]>
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 | |
b11f6e68 | 269 | iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len); |
1da177e4 | 270 | |
03d95eb2 | 271 | file_start_write(file); |
abbb6589 | 272 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 273 | file_end_write(file); |
aa4d8616 CH |
274 | |
275 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 276 | return 0; |
aa4d8616 CH |
277 | |
278 | printk_ratelimited(KERN_ERR | |
279 | "loop: Write error at byte offset %llu, length %i.\n", | |
280 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
281 | if (bw >= 0) |
282 | bw = -EIO; | |
283 | return bw; | |
284 | } | |
285 | ||
aa4d8616 CH |
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 | |
d4478e92 SL |
479 | if (cmd->css) |
480 | css_put(cmd->css); | |
fe2cb290 | 481 | cmd->ret = ret; |
92d77332 | 482 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
483 | } |
484 | ||
485 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
486 | loff_t pos, bool rw) | |
487 | { | |
488 | struct iov_iter iter; | |
489 | struct bio_vec *bvec; | |
40326d8a SL |
490 | struct request *rq = cmd->rq; |
491 | struct bio *bio = rq->bio; | |
bc07c10a | 492 | struct file *file = lo->lo_backing_file; |
40326d8a SL |
493 | unsigned int offset; |
494 | int segments = 0; | |
bc07c10a ML |
495 | int ret; |
496 | ||
40326d8a SL |
497 | if (rq->bio != rq->biotail) { |
498 | struct req_iterator iter; | |
499 | struct bio_vec tmp; | |
500 | ||
501 | __rq_for_each_bio(bio, rq) | |
502 | segments += bio_segments(bio); | |
503 | bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO); | |
504 | if (!bvec) | |
505 | return -EIO; | |
506 | cmd->bvec = bvec; | |
507 | ||
508 | /* | |
509 | * The bios of the request may be started from the middle of | |
510 | * the 'bvec' because of bio splitting, so we can't directly | |
511 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment | |
512 | * API will take care of all details for us. | |
513 | */ | |
514 | rq_for_each_segment(tmp, rq, iter) { | |
515 | *bvec = tmp; | |
516 | bvec++; | |
517 | } | |
518 | bvec = cmd->bvec; | |
519 | offset = 0; | |
520 | } else { | |
521 | /* | |
522 | * Same here, this bio may be started from the middle of the | |
523 | * 'bvec' because of bio splitting, so offset from the bvec | |
524 | * must be passed to iov iterator | |
525 | */ | |
526 | offset = bio->bi_iter.bi_bvec_done; | |
527 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
528 | segments = bio_segments(bio); | |
529 | } | |
92d77332 | 530 | atomic_set(&cmd->ref, 2); |
bc07c10a | 531 | |
bc07c10a | 532 | iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, |
40326d8a SL |
533 | segments, blk_rq_bytes(rq)); |
534 | iter.iov_offset = offset; | |
bc07c10a ML |
535 | |
536 | cmd->iocb.ki_pos = pos; | |
537 | cmd->iocb.ki_filp = file; | |
538 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
539 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d4478e92 SL |
540 | if (cmd->css) |
541 | kthread_associate_blkcg(cmd->css); | |
bc07c10a ML |
542 | |
543 | if (rw == WRITE) | |
bb7462b6 | 544 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 545 | else |
bb7462b6 | 546 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 547 | |
92d77332 | 548 | lo_rw_aio_do_completion(cmd); |
d4478e92 | 549 | kthread_associate_blkcg(NULL); |
92d77332 | 550 | |
bc07c10a ML |
551 | if (ret != -EIOCBQUEUED) |
552 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
553 | return 0; | |
554 | } | |
555 | ||
c1c87c2b | 556 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
557 | { |
558 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 559 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
560 | |
561 | /* | |
562 | * lo_write_simple and lo_read_simple should have been covered | |
563 | * by io submit style function like lo_rw_aio(), one blocker | |
564 | * is that lo_read_simple() need to call flush_dcache_page after | |
565 | * the page is written from kernel, and it isn't easy to handle | |
566 | * this in io submit style function which submits all segments | |
567 | * of the req at one time. And direct read IO doesn't need to | |
568 | * run flush_dcache_page(). | |
569 | */ | |
c1c87c2b CH |
570 | switch (req_op(rq)) { |
571 | case REQ_OP_FLUSH: | |
572 | return lo_req_flush(lo, rq); | |
573 | case REQ_OP_DISCARD: | |
19372e27 | 574 | case REQ_OP_WRITE_ZEROES: |
c1c87c2b CH |
575 | return lo_discard(lo, rq, pos); |
576 | case REQ_OP_WRITE: | |
577 | if (lo->transfer) | |
578 | return lo_write_transfer(lo, rq, pos); | |
579 | else if (cmd->use_aio) | |
580 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 581 | else |
c1c87c2b CH |
582 | return lo_write_simple(lo, rq, pos); |
583 | case REQ_OP_READ: | |
aa4d8616 | 584 | if (lo->transfer) |
c1c87c2b CH |
585 | return lo_read_transfer(lo, rq, pos); |
586 | else if (cmd->use_aio) | |
587 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 588 | else |
c1c87c2b CH |
589 | return lo_read_simple(lo, rq, pos); |
590 | default: | |
591 | WARN_ON_ONCE(1); | |
592 | return -EIO; | |
593 | break; | |
aa4d8616 | 594 | } |
1da177e4 LT |
595 | } |
596 | ||
2e5ab5f3 ML |
597 | static inline void loop_update_dio(struct loop_device *lo) |
598 | { | |
599 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
600 | lo->use_dio); | |
601 | } | |
602 | ||
c088e31d SF |
603 | static struct file *loop_real_file(struct file *file) |
604 | { | |
605 | struct file *f = NULL; | |
606 | ||
607 | if (file->f_path.dentry->d_sb->s_op->real_loop) | |
608 | f = file->f_path.dentry->d_sb->s_op->real_loop(file); | |
609 | return f; | |
610 | } | |
611 | ||
06f0e9e6 ML |
612 | static void loop_reread_partitions(struct loop_device *lo, |
613 | struct block_device *bdev) | |
614 | { | |
615 | int rc; | |
616 | ||
617 | /* | |
618 | * bd_mutex has been held already in release path, so don't | |
619 | * acquire it if this function is called in such case. | |
620 | * | |
621 | * If the reread partition isn't from release path, lo_refcnt | |
622 | * must be at least one and it can only become zero when the | |
623 | * current holder is released. | |
624 | */ | |
625 | if (!atomic_read(&lo->lo_refcnt)) | |
626 | rc = __blkdev_reread_part(bdev); | |
627 | else | |
628 | rc = blkdev_reread_part(bdev); | |
629 | if (rc) | |
630 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
631 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
632 | } | |
633 | ||
9c6b9fb1 TT |
634 | static inline int is_loop_device(struct file *file) |
635 | { | |
636 | struct inode *i = file->f_mapping->host; | |
637 | ||
638 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
639 | } | |
640 | ||
641 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
642 | { | |
643 | struct inode *inode = file->f_mapping->host; | |
644 | struct file *f = file; | |
645 | ||
646 | /* Avoid recursion */ | |
647 | while (is_loop_device(f)) { | |
648 | struct loop_device *l; | |
649 | ||
650 | if (f->f_mapping->host->i_bdev == bdev) | |
651 | return -EBADF; | |
652 | ||
653 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
654 | if (l->lo_state == Lo_unbound) { | |
655 | return -EINVAL; | |
656 | } | |
657 | f = l->lo_backing_file; | |
658 | } | |
659 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
660 | return -EINVAL; | |
661 | return 0; | |
662 | } | |
663 | ||
1da177e4 LT |
664 | /* |
665 | * loop_change_fd switched the backing store of a loopback device to | |
666 | * a new file. This is useful for operating system installers to free up | |
667 | * the original file and in High Availability environments to switch to | |
668 | * an alternative location for the content in case of server meltdown. | |
669 | * This can only work if the loop device is used read-only, and if the | |
670 | * new backing store is the same size and type as the old backing store. | |
671 | */ | |
bb214884 AV |
672 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
673 | unsigned int arg) | |
1da177e4 LT |
674 | { |
675 | struct file *file, *old_file; | |
c088e31d | 676 | struct file *f, *virt_file = NULL, *old_virt_file; |
1da177e4 LT |
677 | struct inode *inode; |
678 | int error; | |
679 | ||
680 | error = -ENXIO; | |
681 | if (lo->lo_state != Lo_bound) | |
682 | goto out; | |
683 | ||
684 | /* the loop device has to be read-only */ | |
685 | error = -EINVAL; | |
686 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
687 | goto out; | |
688 | ||
689 | error = -EBADF; | |
690 | file = fget(arg); | |
691 | if (!file) | |
692 | goto out; | |
c088e31d SF |
693 | f = loop_real_file(file); |
694 | if (f) { | |
695 | virt_file = file; | |
696 | file = f; | |
697 | get_file(file); | |
698 | } | |
1da177e4 | 699 | |
9c6b9fb1 TT |
700 | error = loop_validate_file(file, bdev); |
701 | if (error) | |
702 | goto out_putf; | |
703 | ||
1da177e4 LT |
704 | inode = file->f_mapping->host; |
705 | old_file = lo->lo_backing_file; | |
c088e31d | 706 | old_virt_file = lo->lo_backing_virt_file; |
1da177e4 LT |
707 | |
708 | error = -EINVAL; | |
709 | ||
1da177e4 LT |
710 | /* size of the new backing store needs to be the same */ |
711 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
712 | goto out_putf; | |
713 | ||
714 | /* and ... switch */ | |
43cade80 OS |
715 | blk_mq_freeze_queue(lo->lo_queue); |
716 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
717 | lo->lo_backing_file = file; | |
c088e31d | 718 | lo->lo_backing_virt_file = virt_file; |
43cade80 OS |
719 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); |
720 | mapping_set_gfp_mask(file->f_mapping, | |
721 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
722 | loop_update_dio(lo); | |
723 | blk_mq_unfreeze_queue(lo->lo_queue); | |
1da177e4 LT |
724 | |
725 | fput(old_file); | |
c088e31d SF |
726 | if (old_virt_file) |
727 | fput(old_virt_file); | |
e03c8dd1 | 728 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
06f0e9e6 | 729 | loop_reread_partitions(lo, bdev); |
1da177e4 LT |
730 | return 0; |
731 | ||
732 | out_putf: | |
733 | fput(file); | |
c088e31d SF |
734 | if (virt_file) |
735 | fput(virt_file); | |
1da177e4 LT |
736 | out: |
737 | return error; | |
738 | } | |
739 | ||
c088e31d SF |
740 | /* |
741 | * for AUFS | |
742 | * no get/put for file. | |
743 | */ | |
744 | struct file *loop_backing_file(struct super_block *sb) | |
745 | { | |
746 | struct file *ret; | |
747 | struct loop_device *l; | |
748 | ||
749 | ret = NULL; | |
750 | if (MAJOR(sb->s_dev) == LOOP_MAJOR) { | |
751 | l = sb->s_bdev->bd_disk->private_data; | |
752 | ret = l->lo_backing_file; | |
753 | } | |
754 | return ret; | |
755 | } | |
756 | EXPORT_SYMBOL_GPL(loop_backing_file); | |
757 | ||
ee862730 MB |
758 | /* loop sysfs attributes */ |
759 | ||
760 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
761 | ssize_t (*callback)(struct loop_device *, char *)) | |
762 | { | |
34dd82af KS |
763 | struct gendisk *disk = dev_to_disk(dev); |
764 | struct loop_device *lo = disk->private_data; | |
ee862730 | 765 | |
34dd82af | 766 | return callback(lo, page); |
ee862730 MB |
767 | } |
768 | ||
769 | #define LOOP_ATTR_RO(_name) \ | |
770 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
771 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
772 | struct device_attribute *attr, char *b) \ | |
773 | { \ | |
774 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
775 | } \ | |
776 | static struct device_attribute loop_attr_##_name = \ | |
777 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
778 | ||
779 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
780 | { | |
781 | ssize_t ret; | |
782 | char *p = NULL; | |
783 | ||
05eb0f25 | 784 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 785 | if (lo->lo_backing_file) |
9bf39ab2 | 786 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 787 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
788 | |
789 | if (IS_ERR_OR_NULL(p)) | |
790 | ret = PTR_ERR(p); | |
791 | else { | |
792 | ret = strlen(p); | |
793 | memmove(buf, p, ret); | |
794 | buf[ret++] = '\n'; | |
795 | buf[ret] = 0; | |
796 | } | |
797 | ||
798 | return ret; | |
799 | } | |
800 | ||
801 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
802 | { | |
803 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
804 | } | |
805 | ||
806 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
807 | { | |
808 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
809 | } | |
810 | ||
811 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
812 | { | |
813 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
814 | ||
815 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
816 | } | |
817 | ||
e03c8dd1 KS |
818 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
819 | { | |
820 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
821 | ||
822 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
823 | } | |
824 | ||
2e5ab5f3 ML |
825 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
826 | { | |
827 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
828 | ||
829 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
830 | } | |
831 | ||
ee862730 MB |
832 | LOOP_ATTR_RO(backing_file); |
833 | LOOP_ATTR_RO(offset); | |
834 | LOOP_ATTR_RO(sizelimit); | |
835 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 836 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 837 | LOOP_ATTR_RO(dio); |
ee862730 MB |
838 | |
839 | static struct attribute *loop_attrs[] = { | |
840 | &loop_attr_backing_file.attr, | |
841 | &loop_attr_offset.attr, | |
842 | &loop_attr_sizelimit.attr, | |
843 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 844 | &loop_attr_partscan.attr, |
2e5ab5f3 | 845 | &loop_attr_dio.attr, |
ee862730 MB |
846 | NULL, |
847 | }; | |
848 | ||
849 | static struct attribute_group loop_attribute_group = { | |
850 | .name = "loop", | |
851 | .attrs= loop_attrs, | |
852 | }; | |
853 | ||
5a700051 | 854 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 855 | { |
5a700051 TH |
856 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
857 | &loop_attribute_group); | |
ee862730 MB |
858 | } |
859 | ||
860 | static void loop_sysfs_exit(struct loop_device *lo) | |
861 | { | |
5a700051 TH |
862 | if (lo->sysfs_inited) |
863 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
864 | &loop_attribute_group); | |
ee862730 MB |
865 | } |
866 | ||
dfaa2ef6 LC |
867 | static void loop_config_discard(struct loop_device *lo) |
868 | { | |
869 | struct file *file = lo->lo_backing_file; | |
870 | struct inode *inode = file->f_mapping->host; | |
871 | struct request_queue *q = lo->lo_queue; | |
872 | ||
873 | /* | |
874 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 875 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
876 | * encryption is enabled, because it may give an attacker |
877 | * useful information. | |
878 | */ | |
879 | if ((!file->f_op->fallocate) || | |
880 | lo->lo_encrypt_key_size) { | |
881 | q->limits.discard_granularity = 0; | |
882 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 883 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 884 | blk_queue_max_write_zeroes_sectors(q, 0); |
dfaa2ef6 LC |
885 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
886 | return; | |
887 | } | |
888 | ||
889 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 890 | q->limits.discard_alignment = 0; |
f2c6df7d | 891 | |
1e6ec9ea OS |
892 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
893 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
dfaa2ef6 LC |
894 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
895 | } | |
896 | ||
e03a3d7a ML |
897 | static void loop_unprepare_queue(struct loop_device *lo) |
898 | { | |
3989144f | 899 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
900 | kthread_stop(lo->worker_task); |
901 | } | |
902 | ||
b2ee7d46 N |
903 | static int loop_kthread_worker_fn(void *worker_ptr) |
904 | { | |
905 | current->flags |= PF_LESS_THROTTLE; | |
906 | return kthread_worker_fn(worker_ptr); | |
907 | } | |
908 | ||
e03a3d7a ML |
909 | static int loop_prepare_queue(struct loop_device *lo) |
910 | { | |
3989144f | 911 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 912 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
913 | &lo->worker, "loop%d", lo->lo_number); |
914 | if (IS_ERR(lo->worker_task)) | |
915 | return -ENOMEM; | |
916 | set_user_nice(lo->worker_task, MIN_NICE); | |
917 | return 0; | |
918 | } | |
919 | ||
bb214884 | 920 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
921 | struct block_device *bdev, unsigned int arg) |
922 | { | |
c088e31d | 923 | struct file *file, *f, *virt_file = NULL; |
1da177e4 LT |
924 | struct inode *inode; |
925 | struct address_space *mapping; | |
1da177e4 LT |
926 | int lo_flags = 0; |
927 | int error; | |
928 | loff_t size; | |
929 | ||
930 | /* This is safe, since we have a reference from open(). */ | |
931 | __module_get(THIS_MODULE); | |
932 | ||
933 | error = -EBADF; | |
934 | file = fget(arg); | |
935 | if (!file) | |
936 | goto out; | |
c088e31d SF |
937 | f = loop_real_file(file); |
938 | if (f) { | |
939 | virt_file = file; | |
940 | file = f; | |
941 | get_file(file); | |
942 | } | |
1da177e4 LT |
943 | |
944 | error = -EBUSY; | |
945 | if (lo->lo_state != Lo_unbound) | |
946 | goto out_putf; | |
947 | ||
9c6b9fb1 TT |
948 | error = loop_validate_file(file, bdev); |
949 | if (error) | |
950 | goto out_putf; | |
1da177e4 LT |
951 | |
952 | mapping = file->f_mapping; | |
953 | inode = mapping->host; | |
954 | ||
456be148 | 955 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 956 | !file->f_op->write_iter) |
456be148 | 957 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 958 | |
456be148 | 959 | error = -EFBIG; |
1da177e4 | 960 | size = get_loop_size(lo, file); |
456be148 | 961 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 962 | goto out_putf; |
e03a3d7a ML |
963 | error = loop_prepare_queue(lo); |
964 | if (error) | |
f4aa4c7b | 965 | goto out_putf; |
1da177e4 | 966 | |
456be148 | 967 | error = 0; |
1da177e4 LT |
968 | |
969 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
970 | ||
2e5ab5f3 | 971 | lo->use_dio = false; |
1da177e4 LT |
972 | lo->lo_device = bdev; |
973 | lo->lo_flags = lo_flags; | |
974 | lo->lo_backing_file = file; | |
c088e31d | 975 | lo->lo_backing_virt_file = virt_file; |
aa4d8616 | 976 | lo->transfer = NULL; |
1da177e4 LT |
977 | lo->ioctl = NULL; |
978 | lo->lo_sizelimit = 0; | |
979 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
980 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
981 | ||
68db1961 | 982 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 983 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 984 | |
2e5ab5f3 | 985 | loop_update_dio(lo); |
73285082 | 986 | set_capacity(lo->lo_disk, size); |
1da177e4 | 987 | bd_set_size(bdev, size << 9); |
ee862730 | 988 | loop_sysfs_init(lo); |
c3473c63 DZ |
989 | /* let user-space know about the new size */ |
990 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 991 | |
8a0740c4 OS |
992 | set_blocksize(bdev, S_ISBLK(inode->i_mode) ? |
993 | block_size(inode->i_bdev) : PAGE_SIZE); | |
1da177e4 | 994 | |
6c997918 | 995 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
996 | if (part_shift) |
997 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
998 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
06f0e9e6 | 999 | loop_reread_partitions(lo, bdev); |
c1681bf8 AP |
1000 | |
1001 | /* Grab the block_device to prevent its destruction after we | |
1002 | * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). | |
1003 | */ | |
1004 | bdgrab(bdev); | |
1da177e4 LT |
1005 | return 0; |
1006 | ||
1007 | out_putf: | |
1008 | fput(file); | |
c088e31d SF |
1009 | if (virt_file) |
1010 | fput(virt_file); | |
1da177e4 LT |
1011 | out: |
1012 | /* This is safe: open() is still holding a reference. */ | |
1013 | module_put(THIS_MODULE); | |
1014 | return error; | |
1015 | } | |
1016 | ||
1017 | static int | |
1018 | loop_release_xfer(struct loop_device *lo) | |
1019 | { | |
1020 | int err = 0; | |
1021 | struct loop_func_table *xfer = lo->lo_encryption; | |
1022 | ||
1023 | if (xfer) { | |
1024 | if (xfer->release) | |
1025 | err = xfer->release(lo); | |
1026 | lo->transfer = NULL; | |
1027 | lo->lo_encryption = NULL; | |
1028 | module_put(xfer->owner); | |
1029 | } | |
1030 | return err; | |
1031 | } | |
1032 | ||
1033 | static int | |
1034 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1035 | const struct loop_info64 *i) | |
1036 | { | |
1037 | int err = 0; | |
1038 | ||
1039 | if (xfer) { | |
1040 | struct module *owner = xfer->owner; | |
1041 | ||
1042 | if (!try_module_get(owner)) | |
1043 | return -EINVAL; | |
1044 | if (xfer->init) | |
1045 | err = xfer->init(lo, i); | |
1046 | if (err) | |
1047 | module_put(owner); | |
1048 | else | |
1049 | lo->lo_encryption = xfer; | |
1050 | } | |
1051 | return err; | |
1052 | } | |
1053 | ||
4c823cc3 | 1054 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
1055 | { |
1056 | struct file *filp = lo->lo_backing_file; | |
c088e31d | 1057 | struct file *virt_filp = lo->lo_backing_virt_file; |
b4e3ca1a | 1058 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1059 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
1060 | |
1061 | if (lo->lo_state != Lo_bound) | |
1062 | return -ENXIO; | |
1063 | ||
a1ecac3b DC |
1064 | /* |
1065 | * If we've explicitly asked to tear down the loop device, | |
1066 | * and it has an elevated reference count, set it for auto-teardown when | |
1067 | * the last reference goes away. This stops $!~#$@ udev from | |
1068 | * preventing teardown because it decided that it needs to run blkid on | |
1069 | * the loopback device whenever they appear. xfstests is notorious for | |
1070 | * failing tests because blkid via udev races with a losetup | |
1071 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1072 | * command to fail with EBUSY. | |
1073 | */ | |
f8933667 | 1074 | if (atomic_read(&lo->lo_refcnt) > 1) { |
a1ecac3b DC |
1075 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
1076 | mutex_unlock(&lo->lo_ctl_mutex); | |
1077 | return 0; | |
1078 | } | |
1da177e4 LT |
1079 | |
1080 | if (filp == NULL) | |
1081 | return -EINVAL; | |
1082 | ||
f8933667 ML |
1083 | /* freeze request queue during the transition */ |
1084 | blk_mq_freeze_queue(lo->lo_queue); | |
1085 | ||
1da177e4 LT |
1086 | spin_lock_irq(&lo->lo_lock); |
1087 | lo->lo_state = Lo_rundown; | |
1da177e4 | 1088 | lo->lo_backing_file = NULL; |
c088e31d | 1089 | lo->lo_backing_virt_file = NULL; |
05eb0f25 | 1090 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1091 | |
1092 | loop_release_xfer(lo); | |
1093 | lo->transfer = NULL; | |
1094 | lo->ioctl = NULL; | |
1095 | lo->lo_device = NULL; | |
1096 | lo->lo_encryption = NULL; | |
1097 | lo->lo_offset = 0; | |
1098 | lo->lo_sizelimit = 0; | |
1099 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1100 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1101 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1102 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1103 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1104 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1105 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1106 | if (bdev) { |
1107 | bdput(bdev); | |
bb214884 | 1108 | invalidate_bdev(bdev); |
c1681bf8 | 1109 | } |
73285082 | 1110 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1111 | loop_sysfs_exit(lo); |
c3473c63 | 1112 | if (bdev) { |
bb214884 | 1113 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1114 | /* let user-space know about this change */ |
1115 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1116 | } | |
1da177e4 LT |
1117 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1118 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1119 | /* This is safe: open() is still holding a reference. */ |
1120 | module_put(THIS_MODULE); | |
f8933667 ML |
1121 | blk_mq_unfreeze_queue(lo->lo_queue); |
1122 | ||
c2fccc1c | 1123 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
06f0e9e6 | 1124 | loop_reread_partitions(lo, bdev); |
e03c8dd1 KS |
1125 | lo->lo_flags = 0; |
1126 | if (!part_shift) | |
1127 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
e03a3d7a | 1128 | loop_unprepare_queue(lo); |
f028f3b2 NK |
1129 | mutex_unlock(&lo->lo_ctl_mutex); |
1130 | /* | |
1131 | * Need not hold lo_ctl_mutex to fput backing file. | |
1132 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1133 | * lock dependency possibility warning as fput can take | |
1134 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1135 | */ | |
1136 | fput(filp); | |
c088e31d SF |
1137 | if (virt_filp) |
1138 | fput(virt_filp); | |
1da177e4 LT |
1139 | return 0; |
1140 | } | |
1141 | ||
1142 | static int | |
1143 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1144 | { | |
1145 | int err; | |
1146 | struct loop_func_table *xfer; | |
e4849737 | 1147 | kuid_t uid = current_uid(); |
1da177e4 | 1148 | |
b0fafa81 | 1149 | if (lo->lo_encrypt_key_size && |
e4849737 | 1150 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1151 | !capable(CAP_SYS_ADMIN)) |
1152 | return -EPERM; | |
1153 | if (lo->lo_state != Lo_bound) | |
1154 | return -ENXIO; | |
1155 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1156 | return -EINVAL; | |
1157 | ||
ecdd0959 ML |
1158 | /* I/O need to be drained during transfer transition */ |
1159 | blk_mq_freeze_queue(lo->lo_queue); | |
1160 | ||
1da177e4 LT |
1161 | err = loop_release_xfer(lo); |
1162 | if (err) | |
ecdd0959 | 1163 | goto exit; |
1da177e4 LT |
1164 | |
1165 | if (info->lo_encrypt_type) { | |
1166 | unsigned int type = info->lo_encrypt_type; | |
1167 | ||
d22d7c3d TH |
1168 | if (type >= MAX_LO_CRYPT) { |
1169 | err = -EINVAL; | |
1170 | goto exit; | |
1171 | } | |
1da177e4 | 1172 | xfer = xfer_funcs[type]; |
d22d7c3d TH |
1173 | if (xfer == NULL) { |
1174 | err = -EINVAL; | |
1175 | goto exit; | |
1176 | } | |
1da177e4 LT |
1177 | } else |
1178 | xfer = NULL; | |
1179 | ||
1180 | err = loop_init_xfer(lo, xfer, info); | |
1181 | if (err) | |
ecdd0959 | 1182 | goto exit; |
1da177e4 LT |
1183 | |
1184 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea OS |
1185 | lo->lo_sizelimit != info->lo_sizelimit) { |
1186 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { | |
ecdd0959 ML |
1187 | err = -EFBIG; |
1188 | goto exit; | |
1189 | } | |
b040ad9c | 1190 | } |
541c742a | 1191 | |
dfaa2ef6 | 1192 | loop_config_discard(lo); |
1da177e4 LT |
1193 | |
1194 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1195 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1196 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1197 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1198 | ||
1199 | if (!xfer) | |
1200 | xfer = &none_funcs; | |
1201 | lo->transfer = xfer->transfer; | |
1202 | lo->ioctl = xfer->ioctl; | |
1203 | ||
96c58655 DW |
1204 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1205 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1206 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1207 | ||
1da177e4 LT |
1208 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1209 | lo->lo_init[0] = info->lo_init[0]; | |
1210 | lo->lo_init[1] = info->lo_init[1]; | |
1211 | if (info->lo_encrypt_key_size) { | |
1212 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1213 | info->lo_encrypt_key_size); | |
b0fafa81 | 1214 | lo->lo_key_owner = uid; |
aa4d8616 | 1215 | } |
1da177e4 | 1216 | |
2e5ab5f3 ML |
1217 | /* update dio if lo_offset or transfer is changed */ |
1218 | __loop_update_dio(lo, lo->use_dio); | |
1219 | ||
ecdd0959 ML |
1220 | exit: |
1221 | blk_mq_unfreeze_queue(lo->lo_queue); | |
e02898b4 OS |
1222 | |
1223 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1224 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1225 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1226 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1227 | loop_reread_partitions(lo, lo->lo_device); | |
1228 | } | |
1229 | ||
ecdd0959 | 1230 | return err; |
1da177e4 LT |
1231 | } |
1232 | ||
1233 | static int | |
1234 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1235 | { | |
a4f4c52d | 1236 | struct file *file; |
1da177e4 | 1237 | struct kstat stat; |
a4f4c52d | 1238 | int ret; |
1da177e4 | 1239 | |
a4f4c52d OS |
1240 | if (lo->lo_state != Lo_bound) { |
1241 | mutex_unlock(&lo->lo_ctl_mutex); | |
1da177e4 | 1242 | return -ENXIO; |
a4f4c52d OS |
1243 | } |
1244 | ||
1da177e4 LT |
1245 | memset(info, 0, sizeof(*info)); |
1246 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1247 | info->lo_offset = lo->lo_offset; |
1248 | info->lo_sizelimit = lo->lo_sizelimit; | |
1249 | info->lo_flags = lo->lo_flags; | |
1250 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1251 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1252 | info->lo_encrypt_type = | |
1253 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1254 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1255 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1256 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1257 | lo->lo_encrypt_key_size); | |
1258 | } | |
a4f4c52d OS |
1259 | |
1260 | /* Drop lo_ctl_mutex while we call into the filesystem. */ | |
1261 | file = get_file(lo->lo_backing_file); | |
1262 | mutex_unlock(&lo->lo_ctl_mutex); | |
1263 | ret = vfs_getattr(&file->f_path, &stat, STATX_INO, | |
1264 | AT_STATX_SYNC_AS_STAT); | |
1265 | if (!ret) { | |
1266 | info->lo_device = huge_encode_dev(stat.dev); | |
1267 | info->lo_inode = stat.ino; | |
1268 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1269 | } | |
1270 | fput(file); | |
1271 | return ret; | |
1da177e4 LT |
1272 | } |
1273 | ||
1274 | static void | |
1275 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1276 | { | |
1277 | memset(info64, 0, sizeof(*info64)); | |
1278 | info64->lo_number = info->lo_number; | |
1279 | info64->lo_device = info->lo_device; | |
1280 | info64->lo_inode = info->lo_inode; | |
1281 | info64->lo_rdevice = info->lo_rdevice; | |
1282 | info64->lo_offset = info->lo_offset; | |
1283 | info64->lo_sizelimit = 0; | |
1284 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1285 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1286 | info64->lo_flags = info->lo_flags; | |
1287 | info64->lo_init[0] = info->lo_init[0]; | |
1288 | info64->lo_init[1] = info->lo_init[1]; | |
1289 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1290 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1291 | else | |
1292 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1293 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1294 | } | |
1295 | ||
1296 | static int | |
1297 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1298 | { | |
1299 | memset(info, 0, sizeof(*info)); | |
1300 | info->lo_number = info64->lo_number; | |
1301 | info->lo_device = info64->lo_device; | |
1302 | info->lo_inode = info64->lo_inode; | |
1303 | info->lo_rdevice = info64->lo_rdevice; | |
1304 | info->lo_offset = info64->lo_offset; | |
1305 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1306 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1307 | info->lo_flags = info64->lo_flags; | |
1308 | info->lo_init[0] = info64->lo_init[0]; | |
1309 | info->lo_init[1] = info64->lo_init[1]; | |
1310 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1311 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1312 | else | |
1313 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1314 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1315 | ||
1316 | /* error in case values were truncated */ | |
1317 | if (info->lo_device != info64->lo_device || | |
1318 | info->lo_rdevice != info64->lo_rdevice || | |
1319 | info->lo_inode != info64->lo_inode || | |
1320 | info->lo_offset != info64->lo_offset) | |
1321 | return -EOVERFLOW; | |
1322 | ||
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | static int | |
1327 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1328 | { | |
1329 | struct loop_info info; | |
1330 | struct loop_info64 info64; | |
1331 | ||
1332 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1333 | return -EFAULT; | |
1334 | loop_info64_from_old(&info, &info64); | |
1335 | return loop_set_status(lo, &info64); | |
1336 | } | |
1337 | ||
1338 | static int | |
1339 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1340 | { | |
1341 | struct loop_info64 info64; | |
1342 | ||
1343 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1344 | return -EFAULT; | |
1345 | return loop_set_status(lo, &info64); | |
1346 | } | |
1347 | ||
1348 | static int | |
1349 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1350 | struct loop_info info; | |
1351 | struct loop_info64 info64; | |
f118686d | 1352 | int err; |
1da177e4 | 1353 | |
f118686d OS |
1354 | if (!arg) { |
1355 | mutex_unlock(&lo->lo_ctl_mutex); | |
1356 | return -EINVAL; | |
1357 | } | |
1358 | err = loop_get_status(lo, &info64); | |
1da177e4 LT |
1359 | if (!err) |
1360 | err = loop_info64_to_old(&info64, &info); | |
1361 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1362 | err = -EFAULT; | |
1363 | ||
1364 | return err; | |
1365 | } | |
1366 | ||
1367 | static int | |
1368 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1369 | struct loop_info64 info64; | |
f118686d | 1370 | int err; |
1da177e4 | 1371 | |
f118686d OS |
1372 | if (!arg) { |
1373 | mutex_unlock(&lo->lo_ctl_mutex); | |
1374 | return -EINVAL; | |
1375 | } | |
1376 | err = loop_get_status(lo, &info64); | |
1da177e4 LT |
1377 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1378 | err = -EFAULT; | |
1379 | ||
1380 | return err; | |
1381 | } | |
1382 | ||
51001b7d | 1383 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1384 | { |
53d66608 | 1385 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1386 | return -ENXIO; |
53d66608 | 1387 | |
1e6ec9ea | 1388 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1389 | } |
1390 | ||
ab1cb278 ML |
1391 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1392 | { | |
1393 | int error = -ENXIO; | |
1394 | if (lo->lo_state != Lo_bound) | |
1395 | goto out; | |
1396 | ||
1397 | __loop_update_dio(lo, !!arg); | |
1398 | if (lo->use_dio == !!arg) | |
1399 | return 0; | |
1400 | error = -EINVAL; | |
1401 | out: | |
1402 | return error; | |
1403 | } | |
1404 | ||
89e4fdec OS |
1405 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1406 | { | |
1407 | if (lo->lo_state != Lo_bound) | |
1408 | return -ENXIO; | |
1409 | ||
1410 | if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) | |
1411 | return -EINVAL; | |
1412 | ||
1413 | blk_mq_freeze_queue(lo->lo_queue); | |
1414 | ||
1415 | blk_queue_logical_block_size(lo->lo_queue, arg); | |
bf093753 OS |
1416 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1417 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec OS |
1418 | loop_update_dio(lo); |
1419 | ||
1420 | blk_mq_unfreeze_queue(lo->lo_queue); | |
1421 | ||
1422 | return 0; | |
1423 | } | |
1424 | ||
bb214884 | 1425 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1426 | unsigned int cmd, unsigned long arg) |
1427 | { | |
bb214884 | 1428 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1429 | int err; |
1430 | ||
f028f3b2 | 1431 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1432 | switch (cmd) { |
1433 | case LOOP_SET_FD: | |
bb214884 | 1434 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1435 | break; |
1436 | case LOOP_CHANGE_FD: | |
bb214884 | 1437 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1438 | break; |
1439 | case LOOP_CLR_FD: | |
f028f3b2 | 1440 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1441 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1442 | if (!err) |
1443 | goto out_unlocked; | |
1da177e4 LT |
1444 | break; |
1445 | case LOOP_SET_STATUS: | |
7035b5df DM |
1446 | err = -EPERM; |
1447 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1448 | err = loop_set_status_old(lo, | |
1449 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1450 | break; |
1451 | case LOOP_GET_STATUS: | |
1452 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
a4f4c52d OS |
1453 | /* loop_get_status() unlocks lo_ctl_mutex */ |
1454 | goto out_unlocked; | |
1da177e4 | 1455 | case LOOP_SET_STATUS64: |
7035b5df DM |
1456 | err = -EPERM; |
1457 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1458 | err = loop_set_status64(lo, | |
1459 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1460 | break; |
1461 | case LOOP_GET_STATUS64: | |
1462 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
a4f4c52d OS |
1463 | /* loop_get_status() unlocks lo_ctl_mutex */ |
1464 | goto out_unlocked; | |
53d66608 O |
1465 | case LOOP_SET_CAPACITY: |
1466 | err = -EPERM; | |
1467 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
51001b7d | 1468 | err = loop_set_capacity(lo); |
53d66608 | 1469 | break; |
ab1cb278 ML |
1470 | case LOOP_SET_DIRECT_IO: |
1471 | err = -EPERM; | |
1472 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1473 | err = loop_set_dio(lo, arg); | |
1474 | break; | |
89e4fdec OS |
1475 | case LOOP_SET_BLOCK_SIZE: |
1476 | err = -EPERM; | |
1477 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1478 | err = loop_set_block_size(lo, arg); | |
1479 | break; | |
1da177e4 LT |
1480 | default: |
1481 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1482 | } | |
f85221dd | 1483 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1484 | |
1485 | out_unlocked: | |
1da177e4 LT |
1486 | return err; |
1487 | } | |
1488 | ||
863d5b82 DH |
1489 | #ifdef CONFIG_COMPAT |
1490 | struct compat_loop_info { | |
1491 | compat_int_t lo_number; /* ioctl r/o */ | |
1492 | compat_dev_t lo_device; /* ioctl r/o */ | |
1493 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1494 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1495 | compat_int_t lo_offset; | |
1496 | compat_int_t lo_encrypt_type; | |
1497 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1498 | compat_int_t lo_flags; /* ioctl r/o */ | |
1499 | char lo_name[LO_NAME_SIZE]; | |
1500 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1501 | compat_ulong_t lo_init[2]; | |
1502 | char reserved[4]; | |
1503 | }; | |
1504 | ||
1505 | /* | |
1506 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1507 | * - noinlined to reduce stack space usage in main part of driver | |
1508 | */ | |
1509 | static noinline int | |
ba674cfc | 1510 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1511 | struct loop_info64 *info64) |
1512 | { | |
1513 | struct compat_loop_info info; | |
1514 | ||
1515 | if (copy_from_user(&info, arg, sizeof(info))) | |
1516 | return -EFAULT; | |
1517 | ||
1518 | memset(info64, 0, sizeof(*info64)); | |
1519 | info64->lo_number = info.lo_number; | |
1520 | info64->lo_device = info.lo_device; | |
1521 | info64->lo_inode = info.lo_inode; | |
1522 | info64->lo_rdevice = info.lo_rdevice; | |
1523 | info64->lo_offset = info.lo_offset; | |
1524 | info64->lo_sizelimit = 0; | |
1525 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1526 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1527 | info64->lo_flags = info.lo_flags; | |
1528 | info64->lo_init[0] = info.lo_init[0]; | |
1529 | info64->lo_init[1] = info.lo_init[1]; | |
1530 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1531 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1532 | else | |
1533 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1534 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1540 | * - noinlined to reduce stack space usage in main part of driver | |
1541 | */ | |
1542 | static noinline int | |
1543 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1544 | struct compat_loop_info __user *arg) | |
1545 | { | |
1546 | struct compat_loop_info info; | |
1547 | ||
1548 | memset(&info, 0, sizeof(info)); | |
1549 | info.lo_number = info64->lo_number; | |
1550 | info.lo_device = info64->lo_device; | |
1551 | info.lo_inode = info64->lo_inode; | |
1552 | info.lo_rdevice = info64->lo_rdevice; | |
1553 | info.lo_offset = info64->lo_offset; | |
1554 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1555 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1556 | info.lo_flags = info64->lo_flags; | |
1557 | info.lo_init[0] = info64->lo_init[0]; | |
1558 | info.lo_init[1] = info64->lo_init[1]; | |
1559 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1560 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1561 | else | |
1562 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1563 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1564 | ||
1565 | /* error in case values were truncated */ | |
1566 | if (info.lo_device != info64->lo_device || | |
1567 | info.lo_rdevice != info64->lo_rdevice || | |
1568 | info.lo_inode != info64->lo_inode || | |
1569 | info.lo_offset != info64->lo_offset || | |
1570 | info.lo_init[0] != info64->lo_init[0] || | |
1571 | info.lo_init[1] != info64->lo_init[1]) | |
1572 | return -EOVERFLOW; | |
1573 | ||
1574 | if (copy_to_user(arg, &info, sizeof(info))) | |
1575 | return -EFAULT; | |
1576 | return 0; | |
1577 | } | |
1578 | ||
1579 | static int | |
1580 | loop_set_status_compat(struct loop_device *lo, | |
1581 | const struct compat_loop_info __user *arg) | |
1582 | { | |
1583 | struct loop_info64 info64; | |
1584 | int ret; | |
1585 | ||
1586 | ret = loop_info64_from_compat(arg, &info64); | |
1587 | if (ret < 0) | |
1588 | return ret; | |
1589 | return loop_set_status(lo, &info64); | |
1590 | } | |
1591 | ||
1592 | static int | |
1593 | loop_get_status_compat(struct loop_device *lo, | |
1594 | struct compat_loop_info __user *arg) | |
1595 | { | |
1596 | struct loop_info64 info64; | |
f118686d | 1597 | int err; |
863d5b82 | 1598 | |
f118686d OS |
1599 | if (!arg) { |
1600 | mutex_unlock(&lo->lo_ctl_mutex); | |
1601 | return -EINVAL; | |
1602 | } | |
1603 | err = loop_get_status(lo, &info64); | |
863d5b82 DH |
1604 | if (!err) |
1605 | err = loop_info64_to_compat(&info64, arg); | |
1606 | return err; | |
1607 | } | |
1608 | ||
bb214884 AV |
1609 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1610 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1611 | { |
bb214884 | 1612 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1613 | int err; |
1614 | ||
863d5b82 DH |
1615 | switch(cmd) { |
1616 | case LOOP_SET_STATUS: | |
1617 | mutex_lock(&lo->lo_ctl_mutex); | |
1618 | err = loop_set_status_compat( | |
1619 | lo, (const struct compat_loop_info __user *) arg); | |
1620 | mutex_unlock(&lo->lo_ctl_mutex); | |
1621 | break; | |
1622 | case LOOP_GET_STATUS: | |
1623 | mutex_lock(&lo->lo_ctl_mutex); | |
1624 | err = loop_get_status_compat( | |
1625 | lo, (struct compat_loop_info __user *) arg); | |
a4f4c52d | 1626 | /* loop_get_status() unlocks lo_ctl_mutex */ |
863d5b82 | 1627 | break; |
53d66608 | 1628 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1629 | case LOOP_CLR_FD: |
1630 | case LOOP_GET_STATUS64: | |
1631 | case LOOP_SET_STATUS64: | |
1632 | arg = (unsigned long) compat_ptr(arg); | |
1633 | case LOOP_SET_FD: | |
1634 | case LOOP_CHANGE_FD: | |
bb214884 | 1635 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1636 | break; |
1637 | default: | |
1638 | err = -ENOIOCTLCMD; | |
1639 | break; | |
1640 | } | |
863d5b82 DH |
1641 | return err; |
1642 | } | |
1643 | #endif | |
1644 | ||
bb214884 | 1645 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1646 | { |
770fe30a KS |
1647 | struct loop_device *lo; |
1648 | int err = 0; | |
1649 | ||
1650 | mutex_lock(&loop_index_mutex); | |
1651 | lo = bdev->bd_disk->private_data; | |
1652 | if (!lo) { | |
1653 | err = -ENXIO; | |
1654 | goto out; | |
1655 | } | |
1da177e4 | 1656 | |
f8933667 | 1657 | atomic_inc(&lo->lo_refcnt); |
770fe30a KS |
1658 | out: |
1659 | mutex_unlock(&loop_index_mutex); | |
1660 | return err; | |
1da177e4 LT |
1661 | } |
1662 | ||
ae665016 | 1663 | static void __lo_release(struct loop_device *lo) |
1da177e4 | 1664 | { |
ffcd7dca | 1665 | int err; |
1da177e4 | 1666 | |
f8933667 ML |
1667 | if (atomic_dec_return(&lo->lo_refcnt)) |
1668 | return; | |
14f27939 | 1669 | |
f8933667 | 1670 | mutex_lock(&lo->lo_ctl_mutex); |
14f27939 MB |
1671 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { |
1672 | /* | |
1673 | * In autoclear mode, stop the loop thread | |
1674 | * and remove configuration after last close. | |
1675 | */ | |
4c823cc3 | 1676 | err = loop_clr_fd(lo); |
ffcd7dca | 1677 | if (!err) |
db2a144b | 1678 | return; |
43cade80 | 1679 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
1680 | /* |
1681 | * Otherwise keep thread (if running) and config, | |
1682 | * but flush possible ongoing bios in thread. | |
1683 | */ | |
43cade80 OS |
1684 | blk_mq_freeze_queue(lo->lo_queue); |
1685 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 1686 | } |
96c58655 | 1687 | |
f85221dd | 1688 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1689 | } |
1690 | ||
ae665016 LT |
1691 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1692 | { | |
1693 | mutex_lock(&loop_index_mutex); | |
1694 | __lo_release(disk->private_data); | |
1695 | mutex_unlock(&loop_index_mutex); | |
1696 | } | |
1697 | ||
83d5cde4 | 1698 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1699 | .owner = THIS_MODULE, |
bb214884 AV |
1700 | .open = lo_open, |
1701 | .release = lo_release, | |
1702 | .ioctl = lo_ioctl, | |
863d5b82 | 1703 | #ifdef CONFIG_COMPAT |
bb214884 | 1704 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1705 | #endif |
1da177e4 LT |
1706 | }; |
1707 | ||
1708 | /* | |
1709 | * And now the modules code and kernel interface. | |
1710 | */ | |
73285082 | 1711 | static int max_loop; |
ac04fee0 | 1712 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1713 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1714 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1715 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1716 | MODULE_LICENSE("GPL"); |
1717 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1718 | ||
1719 | int loop_register_transfer(struct loop_func_table *funcs) | |
1720 | { | |
1721 | unsigned int n = funcs->number; | |
1722 | ||
1723 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1724 | return -EINVAL; | |
1725 | xfer_funcs[n] = funcs; | |
1726 | return 0; | |
1727 | } | |
1728 | ||
34dd82af KS |
1729 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1730 | { | |
1731 | struct loop_device *lo = ptr; | |
1732 | struct loop_func_table *xfer = data; | |
1733 | ||
1734 | mutex_lock(&lo->lo_ctl_mutex); | |
1735 | if (lo->lo_encryption == xfer) | |
1736 | loop_release_xfer(lo); | |
1737 | mutex_unlock(&lo->lo_ctl_mutex); | |
1738 | return 0; | |
1739 | } | |
1740 | ||
1da177e4 LT |
1741 | int loop_unregister_transfer(int number) |
1742 | { | |
1743 | unsigned int n = number; | |
1da177e4 LT |
1744 | struct loop_func_table *xfer; |
1745 | ||
1746 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1747 | return -EINVAL; | |
1748 | ||
1749 | xfer_funcs[n] = NULL; | |
34dd82af | 1750 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1751 | return 0; |
1752 | } | |
1753 | ||
1754 | EXPORT_SYMBOL(loop_register_transfer); | |
1755 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1756 | ||
fc17b653 | 1757 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1758 | const struct blk_mq_queue_data *bd) |
1759 | { | |
1760 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
f4aa4c7b | 1761 | struct loop_device *lo = cmd->rq->q->queuedata; |
b5dd2f60 ML |
1762 | |
1763 | blk_mq_start_request(bd->rq); | |
1764 | ||
f4aa4c7b | 1765 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1766 | return BLK_STS_IOERR; |
f4aa4c7b | 1767 | |
f0225cac CH |
1768 | switch (req_op(cmd->rq)) { |
1769 | case REQ_OP_FLUSH: | |
1770 | case REQ_OP_DISCARD: | |
19372e27 | 1771 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1772 | cmd->use_aio = false; |
f0225cac CH |
1773 | break; |
1774 | default: | |
1775 | cmd->use_aio = lo->use_dio; | |
1776 | break; | |
1777 | } | |
bc07c10a | 1778 | |
d4478e92 | 1779 | /* always use the first bio's css */ |
0b508bc9 | 1780 | #ifdef CONFIG_BLK_CGROUP |
d4478e92 SL |
1781 | if (cmd->use_aio && cmd->rq->bio && cmd->rq->bio->bi_css) { |
1782 | cmd->css = cmd->rq->bio->bi_css; | |
1783 | css_get(cmd->css); | |
1784 | } else | |
1785 | #endif | |
1786 | cmd->css = NULL; | |
3989144f | 1787 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1788 | |
fc17b653 | 1789 | return BLK_STS_OK; |
b5dd2f60 ML |
1790 | } |
1791 | ||
1792 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1793 | { | |
a8ebb056 | 1794 | const bool write = op_is_write(req_op(cmd->rq)); |
b5dd2f60 | 1795 | struct loop_device *lo = cmd->rq->q->queuedata; |
f4829a9b | 1796 | int ret = 0; |
b5dd2f60 | 1797 | |
f4829a9b CH |
1798 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1799 | ret = -EIO; | |
b5dd2f60 | 1800 | goto failed; |
f4829a9b | 1801 | } |
b5dd2f60 | 1802 | |
30112013 | 1803 | ret = do_req_filebacked(lo, cmd->rq); |
b5dd2f60 | 1804 | failed: |
bc07c10a | 1805 | /* complete non-aio request */ |
fe2cb290 CH |
1806 | if (!cmd->use_aio || ret) { |
1807 | cmd->ret = ret ? -EIO : 0; | |
08e0029a | 1808 | blk_mq_complete_request(cmd->rq); |
fe2cb290 | 1809 | } |
b5dd2f60 ML |
1810 | } |
1811 | ||
e03a3d7a | 1812 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1813 | { |
1814 | struct loop_cmd *cmd = | |
e03a3d7a | 1815 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1816 | |
1817 | loop_handle_cmd(cmd); | |
1818 | } | |
1819 | ||
d6296d39 CH |
1820 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1821 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1822 | { |
1823 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1824 | ||
1825 | cmd->rq = rq; | |
3989144f | 1826 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1827 | |
1828 | return 0; | |
1829 | } | |
1830 | ||
f363b089 | 1831 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1832 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 1833 | .init_request = loop_init_request, |
fe2cb290 | 1834 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1835 | }; |
1836 | ||
34dd82af | 1837 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1838 | { |
1839 | struct loop_device *lo; | |
1840 | struct gendisk *disk; | |
34dd82af | 1841 | int err; |
73285082 | 1842 | |
68d740d7 | 1843 | err = -ENOMEM; |
73285082 | 1844 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1845 | if (!lo) |
73285082 | 1846 | goto out; |
34dd82af | 1847 | |
ef7e7c82 MP |
1848 | lo->lo_state = Lo_unbound; |
1849 | ||
c718aa65 | 1850 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1851 | if (i >= 0) { |
c718aa65 TH |
1852 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1853 | if (err == -ENOSPC) | |
34dd82af | 1854 | err = -EEXIST; |
34dd82af | 1855 | } else { |
c718aa65 | 1856 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1857 | } |
1858 | if (err < 0) | |
1859 | goto out_free_dev; | |
c718aa65 | 1860 | i = err; |
73285082 | 1861 | |
183cfb57 | 1862 | err = -ENOMEM; |
b5dd2f60 ML |
1863 | lo->tag_set.ops = &loop_mq_ops; |
1864 | lo->tag_set.nr_hw_queues = 1; | |
1865 | lo->tag_set.queue_depth = 128; | |
1866 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1867 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
1868 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
1869 | lo->tag_set.driver_data = lo; | |
1870 | ||
1871 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1872 | if (err) | |
3ec981e3 | 1873 | goto out_free_idr; |
73285082 | 1874 | |
b5dd2f60 ML |
1875 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
1876 | if (IS_ERR_OR_NULL(lo->lo_queue)) { | |
1877 | err = PTR_ERR(lo->lo_queue); | |
1878 | goto out_cleanup_tags; | |
1879 | } | |
ef7e7c82 MP |
1880 | lo->lo_queue->queuedata = lo; |
1881 | ||
54bb0ade | 1882 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 1883 | |
5b5e20f4 | 1884 | /* |
40326d8a SL |
1885 | * By default, we do buffer IO, so it doesn't make sense to enable |
1886 | * merge because the I/O submitted to backing file is handled page by | |
1887 | * page. For directio mode, merge does help to dispatch bigger request | |
1888 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 ML |
1889 | */ |
1890 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
1891 | ||
7a649737 | 1892 | err = -ENOMEM; |
476a4813 | 1893 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1894 | if (!disk) |
1895 | goto out_free_queue; | |
1896 | ||
e03c8dd1 KS |
1897 | /* |
1898 | * Disable partition scanning by default. The in-kernel partition | |
1899 | * scanning can be requested individually per-device during its | |
1900 | * setup. Userspace can always add and remove partitions from all | |
1901 | * devices. The needed partition minors are allocated from the | |
1902 | * extended minor space, the main loop device numbers will continue | |
1903 | * to match the loop minors, regardless of the number of partitions | |
1904 | * used. | |
1905 | * | |
1906 | * If max_part is given, partition scanning is globally enabled for | |
1907 | * all loop devices. The minors for the main loop devices will be | |
1908 | * multiples of max_part. | |
1909 | * | |
1910 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1911 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1912 | * complicated, are too static, inflexible and may surprise | |
1913 | * userspace tools. Parameters like this in general should be avoided. | |
1914 | */ | |
1915 | if (!part_shift) | |
1916 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1917 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 | 1918 | mutex_init(&lo->lo_ctl_mutex); |
f8933667 | 1919 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 1920 | lo->lo_number = i; |
73285082 KC |
1921 | spin_lock_init(&lo->lo_lock); |
1922 | disk->major = LOOP_MAJOR; | |
476a4813 | 1923 | disk->first_minor = i << part_shift; |
73285082 KC |
1924 | disk->fops = &lo_fops; |
1925 | disk->private_data = lo; | |
1926 | disk->queue = lo->lo_queue; | |
1927 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1928 | add_disk(disk); |
1929 | *l = lo; | |
1930 | return lo->lo_number; | |
73285082 KC |
1931 | |
1932 | out_free_queue: | |
1933 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
1934 | out_cleanup_tags: |
1935 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
1936 | out_free_idr: |
1937 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
1938 | out_free_dev: |
1939 | kfree(lo); | |
1940 | out: | |
34dd82af | 1941 | return err; |
73285082 KC |
1942 | } |
1943 | ||
34dd82af | 1944 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1945 | { |
73285082 | 1946 | blk_cleanup_queue(lo->lo_queue); |
6cd18e71 | 1947 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 1948 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 1949 | put_disk(lo->lo_disk); |
73285082 KC |
1950 | kfree(lo); |
1951 | } | |
1da177e4 | 1952 | |
770fe30a KS |
1953 | static int find_free_cb(int id, void *ptr, void *data) |
1954 | { | |
1955 | struct loop_device *lo = ptr; | |
1956 | struct loop_device **l = data; | |
1957 | ||
1958 | if (lo->lo_state == Lo_unbound) { | |
1959 | *l = lo; | |
1960 | return 1; | |
1961 | } | |
1962 | return 0; | |
1963 | } | |
1964 | ||
34dd82af | 1965 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1966 | { |
1967 | struct loop_device *lo; | |
34dd82af | 1968 | int ret = -ENODEV; |
a47653fc | 1969 | |
770fe30a KS |
1970 | if (i < 0) { |
1971 | int err; | |
1972 | ||
1973 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1974 | if (err == 1) { | |
1975 | *l = lo; | |
1976 | ret = lo->lo_number; | |
1977 | } | |
1978 | goto out; | |
a47653fc KC |
1979 | } |
1980 | ||
770fe30a | 1981 | /* lookup and return a specific i */ |
34dd82af | 1982 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1983 | if (lo) { |
34dd82af KS |
1984 | *l = lo; |
1985 | ret = lo->lo_number; | |
a47653fc | 1986 | } |
770fe30a | 1987 | out: |
34dd82af | 1988 | return ret; |
a47653fc KC |
1989 | } |
1990 | ||
73285082 KC |
1991 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1992 | { | |
705962cc | 1993 | struct loop_device *lo; |
07002e99 | 1994 | struct kobject *kobj; |
34dd82af | 1995 | int err; |
73285082 | 1996 | |
34dd82af KS |
1997 | mutex_lock(&loop_index_mutex); |
1998 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1999 | if (err < 0) | |
2000 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
2001 | if (err < 0) | |
a207f593 | 2002 | kobj = NULL; |
34dd82af KS |
2003 | else |
2004 | kobj = get_disk(lo->lo_disk); | |
2005 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
2006 | |
2007 | *part = 0; | |
07002e99 | 2008 | return kobj; |
73285082 KC |
2009 | } |
2010 | ||
770fe30a KS |
2011 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2012 | unsigned long parm) | |
2013 | { | |
2014 | struct loop_device *lo; | |
2015 | int ret = -ENOSYS; | |
2016 | ||
2017 | mutex_lock(&loop_index_mutex); | |
2018 | switch (cmd) { | |
2019 | case LOOP_CTL_ADD: | |
2020 | ret = loop_lookup(&lo, parm); | |
2021 | if (ret >= 0) { | |
2022 | ret = -EEXIST; | |
2023 | break; | |
2024 | } | |
2025 | ret = loop_add(&lo, parm); | |
2026 | break; | |
2027 | case LOOP_CTL_REMOVE: | |
2028 | ret = loop_lookup(&lo, parm); | |
2029 | if (ret < 0) | |
2030 | break; | |
2031 | mutex_lock(&lo->lo_ctl_mutex); | |
2032 | if (lo->lo_state != Lo_unbound) { | |
2033 | ret = -EBUSY; | |
2034 | mutex_unlock(&lo->lo_ctl_mutex); | |
2035 | break; | |
2036 | } | |
f8933667 | 2037 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a KS |
2038 | ret = -EBUSY; |
2039 | mutex_unlock(&lo->lo_ctl_mutex); | |
2040 | break; | |
2041 | } | |
2042 | lo->lo_disk->private_data = NULL; | |
2043 | mutex_unlock(&lo->lo_ctl_mutex); | |
2044 | idr_remove(&loop_index_idr, lo->lo_number); | |
2045 | loop_remove(lo); | |
2046 | break; | |
2047 | case LOOP_CTL_GET_FREE: | |
2048 | ret = loop_lookup(&lo, -1); | |
2049 | if (ret >= 0) | |
2050 | break; | |
2051 | ret = loop_add(&lo, -1); | |
2052 | } | |
2053 | mutex_unlock(&loop_index_mutex); | |
2054 | ||
2055 | return ret; | |
2056 | } | |
2057 | ||
2058 | static const struct file_operations loop_ctl_fops = { | |
2059 | .open = nonseekable_open, | |
2060 | .unlocked_ioctl = loop_control_ioctl, | |
2061 | .compat_ioctl = loop_control_ioctl, | |
2062 | .owner = THIS_MODULE, | |
2063 | .llseek = noop_llseek, | |
2064 | }; | |
2065 | ||
2066 | static struct miscdevice loop_misc = { | |
2067 | .minor = LOOP_CTRL_MINOR, | |
2068 | .name = "loop-control", | |
2069 | .fops = &loop_ctl_fops, | |
2070 | }; | |
2071 | ||
2072 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2073 | MODULE_ALIAS("devname:loop-control"); | |
2074 | ||
73285082 KC |
2075 | static int __init loop_init(void) |
2076 | { | |
a47653fc KC |
2077 | int i, nr; |
2078 | unsigned long range; | |
34dd82af | 2079 | struct loop_device *lo; |
770fe30a | 2080 | int err; |
a47653fc | 2081 | |
476a4813 | 2082 | part_shift = 0; |
ac04fee0 | 2083 | if (max_part > 0) { |
476a4813 LV |
2084 | part_shift = fls(max_part); |
2085 | ||
ac04fee0 NK |
2086 | /* |
2087 | * Adjust max_part according to part_shift as it is exported | |
2088 | * to user space so that user can decide correct minor number | |
2089 | * if [s]he want to create more devices. | |
2090 | * | |
2091 | * Note that -1 is required because partition 0 is reserved | |
2092 | * for the whole disk. | |
2093 | */ | |
2094 | max_part = (1UL << part_shift) - 1; | |
2095 | } | |
2096 | ||
b1a66504 GC |
2097 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2098 | err = -EINVAL; | |
a8c1d064 | 2099 | goto err_out; |
b1a66504 | 2100 | } |
78f4bb36 | 2101 | |
b1a66504 GC |
2102 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2103 | err = -EINVAL; | |
a8c1d064 | 2104 | goto err_out; |
b1a66504 | 2105 | } |
1da177e4 | 2106 | |
d134b00b KS |
2107 | /* |
2108 | * If max_loop is specified, create that many devices upfront. | |
2109 | * This also becomes a hard limit. If max_loop is not specified, | |
2110 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2111 | * init time. Loop devices can be requested on-demand with the | |
2112 | * /dev/loop-control interface, or be instantiated by accessing | |
2113 | * a 'dead' device node. | |
2114 | */ | |
73285082 | 2115 | if (max_loop) { |
a47653fc | 2116 | nr = max_loop; |
a1c15c59 | 2117 | range = max_loop << part_shift; |
a47653fc | 2118 | } else { |
d134b00b | 2119 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2120 | range = 1UL << MINORBITS; |
a47653fc KC |
2121 | } |
2122 | ||
a8c1d064 AV |
2123 | err = misc_register(&loop_misc); |
2124 | if (err < 0) | |
2125 | goto err_out; | |
2126 | ||
2127 | ||
b1a66504 GC |
2128 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2129 | err = -EIO; | |
2130 | goto misc_out; | |
2131 | } | |
1da177e4 | 2132 | |
a47653fc KC |
2133 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2134 | THIS_MODULE, loop_probe, NULL, NULL); | |
2135 | ||
d134b00b | 2136 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
2137 | mutex_lock(&loop_index_mutex); |
2138 | for (i = 0; i < nr; i++) | |
2139 | loop_add(&lo, i); | |
2140 | mutex_unlock(&loop_index_mutex); | |
2141 | ||
73285082 | 2142 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2143 | return 0; |
b1a66504 GC |
2144 | |
2145 | misc_out: | |
2146 | misc_deregister(&loop_misc); | |
a8c1d064 | 2147 | err_out: |
b1a66504 | 2148 | return err; |
34dd82af | 2149 | } |
a47653fc | 2150 | |
34dd82af KS |
2151 | static int loop_exit_cb(int id, void *ptr, void *data) |
2152 | { | |
2153 | struct loop_device *lo = ptr; | |
a47653fc | 2154 | |
34dd82af KS |
2155 | loop_remove(lo); |
2156 | return 0; | |
1da177e4 LT |
2157 | } |
2158 | ||
73285082 | 2159 | static void __exit loop_exit(void) |
1da177e4 | 2160 | { |
a47653fc | 2161 | unsigned long range; |
1da177e4 | 2162 | |
a1c15c59 | 2163 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2164 | |
34dd82af | 2165 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2166 | idr_destroy(&loop_index_idr); |
73285082 | 2167 | |
a47653fc | 2168 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2169 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2170 | |
2171 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2172 | } |
2173 | ||
2174 | module_init(loop_init); | |
2175 | module_exit(loop_exit); | |
2176 | ||
2177 | #ifndef MODULE | |
2178 | static int __init max_loop_setup(char *str) | |
2179 | { | |
2180 | max_loop = simple_strtol(str, NULL, 0); | |
2181 | return 1; | |
2182 | } | |
2183 | ||
2184 | __setup("max_loop=", max_loop_setup); | |
2185 | #endif |