]>
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 | 80 | |
7c0f6ba6 | 81 | #include <linux/uaccess.h> |
1da177e4 | 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 | ||
bc07c10a ML |
448 | static inline void handle_partial_read(struct loop_cmd *cmd, long bytes) |
449 | { | |
a8ebb056 | 450 | if (bytes < 0 || op_is_write(req_op(cmd->rq))) |
bc07c10a ML |
451 | return; |
452 | ||
453 | if (unlikely(bytes < blk_rq_bytes(cmd->rq))) { | |
454 | struct bio *bio = cmd->rq->bio; | |
455 | ||
456 | bio_advance(bio, bytes); | |
457 | zero_fill_bio(bio); | |
458 | } | |
459 | } | |
460 | ||
461 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) | |
462 | { | |
463 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
464 | struct request *rq = cmd->rq; | |
465 | ||
466 | handle_partial_read(cmd, ret); | |
467 | ||
468 | if (ret > 0) | |
469 | ret = 0; | |
470 | else if (ret < 0) | |
471 | ret = -EIO; | |
472 | ||
54ef2b96 | 473 | blk_mq_complete_request(rq, ret); |
bc07c10a ML |
474 | } |
475 | ||
476 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
477 | loff_t pos, bool rw) | |
478 | { | |
479 | struct iov_iter iter; | |
480 | struct bio_vec *bvec; | |
481 | struct bio *bio = cmd->rq->bio; | |
482 | struct file *file = lo->lo_backing_file; | |
483 | int ret; | |
484 | ||
485 | /* nomerge for loop request queue */ | |
486 | WARN_ON(cmd->rq->bio != cmd->rq->biotail); | |
487 | ||
488 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
489 | iov_iter_bvec(&iter, ITER_BVEC | rw, bvec, | |
490 | bio_segments(bio), blk_rq_bytes(cmd->rq)); | |
a7297a6a ML |
491 | /* |
492 | * This bio may be started from the middle of the 'bvec' | |
493 | * because of bio splitting, so offset from the bvec must | |
494 | * be passed to iov iterator | |
495 | */ | |
496 | iter.iov_offset = bio->bi_iter.bi_bvec_done; | |
bc07c10a ML |
497 | |
498 | cmd->iocb.ki_pos = pos; | |
499 | cmd->iocb.ki_filp = file; | |
500 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
501 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
502 | ||
503 | if (rw == WRITE) | |
504 | ret = file->f_op->write_iter(&cmd->iocb, &iter); | |
505 | else | |
506 | ret = file->f_op->read_iter(&cmd->iocb, &iter); | |
507 | ||
508 | if (ret != -EIOCBQUEUED) | |
509 | cmd->iocb.ki_complete(&cmd->iocb, ret, 0); | |
510 | return 0; | |
511 | } | |
512 | ||
c1c87c2b | 513 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
514 | { |
515 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 516 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
517 | |
518 | /* | |
519 | * lo_write_simple and lo_read_simple should have been covered | |
520 | * by io submit style function like lo_rw_aio(), one blocker | |
521 | * is that lo_read_simple() need to call flush_dcache_page after | |
522 | * the page is written from kernel, and it isn't easy to handle | |
523 | * this in io submit style function which submits all segments | |
524 | * of the req at one time. And direct read IO doesn't need to | |
525 | * run flush_dcache_page(). | |
526 | */ | |
c1c87c2b CH |
527 | switch (req_op(rq)) { |
528 | case REQ_OP_FLUSH: | |
529 | return lo_req_flush(lo, rq); | |
530 | case REQ_OP_DISCARD: | |
531 | return lo_discard(lo, rq, pos); | |
532 | case REQ_OP_WRITE: | |
533 | if (lo->transfer) | |
534 | return lo_write_transfer(lo, rq, pos); | |
535 | else if (cmd->use_aio) | |
536 | return lo_rw_aio(lo, cmd, pos, WRITE); | |
af65aa8e | 537 | else |
c1c87c2b CH |
538 | return lo_write_simple(lo, rq, pos); |
539 | case REQ_OP_READ: | |
aa4d8616 | 540 | if (lo->transfer) |
c1c87c2b CH |
541 | return lo_read_transfer(lo, rq, pos); |
542 | else if (cmd->use_aio) | |
543 | return lo_rw_aio(lo, cmd, pos, READ); | |
aa4d8616 | 544 | else |
c1c87c2b CH |
545 | return lo_read_simple(lo, rq, pos); |
546 | default: | |
547 | WARN_ON_ONCE(1); | |
548 | return -EIO; | |
549 | break; | |
aa4d8616 | 550 | } |
1da177e4 LT |
551 | } |
552 | ||
1da177e4 LT |
553 | struct switch_request { |
554 | struct file *file; | |
555 | struct completion wait; | |
556 | }; | |
557 | ||
2e5ab5f3 ML |
558 | static inline void loop_update_dio(struct loop_device *lo) |
559 | { | |
560 | __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | | |
561 | lo->use_dio); | |
562 | } | |
563 | ||
1da177e4 | 564 | /* |
b5dd2f60 | 565 | * Do the actual switch; called from the BIO completion routine |
1da177e4 | 566 | */ |
b5dd2f60 | 567 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) |
1da177e4 | 568 | { |
b5dd2f60 ML |
569 | struct file *file = p->file; |
570 | struct file *old_file = lo->lo_backing_file; | |
571 | struct address_space *mapping; | |
35a82d1a | 572 | |
b5dd2f60 ML |
573 | /* if no new file, only flush of queued bios requested */ |
574 | if (!file) | |
575 | return; | |
1da177e4 | 576 | |
b5dd2f60 ML |
577 | mapping = file->f_mapping; |
578 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
579 | lo->lo_backing_file = file; | |
580 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? | |
581 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
582 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
583 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
2e5ab5f3 | 584 | loop_update_dio(lo); |
1da177e4 LT |
585 | } |
586 | ||
587 | /* | |
588 | * loop_switch performs the hard work of switching a backing store. | |
589 | * First it needs to flush existing IO, it does this by sending a magic | |
590 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
591 | */ | |
592 | static int loop_switch(struct loop_device *lo, struct file *file) | |
593 | { | |
594 | struct switch_request w; | |
b5dd2f60 | 595 | |
1da177e4 | 596 | w.file = file; |
b5dd2f60 ML |
597 | |
598 | /* freeze queue and wait for completion of scheduled requests */ | |
599 | blk_mq_freeze_queue(lo->lo_queue); | |
600 | ||
601 | /* do the switch action */ | |
602 | do_loop_switch(lo, &w); | |
603 | ||
604 | /* unfreeze */ | |
605 | blk_mq_unfreeze_queue(lo->lo_queue); | |
606 | ||
1da177e4 LT |
607 | return 0; |
608 | } | |
609 | ||
14f27939 MB |
610 | /* |
611 | * Helper to flush the IOs in loop, but keeping loop thread running | |
612 | */ | |
613 | static int loop_flush(struct loop_device *lo) | |
614 | { | |
14f27939 MB |
615 | return loop_switch(lo, NULL); |
616 | } | |
617 | ||
06f0e9e6 ML |
618 | static void loop_reread_partitions(struct loop_device *lo, |
619 | struct block_device *bdev) | |
620 | { | |
621 | int rc; | |
622 | ||
623 | /* | |
624 | * bd_mutex has been held already in release path, so don't | |
625 | * acquire it if this function is called in such case. | |
626 | * | |
627 | * If the reread partition isn't from release path, lo_refcnt | |
628 | * must be at least one and it can only become zero when the | |
629 | * current holder is released. | |
630 | */ | |
631 | if (!atomic_read(&lo->lo_refcnt)) | |
632 | rc = __blkdev_reread_part(bdev); | |
633 | else | |
634 | rc = blkdev_reread_part(bdev); | |
635 | if (rc) | |
636 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
637 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
638 | } | |
639 | ||
1da177e4 LT |
640 | /* |
641 | * loop_change_fd switched the backing store of a loopback device to | |
642 | * a new file. This is useful for operating system installers to free up | |
643 | * the original file and in High Availability environments to switch to | |
644 | * an alternative location for the content in case of server meltdown. | |
645 | * This can only work if the loop device is used read-only, and if the | |
646 | * new backing store is the same size and type as the old backing store. | |
647 | */ | |
bb214884 AV |
648 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
649 | unsigned int arg) | |
1da177e4 LT |
650 | { |
651 | struct file *file, *old_file; | |
652 | struct inode *inode; | |
653 | int error; | |
654 | ||
655 | error = -ENXIO; | |
656 | if (lo->lo_state != Lo_bound) | |
657 | goto out; | |
658 | ||
659 | /* the loop device has to be read-only */ | |
660 | error = -EINVAL; | |
661 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
662 | goto out; | |
663 | ||
664 | error = -EBADF; | |
665 | file = fget(arg); | |
666 | if (!file) | |
667 | goto out; | |
668 | ||
669 | inode = file->f_mapping->host; | |
670 | old_file = lo->lo_backing_file; | |
671 | ||
672 | error = -EINVAL; | |
673 | ||
674 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
675 | goto out_putf; | |
676 | ||
1da177e4 LT |
677 | /* size of the new backing store needs to be the same */ |
678 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
679 | goto out_putf; | |
680 | ||
681 | /* and ... switch */ | |
682 | error = loop_switch(lo, file); | |
683 | if (error) | |
684 | goto out_putf; | |
685 | ||
686 | fput(old_file); | |
e03c8dd1 | 687 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
06f0e9e6 | 688 | loop_reread_partitions(lo, bdev); |
1da177e4 LT |
689 | return 0; |
690 | ||
691 | out_putf: | |
692 | fput(file); | |
693 | out: | |
694 | return error; | |
695 | } | |
696 | ||
697 | static inline int is_loop_device(struct file *file) | |
698 | { | |
699 | struct inode *i = file->f_mapping->host; | |
700 | ||
701 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
702 | } | |
703 | ||
e14748e8 SF |
704 | /* |
705 | * for AUFS | |
706 | * no get/put for file. | |
707 | */ | |
708 | struct file *loop_backing_file(struct super_block *sb) | |
709 | { | |
710 | struct file *ret; | |
711 | struct loop_device *l; | |
712 | ||
713 | ret = NULL; | |
714 | if (MAJOR(sb->s_dev) == LOOP_MAJOR) { | |
715 | l = sb->s_bdev->bd_disk->private_data; | |
716 | ret = l->lo_backing_file; | |
717 | } | |
718 | return ret; | |
719 | } | |
720 | EXPORT_SYMBOL_GPL(loop_backing_file); | |
721 | ||
ee862730 MB |
722 | /* loop sysfs attributes */ |
723 | ||
724 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
725 | ssize_t (*callback)(struct loop_device *, char *)) | |
726 | { | |
34dd82af KS |
727 | struct gendisk *disk = dev_to_disk(dev); |
728 | struct loop_device *lo = disk->private_data; | |
ee862730 | 729 | |
34dd82af | 730 | return callback(lo, page); |
ee862730 MB |
731 | } |
732 | ||
733 | #define LOOP_ATTR_RO(_name) \ | |
734 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
735 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
736 | struct device_attribute *attr, char *b) \ | |
737 | { \ | |
738 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
739 | } \ | |
740 | static struct device_attribute loop_attr_##_name = \ | |
741 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
742 | ||
743 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
744 | { | |
745 | ssize_t ret; | |
746 | char *p = NULL; | |
747 | ||
05eb0f25 | 748 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 749 | if (lo->lo_backing_file) |
9bf39ab2 | 750 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 751 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
752 | |
753 | if (IS_ERR_OR_NULL(p)) | |
754 | ret = PTR_ERR(p); | |
755 | else { | |
756 | ret = strlen(p); | |
757 | memmove(buf, p, ret); | |
758 | buf[ret++] = '\n'; | |
759 | buf[ret] = 0; | |
760 | } | |
761 | ||
762 | return ret; | |
763 | } | |
764 | ||
765 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
766 | { | |
767 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
768 | } | |
769 | ||
770 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
771 | { | |
772 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
773 | } | |
774 | ||
775 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
776 | { | |
777 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
778 | ||
779 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
780 | } | |
781 | ||
e03c8dd1 KS |
782 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
783 | { | |
784 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
785 | ||
786 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
787 | } | |
788 | ||
2e5ab5f3 ML |
789 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
790 | { | |
791 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
792 | ||
793 | return sprintf(buf, "%s\n", dio ? "1" : "0"); | |
794 | } | |
795 | ||
ee862730 MB |
796 | LOOP_ATTR_RO(backing_file); |
797 | LOOP_ATTR_RO(offset); | |
798 | LOOP_ATTR_RO(sizelimit); | |
799 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 800 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 801 | LOOP_ATTR_RO(dio); |
ee862730 MB |
802 | |
803 | static struct attribute *loop_attrs[] = { | |
804 | &loop_attr_backing_file.attr, | |
805 | &loop_attr_offset.attr, | |
806 | &loop_attr_sizelimit.attr, | |
807 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 808 | &loop_attr_partscan.attr, |
2e5ab5f3 | 809 | &loop_attr_dio.attr, |
ee862730 MB |
810 | NULL, |
811 | }; | |
812 | ||
813 | static struct attribute_group loop_attribute_group = { | |
814 | .name = "loop", | |
815 | .attrs= loop_attrs, | |
816 | }; | |
817 | ||
818 | static int loop_sysfs_init(struct loop_device *lo) | |
819 | { | |
820 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
821 | &loop_attribute_group); | |
822 | } | |
823 | ||
824 | static void loop_sysfs_exit(struct loop_device *lo) | |
825 | { | |
826 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
827 | &loop_attribute_group); | |
828 | } | |
829 | ||
dfaa2ef6 LC |
830 | static void loop_config_discard(struct loop_device *lo) |
831 | { | |
832 | struct file *file = lo->lo_backing_file; | |
833 | struct inode *inode = file->f_mapping->host; | |
834 | struct request_queue *q = lo->lo_queue; | |
835 | ||
836 | /* | |
837 | * We use punch hole to reclaim the free space used by the | |
12a64d2f | 838 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
839 | * encryption is enabled, because it may give an attacker |
840 | * useful information. | |
841 | */ | |
842 | if ((!file->f_op->fallocate) || | |
843 | lo->lo_encrypt_key_size) { | |
844 | q->limits.discard_granularity = 0; | |
845 | q->limits.discard_alignment = 0; | |
2bb4cd5c | 846 | blk_queue_max_discard_sectors(q, 0); |
dfaa2ef6 LC |
847 | q->limits.discard_zeroes_data = 0; |
848 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
849 | return; | |
850 | } | |
851 | ||
852 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 853 | q->limits.discard_alignment = 0; |
2bb4cd5c | 854 | blk_queue_max_discard_sectors(q, UINT_MAX >> 9); |
dfaa2ef6 LC |
855 | q->limits.discard_zeroes_data = 1; |
856 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
857 | } | |
858 | ||
e03a3d7a ML |
859 | static void loop_unprepare_queue(struct loop_device *lo) |
860 | { | |
3989144f | 861 | kthread_flush_worker(&lo->worker); |
e03a3d7a ML |
862 | kthread_stop(lo->worker_task); |
863 | } | |
864 | ||
865 | static int loop_prepare_queue(struct loop_device *lo) | |
866 | { | |
3989144f | 867 | kthread_init_worker(&lo->worker); |
e03a3d7a ML |
868 | lo->worker_task = kthread_run(kthread_worker_fn, |
869 | &lo->worker, "loop%d", lo->lo_number); | |
870 | if (IS_ERR(lo->worker_task)) | |
871 | return -ENOMEM; | |
872 | set_user_nice(lo->worker_task, MIN_NICE); | |
873 | return 0; | |
874 | } | |
875 | ||
bb214884 | 876 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
877 | struct block_device *bdev, unsigned int arg) |
878 | { | |
879 | struct file *file, *f; | |
880 | struct inode *inode; | |
881 | struct address_space *mapping; | |
882 | unsigned lo_blocksize; | |
883 | int lo_flags = 0; | |
884 | int error; | |
885 | loff_t size; | |
886 | ||
887 | /* This is safe, since we have a reference from open(). */ | |
888 | __module_get(THIS_MODULE); | |
889 | ||
890 | error = -EBADF; | |
891 | file = fget(arg); | |
892 | if (!file) | |
893 | goto out; | |
894 | ||
895 | error = -EBUSY; | |
896 | if (lo->lo_state != Lo_unbound) | |
897 | goto out_putf; | |
898 | ||
899 | /* Avoid recursion */ | |
900 | f = file; | |
901 | while (is_loop_device(f)) { | |
902 | struct loop_device *l; | |
903 | ||
bb214884 | 904 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
905 | goto out_putf; |
906 | ||
907 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
908 | if (l->lo_state == Lo_unbound) { | |
909 | error = -EINVAL; | |
910 | goto out_putf; | |
911 | } | |
912 | f = l->lo_backing_file; | |
913 | } | |
914 | ||
915 | mapping = file->f_mapping; | |
916 | inode = mapping->host; | |
917 | ||
1da177e4 | 918 | error = -EINVAL; |
456be148 CH |
919 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
920 | goto out_putf; | |
1da177e4 | 921 | |
456be148 | 922 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
283e7e5d | 923 | !file->f_op->write_iter) |
456be148 | 924 | lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 925 | |
456be148 CH |
926 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
927 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 | 928 | |
456be148 | 929 | error = -EFBIG; |
1da177e4 | 930 | size = get_loop_size(lo, file); |
456be148 | 931 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 932 | goto out_putf; |
e03a3d7a ML |
933 | error = loop_prepare_queue(lo); |
934 | if (error) | |
f4aa4c7b | 935 | goto out_putf; |
1da177e4 | 936 | |
456be148 | 937 | error = 0; |
1da177e4 LT |
938 | |
939 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
940 | ||
2e5ab5f3 | 941 | lo->use_dio = false; |
1da177e4 LT |
942 | lo->lo_blocksize = lo_blocksize; |
943 | lo->lo_device = bdev; | |
944 | lo->lo_flags = lo_flags; | |
945 | lo->lo_backing_file = file; | |
aa4d8616 | 946 | lo->transfer = NULL; |
1da177e4 LT |
947 | lo->ioctl = NULL; |
948 | lo->lo_sizelimit = 0; | |
949 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
950 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
951 | ||
68db1961 | 952 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 953 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 954 | |
2e5ab5f3 | 955 | loop_update_dio(lo); |
73285082 | 956 | set_capacity(lo->lo_disk, size); |
1da177e4 | 957 | bd_set_size(bdev, size << 9); |
ee862730 | 958 | loop_sysfs_init(lo); |
c3473c63 DZ |
959 | /* let user-space know about the new size */ |
960 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 LT |
961 | |
962 | set_blocksize(bdev, lo_blocksize); | |
963 | ||
6c997918 | 964 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
965 | if (part_shift) |
966 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
967 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
06f0e9e6 | 968 | loop_reread_partitions(lo, bdev); |
c1681bf8 AP |
969 | |
970 | /* Grab the block_device to prevent its destruction after we | |
971 | * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev). | |
972 | */ | |
973 | bdgrab(bdev); | |
1da177e4 LT |
974 | return 0; |
975 | ||
976 | out_putf: | |
977 | fput(file); | |
978 | out: | |
979 | /* This is safe: open() is still holding a reference. */ | |
980 | module_put(THIS_MODULE); | |
981 | return error; | |
982 | } | |
983 | ||
984 | static int | |
985 | loop_release_xfer(struct loop_device *lo) | |
986 | { | |
987 | int err = 0; | |
988 | struct loop_func_table *xfer = lo->lo_encryption; | |
989 | ||
990 | if (xfer) { | |
991 | if (xfer->release) | |
992 | err = xfer->release(lo); | |
993 | lo->transfer = NULL; | |
994 | lo->lo_encryption = NULL; | |
995 | module_put(xfer->owner); | |
996 | } | |
997 | return err; | |
998 | } | |
999 | ||
1000 | static int | |
1001 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1002 | const struct loop_info64 *i) | |
1003 | { | |
1004 | int err = 0; | |
1005 | ||
1006 | if (xfer) { | |
1007 | struct module *owner = xfer->owner; | |
1008 | ||
1009 | if (!try_module_get(owner)) | |
1010 | return -EINVAL; | |
1011 | if (xfer->init) | |
1012 | err = xfer->init(lo, i); | |
1013 | if (err) | |
1014 | module_put(owner); | |
1015 | else | |
1016 | lo->lo_encryption = xfer; | |
1017 | } | |
1018 | return err; | |
1019 | } | |
1020 | ||
4c823cc3 | 1021 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
1022 | { |
1023 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 1024 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 1025 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
1026 | |
1027 | if (lo->lo_state != Lo_bound) | |
1028 | return -ENXIO; | |
1029 | ||
a1ecac3b DC |
1030 | /* |
1031 | * If we've explicitly asked to tear down the loop device, | |
1032 | * and it has an elevated reference count, set it for auto-teardown when | |
1033 | * the last reference goes away. This stops $!~#$@ udev from | |
1034 | * preventing teardown because it decided that it needs to run blkid on | |
1035 | * the loopback device whenever they appear. xfstests is notorious for | |
1036 | * failing tests because blkid via udev races with a losetup | |
1037 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1038 | * command to fail with EBUSY. | |
1039 | */ | |
f8933667 | 1040 | if (atomic_read(&lo->lo_refcnt) > 1) { |
a1ecac3b DC |
1041 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
1042 | mutex_unlock(&lo->lo_ctl_mutex); | |
1043 | return 0; | |
1044 | } | |
1da177e4 LT |
1045 | |
1046 | if (filp == NULL) | |
1047 | return -EINVAL; | |
1048 | ||
f8933667 ML |
1049 | /* freeze request queue during the transition */ |
1050 | blk_mq_freeze_queue(lo->lo_queue); | |
1051 | ||
1da177e4 LT |
1052 | spin_lock_irq(&lo->lo_lock); |
1053 | lo->lo_state = Lo_rundown; | |
1da177e4 | 1054 | lo->lo_backing_file = NULL; |
05eb0f25 | 1055 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1056 | |
1057 | loop_release_xfer(lo); | |
1058 | lo->transfer = NULL; | |
1059 | lo->ioctl = NULL; | |
1060 | lo->lo_device = NULL; | |
1061 | lo->lo_encryption = NULL; | |
1062 | lo->lo_offset = 0; | |
1063 | lo->lo_sizelimit = 0; | |
1064 | lo->lo_encrypt_key_size = 0; | |
1da177e4 LT |
1065 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1066 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1067 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
c1681bf8 AP |
1068 | if (bdev) { |
1069 | bdput(bdev); | |
bb214884 | 1070 | invalidate_bdev(bdev); |
c1681bf8 | 1071 | } |
73285082 | 1072 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1073 | loop_sysfs_exit(lo); |
c3473c63 | 1074 | if (bdev) { |
bb214884 | 1075 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1076 | /* let user-space know about this change */ |
1077 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1078 | } | |
1da177e4 LT |
1079 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1080 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1081 | /* This is safe: open() is still holding a reference. */ |
1082 | module_put(THIS_MODULE); | |
f8933667 ML |
1083 | blk_mq_unfreeze_queue(lo->lo_queue); |
1084 | ||
c2fccc1c | 1085 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
06f0e9e6 | 1086 | loop_reread_partitions(lo, bdev); |
e03c8dd1 KS |
1087 | lo->lo_flags = 0; |
1088 | if (!part_shift) | |
1089 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
e03a3d7a | 1090 | loop_unprepare_queue(lo); |
f028f3b2 NK |
1091 | mutex_unlock(&lo->lo_ctl_mutex); |
1092 | /* | |
1093 | * Need not hold lo_ctl_mutex to fput backing file. | |
1094 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1095 | * lock dependency possibility warning as fput can take | |
1096 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1097 | */ | |
1098 | fput(filp); | |
1da177e4 LT |
1099 | return 0; |
1100 | } | |
1101 | ||
1102 | static int | |
1103 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1104 | { | |
1105 | int err; | |
1106 | struct loop_func_table *xfer; | |
e4849737 | 1107 | kuid_t uid = current_uid(); |
1da177e4 | 1108 | |
b0fafa81 | 1109 | if (lo->lo_encrypt_key_size && |
e4849737 | 1110 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1111 | !capable(CAP_SYS_ADMIN)) |
1112 | return -EPERM; | |
1113 | if (lo->lo_state != Lo_bound) | |
1114 | return -ENXIO; | |
1115 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1116 | return -EINVAL; | |
1117 | ||
df8ca830 ML |
1118 | /* I/O need to be drained during transfer transition */ |
1119 | blk_mq_freeze_queue(lo->lo_queue); | |
1120 | ||
1da177e4 LT |
1121 | err = loop_release_xfer(lo); |
1122 | if (err) | |
df8ca830 | 1123 | goto exit; |
1da177e4 LT |
1124 | |
1125 | if (info->lo_encrypt_type) { | |
1126 | unsigned int type = info->lo_encrypt_type; | |
1127 | ||
1128 | if (type >= MAX_LO_CRYPT) | |
1129 | return -EINVAL; | |
1130 | xfer = xfer_funcs[type]; | |
1131 | if (xfer == NULL) | |
1132 | return -EINVAL; | |
1133 | } else | |
1134 | xfer = NULL; | |
1135 | ||
1136 | err = loop_init_xfer(lo, xfer, info); | |
1137 | if (err) | |
df8ca830 | 1138 | goto exit; |
1da177e4 LT |
1139 | |
1140 | if (lo->lo_offset != info->lo_offset || | |
7b0576a3 | 1141 | lo->lo_sizelimit != info->lo_sizelimit) |
df8ca830 ML |
1142 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) { |
1143 | err = -EFBIG; | |
1144 | goto exit; | |
1145 | } | |
541c742a | 1146 | |
dfaa2ef6 | 1147 | loop_config_discard(lo); |
1da177e4 LT |
1148 | |
1149 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1150 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1151 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1152 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1153 | ||
1154 | if (!xfer) | |
1155 | xfer = &none_funcs; | |
1156 | lo->transfer = xfer->transfer; | |
1157 | lo->ioctl = xfer->ioctl; | |
1158 | ||
96c58655 DW |
1159 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1160 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1161 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1162 | ||
1da177e4 LT |
1163 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1164 | lo->lo_init[0] = info->lo_init[0]; | |
1165 | lo->lo_init[1] = info->lo_init[1]; | |
1166 | if (info->lo_encrypt_key_size) { | |
1167 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1168 | info->lo_encrypt_key_size); | |
b0fafa81 | 1169 | lo->lo_key_owner = uid; |
aa4d8616 | 1170 | } |
1da177e4 | 1171 | |
2e5ab5f3 ML |
1172 | /* update dio if lo_offset or transfer is changed */ |
1173 | __loop_update_dio(lo, lo->use_dio); | |
1174 | ||
df8ca830 ML |
1175 | exit: |
1176 | blk_mq_unfreeze_queue(lo->lo_queue); | |
bf1447c8 OS |
1177 | |
1178 | if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && | |
1179 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1180 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1181 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1182 | loop_reread_partitions(lo, lo->lo_device); | |
1183 | } | |
1184 | ||
df8ca830 | 1185 | return err; |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | static int | |
1189 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1190 | { | |
1191 | struct file *file = lo->lo_backing_file; | |
1192 | struct kstat stat; | |
1193 | int error; | |
1194 | ||
1195 | if (lo->lo_state != Lo_bound) | |
1196 | return -ENXIO; | |
3dadecce | 1197 | error = vfs_getattr(&file->f_path, &stat); |
1da177e4 LT |
1198 | if (error) |
1199 | return error; | |
1200 | memset(info, 0, sizeof(*info)); | |
1201 | info->lo_number = lo->lo_number; | |
1202 | info->lo_device = huge_encode_dev(stat.dev); | |
1203 | info->lo_inode = stat.ino; | |
1204 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1205 | info->lo_offset = lo->lo_offset; | |
1206 | info->lo_sizelimit = lo->lo_sizelimit; | |
1207 | info->lo_flags = lo->lo_flags; | |
1208 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1209 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1210 | info->lo_encrypt_type = | |
1211 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1212 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1213 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1214 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1215 | lo->lo_encrypt_key_size); | |
1216 | } | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | static void | |
1221 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1222 | { | |
1223 | memset(info64, 0, sizeof(*info64)); | |
1224 | info64->lo_number = info->lo_number; | |
1225 | info64->lo_device = info->lo_device; | |
1226 | info64->lo_inode = info->lo_inode; | |
1227 | info64->lo_rdevice = info->lo_rdevice; | |
1228 | info64->lo_offset = info->lo_offset; | |
1229 | info64->lo_sizelimit = 0; | |
1230 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1231 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1232 | info64->lo_flags = info->lo_flags; | |
1233 | info64->lo_init[0] = info->lo_init[0]; | |
1234 | info64->lo_init[1] = info->lo_init[1]; | |
1235 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1236 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1237 | else | |
1238 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1239 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1240 | } | |
1241 | ||
1242 | static int | |
1243 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1244 | { | |
1245 | memset(info, 0, sizeof(*info)); | |
1246 | info->lo_number = info64->lo_number; | |
1247 | info->lo_device = info64->lo_device; | |
1248 | info->lo_inode = info64->lo_inode; | |
1249 | info->lo_rdevice = info64->lo_rdevice; | |
1250 | info->lo_offset = info64->lo_offset; | |
1251 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1252 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1253 | info->lo_flags = info64->lo_flags; | |
1254 | info->lo_init[0] = info64->lo_init[0]; | |
1255 | info->lo_init[1] = info64->lo_init[1]; | |
1256 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1257 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1258 | else | |
1259 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1260 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1261 | ||
1262 | /* error in case values were truncated */ | |
1263 | if (info->lo_device != info64->lo_device || | |
1264 | info->lo_rdevice != info64->lo_rdevice || | |
1265 | info->lo_inode != info64->lo_inode || | |
1266 | info->lo_offset != info64->lo_offset) | |
1267 | return -EOVERFLOW; | |
1268 | ||
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | static int | |
1273 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1274 | { | |
1275 | struct loop_info info; | |
1276 | struct loop_info64 info64; | |
1277 | ||
1278 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1279 | return -EFAULT; | |
1280 | loop_info64_from_old(&info, &info64); | |
1281 | return loop_set_status(lo, &info64); | |
1282 | } | |
1283 | ||
1284 | static int | |
1285 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1286 | { | |
1287 | struct loop_info64 info64; | |
1288 | ||
1289 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1290 | return -EFAULT; | |
1291 | return loop_set_status(lo, &info64); | |
1292 | } | |
1293 | ||
1294 | static int | |
1295 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1296 | struct loop_info info; | |
1297 | struct loop_info64 info64; | |
1298 | int err = 0; | |
1299 | ||
1300 | if (!arg) | |
1301 | err = -EINVAL; | |
1302 | if (!err) | |
1303 | err = loop_get_status(lo, &info64); | |
1304 | if (!err) | |
1305 | err = loop_info64_to_old(&info64, &info); | |
1306 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1307 | err = -EFAULT; | |
1308 | ||
1309 | return err; | |
1310 | } | |
1311 | ||
1312 | static int | |
1313 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1314 | struct loop_info64 info64; | |
1315 | int err = 0; | |
1316 | ||
1317 | if (!arg) | |
1318 | err = -EINVAL; | |
1319 | if (!err) | |
1320 | err = loop_get_status(lo, &info64); | |
1321 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1322 | err = -EFAULT; | |
1323 | ||
1324 | return err; | |
1325 | } | |
1326 | ||
53d66608 O |
1327 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) |
1328 | { | |
53d66608 | 1329 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1330 | return -ENXIO; |
53d66608 | 1331 | |
7b0576a3 | 1332 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1333 | } |
1334 | ||
ab1cb278 ML |
1335 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1336 | { | |
1337 | int error = -ENXIO; | |
1338 | if (lo->lo_state != Lo_bound) | |
1339 | goto out; | |
1340 | ||
1341 | __loop_update_dio(lo, !!arg); | |
1342 | if (lo->use_dio == !!arg) | |
1343 | return 0; | |
1344 | error = -EINVAL; | |
1345 | out: | |
1346 | return error; | |
1347 | } | |
1348 | ||
bb214884 | 1349 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1350 | unsigned int cmd, unsigned long arg) |
1351 | { | |
bb214884 | 1352 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1353 | int err; |
1354 | ||
f028f3b2 | 1355 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1356 | switch (cmd) { |
1357 | case LOOP_SET_FD: | |
bb214884 | 1358 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1359 | break; |
1360 | case LOOP_CHANGE_FD: | |
bb214884 | 1361 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1362 | break; |
1363 | case LOOP_CLR_FD: | |
f028f3b2 | 1364 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1365 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1366 | if (!err) |
1367 | goto out_unlocked; | |
1da177e4 LT |
1368 | break; |
1369 | case LOOP_SET_STATUS: | |
7035b5df DM |
1370 | err = -EPERM; |
1371 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1372 | err = loop_set_status_old(lo, | |
1373 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1374 | break; |
1375 | case LOOP_GET_STATUS: | |
1376 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1377 | break; | |
1378 | case LOOP_SET_STATUS64: | |
7035b5df DM |
1379 | err = -EPERM; |
1380 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1381 | err = loop_set_status64(lo, | |
1382 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1383 | break; |
1384 | case LOOP_GET_STATUS64: | |
1385 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1386 | break; | |
53d66608 O |
1387 | case LOOP_SET_CAPACITY: |
1388 | err = -EPERM; | |
1389 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1390 | err = loop_set_capacity(lo, bdev); | |
1391 | break; | |
ab1cb278 ML |
1392 | case LOOP_SET_DIRECT_IO: |
1393 | err = -EPERM; | |
1394 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1395 | err = loop_set_dio(lo, arg); | |
1396 | break; | |
1da177e4 LT |
1397 | default: |
1398 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1399 | } | |
f85221dd | 1400 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1401 | |
1402 | out_unlocked: | |
1da177e4 LT |
1403 | return err; |
1404 | } | |
1405 | ||
863d5b82 DH |
1406 | #ifdef CONFIG_COMPAT |
1407 | struct compat_loop_info { | |
1408 | compat_int_t lo_number; /* ioctl r/o */ | |
1409 | compat_dev_t lo_device; /* ioctl r/o */ | |
1410 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1411 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1412 | compat_int_t lo_offset; | |
1413 | compat_int_t lo_encrypt_type; | |
1414 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1415 | compat_int_t lo_flags; /* ioctl r/o */ | |
1416 | char lo_name[LO_NAME_SIZE]; | |
1417 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1418 | compat_ulong_t lo_init[2]; | |
1419 | char reserved[4]; | |
1420 | }; | |
1421 | ||
1422 | /* | |
1423 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1424 | * - noinlined to reduce stack space usage in main part of driver | |
1425 | */ | |
1426 | static noinline int | |
ba674cfc | 1427 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1428 | struct loop_info64 *info64) |
1429 | { | |
1430 | struct compat_loop_info info; | |
1431 | ||
1432 | if (copy_from_user(&info, arg, sizeof(info))) | |
1433 | return -EFAULT; | |
1434 | ||
1435 | memset(info64, 0, sizeof(*info64)); | |
1436 | info64->lo_number = info.lo_number; | |
1437 | info64->lo_device = info.lo_device; | |
1438 | info64->lo_inode = info.lo_inode; | |
1439 | info64->lo_rdevice = info.lo_rdevice; | |
1440 | info64->lo_offset = info.lo_offset; | |
1441 | info64->lo_sizelimit = 0; | |
1442 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1443 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1444 | info64->lo_flags = info.lo_flags; | |
1445 | info64->lo_init[0] = info.lo_init[0]; | |
1446 | info64->lo_init[1] = info.lo_init[1]; | |
1447 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1448 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1449 | else | |
1450 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1451 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1452 | return 0; | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1457 | * - noinlined to reduce stack space usage in main part of driver | |
1458 | */ | |
1459 | static noinline int | |
1460 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1461 | struct compat_loop_info __user *arg) | |
1462 | { | |
1463 | struct compat_loop_info info; | |
1464 | ||
1465 | memset(&info, 0, sizeof(info)); | |
1466 | info.lo_number = info64->lo_number; | |
1467 | info.lo_device = info64->lo_device; | |
1468 | info.lo_inode = info64->lo_inode; | |
1469 | info.lo_rdevice = info64->lo_rdevice; | |
1470 | info.lo_offset = info64->lo_offset; | |
1471 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1472 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1473 | info.lo_flags = info64->lo_flags; | |
1474 | info.lo_init[0] = info64->lo_init[0]; | |
1475 | info.lo_init[1] = info64->lo_init[1]; | |
1476 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1477 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1478 | else | |
1479 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1480 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1481 | ||
1482 | /* error in case values were truncated */ | |
1483 | if (info.lo_device != info64->lo_device || | |
1484 | info.lo_rdevice != info64->lo_rdevice || | |
1485 | info.lo_inode != info64->lo_inode || | |
1486 | info.lo_offset != info64->lo_offset || | |
1487 | info.lo_init[0] != info64->lo_init[0] || | |
1488 | info.lo_init[1] != info64->lo_init[1]) | |
1489 | return -EOVERFLOW; | |
1490 | ||
1491 | if (copy_to_user(arg, &info, sizeof(info))) | |
1492 | return -EFAULT; | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | static int | |
1497 | loop_set_status_compat(struct loop_device *lo, | |
1498 | const struct compat_loop_info __user *arg) | |
1499 | { | |
1500 | struct loop_info64 info64; | |
1501 | int ret; | |
1502 | ||
1503 | ret = loop_info64_from_compat(arg, &info64); | |
1504 | if (ret < 0) | |
1505 | return ret; | |
1506 | return loop_set_status(lo, &info64); | |
1507 | } | |
1508 | ||
1509 | static int | |
1510 | loop_get_status_compat(struct loop_device *lo, | |
1511 | struct compat_loop_info __user *arg) | |
1512 | { | |
1513 | struct loop_info64 info64; | |
1514 | int err = 0; | |
1515 | ||
1516 | if (!arg) | |
1517 | err = -EINVAL; | |
1518 | if (!err) | |
1519 | err = loop_get_status(lo, &info64); | |
1520 | if (!err) | |
1521 | err = loop_info64_to_compat(&info64, arg); | |
1522 | return err; | |
1523 | } | |
1524 | ||
bb214884 AV |
1525 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1526 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1527 | { |
bb214884 | 1528 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1529 | int err; |
1530 | ||
863d5b82 DH |
1531 | switch(cmd) { |
1532 | case LOOP_SET_STATUS: | |
1533 | mutex_lock(&lo->lo_ctl_mutex); | |
1534 | err = loop_set_status_compat( | |
1535 | lo, (const struct compat_loop_info __user *) arg); | |
1536 | mutex_unlock(&lo->lo_ctl_mutex); | |
1537 | break; | |
1538 | case LOOP_GET_STATUS: | |
1539 | mutex_lock(&lo->lo_ctl_mutex); | |
1540 | err = loop_get_status_compat( | |
1541 | lo, (struct compat_loop_info __user *) arg); | |
1542 | mutex_unlock(&lo->lo_ctl_mutex); | |
1543 | break; | |
53d66608 | 1544 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1545 | case LOOP_CLR_FD: |
1546 | case LOOP_GET_STATUS64: | |
1547 | case LOOP_SET_STATUS64: | |
1548 | arg = (unsigned long) compat_ptr(arg); | |
1549 | case LOOP_SET_FD: | |
1550 | case LOOP_CHANGE_FD: | |
bb214884 | 1551 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1552 | break; |
1553 | default: | |
1554 | err = -ENOIOCTLCMD; | |
1555 | break; | |
1556 | } | |
863d5b82 DH |
1557 | return err; |
1558 | } | |
1559 | #endif | |
1560 | ||
bb214884 | 1561 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1562 | { |
770fe30a KS |
1563 | struct loop_device *lo; |
1564 | int err = 0; | |
1565 | ||
1566 | mutex_lock(&loop_index_mutex); | |
1567 | lo = bdev->bd_disk->private_data; | |
1568 | if (!lo) { | |
1569 | err = -ENXIO; | |
1570 | goto out; | |
1571 | } | |
1da177e4 | 1572 | |
f8933667 | 1573 | atomic_inc(&lo->lo_refcnt); |
770fe30a KS |
1574 | out: |
1575 | mutex_unlock(&loop_index_mutex); | |
1576 | return err; | |
1da177e4 LT |
1577 | } |
1578 | ||
db2a144b | 1579 | static void lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1580 | { |
bb214884 | 1581 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1582 | int err; |
1da177e4 | 1583 | |
f8933667 ML |
1584 | if (atomic_dec_return(&lo->lo_refcnt)) |
1585 | return; | |
14f27939 | 1586 | |
f8933667 | 1587 | mutex_lock(&lo->lo_ctl_mutex); |
14f27939 MB |
1588 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { |
1589 | /* | |
1590 | * In autoclear mode, stop the loop thread | |
1591 | * and remove configuration after last close. | |
1592 | */ | |
4c823cc3 | 1593 | err = loop_clr_fd(lo); |
ffcd7dca | 1594 | if (!err) |
db2a144b | 1595 | return; |
14f27939 MB |
1596 | } else { |
1597 | /* | |
1598 | * Otherwise keep thread (if running) and config, | |
1599 | * but flush possible ongoing bios in thread. | |
1600 | */ | |
1601 | loop_flush(lo); | |
1602 | } | |
96c58655 | 1603 | |
f85221dd | 1604 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1605 | } |
1606 | ||
83d5cde4 | 1607 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1608 | .owner = THIS_MODULE, |
bb214884 AV |
1609 | .open = lo_open, |
1610 | .release = lo_release, | |
1611 | .ioctl = lo_ioctl, | |
863d5b82 | 1612 | #ifdef CONFIG_COMPAT |
bb214884 | 1613 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1614 | #endif |
1da177e4 LT |
1615 | }; |
1616 | ||
1617 | /* | |
1618 | * And now the modules code and kernel interface. | |
1619 | */ | |
73285082 | 1620 | static int max_loop; |
ac04fee0 | 1621 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1622 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1623 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1624 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1625 | MODULE_LICENSE("GPL"); |
1626 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1627 | ||
1628 | int loop_register_transfer(struct loop_func_table *funcs) | |
1629 | { | |
1630 | unsigned int n = funcs->number; | |
1631 | ||
1632 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1633 | return -EINVAL; | |
1634 | xfer_funcs[n] = funcs; | |
1635 | return 0; | |
1636 | } | |
1637 | ||
34dd82af KS |
1638 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1639 | { | |
1640 | struct loop_device *lo = ptr; | |
1641 | struct loop_func_table *xfer = data; | |
1642 | ||
1643 | mutex_lock(&lo->lo_ctl_mutex); | |
1644 | if (lo->lo_encryption == xfer) | |
1645 | loop_release_xfer(lo); | |
1646 | mutex_unlock(&lo->lo_ctl_mutex); | |
1647 | return 0; | |
1648 | } | |
1649 | ||
1da177e4 LT |
1650 | int loop_unregister_transfer(int number) |
1651 | { | |
1652 | unsigned int n = number; | |
1da177e4 LT |
1653 | struct loop_func_table *xfer; |
1654 | ||
1655 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1656 | return -EINVAL; | |
1657 | ||
1658 | xfer_funcs[n] = NULL; | |
34dd82af | 1659 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1660 | return 0; |
1661 | } | |
1662 | ||
1663 | EXPORT_SYMBOL(loop_register_transfer); | |
1664 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1665 | ||
b5dd2f60 ML |
1666 | static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
1667 | const struct blk_mq_queue_data *bd) | |
1668 | { | |
1669 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
f4aa4c7b | 1670 | struct loop_device *lo = cmd->rq->q->queuedata; |
b5dd2f60 ML |
1671 | |
1672 | blk_mq_start_request(bd->rq); | |
1673 | ||
f4aa4c7b | 1674 | if (lo->lo_state != Lo_bound) |
b4a567e8 | 1675 | return BLK_MQ_RQ_QUEUE_ERROR; |
f4aa4c7b | 1676 | |
f0225cac CH |
1677 | switch (req_op(cmd->rq)) { |
1678 | case REQ_OP_FLUSH: | |
1679 | case REQ_OP_DISCARD: | |
bc07c10a | 1680 | cmd->use_aio = false; |
f0225cac CH |
1681 | break; |
1682 | default: | |
1683 | cmd->use_aio = lo->use_dio; | |
1684 | break; | |
1685 | } | |
bc07c10a | 1686 | |
3989144f | 1687 | kthread_queue_work(&lo->worker, &cmd->work); |
b5dd2f60 ML |
1688 | |
1689 | return BLK_MQ_RQ_QUEUE_OK; | |
1690 | } | |
1691 | ||
1692 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1693 | { | |
a8ebb056 | 1694 | const bool write = op_is_write(req_op(cmd->rq)); |
b5dd2f60 | 1695 | struct loop_device *lo = cmd->rq->q->queuedata; |
f4829a9b | 1696 | int ret = 0; |
b5dd2f60 | 1697 | |
f4829a9b CH |
1698 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1699 | ret = -EIO; | |
b5dd2f60 | 1700 | goto failed; |
f4829a9b | 1701 | } |
b5dd2f60 | 1702 | |
30112013 | 1703 | ret = do_req_filebacked(lo, cmd->rq); |
b5dd2f60 | 1704 | failed: |
bc07c10a ML |
1705 | /* complete non-aio request */ |
1706 | if (!cmd->use_aio || ret) | |
54ef2b96 | 1707 | blk_mq_complete_request(cmd->rq, ret ? -EIO : 0); |
b5dd2f60 ML |
1708 | } |
1709 | ||
e03a3d7a | 1710 | static void loop_queue_work(struct kthread_work *work) |
b5dd2f60 ML |
1711 | { |
1712 | struct loop_cmd *cmd = | |
e03a3d7a | 1713 | container_of(work, struct loop_cmd, work); |
b5dd2f60 ML |
1714 | |
1715 | loop_handle_cmd(cmd); | |
1716 | } | |
1717 | ||
1718 | static int loop_init_request(void *data, struct request *rq, | |
1719 | unsigned int hctx_idx, unsigned int request_idx, | |
1720 | unsigned int numa_node) | |
1721 | { | |
1722 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1723 | ||
1724 | cmd->rq = rq; | |
3989144f | 1725 | kthread_init_work(&cmd->work, loop_queue_work); |
b5dd2f60 ML |
1726 | |
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | static struct blk_mq_ops loop_mq_ops = { | |
1731 | .queue_rq = loop_queue_rq, | |
b5dd2f60 ML |
1732 | .init_request = loop_init_request, |
1733 | }; | |
1734 | ||
34dd82af | 1735 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1736 | { |
1737 | struct loop_device *lo; | |
1738 | struct gendisk *disk; | |
34dd82af | 1739 | int err; |
73285082 | 1740 | |
68d740d7 | 1741 | err = -ENOMEM; |
73285082 | 1742 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1743 | if (!lo) |
73285082 | 1744 | goto out; |
34dd82af | 1745 | |
ef7e7c82 MP |
1746 | lo->lo_state = Lo_unbound; |
1747 | ||
c718aa65 | 1748 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1749 | if (i >= 0) { |
c718aa65 TH |
1750 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1751 | if (err == -ENOSPC) | |
34dd82af | 1752 | err = -EEXIST; |
34dd82af | 1753 | } else { |
c718aa65 | 1754 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1755 | } |
1756 | if (err < 0) | |
1757 | goto out_free_dev; | |
c718aa65 | 1758 | i = err; |
73285082 | 1759 | |
183cfb57 | 1760 | err = -ENOMEM; |
b5dd2f60 ML |
1761 | lo->tag_set.ops = &loop_mq_ops; |
1762 | lo->tag_set.nr_hw_queues = 1; | |
1763 | lo->tag_set.queue_depth = 128; | |
1764 | lo->tag_set.numa_node = NUMA_NO_NODE; | |
1765 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
1766 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
1767 | lo->tag_set.driver_data = lo; | |
1768 | ||
1769 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
1770 | if (err) | |
3ec981e3 | 1771 | goto out_free_idr; |
73285082 | 1772 | |
b5dd2f60 ML |
1773 | lo->lo_queue = blk_mq_init_queue(&lo->tag_set); |
1774 | if (IS_ERR_OR_NULL(lo->lo_queue)) { | |
1775 | err = PTR_ERR(lo->lo_queue); | |
1776 | goto out_cleanup_tags; | |
1777 | } | |
ef7e7c82 MP |
1778 | lo->lo_queue->queuedata = lo; |
1779 | ||
5b5e20f4 ML |
1780 | /* |
1781 | * It doesn't make sense to enable merge because the I/O | |
1782 | * submitted to backing file is handled page by page. | |
1783 | */ | |
1784 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue); | |
1785 | ||
7a649737 | 1786 | err = -ENOMEM; |
476a4813 | 1787 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1788 | if (!disk) |
1789 | goto out_free_queue; | |
1790 | ||
e03c8dd1 KS |
1791 | /* |
1792 | * Disable partition scanning by default. The in-kernel partition | |
1793 | * scanning can be requested individually per-device during its | |
1794 | * setup. Userspace can always add and remove partitions from all | |
1795 | * devices. The needed partition minors are allocated from the | |
1796 | * extended minor space, the main loop device numbers will continue | |
1797 | * to match the loop minors, regardless of the number of partitions | |
1798 | * used. | |
1799 | * | |
1800 | * If max_part is given, partition scanning is globally enabled for | |
1801 | * all loop devices. The minors for the main loop devices will be | |
1802 | * multiples of max_part. | |
1803 | * | |
1804 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1805 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1806 | * complicated, are too static, inflexible and may surprise | |
1807 | * userspace tools. Parameters like this in general should be avoided. | |
1808 | */ | |
1809 | if (!part_shift) | |
1810 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1811 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 | 1812 | mutex_init(&lo->lo_ctl_mutex); |
f8933667 | 1813 | atomic_set(&lo->lo_refcnt, 0); |
73285082 | 1814 | lo->lo_number = i; |
73285082 KC |
1815 | spin_lock_init(&lo->lo_lock); |
1816 | disk->major = LOOP_MAJOR; | |
476a4813 | 1817 | disk->first_minor = i << part_shift; |
73285082 KC |
1818 | disk->fops = &lo_fops; |
1819 | disk->private_data = lo; | |
1820 | disk->queue = lo->lo_queue; | |
1821 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1822 | add_disk(disk); |
1823 | *l = lo; | |
1824 | return lo->lo_number; | |
73285082 KC |
1825 | |
1826 | out_free_queue: | |
1827 | blk_cleanup_queue(lo->lo_queue); | |
b5dd2f60 ML |
1828 | out_cleanup_tags: |
1829 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 MP |
1830 | out_free_idr: |
1831 | idr_remove(&loop_index_idr, i); | |
73285082 KC |
1832 | out_free_dev: |
1833 | kfree(lo); | |
1834 | out: | |
34dd82af | 1835 | return err; |
73285082 KC |
1836 | } |
1837 | ||
34dd82af | 1838 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1839 | { |
73285082 | 1840 | blk_cleanup_queue(lo->lo_queue); |
6cd18e71 | 1841 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 1842 | blk_mq_free_tag_set(&lo->tag_set); |
73285082 | 1843 | put_disk(lo->lo_disk); |
73285082 KC |
1844 | kfree(lo); |
1845 | } | |
1da177e4 | 1846 | |
770fe30a KS |
1847 | static int find_free_cb(int id, void *ptr, void *data) |
1848 | { | |
1849 | struct loop_device *lo = ptr; | |
1850 | struct loop_device **l = data; | |
1851 | ||
1852 | if (lo->lo_state == Lo_unbound) { | |
1853 | *l = lo; | |
1854 | return 1; | |
1855 | } | |
1856 | return 0; | |
1857 | } | |
1858 | ||
34dd82af | 1859 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1860 | { |
1861 | struct loop_device *lo; | |
34dd82af | 1862 | int ret = -ENODEV; |
a47653fc | 1863 | |
770fe30a KS |
1864 | if (i < 0) { |
1865 | int err; | |
1866 | ||
1867 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1868 | if (err == 1) { | |
1869 | *l = lo; | |
1870 | ret = lo->lo_number; | |
1871 | } | |
1872 | goto out; | |
a47653fc KC |
1873 | } |
1874 | ||
770fe30a | 1875 | /* lookup and return a specific i */ |
34dd82af | 1876 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1877 | if (lo) { |
34dd82af KS |
1878 | *l = lo; |
1879 | ret = lo->lo_number; | |
a47653fc | 1880 | } |
770fe30a | 1881 | out: |
34dd82af | 1882 | return ret; |
a47653fc KC |
1883 | } |
1884 | ||
73285082 KC |
1885 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1886 | { | |
705962cc | 1887 | struct loop_device *lo; |
07002e99 | 1888 | struct kobject *kobj; |
34dd82af | 1889 | int err; |
73285082 | 1890 | |
34dd82af KS |
1891 | mutex_lock(&loop_index_mutex); |
1892 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1893 | if (err < 0) | |
1894 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
1895 | if (err < 0) | |
a207f593 | 1896 | kobj = NULL; |
34dd82af KS |
1897 | else |
1898 | kobj = get_disk(lo->lo_disk); | |
1899 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
1900 | |
1901 | *part = 0; | |
07002e99 | 1902 | return kobj; |
73285082 KC |
1903 | } |
1904 | ||
770fe30a KS |
1905 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
1906 | unsigned long parm) | |
1907 | { | |
1908 | struct loop_device *lo; | |
1909 | int ret = -ENOSYS; | |
1910 | ||
1911 | mutex_lock(&loop_index_mutex); | |
1912 | switch (cmd) { | |
1913 | case LOOP_CTL_ADD: | |
1914 | ret = loop_lookup(&lo, parm); | |
1915 | if (ret >= 0) { | |
1916 | ret = -EEXIST; | |
1917 | break; | |
1918 | } | |
1919 | ret = loop_add(&lo, parm); | |
1920 | break; | |
1921 | case LOOP_CTL_REMOVE: | |
1922 | ret = loop_lookup(&lo, parm); | |
1923 | if (ret < 0) | |
1924 | break; | |
1925 | mutex_lock(&lo->lo_ctl_mutex); | |
1926 | if (lo->lo_state != Lo_unbound) { | |
1927 | ret = -EBUSY; | |
1928 | mutex_unlock(&lo->lo_ctl_mutex); | |
1929 | break; | |
1930 | } | |
f8933667 | 1931 | if (atomic_read(&lo->lo_refcnt) > 0) { |
770fe30a KS |
1932 | ret = -EBUSY; |
1933 | mutex_unlock(&lo->lo_ctl_mutex); | |
1934 | break; | |
1935 | } | |
1936 | lo->lo_disk->private_data = NULL; | |
1937 | mutex_unlock(&lo->lo_ctl_mutex); | |
1938 | idr_remove(&loop_index_idr, lo->lo_number); | |
1939 | loop_remove(lo); | |
1940 | break; | |
1941 | case LOOP_CTL_GET_FREE: | |
1942 | ret = loop_lookup(&lo, -1); | |
1943 | if (ret >= 0) | |
1944 | break; | |
1945 | ret = loop_add(&lo, -1); | |
1946 | } | |
1947 | mutex_unlock(&loop_index_mutex); | |
1948 | ||
1949 | return ret; | |
1950 | } | |
1951 | ||
1952 | static const struct file_operations loop_ctl_fops = { | |
1953 | .open = nonseekable_open, | |
1954 | .unlocked_ioctl = loop_control_ioctl, | |
1955 | .compat_ioctl = loop_control_ioctl, | |
1956 | .owner = THIS_MODULE, | |
1957 | .llseek = noop_llseek, | |
1958 | }; | |
1959 | ||
1960 | static struct miscdevice loop_misc = { | |
1961 | .minor = LOOP_CTRL_MINOR, | |
1962 | .name = "loop-control", | |
1963 | .fops = &loop_ctl_fops, | |
1964 | }; | |
1965 | ||
1966 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
1967 | MODULE_ALIAS("devname:loop-control"); | |
1968 | ||
73285082 KC |
1969 | static int __init loop_init(void) |
1970 | { | |
a47653fc KC |
1971 | int i, nr; |
1972 | unsigned long range; | |
34dd82af | 1973 | struct loop_device *lo; |
770fe30a | 1974 | int err; |
a47653fc | 1975 | |
770fe30a KS |
1976 | err = misc_register(&loop_misc); |
1977 | if (err < 0) | |
1978 | return err; | |
476a4813 LV |
1979 | |
1980 | part_shift = 0; | |
ac04fee0 | 1981 | if (max_part > 0) { |
476a4813 LV |
1982 | part_shift = fls(max_part); |
1983 | ||
ac04fee0 NK |
1984 | /* |
1985 | * Adjust max_part according to part_shift as it is exported | |
1986 | * to user space so that user can decide correct minor number | |
1987 | * if [s]he want to create more devices. | |
1988 | * | |
1989 | * Note that -1 is required because partition 0 is reserved | |
1990 | * for the whole disk. | |
1991 | */ | |
1992 | max_part = (1UL << part_shift) - 1; | |
1993 | } | |
1994 | ||
b1a66504 GC |
1995 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
1996 | err = -EINVAL; | |
1997 | goto misc_out; | |
1998 | } | |
78f4bb36 | 1999 | |
b1a66504 GC |
2000 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2001 | err = -EINVAL; | |
2002 | goto misc_out; | |
2003 | } | |
1da177e4 | 2004 | |
d134b00b KS |
2005 | /* |
2006 | * If max_loop is specified, create that many devices upfront. | |
2007 | * This also becomes a hard limit. If max_loop is not specified, | |
2008 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
2009 | * init time. Loop devices can be requested on-demand with the | |
2010 | * /dev/loop-control interface, or be instantiated by accessing | |
2011 | * a 'dead' device node. | |
2012 | */ | |
73285082 | 2013 | if (max_loop) { |
a47653fc | 2014 | nr = max_loop; |
a1c15c59 | 2015 | range = max_loop << part_shift; |
a47653fc | 2016 | } else { |
d134b00b | 2017 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 2018 | range = 1UL << MINORBITS; |
a47653fc KC |
2019 | } |
2020 | ||
b1a66504 GC |
2021 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
2022 | err = -EIO; | |
2023 | goto misc_out; | |
2024 | } | |
1da177e4 | 2025 | |
a47653fc KC |
2026 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
2027 | THIS_MODULE, loop_probe, NULL, NULL); | |
2028 | ||
d134b00b | 2029 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
2030 | mutex_lock(&loop_index_mutex); |
2031 | for (i = 0; i < nr; i++) | |
2032 | loop_add(&lo, i); | |
2033 | mutex_unlock(&loop_index_mutex); | |
2034 | ||
73285082 | 2035 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2036 | return 0; |
b1a66504 GC |
2037 | |
2038 | misc_out: | |
2039 | misc_deregister(&loop_misc); | |
2040 | return err; | |
34dd82af | 2041 | } |
a47653fc | 2042 | |
34dd82af KS |
2043 | static int loop_exit_cb(int id, void *ptr, void *data) |
2044 | { | |
2045 | struct loop_device *lo = ptr; | |
a47653fc | 2046 | |
34dd82af KS |
2047 | loop_remove(lo); |
2048 | return 0; | |
1da177e4 LT |
2049 | } |
2050 | ||
73285082 | 2051 | static void __exit loop_exit(void) |
1da177e4 | 2052 | { |
a47653fc | 2053 | unsigned long range; |
1da177e4 | 2054 | |
a1c15c59 | 2055 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 2056 | |
34dd82af | 2057 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 2058 | idr_destroy(&loop_index_idr); |
73285082 | 2059 | |
a47653fc | 2060 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2061 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
2062 | |
2063 | misc_deregister(&loop_misc); | |
1da177e4 LT |
2064 | } |
2065 | ||
2066 | module_init(loop_init); | |
2067 | module_exit(loop_exit); | |
2068 | ||
2069 | #ifndef MODULE | |
2070 | static int __init max_loop_setup(char *str) | |
2071 | { | |
2072 | max_loop = simple_strtol(str, NULL, 0); | |
2073 | return 1; | |
2074 | } | |
2075 | ||
2076 | __setup("max_loop=", max_loop_setup); | |
2077 | #endif |