]>
Commit | Line | Data |
---|---|---|
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. | |
32 | * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 | |
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 | |
43 | * operations write_begin is not available on the backing filesystem. | |
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 | ||
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> | |
64 | #include <linux/swap.h> | |
65 | #include <linux/slab.h> | |
66 | #include <linux/loop.h> | |
67 | #include <linux/compat.h> | |
68 | #include <linux/suspend.h> | |
69 | #include <linux/freezer.h> | |
70 | #include <linux/mutex.h> | |
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/kthread.h> | |
76 | #include <linux/splice.h> | |
77 | ||
78 | #include <asm/uaccess.h> | |
79 | ||
80 | static DEFINE_MUTEX(loop_mutex); | |
81 | static LIST_HEAD(loop_devices); | |
82 | static DEFINE_MUTEX(loop_devices_mutex); | |
83 | ||
84 | static int max_part; | |
85 | static int part_shift; | |
86 | ||
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 | ||
189 | set_capacity(lo->lo_disk, x); | |
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 | |
209 | * space operations write_begin and write_end. | |
210 | */ | |
211 | static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec, | |
212 | loff_t pos, struct page *unused) | |
213 | { | |
214 | struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */ | |
215 | struct address_space *mapping = file->f_mapping; | |
216 | pgoff_t index; | |
217 | unsigned offset, bv_offs; | |
218 | int len, ret; | |
219 | ||
220 | mutex_lock(&mapping->host->i_mutex); | |
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; | |
227 | unsigned size, copied; | |
228 | int transfer_result; | |
229 | struct page *page; | |
230 | void *fsdata; | |
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; | |
236 | ||
237 | ret = pagecache_write_begin(file, mapping, pos, size, 0, | |
238 | &page, &fsdata); | |
239 | if (ret) | |
240 | goto fail; | |
241 | ||
242 | file_update_time(file); | |
243 | ||
244 | transfer_result = lo_do_transfer(lo, WRITE, page, offset, | |
245 | bvec->bv_page, bv_offs, size, IV); | |
246 | copied = size; | |
247 | if (unlikely(transfer_result)) | |
248 | copied = 0; | |
249 | ||
250 | ret = pagecache_write_end(file, mapping, pos, size, copied, | |
251 | page, fsdata); | |
252 | if (ret < 0 || ret != copied) | |
253 | goto fail; | |
254 | ||
255 | if (unlikely(transfer_result)) | |
256 | goto fail; | |
257 | ||
258 | bv_offs += copied; | |
259 | len -= copied; | |
260 | offset = 0; | |
261 | index++; | |
262 | pos += copied; | |
263 | } | |
264 | ret = 0; | |
265 | out: | |
266 | mutex_unlock(&mapping->host->i_mutex); | |
267 | return ret; | |
268 | fail: | |
269 | ret = -1; | |
270 | goto out; | |
271 | } | |
272 | ||
273 | /** | |
274 | * __do_lo_send_write - helper for writing data to a loop device | |
275 | * | |
276 | * This helper just factors out common code between do_lo_send_direct_write() | |
277 | * and do_lo_send_write(). | |
278 | */ | |
279 | static int __do_lo_send_write(struct file *file, | |
280 | u8 *buf, const int len, loff_t pos) | |
281 | { | |
282 | ssize_t bw; | |
283 | mm_segment_t old_fs = get_fs(); | |
284 | ||
285 | set_fs(get_ds()); | |
286 | bw = file->f_op->write(file, buf, len, &pos); | |
287 | set_fs(old_fs); | |
288 | if (likely(bw == len)) | |
289 | return 0; | |
290 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", | |
291 | (unsigned long long)pos, len); | |
292 | if (bw >= 0) | |
293 | bw = -EIO; | |
294 | return bw; | |
295 | } | |
296 | ||
297 | /** | |
298 | * do_lo_send_direct_write - helper for writing data to a loop device | |
299 | * | |
300 | * This is the fast, non-transforming version for backing filesystems which do | |
301 | * not implement the address space operations write_begin and write_end. | |
302 | * It uses the write file operation which should be present on all writeable | |
303 | * filesystems. | |
304 | */ | |
305 | static int do_lo_send_direct_write(struct loop_device *lo, | |
306 | struct bio_vec *bvec, loff_t pos, struct page *page) | |
307 | { | |
308 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file, | |
309 | kmap(bvec->bv_page) + bvec->bv_offset, | |
310 | bvec->bv_len, pos); | |
311 | kunmap(bvec->bv_page); | |
312 | cond_resched(); | |
313 | return bw; | |
314 | } | |
315 | ||
316 | /** | |
317 | * do_lo_send_write - helper for writing data to a loop device | |
318 | * | |
319 | * This is the slow, transforming version for filesystems which do not | |
320 | * implement the address space operations write_begin and write_end. It | |
321 | * uses the write file operation which should be present on all writeable | |
322 | * filesystems. | |
323 | * | |
324 | * Using fops->write is slower than using aops->{prepare,commit}_write in the | |
325 | * transforming case because we need to double buffer the data as we cannot do | |
326 | * the transformations in place as we do not have direct access to the | |
327 | * destination pages of the backing file. | |
328 | */ | |
329 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, | |
330 | loff_t pos, struct page *page) | |
331 | { | |
332 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, | |
333 | bvec->bv_offset, bvec->bv_len, pos >> 9); | |
334 | if (likely(!ret)) | |
335 | return __do_lo_send_write(lo->lo_backing_file, | |
336 | page_address(page), bvec->bv_len, | |
337 | pos); | |
338 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, " | |
339 | "length %i.\n", (unsigned long long)pos, bvec->bv_len); | |
340 | if (ret > 0) | |
341 | ret = -EIO; | |
342 | return ret; | |
343 | } | |
344 | ||
345 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) | |
346 | { | |
347 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, | |
348 | struct page *page); | |
349 | struct bio_vec *bvec; | |
350 | struct page *page = NULL; | |
351 | int i, ret = 0; | |
352 | ||
353 | do_lo_send = do_lo_send_aops; | |
354 | if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) { | |
355 | do_lo_send = do_lo_send_direct_write; | |
356 | if (lo->transfer != transfer_none) { | |
357 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
358 | if (unlikely(!page)) | |
359 | goto fail; | |
360 | kmap(page); | |
361 | do_lo_send = do_lo_send_write; | |
362 | } | |
363 | } | |
364 | bio_for_each_segment(bvec, bio, i) { | |
365 | ret = do_lo_send(lo, bvec, pos, page); | |
366 | if (ret < 0) | |
367 | break; | |
368 | pos += bvec->bv_len; | |
369 | } | |
370 | if (page) { | |
371 | kunmap(page); | |
372 | __free_page(page); | |
373 | } | |
374 | out: | |
375 | return ret; | |
376 | fail: | |
377 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); | |
378 | ret = -ENOMEM; | |
379 | goto out; | |
380 | } | |
381 | ||
382 | struct lo_read_data { | |
383 | struct loop_device *lo; | |
384 | struct page *page; | |
385 | unsigned offset; | |
386 | int bsize; | |
387 | }; | |
388 | ||
389 | static int | |
390 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, | |
391 | struct splice_desc *sd) | |
392 | { | |
393 | struct lo_read_data *p = sd->u.data; | |
394 | struct loop_device *lo = p->lo; | |
395 | struct page *page = buf->page; | |
396 | sector_t IV; | |
397 | int size, ret; | |
398 | ||
399 | ret = buf->ops->confirm(pipe, buf); | |
400 | if (unlikely(ret)) | |
401 | return ret; | |
402 | ||
403 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + | |
404 | (buf->offset >> 9); | |
405 | size = sd->len; | |
406 | if (size > p->bsize) | |
407 | size = p->bsize; | |
408 | ||
409 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { | |
410 | printk(KERN_ERR "loop: transfer error block %ld\n", | |
411 | page->index); | |
412 | size = -EINVAL; | |
413 | } | |
414 | ||
415 | flush_dcache_page(p->page); | |
416 | ||
417 | if (size > 0) | |
418 | p->offset += size; | |
419 | ||
420 | return size; | |
421 | } | |
422 | ||
423 | static int | |
424 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
425 | { | |
426 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
427 | } | |
428 | ||
429 | static int | |
430 | do_lo_receive(struct loop_device *lo, | |
431 | struct bio_vec *bvec, int bsize, loff_t pos) | |
432 | { | |
433 | struct lo_read_data cookie; | |
434 | struct splice_desc sd; | |
435 | struct file *file; | |
436 | long retval; | |
437 | ||
438 | cookie.lo = lo; | |
439 | cookie.page = bvec->bv_page; | |
440 | cookie.offset = bvec->bv_offset; | |
441 | cookie.bsize = bsize; | |
442 | ||
443 | sd.len = 0; | |
444 | sd.total_len = bvec->bv_len; | |
445 | sd.flags = 0; | |
446 | sd.pos = pos; | |
447 | sd.u.data = &cookie; | |
448 | ||
449 | file = lo->lo_backing_file; | |
450 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); | |
451 | ||
452 | if (retval < 0) | |
453 | return retval; | |
454 | ||
455 | return 0; | |
456 | } | |
457 | ||
458 | static int | |
459 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
460 | { | |
461 | struct bio_vec *bvec; | |
462 | int i, ret = 0; | |
463 | ||
464 | bio_for_each_segment(bvec, bio, i) { | |
465 | ret = do_lo_receive(lo, bvec, bsize, pos); | |
466 | if (ret < 0) | |
467 | break; | |
468 | pos += bvec->bv_len; | |
469 | } | |
470 | return ret; | |
471 | } | |
472 | ||
473 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
474 | { | |
475 | loff_t pos; | |
476 | int ret; | |
477 | ||
478 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
479 | ||
480 | if (bio_rw(bio) == WRITE) { | |
481 | bool barrier = !!(bio->bi_rw & REQ_HARDBARRIER); | |
482 | struct file *file = lo->lo_backing_file; | |
483 | ||
484 | if (barrier) { | |
485 | if (unlikely(!file->f_op->fsync)) { | |
486 | ret = -EOPNOTSUPP; | |
487 | goto out; | |
488 | } | |
489 | ||
490 | ret = vfs_fsync(file, 0); | |
491 | if (unlikely(ret)) { | |
492 | ret = -EIO; | |
493 | goto out; | |
494 | } | |
495 | } | |
496 | ||
497 | ret = lo_send(lo, bio, pos); | |
498 | ||
499 | if (barrier && !ret) { | |
500 | ret = vfs_fsync(file, 0); | |
501 | if (unlikely(ret)) | |
502 | ret = -EIO; | |
503 | } | |
504 | } else | |
505 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); | |
506 | ||
507 | out: | |
508 | return ret; | |
509 | } | |
510 | ||
511 | /* | |
512 | * Add bio to back of pending list | |
513 | */ | |
514 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
515 | { | |
516 | bio_list_add(&lo->lo_bio_list, bio); | |
517 | } | |
518 | ||
519 | /* | |
520 | * Grab first pending buffer | |
521 | */ | |
522 | static struct bio *loop_get_bio(struct loop_device *lo) | |
523 | { | |
524 | return bio_list_pop(&lo->lo_bio_list); | |
525 | } | |
526 | ||
527 | static int loop_make_request(struct request_queue *q, struct bio *old_bio) | |
528 | { | |
529 | struct loop_device *lo = q->queuedata; | |
530 | int rw = bio_rw(old_bio); | |
531 | ||
532 | if (rw == READA) | |
533 | rw = READ; | |
534 | ||
535 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
536 | ||
537 | spin_lock_irq(&lo->lo_lock); | |
538 | if (lo->lo_state != Lo_bound) | |
539 | goto out; | |
540 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
541 | goto out; | |
542 | loop_add_bio(lo, old_bio); | |
543 | wake_up(&lo->lo_event); | |
544 | spin_unlock_irq(&lo->lo_lock); | |
545 | return 0; | |
546 | ||
547 | out: | |
548 | spin_unlock_irq(&lo->lo_lock); | |
549 | bio_io_error(old_bio); | |
550 | return 0; | |
551 | } | |
552 | ||
553 | /* | |
554 | * kick off io on the underlying address space | |
555 | */ | |
556 | static void loop_unplug(struct request_queue *q) | |
557 | { | |
558 | struct loop_device *lo = q->queuedata; | |
559 | ||
560 | queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q); | |
561 | blk_run_address_space(lo->lo_backing_file->f_mapping); | |
562 | } | |
563 | ||
564 | struct switch_request { | |
565 | struct file *file; | |
566 | struct completion wait; | |
567 | }; | |
568 | ||
569 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
570 | ||
571 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
572 | { | |
573 | if (unlikely(!bio->bi_bdev)) { | |
574 | do_loop_switch(lo, bio->bi_private); | |
575 | bio_put(bio); | |
576 | } else { | |
577 | int ret = do_bio_filebacked(lo, bio); | |
578 | bio_endio(bio, ret); | |
579 | } | |
580 | } | |
581 | ||
582 | /* | |
583 | * worker thread that handles reads/writes to file backed loop devices, | |
584 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
585 | * on reads for block backed loop, as that is too heavy to do from | |
586 | * b_end_io context where irqs may be disabled. | |
587 | * | |
588 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
589 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
590 | * true, make_request will not place any more requests. Therefore | |
591 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
592 | * done with the loop. | |
593 | */ | |
594 | static int loop_thread(void *data) | |
595 | { | |
596 | struct loop_device *lo = data; | |
597 | struct bio *bio; | |
598 | ||
599 | set_user_nice(current, -20); | |
600 | ||
601 | while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) { | |
602 | ||
603 | wait_event_interruptible(lo->lo_event, | |
604 | !bio_list_empty(&lo->lo_bio_list) || | |
605 | kthread_should_stop()); | |
606 | ||
607 | if (bio_list_empty(&lo->lo_bio_list)) | |
608 | continue; | |
609 | spin_lock_irq(&lo->lo_lock); | |
610 | bio = loop_get_bio(lo); | |
611 | spin_unlock_irq(&lo->lo_lock); | |
612 | ||
613 | BUG_ON(!bio); | |
614 | loop_handle_bio(lo, bio); | |
615 | } | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
620 | /* | |
621 | * loop_switch performs the hard work of switching a backing store. | |
622 | * First it needs to flush existing IO, it does this by sending a magic | |
623 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
624 | */ | |
625 | static int loop_switch(struct loop_device *lo, struct file *file) | |
626 | { | |
627 | struct switch_request w; | |
628 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); | |
629 | if (!bio) | |
630 | return -ENOMEM; | |
631 | init_completion(&w.wait); | |
632 | w.file = file; | |
633 | bio->bi_private = &w; | |
634 | bio->bi_bdev = NULL; | |
635 | loop_make_request(lo->lo_queue, bio); | |
636 | wait_for_completion(&w.wait); | |
637 | return 0; | |
638 | } | |
639 | ||
640 | /* | |
641 | * Helper to flush the IOs in loop, but keeping loop thread running | |
642 | */ | |
643 | static int loop_flush(struct loop_device *lo) | |
644 | { | |
645 | /* loop not yet configured, no running thread, nothing to flush */ | |
646 | if (!lo->lo_thread) | |
647 | return 0; | |
648 | ||
649 | return loop_switch(lo, NULL); | |
650 | } | |
651 | ||
652 | /* | |
653 | * Do the actual switch; called from the BIO completion routine | |
654 | */ | |
655 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
656 | { | |
657 | struct file *file = p->file; | |
658 | struct file *old_file = lo->lo_backing_file; | |
659 | struct address_space *mapping; | |
660 | ||
661 | /* if no new file, only flush of queued bios requested */ | |
662 | if (!file) | |
663 | goto out; | |
664 | ||
665 | mapping = file->f_mapping; | |
666 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
667 | lo->lo_backing_file = file; | |
668 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? | |
669 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
670 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
671 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
672 | out: | |
673 | complete(&p->wait); | |
674 | } | |
675 | ||
676 | ||
677 | /* | |
678 | * loop_change_fd switched the backing store of a loopback device to | |
679 | * a new file. This is useful for operating system installers to free up | |
680 | * the original file and in High Availability environments to switch to | |
681 | * an alternative location for the content in case of server meltdown. | |
682 | * This can only work if the loop device is used read-only, and if the | |
683 | * new backing store is the same size and type as the old backing store. | |
684 | */ | |
685 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, | |
686 | unsigned int arg) | |
687 | { | |
688 | struct file *file, *old_file; | |
689 | struct inode *inode; | |
690 | int error; | |
691 | ||
692 | error = -ENXIO; | |
693 | if (lo->lo_state != Lo_bound) | |
694 | goto out; | |
695 | ||
696 | /* the loop device has to be read-only */ | |
697 | error = -EINVAL; | |
698 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
699 | goto out; | |
700 | ||
701 | error = -EBADF; | |
702 | file = fget(arg); | |
703 | if (!file) | |
704 | goto out; | |
705 | ||
706 | inode = file->f_mapping->host; | |
707 | old_file = lo->lo_backing_file; | |
708 | ||
709 | error = -EINVAL; | |
710 | ||
711 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
712 | goto out_putf; | |
713 | ||
714 | /* size of the new backing store needs to be the same */ | |
715 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
716 | goto out_putf; | |
717 | ||
718 | /* and ... switch */ | |
719 | error = loop_switch(lo, file); | |
720 | if (error) | |
721 | goto out_putf; | |
722 | ||
723 | fput(old_file); | |
724 | if (max_part > 0) | |
725 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
726 | return 0; | |
727 | ||
728 | out_putf: | |
729 | fput(file); | |
730 | out: | |
731 | return error; | |
732 | } | |
733 | ||
734 | static inline int is_loop_device(struct file *file) | |
735 | { | |
736 | struct inode *i = file->f_mapping->host; | |
737 | ||
738 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
739 | } | |
740 | ||
741 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, | |
742 | struct block_device *bdev, unsigned int arg) | |
743 | { | |
744 | struct file *file, *f; | |
745 | struct inode *inode; | |
746 | struct address_space *mapping; | |
747 | unsigned lo_blocksize; | |
748 | int lo_flags = 0; | |
749 | int error; | |
750 | loff_t size; | |
751 | ||
752 | /* This is safe, since we have a reference from open(). */ | |
753 | __module_get(THIS_MODULE); | |
754 | ||
755 | error = -EBADF; | |
756 | file = fget(arg); | |
757 | if (!file) | |
758 | goto out; | |
759 | ||
760 | error = -EBUSY; | |
761 | if (lo->lo_state != Lo_unbound) | |
762 | goto out_putf; | |
763 | ||
764 | /* Avoid recursion */ | |
765 | f = file; | |
766 | while (is_loop_device(f)) { | |
767 | struct loop_device *l; | |
768 | ||
769 | if (f->f_mapping->host->i_bdev == bdev) | |
770 | goto out_putf; | |
771 | ||
772 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
773 | if (l->lo_state == Lo_unbound) { | |
774 | error = -EINVAL; | |
775 | goto out_putf; | |
776 | } | |
777 | f = l->lo_backing_file; | |
778 | } | |
779 | ||
780 | mapping = file->f_mapping; | |
781 | inode = mapping->host; | |
782 | ||
783 | if (!(file->f_mode & FMODE_WRITE)) | |
784 | lo_flags |= LO_FLAGS_READ_ONLY; | |
785 | ||
786 | error = -EINVAL; | |
787 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
788 | const struct address_space_operations *aops = mapping->a_ops; | |
789 | ||
790 | if (aops->write_begin) | |
791 | lo_flags |= LO_FLAGS_USE_AOPS; | |
792 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) | |
793 | lo_flags |= LO_FLAGS_READ_ONLY; | |
794 | ||
795 | lo_blocksize = S_ISBLK(inode->i_mode) ? | |
796 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
797 | ||
798 | error = 0; | |
799 | } else { | |
800 | goto out_putf; | |
801 | } | |
802 | ||
803 | size = get_loop_size(lo, file); | |
804 | ||
805 | if ((loff_t)(sector_t)size != size) { | |
806 | error = -EFBIG; | |
807 | goto out_putf; | |
808 | } | |
809 | ||
810 | if (!(mode & FMODE_WRITE)) | |
811 | lo_flags |= LO_FLAGS_READ_ONLY; | |
812 | ||
813 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
814 | ||
815 | lo->lo_blocksize = lo_blocksize; | |
816 | lo->lo_device = bdev; | |
817 | lo->lo_flags = lo_flags; | |
818 | lo->lo_backing_file = file; | |
819 | lo->transfer = transfer_none; | |
820 | lo->ioctl = NULL; | |
821 | lo->lo_sizelimit = 0; | |
822 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
823 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
824 | ||
825 | bio_list_init(&lo->lo_bio_list); | |
826 | ||
827 | /* | |
828 | * set queue make_request_fn, and add limits based on lower level | |
829 | * device | |
830 | */ | |
831 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
832 | lo->lo_queue->queuedata = lo; | |
833 | lo->lo_queue->unplug_fn = loop_unplug; | |
834 | ||
835 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) | |
836 | blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN); | |
837 | ||
838 | set_capacity(lo->lo_disk, size); | |
839 | bd_set_size(bdev, size << 9); | |
840 | /* let user-space know about the new size */ | |
841 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
842 | ||
843 | set_blocksize(bdev, lo_blocksize); | |
844 | ||
845 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", | |
846 | lo->lo_number); | |
847 | if (IS_ERR(lo->lo_thread)) { | |
848 | error = PTR_ERR(lo->lo_thread); | |
849 | goto out_clr; | |
850 | } | |
851 | lo->lo_state = Lo_bound; | |
852 | wake_up_process(lo->lo_thread); | |
853 | if (max_part > 0) | |
854 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
855 | return 0; | |
856 | ||
857 | out_clr: | |
858 | lo->lo_thread = NULL; | |
859 | lo->lo_device = NULL; | |
860 | lo->lo_backing_file = NULL; | |
861 | lo->lo_flags = 0; | |
862 | set_capacity(lo->lo_disk, 0); | |
863 | invalidate_bdev(bdev); | |
864 | bd_set_size(bdev, 0); | |
865 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
866 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); | |
867 | lo->lo_state = Lo_unbound; | |
868 | out_putf: | |
869 | fput(file); | |
870 | out: | |
871 | /* This is safe: open() is still holding a reference. */ | |
872 | module_put(THIS_MODULE); | |
873 | return error; | |
874 | } | |
875 | ||
876 | static int | |
877 | loop_release_xfer(struct loop_device *lo) | |
878 | { | |
879 | int err = 0; | |
880 | struct loop_func_table *xfer = lo->lo_encryption; | |
881 | ||
882 | if (xfer) { | |
883 | if (xfer->release) | |
884 | err = xfer->release(lo); | |
885 | lo->transfer = NULL; | |
886 | lo->lo_encryption = NULL; | |
887 | module_put(xfer->owner); | |
888 | } | |
889 | return err; | |
890 | } | |
891 | ||
892 | static int | |
893 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
894 | const struct loop_info64 *i) | |
895 | { | |
896 | int err = 0; | |
897 | ||
898 | if (xfer) { | |
899 | struct module *owner = xfer->owner; | |
900 | ||
901 | if (!try_module_get(owner)) | |
902 | return -EINVAL; | |
903 | if (xfer->init) | |
904 | err = xfer->init(lo, i); | |
905 | if (err) | |
906 | module_put(owner); | |
907 | else | |
908 | lo->lo_encryption = xfer; | |
909 | } | |
910 | return err; | |
911 | } | |
912 | ||
913 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev) | |
914 | { | |
915 | struct file *filp = lo->lo_backing_file; | |
916 | gfp_t gfp = lo->old_gfp_mask; | |
917 | ||
918 | if (lo->lo_state != Lo_bound) | |
919 | return -ENXIO; | |
920 | ||
921 | if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */ | |
922 | return -EBUSY; | |
923 | ||
924 | if (filp == NULL) | |
925 | return -EINVAL; | |
926 | ||
927 | spin_lock_irq(&lo->lo_lock); | |
928 | lo->lo_state = Lo_rundown; | |
929 | spin_unlock_irq(&lo->lo_lock); | |
930 | ||
931 | kthread_stop(lo->lo_thread); | |
932 | ||
933 | lo->lo_queue->unplug_fn = NULL; | |
934 | lo->lo_backing_file = NULL; | |
935 | ||
936 | loop_release_xfer(lo); | |
937 | lo->transfer = NULL; | |
938 | lo->ioctl = NULL; | |
939 | lo->lo_device = NULL; | |
940 | lo->lo_encryption = NULL; | |
941 | lo->lo_offset = 0; | |
942 | lo->lo_sizelimit = 0; | |
943 | lo->lo_encrypt_key_size = 0; | |
944 | lo->lo_flags = 0; | |
945 | lo->lo_thread = NULL; | |
946 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); | |
947 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
948 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
949 | if (bdev) | |
950 | invalidate_bdev(bdev); | |
951 | set_capacity(lo->lo_disk, 0); | |
952 | if (bdev) { | |
953 | bd_set_size(bdev, 0); | |
954 | /* let user-space know about this change */ | |
955 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
956 | } | |
957 | mapping_set_gfp_mask(filp->f_mapping, gfp); | |
958 | lo->lo_state = Lo_unbound; | |
959 | /* This is safe: open() is still holding a reference. */ | |
960 | module_put(THIS_MODULE); | |
961 | if (max_part > 0 && bdev) | |
962 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
963 | mutex_unlock(&lo->lo_ctl_mutex); | |
964 | /* | |
965 | * Need not hold lo_ctl_mutex to fput backing file. | |
966 | * Calling fput holding lo_ctl_mutex triggers a circular | |
967 | * lock dependency possibility warning as fput can take | |
968 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
969 | */ | |
970 | fput(filp); | |
971 | return 0; | |
972 | } | |
973 | ||
974 | static int | |
975 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
976 | { | |
977 | int err; | |
978 | struct loop_func_table *xfer; | |
979 | uid_t uid = current_uid(); | |
980 | ||
981 | if (lo->lo_encrypt_key_size && | |
982 | lo->lo_key_owner != uid && | |
983 | !capable(CAP_SYS_ADMIN)) | |
984 | return -EPERM; | |
985 | if (lo->lo_state != Lo_bound) | |
986 | return -ENXIO; | |
987 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
988 | return -EINVAL; | |
989 | ||
990 | err = loop_release_xfer(lo); | |
991 | if (err) | |
992 | return err; | |
993 | ||
994 | if (info->lo_encrypt_type) { | |
995 | unsigned int type = info->lo_encrypt_type; | |
996 | ||
997 | if (type >= MAX_LO_CRYPT) | |
998 | return -EINVAL; | |
999 | xfer = xfer_funcs[type]; | |
1000 | if (xfer == NULL) | |
1001 | return -EINVAL; | |
1002 | } else | |
1003 | xfer = NULL; | |
1004 | ||
1005 | err = loop_init_xfer(lo, xfer, info); | |
1006 | if (err) | |
1007 | return err; | |
1008 | ||
1009 | if (lo->lo_offset != info->lo_offset || | |
1010 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1011 | lo->lo_offset = info->lo_offset; | |
1012 | lo->lo_sizelimit = info->lo_sizelimit; | |
1013 | if (figure_loop_size(lo)) | |
1014 | return -EFBIG; | |
1015 | } | |
1016 | ||
1017 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1018 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1019 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1020 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1021 | ||
1022 | if (!xfer) | |
1023 | xfer = &none_funcs; | |
1024 | lo->transfer = xfer->transfer; | |
1025 | lo->ioctl = xfer->ioctl; | |
1026 | ||
1027 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != | |
1028 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1029 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1030 | ||
1031 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1032 | lo->lo_init[0] = info->lo_init[0]; | |
1033 | lo->lo_init[1] = info->lo_init[1]; | |
1034 | if (info->lo_encrypt_key_size) { | |
1035 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1036 | info->lo_encrypt_key_size); | |
1037 | lo->lo_key_owner = uid; | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
1043 | static int | |
1044 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1045 | { | |
1046 | struct file *file = lo->lo_backing_file; | |
1047 | struct kstat stat; | |
1048 | int error; | |
1049 | ||
1050 | if (lo->lo_state != Lo_bound) | |
1051 | return -ENXIO; | |
1052 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); | |
1053 | if (error) | |
1054 | return error; | |
1055 | memset(info, 0, sizeof(*info)); | |
1056 | info->lo_number = lo->lo_number; | |
1057 | info->lo_device = huge_encode_dev(stat.dev); | |
1058 | info->lo_inode = stat.ino; | |
1059 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1060 | info->lo_offset = lo->lo_offset; | |
1061 | info->lo_sizelimit = lo->lo_sizelimit; | |
1062 | info->lo_flags = lo->lo_flags; | |
1063 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1064 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1065 | info->lo_encrypt_type = | |
1066 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1067 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1068 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1069 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1070 | lo->lo_encrypt_key_size); | |
1071 | } | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | static void | |
1076 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1077 | { | |
1078 | memset(info64, 0, sizeof(*info64)); | |
1079 | info64->lo_number = info->lo_number; | |
1080 | info64->lo_device = info->lo_device; | |
1081 | info64->lo_inode = info->lo_inode; | |
1082 | info64->lo_rdevice = info->lo_rdevice; | |
1083 | info64->lo_offset = info->lo_offset; | |
1084 | info64->lo_sizelimit = 0; | |
1085 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1086 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1087 | info64->lo_flags = info->lo_flags; | |
1088 | info64->lo_init[0] = info->lo_init[0]; | |
1089 | info64->lo_init[1] = info->lo_init[1]; | |
1090 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1091 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1092 | else | |
1093 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1094 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1095 | } | |
1096 | ||
1097 | static int | |
1098 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1099 | { | |
1100 | memset(info, 0, sizeof(*info)); | |
1101 | info->lo_number = info64->lo_number; | |
1102 | info->lo_device = info64->lo_device; | |
1103 | info->lo_inode = info64->lo_inode; | |
1104 | info->lo_rdevice = info64->lo_rdevice; | |
1105 | info->lo_offset = info64->lo_offset; | |
1106 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1107 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1108 | info->lo_flags = info64->lo_flags; | |
1109 | info->lo_init[0] = info64->lo_init[0]; | |
1110 | info->lo_init[1] = info64->lo_init[1]; | |
1111 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1112 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1113 | else | |
1114 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1115 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1116 | ||
1117 | /* error in case values were truncated */ | |
1118 | if (info->lo_device != info64->lo_device || | |
1119 | info->lo_rdevice != info64->lo_rdevice || | |
1120 | info->lo_inode != info64->lo_inode || | |
1121 | info->lo_offset != info64->lo_offset) | |
1122 | return -EOVERFLOW; | |
1123 | ||
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | static int | |
1128 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1129 | { | |
1130 | struct loop_info info; | |
1131 | struct loop_info64 info64; | |
1132 | ||
1133 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1134 | return -EFAULT; | |
1135 | loop_info64_from_old(&info, &info64); | |
1136 | return loop_set_status(lo, &info64); | |
1137 | } | |
1138 | ||
1139 | static int | |
1140 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1141 | { | |
1142 | struct loop_info64 info64; | |
1143 | ||
1144 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1145 | return -EFAULT; | |
1146 | return loop_set_status(lo, &info64); | |
1147 | } | |
1148 | ||
1149 | static int | |
1150 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1151 | struct loop_info info; | |
1152 | struct loop_info64 info64; | |
1153 | int err = 0; | |
1154 | ||
1155 | if (!arg) | |
1156 | err = -EINVAL; | |
1157 | if (!err) | |
1158 | err = loop_get_status(lo, &info64); | |
1159 | if (!err) | |
1160 | err = loop_info64_to_old(&info64, &info); | |
1161 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1162 | err = -EFAULT; | |
1163 | ||
1164 | return err; | |
1165 | } | |
1166 | ||
1167 | static int | |
1168 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1169 | struct loop_info64 info64; | |
1170 | int err = 0; | |
1171 | ||
1172 | if (!arg) | |
1173 | err = -EINVAL; | |
1174 | if (!err) | |
1175 | err = loop_get_status(lo, &info64); | |
1176 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1177 | err = -EFAULT; | |
1178 | ||
1179 | return err; | |
1180 | } | |
1181 | ||
1182 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) | |
1183 | { | |
1184 | int err; | |
1185 | sector_t sec; | |
1186 | loff_t sz; | |
1187 | ||
1188 | err = -ENXIO; | |
1189 | if (unlikely(lo->lo_state != Lo_bound)) | |
1190 | goto out; | |
1191 | err = figure_loop_size(lo); | |
1192 | if (unlikely(err)) | |
1193 | goto out; | |
1194 | sec = get_capacity(lo->lo_disk); | |
1195 | /* the width of sector_t may be narrow for bit-shift */ | |
1196 | sz = sec; | |
1197 | sz <<= 9; | |
1198 | mutex_lock(&bdev->bd_mutex); | |
1199 | bd_set_size(bdev, sz); | |
1200 | /* let user-space know about the new size */ | |
1201 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); | |
1202 | mutex_unlock(&bdev->bd_mutex); | |
1203 | ||
1204 | out: | |
1205 | return err; | |
1206 | } | |
1207 | ||
1208 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, | |
1209 | unsigned int cmd, unsigned long arg) | |
1210 | { | |
1211 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1212 | int err; | |
1213 | ||
1214 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); | |
1215 | switch (cmd) { | |
1216 | case LOOP_SET_FD: | |
1217 | err = loop_set_fd(lo, mode, bdev, arg); | |
1218 | break; | |
1219 | case LOOP_CHANGE_FD: | |
1220 | err = loop_change_fd(lo, bdev, arg); | |
1221 | break; | |
1222 | case LOOP_CLR_FD: | |
1223 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ | |
1224 | err = loop_clr_fd(lo, bdev); | |
1225 | if (!err) | |
1226 | goto out_unlocked; | |
1227 | break; | |
1228 | case LOOP_SET_STATUS: | |
1229 | err = loop_set_status_old(lo, (struct loop_info __user *) arg); | |
1230 | break; | |
1231 | case LOOP_GET_STATUS: | |
1232 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1233 | break; | |
1234 | case LOOP_SET_STATUS64: | |
1235 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg); | |
1236 | break; | |
1237 | case LOOP_GET_STATUS64: | |
1238 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1239 | break; | |
1240 | case LOOP_SET_CAPACITY: | |
1241 | err = -EPERM; | |
1242 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1243 | err = loop_set_capacity(lo, bdev); | |
1244 | break; | |
1245 | default: | |
1246 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1247 | } | |
1248 | mutex_unlock(&lo->lo_ctl_mutex); | |
1249 | ||
1250 | out_unlocked: | |
1251 | return err; | |
1252 | } | |
1253 | ||
1254 | #ifdef CONFIG_COMPAT | |
1255 | struct compat_loop_info { | |
1256 | compat_int_t lo_number; /* ioctl r/o */ | |
1257 | compat_dev_t lo_device; /* ioctl r/o */ | |
1258 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1259 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1260 | compat_int_t lo_offset; | |
1261 | compat_int_t lo_encrypt_type; | |
1262 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1263 | compat_int_t lo_flags; /* ioctl r/o */ | |
1264 | char lo_name[LO_NAME_SIZE]; | |
1265 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1266 | compat_ulong_t lo_init[2]; | |
1267 | char reserved[4]; | |
1268 | }; | |
1269 | ||
1270 | /* | |
1271 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1272 | * - noinlined to reduce stack space usage in main part of driver | |
1273 | */ | |
1274 | static noinline int | |
1275 | loop_info64_from_compat(const struct compat_loop_info __user *arg, | |
1276 | struct loop_info64 *info64) | |
1277 | { | |
1278 | struct compat_loop_info info; | |
1279 | ||
1280 | if (copy_from_user(&info, arg, sizeof(info))) | |
1281 | return -EFAULT; | |
1282 | ||
1283 | memset(info64, 0, sizeof(*info64)); | |
1284 | info64->lo_number = info.lo_number; | |
1285 | info64->lo_device = info.lo_device; | |
1286 | info64->lo_inode = info.lo_inode; | |
1287 | info64->lo_rdevice = info.lo_rdevice; | |
1288 | info64->lo_offset = info.lo_offset; | |
1289 | info64->lo_sizelimit = 0; | |
1290 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1291 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1292 | info64->lo_flags = info.lo_flags; | |
1293 | info64->lo_init[0] = info.lo_init[0]; | |
1294 | info64->lo_init[1] = info.lo_init[1]; | |
1295 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1296 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1297 | else | |
1298 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1299 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1305 | * - noinlined to reduce stack space usage in main part of driver | |
1306 | */ | |
1307 | static noinline int | |
1308 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1309 | struct compat_loop_info __user *arg) | |
1310 | { | |
1311 | struct compat_loop_info info; | |
1312 | ||
1313 | memset(&info, 0, sizeof(info)); | |
1314 | info.lo_number = info64->lo_number; | |
1315 | info.lo_device = info64->lo_device; | |
1316 | info.lo_inode = info64->lo_inode; | |
1317 | info.lo_rdevice = info64->lo_rdevice; | |
1318 | info.lo_offset = info64->lo_offset; | |
1319 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1320 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1321 | info.lo_flags = info64->lo_flags; | |
1322 | info.lo_init[0] = info64->lo_init[0]; | |
1323 | info.lo_init[1] = info64->lo_init[1]; | |
1324 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1325 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1326 | else | |
1327 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1328 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1329 | ||
1330 | /* error in case values were truncated */ | |
1331 | if (info.lo_device != info64->lo_device || | |
1332 | info.lo_rdevice != info64->lo_rdevice || | |
1333 | info.lo_inode != info64->lo_inode || | |
1334 | info.lo_offset != info64->lo_offset || | |
1335 | info.lo_init[0] != info64->lo_init[0] || | |
1336 | info.lo_init[1] != info64->lo_init[1]) | |
1337 | return -EOVERFLOW; | |
1338 | ||
1339 | if (copy_to_user(arg, &info, sizeof(info))) | |
1340 | return -EFAULT; | |
1341 | return 0; | |
1342 | } | |
1343 | ||
1344 | static int | |
1345 | loop_set_status_compat(struct loop_device *lo, | |
1346 | const struct compat_loop_info __user *arg) | |
1347 | { | |
1348 | struct loop_info64 info64; | |
1349 | int ret; | |
1350 | ||
1351 | ret = loop_info64_from_compat(arg, &info64); | |
1352 | if (ret < 0) | |
1353 | return ret; | |
1354 | return loop_set_status(lo, &info64); | |
1355 | } | |
1356 | ||
1357 | static int | |
1358 | loop_get_status_compat(struct loop_device *lo, | |
1359 | struct compat_loop_info __user *arg) | |
1360 | { | |
1361 | struct loop_info64 info64; | |
1362 | int err = 0; | |
1363 | ||
1364 | if (!arg) | |
1365 | err = -EINVAL; | |
1366 | if (!err) | |
1367 | err = loop_get_status(lo, &info64); | |
1368 | if (!err) | |
1369 | err = loop_info64_to_compat(&info64, arg); | |
1370 | return err; | |
1371 | } | |
1372 | ||
1373 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1374 | unsigned int cmd, unsigned long arg) | |
1375 | { | |
1376 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1377 | int err; | |
1378 | ||
1379 | switch(cmd) { | |
1380 | case LOOP_SET_STATUS: | |
1381 | mutex_lock(&lo->lo_ctl_mutex); | |
1382 | err = loop_set_status_compat( | |
1383 | lo, (const struct compat_loop_info __user *) arg); | |
1384 | mutex_unlock(&lo->lo_ctl_mutex); | |
1385 | break; | |
1386 | case LOOP_GET_STATUS: | |
1387 | mutex_lock(&lo->lo_ctl_mutex); | |
1388 | err = loop_get_status_compat( | |
1389 | lo, (struct compat_loop_info __user *) arg); | |
1390 | mutex_unlock(&lo->lo_ctl_mutex); | |
1391 | break; | |
1392 | case LOOP_SET_CAPACITY: | |
1393 | case LOOP_CLR_FD: | |
1394 | case LOOP_GET_STATUS64: | |
1395 | case LOOP_SET_STATUS64: | |
1396 | arg = (unsigned long) compat_ptr(arg); | |
1397 | case LOOP_SET_FD: | |
1398 | case LOOP_CHANGE_FD: | |
1399 | err = lo_ioctl(bdev, mode, cmd, arg); | |
1400 | break; | |
1401 | default: | |
1402 | err = -ENOIOCTLCMD; | |
1403 | break; | |
1404 | } | |
1405 | return err; | |
1406 | } | |
1407 | #endif | |
1408 | ||
1409 | static int lo_open(struct block_device *bdev, fmode_t mode) | |
1410 | { | |
1411 | struct loop_device *lo = bdev->bd_disk->private_data; | |
1412 | ||
1413 | mutex_lock(&loop_mutex); | |
1414 | mutex_lock(&lo->lo_ctl_mutex); | |
1415 | lo->lo_refcnt++; | |
1416 | mutex_unlock(&lo->lo_ctl_mutex); | |
1417 | mutex_unlock(&loop_mutex); | |
1418 | ||
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | static int lo_release(struct gendisk *disk, fmode_t mode) | |
1423 | { | |
1424 | struct loop_device *lo = disk->private_data; | |
1425 | int err; | |
1426 | ||
1427 | mutex_lock(&loop_mutex); | |
1428 | mutex_lock(&lo->lo_ctl_mutex); | |
1429 | ||
1430 | if (--lo->lo_refcnt) | |
1431 | goto out; | |
1432 | ||
1433 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
1434 | /* | |
1435 | * In autoclear mode, stop the loop thread | |
1436 | * and remove configuration after last close. | |
1437 | */ | |
1438 | err = loop_clr_fd(lo, NULL); | |
1439 | if (!err) | |
1440 | goto out_unlocked; | |
1441 | } else { | |
1442 | /* | |
1443 | * Otherwise keep thread (if running) and config, | |
1444 | * but flush possible ongoing bios in thread. | |
1445 | */ | |
1446 | loop_flush(lo); | |
1447 | } | |
1448 | ||
1449 | out: | |
1450 | mutex_unlock(&lo->lo_ctl_mutex); | |
1451 | out_unlocked: | |
1452 | mutex_unlock(&loop_mutex); | |
1453 | return 0; | |
1454 | } | |
1455 | ||
1456 | static const struct block_device_operations lo_fops = { | |
1457 | .owner = THIS_MODULE, | |
1458 | .open = lo_open, | |
1459 | .release = lo_release, | |
1460 | .ioctl = lo_ioctl, | |
1461 | #ifdef CONFIG_COMPAT | |
1462 | .compat_ioctl = lo_compat_ioctl, | |
1463 | #endif | |
1464 | }; | |
1465 | ||
1466 | /* | |
1467 | * And now the modules code and kernel interface. | |
1468 | */ | |
1469 | static int max_loop; | |
1470 | module_param(max_loop, int, 0); | |
1471 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); | |
1472 | module_param(max_part, int, 0); | |
1473 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); | |
1474 | MODULE_LICENSE("GPL"); | |
1475 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1476 | ||
1477 | int loop_register_transfer(struct loop_func_table *funcs) | |
1478 | { | |
1479 | unsigned int n = funcs->number; | |
1480 | ||
1481 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1482 | return -EINVAL; | |
1483 | xfer_funcs[n] = funcs; | |
1484 | return 0; | |
1485 | } | |
1486 | ||
1487 | int loop_unregister_transfer(int number) | |
1488 | { | |
1489 | unsigned int n = number; | |
1490 | struct loop_device *lo; | |
1491 | struct loop_func_table *xfer; | |
1492 | ||
1493 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1494 | return -EINVAL; | |
1495 | ||
1496 | xfer_funcs[n] = NULL; | |
1497 | ||
1498 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1499 | mutex_lock(&lo->lo_ctl_mutex); | |
1500 | ||
1501 | if (lo->lo_encryption == xfer) | |
1502 | loop_release_xfer(lo); | |
1503 | ||
1504 | mutex_unlock(&lo->lo_ctl_mutex); | |
1505 | } | |
1506 | ||
1507 | return 0; | |
1508 | } | |
1509 | ||
1510 | EXPORT_SYMBOL(loop_register_transfer); | |
1511 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1512 | ||
1513 | static struct loop_device *loop_alloc(int i) | |
1514 | { | |
1515 | struct loop_device *lo; | |
1516 | struct gendisk *disk; | |
1517 | ||
1518 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); | |
1519 | if (!lo) | |
1520 | goto out; | |
1521 | ||
1522 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); | |
1523 | if (!lo->lo_queue) | |
1524 | goto out_free_dev; | |
1525 | ||
1526 | disk = lo->lo_disk = alloc_disk(1 << part_shift); | |
1527 | if (!disk) | |
1528 | goto out_free_queue; | |
1529 | ||
1530 | mutex_init(&lo->lo_ctl_mutex); | |
1531 | lo->lo_number = i; | |
1532 | lo->lo_thread = NULL; | |
1533 | init_waitqueue_head(&lo->lo_event); | |
1534 | spin_lock_init(&lo->lo_lock); | |
1535 | disk->major = LOOP_MAJOR; | |
1536 | disk->first_minor = i << part_shift; | |
1537 | disk->fops = &lo_fops; | |
1538 | disk->private_data = lo; | |
1539 | disk->queue = lo->lo_queue; | |
1540 | sprintf(disk->disk_name, "loop%d", i); | |
1541 | return lo; | |
1542 | ||
1543 | out_free_queue: | |
1544 | blk_cleanup_queue(lo->lo_queue); | |
1545 | out_free_dev: | |
1546 | kfree(lo); | |
1547 | out: | |
1548 | return NULL; | |
1549 | } | |
1550 | ||
1551 | static void loop_free(struct loop_device *lo) | |
1552 | { | |
1553 | blk_cleanup_queue(lo->lo_queue); | |
1554 | put_disk(lo->lo_disk); | |
1555 | list_del(&lo->lo_list); | |
1556 | kfree(lo); | |
1557 | } | |
1558 | ||
1559 | static struct loop_device *loop_init_one(int i) | |
1560 | { | |
1561 | struct loop_device *lo; | |
1562 | ||
1563 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1564 | if (lo->lo_number == i) | |
1565 | return lo; | |
1566 | } | |
1567 | ||
1568 | lo = loop_alloc(i); | |
1569 | if (lo) { | |
1570 | add_disk(lo->lo_disk); | |
1571 | list_add_tail(&lo->lo_list, &loop_devices); | |
1572 | } | |
1573 | return lo; | |
1574 | } | |
1575 | ||
1576 | static void loop_del_one(struct loop_device *lo) | |
1577 | { | |
1578 | del_gendisk(lo->lo_disk); | |
1579 | loop_free(lo); | |
1580 | } | |
1581 | ||
1582 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) | |
1583 | { | |
1584 | struct loop_device *lo; | |
1585 | struct kobject *kobj; | |
1586 | ||
1587 | mutex_lock(&loop_devices_mutex); | |
1588 | lo = loop_init_one(dev & MINORMASK); | |
1589 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM); | |
1590 | mutex_unlock(&loop_devices_mutex); | |
1591 | ||
1592 | *part = 0; | |
1593 | return kobj; | |
1594 | } | |
1595 | ||
1596 | static int __init loop_init(void) | |
1597 | { | |
1598 | int i, nr; | |
1599 | unsigned long range; | |
1600 | struct loop_device *lo, *next; | |
1601 | ||
1602 | /* | |
1603 | * loop module now has a feature to instantiate underlying device | |
1604 | * structure on-demand, provided that there is an access dev node. | |
1605 | * However, this will not work well with user space tool that doesn't | |
1606 | * know about such "feature". In order to not break any existing | |
1607 | * tool, we do the following: | |
1608 | * | |
1609 | * (1) if max_loop is specified, create that many upfront, and this | |
1610 | * also becomes a hard limit. | |
1611 | * (2) if max_loop is not specified, create 8 loop device on module | |
1612 | * load, user can further extend loop device by create dev node | |
1613 | * themselves and have kernel automatically instantiate actual | |
1614 | * device on-demand. | |
1615 | */ | |
1616 | ||
1617 | part_shift = 0; | |
1618 | if (max_part > 0) | |
1619 | part_shift = fls(max_part); | |
1620 | ||
1621 | if (max_loop > 1UL << (MINORBITS - part_shift)) | |
1622 | return -EINVAL; | |
1623 | ||
1624 | if (max_loop) { | |
1625 | nr = max_loop; | |
1626 | range = max_loop; | |
1627 | } else { | |
1628 | nr = 8; | |
1629 | range = 1UL << (MINORBITS - part_shift); | |
1630 | } | |
1631 | ||
1632 | if (register_blkdev(LOOP_MAJOR, "loop")) | |
1633 | return -EIO; | |
1634 | ||
1635 | for (i = 0; i < nr; i++) { | |
1636 | lo = loop_alloc(i); | |
1637 | if (!lo) | |
1638 | goto Enomem; | |
1639 | list_add_tail(&lo->lo_list, &loop_devices); | |
1640 | } | |
1641 | ||
1642 | /* point of no return */ | |
1643 | ||
1644 | list_for_each_entry(lo, &loop_devices, lo_list) | |
1645 | add_disk(lo->lo_disk); | |
1646 | ||
1647 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, | |
1648 | THIS_MODULE, loop_probe, NULL, NULL); | |
1649 | ||
1650 | printk(KERN_INFO "loop: module loaded\n"); | |
1651 | return 0; | |
1652 | ||
1653 | Enomem: | |
1654 | printk(KERN_INFO "loop: out of memory\n"); | |
1655 | ||
1656 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
1657 | loop_free(lo); | |
1658 | ||
1659 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
1660 | return -ENOMEM; | |
1661 | } | |
1662 | ||
1663 | static void __exit loop_exit(void) | |
1664 | { | |
1665 | unsigned long range; | |
1666 | struct loop_device *lo, *next; | |
1667 | ||
1668 | range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift); | |
1669 | ||
1670 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
1671 | loop_del_one(lo); | |
1672 | ||
1673 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); | |
1674 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
1675 | } | |
1676 | ||
1677 | module_init(loop_init); | |
1678 | module_exit(loop_exit); | |
1679 | ||
1680 | #ifndef MODULE | |
1681 | static int __init max_loop_setup(char *str) | |
1682 | { | |
1683 | max_loop = simple_strtol(str, NULL, 0); | |
1684 | return 1; | |
1685 | } | |
1686 | ||
1687 | __setup("max_loop=", max_loop_setup); | |
1688 | #endif |