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 | |
aa563d7b | 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) { |
aa563d7b | 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 | |
aa563d7b | 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 | ||
cf655d95 ML |
420 | static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) |
421 | { | |
422 | /* | |
423 | * We use punch hole to reclaim the free space used by the | |
424 | * image a.k.a. discard. However we do not support discard if | |
425 | * encryption is enabled, because it may give an attacker | |
426 | * useful information. | |
427 | */ | |
428 | struct file *file = lo->lo_backing_file; | |
429 | int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; | |
430 | int ret; | |
431 | ||
432 | if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { | |
433 | ret = -EOPNOTSUPP; | |
434 | goto out; | |
435 | } | |
436 | ||
437 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
438 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
439 | ret = -EIO; | |
440 | out: | |
441 | return ret; | |
442 | } | |
443 | ||
444 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
445 | { | |
446 | struct file *file = lo->lo_backing_file; | |
447 | int ret = vfs_fsync(file, 0); | |
448 | if (unlikely(ret && ret != -EINVAL)) | |
449 | ret = -EIO; | |
450 | ||
451 | return ret; | |
452 | } | |
453 | ||
fe2cb290 | 454 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 455 | { |
fe2cb290 | 456 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 457 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 458 | |
f9de14bc JA |
459 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
460 | req_op(rq) != REQ_OP_READ) { | |
461 | if (cmd->ret < 0) | |
462 | ret = BLK_STS_IOERR; | |
463 | goto end_io; | |
bc07c10a | 464 | } |
fe2cb290 | 465 | |
f9de14bc JA |
466 | /* |
467 | * Short READ - if we got some data, advance our request and | |
468 | * retry it. If we got no data, end the rest with EIO. | |
469 | */ | |
470 | if (cmd->ret) { | |
471 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
472 | cmd->ret = 0; | |
473 | blk_mq_requeue_request(rq, true); | |
474 | } else { | |
475 | if (cmd->use_aio) { | |
476 | struct bio *bio = rq->bio; | |
477 | ||
478 | while (bio) { | |
479 | zero_fill_bio(bio); | |
480 | bio = bio->bi_next; | |
481 | } | |
482 | } | |
483 | ret = BLK_STS_IOERR; | |
484 | end_io: | |
485 | blk_mq_end_request(rq, ret); | |
486 | } | |
bc07c10a ML |
487 | } |
488 | ||
92d77332 SL |
489 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
490 | { | |
1894e916 JA |
491 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
492 | ||
92d77332 SL |
493 | if (!atomic_dec_and_test(&cmd->ref)) |
494 | return; | |
495 | kfree(cmd->bvec); | |
496 | cmd->bvec = NULL; | |
1894e916 | 497 | blk_mq_complete_request(rq); |
92d77332 SL |
498 | } |
499 | ||
bc07c10a ML |
500 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) |
501 | { | |
502 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 503 | |
d4478e92 SL |
504 | if (cmd->css) |
505 | css_put(cmd->css); | |
fe2cb290 | 506 | cmd->ret = ret; |
92d77332 | 507 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
508 | } |
509 | ||
510 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
511 | loff_t pos, bool rw) | |
512 | { | |
513 | struct iov_iter iter; | |
86af5952 | 514 | struct req_iterator rq_iter; |
bc07c10a | 515 | struct bio_vec *bvec; |
1894e916 | 516 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 517 | struct bio *bio = rq->bio; |
bc07c10a | 518 | struct file *file = lo->lo_backing_file; |
86af5952 | 519 | struct bio_vec tmp; |
40326d8a | 520 | unsigned int offset; |
86af5952 | 521 | int nr_bvec = 0; |
bc07c10a ML |
522 | int ret; |
523 | ||
86af5952 ML |
524 | rq_for_each_bvec(tmp, rq, rq_iter) |
525 | nr_bvec++; | |
526 | ||
40326d8a | 527 | if (rq->bio != rq->biotail) { |
40326d8a | 528 | |
86af5952 | 529 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 530 | GFP_NOIO); |
40326d8a SL |
531 | if (!bvec) |
532 | return -EIO; | |
533 | cmd->bvec = bvec; | |
534 | ||
535 | /* | |
536 | * The bios of the request may be started from the middle of | |
537 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 538 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
539 | * API will take care of all details for us. |
540 | */ | |
86af5952 | 541 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
542 | *bvec = tmp; |
543 | bvec++; | |
544 | } | |
545 | bvec = cmd->bvec; | |
546 | offset = 0; | |
547 | } else { | |
548 | /* | |
549 | * Same here, this bio may be started from the middle of the | |
550 | * 'bvec' because of bio splitting, so offset from the bvec | |
551 | * must be passed to iov iterator | |
552 | */ | |
553 | offset = bio->bi_iter.bi_bvec_done; | |
554 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 555 | } |
92d77332 | 556 | atomic_set(&cmd->ref, 2); |
bc07c10a | 557 | |
86af5952 | 558 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 559 | iter.iov_offset = offset; |
bc07c10a ML |
560 | |
561 | cmd->iocb.ki_pos = pos; | |
562 | cmd->iocb.ki_filp = file; | |
563 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
564 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 565 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
d4478e92 SL |
566 | if (cmd->css) |
567 | kthread_associate_blkcg(cmd->css); | |
bc07c10a ML |
568 | |
569 | if (rw == WRITE) | |
bb7462b6 | 570 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 571 | else |
bb7462b6 | 572 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 573 | |
92d77332 | 574 | lo_rw_aio_do_completion(cmd); |
d4478e92 | 575 | kthread_associate_blkcg(NULL); |
92d77332 | 576 | |
bc07c10a ML |
577 | if (ret != -EIOCBQUEUED) |
578 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
579 | return 0; | |
580 | } | |
581 | ||
c1c87c2b | 582 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
583 | { |
584 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 585 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
586 | |
587 | /* | |
588 | * lo_write_simple and lo_read_simple should have been covered | |
589 | * by io submit style function like lo_rw_aio(), one blocker | |
590 | * is that lo_read_simple() need to call flush_dcache_page after | |
591 | * the page is written from kernel, and it isn't easy to handle | |
592 | * this in io submit style function which submits all segments | |
593 | * of the req at one time. And direct read IO doesn't need to | |
594 | * run flush_dcache_page(). | |
595 | */ | |
c1c87c2b CH |
596 | switch (req_op(rq)) { |
597 | case REQ_OP_FLUSH: | |
598 | return lo_req_flush(lo, rq); | |
599 | case REQ_OP_DISCARD: | |
19372e27 | 600 | case REQ_OP_WRITE_ZEROES: |
c1c87c2b CH |
601 | return lo_discard(lo, rq, pos); |
602 | case REQ_OP_WRITE: | |
603 | if (lo->transfer) | |
604 | return lo_write_transfer(lo, rq, pos); | |
605 | else if (cmd->use_aio) | |
606 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 607 | else |
c1c87c2b CH |
608 | return lo_write_simple(lo, rq, pos); |
609 | case REQ_OP_READ: | |
aa4d8616 | 610 | if (lo->transfer) |
c1c87c2b CH |
611 | return lo_read_transfer(lo, rq, pos); |
612 | else if (cmd->use_aio) | |
613 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 614 | else |
c1c87c2b CH |
615 | return lo_read_simple(lo, rq, pos); |
616 | default: | |
617 | WARN_ON_ONCE(1); | |
618 | return -EIO; | |
aa4d8616 | 619 | } |
1da177e4 LT |
620 | } |
621 | ||
2e5ab5f3 ML |
622 | static inline void loop_update_dio(struct loop_device *lo) |
623 | { | |
624 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
625 | lo->use_dio); | |
626 | } | |
627 | ||
06f0e9e6 ML |
628 | static void loop_reread_partitions(struct loop_device *lo, |
629 | struct block_device *bdev) | |
630 | { | |
631 | int rc; | |
632 | ||
d57f3374 | 633 | rc = blkdev_reread_part(bdev); |
06f0e9e6 ML |
634 | if (rc) |
635 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
636 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
637 | } | |
638 | ||
d2ac838e TT |
639 | static inline int is_loop_device(struct file *file) |
640 | { | |
641 | struct inode *i = file->f_mapping->host; | |
642 | ||
643 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
644 | } | |
645 | ||
646 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
647 | { | |
648 | struct inode *inode = file->f_mapping->host; | |
649 | struct file *f = file; | |
650 | ||
651 | /* Avoid recursion */ | |
652 | while (is_loop_device(f)) { | |
653 | struct loop_device *l; | |
654 | ||
655 | if (f->f_mapping->host->i_bdev == bdev) | |
656 | return -EBADF; | |
657 | ||
658 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
f7c8a412 | 659 | if (l->lo_state != Lo_bound) { |
d2ac838e TT |
660 | return -EINVAL; |
661 | } | |
662 | f = l->lo_backing_file; | |
663 | } | |
664 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
665 | return -EINVAL; | |
666 | return 0; | |
667 | } | |
668 | ||
1da177e4 LT |
669 | /* |
670 | * loop_change_fd switched the backing store of a loopback device to | |
671 | * a new file. This is useful for operating system installers to free up | |
672 | * the original file and in High Availability environments to switch to | |
673 | * an alternative location for the content in case of server meltdown. | |
674 | * This can only work if the loop device is used read-only, and if the | |
675 | * new backing store is the same size and type as the old backing store. | |
676 | */ | |
bb214884 AV |
677 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
678 | unsigned int arg) | |
1da177e4 | 679 | { |
1dded9ac | 680 | struct file *file = NULL, *old_file; |
1da177e4 | 681 | int error; |
85b0a54a | 682 | bool partscan; |
1da177e4 | 683 | |
c28445fa | 684 | error = mutex_lock_killable(&loop_ctl_mutex); |
c3710770 JK |
685 | if (error) |
686 | return error; | |
1da177e4 LT |
687 | error = -ENXIO; |
688 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 689 | goto out_err; |
1da177e4 LT |
690 | |
691 | /* the loop device has to be read-only */ | |
692 | error = -EINVAL; | |
693 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 694 | goto out_err; |
1da177e4 LT |
695 | |
696 | error = -EBADF; | |
697 | file = fget(arg); | |
698 | if (!file) | |
1dded9ac | 699 | goto out_err; |
1da177e4 | 700 | |
d2ac838e TT |
701 | error = loop_validate_file(file, bdev); |
702 | if (error) | |
1dded9ac | 703 | goto out_err; |
d2ac838e | 704 | |
1da177e4 LT |
705 | old_file = lo->lo_backing_file; |
706 | ||
707 | error = -EINVAL; | |
708 | ||
1da177e4 LT |
709 | /* size of the new backing store needs to be the same */ |
710 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 711 | goto out_err; |
1da177e4 LT |
712 | |
713 | /* and ... switch */ | |
43cade80 OS |
714 | blk_mq_freeze_queue(lo->lo_queue); |
715 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
716 | lo->lo_backing_file = file; | |
717 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
718 | mapping_set_gfp_mask(file->f_mapping, | |
719 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
720 | loop_update_dio(lo); | |
721 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 722 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c3710770 | 723 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
724 | /* |
725 | * We must drop file reference outside of loop_ctl_mutex as dropping | |
726 | * the file ref can take bd_mutex which creates circular locking | |
727 | * dependency. | |
728 | */ | |
729 | fput(old_file); | |
85b0a54a JK |
730 | if (partscan) |
731 | loop_reread_partitions(lo, bdev); | |
1da177e4 LT |
732 | return 0; |
733 | ||
1dded9ac | 734 | out_err: |
c3710770 | 735 | mutex_unlock(&loop_ctl_mutex); |
1dded9ac JK |
736 | if (file) |
737 | fput(file); | |
1da177e4 LT |
738 | return error; |
739 | } | |
740 | ||
ee862730 MB |
741 | /* loop sysfs attributes */ |
742 | ||
743 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
744 | ssize_t (*callback)(struct loop_device *, char *)) | |
745 | { | |
34dd82af KS |
746 | struct gendisk *disk = dev_to_disk(dev); |
747 | struct loop_device *lo = disk->private_data; | |
ee862730 | 748 | |
34dd82af | 749 | return callback(lo, page); |
ee862730 MB |
750 | } |
751 | ||
752 | #define LOOP_ATTR_RO(_name) \ | |
753 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
754 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
755 | struct device_attribute *attr, char *b) \ | |
756 | { \ | |
757 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
758 | } \ | |
759 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 760 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
761 | |
762 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
763 | { | |
764 | ssize_t ret; | |
765 | char *p = NULL; | |
766 | ||
05eb0f25 | 767 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 768 | if (lo->lo_backing_file) |
9bf39ab2 | 769 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 770 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
771 | |
772 | if (IS_ERR_OR_NULL(p)) | |
773 | ret = PTR_ERR(p); | |
774 | else { | |
775 | ret = strlen(p); | |
776 | memmove(buf, p, ret); | |
777 | buf[ret++] = '\n'; | |
778 | buf[ret] = 0; | |
779 | } | |
780 | ||
781 | return ret; | |
782 | } | |
783 | ||
784 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
785 | { | |
786 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
787 | } | |
788 | ||
789 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
790 | { | |
791 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
792 | } | |
793 | ||
794 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
795 | { | |
796 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
797 | ||
798 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
799 | } | |
800 | ||
e03c8dd1 KS |
801 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
802 | { | |
803 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
804 | ||
805 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
806 | } | |
807 | ||
2e5ab5f3 ML |
808 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
809 | { | |
810 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
811 | ||
812 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
813 | } | |
814 | ||
ee862730 MB |
815 | LOOP_ATTR_RO(backing_file); |
816 | LOOP_ATTR_RO(offset); | |
817 | LOOP_ATTR_RO(sizelimit); | |
818 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 819 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 820 | LOOP_ATTR_RO(dio); |
ee862730 MB |
821 | |
822 | static struct attribute *loop_attrs[] = { | |
823 | &loop_attr_backing_file.attr, | |
824 | &loop_attr_offset.attr, | |
825 | &loop_attr_sizelimit.attr, | |
826 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 827 | &loop_attr_partscan.attr, |
2e5ab5f3 | 828 | &loop_attr_dio.attr, |
ee862730 MB |
829 | NULL, |
830 | }; | |
831 | ||
832 | static struct attribute_group loop_attribute_group = { | |
833 | .name = "loop", | |
834 | .attrs= loop_attrs, | |
835 | }; | |
836 | ||
d3349b6b | 837 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 838 | { |
d3349b6b TH |
839 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
840 | &loop_attribute_group); | |
ee862730 MB |
841 | } |
842 | ||
843 | static void loop_sysfs_exit(struct loop_device *lo) | |
844 | { | |
d3349b6b TH |
845 | if (lo->sysfs_inited) |
846 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
847 | &loop_attribute_group); | |
ee862730 MB |
848 | } |
849 | ||
dfaa2ef6 LC |
850 | static void loop_config_discard(struct loop_device *lo) |
851 | { | |
852 | struct file *file = lo->lo_backing_file; | |
853 | struct inode *inode = file->f_mapping->host; | |
854 | struct request_queue *q = lo->lo_queue; | |
855 | ||
856 | /* | |
857 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 858 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
859 | * encryption is enabled, because it may give an attacker |
860 | * useful information. | |
861 | */ | |
862 | if ((!file->f_op->fallocate) || | |
863 | lo->lo_encrypt_key_size) { | |
864 | q->limits.discard_granularity = 0; | |
865 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 866 | blk_queue_max_discard_sectors(q, 0); |
19372e27 | 867 | blk_queue_max_write_zeroes_sectors(q, 0); |
8b904b5b | 868 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); |
dfaa2ef6 LC |
869 | return; |
870 | } | |
871 | ||
872 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 873 | q->limits.discard_alignment = 0; |
f2c6df7d | 874 | |
1e6ec9ea OS |
875 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
876 | blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); | |
8b904b5b | 877 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); |
dfaa2ef6 LC |
878 | } |
879 | ||
e03a3d7a ML |
880 | static void loop_unprepare_queue(struct loop_device *lo) |
881 | { | |
3989144f | 882 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
883 | kthread_stop(lo->worker_task); |
884 | } | |
885 | ||
b2ee7d46 N |
886 | static int loop_kthread_worker_fn(void *worker_ptr) |
887 | { | |
888 | current->flags |= PF_LESS_THROTTLE; | |
889 | return kthread_worker_fn(worker_ptr); | |
890 | } | |
891 | ||
e03a3d7a ML |
892 | static int loop_prepare_queue(struct loop_device *lo) |
893 | { | |
3989144f | 894 | kthread_init_worker(&lo->worker); |
b2ee7d46 | 895 | lo->worker_task = kthread_run(loop_kthread_worker_fn, |
e03a3d7a ML |
896 | &lo->worker, "loop%d", lo->lo_number); |
897 | if (IS_ERR(lo->worker_task)) | |
898 | return -ENOMEM; | |
899 | set_user_nice(lo->worker_task, MIN_NICE); | |
900 | return 0; | |
901 | } | |
902 | ||
bb214884 | 903 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
904 | struct block_device *bdev, unsigned int arg) |
905 | { | |
d2ac838e | 906 | struct file *file; |
1da177e4 LT |
907 | struct inode *inode; |
908 | struct address_space *mapping; | |
1da177e4 LT |
909 | int lo_flags = 0; |
910 | int error; | |
911 | loff_t size; | |
85b0a54a | 912 | bool partscan; |
1da177e4 LT |
913 | |
914 | /* This is safe, since we have a reference from open(). */ | |
915 | __module_get(THIS_MODULE); | |
916 | ||
917 | error = -EBADF; | |
918 | file = fget(arg); | |
919 | if (!file) | |
920 | goto out; | |
921 | ||
c28445fa | 922 | error = mutex_lock_killable(&loop_ctl_mutex); |
757ecf40 JK |
923 | if (error) |
924 | goto out_putf; | |
925 | ||
1da177e4 LT |
926 | error = -EBUSY; |
927 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 928 | goto out_unlock; |
1da177e4 | 929 | |
d2ac838e TT |
930 | error = loop_validate_file(file, bdev); |
931 | if (error) | |
757ecf40 | 932 | goto out_unlock; |
1da177e4 LT |
933 | |
934 | mapping = file->f_mapping; | |
935 | inode = mapping->host; | |
936 | ||
456be148 | 937 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 938 | !file->f_op->write_iter) |
456be148 | 939 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 940 | |
456be148 | 941 | error = -EFBIG; |
1da177e4 | 942 | size = get_loop_size(lo, file); |
456be148 | 943 | if ((loff_t)(sector_t)size != size) |
757ecf40 | 944 | goto out_unlock; |
e03a3d7a ML |
945 | error = loop_prepare_queue(lo); |
946 | if (error) | |
757ecf40 | 947 | goto out_unlock; |
1da177e4 | 948 | |
456be148 | 949 | error = 0; |
1da177e4 LT |
950 | |
951 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
952 | ||
2e5ab5f3 | 953 | lo->use_dio = false; |
1da177e4 LT |
954 | lo->lo_device = bdev; |
955 | lo->lo_flags = lo_flags; | |
956 | lo->lo_backing_file = file; | |
aa4d8616 | 957 | lo->transfer = NULL; |
1da177e4 LT |
958 | lo->ioctl = NULL; |
959 | lo->lo_sizelimit = 0; | |
960 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
961 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
962 | ||
68db1961 | 963 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 964 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 965 | |
2e5ab5f3 | 966 | loop_update_dio(lo); |
73285082 | 967 | set_capacity(lo->lo_disk, size); |
1da177e4 | 968 | bd_set_size(bdev, size << 9); |
ee862730 | 969 | loop_sysfs_init(lo); |
c3473c63 DZ |
970 | /* let user-space know about the new size */ |
971 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 972 | |
8a0740c4 OS |
973 | set_blocksize(bdev, S_ISBLK(inode->i_mode) ? |
974 | block_size(inode->i_bdev) : PAGE_SIZE); | |
1da177e4 | 975 | |
6c997918 | 976 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
977 | if (part_shift) |
978 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 979 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
c1681bf8 AP |
980 | |
981 | /* Grab the block_device to prevent its destruction after we | |
a2505b79 | 982 | * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). |
c1681bf8 AP |
983 | */ |
984 | bdgrab(bdev); | |
757ecf40 | 985 | mutex_unlock(&loop_ctl_mutex); |
85b0a54a JK |
986 | if (partscan) |
987 | loop_reread_partitions(lo, bdev); | |
1da177e4 LT |
988 | return 0; |
989 | ||
757ecf40 JK |
990 | out_unlock: |
991 | mutex_unlock(&loop_ctl_mutex); | |
992 | out_putf: | |
1da177e4 | 993 | fput(file); |
757ecf40 | 994 | out: |
1da177e4 LT |
995 | /* This is safe: open() is still holding a reference. */ |
996 | module_put(THIS_MODULE); | |
997 | return error; | |
998 | } | |
999 | ||
1000 | static int | |
1001 | loop_release_xfer(struct loop_device *lo) | |
1002 | { | |
1003 | int err = 0; | |
1004 | struct loop_func_table *xfer = lo->lo_encryption; | |
1005 | ||
1006 | if (xfer) { | |
1007 | if (xfer->release) | |
1008 | err = xfer->release(lo); | |
1009 | lo->transfer = NULL; | |
1010 | lo->lo_encryption = NULL; | |
1011 | module_put(xfer->owner); | |
1012 | } | |
1013 | return err; | |
1014 | } | |
1015 | ||
1016 | static int | |
1017 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1018 | const struct loop_info64 *i) | |
1019 | { | |
1020 | int err = 0; | |
1021 | ||
1022 | if (xfer) { | |
1023 | struct module *owner = xfer->owner; | |
1024 | ||
1025 | if (!try_module_get(owner)) | |
1026 | return -EINVAL; | |
1027 | if (xfer->init) | |
1028 | err = xfer->init(lo, i); | |
1029 | if (err) | |
1030 | module_put(owner); | |
1031 | else | |
1032 | lo->lo_encryption = xfer; | |
1033 | } | |
1034 | return err; | |
1035 | } | |
1036 | ||
0da03cab | 1037 | static int __loop_clr_fd(struct loop_device *lo, bool release) |
1da177e4 | 1038 | { |
7ccd0791 | 1039 | struct file *filp = NULL; |
b4e3ca1a | 1040 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1041 | struct block_device *bdev = lo->lo_device; |
7ccd0791 | 1042 | int err = 0; |
0da03cab JK |
1043 | bool partscan = false; |
1044 | int lo_number; | |
1da177e4 | 1045 | |
7ccd0791 JK |
1046 | mutex_lock(&loop_ctl_mutex); |
1047 | if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { | |
1048 | err = -ENXIO; | |
1049 | goto out_unlock; | |
1050 | } | |
1da177e4 | 1051 | |
7ccd0791 JK |
1052 | filp = lo->lo_backing_file; |
1053 | if (filp == NULL) { | |
1054 | err = -EINVAL; | |
1055 | goto out_unlock; | |
1056 | } | |
1da177e4 | 1057 | |
f8933667 ML |
1058 | /* freeze request queue during the transition */ |
1059 | blk_mq_freeze_queue(lo->lo_queue); | |
1060 | ||
1da177e4 | 1061 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 1062 | lo->lo_backing_file = NULL; |
05eb0f25 | 1063 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1064 | |
1065 | loop_release_xfer(lo); | |
1066 | lo->transfer = NULL; | |
1067 | lo->ioctl = NULL; | |
1068 | lo->lo_device = NULL; | |
1069 | lo->lo_encryption = NULL; | |
1070 | lo->lo_offset = 0; | |
1071 | lo->lo_sizelimit = 0; | |
1072 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1073 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1074 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1075 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
89e4fdec | 1076 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1077 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1078 | blk_queue_io_min(lo->lo_queue, 512); | |
c1681bf8 AP |
1079 | if (bdev) { |
1080 | bdput(bdev); | |
bb214884 | 1081 | invalidate_bdev(bdev); |
eedffa28 | 1082 | bdev->bd_inode->i_mapping->wb_err = 0; |
c1681bf8 | 1083 | } |
73285082 | 1084 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1085 | loop_sysfs_exit(lo); |
c3473c63 | 1086 | if (bdev) { |
bb214884 | 1087 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1088 | /* let user-space know about this change */ |
1089 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1090 | } | |
1da177e4 | 1091 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1da177e4 LT |
1092 | /* This is safe: open() is still holding a reference. */ |
1093 | module_put(THIS_MODULE); | |
f8933667 ML |
1094 | blk_mq_unfreeze_queue(lo->lo_queue); |
1095 | ||
0da03cab JK |
1096 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; |
1097 | lo_number = lo->lo_number; | |
0da03cab JK |
1098 | loop_unprepare_queue(lo); |
1099 | out_unlock: | |
1100 | mutex_unlock(&loop_ctl_mutex); | |
1101 | if (partscan) { | |
d57f3374 JK |
1102 | /* |
1103 | * bd_mutex has been held already in release path, so don't | |
1104 | * acquire it if this function is called in such case. | |
1105 | * | |
1106 | * If the reread partition isn't from release path, lo_refcnt | |
1107 | * must be at least one and it can only become zero when the | |
1108 | * current holder is released. | |
1109 | */ | |
0da03cab | 1110 | if (release) |
d57f3374 JK |
1111 | err = __blkdev_reread_part(bdev); |
1112 | else | |
1113 | err = blkdev_reread_part(bdev); | |
40853d6f DZ |
1114 | if (err) |
1115 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
1116 | __func__, lo_number, err); | |
d57f3374 JK |
1117 | /* Device is gone, no point in returning error */ |
1118 | err = 0; | |
1119 | } | |
758a58d0 DZ |
1120 | |
1121 | /* | |
1122 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1123 | * finished. | |
1124 | * | |
1125 | * There cannot be anybody else entering __loop_clr_fd() as | |
1126 | * lo->lo_backing_file is already cleared and Lo_rundown state | |
1127 | * protects us from all the other places trying to change the 'lo' | |
1128 | * device. | |
1129 | */ | |
1130 | mutex_lock(&loop_ctl_mutex); | |
1131 | lo->lo_flags = 0; | |
1132 | if (!part_shift) | |
1133 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1134 | lo->lo_state = Lo_unbound; | |
1135 | mutex_unlock(&loop_ctl_mutex); | |
1136 | ||
f028f3b2 | 1137 | /* |
310ca162 TH |
1138 | * Need not hold loop_ctl_mutex to fput backing file. |
1139 | * Calling fput holding loop_ctl_mutex triggers a circular | |
f028f3b2 | 1140 | * lock dependency possibility warning as fput can take |
310ca162 | 1141 | * bd_mutex which is usually taken before loop_ctl_mutex. |
f028f3b2 | 1142 | */ |
7ccd0791 JK |
1143 | if (filp) |
1144 | fput(filp); | |
1145 | return err; | |
1da177e4 LT |
1146 | } |
1147 | ||
a2505b79 JK |
1148 | static int loop_clr_fd(struct loop_device *lo) |
1149 | { | |
7ccd0791 JK |
1150 | int err; |
1151 | ||
c28445fa | 1152 | err = mutex_lock_killable(&loop_ctl_mutex); |
7ccd0791 JK |
1153 | if (err) |
1154 | return err; | |
1155 | if (lo->lo_state != Lo_bound) { | |
1156 | mutex_unlock(&loop_ctl_mutex); | |
a2505b79 | 1157 | return -ENXIO; |
7ccd0791 | 1158 | } |
a2505b79 JK |
1159 | /* |
1160 | * If we've explicitly asked to tear down the loop device, | |
1161 | * and it has an elevated reference count, set it for auto-teardown when | |
1162 | * the last reference goes away. This stops $!~#$@ udev from | |
1163 | * preventing teardown because it decided that it needs to run blkid on | |
1164 | * the loopback device whenever they appear. xfstests is notorious for | |
1165 | * failing tests because blkid via udev races with a losetup | |
1166 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1167 | * command to fail with EBUSY. | |
1168 | */ | |
1169 | if (atomic_read(&lo->lo_refcnt) > 1) { | |
1170 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
1171 | mutex_unlock(&loop_ctl_mutex); | |
1172 | return 0; | |
1173 | } | |
1174 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1175 | mutex_unlock(&loop_ctl_mutex); |
a2505b79 | 1176 | |
0da03cab | 1177 | return __loop_clr_fd(lo, false); |
a2505b79 JK |
1178 | } |
1179 | ||
1da177e4 LT |
1180 | static int |
1181 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1182 | { | |
1183 | int err; | |
1184 | struct loop_func_table *xfer; | |
e4849737 | 1185 | kuid_t uid = current_uid(); |
85b0a54a JK |
1186 | struct block_device *bdev; |
1187 | bool partscan = false; | |
1da177e4 | 1188 | |
c28445fa | 1189 | err = mutex_lock_killable(&loop_ctl_mutex); |
550df5fd JK |
1190 | if (err) |
1191 | return err; | |
b0fafa81 | 1192 | if (lo->lo_encrypt_key_size && |
e4849737 | 1193 | !uid_eq(lo->lo_key_owner, uid) && |
550df5fd JK |
1194 | !capable(CAP_SYS_ADMIN)) { |
1195 | err = -EPERM; | |
1196 | goto out_unlock; | |
1197 | } | |
1198 | if (lo->lo_state != Lo_bound) { | |
1199 | err = -ENXIO; | |
1200 | goto out_unlock; | |
1201 | } | |
1202 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { | |
1203 | err = -EINVAL; | |
1204 | goto out_unlock; | |
1205 | } | |
1da177e4 | 1206 | |
5db470e2 JK |
1207 | if (lo->lo_offset != info->lo_offset || |
1208 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1209 | sync_blockdev(lo->lo_device); | |
1210 | kill_bdev(lo->lo_device); | |
1211 | } | |
1212 | ||
ecdd0959 ML |
1213 | /* I/O need to be drained during transfer transition */ |
1214 | blk_mq_freeze_queue(lo->lo_queue); | |
1215 | ||
1da177e4 LT |
1216 | err = loop_release_xfer(lo); |
1217 | if (err) | |
550df5fd | 1218 | goto out_unfreeze; |
1da177e4 LT |
1219 | |
1220 | if (info->lo_encrypt_type) { | |
1221 | unsigned int type = info->lo_encrypt_type; | |
1222 | ||
1e047eaa TH |
1223 | if (type >= MAX_LO_CRYPT) { |
1224 | err = -EINVAL; | |
550df5fd | 1225 | goto out_unfreeze; |
1e047eaa | 1226 | } |
1da177e4 | 1227 | xfer = xfer_funcs[type]; |
1e047eaa TH |
1228 | if (xfer == NULL) { |
1229 | err = -EINVAL; | |
550df5fd | 1230 | goto out_unfreeze; |
1e047eaa | 1231 | } |
1da177e4 LT |
1232 | } else |
1233 | xfer = NULL; | |
1234 | ||
1235 | err = loop_init_xfer(lo, xfer, info); | |
1236 | if (err) | |
550df5fd | 1237 | goto out_unfreeze; |
1da177e4 LT |
1238 | |
1239 | if (lo->lo_offset != info->lo_offset || | |
1e6ec9ea | 1240 | lo->lo_sizelimit != info->lo_sizelimit) { |
5db470e2 JK |
1241 | /* kill_bdev should have truncated all the pages */ |
1242 | if (lo->lo_device->bd_inode->i_mapping->nrpages) { | |
1243 | err = -EAGAIN; | |
1244 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1245 | __func__, lo->lo_number, lo->lo_file_name, | |
1246 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1247 | goto out_unfreeze; | |
1248 | } | |
1e6ec9ea | 1249 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { |
ecdd0959 | 1250 | err = -EFBIG; |
550df5fd | 1251 | goto out_unfreeze; |
ecdd0959 | 1252 | } |
b040ad9c | 1253 | } |
541c742a | 1254 | |
dfaa2ef6 | 1255 | loop_config_discard(lo); |
1da177e4 LT |
1256 | |
1257 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1258 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1259 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1260 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1261 | ||
1262 | if (!xfer) | |
1263 | xfer = &none_funcs; | |
1264 | lo->transfer = xfer->transfer; | |
1265 | lo->ioctl = xfer->ioctl; | |
1266 | ||
96c58655 DW |
1267 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1268 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1269 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1270 | ||
1da177e4 LT |
1271 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1272 | lo->lo_init[0] = info->lo_init[0]; | |
1273 | lo->lo_init[1] = info->lo_init[1]; | |
1274 | if (info->lo_encrypt_key_size) { | |
1275 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1276 | info->lo_encrypt_key_size); | |
b0fafa81 | 1277 | lo->lo_key_owner = uid; |
aa4d8616 | 1278 | } |
1da177e4 | 1279 | |
2e5ab5f3 ML |
1280 | /* update dio if lo_offset or transfer is changed */ |
1281 | __loop_update_dio(lo, lo->use_dio); | |
1282 | ||
550df5fd | 1283 | out_unfreeze: |
ecdd0959 | 1284 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 OS |
1285 | |
1286 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1287 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1288 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1289 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
85b0a54a JK |
1290 | bdev = lo->lo_device; |
1291 | partscan = true; | |
e02898b4 | 1292 | } |
550df5fd JK |
1293 | out_unlock: |
1294 | mutex_unlock(&loop_ctl_mutex); | |
85b0a54a JK |
1295 | if (partscan) |
1296 | loop_reread_partitions(lo, bdev); | |
e02898b4 | 1297 | |
ecdd0959 | 1298 | return err; |
1da177e4 LT |
1299 | } |
1300 | ||
1301 | static int | |
1302 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1303 | { | |
b1ab5fa3 | 1304 | struct path path; |
1da177e4 | 1305 | struct kstat stat; |
2d1d4c1e | 1306 | int ret; |
1da177e4 | 1307 | |
c28445fa | 1308 | ret = mutex_lock_killable(&loop_ctl_mutex); |
4a5ce9ba JK |
1309 | if (ret) |
1310 | return ret; | |
2d1d4c1e | 1311 | if (lo->lo_state != Lo_bound) { |
310ca162 | 1312 | mutex_unlock(&loop_ctl_mutex); |
1da177e4 | 1313 | return -ENXIO; |
2d1d4c1e OS |
1314 | } |
1315 | ||
1da177e4 LT |
1316 | memset(info, 0, sizeof(*info)); |
1317 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1318 | info->lo_offset = lo->lo_offset; |
1319 | info->lo_sizelimit = lo->lo_sizelimit; | |
1320 | info->lo_flags = lo->lo_flags; | |
1321 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1322 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1323 | info->lo_encrypt_type = | |
1324 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1325 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1326 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1327 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1328 | lo->lo_encrypt_key_size); | |
1329 | } | |
2d1d4c1e | 1330 | |
310ca162 | 1331 | /* Drop loop_ctl_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1332 | path = lo->lo_backing_file->f_path; |
1333 | path_get(&path); | |
310ca162 | 1334 | mutex_unlock(&loop_ctl_mutex); |
b1ab5fa3 | 1335 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1336 | if (!ret) { |
1337 | info->lo_device = huge_encode_dev(stat.dev); | |
1338 | info->lo_inode = stat.ino; | |
1339 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1340 | } | |
b1ab5fa3 | 1341 | path_put(&path); |
2d1d4c1e | 1342 | return ret; |
1da177e4 LT |
1343 | } |
1344 | ||
1345 | static void | |
1346 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1347 | { | |
1348 | memset(info64, 0, sizeof(*info64)); | |
1349 | info64->lo_number = info->lo_number; | |
1350 | info64->lo_device = info->lo_device; | |
1351 | info64->lo_inode = info->lo_inode; | |
1352 | info64->lo_rdevice = info->lo_rdevice; | |
1353 | info64->lo_offset = info->lo_offset; | |
1354 | info64->lo_sizelimit = 0; | |
1355 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1356 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1357 | info64->lo_flags = info->lo_flags; | |
1358 | info64->lo_init[0] = info->lo_init[0]; | |
1359 | info64->lo_init[1] = info->lo_init[1]; | |
1360 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1361 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1362 | else | |
1363 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1364 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1365 | } | |
1366 | ||
1367 | static int | |
1368 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1369 | { | |
1370 | memset(info, 0, sizeof(*info)); | |
1371 | info->lo_number = info64->lo_number; | |
1372 | info->lo_device = info64->lo_device; | |
1373 | info->lo_inode = info64->lo_inode; | |
1374 | info->lo_rdevice = info64->lo_rdevice; | |
1375 | info->lo_offset = info64->lo_offset; | |
1376 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1377 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1378 | info->lo_flags = info64->lo_flags; | |
1379 | info->lo_init[0] = info64->lo_init[0]; | |
1380 | info->lo_init[1] = info64->lo_init[1]; | |
1381 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1382 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1383 | else | |
1384 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1385 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1386 | ||
1387 | /* error in case values were truncated */ | |
1388 | if (info->lo_device != info64->lo_device || | |
1389 | info->lo_rdevice != info64->lo_rdevice || | |
1390 | info->lo_inode != info64->lo_inode || | |
1391 | info->lo_offset != info64->lo_offset) | |
1392 | return -EOVERFLOW; | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
1397 | static int | |
1398 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1399 | { | |
1400 | struct loop_info info; | |
1401 | struct loop_info64 info64; | |
1402 | ||
1403 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1404 | return -EFAULT; | |
1405 | loop_info64_from_old(&info, &info64); | |
1406 | return loop_set_status(lo, &info64); | |
1407 | } | |
1408 | ||
1409 | static int | |
1410 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1411 | { | |
1412 | struct loop_info64 info64; | |
1413 | ||
1414 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1415 | return -EFAULT; | |
1416 | return loop_set_status(lo, &info64); | |
1417 | } | |
1418 | ||
1419 | static int | |
1420 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1421 | struct loop_info info; | |
1422 | struct loop_info64 info64; | |
bdac616d | 1423 | int err; |
1da177e4 | 1424 | |
4a5ce9ba | 1425 | if (!arg) |
bdac616d | 1426 | return -EINVAL; |
bdac616d | 1427 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1428 | if (!err) |
1429 | err = loop_info64_to_old(&info64, &info); | |
1430 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1431 | err = -EFAULT; | |
1432 | ||
1433 | return err; | |
1434 | } | |
1435 | ||
1436 | static int | |
1437 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1438 | struct loop_info64 info64; | |
bdac616d | 1439 | int err; |
1da177e4 | 1440 | |
4a5ce9ba | 1441 | if (!arg) |
bdac616d | 1442 | return -EINVAL; |
bdac616d | 1443 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1444 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1445 | err = -EFAULT; | |
1446 | ||
1447 | return err; | |
1448 | } | |
1449 | ||
51001b7d | 1450 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1451 | { |
53d66608 | 1452 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1453 | return -ENXIO; |
53d66608 | 1454 | |
1e6ec9ea | 1455 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1456 | } |
1457 | ||
ab1cb278 ML |
1458 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1459 | { | |
1460 | int error = -ENXIO; | |
1461 | if (lo->lo_state != Lo_bound) | |
1462 | goto out; | |
1463 | ||
1464 | __loop_update_dio(lo, !!arg); | |
1465 | if (lo->use_dio == !!arg) | |
1466 | return 0; | |
1467 | error = -EINVAL; | |
1468 | out: | |
1469 | return error; | |
1470 | } | |
1471 | ||
89e4fdec OS |
1472 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1473 | { | |
5db470e2 JK |
1474 | int err = 0; |
1475 | ||
89e4fdec OS |
1476 | if (lo->lo_state != Lo_bound) |
1477 | return -ENXIO; | |
1478 | ||
1479 | if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) | |
1480 | return -EINVAL; | |
1481 | ||
5db470e2 JK |
1482 | if (lo->lo_queue->limits.logical_block_size != arg) { |
1483 | sync_blockdev(lo->lo_device); | |
1484 | kill_bdev(lo->lo_device); | |
1485 | } | |
1486 | ||
89e4fdec OS |
1487 | blk_mq_freeze_queue(lo->lo_queue); |
1488 | ||
5db470e2 JK |
1489 | /* kill_bdev should have truncated all the pages */ |
1490 | if (lo->lo_queue->limits.logical_block_size != arg && | |
1491 | lo->lo_device->bd_inode->i_mapping->nrpages) { | |
1492 | err = -EAGAIN; | |
1493 | pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", | |
1494 | __func__, lo->lo_number, lo->lo_file_name, | |
1495 | lo->lo_device->bd_inode->i_mapping->nrpages); | |
1496 | goto out_unfreeze; | |
1497 | } | |
1498 | ||
89e4fdec | 1499 | blk_queue_logical_block_size(lo->lo_queue, arg); |
bf093753 OS |
1500 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1501 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec | 1502 | loop_update_dio(lo); |
5db470e2 | 1503 | out_unfreeze: |
89e4fdec OS |
1504 | blk_mq_unfreeze_queue(lo->lo_queue); |
1505 | ||
5db470e2 | 1506 | return err; |
89e4fdec OS |
1507 | } |
1508 | ||
a1316544 JK |
1509 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1510 | unsigned long arg) | |
1da177e4 | 1511 | { |
1da177e4 LT |
1512 | int err; |
1513 | ||
c28445fa | 1514 | err = mutex_lock_killable(&loop_ctl_mutex); |
3148ffbd | 1515 | if (err) |
a1316544 JK |
1516 | return err; |
1517 | switch (cmd) { | |
1518 | case LOOP_SET_CAPACITY: | |
1519 | err = loop_set_capacity(lo); | |
1520 | break; | |
1521 | case LOOP_SET_DIRECT_IO: | |
1522 | err = loop_set_dio(lo, arg); | |
1523 | break; | |
1524 | case LOOP_SET_BLOCK_SIZE: | |
1525 | err = loop_set_block_size(lo, arg); | |
1526 | break; | |
1527 | default: | |
1528 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1529 | } | |
1530 | mutex_unlock(&loop_ctl_mutex); | |
1531 | return err; | |
1532 | } | |
1533 | ||
1534 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, | |
1535 | unsigned int cmd, unsigned long arg) | |
1536 | { | |
1537 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1538 | int err; | |
3148ffbd | 1539 | |
1da177e4 LT |
1540 | switch (cmd) { |
1541 | case LOOP_SET_FD: | |
757ecf40 | 1542 | return loop_set_fd(lo, mode, bdev, arg); |
1da177e4 | 1543 | case LOOP_CHANGE_FD: |
c3710770 | 1544 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1545 | case LOOP_CLR_FD: |
7ccd0791 | 1546 | return loop_clr_fd(lo); |
1da177e4 | 1547 | case LOOP_SET_STATUS: |
7035b5df | 1548 | err = -EPERM; |
a1316544 | 1549 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1550 | err = loop_set_status_old(lo, |
1551 | (struct loop_info __user *)arg); | |
a1316544 | 1552 | } |
1da177e4 LT |
1553 | break; |
1554 | case LOOP_GET_STATUS: | |
4a5ce9ba | 1555 | return loop_get_status_old(lo, (struct loop_info __user *) arg); |
1da177e4 | 1556 | case LOOP_SET_STATUS64: |
7035b5df | 1557 | err = -EPERM; |
a1316544 | 1558 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { |
7035b5df DM |
1559 | err = loop_set_status64(lo, |
1560 | (struct loop_info64 __user *) arg); | |
a1316544 | 1561 | } |
1da177e4 LT |
1562 | break; |
1563 | case LOOP_GET_STATUS64: | |
4a5ce9ba | 1564 | return loop_get_status64(lo, (struct loop_info64 __user *) arg); |
a1316544 | 1565 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1566 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1567 | case LOOP_SET_BLOCK_SIZE: |
a1316544 JK |
1568 | if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) |
1569 | return -EPERM; | |
1570 | /* Fall through */ | |
1da177e4 | 1571 | default: |
a1316544 JK |
1572 | err = lo_simple_ioctl(lo, cmd, arg); |
1573 | break; | |
1da177e4 | 1574 | } |
f028f3b2 | 1575 | |
1da177e4 LT |
1576 | return err; |
1577 | } | |
1578 | ||
863d5b82 DH |
1579 | #ifdef CONFIG_COMPAT |
1580 | struct compat_loop_info { | |
1581 | compat_int_t lo_number; /* ioctl r/o */ | |
1582 | compat_dev_t lo_device; /* ioctl r/o */ | |
1583 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1584 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1585 | compat_int_t lo_offset; | |
1586 | compat_int_t lo_encrypt_type; | |
1587 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1588 | compat_int_t lo_flags; /* ioctl r/o */ | |
1589 | char lo_name[LO_NAME_SIZE]; | |
1590 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1591 | compat_ulong_t lo_init[2]; | |
1592 | char reserved[4]; | |
1593 | }; | |
1594 | ||
1595 | /* | |
1596 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1597 | * - noinlined to reduce stack space usage in main part of driver | |
1598 | */ | |
1599 | static noinline int | |
ba674cfc | 1600 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1601 | struct loop_info64 *info64) |
1602 | { | |
1603 | struct compat_loop_info info; | |
1604 | ||
1605 | if (copy_from_user(&info, arg, sizeof(info))) | |
1606 | return -EFAULT; | |
1607 | ||
1608 | memset(info64, 0, sizeof(*info64)); | |
1609 | info64->lo_number = info.lo_number; | |
1610 | info64->lo_device = info.lo_device; | |
1611 | info64->lo_inode = info.lo_inode; | |
1612 | info64->lo_rdevice = info.lo_rdevice; | |
1613 | info64->lo_offset = info.lo_offset; | |
1614 | info64->lo_sizelimit = 0; | |
1615 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1616 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1617 | info64->lo_flags = info.lo_flags; | |
1618 | info64->lo_init[0] = info.lo_init[0]; | |
1619 | info64->lo_init[1] = info.lo_init[1]; | |
1620 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1621 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1622 | else | |
1623 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1624 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1625 | return 0; | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1630 | * - noinlined to reduce stack space usage in main part of driver | |
1631 | */ | |
1632 | static noinline int | |
1633 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1634 | struct compat_loop_info __user *arg) | |
1635 | { | |
1636 | struct compat_loop_info info; | |
1637 | ||
1638 | memset(&info, 0, sizeof(info)); | |
1639 | info.lo_number = info64->lo_number; | |
1640 | info.lo_device = info64->lo_device; | |
1641 | info.lo_inode = info64->lo_inode; | |
1642 | info.lo_rdevice = info64->lo_rdevice; | |
1643 | info.lo_offset = info64->lo_offset; | |
1644 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1645 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1646 | info.lo_flags = info64->lo_flags; | |
1647 | info.lo_init[0] = info64->lo_init[0]; | |
1648 | info.lo_init[1] = info64->lo_init[1]; | |
1649 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1650 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1651 | else | |
1652 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1653 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1654 | ||
1655 | /* error in case values were truncated */ | |
1656 | if (info.lo_device != info64->lo_device || | |
1657 | info.lo_rdevice != info64->lo_rdevice || | |
1658 | info.lo_inode != info64->lo_inode || | |
1659 | info.lo_offset != info64->lo_offset || | |
1660 | info.lo_init[0] != info64->lo_init[0] || | |
1661 | info.lo_init[1] != info64->lo_init[1]) | |
1662 | return -EOVERFLOW; | |
1663 | ||
1664 | if (copy_to_user(arg, &info, sizeof(info))) | |
1665 | return -EFAULT; | |
1666 | return 0; | |
1667 | } | |
1668 | ||
1669 | static int | |
1670 | loop_set_status_compat(struct loop_device *lo, | |
1671 | const struct compat_loop_info __user *arg) | |
1672 | { | |
1673 | struct loop_info64 info64; | |
1674 | int ret; | |
1675 | ||
1676 | ret = loop_info64_from_compat(arg, &info64); | |
1677 | if (ret < 0) | |
1678 | return ret; | |
1679 | return loop_set_status(lo, &info64); | |
1680 | } | |
1681 | ||
1682 | static int | |
1683 | loop_get_status_compat(struct loop_device *lo, | |
1684 | struct compat_loop_info __user *arg) | |
1685 | { | |
1686 | struct loop_info64 info64; | |
bdac616d | 1687 | int err; |
863d5b82 | 1688 | |
4a5ce9ba | 1689 | if (!arg) |
bdac616d | 1690 | return -EINVAL; |
bdac616d | 1691 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1692 | if (!err) |
1693 | err = loop_info64_to_compat(&info64, arg); | |
1694 | return err; | |
1695 | } | |
1696 | ||
bb214884 AV |
1697 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1698 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1699 | { |
bb214884 | 1700 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1701 | int err; |
1702 | ||
863d5b82 DH |
1703 | switch(cmd) { |
1704 | case LOOP_SET_STATUS: | |
550df5fd JK |
1705 | err = loop_set_status_compat(lo, |
1706 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1707 | break; |
1708 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1709 | err = loop_get_status_compat(lo, |
1710 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1711 | break; |
53d66608 | 1712 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1713 | case LOOP_CLR_FD: |
1714 | case LOOP_GET_STATUS64: | |
1715 | case LOOP_SET_STATUS64: | |
1716 | arg = (unsigned long) compat_ptr(arg); | |
d893ff86 | 1717 | /* fall through */ |
863d5b82 DH |
1718 | case LOOP_SET_FD: |
1719 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1720 | case LOOP_SET_BLOCK_SIZE: |
bb214884 | 1721 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1722 | break; |
1723 | default: | |
1724 | err = -ENOIOCTLCMD; | |
1725 | break; | |
1726 | } | |
863d5b82 DH |
1727 | return err; |
1728 | } | |
1729 | #endif | |
1730 | ||
bb214884 | 1731 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1732 | { |
770fe30a | 1733 | struct loop_device *lo; |
0a42e99b | 1734 | int err; |
770fe30a | 1735 | |
0a42e99b JK |
1736 | err = mutex_lock_killable(&loop_ctl_mutex); |
1737 | if (err) | |
1738 | return err; | |
770fe30a KS |
1739 | lo = bdev->bd_disk->private_data; |
1740 | if (!lo) { | |
1741 | err = -ENXIO; | |
1742 | goto out; | |
1743 | } | |
1da177e4 | 1744 | |
f8933667 | 1745 | atomic_inc(&lo->lo_refcnt); |
770fe30a | 1746 | out: |
0a42e99b | 1747 | mutex_unlock(&loop_ctl_mutex); |
770fe30a | 1748 | return err; |
1da177e4 LT |
1749 | } |
1750 | ||
967d1dc1 | 1751 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1752 | { |
967d1dc1 | 1753 | struct loop_device *lo; |
1da177e4 | 1754 | |
0a42e99b | 1755 | mutex_lock(&loop_ctl_mutex); |
967d1dc1 | 1756 | lo = disk->private_data; |
f8933667 | 1757 | if (atomic_dec_return(&lo->lo_refcnt)) |
0a42e99b | 1758 | goto out_unlock; |
14f27939 MB |
1759 | |
1760 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
a2505b79 JK |
1761 | if (lo->lo_state != Lo_bound) |
1762 | goto out_unlock; | |
1763 | lo->lo_state = Lo_rundown; | |
7ccd0791 | 1764 | mutex_unlock(&loop_ctl_mutex); |
14f27939 MB |
1765 | /* |
1766 | * In autoclear mode, stop the loop thread | |
1767 | * and remove configuration after last close. | |
1768 | */ | |
0da03cab | 1769 | __loop_clr_fd(lo, true); |
7ccd0791 | 1770 | return; |
43cade80 | 1771 | } else if (lo->lo_state == Lo_bound) { |
14f27939 MB |
1772 | /* |
1773 | * Otherwise keep thread (if running) and config, | |
1774 | * but flush possible ongoing bios in thread. | |
1775 | */ | |
43cade80 OS |
1776 | blk_mq_freeze_queue(lo->lo_queue); |
1777 | blk_mq_unfreeze_queue(lo->lo_queue); | |
14f27939 | 1778 | } |
96c58655 | 1779 | |
0a42e99b | 1780 | out_unlock: |
310ca162 | 1781 | mutex_unlock(&loop_ctl_mutex); |
ae665016 LT |
1782 | } |
1783 | ||
83d5cde4 | 1784 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1785 | .owner = THIS_MODULE, |
bb214884 AV |
1786 | .open = lo_open, |
1787 | .release = lo_release, | |
1788 | .ioctl = lo_ioctl, | |
863d5b82 | 1789 | #ifdef CONFIG_COMPAT |
bb214884 | 1790 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1791 | #endif |
1da177e4 LT |
1792 | }; |
1793 | ||
1794 | /* | |
1795 | * And now the modules code and kernel interface. | |
1796 | */ | |
73285082 | 1797 | static int max_loop; |
5657a819 | 1798 | module_param(max_loop, int, 0444); |
a47653fc | 1799 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
5657a819 | 1800 | module_param(max_part, int, 0444); |
476a4813 | 1801 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1802 | MODULE_LICENSE("GPL"); |
1803 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1804 | ||
1805 | int loop_register_transfer(struct loop_func_table *funcs) | |
1806 | { | |
1807 | unsigned int n = funcs->number; | |
1808 | ||
1809 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1810 | return -EINVAL; | |
1811 | xfer_funcs[n] = funcs; | |
1812 | return 0; | |
1813 | } | |
1814 | ||
34dd82af KS |
1815 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1816 | { | |
1817 | struct loop_device *lo = ptr; | |
1818 | struct loop_func_table *xfer = data; | |
1819 | ||
310ca162 | 1820 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
1821 | if (lo->lo_encryption == xfer) |
1822 | loop_release_xfer(lo); | |
310ca162 | 1823 | mutex_unlock(&loop_ctl_mutex); |
34dd82af KS |
1824 | return 0; |
1825 | } | |
1826 | ||
1da177e4 LT |
1827 | int loop_unregister_transfer(int number) |
1828 | { | |
1829 | unsigned int n = number; | |
1da177e4 LT |
1830 | struct loop_func_table *xfer; |
1831 | ||
1832 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1833 | return -EINVAL; | |
1834 | ||
1835 | xfer_funcs[n] = NULL; | |
34dd82af | 1836 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1837 | return 0; |
1838 | } | |
1839 | ||
1840 | EXPORT_SYMBOL(loop_register_transfer); | |
1841 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1842 | ||
fc17b653 | 1843 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1844 | const struct blk_mq_queue_data *bd) |
1845 | { | |
1894e916 JA |
1846 | struct request *rq = bd->rq; |
1847 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1848 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 1849 | |
1894e916 | 1850 | blk_mq_start_request(rq); |
b5dd2f60 | 1851 | |
f4aa4c7b | 1852 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1853 | return BLK_STS_IOERR; |
f4aa4c7b | 1854 | |
1894e916 | 1855 | switch (req_op(rq)) { |
f0225cac CH |
1856 | case REQ_OP_FLUSH: |
1857 | case REQ_OP_DISCARD: | |
19372e27 | 1858 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1859 | cmd->use_aio = false; |
f0225cac CH |
1860 | break; |
1861 | default: | |
1862 | cmd->use_aio = lo->use_dio; | |
1863 | break; | |
1864 | } | |
bc07c10a | 1865 | |
d4478e92 | 1866 | /* always use the first bio's css */ |
0b508bc9 | 1867 | #ifdef CONFIG_BLK_CGROUP |
db6638d7 DZ |
1868 | if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { |
1869 | cmd->css = &bio_blkcg(rq->bio)->css; | |
d4478e92 SL |
1870 | css_get(cmd->css); |
1871 | } else | |
1872 | #endif | |
1873 | cmd->css = NULL; | |
3989144f | 1874 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 | 1875 | |
fc17b653 | 1876 | return BLK_STS_OK; |
b5dd2f60 ML |
1877 | } |
1878 | ||
1879 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1880 | { | |
1894e916 JA |
1881 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
1882 | const bool write = op_is_write(req_op(rq)); | |
1883 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 1884 | int ret = 0; |
b5dd2f60 | 1885 | |
f4829a9b CH |
1886 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1887 | ret = -EIO; | |
b5dd2f60 | 1888 | goto failed; |
f4829a9b | 1889 | } |
b5dd2f60 | 1890 | |
1894e916 | 1891 | ret = do_req_filebacked(lo, rq); |
b5dd2f60 | 1892 | failed: |
bc07c10a | 1893 | /* complete non-aio request */ |
fe2cb290 CH |
1894 | if (!cmd->use_aio || ret) { |
1895 | cmd->ret = ret ? -EIO : 0; | |
1894e916 | 1896 | blk_mq_complete_request(rq); |
fe2cb290 | 1897 | } |
b5dd2f60 ML |
1898 | } |
1899 | ||
e03a3d7a | 1900 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1901 | { |
1902 | struct loop_cmd *cmd = | |
e03a3d7a | 1903 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1904 | |
1905 | loop_handle_cmd(cmd); | |
1906 | } | |
1907 | ||
d6296d39 CH |
1908 | static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1909 | unsigned int hctx_idx, unsigned int numa_node) | |
b5dd2f60 ML |
1910 | { |
1911 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1912 | ||
3989144f | 1913 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1914 | return 0; |
1915 | } | |
1916 | ||
f363b089 | 1917 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1918 | .queue_rq = loop_queue_rq, |
b5dd2f60 | 1919 | .init_request = loop_init_request, |
fe2cb290 | 1920 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1921 | }; |
1922 | ||
34dd82af | 1923 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1924 | { |
1925 | struct loop_device *lo; | |
1926 | struct gendisk *disk; | |
34dd82af | 1927 | int err; |
73285082 | 1928 | |
68d740d7 | 1929 | err = -ENOMEM; |
73285082 | 1930 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1931 | if (!lo) |
73285082 | 1932 | goto out; |
34dd82af | 1933 | |
ef7e7c82 MP |
1934 | lo->lo_state = Lo_unbound; |
1935 | ||
c718aa65 | 1936 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1937 | if (i >= 0) { |
c718aa65 TH |
1938 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1939 | if (err == -ENOSPC) | |
34dd82af | 1940 | err = -EEXIST; |
34dd82af | 1941 | } else { |
c718aa65 | 1942 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1943 | } |
1944 | if (err < 0) | |
1945 | goto out_free_dev; | |
c718aa65 | 1946 | i = err; |
73285082 | 1947 | |
183cfb57 | 1948 | err = -ENOMEM; |
b5dd2f60 ML |
1949 | lo->tag_set.ops = &loop_mq_ops; |
1950 | lo->tag_set.nr_hw_queues = 1; | |
1951 | lo->tag_set.queue_depth = 128; | |
1952 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1953 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
56d18f62 | 1954 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; |
b5dd2f60 ML |
1955 | lo->tag_set.driver_data = lo; |
1956 | ||
1957 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1958 | if (err) | |
3ec981e3 | 1959 | goto out_free_idr; |
73285082 | 1960 | |
b5dd2f60 | 1961 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
38a3499f | 1962 | if (IS_ERR(lo->lo_queue)) { |
b5dd2f60 ML |
1963 | err = PTR_ERR(lo->lo_queue); |
1964 | goto out_cleanup_tags; | |
1965 | } | |
ef7e7c82 MP |
1966 | lo->lo_queue->queuedata = lo; |
1967 | ||
54bb0ade | 1968 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 1969 | |
5b5e20f4 | 1970 | /* |
40326d8a SL |
1971 | * By default, we do buffer IO, so it doesn't make sense to enable |
1972 | * merge because the I/O submitted to backing file is handled page by | |
1973 | * page. For directio mode, merge does help to dispatch bigger request | |
1974 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 | 1975 | */ |
8b904b5b | 1976 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
5b5e20f4 | 1977 | |
7a649737 | 1978 | err = -ENOMEM; |
476a4813 | 1979 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1980 | if (!disk) |
1981 | goto out_free_queue; | |
1982 | ||
e03c8dd1 KS |
1983 | /* |
1984 | * Disable partition scanning by default. The in-kernel partition | |
1985 | * scanning can be requested individually per-device during its | |
1986 | * setup. Userspace can always add and remove partitions from all | |
1987 | * devices. The needed partition minors are allocated from the | |
1988 | * extended minor space, the main loop device numbers will continue | |
1989 | * to match the loop minors, regardless of the number of partitions | |
1990 | * used. | |
1991 | * | |
1992 | * If max_part is given, partition scanning is globally enabled for | |
1993 | * all loop devices. The minors for the main loop devices will be | |
1994 | * multiples of max_part. | |
1995 | * | |
1996 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1997 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1998 | * complicated, are too static, inflexible and may surprise | |
1999 | * userspace tools. Parameters like this in general should be avoided. | |
2000 | */ | |
2001 | if (!part_shift) | |
2002 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
2003 | disk->flags |= GENHD_FL_EXT_DEVT; | |
f8933667 | 2004 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 2005 | lo->lo_number = i; |
73285082 KC |
2006 | spin_lock_init(&lo->lo_lock); |
2007 | disk->major = LOOP_MAJOR; | |
476a4813 | 2008 | disk->first_minor = i << part_shift; |
73285082 KC |
2009 | disk->fops = &lo_fops; |
2010 | disk->private_data = lo; | |
2011 | disk->queue = lo->lo_queue; | |
2012 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
2013 | add_disk(disk); |
2014 | *l = lo; | |
2015 | return lo->lo_number; | |
73285082 KC |
2016 | |
2017 | out_free_queue: | |
2018 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
2019 | out_cleanup_tags: |
2020 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
2021 | out_free_idr: |
2022 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
2023 | out_free_dev: |
2024 | kfree(lo); | |
2025 | out: | |
34dd82af | 2026 | return err; |
73285082 KC |
2027 | } |
2028 | ||
34dd82af | 2029 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2030 | { |
6cd18e71 | 2031 | del_gendisk(lo->lo_disk); |
0fa8ebdd | 2032 | blk_cleanup_queue(lo->lo_queue); |
b5dd2f60 | 2033 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 2034 | put_disk(lo->lo_disk); |
73285082 KC |
2035 | kfree(lo); |
2036 | } | |
1da177e4 | 2037 | |
770fe30a KS |
2038 | static int find_free_cb(int id, void *ptr, void *data) |
2039 | { | |
2040 | struct loop_device *lo = ptr; | |
2041 | struct loop_device **l = data; | |
2042 | ||
2043 | if (lo->lo_state == Lo_unbound) { | |
2044 | *l = lo; | |
2045 | return 1; | |
2046 | } | |
2047 | return 0; | |
2048 | } | |
2049 | ||
34dd82af | 2050 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
2051 | { |
2052 | struct loop_device *lo; | |
34dd82af | 2053 | int ret = -ENODEV; |
a47653fc | 2054 | |
770fe30a KS |
2055 | if (i < 0) { |
2056 | int err; | |
2057 | ||
2058 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
2059 | if (err == 1) { | |
2060 | *l = lo; | |
2061 | ret = lo->lo_number; | |
2062 | } | |
2063 | goto out; | |
a47653fc KC |
2064 | } |
2065 | ||
770fe30a | 2066 | /* lookup and return a specific i */ |
34dd82af | 2067 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 2068 | if (lo) { |
34dd82af KS |
2069 | *l = lo; |
2070 | ret = lo->lo_number; | |
a47653fc | 2071 | } |
770fe30a | 2072 | out: |
34dd82af | 2073 | return ret; |
a47653fc KC |
2074 | } |
2075 | ||
73285082 KC |
2076 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
2077 | { | |
705962cc | 2078 | struct loop_device *lo; |
07002e99 | 2079 | struct kobject *kobj; |
34dd82af | 2080 | int err; |
73285082 | 2081 | |
0a42e99b | 2082 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2083 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); |
2084 | if (err < 0) | |
2085 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
2086 | if (err < 0) | |
a207f593 | 2087 | kobj = NULL; |
34dd82af | 2088 | else |
3079c22e | 2089 | kobj = get_disk_and_module(lo->lo_disk); |
0a42e99b | 2090 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2091 | |
2092 | *part = 0; | |
07002e99 | 2093 | return kobj; |
73285082 KC |
2094 | } |
2095 | ||
770fe30a KS |
2096 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2097 | unsigned long parm) | |
2098 | { | |
2099 | struct loop_device *lo; | |
0a42e99b | 2100 | int ret; |
770fe30a | 2101 | |
0a42e99b JK |
2102 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2103 | if (ret) | |
2104 | return ret; | |
2105 | ||
2106 | ret = -ENOSYS; | |
770fe30a KS |
2107 | switch (cmd) { |
2108 | case LOOP_CTL_ADD: | |
2109 | ret = loop_lookup(&lo, parm); | |
2110 | if (ret >= 0) { | |
2111 | ret = -EEXIST; | |
2112 | break; | |
2113 | } | |
2114 | ret = loop_add(&lo, parm); | |
2115 | break; | |
2116 | case LOOP_CTL_REMOVE: | |
2117 | ret = loop_lookup(&lo, parm); | |
2118 | if (ret < 0) | |
2119 | break; | |
770fe30a KS |
2120 | if (lo->lo_state != Lo_unbound) { |
2121 | ret = -EBUSY; | |
770fe30a KS |
2122 | break; |
2123 | } | |
f8933667 | 2124 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a | 2125 | ret = -EBUSY; |
770fe30a KS |
2126 | break; |
2127 | } | |
2128 | lo->lo_disk->private_data = NULL; | |
770fe30a KS |
2129 | idr_remove(&loop_index_idr, lo->lo_number); |
2130 | loop_remove(lo); | |
2131 | break; | |
2132 | case LOOP_CTL_GET_FREE: | |
2133 | ret = loop_lookup(&lo, -1); | |
2134 | if (ret >= 0) | |
2135 | break; | |
2136 | ret = loop_add(&lo, -1); | |
2137 | } | |
0a42e99b | 2138 | mutex_unlock(&loop_ctl_mutex); |
770fe30a KS |
2139 | |
2140 | return ret; | |
2141 | } | |
2142 | ||
2143 | static const struct file_operations loop_ctl_fops = { | |
2144 | .open = nonseekable_open, | |
2145 | .unlocked_ioctl = loop_control_ioctl, | |
2146 | .compat_ioctl = loop_control_ioctl, | |
2147 | .owner = THIS_MODULE, | |
2148 | .llseek = noop_llseek, | |
2149 | }; | |
2150 | ||
2151 | static struct miscdevice loop_misc = { | |
2152 | .minor = LOOP_CTRL_MINOR, | |
2153 | .name = "loop-control", | |
2154 | .fops = &loop_ctl_fops, | |
2155 | }; | |
2156 | ||
2157 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2158 | MODULE_ALIAS("devname:loop-control"); | |
2159 | ||
73285082 KC |
2160 | static int __init loop_init(void) |
2161 | { | |
a47653fc KC |
2162 | int i, nr; |
2163 | unsigned long range; | |
34dd82af | 2164 | struct loop_device *lo; |
770fe30a | 2165 | int err; |
a47653fc | 2166 | |
476a4813 | 2167 | part_shift = 0; |
ac04fee0 | 2168 | if (max_part > 0) { |
476a4813 LV |
2169 | part_shift = fls(max_part); |
2170 | ||
ac04fee0 NK |
2171 | /* |
2172 | * Adjust max_part according to part_shift as it is exported | |
2173 | * to user space so that user can decide correct minor number | |
2174 | * if [s]he want to create more devices. | |
2175 | * | |
2176 | * Note that -1 is required because partition 0 is reserved | |
2177 | * for the whole disk. | |
2178 | */ | |
2179 | max_part = (1UL << part_shift) - 1; | |
2180 | } | |
2181 | ||
b1a66504 GC |
2182 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2183 | err = -EINVAL; | |
a8c1d064 | 2184 | goto err_out; |
b1a66504 | 2185 | } |
78f4bb36 | 2186 | |
b1a66504 GC |
2187 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2188 | err = -EINVAL; | |
a8c1d064 | 2189 | goto err_out; |
b1a66504 | 2190 | } |
1da177e4 | 2191 | |
d134b00b KS |
2192 | /* |
2193 | * If max_loop is specified, create that many devices upfront. | |
2194 | * This also becomes a hard limit. If max_loop is not specified, | |
2195 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2196 | * init time. Loop devices can be requested on-demand with the | |
2197 | * /dev/loop-control interface, or be instantiated by accessing | |
2198 | * a 'dead' device node. | |
2199 | */ | |
73285082 | 2200 | if (max_loop) { |
a47653fc | 2201 | nr = max_loop; |
a1c15c59 | 2202 | range = max_loop << part_shift; |
a47653fc | 2203 | } else { |
d134b00b | 2204 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2205 | range = 1UL << MINORBITS; |
a47653fc KC |
2206 | } |
2207 | ||
a8c1d064 AV |
2208 | err = misc_register(&loop_misc); |
2209 | if (err < 0) | |
2210 | goto err_out; | |
2211 | ||
2212 | ||
b1a66504 GC |
2213 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2214 | err = -EIO; | |
2215 | goto misc_out; | |
2216 | } | |
1da177e4 | 2217 | |
a47653fc KC |
2218 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2219 | THIS_MODULE, loop_probe, NULL, NULL); | |
2220 | ||
d134b00b | 2221 | /* pre-create number of devices given by config or max_loop */ |
0a42e99b | 2222 | mutex_lock(&loop_ctl_mutex); |
34dd82af KS |
2223 | for (i = 0; i < nr; i++) |
2224 | loop_add(&lo, i); | |
0a42e99b | 2225 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 2226 | |
73285082 | 2227 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2228 | return 0; |
b1a66504 GC |
2229 | |
2230 | misc_out: | |
2231 | misc_deregister(&loop_misc); | |
a8c1d064 | 2232 | err_out: |
b1a66504 | 2233 | return err; |
34dd82af | 2234 | } |
a47653fc | 2235 | |
34dd82af KS |
2236 | static int loop_exit_cb(int id, void *ptr, void *data) |
2237 | { | |
2238 | struct loop_device *lo = ptr; | |
a47653fc | 2239 | |
34dd82af KS |
2240 | loop_remove(lo); |
2241 | return 0; | |
1da177e4 LT |
2242 | } |
2243 | ||
73285082 | 2244 | static void __exit loop_exit(void) |
1da177e4 | 2245 | { |
a47653fc | 2246 | unsigned long range; |
1da177e4 | 2247 | |
a1c15c59 | 2248 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2249 | |
34dd82af | 2250 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2251 | idr_destroy(&loop_index_idr); |
73285082 | 2252 | |
a47653fc | 2253 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2254 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2255 | |
2256 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2257 | } |
2258 | ||
2259 | module_init(loop_init); | |
2260 | module_exit(loop_exit); | |
2261 | ||
2262 | #ifndef MODULE | |
2263 | static int __init max_loop_setup(char *str) | |
2264 | { | |
2265 | max_loop = simple_strtol(str, NULL, 0); | |
2266 | return 1; | |
2267 | } | |
2268 | ||
2269 | __setup("max_loop=", max_loop_setup); | |
2270 | #endif |