Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/block/loop.c | |
3 | * | |
4 | * Written by Theodore Ts'o, 3/29/93 | |
5 | * | |
6 | * Copyright 1993 by Theodore Ts'o. Redistribution of this file is | |
7 | * permitted under the GNU General Public License. | |
8 | * | |
9 | * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 | |
10 | * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 | |
11 | * | |
12 | * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 | |
13 | * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 | |
14 | * | |
15 | * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 | |
16 | * | |
17 | * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 | |
18 | * | |
19 | * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 | |
20 | * | |
21 | * Loadable modules and other fixes by AK, 1998 | |
22 | * | |
23 | * Make real block number available to downstream transfer functions, enables | |
24 | * CBC (and relatives) mode encryption requiring unique IVs per data block. | |
25 | * Reed H. Petty, rhp@draper.net | |
26 | * | |
27 | * Maximum number of loop devices now dynamic via max_loop module parameter. | |
28 | * Russell Kroll <rkroll@exploits.org> 19990701 | |
29 | * | |
30 | * Maximum number of loop devices when compiled-in now selectable by passing | |
31 | * max_loop=<1-255> to the kernel on boot. | |
96de0e25 | 32 | * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 |
1da177e4 LT |
33 | * |
34 | * Completely rewrite request handling to be make_request_fn style and | |
35 | * non blocking, pushing work to a helper thread. Lots of fixes from | |
36 | * Al Viro too. | |
37 | * Jens Axboe <axboe@suse.de>, Nov 2000 | |
38 | * | |
39 | * Support up to 256 loop devices | |
40 | * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 | |
41 | * | |
42 | * Support for falling back on the write file operation when the address space | |
4e02ed4b | 43 | * operations write_begin is not available on the backing filesystem. |
1da177e4 LT |
44 | * Anton Altaparmakov, 16 Feb 2005 |
45 | * | |
46 | * Still To Fix: | |
47 | * - Advisory locking is ignored here. | |
48 | * - Should use an own CAP_* category instead of CAP_SYS_ADMIN | |
49 | * | |
50 | */ | |
51 | ||
1da177e4 LT |
52 | #include <linux/module.h> |
53 | #include <linux/moduleparam.h> | |
54 | #include <linux/sched.h> | |
55 | #include <linux/fs.h> | |
56 | #include <linux/file.h> | |
57 | #include <linux/stat.h> | |
58 | #include <linux/errno.h> | |
59 | #include <linux/major.h> | |
60 | #include <linux/wait.h> | |
61 | #include <linux/blkdev.h> | |
62 | #include <linux/blkpg.h> | |
63 | #include <linux/init.h> | |
1da177e4 LT |
64 | #include <linux/swap.h> |
65 | #include <linux/slab.h> | |
863d5b82 | 66 | #include <linux/compat.h> |
1da177e4 | 67 | #include <linux/suspend.h> |
83144186 | 68 | #include <linux/freezer.h> |
2a48fc0a | 69 | #include <linux/mutex.h> |
1da177e4 | 70 | #include <linux/writeback.h> |
1da177e4 LT |
71 | #include <linux/completion.h> |
72 | #include <linux/highmem.h> | |
6c997918 | 73 | #include <linux/kthread.h> |
d6b29d7c | 74 | #include <linux/splice.h> |
ee862730 | 75 | #include <linux/sysfs.h> |
770fe30a | 76 | #include <linux/miscdevice.h> |
dfaa2ef6 | 77 | #include <linux/falloc.h> |
283e7e5d | 78 | #include <linux/uio.h> |
d9a08a9e | 79 | #include <linux/ioprio.h> |
db6638d7 | 80 | #include <linux/blk-cgroup.h> |
d9a08a9e | 81 | |
83a87611 | 82 | #include "loop.h" |
1da177e4 | 83 | |
7c0f6ba6 | 84 | #include <linux/uaccess.h> |
1da177e4 | 85 | |
34dd82af | 86 | static DEFINE_IDR(loop_index_idr); |
310ca162 | 87 | static DEFINE_MUTEX(loop_ctl_mutex); |
1da177e4 | 88 | |
476a4813 LV |
89 | static int max_part; |
90 | static int part_shift; | |
91 | ||
1da177e4 LT |
92 | static int transfer_xor(struct loop_device *lo, int cmd, |
93 | struct page *raw_page, unsigned raw_off, | |
94 | struct page *loop_page, unsigned loop_off, | |
95 | int size, sector_t real_block) | |
96 | { | |
cfd8005c CW |
97 | char *raw_buf = kmap_atomic(raw_page) + raw_off; |
98 | char *loop_buf = kmap_atomic(loop_page) + loop_off; | |
1da177e4 LT |
99 | char *in, *out, *key; |
100 | int i, keysize; | |
101 | ||
102 | if (cmd == READ) { | |
103 | in = raw_buf; | |
104 | out = loop_buf; | |
105 | } else { | |
106 | in = loop_buf; | |
107 | out = raw_buf; | |
108 | } | |
109 | ||
110 | key = lo->lo_encrypt_key; | |
111 | keysize = lo->lo_encrypt_key_size; | |
112 | for (i = 0; i < size; i++) | |
113 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
114 | ||
cfd8005c CW |
115 | kunmap_atomic(loop_buf); |
116 | kunmap_atomic(raw_buf); | |
1da177e4 LT |
117 | cond_resched(); |
118 | return 0; | |
119 | } | |
120 | ||
121 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
122 | { | |
123 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
124 | return -EINVAL; | |
125 | return 0; | |
126 | } | |
127 | ||
128 | static struct loop_func_table none_funcs = { | |
129 | .number = LO_CRYPT_NONE, | |
aa4d8616 | 130 | }; |
1da177e4 LT |
131 | |
132 | static struct loop_func_table xor_funcs = { | |
133 | .number = LO_CRYPT_XOR, | |
134 | .transfer = transfer_xor, | |
135 | .init = xor_init | |
aa4d8616 | 136 | }; |
1da177e4 LT |
137 | |
138 | /* xfer_funcs[0] is special - its release function is never called */ | |
139 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
140 | &none_funcs, | |
141 | &xor_funcs | |
142 | }; | |
143 | ||
7035b5df | 144 | static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) |
1da177e4 | 145 | { |
b7a1da69 | 146 | loff_t loopsize; |
1da177e4 LT |
147 | |
148 | /* Compute loopsize in bytes */ | |
b7a1da69 GC |
149 | loopsize = i_size_read(file->f_mapping->host); |
150 | if (offset > 0) | |
151 | loopsize -= offset; | |
152 | /* offset is beyond i_size, weird but possible */ | |
7035b5df DM |
153 | if (loopsize < 0) |
154 | return 0; | |
1da177e4 | 155 | |
7035b5df DM |
156 | if (sizelimit > 0 && sizelimit < loopsize) |
157 | loopsize = sizelimit; | |
1da177e4 LT |
158 | /* |
159 | * Unfortunately, if we want to do I/O on the device, | |
160 | * the number of 512-byte sectors has to fit into a sector_t. | |
161 | */ | |
162 | return loopsize >> 9; | |
163 | } | |
164 | ||
7035b5df DM |
165 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) |
166 | { | |
167 | return get_size(lo->lo_offset, lo->lo_sizelimit, file); | |
168 | } | |
169 | ||
2e5ab5f3 ML |
170 | static void __loop_update_dio(struct loop_device *lo, bool dio) |
171 | { | |
172 | struct file *file = lo->lo_backing_file; | |
173 | struct address_space *mapping = file->f_mapping; | |
174 | struct inode *inode = mapping->host; | |
175 | unsigned short sb_bsize = 0; | |
176 | unsigned dio_align = 0; | |
177 | bool use_dio; | |
178 | ||
179 | if (inode->i_sb->s_bdev) { | |
180 | sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); | |
181 | dio_align = sb_bsize - 1; | |
182 | } | |
183 | ||
184 | /* | |
185 | * We support direct I/O only if lo_offset is aligned with the | |
186 | * logical I/O size of backing device, and the logical block | |
187 | * size of loop is bigger than the backing device's and the loop | |
188 | * needn't transform transfer. | |
189 | * | |
190 | * TODO: the above condition may be loosed in the future, and | |
191 | * direct I/O may be switched runtime at that time because most | |
89d790ab | 192 | * of requests in sane applications should be PAGE_SIZE aligned |
2e5ab5f3 ML |
193 | */ |
194 | if (dio) { | |
195 | if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && | |
196 | !(lo->lo_offset & dio_align) && | |
197 | mapping->a_ops->direct_IO && | |
198 | !lo->transfer) | |
199 | use_dio = true; | |
200 | else | |
201 | use_dio = false; | |
202 | } else { | |
203 | use_dio = false; | |
204 | } | |
205 | ||
206 | if (lo->use_dio == use_dio) | |
207 | return; | |
208 | ||
209 | /* flush dirty pages before changing direct IO */ | |
210 | vfs_fsync(file, 0); | |
211 | ||
212 | /* | |
213 | * The flag of LO_FLAGS_DIRECT_IO is handled similarly with | |
214 | * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup | |
215 | * will get updated by ioctl(LOOP_GET_STATUS) | |
216 | */ | |
0fbcf579 MC |
217 | if (lo->lo_state == Lo_bound) |
218 | blk_mq_freeze_queue(lo->lo_queue); | |
2e5ab5f3 | 219 | lo->use_dio = use_dio; |
40326d8a | 220 | if (use_dio) { |
8b904b5b | 221 | blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 222 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
40326d8a | 223 | } else { |
8b904b5b | 224 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 225 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; |
40326d8a | 226 | } |
0fbcf579 MC |
227 | if (lo->lo_state == Lo_bound) |
228 | blk_mq_unfreeze_queue(lo->lo_queue); | |
2e5ab5f3 ML |
229 | } |
230 | ||
1da177e4 | 231 | static int |
1e6ec9ea | 232 | figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) |
1da177e4 | 233 | { |
7035b5df | 234 | loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); |
1da177e4 | 235 | sector_t x = (sector_t)size; |
7b0576a3 | 236 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
237 | |
238 | if (unlikely((loff_t)x != size)) | |
239 | return -EFBIG; | |
7035b5df DM |
240 | if (lo->lo_offset != offset) |
241 | lo->lo_offset = offset; | |
242 | if (lo->lo_sizelimit != sizelimit) | |
243 | lo->lo_sizelimit = sizelimit; | |
73285082 | 244 | set_capacity(lo->lo_disk, x); |
7b0576a3 GC |
245 | bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); |
246 | /* let user-space know about the new size */ | |
247 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
7035b5df | 248 | return 0; |
1da177e4 LT |
249 | } |
250 | ||
251 | static inline int | |
252 | lo_do_transfer(struct loop_device *lo, int cmd, | |
253 | struct page *rpage, unsigned roffs, | |
254 | struct page *lpage, unsigned loffs, | |
255 | int size, sector_t rblock) | |
256 | { | |
aa4d8616 CH |
257 | int ret; |
258 | ||
259 | ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
260 | if (likely(!ret)) | |
1da177e4 LT |
261 | return 0; |
262 | ||
aa4d8616 CH |
263 | printk_ratelimited(KERN_ERR |
264 | "loop: Transfer error at byte offset %llu, length %i.\n", | |
265 | (unsigned long long)rblock << 9, size); | |
266 | return ret; | |
1da177e4 LT |
267 | } |
268 | ||
aa4d8616 | 269 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 270 | { |
aa4d8616 | 271 | struct iov_iter i; |
1da177e4 | 272 | ssize_t bw; |
283e7e5d | 273 | |
b6207430 | 274 | iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); |
1da177e4 | 275 | |
03d95eb2 | 276 | file_start_write(file); |
abbb6589 | 277 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 278 | file_end_write(file); |
aa4d8616 CH |
279 | |
280 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 281 | return 0; |
aa4d8616 CH |
282 | |
283 | printk_ratelimited(KERN_ERR | |
284 | "loop: Write error at byte offset %llu, length %i.\n", | |
285 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
286 | if (bw >= 0) |
287 | bw = -EIO; | |
288 | return bw; | |
289 | } | |
290 | ||
aa4d8616 CH |
291 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
292 | loff_t pos) | |
1da177e4 | 293 | { |
aa4d8616 CH |
294 | struct bio_vec bvec; |
295 | struct req_iterator iter; | |
296 | int ret = 0; | |
297 | ||
298 | rq_for_each_segment(bvec, rq, iter) { | |
299 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
300 | if (ret < 0) | |
301 | break; | |
302 | cond_resched(); | |
303 | } | |
304 | ||
305 | return ret; | |
1da177e4 LT |
306 | } |
307 | ||
aa4d8616 | 308 | /* |
456be148 CH |
309 | * This is the slow, transforming version that needs to double buffer the |
310 | * data as it cannot do the transformations in place without having direct | |
311 | * access to the destination pages of the backing file. | |
1da177e4 | 312 | */ |
aa4d8616 CH |
313 | static int lo_write_transfer(struct loop_device *lo, struct request *rq, |
314 | loff_t pos) | |
1da177e4 | 315 | { |
aa4d8616 | 316 | struct bio_vec bvec, b; |
30112013 | 317 | struct req_iterator iter; |
aa4d8616 | 318 | struct page *page; |
7988613b | 319 | int ret = 0; |
1da177e4 | 320 | |
aa4d8616 CH |
321 | page = alloc_page(GFP_NOIO); |
322 | if (unlikely(!page)) | |
323 | return -ENOMEM; | |
456be148 | 324 | |
30112013 | 325 | rq_for_each_segment(bvec, rq, iter) { |
aa4d8616 CH |
326 | ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, |
327 | bvec.bv_offset, bvec.bv_len, pos >> 9); | |
328 | if (unlikely(ret)) | |
329 | break; | |
330 | ||
331 | b.bv_page = page; | |
332 | b.bv_offset = 0; | |
333 | b.bv_len = bvec.bv_len; | |
334 | ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); | |
1da177e4 LT |
335 | if (ret < 0) |
336 | break; | |
1da177e4 | 337 | } |
aa4d8616 CH |
338 | |
339 | __free_page(page); | |
1da177e4 | 340 | return ret; |
1da177e4 LT |
341 | } |
342 | ||
aa4d8616 CH |
343 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
344 | loff_t pos) | |
1da177e4 | 345 | { |
aa4d8616 CH |
346 | struct bio_vec bvec; |
347 | struct req_iterator iter; | |
348 | struct iov_iter i; | |
349 | ssize_t len; | |
1da177e4 | 350 | |
aa4d8616 | 351 | rq_for_each_segment(bvec, rq, iter) { |
b6207430 | 352 | iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); |
18e9710e | 353 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
354 | if (len < 0) |
355 | return len; | |
1da177e4 | 356 | |
aa4d8616 | 357 | flush_dcache_page(bvec.bv_page); |
fd582140 | 358 | |
aa4d8616 CH |
359 | if (len != bvec.bv_len) { |
360 | struct bio *bio; | |
1da177e4 | 361 | |
aa4d8616 CH |
362 | __rq_for_each_bio(bio, rq) |
363 | zero_fill_bio(bio); | |
364 | break; | |
365 | } | |
366 | cond_resched(); | |
367 | } | |
368 | ||
369 | return 0; | |
fd582140 JA |
370 | } |
371 | ||
aa4d8616 CH |
372 | static int lo_read_transfer(struct loop_device *lo, struct request *rq, |
373 | loff_t pos) | |
1da177e4 | 374 | { |
aa4d8616 CH |
375 | struct bio_vec bvec, b; |
376 | struct req_iterator iter; | |
377 | struct iov_iter i; | |
378 | struct page *page; | |
379 | ssize_t len; | |
380 | int ret = 0; | |
1da177e4 | 381 | |
aa4d8616 CH |
382 | page = alloc_page(GFP_NOIO); |
383 | if (unlikely(!page)) | |
384 | return -ENOMEM; | |
fd582140 | 385 | |
aa4d8616 CH |
386 | rq_for_each_segment(bvec, rq, iter) { |
387 | loff_t offset = pos; | |
fd582140 | 388 | |
aa4d8616 CH |
389 | b.bv_page = page; |
390 | b.bv_offset = 0; | |
391 | b.bv_len = bvec.bv_len; | |
fd582140 | 392 | |
b6207430 | 393 | iov_iter_bvec(&i, READ, &b, 1, b.bv_len); |
18e9710e | 394 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
395 | if (len < 0) { |
396 | ret = len; | |
397 | goto out_free_page; | |
398 | } | |
1da177e4 | 399 | |
aa4d8616 CH |
400 | ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, |
401 | bvec.bv_offset, len, offset >> 9); | |
402 | if (ret) | |
403 | goto out_free_page; | |
1da177e4 | 404 | |
aa4d8616 | 405 | flush_dcache_page(bvec.bv_page); |
306df071 | 406 | |
aa4d8616 | 407 | if (len != bvec.bv_len) { |
30112013 ML |
408 | struct bio *bio; |
409 | ||
410 | __rq_for_each_bio(bio, rq) | |
411 | zero_fill_bio(bio); | |
1da177e4 | 412 | break; |
306df071 | 413 | } |
1da177e4 | 414 | } |
aa4d8616 CH |
415 | |
416 | ret = 0; | |
417 | out_free_page: | |
418 | __free_page(page); | |
419 | return ret; | |
1da177e4 LT |
420 | } |
421 | ||
efcfec57 DW |
422 | static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, |
423 | int mode) | |
cf655d95 ML |
424 | { |
425 | /* | |
efcfec57 DW |
426 | * We use fallocate to manipulate the space mappings used by the image |
427 | * a.k.a. discard/zerorange. However we do not support this if | |
428 | * encryption is enabled, because it may give an attacker useful | |
429 | * information. | |
cf655d95 ML |
430 | */ |
431 | struct file *file = lo->lo_backing_file; | |
c52abf56 | 432 | struct request_queue *q = lo->lo_queue; |
cf655d95 ML |
433 | int ret; |
434 | ||
efcfec57 DW |
435 | mode |= FALLOC_FL_KEEP_SIZE; |
436 | ||
c52abf56 | 437 | if (!blk_queue_discard(q)) { |
cf655d95 ML |
438 | ret = -EOPNOTSUPP; |
439 | goto out; | |
440 | } | |
441 | ||
442 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
443 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
444 | ret = -EIO; | |
445 | out: | |
446 | return ret; | |
447 | } | |
448 | ||
449 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
450 | { | |
451 | struct file *file = lo->lo_backing_file; | |
452 | int ret = vfs_fsync(file, 0); | |
453 | if (unlikely(ret && ret != -EINVAL)) | |
454 | ret = -EIO; | |
455 | ||
456 | return ret; | |
457 | } | |
458 | ||
fe2cb290 | 459 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 460 | { |
fe2cb290 | 461 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 462 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 463 | |
f9de14bc JA |
464 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
465 | req_op(rq) != REQ_OP_READ) { | |
466 | if (cmd->ret < 0) | |
8cd55087 | 467 | ret = errno_to_blk_status(cmd->ret); |
f9de14bc | 468 | goto end_io; |
bc07c10a | 469 | } |
fe2cb290 | 470 | |
f9de14bc JA |
471 | /* |
472 | * Short READ - if we got some data, advance our request and | |
473 | * retry it. If we got no data, end the rest with EIO. | |
474 | */ | |
475 | if (cmd->ret) { | |
476 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
477 | cmd->ret = 0; | |
478 | blk_mq_requeue_request(rq, true); | |
479 | } else { | |
480 | if (cmd->use_aio) { | |
481 | struct bio *bio = rq->bio; | |
482 | ||
483 | while (bio) { | |
484 | zero_fill_bio(bio); | |
485 | bio = bio->bi_next; | |
486 | } | |
487 | } | |
488 | ret = BLK_STS_IOERR; | |
489 | end_io: | |
490 | blk_mq_end_request(rq, ret); | |
491 | } | |
bc07c10a ML |
492 | } |
493 | ||
92d77332 SL |
494 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
495 | { | |
1894e916 JA |
496 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
497 | ||
92d77332 SL |
498 | if (!atomic_dec_and_test(&cmd->ref)) |
499 | return; | |
500 | kfree(cmd->bvec); | |
501 | cmd->bvec = NULL; | |
1894e916 | 502 | blk_mq_complete_request(rq); |
92d77332 SL |
503 | } |
504 | ||
bc07c10a ML |
505 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) |
506 | { | |
507 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 508 | |
d4478e92 SL |
509 | if (cmd->css) |
510 | css_put(cmd->css); | |
fe2cb290 | 511 | cmd->ret = ret; |
92d77332 | 512 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
513 | } |
514 | ||
515 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
516 | loff_t pos, bool rw) | |
517 | { | |
518 | struct iov_iter iter; | |
86af5952 | 519 | struct req_iterator rq_iter; |
bc07c10a | 520 | struct bio_vec *bvec; |
1894e916 | 521 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 522 | struct bio *bio = rq->bio; |
bc07c10a | 523 | struct file *file = lo->lo_backing_file; |
86af5952 | 524 | struct bio_vec tmp; |
40326d8a | 525 | unsigned int offset; |
86af5952 | 526 | int nr_bvec = 0; |
bc07c10a ML |
527 | int ret; |
528 | ||
86af5952 ML |
529 | rq_for_each_bvec(tmp, rq, rq_iter) |
530 | nr_bvec++; | |
531 | ||
40326d8a | 532 | if (rq->bio != rq->biotail) { |
40326d8a | 533 | |
86af5952 | 534 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 535 | GFP_NOIO); |
40326d8a SL |
536 | if (!bvec) |
537 | return -EIO; | |
538 | cmd->bvec = bvec; | |
539 | ||
540 | /* | |
541 | * The bios of the request may be started from the middle of | |
542 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 543 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
544 | * API will take care of all details for us. |
545 | */ | |
86af5952 | 546 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
547 | *bvec = tmp; |
548 | bvec++; | |
549 | } | |
550 | bvec = cmd->bvec; | |
551 | offset = 0; | |
552 | } else { | |
553 | /* | |
554 | * Same here, this bio may be started from the middle of the | |
555 | * 'bvec' because of bio splitting, so offset from the bvec | |
556 | * must be passed to iov iterator | |
557 | */ | |
558 | offset = bio->bi_iter.bi_bvec_done; | |
559 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 560 | } |
92d77332 | 561 | atomic_set(&cmd->ref, 2); |
bc07c10a | 562 | |
b6207430 | 563 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 564 | iter.iov_offset = offset; |
bc07c10a ML |
565 | |
566 | cmd->iocb.ki_pos = pos; | |
567 | cmd->iocb.ki_filp = file; | |
568 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
569 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 570 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
d4478e92 SL |
571 | if (cmd->css) |
572 | kthread_associate_blkcg(cmd->css); | |
bc07c10a ML |
573 | |
574 | if (rw == WRITE) | |
bb7462b6 | 575 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 576 | else |
bb7462b6 | 577 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 578 | |
92d77332 | 579 | lo_rw_aio_do_completion(cmd); |
d4478e92 | 580 | kthread_associate_blkcg(NULL); |
92d77332 | 581 | |
bc07c10a ML |
582 | if (ret != -EIOCBQUEUED) |
583 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
584 | return 0; | |
585 | } | |
586 | ||
c1c87c2b | 587 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
588 | { |
589 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 590 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
591 | |
592 | /* | |
593 | * lo_write_simple and lo_read_simple should have been covered | |
594 | * by io submit style function like lo_rw_aio(), one blocker | |
595 | * is that lo_read_simple() need to call flush_dcache_page after | |
596 | * the page is written from kernel, and it isn't easy to handle | |
597 | * this in io submit style function which submits all segments | |
598 | * of the req at one time. And direct read IO doesn't need to | |
599 | * run flush_dcache_page(). | |
600 | */ | |
c1c87c2b CH |
601 | switch (req_op(rq)) { |
602 | case REQ_OP_FLUSH: | |
603 | return lo_req_flush(lo, rq); | |
19372e27 | 604 | case REQ_OP_WRITE_ZEROES: |
efcfec57 DW |
605 | /* |
606 | * If the caller doesn't want deallocation, call zeroout to | |
607 | * write zeroes the range. Otherwise, punch them out. | |
608 | */ | |
609 | return lo_fallocate(lo, rq, pos, | |
610 | (rq->cmd_flags & REQ_NOUNMAP) ? | |
611 | FALLOC_FL_ZERO_RANGE : | |
612 | FALLOC_FL_PUNCH_HOLE); | |
613 | case REQ_OP_DISCARD: | |
614 | return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); | |
c1c87c2b CH |
615 | case REQ_OP_WRITE: |
616 | if (lo->transfer) | |
617 | return lo_write_transfer(lo, rq, pos); | |
618 | else if (cmd->use_aio) | |
619 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 620 | else |
c1c87c2b CH |
621 | return lo_write_simple(lo, rq, pos); |
622 | case REQ_OP_READ: | |
aa4d8616 | 623 | if (lo->transfer) |
c1c87c2b CH |
624 | return lo_read_transfer(lo, rq, pos); |
625 | else if (cmd->use_aio) | |
626 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 627 | else |
c1c87c2b CH |
628 | return lo_read_simple(lo, rq, pos); |
629 | default: | |
630 | WARN_ON_ONCE(1); | |
631 | return -EIO; | |
aa4d8616 | 632 | } |
1da177e4 LT |
633 | } |
634 | ||
2e5ab5f3 ML |
635 | static inline void loop_update_dio(struct loop_device *lo) |
636 | { | |
637 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
638 | lo->use_dio); | |
639 | } | |
640 | ||
06f0e9e6 ML |
641 | static void loop_reread_partitions(struct loop_device *lo, |
642 | struct block_device *bdev) | |
643 | { | |
644 | int rc; | |
645 | ||
f0b870df CH |
646 | mutex_lock(&bdev->bd_mutex); |
647 | rc = bdev_disk_changed(bdev, false); | |
648 | mutex_unlock(&bdev->bd_mutex); | |
06f0e9e6 ML |
649 | if (rc) |
650 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
651 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
652 | } | |
653 | ||
d2ac838e TT |
654 | static inline int is_loop_device(struct file *file) |
655 | { | |
656 | struct inode *i = file->f_mapping->host; | |
657 | ||
658 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
659 | } | |
660 | ||
661 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
662 | { | |
663 | struct inode *inode = file->f_mapping->host; | |
664 | struct file *f = file; | |
665 | ||
666 | /* Avoid recursion */ | |
667 | while (is_loop_device(f)) { | |
668 | struct loop_device *l; | |
669 | ||
670 | if (f->f_mapping->host->i_bdev == bdev) | |
671 | return -EBADF; | |
672 | ||
673 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
f7c8a412 | 674 | if (l->lo_state != Lo_bound) { |
d2ac838e TT |
675 | return -EINVAL; |
676 | } | |
677 | f = l->lo_backing_file; | |
678 | } | |
679 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
680 | return -EINVAL; | |
681 | return 0; | |
682 | } | |
683 | ||
1da177e4 LT |
684 | /* |
685 | * loop_change_fd switched the backing store of a loopback device to | |
686 | * a new file. This is useful for operating system installers to free up | |
687 | * the original file and in High Availability environments to switch to | |
688 | * an alternative location for the content in case of server meltdown. | |
689 | * This can only work if the loop device is used read-only, and if the | |
690 | * new backing store is the same size and type as the old backing store. | |
691 | */ | |
bb214884 AV |
692 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
693 | unsigned int arg) | |
1da177e4 | 694 | { |
1dded9ac | 695 | struct file *file = NULL, *old_file; |
1da177e4 | 696 | int error; |
85b0a54a | 697 | bool partscan; |
1da177e4 | 698 | |
c28445fa | 699 | error = mutex_lock_killable(&loop_ctl_mutex); |
c3710770 JK |
700 | if (error) |
701 | return error; | |
1da177e4 LT |
702 | error = -ENXIO; |
703 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 704 | goto out_err; |
1da177e4 LT |
705 | |
706 | /* the loop device has to be read-only */ | |
707 | error = -EINVAL; | |
708 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 709 | goto out_err; |
1da177e4 LT |
710 | |
711 | error = -EBADF; | |
712 | file = fget(arg); | |
713 | if (!file) | |
1dded9ac | 714 | goto out_err; |
1da177e4 | 715 | |
d2ac838e TT |
716 | error = loop_validate_file(file, bdev); |
717 | if (error) | |
1dded9ac | 718 | goto out_err; |
d2ac838e | 719 | |
1da177e4 LT |
720 | old_file = lo->lo_backing_file; |
721 | ||
722 | error = -EINVAL; | |
723 | ||
1da177e4 LT |
724 | /* size of the new backing store needs to be the same */ |
725 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 726 | goto out_err; |
1da177e4 LT |
727 | |
728 | /* and ... switch */ | |
43cade80 OS |
729 | blk_mq_freeze_queue(lo->lo_queue); |
730 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
731 | lo->lo_backing_file = file; | |
732 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
733 | mapping_set_gfp_mask(file->f_mapping, | |
734 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
735 | loop_update_dio(lo); | |
736 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 737 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c3710770 | 738 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
739 | /* |
740 | * We must drop file reference outside of loop_ctl_mutex as dropping | |
741 | * the file ref can take bd_mutex which creates circular locking | |
742 | * dependency. | |
743 | */ | |
744 | fput(old_file); | |
85b0a54a JK |
745 | if (partscan) |
746 | loop_reread_partitions(lo, bdev); | |
1da177e4 LT |
747 | return 0; |
748 | ||
1dded9ac | 749 | out_err: |
c3710770 | 750 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
751 | if (file) |
752 | fput(file); | |
1da177e4 LT |
753 | return error; |
754 | } | |
755 | ||
ee862730 MB |
756 | /* loop sysfs attributes */ |
757 | ||
758 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
759 | ssize_t (*callback)(struct loop_device *, char *)) | |
760 | { | |
34dd82af KS |
761 | struct gendisk *disk = dev_to_disk(dev); |
762 | struct loop_device *lo = disk->private_data; | |
ee862730 | 763 | |
34dd82af | 764 | return callback(lo, page); |
ee862730 MB |
765 | } |
766 | ||
767 | #define LOOP_ATTR_RO(_name) \ | |
768 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
769 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
770 | struct device_attribute *attr, char *b) \ | |
771 | { \ | |
772 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
773 | } \ | |
774 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 775 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
776 | |
777 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
778 | { | |
779 | ssize_t ret; | |
780 | char *p = NULL; | |
781 | ||
05eb0f25 | 782 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 783 | if (lo->lo_backing_file) |
9bf39ab2 | 784 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 785 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
786 | |
787 | if (IS_ERR_OR_NULL(p)) | |
788 | ret = PTR_ERR(p); | |
789 | else { | |
790 | ret = strlen(p); | |
791 | memmove(buf, p, ret); | |
792 | buf[ret++] = '\n'; | |
793 | buf[ret] = 0; | |
794 | } | |
795 | ||
796 | return ret; | |
797 | } | |
798 | ||
799 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
800 | { | |
801 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
802 | } | |
803 | ||
804 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
805 | { | |
806 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
807 | } | |
808 | ||
809 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
810 | { | |
811 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
812 | ||
813 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
814 | } | |
815 | ||
e03c8dd1 KS |
816 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
817 | { | |
818 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
819 | ||
820 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
821 | } | |
822 | ||
2e5ab5f3 ML |
823 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
824 | { | |
825 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
826 | ||
827 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
828 | } | |
829 | ||
ee862730 MB |
830 | LOOP_ATTR_RO(backing_file); |
831 | LOOP_ATTR_RO(offset); | |
832 | LOOP_ATTR_RO(sizelimit); | |
833 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 834 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 835 | LOOP_ATTR_RO(dio); |
ee862730 MB |
836 | |
837 | static struct attribute *loop_attrs[] = { | |
838 | &loop_attr_backing_file.attr, | |
839 | &loop_attr_offset.attr, | |
840 | &loop_attr_sizelimit.attr, | |
841 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 842 | &loop_attr_partscan.attr, |
2e5ab5f3 | 843 | &loop_attr_dio.attr, |
ee862730 MB |
844 | NULL, |
845 | }; | |
846 | ||
847 | static struct attribute_group loop_attribute_group = { | |
848 | .name = "loop", | |
849 | .attrs= loop_attrs, | |
850 | }; | |
851 | ||
d3349b6b | 852 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 853 | { |
d3349b6b TH |
854 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
855 | &loop_attribute_group); | |
ee862730 MB |
856 | } |
857 | ||
858 | static void loop_sysfs_exit(struct loop_device *lo) | |
859 | { | |
d3349b6b TH |
860 | if (lo->sysfs_inited) |
861 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
862 | &loop_attribute_group); | |
ee862730 MB |
863 | } |
864 | ||
dfaa2ef6 LC |
865 | static void loop_config_discard(struct loop_device *lo) |
866 | { | |
867 | struct file *file = lo->lo_backing_file; | |
868 | struct inode *inode = file->f_mapping->host; | |
869 | struct request_queue *q = lo->lo_queue; | |
870 | ||
c52abf56 EG |
871 | /* |
872 | * If the backing device is a block device, mirror its zeroing | |
873 | * capability. Set the discard sectors to the block device's zeroing | |
874 | * capabilities because loop discards result in blkdev_issue_zeroout(), | |
875 | * not blkdev_issue_discard(). This maintains consistent behavior with | |
876 | * file-backed loop devices: discarded regions read back as zero. | |
877 | */ | |
878 | if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { | |
879 | struct request_queue *backingq; | |
880 | ||
881 | backingq = bdev_get_queue(inode->i_bdev); | |
882 | blk_queue_max_discard_sectors(q, | |
883 | backingq->limits.max_write_zeroes_sectors); | |
884 | ||
885 | blk_queue_max_write_zeroes_sectors(q, | |
886 | backingq->limits.max_write_zeroes_sectors); | |
887 | ||
dfaa2ef6 LC |
888 | /* |
889 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 890 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
891 | * encryption is enabled, because it may give an attacker |
892 | * useful information. | |
893 | */ | |
c52abf56 | 894 | } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { |
dfaa2ef6 LC |
895 | q->limits.discard_granularity = 0; |
896 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 897 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 898 | blk_queue_max_write_zeroes_sectors(q, 0); |
dfaa2ef6 | 899 | |
c52abf56 EG |
900 | } else { |
901 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
902 | q->limits.discard_alignment = 0; | |
f2c6df7d | 903 | |
c52abf56 EG |
904 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
905 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
906 | } | |
907 | ||
908 | if (q->limits.max_write_zeroes_sectors) | |
909 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); | |
910 | else | |
911 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); | |
dfaa2ef6 LC |
912 | } |
913 | ||
e03a3d7a ML |
914 | static void loop_unprepare_queue(struct loop_device *lo) |
915 | { | |
3989144f | 916 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
917 | kthread_stop(lo->worker_task); |
918 | } | |
919 | ||
b2ee7d46 N |
920 | static int loop_kthread_worker_fn(void *worker_ptr) |
921 | { | |
d0a255e7 | 922 | current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO; |
b2ee7d46 N |
923 | return kthread_worker_fn(worker_ptr); |
924 | } | |
925 | ||
e03a3d7a ML |
926 | static int loop_prepare_queue(struct loop_device *lo) |
927 | { | |
3989144f | 928 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 929 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
930 | &lo->worker, "loop%d", lo->lo_number); |
931 | if (IS_ERR(lo->worker_task)) | |
932 | return -ENOMEM; | |
933 | set_user_nice(lo->worker_task, MIN_NICE); | |
934 | return 0; | |
935 | } | |
936 | ||
56a85fd8 HH |
937 | static void loop_update_rotational(struct loop_device *lo) |
938 | { | |
939 | struct file *file = lo->lo_backing_file; | |
940 | struct inode *file_inode = file->f_mapping->host; | |
941 | struct block_device *file_bdev = file_inode->i_sb->s_bdev; | |
942 | struct request_queue *q = lo->lo_queue; | |
943 | bool nonrot = true; | |
944 | ||
945 | /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ | |
946 | if (file_bdev) | |
947 | nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); | |
948 | ||
949 | if (nonrot) | |
950 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q); | |
951 | else | |
952 | blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); | |
953 | } | |
954 | ||
bb214884 | 955 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
956 | struct block_device *bdev, unsigned int arg) |
957 | { | |
d2ac838e | 958 | struct file *file; |
1da177e4 LT |
959 | struct inode *inode; |
960 | struct address_space *mapping; | |
89e524c0 | 961 | struct block_device *claimed_bdev = NULL; |
1da177e4 LT |
962 | int lo_flags = 0; |
963 | int error; | |
964 | loff_t size; | |
85b0a54a | 965 | bool partscan; |
1da177e4 LT |
966 | |
967 | /* This is safe, since we have a reference from open(). */ | |
968 | __module_get(THIS_MODULE); | |
969 | ||
970 | error = -EBADF; | |
971 | file = fget(arg); | |
972 | if (!file) | |
973 | goto out; | |
974 | ||
33ec3e53 JK |
975 | /* |
976 | * If we don't hold exclusive handle for the device, upgrade to it | |
977 | * here to avoid changing device under exclusive owner. | |
978 | */ | |
979 | if (!(mode & FMODE_EXCL)) { | |
89e524c0 JK |
980 | claimed_bdev = bd_start_claiming(bdev, loop_set_fd); |
981 | if (IS_ERR(claimed_bdev)) { | |
982 | error = PTR_ERR(claimed_bdev); | |
33ec3e53 | 983 | goto out_putf; |
89e524c0 | 984 | } |
33ec3e53 JK |
985 | } |
986 | ||
c28445fa | 987 | error = mutex_lock_killable(&loop_ctl_mutex); |
757ecf40 | 988 | if (error) |
33ec3e53 | 989 | goto out_bdev; |
757ecf40 | 990 | |
1da177e4 LT |
991 | error = -EBUSY; |
992 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 993 | goto out_unlock; |
1da177e4 | 994 | |
d2ac838e TT |
995 | error = loop_validate_file(file, bdev); |
996 | if (error) | |
757ecf40 | 997 | goto out_unlock; |
1da177e4 LT |
998 | |
999 | mapping = file->f_mapping; | |
1000 | inode = mapping->host; | |
1001 | ||
456be148 | 1002 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 1003 | !file->f_op->write_iter) |
456be148 | 1004 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 1005 | |
456be148 | 1006 | error = -EFBIG; |
1da177e4 | 1007 | size = get_loop_size(lo, file); |
456be148 | 1008 | if ((loff_t)(sector_t)size != size) |
757ecf40 | 1009 | goto out_unlock; |
e03a3d7a ML |
1010 | error = loop_prepare_queue(lo); |
1011 | if (error) | |
757ecf40 | 1012 | goto out_unlock; |
1da177e4 | 1013 | |
456be148 | 1014 | error = 0; |
1da177e4 LT |
1015 | |
1016 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
1017 | ||
2e5ab5f3 | 1018 | lo->use_dio = false; |
1da177e4 LT |
1019 | lo->lo_device = bdev; |
1020 | lo->lo_flags = lo_flags; | |
1021 | lo->lo_backing_file = file; | |
aa4d8616 | 1022 | lo->transfer = NULL; |
1da177e4 LT |
1023 | lo->ioctl = NULL; |
1024 | lo->lo_sizelimit = 0; | |
1025 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
1026 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1027 | ||
68db1961 | 1028 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 1029 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 1030 | |
85560117 MC |
1031 | if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) { |
1032 | /* In case of direct I/O, match underlying block size */ | |
1033 | unsigned short bsize = bdev_logical_block_size( | |
1034 | inode->i_sb->s_bdev); | |
1035 | ||
1036 | blk_queue_logical_block_size(lo->lo_queue, bsize); | |
1037 | blk_queue_physical_block_size(lo->lo_queue, bsize); | |
1038 | blk_queue_io_min(lo->lo_queue, bsize); | |
1039 | } | |
1040 | ||
56a85fd8 | 1041 | loop_update_rotational(lo); |
2e5ab5f3 | 1042 | loop_update_dio(lo); |
73285082 | 1043 | set_capacity(lo->lo_disk, size); |
1da177e4 | 1044 | bd_set_size(bdev, size << 9); |
ee862730 | 1045 | loop_sysfs_init(lo); |
c3473c63 DZ |
1046 | /* let user-space know about the new size */ |
1047 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 1048 | |
8a0740c4 OS |
1049 | set_blocksize(bdev, S_ISBLK(inode->i_mode) ? |
1050 | block_size(inode->i_bdev) : PAGE_SIZE); | |
1da177e4 | 1051 | |
6c997918 | 1052 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1053 | if (part_shift) |
1054 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 1055 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c1681bf8 AP |
1056 | |
1057 | /* Grab the block_device to prevent its destruction after we | |
a2505b79 | 1058 | * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). |
c1681bf8 AP |
1059 | */ |
1060 | bdgrab(bdev); | |
757ecf40 | 1061 | mutex_unlock(&loop_ctl_mutex); |
85b0a54a JK |
1062 | if (partscan) |
1063 | loop_reread_partitions(lo, bdev); | |
89e524c0 JK |
1064 | if (claimed_bdev) |
1065 | bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); | |
1da177e4 LT |
1066 | return 0; |
1067 | ||
757ecf40 JK |
1068 | out_unlock: |
1069 | mutex_unlock(&loop_ctl_mutex); | |
33ec3e53 | 1070 | out_bdev: |
89e524c0 JK |
1071 | if (claimed_bdev) |
1072 | bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); | |
757ecf40 | 1073 | out_putf: |
1da177e4 | 1074 | fput(file); |
757ecf40 | 1075 | out: |
1da177e4 LT |
1076 | /* This is safe: open() is still holding a reference. */ |
1077 | module_put(THIS_MODULE); | |
1078 | return error; | |
1079 | } | |
1080 | ||
1081 | static int | |
1082 | loop_release_xfer(struct loop_device *lo) | |
1083 | { | |
1084 | int err = 0; | |
1085 | struct loop_func_table *xfer = lo->lo_encryption; | |
1086 | ||
1087 | if (xfer) { | |
1088 | if (xfer->release) | |
1089 | err = xfer->release(lo); | |
1090 | lo->transfer = NULL; | |
1091 | lo->lo_encryption = NULL; | |
1092 | module_put(xfer->owner); | |
1093 | } | |
1094 | return err; | |
1095 | } | |
1096 | ||
1097 | static int | |
1098 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1099 | const struct loop_info64 *i) | |
1100 | { | |
1101 | int err = 0; | |
1102 | ||
1103 | if (xfer) { | |
1104 | struct module *owner = xfer->owner; | |
1105 | ||
1106 | if (!try_module_get(owner)) | |
1107 | return -EINVAL; | |
1108 | if (xfer->init) | |
1109 | err = xfer->init(lo, i); | |
1110 | if (err) | |
1111 | module_put(owner); | |
1112 | else | |
1113 | lo->lo_encryption = xfer; | |
1114 | } | |
1115 | return err; | |
1116 | } | |
1117 | ||
0da03cab | 1118 | static int __loop_clr_fd(struct loop_device *lo, bool release) |
1da177e4 | 1119 | { |
7ccd0791 | 1120 | struct file *filp = NULL; |
b4e3ca1a | 1121 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1122 | struct block_device *bdev = lo->lo_device; |
7ccd0791 | 1123 | int err = 0; |
0da03cab JK |
1124 | bool partscan = false; |
1125 | int lo_number; | |
1da177e4 | 1126 | |
7ccd0791 JK |
1127 | mutex_lock(&loop_ctl_mutex); |
1128 | if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { | |
1129 | err = -ENXIO; | |
1130 | goto out_unlock; | |
1131 | } | |
1da177e4 | 1132 | |
7ccd0791 JK |
1133 | filp = lo->lo_backing_file; |
1134 | if (filp == NULL) { | |
1135 | err = -EINVAL; | |
1136 | goto out_unlock; | |
1137 | } | |
1da177e4 | 1138 | |
f8933667 ML |
1139 | /* freeze request queue during the transition */ |
1140 | blk_mq_freeze_queue(lo->lo_queue); | |
1141 | ||
1da177e4 | 1142 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 1143 | lo->lo_backing_file = NULL; |
05eb0f25 | 1144 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1145 | |
1146 | loop_release_xfer(lo); | |
1147 | lo->transfer = NULL; | |
1148 | lo->ioctl = NULL; | |
1149 | lo->lo_device = NULL; | |
1150 | lo->lo_encryption = NULL; | |
1151 | lo->lo_offset = 0; | |
1152 | lo->lo_sizelimit = 0; | |
1153 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1154 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1155 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1156 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1157 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1158 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1159 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1160 | if (bdev) { |
1161 | bdput(bdev); | |
bb214884 | 1162 | invalidate_bdev(bdev); |
eedffa28 | 1163 | bdev->bd_inode->i_mapping->wb_err = 0; |
c1681bf8 | 1164 | } |
73285082 | 1165 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1166 | loop_sysfs_exit(lo); |
c3473c63 | 1167 | if (bdev) { |
bb214884 | 1168 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1169 | /* let user-space know about this change */ |
1170 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1171 | } | |
1da177e4 | 1172 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1da177e4 LT |
1173 | /* This is safe: open() is still holding a reference. */ |
1174 | module_put(THIS_MODULE); | |
f8933667 ML |
1175 | blk_mq_unfreeze_queue(lo->lo_queue); |
1176 | ||
0da03cab JK |
1177 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; |
1178 | lo_number = lo->lo_number; | |
0da03cab JK |
1179 | loop_unprepare_queue(lo); |
1180 | out_unlock: | |
1181 | mutex_unlock(&loop_ctl_mutex); | |
1182 | if (partscan) { | |
d57f3374 JK |
1183 | /* |
1184 | * bd_mutex has been held already in release path, so don't | |
1185 | * acquire it if this function is called in such case. | |
1186 | * | |
1187 | * If the reread partition isn't from release path, lo_refcnt | |
1188 | * must be at least one and it can only become zero when the | |
1189 | * current holder is released. | |
1190 | */ | |
f0b870df CH |
1191 | if (!release) |
1192 | mutex_lock(&bdev->bd_mutex); | |
1193 | err = bdev_disk_changed(bdev, false); | |
1194 | if (!release) | |
1195 | mutex_unlock(&bdev->bd_mutex); | |
40853d6f DZ |
1196 | if (err) |
1197 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
1198 | __func__, lo_number, err); | |
d57f3374 JK |
1199 | /* Device is gone, no point in returning error */ |
1200 | err = 0; | |
1201 | } | |
758a58d0 DZ |
1202 | |
1203 | /* | |
1204 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1205 | * finished. | |
1206 | * | |
1207 | * There cannot be anybody else entering __loop_clr_fd() as | |
1208 | * lo->lo_backing_file is already cleared and Lo_rundown state | |
1209 | * protects us from all the other places trying to change the 'lo' | |
1210 | * device. | |
1211 | */ | |
1212 | mutex_lock(&loop_ctl_mutex); | |
1213 | lo->lo_flags = 0; | |
1214 | if (!part_shift) | |
1215 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1216 | lo->lo_state = Lo_unbound; | |
1217 | mutex_unlock(&loop_ctl_mutex); | |
1218 | ||
f028f3b2 | 1219 | /* |
310ca162 TH |
1220 | * Need not hold loop_ctl_mutex to fput backing file. |
1221 | * Calling fput holding loop_ctl_mutex triggers a circular | |
f028f3b2 | 1222 | * lock dependency possibility warning as fput can take |
310ca162 | 1223 | * bd_mutex which is usually taken before loop_ctl_mutex. |
f028f3b2 | 1224 | */ |
7ccd0791 JK |
1225 | if (filp) |
1226 | fput(filp); | |
1227 | return err; | |
1da177e4 LT |
1228 | } |
1229 | ||
a2505b79 JK |
1230 | static int loop_clr_fd(struct loop_device *lo) |
1231 | { | |
7ccd0791 JK |
1232 | int err; |
1233 | ||
c28445fa | 1234 | err = mutex_lock_killable(&loop_ctl_mutex); |
7ccd0791 JK |
1235 | if (err) |
1236 | return err; | |
1237 | if (lo->lo_state != Lo_bound) { | |
1238 | mutex_unlock(&loop_ctl_mutex); | |
a2505b79 | 1239 | return -ENXIO; |
7ccd0791 | 1240 | } |
a2505b79 JK |
1241 | /* |
1242 | * If we've explicitly asked to tear down the loop device, | |
1243 | * and it has an elevated reference count, set it for auto-teardown when | |
1244 | * the last reference goes away. This stops $!~#$@ udev from | |
1245 | * preventing teardown because it decided that it needs to run blkid on | |
1246 | * the loopback device whenever they appear. xfstests is notorious for | |
1247 | * failing tests because blkid via udev races with a losetup | |
1248 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1249 | * command to fail with EBUSY. | |
1250 | */ | |
1251 | if (atomic_read(&lo->lo_refcnt) > 1) { | |
1252 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
1253 | mutex_unlock(&loop_ctl_mutex); | |
1254 | return 0; | |
1255 | } | |
1256 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1257 | mutex_unlock(&loop_ctl_mutex); |
a2505b79 | 1258 | |
0da03cab | 1259 | return __loop_clr_fd(lo, false); |
a2505b79 JK |
1260 | } |
1261 | ||
1da177e4 LT |
1262 | static int |
1263 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1264 | { | |
1265 | int err; | |
1266 | struct loop_func_table *xfer; | |
e4849737 | 1267 | kuid_t uid = current_uid(); |
85b0a54a JK |
1268 | struct block_device *bdev; |
1269 | bool partscan = false; | |
1da177e4 | 1270 | |
c28445fa | 1271 | err = mutex_lock_killable(&loop_ctl_mutex); |
550df5fd JK |
1272 | if (err) |
1273 | return err; | |
b0fafa81 | 1274 | if (lo->lo_encrypt_key_size && |
e4849737 | 1275 | !uid_eq(lo->lo_key_owner, uid) && |
550df5fd JK |
1276 | !capable(CAP_SYS_ADMIN)) { |
1277 | err = -EPERM; | |
1278 | goto out_unlock; | |
1279 | } | |
1280 | if (lo->lo_state != Lo_bound) { | |
1281 | err = -ENXIO; | |
1282 | goto out_unlock; | |
1283 | } | |
1284 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { | |
1285 | err = -EINVAL; | |
1286 | goto out_unlock; | |
1287 | } | |
1da177e4 | 1288 | |
5db470e2 JK |
1289 | if (lo->lo_offset != info->lo_offset || |
1290 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1291 | sync_blockdev(lo->lo_device); | |
1292 | kill_bdev(lo->lo_device); | |
1293 | } | |
1294 | ||
ecdd0959 ML |
1295 | /* I/O need to be drained during transfer transition */ |
1296 | blk_mq_freeze_queue(lo->lo_queue); | |
1297 | ||
1da177e4 LT |
1298 | err = loop_release_xfer(lo); |
1299 | if (err) | |
550df5fd | 1300 | goto out_unfreeze; |
1da177e4 LT |
1301 | |
1302 | if (info->lo_encrypt_type) { | |
1303 | unsigned int type = info->lo_encrypt_type; | |
1304 | ||
1e047eaa TH |
1305 | if (type >= MAX_LO_CRYPT) { |
1306 | err = -EINVAL; | |
550df5fd | 1307 | goto out_unfreeze; |
1e047eaa | 1308 | } |
1da177e4 | 1309 | xfer = xfer_funcs[type]; |
1e047eaa TH |
1310 | if (xfer == NULL) { |
1311 | err = -EINVAL; | |
550df5fd | 1312 | goto out_unfreeze; |
1e047eaa | 1313 | } |
1da177e4 LT |
1314 | } else |
1315 | xfer = NULL; | |
1316 | ||
1317 | err = loop_init_xfer(lo, xfer, info); | |
1318 | if (err) | |
550df5fd | 1319 | goto out_unfreeze; |
1da177e4 LT |
1320 | |
1321 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea | 1322 | lo->lo_sizelimit != info->lo_sizelimit) { |
5db470e2 JK |
1323 | /* kill_bdev should have truncated all the pages */ |
1324 | if (lo->lo_device->bd_inode->i_mapping->nrpages) { | |
1325 | err = -EAGAIN; | |
1326 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1327 | __func__, lo->lo_number, lo->lo_file_name, | |
1328 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1329 | goto out_unfreeze; | |
1330 | } | |
1e6ec9ea | 1331 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { |
ecdd0959 | 1332 | err = -EFBIG; |
550df5fd | 1333 | goto out_unfreeze; |
ecdd0959 | 1334 | } |
b040ad9c | 1335 | } |
541c742a | 1336 | |
1da177e4 LT |
1337 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); |
1338 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1339 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1340 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1341 | ||
1342 | if (!xfer) | |
1343 | xfer = &none_funcs; | |
1344 | lo->transfer = xfer->transfer; | |
1345 | lo->ioctl = xfer->ioctl; | |
1346 | ||
96c58655 DW |
1347 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1348 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1349 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1350 | ||
1da177e4 LT |
1351 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1352 | lo->lo_init[0] = info->lo_init[0]; | |
1353 | lo->lo_init[1] = info->lo_init[1]; | |
1354 | if (info->lo_encrypt_key_size) { | |
1355 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1356 | info->lo_encrypt_key_size); | |
b0fafa81 | 1357 | lo->lo_key_owner = uid; |
aa4d8616 | 1358 | } |
1da177e4 | 1359 | |
7c5014b0 MC |
1360 | loop_config_discard(lo); |
1361 | ||
2e5ab5f3 ML |
1362 | /* update dio if lo_offset or transfer is changed */ |
1363 | __loop_update_dio(lo, lo->use_dio); | |
1364 | ||
550df5fd | 1365 | out_unfreeze: |
ecdd0959 | 1366 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 OS |
1367 | |
1368 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1369 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1370 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1371 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
85b0a54a JK |
1372 | bdev = lo->lo_device; |
1373 | partscan = true; | |
e02898b4 | 1374 | } |
550df5fd JK |
1375 | out_unlock: |
1376 | mutex_unlock(&loop_ctl_mutex); | |
85b0a54a JK |
1377 | if (partscan) |
1378 | loop_reread_partitions(lo, bdev); | |
e02898b4 | 1379 | |
ecdd0959 | 1380 | return err; |
1da177e4 LT |
1381 | } |
1382 | ||
1383 | static int | |
1384 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1385 | { | |
b1ab5fa3 | 1386 | struct path path; |
1da177e4 | 1387 | struct kstat stat; |
2d1d4c1e | 1388 | int ret; |
1da177e4 | 1389 | |
c28445fa | 1390 | ret = mutex_lock_killable(&loop_ctl_mutex); |
4a5ce9ba JK |
1391 | if (ret) |
1392 | return ret; | |
2d1d4c1e | 1393 | if (lo->lo_state != Lo_bound) { |
310ca162 | 1394 | mutex_unlock(&loop_ctl_mutex); |
1da177e4 | 1395 | return -ENXIO; |
2d1d4c1e OS |
1396 | } |
1397 | ||
1da177e4 LT |
1398 | memset(info, 0, sizeof(*info)); |
1399 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1400 | info->lo_offset = lo->lo_offset; |
1401 | info->lo_sizelimit = lo->lo_sizelimit; | |
1402 | info->lo_flags = lo->lo_flags; | |
1403 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1404 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1405 | info->lo_encrypt_type = | |
1406 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1407 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1408 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1409 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1410 | lo->lo_encrypt_key_size); | |
1411 | } | |
2d1d4c1e | 1412 | |
310ca162 | 1413 | /* Drop loop_ctl_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1414 | path = lo->lo_backing_file->f_path; |
1415 | path_get(&path); | |
310ca162 | 1416 | mutex_unlock(&loop_ctl_mutex); |
b1ab5fa3 | 1417 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1418 | if (!ret) { |
1419 | info->lo_device = huge_encode_dev(stat.dev); | |
1420 | info->lo_inode = stat.ino; | |
1421 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1422 | } | |
b1ab5fa3 | 1423 | path_put(&path); |
2d1d4c1e | 1424 | return ret; |
1da177e4 LT |
1425 | } |
1426 | ||
1427 | static void | |
1428 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1429 | { | |
1430 | memset(info64, 0, sizeof(*info64)); | |
1431 | info64->lo_number = info->lo_number; | |
1432 | info64->lo_device = info->lo_device; | |
1433 | info64->lo_inode = info->lo_inode; | |
1434 | info64->lo_rdevice = info->lo_rdevice; | |
1435 | info64->lo_offset = info->lo_offset; | |
1436 | info64->lo_sizelimit = 0; | |
1437 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1438 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1439 | info64->lo_flags = info->lo_flags; | |
1440 | info64->lo_init[0] = info->lo_init[0]; | |
1441 | info64->lo_init[1] = info->lo_init[1]; | |
1442 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1443 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1444 | else | |
1445 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1446 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1447 | } | |
1448 | ||
1449 | static int | |
1450 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1451 | { | |
1452 | memset(info, 0, sizeof(*info)); | |
1453 | info->lo_number = info64->lo_number; | |
1454 | info->lo_device = info64->lo_device; | |
1455 | info->lo_inode = info64->lo_inode; | |
1456 | info->lo_rdevice = info64->lo_rdevice; | |
1457 | info->lo_offset = info64->lo_offset; | |
1458 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1459 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1460 | info->lo_flags = info64->lo_flags; | |
1461 | info->lo_init[0] = info64->lo_init[0]; | |
1462 | info->lo_init[1] = info64->lo_init[1]; | |
1463 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1464 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1465 | else | |
1466 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1467 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1468 | ||
1469 | /* error in case values were truncated */ | |
1470 | if (info->lo_device != info64->lo_device || | |
1471 | info->lo_rdevice != info64->lo_rdevice || | |
1472 | info->lo_inode != info64->lo_inode || | |
1473 | info->lo_offset != info64->lo_offset) | |
1474 | return -EOVERFLOW; | |
1475 | ||
1476 | return 0; | |
1477 | } | |
1478 | ||
1479 | static int | |
1480 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1481 | { | |
1482 | struct loop_info info; | |
1483 | struct loop_info64 info64; | |
1484 | ||
1485 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1486 | return -EFAULT; | |
1487 | loop_info64_from_old(&info, &info64); | |
1488 | return loop_set_status(lo, &info64); | |
1489 | } | |
1490 | ||
1491 | static int | |
1492 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1493 | { | |
1494 | struct loop_info64 info64; | |
1495 | ||
1496 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1497 | return -EFAULT; | |
1498 | return loop_set_status(lo, &info64); | |
1499 | } | |
1500 | ||
1501 | static int | |
1502 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1503 | struct loop_info info; | |
1504 | struct loop_info64 info64; | |
bdac616d | 1505 | int err; |
1da177e4 | 1506 | |
4a5ce9ba | 1507 | if (!arg) |
bdac616d | 1508 | return -EINVAL; |
bdac616d | 1509 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1510 | if (!err) |
1511 | err = loop_info64_to_old(&info64, &info); | |
1512 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1513 | err = -EFAULT; | |
1514 | ||
1515 | return err; | |
1516 | } | |
1517 | ||
1518 | static int | |
1519 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1520 | struct loop_info64 info64; | |
bdac616d | 1521 | int err; |
1da177e4 | 1522 | |
4a5ce9ba | 1523 | if (!arg) |
bdac616d | 1524 | return -EINVAL; |
bdac616d | 1525 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1526 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1527 | err = -EFAULT; | |
1528 | ||
1529 | return err; | |
1530 | } | |
1531 | ||
51001b7d | 1532 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1533 | { |
53d66608 | 1534 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1535 | return -ENXIO; |
53d66608 | 1536 | |
1e6ec9ea | 1537 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1538 | } |
1539 | ||
ab1cb278 ML |
1540 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1541 | { | |
1542 | int error = -ENXIO; | |
1543 | if (lo->lo_state != Lo_bound) | |
1544 | goto out; | |
1545 | ||
1546 | __loop_update_dio(lo, !!arg); | |
1547 | if (lo->use_dio == !!arg) | |
1548 | return 0; | |
1549 | error = -EINVAL; | |
1550 | out: | |
1551 | return error; | |
1552 | } | |
1553 | ||
89e4fdec OS |
1554 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1555 | { | |
5db470e2 JK |
1556 | int err = 0; |
1557 | ||
89e4fdec OS |
1558 | if (lo->lo_state != Lo_bound) |
1559 | return -ENXIO; | |
1560 | ||
1561 | if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) | |
1562 | return -EINVAL; | |
1563 | ||
7e81f99a MC |
1564 | if (lo->lo_queue->limits.logical_block_size == arg) |
1565 | return 0; | |
1566 | ||
1567 | sync_blockdev(lo->lo_device); | |
1568 | kill_bdev(lo->lo_device); | |
5db470e2 | 1569 | |
89e4fdec OS |
1570 | blk_mq_freeze_queue(lo->lo_queue); |
1571 | ||
5db470e2 | 1572 | /* kill_bdev should have truncated all the pages */ |
7e81f99a | 1573 | if (lo->lo_device->bd_inode->i_mapping->nrpages) { |
5db470e2 JK |
1574 | err = -EAGAIN; |
1575 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1576 | __func__, lo->lo_number, lo->lo_file_name, | |
1577 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1578 | goto out_unfreeze; | |
1579 | } | |
1580 | ||
89e4fdec | 1581 | blk_queue_logical_block_size(lo->lo_queue, arg); |
bf093753 OS |
1582 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1583 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec | 1584 | loop_update_dio(lo); |
5db470e2 | 1585 | out_unfreeze: |
89e4fdec OS |
1586 | blk_mq_unfreeze_queue(lo->lo_queue); |
1587 | ||
5db470e2 | 1588 | return err; |
89e4fdec OS |
1589 | } |
1590 | ||
a1316544 JK |
1591 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1592 | unsigned long arg) | |
1da177e4 | 1593 | { |
1da177e4 LT |
1594 | int err; |
1595 | ||
c28445fa | 1596 | err = mutex_lock_killable(&loop_ctl_mutex); |
3148ffbd | 1597 | if (err) |
a1316544 JK |
1598 | return err; |
1599 | switch (cmd) { | |
1600 | case LOOP_SET_CAPACITY: | |
1601 | err = loop_set_capacity(lo); | |
1602 | break; | |
1603 | case LOOP_SET_DIRECT_IO: | |
1604 | err = loop_set_dio(lo, arg); | |
1605 | break; | |
1606 | case LOOP_SET_BLOCK_SIZE: | |
1607 | err = loop_set_block_size(lo, arg); | |
1608 | break; | |
1609 | default: | |
1610 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1611 | } | |
1612 | mutex_unlock(&loop_ctl_mutex); | |
1613 | return err; | |
1614 | } | |
1615 | ||
1616 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, | |
1617 | unsigned int cmd, unsigned long arg) | |
1618 | { | |
1619 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1620 | int err; | |
3148ffbd | 1621 | |
1da177e4 LT |
1622 | switch (cmd) { |
1623 | case LOOP_SET_FD: | |
757ecf40 | 1624 | return loop_set_fd(lo, mode, bdev, arg); |
1da177e4 | 1625 | case LOOP_CHANGE_FD: |
c3710770 | 1626 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1627 | case LOOP_CLR_FD: |
7ccd0791 | 1628 | return loop_clr_fd(lo); |
1da177e4 | 1629 | case LOOP_SET_STATUS: |
7035b5df | 1630 | err = -EPERM; |
a1316544 | 1631 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1632 | err = loop_set_status_old(lo, |
1633 | (struct loop_info __user *)arg); | |
a1316544 | 1634 | } |
1da177e4 LT |
1635 | break; |
1636 | case LOOP_GET_STATUS: | |
4a5ce9ba | 1637 | return loop_get_status_old(lo, (struct loop_info __user *) arg); |
1da177e4 | 1638 | case LOOP_SET_STATUS64: |
7035b5df | 1639 | err = -EPERM; |
a1316544 | 1640 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1641 | err = loop_set_status64(lo, |
1642 | (struct loop_info64 __user *) arg); | |
a1316544 | 1643 | } |
1da177e4 LT |
1644 | break; |
1645 | case LOOP_GET_STATUS64: | |
4a5ce9ba | 1646 | return loop_get_status64(lo, (struct loop_info64 __user *) arg); |
a1316544 | 1647 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1648 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1649 | case LOOP_SET_BLOCK_SIZE: |
a1316544 JK |
1650 | if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) |
1651 | return -EPERM; | |
1652 | /* Fall through */ | |
1da177e4 | 1653 | default: |
a1316544 JK |
1654 | err = lo_simple_ioctl(lo, cmd, arg); |
1655 | break; | |
1da177e4 | 1656 | } |
f028f3b2 | 1657 | |
1da177e4 LT |
1658 | return err; |
1659 | } | |
1660 | ||
863d5b82 DH |
1661 | #ifdef CONFIG_COMPAT |
1662 | struct compat_loop_info { | |
1663 | compat_int_t lo_number; /* ioctl r/o */ | |
1664 | compat_dev_t lo_device; /* ioctl r/o */ | |
1665 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1666 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1667 | compat_int_t lo_offset; | |
1668 | compat_int_t lo_encrypt_type; | |
1669 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1670 | compat_int_t lo_flags; /* ioctl r/o */ | |
1671 | char lo_name[LO_NAME_SIZE]; | |
1672 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1673 | compat_ulong_t lo_init[2]; | |
1674 | char reserved[4]; | |
1675 | }; | |
1676 | ||
1677 | /* | |
1678 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1679 | * - noinlined to reduce stack space usage in main part of driver | |
1680 | */ | |
1681 | static noinline int | |
ba674cfc | 1682 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1683 | struct loop_info64 *info64) |
1684 | { | |
1685 | struct compat_loop_info info; | |
1686 | ||
1687 | if (copy_from_user(&info, arg, sizeof(info))) | |
1688 | return -EFAULT; | |
1689 | ||
1690 | memset(info64, 0, sizeof(*info64)); | |
1691 | info64->lo_number = info.lo_number; | |
1692 | info64->lo_device = info.lo_device; | |
1693 | info64->lo_inode = info.lo_inode; | |
1694 | info64->lo_rdevice = info.lo_rdevice; | |
1695 | info64->lo_offset = info.lo_offset; | |
1696 | info64->lo_sizelimit = 0; | |
1697 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1698 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1699 | info64->lo_flags = info.lo_flags; | |
1700 | info64->lo_init[0] = info.lo_init[0]; | |
1701 | info64->lo_init[1] = info.lo_init[1]; | |
1702 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1703 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1704 | else | |
1705 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1706 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1707 | return 0; | |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1712 | * - noinlined to reduce stack space usage in main part of driver | |
1713 | */ | |
1714 | static noinline int | |
1715 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1716 | struct compat_loop_info __user *arg) | |
1717 | { | |
1718 | struct compat_loop_info info; | |
1719 | ||
1720 | memset(&info, 0, sizeof(info)); | |
1721 | info.lo_number = info64->lo_number; | |
1722 | info.lo_device = info64->lo_device; | |
1723 | info.lo_inode = info64->lo_inode; | |
1724 | info.lo_rdevice = info64->lo_rdevice; | |
1725 | info.lo_offset = info64->lo_offset; | |
1726 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1727 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1728 | info.lo_flags = info64->lo_flags; | |
1729 | info.lo_init[0] = info64->lo_init[0]; | |
1730 | info.lo_init[1] = info64->lo_init[1]; | |
1731 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1732 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1733 | else | |
1734 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1735 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1736 | ||
1737 | /* error in case values were truncated */ | |
1738 | if (info.lo_device != info64->lo_device || | |
1739 | info.lo_rdevice != info64->lo_rdevice || | |
1740 | info.lo_inode != info64->lo_inode || | |
1741 | info.lo_offset != info64->lo_offset || | |
1742 | info.lo_init[0] != info64->lo_init[0] || | |
1743 | info.lo_init[1] != info64->lo_init[1]) | |
1744 | return -EOVERFLOW; | |
1745 | ||
1746 | if (copy_to_user(arg, &info, sizeof(info))) | |
1747 | return -EFAULT; | |
1748 | return 0; | |
1749 | } | |
1750 | ||
1751 | static int | |
1752 | loop_set_status_compat(struct loop_device *lo, | |
1753 | const struct compat_loop_info __user *arg) | |
1754 | { | |
1755 | struct loop_info64 info64; | |
1756 | int ret; | |
1757 | ||
1758 | ret = loop_info64_from_compat(arg, &info64); | |
1759 | if (ret < 0) | |
1760 | return ret; | |
1761 | return loop_set_status(lo, &info64); | |
1762 | } | |
1763 | ||
1764 | static int | |
1765 | loop_get_status_compat(struct loop_device *lo, | |
1766 | struct compat_loop_info __user *arg) | |
1767 | { | |
1768 | struct loop_info64 info64; | |
bdac616d | 1769 | int err; |
863d5b82 | 1770 | |
4a5ce9ba | 1771 | if (!arg) |
bdac616d | 1772 | return -EINVAL; |
bdac616d | 1773 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1774 | if (!err) |
1775 | err = loop_info64_to_compat(&info64, arg); | |
1776 | return err; | |
1777 | } | |
1778 | ||
bb214884 AV |
1779 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1780 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1781 | { |
bb214884 | 1782 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1783 | int err; |
1784 | ||
863d5b82 DH |
1785 | switch(cmd) { |
1786 | case LOOP_SET_STATUS: | |
550df5fd JK |
1787 | err = loop_set_status_compat(lo, |
1788 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1789 | break; |
1790 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1791 | err = loop_get_status_compat(lo, |
1792 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1793 | break; |
53d66608 | 1794 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1795 | case LOOP_CLR_FD: |
1796 | case LOOP_GET_STATUS64: | |
1797 | case LOOP_SET_STATUS64: | |
1798 | arg = (unsigned long) compat_ptr(arg); | |
d893ff86 | 1799 | /* fall through */ |
863d5b82 DH |
1800 | case LOOP_SET_FD: |
1801 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1802 | case LOOP_SET_BLOCK_SIZE: |
fdbe4eee | 1803 | case LOOP_SET_DIRECT_IO: |
bb214884 | 1804 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1805 | break; |
1806 | default: | |
1807 | err = -ENOIOCTLCMD; | |
1808 | break; | |
1809 | } | |
863d5b82 DH |
1810 | return err; |
1811 | } | |
1812 | #endif | |
1813 | ||
bb214884 | 1814 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1815 | { |
770fe30a | 1816 | struct loop_device *lo; |
0a42e99b | 1817 | int err; |
770fe30a | 1818 | |
0a42e99b JK |
1819 | err = mutex_lock_killable(&loop_ctl_mutex); |
1820 | if (err) | |
1821 | return err; | |
770fe30a KS |
1822 | lo = bdev->bd_disk->private_data; |
1823 | if (!lo) { | |
1824 | err = -ENXIO; | |
1825 | goto out; | |
1826 | } | |
1da177e4 | 1827 | |
f8933667 | 1828 | atomic_inc(&lo->lo_refcnt); |
770fe30a | 1829 | out: |
0a42e99b | 1830 | mutex_unlock(&loop_ctl_mutex); |
770fe30a | 1831 | return err; |
1da177e4 LT |
1832 | } |
1833 | ||
967d1dc1 | 1834 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1835 | { |
967d1dc1 | 1836 | struct loop_device *lo; |
1da177e4 | 1837 | |
0a42e99b | 1838 | mutex_lock(&loop_ctl_mutex); |
967d1dc1 | 1839 | lo = disk->private_data; |
f8933667 | 1840 | if (atomic_dec_return(&lo->lo_refcnt)) |
0a42e99b | 1841 | goto out_unlock; |
14f27939 MB |
1842 | |
1843 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
a2505b79 JK |
1844 | if (lo->lo_state != Lo_bound) |
1845 | goto out_unlock; | |
1846 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1847 | mutex_unlock(&loop_ctl_mutex); |
14f27939 MB |
1848 | /* |
1849 | * In autoclear mode, stop the loop thread | |
1850 | * and remove configuration after last close. | |
1851 | */ | |
0da03cab | 1852 | __loop_clr_fd(lo, true); |
7ccd0791 | 1853 | return; |
43cade80 | 1854 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
1855 | /* |
1856 | * Otherwise keep thread (if running) and config, | |
1857 | * but flush possible ongoing bios in thread. | |
1858 | */ | |
43cade80 OS |
1859 | blk_mq_freeze_queue(lo->lo_queue); |
1860 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 1861 | } |
96c58655 | 1862 | |
0a42e99b | 1863 | out_unlock: |
310ca162 | 1864 | mutex_unlock(&loop_ctl_mutex); |
ae665016 LT |
1865 | } |
1866 | ||
83d5cde4 | 1867 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1868 | .owner = THIS_MODULE, |
bb214884 AV |
1869 | .open = lo_open, |
1870 | .release = lo_release, | |
1871 | .ioctl = lo_ioctl, | |
863d5b82 | 1872 | #ifdef CONFIG_COMPAT |
bb214884 | 1873 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1874 | #endif |
1da177e4 LT |
1875 | }; |
1876 | ||
1877 | /* | |
1878 | * And now the modules code and kernel interface. | |
1879 | */ | |
73285082 | 1880 | static int max_loop; |
5657a819 | 1881 | module_param(max_loop, int, 0444); |
a47653fc | 1882 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
5657a819 | 1883 | module_param(max_part, int, 0444); |
476a4813 | 1884 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1885 | MODULE_LICENSE("GPL"); |
1886 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1887 | ||
1888 | int loop_register_transfer(struct loop_func_table *funcs) | |
1889 | { | |
1890 | unsigned int n = funcs->number; | |
1891 | ||
1892 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1893 | return -EINVAL; | |
1894 | xfer_funcs[n] = funcs; | |
1895 | return 0; | |
1896 | } | |
1897 | ||
34dd82af KS |
1898 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1899 | { | |
1900 | struct loop_device *lo = ptr; | |
1901 | struct loop_func_table *xfer = data; | |
1902 | ||
310ca162 | 1903 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
1904 | if (lo->lo_encryption == xfer) |
1905 | loop_release_xfer(lo); | |
310ca162 | 1906 | mutex_unlock(&loop_ctl_mutex); |
34dd82af KS |
1907 | return 0; |
1908 | } | |
1909 | ||
1da177e4 LT |
1910 | int loop_unregister_transfer(int number) |
1911 | { | |
1912 | unsigned int n = number; | |
1da177e4 LT |
1913 | struct loop_func_table *xfer; |
1914 | ||
1915 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1916 | return -EINVAL; | |
1917 | ||
1918 | xfer_funcs[n] = NULL; | |
34dd82af | 1919 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1920 | return 0; |
1921 | } | |
1922 | ||
1923 | EXPORT_SYMBOL(loop_register_transfer); | |
1924 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1925 | ||
fc17b653 | 1926 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1927 | const struct blk_mq_queue_data *bd) |
1928 | { | |
1894e916 JA |
1929 | struct request *rq = bd->rq; |
1930 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1931 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 1932 | |
1894e916 | 1933 | blk_mq_start_request(rq); |
b5dd2f60 | 1934 | |
f4aa4c7b | 1935 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1936 | return BLK_STS_IOERR; |
f4aa4c7b | 1937 | |
1894e916 | 1938 | switch (req_op(rq)) { |
f0225cac CH |
1939 | case REQ_OP_FLUSH: |
1940 | case REQ_OP_DISCARD: | |
19372e27 | 1941 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1942 | cmd->use_aio = false; |
f0225cac CH |
1943 | break; |
1944 | default: | |
1945 | cmd->use_aio = lo->use_dio; | |
1946 | break; | |
1947 | } | |
bc07c10a | 1948 | |
d4478e92 | 1949 | /* always use the first bio's css */ |
0b508bc9 | 1950 | #ifdef CONFIG_BLK_CGROUP |
db6638d7 DZ |
1951 | if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { |
1952 | cmd->css = &bio_blkcg(rq->bio)->css; | |
d4478e92 SL |
1953 | css_get(cmd->css); |
1954 | } else | |
1955 | #endif | |
1956 | cmd->css = NULL; | |
3989144f | 1957 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1958 | |
fc17b653 | 1959 | return BLK_STS_OK; |
b5dd2f60 ML |
1960 | } |
1961 | ||
1962 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1963 | { | |
1894e916 JA |
1964 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
1965 | const bool write = op_is_write(req_op(rq)); | |
1966 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 1967 | int ret = 0; |
b5dd2f60 | 1968 | |
f4829a9b CH |
1969 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1970 | ret = -EIO; | |
b5dd2f60 | 1971 | goto failed; |
f4829a9b | 1972 | } |
b5dd2f60 | 1973 | |
1894e916 | 1974 | ret = do_req_filebacked(lo, rq); |
b5dd2f60 | 1975 | failed: |
bc07c10a | 1976 | /* complete non-aio request */ |
fe2cb290 | 1977 | if (!cmd->use_aio || ret) { |
8cd55087 EG |
1978 | if (ret == -EOPNOTSUPP) |
1979 | cmd->ret = ret; | |
1980 | else | |
1981 | cmd->ret = ret ? -EIO : 0; | |
1894e916 | 1982 | blk_mq_complete_request(rq); |
fe2cb290 | 1983 | } |
b5dd2f60 ML |
1984 | } |
1985 | ||
e03a3d7a | 1986 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1987 | { |
1988 | struct loop_cmd *cmd = | |
e03a3d7a | 1989 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1990 | |
1991 | loop_handle_cmd(cmd); | |
1992 | } | |
1993 | ||
d6296d39 CH |
1994 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1995 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1996 | { |
1997 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1998 | ||
3989144f | 1999 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
2000 | return 0; |
2001 | } | |
2002 | ||
f363b089 | 2003 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 2004 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 2005 | .init_request = loop_init_request, |
fe2cb290 | 2006 | .complete = lo_complete_rq, |
b5dd2f60 ML |
2007 | }; |
2008 | ||
34dd82af | 2009 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
2010 | { |
2011 | struct loop_device *lo; | |
2012 | struct gendisk *disk; | |
34dd82af | 2013 | int err; |
73285082 | 2014 | |
68d740d7 | 2015 | err = -ENOMEM; |
73285082 | 2016 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 2017 | if (!lo) |
73285082 | 2018 | goto out; |
34dd82af | 2019 | |
ef7e7c82 MP |
2020 | lo->lo_state = Lo_unbound; |
2021 | ||
c718aa65 | 2022 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 2023 | if (i >= 0) { |
c718aa65 TH |
2024 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
2025 | if (err == -ENOSPC) | |
34dd82af | 2026 | err = -EEXIST; |
34dd82af | 2027 | } else { |
c718aa65 | 2028 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
2029 | } |
2030 | if (err < 0) | |
2031 | goto out_free_dev; | |
c718aa65 | 2032 | i = err; |
73285082 | 2033 | |
183cfb57 | 2034 | err = -ENOMEM; |
b5dd2f60 ML |
2035 | lo->tag_set.ops = &loop_mq_ops; |
2036 | lo->tag_set.nr_hw_queues = 1; | |
2037 | lo->tag_set.queue_depth = 128; | |
2038 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
2039 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
56d18f62 | 2040 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; |
b5dd2f60 ML |
2041 | lo->tag_set.driver_data = lo; |
2042 | ||
2043 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
2044 | if (err) | |
3ec981e3 | 2045 | goto out_free_idr; |
73285082 | 2046 | |
b5dd2f60 | 2047 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
38a3499f | 2048 | if (IS_ERR(lo->lo_queue)) { |
b5dd2f60 ML |
2049 | err = PTR_ERR(lo->lo_queue); |
2050 | goto out_cleanup_tags; | |
2051 | } | |
ef7e7c82 MP |
2052 | lo->lo_queue->queuedata = lo; |
2053 | ||
54bb0ade | 2054 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 2055 | |
5b5e20f4 | 2056 | /* |
40326d8a SL |
2057 | * By default, we do buffer IO, so it doesn't make sense to enable |
2058 | * merge because the I/O submitted to backing file is handled page by | |
2059 | * page. For directio mode, merge does help to dispatch bigger request | |
2060 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 | 2061 | */ |
8b904b5b | 2062 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
5b5e20f4 | 2063 | |
7a649737 | 2064 | err = -ENOMEM; |
476a4813 | 2065 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
2066 | if (!disk) |
2067 | goto out_free_queue; | |
2068 | ||
e03c8dd1 KS |
2069 | /* |
2070 | * Disable partition scanning by default. The in-kernel partition | |
2071 | * scanning can be requested individually per-device during its | |
2072 | * setup. Userspace can always add and remove partitions from all | |
2073 | * devices. The needed partition minors are allocated from the | |
2074 | * extended minor space, the main loop device numbers will continue | |
2075 | * to match the loop minors, regardless of the number of partitions | |
2076 | * used. | |
2077 | * | |
2078 | * If max_part is given, partition scanning is globally enabled for | |
2079 | * all loop devices. The minors for the main loop devices will be | |
2080 | * multiples of max_part. | |
2081 | * | |
2082 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
2083 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
2084 | * complicated, are too static, inflexible and may surprise | |
2085 | * userspace tools. Parameters like this in general should be avoided. | |
2086 | */ | |
2087 | if (!part_shift) | |
2088 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
2089 | disk->flags |= GENHD_FL_EXT_DEVT; | |
f8933667 | 2090 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 2091 | lo->lo_number = i; |
73285082 KC |
2092 | spin_lock_init(&lo->lo_lock); |
2093 | disk->major = LOOP_MAJOR; | |
476a4813 | 2094 | disk->first_minor = i << part_shift; |
73285082 KC |
2095 | disk->fops = &lo_fops; |
2096 | disk->private_data = lo; | |
2097 | disk->queue = lo->lo_queue; | |
2098 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
2099 | add_disk(disk); |
2100 | *l = lo; | |
2101 | return lo->lo_number; | |
73285082 KC |
2102 | |
2103 | out_free_queue: | |
2104 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
2105 | out_cleanup_tags: |
2106 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
2107 | out_free_idr: |
2108 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
2109 | out_free_dev: |
2110 | kfree(lo); | |
2111 | out: | |
34dd82af | 2112 | return err; |
73285082 KC |
2113 | } |
2114 | ||
34dd82af | 2115 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2116 | { |
6cd18e71 | 2117 | del_gendisk(lo->lo_disk); |
0fa8ebdd | 2118 | blk_cleanup_queue(lo->lo_queue); |
b5dd2f60 | 2119 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 2120 | put_disk(lo->lo_disk); |
73285082 KC |
2121 | kfree(lo); |
2122 | } | |
1da177e4 | 2123 | |
770fe30a KS |
2124 | static int find_free_cb(int id, void *ptr, void *data) |
2125 | { | |
2126 | struct loop_device *lo = ptr; | |
2127 | struct loop_device **l = data; | |
2128 | ||
2129 | if (lo->lo_state == Lo_unbound) { | |
2130 | *l = lo; | |
2131 | return 1; | |
2132 | } | |
2133 | return 0; | |
2134 | } | |
2135 | ||
34dd82af | 2136 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
2137 | { |
2138 | struct loop_device *lo; | |
34dd82af | 2139 | int ret = -ENODEV; |
a47653fc | 2140 | |
770fe30a KS |
2141 | if (i < 0) { |
2142 | int err; | |
2143 | ||
2144 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
2145 | if (err == 1) { | |
2146 | *l = lo; | |
2147 | ret = lo->lo_number; | |
2148 | } | |
2149 | goto out; | |
a47653fc KC |
2150 | } |
2151 | ||
770fe30a | 2152 | /* lookup and return a specific i */ |
34dd82af | 2153 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 2154 | if (lo) { |
34dd82af KS |
2155 | *l = lo; |
2156 | ret = lo->lo_number; | |
a47653fc | 2157 | } |
770fe30a | 2158 | out: |
34dd82af | 2159 | return ret; |
a47653fc KC |
2160 | } |
2161 | ||
73285082 KC |
2162 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
2163 | { | |
705962cc | 2164 | struct loop_device *lo; |
07002e99 | 2165 | struct kobject *kobj; |
34dd82af | 2166 | int err; |
73285082 | 2167 | |
0a42e99b | 2168 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2169 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); |
2170 | if (err < 0) | |
2171 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
2172 | if (err < 0) | |
a207f593 | 2173 | kobj = NULL; |
34dd82af | 2174 | else |
3079c22e | 2175 | kobj = get_disk_and_module(lo->lo_disk); |
0a42e99b | 2176 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2177 | |
2178 | *part = 0; | |
07002e99 | 2179 | return kobj; |
73285082 KC |
2180 | } |
2181 | ||
770fe30a KS |
2182 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2183 | unsigned long parm) | |
2184 | { | |
2185 | struct loop_device *lo; | |
0a42e99b | 2186 | int ret; |
770fe30a | 2187 | |
0a42e99b JK |
2188 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2189 | if (ret) | |
2190 | return ret; | |
2191 | ||
2192 | ret = -ENOSYS; | |
770fe30a KS |
2193 | switch (cmd) { |
2194 | case LOOP_CTL_ADD: | |
2195 | ret = loop_lookup(&lo, parm); | |
2196 | if (ret >= 0) { | |
2197 | ret = -EEXIST; | |
2198 | break; | |
2199 | } | |
2200 | ret = loop_add(&lo, parm); | |
2201 | break; | |
2202 | case LOOP_CTL_REMOVE: | |
2203 | ret = loop_lookup(&lo, parm); | |
2204 | if (ret < 0) | |
2205 | break; | |
770fe30a KS |
2206 | if (lo->lo_state != Lo_unbound) { |
2207 | ret = -EBUSY; | |
770fe30a KS |
2208 | break; |
2209 | } | |
f8933667 | 2210 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a | 2211 | ret = -EBUSY; |
770fe30a KS |
2212 | break; |
2213 | } | |
2214 | lo->lo_disk->private_data = NULL; | |
770fe30a KS |
2215 | idr_remove(&loop_index_idr, lo->lo_number); |
2216 | loop_remove(lo); | |
2217 | break; | |
2218 | case LOOP_CTL_GET_FREE: | |
2219 | ret = loop_lookup(&lo, -1); | |
2220 | if (ret >= 0) | |
2221 | break; | |
2222 | ret = loop_add(&lo, -1); | |
2223 | } | |
0a42e99b | 2224 | mutex_unlock(&loop_ctl_mutex); |
770fe30a KS |
2225 | |
2226 | return ret; | |
2227 | } | |
2228 | ||
2229 | static const struct file_operations loop_ctl_fops = { | |
2230 | .open = nonseekable_open, | |
2231 | .unlocked_ioctl = loop_control_ioctl, | |
2232 | .compat_ioctl = loop_control_ioctl, | |
2233 | .owner = THIS_MODULE, | |
2234 | .llseek = noop_llseek, | |
2235 | }; | |
2236 | ||
2237 | static struct miscdevice loop_misc = { | |
2238 | .minor = LOOP_CTRL_MINOR, | |
2239 | .name = "loop-control", | |
2240 | .fops = &loop_ctl_fops, | |
2241 | }; | |
2242 | ||
2243 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2244 | MODULE_ALIAS("devname:loop-control"); | |
2245 | ||
73285082 KC |
2246 | static int __init loop_init(void) |
2247 | { | |
a47653fc KC |
2248 | int i, nr; |
2249 | unsigned long range; | |
34dd82af | 2250 | struct loop_device *lo; |
770fe30a | 2251 | int err; |
a47653fc | 2252 | |
476a4813 | 2253 | part_shift = 0; |
ac04fee0 | 2254 | if (max_part > 0) { |
476a4813 LV |
2255 | part_shift = fls(max_part); |
2256 | ||
ac04fee0 NK |
2257 | /* |
2258 | * Adjust max_part according to part_shift as it is exported | |
2259 | * to user space so that user can decide correct minor number | |
2260 | * if [s]he want to create more devices. | |
2261 | * | |
2262 | * Note that -1 is required because partition 0 is reserved | |
2263 | * for the whole disk. | |
2264 | */ | |
2265 | max_part = (1UL << part_shift) - 1; | |
2266 | } | |
2267 | ||
b1a66504 GC |
2268 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2269 | err = -EINVAL; | |
a8c1d064 | 2270 | goto err_out; |
b1a66504 | 2271 | } |
78f4bb36 | 2272 | |
b1a66504 GC |
2273 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2274 | err = -EINVAL; | |
a8c1d064 | 2275 | goto err_out; |
b1a66504 | 2276 | } |
1da177e4 | 2277 | |
d134b00b KS |
2278 | /* |
2279 | * If max_loop is specified, create that many devices upfront. | |
2280 | * This also becomes a hard limit. If max_loop is not specified, | |
2281 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2282 | * init time. Loop devices can be requested on-demand with the | |
2283 | * /dev/loop-control interface, or be instantiated by accessing | |
2284 | * a 'dead' device node. | |
2285 | */ | |
73285082 | 2286 | if (max_loop) { |
a47653fc | 2287 | nr = max_loop; |
a1c15c59 | 2288 | range = max_loop << part_shift; |
a47653fc | 2289 | } else { |
d134b00b | 2290 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2291 | range = 1UL << MINORBITS; |
a47653fc KC |
2292 | } |
2293 | ||
a8c1d064 AV |
2294 | err = misc_register(&loop_misc); |
2295 | if (err < 0) | |
2296 | goto err_out; | |
2297 | ||
2298 | ||
b1a66504 GC |
2299 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2300 | err = -EIO; | |
2301 | goto misc_out; | |
2302 | } | |
1da177e4 | 2303 | |
a47653fc KC |
2304 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2305 | THIS_MODULE, loop_probe, NULL, NULL); | |
2306 | ||
d134b00b | 2307 | /* pre-create number of devices given by config or max_loop */ |
0a42e99b | 2308 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2309 | for (i = 0; i < nr; i++) |
2310 | loop_add(&lo, i); | |
0a42e99b | 2311 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 2312 | |
73285082 | 2313 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2314 | return 0; |
b1a66504 GC |
2315 | |
2316 | misc_out: | |
2317 | misc_deregister(&loop_misc); | |
a8c1d064 | 2318 | err_out: |
b1a66504 | 2319 | return err; |
34dd82af | 2320 | } |
a47653fc | 2321 | |
34dd82af KS |
2322 | static int loop_exit_cb(int id, void *ptr, void *data) |
2323 | { | |
2324 | struct loop_device *lo = ptr; | |
a47653fc | 2325 | |
34dd82af KS |
2326 | loop_remove(lo); |
2327 | return 0; | |
1da177e4 LT |
2328 | } |
2329 | ||
73285082 | 2330 | static void __exit loop_exit(void) |
1da177e4 | 2331 | { |
a47653fc | 2332 | unsigned long range; |
1da177e4 | 2333 | |
a1c15c59 | 2334 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2335 | |
34dd82af | 2336 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2337 | idr_destroy(&loop_index_idr); |
73285082 | 2338 | |
a47653fc | 2339 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2340 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2341 | |
2342 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2343 | } |
2344 | ||
2345 | module_init(loop_init); | |
2346 | module_exit(loop_exit); | |
2347 | ||
2348 | #ifndef MODULE | |
2349 | static int __init max_loop_setup(char *str) | |
2350 | { | |
2351 | max_loop = simple_strtol(str, NULL, 0); | |
2352 | return 1; | |
2353 | } | |
2354 | ||
2355 | __setup("max_loop=", max_loop_setup); | |
2356 | #endif |