]>
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/smp_lock.h> |
65 | #include <linux/swap.h> | |
66 | #include <linux/slab.h> | |
67 | #include <linux/loop.h> | |
863d5b82 | 68 | #include <linux/compat.h> |
1da177e4 | 69 | #include <linux/suspend.h> |
83144186 | 70 | #include <linux/freezer.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> | |
75 | #include <linux/gfp.h> | |
6c997918 | 76 | #include <linux/kthread.h> |
d6b29d7c | 77 | #include <linux/splice.h> |
1da177e4 LT |
78 | |
79 | #include <asm/uaccess.h> | |
80 | ||
73285082 KC |
81 | static LIST_HEAD(loop_devices); |
82 | static DEFINE_MUTEX(loop_devices_mutex); | |
1da177e4 | 83 | |
476a4813 LV |
84 | static int max_part; |
85 | static int part_shift; | |
86 | ||
1da177e4 LT |
87 | /* |
88 | * Transfer functions | |
89 | */ | |
90 | static int transfer_none(struct loop_device *lo, int cmd, | |
91 | struct page *raw_page, unsigned raw_off, | |
92 | struct page *loop_page, unsigned loop_off, | |
93 | int size, sector_t real_block) | |
94 | { | |
95 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
96 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
97 | ||
98 | if (cmd == READ) | |
99 | memcpy(loop_buf, raw_buf, size); | |
100 | else | |
101 | memcpy(raw_buf, loop_buf, size); | |
102 | ||
103 | kunmap_atomic(raw_buf, KM_USER0); | |
104 | kunmap_atomic(loop_buf, KM_USER1); | |
105 | cond_resched(); | |
106 | return 0; | |
107 | } | |
108 | ||
109 | static int transfer_xor(struct loop_device *lo, int cmd, | |
110 | struct page *raw_page, unsigned raw_off, | |
111 | struct page *loop_page, unsigned loop_off, | |
112 | int size, sector_t real_block) | |
113 | { | |
114 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
115 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
116 | char *in, *out, *key; | |
117 | int i, keysize; | |
118 | ||
119 | if (cmd == READ) { | |
120 | in = raw_buf; | |
121 | out = loop_buf; | |
122 | } else { | |
123 | in = loop_buf; | |
124 | out = raw_buf; | |
125 | } | |
126 | ||
127 | key = lo->lo_encrypt_key; | |
128 | keysize = lo->lo_encrypt_key_size; | |
129 | for (i = 0; i < size; i++) | |
130 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
131 | ||
132 | kunmap_atomic(raw_buf, KM_USER0); | |
133 | kunmap_atomic(loop_buf, KM_USER1); | |
134 | cond_resched(); | |
135 | return 0; | |
136 | } | |
137 | ||
138 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
139 | { | |
140 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
141 | return -EINVAL; | |
142 | return 0; | |
143 | } | |
144 | ||
145 | static struct loop_func_table none_funcs = { | |
146 | .number = LO_CRYPT_NONE, | |
147 | .transfer = transfer_none, | |
148 | }; | |
149 | ||
150 | static struct loop_func_table xor_funcs = { | |
151 | .number = LO_CRYPT_XOR, | |
152 | .transfer = transfer_xor, | |
153 | .init = xor_init | |
154 | }; | |
155 | ||
156 | /* xfer_funcs[0] is special - its release function is never called */ | |
157 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
158 | &none_funcs, | |
159 | &xor_funcs | |
160 | }; | |
161 | ||
162 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) | |
163 | { | |
164 | loff_t size, offset, loopsize; | |
165 | ||
166 | /* Compute loopsize in bytes */ | |
167 | size = i_size_read(file->f_mapping->host); | |
168 | offset = lo->lo_offset; | |
169 | loopsize = size - offset; | |
170 | if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize) | |
171 | loopsize = lo->lo_sizelimit; | |
172 | ||
173 | /* | |
174 | * Unfortunately, if we want to do I/O on the device, | |
175 | * the number of 512-byte sectors has to fit into a sector_t. | |
176 | */ | |
177 | return loopsize >> 9; | |
178 | } | |
179 | ||
180 | static int | |
181 | figure_loop_size(struct loop_device *lo) | |
182 | { | |
183 | loff_t size = get_loop_size(lo, lo->lo_backing_file); | |
184 | sector_t x = (sector_t)size; | |
185 | ||
186 | if (unlikely((loff_t)x != size)) | |
187 | return -EFBIG; | |
188 | ||
73285082 | 189 | set_capacity(lo->lo_disk, x); |
1da177e4 LT |
190 | return 0; |
191 | } | |
192 | ||
193 | static inline int | |
194 | lo_do_transfer(struct loop_device *lo, int cmd, | |
195 | struct page *rpage, unsigned roffs, | |
196 | struct page *lpage, unsigned loffs, | |
197 | int size, sector_t rblock) | |
198 | { | |
199 | if (unlikely(!lo->transfer)) | |
200 | return 0; | |
201 | ||
202 | return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
203 | } | |
204 | ||
205 | /** | |
206 | * do_lo_send_aops - helper for writing data to a loop device | |
207 | * | |
208 | * This is the fast version for backing filesystems which implement the address | |
afddba49 | 209 | * space operations write_begin and write_end. |
1da177e4 LT |
210 | */ |
211 | static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 212 | loff_t pos, struct page *unused) |
1da177e4 LT |
213 | { |
214 | struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */ | |
215 | struct address_space *mapping = file->f_mapping; | |
1da177e4 LT |
216 | pgoff_t index; |
217 | unsigned offset, bv_offs; | |
994fc28c | 218 | int len, ret; |
1da177e4 | 219 | |
1b1dcc1b | 220 | mutex_lock(&mapping->host->i_mutex); |
1da177e4 LT |
221 | index = pos >> PAGE_CACHE_SHIFT; |
222 | offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1); | |
223 | bv_offs = bvec->bv_offset; | |
224 | len = bvec->bv_len; | |
225 | while (len > 0) { | |
226 | sector_t IV; | |
afddba49 | 227 | unsigned size, copied; |
1da177e4 | 228 | int transfer_result; |
afddba49 NP |
229 | struct page *page; |
230 | void *fsdata; | |
1da177e4 LT |
231 | |
232 | IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9); | |
233 | size = PAGE_CACHE_SIZE - offset; | |
234 | if (size > len) | |
235 | size = len; | |
afddba49 NP |
236 | |
237 | ret = pagecache_write_begin(file, mapping, pos, size, 0, | |
238 | &page, &fsdata); | |
239 | if (ret) | |
1da177e4 | 240 | goto fail; |
afddba49 | 241 | |
1da177e4 LT |
242 | transfer_result = lo_do_transfer(lo, WRITE, page, offset, |
243 | bvec->bv_page, bv_offs, size, IV); | |
afddba49 | 244 | copied = size; |
1da177e4 | 245 | if (unlikely(transfer_result)) |
afddba49 NP |
246 | copied = 0; |
247 | ||
248 | ret = pagecache_write_end(file, mapping, pos, size, copied, | |
249 | page, fsdata); | |
8268f5a7 | 250 | if (ret < 0 || ret != copied) |
afddba49 | 251 | goto fail; |
afddba49 NP |
252 | |
253 | if (unlikely(transfer_result)) | |
254 | goto fail; | |
255 | ||
256 | bv_offs += copied; | |
257 | len -= copied; | |
1da177e4 LT |
258 | offset = 0; |
259 | index++; | |
afddba49 | 260 | pos += copied; |
1da177e4 | 261 | } |
994fc28c | 262 | ret = 0; |
1da177e4 | 263 | out: |
1b1dcc1b | 264 | mutex_unlock(&mapping->host->i_mutex); |
1da177e4 | 265 | return ret; |
1da177e4 LT |
266 | fail: |
267 | ret = -1; | |
268 | goto out; | |
269 | } | |
270 | ||
271 | /** | |
272 | * __do_lo_send_write - helper for writing data to a loop device | |
273 | * | |
274 | * This helper just factors out common code between do_lo_send_direct_write() | |
275 | * and do_lo_send_write(). | |
276 | */ | |
858119e1 | 277 | static int __do_lo_send_write(struct file *file, |
98ae6ccd | 278 | u8 *buf, const int len, loff_t pos) |
1da177e4 LT |
279 | { |
280 | ssize_t bw; | |
281 | mm_segment_t old_fs = get_fs(); | |
282 | ||
283 | set_fs(get_ds()); | |
284 | bw = file->f_op->write(file, buf, len, &pos); | |
285 | set_fs(old_fs); | |
286 | if (likely(bw == len)) | |
287 | return 0; | |
288 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", | |
289 | (unsigned long long)pos, len); | |
290 | if (bw >= 0) | |
291 | bw = -EIO; | |
292 | return bw; | |
293 | } | |
294 | ||
295 | /** | |
296 | * do_lo_send_direct_write - helper for writing data to a loop device | |
297 | * | |
298 | * This is the fast, non-transforming version for backing filesystems which do | |
afddba49 | 299 | * not implement the address space operations write_begin and write_end. |
1da177e4 LT |
300 | * It uses the write file operation which should be present on all writeable |
301 | * filesystems. | |
302 | */ | |
303 | static int do_lo_send_direct_write(struct loop_device *lo, | |
511de73f | 304 | struct bio_vec *bvec, loff_t pos, struct page *page) |
1da177e4 LT |
305 | { |
306 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 307 | kmap(bvec->bv_page) + bvec->bv_offset, |
1da177e4 LT |
308 | bvec->bv_len, pos); |
309 | kunmap(bvec->bv_page); | |
310 | cond_resched(); | |
311 | return bw; | |
312 | } | |
313 | ||
314 | /** | |
315 | * do_lo_send_write - helper for writing data to a loop device | |
316 | * | |
317 | * This is the slow, transforming version for filesystems which do not | |
afddba49 | 318 | * implement the address space operations write_begin and write_end. It |
1da177e4 LT |
319 | * uses the write file operation which should be present on all writeable |
320 | * filesystems. | |
321 | * | |
322 | * Using fops->write is slower than using aops->{prepare,commit}_write in the | |
323 | * transforming case because we need to double buffer the data as we cannot do | |
324 | * the transformations in place as we do not have direct access to the | |
325 | * destination pages of the backing file. | |
326 | */ | |
327 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 328 | loff_t pos, struct page *page) |
1da177e4 LT |
329 | { |
330 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, | |
331 | bvec->bv_offset, bvec->bv_len, pos >> 9); | |
332 | if (likely(!ret)) | |
333 | return __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 334 | page_address(page), bvec->bv_len, |
1da177e4 LT |
335 | pos); |
336 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, " | |
337 | "length %i.\n", (unsigned long long)pos, bvec->bv_len); | |
338 | if (ret > 0) | |
339 | ret = -EIO; | |
340 | return ret; | |
341 | } | |
342 | ||
511de73f | 343 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) |
1da177e4 | 344 | { |
511de73f | 345 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, |
1da177e4 LT |
346 | struct page *page); |
347 | struct bio_vec *bvec; | |
348 | struct page *page = NULL; | |
349 | int i, ret = 0; | |
350 | ||
351 | do_lo_send = do_lo_send_aops; | |
352 | if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) { | |
353 | do_lo_send = do_lo_send_direct_write; | |
354 | if (lo->transfer != transfer_none) { | |
355 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
356 | if (unlikely(!page)) | |
357 | goto fail; | |
358 | kmap(page); | |
359 | do_lo_send = do_lo_send_write; | |
360 | } | |
361 | } | |
362 | bio_for_each_segment(bvec, bio, i) { | |
511de73f | 363 | ret = do_lo_send(lo, bvec, pos, page); |
1da177e4 LT |
364 | if (ret < 0) |
365 | break; | |
366 | pos += bvec->bv_len; | |
367 | } | |
368 | if (page) { | |
369 | kunmap(page); | |
370 | __free_page(page); | |
371 | } | |
372 | out: | |
373 | return ret; | |
374 | fail: | |
375 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); | |
376 | ret = -ENOMEM; | |
377 | goto out; | |
378 | } | |
379 | ||
380 | struct lo_read_data { | |
381 | struct loop_device *lo; | |
382 | struct page *page; | |
383 | unsigned offset; | |
384 | int bsize; | |
385 | }; | |
386 | ||
387 | static int | |
fd582140 JA |
388 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
389 | struct splice_desc *sd) | |
1da177e4 | 390 | { |
fd582140 | 391 | struct lo_read_data *p = sd->u.data; |
1da177e4 | 392 | struct loop_device *lo = p->lo; |
fd582140 | 393 | struct page *page = buf->page; |
1da177e4 | 394 | sector_t IV; |
fd582140 JA |
395 | size_t size; |
396 | int ret; | |
1da177e4 | 397 | |
cac36bb0 | 398 | ret = buf->ops->confirm(pipe, buf); |
fd582140 JA |
399 | if (unlikely(ret)) |
400 | return ret; | |
1da177e4 | 401 | |
fd582140 JA |
402 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + |
403 | (buf->offset >> 9); | |
404 | size = sd->len; | |
405 | if (size > p->bsize) | |
406 | size = p->bsize; | |
1da177e4 | 407 | |
fd582140 | 408 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { |
1da177e4 LT |
409 | printk(KERN_ERR "loop: transfer error block %ld\n", |
410 | page->index); | |
fd582140 | 411 | size = -EINVAL; |
1da177e4 LT |
412 | } |
413 | ||
414 | flush_dcache_page(p->page); | |
415 | ||
fd582140 JA |
416 | if (size > 0) |
417 | p->offset += size; | |
418 | ||
1da177e4 LT |
419 | return size; |
420 | } | |
421 | ||
fd582140 JA |
422 | static int |
423 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
424 | { | |
425 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
426 | } | |
427 | ||
1da177e4 LT |
428 | static int |
429 | do_lo_receive(struct loop_device *lo, | |
430 | struct bio_vec *bvec, int bsize, loff_t pos) | |
431 | { | |
432 | struct lo_read_data cookie; | |
fd582140 | 433 | struct splice_desc sd; |
1da177e4 | 434 | struct file *file; |
fd582140 | 435 | long retval; |
1da177e4 LT |
436 | |
437 | cookie.lo = lo; | |
438 | cookie.page = bvec->bv_page; | |
439 | cookie.offset = bvec->bv_offset; | |
440 | cookie.bsize = bsize; | |
fd582140 JA |
441 | |
442 | sd.len = 0; | |
443 | sd.total_len = bvec->bv_len; | |
444 | sd.flags = 0; | |
445 | sd.pos = pos; | |
446 | sd.u.data = &cookie; | |
447 | ||
1da177e4 | 448 | file = lo->lo_backing_file; |
fd582140 JA |
449 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); |
450 | ||
451 | if (retval < 0) | |
452 | return retval; | |
453 | ||
454 | return 0; | |
1da177e4 LT |
455 | } |
456 | ||
457 | static int | |
458 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
459 | { | |
460 | struct bio_vec *bvec; | |
461 | int i, ret = 0; | |
462 | ||
463 | bio_for_each_segment(bvec, bio, i) { | |
464 | ret = do_lo_receive(lo, bvec, bsize, pos); | |
465 | if (ret < 0) | |
466 | break; | |
467 | pos += bvec->bv_len; | |
468 | } | |
469 | return ret; | |
470 | } | |
471 | ||
472 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
473 | { | |
474 | loff_t pos; | |
475 | int ret; | |
476 | ||
477 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
478 | if (bio_rw(bio) == WRITE) | |
511de73f | 479 | ret = lo_send(lo, bio, pos); |
1da177e4 LT |
480 | else |
481 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); | |
482 | return ret; | |
483 | } | |
484 | ||
485 | /* | |
486 | * Add bio to back of pending list | |
487 | */ | |
488 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
489 | { | |
1da177e4 LT |
490 | if (lo->lo_biotail) { |
491 | lo->lo_biotail->bi_next = bio; | |
492 | lo->lo_biotail = bio; | |
493 | } else | |
494 | lo->lo_bio = lo->lo_biotail = bio; | |
1da177e4 LT |
495 | } |
496 | ||
497 | /* | |
498 | * Grab first pending buffer | |
499 | */ | |
500 | static struct bio *loop_get_bio(struct loop_device *lo) | |
501 | { | |
502 | struct bio *bio; | |
503 | ||
1da177e4 LT |
504 | if ((bio = lo->lo_bio)) { |
505 | if (bio == lo->lo_biotail) | |
506 | lo->lo_biotail = NULL; | |
507 | lo->lo_bio = bio->bi_next; | |
508 | bio->bi_next = NULL; | |
509 | } | |
1da177e4 LT |
510 | |
511 | return bio; | |
512 | } | |
513 | ||
165125e1 | 514 | static int loop_make_request(struct request_queue *q, struct bio *old_bio) |
1da177e4 LT |
515 | { |
516 | struct loop_device *lo = q->queuedata; | |
517 | int rw = bio_rw(old_bio); | |
518 | ||
35a82d1a NP |
519 | if (rw == READA) |
520 | rw = READ; | |
521 | ||
522 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
1da177e4 LT |
523 | |
524 | spin_lock_irq(&lo->lo_lock); | |
525 | if (lo->lo_state != Lo_bound) | |
35a82d1a NP |
526 | goto out; |
527 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
528 | goto out; | |
1da177e4 | 529 | loop_add_bio(lo, old_bio); |
6c997918 | 530 | wake_up(&lo->lo_event); |
35a82d1a | 531 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 532 | return 0; |
35a82d1a | 533 | |
1da177e4 | 534 | out: |
35a82d1a | 535 | spin_unlock_irq(&lo->lo_lock); |
6712ecf8 | 536 | bio_io_error(old_bio); |
1da177e4 | 537 | return 0; |
1da177e4 LT |
538 | } |
539 | ||
540 | /* | |
541 | * kick off io on the underlying address space | |
542 | */ | |
165125e1 | 543 | static void loop_unplug(struct request_queue *q) |
1da177e4 LT |
544 | { |
545 | struct loop_device *lo = q->queuedata; | |
546 | ||
75ad23bc | 547 | queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q); |
1da177e4 LT |
548 | blk_run_address_space(lo->lo_backing_file->f_mapping); |
549 | } | |
550 | ||
551 | struct switch_request { | |
552 | struct file *file; | |
553 | struct completion wait; | |
554 | }; | |
555 | ||
556 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
557 | ||
558 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
559 | { | |
1da177e4 LT |
560 | if (unlikely(!bio->bi_bdev)) { |
561 | do_loop_switch(lo, bio->bi_private); | |
562 | bio_put(bio); | |
563 | } else { | |
35a82d1a | 564 | int ret = do_bio_filebacked(lo, bio); |
6712ecf8 | 565 | bio_endio(bio, ret); |
1da177e4 LT |
566 | } |
567 | } | |
568 | ||
569 | /* | |
570 | * worker thread that handles reads/writes to file backed loop devices, | |
571 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
572 | * on reads for block backed loop, as that is too heavy to do from | |
573 | * b_end_io context where irqs may be disabled. | |
6c997918 SH |
574 | * |
575 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
576 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
577 | * true, make_request will not place any more requests. Therefore | |
578 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
579 | * done with the loop. | |
1da177e4 LT |
580 | */ |
581 | static int loop_thread(void *data) | |
582 | { | |
583 | struct loop_device *lo = data; | |
584 | struct bio *bio; | |
585 | ||
1da177e4 LT |
586 | set_user_nice(current, -20); |
587 | ||
6c997918 | 588 | while (!kthread_should_stop() || lo->lo_bio) { |
09c0dc68 | 589 | |
6c997918 SH |
590 | wait_event_interruptible(lo->lo_event, |
591 | lo->lo_bio || kthread_should_stop()); | |
35a82d1a | 592 | |
6c997918 | 593 | if (!lo->lo_bio) |
35a82d1a | 594 | continue; |
35a82d1a | 595 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 596 | bio = loop_get_bio(lo); |
35a82d1a NP |
597 | spin_unlock_irq(&lo->lo_lock); |
598 | ||
599 | BUG_ON(!bio); | |
1da177e4 | 600 | loop_handle_bio(lo, bio); |
1da177e4 LT |
601 | } |
602 | ||
1da177e4 LT |
603 | return 0; |
604 | } | |
605 | ||
606 | /* | |
607 | * loop_switch performs the hard work of switching a backing store. | |
608 | * First it needs to flush existing IO, it does this by sending a magic | |
609 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
610 | */ | |
611 | static int loop_switch(struct loop_device *lo, struct file *file) | |
612 | { | |
613 | struct switch_request w; | |
a24eab1e | 614 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); |
1da177e4 LT |
615 | if (!bio) |
616 | return -ENOMEM; | |
617 | init_completion(&w.wait); | |
618 | w.file = file; | |
619 | bio->bi_private = &w; | |
620 | bio->bi_bdev = NULL; | |
621 | loop_make_request(lo->lo_queue, bio); | |
622 | wait_for_completion(&w.wait); | |
623 | return 0; | |
624 | } | |
625 | ||
626 | /* | |
627 | * Do the actual switch; called from the BIO completion routine | |
628 | */ | |
629 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
630 | { | |
631 | struct file *file = p->file; | |
632 | struct file *old_file = lo->lo_backing_file; | |
633 | struct address_space *mapping = file->f_mapping; | |
634 | ||
635 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
636 | lo->lo_backing_file = file; | |
ba52de12 TT |
637 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
638 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 LT |
639 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
640 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
641 | complete(&p->wait); | |
642 | } | |
643 | ||
644 | ||
645 | /* | |
646 | * loop_change_fd switched the backing store of a loopback device to | |
647 | * a new file. This is useful for operating system installers to free up | |
648 | * the original file and in High Availability environments to switch to | |
649 | * an alternative location for the content in case of server meltdown. | |
650 | * This can only work if the loop device is used read-only, and if the | |
651 | * new backing store is the same size and type as the old backing store. | |
652 | */ | |
bb214884 AV |
653 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
654 | unsigned int arg) | |
1da177e4 LT |
655 | { |
656 | struct file *file, *old_file; | |
657 | struct inode *inode; | |
658 | int error; | |
659 | ||
660 | error = -ENXIO; | |
661 | if (lo->lo_state != Lo_bound) | |
662 | goto out; | |
663 | ||
664 | /* the loop device has to be read-only */ | |
665 | error = -EINVAL; | |
666 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
667 | goto out; | |
668 | ||
669 | error = -EBADF; | |
670 | file = fget(arg); | |
671 | if (!file) | |
672 | goto out; | |
673 | ||
674 | inode = file->f_mapping->host; | |
675 | old_file = lo->lo_backing_file; | |
676 | ||
677 | error = -EINVAL; | |
678 | ||
679 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
680 | goto out_putf; | |
681 | ||
fd582140 JA |
682 | /* new backing store needs to support loop (eg splice_read) */ |
683 | if (!inode->i_fop->splice_read) | |
1da177e4 LT |
684 | goto out_putf; |
685 | ||
686 | /* size of the new backing store needs to be the same */ | |
687 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
688 | goto out_putf; | |
689 | ||
690 | /* and ... switch */ | |
691 | error = loop_switch(lo, file); | |
692 | if (error) | |
693 | goto out_putf; | |
694 | ||
695 | fput(old_file); | |
476a4813 LV |
696 | if (max_part > 0) |
697 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
698 | return 0; |
699 | ||
700 | out_putf: | |
701 | fput(file); | |
702 | out: | |
703 | return error; | |
704 | } | |
705 | ||
706 | static inline int is_loop_device(struct file *file) | |
707 | { | |
708 | struct inode *i = file->f_mapping->host; | |
709 | ||
710 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
711 | } | |
712 | ||
bb214884 | 713 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
714 | struct block_device *bdev, unsigned int arg) |
715 | { | |
716 | struct file *file, *f; | |
717 | struct inode *inode; | |
718 | struct address_space *mapping; | |
719 | unsigned lo_blocksize; | |
720 | int lo_flags = 0; | |
721 | int error; | |
722 | loff_t size; | |
723 | ||
724 | /* This is safe, since we have a reference from open(). */ | |
725 | __module_get(THIS_MODULE); | |
726 | ||
727 | error = -EBADF; | |
728 | file = fget(arg); | |
729 | if (!file) | |
730 | goto out; | |
731 | ||
732 | error = -EBUSY; | |
733 | if (lo->lo_state != Lo_unbound) | |
734 | goto out_putf; | |
735 | ||
736 | /* Avoid recursion */ | |
737 | f = file; | |
738 | while (is_loop_device(f)) { | |
739 | struct loop_device *l; | |
740 | ||
bb214884 | 741 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
742 | goto out_putf; |
743 | ||
744 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
745 | if (l->lo_state == Lo_unbound) { | |
746 | error = -EINVAL; | |
747 | goto out_putf; | |
748 | } | |
749 | f = l->lo_backing_file; | |
750 | } | |
751 | ||
752 | mapping = file->f_mapping; | |
753 | inode = mapping->host; | |
754 | ||
755 | if (!(file->f_mode & FMODE_WRITE)) | |
756 | lo_flags |= LO_FLAGS_READ_ONLY; | |
757 | ||
758 | error = -EINVAL; | |
759 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
f5e54d6e | 760 | const struct address_space_operations *aops = mapping->a_ops; |
1da177e4 LT |
761 | /* |
762 | * If we can't read - sorry. If we only can't write - well, | |
763 | * it's going to be read-only. | |
764 | */ | |
fd582140 | 765 | if (!file->f_op->splice_read) |
1da177e4 | 766 | goto out_putf; |
4e02ed4b | 767 | if (aops->write_begin) |
1da177e4 LT |
768 | lo_flags |= LO_FLAGS_USE_AOPS; |
769 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) | |
770 | lo_flags |= LO_FLAGS_READ_ONLY; | |
771 | ||
ba52de12 TT |
772 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
773 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
774 | ||
1da177e4 LT |
775 | error = 0; |
776 | } else { | |
777 | goto out_putf; | |
778 | } | |
779 | ||
780 | size = get_loop_size(lo, file); | |
781 | ||
782 | if ((loff_t)(sector_t)size != size) { | |
783 | error = -EFBIG; | |
784 | goto out_putf; | |
785 | } | |
786 | ||
bb214884 | 787 | if (!(mode & FMODE_WRITE)) |
1da177e4 LT |
788 | lo_flags |= LO_FLAGS_READ_ONLY; |
789 | ||
790 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
791 | ||
792 | lo->lo_blocksize = lo_blocksize; | |
793 | lo->lo_device = bdev; | |
794 | lo->lo_flags = lo_flags; | |
795 | lo->lo_backing_file = file; | |
eefe85ee | 796 | lo->transfer = transfer_none; |
1da177e4 LT |
797 | lo->ioctl = NULL; |
798 | lo->lo_sizelimit = 0; | |
799 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
800 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
801 | ||
802 | lo->lo_bio = lo->lo_biotail = NULL; | |
803 | ||
804 | /* | |
805 | * set queue make_request_fn, and add limits based on lower level | |
806 | * device | |
807 | */ | |
808 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
809 | lo->lo_queue->queuedata = lo; | |
810 | lo->lo_queue->unplug_fn = loop_unplug; | |
811 | ||
73285082 | 812 | set_capacity(lo->lo_disk, size); |
1da177e4 LT |
813 | bd_set_size(bdev, size << 9); |
814 | ||
815 | set_blocksize(bdev, lo_blocksize); | |
816 | ||
6c997918 SH |
817 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", |
818 | lo->lo_number); | |
819 | if (IS_ERR(lo->lo_thread)) { | |
820 | error = PTR_ERR(lo->lo_thread); | |
a7422bf8 | 821 | goto out_clr; |
6c997918 SH |
822 | } |
823 | lo->lo_state = Lo_bound; | |
824 | wake_up_process(lo->lo_thread); | |
476a4813 LV |
825 | if (max_part > 0) |
826 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
827 | return 0; |
828 | ||
a7422bf8 SH |
829 | out_clr: |
830 | lo->lo_thread = NULL; | |
831 | lo->lo_device = NULL; | |
832 | lo->lo_backing_file = NULL; | |
833 | lo->lo_flags = 0; | |
73285082 | 834 | set_capacity(lo->lo_disk, 0); |
f98393a6 | 835 | invalidate_bdev(bdev); |
a7422bf8 SH |
836 | bd_set_size(bdev, 0); |
837 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); | |
838 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
839 | out_putf: |
840 | fput(file); | |
841 | out: | |
842 | /* This is safe: open() is still holding a reference. */ | |
843 | module_put(THIS_MODULE); | |
844 | return error; | |
845 | } | |
846 | ||
847 | static int | |
848 | loop_release_xfer(struct loop_device *lo) | |
849 | { | |
850 | int err = 0; | |
851 | struct loop_func_table *xfer = lo->lo_encryption; | |
852 | ||
853 | if (xfer) { | |
854 | if (xfer->release) | |
855 | err = xfer->release(lo); | |
856 | lo->transfer = NULL; | |
857 | lo->lo_encryption = NULL; | |
858 | module_put(xfer->owner); | |
859 | } | |
860 | return err; | |
861 | } | |
862 | ||
863 | static int | |
864 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
865 | const struct loop_info64 *i) | |
866 | { | |
867 | int err = 0; | |
868 | ||
869 | if (xfer) { | |
870 | struct module *owner = xfer->owner; | |
871 | ||
872 | if (!try_module_get(owner)) | |
873 | return -EINVAL; | |
874 | if (xfer->init) | |
875 | err = xfer->init(lo, i); | |
876 | if (err) | |
877 | module_put(owner); | |
878 | else | |
879 | lo->lo_encryption = xfer; | |
880 | } | |
881 | return err; | |
882 | } | |
883 | ||
884 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev) | |
885 | { | |
886 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 887 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 LT |
888 | |
889 | if (lo->lo_state != Lo_bound) | |
890 | return -ENXIO; | |
891 | ||
892 | if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */ | |
893 | return -EBUSY; | |
894 | ||
895 | if (filp == NULL) | |
896 | return -EINVAL; | |
897 | ||
898 | spin_lock_irq(&lo->lo_lock); | |
899 | lo->lo_state = Lo_rundown; | |
1da177e4 LT |
900 | spin_unlock_irq(&lo->lo_lock); |
901 | ||
6c997918 | 902 | kthread_stop(lo->lo_thread); |
1da177e4 LT |
903 | |
904 | lo->lo_backing_file = NULL; | |
905 | ||
906 | loop_release_xfer(lo); | |
907 | lo->transfer = NULL; | |
908 | lo->ioctl = NULL; | |
909 | lo->lo_device = NULL; | |
910 | lo->lo_encryption = NULL; | |
911 | lo->lo_offset = 0; | |
912 | lo->lo_sizelimit = 0; | |
913 | lo->lo_encrypt_key_size = 0; | |
914 | lo->lo_flags = 0; | |
6c997918 | 915 | lo->lo_thread = NULL; |
1da177e4 LT |
916 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
917 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
918 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
bb214884 AV |
919 | if (bdev) |
920 | invalidate_bdev(bdev); | |
73285082 | 921 | set_capacity(lo->lo_disk, 0); |
bb214884 AV |
922 | if (bdev) |
923 | bd_set_size(bdev, 0); | |
1da177e4 LT |
924 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
925 | lo->lo_state = Lo_unbound; | |
926 | fput(filp); | |
927 | /* This is safe: open() is still holding a reference. */ | |
928 | module_put(THIS_MODULE); | |
476a4813 LV |
929 | if (max_part > 0) |
930 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
931 | return 0; |
932 | } | |
933 | ||
934 | static int | |
935 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
936 | { | |
937 | int err; | |
938 | struct loop_func_table *xfer; | |
b0fafa81 | 939 | uid_t uid = current_uid(); |
1da177e4 | 940 | |
b0fafa81 DH |
941 | if (lo->lo_encrypt_key_size && |
942 | lo->lo_key_owner != uid && | |
1da177e4 LT |
943 | !capable(CAP_SYS_ADMIN)) |
944 | return -EPERM; | |
945 | if (lo->lo_state != Lo_bound) | |
946 | return -ENXIO; | |
947 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
948 | return -EINVAL; | |
949 | ||
950 | err = loop_release_xfer(lo); | |
951 | if (err) | |
952 | return err; | |
953 | ||
954 | if (info->lo_encrypt_type) { | |
955 | unsigned int type = info->lo_encrypt_type; | |
956 | ||
957 | if (type >= MAX_LO_CRYPT) | |
958 | return -EINVAL; | |
959 | xfer = xfer_funcs[type]; | |
960 | if (xfer == NULL) | |
961 | return -EINVAL; | |
962 | } else | |
963 | xfer = NULL; | |
964 | ||
965 | err = loop_init_xfer(lo, xfer, info); | |
966 | if (err) | |
967 | return err; | |
968 | ||
969 | if (lo->lo_offset != info->lo_offset || | |
970 | lo->lo_sizelimit != info->lo_sizelimit) { | |
971 | lo->lo_offset = info->lo_offset; | |
972 | lo->lo_sizelimit = info->lo_sizelimit; | |
973 | if (figure_loop_size(lo)) | |
974 | return -EFBIG; | |
975 | } | |
976 | ||
977 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
978 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
979 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
980 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
981 | ||
982 | if (!xfer) | |
983 | xfer = &none_funcs; | |
984 | lo->transfer = xfer->transfer; | |
985 | lo->ioctl = xfer->ioctl; | |
986 | ||
96c58655 DW |
987 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
988 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
989 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
990 | ||
1da177e4 LT |
991 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
992 | lo->lo_init[0] = info->lo_init[0]; | |
993 | lo->lo_init[1] = info->lo_init[1]; | |
994 | if (info->lo_encrypt_key_size) { | |
995 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
996 | info->lo_encrypt_key_size); | |
b0fafa81 | 997 | lo->lo_key_owner = uid; |
1da177e4 LT |
998 | } |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | static int | |
1004 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1005 | { | |
1006 | struct file *file = lo->lo_backing_file; | |
1007 | struct kstat stat; | |
1008 | int error; | |
1009 | ||
1010 | if (lo->lo_state != Lo_bound) | |
1011 | return -ENXIO; | |
6c648be6 | 1012 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); |
1da177e4 LT |
1013 | if (error) |
1014 | return error; | |
1015 | memset(info, 0, sizeof(*info)); | |
1016 | info->lo_number = lo->lo_number; | |
1017 | info->lo_device = huge_encode_dev(stat.dev); | |
1018 | info->lo_inode = stat.ino; | |
1019 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1020 | info->lo_offset = lo->lo_offset; | |
1021 | info->lo_sizelimit = lo->lo_sizelimit; | |
1022 | info->lo_flags = lo->lo_flags; | |
1023 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1024 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1025 | info->lo_encrypt_type = | |
1026 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1027 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1028 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1029 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1030 | lo->lo_encrypt_key_size); | |
1031 | } | |
1032 | return 0; | |
1033 | } | |
1034 | ||
1035 | static void | |
1036 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1037 | { | |
1038 | memset(info64, 0, sizeof(*info64)); | |
1039 | info64->lo_number = info->lo_number; | |
1040 | info64->lo_device = info->lo_device; | |
1041 | info64->lo_inode = info->lo_inode; | |
1042 | info64->lo_rdevice = info->lo_rdevice; | |
1043 | info64->lo_offset = info->lo_offset; | |
1044 | info64->lo_sizelimit = 0; | |
1045 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1046 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1047 | info64->lo_flags = info->lo_flags; | |
1048 | info64->lo_init[0] = info->lo_init[0]; | |
1049 | info64->lo_init[1] = info->lo_init[1]; | |
1050 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1051 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1052 | else | |
1053 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1054 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1055 | } | |
1056 | ||
1057 | static int | |
1058 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1059 | { | |
1060 | memset(info, 0, sizeof(*info)); | |
1061 | info->lo_number = info64->lo_number; | |
1062 | info->lo_device = info64->lo_device; | |
1063 | info->lo_inode = info64->lo_inode; | |
1064 | info->lo_rdevice = info64->lo_rdevice; | |
1065 | info->lo_offset = info64->lo_offset; | |
1066 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1067 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1068 | info->lo_flags = info64->lo_flags; | |
1069 | info->lo_init[0] = info64->lo_init[0]; | |
1070 | info->lo_init[1] = info64->lo_init[1]; | |
1071 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1072 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1073 | else | |
1074 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1075 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1076 | ||
1077 | /* error in case values were truncated */ | |
1078 | if (info->lo_device != info64->lo_device || | |
1079 | info->lo_rdevice != info64->lo_rdevice || | |
1080 | info->lo_inode != info64->lo_inode || | |
1081 | info->lo_offset != info64->lo_offset) | |
1082 | return -EOVERFLOW; | |
1083 | ||
1084 | return 0; | |
1085 | } | |
1086 | ||
1087 | static int | |
1088 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1089 | { | |
1090 | struct loop_info info; | |
1091 | struct loop_info64 info64; | |
1092 | ||
1093 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1094 | return -EFAULT; | |
1095 | loop_info64_from_old(&info, &info64); | |
1096 | return loop_set_status(lo, &info64); | |
1097 | } | |
1098 | ||
1099 | static int | |
1100 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1101 | { | |
1102 | struct loop_info64 info64; | |
1103 | ||
1104 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1105 | return -EFAULT; | |
1106 | return loop_set_status(lo, &info64); | |
1107 | } | |
1108 | ||
1109 | static int | |
1110 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1111 | struct loop_info info; | |
1112 | struct loop_info64 info64; | |
1113 | int err = 0; | |
1114 | ||
1115 | if (!arg) | |
1116 | err = -EINVAL; | |
1117 | if (!err) | |
1118 | err = loop_get_status(lo, &info64); | |
1119 | if (!err) | |
1120 | err = loop_info64_to_old(&info64, &info); | |
1121 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1122 | err = -EFAULT; | |
1123 | ||
1124 | return err; | |
1125 | } | |
1126 | ||
1127 | static int | |
1128 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1129 | struct loop_info64 info64; | |
1130 | int err = 0; | |
1131 | ||
1132 | if (!arg) | |
1133 | err = -EINVAL; | |
1134 | if (!err) | |
1135 | err = loop_get_status(lo, &info64); | |
1136 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1137 | err = -EFAULT; | |
1138 | ||
1139 | return err; | |
1140 | } | |
1141 | ||
bb214884 | 1142 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1143 | unsigned int cmd, unsigned long arg) |
1144 | { | |
bb214884 | 1145 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1146 | int err; |
1147 | ||
f85221dd | 1148 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1149 | switch (cmd) { |
1150 | case LOOP_SET_FD: | |
bb214884 | 1151 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1152 | break; |
1153 | case LOOP_CHANGE_FD: | |
bb214884 | 1154 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1155 | break; |
1156 | case LOOP_CLR_FD: | |
bb214884 | 1157 | err = loop_clr_fd(lo, bdev); |
1da177e4 LT |
1158 | break; |
1159 | case LOOP_SET_STATUS: | |
1160 | err = loop_set_status_old(lo, (struct loop_info __user *) arg); | |
1161 | break; | |
1162 | case LOOP_GET_STATUS: | |
1163 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1164 | break; | |
1165 | case LOOP_SET_STATUS64: | |
1166 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg); | |
1167 | break; | |
1168 | case LOOP_GET_STATUS64: | |
1169 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1170 | break; | |
1171 | default: | |
1172 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1173 | } | |
f85221dd | 1174 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1175 | return err; |
1176 | } | |
1177 | ||
863d5b82 DH |
1178 | #ifdef CONFIG_COMPAT |
1179 | struct compat_loop_info { | |
1180 | compat_int_t lo_number; /* ioctl r/o */ | |
1181 | compat_dev_t lo_device; /* ioctl r/o */ | |
1182 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1183 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1184 | compat_int_t lo_offset; | |
1185 | compat_int_t lo_encrypt_type; | |
1186 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1187 | compat_int_t lo_flags; /* ioctl r/o */ | |
1188 | char lo_name[LO_NAME_SIZE]; | |
1189 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1190 | compat_ulong_t lo_init[2]; | |
1191 | char reserved[4]; | |
1192 | }; | |
1193 | ||
1194 | /* | |
1195 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1196 | * - noinlined to reduce stack space usage in main part of driver | |
1197 | */ | |
1198 | static noinline int | |
ba674cfc | 1199 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1200 | struct loop_info64 *info64) |
1201 | { | |
1202 | struct compat_loop_info info; | |
1203 | ||
1204 | if (copy_from_user(&info, arg, sizeof(info))) | |
1205 | return -EFAULT; | |
1206 | ||
1207 | memset(info64, 0, sizeof(*info64)); | |
1208 | info64->lo_number = info.lo_number; | |
1209 | info64->lo_device = info.lo_device; | |
1210 | info64->lo_inode = info.lo_inode; | |
1211 | info64->lo_rdevice = info.lo_rdevice; | |
1212 | info64->lo_offset = info.lo_offset; | |
1213 | info64->lo_sizelimit = 0; | |
1214 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1215 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1216 | info64->lo_flags = info.lo_flags; | |
1217 | info64->lo_init[0] = info.lo_init[0]; | |
1218 | info64->lo_init[1] = info.lo_init[1]; | |
1219 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1220 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1221 | else | |
1222 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1223 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1229 | * - noinlined to reduce stack space usage in main part of driver | |
1230 | */ | |
1231 | static noinline int | |
1232 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1233 | struct compat_loop_info __user *arg) | |
1234 | { | |
1235 | struct compat_loop_info info; | |
1236 | ||
1237 | memset(&info, 0, sizeof(info)); | |
1238 | info.lo_number = info64->lo_number; | |
1239 | info.lo_device = info64->lo_device; | |
1240 | info.lo_inode = info64->lo_inode; | |
1241 | info.lo_rdevice = info64->lo_rdevice; | |
1242 | info.lo_offset = info64->lo_offset; | |
1243 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1244 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1245 | info.lo_flags = info64->lo_flags; | |
1246 | info.lo_init[0] = info64->lo_init[0]; | |
1247 | info.lo_init[1] = info64->lo_init[1]; | |
1248 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1249 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1250 | else | |
1251 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1252 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1253 | ||
1254 | /* error in case values were truncated */ | |
1255 | if (info.lo_device != info64->lo_device || | |
1256 | info.lo_rdevice != info64->lo_rdevice || | |
1257 | info.lo_inode != info64->lo_inode || | |
1258 | info.lo_offset != info64->lo_offset || | |
1259 | info.lo_init[0] != info64->lo_init[0] || | |
1260 | info.lo_init[1] != info64->lo_init[1]) | |
1261 | return -EOVERFLOW; | |
1262 | ||
1263 | if (copy_to_user(arg, &info, sizeof(info))) | |
1264 | return -EFAULT; | |
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | static int | |
1269 | loop_set_status_compat(struct loop_device *lo, | |
1270 | const struct compat_loop_info __user *arg) | |
1271 | { | |
1272 | struct loop_info64 info64; | |
1273 | int ret; | |
1274 | ||
1275 | ret = loop_info64_from_compat(arg, &info64); | |
1276 | if (ret < 0) | |
1277 | return ret; | |
1278 | return loop_set_status(lo, &info64); | |
1279 | } | |
1280 | ||
1281 | static int | |
1282 | loop_get_status_compat(struct loop_device *lo, | |
1283 | struct compat_loop_info __user *arg) | |
1284 | { | |
1285 | struct loop_info64 info64; | |
1286 | int err = 0; | |
1287 | ||
1288 | if (!arg) | |
1289 | err = -EINVAL; | |
1290 | if (!err) | |
1291 | err = loop_get_status(lo, &info64); | |
1292 | if (!err) | |
1293 | err = loop_info64_to_compat(&info64, arg); | |
1294 | return err; | |
1295 | } | |
1296 | ||
bb214884 AV |
1297 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1298 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1299 | { |
bb214884 | 1300 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1301 | int err; |
1302 | ||
863d5b82 DH |
1303 | switch(cmd) { |
1304 | case LOOP_SET_STATUS: | |
1305 | mutex_lock(&lo->lo_ctl_mutex); | |
1306 | err = loop_set_status_compat( | |
1307 | lo, (const struct compat_loop_info __user *) arg); | |
1308 | mutex_unlock(&lo->lo_ctl_mutex); | |
1309 | break; | |
1310 | case LOOP_GET_STATUS: | |
1311 | mutex_lock(&lo->lo_ctl_mutex); | |
1312 | err = loop_get_status_compat( | |
1313 | lo, (struct compat_loop_info __user *) arg); | |
1314 | mutex_unlock(&lo->lo_ctl_mutex); | |
1315 | break; | |
1316 | case LOOP_CLR_FD: | |
1317 | case LOOP_GET_STATUS64: | |
1318 | case LOOP_SET_STATUS64: | |
1319 | arg = (unsigned long) compat_ptr(arg); | |
1320 | case LOOP_SET_FD: | |
1321 | case LOOP_CHANGE_FD: | |
bb214884 | 1322 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1323 | break; |
1324 | default: | |
1325 | err = -ENOIOCTLCMD; | |
1326 | break; | |
1327 | } | |
863d5b82 DH |
1328 | return err; |
1329 | } | |
1330 | #endif | |
1331 | ||
bb214884 | 1332 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1333 | { |
bb214884 | 1334 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 | 1335 | |
f85221dd | 1336 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1337 | lo->lo_refcnt++; |
f85221dd | 1338 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1339 | |
1340 | return 0; | |
1341 | } | |
1342 | ||
bb214884 | 1343 | static int lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1344 | { |
bb214884 | 1345 | struct loop_device *lo = disk->private_data; |
1da177e4 | 1346 | |
f85221dd | 1347 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1348 | --lo->lo_refcnt; |
96c58655 DW |
1349 | |
1350 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt) | |
bb214884 | 1351 | loop_clr_fd(lo, NULL); |
96c58655 | 1352 | |
f85221dd | 1353 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1354 | |
1355 | return 0; | |
1356 | } | |
1357 | ||
1358 | static struct block_device_operations lo_fops = { | |
1359 | .owner = THIS_MODULE, | |
bb214884 AV |
1360 | .open = lo_open, |
1361 | .release = lo_release, | |
1362 | .ioctl = lo_ioctl, | |
863d5b82 | 1363 | #ifdef CONFIG_COMPAT |
bb214884 | 1364 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1365 | #endif |
1da177e4 LT |
1366 | }; |
1367 | ||
1368 | /* | |
1369 | * And now the modules code and kernel interface. | |
1370 | */ | |
73285082 | 1371 | static int max_loop; |
1da177e4 | 1372 | module_param(max_loop, int, 0); |
a47653fc | 1373 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
476a4813 LV |
1374 | module_param(max_part, int, 0); |
1375 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); | |
1da177e4 LT |
1376 | MODULE_LICENSE("GPL"); |
1377 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1378 | ||
1379 | int loop_register_transfer(struct loop_func_table *funcs) | |
1380 | { | |
1381 | unsigned int n = funcs->number; | |
1382 | ||
1383 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1384 | return -EINVAL; | |
1385 | xfer_funcs[n] = funcs; | |
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | int loop_unregister_transfer(int number) | |
1390 | { | |
1391 | unsigned int n = number; | |
1392 | struct loop_device *lo; | |
1393 | struct loop_func_table *xfer; | |
1394 | ||
1395 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1396 | return -EINVAL; | |
1397 | ||
1398 | xfer_funcs[n] = NULL; | |
1399 | ||
73285082 | 1400 | list_for_each_entry(lo, &loop_devices, lo_list) { |
f85221dd | 1401 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1402 | |
1403 | if (lo->lo_encryption == xfer) | |
1404 | loop_release_xfer(lo); | |
1405 | ||
f85221dd | 1406 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | EXPORT_SYMBOL(loop_register_transfer); | |
1413 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1414 | ||
a47653fc | 1415 | static struct loop_device *loop_alloc(int i) |
73285082 KC |
1416 | { |
1417 | struct loop_device *lo; | |
1418 | struct gendisk *disk; | |
1419 | ||
1420 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); | |
1421 | if (!lo) | |
1422 | goto out; | |
1423 | ||
1424 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); | |
1425 | if (!lo->lo_queue) | |
1426 | goto out_free_dev; | |
1427 | ||
476a4813 | 1428 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1429 | if (!disk) |
1430 | goto out_free_queue; | |
1431 | ||
1432 | mutex_init(&lo->lo_ctl_mutex); | |
1433 | lo->lo_number = i; | |
1434 | lo->lo_thread = NULL; | |
1435 | init_waitqueue_head(&lo->lo_event); | |
1436 | spin_lock_init(&lo->lo_lock); | |
1437 | disk->major = LOOP_MAJOR; | |
476a4813 | 1438 | disk->first_minor = i << part_shift; |
73285082 KC |
1439 | disk->fops = &lo_fops; |
1440 | disk->private_data = lo; | |
1441 | disk->queue = lo->lo_queue; | |
1442 | sprintf(disk->disk_name, "loop%d", i); | |
73285082 KC |
1443 | return lo; |
1444 | ||
1445 | out_free_queue: | |
1446 | blk_cleanup_queue(lo->lo_queue); | |
1447 | out_free_dev: | |
1448 | kfree(lo); | |
1449 | out: | |
07002e99 | 1450 | return NULL; |
73285082 KC |
1451 | } |
1452 | ||
a47653fc | 1453 | static void loop_free(struct loop_device *lo) |
1da177e4 | 1454 | { |
73285082 KC |
1455 | blk_cleanup_queue(lo->lo_queue); |
1456 | put_disk(lo->lo_disk); | |
1457 | list_del(&lo->lo_list); | |
1458 | kfree(lo); | |
1459 | } | |
1da177e4 | 1460 | |
a47653fc KC |
1461 | static struct loop_device *loop_init_one(int i) |
1462 | { | |
1463 | struct loop_device *lo; | |
1464 | ||
1465 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1466 | if (lo->lo_number == i) | |
1467 | return lo; | |
1468 | } | |
1469 | ||
1470 | lo = loop_alloc(i); | |
1471 | if (lo) { | |
1472 | add_disk(lo->lo_disk); | |
1473 | list_add_tail(&lo->lo_list, &loop_devices); | |
1474 | } | |
1475 | return lo; | |
1476 | } | |
1477 | ||
1478 | static void loop_del_one(struct loop_device *lo) | |
1479 | { | |
1480 | del_gendisk(lo->lo_disk); | |
1481 | loop_free(lo); | |
1482 | } | |
1483 | ||
73285082 KC |
1484 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1485 | { | |
705962cc | 1486 | struct loop_device *lo; |
07002e99 | 1487 | struct kobject *kobj; |
73285082 | 1488 | |
705962cc AV |
1489 | mutex_lock(&loop_devices_mutex); |
1490 | lo = loop_init_one(dev & MINORMASK); | |
07002e99 | 1491 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM); |
73285082 KC |
1492 | mutex_unlock(&loop_devices_mutex); |
1493 | ||
1494 | *part = 0; | |
07002e99 | 1495 | return kobj; |
73285082 KC |
1496 | } |
1497 | ||
1498 | static int __init loop_init(void) | |
1499 | { | |
a47653fc KC |
1500 | int i, nr; |
1501 | unsigned long range; | |
1502 | struct loop_device *lo, *next; | |
1503 | ||
1504 | /* | |
1505 | * loop module now has a feature to instantiate underlying device | |
1506 | * structure on-demand, provided that there is an access dev node. | |
1507 | * However, this will not work well with user space tool that doesn't | |
1508 | * know about such "feature". In order to not break any existing | |
1509 | * tool, we do the following: | |
1510 | * | |
1511 | * (1) if max_loop is specified, create that many upfront, and this | |
1512 | * also becomes a hard limit. | |
1513 | * (2) if max_loop is not specified, create 8 loop device on module | |
1514 | * load, user can further extend loop device by create dev node | |
1515 | * themselves and have kernel automatically instantiate actual | |
1516 | * device on-demand. | |
1517 | */ | |
476a4813 LV |
1518 | |
1519 | part_shift = 0; | |
1520 | if (max_part > 0) | |
1521 | part_shift = fls(max_part); | |
1522 | ||
1523 | if (max_loop > 1UL << (MINORBITS - part_shift)) | |
a47653fc | 1524 | return -EINVAL; |
1da177e4 | 1525 | |
73285082 | 1526 | if (max_loop) { |
a47653fc KC |
1527 | nr = max_loop; |
1528 | range = max_loop; | |
1529 | } else { | |
1530 | nr = 8; | |
476a4813 | 1531 | range = 1UL << (MINORBITS - part_shift); |
a47653fc KC |
1532 | } |
1533 | ||
1534 | if (register_blkdev(LOOP_MAJOR, "loop")) | |
1535 | return -EIO; | |
1da177e4 | 1536 | |
a47653fc KC |
1537 | for (i = 0; i < nr; i++) { |
1538 | lo = loop_alloc(i); | |
1539 | if (!lo) | |
1540 | goto Enomem; | |
1541 | list_add_tail(&lo->lo_list, &loop_devices); | |
1da177e4 | 1542 | } |
a47653fc KC |
1543 | |
1544 | /* point of no return */ | |
1545 | ||
1546 | list_for_each_entry(lo, &loop_devices, lo_list) | |
1547 | add_disk(lo->lo_disk); | |
1548 | ||
1549 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, | |
1550 | THIS_MODULE, loop_probe, NULL, NULL); | |
1551 | ||
73285082 | 1552 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 1553 | return 0; |
a47653fc KC |
1554 | |
1555 | Enomem: | |
1556 | printk(KERN_INFO "loop: out of memory\n"); | |
1557 | ||
1558 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
1559 | loop_free(lo); | |
1560 | ||
1561 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
1562 | return -ENOMEM; | |
1da177e4 LT |
1563 | } |
1564 | ||
73285082 | 1565 | static void __exit loop_exit(void) |
1da177e4 | 1566 | { |
a47653fc | 1567 | unsigned long range; |
73285082 | 1568 | struct loop_device *lo, *next; |
1da177e4 | 1569 | |
476a4813 | 1570 | range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift); |
a47653fc | 1571 | |
73285082 KC |
1572 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) |
1573 | loop_del_one(lo); | |
1574 | ||
a47653fc | 1575 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 1576 | unregister_blkdev(LOOP_MAJOR, "loop"); |
1da177e4 LT |
1577 | } |
1578 | ||
1579 | module_init(loop_init); | |
1580 | module_exit(loop_exit); | |
1581 | ||
1582 | #ifndef MODULE | |
1583 | static int __init max_loop_setup(char *str) | |
1584 | { | |
1585 | max_loop = simple_strtol(str, NULL, 0); | |
1586 | return 1; | |
1587 | } | |
1588 | ||
1589 | __setup("max_loop=", max_loop_setup); | |
1590 | #endif |