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> | |
66 | #include <linux/loop.h> | |
863d5b82 | 67 | #include <linux/compat.h> |
1da177e4 | 68 | #include <linux/suspend.h> |
83144186 | 69 | #include <linux/freezer.h> |
2a48fc0a | 70 | #include <linux/mutex.h> |
1da177e4 | 71 | #include <linux/writeback.h> |
1da177e4 LT |
72 | #include <linux/completion.h> |
73 | #include <linux/highmem.h> | |
6c997918 | 74 | #include <linux/kthread.h> |
d6b29d7c | 75 | #include <linux/splice.h> |
ee862730 | 76 | #include <linux/sysfs.h> |
770fe30a | 77 | #include <linux/miscdevice.h> |
dfaa2ef6 | 78 | #include <linux/falloc.h> |
1da177e4 LT |
79 | |
80 | #include <asm/uaccess.h> | |
81 | ||
34dd82af KS |
82 | static DEFINE_IDR(loop_index_idr); |
83 | static DEFINE_MUTEX(loop_index_mutex); | |
1da177e4 | 84 | |
476a4813 LV |
85 | static int max_part; |
86 | static int part_shift; | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * Transfer functions | |
90 | */ | |
91 | static int transfer_none(struct loop_device *lo, int cmd, | |
92 | struct page *raw_page, unsigned raw_off, | |
93 | struct page *loop_page, unsigned loop_off, | |
94 | int size, sector_t real_block) | |
95 | { | |
cfd8005c CW |
96 | char *raw_buf = kmap_atomic(raw_page) + raw_off; |
97 | char *loop_buf = kmap_atomic(loop_page) + loop_off; | |
1da177e4 LT |
98 | |
99 | if (cmd == READ) | |
100 | memcpy(loop_buf, raw_buf, size); | |
101 | else | |
102 | memcpy(raw_buf, loop_buf, size); | |
103 | ||
cfd8005c CW |
104 | kunmap_atomic(loop_buf); |
105 | kunmap_atomic(raw_buf); | |
1da177e4 LT |
106 | cond_resched(); |
107 | return 0; | |
108 | } | |
109 | ||
110 | static int transfer_xor(struct loop_device *lo, int cmd, | |
111 | struct page *raw_page, unsigned raw_off, | |
112 | struct page *loop_page, unsigned loop_off, | |
113 | int size, sector_t real_block) | |
114 | { | |
cfd8005c CW |
115 | char *raw_buf = kmap_atomic(raw_page) + raw_off; |
116 | char *loop_buf = kmap_atomic(loop_page) + loop_off; | |
1da177e4 LT |
117 | char *in, *out, *key; |
118 | int i, keysize; | |
119 | ||
120 | if (cmd == READ) { | |
121 | in = raw_buf; | |
122 | out = loop_buf; | |
123 | } else { | |
124 | in = loop_buf; | |
125 | out = raw_buf; | |
126 | } | |
127 | ||
128 | key = lo->lo_encrypt_key; | |
129 | keysize = lo->lo_encrypt_key_size; | |
130 | for (i = 0; i < size; i++) | |
131 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
132 | ||
cfd8005c CW |
133 | kunmap_atomic(loop_buf); |
134 | kunmap_atomic(raw_buf); | |
1da177e4 LT |
135 | cond_resched(); |
136 | return 0; | |
137 | } | |
138 | ||
139 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
140 | { | |
141 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
142 | return -EINVAL; | |
143 | return 0; | |
144 | } | |
145 | ||
146 | static struct loop_func_table none_funcs = { | |
147 | .number = LO_CRYPT_NONE, | |
148 | .transfer = transfer_none, | |
149 | }; | |
150 | ||
151 | static struct loop_func_table xor_funcs = { | |
152 | .number = LO_CRYPT_XOR, | |
153 | .transfer = transfer_xor, | |
154 | .init = xor_init | |
155 | }; | |
156 | ||
157 | /* xfer_funcs[0] is special - its release function is never called */ | |
158 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
159 | &none_funcs, | |
160 | &xor_funcs | |
161 | }; | |
162 | ||
7035b5df | 163 | static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) |
1da177e4 | 164 | { |
b7a1da69 | 165 | loff_t loopsize; |
1da177e4 LT |
166 | |
167 | /* Compute loopsize in bytes */ | |
b7a1da69 GC |
168 | loopsize = i_size_read(file->f_mapping->host); |
169 | if (offset > 0) | |
170 | loopsize -= offset; | |
171 | /* offset is beyond i_size, weird but possible */ | |
7035b5df DM |
172 | if (loopsize < 0) |
173 | return 0; | |
1da177e4 | 174 | |
7035b5df DM |
175 | if (sizelimit > 0 && sizelimit < loopsize) |
176 | loopsize = sizelimit; | |
1da177e4 LT |
177 | /* |
178 | * Unfortunately, if we want to do I/O on the device, | |
179 | * the number of 512-byte sectors has to fit into a sector_t. | |
180 | */ | |
181 | return loopsize >> 9; | |
182 | } | |
183 | ||
7035b5df DM |
184 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) |
185 | { | |
186 | return get_size(lo->lo_offset, lo->lo_sizelimit, file); | |
187 | } | |
188 | ||
1da177e4 | 189 | static int |
7035b5df | 190 | figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) |
1da177e4 | 191 | { |
7035b5df | 192 | loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); |
1da177e4 | 193 | sector_t x = (sector_t)size; |
7b0576a3 | 194 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
195 | |
196 | if (unlikely((loff_t)x != size)) | |
197 | return -EFBIG; | |
7035b5df DM |
198 | if (lo->lo_offset != offset) |
199 | lo->lo_offset = offset; | |
200 | if (lo->lo_sizelimit != sizelimit) | |
201 | lo->lo_sizelimit = sizelimit; | |
73285082 | 202 | set_capacity(lo->lo_disk, x); |
7b0576a3 GC |
203 | bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9); |
204 | /* let user-space know about the new size */ | |
205 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
7035b5df | 206 | return 0; |
1da177e4 LT |
207 | } |
208 | ||
209 | static inline int | |
210 | lo_do_transfer(struct loop_device *lo, int cmd, | |
211 | struct page *rpage, unsigned roffs, | |
212 | struct page *lpage, unsigned loffs, | |
213 | int size, sector_t rblock) | |
214 | { | |
215 | if (unlikely(!lo->transfer)) | |
216 | return 0; | |
217 | ||
218 | return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
219 | } | |
220 | ||
1da177e4 LT |
221 | /** |
222 | * __do_lo_send_write - helper for writing data to a loop device | |
223 | * | |
224 | * This helper just factors out common code between do_lo_send_direct_write() | |
225 | * and do_lo_send_write(). | |
226 | */ | |
858119e1 | 227 | static int __do_lo_send_write(struct file *file, |
98ae6ccd | 228 | u8 *buf, const int len, loff_t pos) |
1da177e4 LT |
229 | { |
230 | ssize_t bw; | |
231 | mm_segment_t old_fs = get_fs(); | |
232 | ||
233 | set_fs(get_ds()); | |
234 | bw = file->f_op->write(file, buf, len, &pos); | |
235 | set_fs(old_fs); | |
236 | if (likely(bw == len)) | |
237 | return 0; | |
238 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", | |
239 | (unsigned long long)pos, len); | |
240 | if (bw >= 0) | |
241 | bw = -EIO; | |
242 | return bw; | |
243 | } | |
244 | ||
245 | /** | |
246 | * do_lo_send_direct_write - helper for writing data to a loop device | |
247 | * | |
456be148 CH |
248 | * This is the fast, non-transforming version that does not need double |
249 | * buffering. | |
1da177e4 LT |
250 | */ |
251 | static int do_lo_send_direct_write(struct loop_device *lo, | |
511de73f | 252 | struct bio_vec *bvec, loff_t pos, struct page *page) |
1da177e4 LT |
253 | { |
254 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 255 | kmap(bvec->bv_page) + bvec->bv_offset, |
1da177e4 LT |
256 | bvec->bv_len, pos); |
257 | kunmap(bvec->bv_page); | |
258 | cond_resched(); | |
259 | return bw; | |
260 | } | |
261 | ||
262 | /** | |
263 | * do_lo_send_write - helper for writing data to a loop device | |
264 | * | |
456be148 CH |
265 | * This is the slow, transforming version that needs to double buffer the |
266 | * data as it cannot do the transformations in place without having direct | |
267 | * access to the destination pages of the backing file. | |
1da177e4 LT |
268 | */ |
269 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 270 | loff_t pos, struct page *page) |
1da177e4 LT |
271 | { |
272 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, | |
273 | bvec->bv_offset, bvec->bv_len, pos >> 9); | |
274 | if (likely(!ret)) | |
275 | return __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 276 | page_address(page), bvec->bv_len, |
1da177e4 LT |
277 | pos); |
278 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, " | |
279 | "length %i.\n", (unsigned long long)pos, bvec->bv_len); | |
280 | if (ret > 0) | |
281 | ret = -EIO; | |
282 | return ret; | |
283 | } | |
284 | ||
511de73f | 285 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) |
1da177e4 | 286 | { |
511de73f | 287 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, |
1da177e4 LT |
288 | struct page *page); |
289 | struct bio_vec *bvec; | |
290 | struct page *page = NULL; | |
291 | int i, ret = 0; | |
292 | ||
456be148 CH |
293 | if (lo->transfer != transfer_none) { |
294 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
295 | if (unlikely(!page)) | |
296 | goto fail; | |
297 | kmap(page); | |
298 | do_lo_send = do_lo_send_write; | |
299 | } else { | |
1da177e4 | 300 | do_lo_send = do_lo_send_direct_write; |
1da177e4 | 301 | } |
456be148 | 302 | |
1da177e4 | 303 | bio_for_each_segment(bvec, bio, i) { |
511de73f | 304 | ret = do_lo_send(lo, bvec, pos, page); |
1da177e4 LT |
305 | if (ret < 0) |
306 | break; | |
307 | pos += bvec->bv_len; | |
308 | } | |
309 | if (page) { | |
310 | kunmap(page); | |
311 | __free_page(page); | |
312 | } | |
313 | out: | |
314 | return ret; | |
315 | fail: | |
316 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); | |
317 | ret = -ENOMEM; | |
318 | goto out; | |
319 | } | |
320 | ||
321 | struct lo_read_data { | |
322 | struct loop_device *lo; | |
323 | struct page *page; | |
324 | unsigned offset; | |
325 | int bsize; | |
326 | }; | |
327 | ||
328 | static int | |
fd582140 JA |
329 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
330 | struct splice_desc *sd) | |
1da177e4 | 331 | { |
fd582140 | 332 | struct lo_read_data *p = sd->u.data; |
1da177e4 | 333 | struct loop_device *lo = p->lo; |
fd582140 | 334 | struct page *page = buf->page; |
1da177e4 | 335 | sector_t IV; |
3603b8ea | 336 | int size; |
1da177e4 | 337 | |
fd582140 JA |
338 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + |
339 | (buf->offset >> 9); | |
340 | size = sd->len; | |
341 | if (size > p->bsize) | |
342 | size = p->bsize; | |
1da177e4 | 343 | |
fd582140 | 344 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { |
1da177e4 LT |
345 | printk(KERN_ERR "loop: transfer error block %ld\n", |
346 | page->index); | |
fd582140 | 347 | size = -EINVAL; |
1da177e4 LT |
348 | } |
349 | ||
350 | flush_dcache_page(p->page); | |
351 | ||
fd582140 JA |
352 | if (size > 0) |
353 | p->offset += size; | |
354 | ||
1da177e4 LT |
355 | return size; |
356 | } | |
357 | ||
fd582140 JA |
358 | static int |
359 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
360 | { | |
361 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
362 | } | |
363 | ||
306df071 | 364 | static ssize_t |
1da177e4 LT |
365 | do_lo_receive(struct loop_device *lo, |
366 | struct bio_vec *bvec, int bsize, loff_t pos) | |
367 | { | |
368 | struct lo_read_data cookie; | |
fd582140 | 369 | struct splice_desc sd; |
1da177e4 | 370 | struct file *file; |
306df071 | 371 | ssize_t retval; |
1da177e4 LT |
372 | |
373 | cookie.lo = lo; | |
374 | cookie.page = bvec->bv_page; | |
375 | cookie.offset = bvec->bv_offset; | |
376 | cookie.bsize = bsize; | |
fd582140 JA |
377 | |
378 | sd.len = 0; | |
379 | sd.total_len = bvec->bv_len; | |
380 | sd.flags = 0; | |
381 | sd.pos = pos; | |
382 | sd.u.data = &cookie; | |
383 | ||
1da177e4 | 384 | file = lo->lo_backing_file; |
fd582140 JA |
385 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); |
386 | ||
306df071 | 387 | return retval; |
1da177e4 LT |
388 | } |
389 | ||
390 | static int | |
391 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
392 | { | |
393 | struct bio_vec *bvec; | |
306df071 DY |
394 | ssize_t s; |
395 | int i; | |
1da177e4 LT |
396 | |
397 | bio_for_each_segment(bvec, bio, i) { | |
306df071 DY |
398 | s = do_lo_receive(lo, bvec, bsize, pos); |
399 | if (s < 0) | |
400 | return s; | |
401 | ||
402 | if (s != bvec->bv_len) { | |
403 | zero_fill_bio(bio); | |
1da177e4 | 404 | break; |
306df071 | 405 | } |
1da177e4 LT |
406 | pos += bvec->bv_len; |
407 | } | |
306df071 | 408 | return 0; |
1da177e4 LT |
409 | } |
410 | ||
411 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
412 | { | |
413 | loff_t pos; | |
414 | int ret; | |
415 | ||
416 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
68db1961 NK |
417 | |
418 | if (bio_rw(bio) == WRITE) { | |
68db1961 NK |
419 | struct file *file = lo->lo_backing_file; |
420 | ||
6259f284 | 421 | if (bio->bi_rw & REQ_FLUSH) { |
8018ab05 | 422 | ret = vfs_fsync(file, 0); |
6259f284 | 423 | if (unlikely(ret && ret != -EINVAL)) { |
68db1961 NK |
424 | ret = -EIO; |
425 | goto out; | |
426 | } | |
427 | } | |
428 | ||
dfaa2ef6 LC |
429 | /* |
430 | * We use punch hole to reclaim the free space used by the | |
ae95757a | 431 | * image a.k.a. discard. However we do not support discard if |
dfaa2ef6 LC |
432 | * encryption is enabled, because it may give an attacker |
433 | * useful information. | |
434 | */ | |
435 | if (bio->bi_rw & REQ_DISCARD) { | |
436 | struct file *file = lo->lo_backing_file; | |
437 | int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; | |
438 | ||
439 | if ((!file->f_op->fallocate) || | |
440 | lo->lo_encrypt_key_size) { | |
441 | ret = -EOPNOTSUPP; | |
442 | goto out; | |
443 | } | |
444 | ret = file->f_op->fallocate(file, mode, pos, | |
445 | bio->bi_size); | |
446 | if (unlikely(ret && ret != -EINVAL && | |
447 | ret != -EOPNOTSUPP)) | |
448 | ret = -EIO; | |
449 | goto out; | |
450 | } | |
451 | ||
511de73f | 452 | ret = lo_send(lo, bio, pos); |
68db1961 | 453 | |
6259f284 | 454 | if ((bio->bi_rw & REQ_FUA) && !ret) { |
8018ab05 | 455 | ret = vfs_fsync(file, 0); |
6259f284 | 456 | if (unlikely(ret && ret != -EINVAL)) |
68db1961 NK |
457 | ret = -EIO; |
458 | } | |
459 | } else | |
1da177e4 | 460 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); |
68db1961 NK |
461 | |
462 | out: | |
1da177e4 LT |
463 | return ret; |
464 | } | |
465 | ||
466 | /* | |
467 | * Add bio to back of pending list | |
468 | */ | |
469 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
470 | { | |
7b5a3522 | 471 | lo->lo_bio_count++; |
e686307f | 472 | bio_list_add(&lo->lo_bio_list, bio); |
1da177e4 LT |
473 | } |
474 | ||
475 | /* | |
476 | * Grab first pending buffer | |
477 | */ | |
478 | static struct bio *loop_get_bio(struct loop_device *lo) | |
479 | { | |
7b5a3522 | 480 | lo->lo_bio_count--; |
e686307f | 481 | return bio_list_pop(&lo->lo_bio_list); |
1da177e4 LT |
482 | } |
483 | ||
5a7bbad2 | 484 | static void loop_make_request(struct request_queue *q, struct bio *old_bio) |
1da177e4 LT |
485 | { |
486 | struct loop_device *lo = q->queuedata; | |
487 | int rw = bio_rw(old_bio); | |
488 | ||
35a82d1a NP |
489 | if (rw == READA) |
490 | rw = READ; | |
491 | ||
492 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
1da177e4 LT |
493 | |
494 | spin_lock_irq(&lo->lo_lock); | |
495 | if (lo->lo_state != Lo_bound) | |
35a82d1a NP |
496 | goto out; |
497 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
498 | goto out; | |
7b5a3522 LC |
499 | if (lo->lo_bio_count >= q->nr_congestion_on) |
500 | wait_event_lock_irq(lo->lo_req_wait, | |
501 | lo->lo_bio_count < q->nr_congestion_off, | |
502 | lo->lo_lock); | |
1da177e4 | 503 | loop_add_bio(lo, old_bio); |
6c997918 | 504 | wake_up(&lo->lo_event); |
35a82d1a | 505 | spin_unlock_irq(&lo->lo_lock); |
5a7bbad2 | 506 | return; |
35a82d1a | 507 | |
1da177e4 | 508 | out: |
35a82d1a | 509 | spin_unlock_irq(&lo->lo_lock); |
6712ecf8 | 510 | bio_io_error(old_bio); |
1da177e4 LT |
511 | } |
512 | ||
1da177e4 LT |
513 | struct switch_request { |
514 | struct file *file; | |
515 | struct completion wait; | |
516 | }; | |
517 | ||
518 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
519 | ||
520 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
521 | { | |
1da177e4 LT |
522 | if (unlikely(!bio->bi_bdev)) { |
523 | do_loop_switch(lo, bio->bi_private); | |
524 | bio_put(bio); | |
525 | } else { | |
35a82d1a | 526 | int ret = do_bio_filebacked(lo, bio); |
6712ecf8 | 527 | bio_endio(bio, ret); |
1da177e4 LT |
528 | } |
529 | } | |
530 | ||
531 | /* | |
532 | * worker thread that handles reads/writes to file backed loop devices, | |
533 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
534 | * on reads for block backed loop, as that is too heavy to do from | |
535 | * b_end_io context where irqs may be disabled. | |
6c997918 SH |
536 | * |
537 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
538 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
539 | * true, make_request will not place any more requests. Therefore | |
540 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
541 | * done with the loop. | |
1da177e4 LT |
542 | */ |
543 | static int loop_thread(void *data) | |
544 | { | |
545 | struct loop_device *lo = data; | |
546 | struct bio *bio; | |
547 | ||
1da177e4 LT |
548 | set_user_nice(current, -20); |
549 | ||
e686307f | 550 | while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) { |
09c0dc68 | 551 | |
6c997918 | 552 | wait_event_interruptible(lo->lo_event, |
e686307f AM |
553 | !bio_list_empty(&lo->lo_bio_list) || |
554 | kthread_should_stop()); | |
35a82d1a | 555 | |
e686307f | 556 | if (bio_list_empty(&lo->lo_bio_list)) |
35a82d1a | 557 | continue; |
35a82d1a | 558 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 559 | bio = loop_get_bio(lo); |
7b5a3522 LC |
560 | if (lo->lo_bio_count < lo->lo_queue->nr_congestion_off) |
561 | wake_up(&lo->lo_req_wait); | |
35a82d1a NP |
562 | spin_unlock_irq(&lo->lo_lock); |
563 | ||
564 | BUG_ON(!bio); | |
1da177e4 | 565 | loop_handle_bio(lo, bio); |
1da177e4 LT |
566 | } |
567 | ||
1da177e4 LT |
568 | return 0; |
569 | } | |
570 | ||
571 | /* | |
572 | * loop_switch performs the hard work of switching a backing store. | |
573 | * First it needs to flush existing IO, it does this by sending a magic | |
574 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
575 | */ | |
576 | static int loop_switch(struct loop_device *lo, struct file *file) | |
577 | { | |
578 | struct switch_request w; | |
a24eab1e | 579 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); |
1da177e4 LT |
580 | if (!bio) |
581 | return -ENOMEM; | |
582 | init_completion(&w.wait); | |
583 | w.file = file; | |
584 | bio->bi_private = &w; | |
585 | bio->bi_bdev = NULL; | |
586 | loop_make_request(lo->lo_queue, bio); | |
587 | wait_for_completion(&w.wait); | |
588 | return 0; | |
589 | } | |
590 | ||
14f27939 MB |
591 | /* |
592 | * Helper to flush the IOs in loop, but keeping loop thread running | |
593 | */ | |
594 | static int loop_flush(struct loop_device *lo) | |
595 | { | |
596 | /* loop not yet configured, no running thread, nothing to flush */ | |
597 | if (!lo->lo_thread) | |
598 | return 0; | |
599 | ||
600 | return loop_switch(lo, NULL); | |
601 | } | |
602 | ||
1da177e4 LT |
603 | /* |
604 | * Do the actual switch; called from the BIO completion routine | |
605 | */ | |
606 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
607 | { | |
608 | struct file *file = p->file; | |
609 | struct file *old_file = lo->lo_backing_file; | |
14f27939 MB |
610 | struct address_space *mapping; |
611 | ||
612 | /* if no new file, only flush of queued bios requested */ | |
613 | if (!file) | |
614 | goto out; | |
1da177e4 | 615 | |
14f27939 | 616 | mapping = file->f_mapping; |
1da177e4 LT |
617 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); |
618 | lo->lo_backing_file = file; | |
ba52de12 TT |
619 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
620 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 LT |
621 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
622 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
14f27939 | 623 | out: |
1da177e4 LT |
624 | complete(&p->wait); |
625 | } | |
626 | ||
627 | ||
628 | /* | |
629 | * loop_change_fd switched the backing store of a loopback device to | |
630 | * a new file. This is useful for operating system installers to free up | |
631 | * the original file and in High Availability environments to switch to | |
632 | * an alternative location for the content in case of server meltdown. | |
633 | * This can only work if the loop device is used read-only, and if the | |
634 | * new backing store is the same size and type as the old backing store. | |
635 | */ | |
bb214884 AV |
636 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
637 | unsigned int arg) | |
1da177e4 LT |
638 | { |
639 | struct file *file, *old_file; | |
640 | struct inode *inode; | |
641 | int error; | |
642 | ||
643 | error = -ENXIO; | |
644 | if (lo->lo_state != Lo_bound) | |
645 | goto out; | |
646 | ||
647 | /* the loop device has to be read-only */ | |
648 | error = -EINVAL; | |
649 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
650 | goto out; | |
651 | ||
652 | error = -EBADF; | |
653 | file = fget(arg); | |
654 | if (!file) | |
655 | goto out; | |
656 | ||
657 | inode = file->f_mapping->host; | |
658 | old_file = lo->lo_backing_file; | |
659 | ||
660 | error = -EINVAL; | |
661 | ||
662 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
663 | goto out_putf; | |
664 | ||
1da177e4 LT |
665 | /* size of the new backing store needs to be the same */ |
666 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
667 | goto out_putf; | |
668 | ||
669 | /* and ... switch */ | |
670 | error = loop_switch(lo, file); | |
671 | if (error) | |
672 | goto out_putf; | |
673 | ||
674 | fput(old_file); | |
e03c8dd1 | 675 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
476a4813 | 676 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
1da177e4 LT |
677 | return 0; |
678 | ||
679 | out_putf: | |
680 | fput(file); | |
681 | out: | |
682 | return error; | |
683 | } | |
684 | ||
685 | static inline int is_loop_device(struct file *file) | |
686 | { | |
687 | struct inode *i = file->f_mapping->host; | |
688 | ||
689 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
690 | } | |
691 | ||
ee862730 MB |
692 | /* loop sysfs attributes */ |
693 | ||
694 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
695 | ssize_t (*callback)(struct loop_device *, char *)) | |
696 | { | |
34dd82af KS |
697 | struct gendisk *disk = dev_to_disk(dev); |
698 | struct loop_device *lo = disk->private_data; | |
ee862730 | 699 | |
34dd82af | 700 | return callback(lo, page); |
ee862730 MB |
701 | } |
702 | ||
703 | #define LOOP_ATTR_RO(_name) \ | |
704 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
705 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
706 | struct device_attribute *attr, char *b) \ | |
707 | { \ | |
708 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
709 | } \ | |
710 | static struct device_attribute loop_attr_##_name = \ | |
711 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
712 | ||
713 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
714 | { | |
715 | ssize_t ret; | |
716 | char *p = NULL; | |
717 | ||
05eb0f25 | 718 | spin_lock_irq(&lo->lo_lock); |
ee862730 MB |
719 | if (lo->lo_backing_file) |
720 | p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1); | |
05eb0f25 | 721 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
722 | |
723 | if (IS_ERR_OR_NULL(p)) | |
724 | ret = PTR_ERR(p); | |
725 | else { | |
726 | ret = strlen(p); | |
727 | memmove(buf, p, ret); | |
728 | buf[ret++] = '\n'; | |
729 | buf[ret] = 0; | |
730 | } | |
731 | ||
732 | return ret; | |
733 | } | |
734 | ||
735 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
736 | { | |
737 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
738 | } | |
739 | ||
740 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
741 | { | |
742 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
743 | } | |
744 | ||
745 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
746 | { | |
747 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
748 | ||
749 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
750 | } | |
751 | ||
e03c8dd1 KS |
752 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
753 | { | |
754 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
755 | ||
756 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
757 | } | |
758 | ||
ee862730 MB |
759 | LOOP_ATTR_RO(backing_file); |
760 | LOOP_ATTR_RO(offset); | |
761 | LOOP_ATTR_RO(sizelimit); | |
762 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 763 | LOOP_ATTR_RO(partscan); |
ee862730 MB |
764 | |
765 | static struct attribute *loop_attrs[] = { | |
766 | &loop_attr_backing_file.attr, | |
767 | &loop_attr_offset.attr, | |
768 | &loop_attr_sizelimit.attr, | |
769 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 770 | &loop_attr_partscan.attr, |
ee862730 MB |
771 | NULL, |
772 | }; | |
773 | ||
774 | static struct attribute_group loop_attribute_group = { | |
775 | .name = "loop", | |
776 | .attrs= loop_attrs, | |
777 | }; | |
778 | ||
779 | static int loop_sysfs_init(struct loop_device *lo) | |
780 | { | |
781 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
782 | &loop_attribute_group); | |
783 | } | |
784 | ||
785 | static void loop_sysfs_exit(struct loop_device *lo) | |
786 | { | |
787 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
788 | &loop_attribute_group); | |
789 | } | |
790 | ||
dfaa2ef6 LC |
791 | static void loop_config_discard(struct loop_device *lo) |
792 | { | |
793 | struct file *file = lo->lo_backing_file; | |
794 | struct inode *inode = file->f_mapping->host; | |
795 | struct request_queue *q = lo->lo_queue; | |
796 | ||
797 | /* | |
798 | * We use punch hole to reclaim the free space used by the | |
799 | * image a.k.a. discard. However we do support discard if | |
800 | * encryption is enabled, because it may give an attacker | |
801 | * useful information. | |
802 | */ | |
803 | if ((!file->f_op->fallocate) || | |
804 | lo->lo_encrypt_key_size) { | |
805 | q->limits.discard_granularity = 0; | |
806 | q->limits.discard_alignment = 0; | |
807 | q->limits.max_discard_sectors = 0; | |
808 | q->limits.discard_zeroes_data = 0; | |
809 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
810 | return; | |
811 | } | |
812 | ||
813 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
dfaf3c03 | 814 | q->limits.discard_alignment = 0; |
dfaa2ef6 LC |
815 | q->limits.max_discard_sectors = UINT_MAX >> 9; |
816 | q->limits.discard_zeroes_data = 1; | |
817 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
818 | } | |
819 | ||
bb214884 | 820 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
821 | struct block_device *bdev, unsigned int arg) |
822 | { | |
823 | struct file *file, *f; | |
824 | struct inode *inode; | |
825 | struct address_space *mapping; | |
826 | unsigned lo_blocksize; | |
827 | int lo_flags = 0; | |
828 | int error; | |
829 | loff_t size; | |
830 | ||
831 | /* This is safe, since we have a reference from open(). */ | |
832 | __module_get(THIS_MODULE); | |
833 | ||
834 | error = -EBADF; | |
835 | file = fget(arg); | |
836 | if (!file) | |
837 | goto out; | |
838 | ||
839 | error = -EBUSY; | |
840 | if (lo->lo_state != Lo_unbound) | |
841 | goto out_putf; | |
842 | ||
843 | /* Avoid recursion */ | |
844 | f = file; | |
845 | while (is_loop_device(f)) { | |
846 | struct loop_device *l; | |
847 | ||
bb214884 | 848 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
849 | goto out_putf; |
850 | ||
851 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
852 | if (l->lo_state == Lo_unbound) { | |
853 | error = -EINVAL; | |
854 | goto out_putf; | |
855 | } | |
856 | f = l->lo_backing_file; | |
857 | } | |
858 | ||
859 | mapping = file->f_mapping; | |
860 | inode = mapping->host; | |
861 | ||
1da177e4 | 862 | error = -EINVAL; |
456be148 CH |
863 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) |
864 | goto out_putf; | |
1da177e4 | 865 | |
456be148 CH |
866 | if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || |
867 | !file->f_op->write) | |
868 | lo_flags |= LO_FLAGS_READ_ONLY; | |
ba52de12 | 869 | |
456be148 CH |
870 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
871 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 | 872 | |
456be148 | 873 | error = -EFBIG; |
1da177e4 | 874 | size = get_loop_size(lo, file); |
456be148 | 875 | if ((loff_t)(sector_t)size != size) |
1da177e4 | 876 | goto out_putf; |
1da177e4 | 877 | |
456be148 | 878 | error = 0; |
1da177e4 LT |
879 | |
880 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
881 | ||
882 | lo->lo_blocksize = lo_blocksize; | |
883 | lo->lo_device = bdev; | |
884 | lo->lo_flags = lo_flags; | |
885 | lo->lo_backing_file = file; | |
eefe85ee | 886 | lo->transfer = transfer_none; |
1da177e4 LT |
887 | lo->ioctl = NULL; |
888 | lo->lo_sizelimit = 0; | |
7b5a3522 | 889 | lo->lo_bio_count = 0; |
1da177e4 LT |
890 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
891 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
892 | ||
e686307f | 893 | bio_list_init(&lo->lo_bio_list); |
1da177e4 LT |
894 | |
895 | /* | |
896 | * set queue make_request_fn, and add limits based on lower level | |
897 | * device | |
898 | */ | |
899 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
900 | lo->lo_queue->queuedata = lo; | |
1da177e4 | 901 | |
68db1961 | 902 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
4913efe4 | 903 | blk_queue_flush(lo->lo_queue, REQ_FLUSH); |
68db1961 | 904 | |
73285082 | 905 | set_capacity(lo->lo_disk, size); |
1da177e4 | 906 | bd_set_size(bdev, size << 9); |
ee862730 | 907 | loop_sysfs_init(lo); |
c3473c63 DZ |
908 | /* let user-space know about the new size */ |
909 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 LT |
910 | |
911 | set_blocksize(bdev, lo_blocksize); | |
912 | ||
6c997918 SH |
913 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", |
914 | lo->lo_number); | |
915 | if (IS_ERR(lo->lo_thread)) { | |
916 | error = PTR_ERR(lo->lo_thread); | |
a7422bf8 | 917 | goto out_clr; |
6c997918 SH |
918 | } |
919 | lo->lo_state = Lo_bound; | |
920 | wake_up_process(lo->lo_thread); | |
e03c8dd1 KS |
921 | if (part_shift) |
922 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
923 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
476a4813 | 924 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
1da177e4 LT |
925 | return 0; |
926 | ||
a7422bf8 | 927 | out_clr: |
ee862730 | 928 | loop_sysfs_exit(lo); |
a7422bf8 SH |
929 | lo->lo_thread = NULL; |
930 | lo->lo_device = NULL; | |
931 | lo->lo_backing_file = NULL; | |
932 | lo->lo_flags = 0; | |
73285082 | 933 | set_capacity(lo->lo_disk, 0); |
f98393a6 | 934 | invalidate_bdev(bdev); |
a7422bf8 | 935 | bd_set_size(bdev, 0); |
c3473c63 | 936 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); |
a7422bf8 SH |
937 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); |
938 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
939 | out_putf: |
940 | fput(file); | |
941 | out: | |
942 | /* This is safe: open() is still holding a reference. */ | |
943 | module_put(THIS_MODULE); | |
944 | return error; | |
945 | } | |
946 | ||
947 | static int | |
948 | loop_release_xfer(struct loop_device *lo) | |
949 | { | |
950 | int err = 0; | |
951 | struct loop_func_table *xfer = lo->lo_encryption; | |
952 | ||
953 | if (xfer) { | |
954 | if (xfer->release) | |
955 | err = xfer->release(lo); | |
956 | lo->transfer = NULL; | |
957 | lo->lo_encryption = NULL; | |
958 | module_put(xfer->owner); | |
959 | } | |
960 | return err; | |
961 | } | |
962 | ||
963 | static int | |
964 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
965 | const struct loop_info64 *i) | |
966 | { | |
967 | int err = 0; | |
968 | ||
969 | if (xfer) { | |
970 | struct module *owner = xfer->owner; | |
971 | ||
972 | if (!try_module_get(owner)) | |
973 | return -EINVAL; | |
974 | if (xfer->init) | |
975 | err = xfer->init(lo, i); | |
976 | if (err) | |
977 | module_put(owner); | |
978 | else | |
979 | lo->lo_encryption = xfer; | |
980 | } | |
981 | return err; | |
982 | } | |
983 | ||
4c823cc3 | 984 | static int loop_clr_fd(struct loop_device *lo) |
1da177e4 LT |
985 | { |
986 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 987 | gfp_t gfp = lo->old_gfp_mask; |
4c823cc3 | 988 | struct block_device *bdev = lo->lo_device; |
1da177e4 LT |
989 | |
990 | if (lo->lo_state != Lo_bound) | |
991 | return -ENXIO; | |
992 | ||
a1ecac3b DC |
993 | /* |
994 | * If we've explicitly asked to tear down the loop device, | |
995 | * and it has an elevated reference count, set it for auto-teardown when | |
996 | * the last reference goes away. This stops $!~#$@ udev from | |
997 | * preventing teardown because it decided that it needs to run blkid on | |
998 | * the loopback device whenever they appear. xfstests is notorious for | |
999 | * failing tests because blkid via udev races with a losetup | |
1000 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1001 | * command to fail with EBUSY. | |
1002 | */ | |
1003 | if (lo->lo_refcnt > 1) { | |
1004 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
1005 | mutex_unlock(&lo->lo_ctl_mutex); | |
1006 | return 0; | |
1007 | } | |
1da177e4 LT |
1008 | |
1009 | if (filp == NULL) | |
1010 | return -EINVAL; | |
1011 | ||
1012 | spin_lock_irq(&lo->lo_lock); | |
1013 | lo->lo_state = Lo_rundown; | |
1da177e4 LT |
1014 | spin_unlock_irq(&lo->lo_lock); |
1015 | ||
6c997918 | 1016 | kthread_stop(lo->lo_thread); |
1da177e4 | 1017 | |
05eb0f25 | 1018 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 1019 | lo->lo_backing_file = NULL; |
05eb0f25 | 1020 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 LT |
1021 | |
1022 | loop_release_xfer(lo); | |
1023 | lo->transfer = NULL; | |
1024 | lo->ioctl = NULL; | |
1025 | lo->lo_device = NULL; | |
1026 | lo->lo_encryption = NULL; | |
1027 | lo->lo_offset = 0; | |
1028 | lo->lo_sizelimit = 0; | |
1029 | lo->lo_encrypt_key_size = 0; | |
6c997918 | 1030 | lo->lo_thread = NULL; |
1da177e4 LT |
1031 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1032 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1033 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
bb214884 AV |
1034 | if (bdev) |
1035 | invalidate_bdev(bdev); | |
73285082 | 1036 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1037 | loop_sysfs_exit(lo); |
c3473c63 | 1038 | if (bdev) { |
bb214884 | 1039 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1040 | /* let user-space know about this change */ |
1041 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1042 | } | |
1da177e4 LT |
1043 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1044 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1045 | /* This is safe: open() is still holding a reference. */ |
1046 | module_put(THIS_MODULE); | |
e03c8dd1 KS |
1047 | lo->lo_flags = 0; |
1048 | if (!part_shift) | |
1049 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
f028f3b2 | 1050 | mutex_unlock(&lo->lo_ctl_mutex); |
8761a3dc PS |
1051 | |
1052 | /* | |
1053 | * Remove all partitions, since BLKRRPART won't remove user | |
1054 | * added partitions when max_part=0 | |
1055 | */ | |
1056 | if (bdev) { | |
1057 | struct disk_part_iter piter; | |
1058 | struct hd_struct *part; | |
1059 | ||
1060 | mutex_lock_nested(&bdev->bd_mutex, 1); | |
1061 | invalidate_partition(bdev->bd_disk, 0); | |
1062 | disk_part_iter_init(&piter, bdev->bd_disk, | |
1063 | DISK_PITER_INCL_EMPTY); | |
1064 | while ((part = disk_part_iter_next(&piter))) | |
1065 | delete_partition(bdev->bd_disk, part->partno); | |
1066 | disk_part_iter_exit(&piter); | |
1067 | mutex_unlock(&bdev->bd_mutex); | |
1068 | } | |
1069 | ||
f028f3b2 NK |
1070 | /* |
1071 | * Need not hold lo_ctl_mutex to fput backing file. | |
1072 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1073 | * lock dependency possibility warning as fput can take | |
1074 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1075 | */ | |
1076 | fput(filp); | |
1da177e4 LT |
1077 | return 0; |
1078 | } | |
1079 | ||
1080 | static int | |
1081 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1082 | { | |
1083 | int err; | |
1084 | struct loop_func_table *xfer; | |
e4849737 | 1085 | kuid_t uid = current_uid(); |
1da177e4 | 1086 | |
b0fafa81 | 1087 | if (lo->lo_encrypt_key_size && |
e4849737 | 1088 | !uid_eq(lo->lo_key_owner, uid) && |
1da177e4 LT |
1089 | !capable(CAP_SYS_ADMIN)) |
1090 | return -EPERM; | |
1091 | if (lo->lo_state != Lo_bound) | |
1092 | return -ENXIO; | |
1093 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1094 | return -EINVAL; | |
1095 | ||
1096 | err = loop_release_xfer(lo); | |
1097 | if (err) | |
1098 | return err; | |
1099 | ||
1100 | if (info->lo_encrypt_type) { | |
1101 | unsigned int type = info->lo_encrypt_type; | |
1102 | ||
1103 | if (type >= MAX_LO_CRYPT) | |
1104 | return -EINVAL; | |
1105 | xfer = xfer_funcs[type]; | |
1106 | if (xfer == NULL) | |
1107 | return -EINVAL; | |
1108 | } else | |
1109 | xfer = NULL; | |
1110 | ||
1111 | err = loop_init_xfer(lo, xfer, info); | |
1112 | if (err) | |
1113 | return err; | |
1114 | ||
1115 | if (lo->lo_offset != info->lo_offset || | |
7b0576a3 | 1116 | lo->lo_sizelimit != info->lo_sizelimit) |
7035b5df | 1117 | if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) |
1da177e4 | 1118 | return -EFBIG; |
541c742a | 1119 | |
dfaa2ef6 | 1120 | loop_config_discard(lo); |
1da177e4 LT |
1121 | |
1122 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1123 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1124 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1125 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1126 | ||
1127 | if (!xfer) | |
1128 | xfer = &none_funcs; | |
1129 | lo->transfer = xfer->transfer; | |
1130 | lo->ioctl = xfer->ioctl; | |
1131 | ||
96c58655 DW |
1132 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1133 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1134 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1135 | ||
e03c8dd1 KS |
1136 | if ((info->lo_flags & LO_FLAGS_PARTSCAN) && |
1137 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1138 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1139 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1140 | ioctl_by_bdev(lo->lo_device, BLKRRPART, 0); | |
1141 | } | |
1142 | ||
1da177e4 LT |
1143 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1144 | lo->lo_init[0] = info->lo_init[0]; | |
1145 | lo->lo_init[1] = info->lo_init[1]; | |
1146 | if (info->lo_encrypt_key_size) { | |
1147 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1148 | info->lo_encrypt_key_size); | |
b0fafa81 | 1149 | lo->lo_key_owner = uid; |
1da177e4 LT |
1150 | } |
1151 | ||
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | static int | |
1156 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1157 | { | |
1158 | struct file *file = lo->lo_backing_file; | |
1159 | struct kstat stat; | |
1160 | int error; | |
1161 | ||
1162 | if (lo->lo_state != Lo_bound) | |
1163 | return -ENXIO; | |
3dadecce | 1164 | error = vfs_getattr(&file->f_path, &stat); |
1da177e4 LT |
1165 | if (error) |
1166 | return error; | |
1167 | memset(info, 0, sizeof(*info)); | |
1168 | info->lo_number = lo->lo_number; | |
1169 | info->lo_device = huge_encode_dev(stat.dev); | |
1170 | info->lo_inode = stat.ino; | |
1171 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1172 | info->lo_offset = lo->lo_offset; | |
1173 | info->lo_sizelimit = lo->lo_sizelimit; | |
1174 | info->lo_flags = lo->lo_flags; | |
1175 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1176 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1177 | info->lo_encrypt_type = | |
1178 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1179 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1180 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1181 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1182 | lo->lo_encrypt_key_size); | |
1183 | } | |
1184 | return 0; | |
1185 | } | |
1186 | ||
1187 | static void | |
1188 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1189 | { | |
1190 | memset(info64, 0, sizeof(*info64)); | |
1191 | info64->lo_number = info->lo_number; | |
1192 | info64->lo_device = info->lo_device; | |
1193 | info64->lo_inode = info->lo_inode; | |
1194 | info64->lo_rdevice = info->lo_rdevice; | |
1195 | info64->lo_offset = info->lo_offset; | |
1196 | info64->lo_sizelimit = 0; | |
1197 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1198 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1199 | info64->lo_flags = info->lo_flags; | |
1200 | info64->lo_init[0] = info->lo_init[0]; | |
1201 | info64->lo_init[1] = info->lo_init[1]; | |
1202 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1203 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1204 | else | |
1205 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1206 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1207 | } | |
1208 | ||
1209 | static int | |
1210 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1211 | { | |
1212 | memset(info, 0, sizeof(*info)); | |
1213 | info->lo_number = info64->lo_number; | |
1214 | info->lo_device = info64->lo_device; | |
1215 | info->lo_inode = info64->lo_inode; | |
1216 | info->lo_rdevice = info64->lo_rdevice; | |
1217 | info->lo_offset = info64->lo_offset; | |
1218 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1219 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1220 | info->lo_flags = info64->lo_flags; | |
1221 | info->lo_init[0] = info64->lo_init[0]; | |
1222 | info->lo_init[1] = info64->lo_init[1]; | |
1223 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1224 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1225 | else | |
1226 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1227 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1228 | ||
1229 | /* error in case values were truncated */ | |
1230 | if (info->lo_device != info64->lo_device || | |
1231 | info->lo_rdevice != info64->lo_rdevice || | |
1232 | info->lo_inode != info64->lo_inode || | |
1233 | info->lo_offset != info64->lo_offset) | |
1234 | return -EOVERFLOW; | |
1235 | ||
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | static int | |
1240 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1241 | { | |
1242 | struct loop_info info; | |
1243 | struct loop_info64 info64; | |
1244 | ||
1245 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1246 | return -EFAULT; | |
1247 | loop_info64_from_old(&info, &info64); | |
1248 | return loop_set_status(lo, &info64); | |
1249 | } | |
1250 | ||
1251 | static int | |
1252 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1253 | { | |
1254 | struct loop_info64 info64; | |
1255 | ||
1256 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1257 | return -EFAULT; | |
1258 | return loop_set_status(lo, &info64); | |
1259 | } | |
1260 | ||
1261 | static int | |
1262 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1263 | struct loop_info info; | |
1264 | struct loop_info64 info64; | |
1265 | int err = 0; | |
1266 | ||
1267 | if (!arg) | |
1268 | err = -EINVAL; | |
1269 | if (!err) | |
1270 | err = loop_get_status(lo, &info64); | |
1271 | if (!err) | |
1272 | err = loop_info64_to_old(&info64, &info); | |
1273 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1274 | err = -EFAULT; | |
1275 | ||
1276 | return err; | |
1277 | } | |
1278 | ||
1279 | static int | |
1280 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1281 | struct loop_info64 info64; | |
1282 | int err = 0; | |
1283 | ||
1284 | if (!arg) | |
1285 | err = -EINVAL; | |
1286 | if (!err) | |
1287 | err = loop_get_status(lo, &info64); | |
1288 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1289 | err = -EFAULT; | |
1290 | ||
1291 | return err; | |
1292 | } | |
1293 | ||
53d66608 O |
1294 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) |
1295 | { | |
53d66608 | 1296 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1297 | return -ENXIO; |
53d66608 | 1298 | |
7b0576a3 | 1299 | return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); |
53d66608 O |
1300 | } |
1301 | ||
bb214884 | 1302 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1303 | unsigned int cmd, unsigned long arg) |
1304 | { | |
bb214884 | 1305 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1306 | int err; |
1307 | ||
f028f3b2 | 1308 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1309 | switch (cmd) { |
1310 | case LOOP_SET_FD: | |
bb214884 | 1311 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1312 | break; |
1313 | case LOOP_CHANGE_FD: | |
bb214884 | 1314 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1315 | break; |
1316 | case LOOP_CLR_FD: | |
f028f3b2 | 1317 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
4c823cc3 | 1318 | err = loop_clr_fd(lo); |
f028f3b2 NK |
1319 | if (!err) |
1320 | goto out_unlocked; | |
1da177e4 LT |
1321 | break; |
1322 | case LOOP_SET_STATUS: | |
7035b5df DM |
1323 | err = -EPERM; |
1324 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1325 | err = loop_set_status_old(lo, | |
1326 | (struct loop_info __user *)arg); | |
1da177e4 LT |
1327 | break; |
1328 | case LOOP_GET_STATUS: | |
1329 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1330 | break; | |
1331 | case LOOP_SET_STATUS64: | |
7035b5df DM |
1332 | err = -EPERM; |
1333 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1334 | err = loop_set_status64(lo, | |
1335 | (struct loop_info64 __user *) arg); | |
1da177e4 LT |
1336 | break; |
1337 | case LOOP_GET_STATUS64: | |
1338 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1339 | break; | |
53d66608 O |
1340 | case LOOP_SET_CAPACITY: |
1341 | err = -EPERM; | |
1342 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1343 | err = loop_set_capacity(lo, bdev); | |
1344 | break; | |
1da177e4 LT |
1345 | default: |
1346 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1347 | } | |
f85221dd | 1348 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1349 | |
1350 | out_unlocked: | |
1da177e4 LT |
1351 | return err; |
1352 | } | |
1353 | ||
863d5b82 DH |
1354 | #ifdef CONFIG_COMPAT |
1355 | struct compat_loop_info { | |
1356 | compat_int_t lo_number; /* ioctl r/o */ | |
1357 | compat_dev_t lo_device; /* ioctl r/o */ | |
1358 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1359 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1360 | compat_int_t lo_offset; | |
1361 | compat_int_t lo_encrypt_type; | |
1362 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1363 | compat_int_t lo_flags; /* ioctl r/o */ | |
1364 | char lo_name[LO_NAME_SIZE]; | |
1365 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1366 | compat_ulong_t lo_init[2]; | |
1367 | char reserved[4]; | |
1368 | }; | |
1369 | ||
1370 | /* | |
1371 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1372 | * - noinlined to reduce stack space usage in main part of driver | |
1373 | */ | |
1374 | static noinline int | |
ba674cfc | 1375 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1376 | struct loop_info64 *info64) |
1377 | { | |
1378 | struct compat_loop_info info; | |
1379 | ||
1380 | if (copy_from_user(&info, arg, sizeof(info))) | |
1381 | return -EFAULT; | |
1382 | ||
1383 | memset(info64, 0, sizeof(*info64)); | |
1384 | info64->lo_number = info.lo_number; | |
1385 | info64->lo_device = info.lo_device; | |
1386 | info64->lo_inode = info.lo_inode; | |
1387 | info64->lo_rdevice = info.lo_rdevice; | |
1388 | info64->lo_offset = info.lo_offset; | |
1389 | info64->lo_sizelimit = 0; | |
1390 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1391 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1392 | info64->lo_flags = info.lo_flags; | |
1393 | info64->lo_init[0] = info.lo_init[0]; | |
1394 | info64->lo_init[1] = info.lo_init[1]; | |
1395 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1396 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1397 | else | |
1398 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1399 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1405 | * - noinlined to reduce stack space usage in main part of driver | |
1406 | */ | |
1407 | static noinline int | |
1408 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1409 | struct compat_loop_info __user *arg) | |
1410 | { | |
1411 | struct compat_loop_info info; | |
1412 | ||
1413 | memset(&info, 0, sizeof(info)); | |
1414 | info.lo_number = info64->lo_number; | |
1415 | info.lo_device = info64->lo_device; | |
1416 | info.lo_inode = info64->lo_inode; | |
1417 | info.lo_rdevice = info64->lo_rdevice; | |
1418 | info.lo_offset = info64->lo_offset; | |
1419 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1420 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1421 | info.lo_flags = info64->lo_flags; | |
1422 | info.lo_init[0] = info64->lo_init[0]; | |
1423 | info.lo_init[1] = info64->lo_init[1]; | |
1424 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1425 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1426 | else | |
1427 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1428 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1429 | ||
1430 | /* error in case values were truncated */ | |
1431 | if (info.lo_device != info64->lo_device || | |
1432 | info.lo_rdevice != info64->lo_rdevice || | |
1433 | info.lo_inode != info64->lo_inode || | |
1434 | info.lo_offset != info64->lo_offset || | |
1435 | info.lo_init[0] != info64->lo_init[0] || | |
1436 | info.lo_init[1] != info64->lo_init[1]) | |
1437 | return -EOVERFLOW; | |
1438 | ||
1439 | if (copy_to_user(arg, &info, sizeof(info))) | |
1440 | return -EFAULT; | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | static int | |
1445 | loop_set_status_compat(struct loop_device *lo, | |
1446 | const struct compat_loop_info __user *arg) | |
1447 | { | |
1448 | struct loop_info64 info64; | |
1449 | int ret; | |
1450 | ||
1451 | ret = loop_info64_from_compat(arg, &info64); | |
1452 | if (ret < 0) | |
1453 | return ret; | |
1454 | return loop_set_status(lo, &info64); | |
1455 | } | |
1456 | ||
1457 | static int | |
1458 | loop_get_status_compat(struct loop_device *lo, | |
1459 | struct compat_loop_info __user *arg) | |
1460 | { | |
1461 | struct loop_info64 info64; | |
1462 | int err = 0; | |
1463 | ||
1464 | if (!arg) | |
1465 | err = -EINVAL; | |
1466 | if (!err) | |
1467 | err = loop_get_status(lo, &info64); | |
1468 | if (!err) | |
1469 | err = loop_info64_to_compat(&info64, arg); | |
1470 | return err; | |
1471 | } | |
1472 | ||
bb214884 AV |
1473 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1474 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1475 | { |
bb214884 | 1476 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1477 | int err; |
1478 | ||
863d5b82 DH |
1479 | switch(cmd) { |
1480 | case LOOP_SET_STATUS: | |
1481 | mutex_lock(&lo->lo_ctl_mutex); | |
1482 | err = loop_set_status_compat( | |
1483 | lo, (const struct compat_loop_info __user *) arg); | |
1484 | mutex_unlock(&lo->lo_ctl_mutex); | |
1485 | break; | |
1486 | case LOOP_GET_STATUS: | |
1487 | mutex_lock(&lo->lo_ctl_mutex); | |
1488 | err = loop_get_status_compat( | |
1489 | lo, (struct compat_loop_info __user *) arg); | |
1490 | mutex_unlock(&lo->lo_ctl_mutex); | |
1491 | break; | |
53d66608 | 1492 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1493 | case LOOP_CLR_FD: |
1494 | case LOOP_GET_STATUS64: | |
1495 | case LOOP_SET_STATUS64: | |
1496 | arg = (unsigned long) compat_ptr(arg); | |
1497 | case LOOP_SET_FD: | |
1498 | case LOOP_CHANGE_FD: | |
bb214884 | 1499 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1500 | break; |
1501 | default: | |
1502 | err = -ENOIOCTLCMD; | |
1503 | break; | |
1504 | } | |
863d5b82 DH |
1505 | return err; |
1506 | } | |
1507 | #endif | |
1508 | ||
bb214884 | 1509 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1510 | { |
770fe30a KS |
1511 | struct loop_device *lo; |
1512 | int err = 0; | |
1513 | ||
1514 | mutex_lock(&loop_index_mutex); | |
1515 | lo = bdev->bd_disk->private_data; | |
1516 | if (!lo) { | |
1517 | err = -ENXIO; | |
1518 | goto out; | |
1519 | } | |
1da177e4 | 1520 | |
f85221dd | 1521 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1522 | lo->lo_refcnt++; |
f85221dd | 1523 | mutex_unlock(&lo->lo_ctl_mutex); |
770fe30a KS |
1524 | out: |
1525 | mutex_unlock(&loop_index_mutex); | |
1526 | return err; | |
1da177e4 LT |
1527 | } |
1528 | ||
bb214884 | 1529 | static int lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1530 | { |
bb214884 | 1531 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1532 | int err; |
1da177e4 | 1533 | |
f85221dd | 1534 | mutex_lock(&lo->lo_ctl_mutex); |
96c58655 | 1535 | |
14f27939 MB |
1536 | if (--lo->lo_refcnt) |
1537 | goto out; | |
1538 | ||
1539 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
1540 | /* | |
1541 | * In autoclear mode, stop the loop thread | |
1542 | * and remove configuration after last close. | |
1543 | */ | |
4c823cc3 | 1544 | err = loop_clr_fd(lo); |
ffcd7dca AB |
1545 | if (!err) |
1546 | goto out_unlocked; | |
14f27939 MB |
1547 | } else { |
1548 | /* | |
1549 | * Otherwise keep thread (if running) and config, | |
1550 | * but flush possible ongoing bios in thread. | |
1551 | */ | |
1552 | loop_flush(lo); | |
1553 | } | |
96c58655 | 1554 | |
14f27939 | 1555 | out: |
f85221dd | 1556 | mutex_unlock(&lo->lo_ctl_mutex); |
ffcd7dca | 1557 | out_unlocked: |
1da177e4 LT |
1558 | return 0; |
1559 | } | |
1560 | ||
83d5cde4 | 1561 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1562 | .owner = THIS_MODULE, |
bb214884 AV |
1563 | .open = lo_open, |
1564 | .release = lo_release, | |
1565 | .ioctl = lo_ioctl, | |
863d5b82 | 1566 | #ifdef CONFIG_COMPAT |
bb214884 | 1567 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1568 | #endif |
1da177e4 LT |
1569 | }; |
1570 | ||
1571 | /* | |
1572 | * And now the modules code and kernel interface. | |
1573 | */ | |
73285082 | 1574 | static int max_loop; |
ac04fee0 | 1575 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1576 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1577 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1578 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1579 | MODULE_LICENSE("GPL"); |
1580 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1581 | ||
1582 | int loop_register_transfer(struct loop_func_table *funcs) | |
1583 | { | |
1584 | unsigned int n = funcs->number; | |
1585 | ||
1586 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1587 | return -EINVAL; | |
1588 | xfer_funcs[n] = funcs; | |
1589 | return 0; | |
1590 | } | |
1591 | ||
34dd82af KS |
1592 | static int unregister_transfer_cb(int id, void *ptr, void *data) |
1593 | { | |
1594 | struct loop_device *lo = ptr; | |
1595 | struct loop_func_table *xfer = data; | |
1596 | ||
1597 | mutex_lock(&lo->lo_ctl_mutex); | |
1598 | if (lo->lo_encryption == xfer) | |
1599 | loop_release_xfer(lo); | |
1600 | mutex_unlock(&lo->lo_ctl_mutex); | |
1601 | return 0; | |
1602 | } | |
1603 | ||
1da177e4 LT |
1604 | int loop_unregister_transfer(int number) |
1605 | { | |
1606 | unsigned int n = number; | |
1da177e4 LT |
1607 | struct loop_func_table *xfer; |
1608 | ||
1609 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1610 | return -EINVAL; | |
1611 | ||
1612 | xfer_funcs[n] = NULL; | |
34dd82af | 1613 | idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); |
1da177e4 LT |
1614 | return 0; |
1615 | } | |
1616 | ||
1617 | EXPORT_SYMBOL(loop_register_transfer); | |
1618 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1619 | ||
34dd82af | 1620 | static int loop_add(struct loop_device **l, int i) |
73285082 KC |
1621 | { |
1622 | struct loop_device *lo; | |
1623 | struct gendisk *disk; | |
34dd82af | 1624 | int err; |
73285082 | 1625 | |
68d740d7 | 1626 | err = -ENOMEM; |
73285082 | 1627 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1628 | if (!lo) |
73285082 | 1629 | goto out; |
34dd82af | 1630 | |
c718aa65 | 1631 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1632 | if (i >= 0) { |
c718aa65 TH |
1633 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1634 | if (err == -ENOSPC) | |
34dd82af | 1635 | err = -EEXIST; |
34dd82af | 1636 | } else { |
c718aa65 | 1637 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af KS |
1638 | } |
1639 | if (err < 0) | |
1640 | goto out_free_dev; | |
c718aa65 | 1641 | i = err; |
73285082 | 1642 | |
183cfb57 | 1643 | err = -ENOMEM; |
73285082 KC |
1644 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); |
1645 | if (!lo->lo_queue) | |
1646 | goto out_free_dev; | |
1647 | ||
476a4813 | 1648 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1649 | if (!disk) |
1650 | goto out_free_queue; | |
1651 | ||
e03c8dd1 KS |
1652 | /* |
1653 | * Disable partition scanning by default. The in-kernel partition | |
1654 | * scanning can be requested individually per-device during its | |
1655 | * setup. Userspace can always add and remove partitions from all | |
1656 | * devices. The needed partition minors are allocated from the | |
1657 | * extended minor space, the main loop device numbers will continue | |
1658 | * to match the loop minors, regardless of the number of partitions | |
1659 | * used. | |
1660 | * | |
1661 | * If max_part is given, partition scanning is globally enabled for | |
1662 | * all loop devices. The minors for the main loop devices will be | |
1663 | * multiples of max_part. | |
1664 | * | |
1665 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1666 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1667 | * complicated, are too static, inflexible and may surprise | |
1668 | * userspace tools. Parameters like this in general should be avoided. | |
1669 | */ | |
1670 | if (!part_shift) | |
1671 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1672 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 KC |
1673 | mutex_init(&lo->lo_ctl_mutex); |
1674 | lo->lo_number = i; | |
1675 | lo->lo_thread = NULL; | |
1676 | init_waitqueue_head(&lo->lo_event); | |
7b5a3522 | 1677 | init_waitqueue_head(&lo->lo_req_wait); |
73285082 KC |
1678 | spin_lock_init(&lo->lo_lock); |
1679 | disk->major = LOOP_MAJOR; | |
476a4813 | 1680 | disk->first_minor = i << part_shift; |
73285082 KC |
1681 | disk->fops = &lo_fops; |
1682 | disk->private_data = lo; | |
1683 | disk->queue = lo->lo_queue; | |
1684 | sprintf(disk->disk_name, "loop%d", i); | |
34dd82af KS |
1685 | add_disk(disk); |
1686 | *l = lo; | |
1687 | return lo->lo_number; | |
73285082 KC |
1688 | |
1689 | out_free_queue: | |
1690 | blk_cleanup_queue(lo->lo_queue); | |
1691 | out_free_dev: | |
1692 | kfree(lo); | |
1693 | out: | |
34dd82af | 1694 | return err; |
73285082 KC |
1695 | } |
1696 | ||
34dd82af | 1697 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 1698 | { |
34dd82af | 1699 | del_gendisk(lo->lo_disk); |
73285082 KC |
1700 | blk_cleanup_queue(lo->lo_queue); |
1701 | put_disk(lo->lo_disk); | |
73285082 KC |
1702 | kfree(lo); |
1703 | } | |
1da177e4 | 1704 | |
770fe30a KS |
1705 | static int find_free_cb(int id, void *ptr, void *data) |
1706 | { | |
1707 | struct loop_device *lo = ptr; | |
1708 | struct loop_device **l = data; | |
1709 | ||
1710 | if (lo->lo_state == Lo_unbound) { | |
1711 | *l = lo; | |
1712 | return 1; | |
1713 | } | |
1714 | return 0; | |
1715 | } | |
1716 | ||
34dd82af | 1717 | static int loop_lookup(struct loop_device **l, int i) |
a47653fc KC |
1718 | { |
1719 | struct loop_device *lo; | |
34dd82af | 1720 | int ret = -ENODEV; |
a47653fc | 1721 | |
770fe30a KS |
1722 | if (i < 0) { |
1723 | int err; | |
1724 | ||
1725 | err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); | |
1726 | if (err == 1) { | |
1727 | *l = lo; | |
1728 | ret = lo->lo_number; | |
1729 | } | |
1730 | goto out; | |
a47653fc KC |
1731 | } |
1732 | ||
770fe30a | 1733 | /* lookup and return a specific i */ |
34dd82af | 1734 | lo = idr_find(&loop_index_idr, i); |
a47653fc | 1735 | if (lo) { |
34dd82af KS |
1736 | *l = lo; |
1737 | ret = lo->lo_number; | |
a47653fc | 1738 | } |
770fe30a | 1739 | out: |
34dd82af | 1740 | return ret; |
a47653fc KC |
1741 | } |
1742 | ||
73285082 KC |
1743 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1744 | { | |
705962cc | 1745 | struct loop_device *lo; |
07002e99 | 1746 | struct kobject *kobj; |
34dd82af | 1747 | int err; |
73285082 | 1748 | |
34dd82af KS |
1749 | mutex_lock(&loop_index_mutex); |
1750 | err = loop_lookup(&lo, MINOR(dev) >> part_shift); | |
1751 | if (err < 0) | |
1752 | err = loop_add(&lo, MINOR(dev) >> part_shift); | |
1753 | if (err < 0) | |
1754 | kobj = ERR_PTR(err); | |
1755 | else | |
1756 | kobj = get_disk(lo->lo_disk); | |
1757 | mutex_unlock(&loop_index_mutex); | |
73285082 KC |
1758 | |
1759 | *part = 0; | |
07002e99 | 1760 | return kobj; |
73285082 KC |
1761 | } |
1762 | ||
770fe30a KS |
1763 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
1764 | unsigned long parm) | |
1765 | { | |
1766 | struct loop_device *lo; | |
1767 | int ret = -ENOSYS; | |
1768 | ||
1769 | mutex_lock(&loop_index_mutex); | |
1770 | switch (cmd) { | |
1771 | case LOOP_CTL_ADD: | |
1772 | ret = loop_lookup(&lo, parm); | |
1773 | if (ret >= 0) { | |
1774 | ret = -EEXIST; | |
1775 | break; | |
1776 | } | |
1777 | ret = loop_add(&lo, parm); | |
1778 | break; | |
1779 | case LOOP_CTL_REMOVE: | |
1780 | ret = loop_lookup(&lo, parm); | |
1781 | if (ret < 0) | |
1782 | break; | |
1783 | mutex_lock(&lo->lo_ctl_mutex); | |
1784 | if (lo->lo_state != Lo_unbound) { | |
1785 | ret = -EBUSY; | |
1786 | mutex_unlock(&lo->lo_ctl_mutex); | |
1787 | break; | |
1788 | } | |
1789 | if (lo->lo_refcnt > 0) { | |
1790 | ret = -EBUSY; | |
1791 | mutex_unlock(&lo->lo_ctl_mutex); | |
1792 | break; | |
1793 | } | |
1794 | lo->lo_disk->private_data = NULL; | |
1795 | mutex_unlock(&lo->lo_ctl_mutex); | |
1796 | idr_remove(&loop_index_idr, lo->lo_number); | |
1797 | loop_remove(lo); | |
1798 | break; | |
1799 | case LOOP_CTL_GET_FREE: | |
1800 | ret = loop_lookup(&lo, -1); | |
1801 | if (ret >= 0) | |
1802 | break; | |
1803 | ret = loop_add(&lo, -1); | |
1804 | } | |
1805 | mutex_unlock(&loop_index_mutex); | |
1806 | ||
1807 | return ret; | |
1808 | } | |
1809 | ||
1810 | static const struct file_operations loop_ctl_fops = { | |
1811 | .open = nonseekable_open, | |
1812 | .unlocked_ioctl = loop_control_ioctl, | |
1813 | .compat_ioctl = loop_control_ioctl, | |
1814 | .owner = THIS_MODULE, | |
1815 | .llseek = noop_llseek, | |
1816 | }; | |
1817 | ||
1818 | static struct miscdevice loop_misc = { | |
1819 | .minor = LOOP_CTRL_MINOR, | |
1820 | .name = "loop-control", | |
1821 | .fops = &loop_ctl_fops, | |
1822 | }; | |
1823 | ||
1824 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
1825 | MODULE_ALIAS("devname:loop-control"); | |
1826 | ||
73285082 KC |
1827 | static int __init loop_init(void) |
1828 | { | |
a47653fc KC |
1829 | int i, nr; |
1830 | unsigned long range; | |
34dd82af | 1831 | struct loop_device *lo; |
770fe30a | 1832 | int err; |
a47653fc | 1833 | |
770fe30a KS |
1834 | err = misc_register(&loop_misc); |
1835 | if (err < 0) | |
1836 | return err; | |
476a4813 LV |
1837 | |
1838 | part_shift = 0; | |
ac04fee0 | 1839 | if (max_part > 0) { |
476a4813 LV |
1840 | part_shift = fls(max_part); |
1841 | ||
ac04fee0 NK |
1842 | /* |
1843 | * Adjust max_part according to part_shift as it is exported | |
1844 | * to user space so that user can decide correct minor number | |
1845 | * if [s]he want to create more devices. | |
1846 | * | |
1847 | * Note that -1 is required because partition 0 is reserved | |
1848 | * for the whole disk. | |
1849 | */ | |
1850 | max_part = (1UL << part_shift) - 1; | |
1851 | } | |
1852 | ||
b1a66504 GC |
1853 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
1854 | err = -EINVAL; | |
1855 | goto misc_out; | |
1856 | } | |
78f4bb36 | 1857 | |
b1a66504 GC |
1858 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
1859 | err = -EINVAL; | |
1860 | goto misc_out; | |
1861 | } | |
1da177e4 | 1862 | |
d134b00b KS |
1863 | /* |
1864 | * If max_loop is specified, create that many devices upfront. | |
1865 | * This also becomes a hard limit. If max_loop is not specified, | |
1866 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
1867 | * init time. Loop devices can be requested on-demand with the | |
1868 | * /dev/loop-control interface, or be instantiated by accessing | |
1869 | * a 'dead' device node. | |
1870 | */ | |
73285082 | 1871 | if (max_loop) { |
a47653fc | 1872 | nr = max_loop; |
a1c15c59 | 1873 | range = max_loop << part_shift; |
a47653fc | 1874 | } else { |
d134b00b | 1875 | nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; |
a1c15c59 | 1876 | range = 1UL << MINORBITS; |
a47653fc KC |
1877 | } |
1878 | ||
b1a66504 GC |
1879 | if (register_blkdev(LOOP_MAJOR, "loop")) { |
1880 | err = -EIO; | |
1881 | goto misc_out; | |
1882 | } | |
1da177e4 | 1883 | |
a47653fc KC |
1884 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, |
1885 | THIS_MODULE, loop_probe, NULL, NULL); | |
1886 | ||
d134b00b | 1887 | /* pre-create number of devices given by config or max_loop */ |
34dd82af KS |
1888 | mutex_lock(&loop_index_mutex); |
1889 | for (i = 0; i < nr; i++) | |
1890 | loop_add(&lo, i); | |
1891 | mutex_unlock(&loop_index_mutex); | |
1892 | ||
73285082 | 1893 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 1894 | return 0; |
b1a66504 GC |
1895 | |
1896 | misc_out: | |
1897 | misc_deregister(&loop_misc); | |
1898 | return err; | |
34dd82af | 1899 | } |
a47653fc | 1900 | |
34dd82af KS |
1901 | static int loop_exit_cb(int id, void *ptr, void *data) |
1902 | { | |
1903 | struct loop_device *lo = ptr; | |
a47653fc | 1904 | |
34dd82af KS |
1905 | loop_remove(lo); |
1906 | return 0; | |
1da177e4 LT |
1907 | } |
1908 | ||
73285082 | 1909 | static void __exit loop_exit(void) |
1da177e4 | 1910 | { |
a47653fc | 1911 | unsigned long range; |
1da177e4 | 1912 | |
a1c15c59 | 1913 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 1914 | |
34dd82af | 1915 | idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); |
34dd82af | 1916 | idr_destroy(&loop_index_idr); |
73285082 | 1917 | |
a47653fc | 1918 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 1919 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a KS |
1920 | |
1921 | misc_deregister(&loop_misc); | |
1da177e4 LT |
1922 | } |
1923 | ||
1924 | module_init(loop_init); | |
1925 | module_exit(loop_exit); | |
1926 | ||
1927 | #ifndef MODULE | |
1928 | static int __init max_loop_setup(char *str) | |
1929 | { | |
1930 | max_loop = simple_strtol(str, NULL, 0); | |
1931 | return 1; | |
1932 | } | |
1933 | ||
1934 | __setup("max_loop=", max_loop_setup); | |
1935 | #endif |