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> | |
4ee60ec1 | 56 | #include <linux/pagemap.h> |
1da177e4 LT |
57 | #include <linux/file.h> |
58 | #include <linux/stat.h> | |
59 | #include <linux/errno.h> | |
60 | #include <linux/major.h> | |
61 | #include <linux/wait.h> | |
62 | #include <linux/blkdev.h> | |
63 | #include <linux/blkpg.h> | |
64 | #include <linux/init.h> | |
1da177e4 LT |
65 | #include <linux/swap.h> |
66 | #include <linux/slab.h> | |
863d5b82 | 67 | #include <linux/compat.h> |
1da177e4 | 68 | #include <linux/suspend.h> |
83144186 | 69 | #include <linux/freezer.h> |
2a48fc0a | 70 | #include <linux/mutex.h> |
1da177e4 | 71 | #include <linux/writeback.h> |
1da177e4 LT |
72 | #include <linux/completion.h> |
73 | #include <linux/highmem.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> |
d9a08a9e | 79 | #include <linux/ioprio.h> |
db6638d7 | 80 | #include <linux/blk-cgroup.h> |
c74d40e8 | 81 | #include <linux/sched/mm.h> |
d9a08a9e | 82 | |
83a87611 | 83 | #include "loop.h" |
1da177e4 | 84 | |
7c0f6ba6 | 85 | #include <linux/uaccess.h> |
1da177e4 | 86 | |
87579e9b DS |
87 | #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) |
88 | ||
34dd82af | 89 | static DEFINE_IDR(loop_index_idr); |
310ca162 | 90 | static DEFINE_MUTEX(loop_ctl_mutex); |
1da177e4 | 91 | |
476a4813 LV |
92 | static int max_part; |
93 | static int part_shift; | |
94 | ||
1da177e4 LT |
95 | static int transfer_xor(struct loop_device *lo, int cmd, |
96 | struct page *raw_page, unsigned raw_off, | |
97 | struct page *loop_page, unsigned loop_off, | |
98 | int size, sector_t real_block) | |
99 | { | |
cfd8005c CW |
100 | char *raw_buf = kmap_atomic(raw_page) + raw_off; |
101 | char *loop_buf = kmap_atomic(loop_page) + loop_off; | |
1da177e4 LT |
102 | char *in, *out, *key; |
103 | int i, keysize; | |
104 | ||
105 | if (cmd == READ) { | |
106 | in = raw_buf; | |
107 | out = loop_buf; | |
108 | } else { | |
109 | in = loop_buf; | |
110 | out = raw_buf; | |
111 | } | |
112 | ||
113 | key = lo->lo_encrypt_key; | |
114 | keysize = lo->lo_encrypt_key_size; | |
115 | for (i = 0; i < size; i++) | |
116 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
117 | ||
cfd8005c CW |
118 | kunmap_atomic(loop_buf); |
119 | kunmap_atomic(raw_buf); | |
1da177e4 LT |
120 | cond_resched(); |
121 | return 0; | |
122 | } | |
123 | ||
124 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
125 | { | |
126 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
127 | return -EINVAL; | |
128 | return 0; | |
129 | } | |
130 | ||
131 | static struct loop_func_table none_funcs = { | |
132 | .number = LO_CRYPT_NONE, | |
aa4d8616 | 133 | }; |
1da177e4 LT |
134 | |
135 | static struct loop_func_table xor_funcs = { | |
136 | .number = LO_CRYPT_XOR, | |
137 | .transfer = transfer_xor, | |
138 | .init = xor_init | |
aa4d8616 | 139 | }; |
1da177e4 LT |
140 | |
141 | /* xfer_funcs[0] is special - its release function is never called */ | |
142 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
143 | &none_funcs, | |
144 | &xor_funcs | |
145 | }; | |
146 | ||
7035b5df | 147 | static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) |
1da177e4 | 148 | { |
b7a1da69 | 149 | loff_t loopsize; |
1da177e4 LT |
150 | |
151 | /* Compute loopsize in bytes */ | |
b7a1da69 GC |
152 | loopsize = i_size_read(file->f_mapping->host); |
153 | if (offset > 0) | |
154 | loopsize -= offset; | |
155 | /* offset is beyond i_size, weird but possible */ | |
7035b5df DM |
156 | if (loopsize < 0) |
157 | return 0; | |
1da177e4 | 158 | |
7035b5df DM |
159 | if (sizelimit > 0 && sizelimit < loopsize) |
160 | loopsize = sizelimit; | |
1da177e4 LT |
161 | /* |
162 | * Unfortunately, if we want to do I/O on the device, | |
163 | * the number of 512-byte sectors has to fit into a sector_t. | |
164 | */ | |
165 | return loopsize >> 9; | |
166 | } | |
167 | ||
7035b5df DM |
168 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) |
169 | { | |
170 | return get_size(lo->lo_offset, lo->lo_sizelimit, file); | |
171 | } | |
172 | ||
2e5ab5f3 ML |
173 | static void __loop_update_dio(struct loop_device *lo, bool dio) |
174 | { | |
175 | struct file *file = lo->lo_backing_file; | |
176 | struct address_space *mapping = file->f_mapping; | |
177 | struct inode *inode = mapping->host; | |
178 | unsigned short sb_bsize = 0; | |
179 | unsigned dio_align = 0; | |
180 | bool use_dio; | |
181 | ||
182 | if (inode->i_sb->s_bdev) { | |
183 | sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); | |
184 | dio_align = sb_bsize - 1; | |
185 | } | |
186 | ||
187 | /* | |
188 | * We support direct I/O only if lo_offset is aligned with the | |
189 | * logical I/O size of backing device, and the logical block | |
190 | * size of loop is bigger than the backing device's and the loop | |
191 | * needn't transform transfer. | |
192 | * | |
193 | * TODO: the above condition may be loosed in the future, and | |
194 | * direct I/O may be switched runtime at that time because most | |
89d790ab | 195 | * of requests in sane applications should be PAGE_SIZE aligned |
2e5ab5f3 ML |
196 | */ |
197 | if (dio) { | |
198 | if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && | |
199 | !(lo->lo_offset & dio_align) && | |
200 | mapping->a_ops->direct_IO && | |
201 | !lo->transfer) | |
202 | use_dio = true; | |
203 | else | |
204 | use_dio = false; | |
205 | } else { | |
206 | use_dio = false; | |
207 | } | |
208 | ||
209 | if (lo->use_dio == use_dio) | |
210 | return; | |
211 | ||
212 | /* flush dirty pages before changing direct IO */ | |
213 | vfs_fsync(file, 0); | |
214 | ||
215 | /* | |
216 | * The flag of LO_FLAGS_DIRECT_IO is handled similarly with | |
217 | * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup | |
218 | * will get updated by ioctl(LOOP_GET_STATUS) | |
219 | */ | |
0fbcf579 MC |
220 | if (lo->lo_state == Lo_bound) |
221 | blk_mq_freeze_queue(lo->lo_queue); | |
2e5ab5f3 | 222 | lo->use_dio = use_dio; |
40326d8a | 223 | if (use_dio) { |
8b904b5b | 224 | blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 225 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
40326d8a | 226 | } else { |
8b904b5b | 227 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 228 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; |
40326d8a | 229 | } |
0fbcf579 MC |
230 | if (lo->lo_state == Lo_bound) |
231 | blk_mq_unfreeze_queue(lo->lo_queue); | |
2e5ab5f3 ML |
232 | } |
233 | ||
3448914e MC |
234 | /** |
235 | * loop_validate_block_size() - validates the passed in block size | |
236 | * @bsize: size to validate | |
237 | */ | |
1da177e4 | 238 | static int |
3448914e | 239 | loop_validate_block_size(unsigned short bsize) |
1da177e4 | 240 | { |
3448914e MC |
241 | if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize)) |
242 | return -EINVAL; | |
1da177e4 | 243 | |
7035b5df | 244 | return 0; |
1da177e4 LT |
245 | } |
246 | ||
5795b6f5 MC |
247 | /** |
248 | * loop_set_size() - sets device size and notifies userspace | |
249 | * @lo: struct loop_device to set the size for | |
250 | * @size: new size of the loop device | |
251 | * | |
252 | * Callers must validate that the size passed into this function fits into | |
253 | * a sector_t, eg using loop_validate_size() | |
254 | */ | |
255 | static void loop_set_size(struct loop_device *lo, loff_t size) | |
256 | { | |
449f4ec9 | 257 | if (!set_capacity_and_notify(lo->lo_disk, size)) |
3b4f85d0 | 258 | kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); |
5795b6f5 MC |
259 | } |
260 | ||
1da177e4 LT |
261 | static inline int |
262 | lo_do_transfer(struct loop_device *lo, int cmd, | |
263 | struct page *rpage, unsigned roffs, | |
264 | struct page *lpage, unsigned loffs, | |
265 | int size, sector_t rblock) | |
266 | { | |
aa4d8616 CH |
267 | int ret; |
268 | ||
269 | ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
270 | if (likely(!ret)) | |
1da177e4 LT |
271 | return 0; |
272 | ||
aa4d8616 CH |
273 | printk_ratelimited(KERN_ERR |
274 | "loop: Transfer error at byte offset %llu, length %i.\n", | |
275 | (unsigned long long)rblock << 9, size); | |
276 | return ret; | |
1da177e4 LT |
277 | } |
278 | ||
aa4d8616 | 279 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 280 | { |
aa4d8616 | 281 | struct iov_iter i; |
1da177e4 | 282 | ssize_t bw; |
283e7e5d | 283 | |
b6207430 | 284 | iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); |
1da177e4 | 285 | |
03d95eb2 | 286 | file_start_write(file); |
abbb6589 | 287 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 288 | file_end_write(file); |
aa4d8616 CH |
289 | |
290 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 291 | return 0; |
aa4d8616 CH |
292 | |
293 | printk_ratelimited(KERN_ERR | |
294 | "loop: Write error at byte offset %llu, length %i.\n", | |
295 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
296 | if (bw >= 0) |
297 | bw = -EIO; | |
298 | return bw; | |
299 | } | |
300 | ||
aa4d8616 CH |
301 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
302 | loff_t pos) | |
1da177e4 | 303 | { |
aa4d8616 CH |
304 | struct bio_vec bvec; |
305 | struct req_iterator iter; | |
306 | int ret = 0; | |
307 | ||
308 | rq_for_each_segment(bvec, rq, iter) { | |
309 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
310 | if (ret < 0) | |
311 | break; | |
312 | cond_resched(); | |
313 | } | |
314 | ||
315 | return ret; | |
1da177e4 LT |
316 | } |
317 | ||
aa4d8616 | 318 | /* |
456be148 CH |
319 | * This is the slow, transforming version that needs to double buffer the |
320 | * data as it cannot do the transformations in place without having direct | |
321 | * access to the destination pages of the backing file. | |
1da177e4 | 322 | */ |
aa4d8616 CH |
323 | static int lo_write_transfer(struct loop_device *lo, struct request *rq, |
324 | loff_t pos) | |
1da177e4 | 325 | { |
aa4d8616 | 326 | struct bio_vec bvec, b; |
30112013 | 327 | struct req_iterator iter; |
aa4d8616 | 328 | struct page *page; |
7988613b | 329 | int ret = 0; |
1da177e4 | 330 | |
aa4d8616 CH |
331 | page = alloc_page(GFP_NOIO); |
332 | if (unlikely(!page)) | |
333 | return -ENOMEM; | |
456be148 | 334 | |
30112013 | 335 | rq_for_each_segment(bvec, rq, iter) { |
aa4d8616 CH |
336 | ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, |
337 | bvec.bv_offset, bvec.bv_len, pos >> 9); | |
338 | if (unlikely(ret)) | |
339 | break; | |
340 | ||
341 | b.bv_page = page; | |
342 | b.bv_offset = 0; | |
343 | b.bv_len = bvec.bv_len; | |
344 | ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); | |
1da177e4 LT |
345 | if (ret < 0) |
346 | break; | |
1da177e4 | 347 | } |
aa4d8616 CH |
348 | |
349 | __free_page(page); | |
1da177e4 | 350 | return ret; |
1da177e4 LT |
351 | } |
352 | ||
aa4d8616 CH |
353 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
354 | loff_t pos) | |
1da177e4 | 355 | { |
aa4d8616 CH |
356 | struct bio_vec bvec; |
357 | struct req_iterator iter; | |
358 | struct iov_iter i; | |
359 | ssize_t len; | |
1da177e4 | 360 | |
aa4d8616 | 361 | rq_for_each_segment(bvec, rq, iter) { |
b6207430 | 362 | iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); |
18e9710e | 363 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
364 | if (len < 0) |
365 | return len; | |
1da177e4 | 366 | |
aa4d8616 | 367 | flush_dcache_page(bvec.bv_page); |
fd582140 | 368 | |
aa4d8616 CH |
369 | if (len != bvec.bv_len) { |
370 | struct bio *bio; | |
1da177e4 | 371 | |
aa4d8616 CH |
372 | __rq_for_each_bio(bio, rq) |
373 | zero_fill_bio(bio); | |
374 | break; | |
375 | } | |
376 | cond_resched(); | |
377 | } | |
378 | ||
379 | return 0; | |
fd582140 JA |
380 | } |
381 | ||
aa4d8616 CH |
382 | static int lo_read_transfer(struct loop_device *lo, struct request *rq, |
383 | loff_t pos) | |
1da177e4 | 384 | { |
aa4d8616 CH |
385 | struct bio_vec bvec, b; |
386 | struct req_iterator iter; | |
387 | struct iov_iter i; | |
388 | struct page *page; | |
389 | ssize_t len; | |
390 | int ret = 0; | |
1da177e4 | 391 | |
aa4d8616 CH |
392 | page = alloc_page(GFP_NOIO); |
393 | if (unlikely(!page)) | |
394 | return -ENOMEM; | |
fd582140 | 395 | |
aa4d8616 CH |
396 | rq_for_each_segment(bvec, rq, iter) { |
397 | loff_t offset = pos; | |
fd582140 | 398 | |
aa4d8616 CH |
399 | b.bv_page = page; |
400 | b.bv_offset = 0; | |
401 | b.bv_len = bvec.bv_len; | |
fd582140 | 402 | |
b6207430 | 403 | iov_iter_bvec(&i, READ, &b, 1, b.bv_len); |
18e9710e | 404 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
405 | if (len < 0) { |
406 | ret = len; | |
407 | goto out_free_page; | |
408 | } | |
1da177e4 | 409 | |
aa4d8616 CH |
410 | ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, |
411 | bvec.bv_offset, len, offset >> 9); | |
412 | if (ret) | |
413 | goto out_free_page; | |
1da177e4 | 414 | |
aa4d8616 | 415 | flush_dcache_page(bvec.bv_page); |
306df071 | 416 | |
aa4d8616 | 417 | if (len != bvec.bv_len) { |
30112013 ML |
418 | struct bio *bio; |
419 | ||
420 | __rq_for_each_bio(bio, rq) | |
421 | zero_fill_bio(bio); | |
1da177e4 | 422 | break; |
306df071 | 423 | } |
1da177e4 | 424 | } |
aa4d8616 CH |
425 | |
426 | ret = 0; | |
427 | out_free_page: | |
428 | __free_page(page); | |
429 | return ret; | |
1da177e4 LT |
430 | } |
431 | ||
efcfec57 DW |
432 | static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, |
433 | int mode) | |
cf655d95 ML |
434 | { |
435 | /* | |
efcfec57 DW |
436 | * We use fallocate to manipulate the space mappings used by the image |
437 | * a.k.a. discard/zerorange. However we do not support this if | |
438 | * encryption is enabled, because it may give an attacker useful | |
439 | * information. | |
cf655d95 ML |
440 | */ |
441 | struct file *file = lo->lo_backing_file; | |
c52abf56 | 442 | struct request_queue *q = lo->lo_queue; |
cf655d95 ML |
443 | int ret; |
444 | ||
efcfec57 DW |
445 | mode |= FALLOC_FL_KEEP_SIZE; |
446 | ||
c52abf56 | 447 | if (!blk_queue_discard(q)) { |
cf655d95 ML |
448 | ret = -EOPNOTSUPP; |
449 | goto out; | |
450 | } | |
451 | ||
452 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
453 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
454 | ret = -EIO; | |
455 | out: | |
456 | return ret; | |
457 | } | |
458 | ||
459 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
460 | { | |
461 | struct file *file = lo->lo_backing_file; | |
462 | int ret = vfs_fsync(file, 0); | |
463 | if (unlikely(ret && ret != -EINVAL)) | |
464 | ret = -EIO; | |
465 | ||
466 | return ret; | |
467 | } | |
468 | ||
fe2cb290 | 469 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 470 | { |
fe2cb290 | 471 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 472 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 473 | |
f9de14bc JA |
474 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
475 | req_op(rq) != REQ_OP_READ) { | |
476 | if (cmd->ret < 0) | |
8cd55087 | 477 | ret = errno_to_blk_status(cmd->ret); |
f9de14bc | 478 | goto end_io; |
bc07c10a | 479 | } |
fe2cb290 | 480 | |
f9de14bc JA |
481 | /* |
482 | * Short READ - if we got some data, advance our request and | |
483 | * retry it. If we got no data, end the rest with EIO. | |
484 | */ | |
485 | if (cmd->ret) { | |
486 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
487 | cmd->ret = 0; | |
488 | blk_mq_requeue_request(rq, true); | |
489 | } else { | |
490 | if (cmd->use_aio) { | |
491 | struct bio *bio = rq->bio; | |
492 | ||
493 | while (bio) { | |
494 | zero_fill_bio(bio); | |
495 | bio = bio->bi_next; | |
496 | } | |
497 | } | |
498 | ret = BLK_STS_IOERR; | |
499 | end_io: | |
500 | blk_mq_end_request(rq, ret); | |
501 | } | |
bc07c10a ML |
502 | } |
503 | ||
92d77332 SL |
504 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
505 | { | |
1894e916 JA |
506 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
507 | ||
92d77332 SL |
508 | if (!atomic_dec_and_test(&cmd->ref)) |
509 | return; | |
510 | kfree(cmd->bvec); | |
511 | cmd->bvec = NULL; | |
15f73f5b CH |
512 | if (likely(!blk_should_fake_timeout(rq->q))) |
513 | blk_mq_complete_request(rq); | |
92d77332 SL |
514 | } |
515 | ||
bc07c10a ML |
516 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) |
517 | { | |
518 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 519 | |
fe2cb290 | 520 | cmd->ret = ret; |
92d77332 | 521 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
522 | } |
523 | ||
524 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
525 | loff_t pos, bool rw) | |
526 | { | |
527 | struct iov_iter iter; | |
86af5952 | 528 | struct req_iterator rq_iter; |
bc07c10a | 529 | struct bio_vec *bvec; |
1894e916 | 530 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 531 | struct bio *bio = rq->bio; |
bc07c10a | 532 | struct file *file = lo->lo_backing_file; |
86af5952 | 533 | struct bio_vec tmp; |
40326d8a | 534 | unsigned int offset; |
86af5952 | 535 | int nr_bvec = 0; |
bc07c10a ML |
536 | int ret; |
537 | ||
86af5952 ML |
538 | rq_for_each_bvec(tmp, rq, rq_iter) |
539 | nr_bvec++; | |
540 | ||
40326d8a | 541 | if (rq->bio != rq->biotail) { |
40326d8a | 542 | |
86af5952 | 543 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 544 | GFP_NOIO); |
40326d8a SL |
545 | if (!bvec) |
546 | return -EIO; | |
547 | cmd->bvec = bvec; | |
548 | ||
549 | /* | |
550 | * The bios of the request may be started from the middle of | |
551 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 552 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
553 | * API will take care of all details for us. |
554 | */ | |
86af5952 | 555 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
556 | *bvec = tmp; |
557 | bvec++; | |
558 | } | |
559 | bvec = cmd->bvec; | |
560 | offset = 0; | |
561 | } else { | |
562 | /* | |
563 | * Same here, this bio may be started from the middle of the | |
564 | * 'bvec' because of bio splitting, so offset from the bvec | |
565 | * must be passed to iov iterator | |
566 | */ | |
567 | offset = bio->bi_iter.bi_bvec_done; | |
568 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 569 | } |
92d77332 | 570 | atomic_set(&cmd->ref, 2); |
bc07c10a | 571 | |
b6207430 | 572 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 573 | iter.iov_offset = offset; |
bc07c10a ML |
574 | |
575 | cmd->iocb.ki_pos = pos; | |
576 | cmd->iocb.ki_filp = file; | |
577 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
578 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 579 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
bc07c10a ML |
580 | |
581 | if (rw == WRITE) | |
bb7462b6 | 582 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 583 | else |
bb7462b6 | 584 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 585 | |
92d77332 SL |
586 | lo_rw_aio_do_completion(cmd); |
587 | ||
bc07c10a ML |
588 | if (ret != -EIOCBQUEUED) |
589 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
590 | return 0; | |
591 | } | |
592 | ||
c1c87c2b | 593 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
594 | { |
595 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 596 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
597 | |
598 | /* | |
599 | * lo_write_simple and lo_read_simple should have been covered | |
600 | * by io submit style function like lo_rw_aio(), one blocker | |
601 | * is that lo_read_simple() need to call flush_dcache_page after | |
602 | * the page is written from kernel, and it isn't easy to handle | |
603 | * this in io submit style function which submits all segments | |
604 | * of the req at one time. And direct read IO doesn't need to | |
605 | * run flush_dcache_page(). | |
606 | */ | |
c1c87c2b CH |
607 | switch (req_op(rq)) { |
608 | case REQ_OP_FLUSH: | |
609 | return lo_req_flush(lo, rq); | |
19372e27 | 610 | case REQ_OP_WRITE_ZEROES: |
efcfec57 DW |
611 | /* |
612 | * If the caller doesn't want deallocation, call zeroout to | |
613 | * write zeroes the range. Otherwise, punch them out. | |
614 | */ | |
615 | return lo_fallocate(lo, rq, pos, | |
616 | (rq->cmd_flags & REQ_NOUNMAP) ? | |
617 | FALLOC_FL_ZERO_RANGE : | |
618 | FALLOC_FL_PUNCH_HOLE); | |
619 | case REQ_OP_DISCARD: | |
620 | return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); | |
c1c87c2b CH |
621 | case REQ_OP_WRITE: |
622 | if (lo->transfer) | |
623 | return lo_write_transfer(lo, rq, pos); | |
624 | else if (cmd->use_aio) | |
625 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 626 | else |
c1c87c2b CH |
627 | return lo_write_simple(lo, rq, pos); |
628 | case REQ_OP_READ: | |
aa4d8616 | 629 | if (lo->transfer) |
c1c87c2b CH |
630 | return lo_read_transfer(lo, rq, pos); |
631 | else if (cmd->use_aio) | |
632 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 633 | else |
c1c87c2b CH |
634 | return lo_read_simple(lo, rq, pos); |
635 | default: | |
636 | WARN_ON_ONCE(1); | |
637 | return -EIO; | |
aa4d8616 | 638 | } |
1da177e4 LT |
639 | } |
640 | ||
2e5ab5f3 ML |
641 | static inline void loop_update_dio(struct loop_device *lo) |
642 | { | |
efbe3c24 IW |
643 | __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | |
644 | lo->use_dio); | |
2e5ab5f3 ML |
645 | } |
646 | ||
0384264e | 647 | static void loop_reread_partitions(struct loop_device *lo) |
06f0e9e6 ML |
648 | { |
649 | int rc; | |
650 | ||
0384264e CH |
651 | mutex_lock(&lo->lo_disk->open_mutex); |
652 | rc = bdev_disk_changed(lo->lo_disk, false); | |
653 | mutex_unlock(&lo->lo_disk->open_mutex); | |
06f0e9e6 ML |
654 | if (rc) |
655 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
656 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
657 | } | |
658 | ||
d2ac838e TT |
659 | static inline int is_loop_device(struct file *file) |
660 | { | |
661 | struct inode *i = file->f_mapping->host; | |
662 | ||
6f24784f | 663 | return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; |
d2ac838e TT |
664 | } |
665 | ||
666 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
667 | { | |
668 | struct inode *inode = file->f_mapping->host; | |
669 | struct file *f = file; | |
670 | ||
671 | /* Avoid recursion */ | |
672 | while (is_loop_device(f)) { | |
673 | struct loop_device *l; | |
674 | ||
4e7b5671 | 675 | if (f->f_mapping->host->i_rdev == bdev->bd_dev) |
d2ac838e TT |
676 | return -EBADF; |
677 | ||
4e7b5671 | 678 | l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; |
f7c8a412 | 679 | if (l->lo_state != Lo_bound) { |
d2ac838e TT |
680 | return -EINVAL; |
681 | } | |
682 | f = l->lo_backing_file; | |
683 | } | |
684 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
685 | return -EINVAL; | |
686 | return 0; | |
687 | } | |
688 | ||
1da177e4 LT |
689 | /* |
690 | * loop_change_fd switched the backing store of a loopback device to | |
691 | * a new file. This is useful for operating system installers to free up | |
692 | * the original file and in High Availability environments to switch to | |
693 | * an alternative location for the content in case of server meltdown. | |
694 | * This can only work if the loop device is used read-only, and if the | |
695 | * new backing store is the same size and type as the old backing store. | |
696 | */ | |
bb214884 AV |
697 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
698 | unsigned int arg) | |
1da177e4 | 699 | { |
1dded9ac | 700 | struct file *file = NULL, *old_file; |
1da177e4 | 701 | int error; |
85b0a54a | 702 | bool partscan; |
1da177e4 | 703 | |
6cc8e743 | 704 | error = mutex_lock_killable(&lo->lo_mutex); |
c3710770 JK |
705 | if (error) |
706 | return error; | |
1da177e4 LT |
707 | error = -ENXIO; |
708 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 709 | goto out_err; |
1da177e4 LT |
710 | |
711 | /* the loop device has to be read-only */ | |
712 | error = -EINVAL; | |
713 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 714 | goto out_err; |
1da177e4 LT |
715 | |
716 | error = -EBADF; | |
717 | file = fget(arg); | |
718 | if (!file) | |
1dded9ac | 719 | goto out_err; |
1da177e4 | 720 | |
d2ac838e TT |
721 | error = loop_validate_file(file, bdev); |
722 | if (error) | |
1dded9ac | 723 | goto out_err; |
d2ac838e | 724 | |
1da177e4 LT |
725 | old_file = lo->lo_backing_file; |
726 | ||
727 | error = -EINVAL; | |
728 | ||
1da177e4 LT |
729 | /* size of the new backing store needs to be the same */ |
730 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 731 | goto out_err; |
1da177e4 LT |
732 | |
733 | /* and ... switch */ | |
43cade80 OS |
734 | blk_mq_freeze_queue(lo->lo_queue); |
735 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
736 | lo->lo_backing_file = file; | |
737 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
738 | mapping_set_gfp_mask(file->f_mapping, | |
739 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
740 | loop_update_dio(lo); | |
741 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 742 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
6cc8e743 | 743 | mutex_unlock(&lo->lo_mutex); |
1dded9ac | 744 | /* |
6cc8e743 | 745 | * We must drop file reference outside of lo_mutex as dropping |
a8698707 | 746 | * the file ref can take open_mutex which creates circular locking |
1dded9ac JK |
747 | * dependency. |
748 | */ | |
749 | fput(old_file); | |
85b0a54a | 750 | if (partscan) |
0384264e | 751 | loop_reread_partitions(lo); |
1da177e4 LT |
752 | return 0; |
753 | ||
1dded9ac | 754 | out_err: |
6cc8e743 | 755 | mutex_unlock(&lo->lo_mutex); |
1dded9ac JK |
756 | if (file) |
757 | fput(file); | |
1da177e4 LT |
758 | return error; |
759 | } | |
760 | ||
ee862730 MB |
761 | /* loop sysfs attributes */ |
762 | ||
763 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
764 | ssize_t (*callback)(struct loop_device *, char *)) | |
765 | { | |
34dd82af KS |
766 | struct gendisk *disk = dev_to_disk(dev); |
767 | struct loop_device *lo = disk->private_data; | |
ee862730 | 768 | |
34dd82af | 769 | return callback(lo, page); |
ee862730 MB |
770 | } |
771 | ||
772 | #define LOOP_ATTR_RO(_name) \ | |
773 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
774 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
775 | struct device_attribute *attr, char *b) \ | |
776 | { \ | |
777 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
778 | } \ | |
779 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 780 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
781 | |
782 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
783 | { | |
784 | ssize_t ret; | |
785 | char *p = NULL; | |
786 | ||
05eb0f25 | 787 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 788 | if (lo->lo_backing_file) |
9bf39ab2 | 789 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 790 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
791 | |
792 | if (IS_ERR_OR_NULL(p)) | |
793 | ret = PTR_ERR(p); | |
794 | else { | |
795 | ret = strlen(p); | |
796 | memmove(buf, p, ret); | |
797 | buf[ret++] = '\n'; | |
798 | buf[ret] = 0; | |
799 | } | |
800 | ||
801 | return ret; | |
802 | } | |
803 | ||
804 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
805 | { | |
806 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
807 | } | |
808 | ||
809 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
810 | { | |
811 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
812 | } | |
813 | ||
814 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
815 | { | |
816 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
817 | ||
818 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
819 | } | |
820 | ||
e03c8dd1 KS |
821 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
822 | { | |
823 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
824 | ||
825 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
826 | } | |
827 | ||
2e5ab5f3 ML |
828 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
829 | { | |
830 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
831 | ||
832 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
833 | } | |
834 | ||
ee862730 MB |
835 | LOOP_ATTR_RO(backing_file); |
836 | LOOP_ATTR_RO(offset); | |
837 | LOOP_ATTR_RO(sizelimit); | |
838 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 839 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 840 | LOOP_ATTR_RO(dio); |
ee862730 MB |
841 | |
842 | static struct attribute *loop_attrs[] = { | |
843 | &loop_attr_backing_file.attr, | |
844 | &loop_attr_offset.attr, | |
845 | &loop_attr_sizelimit.attr, | |
846 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 847 | &loop_attr_partscan.attr, |
2e5ab5f3 | 848 | &loop_attr_dio.attr, |
ee862730 MB |
849 | NULL, |
850 | }; | |
851 | ||
852 | static struct attribute_group loop_attribute_group = { | |
853 | .name = "loop", | |
854 | .attrs= loop_attrs, | |
855 | }; | |
856 | ||
d3349b6b | 857 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 858 | { |
d3349b6b TH |
859 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
860 | &loop_attribute_group); | |
ee862730 MB |
861 | } |
862 | ||
863 | static void loop_sysfs_exit(struct loop_device *lo) | |
864 | { | |
d3349b6b TH |
865 | if (lo->sysfs_inited) |
866 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
867 | &loop_attribute_group); | |
ee862730 MB |
868 | } |
869 | ||
dfaa2ef6 LC |
870 | static void loop_config_discard(struct loop_device *lo) |
871 | { | |
872 | struct file *file = lo->lo_backing_file; | |
873 | struct inode *inode = file->f_mapping->host; | |
874 | struct request_queue *q = lo->lo_queue; | |
bcb21c8c | 875 | u32 granularity, max_discard_sectors; |
dfaa2ef6 | 876 | |
c52abf56 EG |
877 | /* |
878 | * If the backing device is a block device, mirror its zeroing | |
879 | * capability. Set the discard sectors to the block device's zeroing | |
880 | * capabilities because loop discards result in blkdev_issue_zeroout(), | |
881 | * not blkdev_issue_discard(). This maintains consistent behavior with | |
882 | * file-backed loop devices: discarded regions read back as zero. | |
883 | */ | |
884 | if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { | |
4e7b5671 | 885 | struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); |
c52abf56 | 886 | |
bcb21c8c ML |
887 | max_discard_sectors = backingq->limits.max_write_zeroes_sectors; |
888 | granularity = backingq->limits.discard_granularity ?: | |
889 | queue_physical_block_size(backingq); | |
c52abf56 | 890 | |
dfaa2ef6 LC |
891 | /* |
892 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 893 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
894 | * encryption is enabled, because it may give an attacker |
895 | * useful information. | |
896 | */ | |
c52abf56 | 897 | } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { |
bcb21c8c ML |
898 | max_discard_sectors = 0; |
899 | granularity = 0; | |
dfaa2ef6 | 900 | |
c52abf56 | 901 | } else { |
bcb21c8c ML |
902 | max_discard_sectors = UINT_MAX >> 9; |
903 | granularity = inode->i_sb->s_blocksize; | |
c52abf56 EG |
904 | } |
905 | ||
bcb21c8c ML |
906 | if (max_discard_sectors) { |
907 | q->limits.discard_granularity = granularity; | |
908 | blk_queue_max_discard_sectors(q, max_discard_sectors); | |
909 | blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); | |
c52abf56 | 910 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); |
bcb21c8c ML |
911 | } else { |
912 | q->limits.discard_granularity = 0; | |
913 | blk_queue_max_discard_sectors(q, 0); | |
914 | blk_queue_max_write_zeroes_sectors(q, 0); | |
c52abf56 | 915 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); |
bcb21c8c ML |
916 | } |
917 | q->limits.discard_alignment = 0; | |
dfaa2ef6 LC |
918 | } |
919 | ||
87579e9b DS |
920 | struct loop_worker { |
921 | struct rb_node rb_node; | |
922 | struct work_struct work; | |
923 | struct list_head cmd_list; | |
924 | struct list_head idle_list; | |
925 | struct loop_device *lo; | |
c74d40e8 | 926 | struct cgroup_subsys_state *blkcg_css; |
87579e9b DS |
927 | unsigned long last_ran_at; |
928 | }; | |
929 | ||
930 | static void loop_workfn(struct work_struct *work); | |
931 | static void loop_rootcg_workfn(struct work_struct *work); | |
932 | static void loop_free_idle_workers(struct timer_list *timer); | |
933 | ||
934 | #ifdef CONFIG_BLK_CGROUP | |
935 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
e03a3d7a | 936 | { |
87579e9b | 937 | return !css || css == blkcg_root_css; |
e03a3d7a | 938 | } |
87579e9b DS |
939 | #else |
940 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
b2ee7d46 | 941 | { |
87579e9b | 942 | return !css; |
b2ee7d46 | 943 | } |
87579e9b | 944 | #endif |
b2ee7d46 | 945 | |
87579e9b | 946 | static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) |
e03a3d7a | 947 | { |
87579e9b DS |
948 | struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL; |
949 | struct loop_worker *cur_worker, *worker = NULL; | |
950 | struct work_struct *work; | |
951 | struct list_head *cmd_list; | |
952 | ||
953 | spin_lock_irq(&lo->lo_work_lock); | |
954 | ||
c74d40e8 | 955 | if (queue_on_root_worker(cmd->blkcg_css)) |
87579e9b DS |
956 | goto queue_work; |
957 | ||
958 | node = &lo->worker_tree.rb_node; | |
959 | ||
960 | while (*node) { | |
961 | parent = *node; | |
962 | cur_worker = container_of(*node, struct loop_worker, rb_node); | |
c74d40e8 | 963 | if (cur_worker->blkcg_css == cmd->blkcg_css) { |
87579e9b DS |
964 | worker = cur_worker; |
965 | break; | |
c74d40e8 | 966 | } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { |
87579e9b DS |
967 | node = &(*node)->rb_left; |
968 | } else { | |
969 | node = &(*node)->rb_right; | |
970 | } | |
971 | } | |
972 | if (worker) | |
973 | goto queue_work; | |
974 | ||
975 | worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); | |
976 | /* | |
977 | * In the event we cannot allocate a worker, just queue on the | |
c74d40e8 | 978 | * rootcg worker and issue the I/O as the rootcg |
87579e9b | 979 | */ |
c74d40e8 DS |
980 | if (!worker) { |
981 | cmd->blkcg_css = NULL; | |
982 | if (cmd->memcg_css) | |
983 | css_put(cmd->memcg_css); | |
984 | cmd->memcg_css = NULL; | |
87579e9b | 985 | goto queue_work; |
c74d40e8 | 986 | } |
87579e9b | 987 | |
c74d40e8 DS |
988 | worker->blkcg_css = cmd->blkcg_css; |
989 | css_get(worker->blkcg_css); | |
87579e9b DS |
990 | INIT_WORK(&worker->work, loop_workfn); |
991 | INIT_LIST_HEAD(&worker->cmd_list); | |
992 | INIT_LIST_HEAD(&worker->idle_list); | |
993 | worker->lo = lo; | |
994 | rb_link_node(&worker->rb_node, parent, node); | |
995 | rb_insert_color(&worker->rb_node, &lo->worker_tree); | |
996 | queue_work: | |
997 | if (worker) { | |
998 | /* | |
999 | * We need to remove from the idle list here while | |
1000 | * holding the lock so that the idle timer doesn't | |
1001 | * free the worker | |
1002 | */ | |
1003 | if (!list_empty(&worker->idle_list)) | |
1004 | list_del_init(&worker->idle_list); | |
1005 | work = &worker->work; | |
1006 | cmd_list = &worker->cmd_list; | |
1007 | } else { | |
1008 | work = &lo->rootcg_work; | |
1009 | cmd_list = &lo->rootcg_cmd_list; | |
1010 | } | |
1011 | list_add_tail(&cmd->list_entry, cmd_list); | |
1012 | queue_work(lo->workqueue, work); | |
1013 | spin_unlock_irq(&lo->lo_work_lock); | |
e03a3d7a ML |
1014 | } |
1015 | ||
56a85fd8 HH |
1016 | static void loop_update_rotational(struct loop_device *lo) |
1017 | { | |
1018 | struct file *file = lo->lo_backing_file; | |
1019 | struct inode *file_inode = file->f_mapping->host; | |
1020 | struct block_device *file_bdev = file_inode->i_sb->s_bdev; | |
1021 | struct request_queue *q = lo->lo_queue; | |
1022 | bool nonrot = true; | |
1023 | ||
1024 | /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ | |
1025 | if (file_bdev) | |
1026 | nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); | |
1027 | ||
1028 | if (nonrot) | |
1029 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q); | |
1030 | else | |
1031 | blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); | |
1032 | } | |
1033 | ||
62ab466c MC |
1034 | static int |
1035 | loop_release_xfer(struct loop_device *lo) | |
1036 | { | |
1037 | int err = 0; | |
1038 | struct loop_func_table *xfer = lo->lo_encryption; | |
1039 | ||
1040 | if (xfer) { | |
1041 | if (xfer->release) | |
1042 | err = xfer->release(lo); | |
1043 | lo->transfer = NULL; | |
1044 | lo->lo_encryption = NULL; | |
1045 | module_put(xfer->owner); | |
1046 | } | |
1047 | return err; | |
1048 | } | |
1049 | ||
1050 | static int | |
1051 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1052 | const struct loop_info64 *i) | |
1053 | { | |
1054 | int err = 0; | |
1055 | ||
1056 | if (xfer) { | |
1057 | struct module *owner = xfer->owner; | |
1058 | ||
1059 | if (!try_module_get(owner)) | |
1060 | return -EINVAL; | |
1061 | if (xfer->init) | |
1062 | err = xfer->init(lo, i); | |
1063 | if (err) | |
1064 | module_put(owner); | |
1065 | else | |
1066 | lo->lo_encryption = xfer; | |
1067 | } | |
1068 | return err; | |
1069 | } | |
1070 | ||
1071 | /** | |
1072 | * loop_set_status_from_info - configure device from loop_info | |
1073 | * @lo: struct loop_device to configure | |
1074 | * @info: struct loop_info64 to configure the device with | |
1075 | * | |
1076 | * Configures the loop device parameters according to the passed | |
1077 | * in loop_info64 configuration. | |
1078 | */ | |
1079 | static int | |
1080 | loop_set_status_from_info(struct loop_device *lo, | |
1081 | const struct loop_info64 *info) | |
1082 | { | |
1083 | int err; | |
1084 | struct loop_func_table *xfer; | |
1085 | kuid_t uid = current_uid(); | |
1086 | ||
1087 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1088 | return -EINVAL; | |
1089 | ||
1090 | err = loop_release_xfer(lo); | |
1091 | if (err) | |
1092 | return err; | |
1093 | ||
1094 | if (info->lo_encrypt_type) { | |
1095 | unsigned int type = info->lo_encrypt_type; | |
1096 | ||
1097 | if (type >= MAX_LO_CRYPT) | |
1098 | return -EINVAL; | |
1099 | xfer = xfer_funcs[type]; | |
1100 | if (xfer == NULL) | |
1101 | return -EINVAL; | |
1102 | } else | |
1103 | xfer = NULL; | |
1104 | ||
1105 | err = loop_init_xfer(lo, xfer, info); | |
1106 | if (err) | |
1107 | return err; | |
1108 | ||
1109 | lo->lo_offset = info->lo_offset; | |
1110 | lo->lo_sizelimit = info->lo_sizelimit; | |
1111 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1112 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1113 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1114 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1115 | ||
1116 | if (!xfer) | |
1117 | xfer = &none_funcs; | |
1118 | lo->transfer = xfer->transfer; | |
1119 | lo->ioctl = xfer->ioctl; | |
1120 | ||
faf1d254 | 1121 | lo->lo_flags = info->lo_flags; |
62ab466c MC |
1122 | |
1123 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1124 | lo->lo_init[0] = info->lo_init[0]; | |
1125 | lo->lo_init[1] = info->lo_init[1]; | |
1126 | if (info->lo_encrypt_key_size) { | |
1127 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1128 | info->lo_encrypt_key_size); | |
1129 | lo->lo_key_owner = uid; | |
1130 | } | |
1131 | ||
1132 | return 0; | |
1133 | } | |
1134 | ||
3448914e MC |
1135 | static int loop_configure(struct loop_device *lo, fmode_t mode, |
1136 | struct block_device *bdev, | |
1137 | const struct loop_config *config) | |
1da177e4 | 1138 | { |
d2ac838e | 1139 | struct file *file; |
1da177e4 LT |
1140 | struct inode *inode; |
1141 | struct address_space *mapping; | |
1da177e4 LT |
1142 | int error; |
1143 | loff_t size; | |
85b0a54a | 1144 | bool partscan; |
3448914e | 1145 | unsigned short bsize; |
1da177e4 LT |
1146 | |
1147 | /* This is safe, since we have a reference from open(). */ | |
1148 | __module_get(THIS_MODULE); | |
1149 | ||
1150 | error = -EBADF; | |
3448914e | 1151 | file = fget(config->fd); |
1da177e4 LT |
1152 | if (!file) |
1153 | goto out; | |
1154 | ||
33ec3e53 JK |
1155 | /* |
1156 | * If we don't hold exclusive handle for the device, upgrade to it | |
1157 | * here to avoid changing device under exclusive owner. | |
1158 | */ | |
1159 | if (!(mode & FMODE_EXCL)) { | |
37c3fc9a | 1160 | error = bd_prepare_to_claim(bdev, loop_configure); |
ecbe6bc0 | 1161 | if (error) |
33ec3e53 JK |
1162 | goto out_putf; |
1163 | } | |
1164 | ||
6cc8e743 | 1165 | error = mutex_lock_killable(&lo->lo_mutex); |
757ecf40 | 1166 | if (error) |
33ec3e53 | 1167 | goto out_bdev; |
757ecf40 | 1168 | |
1da177e4 LT |
1169 | error = -EBUSY; |
1170 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 1171 | goto out_unlock; |
1da177e4 | 1172 | |
d2ac838e TT |
1173 | error = loop_validate_file(file, bdev); |
1174 | if (error) | |
757ecf40 | 1175 | goto out_unlock; |
1da177e4 LT |
1176 | |
1177 | mapping = file->f_mapping; | |
1178 | inode = mapping->host; | |
1179 | ||
3448914e MC |
1180 | if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { |
1181 | error = -EINVAL; | |
1182 | goto out_unlock; | |
1183 | } | |
1184 | ||
1185 | if (config->block_size) { | |
1186 | error = loop_validate_block_size(config->block_size); | |
1187 | if (error) | |
1188 | goto out_unlock; | |
1189 | } | |
1190 | ||
1191 | error = loop_set_status_from_info(lo, &config->info); | |
1192 | if (error) | |
1193 | goto out_unlock; | |
1194 | ||
456be148 | 1195 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 1196 | !file->f_op->write_iter) |
3448914e | 1197 | lo->lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 1198 | |
87579e9b DS |
1199 | lo->workqueue = alloc_workqueue("loop%d", |
1200 | WQ_UNBOUND | WQ_FREEZABLE, | |
1201 | 0, | |
1202 | lo->lo_number); | |
1203 | if (!lo->workqueue) { | |
1204 | error = -ENOMEM; | |
757ecf40 | 1205 | goto out_unlock; |
87579e9b | 1206 | } |
1da177e4 | 1207 | |
7a2f0ce1 | 1208 | set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); |
1da177e4 | 1209 | |
87579e9b DS |
1210 | INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); |
1211 | INIT_LIST_HEAD(&lo->rootcg_cmd_list); | |
1212 | INIT_LIST_HEAD(&lo->idle_worker_list); | |
1213 | lo->worker_tree = RB_ROOT; | |
1214 | timer_setup(&lo->timer, loop_free_idle_workers, | |
1215 | TIMER_DEFERRABLE); | |
3448914e | 1216 | lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; |
1da177e4 | 1217 | lo->lo_device = bdev; |
1da177e4 | 1218 | lo->lo_backing_file = file; |
1da177e4 LT |
1219 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
1220 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1221 | ||
3448914e | 1222 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 1223 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 1224 | |
3448914e MC |
1225 | if (config->block_size) |
1226 | bsize = config->block_size; | |
96ed320d | 1227 | else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) |
85560117 | 1228 | /* In case of direct I/O, match underlying block size */ |
3448914e MC |
1229 | bsize = bdev_logical_block_size(inode->i_sb->s_bdev); |
1230 | else | |
1231 | bsize = 512; | |
85560117 | 1232 | |
3448914e MC |
1233 | blk_queue_logical_block_size(lo->lo_queue, bsize); |
1234 | blk_queue_physical_block_size(lo->lo_queue, bsize); | |
1235 | blk_queue_io_min(lo->lo_queue, bsize); | |
85560117 | 1236 | |
2b9ac22b | 1237 | loop_config_discard(lo); |
56a85fd8 | 1238 | loop_update_rotational(lo); |
2e5ab5f3 | 1239 | loop_update_dio(lo); |
ee862730 | 1240 | loop_sysfs_init(lo); |
79e5dc59 MC |
1241 | |
1242 | size = get_loop_size(lo, file); | |
5795b6f5 | 1243 | loop_set_size(lo, size); |
1da177e4 | 1244 | |
6c997918 | 1245 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1246 | if (part_shift) |
1247 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 1248 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
fe6a8fc5 LP |
1249 | if (partscan) |
1250 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
c1681bf8 AP |
1251 | |
1252 | /* Grab the block_device to prevent its destruction after we | |
a2505b79 | 1253 | * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). |
c1681bf8 AP |
1254 | */ |
1255 | bdgrab(bdev); | |
6cc8e743 | 1256 | mutex_unlock(&lo->lo_mutex); |
85b0a54a | 1257 | if (partscan) |
0384264e | 1258 | loop_reread_partitions(lo); |
37c3fc9a CH |
1259 | if (!(mode & FMODE_EXCL)) |
1260 | bd_abort_claiming(bdev, loop_configure); | |
1da177e4 LT |
1261 | return 0; |
1262 | ||
757ecf40 | 1263 | out_unlock: |
6cc8e743 | 1264 | mutex_unlock(&lo->lo_mutex); |
33ec3e53 | 1265 | out_bdev: |
37c3fc9a CH |
1266 | if (!(mode & FMODE_EXCL)) |
1267 | bd_abort_claiming(bdev, loop_configure); | |
757ecf40 | 1268 | out_putf: |
1da177e4 | 1269 | fput(file); |
757ecf40 | 1270 | out: |
1da177e4 LT |
1271 | /* This is safe: open() is still holding a reference. */ |
1272 | module_put(THIS_MODULE); | |
1273 | return error; | |
1274 | } | |
1275 | ||
0da03cab | 1276 | static int __loop_clr_fd(struct loop_device *lo, bool release) |
1da177e4 | 1277 | { |
7ccd0791 | 1278 | struct file *filp = NULL; |
b4e3ca1a | 1279 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1280 | struct block_device *bdev = lo->lo_device; |
7ccd0791 | 1281 | int err = 0; |
0da03cab JK |
1282 | bool partscan = false; |
1283 | int lo_number; | |
87579e9b | 1284 | struct loop_worker *pos, *worker; |
1da177e4 | 1285 | |
6cc8e743 | 1286 | mutex_lock(&lo->lo_mutex); |
7ccd0791 JK |
1287 | if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { |
1288 | err = -ENXIO; | |
1289 | goto out_unlock; | |
1290 | } | |
1da177e4 | 1291 | |
7ccd0791 JK |
1292 | filp = lo->lo_backing_file; |
1293 | if (filp == NULL) { | |
1294 | err = -EINVAL; | |
1295 | goto out_unlock; | |
1296 | } | |
1da177e4 | 1297 | |
4ceddce5 MFO |
1298 | if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) |
1299 | blk_queue_write_cache(lo->lo_queue, false, false); | |
1300 | ||
f8933667 ML |
1301 | /* freeze request queue during the transition */ |
1302 | blk_mq_freeze_queue(lo->lo_queue); | |
1303 | ||
87579e9b DS |
1304 | destroy_workqueue(lo->workqueue); |
1305 | spin_lock_irq(&lo->lo_work_lock); | |
1306 | list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, | |
1307 | idle_list) { | |
1308 | list_del(&worker->idle_list); | |
1309 | rb_erase(&worker->rb_node, &lo->worker_tree); | |
c74d40e8 | 1310 | css_put(worker->blkcg_css); |
87579e9b DS |
1311 | kfree(worker); |
1312 | } | |
1313 | spin_unlock_irq(&lo->lo_work_lock); | |
1314 | del_timer_sync(&lo->timer); | |
1315 | ||
1da177e4 | 1316 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 1317 | lo->lo_backing_file = NULL; |
05eb0f25 | 1318 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1319 | |
1320 | loop_release_xfer(lo); | |
1321 | lo->transfer = NULL; | |
1322 | lo->ioctl = NULL; | |
1323 | lo->lo_device = NULL; | |
1324 | lo->lo_encryption = NULL; | |
1325 | lo->lo_offset = 0; | |
1326 | lo->lo_sizelimit = 0; | |
1327 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1328 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1329 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1330 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1331 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1332 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1333 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1334 | if (bdev) { |
1335 | bdput(bdev); | |
bb214884 | 1336 | invalidate_bdev(bdev); |
eedffa28 | 1337 | bdev->bd_inode->i_mapping->wb_err = 0; |
c1681bf8 | 1338 | } |
73285082 | 1339 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1340 | loop_sysfs_exit(lo); |
c3473c63 | 1341 | if (bdev) { |
c3473c63 DZ |
1342 | /* let user-space know about this change */ |
1343 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1344 | } | |
1da177e4 | 1345 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1da177e4 LT |
1346 | /* This is safe: open() is still holding a reference. */ |
1347 | module_put(THIS_MODULE); | |
f8933667 ML |
1348 | blk_mq_unfreeze_queue(lo->lo_queue); |
1349 | ||
0da03cab JK |
1350 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; |
1351 | lo_number = lo->lo_number; | |
0da03cab | 1352 | out_unlock: |
6cc8e743 | 1353 | mutex_unlock(&lo->lo_mutex); |
0da03cab | 1354 | if (partscan) { |
d57f3374 | 1355 | /* |
a8698707 | 1356 | * open_mutex has been held already in release path, so don't |
d57f3374 JK |
1357 | * acquire it if this function is called in such case. |
1358 | * | |
1359 | * If the reread partition isn't from release path, lo_refcnt | |
1360 | * must be at least one and it can only become zero when the | |
1361 | * current holder is released. | |
1362 | */ | |
f0b870df | 1363 | if (!release) |
0384264e CH |
1364 | mutex_lock(&lo->lo_disk->open_mutex); |
1365 | err = bdev_disk_changed(lo->lo_disk, false); | |
f0b870df | 1366 | if (!release) |
0384264e | 1367 | mutex_unlock(&lo->lo_disk->open_mutex); |
40853d6f DZ |
1368 | if (err) |
1369 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
1370 | __func__, lo_number, err); | |
d57f3374 JK |
1371 | /* Device is gone, no point in returning error */ |
1372 | err = 0; | |
1373 | } | |
758a58d0 DZ |
1374 | |
1375 | /* | |
1376 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1377 | * finished. | |
1378 | * | |
1379 | * There cannot be anybody else entering __loop_clr_fd() as | |
1380 | * lo->lo_backing_file is already cleared and Lo_rundown state | |
1381 | * protects us from all the other places trying to change the 'lo' | |
1382 | * device. | |
1383 | */ | |
6cc8e743 | 1384 | mutex_lock(&lo->lo_mutex); |
758a58d0 DZ |
1385 | lo->lo_flags = 0; |
1386 | if (!part_shift) | |
1387 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1388 | lo->lo_state = Lo_unbound; | |
6cc8e743 | 1389 | mutex_unlock(&lo->lo_mutex); |
758a58d0 | 1390 | |
f028f3b2 | 1391 | /* |
6cc8e743 PT |
1392 | * Need not hold lo_mutex to fput backing file. Calling fput holding |
1393 | * lo_mutex triggers a circular lock dependency possibility warning as | |
a8698707 | 1394 | * fput can take open_mutex which is usually taken before lo_mutex. |
f028f3b2 | 1395 | */ |
7ccd0791 JK |
1396 | if (filp) |
1397 | fput(filp); | |
1398 | return err; | |
1da177e4 LT |
1399 | } |
1400 | ||
a2505b79 JK |
1401 | static int loop_clr_fd(struct loop_device *lo) |
1402 | { | |
7ccd0791 JK |
1403 | int err; |
1404 | ||
6cc8e743 | 1405 | err = mutex_lock_killable(&lo->lo_mutex); |
7ccd0791 JK |
1406 | if (err) |
1407 | return err; | |
1408 | if (lo->lo_state != Lo_bound) { | |
6cc8e743 | 1409 | mutex_unlock(&lo->lo_mutex); |
a2505b79 | 1410 | return -ENXIO; |
7ccd0791 | 1411 | } |
a2505b79 JK |
1412 | /* |
1413 | * If we've explicitly asked to tear down the loop device, | |
1414 | * and it has an elevated reference count, set it for auto-teardown when | |
1415 | * the last reference goes away. This stops $!~#$@ udev from | |
1416 | * preventing teardown because it decided that it needs to run blkid on | |
1417 | * the loopback device whenever they appear. xfstests is notorious for | |
1418 | * failing tests because blkid via udev races with a losetup | |
1419 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1420 | * command to fail with EBUSY. | |
1421 | */ | |
1422 | if (atomic_read(&lo->lo_refcnt) > 1) { | |
1423 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
6cc8e743 | 1424 | mutex_unlock(&lo->lo_mutex); |
a2505b79 JK |
1425 | return 0; |
1426 | } | |
1427 | lo->lo_state = Lo_rundown; | |
6cc8e743 | 1428 | mutex_unlock(&lo->lo_mutex); |
a2505b79 | 1429 | |
0da03cab | 1430 | return __loop_clr_fd(lo, false); |
a2505b79 JK |
1431 | } |
1432 | ||
1da177e4 LT |
1433 | static int |
1434 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1435 | { | |
1436 | int err; | |
85b0a54a | 1437 | struct block_device *bdev; |
0c3796c2 | 1438 | kuid_t uid = current_uid(); |
faf1d254 | 1439 | int prev_lo_flags; |
85b0a54a | 1440 | bool partscan = false; |
0c3796c2 | 1441 | bool size_changed = false; |
1da177e4 | 1442 | |
6cc8e743 | 1443 | err = mutex_lock_killable(&lo->lo_mutex); |
550df5fd JK |
1444 | if (err) |
1445 | return err; | |
b0fafa81 | 1446 | if (lo->lo_encrypt_key_size && |
e4849737 | 1447 | !uid_eq(lo->lo_key_owner, uid) && |
550df5fd JK |
1448 | !capable(CAP_SYS_ADMIN)) { |
1449 | err = -EPERM; | |
1450 | goto out_unlock; | |
1451 | } | |
1452 | if (lo->lo_state != Lo_bound) { | |
1453 | err = -ENXIO; | |
1454 | goto out_unlock; | |
1455 | } | |
1da177e4 | 1456 | |
5db470e2 JK |
1457 | if (lo->lo_offset != info->lo_offset || |
1458 | lo->lo_sizelimit != info->lo_sizelimit) { | |
0c3796c2 | 1459 | size_changed = true; |
5db470e2 | 1460 | sync_blockdev(lo->lo_device); |
f4bd34b1 | 1461 | invalidate_bdev(lo->lo_device); |
5db470e2 JK |
1462 | } |
1463 | ||
ecdd0959 ML |
1464 | /* I/O need to be drained during transfer transition */ |
1465 | blk_mq_freeze_queue(lo->lo_queue); | |
1466 | ||
0c3796c2 | 1467 | if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { |
f4bd34b1 | 1468 | /* If any pages were dirtied after invalidate_bdev(), try again */ |
0c3796c2 MC |
1469 | err = -EAGAIN; |
1470 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1471 | __func__, lo->lo_number, lo->lo_file_name, | |
1472 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
550df5fd | 1473 | goto out_unfreeze; |
0c3796c2 | 1474 | } |
1da177e4 | 1475 | |
faf1d254 | 1476 | prev_lo_flags = lo->lo_flags; |
1da177e4 | 1477 | |
0c3796c2 | 1478 | err = loop_set_status_from_info(lo, info); |
1da177e4 | 1479 | if (err) |
550df5fd | 1480 | goto out_unfreeze; |
1da177e4 | 1481 | |
faf1d254 | 1482 | /* Mask out flags that can't be set using LOOP_SET_STATUS. */ |
6ac92fb5 | 1483 | lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; |
faf1d254 MC |
1484 | /* For those flags, use the previous values instead */ |
1485 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; | |
1486 | /* For flags that can't be cleared, use previous values too */ | |
1487 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; | |
1488 | ||
b0bd158d MC |
1489 | if (size_changed) { |
1490 | loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, | |
1491 | lo->lo_backing_file); | |
1492 | loop_set_size(lo, new_size); | |
b040ad9c | 1493 | } |
541c742a | 1494 | |
dfaa2ef6 | 1495 | loop_config_discard(lo); |
1da177e4 | 1496 | |
2e5ab5f3 ML |
1497 | /* update dio if lo_offset or transfer is changed */ |
1498 | __loop_update_dio(lo, lo->use_dio); | |
1499 | ||
550df5fd | 1500 | out_unfreeze: |
ecdd0959 | 1501 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 | 1502 | |
faf1d254 MC |
1503 | if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && |
1504 | !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { | |
e02898b4 | 1505 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; |
85b0a54a JK |
1506 | bdev = lo->lo_device; |
1507 | partscan = true; | |
e02898b4 | 1508 | } |
550df5fd | 1509 | out_unlock: |
6cc8e743 | 1510 | mutex_unlock(&lo->lo_mutex); |
85b0a54a | 1511 | if (partscan) |
0384264e | 1512 | loop_reread_partitions(lo); |
e02898b4 | 1513 | |
ecdd0959 | 1514 | return err; |
1da177e4 LT |
1515 | } |
1516 | ||
1517 | static int | |
1518 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1519 | { | |
b1ab5fa3 | 1520 | struct path path; |
1da177e4 | 1521 | struct kstat stat; |
2d1d4c1e | 1522 | int ret; |
1da177e4 | 1523 | |
6cc8e743 | 1524 | ret = mutex_lock_killable(&lo->lo_mutex); |
4a5ce9ba JK |
1525 | if (ret) |
1526 | return ret; | |
2d1d4c1e | 1527 | if (lo->lo_state != Lo_bound) { |
6cc8e743 | 1528 | mutex_unlock(&lo->lo_mutex); |
1da177e4 | 1529 | return -ENXIO; |
2d1d4c1e OS |
1530 | } |
1531 | ||
1da177e4 LT |
1532 | memset(info, 0, sizeof(*info)); |
1533 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1534 | info->lo_offset = lo->lo_offset; |
1535 | info->lo_sizelimit = lo->lo_sizelimit; | |
1536 | info->lo_flags = lo->lo_flags; | |
1537 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1538 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1539 | info->lo_encrypt_type = | |
1540 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1541 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1542 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1543 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1544 | lo->lo_encrypt_key_size); | |
1545 | } | |
2d1d4c1e | 1546 | |
6cc8e743 | 1547 | /* Drop lo_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1548 | path = lo->lo_backing_file->f_path; |
1549 | path_get(&path); | |
6cc8e743 | 1550 | mutex_unlock(&lo->lo_mutex); |
b1ab5fa3 | 1551 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1552 | if (!ret) { |
1553 | info->lo_device = huge_encode_dev(stat.dev); | |
1554 | info->lo_inode = stat.ino; | |
1555 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1556 | } | |
b1ab5fa3 | 1557 | path_put(&path); |
2d1d4c1e | 1558 | return ret; |
1da177e4 LT |
1559 | } |
1560 | ||
1561 | static void | |
1562 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1563 | { | |
1564 | memset(info64, 0, sizeof(*info64)); | |
1565 | info64->lo_number = info->lo_number; | |
1566 | info64->lo_device = info->lo_device; | |
1567 | info64->lo_inode = info->lo_inode; | |
1568 | info64->lo_rdevice = info->lo_rdevice; | |
1569 | info64->lo_offset = info->lo_offset; | |
1570 | info64->lo_sizelimit = 0; | |
1571 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1572 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1573 | info64->lo_flags = info->lo_flags; | |
1574 | info64->lo_init[0] = info->lo_init[0]; | |
1575 | info64->lo_init[1] = info->lo_init[1]; | |
1576 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1577 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1578 | else | |
1579 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1580 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1581 | } | |
1582 | ||
1583 | static int | |
1584 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1585 | { | |
1586 | memset(info, 0, sizeof(*info)); | |
1587 | info->lo_number = info64->lo_number; | |
1588 | info->lo_device = info64->lo_device; | |
1589 | info->lo_inode = info64->lo_inode; | |
1590 | info->lo_rdevice = info64->lo_rdevice; | |
1591 | info->lo_offset = info64->lo_offset; | |
1592 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1593 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1594 | info->lo_flags = info64->lo_flags; | |
1595 | info->lo_init[0] = info64->lo_init[0]; | |
1596 | info->lo_init[1] = info64->lo_init[1]; | |
1597 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1598 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1599 | else | |
1600 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1601 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1602 | ||
1603 | /* error in case values were truncated */ | |
1604 | if (info->lo_device != info64->lo_device || | |
1605 | info->lo_rdevice != info64->lo_rdevice || | |
1606 | info->lo_inode != info64->lo_inode || | |
1607 | info->lo_offset != info64->lo_offset) | |
1608 | return -EOVERFLOW; | |
1609 | ||
1610 | return 0; | |
1611 | } | |
1612 | ||
1613 | static int | |
1614 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1615 | { | |
1616 | struct loop_info info; | |
1617 | struct loop_info64 info64; | |
1618 | ||
1619 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1620 | return -EFAULT; | |
1621 | loop_info64_from_old(&info, &info64); | |
1622 | return loop_set_status(lo, &info64); | |
1623 | } | |
1624 | ||
1625 | static int | |
1626 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1627 | { | |
1628 | struct loop_info64 info64; | |
1629 | ||
1630 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1631 | return -EFAULT; | |
1632 | return loop_set_status(lo, &info64); | |
1633 | } | |
1634 | ||
1635 | static int | |
1636 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1637 | struct loop_info info; | |
1638 | struct loop_info64 info64; | |
bdac616d | 1639 | int err; |
1da177e4 | 1640 | |
4a5ce9ba | 1641 | if (!arg) |
bdac616d | 1642 | return -EINVAL; |
bdac616d | 1643 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1644 | if (!err) |
1645 | err = loop_info64_to_old(&info64, &info); | |
1646 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1647 | err = -EFAULT; | |
1648 | ||
1649 | return err; | |
1650 | } | |
1651 | ||
1652 | static int | |
1653 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1654 | struct loop_info64 info64; | |
bdac616d | 1655 | int err; |
1da177e4 | 1656 | |
4a5ce9ba | 1657 | if (!arg) |
bdac616d | 1658 | return -EINVAL; |
bdac616d | 1659 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1660 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1661 | err = -EFAULT; | |
1662 | ||
1663 | return err; | |
1664 | } | |
1665 | ||
51001b7d | 1666 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1667 | { |
0a6ed1b5 MC |
1668 | loff_t size; |
1669 | ||
53d66608 | 1670 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1671 | return -ENXIO; |
53d66608 | 1672 | |
0a6ed1b5 MC |
1673 | size = get_loop_size(lo, lo->lo_backing_file); |
1674 | loop_set_size(lo, size); | |
083a6a50 MC |
1675 | |
1676 | return 0; | |
53d66608 O |
1677 | } |
1678 | ||
ab1cb278 ML |
1679 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1680 | { | |
1681 | int error = -ENXIO; | |
1682 | if (lo->lo_state != Lo_bound) | |
1683 | goto out; | |
1684 | ||
1685 | __loop_update_dio(lo, !!arg); | |
1686 | if (lo->use_dio == !!arg) | |
1687 | return 0; | |
1688 | error = -EINVAL; | |
1689 | out: | |
1690 | return error; | |
1691 | } | |
1692 | ||
89e4fdec OS |
1693 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1694 | { | |
5db470e2 JK |
1695 | int err = 0; |
1696 | ||
89e4fdec OS |
1697 | if (lo->lo_state != Lo_bound) |
1698 | return -ENXIO; | |
1699 | ||
3448914e MC |
1700 | err = loop_validate_block_size(arg); |
1701 | if (err) | |
1702 | return err; | |
89e4fdec | 1703 | |
7e81f99a MC |
1704 | if (lo->lo_queue->limits.logical_block_size == arg) |
1705 | return 0; | |
1706 | ||
1707 | sync_blockdev(lo->lo_device); | |
f4bd34b1 | 1708 | invalidate_bdev(lo->lo_device); |
5db470e2 | 1709 | |
89e4fdec OS |
1710 | blk_mq_freeze_queue(lo->lo_queue); |
1711 | ||
f4bd34b1 | 1712 | /* invalidate_bdev should have truncated all the pages */ |
7e81f99a | 1713 | if (lo->lo_device->bd_inode->i_mapping->nrpages) { |
5db470e2 JK |
1714 | err = -EAGAIN; |
1715 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1716 | __func__, lo->lo_number, lo->lo_file_name, | |
1717 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1718 | goto out_unfreeze; | |
1719 | } | |
1720 | ||
89e4fdec | 1721 | blk_queue_logical_block_size(lo->lo_queue, arg); |
bf093753 OS |
1722 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1723 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec | 1724 | loop_update_dio(lo); |
5db470e2 | 1725 | out_unfreeze: |
89e4fdec OS |
1726 | blk_mq_unfreeze_queue(lo->lo_queue); |
1727 | ||
5db470e2 | 1728 | return err; |
89e4fdec OS |
1729 | } |
1730 | ||
a1316544 JK |
1731 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1732 | unsigned long arg) | |
1da177e4 | 1733 | { |
1da177e4 LT |
1734 | int err; |
1735 | ||
6cc8e743 | 1736 | err = mutex_lock_killable(&lo->lo_mutex); |
3148ffbd | 1737 | if (err) |
a1316544 JK |
1738 | return err; |
1739 | switch (cmd) { | |
1740 | case LOOP_SET_CAPACITY: | |
1741 | err = loop_set_capacity(lo); | |
1742 | break; | |
1743 | case LOOP_SET_DIRECT_IO: | |
1744 | err = loop_set_dio(lo, arg); | |
1745 | break; | |
1746 | case LOOP_SET_BLOCK_SIZE: | |
1747 | err = loop_set_block_size(lo, arg); | |
1748 | break; | |
1749 | default: | |
1750 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1751 | } | |
6cc8e743 | 1752 | mutex_unlock(&lo->lo_mutex); |
a1316544 JK |
1753 | return err; |
1754 | } | |
1755 | ||
1756 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, | |
1757 | unsigned int cmd, unsigned long arg) | |
1758 | { | |
1759 | struct loop_device *lo = bdev->bd_disk->private_data; | |
571fae6e | 1760 | void __user *argp = (void __user *) arg; |
a1316544 | 1761 | int err; |
3148ffbd | 1762 | |
1da177e4 | 1763 | switch (cmd) { |
3448914e MC |
1764 | case LOOP_SET_FD: { |
1765 | /* | |
1766 | * Legacy case - pass in a zeroed out struct loop_config with | |
1767 | * only the file descriptor set , which corresponds with the | |
1768 | * default parameters we'd have used otherwise. | |
1769 | */ | |
1770 | struct loop_config config; | |
1771 | ||
1772 | memset(&config, 0, sizeof(config)); | |
1773 | config.fd = arg; | |
1774 | ||
1775 | return loop_configure(lo, mode, bdev, &config); | |
1776 | } | |
1777 | case LOOP_CONFIGURE: { | |
1778 | struct loop_config config; | |
1779 | ||
1780 | if (copy_from_user(&config, argp, sizeof(config))) | |
1781 | return -EFAULT; | |
1782 | ||
1783 | return loop_configure(lo, mode, bdev, &config); | |
1784 | } | |
1da177e4 | 1785 | case LOOP_CHANGE_FD: |
c3710770 | 1786 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1787 | case LOOP_CLR_FD: |
7ccd0791 | 1788 | return loop_clr_fd(lo); |
1da177e4 | 1789 | case LOOP_SET_STATUS: |
7035b5df | 1790 | err = -EPERM; |
a1316544 | 1791 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
571fae6e | 1792 | err = loop_set_status_old(lo, argp); |
a1316544 | 1793 | } |
1da177e4 LT |
1794 | break; |
1795 | case LOOP_GET_STATUS: | |
571fae6e | 1796 | return loop_get_status_old(lo, argp); |
1da177e4 | 1797 | case LOOP_SET_STATUS64: |
7035b5df | 1798 | err = -EPERM; |
a1316544 | 1799 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
571fae6e | 1800 | err = loop_set_status64(lo, argp); |
a1316544 | 1801 | } |
1da177e4 LT |
1802 | break; |
1803 | case LOOP_GET_STATUS64: | |
571fae6e | 1804 | return loop_get_status64(lo, argp); |
a1316544 | 1805 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1806 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1807 | case LOOP_SET_BLOCK_SIZE: |
a1316544 JK |
1808 | if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) |
1809 | return -EPERM; | |
df561f66 | 1810 | fallthrough; |
1da177e4 | 1811 | default: |
a1316544 JK |
1812 | err = lo_simple_ioctl(lo, cmd, arg); |
1813 | break; | |
1da177e4 | 1814 | } |
f028f3b2 | 1815 | |
1da177e4 LT |
1816 | return err; |
1817 | } | |
1818 | ||
863d5b82 DH |
1819 | #ifdef CONFIG_COMPAT |
1820 | struct compat_loop_info { | |
1821 | compat_int_t lo_number; /* ioctl r/o */ | |
1822 | compat_dev_t lo_device; /* ioctl r/o */ | |
1823 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1824 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1825 | compat_int_t lo_offset; | |
1826 | compat_int_t lo_encrypt_type; | |
1827 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1828 | compat_int_t lo_flags; /* ioctl r/o */ | |
1829 | char lo_name[LO_NAME_SIZE]; | |
1830 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1831 | compat_ulong_t lo_init[2]; | |
1832 | char reserved[4]; | |
1833 | }; | |
1834 | ||
1835 | /* | |
1836 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1837 | * - noinlined to reduce stack space usage in main part of driver | |
1838 | */ | |
1839 | static noinline int | |
ba674cfc | 1840 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1841 | struct loop_info64 *info64) |
1842 | { | |
1843 | struct compat_loop_info info; | |
1844 | ||
1845 | if (copy_from_user(&info, arg, sizeof(info))) | |
1846 | return -EFAULT; | |
1847 | ||
1848 | memset(info64, 0, sizeof(*info64)); | |
1849 | info64->lo_number = info.lo_number; | |
1850 | info64->lo_device = info.lo_device; | |
1851 | info64->lo_inode = info.lo_inode; | |
1852 | info64->lo_rdevice = info.lo_rdevice; | |
1853 | info64->lo_offset = info.lo_offset; | |
1854 | info64->lo_sizelimit = 0; | |
1855 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1856 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1857 | info64->lo_flags = info.lo_flags; | |
1858 | info64->lo_init[0] = info.lo_init[0]; | |
1859 | info64->lo_init[1] = info.lo_init[1]; | |
1860 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1861 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1862 | else | |
1863 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1864 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | /* | |
1869 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1870 | * - noinlined to reduce stack space usage in main part of driver | |
1871 | */ | |
1872 | static noinline int | |
1873 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1874 | struct compat_loop_info __user *arg) | |
1875 | { | |
1876 | struct compat_loop_info info; | |
1877 | ||
1878 | memset(&info, 0, sizeof(info)); | |
1879 | info.lo_number = info64->lo_number; | |
1880 | info.lo_device = info64->lo_device; | |
1881 | info.lo_inode = info64->lo_inode; | |
1882 | info.lo_rdevice = info64->lo_rdevice; | |
1883 | info.lo_offset = info64->lo_offset; | |
1884 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1885 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1886 | info.lo_flags = info64->lo_flags; | |
1887 | info.lo_init[0] = info64->lo_init[0]; | |
1888 | info.lo_init[1] = info64->lo_init[1]; | |
1889 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1890 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1891 | else | |
1892 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1893 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1894 | ||
1895 | /* error in case values were truncated */ | |
1896 | if (info.lo_device != info64->lo_device || | |
1897 | info.lo_rdevice != info64->lo_rdevice || | |
1898 | info.lo_inode != info64->lo_inode || | |
1899 | info.lo_offset != info64->lo_offset || | |
1900 | info.lo_init[0] != info64->lo_init[0] || | |
1901 | info.lo_init[1] != info64->lo_init[1]) | |
1902 | return -EOVERFLOW; | |
1903 | ||
1904 | if (copy_to_user(arg, &info, sizeof(info))) | |
1905 | return -EFAULT; | |
1906 | return 0; | |
1907 | } | |
1908 | ||
1909 | static int | |
1910 | loop_set_status_compat(struct loop_device *lo, | |
1911 | const struct compat_loop_info __user *arg) | |
1912 | { | |
1913 | struct loop_info64 info64; | |
1914 | int ret; | |
1915 | ||
1916 | ret = loop_info64_from_compat(arg, &info64); | |
1917 | if (ret < 0) | |
1918 | return ret; | |
1919 | return loop_set_status(lo, &info64); | |
1920 | } | |
1921 | ||
1922 | static int | |
1923 | loop_get_status_compat(struct loop_device *lo, | |
1924 | struct compat_loop_info __user *arg) | |
1925 | { | |
1926 | struct loop_info64 info64; | |
bdac616d | 1927 | int err; |
863d5b82 | 1928 | |
4a5ce9ba | 1929 | if (!arg) |
bdac616d | 1930 | return -EINVAL; |
bdac616d | 1931 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1932 | if (!err) |
1933 | err = loop_info64_to_compat(&info64, arg); | |
1934 | return err; | |
1935 | } | |
1936 | ||
bb214884 AV |
1937 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1938 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1939 | { |
bb214884 | 1940 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1941 | int err; |
1942 | ||
863d5b82 DH |
1943 | switch(cmd) { |
1944 | case LOOP_SET_STATUS: | |
550df5fd JK |
1945 | err = loop_set_status_compat(lo, |
1946 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1947 | break; |
1948 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1949 | err = loop_get_status_compat(lo, |
1950 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1951 | break; |
53d66608 | 1952 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1953 | case LOOP_CLR_FD: |
1954 | case LOOP_GET_STATUS64: | |
1955 | case LOOP_SET_STATUS64: | |
3448914e | 1956 | case LOOP_CONFIGURE: |
863d5b82 | 1957 | arg = (unsigned long) compat_ptr(arg); |
df561f66 | 1958 | fallthrough; |
863d5b82 DH |
1959 | case LOOP_SET_FD: |
1960 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1961 | case LOOP_SET_BLOCK_SIZE: |
fdbe4eee | 1962 | case LOOP_SET_DIRECT_IO: |
bb214884 | 1963 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1964 | break; |
1965 | default: | |
1966 | err = -ENOIOCTLCMD; | |
1967 | break; | |
1968 | } | |
863d5b82 DH |
1969 | return err; |
1970 | } | |
1971 | #endif | |
1972 | ||
bb214884 | 1973 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1974 | { |
990e7811 | 1975 | struct loop_device *lo = bdev->bd_disk->private_data; |
0a42e99b | 1976 | int err; |
770fe30a | 1977 | |
6cc8e743 | 1978 | err = mutex_lock_killable(&lo->lo_mutex); |
6cc8e743 PT |
1979 | if (err) |
1980 | return err; | |
990e7811 CH |
1981 | if (lo->lo_state == Lo_deleting) |
1982 | err = -ENXIO; | |
1983 | else | |
1984 | atomic_inc(&lo->lo_refcnt); | |
6cc8e743 | 1985 | mutex_unlock(&lo->lo_mutex); |
990e7811 | 1986 | return err; |
1da177e4 LT |
1987 | } |
1988 | ||
967d1dc1 | 1989 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1990 | { |
6cc8e743 | 1991 | struct loop_device *lo = disk->private_data; |
1da177e4 | 1992 | |
6cc8e743 | 1993 | mutex_lock(&lo->lo_mutex); |
f8933667 | 1994 | if (atomic_dec_return(&lo->lo_refcnt)) |
0a42e99b | 1995 | goto out_unlock; |
14f27939 MB |
1996 | |
1997 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
a2505b79 JK |
1998 | if (lo->lo_state != Lo_bound) |
1999 | goto out_unlock; | |
2000 | lo->lo_state = Lo_rundown; | |
6cc8e743 | 2001 | mutex_unlock(&lo->lo_mutex); |
14f27939 MB |
2002 | /* |
2003 | * In autoclear mode, stop the loop thread | |
2004 | * and remove configuration after last close. | |
2005 | */ | |
0da03cab | 2006 | __loop_clr_fd(lo, true); |
7ccd0791 | 2007 | return; |
43cade80 | 2008 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
2009 | /* |
2010 | * Otherwise keep thread (if running) and config, | |
2011 | * but flush possible ongoing bios in thread. | |
2012 | */ | |
43cade80 OS |
2013 | blk_mq_freeze_queue(lo->lo_queue); |
2014 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 2015 | } |
96c58655 | 2016 | |
0a42e99b | 2017 | out_unlock: |
6cc8e743 | 2018 | mutex_unlock(&lo->lo_mutex); |
ae665016 LT |
2019 | } |
2020 | ||
83d5cde4 | 2021 | static const struct block_device_operations lo_fops = { |
1da177e4 | 2022 | .owner = THIS_MODULE, |
bb214884 AV |
2023 | .open = lo_open, |
2024 | .release = lo_release, | |
2025 | .ioctl = lo_ioctl, | |
863d5b82 | 2026 | #ifdef CONFIG_COMPAT |
bb214884 | 2027 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 2028 | #endif |
1da177e4 LT |
2029 | }; |
2030 | ||
2031 | /* | |
2032 | * And now the modules code and kernel interface. | |
2033 | */ | |
73285082 | 2034 | static int max_loop; |
5657a819 | 2035 | module_param(max_loop, int, 0444); |
a47653fc | 2036 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
5657a819 | 2037 | module_param(max_part, int, 0444); |
476a4813 | 2038 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
2039 | MODULE_LICENSE("GPL"); |
2040 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
2041 | ||
2042 | int loop_register_transfer(struct loop_func_table *funcs) | |
2043 | { | |
2044 | unsigned int n = funcs->number; | |
2045 | ||
2046 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
2047 | return -EINVAL; | |
2048 | xfer_funcs[n] = funcs; | |
2049 | return 0; | |
2050 | } | |
2051 | ||
34dd82af KS |
2052 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
2053 | { | |
2054 | struct loop_device *lo = ptr; | |
2055 | struct loop_func_table *xfer = data; | |
2056 | ||
6cc8e743 | 2057 | mutex_lock(&lo->lo_mutex); |
34dd82af KS |
2058 | if (lo->lo_encryption == xfer) |
2059 | loop_release_xfer(lo); | |
6cc8e743 | 2060 | mutex_unlock(&lo->lo_mutex); |
34dd82af KS |
2061 | return 0; |
2062 | } | |
2063 | ||
1da177e4 LT |
2064 | int loop_unregister_transfer(int number) |
2065 | { | |
2066 | unsigned int n = number; | |
1da177e4 LT |
2067 | struct loop_func_table *xfer; |
2068 | ||
2069 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
2070 | return -EINVAL; | |
2071 | ||
2072 | xfer_funcs[n] = NULL; | |
34dd82af | 2073 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
2074 | return 0; |
2075 | } | |
2076 | ||
2077 | EXPORT_SYMBOL(loop_register_transfer); | |
2078 | EXPORT_SYMBOL(loop_unregister_transfer); | |
2079 | ||
fc17b653 | 2080 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
2081 | const struct blk_mq_queue_data *bd) |
2082 | { | |
1894e916 JA |
2083 | struct request *rq = bd->rq; |
2084 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
2085 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 2086 | |
1894e916 | 2087 | blk_mq_start_request(rq); |
b5dd2f60 | 2088 | |
f4aa4c7b | 2089 | if (lo->lo_state != Lo_bound) |
fc17b653 | 2090 | return BLK_STS_IOERR; |
f4aa4c7b | 2091 | |
1894e916 | 2092 | switch (req_op(rq)) { |
f0225cac CH |
2093 | case REQ_OP_FLUSH: |
2094 | case REQ_OP_DISCARD: | |
19372e27 | 2095 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 2096 | cmd->use_aio = false; |
f0225cac CH |
2097 | break; |
2098 | default: | |
2099 | cmd->use_aio = lo->use_dio; | |
2100 | break; | |
2101 | } | |
bc07c10a | 2102 | |
d4478e92 | 2103 | /* always use the first bio's css */ |
c74d40e8 DS |
2104 | cmd->blkcg_css = NULL; |
2105 | cmd->memcg_css = NULL; | |
0b508bc9 | 2106 | #ifdef CONFIG_BLK_CGROUP |
c74d40e8 DS |
2107 | if (rq->bio && rq->bio->bi_blkg) { |
2108 | cmd->blkcg_css = &bio_blkcg(rq->bio)->css; | |
2109 | #ifdef CONFIG_MEMCG | |
2110 | cmd->memcg_css = | |
2111 | cgroup_get_e_css(cmd->blkcg_css->cgroup, | |
2112 | &memory_cgrp_subsys); | |
2113 | #endif | |
2114 | } | |
d4478e92 | 2115 | #endif |
87579e9b | 2116 | loop_queue_work(lo, cmd); |
b5dd2f60 | 2117 | |
fc17b653 | 2118 | return BLK_STS_OK; |
b5dd2f60 ML |
2119 | } |
2120 | ||
2121 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
2122 | { | |
1894e916 JA |
2123 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
2124 | const bool write = op_is_write(req_op(rq)); | |
2125 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 2126 | int ret = 0; |
c74d40e8 | 2127 | struct mem_cgroup *old_memcg = NULL; |
b5dd2f60 | 2128 | |
f4829a9b CH |
2129 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
2130 | ret = -EIO; | |
b5dd2f60 | 2131 | goto failed; |
f4829a9b | 2132 | } |
b5dd2f60 | 2133 | |
c74d40e8 DS |
2134 | if (cmd->blkcg_css) |
2135 | kthread_associate_blkcg(cmd->blkcg_css); | |
2136 | if (cmd->memcg_css) | |
2137 | old_memcg = set_active_memcg( | |
2138 | mem_cgroup_from_css(cmd->memcg_css)); | |
2139 | ||
1894e916 | 2140 | ret = do_req_filebacked(lo, rq); |
c74d40e8 DS |
2141 | |
2142 | if (cmd->blkcg_css) | |
2143 | kthread_associate_blkcg(NULL); | |
2144 | ||
2145 | if (cmd->memcg_css) { | |
2146 | set_active_memcg(old_memcg); | |
2147 | css_put(cmd->memcg_css); | |
2148 | } | |
b5dd2f60 | 2149 | failed: |
bc07c10a | 2150 | /* complete non-aio request */ |
fe2cb290 | 2151 | if (!cmd->use_aio || ret) { |
8cd55087 EG |
2152 | if (ret == -EOPNOTSUPP) |
2153 | cmd->ret = ret; | |
2154 | else | |
2155 | cmd->ret = ret ? -EIO : 0; | |
15f73f5b CH |
2156 | if (likely(!blk_should_fake_timeout(rq->q))) |
2157 | blk_mq_complete_request(rq); | |
fe2cb290 | 2158 | } |
b5dd2f60 ML |
2159 | } |
2160 | ||
87579e9b DS |
2161 | static void loop_set_timer(struct loop_device *lo) |
2162 | { | |
2163 | timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); | |
2164 | } | |
2165 | ||
2166 | static void loop_process_work(struct loop_worker *worker, | |
2167 | struct list_head *cmd_list, struct loop_device *lo) | |
b5dd2f60 | 2168 | { |
87579e9b DS |
2169 | int orig_flags = current->flags; |
2170 | struct loop_cmd *cmd; | |
b5dd2f60 | 2171 | |
87579e9b DS |
2172 | current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; |
2173 | spin_lock_irq(&lo->lo_work_lock); | |
2174 | while (!list_empty(cmd_list)) { | |
2175 | cmd = container_of( | |
2176 | cmd_list->next, struct loop_cmd, list_entry); | |
2177 | list_del(cmd_list->next); | |
2178 | spin_unlock_irq(&lo->lo_work_lock); | |
2179 | ||
2180 | loop_handle_cmd(cmd); | |
2181 | cond_resched(); | |
2182 | ||
2183 | spin_lock_irq(&lo->lo_work_lock); | |
2184 | } | |
2185 | ||
2186 | /* | |
2187 | * We only add to the idle list if there are no pending cmds | |
2188 | * *and* the worker will not run again which ensures that it | |
2189 | * is safe to free any worker on the idle list | |
2190 | */ | |
2191 | if (worker && !work_pending(&worker->work)) { | |
2192 | worker->last_ran_at = jiffies; | |
2193 | list_add_tail(&worker->idle_list, &lo->idle_worker_list); | |
2194 | loop_set_timer(lo); | |
2195 | } | |
2196 | spin_unlock_irq(&lo->lo_work_lock); | |
2197 | current->flags = orig_flags; | |
b5dd2f60 ML |
2198 | } |
2199 | ||
87579e9b | 2200 | static void loop_workfn(struct work_struct *work) |
b5dd2f60 | 2201 | { |
87579e9b DS |
2202 | struct loop_worker *worker = |
2203 | container_of(work, struct loop_worker, work); | |
2204 | loop_process_work(worker, &worker->cmd_list, worker->lo); | |
2205 | } | |
b5dd2f60 | 2206 | |
87579e9b DS |
2207 | static void loop_rootcg_workfn(struct work_struct *work) |
2208 | { | |
2209 | struct loop_device *lo = | |
2210 | container_of(work, struct loop_device, rootcg_work); | |
2211 | loop_process_work(NULL, &lo->rootcg_cmd_list, lo); | |
2212 | } | |
2213 | ||
2214 | static void loop_free_idle_workers(struct timer_list *timer) | |
2215 | { | |
2216 | struct loop_device *lo = container_of(timer, struct loop_device, timer); | |
2217 | struct loop_worker *pos, *worker; | |
2218 | ||
2219 | spin_lock_irq(&lo->lo_work_lock); | |
2220 | list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, | |
2221 | idle_list) { | |
2222 | if (time_is_after_jiffies(worker->last_ran_at + | |
2223 | LOOP_IDLE_WORKER_TIMEOUT)) | |
2224 | break; | |
2225 | list_del(&worker->idle_list); | |
2226 | rb_erase(&worker->rb_node, &lo->worker_tree); | |
c74d40e8 | 2227 | css_put(worker->blkcg_css); |
87579e9b DS |
2228 | kfree(worker); |
2229 | } | |
2230 | if (!list_empty(&lo->idle_worker_list)) | |
2231 | loop_set_timer(lo); | |
2232 | spin_unlock_irq(&lo->lo_work_lock); | |
b5dd2f60 ML |
2233 | } |
2234 | ||
f363b089 | 2235 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 2236 | .queue_rq = loop_queue_rq, |
fe2cb290 | 2237 | .complete = lo_complete_rq, |
b5dd2f60 ML |
2238 | }; |
2239 | ||
34dd82af | 2240 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
2241 | { |
2242 | struct loop_device *lo; | |
2243 | struct gendisk *disk; | |
34dd82af | 2244 | int err; |
73285082 | 2245 | |
68d740d7 | 2246 | err = -ENOMEM; |
73285082 | 2247 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 2248 | if (!lo) |
73285082 | 2249 | goto out; |
34dd82af | 2250 | |
ef7e7c82 MP |
2251 | lo->lo_state = Lo_unbound; |
2252 | ||
c718aa65 | 2253 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 2254 | if (i >= 0) { |
c718aa65 TH |
2255 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
2256 | if (err == -ENOSPC) | |
34dd82af | 2257 | err = -EEXIST; |
34dd82af | 2258 | } else { |
c718aa65 | 2259 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
2260 | } |
2261 | if (err < 0) | |
2262 | goto out_free_dev; | |
c718aa65 | 2263 | i = err; |
73285082 | 2264 | |
183cfb57 | 2265 | err = -ENOMEM; |
b5dd2f60 ML |
2266 | lo->tag_set.ops = &loop_mq_ops; |
2267 | lo->tag_set.nr_hw_queues = 1; | |
2268 | lo->tag_set.queue_depth = 128; | |
2269 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
2270 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
bf0beec0 | 2271 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; |
b5dd2f60 ML |
2272 | lo->tag_set.driver_data = lo; |
2273 | ||
2274 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
2275 | if (err) | |
3ec981e3 | 2276 | goto out_free_idr; |
73285082 | 2277 | |
1c99502f CH |
2278 | disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); |
2279 | if (IS_ERR(disk)) { | |
2280 | err = PTR_ERR(disk); | |
b5dd2f60 ML |
2281 | goto out_cleanup_tags; |
2282 | } | |
1c99502f | 2283 | lo->lo_queue = lo->lo_disk->queue; |
ef7e7c82 | 2284 | |
54bb0ade | 2285 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 2286 | |
5b5e20f4 | 2287 | /* |
40326d8a SL |
2288 | * By default, we do buffer IO, so it doesn't make sense to enable |
2289 | * merge because the I/O submitted to backing file is handled page by | |
2290 | * page. For directio mode, merge does help to dispatch bigger request | |
2291 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 | 2292 | */ |
8b904b5b | 2293 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
5b5e20f4 | 2294 | |
e03c8dd1 KS |
2295 | /* |
2296 | * Disable partition scanning by default. The in-kernel partition | |
2297 | * scanning can be requested individually per-device during its | |
2298 | * setup. Userspace can always add and remove partitions from all | |
2299 | * devices. The needed partition minors are allocated from the | |
2300 | * extended minor space, the main loop device numbers will continue | |
2301 | * to match the loop minors, regardless of the number of partitions | |
2302 | * used. | |
2303 | * | |
2304 | * If max_part is given, partition scanning is globally enabled for | |
2305 | * all loop devices. The minors for the main loop devices will be | |
2306 | * multiples of max_part. | |
2307 | * | |
2308 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
2309 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
2310 | * complicated, are too static, inflexible and may surprise | |
2311 | * userspace tools. Parameters like this in general should be avoided. | |
2312 | */ | |
2313 | if (!part_shift) | |
2314 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
2315 | disk->flags |= GENHD_FL_EXT_DEVT; | |
f8933667 | 2316 | atomic_set(&lo->lo_refcnt, 0); |
6cc8e743 | 2317 | mutex_init(&lo->lo_mutex); |
73285082 | 2318 | lo->lo_number = i; |
73285082 | 2319 | spin_lock_init(&lo->lo_lock); |
87579e9b | 2320 | spin_lock_init(&lo->lo_work_lock); |
73285082 | 2321 | disk->major = LOOP_MAJOR; |
476a4813 | 2322 | disk->first_minor = i << part_shift; |
1c99502f | 2323 | disk->minors = 1 << part_shift; |
73285082 KC |
2324 | disk->fops = &lo_fops; |
2325 | disk->private_data = lo; | |
2326 | disk->queue = lo->lo_queue; | |
2327 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
2328 | add_disk(disk); |
2329 | *l = lo; | |
2330 | return lo->lo_number; | |
73285082 | 2331 | |
b5dd2f60 ML |
2332 | out_cleanup_tags: |
2333 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
2334 | out_free_idr: |
2335 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
2336 | out_free_dev: |
2337 | kfree(lo); | |
2338 | out: | |
34dd82af | 2339 | return err; |
73285082 KC |
2340 | } |
2341 | ||
34dd82af | 2342 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2343 | { |
6cd18e71 | 2344 | del_gendisk(lo->lo_disk); |
1c99502f | 2345 | blk_cleanup_disk(lo->lo_disk); |
b5dd2f60 | 2346 | blk_mq_free_tag_set(&lo->tag_set); |
6cc8e743 | 2347 | mutex_destroy(&lo->lo_mutex); |
73285082 KC |
2348 | kfree(lo); |
2349 | } | |
1da177e4 | 2350 | |
770fe30a KS |
2351 | static int find_free_cb(int id, void *ptr, void *data) |
2352 | { | |
2353 | struct loop_device *lo = ptr; | |
2354 | struct loop_device **l = data; | |
2355 | ||
2356 | if (lo->lo_state == Lo_unbound) { | |
2357 | *l = lo; | |
2358 | return 1; | |
2359 | } | |
2360 | return 0; | |
2361 | } | |
2362 | ||
34dd82af | 2363 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
2364 | { |
2365 | struct loop_device *lo; | |
34dd82af | 2366 | int ret = -ENODEV; |
a47653fc | 2367 | |
770fe30a KS |
2368 | if (i < 0) { |
2369 | int err; | |
2370 | ||
2371 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
2372 | if (err == 1) { | |
2373 | *l = lo; | |
2374 | ret = lo->lo_number; | |
2375 | } | |
2376 | goto out; | |
a47653fc KC |
2377 | } |
2378 | ||
770fe30a | 2379 | /* lookup and return a specific i */ |
34dd82af | 2380 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 2381 | if (lo) { |
34dd82af KS |
2382 | *l = lo; |
2383 | ret = lo->lo_number; | |
a47653fc | 2384 | } |
770fe30a | 2385 | out: |
34dd82af | 2386 | return ret; |
a47653fc KC |
2387 | } |
2388 | ||
8410d38c | 2389 | static void loop_probe(dev_t dev) |
73285082 | 2390 | { |
8410d38c | 2391 | int idx = MINOR(dev) >> part_shift; |
705962cc | 2392 | struct loop_device *lo; |
8410d38c CH |
2393 | |
2394 | if (max_loop && idx >= max_loop) | |
2395 | return; | |
73285082 | 2396 | |
0a42e99b | 2397 | mutex_lock(&loop_ctl_mutex); |
8410d38c CH |
2398 | if (loop_lookup(&lo, idx) < 0) |
2399 | loop_add(&lo, idx); | |
0a42e99b | 2400 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2401 | } |
2402 | ||
770fe30a KS |
2403 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2404 | unsigned long parm) | |
2405 | { | |
2406 | struct loop_device *lo; | |
0a42e99b | 2407 | int ret; |
770fe30a | 2408 | |
0a42e99b JK |
2409 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2410 | if (ret) | |
2411 | return ret; | |
2412 | ||
2413 | ret = -ENOSYS; | |
770fe30a KS |
2414 | switch (cmd) { |
2415 | case LOOP_CTL_ADD: | |
2416 | ret = loop_lookup(&lo, parm); | |
2417 | if (ret >= 0) { | |
2418 | ret = -EEXIST; | |
2419 | break; | |
2420 | } | |
2421 | ret = loop_add(&lo, parm); | |
2422 | break; | |
2423 | case LOOP_CTL_REMOVE: | |
2424 | ret = loop_lookup(&lo, parm); | |
2425 | if (ret < 0) | |
2426 | break; | |
6cc8e743 PT |
2427 | ret = mutex_lock_killable(&lo->lo_mutex); |
2428 | if (ret) | |
2429 | break; | |
770fe30a KS |
2430 | if (lo->lo_state != Lo_unbound) { |
2431 | ret = -EBUSY; | |
6cc8e743 | 2432 | mutex_unlock(&lo->lo_mutex); |
770fe30a KS |
2433 | break; |
2434 | } | |
f8933667 | 2435 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a | 2436 | ret = -EBUSY; |
6cc8e743 | 2437 | mutex_unlock(&lo->lo_mutex); |
770fe30a KS |
2438 | break; |
2439 | } | |
990e7811 | 2440 | lo->lo_state = Lo_deleting; |
6cc8e743 | 2441 | mutex_unlock(&lo->lo_mutex); |
770fe30a KS |
2442 | idr_remove(&loop_index_idr, lo->lo_number); |
2443 | loop_remove(lo); | |
2444 | break; | |
2445 | case LOOP_CTL_GET_FREE: | |
2446 | ret = loop_lookup(&lo, -1); | |
2447 | if (ret >= 0) | |
2448 | break; | |
2449 | ret = loop_add(&lo, -1); | |
2450 | } | |
0a42e99b | 2451 | mutex_unlock(&loop_ctl_mutex); |
770fe30a KS |
2452 | |
2453 | return ret; | |
2454 | } | |
2455 | ||
2456 | static const struct file_operations loop_ctl_fops = { | |
2457 | .open = nonseekable_open, | |
2458 | .unlocked_ioctl = loop_control_ioctl, | |
2459 | .compat_ioctl = loop_control_ioctl, | |
2460 | .owner = THIS_MODULE, | |
2461 | .llseek = noop_llseek, | |
2462 | }; | |
2463 | ||
2464 | static struct miscdevice loop_misc = { | |
2465 | .minor = LOOP_CTRL_MINOR, | |
2466 | .name = "loop-control", | |
2467 | .fops = &loop_ctl_fops, | |
2468 | }; | |
2469 | ||
2470 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2471 | MODULE_ALIAS("devname:loop-control"); | |
2472 | ||
73285082 KC |
2473 | static int __init loop_init(void) |
2474 | { | |
a47653fc | 2475 | int i, nr; |
34dd82af | 2476 | struct loop_device *lo; |
770fe30a | 2477 | int err; |
a47653fc | 2478 | |
476a4813 | 2479 | part_shift = 0; |
ac04fee0 | 2480 | if (max_part > 0) { |
476a4813 LV |
2481 | part_shift = fls(max_part); |
2482 | ||
ac04fee0 NK |
2483 | /* |
2484 | * Adjust max_part according to part_shift as it is exported | |
2485 | * to user space so that user can decide correct minor number | |
2486 | * if [s]he want to create more devices. | |
2487 | * | |
2488 | * Note that -1 is required because partition 0 is reserved | |
2489 | * for the whole disk. | |
2490 | */ | |
2491 | max_part = (1UL << part_shift) - 1; | |
2492 | } | |
2493 | ||
b1a66504 GC |
2494 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2495 | err = -EINVAL; | |
a8c1d064 | 2496 | goto err_out; |
b1a66504 | 2497 | } |
78f4bb36 | 2498 | |
b1a66504 GC |
2499 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2500 | err = -EINVAL; | |
a8c1d064 | 2501 | goto err_out; |
b1a66504 | 2502 | } |
1da177e4 | 2503 | |
d134b00b KS |
2504 | /* |
2505 | * If max_loop is specified, create that many devices upfront. | |
2506 | * This also becomes a hard limit. If max_loop is not specified, | |
2507 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2508 | * init time. Loop devices can be requested on-demand with the | |
2509 | * /dev/loop-control interface, or be instantiated by accessing | |
2510 | * a 'dead' device node. | |
2511 | */ | |
aeb2b0b1 | 2512 | if (max_loop) |
a47653fc | 2513 | nr = max_loop; |
aeb2b0b1 | 2514 | else |
d134b00b | 2515 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a47653fc | 2516 | |
a8c1d064 AV |
2517 | err = misc_register(&loop_misc); |
2518 | if (err < 0) | |
2519 | goto err_out; | |
2520 | ||
2521 | ||
8410d38c | 2522 | if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { |
b1a66504 GC |
2523 | err = -EIO; |
2524 | goto misc_out; | |
2525 | } | |
1da177e4 | 2526 | |
d134b00b | 2527 | /* pre-create number of devices given by config or max_loop */ |
0a42e99b | 2528 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2529 | for (i = 0; i < nr; i++) |
2530 | loop_add(&lo, i); | |
0a42e99b | 2531 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 2532 | |
73285082 | 2533 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2534 | return 0; |
b1a66504 GC |
2535 | |
2536 | misc_out: | |
2537 | misc_deregister(&loop_misc); | |
a8c1d064 | 2538 | err_out: |
b1a66504 | 2539 | return err; |
34dd82af | 2540 | } |
a47653fc | 2541 | |
34dd82af KS |
2542 | static int loop_exit_cb(int id, void *ptr, void *data) |
2543 | { | |
2544 | struct loop_device *lo = ptr; | |
a47653fc | 2545 | |
34dd82af KS |
2546 | loop_remove(lo); |
2547 | return 0; | |
1da177e4 LT |
2548 | } |
2549 | ||
73285082 | 2550 | static void __exit loop_exit(void) |
1da177e4 | 2551 | { |
200f9337 LC |
2552 | mutex_lock(&loop_ctl_mutex); |
2553 | ||
34dd82af | 2554 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2555 | idr_destroy(&loop_index_idr); |
73285082 | 2556 | |
00d59405 | 2557 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2558 | |
2559 | misc_deregister(&loop_misc); | |
200f9337 LC |
2560 | |
2561 | mutex_unlock(&loop_ctl_mutex); | |
1da177e4 LT |
2562 | } |
2563 | ||
2564 | module_init(loop_init); | |
2565 | module_exit(loop_exit); | |
2566 | ||
2567 | #ifndef MODULE | |
2568 | static int __init max_loop_setup(char *str) | |
2569 | { | |
2570 | max_loop = simple_strtol(str, NULL, 0); | |
2571 | return 1; | |
2572 | } | |
2573 | ||
2574 | __setup("max_loop=", max_loop_setup); | |
2575 | #endif |