]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/block/loop.c | |
3 | * | |
4 | * Written by Theodore Ts'o, 3/29/93 | |
5 | * | |
6 | * Copyright 1993 by Theodore Ts'o. Redistribution of this file is | |
7 | * permitted under the GNU General Public License. | |
8 | * | |
9 | * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 | |
10 | * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 | |
11 | * | |
12 | * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 | |
13 | * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 | |
14 | * | |
15 | * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 | |
16 | * | |
17 | * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 | |
18 | * | |
19 | * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 | |
20 | * | |
21 | * Loadable modules and other fixes by AK, 1998 | |
22 | * | |
23 | * Make real block number available to downstream transfer functions, enables | |
24 | * CBC (and relatives) mode encryption requiring unique IVs per data block. | |
25 | * Reed H. Petty, rhp@draper.net | |
26 | * | |
27 | * Maximum number of loop devices now dynamic via max_loop module parameter. | |
28 | * Russell Kroll <rkroll@exploits.org> 19990701 | |
29 | * | |
30 | * Maximum number of loop devices when compiled-in now selectable by passing | |
31 | * max_loop=<1-255> to the kernel on boot. | |
96de0e25 | 32 | * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 |
1da177e4 LT |
33 | * |
34 | * Completely rewrite request handling to be make_request_fn style and | |
35 | * non blocking, pushing work to a helper thread. Lots of fixes from | |
36 | * Al Viro too. | |
37 | * Jens Axboe <axboe@suse.de>, Nov 2000 | |
38 | * | |
39 | * Support up to 256 loop devices | |
40 | * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 | |
41 | * | |
42 | * Support for falling back on the write file operation when the address space | |
4e02ed4b | 43 | * operations write_begin is not available on the backing filesystem. |
1da177e4 LT |
44 | * Anton Altaparmakov, 16 Feb 2005 |
45 | * | |
46 | * Still To Fix: | |
47 | * - Advisory locking is ignored here. | |
48 | * - Should use an own CAP_* category instead of CAP_SYS_ADMIN | |
49 | * | |
50 | */ | |
51 | ||
1da177e4 LT |
52 | #include <linux/module.h> |
53 | #include <linux/moduleparam.h> | |
54 | #include <linux/sched.h> | |
55 | #include <linux/fs.h> | |
56 | #include <linux/file.h> | |
57 | #include <linux/stat.h> | |
58 | #include <linux/errno.h> | |
59 | #include <linux/major.h> | |
60 | #include <linux/wait.h> | |
61 | #include <linux/blkdev.h> | |
62 | #include <linux/blkpg.h> | |
63 | #include <linux/init.h> | |
1da177e4 LT |
64 | #include <linux/swap.h> |
65 | #include <linux/slab.h> | |
863d5b82 | 66 | #include <linux/compat.h> |
1da177e4 | 67 | #include <linux/suspend.h> |
83144186 | 68 | #include <linux/freezer.h> |
2a48fc0a | 69 | #include <linux/mutex.h> |
1da177e4 | 70 | #include <linux/writeback.h> |
1da177e4 LT |
71 | #include <linux/completion.h> |
72 | #include <linux/highmem.h> | |
6c997918 | 73 | #include <linux/kthread.h> |
d6b29d7c | 74 | #include <linux/splice.h> |
ee862730 | 75 | #include <linux/sysfs.h> |
770fe30a | 76 | #include <linux/miscdevice.h> |
dfaa2ef6 | 77 | #include <linux/falloc.h> |
283e7e5d | 78 | #include <linux/uio.h> |
83a87611 | 79 | #include "loop.h" |
1da177e4 LT |
80 | |
81 | #include <asm/uaccess.h> | |
82 | ||
34dd82af KS |
83 | static DEFINE_IDR(loop_index_idr); |
84 | static DEFINE_MUTEX(loop_index_mutex); | |
1da177e4 | 85 | |
476a4813 LV |
86 | static int max_part; |
87 | static int part_shift; | |
88 | ||
1da177e4 LT |
89 | static int transfer_xor(struct loop_device *lo, int cmd, |
90 | struct page *raw_page, unsigned raw_off, | |
91 | struct page *loop_page, unsigned loop_off, | |
92 | int size, sector_t real_block) | |
93 | { | |
cfd8005c CW |
94 | char *raw_buf = kmap_atomic(raw_page) + raw_off; |
95 | char *loop_buf = kmap_atomic(loop_page) + loop_off; | |
1da177e4 LT |
96 | char *in, *out, *key; |
97 | int i, keysize; | |
98 | ||
99 | if (cmd == READ) { | |
100 | in = raw_buf; | |
101 | out = loop_buf; | |
102 | } else { | |
103 | in = loop_buf; | |
104 | out = raw_buf; | |
105 | } | |
106 | ||
107 | key = lo->lo_encrypt_key; | |
108 | keysize = lo->lo_encrypt_key_size; | |
109 | for (i = 0; i < size; i++) | |
110 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
111 | ||
cfd8005c CW |
112 | kunmap_atomic(loop_buf); |
113 | kunmap_atomic(raw_buf); | |
1da177e4 LT |
114 | cond_resched(); |
115 | return 0; | |
116 | } | |
117 | ||
118 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
119 | { | |
120 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
121 | return -EINVAL; | |
122 | return 0; | |
123 | } | |
124 | ||
125 | static struct loop_func_table none_funcs = { | |
126 | .number = LO_CRYPT_NONE, | |
aa4d8616 | 127 | }; |
1da177e4 LT |
128 | |
129 | static struct loop_func_table xor_funcs = { | |
130 | .number = LO_CRYPT_XOR, | |
131 | .transfer = transfer_xor, | |
132 | .init = xor_init | |
aa4d8616 | 133 | }; |
1da177e4 LT |
134 | |
135 | /* xfer_funcs[0] is special - its release function is never called */ | |
136 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
137 | &none_funcs, | |
138 | &xor_funcs | |
139 | }; | |
140 | ||
7035b5df | 141 | static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) |
1da177e4 | 142 | { |
b7a1da69 | 143 | loff_t loopsize; |
1da177e4 LT |
144 | |
145 | /* Compute loopsize in bytes */ | |
b7a1da69 GC |
146 | loopsize = i_size_read(file->f_mapping->host); |
147 | if (offset > 0) | |
148 | loopsize -= offset; | |
149 | /* offset is beyond i_size, weird but possible */ | |
7035b5df DM |
150 | if (loopsize < 0) |
151 | return 0; | |
1da177e4 | 152 | |
7035b5df DM |
153 | if (sizelimit > 0 && sizelimit < loopsize) |
154 | loopsize = sizelimit; | |
1da177e4 LT |
155 | /* |
156 | * Unfortunately, if we want to do I/O on the device, | |
157 | * the number of 512-byte sectors has to fit into a sector_t. | |
158 | */ | |
159 | return loopsize >> 9; | |
160 | } | |
161 | ||
7035b5df DM |
162 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) |
163 | { | |
164 | return get_size(lo->lo_offset, lo->lo_sizelimit, file); | |
165 | } | |
166 | ||
2e5ab5f3 ML |
167 | static void __loop_update_dio(struct loop_device *lo, bool dio) |
168 | { | |
169 | struct file *file = lo->lo_backing_file; | |
170 | struct address_space *mapping = file->f_mapping; | |
171 | struct inode *inode = mapping->host; | |
172 | unsigned short sb_bsize = 0; | |
173 | unsigned dio_align = 0; | |
174 | bool use_dio; | |
175 | ||
176 | if (inode->i_sb->s_bdev) { | |
177 | sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); | |
178 | dio_align = sb_bsize - 1; | |
179 | } | |
180 | ||
181 | /* | |
182 | * We support direct I/O only if lo_offset is aligned with the | |
183 | * logical I/O size of backing device, and the logical block | |
184 | * size of loop is bigger than the backing device's and the loop | |
185 | * needn't transform transfer. | |
186 | * | |
187 | * TODO: the above condition may be loosed in the future, and | |
188 | * direct I/O may be switched runtime at that time because most | |
189 | * of requests in sane appplications should be PAGE_SIZE algined | |
190 | */ | |
191 | if (dio) { | |
192 | if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && | |
193 | !(lo->lo_offset & dio_align) && | |
194 | mapping->a_ops->direct_IO && | |
195 | !lo->transfer) | |
196 | use_dio = true; | |
197 | else | |
198 | use_dio = false; | |
199 | } else { | |
200 | use_dio = false; | |
201 | } | |
202 | ||
203 | if (lo->use_dio == use_dio) | |
204 | return; | |
205 | ||
206 | /* flush dirty pages before changing direct IO */ | |
207 | vfs_fsync(file, 0); | |
208 | ||
209 | /* | |
210 | * The flag of LO_FLAGS_DIRECT_IO is handled similarly with | |
211 | * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup | |
212 | * will get updated by ioctl(LOOP_GET_STATUS) | |
213 | */ | |
214 | blk_mq_freeze_queue(lo->lo_queue); | |
215 | lo->use_dio = use_dio; | |
216 | if (use_dio) | |
217 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; | |
218 | else | |
219 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; | |
220 | blk_mq_unfreeze_queue(lo->lo_queue); | |
221 | } | |
222 | ||
1da177e4 | 223 | static int |
7035b5df | 224 | figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) |
1da177e4 | 225 | { |
7035b5df | 226 | loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); |
1da177e4 | 227 | sector_t x = (sector_t)size; |
7b0576a3 | 228 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
229 | |
230 | if (unlikely((loff_t)x != size)) | |
231 | return -EFBIG; | |
7035b5df DM |
232 | if (lo->lo_offset != offset) |
233 | lo->lo_offset = offset; | |
234 | if (lo->lo_sizelimit != sizelimit) | |
235 | lo->lo_sizelimit = sizelimit; | |
73285082 | 236 | set_capacity(lo->lo_disk, x); |
7b0576a3 GC |
237 | bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); |
238 | /* let user-space know about the new size */ | |
239 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
7035b5df | 240 | return 0; |
1da177e4 LT |
241 | } |
242 | ||
243 | static inline int | |
244 | lo_do_transfer(struct loop_device *lo, int cmd, | |
245 | struct page *rpage, unsigned roffs, | |
246 | struct page *lpage, unsigned loffs, | |
247 | int size, sector_t rblock) | |
248 | { | |
aa4d8616 CH |
249 | int ret; |
250 | ||
251 | ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
252 | if (likely(!ret)) | |
1da177e4 LT |
253 | return 0; |
254 | ||
aa4d8616 CH |
255 | printk_ratelimited(KERN_ERR |
256 | "loop: Transfer error at byte offset %llu, length %i.\n", | |
257 | (unsigned long long)rblock << 9, size); | |
258 | return ret; | |
1da177e4 LT |
259 | } |
260 | ||
aa4d8616 | 261 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 262 | { |
aa4d8616 | 263 | struct iov_iter i; |
1da177e4 | 264 | ssize_t bw; |
283e7e5d | 265 | |
aa4d8616 | 266 | iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len); |
1da177e4 | 267 | |
03d95eb2 | 268 | file_start_write(file); |
aa4d8616 | 269 | bw = vfs_iter_write(file, &i, ppos); |
03d95eb2 | 270 | file_end_write(file); |
aa4d8616 CH |
271 | |
272 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 273 | return 0; |
aa4d8616 CH |
274 | |
275 | printk_ratelimited(KERN_ERR | |
276 | "loop: Write error at byte offset %llu, length %i.\n", | |
277 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
278 | if (bw >= 0) |
279 | bw = -EIO; | |
280 | return bw; | |
281 | } | |
282 | ||
aa4d8616 CH |
283 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
284 | loff_t pos) | |
1da177e4 | 285 | { |
aa4d8616 CH |
286 | struct bio_vec bvec; |
287 | struct req_iterator iter; | |
288 | int ret = 0; | |
289 | ||
290 | rq_for_each_segment(bvec, rq, iter) { | |
291 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
292 | if (ret < 0) | |
293 | break; | |
294 | cond_resched(); | |
295 | } | |
296 | ||
297 | return ret; | |
1da177e4 LT |
298 | } |
299 | ||
aa4d8616 | 300 | /* |
456be148 CH |
301 | * This is the slow, transforming version that needs to double buffer the |
302 | * data as it cannot do the transformations in place without having direct | |
303 | * access to the destination pages of the backing file. | |
1da177e4 | 304 | */ |
aa4d8616 CH |
305 | static int lo_write_transfer(struct loop_device *lo, struct request *rq, |
306 | loff_t pos) | |
1da177e4 | 307 | { |
aa4d8616 | 308 | struct bio_vec bvec, b; |
30112013 | 309 | struct req_iterator iter; |
aa4d8616 | 310 | struct page *page; |
7988613b | 311 | int ret = 0; |
1da177e4 | 312 | |
aa4d8616 CH |
313 | page = alloc_page(GFP_NOIO); |
314 | if (unlikely(!page)) | |
315 | return -ENOMEM; | |
456be148 | 316 | |
30112013 | 317 | rq_for_each_segment(bvec, rq, iter) { |
aa4d8616 CH |
318 | ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, |
319 | bvec.bv_offset, bvec.bv_len, pos >> 9); | |
320 | if (unlikely(ret)) | |
321 | break; | |
322 | ||
323 | b.bv_page = page; | |
324 | b.bv_offset = 0; | |
325 | b.bv_len = bvec.bv_len; | |
326 | ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); | |
1da177e4 LT |
327 | if (ret < 0) |
328 | break; | |
1da177e4 | 329 | } |
aa4d8616 CH |
330 | |
331 | __free_page(page); | |
1da177e4 | 332 | return ret; |
1da177e4 LT |
333 | } |
334 | ||
aa4d8616 CH |
335 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
336 | loff_t pos) | |
1da177e4 | 337 | { |
aa4d8616 CH |
338 | struct bio_vec bvec; |
339 | struct req_iterator iter; | |
340 | struct iov_iter i; | |
341 | ssize_t len; | |
1da177e4 | 342 | |
aa4d8616 CH |
343 | rq_for_each_segment(bvec, rq, iter) { |
344 | iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len); | |
345 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos); | |
346 | if (len < 0) | |
347 | return len; | |
1da177e4 | 348 | |
aa4d8616 | 349 | flush_dcache_page(bvec.bv_page); |
fd582140 | 350 | |
aa4d8616 CH |
351 | if (len != bvec.bv_len) { |
352 | struct bio *bio; | |
1da177e4 | 353 | |
aa4d8616 CH |
354 | __rq_for_each_bio(bio, rq) |
355 | zero_fill_bio(bio); | |
356 | break; | |
357 | } | |
358 | cond_resched(); | |
359 | } | |
360 | ||
361 | return 0; | |
fd582140 JA |
362 | } |
363 | ||
aa4d8616 CH |
364 | static int lo_read_transfer(struct loop_device *lo, struct request *rq, |
365 | loff_t pos) | |
1da177e4 | 366 | { |
aa4d8616 CH |
367 | struct bio_vec bvec, b; |
368 | struct req_iterator iter; | |
369 | struct iov_iter i; | |
370 | struct page *page; | |
371 | ssize_t len; | |
372 | int ret = 0; | |
1da177e4 | 373 | |
aa4d8616 CH |
374 | page = alloc_page(GFP_NOIO); |
375 | if (unlikely(!page)) | |
376 | return -ENOMEM; | |
fd582140 | 377 | |
aa4d8616 CH |
378 | rq_for_each_segment(bvec, rq, iter) { |
379 | loff_t offset = pos; | |
fd582140 | 380 | |
aa4d8616 CH |
381 | b.bv_page = page; |
382 | b.bv_offset = 0; | |
383 | b.bv_len = bvec.bv_len; | |
fd582140 | 384 | |
aa4d8616 CH |
385 | iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len); |
386 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos); | |
387 | if (len < 0) { | |
388 | ret = len; | |
389 | goto out_free_page; | |
390 | } | |
1da177e4 | 391 | |
aa4d8616 CH |
392 | ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, |
393 | bvec.bv_offset, len, offset >> 9); | |
394 | if (ret) | |
395 | goto out_free_page; | |
1da177e4 | 396 | |
aa4d8616 | 397 | flush_dcache_page(bvec.bv_page); |
306df071 | 398 | |
aa4d8616 | 399 | if (len != bvec.bv_len) { |
30112013 ML |
400 | struct bio *bio; |
401 | ||
402 | __rq_for_each_bio(bio, rq) | |
403 | zero_fill_bio(bio); | |
1da177e4 | 404 | break; |
306df071 | 405 | } |
1da177e4 | 406 | } |
aa4d8616 CH |
407 | |
408 | ret = 0; | |
409 | out_free_page: | |
410 | __free_page(page); | |
411 | return ret; | |
1da177e4 LT |
412 | } |
413 | ||
cf655d95 ML |
414 | static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos) |
415 | { | |
416 | /* | |
417 | * We use punch hole to reclaim the free space used by the | |
418 | * image a.k.a. discard. However we do not support discard if | |
419 | * encryption is enabled, because it may give an attacker | |
420 | * useful information. | |
421 | */ | |
422 | struct file *file = lo->lo_backing_file; | |
423 | int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; | |
424 | int ret; | |
425 | ||
426 | if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) { | |
427 | ret = -EOPNOTSUPP; | |
428 | goto out; | |
429 | } | |
430 | ||
431 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
432 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
433 | ret = -EIO; | |
434 | out: | |
435 | return ret; | |
436 | } | |
437 | ||
438 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
439 | { | |
440 | struct file *file = lo->lo_backing_file; | |
441 | int ret = vfs_fsync(file, 0); | |
442 | if (unlikely(ret && ret != -EINVAL)) | |
443 | ret = -EIO; | |
444 | ||
445 | return ret; | |
446 | } | |
447 | ||
30112013 | 448 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
1da177e4 LT |
449 | { |
450 | loff_t pos; | |
451 | int ret; | |
452 | ||
30112013 | 453 | pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
68db1961 | 454 | |
30112013 | 455 | if (rq->cmd_flags & REQ_WRITE) { |
cf655d95 ML |
456 | if (rq->cmd_flags & REQ_FLUSH) |
457 | ret = lo_req_flush(lo, rq); | |
af65aa8e | 458 | else if (rq->cmd_flags & REQ_DISCARD) |
cf655d95 | 459 | ret = lo_discard(lo, rq, pos); |
aa4d8616 CH |
460 | else if (lo->transfer) |
461 | ret = lo_write_transfer(lo, rq, pos); | |
af65aa8e | 462 | else |
aa4d8616 CH |
463 | ret = lo_write_simple(lo, rq, pos); |
464 | ||
465 | } else { | |
466 | if (lo->transfer) | |
467 | ret = lo_read_transfer(lo, rq, pos); | |
468 | else | |
469 | ret = lo_read_simple(lo, rq, pos); | |
470 | } | |
68db1961 | 471 | |
1da177e4 LT |
472 | return ret; |
473 | } | |
474 | ||
1da177e4 LT |
475 | struct switch_request { |
476 | struct file *file; | |
477 | struct completion wait; | |
478 | }; | |
479 | ||
2e5ab5f3 ML |
480 | static inline void loop_update_dio(struct loop_device *lo) |
481 | { | |
482 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
483 | lo->use_dio); | |
484 | } | |
485 | ||
1da177e4 | 486 | /* |
b5dd2f60 | 487 | * Do the actual switch; called from the BIO completion routine |
1da177e4 | 488 | */ |
b5dd2f60 | 489 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) |
1da177e4 | 490 | { |
b5dd2f60 ML |
491 | struct file *file = p->file; |
492 | struct file *old_file = lo->lo_backing_file; | |
493 | struct address_space *mapping; | |
35a82d1a | 494 | |
b5dd2f60 ML |
495 | /* if no new file, only flush of queued bios requested */ |
496 | if (!file) | |
497 | return; | |
1da177e4 | 498 | |
b5dd2f60 ML |
499 | mapping = file->f_mapping; |
500 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
501 | lo->lo_backing_file = file; | |
502 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? | |
503 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
504 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
505 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
2e5ab5f3 | 506 | loop_update_dio(lo); |
1da177e4 LT |
507 | } |
508 | ||
509 | /* | |
510 | * loop_switch performs the hard work of switching a backing store. | |
511 | * First it needs to flush existing IO, it does this by sending a magic | |
512 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
513 | */ | |
514 | static int loop_switch(struct loop_device *lo, struct file *file) | |
515 | { | |
516 | struct switch_request w; | |
b5dd2f60 | 517 | |
1da177e4 | 518 | w.file = file; |
b5dd2f60 ML |
519 | |
520 | /* freeze queue and wait for completion of scheduled requests */ | |
521 | blk_mq_freeze_queue(lo->lo_queue); | |
522 | ||
523 | /* do the switch action */ | |
524 | do_loop_switch(lo, &w); | |
525 | ||
526 | /* unfreeze */ | |
527 | blk_mq_unfreeze_queue(lo->lo_queue); | |
528 | ||
1da177e4 LT |
529 | return 0; |
530 | } | |
531 | ||
14f27939 MB |
532 | /* |
533 | * Helper to flush the IOs in loop, but keeping loop thread running | |
534 | */ | |
535 | static int loop_flush(struct loop_device *lo) | |
536 | { | |
14f27939 MB |
537 | return loop_switch(lo, NULL); |
538 | } | |
539 | ||
06f0e9e6 ML |
540 | static void loop_reread_partitions(struct loop_device *lo, |
541 | struct block_device *bdev) | |
542 | { | |
543 | int rc; | |
544 | ||
545 | /* | |
546 | * bd_mutex has been held already in release path, so don't | |
547 | * acquire it if this function is called in such case. | |
548 | * | |
549 | * If the reread partition isn't from release path, lo_refcnt | |
550 | * must be at least one and it can only become zero when the | |
551 | * current holder is released. | |
552 | */ | |
553 | if (!atomic_read(&lo->lo_refcnt)) | |
554 | rc = __blkdev_reread_part(bdev); | |
555 | else | |
556 | rc = blkdev_reread_part(bdev); | |
557 | if (rc) | |
558 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
559 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
560 | } | |
561 | ||
1da177e4 LT |
562 | /* |
563 | * loop_change_fd switched the backing store of a loopback device to | |
564 | * a new file. This is useful for operating system installers to free up | |
565 | * the original file and in High Availability environments to switch to | |
566 | * an alternative location for the content in case of server meltdown. | |
567 | * This can only work if the loop device is used read-only, and if the | |
568 | * new backing store is the same size and type as the old backing store. | |
569 | */ | |
bb214884 AV |
570 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
571 | unsigned int arg) | |
1da177e4 LT |
572 | { |
573 | struct file *file, *old_file; | |
574 | struct inode *inode; | |
575 | int error; | |
576 | ||
577 | error = -ENXIO; | |
578 | if (lo->lo_state != Lo_bound) | |
579 | goto out; | |
580 | ||
581 | /* the loop device has to be read-only */ | |
582 | error = -EINVAL; | |
583 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
584 | goto out; | |
585 | ||
586 | error = -EBADF; | |
587 | file = fget(arg); | |
588 | if (!file) | |
589 | goto out; | |
590 | ||
591 | inode = file->f_mapping->host; | |
592 | old_file = lo->lo_backing_file; | |
593 | ||
594 | error = -EINVAL; | |
595 | ||
596 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
597 | goto out_putf; | |
598 | ||
1da177e4 LT |
599 | /* size of the new backing store needs to be the same */ |
600 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
601 | goto out_putf; | |
602 | ||
603 | /* and ... switch */ | |
604 | error = loop_switch(lo, file); | |
605 | if (error) | |
606 | goto out_putf; | |
607 | ||
608 | fput(old_file); | |
e03c8dd1 | 609 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
06f0e9e6 | 610 | loop_reread_partitions(lo, bdev); |
1da177e4 LT |
611 | return 0; |
612 | ||
613 | out_putf: | |
614 | fput(file); | |
615 | out: | |
616 | return error; | |
617 | } | |
618 | ||
619 | static inline int is_loop_device(struct file *file) | |
620 | { | |
621 | struct inode *i = file->f_mapping->host; | |
622 | ||
623 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
624 | } | |
625 | ||
ee862730 MB |
626 | /* loop sysfs attributes */ |
627 | ||
628 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
629 | ssize_t (*callback)(struct loop_device *, char *)) | |
630 | { | |
34dd82af KS |
631 | struct gendisk *disk = dev_to_disk(dev); |
632 | struct loop_device *lo = disk->private_data; | |
ee862730 | 633 | |
34dd82af | 634 | return callback(lo, page); |
ee862730 MB |
635 | } |
636 | ||
637 | #define LOOP_ATTR_RO(_name) \ | |
638 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
639 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
640 | struct device_attribute *attr, char *b) \ | |
641 | { \ | |
642 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
643 | } \ | |
644 | static struct device_attribute loop_attr_##_name = \ | |
645 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
646 | ||
647 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
648 | { | |
649 | ssize_t ret; | |
650 | char *p = NULL; | |
651 | ||
05eb0f25 | 652 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 653 | if (lo->lo_backing_file) |
9bf39ab2 | 654 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 655 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
656 | |
657 | if (IS_ERR_OR_NULL(p)) | |
658 | ret = PTR_ERR(p); | |
659 | else { | |
660 | ret = strlen(p); | |
661 | memmove(buf, p, ret); | |
662 | buf[ret++] = '\n'; | |
663 | buf[ret] = 0; | |
664 | } | |
665 | ||
666 | return ret; | |
667 | } | |
668 | ||
669 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
670 | { | |
671 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
672 | } | |
673 | ||
674 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
675 | { | |
676 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
677 | } | |
678 | ||
679 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
680 | { | |
681 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
682 | ||
683 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
684 | } | |
685 | ||
e03c8dd1 KS |
686 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
687 | { | |
688 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
689 | ||
690 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
691 | } | |
692 | ||
2e5ab5f3 ML |
693 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
694 | { | |
695 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
696 | ||
697 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
698 | } | |
699 | ||
ee862730 MB |
700 | LOOP_ATTR_RO(backing_file); |
701 | LOOP_ATTR_RO(offset); | |
702 | LOOP_ATTR_RO(sizelimit); | |
703 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 704 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 705 | LOOP_ATTR_RO(dio); |
ee862730 MB |
706 | |
707 | static struct attribute *loop_attrs[] = { | |
708 | &loop_attr_backing_file.attr, | |
709 | &loop_attr_offset.attr, | |
710 | &loop_attr_sizelimit.attr, | |
711 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 712 | &loop_attr_partscan.attr, |
2e5ab5f3 | 713 | &loop_attr_dio.attr, |
ee862730 MB |
714 | NULL, |
715 | }; | |
716 | ||
717 | static struct attribute_group loop_attribute_group = { | |
718 | .name = "loop", | |
719 | .attrs= loop_attrs, | |
720 | }; | |
721 | ||
722 | static int loop_sysfs_init(struct loop_device *lo) | |
723 | { | |
724 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
725 | &loop_attribute_group); | |
726 | } | |
727 | ||
728 | static void loop_sysfs_exit(struct loop_device *lo) | |
729 | { | |
730 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
731 | &loop_attribute_group); | |
732 | } | |
733 | ||
dfaa2ef6 LC |
734 | static void loop_config_discard(struct loop_device *lo) |
735 | { | |
736 | struct file *file = lo->lo_backing_file; | |
737 | struct inode *inode = file->f_mapping->host; | |
738 | struct request_queue *q = lo->lo_queue; | |
739 | ||
740 | /* | |
741 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 742 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
743 | * encryption is enabled, because it may give an attacker |
744 | * useful information. | |
745 | */ | |
746 | if ((!file->f_op->fallocate) || | |
747 | lo->lo_encrypt_key_size) { | |
748 | q->limits.discard_granularity = 0; | |
749 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 750 | blk_queue_max_discard_sectors(q, 0); |
dfaa2ef6 LC |
751 | q->limits.discard_zeroes_data = 0; |
752 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
753 | return; | |
754 | } | |
755 | ||
756 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 757 | q->limits.discard_alignment = 0; |
2bb4cd5c | 758 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
dfaa2ef6 LC |
759 | q->limits.discard_zeroes_data = 1; |
760 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
761 | } | |
762 | ||
e03a3d7a ML |
763 | static void loop_unprepare_queue(struct loop_device *lo) |
764 | { | |
765 | flush_kthread_worker(&lo->worker); | |
766 | kthread_stop(lo->worker_task); | |
767 | } | |
768 | ||
769 | static int loop_prepare_queue(struct loop_device *lo) | |
770 | { | |
771 | init_kthread_worker(&lo->worker); | |
772 | lo->worker_task = kthread_run(kthread_worker_fn, | |
773 | &lo->worker, "loop%d", lo->lo_number); | |
774 | if (IS_ERR(lo->worker_task)) | |
775 | return -ENOMEM; | |
776 | set_user_nice(lo->worker_task, MIN_NICE); | |
777 | return 0; | |
778 | } | |
779 | ||
bb214884 | 780 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
781 | struct block_device *bdev, unsigned int arg) |
782 | { | |
783 | struct file *file, *f; | |
784 | struct inode *inode; | |
785 | struct address_space *mapping; | |
786 | unsigned lo_blocksize; | |
787 | int lo_flags = 0; | |
788 | int error; | |
789 | loff_t size; | |
790 | ||
791 | /* This is safe, since we have a reference from open(). */ | |
792 | __module_get(THIS_MODULE); | |
793 | ||
794 | error = -EBADF; | |
795 | file = fget(arg); | |
796 | if (!file) | |
797 | goto out; | |
798 | ||
799 | error = -EBUSY; | |
800 | if (lo->lo_state != Lo_unbound) | |
801 | goto out_putf; | |
802 | ||
803 | /* Avoid recursion */ | |
804 | f = file; | |
805 | while (is_loop_device(f)) { | |
806 | struct loop_device *l; | |
807 | ||
bb214884 | 808 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
809 | goto out_putf; |
810 | ||
811 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
812 | if (l->lo_state == Lo_unbound) { | |
813 | error = -EINVAL; | |
814 | goto out_putf; | |
815 | } | |
816 | f = l->lo_backing_file; | |
817 | } | |
818 | ||
819 | mapping = file->f_mapping; | |
820 | inode = mapping->host; | |
821 | ||
1da177e4 | 822 | error = -EINVAL; |
456be148 CH |
823 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
824 | goto out_putf; | |
1da177e4 | 825 | |
456be148 | 826 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 827 | !file->f_op->write_iter) |
456be148 | 828 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 829 | |
456be148 CH |
830 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
831 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 | 832 | |
456be148 | 833 | error = -EFBIG; |
1da177e4 | 834 | size = get_loop_size(lo, file); |
456be148 | 835 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 836 | goto out_putf; |
e03a3d7a ML |
837 | error = loop_prepare_queue(lo); |
838 | if (error) | |
f4aa4c7b | 839 | goto out_putf; |
1da177e4 | 840 | |
456be148 | 841 | error = 0; |
1da177e4 LT |
842 | |
843 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
844 | ||
2e5ab5f3 | 845 | lo->use_dio = false; |
1da177e4 LT |
846 | lo->lo_blocksize = lo_blocksize; |
847 | lo->lo_device = bdev; | |
848 | lo->lo_flags = lo_flags; | |
849 | lo->lo_backing_file = file; | |
aa4d8616 | 850 | lo->transfer = NULL; |
1da177e4 LT |
851 | lo->ioctl = NULL; |
852 | lo->lo_sizelimit = 0; | |
853 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
854 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
855 | ||
68db1961 | 856 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
4913efe4 | 857 | blk_queue_flush(lo->lo_queue, REQ_FLUSH); |
68db1961 | 858 | |
2e5ab5f3 | 859 | loop_update_dio(lo); |
73285082 | 860 | set_capacity(lo->lo_disk, size); |
1da177e4 | 861 | bd_set_size(bdev, size << 9); |
ee862730 | 862 | loop_sysfs_init(lo); |
c3473c63 DZ |
863 | /* let user-space know about the new size */ |
864 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 LT |
865 | |
866 | set_blocksize(bdev, lo_blocksize); | |
867 | ||
6c997918 | 868 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
869 | if (part_shift) |
870 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
871 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
06f0e9e6 | 872 | loop_reread_partitions(lo, bdev); |
c1681bf8 AP |
873 | |
874 | /* Grab the block_device to prevent its destruction after we | |
875 | * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). | |
876 | */ | |
877 | bdgrab(bdev); | |
1da177e4 LT |
878 | return 0; |
879 | ||
880 | out_putf: | |
881 | fput(file); | |
882 | out: | |
883 | /* This is safe: open() is still holding a reference. */ | |
884 | module_put(THIS_MODULE); | |
885 | return error; | |
886 | } | |
887 | ||
888 | static int | |
889 | loop_release_xfer(struct loop_device *lo) | |
890 | { | |
891 | int err = 0; | |
892 | struct loop_func_table *xfer = lo->lo_encryption; | |
893 | ||
894 | if (xfer) { | |
895 | if (xfer->release) | |
896 | err = xfer->release(lo); | |
897 | lo->transfer = NULL; | |
898 | lo->lo_encryption = NULL; | |
899 | module_put(xfer->owner); | |
900 | } | |
901 | return err; | |
902 | } | |
903 | ||
904 | static int | |
905 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
906 | const struct loop_info64 *i) | |
907 | { | |
908 | int err = 0; | |
909 | ||
910 | if (xfer) { | |
911 | struct module *owner = xfer->owner; | |
912 | ||
913 | if (!try_module_get(owner)) | |
914 | return -EINVAL; | |
915 | if (xfer->init) | |
916 | err = xfer->init(lo, i); | |
917 | if (err) | |
918 | module_put(owner); | |
919 | else | |
920 | lo->lo_encryption = xfer; | |
921 | } | |
922 | return err; | |
923 | } | |
924 | ||
4c823cc3 | 925 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
926 | { |
927 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 928 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 929 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
930 | |
931 | if (lo->lo_state != Lo_bound) | |
932 | return -ENXIO; | |
933 | ||
a1ecac3b DC |
934 | /* |
935 | * If we've explicitly asked to tear down the loop device, | |
936 | * and it has an elevated reference count, set it for auto-teardown when | |
937 | * the last reference goes away. This stops $!~#$@ udev from | |
938 | * preventing teardown because it decided that it needs to run blkid on | |
939 | * the loopback device whenever they appear. xfstests is notorious for | |
940 | * failing tests because blkid via udev races with a losetup | |
941 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
942 | * command to fail with EBUSY. | |
943 | */ | |
f8933667 | 944 | if (atomic_read(&lo->lo_refcnt) > 1) { |
a1ecac3b DC |
945 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
946 | mutex_unlock(&lo->lo_ctl_mutex); | |
947 | return 0; | |
948 | } | |
1da177e4 LT |
949 | |
950 | if (filp == NULL) | |
951 | return -EINVAL; | |
952 | ||
f8933667 ML |
953 | /* freeze request queue during the transition */ |
954 | blk_mq_freeze_queue(lo->lo_queue); | |
955 | ||
1da177e4 LT |
956 | spin_lock_irq(&lo->lo_lock); |
957 | lo->lo_state = Lo_rundown; | |
1da177e4 | 958 | lo->lo_backing_file = NULL; |
05eb0f25 | 959 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
960 | |
961 | loop_release_xfer(lo); | |
962 | lo->transfer = NULL; | |
963 | lo->ioctl = NULL; | |
964 | lo->lo_device = NULL; | |
965 | lo->lo_encryption = NULL; | |
966 | lo->lo_offset = 0; | |
967 | lo->lo_sizelimit = 0; | |
968 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
969 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
970 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
971 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
c1681bf8 AP |
972 | if (bdev) { |
973 | bdput(bdev); | |
bb214884 | 974 | invalidate_bdev(bdev); |
c1681bf8 | 975 | } |
73285082 | 976 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 977 | loop_sysfs_exit(lo); |
c3473c63 | 978 | if (bdev) { |
bb214884 | 979 | bd_set_size(bdev, 0); |
c3473c63 DZ |
980 | /* let user-space know about this change */ |
981 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
982 | } | |
1da177e4 LT |
983 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
984 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
985 | /* This is safe: open() is still holding a reference. */ |
986 | module_put(THIS_MODULE); | |
f8933667 ML |
987 | blk_mq_unfreeze_queue(lo->lo_queue); |
988 | ||
c2fccc1c | 989 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
06f0e9e6 | 990 | loop_reread_partitions(lo, bdev); |
e03c8dd1 KS |
991 | lo->lo_flags = 0; |
992 | if (!part_shift) | |
993 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
e03a3d7a | 994 | loop_unprepare_queue(lo); |
f028f3b2 NK |
995 | mutex_unlock(&lo->lo_ctl_mutex); |
996 | /* | |
997 | * Need not hold lo_ctl_mutex to fput backing file. | |
998 | * Calling fput holding lo_ctl_mutex triggers a circular | |
999 | * lock dependency possibility warning as fput can take | |
1000 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1001 | */ | |
1002 | fput(filp); | |
1da177e4 LT |
1003 | return 0; |
1004 | } | |
1005 | ||
1006 | static int | |
1007 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1008 | { | |
1009 | int err; | |
1010 | struct loop_func_table *xfer; | |
e4849737 | 1011 | kuid_t uid = current_uid(); |
1da177e4 | 1012 | |
b0fafa81 | 1013 | if (lo->lo_encrypt_key_size && |
e4849737 | 1014 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1015 | !capable(CAP_SYS_ADMIN)) |
1016 | return -EPERM; | |
1017 | if (lo->lo_state != Lo_bound) | |
1018 | return -ENXIO; | |
1019 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1020 | return -EINVAL; | |
1021 | ||
1022 | err = loop_release_xfer(lo); | |
1023 | if (err) | |
1024 | return err; | |
1025 | ||
1026 | if (info->lo_encrypt_type) { | |
1027 | unsigned int type = info->lo_encrypt_type; | |
1028 | ||
1029 | if (type >= MAX_LO_CRYPT) | |
1030 | return -EINVAL; | |
1031 | xfer = xfer_funcs[type]; | |
1032 | if (xfer == NULL) | |
1033 | return -EINVAL; | |
1034 | } else | |
1035 | xfer = NULL; | |
1036 | ||
1037 | err = loop_init_xfer(lo, xfer, info); | |
1038 | if (err) | |
1039 | return err; | |
1040 | ||
1041 | if (lo->lo_offset != info->lo_offset || | |
7b0576a3 | 1042 | lo->lo_sizelimit != info->lo_sizelimit) |
7035b5df | 1043 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) |
1da177e4 | 1044 | return -EFBIG; |
541c742a | 1045 | |
dfaa2ef6 | 1046 | loop_config_discard(lo); |
1da177e4 LT |
1047 | |
1048 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1049 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1050 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1051 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1052 | ||
1053 | if (!xfer) | |
1054 | xfer = &none_funcs; | |
1055 | lo->transfer = xfer->transfer; | |
1056 | lo->ioctl = xfer->ioctl; | |
1057 | ||
96c58655 DW |
1058 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1059 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1060 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1061 | ||
e03c8dd1 KS |
1062 | if ((info->lo_flags & LO_FLAGS_PARTSCAN) && |
1063 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1064 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1065 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
06f0e9e6 | 1066 | loop_reread_partitions(lo, lo->lo_device); |
e03c8dd1 KS |
1067 | } |
1068 | ||
1da177e4 LT |
1069 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1070 | lo->lo_init[0] = info->lo_init[0]; | |
1071 | lo->lo_init[1] = info->lo_init[1]; | |
1072 | if (info->lo_encrypt_key_size) { | |
1073 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1074 | info->lo_encrypt_key_size); | |
b0fafa81 | 1075 | lo->lo_key_owner = uid; |
aa4d8616 | 1076 | } |
1da177e4 | 1077 | |
2e5ab5f3 ML |
1078 | /* update dio if lo_offset or transfer is changed */ |
1079 | __loop_update_dio(lo, lo->use_dio); | |
1080 | ||
1da177e4 LT |
1081 | return 0; |
1082 | } | |
1083 | ||
1084 | static int | |
1085 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1086 | { | |
1087 | struct file *file = lo->lo_backing_file; | |
1088 | struct kstat stat; | |
1089 | int error; | |
1090 | ||
1091 | if (lo->lo_state != Lo_bound) | |
1092 | return -ENXIO; | |
3dadecce | 1093 | error = vfs_getattr(&file->f_path, &stat); |
1da177e4 LT |
1094 | if (error) |
1095 | return error; | |
1096 | memset(info, 0, sizeof(*info)); | |
1097 | info->lo_number = lo->lo_number; | |
1098 | info->lo_device = huge_encode_dev(stat.dev); | |
1099 | info->lo_inode = stat.ino; | |
1100 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1101 | info->lo_offset = lo->lo_offset; | |
1102 | info->lo_sizelimit = lo->lo_sizelimit; | |
1103 | info->lo_flags = lo->lo_flags; | |
1104 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1105 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1106 | info->lo_encrypt_type = | |
1107 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1108 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1109 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1110 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1111 | lo->lo_encrypt_key_size); | |
1112 | } | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | static void | |
1117 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1118 | { | |
1119 | memset(info64, 0, sizeof(*info64)); | |
1120 | info64->lo_number = info->lo_number; | |
1121 | info64->lo_device = info->lo_device; | |
1122 | info64->lo_inode = info->lo_inode; | |
1123 | info64->lo_rdevice = info->lo_rdevice; | |
1124 | info64->lo_offset = info->lo_offset; | |
1125 | info64->lo_sizelimit = 0; | |
1126 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1127 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1128 | info64->lo_flags = info->lo_flags; | |
1129 | info64->lo_init[0] = info->lo_init[0]; | |
1130 | info64->lo_init[1] = info->lo_init[1]; | |
1131 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1132 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1133 | else | |
1134 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1135 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1136 | } | |
1137 | ||
1138 | static int | |
1139 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1140 | { | |
1141 | memset(info, 0, sizeof(*info)); | |
1142 | info->lo_number = info64->lo_number; | |
1143 | info->lo_device = info64->lo_device; | |
1144 | info->lo_inode = info64->lo_inode; | |
1145 | info->lo_rdevice = info64->lo_rdevice; | |
1146 | info->lo_offset = info64->lo_offset; | |
1147 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1148 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1149 | info->lo_flags = info64->lo_flags; | |
1150 | info->lo_init[0] = info64->lo_init[0]; | |
1151 | info->lo_init[1] = info64->lo_init[1]; | |
1152 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1153 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1154 | else | |
1155 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1156 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1157 | ||
1158 | /* error in case values were truncated */ | |
1159 | if (info->lo_device != info64->lo_device || | |
1160 | info->lo_rdevice != info64->lo_rdevice || | |
1161 | info->lo_inode != info64->lo_inode || | |
1162 | info->lo_offset != info64->lo_offset) | |
1163 | return -EOVERFLOW; | |
1164 | ||
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | static int | |
1169 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1170 | { | |
1171 | struct loop_info info; | |
1172 | struct loop_info64 info64; | |
1173 | ||
1174 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1175 | return -EFAULT; | |
1176 | loop_info64_from_old(&info, &info64); | |
1177 | return loop_set_status(lo, &info64); | |
1178 | } | |
1179 | ||
1180 | static int | |
1181 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1182 | { | |
1183 | struct loop_info64 info64; | |
1184 | ||
1185 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1186 | return -EFAULT; | |
1187 | return loop_set_status(lo, &info64); | |
1188 | } | |
1189 | ||
1190 | static int | |
1191 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1192 | struct loop_info info; | |
1193 | struct loop_info64 info64; | |
1194 | int err = 0; | |
1195 | ||
1196 | if (!arg) | |
1197 | err = -EINVAL; | |
1198 | if (!err) | |
1199 | err = loop_get_status(lo, &info64); | |
1200 | if (!err) | |
1201 | err = loop_info64_to_old(&info64, &info); | |
1202 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1203 | err = -EFAULT; | |
1204 | ||
1205 | return err; | |
1206 | } | |
1207 | ||
1208 | static int | |
1209 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1210 | struct loop_info64 info64; | |
1211 | int err = 0; | |
1212 | ||
1213 | if (!arg) | |
1214 | err = -EINVAL; | |
1215 | if (!err) | |
1216 | err = loop_get_status(lo, &info64); | |
1217 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1218 | err = -EFAULT; | |
1219 | ||
1220 | return err; | |
1221 | } | |
1222 | ||
53d66608 O |
1223 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) |
1224 | { | |
53d66608 | 1225 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1226 | return -ENXIO; |
53d66608 | 1227 | |
7b0576a3 | 1228 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1229 | } |
1230 | ||
ab1cb278 ML |
1231 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1232 | { | |
1233 | int error = -ENXIO; | |
1234 | if (lo->lo_state != Lo_bound) | |
1235 | goto out; | |
1236 | ||
1237 | __loop_update_dio(lo, !!arg); | |
1238 | if (lo->use_dio == !!arg) | |
1239 | return 0; | |
1240 | error = -EINVAL; | |
1241 | out: | |
1242 | return error; | |
1243 | } | |
1244 | ||
bb214884 | 1245 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1246 | unsigned int cmd, unsigned long arg) |
1247 | { | |
bb214884 | 1248 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1249 | int err; |
1250 | ||
f028f3b2 | 1251 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1252 | switch (cmd) { |
1253 | case LOOP_SET_FD: | |
bb214884 | 1254 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1255 | break; |
1256 | case LOOP_CHANGE_FD: | |
bb214884 | 1257 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1258 | break; |
1259 | case LOOP_CLR_FD: | |
f028f3b2 | 1260 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1261 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1262 | if (!err) |
1263 | goto out_unlocked; | |
1da177e4 LT |
1264 | break; |
1265 | case LOOP_SET_STATUS: | |
7035b5df DM |
1266 | err = -EPERM; |
1267 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1268 | err = loop_set_status_old(lo, | |
1269 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1270 | break; |
1271 | case LOOP_GET_STATUS: | |
1272 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1273 | break; | |
1274 | case LOOP_SET_STATUS64: | |
7035b5df DM |
1275 | err = -EPERM; |
1276 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1277 | err = loop_set_status64(lo, | |
1278 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1279 | break; |
1280 | case LOOP_GET_STATUS64: | |
1281 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1282 | break; | |
53d66608 O |
1283 | case LOOP_SET_CAPACITY: |
1284 | err = -EPERM; | |
1285 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1286 | err = loop_set_capacity(lo, bdev); | |
1287 | break; | |
ab1cb278 ML |
1288 | case LOOP_SET_DIRECT_IO: |
1289 | err = -EPERM; | |
1290 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1291 | err = loop_set_dio(lo, arg); | |
1292 | break; | |
1da177e4 LT |
1293 | default: |
1294 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1295 | } | |
f85221dd | 1296 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1297 | |
1298 | out_unlocked: | |
1da177e4 LT |
1299 | return err; |
1300 | } | |
1301 | ||
863d5b82 DH |
1302 | #ifdef CONFIG_COMPAT |
1303 | struct compat_loop_info { | |
1304 | compat_int_t lo_number; /* ioctl r/o */ | |
1305 | compat_dev_t lo_device; /* ioctl r/o */ | |
1306 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1307 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1308 | compat_int_t lo_offset; | |
1309 | compat_int_t lo_encrypt_type; | |
1310 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1311 | compat_int_t lo_flags; /* ioctl r/o */ | |
1312 | char lo_name[LO_NAME_SIZE]; | |
1313 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1314 | compat_ulong_t lo_init[2]; | |
1315 | char reserved[4]; | |
1316 | }; | |
1317 | ||
1318 | /* | |
1319 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1320 | * - noinlined to reduce stack space usage in main part of driver | |
1321 | */ | |
1322 | static noinline int | |
ba674cfc | 1323 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1324 | struct loop_info64 *info64) |
1325 | { | |
1326 | struct compat_loop_info info; | |
1327 | ||
1328 | if (copy_from_user(&info, arg, sizeof(info))) | |
1329 | return -EFAULT; | |
1330 | ||
1331 | memset(info64, 0, sizeof(*info64)); | |
1332 | info64->lo_number = info.lo_number; | |
1333 | info64->lo_device = info.lo_device; | |
1334 | info64->lo_inode = info.lo_inode; | |
1335 | info64->lo_rdevice = info.lo_rdevice; | |
1336 | info64->lo_offset = info.lo_offset; | |
1337 | info64->lo_sizelimit = 0; | |
1338 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1339 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1340 | info64->lo_flags = info.lo_flags; | |
1341 | info64->lo_init[0] = info.lo_init[0]; | |
1342 | info64->lo_init[1] = info.lo_init[1]; | |
1343 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1344 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1345 | else | |
1346 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1347 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1353 | * - noinlined to reduce stack space usage in main part of driver | |
1354 | */ | |
1355 | static noinline int | |
1356 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1357 | struct compat_loop_info __user *arg) | |
1358 | { | |
1359 | struct compat_loop_info info; | |
1360 | ||
1361 | memset(&info, 0, sizeof(info)); | |
1362 | info.lo_number = info64->lo_number; | |
1363 | info.lo_device = info64->lo_device; | |
1364 | info.lo_inode = info64->lo_inode; | |
1365 | info.lo_rdevice = info64->lo_rdevice; | |
1366 | info.lo_offset = info64->lo_offset; | |
1367 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1368 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1369 | info.lo_flags = info64->lo_flags; | |
1370 | info.lo_init[0] = info64->lo_init[0]; | |
1371 | info.lo_init[1] = info64->lo_init[1]; | |
1372 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1373 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1374 | else | |
1375 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1376 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1377 | ||
1378 | /* error in case values were truncated */ | |
1379 | if (info.lo_device != info64->lo_device || | |
1380 | info.lo_rdevice != info64->lo_rdevice || | |
1381 | info.lo_inode != info64->lo_inode || | |
1382 | info.lo_offset != info64->lo_offset || | |
1383 | info.lo_init[0] != info64->lo_init[0] || | |
1384 | info.lo_init[1] != info64->lo_init[1]) | |
1385 | return -EOVERFLOW; | |
1386 | ||
1387 | if (copy_to_user(arg, &info, sizeof(info))) | |
1388 | return -EFAULT; | |
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | static int | |
1393 | loop_set_status_compat(struct loop_device *lo, | |
1394 | const struct compat_loop_info __user *arg) | |
1395 | { | |
1396 | struct loop_info64 info64; | |
1397 | int ret; | |
1398 | ||
1399 | ret = loop_info64_from_compat(arg, &info64); | |
1400 | if (ret < 0) | |
1401 | return ret; | |
1402 | return loop_set_status(lo, &info64); | |
1403 | } | |
1404 | ||
1405 | static int | |
1406 | loop_get_status_compat(struct loop_device *lo, | |
1407 | struct compat_loop_info __user *arg) | |
1408 | { | |
1409 | struct loop_info64 info64; | |
1410 | int err = 0; | |
1411 | ||
1412 | if (!arg) | |
1413 | err = -EINVAL; | |
1414 | if (!err) | |
1415 | err = loop_get_status(lo, &info64); | |
1416 | if (!err) | |
1417 | err = loop_info64_to_compat(&info64, arg); | |
1418 | return err; | |
1419 | } | |
1420 | ||
bb214884 AV |
1421 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1422 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1423 | { |
bb214884 | 1424 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1425 | int err; |
1426 | ||
863d5b82 DH |
1427 | switch(cmd) { |
1428 | case LOOP_SET_STATUS: | |
1429 | mutex_lock(&lo->lo_ctl_mutex); | |
1430 | err = loop_set_status_compat( | |
1431 | lo, (const struct compat_loop_info __user *) arg); | |
1432 | mutex_unlock(&lo->lo_ctl_mutex); | |
1433 | break; | |
1434 | case LOOP_GET_STATUS: | |
1435 | mutex_lock(&lo->lo_ctl_mutex); | |
1436 | err = loop_get_status_compat( | |
1437 | lo, (struct compat_loop_info __user *) arg); | |
1438 | mutex_unlock(&lo->lo_ctl_mutex); | |
1439 | break; | |
53d66608 | 1440 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1441 | case LOOP_CLR_FD: |
1442 | case LOOP_GET_STATUS64: | |
1443 | case LOOP_SET_STATUS64: | |
1444 | arg = (unsigned long) compat_ptr(arg); | |
1445 | case LOOP_SET_FD: | |
1446 | case LOOP_CHANGE_FD: | |
bb214884 | 1447 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1448 | break; |
1449 | default: | |
1450 | err = -ENOIOCTLCMD; | |
1451 | break; | |
1452 | } | |
863d5b82 DH |
1453 | return err; |
1454 | } | |
1455 | #endif | |
1456 | ||
bb214884 | 1457 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1458 | { |
770fe30a KS |
1459 | struct loop_device *lo; |
1460 | int err = 0; | |
1461 | ||
1462 | mutex_lock(&loop_index_mutex); | |
1463 | lo = bdev->bd_disk->private_data; | |
1464 | if (!lo) { | |
1465 | err = -ENXIO; | |
1466 | goto out; | |
1467 | } | |
1da177e4 | 1468 | |
f8933667 | 1469 | atomic_inc(&lo->lo_refcnt); |
770fe30a KS |
1470 | out: |
1471 | mutex_unlock(&loop_index_mutex); | |
1472 | return err; | |
1da177e4 LT |
1473 | } |
1474 | ||
db2a144b | 1475 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1476 | { |
bb214884 | 1477 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1478 | int err; |
1da177e4 | 1479 | |
f8933667 ML |
1480 | if (atomic_dec_return(&lo->lo_refcnt)) |
1481 | return; | |
14f27939 | 1482 | |
f8933667 | 1483 | mutex_lock(&lo->lo_ctl_mutex); |
14f27939 MB |
1484 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { |
1485 | /* | |
1486 | * In autoclear mode, stop the loop thread | |
1487 | * and remove configuration after last close. | |
1488 | */ | |
4c823cc3 | 1489 | err = loop_clr_fd(lo); |
ffcd7dca | 1490 | if (!err) |
db2a144b | 1491 | return; |
14f27939 MB |
1492 | } else { |
1493 | /* | |
1494 | * Otherwise keep thread (if running) and config, | |
1495 | * but flush possible ongoing bios in thread. | |
1496 | */ | |
1497 | loop_flush(lo); | |
1498 | } | |
96c58655 | 1499 | |
f85221dd | 1500 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1501 | } |
1502 | ||
83d5cde4 | 1503 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1504 | .owner = THIS_MODULE, |
bb214884 AV |
1505 | .open = lo_open, |
1506 | .release = lo_release, | |
1507 | .ioctl = lo_ioctl, | |
863d5b82 | 1508 | #ifdef CONFIG_COMPAT |
bb214884 | 1509 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1510 | #endif |
1da177e4 LT |
1511 | }; |
1512 | ||
1513 | /* | |
1514 | * And now the modules code and kernel interface. | |
1515 | */ | |
73285082 | 1516 | static int max_loop; |
ac04fee0 | 1517 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1518 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1519 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1520 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1521 | MODULE_LICENSE("GPL"); |
1522 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1523 | ||
1524 | int loop_register_transfer(struct loop_func_table *funcs) | |
1525 | { | |
1526 | unsigned int n = funcs->number; | |
1527 | ||
1528 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1529 | return -EINVAL; | |
1530 | xfer_funcs[n] = funcs; | |
1531 | return 0; | |
1532 | } | |
1533 | ||
34dd82af KS |
1534 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1535 | { | |
1536 | struct loop_device *lo = ptr; | |
1537 | struct loop_func_table *xfer = data; | |
1538 | ||
1539 | mutex_lock(&lo->lo_ctl_mutex); | |
1540 | if (lo->lo_encryption == xfer) | |
1541 | loop_release_xfer(lo); | |
1542 | mutex_unlock(&lo->lo_ctl_mutex); | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1da177e4 LT |
1546 | int loop_unregister_transfer(int number) |
1547 | { | |
1548 | unsigned int n = number; | |
1da177e4 LT |
1549 | struct loop_func_table *xfer; |
1550 | ||
1551 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1552 | return -EINVAL; | |
1553 | ||
1554 | xfer_funcs[n] = NULL; | |
34dd82af | 1555 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1556 | return 0; |
1557 | } | |
1558 | ||
1559 | EXPORT_SYMBOL(loop_register_transfer); | |
1560 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1561 | ||
b5dd2f60 ML |
1562 | static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
1563 | const struct blk_mq_queue_data *bd) | |
1564 | { | |
1565 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
f4aa4c7b | 1566 | struct loop_device *lo = cmd->rq->q->queuedata; |
b5dd2f60 ML |
1567 | |
1568 | blk_mq_start_request(bd->rq); | |
1569 | ||
f4aa4c7b ML |
1570 | if (lo->lo_state != Lo_bound) |
1571 | return -EIO; | |
1572 | ||
e03a3d7a | 1573 | queue_kthread_work(&lo->worker, &cmd->work); |
b5dd2f60 ML |
1574 | |
1575 | return BLK_MQ_RQ_QUEUE_OK; | |
1576 | } | |
1577 | ||
1578 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1579 | { | |
1580 | const bool write = cmd->rq->cmd_flags & REQ_WRITE; | |
1581 | struct loop_device *lo = cmd->rq->q->queuedata; | |
1582 | int ret = -EIO; | |
b5dd2f60 | 1583 | |
b5dd2f60 ML |
1584 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) |
1585 | goto failed; | |
1586 | ||
30112013 | 1587 | ret = do_req_filebacked(lo, cmd->rq); |
b5dd2f60 ML |
1588 | |
1589 | failed: | |
1590 | if (ret) | |
1591 | cmd->rq->errors = -EIO; | |
1592 | blk_mq_complete_request(cmd->rq); | |
1593 | } | |
1594 | ||
e03a3d7a | 1595 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1596 | { |
1597 | struct loop_cmd *cmd = | |
e03a3d7a | 1598 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1599 | |
1600 | loop_handle_cmd(cmd); | |
1601 | } | |
1602 | ||
1603 | static int loop_init_request(void *data, struct request *rq, | |
1604 | unsigned int hctx_idx, unsigned int request_idx, | |
1605 | unsigned int numa_node) | |
1606 | { | |
1607 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1608 | ||
1609 | cmd->rq = rq; | |
e03a3d7a | 1610 | init_kthread_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1611 | |
1612 | return 0; | |
1613 | } | |
1614 | ||
1615 | static struct blk_mq_ops loop_mq_ops = { | |
1616 | .queue_rq = loop_queue_rq, | |
1617 | .map_queue = blk_mq_map_queue, | |
1618 | .init_request = loop_init_request, | |
1619 | }; | |
1620 | ||
34dd82af | 1621 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1622 | { |
1623 | struct loop_device *lo; | |
1624 | struct gendisk *disk; | |
34dd82af | 1625 | int err; |
73285082 | 1626 | |
68d740d7 | 1627 | err = -ENOMEM; |
73285082 | 1628 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1629 | if (!lo) |
73285082 | 1630 | goto out; |
34dd82af | 1631 | |
ef7e7c82 MP |
1632 | lo->lo_state = Lo_unbound; |
1633 | ||
c718aa65 | 1634 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1635 | if (i >= 0) { |
c718aa65 TH |
1636 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1637 | if (err == -ENOSPC) | |
34dd82af | 1638 | err = -EEXIST; |
34dd82af | 1639 | } else { |
c718aa65 | 1640 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1641 | } |
1642 | if (err < 0) | |
1643 | goto out_free_dev; | |
c718aa65 | 1644 | i = err; |
73285082 | 1645 | |
183cfb57 | 1646 | err = -ENOMEM; |
b5dd2f60 ML |
1647 | lo->tag_set.ops = &loop_mq_ops; |
1648 | lo->tag_set.nr_hw_queues = 1; | |
1649 | lo->tag_set.queue_depth = 128; | |
1650 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1651 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
1652 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
1653 | lo->tag_set.driver_data = lo; | |
1654 | ||
1655 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1656 | if (err) | |
3ec981e3 | 1657 | goto out_free_idr; |
73285082 | 1658 | |
b5dd2f60 ML |
1659 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
1660 | if (IS_ERR_OR_NULL(lo->lo_queue)) { | |
1661 | err = PTR_ERR(lo->lo_queue); | |
1662 | goto out_cleanup_tags; | |
1663 | } | |
ef7e7c82 MP |
1664 | lo->lo_queue->queuedata = lo; |
1665 | ||
5b5e20f4 ML |
1666 | /* |
1667 | * It doesn't make sense to enable merge because the I/O | |
1668 | * submitted to backing file is handled page by page. | |
1669 | */ | |
1670 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
1671 | ||
476a4813 | 1672 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1673 | if (!disk) |
1674 | goto out_free_queue; | |
1675 | ||
e03c8dd1 KS |
1676 | /* |
1677 | * Disable partition scanning by default. The in-kernel partition | |
1678 | * scanning can be requested individually per-device during its | |
1679 | * setup. Userspace can always add and remove partitions from all | |
1680 | * devices. The needed partition minors are allocated from the | |
1681 | * extended minor space, the main loop device numbers will continue | |
1682 | * to match the loop minors, regardless of the number of partitions | |
1683 | * used. | |
1684 | * | |
1685 | * If max_part is given, partition scanning is globally enabled for | |
1686 | * all loop devices. The minors for the main loop devices will be | |
1687 | * multiples of max_part. | |
1688 | * | |
1689 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1690 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1691 | * complicated, are too static, inflexible and may surprise | |
1692 | * userspace tools. Parameters like this in general should be avoided. | |
1693 | */ | |
1694 | if (!part_shift) | |
1695 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1696 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 | 1697 | mutex_init(&lo->lo_ctl_mutex); |
f8933667 | 1698 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 1699 | lo->lo_number = i; |
73285082 KC |
1700 | spin_lock_init(&lo->lo_lock); |
1701 | disk->major = LOOP_MAJOR; | |
476a4813 | 1702 | disk->first_minor = i << part_shift; |
73285082 KC |
1703 | disk->fops = &lo_fops; |
1704 | disk->private_data = lo; | |
1705 | disk->queue = lo->lo_queue; | |
1706 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1707 | add_disk(disk); |
1708 | *l = lo; | |
1709 | return lo->lo_number; | |
73285082 KC |
1710 | |
1711 | out_free_queue: | |
1712 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
1713 | out_cleanup_tags: |
1714 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
1715 | out_free_idr: |
1716 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
1717 | out_free_dev: |
1718 | kfree(lo); | |
1719 | out: | |
34dd82af | 1720 | return err; |
73285082 KC |
1721 | } |
1722 | ||
34dd82af | 1723 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1724 | { |
73285082 | 1725 | blk_cleanup_queue(lo->lo_queue); |
6cd18e71 | 1726 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 1727 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 1728 | put_disk(lo->lo_disk); |
73285082 KC |
1729 | kfree(lo); |
1730 | } | |
1da177e4 | 1731 | |
770fe30a KS |
1732 | static int find_free_cb(int id, void *ptr, void *data) |
1733 | { | |
1734 | struct loop_device *lo = ptr; | |
1735 | struct loop_device **l = data; | |
1736 | ||
1737 | if (lo->lo_state == Lo_unbound) { | |
1738 | *l = lo; | |
1739 | return 1; | |
1740 | } | |
1741 | return 0; | |
1742 | } | |
1743 | ||
34dd82af | 1744 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1745 | { |
1746 | struct loop_device *lo; | |
34dd82af | 1747 | int ret = -ENODEV; |
a47653fc | 1748 | |
770fe30a KS |
1749 | if (i < 0) { |
1750 | int err; | |
1751 | ||
1752 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1753 | if (err == 1) { | |
1754 | *l = lo; | |
1755 | ret = lo->lo_number; | |
1756 | } | |
1757 | goto out; | |
a47653fc KC |
1758 | } |
1759 | ||
770fe30a | 1760 | /* lookup and return a specific i */ |
34dd82af | 1761 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1762 | if (lo) { |
34dd82af KS |
1763 | *l = lo; |
1764 | ret = lo->lo_number; | |
a47653fc | 1765 | } |
770fe30a | 1766 | out: |
34dd82af | 1767 | return ret; |
a47653fc KC |
1768 | } |
1769 | ||
73285082 KC |
1770 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1771 | { | |
705962cc | 1772 | struct loop_device *lo; |
07002e99 | 1773 | struct kobject *kobj; |
34dd82af | 1774 | int err; |
73285082 | 1775 | |
34dd82af KS |
1776 | mutex_lock(&loop_index_mutex); |
1777 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1778 | if (err < 0) | |
1779 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
1780 | if (err < 0) | |
a207f593 | 1781 | kobj = NULL; |
34dd82af KS |
1782 | else |
1783 | kobj = get_disk(lo->lo_disk); | |
1784 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
1785 | |
1786 | *part = 0; | |
07002e99 | 1787 | return kobj; |
73285082 KC |
1788 | } |
1789 | ||
770fe30a KS |
1790 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
1791 | unsigned long parm) | |
1792 | { | |
1793 | struct loop_device *lo; | |
1794 | int ret = -ENOSYS; | |
1795 | ||
1796 | mutex_lock(&loop_index_mutex); | |
1797 | switch (cmd) { | |
1798 | case LOOP_CTL_ADD: | |
1799 | ret = loop_lookup(&lo, parm); | |
1800 | if (ret >= 0) { | |
1801 | ret = -EEXIST; | |
1802 | break; | |
1803 | } | |
1804 | ret = loop_add(&lo, parm); | |
1805 | break; | |
1806 | case LOOP_CTL_REMOVE: | |
1807 | ret = loop_lookup(&lo, parm); | |
1808 | if (ret < 0) | |
1809 | break; | |
1810 | mutex_lock(&lo->lo_ctl_mutex); | |
1811 | if (lo->lo_state != Lo_unbound) { | |
1812 | ret = -EBUSY; | |
1813 | mutex_unlock(&lo->lo_ctl_mutex); | |
1814 | break; | |
1815 | } | |
f8933667 | 1816 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a KS |
1817 | ret = -EBUSY; |
1818 | mutex_unlock(&lo->lo_ctl_mutex); | |
1819 | break; | |
1820 | } | |
1821 | lo->lo_disk->private_data = NULL; | |
1822 | mutex_unlock(&lo->lo_ctl_mutex); | |
1823 | idr_remove(&loop_index_idr, lo->lo_number); | |
1824 | loop_remove(lo); | |
1825 | break; | |
1826 | case LOOP_CTL_GET_FREE: | |
1827 | ret = loop_lookup(&lo, -1); | |
1828 | if (ret >= 0) | |
1829 | break; | |
1830 | ret = loop_add(&lo, -1); | |
1831 | } | |
1832 | mutex_unlock(&loop_index_mutex); | |
1833 | ||
1834 | return ret; | |
1835 | } | |
1836 | ||
1837 | static const struct file_operations loop_ctl_fops = { | |
1838 | .open = nonseekable_open, | |
1839 | .unlocked_ioctl = loop_control_ioctl, | |
1840 | .compat_ioctl = loop_control_ioctl, | |
1841 | .owner = THIS_MODULE, | |
1842 | .llseek = noop_llseek, | |
1843 | }; | |
1844 | ||
1845 | static struct miscdevice loop_misc = { | |
1846 | .minor = LOOP_CTRL_MINOR, | |
1847 | .name = "loop-control", | |
1848 | .fops = &loop_ctl_fops, | |
1849 | }; | |
1850 | ||
1851 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
1852 | MODULE_ALIAS("devname:loop-control"); | |
1853 | ||
73285082 KC |
1854 | static int __init loop_init(void) |
1855 | { | |
a47653fc KC |
1856 | int i, nr; |
1857 | unsigned long range; | |
34dd82af | 1858 | struct loop_device *lo; |
770fe30a | 1859 | int err; |
a47653fc | 1860 | |
770fe30a KS |
1861 | err = misc_register(&loop_misc); |
1862 | if (err < 0) | |
1863 | return err; | |
476a4813 LV |
1864 | |
1865 | part_shift = 0; | |
ac04fee0 | 1866 | if (max_part > 0) { |
476a4813 LV |
1867 | part_shift = fls(max_part); |
1868 | ||
ac04fee0 NK |
1869 | /* |
1870 | * Adjust max_part according to part_shift as it is exported | |
1871 | * to user space so that user can decide correct minor number | |
1872 | * if [s]he want to create more devices. | |
1873 | * | |
1874 | * Note that -1 is required because partition 0 is reserved | |
1875 | * for the whole disk. | |
1876 | */ | |
1877 | max_part = (1UL << part_shift) - 1; | |
1878 | } | |
1879 | ||
b1a66504 GC |
1880 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
1881 | err = -EINVAL; | |
1882 | goto misc_out; | |
1883 | } | |
78f4bb36 | 1884 | |
b1a66504 GC |
1885 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
1886 | err = -EINVAL; | |
1887 | goto misc_out; | |
1888 | } | |
1da177e4 | 1889 | |
d134b00b KS |
1890 | /* |
1891 | * If max_loop is specified, create that many devices upfront. | |
1892 | * This also becomes a hard limit. If max_loop is not specified, | |
1893 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
1894 | * init time. Loop devices can be requested on-demand with the | |
1895 | * /dev/loop-control interface, or be instantiated by accessing | |
1896 | * a 'dead' device node. | |
1897 | */ | |
73285082 | 1898 | if (max_loop) { |
a47653fc | 1899 | nr = max_loop; |
a1c15c59 | 1900 | range = max_loop << part_shift; |
a47653fc | 1901 | } else { |
d134b00b | 1902 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 1903 | range = 1UL << MINORBITS; |
a47653fc KC |
1904 | } |
1905 | ||
b1a66504 GC |
1906 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
1907 | err = -EIO; | |
1908 | goto misc_out; | |
1909 | } | |
1da177e4 | 1910 | |
a47653fc KC |
1911 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
1912 | THIS_MODULE, loop_probe, NULL, NULL); | |
1913 | ||
d134b00b | 1914 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
1915 | mutex_lock(&loop_index_mutex); |
1916 | for (i = 0; i < nr; i++) | |
1917 | loop_add(&lo, i); | |
1918 | mutex_unlock(&loop_index_mutex); | |
1919 | ||
73285082 | 1920 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 1921 | return 0; |
b1a66504 GC |
1922 | |
1923 | misc_out: | |
1924 | misc_deregister(&loop_misc); | |
1925 | return err; | |
34dd82af | 1926 | } |
a47653fc | 1927 | |
34dd82af KS |
1928 | static int loop_exit_cb(int id, void *ptr, void *data) |
1929 | { | |
1930 | struct loop_device *lo = ptr; | |
a47653fc | 1931 | |
34dd82af KS |
1932 | loop_remove(lo); |
1933 | return 0; | |
1da177e4 LT |
1934 | } |
1935 | ||
73285082 | 1936 | static void __exit loop_exit(void) |
1da177e4 | 1937 | { |
a47653fc | 1938 | unsigned long range; |
1da177e4 | 1939 | |
a1c15c59 | 1940 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 1941 | |
34dd82af | 1942 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 1943 | idr_destroy(&loop_index_idr); |
73285082 | 1944 | |
a47653fc | 1945 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 1946 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
1947 | |
1948 | misc_deregister(&loop_misc); | |
1da177e4 LT |
1949 | } |
1950 | ||
1951 | module_init(loop_init); | |
1952 | module_exit(loop_exit); | |
1953 | ||
1954 | #ifndef MODULE | |
1955 | static int __init max_loop_setup(char *str) | |
1956 | { | |
1957 | max_loop = simple_strtol(str, NULL, 0); | |
1958 | return 1; | |
1959 | } | |
1960 | ||
1961 | __setup("max_loop=", max_loop_setup); | |
1962 | #endif |