret = PTR_ERR(p); | |
722 | else { | |
723 | ret = strlen(p); | |
724 | memmove(buf, p, ret); | |
725 | buf[ret++] = '\n'; | |
726 | buf[ret] = 0; | |
727 | } | |
728 | ||
729 | return ret; | |
730 | } | |
731 | ||
732 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
733 | { | |
734 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
735 | } | |
736 | ||
737 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
738 | { | |
739 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
740 | } | |
741 | ||
742 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
743 | { | |
744 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
745 | ||
746 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
747 | } | |
748 | ||
e03c8dd1 KS |
749 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
750 | { | |
751 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
752 | ||
753 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
754 | } | |
755 | ||
2e5ab5f3 ML |
756 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
757 | { | |
758 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
759 | ||
760 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
761 | } | |
762 | ||
ee862730 MB |
763 | LOOP_ATTR_RO(backing_file); |
764 | LOOP_ATTR_RO(offset); | |
765 | LOOP_ATTR_RO(sizelimit); | |
766 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 767 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 768 | LOOP_ATTR_RO(dio); |
ee862730 MB |
769 | |
770 | static struct attribute *loop_attrs[] = { | |
771 | &loop_attr_backing_file.attr, | |
772 | &loop_attr_offset.attr, | |
773 | &loop_attr_sizelimit.attr, | |
774 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 775 | &loop_attr_partscan.attr, |
2e5ab5f3 | 776 | &loop_attr_dio.attr, |
ee862730 MB |
777 | NULL, |
778 | }; | |
779 | ||
780 | static struct attribute_group loop_attribute_group = { | |
781 | .name = "loop", | |
782 | .attrs= loop_attrs, | |
783 | }; | |
784 | ||
785 | static int loop_sysfs_init(struct loop_device *lo) | |
786 | { | |
787 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
788 | &loop_attribute_group); | |
789 | } | |
790 | ||
791 | static void loop_sysfs_exit(struct loop_device *lo) | |
792 | { | |
793 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
794 | &loop_attribute_group); | |
795 | } | |
796 | ||
dfaa2ef6 LC |
797 | static void loop_config_discard(struct loop_device *lo) |
798 | { | |
799 | struct file *file = lo->lo_backing_file; | |
800 | struct inode *inode = file->f_mapping->host; | |
801 | struct request_queue *q = lo->lo_queue; | |
802 | ||
803 | /* | |
804 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 805 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
806 | * encryption is enabled, because it may give an attacker |
807 | * useful information. | |
808 | */ | |
809 | if ((!file->f_op->fallocate) || | |
810 | lo->lo_encrypt_key_size) { | |
811 | q->limits.discard_granularity = 0; | |
812 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 813 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 814 | blk_queue_max_write_zeroes_sectors(q, 0); |
dfaa2ef6 LC |
815 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
816 | return; | |
817 | } | |
818 | ||
819 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 820 | q->limits.discard_alignment = 0; |
f2c6df7d | 821 | |
1e6ec9ea OS |
822 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
823 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
dfaa2ef6 LC |
824 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
825 | } | |
826 | ||
e03a3d7a ML |
827 | static void loop_unprepare_queue(struct loop_device *lo) |
828 | { | |
3989144f | 829 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
830 | kthread_stop(lo->worker_task); |
831 | } | |
832 | ||
b2ee7d46 N |
833 | static int loop_kthread_worker_fn(void *worker_ptr) |
834 | { | |
835 | current->flags |= PF_LESS_THROTTLE; | |
836 | return kthread_worker_fn(worker_ptr); | |
837 | } | |
838 | ||
e03a3d7a ML |
839 | static int loop_prepare_queue(struct loop_device *lo) |
840 | { | |
3989144f | 841 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 842 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
843 | &lo->worker, "loop%d", lo->lo_number); |
844 | if (IS_ERR(lo->worker_task)) | |
845 | return -ENOMEM; | |
846 | set_user_nice(lo->worker_task, MIN_NICE); | |
847 | return 0; | |
848 | } | |
849 | ||
bb214884 | 850 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
851 | struct block_device *bdev, unsigned int arg) |
852 | { | |
853 | struct file *file, *f; | |
854 | struct inode *inode; | |
855 | struct address_space *mapping; | |
1da177e4 LT |
856 | int lo_flags = 0; |
857 | int error; | |
858 | loff_t size; | |
859 | ||
860 | /* This is safe, since we have a reference from open(). */ | |
861 | __module_get(THIS_MODULE); | |
862 | ||
863 | error = -EBADF; | |
864 | file = fget(arg); | |
865 | if (!file) | |
866 | goto out; | |
867 | ||
868 | error = -EBUSY; | |
869 | if (lo->lo_state != Lo_unbound) | |
870 | goto out_putf; | |
871 | ||
872 | /* Avoid recursion */ | |
873 | f = file; | |
874 | while (is_loop_device(f)) { | |
875 | struct loop_device *l; | |
876 | ||
bb214884 | 877 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
878 | goto out_putf; |
879 | ||
880 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
881 | if (l->lo_state == Lo_unbound) { | |
882 | error = -EINVAL; | |
883 | goto out_putf; | |
884 | } | |
885 | f = l->lo_backing_file; | |
886 | } | |
887 | ||
888 | mapping = file->f_mapping; | |
889 | inode = mapping->host; | |
890 | ||
1da177e4 | 891 | error = -EINVAL; |
456be148 CH |
892 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
893 | goto out_putf; | |
1da177e4 | 894 | |
456be148 | 895 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 896 | !file->f_op->write_iter) |
456be148 | 897 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 898 | |
456be148 | 899 | error = -EFBIG; |
1da177e4 | 900 | size = get_loop_size(lo, file); |
456be148 | 901 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 902 | goto out_putf; |
e03a3d7a ML |
903 | error = loop_prepare_queue(lo); |
904 | if (error) | |
f4aa4c7b | 905 | goto out_putf; |
1da177e4 | 906 | |
456be148 | 907 | error = 0; |
1da177e4 LT |
908 | |
909 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
910 | ||
2e5ab5f3 | 911 | lo->use_dio = false; |
1da177e4 LT |
912 | lo->lo_device = bdev; |
913 | lo->lo_flags = lo_flags; | |
914 | lo->lo_backing_file = file; | |
aa4d8616 | 915 | lo->transfer = NULL; |
1da177e4 LT |
916 | lo->ioctl = NULL; |
917 | lo->lo_sizelimit = 0; | |
918 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
919 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
920 | ||
68db1961 | 921 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 922 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 923 | |
2e5ab5f3 | 924 | loop_update_dio(lo); |
73285082 | 925 | set_capacity(lo->lo_disk, size); |
1da177e4 | 926 | bd_set_size(bdev, size << 9); |
ee862730 | 927 | loop_sysfs_init(lo); |
c3473c63 DZ |
928 | /* let user-space know about the new size */ |
929 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 930 | |
8a0740c4 OS |
931 | set_blocksize(bdev, S_ISBLK(inode->i_mode) ? |
932 | block_size(inode->i_bdev) : PAGE_SIZE); | |
1da177e4 | 933 | |
6c997918 | 934 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
935 | if (part_shift) |
936 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
937 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
06f0e9e6 | 938 | loop_reread_partitions(lo, bdev); |
c1681bf8 AP |
939 | |
940 | /* Grab the block_device to prevent its destruction after we | |
941 | * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). | |
942 | */ | |
943 | bdgrab(bdev); | |
1da177e4 LT |
944 | return 0; |
945 | ||
946 | out_putf: | |
947 | fput(file); | |
948 | out: | |
949 | /* This is safe: open() is still holding a reference. */ | |
950 | module_put(THIS_MODULE); | |
951 | return error; | |
952 | } | |
953 | ||
954 | static int | |
955 | loop_release_xfer(struct loop_device *lo) | |
956 | { | |
957 | int err = 0; | |
958 | struct loop_func_table *xfer = lo->lo_encryption; | |
959 | ||
960 | if (xfer) { | |
961 | if (xfer->release) | |
962 | err = xfer->release(lo); | |
963 | lo->transfer = NULL; | |
964 | lo->lo_encryption = NULL; | |
965 | module_put(xfer->owner); | |
966 | } | |
967 | return err; | |
968 | } | |
969 | ||
970 | static int | |
971 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
972 | const struct loop_info64 *i) | |
973 | { | |
974 | int err = 0; | |
975 | ||
976 | if (xfer) { | |
977 | struct module *owner = xfer->owner; | |
978 | ||
979 | if (!try_module_get(owner)) | |
980 | return -EINVAL; | |
981 | if (xfer->init) | |
982 | err = xfer->init(lo, i); | |
983 | if (err) | |
984 | module_put(owner); | |
985 | else | |
986 | lo->lo_encryption = xfer; | |
987 | } | |
988 | return err; | |
989 | } | |
990 | ||
4c823cc3 | 991 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
992 | { |
993 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 994 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 995 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
996 | |
997 | if (lo->lo_state != Lo_bound) | |
998 | return -ENXIO; | |
999 | ||
a1ecac3b DC |
1000 | /* |
1001 | * If we've explicitly asked to tear down the loop device, | |
1002 | * and it has an elevated reference count, set it for auto-teardown when | |
1003 | * the last reference goes away. This stops $!~#$@ udev from | |
1004 | * preventing teardown because it decided that it needs to run blkid on | |
1005 | * the loopback device whenever they appear. xfstests is notorious for | |
1006 | * failing tests because blkid via udev races with a losetup | |
1007 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1008 | * command to fail with EBUSY. | |
1009 | */ | |
f8933667 | 1010 | if (atomic_read(&lo->lo_refcnt) > 1) { |
a1ecac3b DC |
1011 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
1012 | mutex_unlock(&lo->lo_ctl_mutex); | |
1013 | return 0; | |
1014 | } | |
1da177e4 LT |
1015 | |
1016 | if (filp == NULL) | |
1017 | return -EINVAL; | |
1018 | ||
f8933667 ML |
1019 | /* freeze request queue during the transition */ |
1020 | blk_mq_freeze_queue(lo->lo_queue); | |
1021 | ||
1da177e4 LT |
1022 | spin_lock_irq(&lo->lo_lock); |
1023 | lo->lo_state = Lo_rundown; | |
1da177e4 | 1024 | lo->lo_backing_file = NULL; |
05eb0f25 | 1025 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1026 | |
1027 | loop_release_xfer(lo); | |
1028 | lo->transfer = NULL; | |
1029 | lo->ioctl = NULL; | |
1030 | lo->lo_device = NULL; | |
1031 | lo->lo_encryption = NULL; | |
1032 | lo->lo_offset = 0; | |
1033 | lo->lo_sizelimit = 0; | |
1034 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1035 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1036 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1037 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1038 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1039 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1040 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1041 | if (bdev) { |
1042 | bdput(bdev); | |
bb214884 | 1043 | invalidate_bdev(bdev); |
c1681bf8 | 1044 | } |
73285082 | 1045 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1046 | loop_sysfs_exit(lo); |
c3473c63 | 1047 | if (bdev) { |
bb214884 | 1048 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1049 | /* let user-space know about this change */ |
1050 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1051 | } | |
1da177e4 LT |
1052 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1053 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1054 | /* This is safe: open() is still holding a reference. */ |
1055 | module_put(THIS_MODULE); | |
f8933667 ML |
1056 | blk_mq_unfreeze_queue(lo->lo_queue); |
1057 | ||
c2fccc1c | 1058 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
06f0e9e6 | 1059 | loop_reread_partitions(lo, bdev); |
e03c8dd1 KS |
1060 | lo->lo_flags = 0; |
1061 | if (!part_shift) | |
1062 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
e03a3d7a | 1063 | loop_unprepare_queue(lo); |
f028f3b2 NK |
1064 | mutex_unlock(&lo->lo_ctl_mutex); |
1065 | /* | |
1066 | * Need not hold lo_ctl_mutex to fput backing file. | |
1067 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1068 | * lock dependency possibility warning as fput can take | |
1069 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1070 | */ | |
1071 | fput(filp); | |
1da177e4 LT |
1072 | return 0; |
1073 | } | |
1074 | ||
1075 | static int | |
1076 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1077 | { | |
1078 | int err; | |
1079 | struct loop_func_table *xfer; | |
e4849737 | 1080 | kuid_t uid = current_uid(); |
1da177e4 | 1081 | |
b0fafa81 | 1082 | if (lo->lo_encrypt_key_size && |
e4849737 | 1083 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1084 | !capable(CAP_SYS_ADMIN)) |
1085 | return -EPERM; | |
1086 | if (lo->lo_state != Lo_bound) | |
1087 | return -ENXIO; | |
1088 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1089 | return -EINVAL; | |
1090 | ||
ecdd0959 ML |
1091 | /* I/O need to be drained during transfer transition */ |
1092 | blk_mq_freeze_queue(lo->lo_queue); | |
1093 | ||
1da177e4 LT |
1094 | err = loop_release_xfer(lo); |
1095 | if (err) | |
ecdd0959 | 1096 | goto exit; |
1da177e4 LT |
1097 | |
1098 | if (info->lo_encrypt_type) { | |
1099 | unsigned int type = info->lo_encrypt_type; | |
1100 | ||
1101 | if (type >= MAX_LO_CRYPT) | |
1102 | return -EINVAL; | |
1103 | xfer = xfer_funcs[type]; | |
1104 | if (xfer == NULL) | |
1105 | return -EINVAL; | |
1106 | } else | |
1107 | xfer = NULL; | |
1108 | ||
1109 | err = loop_init_xfer(lo, xfer, info); | |
1110 | if (err) | |
ecdd0959 | 1111 | goto exit; |
1da177e4 LT |
1112 | |
1113 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea OS |
1114 | lo->lo_sizelimit != info->lo_sizelimit) { |
1115 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { | |
ecdd0959 ML |
1116 | err = -EFBIG; |
1117 | goto exit; | |
1118 | } | |
b040ad9c | 1119 | } |
541c742a | 1120 | |
dfaa2ef6 | 1121 | loop_config_discard(lo); |
1da177e4 LT |
1122 | |
1123 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1124 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1125 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1126 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1127 | ||
1128 | if (!xfer) | |
1129 | xfer = &none_funcs; | |
1130 | lo->transfer = xfer->transfer; | |
1131 | lo->ioctl = xfer->ioctl; | |
1132 | ||
96c58655 DW |
1133 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1134 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1135 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1136 | ||
1da177e4 LT |
1137 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1138 | lo->lo_init[0] = info->lo_init[0]; | |
1139 | lo->lo_init[1] = info->lo_init[1]; | |
1140 | if (info->lo_encrypt_key_size) { | |
1141 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1142 | info->lo_encrypt_key_size); | |
b0fafa81 | 1143 | lo->lo_key_owner = uid; |
aa4d8616 | 1144 | } |
1da177e4 | 1145 | |
2e5ab5f3 ML |
1146 | /* update dio if lo_offset or transfer is changed */ |
1147 | __loop_update_dio(lo, lo->use_dio); | |
1148 | ||
ecdd0959 ML |
1149 | exit: |
1150 | blk_mq_unfreeze_queue(lo->lo_queue); | |
e02898b4 OS |
1151 | |
1152 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1153 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1154 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1155 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1156 | loop_reread_partitions(lo, lo->lo_device); | |
1157 | } | |
1158 | ||
ecdd0959 | 1159 | return err; |
1da177e4 LT |
1160 | } |
1161 | ||
1162 | static int | |
1163 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1164 | { | |
1165 | struct file *file = lo->lo_backing_file; | |
1166 | struct kstat stat; | |
1167 | int error; | |
1168 | ||
1169 | if (lo->lo_state != Lo_bound) | |
1170 | return -ENXIO; | |
a528d35e DH |
1171 | error = vfs_getattr(&file->f_path, &stat, |
1172 | STATX_INO, AT_STATX_SYNC_AS_STAT); | |
1da177e4 LT |
1173 | if (error) |
1174 | return error; | |
1175 | memset(info, 0, sizeof(*info)); | |
1176 | info->lo_number = lo->lo_number; | |
1177 | info->lo_device = huge_encode_dev(stat.dev); | |
1178 | info->lo_inode = stat.ino; | |
1179 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1180 | info->lo_offset = lo->lo_offset; | |
1181 | info->lo_sizelimit = lo->lo_sizelimit; | |
1182 | info->lo_flags = lo->lo_flags; | |
1183 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1184 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1185 | info->lo_encrypt_type = | |
1186 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1187 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1188 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1189 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1190 | lo->lo_encrypt_key_size); | |
1191 | } | |
1192 | return 0; | |
1193 | } | |
1194 | ||
1195 | static void | |
1196 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1197 | { | |
1198 | memset(info64, 0, sizeof(*info64)); | |
1199 | info64->lo_number = info->lo_number; | |
1200 | info64->lo_device = info->lo_device; | |
1201 | info64->lo_inode = info->lo_inode; | |
1202 | info64->lo_rdevice = info->lo_rdevice; | |
1203 | info64->lo_offset = info->lo_offset; | |
1204 | info64->lo_sizelimit = 0; | |
1205 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1206 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1207 | info64->lo_flags = info->lo_flags; | |
1208 | info64->lo_init[0] = info->lo_init[0]; | |
1209 | info64->lo_init[1] = info->lo_init[1]; | |
1210 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1211 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1212 | else | |
1213 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1214 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1215 | } | |
1216 | ||
1217 | static int | |
1218 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1219 | { | |
1220 | memset(info, 0, sizeof(*info)); | |
1221 | info->lo_number = info64->lo_number; | |
1222 | info->lo_device = info64->lo_device; | |
1223 | info->lo_inode = info64->lo_inode; | |
1224 | info->lo_rdevice = info64->lo_rdevice; | |
1225 | info->lo_offset = info64->lo_offset; | |
1226 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1227 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1228 | info->lo_flags = info64->lo_flags; | |
1229 | info->lo_init[0] = info64->lo_init[0]; | |
1230 | info->lo_init[1] = info64->lo_init[1]; | |
1231 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1232 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1233 | else | |
1234 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1235 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1236 | ||
1237 | /* error in case values were truncated */ | |
1238 | if (info->lo_device != info64->lo_device || | |
1239 | info->lo_rdevice != info64->lo_rdevice || | |
1240 | info->lo_inode != info64->lo_inode || | |
1241 | info->lo_offset != info64->lo_offset) | |
1242 | return -EOVERFLOW; | |
1243 | ||
1244 | return 0; | |
1245 | } | |
1246 | ||
1247 | static int | |
1248 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1249 | { | |
1250 | struct loop_info info; | |
1251 | struct loop_info64 info64; | |
1252 | ||
1253 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1254 | return -EFAULT; | |
1255 | loop_info64_from_old(&info, &info64); | |
1256 | return loop_set_status(lo, &info64); | |
1257 | } | |
1258 | ||
1259 | static int | |
1260 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1261 | { | |
1262 | struct loop_info64 info64; | |
1263 | ||
1264 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1265 | return -EFAULT; | |
1266 | return loop_set_status(lo, &info64); | |
1267 | } | |
1268 | ||
1269 | static int | |
1270 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1271 | struct loop_info info; | |
1272 | struct loop_info64 info64; | |
1273 | int err = 0; | |
1274 | ||
1275 | if (!arg) | |
1276 | err = -EINVAL; | |
1277 | if (!err) | |
1278 | err = loop_get_status(lo, &info64); | |
1279 | if (!err) | |
1280 | err = loop_info64_to_old(&info64, &info); | |
1281 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1282 | err = -EFAULT; | |
1283 | ||
1284 | return err; | |
1285 | } | |
1286 | ||
1287 | static int | |
1288 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1289 | struct loop_info64 info64; | |
1290 | int err = 0; | |
1291 | ||
1292 | if (!arg) | |
1293 | err = -EINVAL; | |
1294 | if (!err) | |
1295 | err = loop_get_status(lo, &info64); | |
1296 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1297 | err = -EFAULT; | |
1298 | ||
1299 | return err; | |
1300 | } | |
1301 | ||
51001b7d | 1302 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1303 | { |
53d66608 | 1304 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1305 | return -ENXIO; |
53d66608 | 1306 | |
1e6ec9ea | 1307 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1308 | } |
1309 | ||
ab1cb278 ML |
1310 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1311 | { | |
1312 | int error = -ENXIO; | |
1313 | if (lo->lo_state != Lo_bound) | |
1314 | goto out; | |
1315 | ||
1316 | __loop_update_dio(lo, !!arg); | |
1317 | if (lo->use_dio == !!arg) | |
1318 | return 0; | |
1319 | error = -EINVAL; | |
1320 | out: | |
1321 | return error; | |
1322 | } | |
1323 | ||
89e4fdec OS |
1324 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1325 | { | |
1326 | if (lo->lo_state != Lo_bound) | |
1327 | return -ENXIO; | |
1328 | ||
1329 | if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) | |
1330 | return -EINVAL; | |
1331 | ||
1332 | blk_mq_freeze_queue(lo->lo_queue); | |
1333 | ||
1334 | blk_queue_logical_block_size(lo->lo_queue, arg); | |
bf093753 OS |
1335 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1336 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec OS |
1337 | loop_update_dio(lo); |
1338 | ||
1339 | blk_mq_unfreeze_queue(lo->lo_queue); | |
1340 | ||
1341 | return 0; | |
1342 | } | |
1343 | ||
bb214884 | 1344 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1345 | unsigned int cmd, unsigned long arg) |
1346 | { | |
bb214884 | 1347 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1348 | int err; |
1349 | ||
f028f3b2 | 1350 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1351 | switch (cmd) { |
1352 | case LOOP_SET_FD: | |
bb214884 | 1353 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1354 | break; |
1355 | case LOOP_CHANGE_FD: | |
bb214884 | 1356 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1357 | break; |
1358 | case LOOP_CLR_FD: | |
f028f3b2 | 1359 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1360 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1361 | if (!err) |
1362 | goto out_unlocked; | |
1da177e4 LT |
1363 | break; |
1364 | case LOOP_SET_STATUS: | |
7035b5df DM |
1365 | err = -EPERM; |
1366 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1367 | err = loop_set_status_old(lo, | |
1368 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1369 | break; |
1370 | case LOOP_GET_STATUS: | |
1371 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1372 | break; | |
1373 | case LOOP_SET_STATUS64: | |
7035b5df DM |
1374 | err = -EPERM; |
1375 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1376 | err = loop_set_status64(lo, | |
1377 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1378 | break; |
1379 | case LOOP_GET_STATUS64: | |
1380 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1381 | break; | |
53d66608 O |
1382 | case LOOP_SET_CAPACITY: |
1383 | err = -EPERM; | |
1384 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
51001b7d | 1385 | err = loop_set_capacity(lo); |
53d66608 | 1386 | break; |
ab1cb278 ML |
1387 | case LOOP_SET_DIRECT_IO: |
1388 | err = -EPERM; | |
1389 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1390 | err = loop_set_dio(lo, arg); | |
1391 | break; | |
89e4fdec OS |
1392 | case LOOP_SET_BLOCK_SIZE: |
1393 | err = -EPERM; | |
1394 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1395 | err = loop_set_block_size(lo, arg); | |
1396 | break; | |
1da177e4 LT |
1397 | default: |
1398 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1399 | } | |
f85221dd | 1400 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1401 | |
1402 | out_unlocked: | |
1da177e4 LT |
1403 | return err; |
1404 | } | |
1405 | ||
863d5b82 DH |
1406 | #ifdef CONFIG_COMPAT |
1407 | struct compat_loop_info { | |
1408 | compat_int_t lo_number; /* ioctl r/o */ | |
1409 | compat_dev_t lo_device; /* ioctl r/o */ | |
1410 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1411 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1412 | compat_int_t lo_offset; | |
1413 | compat_int_t lo_encrypt_type; | |
1414 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1415 | compat_int_t lo_flags; /* ioctl r/o */ | |
1416 | char lo_name[LO_NAME_SIZE]; | |
1417 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1418 | compat_ulong_t lo_init[2]; | |
1419 | char reserved[4]; | |
1420 | }; | |
1421 | ||
1422 | /* | |
1423 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1424 | * - noinlined to reduce stack space usage in main part of driver | |
1425 | */ | |
1426 | static noinline int | |
ba674cfc | 1427 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1428 | struct loop_info64 *info64) |
1429 | { | |
1430 | struct compat_loop_info info; | |
1431 | ||
1432 | if (copy_from_user(&info, arg, sizeof(info))) | |
1433 | return -EFAULT; | |
1434 | ||
1435 | memset(info64, 0, sizeof(*info64)); | |
1436 | info64->lo_number = info.lo_number; | |
1437 | info64->lo_device = info.lo_device; | |
1438 | info64->lo_inode = info.lo_inode; | |
1439 | info64->lo_rdevice = info.lo_rdevice; | |
1440 | info64->lo_offset = info.lo_offset; | |
1441 | info64->lo_sizelimit = 0; | |
1442 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1443 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1444 | info64->lo_flags = info.lo_flags; | |
1445 | info64->lo_init[0] = info.lo_init[0]; | |
1446 | info64->lo_init[1] = info.lo_init[1]; | |
1447 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1448 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1449 | else | |
1450 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1451 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1452 | return 0; | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1457 | * - noinlined to reduce stack space usage in main part of driver | |
1458 | */ | |
1459 | static noinline int | |
1460 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1461 | struct compat_loop_info __user *arg) | |
1462 | { | |
1463 | struct compat_loop_info info; | |
1464 | ||
1465 | memset(&info, 0, sizeof(info)); | |
1466 | info.lo_number = info64->lo_number; | |
1467 | info.lo_device = info64->lo_device; | |
1468 | info.lo_inode = info64->lo_inode; | |
1469 | info.lo_rdevice = info64->lo_rdevice; | |
1470 | info.lo_offset = info64->lo_offset; | |
1471 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1472 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1473 | info.lo_flags = info64->lo_flags; | |
1474 | info.lo_init[0] = info64->lo_init[0]; | |
1475 | info.lo_init[1] = info64->lo_init[1]; | |
1476 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1477 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1478 | else | |
1479 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1480 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1481 | ||
1482 | /* error in case values were truncated */ | |
1483 | if (info.lo_device != info64->lo_device || | |
1484 | info.lo_rdevice != info64->lo_rdevice || | |
1485 | info.lo_inode != info64->lo_inode || | |
1486 | info.lo_offset != info64->lo_offset || | |
1487 | info.lo_init[0] != info64->lo_init[0] || | |
1488 | info.lo_init[1] != info64->lo_init[1]) | |
1489 | return -EOVERFLOW; | |
1490 | ||
1491 | if (copy_to_user(arg, &info, sizeof(info))) | |
1492 | return -EFAULT; | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | static int | |
1497 | loop_set_status_compat(struct loop_device *lo, | |
1498 | const struct compat_loop_info __user *arg) | |
1499 | { | |
1500 | struct loop_info64 info64; | |
1501 | int ret; | |
1502 | ||
1503 | ret = loop_info64_from_compat(arg, &info64); | |
1504 | if (ret < 0) | |
1505 | return ret; | |
1506 | return loop_set_status(lo, &info64); | |
1507 | } | |
1508 | ||
1509 | static int | |
1510 | loop_get_status_compat(struct loop_device *lo, | |
1511 | struct compat_loop_info __user *arg) | |
1512 | { | |
1513 | struct loop_info64 info64; | |
1514 | int err = 0; | |
1515 | ||
1516 | if (!arg) | |
1517 | err = -EINVAL; | |
1518 | if (!err) | |
1519 | err = loop_get_status(lo, &info64); | |
1520 | if (!err) | |
1521 | err = loop_info64_to_compat(&info64, arg); | |
1522 | return err; | |
1523 | } | |
1524 | ||
bb214884 AV |
1525 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1526 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1527 | { |
bb214884 | 1528 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1529 | int err; |
1530 | ||
863d5b82 DH |
1531 | switch(cmd) { |
1532 | case LOOP_SET_STATUS: | |
1533 | mutex_lock(&lo->lo_ctl_mutex); | |
1534 | err = loop_set_status_compat( | |
1535 | lo, (const struct compat_loop_info __user *) arg); | |
1536 | mutex_unlock(&lo->lo_ctl_mutex); | |
1537 | break; | |
1538 | case LOOP_GET_STATUS: | |
1539 | mutex_lock(&lo->lo_ctl_mutex); | |
1540 | err = loop_get_status_compat( | |
1541 | lo, (struct compat_loop_info __user *) arg); | |
1542 | mutex_unlock(&lo->lo_ctl_mutex); | |
1543 | break; | |
53d66608 | 1544 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1545 | case LOOP_CLR_FD: |
1546 | case LOOP_GET_STATUS64: | |
1547 | case LOOP_SET_STATUS64: | |
1548 | arg = (unsigned long) compat_ptr(arg); | |
1549 | case LOOP_SET_FD: | |
1550 | case LOOP_CHANGE_FD: | |
bb214884 | 1551 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1552 | break; |
1553 | default: | |
1554 | err = -ENOIOCTLCMD; | |
1555 | break; | |
1556 | } | |
863d5b82 DH |
1557 | return err; |
1558 | } | |
1559 | #endif | |
1560 | ||
bb214884 | 1561 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1562 | { |
770fe30a KS |
1563 | struct loop_device *lo; |
1564 | int err = 0; | |
1565 | ||
1566 | mutex_lock(&loop_index_mutex); | |
1567 | lo = bdev->bd_disk->private_data; | |
1568 | if (!lo) { | |
1569 | err = -ENXIO; | |
1570 | goto out; | |
1571 | } | |
1da177e4 | 1572 | |
f8933667 | 1573 | atomic_inc(&lo->lo_refcnt); |
770fe30a KS |
1574 | out: |
1575 | mutex_unlock(&loop_index_mutex); | |
1576 | return err; | |
1da177e4 LT |
1577 | } |
1578 | ||
db2a144b | 1579 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1580 | { |
bb214884 | 1581 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1582 | int err; |
1da177e4 | 1583 | |
f8933667 ML |
1584 | if (atomic_dec_return(&lo->lo_refcnt)) |
1585 | return; | |
14f27939 | 1586 | |
f8933667 | 1587 | mutex_lock(&lo->lo_ctl_mutex); |
14f27939 MB |
1588 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { |
1589 | /* | |
1590 | * In autoclear mode, stop the loop thread | |
1591 | * and remove configuration after last close. | |
1592 | */ | |
4c823cc3 | 1593 | err = loop_clr_fd(lo); |
ffcd7dca | 1594 | if (!err) |
db2a144b | 1595 | return; |
43cade80 | 1596 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
1597 | /* |
1598 | * Otherwise keep thread (if running) and config, | |
1599 | * but flush possible ongoing bios in thread. | |
1600 | */ | |
43cade80 OS |
1601 | blk_mq_freeze_queue(lo->lo_queue); |
1602 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 1603 | } |
96c58655 | 1604 | |
f85221dd | 1605 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1606 | } |
1607 | ||
83d5cde4 | 1608 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1609 | .owner = THIS_MODULE, |
bb214884 AV |
1610 | .open = lo_open, |
1611 | .release = lo_release, | |
1612 | .ioctl = lo_ioctl, | |
863d5b82 | 1613 | #ifdef CONFIG_COMPAT |
bb214884 | 1614 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1615 | #endif |
1da177e4 LT |
1616 | }; |
1617 | ||
1618 | /* | |
1619 | * And now the modules code and kernel interface. | |
1620 | */ | |
73285082 | 1621 | static int max_loop; |
ac04fee0 | 1622 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1623 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1624 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1625 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1626 | MODULE_LICENSE("GPL"); |
1627 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1628 | ||
1629 | int loop_register_transfer(struct loop_func_table *funcs) | |
1630 | { | |
1631 | unsigned int n = funcs->number; | |
1632 | ||
1633 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1634 | return -EINVAL; | |
1635 | xfer_funcs[n] = funcs; | |
1636 | return 0; | |
1637 | } | |
1638 | ||
34dd82af KS |
1639 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1640 | { | |
1641 | struct loop_device *lo = ptr; | |
1642 | struct loop_func_table *xfer = data; | |
1643 | ||
1644 | mutex_lock(&lo->lo_ctl_mutex); | |
1645 | if (lo->lo_encryption == xfer) | |
1646 | loop_release_xfer(lo); | |
1647 | mutex_unlock(&lo->lo_ctl_mutex); | |
1648 | return 0; | |
1649 | } | |
1650 | ||
1da177e4 LT |
1651 | int loop_unregister_transfer(int number) |
1652 | { | |
1653 | unsigned int n = number; | |
1da177e4 LT |
1654 | struct loop_func_table *xfer; |
1655 | ||
1656 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1657 | return -EINVAL; | |
1658 | ||
1659 | xfer_funcs[n] = NULL; | |
34dd82af | 1660 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1661 | return 0; |
1662 | } | |
1663 | ||
1664 | EXPORT_SYMBOL(loop_register_transfer); | |
1665 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1666 | ||
fc17b653 | 1667 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1668 | const struct blk_mq_queue_data *bd) |
1669 | { | |
1670 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
f4aa4c7b | 1671 | struct loop_device *lo = cmd->rq->q->queuedata; |
b5dd2f60 ML |
1672 | |
1673 | blk_mq_start_request(bd->rq); | |
1674 | ||
f4aa4c7b | 1675 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1676 | return BLK_STS_IOERR; |
f4aa4c7b | 1677 | |
f0225cac CH |
1678 | switch (req_op(cmd->rq)) { |
1679 | case REQ_OP_FLUSH: | |
1680 | case REQ_OP_DISCARD: | |
19372e27 | 1681 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1682 | cmd->use_aio = false; |
f0225cac CH |
1683 | break; |
1684 | default: | |
1685 | cmd->use_aio = lo->use_dio; | |
1686 | break; | |
1687 | } | |
bc07c10a | 1688 | |
3989144f | 1689 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1690 | |
fc17b653 | 1691 | return BLK_STS_OK; |
b5dd2f60 ML |
1692 | } |
1693 | ||
1694 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1695 | { | |
a8ebb056 | 1696 | const bool write = op_is_write(req_op(cmd->rq)); |
b5dd2f60 | 1697 | struct loop_device *lo = cmd->rq->q->queuedata; |
f4829a9b | 1698 | int ret = 0; |
b5dd2f60 | 1699 | |
f4829a9b CH |
1700 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1701 | ret = -EIO; | |
b5dd2f60 | 1702 | goto failed; |
f4829a9b | 1703 | } |
b5dd2f60 | 1704 | |
30112013 | 1705 | ret = do_req_filebacked(lo, cmd->rq); |
b5dd2f60 | 1706 | failed: |
bc07c10a | 1707 | /* complete non-aio request */ |
fe2cb290 CH |
1708 | if (!cmd->use_aio || ret) { |
1709 | cmd->ret = ret ? -EIO : 0; | |
08e0029a | 1710 | blk_mq_complete_request(cmd->rq); |
fe2cb290 | 1711 | } |
b5dd2f60 ML |
1712 | } |
1713 | ||
e03a3d7a | 1714 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1715 | { |
1716 | struct loop_cmd *cmd = | |
e03a3d7a | 1717 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1718 | |
1719 | loop_handle_cmd(cmd); | |
1720 | } | |
1721 | ||
d6296d39 CH |
1722 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1723 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1724 | { |
1725 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1726 | ||
1727 | cmd->rq = rq; | |
3989144f | 1728 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1729 | |
1730 | return 0; | |
1731 | } | |
1732 | ||
f363b089 | 1733 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1734 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 1735 | .init_request = loop_init_request, |
fe2cb290 | 1736 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1737 | }; |
1738 | ||
34dd82af | 1739 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1740 | { |
1741 | struct loop_device *lo; | |
1742 | struct gendisk *disk; | |
34dd82af | 1743 | int err; |
73285082 | 1744 | |
68d740d7 | 1745 | err = -ENOMEM; |
73285082 | 1746 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1747 | if (!lo) |
73285082 | 1748 | goto out; |
34dd82af | 1749 | |
ef7e7c82 MP |
1750 | lo->lo_state = Lo_unbound; |
1751 | ||
c718aa65 | 1752 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1753 | if (i >= 0) { |
c718aa65 TH |
1754 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1755 | if (err == -ENOSPC) | |
34dd82af | 1756 | err = -EEXIST; |
34dd82af | 1757 | } else { |
c718aa65 | 1758 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1759 | } |
1760 | if (err < 0) | |
1761 | goto out_free_dev; | |
c718aa65 | 1762 | i = err; |
73285082 | 1763 | |
183cfb57 | 1764 | err = -ENOMEM; |
b5dd2f60 ML |
1765 | lo->tag_set.ops = &loop_mq_ops; |
1766 | lo->tag_set.nr_hw_queues = 1; | |
1767 | lo->tag_set.queue_depth = 128; | |
1768 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1769 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
1770 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
1771 | lo->tag_set.driver_data = lo; | |
1772 | ||
1773 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1774 | if (err) | |
3ec981e3 | 1775 | goto out_free_idr; |
73285082 | 1776 | |
b5dd2f60 ML |
1777 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
1778 | if (IS_ERR_OR_NULL(lo->lo_queue)) { | |
1779 | err = PTR_ERR(lo->lo_queue); | |
1780 | goto out_cleanup_tags; | |
1781 | } | |
ef7e7c82 MP |
1782 | lo->lo_queue->queuedata = lo; |
1783 | ||
54bb0ade | 1784 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 1785 | |
5b5e20f4 | 1786 | /* |
40326d8a SL |
1787 | * By default, we do buffer IO, so it doesn't make sense to enable |
1788 | * merge because the I/O submitted to backing file is handled page by | |
1789 | * page. For directio mode, merge does help to dispatch bigger request | |
1790 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 ML |
1791 | */ |
1792 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
1793 | ||
7a649737 | 1794 | err = -ENOMEM; |
476a4813 | 1795 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1796 | if (!disk) |
1797 | goto out_free_queue; | |
1798 | ||
e03c8dd1 KS |
1799 | /* |
1800 | * Disable partition scanning by default. The in-kernel partition | |
1801 | * scanning can be requested individually per-device during its | |
1802 | * setup. Userspace can always add and remove partitions from all | |
1803 | * devices. The needed partition minors are allocated from the | |
1804 | * extended minor space, the main loop device numbers will continue | |
1805 | * to match the loop minors, regardless of the number of partitions | |
1806 | * used. | |
1807 | * | |
1808 | * If max_part is given, partition scanning is globally enabled for | |
1809 | * all loop devices. The minors for the main loop devices will be | |
1810 | * multiples of max_part. | |
1811 | * | |
1812 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1813 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1814 | * complicated, are too static, inflexible and may surprise | |
1815 | * userspace tools. Parameters like this in general should be avoided. | |
1816 | */ | |
1817 | if (!part_shift) | |
1818 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1819 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 | 1820 | mutex_init(&lo->lo_ctl_mutex); |
f8933667 | 1821 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 1822 | lo->lo_number = i; |
73285082 KC |
1823 | spin_lock_init(&lo->lo_lock); |
1824 | disk->major = LOOP_MAJOR; | |
476a4813 | 1825 | disk->first_minor = i << part_shift; |
73285082 KC |
1826 | disk->fops = &lo_fops; |
1827 | disk->private_data = lo; | |
1828 | disk->queue = lo->lo_queue; | |
1829 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1830 | add_disk(disk); |
1831 | *l = lo; | |
1832 | return lo->lo_number; | |
73285082 KC |
1833 | |
1834 | out_free_queue: | |
1835 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
1836 | out_cleanup_tags: |
1837 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
1838 | out_free_idr: |
1839 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
1840 | out_free_dev: |
1841 | kfree(lo); | |
1842 | out: | |
34dd82af | 1843 | return err; |
73285082 KC |
1844 | } |
1845 | ||
34dd82af | 1846 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1847 | { |
73285082 | 1848 | blk_cleanup_queue(lo->lo_queue); |
6cd18e71 | 1849 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 1850 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 1851 | put_disk(lo->lo_disk); |
73285082 KC |
1852 | kfree(lo); |
1853 | } | |
1da177e4 | 1854 | |
770fe30a KS |
1855 | static int find_free_cb(int id, void *ptr, void *data) |
1856 | { | |
1857 | struct loop_device *lo = ptr; | |
1858 | struct loop_device **l = data; | |
1859 | ||
1860 | if (lo->lo_state == Lo_unbound) { | |
1861 | *l = lo; | |
1862 | return 1; | |
1863 | } | |
1864 | return 0; | |
1865 | } | |
1866 | ||
34dd82af | 1867 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1868 | { |
1869 | struct loop_device *lo; | |
34dd82af | 1870 | int ret = -ENODEV; |
a47653fc | 1871 | |
770fe30a KS |
1872 | if (i < 0) { |
1873 | int err; | |
1874 | ||
1875 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1876 | if (err == 1) { | |
1877 | *l = lo; | |
1878 | ret = lo->lo_number; | |
1879 | } | |
1880 | goto out; | |
a47653fc KC |
1881 | } |
1882 | ||
770fe30a | 1883 | /* lookup and return a specific i */ |
34dd82af | 1884 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1885 | if (lo) { |
34dd82af KS |
1886 | *l = lo; |
1887 | ret = lo->lo_number; | |
a47653fc | 1888 | } |
770fe30a | 1889 | out: |
34dd82af | 1890 | return ret; |
a47653fc KC |
1891 | } |
1892 | ||
73285082 KC |
1893 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1894 | { | |
705962cc | 1895 | struct loop_device *lo; |
07002e99 | 1896 | struct kobject *kobj; |
34dd82af | 1897 | int err; |
73285082 | 1898 | |
34dd82af KS |
1899 | mutex_lock(&loop_index_mutex); |
1900 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1901 | if (err < 0) | |
1902 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
1903 | if (err < 0) | |
a207f593 | 1904 | kobj = NULL; |
34dd82af KS |
1905 | else |
1906 | kobj = get_disk(lo->lo_disk); | |
1907 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
1908 | |
1909 | *part = 0; | |
07002e99 | 1910 | return kobj; |
73285082 KC |
1911 | } |
1912 | ||
770fe30a KS |
1913 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
1914 | unsigned long parm) | |
1915 | { | |
1916 | struct loop_device *lo; | |
1917 | int ret = -ENOSYS; | |
1918 | ||
1919 | mutex_lock(&loop_index_mutex); | |
1920 | switch (cmd) { | |
1921 | case LOOP_CTL_ADD: | |
1922 | ret = loop_lookup(&lo, parm); | |
1923 | if (ret >= 0) { | |
1924 | ret = -EEXIST; | |
1925 | break; | |
1926 | } | |
1927 | ret = loop_add(&lo, parm); | |
1928 | break; | |
1929 | case LOOP_CTL_REMOVE: | |
1930 | ret = loop_lookup(&lo, parm); | |
1931 | if (ret < 0) | |
1932 | break; | |
1933 | mutex_lock(&lo->lo_ctl_mutex); | |
1934 | if (lo->lo_state != Lo_unbound) { | |
1935 | ret = -EBUSY; | |
1936 | mutex_unlock(&lo->lo_ctl_mutex); | |
1937 | break; | |
1938 | } | |
f8933667 | 1939 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a KS |
1940 | ret = -EBUSY; |
1941 | mutex_unlock(&lo->lo_ctl_mutex); | |
1942 | break; | |
1943 | } | |
1944 | lo->lo_disk->private_data = NULL; | |
1945 | mutex_unlock(&lo->lo_ctl_mutex); | |
1946 | idr_remove(&loop_index_idr, lo->lo_number); | |
1947 | loop_remove(lo); | |
1948 | break; | |
1949 | case LOOP_CTL_GET_FREE: | |
1950 | ret = loop_lookup(&lo, -1); | |
1951 | if (ret >= 0) | |
1952 | break; | |
1953 | ret = loop_add(&lo, -1); | |
1954 | } | |
1955 | mutex_unlock(&loop_index_mutex); | |
1956 | ||
1957 | return ret; | |
1958 | } | |
1959 | ||
1960 | static const struct file_operations loop_ctl_fops = { | |
1961 | .open = nonseekable_open, | |
1962 | .unlocked_ioctl = loop_control_ioctl, | |
1963 | .compat_ioctl = loop_control_ioctl, | |
1964 | .owner = THIS_MODULE, | |
1965 | .llseek = noop_llseek, | |
1966 | }; | |
1967 | ||
1968 | static struct miscdevice loop_misc = { | |
1969 | .minor = LOOP_CTRL_MINOR, | |
1970 | .name = "loop-control", | |
1971 | .fops = &loop_ctl_fops, | |
1972 | }; | |
1973 | ||
1974 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
1975 | MODULE_ALIAS("devname:loop-control"); | |
1976 | ||
73285082 KC |
1977 | static int __init loop_init(void) |
1978 | { | |
a47653fc KC |
1979 | int i, nr; |
1980 | unsigned long range; | |
34dd82af | 1981 | struct loop_device *lo; |
770fe30a | 1982 | int err; |
a47653fc | 1983 | |
476a4813 | 1984 | part_shift = 0; |
ac04fee0 | 1985 | if (max_part > 0) { |
476a4813 LV |
1986 | part_shift = fls(max_part); |
1987 | ||
ac04fee0 NK |
1988 | /* |
1989 | * Adjust max_part according to part_shift as it is exported | |
1990 | * to user space so that user can decide correct minor number | |
1991 | * if [s]he want to create more devices. | |
1992 | * | |
1993 | * Note that -1 is required because partition 0 is reserved | |
1994 | * for the whole disk. | |
1995 | */ | |
1996 | max_part = (1UL << part_shift) - 1; | |
1997 | } | |
1998 | ||
b1a66504 GC |
1999 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2000 | err = -EINVAL; | |
a8c1d064 | 2001 | goto err_out; |
b1a66504 | 2002 | } |
78f4bb36 | 2003 | |
b1a66504 GC |
2004 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2005 | err = -EINVAL; | |
a8c1d064 | 2006 | goto err_out; |
b1a66504 | 2007 | } |
1da177e4 | 2008 | |
d134b00b KS |
2009 | /* |
2010 | * If max_loop is specified, create that many devices upfront. | |
2011 | * This also becomes a hard limit. If max_loop is not specified, | |
2012 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2013 | * init time. Loop devices can be requested on-demand with the | |
2014 | * /dev/loop-control interface, or be instantiated by accessing | |
2015 | * a 'dead' device node. | |
2016 | */ | |
73285082 | 2017 | if (max_loop) { |
a47653fc | 2018 | nr = max_loop; |
a1c15c59 | 2019 | range = max_loop << part_shift; |
a47653fc | 2020 | } else { |
d134b00b | 2021 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2022 | range = 1UL << MINORBITS; |
a47653fc KC |
2023 | } |
2024 | ||
a8c1d064 AV |
2025 | err = misc_register(&loop_misc); |
2026 | if (err < 0) | |
2027 | goto err_out; | |
2028 | ||
2029 | ||
b1a66504 GC |
2030 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2031 | err = -EIO; | |
2032 | goto misc_out; | |
2033 | } | |
1da177e4 | 2034 | |
a47653fc KC |
2035 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2036 | THIS_MODULE, loop_probe, NULL, NULL); | |
2037 | ||
d134b00b | 2038 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
2039 | mutex_lock(&loop_index_mutex); |
2040 | for (i = 0; i < nr; i++) | |
2041 | loop_add(&lo, i); | |
2042 | mutex_unlock(&loop_index_mutex); | |
2043 | ||
73285082 | 2044 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2045 | return 0; |
b1a66504 GC |
2046 | |
2047 | misc_out: | |
2048 | misc_deregister(&loop_misc); | |
a8c1d064 | 2049 | err_out: |
b1a66504 | 2050 | return err; |
34dd82af | 2051 | } |
a47653fc | 2052 | |
34dd82af KS |
2053 | static int loop_exit_cb(int id, void *ptr, void *data) |
2054 | { | |
2055 | struct loop_device *lo = ptr; | |
a47653fc | 2056 | |
34dd82af KS |
2057 | loop_remove(lo); |
2058 | return 0; | |
1da177e4 LT |
2059 | } |
2060 | ||
73285082 | 2061 | static void __exit loop_exit(void) |
1da177e4 | 2062 | { |
a47653fc | 2063 | unsigned long range; |
1da177e4 | 2064 | |
a1c15c59 | 2065 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2066 | |
34dd82af | 2067 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2068 | idr_destroy(&loop_index_idr); |
73285082 | 2069 | |
a47653fc | 2070 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2071 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2072 | |
2073 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2074 | } |
2075 | ||
2076 | module_init(loop_init); | |
2077 | module_exit(loop_exit); | |
2078 | ||
2079 | #ifndef MODULE | |
2080 | static int __init max_loop_setup(char *str) | |
2081 | { | |
2082 | max_loop = simple_strtol(str, NULL, 0); | |
2083 | return 1; | |
2084 | } | |
2085 | ||
2086 | __setup("max_loop=", max_loop_setup); | |
2087 | #endif |