]>
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 | */ | |
217 | blk_mq_freeze_queue(lo->lo_queue); | |
218 | lo->use_dio = use_dio; | |
40326d8a | 219 | if (use_dio) { |
8b904b5b | 220 | blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 221 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
40326d8a | 222 | } else { |
8b904b5b | 223 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 224 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; |
40326d8a | 225 | } |
2e5ab5f3 ML |
226 | blk_mq_unfreeze_queue(lo->lo_queue); |
227 | } | |
228 | ||
1da177e4 | 229 | static int |
1e6ec9ea | 230 | figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) |
1da177e4 | 231 | { |
7035b5df | 232 | loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); |
1da177e4 | 233 | sector_t x = (sector_t)size; |
7b0576a3 | 234 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
235 | |
236 | if (unlikely((loff_t)x != size)) | |
237 | return -EFBIG; | |
7035b5df DM |
238 | if (lo->lo_offset != offset) |
239 | lo->lo_offset = offset; | |
240 | if (lo->lo_sizelimit != sizelimit) | |
241 | lo->lo_sizelimit = sizelimit; | |
73285082 | 242 | set_capacity(lo->lo_disk, x); |
7b0576a3 GC |
243 | bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); |
244 | /* let user-space know about the new size */ | |
245 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
7035b5df | 246 | return 0; |
1da177e4 LT |
247 | } |
248 | ||
249 | static inline int | |
250 | lo_do_transfer(struct loop_device *lo, int cmd, | |
251 | struct page *rpage, unsigned roffs, | |
252 | struct page *lpage, unsigned loffs, | |
253 | int size, sector_t rblock) | |
254 | { | |
aa4d8616 CH |
255 | int ret; |
256 | ||
257 | ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
258 | if (likely(!ret)) | |
1da177e4 LT |
259 | return 0; |
260 | ||
aa4d8616 CH |
261 | printk_ratelimited(KERN_ERR |
262 | "loop: Transfer error at byte offset %llu, length %i.\n", | |
263 | (unsigned long long)rblock << 9, size); | |
264 | return ret; | |
1da177e4 LT |
265 | } |
266 | ||
aa4d8616 | 267 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 268 | { |
aa4d8616 | 269 | struct iov_iter i; |
1da177e4 | 270 | ssize_t bw; |
283e7e5d | 271 | |
b6207430 | 272 | iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); |
1da177e4 | 273 | |
03d95eb2 | 274 | file_start_write(file); |
abbb6589 | 275 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 276 | file_end_write(file); |
aa4d8616 CH |
277 | |
278 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 279 | return 0; |
aa4d8616 CH |
280 | |
281 | printk_ratelimited(KERN_ERR | |
282 | "loop: Write error at byte offset %llu, length %i.\n", | |
283 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
284 | if (bw >= 0) |
285 | bw = -EIO; | |
286 | return bw; | |
287 | } | |
288 | ||
aa4d8616 CH |
289 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
290 | loff_t pos) | |
1da177e4 | 291 | { |
aa4d8616 CH |
292 | struct bio_vec bvec; |
293 | struct req_iterator iter; | |
294 | int ret = 0; | |
295 | ||
296 | rq_for_each_segment(bvec, rq, iter) { | |
297 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
298 | if (ret < 0) | |
299 | break; | |
300 | cond_resched(); | |
301 | } | |
302 | ||
303 | return ret; | |
1da177e4 LT |
304 | } |
305 | ||
aa4d8616 | 306 | /* |
456be148 CH |
307 | * This is the slow, transforming version that needs to double buffer the |
308 | * data as it cannot do the transformations in place without having direct | |
309 | * access to the destination pages of the backing file. | |
1da177e4 | 310 | */ |
aa4d8616 CH |
311 | static int lo_write_transfer(struct loop_device *lo, struct request *rq, |
312 | loff_t pos) | |
1da177e4 | 313 | { |
aa4d8616 | 314 | struct bio_vec bvec, b; |
30112013 | 315 | struct req_iterator iter; |
aa4d8616 | 316 | struct page *page; |
7988613b | 317 | int ret = 0; |
1da177e4 | 318 | |
aa4d8616 CH |
319 | page = alloc_page(GFP_NOIO); |
320 | if (unlikely(!page)) | |
321 | return -ENOMEM; | |
456be148 | 322 | |
30112013 | 323 | rq_for_each_segment(bvec, rq, iter) { |
aa4d8616 CH |
324 | ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, |
325 | bvec.bv_offset, bvec.bv_len, pos >> 9); | |
326 | if (unlikely(ret)) | |
327 | break; | |
328 | ||
329 | b.bv_page = page; | |
330 | b.bv_offset = 0; | |
331 | b.bv_len = bvec.bv_len; | |
332 | ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); | |
1da177e4 LT |
333 | if (ret < 0) |
334 | break; | |
1da177e4 | 335 | } |
aa4d8616 CH |
336 | |
337 | __free_page(page); | |
1da177e4 | 338 | return ret; |
1da177e4 LT |
339 | } |
340 | ||
aa4d8616 CH |
341 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
342 | loff_t pos) | |
1da177e4 | 343 | { |
aa4d8616 CH |
344 | struct bio_vec bvec; |
345 | struct req_iterator iter; | |
346 | struct iov_iter i; | |
347 | ssize_t len; | |
1da177e4 | 348 | |
aa4d8616 | 349 | rq_for_each_segment(bvec, rq, iter) { |
b6207430 | 350 | iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); |
18e9710e | 351 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
352 | if (len < 0) |
353 | return len; | |
1da177e4 | 354 | |
aa4d8616 | 355 | flush_dcache_page(bvec.bv_page); |
fd582140 | 356 | |
aa4d8616 CH |
357 | if (len != bvec.bv_len) { |
358 | struct bio *bio; | |
1da177e4 | 359 | |
aa4d8616 CH |
360 | __rq_for_each_bio(bio, rq) |
361 | zero_fill_bio(bio); | |
362 | break; | |
363 | } | |
364 | cond_resched(); | |
365 | } | |
366 | ||
367 | return 0; | |
fd582140 JA |
368 | } |
369 | ||
aa4d8616 CH |
370 | static int lo_read_transfer(struct loop_device *lo, struct request *rq, |
371 | loff_t pos) | |
1da177e4 | 372 | { |
aa4d8616 CH |
373 | struct bio_vec bvec, b; |
374 | struct req_iterator iter; | |
375 | struct iov_iter i; | |
376 | struct page *page; | |
377 | ssize_t len; | |
378 | int ret = 0; | |
1da177e4 | 379 | |
aa4d8616 CH |
380 | page = alloc_page(GFP_NOIO); |
381 | if (unlikely(!page)) | |
382 | return -ENOMEM; | |
fd582140 | 383 | |
aa4d8616 CH |
384 | rq_for_each_segment(bvec, rq, iter) { |
385 | loff_t offset = pos; | |
fd582140 | 386 | |
aa4d8616 CH |
387 | b.bv_page = page; |
388 | b.bv_offset = 0; | |
389 | b.bv_len = bvec.bv_len; | |
fd582140 | 390 | |
b6207430 | 391 | iov_iter_bvec(&i, READ, &b, 1, b.bv_len); |
18e9710e | 392 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
393 | if (len < 0) { |
394 | ret = len; | |
395 | goto out_free_page; | |
396 | } | |
1da177e4 | 397 | |
aa4d8616 CH |
398 | ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, |
399 | bvec.bv_offset, len, offset >> 9); | |
400 | if (ret) | |
401 | goto out_free_page; | |
1da177e4 | 402 | |
aa4d8616 | 403 | flush_dcache_page(bvec.bv_page); |
306df071 | 404 | |
aa4d8616 | 405 | if (len != bvec.bv_len) { |
30112013 ML |
406 | struct bio *bio; |
407 | ||
408 | __rq_for_each_bio(bio, rq) | |
409 | zero_fill_bio(bio); | |
1da177e4 | 410 | break; |
306df071 | 411 | } |
1da177e4 | 412 | } |
aa4d8616 CH |
413 | |
414 | ret = 0; | |
415 | out_free_page: | |
416 | __free_page(page); | |
417 | return ret; | |
1da177e4 LT |
418 | } |
419 | ||
efcfec57 DW |
420 | static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, |
421 | int mode) | |
cf655d95 ML |
422 | { |
423 | /* | |
efcfec57 DW |
424 | * We use fallocate to manipulate the space mappings used by the image |
425 | * a.k.a. discard/zerorange. However we do not support this if | |
426 | * encryption is enabled, because it may give an attacker useful | |
427 | * information. | |
cf655d95 ML |
428 | */ |
429 | struct file *file = lo->lo_backing_file; | |
cf655d95 ML |
430 | int ret; |
431 | ||
efcfec57 DW |
432 | mode |= FALLOC_FL_KEEP_SIZE; |
433 | ||
cf655d95 ML |
434 | if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { |
435 | ret = -EOPNOTSUPP; | |
436 | goto out; | |
437 | } | |
438 | ||
439 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
440 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
441 | ret = -EIO; | |
442 | out: | |
443 | return ret; | |
444 | } | |
445 | ||
446 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
447 | { | |
448 | struct file *file = lo->lo_backing_file; | |
449 | int ret = vfs_fsync(file, 0); | |
450 | if (unlikely(ret && ret != -EINVAL)) | |
451 | ret = -EIO; | |
452 | ||
453 | return ret; | |
454 | } | |
455 | ||
fe2cb290 | 456 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 457 | { |
fe2cb290 | 458 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 459 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 460 | |
f9de14bc JA |
461 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
462 | req_op(rq) != REQ_OP_READ) { | |
463 | if (cmd->ret < 0) | |
464 | ret = BLK_STS_IOERR; | |
465 | goto end_io; | |
bc07c10a | 466 | } |
fe2cb290 | 467 | |
f9de14bc JA |
468 | /* |
469 | * Short READ - if we got some data, advance our request and | |
470 | * retry it. If we got no data, end the rest with EIO. | |
471 | */ | |
472 | if (cmd->ret) { | |
473 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
474 | cmd->ret = 0; | |
475 | blk_mq_requeue_request(rq, true); | |
476 | } else { | |
477 | if (cmd->use_aio) { | |
478 | struct bio *bio = rq->bio; | |
479 | ||
480 | while (bio) { | |
481 | zero_fill_bio(bio); | |
482 | bio = bio->bi_next; | |
483 | } | |
484 | } | |
485 | ret = BLK_STS_IOERR; | |
486 | end_io: | |
487 | blk_mq_end_request(rq, ret); | |
488 | } | |
bc07c10a ML |
489 | } |
490 | ||
92d77332 SL |
491 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
492 | { | |
1894e916 JA |
493 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
494 | ||
92d77332 SL |
495 | if (!atomic_dec_and_test(&cmd->ref)) |
496 | return; | |
497 | kfree(cmd->bvec); | |
498 | cmd->bvec = NULL; | |
1894e916 | 499 | blk_mq_complete_request(rq); |
92d77332 SL |
500 | } |
501 | ||
bc07c10a ML |
502 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) |
503 | { | |
504 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 505 | |
d4478e92 SL |
506 | if (cmd->css) |
507 | css_put(cmd->css); | |
fe2cb290 | 508 | cmd->ret = ret; |
92d77332 | 509 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
510 | } |
511 | ||
512 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
513 | loff_t pos, bool rw) | |
514 | { | |
515 | struct iov_iter iter; | |
86af5952 | 516 | struct req_iterator rq_iter; |
bc07c10a | 517 | struct bio_vec *bvec; |
1894e916 | 518 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 519 | struct bio *bio = rq->bio; |
bc07c10a | 520 | struct file *file = lo->lo_backing_file; |
86af5952 | 521 | struct bio_vec tmp; |
40326d8a | 522 | unsigned int offset; |
86af5952 | 523 | int nr_bvec = 0; |
bc07c10a ML |
524 | int ret; |
525 | ||
86af5952 ML |
526 | rq_for_each_bvec(tmp, rq, rq_iter) |
527 | nr_bvec++; | |
528 | ||
40326d8a | 529 | if (rq->bio != rq->biotail) { |
40326d8a | 530 | |
86af5952 | 531 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 532 | GFP_NOIO); |
40326d8a SL |
533 | if (!bvec) |
534 | return -EIO; | |
535 | cmd->bvec = bvec; | |
536 | ||
537 | /* | |
538 | * The bios of the request may be started from the middle of | |
539 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 540 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
541 | * API will take care of all details for us. |
542 | */ | |
86af5952 | 543 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
544 | *bvec = tmp; |
545 | bvec++; | |
546 | } | |
547 | bvec = cmd->bvec; | |
548 | offset = 0; | |
549 | } else { | |
550 | /* | |
551 | * Same here, this bio may be started from the middle of the | |
552 | * 'bvec' because of bio splitting, so offset from the bvec | |
553 | * must be passed to iov iterator | |
554 | */ | |
555 | offset = bio->bi_iter.bi_bvec_done; | |
556 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 557 | } |
92d77332 | 558 | atomic_set(&cmd->ref, 2); |
bc07c10a | 559 | |
b6207430 | 560 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 561 | iter.iov_offset = offset; |
bc07c10a ML |
562 | |
563 | cmd->iocb.ki_pos = pos; | |
564 | cmd->iocb.ki_filp = file; | |
565 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
566 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 567 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
d4478e92 SL |
568 | if (cmd->css) |
569 | kthread_associate_blkcg(cmd->css); | |
bc07c10a ML |
570 | |
571 | if (rw == WRITE) | |
bb7462b6 | 572 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 573 | else |
bb7462b6 | 574 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 575 | |
92d77332 | 576 | lo_rw_aio_do_completion(cmd); |
d4478e92 | 577 | kthread_associate_blkcg(NULL); |
92d77332 | 578 | |
bc07c10a ML |
579 | if (ret != -EIOCBQUEUED) |
580 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
581 | return 0; | |
582 | } | |
583 | ||
c1c87c2b | 584 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
585 | { |
586 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 587 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
588 | |
589 | /* | |
590 | * lo_write_simple and lo_read_simple should have been covered | |
591 | * by io submit style function like lo_rw_aio(), one blocker | |
592 | * is that lo_read_simple() need to call flush_dcache_page after | |
593 | * the page is written from kernel, and it isn't easy to handle | |
594 | * this in io submit style function which submits all segments | |
595 | * of the req at one time. And direct read IO doesn't need to | |
596 | * run flush_dcache_page(). | |
597 | */ | |
c1c87c2b CH |
598 | switch (req_op(rq)) { |
599 | case REQ_OP_FLUSH: | |
600 | return lo_req_flush(lo, rq); | |
19372e27 | 601 | case REQ_OP_WRITE_ZEROES: |
efcfec57 DW |
602 | /* |
603 | * If the caller doesn't want deallocation, call zeroout to | |
604 | * write zeroes the range. Otherwise, punch them out. | |
605 | */ | |
606 | return lo_fallocate(lo, rq, pos, | |
607 | (rq->cmd_flags & REQ_NOUNMAP) ? | |
608 | FALLOC_FL_ZERO_RANGE : | |
609 | FALLOC_FL_PUNCH_HOLE); | |
610 | case REQ_OP_DISCARD: | |
611 | return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); | |
c1c87c2b CH |
612 | case REQ_OP_WRITE: |
613 | if (lo->transfer) | |
614 | return lo_write_transfer(lo, rq, pos); | |
615 | else if (cmd->use_aio) | |
616 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 617 | else |
c1c87c2b CH |
618 | return lo_write_simple(lo, rq, pos); |
619 | case REQ_OP_READ: | |
aa4d8616 | 620 | if (lo->transfer) |
c1c87c2b CH |
621 | return lo_read_transfer(lo, rq, pos); |
622 | else if (cmd->use_aio) | |
623 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 624 | else |
c1c87c2b CH |
625 | return lo_read_simple(lo, rq, pos); |
626 | default: | |
627 | WARN_ON_ONCE(1); | |
628 | return -EIO; | |
aa4d8616 | 629 | } |
1da177e4 LT |
630 | } |
631 | ||
2e5ab5f3 ML |
632 | static inline void loop_update_dio(struct loop_device *lo) |
633 | { | |
634 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
635 | lo->use_dio); | |
636 | } | |
637 | ||
06f0e9e6 ML |
638 | static void loop_reread_partitions(struct loop_device *lo, |
639 | struct block_device *bdev) | |
640 | { | |
641 | int rc; | |
642 | ||
f0b870df CH |
643 | mutex_lock(&bdev->bd_mutex); |
644 | rc = bdev_disk_changed(bdev, false); | |
645 | mutex_unlock(&bdev->bd_mutex); | |
06f0e9e6 ML |
646 | if (rc) |
647 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
648 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
649 | } | |
650 | ||
d2ac838e TT |
651 | static inline int is_loop_device(struct file *file) |
652 | { | |
653 | struct inode *i = file->f_mapping->host; | |
654 | ||
655 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
656 | } | |
657 | ||
658 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
659 | { | |
660 | struct inode *inode = file->f_mapping->host; | |
661 | struct file *f = file; | |
662 | ||
663 | /* Avoid recursion */ | |
664 | while (is_loop_device(f)) { | |
665 | struct loop_device *l; | |
666 | ||
667 | if (f->f_mapping->host->i_bdev == bdev) | |
668 | return -EBADF; | |
669 | ||
670 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
f7c8a412 | 671 | if (l->lo_state != Lo_bound) { |
d2ac838e TT |
672 | return -EINVAL; |
673 | } | |
674 | f = l->lo_backing_file; | |
675 | } | |
676 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
677 | return -EINVAL; | |
678 | return 0; | |
679 | } | |
680 | ||
1da177e4 LT |
681 | /* |
682 | * loop_change_fd switched the backing store of a loopback device to | |
683 | * a new file. This is useful for operating system installers to free up | |
684 | * the original file and in High Availability environments to switch to | |
685 | * an alternative location for the content in case of server meltdown. | |
686 | * This can only work if the loop device is used read-only, and if the | |
687 | * new backing store is the same size and type as the old backing store. | |
688 | */ | |
bb214884 AV |
689 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
690 | unsigned int arg) | |
1da177e4 | 691 | { |
1dded9ac | 692 | struct file *file = NULL, *old_file; |
1da177e4 | 693 | int error; |
85b0a54a | 694 | bool partscan; |
1da177e4 | 695 | |
c28445fa | 696 | error = mutex_lock_killable(&loop_ctl_mutex); |
c3710770 JK |
697 | if (error) |
698 | return error; | |
1da177e4 LT |
699 | error = -ENXIO; |
700 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 701 | goto out_err; |
1da177e4 LT |
702 | |
703 | /* the loop device has to be read-only */ | |
704 | error = -EINVAL; | |
705 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 706 | goto out_err; |
1da177e4 LT |
707 | |
708 | error = -EBADF; | |
709 | file = fget(arg); | |
710 | if (!file) | |
1dded9ac | 711 | goto out_err; |
1da177e4 | 712 | |
d2ac838e TT |
713 | error = loop_validate_file(file, bdev); |
714 | if (error) | |
1dded9ac | 715 | goto out_err; |
d2ac838e | 716 | |
1da177e4 LT |
717 | old_file = lo->lo_backing_file; |
718 | ||
719 | error = -EINVAL; | |
720 | ||
1da177e4 LT |
721 | /* size of the new backing store needs to be the same */ |
722 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 723 | goto out_err; |
1da177e4 LT |
724 | |
725 | /* and ... switch */ | |
43cade80 OS |
726 | blk_mq_freeze_queue(lo->lo_queue); |
727 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
728 | lo->lo_backing_file = file; | |
729 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
730 | mapping_set_gfp_mask(file->f_mapping, | |
731 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
732 | loop_update_dio(lo); | |
733 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 734 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c3710770 | 735 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
736 | /* |
737 | * We must drop file reference outside of loop_ctl_mutex as dropping | |
738 | * the file ref can take bd_mutex which creates circular locking | |
739 | * dependency. | |
740 | */ | |
741 | fput(old_file); | |
85b0a54a JK |
742 | if (partscan) |
743 | loop_reread_partitions(lo, bdev); | |
1da177e4 LT |
744 | return 0; |
745 | ||
1dded9ac | 746 | out_err: |
c3710770 | 747 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
748 | if (file) |
749 | fput(file); | |
1da177e4 LT |
750 | return error; |
751 | } | |
752 | ||
ee862730 MB |
753 | /* loop sysfs attributes */ |
754 | ||
755 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
756 | ssize_t (*callback)(struct loop_device *, char *)) | |
757 | { | |
34dd82af KS |
758 | struct gendisk *disk = dev_to_disk(dev); |
759 | struct loop_device *lo = disk->private_data; | |
ee862730 | 760 | |
34dd82af | 761 | return callback(lo, page); |
ee862730 MB |
762 | } |
763 | ||
764 | #define LOOP_ATTR_RO(_name) \ | |
765 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
766 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
767 | struct device_attribute *attr, char *b) \ | |
768 | { \ | |
769 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
770 | } \ | |
771 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 772 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
773 | |
774 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
775 | { | |
776 | ssize_t ret; | |
777 | char *p = NULL; | |
778 | ||
05eb0f25 | 779 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 780 | if (lo->lo_backing_file) |
9bf39ab2 | 781 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 782 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
783 | |
784 | if (IS_ERR_OR_NULL(p)) | |
785 | ret = PTR_ERR(p); | |
786 | else { | |
787 | ret = strlen(p); | |
788 | memmove(buf, p, ret); | |
789 | buf[ret++] = '\n'; | |
790 | buf[ret] = 0; | |
791 | } | |
792 | ||
793 | return ret; | |
794 | } | |
795 | ||
796 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
797 | { | |
798 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
799 | } | |
800 | ||
801 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
802 | { | |
803 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
804 | } | |
805 | ||
806 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
807 | { | |
808 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
809 | ||
810 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
811 | } | |
812 | ||
e03c8dd1 KS |
813 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
814 | { | |
815 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
816 | ||
817 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
818 | } | |
819 | ||
2e5ab5f3 ML |
820 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
821 | { | |
822 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
823 | ||
824 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
825 | } | |
826 | ||
ee862730 MB |
827 | LOOP_ATTR_RO(backing_file); |
828 | LOOP_ATTR_RO(offset); | |
829 | LOOP_ATTR_RO(sizelimit); | |
830 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 831 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 832 | LOOP_ATTR_RO(dio); |
ee862730 MB |
833 | |
834 | static struct attribute *loop_attrs[] = { | |
835 | &loop_attr_backing_file.attr, | |
836 | &loop_attr_offset.attr, | |
837 | &loop_attr_sizelimit.attr, | |
838 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 839 | &loop_attr_partscan.attr, |
2e5ab5f3 | 840 | &loop_attr_dio.attr, |
ee862730 MB |
841 | NULL, |
842 | }; | |
843 | ||
844 | static struct attribute_group loop_attribute_group = { | |
845 | .name = "loop", | |
846 | .attrs= loop_attrs, | |
847 | }; | |
848 | ||
d3349b6b | 849 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 850 | { |
d3349b6b TH |
851 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
852 | &loop_attribute_group); | |
ee862730 MB |
853 | } |
854 | ||
855 | static void loop_sysfs_exit(struct loop_device *lo) | |
856 | { | |
d3349b6b TH |
857 | if (lo->sysfs_inited) |
858 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
859 | &loop_attribute_group); | |
ee862730 MB |
860 | } |
861 | ||
dfaa2ef6 LC |
862 | static void loop_config_discard(struct loop_device *lo) |
863 | { | |
864 | struct file *file = lo->lo_backing_file; | |
865 | struct inode *inode = file->f_mapping->host; | |
866 | struct request_queue *q = lo->lo_queue; | |
867 | ||
868 | /* | |
869 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 870 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
871 | * encryption is enabled, because it may give an attacker |
872 | * useful information. | |
873 | */ | |
874 | if ((!file->f_op->fallocate) || | |
875 | lo->lo_encrypt_key_size) { | |
876 | q->limits.discard_granularity = 0; | |
877 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 878 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 879 | blk_queue_max_write_zeroes_sectors(q, 0); |
8b904b5b | 880 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); |
dfaa2ef6 LC |
881 | return; |
882 | } | |
883 | ||
884 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 885 | q->limits.discard_alignment = 0; |
f2c6df7d | 886 | |
1e6ec9ea OS |
887 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
888 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
8b904b5b | 889 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); |
dfaa2ef6 LC |
890 | } |
891 | ||
e03a3d7a ML |
892 | static void loop_unprepare_queue(struct loop_device *lo) |
893 | { | |
3989144f | 894 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
895 | kthread_stop(lo->worker_task); |
896 | } | |
897 | ||
b2ee7d46 N |
898 | static int loop_kthread_worker_fn(void *worker_ptr) |
899 | { | |
d0a255e7 | 900 | current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO; |
b2ee7d46 N |
901 | return kthread_worker_fn(worker_ptr); |
902 | } | |
903 | ||
e03a3d7a ML |
904 | static int loop_prepare_queue(struct loop_device *lo) |
905 | { | |
3989144f | 906 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 907 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
908 | &lo->worker, "loop%d", lo->lo_number); |
909 | if (IS_ERR(lo->worker_task)) | |
910 | return -ENOMEM; | |
911 | set_user_nice(lo->worker_task, MIN_NICE); | |
912 | return 0; | |
913 | } | |
914 | ||
56a85fd8 HH |
915 | static void loop_update_rotational(struct loop_device *lo) |
916 | { | |
917 | struct file *file = lo->lo_backing_file; | |
918 | struct inode *file_inode = file->f_mapping->host; | |
919 | struct block_device *file_bdev = file_inode->i_sb->s_bdev; | |
920 | struct request_queue *q = lo->lo_queue; | |
921 | bool nonrot = true; | |
922 | ||
923 | /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ | |
924 | if (file_bdev) | |
925 | nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); | |
926 | ||
927 | if (nonrot) | |
928 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q); | |
929 | else | |
930 | blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); | |
931 | } | |
932 | ||
bb214884 | 933 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
934 | struct block_device *bdev, unsigned int arg) |
935 | { | |
d2ac838e | 936 | struct file *file; |
1da177e4 LT |
937 | struct inode *inode; |
938 | struct address_space *mapping; | |
89e524c0 | 939 | struct block_device *claimed_bdev = NULL; |
1da177e4 LT |
940 | int lo_flags = 0; |
941 | int error; | |
942 | loff_t size; | |
85b0a54a | 943 | bool partscan; |
1da177e4 LT |
944 | |
945 | /* This is safe, since we have a reference from open(). */ | |
946 | __module_get(THIS_MODULE); | |
947 | ||
948 | error = -EBADF; | |
949 | file = fget(arg); | |
950 | if (!file) | |
951 | goto out; | |
952 | ||
33ec3e53 JK |
953 | /* |
954 | * If we don't hold exclusive handle for the device, upgrade to it | |
955 | * here to avoid changing device under exclusive owner. | |
956 | */ | |
957 | if (!(mode & FMODE_EXCL)) { | |
89e524c0 JK |
958 | claimed_bdev = bd_start_claiming(bdev, loop_set_fd); |
959 | if (IS_ERR(claimed_bdev)) { | |
960 | error = PTR_ERR(claimed_bdev); | |
33ec3e53 | 961 | goto out_putf; |
89e524c0 | 962 | } |
33ec3e53 JK |
963 | } |
964 | ||
c28445fa | 965 | error = mutex_lock_killable(&loop_ctl_mutex); |
757ecf40 | 966 | if (error) |
33ec3e53 | 967 | goto out_bdev; |
757ecf40 | 968 | |
1da177e4 LT |
969 | error = -EBUSY; |
970 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 971 | goto out_unlock; |
1da177e4 | 972 | |
d2ac838e TT |
973 | error = loop_validate_file(file, bdev); |
974 | if (error) | |
757ecf40 | 975 | goto out_unlock; |
1da177e4 LT |
976 | |
977 | mapping = file->f_mapping; | |
978 | inode = mapping->host; | |
979 | ||
456be148 | 980 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 981 | !file->f_op->write_iter) |
456be148 | 982 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 983 | |
456be148 | 984 | error = -EFBIG; |
1da177e4 | 985 | size = get_loop_size(lo, file); |
456be148 | 986 | if ((loff_t)(sector_t)size != size) |
757ecf40 | 987 | goto out_unlock; |
e03a3d7a ML |
988 | error = loop_prepare_queue(lo); |
989 | if (error) | |
757ecf40 | 990 | goto out_unlock; |
1da177e4 | 991 | |
456be148 | 992 | error = 0; |
1da177e4 LT |
993 | |
994 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
995 | ||
2e5ab5f3 | 996 | lo->use_dio = false; |
1da177e4 LT |
997 | lo->lo_device = bdev; |
998 | lo->lo_flags = lo_flags; | |
999 | lo->lo_backing_file = file; | |
aa4d8616 | 1000 | lo->transfer = NULL; |
1da177e4 LT |
1001 | lo->ioctl = NULL; |
1002 | lo->lo_sizelimit = 0; | |
1003 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
1004 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1005 | ||
68db1961 | 1006 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 1007 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 1008 | |
85560117 MC |
1009 | if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) { |
1010 | /* In case of direct I/O, match underlying block size */ | |
1011 | unsigned short bsize = bdev_logical_block_size( | |
1012 | inode->i_sb->s_bdev); | |
1013 | ||
1014 | blk_queue_logical_block_size(lo->lo_queue, bsize); | |
1015 | blk_queue_physical_block_size(lo->lo_queue, bsize); | |
1016 | blk_queue_io_min(lo->lo_queue, bsize); | |
1017 | } | |
1018 | ||
56a85fd8 | 1019 | loop_update_rotational(lo); |
2e5ab5f3 | 1020 | loop_update_dio(lo); |
73285082 | 1021 | set_capacity(lo->lo_disk, size); |
1da177e4 | 1022 | bd_set_size(bdev, size << 9); |
ee862730 | 1023 | loop_sysfs_init(lo); |
c3473c63 DZ |
1024 | /* let user-space know about the new size */ |
1025 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 1026 | |
8a0740c4 OS |
1027 | set_blocksize(bdev, S_ISBLK(inode->i_mode) ? |
1028 | block_size(inode->i_bdev) : PAGE_SIZE); | |
1da177e4 | 1029 | |
6c997918 | 1030 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1031 | if (part_shift) |
1032 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 1033 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c1681bf8 AP |
1034 | |
1035 | /* Grab the block_device to prevent its destruction after we | |
a2505b79 | 1036 | * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). |
c1681bf8 AP |
1037 | */ |
1038 | bdgrab(bdev); | |
757ecf40 | 1039 | mutex_unlock(&loop_ctl_mutex); |
85b0a54a JK |
1040 | if (partscan) |
1041 | loop_reread_partitions(lo, bdev); | |
89e524c0 JK |
1042 | if (claimed_bdev) |
1043 | bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); | |
1da177e4 LT |
1044 | return 0; |
1045 | ||
757ecf40 JK |
1046 | out_unlock: |
1047 | mutex_unlock(&loop_ctl_mutex); | |
33ec3e53 | 1048 | out_bdev: |
89e524c0 JK |
1049 | if (claimed_bdev) |
1050 | bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); | |
757ecf40 | 1051 | out_putf: |
1da177e4 | 1052 | fput(file); |
757ecf40 | 1053 | out: |
1da177e4 LT |
1054 | /* This is safe: open() is still holding a reference. */ |
1055 | module_put(THIS_MODULE); | |
1056 | return error; | |
1057 | } | |
1058 | ||
1059 | static int | |
1060 | loop_release_xfer(struct loop_device *lo) | |
1061 | { | |
1062 | int err = 0; | |
1063 | struct loop_func_table *xfer = lo->lo_encryption; | |
1064 | ||
1065 | if (xfer) { | |
1066 | if (xfer->release) | |
1067 | err = xfer->release(lo); | |
1068 | lo->transfer = NULL; | |
1069 | lo->lo_encryption = NULL; | |
1070 | module_put(xfer->owner); | |
1071 | } | |
1072 | return err; | |
1073 | } | |
1074 | ||
1075 | static int | |
1076 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1077 | const struct loop_info64 *i) | |
1078 | { | |
1079 | int err = 0; | |
1080 | ||
1081 | if (xfer) { | |
1082 | struct module *owner = xfer->owner; | |
1083 | ||
1084 | if (!try_module_get(owner)) | |
1085 | return -EINVAL; | |
1086 | if (xfer->init) | |
1087 | err = xfer->init(lo, i); | |
1088 | if (err) | |
1089 | module_put(owner); | |
1090 | else | |
1091 | lo->lo_encryption = xfer; | |
1092 | } | |
1093 | return err; | |
1094 | } | |
1095 | ||
0da03cab | 1096 | static int __loop_clr_fd(struct loop_device *lo, bool release) |
1da177e4 | 1097 | { |
7ccd0791 | 1098 | struct file *filp = NULL; |
b4e3ca1a | 1099 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1100 | struct block_device *bdev = lo->lo_device; |
7ccd0791 | 1101 | int err = 0; |
0da03cab JK |
1102 | bool partscan = false; |
1103 | int lo_number; | |
1da177e4 | 1104 | |
7ccd0791 JK |
1105 | mutex_lock(&loop_ctl_mutex); |
1106 | if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { | |
1107 | err = -ENXIO; | |
1108 | goto out_unlock; | |
1109 | } | |
1da177e4 | 1110 | |
7ccd0791 JK |
1111 | filp = lo->lo_backing_file; |
1112 | if (filp == NULL) { | |
1113 | err = -EINVAL; | |
1114 | goto out_unlock; | |
1115 | } | |
1da177e4 | 1116 | |
f8933667 ML |
1117 | /* freeze request queue during the transition */ |
1118 | blk_mq_freeze_queue(lo->lo_queue); | |
1119 | ||
1da177e4 | 1120 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 1121 | lo->lo_backing_file = NULL; |
05eb0f25 | 1122 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1123 | |
1124 | loop_release_xfer(lo); | |
1125 | lo->transfer = NULL; | |
1126 | lo->ioctl = NULL; | |
1127 | lo->lo_device = NULL; | |
1128 | lo->lo_encryption = NULL; | |
1129 | lo->lo_offset = 0; | |
1130 | lo->lo_sizelimit = 0; | |
1131 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1132 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1133 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1134 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1135 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1136 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1137 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1138 | if (bdev) { |
1139 | bdput(bdev); | |
bb214884 | 1140 | invalidate_bdev(bdev); |
eedffa28 | 1141 | bdev->bd_inode->i_mapping->wb_err = 0; |
c1681bf8 | 1142 | } |
73285082 | 1143 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1144 | loop_sysfs_exit(lo); |
c3473c63 | 1145 | if (bdev) { |
bb214884 | 1146 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1147 | /* let user-space know about this change */ |
1148 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1149 | } | |
1da177e4 | 1150 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1da177e4 LT |
1151 | /* This is safe: open() is still holding a reference. */ |
1152 | module_put(THIS_MODULE); | |
f8933667 ML |
1153 | blk_mq_unfreeze_queue(lo->lo_queue); |
1154 | ||
0da03cab JK |
1155 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; |
1156 | lo_number = lo->lo_number; | |
0da03cab JK |
1157 | loop_unprepare_queue(lo); |
1158 | out_unlock: | |
1159 | mutex_unlock(&loop_ctl_mutex); | |
1160 | if (partscan) { | |
d57f3374 JK |
1161 | /* |
1162 | * bd_mutex has been held already in release path, so don't | |
1163 | * acquire it if this function is called in such case. | |
1164 | * | |
1165 | * If the reread partition isn't from release path, lo_refcnt | |
1166 | * must be at least one and it can only become zero when the | |
1167 | * current holder is released. | |
1168 | */ | |
f0b870df CH |
1169 | if (!release) |
1170 | mutex_lock(&bdev->bd_mutex); | |
1171 | err = bdev_disk_changed(bdev, false); | |
1172 | if (!release) | |
1173 | mutex_unlock(&bdev->bd_mutex); | |
40853d6f DZ |
1174 | if (err) |
1175 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
1176 | __func__, lo_number, err); | |
d57f3374 JK |
1177 | /* Device is gone, no point in returning error */ |
1178 | err = 0; | |
1179 | } | |
758a58d0 DZ |
1180 | |
1181 | /* | |
1182 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1183 | * finished. | |
1184 | * | |
1185 | * There cannot be anybody else entering __loop_clr_fd() as | |
1186 | * lo->lo_backing_file is already cleared and Lo_rundown state | |
1187 | * protects us from all the other places trying to change the 'lo' | |
1188 | * device. | |
1189 | */ | |
1190 | mutex_lock(&loop_ctl_mutex); | |
1191 | lo->lo_flags = 0; | |
1192 | if (!part_shift) | |
1193 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1194 | lo->lo_state = Lo_unbound; | |
1195 | mutex_unlock(&loop_ctl_mutex); | |
1196 | ||
f028f3b2 | 1197 | /* |
310ca162 TH |
1198 | * Need not hold loop_ctl_mutex to fput backing file. |
1199 | * Calling fput holding loop_ctl_mutex triggers a circular | |
f028f3b2 | 1200 | * lock dependency possibility warning as fput can take |
310ca162 | 1201 | * bd_mutex which is usually taken before loop_ctl_mutex. |
f028f3b2 | 1202 | */ |
7ccd0791 JK |
1203 | if (filp) |
1204 | fput(filp); | |
1205 | return err; | |
1da177e4 LT |
1206 | } |
1207 | ||
a2505b79 JK |
1208 | static int loop_clr_fd(struct loop_device *lo) |
1209 | { | |
7ccd0791 JK |
1210 | int err; |
1211 | ||
c28445fa | 1212 | err = mutex_lock_killable(&loop_ctl_mutex); |
7ccd0791 JK |
1213 | if (err) |
1214 | return err; | |
1215 | if (lo->lo_state != Lo_bound) { | |
1216 | mutex_unlock(&loop_ctl_mutex); | |
a2505b79 | 1217 | return -ENXIO; |
7ccd0791 | 1218 | } |
a2505b79 JK |
1219 | /* |
1220 | * If we've explicitly asked to tear down the loop device, | |
1221 | * and it has an elevated reference count, set it for auto-teardown when | |
1222 | * the last reference goes away. This stops $!~#$@ udev from | |
1223 | * preventing teardown because it decided that it needs to run blkid on | |
1224 | * the loopback device whenever they appear. xfstests is notorious for | |
1225 | * failing tests because blkid via udev races with a losetup | |
1226 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1227 | * command to fail with EBUSY. | |
1228 | */ | |
1229 | if (atomic_read(&lo->lo_refcnt) > 1) { | |
1230 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
1231 | mutex_unlock(&loop_ctl_mutex); | |
1232 | return 0; | |
1233 | } | |
1234 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1235 | mutex_unlock(&loop_ctl_mutex); |
a2505b79 | 1236 | |
0da03cab | 1237 | return __loop_clr_fd(lo, false); |
a2505b79 JK |
1238 | } |
1239 | ||
1da177e4 LT |
1240 | static int |
1241 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1242 | { | |
1243 | int err; | |
1244 | struct loop_func_table *xfer; | |
e4849737 | 1245 | kuid_t uid = current_uid(); |
85b0a54a JK |
1246 | struct block_device *bdev; |
1247 | bool partscan = false; | |
1da177e4 | 1248 | |
c28445fa | 1249 | err = mutex_lock_killable(&loop_ctl_mutex); |
550df5fd JK |
1250 | if (err) |
1251 | return err; | |
b0fafa81 | 1252 | if (lo->lo_encrypt_key_size && |
e4849737 | 1253 | !uid_eq(lo->lo_key_owner, uid) && |
550df5fd JK |
1254 | !capable(CAP_SYS_ADMIN)) { |
1255 | err = -EPERM; | |
1256 | goto out_unlock; | |
1257 | } | |
1258 | if (lo->lo_state != Lo_bound) { | |
1259 | err = -ENXIO; | |
1260 | goto out_unlock; | |
1261 | } | |
1262 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { | |
1263 | err = -EINVAL; | |
1264 | goto out_unlock; | |
1265 | } | |
1da177e4 | 1266 | |
5db470e2 JK |
1267 | if (lo->lo_offset != info->lo_offset || |
1268 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1269 | sync_blockdev(lo->lo_device); | |
1270 | kill_bdev(lo->lo_device); | |
1271 | } | |
1272 | ||
ecdd0959 ML |
1273 | /* I/O need to be drained during transfer transition */ |
1274 | blk_mq_freeze_queue(lo->lo_queue); | |
1275 | ||
1da177e4 LT |
1276 | err = loop_release_xfer(lo); |
1277 | if (err) | |
550df5fd | 1278 | goto out_unfreeze; |
1da177e4 LT |
1279 | |
1280 | if (info->lo_encrypt_type) { | |
1281 | unsigned int type = info->lo_encrypt_type; | |
1282 | ||
1e047eaa TH |
1283 | if (type >= MAX_LO_CRYPT) { |
1284 | err = -EINVAL; | |
550df5fd | 1285 | goto out_unfreeze; |
1e047eaa | 1286 | } |
1da177e4 | 1287 | xfer = xfer_funcs[type]; |
1e047eaa TH |
1288 | if (xfer == NULL) { |
1289 | err = -EINVAL; | |
550df5fd | 1290 | goto out_unfreeze; |
1e047eaa | 1291 | } |
1da177e4 LT |
1292 | } else |
1293 | xfer = NULL; | |
1294 | ||
1295 | err = loop_init_xfer(lo, xfer, info); | |
1296 | if (err) | |
550df5fd | 1297 | goto out_unfreeze; |
1da177e4 LT |
1298 | |
1299 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea | 1300 | lo->lo_sizelimit != info->lo_sizelimit) { |
5db470e2 JK |
1301 | /* kill_bdev should have truncated all the pages */ |
1302 | if (lo->lo_device->bd_inode->i_mapping->nrpages) { | |
1303 | err = -EAGAIN; | |
1304 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1305 | __func__, lo->lo_number, lo->lo_file_name, | |
1306 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1307 | goto out_unfreeze; | |
1308 | } | |
1e6ec9ea | 1309 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { |
ecdd0959 | 1310 | err = -EFBIG; |
550df5fd | 1311 | goto out_unfreeze; |
ecdd0959 | 1312 | } |
b040ad9c | 1313 | } |
541c742a | 1314 | |
dfaa2ef6 | 1315 | loop_config_discard(lo); |
1da177e4 LT |
1316 | |
1317 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1318 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1319 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1320 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1321 | ||
1322 | if (!xfer) | |
1323 | xfer = &none_funcs; | |
1324 | lo->transfer = xfer->transfer; | |
1325 | lo->ioctl = xfer->ioctl; | |
1326 | ||
96c58655 DW |
1327 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1328 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1329 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1330 | ||
1da177e4 LT |
1331 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1332 | lo->lo_init[0] = info->lo_init[0]; | |
1333 | lo->lo_init[1] = info->lo_init[1]; | |
1334 | if (info->lo_encrypt_key_size) { | |
1335 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1336 | info->lo_encrypt_key_size); | |
b0fafa81 | 1337 | lo->lo_key_owner = uid; |
aa4d8616 | 1338 | } |
1da177e4 | 1339 | |
2e5ab5f3 ML |
1340 | /* update dio if lo_offset or transfer is changed */ |
1341 | __loop_update_dio(lo, lo->use_dio); | |
1342 | ||
550df5fd | 1343 | out_unfreeze: |
ecdd0959 | 1344 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 OS |
1345 | |
1346 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1347 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1348 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1349 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
85b0a54a JK |
1350 | bdev = lo->lo_device; |
1351 | partscan = true; | |
e02898b4 | 1352 | } |
550df5fd JK |
1353 | out_unlock: |
1354 | mutex_unlock(&loop_ctl_mutex); | |
85b0a54a JK |
1355 | if (partscan) |
1356 | loop_reread_partitions(lo, bdev); | |
e02898b4 | 1357 | |
ecdd0959 | 1358 | return err; |
1da177e4 LT |
1359 | } |
1360 | ||
1361 | static int | |
1362 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1363 | { | |
b1ab5fa3 | 1364 | struct path path; |
1da177e4 | 1365 | struct kstat stat; |
2d1d4c1e | 1366 | int ret; |
1da177e4 | 1367 | |
c28445fa | 1368 | ret = mutex_lock_killable(&loop_ctl_mutex); |
4a5ce9ba JK |
1369 | if (ret) |
1370 | return ret; | |
2d1d4c1e | 1371 | if (lo->lo_state != Lo_bound) { |
310ca162 | 1372 | mutex_unlock(&loop_ctl_mutex); |
1da177e4 | 1373 | return -ENXIO; |
2d1d4c1e OS |
1374 | } |
1375 | ||
1da177e4 LT |
1376 | memset(info, 0, sizeof(*info)); |
1377 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1378 | info->lo_offset = lo->lo_offset; |
1379 | info->lo_sizelimit = lo->lo_sizelimit; | |
1380 | info->lo_flags = lo->lo_flags; | |
1381 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1382 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1383 | info->lo_encrypt_type = | |
1384 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1385 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1386 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1387 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1388 | lo->lo_encrypt_key_size); | |
1389 | } | |
2d1d4c1e | 1390 | |
310ca162 | 1391 | /* Drop loop_ctl_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1392 | path = lo->lo_backing_file->f_path; |
1393 | path_get(&path); | |
310ca162 | 1394 | mutex_unlock(&loop_ctl_mutex); |
b1ab5fa3 | 1395 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1396 | if (!ret) { |
1397 | info->lo_device = huge_encode_dev(stat.dev); | |
1398 | info->lo_inode = stat.ino; | |
1399 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1400 | } | |
b1ab5fa3 | 1401 | path_put(&path); |
2d1d4c1e | 1402 | return ret; |
1da177e4 LT |
1403 | } |
1404 | ||
1405 | static void | |
1406 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1407 | { | |
1408 | memset(info64, 0, sizeof(*info64)); | |
1409 | info64->lo_number = info->lo_number; | |
1410 | info64->lo_device = info->lo_device; | |
1411 | info64->lo_inode = info->lo_inode; | |
1412 | info64->lo_rdevice = info->lo_rdevice; | |
1413 | info64->lo_offset = info->lo_offset; | |
1414 | info64->lo_sizelimit = 0; | |
1415 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1416 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1417 | info64->lo_flags = info->lo_flags; | |
1418 | info64->lo_init[0] = info->lo_init[0]; | |
1419 | info64->lo_init[1] = info->lo_init[1]; | |
1420 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1421 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1422 | else | |
1423 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1424 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1425 | } | |
1426 | ||
1427 | static int | |
1428 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1429 | { | |
1430 | memset(info, 0, sizeof(*info)); | |
1431 | info->lo_number = info64->lo_number; | |
1432 | info->lo_device = info64->lo_device; | |
1433 | info->lo_inode = info64->lo_inode; | |
1434 | info->lo_rdevice = info64->lo_rdevice; | |
1435 | info->lo_offset = info64->lo_offset; | |
1436 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1437 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1438 | info->lo_flags = info64->lo_flags; | |
1439 | info->lo_init[0] = info64->lo_init[0]; | |
1440 | info->lo_init[1] = info64->lo_init[1]; | |
1441 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1442 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1443 | else | |
1444 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1445 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1446 | ||
1447 | /* error in case values were truncated */ | |
1448 | if (info->lo_device != info64->lo_device || | |
1449 | info->lo_rdevice != info64->lo_rdevice || | |
1450 | info->lo_inode != info64->lo_inode || | |
1451 | info->lo_offset != info64->lo_offset) | |
1452 | return -EOVERFLOW; | |
1453 | ||
1454 | return 0; | |
1455 | } | |
1456 | ||
1457 | static int | |
1458 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1459 | { | |
1460 | struct loop_info info; | |
1461 | struct loop_info64 info64; | |
1462 | ||
1463 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1464 | return -EFAULT; | |
1465 | loop_info64_from_old(&info, &info64); | |
1466 | return loop_set_status(lo, &info64); | |
1467 | } | |
1468 | ||
1469 | static int | |
1470 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1471 | { | |
1472 | struct loop_info64 info64; | |
1473 | ||
1474 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1475 | return -EFAULT; | |
1476 | return loop_set_status(lo, &info64); | |
1477 | } | |
1478 | ||
1479 | static int | |
1480 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1481 | struct loop_info info; | |
1482 | struct loop_info64 info64; | |
bdac616d | 1483 | int err; |
1da177e4 | 1484 | |
4a5ce9ba | 1485 | if (!arg) |
bdac616d | 1486 | return -EINVAL; |
bdac616d | 1487 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1488 | if (!err) |
1489 | err = loop_info64_to_old(&info64, &info); | |
1490 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1491 | err = -EFAULT; | |
1492 | ||
1493 | return err; | |
1494 | } | |
1495 | ||
1496 | static int | |
1497 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1498 | struct loop_info64 info64; | |
bdac616d | 1499 | int err; |
1da177e4 | 1500 | |
4a5ce9ba | 1501 | if (!arg) |
bdac616d | 1502 | return -EINVAL; |
bdac616d | 1503 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1504 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1505 | err = -EFAULT; | |
1506 | ||
1507 | return err; | |
1508 | } | |
1509 | ||
51001b7d | 1510 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1511 | { |
53d66608 | 1512 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1513 | return -ENXIO; |
53d66608 | 1514 | |
1e6ec9ea | 1515 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1516 | } |
1517 | ||
ab1cb278 ML |
1518 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1519 | { | |
1520 | int error = -ENXIO; | |
1521 | if (lo->lo_state != Lo_bound) | |
1522 | goto out; | |
1523 | ||
1524 | __loop_update_dio(lo, !!arg); | |
1525 | if (lo->use_dio == !!arg) | |
1526 | return 0; | |
1527 | error = -EINVAL; | |
1528 | out: | |
1529 | return error; | |
1530 | } | |
1531 | ||
89e4fdec OS |
1532 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1533 | { | |
5db470e2 JK |
1534 | int err = 0; |
1535 | ||
89e4fdec OS |
1536 | if (lo->lo_state != Lo_bound) |
1537 | return -ENXIO; | |
1538 | ||
1539 | if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) | |
1540 | return -EINVAL; | |
1541 | ||
5db470e2 JK |
1542 | if (lo->lo_queue->limits.logical_block_size != arg) { |
1543 | sync_blockdev(lo->lo_device); | |
1544 | kill_bdev(lo->lo_device); | |
1545 | } | |
1546 | ||
89e4fdec OS |
1547 | blk_mq_freeze_queue(lo->lo_queue); |
1548 | ||
5db470e2 JK |
1549 | /* kill_bdev should have truncated all the pages */ |
1550 | if (lo->lo_queue->limits.logical_block_size != arg && | |
1551 | lo->lo_device->bd_inode->i_mapping->nrpages) { | |
1552 | err = -EAGAIN; | |
1553 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1554 | __func__, lo->lo_number, lo->lo_file_name, | |
1555 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1556 | goto out_unfreeze; | |
1557 | } | |
1558 | ||
89e4fdec | 1559 | blk_queue_logical_block_size(lo->lo_queue, arg); |
bf093753 OS |
1560 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1561 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec | 1562 | loop_update_dio(lo); |
5db470e2 | 1563 | out_unfreeze: |
89e4fdec OS |
1564 | blk_mq_unfreeze_queue(lo->lo_queue); |
1565 | ||
5db470e2 | 1566 | return err; |
89e4fdec OS |
1567 | } |
1568 | ||
a1316544 JK |
1569 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1570 | unsigned long arg) | |
1da177e4 | 1571 | { |
1da177e4 LT |
1572 | int err; |
1573 | ||
c28445fa | 1574 | err = mutex_lock_killable(&loop_ctl_mutex); |
3148ffbd | 1575 | if (err) |
a1316544 JK |
1576 | return err; |
1577 | switch (cmd) { | |
1578 | case LOOP_SET_CAPACITY: | |
1579 | err = loop_set_capacity(lo); | |
1580 | break; | |
1581 | case LOOP_SET_DIRECT_IO: | |
1582 | err = loop_set_dio(lo, arg); | |
1583 | break; | |
1584 | case LOOP_SET_BLOCK_SIZE: | |
1585 | err = loop_set_block_size(lo, arg); | |
1586 | break; | |
1587 | default: | |
1588 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1589 | } | |
1590 | mutex_unlock(&loop_ctl_mutex); | |
1591 | return err; | |
1592 | } | |
1593 | ||
1594 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, | |
1595 | unsigned int cmd, unsigned long arg) | |
1596 | { | |
1597 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1598 | int err; | |
3148ffbd | 1599 | |
1da177e4 LT |
1600 | switch (cmd) { |
1601 | case LOOP_SET_FD: | |
757ecf40 | 1602 | return loop_set_fd(lo, mode, bdev, arg); |
1da177e4 | 1603 | case LOOP_CHANGE_FD: |
c3710770 | 1604 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1605 | case LOOP_CLR_FD: |
7ccd0791 | 1606 | return loop_clr_fd(lo); |
1da177e4 | 1607 | case LOOP_SET_STATUS: |
7035b5df | 1608 | err = -EPERM; |
a1316544 | 1609 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1610 | err = loop_set_status_old(lo, |
1611 | (struct loop_info __user *)arg); | |
a1316544 | 1612 | } |
1da177e4 LT |
1613 | break; |
1614 | case LOOP_GET_STATUS: | |
4a5ce9ba | 1615 | return loop_get_status_old(lo, (struct loop_info __user *) arg); |
1da177e4 | 1616 | case LOOP_SET_STATUS64: |
7035b5df | 1617 | err = -EPERM; |
a1316544 | 1618 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1619 | err = loop_set_status64(lo, |
1620 | (struct loop_info64 __user *) arg); | |
a1316544 | 1621 | } |
1da177e4 LT |
1622 | break; |
1623 | case LOOP_GET_STATUS64: | |
4a5ce9ba | 1624 | return loop_get_status64(lo, (struct loop_info64 __user *) arg); |
a1316544 | 1625 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1626 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1627 | case LOOP_SET_BLOCK_SIZE: |
a1316544 JK |
1628 | if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) |
1629 | return -EPERM; | |
1630 | /* Fall through */ | |
1da177e4 | 1631 | default: |
a1316544 JK |
1632 | err = lo_simple_ioctl(lo, cmd, arg); |
1633 | break; | |
1da177e4 | 1634 | } |
f028f3b2 | 1635 | |
1da177e4 LT |
1636 | return err; |
1637 | } | |
1638 | ||
863d5b82 DH |
1639 | #ifdef CONFIG_COMPAT |
1640 | struct compat_loop_info { | |
1641 | compat_int_t lo_number; /* ioctl r/o */ | |
1642 | compat_dev_t lo_device; /* ioctl r/o */ | |
1643 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1644 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1645 | compat_int_t lo_offset; | |
1646 | compat_int_t lo_encrypt_type; | |
1647 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1648 | compat_int_t lo_flags; /* ioctl r/o */ | |
1649 | char lo_name[LO_NAME_SIZE]; | |
1650 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1651 | compat_ulong_t lo_init[2]; | |
1652 | char reserved[4]; | |
1653 | }; | |
1654 | ||
1655 | /* | |
1656 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1657 | * - noinlined to reduce stack space usage in main part of driver | |
1658 | */ | |
1659 | static noinline int | |
ba674cfc | 1660 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1661 | struct loop_info64 *info64) |
1662 | { | |
1663 | struct compat_loop_info info; | |
1664 | ||
1665 | if (copy_from_user(&info, arg, sizeof(info))) | |
1666 | return -EFAULT; | |
1667 | ||
1668 | memset(info64, 0, sizeof(*info64)); | |
1669 | info64->lo_number = info.lo_number; | |
1670 | info64->lo_device = info.lo_device; | |
1671 | info64->lo_inode = info.lo_inode; | |
1672 | info64->lo_rdevice = info.lo_rdevice; | |
1673 | info64->lo_offset = info.lo_offset; | |
1674 | info64->lo_sizelimit = 0; | |
1675 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1676 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1677 | info64->lo_flags = info.lo_flags; | |
1678 | info64->lo_init[0] = info.lo_init[0]; | |
1679 | info64->lo_init[1] = info.lo_init[1]; | |
1680 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1681 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1682 | else | |
1683 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1684 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1685 | return 0; | |
1686 | } | |
1687 | ||
1688 | /* | |
1689 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1690 | * - noinlined to reduce stack space usage in main part of driver | |
1691 | */ | |
1692 | static noinline int | |
1693 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1694 | struct compat_loop_info __user *arg) | |
1695 | { | |
1696 | struct compat_loop_info info; | |
1697 | ||
1698 | memset(&info, 0, sizeof(info)); | |
1699 | info.lo_number = info64->lo_number; | |
1700 | info.lo_device = info64->lo_device; | |
1701 | info.lo_inode = info64->lo_inode; | |
1702 | info.lo_rdevice = info64->lo_rdevice; | |
1703 | info.lo_offset = info64->lo_offset; | |
1704 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1705 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1706 | info.lo_flags = info64->lo_flags; | |
1707 | info.lo_init[0] = info64->lo_init[0]; | |
1708 | info.lo_init[1] = info64->lo_init[1]; | |
1709 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1710 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1711 | else | |
1712 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1713 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1714 | ||
1715 | /* error in case values were truncated */ | |
1716 | if (info.lo_device != info64->lo_device || | |
1717 | info.lo_rdevice != info64->lo_rdevice || | |
1718 | info.lo_inode != info64->lo_inode || | |
1719 | info.lo_offset != info64->lo_offset || | |
1720 | info.lo_init[0] != info64->lo_init[0] || | |
1721 | info.lo_init[1] != info64->lo_init[1]) | |
1722 | return -EOVERFLOW; | |
1723 | ||
1724 | if (copy_to_user(arg, &info, sizeof(info))) | |
1725 | return -EFAULT; | |
1726 | return 0; | |
1727 | } | |
1728 | ||
1729 | static int | |
1730 | loop_set_status_compat(struct loop_device *lo, | |
1731 | const struct compat_loop_info __user *arg) | |
1732 | { | |
1733 | struct loop_info64 info64; | |
1734 | int ret; | |
1735 | ||
1736 | ret = loop_info64_from_compat(arg, &info64); | |
1737 | if (ret < 0) | |
1738 | return ret; | |
1739 | return loop_set_status(lo, &info64); | |
1740 | } | |
1741 | ||
1742 | static int | |
1743 | loop_get_status_compat(struct loop_device *lo, | |
1744 | struct compat_loop_info __user *arg) | |
1745 | { | |
1746 | struct loop_info64 info64; | |
bdac616d | 1747 | int err; |
863d5b82 | 1748 | |
4a5ce9ba | 1749 | if (!arg) |
bdac616d | 1750 | return -EINVAL; |
bdac616d | 1751 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1752 | if (!err) |
1753 | err = loop_info64_to_compat(&info64, arg); | |
1754 | return err; | |
1755 | } | |
1756 | ||
bb214884 AV |
1757 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1758 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1759 | { |
bb214884 | 1760 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1761 | int err; |
1762 | ||
863d5b82 DH |
1763 | switch(cmd) { |
1764 | case LOOP_SET_STATUS: | |
550df5fd JK |
1765 | err = loop_set_status_compat(lo, |
1766 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1767 | break; |
1768 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1769 | err = loop_get_status_compat(lo, |
1770 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1771 | break; |
53d66608 | 1772 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1773 | case LOOP_CLR_FD: |
1774 | case LOOP_GET_STATUS64: | |
1775 | case LOOP_SET_STATUS64: | |
1776 | arg = (unsigned long) compat_ptr(arg); | |
d893ff86 | 1777 | /* fall through */ |
863d5b82 DH |
1778 | case LOOP_SET_FD: |
1779 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1780 | case LOOP_SET_BLOCK_SIZE: |
fdbe4eee | 1781 | case LOOP_SET_DIRECT_IO: |
bb214884 | 1782 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1783 | break; |
1784 | default: | |
1785 | err = -ENOIOCTLCMD; | |
1786 | break; | |
1787 | } | |
863d5b82 DH |
1788 | return err; |
1789 | } | |
1790 | #endif | |
1791 | ||
bb214884 | 1792 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1793 | { |
770fe30a | 1794 | struct loop_device *lo; |
0a42e99b | 1795 | int err; |
770fe30a | 1796 | |
0a42e99b JK |
1797 | err = mutex_lock_killable(&loop_ctl_mutex); |
1798 | if (err) | |
1799 | return err; | |
770fe30a KS |
1800 | lo = bdev->bd_disk->private_data; |
1801 | if (!lo) { | |
1802 | err = -ENXIO; | |
1803 | goto out; | |
1804 | } | |
1da177e4 | 1805 | |
f8933667 | 1806 | atomic_inc(&lo->lo_refcnt); |
770fe30a | 1807 | out: |
0a42e99b | 1808 | mutex_unlock(&loop_ctl_mutex); |
770fe30a | 1809 | return err; |
1da177e4 LT |
1810 | } |
1811 | ||
967d1dc1 | 1812 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1813 | { |
967d1dc1 | 1814 | struct loop_device *lo; |
1da177e4 | 1815 | |
0a42e99b | 1816 | mutex_lock(&loop_ctl_mutex); |
967d1dc1 | 1817 | lo = disk->private_data; |
f8933667 | 1818 | if (atomic_dec_return(&lo->lo_refcnt)) |
0a42e99b | 1819 | goto out_unlock; |
14f27939 MB |
1820 | |
1821 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
a2505b79 JK |
1822 | if (lo->lo_state != Lo_bound) |
1823 | goto out_unlock; | |
1824 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1825 | mutex_unlock(&loop_ctl_mutex); |
14f27939 MB |
1826 | /* |
1827 | * In autoclear mode, stop the loop thread | |
1828 | * and remove configuration after last close. | |
1829 | */ | |
0da03cab | 1830 | __loop_clr_fd(lo, true); |
7ccd0791 | 1831 | return; |
43cade80 | 1832 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
1833 | /* |
1834 | * Otherwise keep thread (if running) and config, | |
1835 | * but flush possible ongoing bios in thread. | |
1836 | */ | |
43cade80 OS |
1837 | blk_mq_freeze_queue(lo->lo_queue); |
1838 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 1839 | } |
96c58655 | 1840 | |
0a42e99b | 1841 | out_unlock: |
310ca162 | 1842 | mutex_unlock(&loop_ctl_mutex); |
ae665016 LT |
1843 | } |
1844 | ||
83d5cde4 | 1845 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1846 | .owner = THIS_MODULE, |
bb214884 AV |
1847 | .open = lo_open, |
1848 | .release = lo_release, | |
1849 | .ioctl = lo_ioctl, | |
863d5b82 | 1850 | #ifdef CONFIG_COMPAT |
bb214884 | 1851 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1852 | #endif |
1da177e4 LT |
1853 | }; |
1854 | ||
1855 | /* | |
1856 | * And now the modules code and kernel interface. | |
1857 | */ | |
73285082 | 1858 | static int max_loop; |
5657a819 | 1859 | module_param(max_loop, int, 0444); |
a47653fc | 1860 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
5657a819 | 1861 | module_param(max_part, int, 0444); |
476a4813 | 1862 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1863 | MODULE_LICENSE("GPL"); |
1864 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1865 | ||
1866 | int loop_register_transfer(struct loop_func_table *funcs) | |
1867 | { | |
1868 | unsigned int n = funcs->number; | |
1869 | ||
1870 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1871 | return -EINVAL; | |
1872 | xfer_funcs[n] = funcs; | |
1873 | return 0; | |
1874 | } | |
1875 | ||
34dd82af KS |
1876 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1877 | { | |
1878 | struct loop_device *lo = ptr; | |
1879 | struct loop_func_table *xfer = data; | |
1880 | ||
310ca162 | 1881 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
1882 | if (lo->lo_encryption == xfer) |
1883 | loop_release_xfer(lo); | |
310ca162 | 1884 | mutex_unlock(&loop_ctl_mutex); |
34dd82af KS |
1885 | return 0; |
1886 | } | |
1887 | ||
1da177e4 LT |
1888 | int loop_unregister_transfer(int number) |
1889 | { | |
1890 | unsigned int n = number; | |
1da177e4 LT |
1891 | struct loop_func_table *xfer; |
1892 | ||
1893 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1894 | return -EINVAL; | |
1895 | ||
1896 | xfer_funcs[n] = NULL; | |
34dd82af | 1897 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1898 | return 0; |
1899 | } | |
1900 | ||
1901 | EXPORT_SYMBOL(loop_register_transfer); | |
1902 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1903 | ||
fc17b653 | 1904 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1905 | const struct blk_mq_queue_data *bd) |
1906 | { | |
1894e916 JA |
1907 | struct request *rq = bd->rq; |
1908 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1909 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 1910 | |
1894e916 | 1911 | blk_mq_start_request(rq); |
b5dd2f60 | 1912 | |
f4aa4c7b | 1913 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1914 | return BLK_STS_IOERR; |
f4aa4c7b | 1915 | |
1894e916 | 1916 | switch (req_op(rq)) { |
f0225cac CH |
1917 | case REQ_OP_FLUSH: |
1918 | case REQ_OP_DISCARD: | |
19372e27 | 1919 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1920 | cmd->use_aio = false; |
f0225cac CH |
1921 | break; |
1922 | default: | |
1923 | cmd->use_aio = lo->use_dio; | |
1924 | break; | |
1925 | } | |
bc07c10a | 1926 | |
d4478e92 | 1927 | /* always use the first bio's css */ |
0b508bc9 | 1928 | #ifdef CONFIG_BLK_CGROUP |
db6638d7 DZ |
1929 | if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { |
1930 | cmd->css = &bio_blkcg(rq->bio)->css; | |
d4478e92 SL |
1931 | css_get(cmd->css); |
1932 | } else | |
1933 | #endif | |
1934 | cmd->css = NULL; | |
3989144f | 1935 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1936 | |
fc17b653 | 1937 | return BLK_STS_OK; |
b5dd2f60 ML |
1938 | } |
1939 | ||
1940 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1941 | { | |
1894e916 JA |
1942 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
1943 | const bool write = op_is_write(req_op(rq)); | |
1944 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 1945 | int ret = 0; |
b5dd2f60 | 1946 | |
f4829a9b CH |
1947 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1948 | ret = -EIO; | |
b5dd2f60 | 1949 | goto failed; |
f4829a9b | 1950 | } |
b5dd2f60 | 1951 | |
1894e916 | 1952 | ret = do_req_filebacked(lo, rq); |
b5dd2f60 | 1953 | failed: |
bc07c10a | 1954 | /* complete non-aio request */ |
fe2cb290 CH |
1955 | if (!cmd->use_aio || ret) { |
1956 | cmd->ret = ret ? -EIO : 0; | |
1894e916 | 1957 | blk_mq_complete_request(rq); |
fe2cb290 | 1958 | } |
b5dd2f60 ML |
1959 | } |
1960 | ||
e03a3d7a | 1961 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1962 | { |
1963 | struct loop_cmd *cmd = | |
e03a3d7a | 1964 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1965 | |
1966 | loop_handle_cmd(cmd); | |
1967 | } | |
1968 | ||
d6296d39 CH |
1969 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1970 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1971 | { |
1972 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1973 | ||
3989144f | 1974 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1975 | return 0; |
1976 | } | |
1977 | ||
f363b089 | 1978 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1979 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 1980 | .init_request = loop_init_request, |
fe2cb290 | 1981 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1982 | }; |
1983 | ||
34dd82af | 1984 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1985 | { |
1986 | struct loop_device *lo; | |
1987 | struct gendisk *disk; | |
34dd82af | 1988 | int err; |
73285082 | 1989 | |
68d740d7 | 1990 | err = -ENOMEM; |
73285082 | 1991 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1992 | if (!lo) |
73285082 | 1993 | goto out; |
34dd82af | 1994 | |
ef7e7c82 MP |
1995 | lo->lo_state = Lo_unbound; |
1996 | ||
c718aa65 | 1997 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1998 | if (i >= 0) { |
c718aa65 TH |
1999 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
2000 | if (err == -ENOSPC) | |
34dd82af | 2001 | err = -EEXIST; |
34dd82af | 2002 | } else { |
c718aa65 | 2003 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
2004 | } |
2005 | if (err < 0) | |
2006 | goto out_free_dev; | |
c718aa65 | 2007 | i = err; |
73285082 | 2008 | |
183cfb57 | 2009 | err = -ENOMEM; |
b5dd2f60 ML |
2010 | lo->tag_set.ops = &loop_mq_ops; |
2011 | lo->tag_set.nr_hw_queues = 1; | |
2012 | lo->tag_set.queue_depth = 128; | |
2013 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
2014 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
56d18f62 | 2015 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; |
b5dd2f60 ML |
2016 | lo->tag_set.driver_data = lo; |
2017 | ||
2018 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
2019 | if (err) | |
3ec981e3 | 2020 | goto out_free_idr; |
73285082 | 2021 | |
b5dd2f60 | 2022 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
38a3499f | 2023 | if (IS_ERR(lo->lo_queue)) { |
b5dd2f60 ML |
2024 | err = PTR_ERR(lo->lo_queue); |
2025 | goto out_cleanup_tags; | |
2026 | } | |
ef7e7c82 MP |
2027 | lo->lo_queue->queuedata = lo; |
2028 | ||
54bb0ade | 2029 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 2030 | |
5b5e20f4 | 2031 | /* |
40326d8a SL |
2032 | * By default, we do buffer IO, so it doesn't make sense to enable |
2033 | * merge because the I/O submitted to backing file is handled page by | |
2034 | * page. For directio mode, merge does help to dispatch bigger request | |
2035 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 | 2036 | */ |
8b904b5b | 2037 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
5b5e20f4 | 2038 | |
7a649737 | 2039 | err = -ENOMEM; |
476a4813 | 2040 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
2041 | if (!disk) |
2042 | goto out_free_queue; | |
2043 | ||
e03c8dd1 KS |
2044 | /* |
2045 | * Disable partition scanning by default. The in-kernel partition | |
2046 | * scanning can be requested individually per-device during its | |
2047 | * setup. Userspace can always add and remove partitions from all | |
2048 | * devices. The needed partition minors are allocated from the | |
2049 | * extended minor space, the main loop device numbers will continue | |
2050 | * to match the loop minors, regardless of the number of partitions | |
2051 | * used. | |
2052 | * | |
2053 | * If max_part is given, partition scanning is globally enabled for | |
2054 | * all loop devices. The minors for the main loop devices will be | |
2055 | * multiples of max_part. | |
2056 | * | |
2057 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
2058 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
2059 | * complicated, are too static, inflexible and may surprise | |
2060 | * userspace tools. Parameters like this in general should be avoided. | |
2061 | */ | |
2062 | if (!part_shift) | |
2063 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
2064 | disk->flags |= GENHD_FL_EXT_DEVT; | |
f8933667 | 2065 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 2066 | lo->lo_number = i; |
73285082 KC |
2067 | spin_lock_init(&lo->lo_lock); |
2068 | disk->major = LOOP_MAJOR; | |
476a4813 | 2069 | disk->first_minor = i << part_shift; |
73285082 KC |
2070 | disk->fops = &lo_fops; |
2071 | disk->private_data = lo; | |
2072 | disk->queue = lo->lo_queue; | |
2073 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
2074 | add_disk(disk); |
2075 | *l = lo; | |
2076 | return lo->lo_number; | |
73285082 KC |
2077 | |
2078 | out_free_queue: | |
2079 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
2080 | out_cleanup_tags: |
2081 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
2082 | out_free_idr: |
2083 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
2084 | out_free_dev: |
2085 | kfree(lo); | |
2086 | out: | |
34dd82af | 2087 | return err; |
73285082 KC |
2088 | } |
2089 | ||
34dd82af | 2090 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2091 | { |
6cd18e71 | 2092 | del_gendisk(lo->lo_disk); |
0fa8ebdd | 2093 | blk_cleanup_queue(lo->lo_queue); |
b5dd2f60 | 2094 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 2095 | put_disk(lo->lo_disk); |
73285082 KC |
2096 | kfree(lo); |
2097 | } | |
1da177e4 | 2098 | |
770fe30a KS |
2099 | static int find_free_cb(int id, void *ptr, void *data) |
2100 | { | |
2101 | struct loop_device *lo = ptr; | |
2102 | struct loop_device **l = data; | |
2103 | ||
2104 | if (lo->lo_state == Lo_unbound) { | |
2105 | *l = lo; | |
2106 | return 1; | |
2107 | } | |
2108 | return 0; | |
2109 | } | |
2110 | ||
34dd82af | 2111 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
2112 | { |
2113 | struct loop_device *lo; | |
34dd82af | 2114 | int ret = -ENODEV; |
a47653fc | 2115 | |
770fe30a KS |
2116 | if (i < 0) { |
2117 | int err; | |
2118 | ||
2119 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
2120 | if (err == 1) { | |
2121 | *l = lo; | |
2122 | ret = lo->lo_number; | |
2123 | } | |
2124 | goto out; | |
a47653fc KC |
2125 | } |
2126 | ||
770fe30a | 2127 | /* lookup and return a specific i */ |
34dd82af | 2128 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 2129 | if (lo) { |
34dd82af KS |
2130 | *l = lo; |
2131 | ret = lo->lo_number; | |
a47653fc | 2132 | } |
770fe30a | 2133 | out: |
34dd82af | 2134 | return ret; |
a47653fc KC |
2135 | } |
2136 | ||
73285082 KC |
2137 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
2138 | { | |
705962cc | 2139 | struct loop_device *lo; |
07002e99 | 2140 | struct kobject *kobj; |
34dd82af | 2141 | int err; |
73285082 | 2142 | |
0a42e99b | 2143 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2144 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); |
2145 | if (err < 0) | |
2146 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
2147 | if (err < 0) | |
a207f593 | 2148 | kobj = NULL; |
34dd82af | 2149 | else |
3079c22e | 2150 | kobj = get_disk_and_module(lo->lo_disk); |
0a42e99b | 2151 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2152 | |
2153 | *part = 0; | |
07002e99 | 2154 | return kobj; |
73285082 KC |
2155 | } |
2156 | ||
770fe30a KS |
2157 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2158 | unsigned long parm) | |
2159 | { | |
2160 | struct loop_device *lo; | |
0a42e99b | 2161 | int ret; |
770fe30a | 2162 | |
0a42e99b JK |
2163 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2164 | if (ret) | |
2165 | return ret; | |
2166 | ||
2167 | ret = -ENOSYS; | |
770fe30a KS |
2168 | switch (cmd) { |
2169 | case LOOP_CTL_ADD: | |
2170 | ret = loop_lookup(&lo, parm); | |
2171 | if (ret >= 0) { | |
2172 | ret = -EEXIST; | |
2173 | break; | |
2174 | } | |
2175 | ret = loop_add(&lo, parm); | |
2176 | break; | |
2177 | case LOOP_CTL_REMOVE: | |
2178 | ret = loop_lookup(&lo, parm); | |
2179 | if (ret < 0) | |
2180 | break; | |
770fe30a KS |
2181 | if (lo->lo_state != Lo_unbound) { |
2182 | ret = -EBUSY; | |
770fe30a KS |
2183 | break; |
2184 | } | |
f8933667 | 2185 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a | 2186 | ret = -EBUSY; |
770fe30a KS |
2187 | break; |
2188 | } | |
2189 | lo->lo_disk->private_data = NULL; | |
770fe30a KS |
2190 | idr_remove(&loop_index_idr, lo->lo_number); |
2191 | loop_remove(lo); | |
2192 | break; | |
2193 | case LOOP_CTL_GET_FREE: | |
2194 | ret = loop_lookup(&lo, -1); | |
2195 | if (ret >= 0) | |
2196 | break; | |
2197 | ret = loop_add(&lo, -1); | |
2198 | } | |
0a42e99b | 2199 | mutex_unlock(&loop_ctl_mutex); |
770fe30a KS |
2200 | |
2201 | return ret; | |
2202 | } | |
2203 | ||
2204 | static const struct file_operations loop_ctl_fops = { | |
2205 | .open = nonseekable_open, | |
2206 | .unlocked_ioctl = loop_control_ioctl, | |
2207 | .compat_ioctl = loop_control_ioctl, | |
2208 | .owner = THIS_MODULE, | |
2209 | .llseek = noop_llseek, | |
2210 | }; | |
2211 | ||
2212 | static struct miscdevice loop_misc = { | |
2213 | .minor = LOOP_CTRL_MINOR, | |
2214 | .name = "loop-control", | |
2215 | .fops = &loop_ctl_fops, | |
2216 | }; | |
2217 | ||
2218 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2219 | MODULE_ALIAS("devname:loop-control"); | |
2220 | ||
73285082 KC |
2221 | static int __init loop_init(void) |
2222 | { | |
a47653fc KC |
2223 | int i, nr; |
2224 | unsigned long range; | |
34dd82af | 2225 | struct loop_device *lo; |
770fe30a | 2226 | int err; |
a47653fc | 2227 | |
476a4813 | 2228 | part_shift = 0; |
ac04fee0 | 2229 | if (max_part > 0) { |
476a4813 LV |
2230 | part_shift = fls(max_part); |
2231 | ||
ac04fee0 NK |
2232 | /* |
2233 | * Adjust max_part according to part_shift as it is exported | |
2234 | * to user space so that user can decide correct minor number | |
2235 | * if [s]he want to create more devices. | |
2236 | * | |
2237 | * Note that -1 is required because partition 0 is reserved | |
2238 | * for the whole disk. | |
2239 | */ | |
2240 | max_part = (1UL << part_shift) - 1; | |
2241 | } | |
2242 | ||
b1a66504 GC |
2243 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2244 | err = -EINVAL; | |
a8c1d064 | 2245 | goto err_out; |
b1a66504 | 2246 | } |
78f4bb36 | 2247 | |
b1a66504 GC |
2248 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2249 | err = -EINVAL; | |
a8c1d064 | 2250 | goto err_out; |
b1a66504 | 2251 | } |
1da177e4 | 2252 | |
d134b00b KS |
2253 | /* |
2254 | * If max_loop is specified, create that many devices upfront. | |
2255 | * This also becomes a hard limit. If max_loop is not specified, | |
2256 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2257 | * init time. Loop devices can be requested on-demand with the | |
2258 | * /dev/loop-control interface, or be instantiated by accessing | |
2259 | * a 'dead' device node. | |
2260 | */ | |
73285082 | 2261 | if (max_loop) { |
a47653fc | 2262 | nr = max_loop; |
a1c15c59 | 2263 | range = max_loop << part_shift; |
a47653fc | 2264 | } else { |
d134b00b | 2265 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2266 | range = 1UL << MINORBITS; |
a47653fc KC |
2267 | } |
2268 | ||
a8c1d064 AV |
2269 | err = misc_register(&loop_misc); |
2270 | if (err < 0) | |
2271 | goto err_out; | |
2272 | ||
2273 | ||
b1a66504 GC |
2274 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2275 | err = -EIO; | |
2276 | goto misc_out; | |
2277 | } | |
1da177e4 | 2278 | |
a47653fc KC |
2279 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2280 | THIS_MODULE, loop_probe, NULL, NULL); | |
2281 | ||
d134b00b | 2282 | /* pre-create number of devices given by config or max_loop */ |
0a42e99b | 2283 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2284 | for (i = 0; i < nr; i++) |
2285 | loop_add(&lo, i); | |
0a42e99b | 2286 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 2287 | |
73285082 | 2288 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2289 | return 0; |
b1a66504 GC |
2290 | |
2291 | misc_out: | |
2292 | misc_deregister(&loop_misc); | |
a8c1d064 | 2293 | err_out: |
b1a66504 | 2294 | return err; |
34dd82af | 2295 | } |
a47653fc | 2296 | |
34dd82af KS |
2297 | static int loop_exit_cb(int id, void *ptr, void *data) |
2298 | { | |
2299 | struct loop_device *lo = ptr; | |
a47653fc | 2300 | |
34dd82af KS |
2301 | loop_remove(lo); |
2302 | return 0; | |
1da177e4 LT |
2303 | } |
2304 | ||
73285082 | 2305 | static void __exit loop_exit(void) |
1da177e4 | 2306 | { |
a47653fc | 2307 | unsigned long range; |
1da177e4 | 2308 | |
a1c15c59 | 2309 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2310 | |
34dd82af | 2311 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2312 | idr_destroy(&loop_index_idr); |
73285082 | 2313 | |
a47653fc | 2314 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2315 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2316 | |
2317 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2318 | } |
2319 | ||
2320 | module_init(loop_init); | |
2321 | module_exit(loop_exit); | |
2322 | ||
2323 | #ifndef MODULE | |
2324 | static int __init max_loop_setup(char *str) | |
2325 | { | |
2326 | max_loop = simple_strtol(str, NULL, 0); | |
2327 | return 1; | |
2328 | } | |
2329 | ||
2330 | __setup("max_loop=", max_loop_setup); | |
2331 | #endif |