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 LT |
71 | #include <linux/writeback.h> |
72 | #include <linux/buffer_head.h> /* for invalidate_bdev() */ | |
73 | #include <linux/completion.h> | |
74 | #include <linux/highmem.h> | |
6c997918 | 75 | #include <linux/kthread.h> |
d6b29d7c | 76 | #include <linux/splice.h> |
ee862730 | 77 | #include <linux/sysfs.h> |
dfaa2ef6 | 78 | #include <linux/falloc.h> |
1da177e4 LT |
79 | |
80 | #include <asm/uaccess.h> | |
81 | ||
73285082 KC |
82 | static LIST_HEAD(loop_devices); |
83 | static DEFINE_MUTEX(loop_devices_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 | { | |
96 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
97 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
98 | ||
99 | if (cmd == READ) | |
100 | memcpy(loop_buf, raw_buf, size); | |
101 | else | |
102 | memcpy(raw_buf, loop_buf, size); | |
103 | ||
1da177e4 | 104 | kunmap_atomic(loop_buf, KM_USER1); |
61ecdb80 | 105 | kunmap_atomic(raw_buf, KM_USER0); |
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 | { | |
115 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
116 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
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 | ||
1da177e4 | 133 | kunmap_atomic(loop_buf, KM_USER1); |
61ecdb80 | 134 | kunmap_atomic(raw_buf, KM_USER0); |
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 | ||
163 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) | |
164 | { | |
165 | loff_t size, offset, loopsize; | |
166 | ||
167 | /* Compute loopsize in bytes */ | |
168 | size = i_size_read(file->f_mapping->host); | |
169 | offset = lo->lo_offset; | |
170 | loopsize = size - offset; | |
171 | if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize) | |
172 | loopsize = lo->lo_sizelimit; | |
173 | ||
174 | /* | |
175 | * Unfortunately, if we want to do I/O on the device, | |
176 | * the number of 512-byte sectors has to fit into a sector_t. | |
177 | */ | |
178 | return loopsize >> 9; | |
179 | } | |
180 | ||
181 | static int | |
182 | figure_loop_size(struct loop_device *lo) | |
183 | { | |
184 | loff_t size = get_loop_size(lo, lo->lo_backing_file); | |
185 | sector_t x = (sector_t)size; | |
186 | ||
187 | if (unlikely((loff_t)x != size)) | |
188 | return -EFBIG; | |
189 | ||
73285082 | 190 | set_capacity(lo->lo_disk, x); |
1da177e4 LT |
191 | return 0; |
192 | } | |
193 | ||
194 | static inline int | |
195 | lo_do_transfer(struct loop_device *lo, int cmd, | |
196 | struct page *rpage, unsigned roffs, | |
197 | struct page *lpage, unsigned loffs, | |
198 | int size, sector_t rblock) | |
199 | { | |
200 | if (unlikely(!lo->transfer)) | |
201 | return 0; | |
202 | ||
203 | return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
204 | } | |
205 | ||
206 | /** | |
207 | * do_lo_send_aops - helper for writing data to a loop device | |
208 | * | |
209 | * This is the fast version for backing filesystems which implement the address | |
afddba49 | 210 | * space operations write_begin and write_end. |
1da177e4 LT |
211 | */ |
212 | static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 213 | loff_t pos, struct page *unused) |
1da177e4 LT |
214 | { |
215 | struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */ | |
216 | struct address_space *mapping = file->f_mapping; | |
1da177e4 LT |
217 | pgoff_t index; |
218 | unsigned offset, bv_offs; | |
994fc28c | 219 | int len, ret; |
1da177e4 | 220 | |
1b1dcc1b | 221 | mutex_lock(&mapping->host->i_mutex); |
1da177e4 LT |
222 | index = pos >> PAGE_CACHE_SHIFT; |
223 | offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1); | |
224 | bv_offs = bvec->bv_offset; | |
225 | len = bvec->bv_len; | |
226 | while (len > 0) { | |
227 | sector_t IV; | |
afddba49 | 228 | unsigned size, copied; |
1da177e4 | 229 | int transfer_result; |
afddba49 NP |
230 | struct page *page; |
231 | void *fsdata; | |
1da177e4 LT |
232 | |
233 | IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9); | |
234 | size = PAGE_CACHE_SIZE - offset; | |
235 | if (size > len) | |
236 | size = len; | |
afddba49 NP |
237 | |
238 | ret = pagecache_write_begin(file, mapping, pos, size, 0, | |
239 | &page, &fsdata); | |
240 | if (ret) | |
1da177e4 | 241 | goto fail; |
afddba49 | 242 | |
02246c41 NK |
243 | file_update_time(file); |
244 | ||
1da177e4 LT |
245 | transfer_result = lo_do_transfer(lo, WRITE, page, offset, |
246 | bvec->bv_page, bv_offs, size, IV); | |
afddba49 | 247 | copied = size; |
1da177e4 | 248 | if (unlikely(transfer_result)) |
afddba49 NP |
249 | copied = 0; |
250 | ||
251 | ret = pagecache_write_end(file, mapping, pos, size, copied, | |
252 | page, fsdata); | |
8268f5a7 | 253 | if (ret < 0 || ret != copied) |
afddba49 | 254 | goto fail; |
afddba49 NP |
255 | |
256 | if (unlikely(transfer_result)) | |
257 | goto fail; | |
258 | ||
259 | bv_offs += copied; | |
260 | len -= copied; | |
1da177e4 LT |
261 | offset = 0; |
262 | index++; | |
afddba49 | 263 | pos += copied; |
1da177e4 | 264 | } |
994fc28c | 265 | ret = 0; |
1da177e4 | 266 | out: |
1b1dcc1b | 267 | mutex_unlock(&mapping->host->i_mutex); |
1da177e4 | 268 | return ret; |
1da177e4 LT |
269 | fail: |
270 | ret = -1; | |
271 | goto out; | |
272 | } | |
273 | ||
274 | /** | |
275 | * __do_lo_send_write - helper for writing data to a loop device | |
276 | * | |
277 | * This helper just factors out common code between do_lo_send_direct_write() | |
278 | * and do_lo_send_write(). | |
279 | */ | |
858119e1 | 280 | static int __do_lo_send_write(struct file *file, |
98ae6ccd | 281 | u8 *buf, const int len, loff_t pos) |
1da177e4 LT |
282 | { |
283 | ssize_t bw; | |
284 | mm_segment_t old_fs = get_fs(); | |
285 | ||
286 | set_fs(get_ds()); | |
287 | bw = file->f_op->write(file, buf, len, &pos); | |
288 | set_fs(old_fs); | |
289 | if (likely(bw == len)) | |
290 | return 0; | |
291 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", | |
292 | (unsigned long long)pos, len); | |
293 | if (bw >= 0) | |
294 | bw = -EIO; | |
295 | return bw; | |
296 | } | |
297 | ||
298 | /** | |
299 | * do_lo_send_direct_write - helper for writing data to a loop device | |
300 | * | |
301 | * This is the fast, non-transforming version for backing filesystems which do | |
afddba49 | 302 | * not implement the address space operations write_begin and write_end. |
1da177e4 LT |
303 | * It uses the write file operation which should be present on all writeable |
304 | * filesystems. | |
305 | */ | |
306 | static int do_lo_send_direct_write(struct loop_device *lo, | |
511de73f | 307 | struct bio_vec *bvec, loff_t pos, struct page *page) |
1da177e4 LT |
308 | { |
309 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 310 | kmap(bvec->bv_page) + bvec->bv_offset, |
1da177e4 LT |
311 | bvec->bv_len, pos); |
312 | kunmap(bvec->bv_page); | |
313 | cond_resched(); | |
314 | return bw; | |
315 | } | |
316 | ||
317 | /** | |
318 | * do_lo_send_write - helper for writing data to a loop device | |
319 | * | |
320 | * This is the slow, transforming version for filesystems which do not | |
afddba49 | 321 | * implement the address space operations write_begin and write_end. It |
1da177e4 LT |
322 | * uses the write file operation which should be present on all writeable |
323 | * filesystems. | |
324 | * | |
325 | * Using fops->write is slower than using aops->{prepare,commit}_write in the | |
326 | * transforming case because we need to double buffer the data as we cannot do | |
327 | * the transformations in place as we do not have direct access to the | |
328 | * destination pages of the backing file. | |
329 | */ | |
330 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 331 | loff_t pos, struct page *page) |
1da177e4 LT |
332 | { |
333 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, | |
334 | bvec->bv_offset, bvec->bv_len, pos >> 9); | |
335 | if (likely(!ret)) | |
336 | return __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 337 | page_address(page), bvec->bv_len, |
1da177e4 LT |
338 | pos); |
339 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, " | |
340 | "length %i.\n", (unsigned long long)pos, bvec->bv_len); | |
341 | if (ret > 0) | |
342 | ret = -EIO; | |
343 | return ret; | |
344 | } | |
345 | ||
511de73f | 346 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) |
1da177e4 | 347 | { |
511de73f | 348 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, |
1da177e4 LT |
349 | struct page *page); |
350 | struct bio_vec *bvec; | |
351 | struct page *page = NULL; | |
352 | int i, ret = 0; | |
353 | ||
354 | do_lo_send = do_lo_send_aops; | |
355 | if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) { | |
356 | do_lo_send = do_lo_send_direct_write; | |
357 | if (lo->transfer != transfer_none) { | |
358 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
359 | if (unlikely(!page)) | |
360 | goto fail; | |
361 | kmap(page); | |
362 | do_lo_send = do_lo_send_write; | |
363 | } | |
364 | } | |
365 | bio_for_each_segment(bvec, bio, i) { | |
511de73f | 366 | ret = do_lo_send(lo, bvec, pos, page); |
1da177e4 LT |
367 | if (ret < 0) |
368 | break; | |
369 | pos += bvec->bv_len; | |
370 | } | |
371 | if (page) { | |
372 | kunmap(page); | |
373 | __free_page(page); | |
374 | } | |
375 | out: | |
376 | return ret; | |
377 | fail: | |
378 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); | |
379 | ret = -ENOMEM; | |
380 | goto out; | |
381 | } | |
382 | ||
383 | struct lo_read_data { | |
384 | struct loop_device *lo; | |
385 | struct page *page; | |
386 | unsigned offset; | |
387 | int bsize; | |
388 | }; | |
389 | ||
390 | static int | |
fd582140 JA |
391 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
392 | struct splice_desc *sd) | |
1da177e4 | 393 | { |
fd582140 | 394 | struct lo_read_data *p = sd->u.data; |
1da177e4 | 395 | struct loop_device *lo = p->lo; |
fd582140 | 396 | struct page *page = buf->page; |
1da177e4 | 397 | sector_t IV; |
3603b8ea | 398 | int size; |
1da177e4 | 399 | |
fd582140 JA |
400 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + |
401 | (buf->offset >> 9); | |
402 | size = sd->len; | |
403 | if (size > p->bsize) | |
404 | size = p->bsize; | |
1da177e4 | 405 | |
fd582140 | 406 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { |
1da177e4 LT |
407 | printk(KERN_ERR "loop: transfer error block %ld\n", |
408 | page->index); | |
fd582140 | 409 | size = -EINVAL; |
1da177e4 LT |
410 | } |
411 | ||
412 | flush_dcache_page(p->page); | |
413 | ||
fd582140 JA |
414 | if (size > 0) |
415 | p->offset += size; | |
416 | ||
1da177e4 LT |
417 | return size; |
418 | } | |
419 | ||
fd582140 JA |
420 | static int |
421 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
422 | { | |
423 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
424 | } | |
425 | ||
1da177e4 LT |
426 | static int |
427 | do_lo_receive(struct loop_device *lo, | |
428 | struct bio_vec *bvec, int bsize, loff_t pos) | |
429 | { | |
430 | struct lo_read_data cookie; | |
fd582140 | 431 | struct splice_desc sd; |
1da177e4 | 432 | struct file *file; |
fd582140 | 433 | long retval; |
1da177e4 LT |
434 | |
435 | cookie.lo = lo; | |
436 | cookie.page = bvec->bv_page; | |
437 | cookie.offset = bvec->bv_offset; | |
438 | cookie.bsize = bsize; | |
fd582140 JA |
439 | |
440 | sd.len = 0; | |
441 | sd.total_len = bvec->bv_len; | |
442 | sd.flags = 0; | |
443 | sd.pos = pos; | |
444 | sd.u.data = &cookie; | |
445 | ||
1da177e4 | 446 | file = lo->lo_backing_file; |
fd582140 JA |
447 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); |
448 | ||
449 | if (retval < 0) | |
450 | return retval; | |
451 | ||
452 | return 0; | |
1da177e4 LT |
453 | } |
454 | ||
455 | static int | |
456 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
457 | { | |
458 | struct bio_vec *bvec; | |
459 | int i, ret = 0; | |
460 | ||
461 | bio_for_each_segment(bvec, bio, i) { | |
462 | ret = do_lo_receive(lo, bvec, bsize, pos); | |
463 | if (ret < 0) | |
464 | break; | |
465 | pos += bvec->bv_len; | |
466 | } | |
467 | return ret; | |
468 | } | |
469 | ||
470 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
471 | { | |
472 | loff_t pos; | |
473 | int ret; | |
474 | ||
475 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
68db1961 NK |
476 | |
477 | if (bio_rw(bio) == WRITE) { | |
68db1961 NK |
478 | struct file *file = lo->lo_backing_file; |
479 | ||
6259f284 | 480 | if (bio->bi_rw & REQ_FLUSH) { |
8018ab05 | 481 | ret = vfs_fsync(file, 0); |
6259f284 | 482 | if (unlikely(ret && ret != -EINVAL)) { |
68db1961 NK |
483 | ret = -EIO; |
484 | goto out; | |
485 | } | |
486 | } | |
487 | ||
dfaa2ef6 LC |
488 | /* |
489 | * We use punch hole to reclaim the free space used by the | |
490 | * image a.k.a. discard. However we do support discard if | |
491 | * encryption is enabled, because it may give an attacker | |
492 | * useful information. | |
493 | */ | |
494 | if (bio->bi_rw & REQ_DISCARD) { | |
495 | struct file *file = lo->lo_backing_file; | |
496 | int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE; | |
497 | ||
498 | if ((!file->f_op->fallocate) || | |
499 | lo->lo_encrypt_key_size) { | |
500 | ret = -EOPNOTSUPP; | |
501 | goto out; | |
502 | } | |
503 | ret = file->f_op->fallocate(file, mode, pos, | |
504 | bio->bi_size); | |
505 | if (unlikely(ret && ret != -EINVAL && | |
506 | ret != -EOPNOTSUPP)) | |
507 | ret = -EIO; | |
508 | goto out; | |
509 | } | |
510 | ||
511de73f | 511 | ret = lo_send(lo, bio, pos); |
68db1961 | 512 | |
6259f284 | 513 | if ((bio->bi_rw & REQ_FUA) && !ret) { |
8018ab05 | 514 | ret = vfs_fsync(file, 0); |
6259f284 | 515 | if (unlikely(ret && ret != -EINVAL)) |
68db1961 NK |
516 | ret = -EIO; |
517 | } | |
518 | } else | |
1da177e4 | 519 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); |
68db1961 NK |
520 | |
521 | out: | |
1da177e4 LT |
522 | return ret; |
523 | } | |
524 | ||
525 | /* | |
526 | * Add bio to back of pending list | |
527 | */ | |
528 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
529 | { | |
e686307f | 530 | bio_list_add(&lo->lo_bio_list, bio); |
1da177e4 LT |
531 | } |
532 | ||
533 | /* | |
534 | * Grab first pending buffer | |
535 | */ | |
536 | static struct bio *loop_get_bio(struct loop_device *lo) | |
537 | { | |
e686307f | 538 | return bio_list_pop(&lo->lo_bio_list); |
1da177e4 LT |
539 | } |
540 | ||
165125e1 | 541 | static int loop_make_request(struct request_queue *q, struct bio *old_bio) |
1da177e4 LT |
542 | { |
543 | struct loop_device *lo = q->queuedata; | |
544 | int rw = bio_rw(old_bio); | |
545 | ||
35a82d1a NP |
546 | if (rw == READA) |
547 | rw = READ; | |
548 | ||
549 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
1da177e4 LT |
550 | |
551 | spin_lock_irq(&lo->lo_lock); | |
552 | if (lo->lo_state != Lo_bound) | |
35a82d1a NP |
553 | goto out; |
554 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
555 | goto out; | |
1da177e4 | 556 | loop_add_bio(lo, old_bio); |
6c997918 | 557 | wake_up(&lo->lo_event); |
35a82d1a | 558 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 559 | return 0; |
35a82d1a | 560 | |
1da177e4 | 561 | out: |
35a82d1a | 562 | spin_unlock_irq(&lo->lo_lock); |
6712ecf8 | 563 | bio_io_error(old_bio); |
1da177e4 | 564 | return 0; |
1da177e4 LT |
565 | } |
566 | ||
1da177e4 LT |
567 | struct switch_request { |
568 | struct file *file; | |
569 | struct completion wait; | |
570 | }; | |
571 | ||
572 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
573 | ||
574 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
575 | { | |
1da177e4 LT |
576 | if (unlikely(!bio->bi_bdev)) { |
577 | do_loop_switch(lo, bio->bi_private); | |
578 | bio_put(bio); | |
579 | } else { | |
35a82d1a | 580 | int ret = do_bio_filebacked(lo, bio); |
6712ecf8 | 581 | bio_endio(bio, ret); |
1da177e4 LT |
582 | } |
583 | } | |
584 | ||
585 | /* | |
586 | * worker thread that handles reads/writes to file backed loop devices, | |
587 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
588 | * on reads for block backed loop, as that is too heavy to do from | |
589 | * b_end_io context where irqs may be disabled. | |
6c997918 SH |
590 | * |
591 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
592 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
593 | * true, make_request will not place any more requests. Therefore | |
594 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
595 | * done with the loop. | |
1da177e4 LT |
596 | */ |
597 | static int loop_thread(void *data) | |
598 | { | |
599 | struct loop_device *lo = data; | |
600 | struct bio *bio; | |
601 | ||
1da177e4 LT |
602 | set_user_nice(current, -20); |
603 | ||
e686307f | 604 | while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) { |
09c0dc68 | 605 | |
6c997918 | 606 | wait_event_interruptible(lo->lo_event, |
e686307f AM |
607 | !bio_list_empty(&lo->lo_bio_list) || |
608 | kthread_should_stop()); | |
35a82d1a | 609 | |
e686307f | 610 | if (bio_list_empty(&lo->lo_bio_list)) |
35a82d1a | 611 | continue; |
35a82d1a | 612 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 613 | bio = loop_get_bio(lo); |
35a82d1a NP |
614 | spin_unlock_irq(&lo->lo_lock); |
615 | ||
616 | BUG_ON(!bio); | |
1da177e4 | 617 | loop_handle_bio(lo, bio); |
1da177e4 LT |
618 | } |
619 | ||
1da177e4 LT |
620 | return 0; |
621 | } | |
622 | ||
623 | /* | |
624 | * loop_switch performs the hard work of switching a backing store. | |
625 | * First it needs to flush existing IO, it does this by sending a magic | |
626 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
627 | */ | |
628 | static int loop_switch(struct loop_device *lo, struct file *file) | |
629 | { | |
630 | struct switch_request w; | |
a24eab1e | 631 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); |
1da177e4 LT |
632 | if (!bio) |
633 | return -ENOMEM; | |
634 | init_completion(&w.wait); | |
635 | w.file = file; | |
636 | bio->bi_private = &w; | |
637 | bio->bi_bdev = NULL; | |
638 | loop_make_request(lo->lo_queue, bio); | |
639 | wait_for_completion(&w.wait); | |
640 | return 0; | |
641 | } | |
642 | ||
14f27939 MB |
643 | /* |
644 | * Helper to flush the IOs in loop, but keeping loop thread running | |
645 | */ | |
646 | static int loop_flush(struct loop_device *lo) | |
647 | { | |
648 | /* loop not yet configured, no running thread, nothing to flush */ | |
649 | if (!lo->lo_thread) | |
650 | return 0; | |
651 | ||
652 | return loop_switch(lo, NULL); | |
653 | } | |
654 | ||
1da177e4 LT |
655 | /* |
656 | * Do the actual switch; called from the BIO completion routine | |
657 | */ | |
658 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
659 | { | |
660 | struct file *file = p->file; | |
661 | struct file *old_file = lo->lo_backing_file; | |
14f27939 MB |
662 | struct address_space *mapping; |
663 | ||
664 | /* if no new file, only flush of queued bios requested */ | |
665 | if (!file) | |
666 | goto out; | |
1da177e4 | 667 | |
14f27939 | 668 | mapping = file->f_mapping; |
1da177e4 LT |
669 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); |
670 | lo->lo_backing_file = file; | |
ba52de12 TT |
671 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
672 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 LT |
673 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
674 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
14f27939 | 675 | out: |
1da177e4 LT |
676 | complete(&p->wait); |
677 | } | |
678 | ||
679 | ||
680 | /* | |
681 | * loop_change_fd switched the backing store of a loopback device to | |
682 | * a new file. This is useful for operating system installers to free up | |
683 | * the original file and in High Availability environments to switch to | |
684 | * an alternative location for the content in case of server meltdown. | |
685 | * This can only work if the loop device is used read-only, and if the | |
686 | * new backing store is the same size and type as the old backing store. | |
687 | */ | |
bb214884 AV |
688 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
689 | unsigned int arg) | |
1da177e4 LT |
690 | { |
691 | struct file *file, *old_file; | |
692 | struct inode *inode; | |
693 | int error; | |
694 | ||
695 | error = -ENXIO; | |
696 | if (lo->lo_state != Lo_bound) | |
697 | goto out; | |
698 | ||
699 | /* the loop device has to be read-only */ | |
700 | error = -EINVAL; | |
701 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
702 | goto out; | |
703 | ||
704 | error = -EBADF; | |
705 | file = fget(arg); | |
706 | if (!file) | |
707 | goto out; | |
708 | ||
709 | inode = file->f_mapping->host; | |
710 | old_file = lo->lo_backing_file; | |
711 | ||
712 | error = -EINVAL; | |
713 | ||
714 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
715 | goto out_putf; | |
716 | ||
1da177e4 LT |
717 | /* size of the new backing store needs to be the same */ |
718 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
719 | goto out_putf; | |
720 | ||
721 | /* and ... switch */ | |
722 | error = loop_switch(lo, file); | |
723 | if (error) | |
724 | goto out_putf; | |
725 | ||
726 | fput(old_file); | |
e03c8dd1 | 727 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) |
476a4813 | 728 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
1da177e4 LT |
729 | return 0; |
730 | ||
731 | out_putf: | |
732 | fput(file); | |
733 | out: | |
734 | return error; | |
735 | } | |
736 | ||
737 | static inline int is_loop_device(struct file *file) | |
738 | { | |
739 | struct inode *i = file->f_mapping->host; | |
740 | ||
741 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
742 | } | |
743 | ||
ee862730 MB |
744 | /* loop sysfs attributes */ |
745 | ||
746 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
747 | ssize_t (*callback)(struct loop_device *, char *)) | |
748 | { | |
749 | struct loop_device *l, *lo = NULL; | |
750 | ||
751 | mutex_lock(&loop_devices_mutex); | |
752 | list_for_each_entry(l, &loop_devices, lo_list) | |
753 | if (disk_to_dev(l->lo_disk) == dev) { | |
754 | lo = l; | |
755 | break; | |
756 | } | |
757 | mutex_unlock(&loop_devices_mutex); | |
758 | ||
759 | return lo ? callback(lo, page) : -EIO; | |
760 | } | |
761 | ||
762 | #define LOOP_ATTR_RO(_name) \ | |
763 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
764 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
765 | struct device_attribute *attr, char *b) \ | |
766 | { \ | |
767 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
768 | } \ | |
769 | static struct device_attribute loop_attr_##_name = \ | |
770 | __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL); | |
771 | ||
772 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
773 | { | |
774 | ssize_t ret; | |
775 | char *p = NULL; | |
776 | ||
777 | mutex_lock(&lo->lo_ctl_mutex); | |
778 | if (lo->lo_backing_file) | |
779 | p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1); | |
780 | mutex_unlock(&lo->lo_ctl_mutex); | |
781 | ||
782 | if (IS_ERR_OR_NULL(p)) | |
783 | ret = PTR_ERR(p); | |
784 | else { | |
785 | ret = strlen(p); | |
786 | memmove(buf, p, ret); | |
787 | buf[ret++] = '\n'; | |
788 | buf[ret] = 0; | |
789 | } | |
790 | ||
791 | return ret; | |
792 | } | |
793 | ||
794 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
795 | { | |
796 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); | |
797 | } | |
798 | ||
799 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
800 | { | |
801 | return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); | |
802 | } | |
803 | ||
804 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
805 | { | |
806 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
807 | ||
808 | return sprintf(buf, "%s\n", autoclear ? "1" : "0"); | |
809 | } | |
810 | ||
e03c8dd1 KS |
811 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
812 | { | |
813 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
814 | ||
815 | return sprintf(buf, "%s\n", partscan ? "1" : "0"); | |
816 | } | |
817 | ||
ee862730 MB |
818 | LOOP_ATTR_RO(backing_file); |
819 | LOOP_ATTR_RO(offset); | |
820 | LOOP_ATTR_RO(sizelimit); | |
821 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 822 | LOOP_ATTR_RO(partscan); |
ee862730 MB |
823 | |
824 | static struct attribute *loop_attrs[] = { | |
825 | &loop_attr_backing_file.attr, | |
826 | &loop_attr_offset.attr, | |
827 | &loop_attr_sizelimit.attr, | |
828 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 829 | &loop_attr_partscan.attr, |
ee862730 MB |
830 | NULL, |
831 | }; | |
832 | ||
833 | static struct attribute_group loop_attribute_group = { | |
834 | .name = "loop", | |
835 | .attrs= loop_attrs, | |
836 | }; | |
837 | ||
838 | static int loop_sysfs_init(struct loop_device *lo) | |
839 | { | |
840 | return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, | |
841 | &loop_attribute_group); | |
842 | } | |
843 | ||
844 | static void loop_sysfs_exit(struct loop_device *lo) | |
845 | { | |
846 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
847 | &loop_attribute_group); | |
848 | } | |
849 | ||
dfaa2ef6 LC |
850 | static void loop_config_discard(struct loop_device *lo) |
851 | { | |
852 | struct file *file = lo->lo_backing_file; | |
853 | struct inode *inode = file->f_mapping->host; | |
854 | struct request_queue *q = lo->lo_queue; | |
855 | ||
856 | /* | |
857 | * We use punch hole to reclaim the free space used by the | |
858 | * image a.k.a. discard. However we do support discard if | |
859 | * encryption is enabled, because it may give an attacker | |
860 | * useful information. | |
861 | */ | |
862 | if ((!file->f_op->fallocate) || | |
863 | lo->lo_encrypt_key_size) { | |
864 | q->limits.discard_granularity = 0; | |
865 | q->limits.discard_alignment = 0; | |
866 | q->limits.max_discard_sectors = 0; | |
867 | q->limits.discard_zeroes_data = 0; | |
868 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
869 | return; | |
870 | } | |
871 | ||
872 | q->limits.discard_granularity = inode->i_sb->s_blocksize; | |
873 | q->limits.discard_alignment = inode->i_sb->s_blocksize; | |
874 | q->limits.max_discard_sectors = UINT_MAX >> 9; | |
875 | q->limits.discard_zeroes_data = 1; | |
876 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
877 | } | |
878 | ||
bb214884 | 879 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
880 | struct block_device *bdev, unsigned int arg) |
881 | { | |
882 | struct file *file, *f; | |
883 | struct inode *inode; | |
884 | struct address_space *mapping; | |
885 | unsigned lo_blocksize; | |
886 | int lo_flags = 0; | |
887 | int error; | |
888 | loff_t size; | |
889 | ||
890 | /* This is safe, since we have a reference from open(). */ | |
891 | __module_get(THIS_MODULE); | |
892 | ||
893 | error = -EBADF; | |
894 | file = fget(arg); | |
895 | if (!file) | |
896 | goto out; | |
897 | ||
898 | error = -EBUSY; | |
899 | if (lo->lo_state != Lo_unbound) | |
900 | goto out_putf; | |
901 | ||
902 | /* Avoid recursion */ | |
903 | f = file; | |
904 | while (is_loop_device(f)) { | |
905 | struct loop_device *l; | |
906 | ||
bb214884 | 907 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
908 | goto out_putf; |
909 | ||
910 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
911 | if (l->lo_state == Lo_unbound) { | |
912 | error = -EINVAL; | |
913 | goto out_putf; | |
914 | } | |
915 | f = l->lo_backing_file; | |
916 | } | |
917 | ||
918 | mapping = file->f_mapping; | |
919 | inode = mapping->host; | |
920 | ||
921 | if (!(file->f_mode & FMODE_WRITE)) | |
922 | lo_flags |= LO_FLAGS_READ_ONLY; | |
923 | ||
924 | error = -EINVAL; | |
925 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
f5e54d6e | 926 | const struct address_space_operations *aops = mapping->a_ops; |
6818173b | 927 | |
4e02ed4b | 928 | if (aops->write_begin) |
1da177e4 LT |
929 | lo_flags |= LO_FLAGS_USE_AOPS; |
930 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) | |
931 | lo_flags |= LO_FLAGS_READ_ONLY; | |
932 | ||
ba52de12 TT |
933 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
934 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
935 | ||
1da177e4 LT |
936 | error = 0; |
937 | } else { | |
938 | goto out_putf; | |
939 | } | |
940 | ||
941 | size = get_loop_size(lo, file); | |
942 | ||
943 | if ((loff_t)(sector_t)size != size) { | |
944 | error = -EFBIG; | |
945 | goto out_putf; | |
946 | } | |
947 | ||
bb214884 | 948 | if (!(mode & FMODE_WRITE)) |
1da177e4 LT |
949 | lo_flags |= LO_FLAGS_READ_ONLY; |
950 | ||
951 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
952 | ||
953 | lo->lo_blocksize = lo_blocksize; | |
954 | lo->lo_device = bdev; | |
955 | lo->lo_flags = lo_flags; | |
956 | lo->lo_backing_file = file; | |
eefe85ee | 957 | lo->transfer = transfer_none; |
1da177e4 LT |
958 | lo->ioctl = NULL; |
959 | lo->lo_sizelimit = 0; | |
960 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
961 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
962 | ||
e686307f | 963 | bio_list_init(&lo->lo_bio_list); |
1da177e4 LT |
964 | |
965 | /* | |
966 | * set queue make_request_fn, and add limits based on lower level | |
967 | * device | |
968 | */ | |
969 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
970 | lo->lo_queue->queuedata = lo; | |
1da177e4 | 971 | |
68db1961 | 972 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
4913efe4 | 973 | blk_queue_flush(lo->lo_queue, REQ_FLUSH); |
68db1961 | 974 | |
73285082 | 975 | set_capacity(lo->lo_disk, size); |
1da177e4 | 976 | bd_set_size(bdev, size << 9); |
ee862730 | 977 | loop_sysfs_init(lo); |
c3473c63 DZ |
978 | /* let user-space know about the new size */ |
979 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 LT |
980 | |
981 | set_blocksize(bdev, lo_blocksize); | |
982 | ||
6c997918 SH |
983 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", |
984 | lo->lo_number); | |
985 | if (IS_ERR(lo->lo_thread)) { | |
986 | error = PTR_ERR(lo->lo_thread); | |
a7422bf8 | 987 | goto out_clr; |
6c997918 SH |
988 | } |
989 | lo->lo_state = Lo_bound; | |
990 | wake_up_process(lo->lo_thread); | |
e03c8dd1 KS |
991 | if (part_shift) |
992 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
993 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) | |
476a4813 | 994 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
1da177e4 LT |
995 | return 0; |
996 | ||
a7422bf8 | 997 | out_clr: |
ee862730 | 998 | loop_sysfs_exit(lo); |
a7422bf8 SH |
999 | lo->lo_thread = NULL; |
1000 | lo->lo_device = NULL; | |
1001 | lo->lo_backing_file = NULL; | |
1002 | lo->lo_flags = 0; | |
73285082 | 1003 | set_capacity(lo->lo_disk, 0); |
f98393a6 | 1004 | invalidate_bdev(bdev); |
a7422bf8 | 1005 | bd_set_size(bdev, 0); |
c3473c63 | 1006 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); |
a7422bf8 SH |
1007 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); |
1008 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1009 | out_putf: |
1010 | fput(file); | |
1011 | out: | |
1012 | /* This is safe: open() is still holding a reference. */ | |
1013 | module_put(THIS_MODULE); | |
1014 | return error; | |
1015 | } | |
1016 | ||
1017 | static int | |
1018 | loop_release_xfer(struct loop_device *lo) | |
1019 | { | |
1020 | int err = 0; | |
1021 | struct loop_func_table *xfer = lo->lo_encryption; | |
1022 | ||
1023 | if (xfer) { | |
1024 | if (xfer->release) | |
1025 | err = xfer->release(lo); | |
1026 | lo->transfer = NULL; | |
1027 | lo->lo_encryption = NULL; | |
1028 | module_put(xfer->owner); | |
1029 | } | |
1030 | return err; | |
1031 | } | |
1032 | ||
1033 | static int | |
1034 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1035 | const struct loop_info64 *i) | |
1036 | { | |
1037 | int err = 0; | |
1038 | ||
1039 | if (xfer) { | |
1040 | struct module *owner = xfer->owner; | |
1041 | ||
1042 | if (!try_module_get(owner)) | |
1043 | return -EINVAL; | |
1044 | if (xfer->init) | |
1045 | err = xfer->init(lo, i); | |
1046 | if (err) | |
1047 | module_put(owner); | |
1048 | else | |
1049 | lo->lo_encryption = xfer; | |
1050 | } | |
1051 | return err; | |
1052 | } | |
1053 | ||
1054 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev) | |
1055 | { | |
1056 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 1057 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 LT |
1058 | |
1059 | if (lo->lo_state != Lo_bound) | |
1060 | return -ENXIO; | |
1061 | ||
1062 | if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */ | |
1063 | return -EBUSY; | |
1064 | ||
1065 | if (filp == NULL) | |
1066 | return -EINVAL; | |
1067 | ||
1068 | spin_lock_irq(&lo->lo_lock); | |
1069 | lo->lo_state = Lo_rundown; | |
1da177e4 LT |
1070 | spin_unlock_irq(&lo->lo_lock); |
1071 | ||
6c997918 | 1072 | kthread_stop(lo->lo_thread); |
1da177e4 LT |
1073 | |
1074 | lo->lo_backing_file = NULL; | |
1075 | ||
1076 | loop_release_xfer(lo); | |
1077 | lo->transfer = NULL; | |
1078 | lo->ioctl = NULL; | |
1079 | lo->lo_device = NULL; | |
1080 | lo->lo_encryption = NULL; | |
1081 | lo->lo_offset = 0; | |
1082 | lo->lo_sizelimit = 0; | |
1083 | lo->lo_encrypt_key_size = 0; | |
6c997918 | 1084 | lo->lo_thread = NULL; |
1da177e4 LT |
1085 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1086 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1087 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
bb214884 AV |
1088 | if (bdev) |
1089 | invalidate_bdev(bdev); | |
73285082 | 1090 | set_capacity(lo->lo_disk, 0); |
51a0bb0c | 1091 | loop_sysfs_exit(lo); |
c3473c63 | 1092 | if (bdev) { |
bb214884 | 1093 | bd_set_size(bdev, 0); |
c3473c63 DZ |
1094 | /* let user-space know about this change */ |
1095 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1096 | } | |
1da177e4 LT |
1097 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1098 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1099 | /* This is safe: open() is still holding a reference. */ |
1100 | module_put(THIS_MODULE); | |
e03c8dd1 | 1101 | if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev) |
476a4813 | 1102 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
e03c8dd1 KS |
1103 | lo->lo_flags = 0; |
1104 | if (!part_shift) | |
1105 | lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; | |
f028f3b2 NK |
1106 | mutex_unlock(&lo->lo_ctl_mutex); |
1107 | /* | |
1108 | * Need not hold lo_ctl_mutex to fput backing file. | |
1109 | * Calling fput holding lo_ctl_mutex triggers a circular | |
1110 | * lock dependency possibility warning as fput can take | |
1111 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
1112 | */ | |
1113 | fput(filp); | |
1da177e4 LT |
1114 | return 0; |
1115 | } | |
1116 | ||
1117 | static int | |
1118 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1119 | { | |
1120 | int err; | |
1121 | struct loop_func_table *xfer; | |
b0fafa81 | 1122 | uid_t uid = current_uid(); |
1da177e4 | 1123 | |
b0fafa81 DH |
1124 | if (lo->lo_encrypt_key_size && |
1125 | lo->lo_key_owner != uid && | |
1da177e4 LT |
1126 | !capable(CAP_SYS_ADMIN)) |
1127 | return -EPERM; | |
1128 | if (lo->lo_state != Lo_bound) | |
1129 | return -ENXIO; | |
1130 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1131 | return -EINVAL; | |
1132 | ||
1133 | err = loop_release_xfer(lo); | |
1134 | if (err) | |
1135 | return err; | |
1136 | ||
1137 | if (info->lo_encrypt_type) { | |
1138 | unsigned int type = info->lo_encrypt_type; | |
1139 | ||
1140 | if (type >= MAX_LO_CRYPT) | |
1141 | return -EINVAL; | |
1142 | xfer = xfer_funcs[type]; | |
1143 | if (xfer == NULL) | |
1144 | return -EINVAL; | |
1145 | } else | |
1146 | xfer = NULL; | |
1147 | ||
1148 | err = loop_init_xfer(lo, xfer, info); | |
1149 | if (err) | |
1150 | return err; | |
1151 | ||
1152 | if (lo->lo_offset != info->lo_offset || | |
1153 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1154 | lo->lo_offset = info->lo_offset; | |
1155 | lo->lo_sizelimit = info->lo_sizelimit; | |
1156 | if (figure_loop_size(lo)) | |
1157 | return -EFBIG; | |
1158 | } | |
dfaa2ef6 | 1159 | loop_config_discard(lo); |
1da177e4 LT |
1160 | |
1161 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1162 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1163 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1164 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1165 | ||
1166 | if (!xfer) | |
1167 | xfer = &none_funcs; | |
1168 | lo->transfer = xfer->transfer; | |
1169 | lo->ioctl = xfer->ioctl; | |
1170 | ||
96c58655 DW |
1171 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1172 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1173 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1174 | ||
e03c8dd1 KS |
1175 | if ((info->lo_flags & LO_FLAGS_PARTSCAN) && |
1176 | !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { | |
1177 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
1178 | lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; | |
1179 | ioctl_by_bdev(lo->lo_device, BLKRRPART, 0); | |
1180 | } | |
1181 | ||
1da177e4 LT |
1182 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1183 | lo->lo_init[0] = info->lo_init[0]; | |
1184 | lo->lo_init[1] = info->lo_init[1]; | |
1185 | if (info->lo_encrypt_key_size) { | |
1186 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1187 | info->lo_encrypt_key_size); | |
b0fafa81 | 1188 | lo->lo_key_owner = uid; |
1da177e4 LT |
1189 | } |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | static int | |
1195 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1196 | { | |
1197 | struct file *file = lo->lo_backing_file; | |
1198 | struct kstat stat; | |
1199 | int error; | |
1200 | ||
1201 | if (lo->lo_state != Lo_bound) | |
1202 | return -ENXIO; | |
6c648be6 | 1203 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); |
1da177e4 LT |
1204 | if (error) |
1205 | return error; | |
1206 | memset(info, 0, sizeof(*info)); | |
1207 | info->lo_number = lo->lo_number; | |
1208 | info->lo_device = huge_encode_dev(stat.dev); | |
1209 | info->lo_inode = stat.ino; | |
1210 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1211 | info->lo_offset = lo->lo_offset; | |
1212 | info->lo_sizelimit = lo->lo_sizelimit; | |
1213 | info->lo_flags = lo->lo_flags; | |
1214 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1215 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1216 | info->lo_encrypt_type = | |
1217 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1218 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1219 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1220 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1221 | lo->lo_encrypt_key_size); | |
1222 | } | |
1223 | return 0; | |
1224 | } | |
1225 | ||
1226 | static void | |
1227 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1228 | { | |
1229 | memset(info64, 0, sizeof(*info64)); | |
1230 | info64->lo_number = info->lo_number; | |
1231 | info64->lo_device = info->lo_device; | |
1232 | info64->lo_inode = info->lo_inode; | |
1233 | info64->lo_rdevice = info->lo_rdevice; | |
1234 | info64->lo_offset = info->lo_offset; | |
1235 | info64->lo_sizelimit = 0; | |
1236 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1237 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1238 | info64->lo_flags = info->lo_flags; | |
1239 | info64->lo_init[0] = info->lo_init[0]; | |
1240 | info64->lo_init[1] = info->lo_init[1]; | |
1241 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1242 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1243 | else | |
1244 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1245 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1246 | } | |
1247 | ||
1248 | static int | |
1249 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1250 | { | |
1251 | memset(info, 0, sizeof(*info)); | |
1252 | info->lo_number = info64->lo_number; | |
1253 | info->lo_device = info64->lo_device; | |
1254 | info->lo_inode = info64->lo_inode; | |
1255 | info->lo_rdevice = info64->lo_rdevice; | |
1256 | info->lo_offset = info64->lo_offset; | |
1257 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1258 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1259 | info->lo_flags = info64->lo_flags; | |
1260 | info->lo_init[0] = info64->lo_init[0]; | |
1261 | info->lo_init[1] = info64->lo_init[1]; | |
1262 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1263 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1264 | else | |
1265 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1266 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1267 | ||
1268 | /* error in case values were truncated */ | |
1269 | if (info->lo_device != info64->lo_device || | |
1270 | info->lo_rdevice != info64->lo_rdevice || | |
1271 | info->lo_inode != info64->lo_inode || | |
1272 | info->lo_offset != info64->lo_offset) | |
1273 | return -EOVERFLOW; | |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | static int | |
1279 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1280 | { | |
1281 | struct loop_info info; | |
1282 | struct loop_info64 info64; | |
1283 | ||
1284 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1285 | return -EFAULT; | |
1286 | loop_info64_from_old(&info, &info64); | |
1287 | return loop_set_status(lo, &info64); | |
1288 | } | |
1289 | ||
1290 | static int | |
1291 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1292 | { | |
1293 | struct loop_info64 info64; | |
1294 | ||
1295 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1296 | return -EFAULT; | |
1297 | return loop_set_status(lo, &info64); | |
1298 | } | |
1299 | ||
1300 | static int | |
1301 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1302 | struct loop_info info; | |
1303 | struct loop_info64 info64; | |
1304 | int err = 0; | |
1305 | ||
1306 | if (!arg) | |
1307 | err = -EINVAL; | |
1308 | if (!err) | |
1309 | err = loop_get_status(lo, &info64); | |
1310 | if (!err) | |
1311 | err = loop_info64_to_old(&info64, &info); | |
1312 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1313 | err = -EFAULT; | |
1314 | ||
1315 | return err; | |
1316 | } | |
1317 | ||
1318 | static int | |
1319 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1320 | struct loop_info64 info64; | |
1321 | int err = 0; | |
1322 | ||
1323 | if (!arg) | |
1324 | err = -EINVAL; | |
1325 | if (!err) | |
1326 | err = loop_get_status(lo, &info64); | |
1327 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1328 | err = -EFAULT; | |
1329 | ||
1330 | return err; | |
1331 | } | |
1332 | ||
53d66608 O |
1333 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) |
1334 | { | |
1335 | int err; | |
1336 | sector_t sec; | |
1337 | loff_t sz; | |
1338 | ||
1339 | err = -ENXIO; | |
1340 | if (unlikely(lo->lo_state != Lo_bound)) | |
1341 | goto out; | |
1342 | err = figure_loop_size(lo); | |
1343 | if (unlikely(err)) | |
1344 | goto out; | |
1345 | sec = get_capacity(lo->lo_disk); | |
1346 | /* the width of sector_t may be narrow for bit-shift */ | |
1347 | sz = sec; | |
1348 | sz <<= 9; | |
1349 | mutex_lock(&bdev->bd_mutex); | |
1350 | bd_set_size(bdev, sz); | |
c3473c63 DZ |
1351 | /* let user-space know about the new size */ |
1352 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
53d66608 O |
1353 | mutex_unlock(&bdev->bd_mutex); |
1354 | ||
1355 | out: | |
1356 | return err; | |
1357 | } | |
1358 | ||
bb214884 | 1359 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1360 | unsigned int cmd, unsigned long arg) |
1361 | { | |
bb214884 | 1362 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1363 | int err; |
1364 | ||
f028f3b2 | 1365 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1366 | switch (cmd) { |
1367 | case LOOP_SET_FD: | |
bb214884 | 1368 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1369 | break; |
1370 | case LOOP_CHANGE_FD: | |
bb214884 | 1371 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1372 | break; |
1373 | case LOOP_CLR_FD: | |
f028f3b2 | 1374 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
bb214884 | 1375 | err = loop_clr_fd(lo, bdev); |
f028f3b2 NK |
1376 | if (!err) |
1377 | goto out_unlocked; | |
1da177e4 LT |
1378 | break; |
1379 | case LOOP_SET_STATUS: | |
1380 | err = loop_set_status_old(lo, (struct loop_info __user *) arg); | |
1381 | break; | |
1382 | case LOOP_GET_STATUS: | |
1383 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1384 | break; | |
1385 | case LOOP_SET_STATUS64: | |
1386 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg); | |
1387 | break; | |
1388 | case LOOP_GET_STATUS64: | |
1389 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1390 | break; | |
53d66608 O |
1391 | case LOOP_SET_CAPACITY: |
1392 | err = -EPERM; | |
1393 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1394 | err = loop_set_capacity(lo, bdev); | |
1395 | break; | |
1da177e4 LT |
1396 | default: |
1397 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1398 | } | |
f85221dd | 1399 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1400 | |
1401 | out_unlocked: | |
1da177e4 LT |
1402 | return err; |
1403 | } | |
1404 | ||
863d5b82 DH |
1405 | #ifdef CONFIG_COMPAT |
1406 | struct compat_loop_info { | |
1407 | compat_int_t lo_number; /* ioctl r/o */ | |
1408 | compat_dev_t lo_device; /* ioctl r/o */ | |
1409 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1410 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1411 | compat_int_t lo_offset; | |
1412 | compat_int_t lo_encrypt_type; | |
1413 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1414 | compat_int_t lo_flags; /* ioctl r/o */ | |
1415 | char lo_name[LO_NAME_SIZE]; | |
1416 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1417 | compat_ulong_t lo_init[2]; | |
1418 | char reserved[4]; | |
1419 | }; | |
1420 | ||
1421 | /* | |
1422 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1423 | * - noinlined to reduce stack space usage in main part of driver | |
1424 | */ | |
1425 | static noinline int | |
ba674cfc | 1426 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1427 | struct loop_info64 *info64) |
1428 | { | |
1429 | struct compat_loop_info info; | |
1430 | ||
1431 | if (copy_from_user(&info, arg, sizeof(info))) | |
1432 | return -EFAULT; | |
1433 | ||
1434 | memset(info64, 0, sizeof(*info64)); | |
1435 | info64->lo_number = info.lo_number; | |
1436 | info64->lo_device = info.lo_device; | |
1437 | info64->lo_inode = info.lo_inode; | |
1438 | info64->lo_rdevice = info.lo_rdevice; | |
1439 | info64->lo_offset = info.lo_offset; | |
1440 | info64->lo_sizelimit = 0; | |
1441 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1442 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1443 | info64->lo_flags = info.lo_flags; | |
1444 | info64->lo_init[0] = info.lo_init[0]; | |
1445 | info64->lo_init[1] = info.lo_init[1]; | |
1446 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1447 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1448 | else | |
1449 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1450 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1451 | return 0; | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1456 | * - noinlined to reduce stack space usage in main part of driver | |
1457 | */ | |
1458 | static noinline int | |
1459 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1460 | struct compat_loop_info __user *arg) | |
1461 | { | |
1462 | struct compat_loop_info info; | |
1463 | ||
1464 | memset(&info, 0, sizeof(info)); | |
1465 | info.lo_number = info64->lo_number; | |
1466 | info.lo_device = info64->lo_device; | |
1467 | info.lo_inode = info64->lo_inode; | |
1468 | info.lo_rdevice = info64->lo_rdevice; | |
1469 | info.lo_offset = info64->lo_offset; | |
1470 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1471 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1472 | info.lo_flags = info64->lo_flags; | |
1473 | info.lo_init[0] = info64->lo_init[0]; | |
1474 | info.lo_init[1] = info64->lo_init[1]; | |
1475 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1476 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1477 | else | |
1478 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1479 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1480 | ||
1481 | /* error in case values were truncated */ | |
1482 | if (info.lo_device != info64->lo_device || | |
1483 | info.lo_rdevice != info64->lo_rdevice || | |
1484 | info.lo_inode != info64->lo_inode || | |
1485 | info.lo_offset != info64->lo_offset || | |
1486 | info.lo_init[0] != info64->lo_init[0] || | |
1487 | info.lo_init[1] != info64->lo_init[1]) | |
1488 | return -EOVERFLOW; | |
1489 | ||
1490 | if (copy_to_user(arg, &info, sizeof(info))) | |
1491 | return -EFAULT; | |
1492 | return 0; | |
1493 | } | |
1494 | ||
1495 | static int | |
1496 | loop_set_status_compat(struct loop_device *lo, | |
1497 | const struct compat_loop_info __user *arg) | |
1498 | { | |
1499 | struct loop_info64 info64; | |
1500 | int ret; | |
1501 | ||
1502 | ret = loop_info64_from_compat(arg, &info64); | |
1503 | if (ret < 0) | |
1504 | return ret; | |
1505 | return loop_set_status(lo, &info64); | |
1506 | } | |
1507 | ||
1508 | static int | |
1509 | loop_get_status_compat(struct loop_device *lo, | |
1510 | struct compat_loop_info __user *arg) | |
1511 | { | |
1512 | struct loop_info64 info64; | |
1513 | int err = 0; | |
1514 | ||
1515 | if (!arg) | |
1516 | err = -EINVAL; | |
1517 | if (!err) | |
1518 | err = loop_get_status(lo, &info64); | |
1519 | if (!err) | |
1520 | err = loop_info64_to_compat(&info64, arg); | |
1521 | return err; | |
1522 | } | |
1523 | ||
bb214884 AV |
1524 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1525 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1526 | { |
bb214884 | 1527 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1528 | int err; |
1529 | ||
863d5b82 DH |
1530 | switch(cmd) { |
1531 | case LOOP_SET_STATUS: | |
1532 | mutex_lock(&lo->lo_ctl_mutex); | |
1533 | err = loop_set_status_compat( | |
1534 | lo, (const struct compat_loop_info __user *) arg); | |
1535 | mutex_unlock(&lo->lo_ctl_mutex); | |
1536 | break; | |
1537 | case LOOP_GET_STATUS: | |
1538 | mutex_lock(&lo->lo_ctl_mutex); | |
1539 | err = loop_get_status_compat( | |
1540 | lo, (struct compat_loop_info __user *) arg); | |
1541 | mutex_unlock(&lo->lo_ctl_mutex); | |
1542 | break; | |
53d66608 | 1543 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1544 | case LOOP_CLR_FD: |
1545 | case LOOP_GET_STATUS64: | |
1546 | case LOOP_SET_STATUS64: | |
1547 | arg = (unsigned long) compat_ptr(arg); | |
1548 | case LOOP_SET_FD: | |
1549 | case LOOP_CHANGE_FD: | |
bb214884 | 1550 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1551 | break; |
1552 | default: | |
1553 | err = -ENOIOCTLCMD; | |
1554 | break; | |
1555 | } | |
863d5b82 DH |
1556 | return err; |
1557 | } | |
1558 | #endif | |
1559 | ||
bb214884 | 1560 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1561 | { |
bb214884 | 1562 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 | 1563 | |
f85221dd | 1564 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1565 | lo->lo_refcnt++; |
f85221dd | 1566 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1567 | |
1568 | return 0; | |
1569 | } | |
1570 | ||
bb214884 | 1571 | static int lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1572 | { |
bb214884 | 1573 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1574 | int err; |
1da177e4 | 1575 | |
f85221dd | 1576 | mutex_lock(&lo->lo_ctl_mutex); |
96c58655 | 1577 | |
14f27939 MB |
1578 | if (--lo->lo_refcnt) |
1579 | goto out; | |
1580 | ||
1581 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
1582 | /* | |
1583 | * In autoclear mode, stop the loop thread | |
1584 | * and remove configuration after last close. | |
1585 | */ | |
8a9c5944 | 1586 | err = loop_clr_fd(lo, lo->lo_device); |
ffcd7dca AB |
1587 | if (!err) |
1588 | goto out_unlocked; | |
14f27939 MB |
1589 | } else { |
1590 | /* | |
1591 | * Otherwise keep thread (if running) and config, | |
1592 | * but flush possible ongoing bios in thread. | |
1593 | */ | |
1594 | loop_flush(lo); | |
1595 | } | |
96c58655 | 1596 | |
14f27939 | 1597 | out: |
f85221dd | 1598 | mutex_unlock(&lo->lo_ctl_mutex); |
ffcd7dca | 1599 | out_unlocked: |
1da177e4 LT |
1600 | return 0; |
1601 | } | |
1602 | ||
83d5cde4 | 1603 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1604 | .owner = THIS_MODULE, |
bb214884 AV |
1605 | .open = lo_open, |
1606 | .release = lo_release, | |
1607 | .ioctl = lo_ioctl, | |
863d5b82 | 1608 | #ifdef CONFIG_COMPAT |
bb214884 | 1609 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1610 | #endif |
1da177e4 LT |
1611 | }; |
1612 | ||
1613 | /* | |
1614 | * And now the modules code and kernel interface. | |
1615 | */ | |
73285082 | 1616 | static int max_loop; |
ac04fee0 | 1617 | module_param(max_loop, int, S_IRUGO); |
a47653fc | 1618 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
ac04fee0 | 1619 | module_param(max_part, int, S_IRUGO); |
476a4813 | 1620 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
1da177e4 LT |
1621 | MODULE_LICENSE("GPL"); |
1622 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1623 | ||
1624 | int loop_register_transfer(struct loop_func_table *funcs) | |
1625 | { | |
1626 | unsigned int n = funcs->number; | |
1627 | ||
1628 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1629 | return -EINVAL; | |
1630 | xfer_funcs[n] = funcs; | |
1631 | return 0; | |
1632 | } | |
1633 | ||
1634 | int loop_unregister_transfer(int number) | |
1635 | { | |
1636 | unsigned int n = number; | |
1637 | struct loop_device *lo; | |
1638 | struct loop_func_table *xfer; | |
1639 | ||
1640 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1641 | return -EINVAL; | |
1642 | ||
1643 | xfer_funcs[n] = NULL; | |
1644 | ||
73285082 | 1645 | list_for_each_entry(lo, &loop_devices, lo_list) { |
f85221dd | 1646 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1647 | |
1648 | if (lo->lo_encryption == xfer) | |
1649 | loop_release_xfer(lo); | |
1650 | ||
f85221dd | 1651 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1652 | } |
1653 | ||
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | EXPORT_SYMBOL(loop_register_transfer); | |
1658 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1659 | ||
a47653fc | 1660 | static struct loop_device *loop_alloc(int i) |
73285082 KC |
1661 | { |
1662 | struct loop_device *lo; | |
1663 | struct gendisk *disk; | |
1664 | ||
1665 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); | |
1666 | if (!lo) | |
1667 | goto out; | |
1668 | ||
1669 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); | |
1670 | if (!lo->lo_queue) | |
1671 | goto out_free_dev; | |
1672 | ||
476a4813 | 1673 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1674 | if (!disk) |
1675 | goto out_free_queue; | |
1676 | ||
e03c8dd1 KS |
1677 | /* |
1678 | * Disable partition scanning by default. The in-kernel partition | |
1679 | * scanning can be requested individually per-device during its | |
1680 | * setup. Userspace can always add and remove partitions from all | |
1681 | * devices. The needed partition minors are allocated from the | |
1682 | * extended minor space, the main loop device numbers will continue | |
1683 | * to match the loop minors, regardless of the number of partitions | |
1684 | * used. | |
1685 | * | |
1686 | * If max_part is given, partition scanning is globally enabled for | |
1687 | * all loop devices. The minors for the main loop devices will be | |
1688 | * multiples of max_part. | |
1689 | * | |
1690 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
1691 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
1692 | * complicated, are too static, inflexible and may surprise | |
1693 | * userspace tools. Parameters like this in general should be avoided. | |
1694 | */ | |
1695 | if (!part_shift) | |
1696 | disk->flags |= GENHD_FL_NO_PART_SCAN; | |
1697 | disk->flags |= GENHD_FL_EXT_DEVT; | |
73285082 KC |
1698 | mutex_init(&lo->lo_ctl_mutex); |
1699 | lo->lo_number = i; | |
1700 | lo->lo_thread = NULL; | |
1701 | init_waitqueue_head(&lo->lo_event); | |
1702 | spin_lock_init(&lo->lo_lock); | |
1703 | disk->major = LOOP_MAJOR; | |
476a4813 | 1704 | disk->first_minor = i << part_shift; |
73285082 KC |
1705 | disk->fops = &lo_fops; |
1706 | disk->private_data = lo; | |
1707 | disk->queue = lo->lo_queue; | |
1708 | sprintf(disk->disk_name, "loop%d", i); | |
73285082 KC |
1709 | return lo; |
1710 | ||
1711 | out_free_queue: | |
1712 | blk_cleanup_queue(lo->lo_queue); | |
1713 | out_free_dev: | |
1714 | kfree(lo); | |
1715 | out: | |
07002e99 | 1716 | return NULL; |
73285082 KC |
1717 | } |
1718 | ||
a47653fc | 1719 | static void loop_free(struct loop_device *lo) |
1da177e4 | 1720 | { |
73285082 KC |
1721 | blk_cleanup_queue(lo->lo_queue); |
1722 | put_disk(lo->lo_disk); | |
1723 | list_del(&lo->lo_list); | |
1724 | kfree(lo); | |
1725 | } | |
1da177e4 | 1726 | |
a47653fc KC |
1727 | static struct loop_device *loop_init_one(int i) |
1728 | { | |
1729 | struct loop_device *lo; | |
1730 | ||
1731 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1732 | if (lo->lo_number == i) | |
1733 | return lo; | |
1734 | } | |
1735 | ||
1736 | lo = loop_alloc(i); | |
1737 | if (lo) { | |
1738 | add_disk(lo->lo_disk); | |
1739 | list_add_tail(&lo->lo_list, &loop_devices); | |
1740 | } | |
1741 | return lo; | |
1742 | } | |
1743 | ||
1744 | static void loop_del_one(struct loop_device *lo) | |
1745 | { | |
1746 | del_gendisk(lo->lo_disk); | |
1747 | loop_free(lo); | |
1748 | } | |
1749 | ||
73285082 KC |
1750 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1751 | { | |
705962cc | 1752 | struct loop_device *lo; |
07002e99 | 1753 | struct kobject *kobj; |
73285082 | 1754 | |
705962cc | 1755 | mutex_lock(&loop_devices_mutex); |
a1c15c59 | 1756 | lo = loop_init_one(MINOR(dev) >> part_shift); |
07002e99 | 1757 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM); |
73285082 KC |
1758 | mutex_unlock(&loop_devices_mutex); |
1759 | ||
1760 | *part = 0; | |
07002e99 | 1761 | return kobj; |
73285082 KC |
1762 | } |
1763 | ||
1764 | static int __init loop_init(void) | |
1765 | { | |
a47653fc KC |
1766 | int i, nr; |
1767 | unsigned long range; | |
1768 | struct loop_device *lo, *next; | |
1769 | ||
1770 | /* | |
1771 | * loop module now has a feature to instantiate underlying device | |
1772 | * structure on-demand, provided that there is an access dev node. | |
1773 | * However, this will not work well with user space tool that doesn't | |
1774 | * know about such "feature". In order to not break any existing | |
1775 | * tool, we do the following: | |
1776 | * | |
1777 | * (1) if max_loop is specified, create that many upfront, and this | |
1778 | * also becomes a hard limit. | |
1779 | * (2) if max_loop is not specified, create 8 loop device on module | |
1780 | * load, user can further extend loop device by create dev node | |
1781 | * themselves and have kernel automatically instantiate actual | |
1782 | * device on-demand. | |
1783 | */ | |
476a4813 LV |
1784 | |
1785 | part_shift = 0; | |
ac04fee0 | 1786 | if (max_part > 0) { |
476a4813 LV |
1787 | part_shift = fls(max_part); |
1788 | ||
ac04fee0 NK |
1789 | /* |
1790 | * Adjust max_part according to part_shift as it is exported | |
1791 | * to user space so that user can decide correct minor number | |
1792 | * if [s]he want to create more devices. | |
1793 | * | |
1794 | * Note that -1 is required because partition 0 is reserved | |
1795 | * for the whole disk. | |
1796 | */ | |
1797 | max_part = (1UL << part_shift) - 1; | |
1798 | } | |
1799 | ||
78f4bb36 NK |
1800 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1801 | return -EINVAL; | |
1802 | ||
476a4813 | 1803 | if (max_loop > 1UL << (MINORBITS - part_shift)) |
a47653fc | 1804 | return -EINVAL; |
1da177e4 | 1805 | |
73285082 | 1806 | if (max_loop) { |
a47653fc | 1807 | nr = max_loop; |
a1c15c59 | 1808 | range = max_loop << part_shift; |
a47653fc KC |
1809 | } else { |
1810 | nr = 8; | |
a1c15c59 | 1811 | range = 1UL << MINORBITS; |
a47653fc KC |
1812 | } |
1813 | ||
1814 | if (register_blkdev(LOOP_MAJOR, "loop")) | |
1815 | return -EIO; | |
1da177e4 | 1816 | |
a47653fc KC |
1817 | for (i = 0; i < nr; i++) { |
1818 | lo = loop_alloc(i); | |
1819 | if (!lo) | |
1820 | goto Enomem; | |
1821 | list_add_tail(&lo->lo_list, &loop_devices); | |
1da177e4 | 1822 | } |
a47653fc KC |
1823 | |
1824 | /* point of no return */ | |
1825 | ||
1826 | list_for_each_entry(lo, &loop_devices, lo_list) | |
1827 | add_disk(lo->lo_disk); | |
1828 | ||
1829 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, | |
1830 | THIS_MODULE, loop_probe, NULL, NULL); | |
1831 | ||
73285082 | 1832 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 1833 | return 0; |
a47653fc KC |
1834 | |
1835 | Enomem: | |
1836 | printk(KERN_INFO "loop: out of memory\n"); | |
1837 | ||
1838 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
1839 | loop_free(lo); | |
1840 | ||
1841 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
1842 | return -ENOMEM; | |
1da177e4 LT |
1843 | } |
1844 | ||
73285082 | 1845 | static void __exit loop_exit(void) |
1da177e4 | 1846 | { |
a47653fc | 1847 | unsigned long range; |
73285082 | 1848 | struct loop_device *lo, *next; |
1da177e4 | 1849 | |
a1c15c59 | 1850 | range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; |
a47653fc | 1851 | |
73285082 KC |
1852 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) |
1853 | loop_del_one(lo); | |
1854 | ||
a47653fc | 1855 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 1856 | unregister_blkdev(LOOP_MAJOR, "loop"); |
1da177e4 LT |
1857 | } |
1858 | ||
1859 | module_init(loop_init); | |
1860 | module_exit(loop_exit); | |
1861 | ||
1862 | #ifndef MODULE | |
1863 | static int __init max_loop_setup(char *str) | |
1864 | { | |
1865 | max_loop = simple_strtol(str, NULL, 0); | |
1866 | return 1; | |
1867 | } | |
1868 | ||
1869 | __setup("max_loop=", max_loop_setup); | |
1870 | #endif |