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; |
a3941ec1 | 395 | int size, ret; |
1da177e4 | 396 | |
cac36bb0 | 397 | ret = buf->ops->confirm(pipe, buf); |
fd582140 JA |
398 | if (unlikely(ret)) |
399 | return ret; | |
1da177e4 | 400 | |
fd582140 JA |
401 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + |
402 | (buf->offset >> 9); | |
403 | size = sd->len; | |
404 | if (size > p->bsize) | |
405 | size = p->bsize; | |
1da177e4 | 406 | |
fd582140 | 407 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { |
1da177e4 LT |
408 | printk(KERN_ERR "loop: transfer error block %ld\n", |
409 | page->index); | |
fd582140 | 410 | size = -EINVAL; |
1da177e4 LT |
411 | } |
412 | ||
413 | flush_dcache_page(p->page); | |
414 | ||
fd582140 JA |
415 | if (size > 0) |
416 | p->offset += size; | |
417 | ||
1da177e4 LT |
418 | return size; |
419 | } | |
420 | ||
fd582140 JA |
421 | static int |
422 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
423 | { | |
424 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
425 | } | |
426 | ||
1da177e4 LT |
427 | static int |
428 | do_lo_receive(struct loop_device *lo, | |
429 | struct bio_vec *bvec, int bsize, loff_t pos) | |
430 | { | |
431 | struct lo_read_data cookie; | |
fd582140 | 432 | struct splice_desc sd; |
1da177e4 | 433 | struct file *file; |
fd582140 | 434 | long retval; |
1da177e4 LT |
435 | |
436 | cookie.lo = lo; | |
437 | cookie.page = bvec->bv_page; | |
438 | cookie.offset = bvec->bv_offset; | |
439 | cookie.bsize = bsize; | |
fd582140 JA |
440 | |
441 | sd.len = 0; | |
442 | sd.total_len = bvec->bv_len; | |
443 | sd.flags = 0; | |
444 | sd.pos = pos; | |
445 | sd.u.data = &cookie; | |
446 | ||
1da177e4 | 447 | file = lo->lo_backing_file; |
fd582140 JA |
448 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); |
449 | ||
450 | if (retval < 0) | |
451 | return retval; | |
452 | ||
453 | return 0; | |
1da177e4 LT |
454 | } |
455 | ||
456 | static int | |
457 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
458 | { | |
459 | struct bio_vec *bvec; | |
460 | int i, ret = 0; | |
461 | ||
462 | bio_for_each_segment(bvec, bio, i) { | |
463 | ret = do_lo_receive(lo, bvec, bsize, pos); | |
464 | if (ret < 0) | |
465 | break; | |
466 | pos += bvec->bv_len; | |
467 | } | |
468 | return ret; | |
469 | } | |
470 | ||
471 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
472 | { | |
473 | loff_t pos; | |
474 | int ret; | |
475 | ||
476 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
68db1961 NK |
477 | |
478 | if (bio_rw(bio) == WRITE) { | |
479 | int barrier = bio_barrier(bio); | |
480 | struct file *file = lo->lo_backing_file; | |
481 | ||
482 | if (barrier) { | |
483 | if (unlikely(!file->f_op->fsync)) { | |
484 | ret = -EOPNOTSUPP; | |
485 | goto out; | |
486 | } | |
487 | ||
488 | ret = vfs_fsync(file, file->f_path.dentry, 0); | |
489 | if (unlikely(ret)) { | |
490 | ret = -EIO; | |
491 | goto out; | |
492 | } | |
493 | } | |
494 | ||
511de73f | 495 | ret = lo_send(lo, bio, pos); |
68db1961 NK |
496 | |
497 | if (barrier && !ret) { | |
498 | ret = vfs_fsync(file, file->f_path.dentry, 0); | |
499 | if (unlikely(ret)) | |
500 | ret = -EIO; | |
501 | } | |
502 | } else | |
1da177e4 | 503 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); |
68db1961 NK |
504 | |
505 | out: | |
1da177e4 LT |
506 | return ret; |
507 | } | |
508 | ||
509 | /* | |
510 | * Add bio to back of pending list | |
511 | */ | |
512 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
513 | { | |
1da177e4 LT |
514 | if (lo->lo_biotail) { |
515 | lo->lo_biotail->bi_next = bio; | |
516 | lo->lo_biotail = bio; | |
517 | } else | |
518 | lo->lo_bio = lo->lo_biotail = bio; | |
1da177e4 LT |
519 | } |
520 | ||
521 | /* | |
522 | * Grab first pending buffer | |
523 | */ | |
524 | static struct bio *loop_get_bio(struct loop_device *lo) | |
525 | { | |
526 | struct bio *bio; | |
527 | ||
1da177e4 LT |
528 | if ((bio = lo->lo_bio)) { |
529 | if (bio == lo->lo_biotail) | |
530 | lo->lo_biotail = NULL; | |
531 | lo->lo_bio = bio->bi_next; | |
532 | bio->bi_next = NULL; | |
533 | } | |
1da177e4 LT |
534 | |
535 | return bio; | |
536 | } | |
537 | ||
165125e1 | 538 | static int loop_make_request(struct request_queue *q, struct bio *old_bio) |
1da177e4 LT |
539 | { |
540 | struct loop_device *lo = q->queuedata; | |
541 | int rw = bio_rw(old_bio); | |
542 | ||
35a82d1a NP |
543 | if (rw == READA) |
544 | rw = READ; | |
545 | ||
546 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
1da177e4 LT |
547 | |
548 | spin_lock_irq(&lo->lo_lock); | |
549 | if (lo->lo_state != Lo_bound) | |
35a82d1a NP |
550 | goto out; |
551 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
552 | goto out; | |
1da177e4 | 553 | loop_add_bio(lo, old_bio); |
6c997918 | 554 | wake_up(&lo->lo_event); |
35a82d1a | 555 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 556 | return 0; |
35a82d1a | 557 | |
1da177e4 | 558 | out: |
35a82d1a | 559 | spin_unlock_irq(&lo->lo_lock); |
6712ecf8 | 560 | bio_io_error(old_bio); |
1da177e4 | 561 | return 0; |
1da177e4 LT |
562 | } |
563 | ||
564 | /* | |
565 | * kick off io on the underlying address space | |
566 | */ | |
165125e1 | 567 | static void loop_unplug(struct request_queue *q) |
1da177e4 LT |
568 | { |
569 | struct loop_device *lo = q->queuedata; | |
570 | ||
75ad23bc | 571 | queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q); |
1da177e4 LT |
572 | blk_run_address_space(lo->lo_backing_file->f_mapping); |
573 | } | |
574 | ||
575 | struct switch_request { | |
576 | struct file *file; | |
577 | struct completion wait; | |
578 | }; | |
579 | ||
580 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
581 | ||
582 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
583 | { | |
1da177e4 LT |
584 | if (unlikely(!bio->bi_bdev)) { |
585 | do_loop_switch(lo, bio->bi_private); | |
586 | bio_put(bio); | |
587 | } else { | |
35a82d1a | 588 | int ret = do_bio_filebacked(lo, bio); |
6712ecf8 | 589 | bio_endio(bio, ret); |
1da177e4 LT |
590 | } |
591 | } | |
592 | ||
593 | /* | |
594 | * worker thread that handles reads/writes to file backed loop devices, | |
595 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
596 | * on reads for block backed loop, as that is too heavy to do from | |
597 | * b_end_io context where irqs may be disabled. | |
6c997918 SH |
598 | * |
599 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
600 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
601 | * true, make_request will not place any more requests. Therefore | |
602 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
603 | * done with the loop. | |
1da177e4 LT |
604 | */ |
605 | static int loop_thread(void *data) | |
606 | { | |
607 | struct loop_device *lo = data; | |
608 | struct bio *bio; | |
609 | ||
1da177e4 LT |
610 | set_user_nice(current, -20); |
611 | ||
6c997918 | 612 | while (!kthread_should_stop() || lo->lo_bio) { |
09c0dc68 | 613 | |
6c997918 SH |
614 | wait_event_interruptible(lo->lo_event, |
615 | lo->lo_bio || kthread_should_stop()); | |
35a82d1a | 616 | |
6c997918 | 617 | if (!lo->lo_bio) |
35a82d1a | 618 | continue; |
35a82d1a | 619 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 620 | bio = loop_get_bio(lo); |
35a82d1a NP |
621 | spin_unlock_irq(&lo->lo_lock); |
622 | ||
623 | BUG_ON(!bio); | |
1da177e4 | 624 | loop_handle_bio(lo, bio); |
1da177e4 LT |
625 | } |
626 | ||
1da177e4 LT |
627 | return 0; |
628 | } | |
629 | ||
630 | /* | |
631 | * loop_switch performs the hard work of switching a backing store. | |
632 | * First it needs to flush existing IO, it does this by sending a magic | |
633 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
634 | */ | |
635 | static int loop_switch(struct loop_device *lo, struct file *file) | |
636 | { | |
637 | struct switch_request w; | |
a24eab1e | 638 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); |
1da177e4 LT |
639 | if (!bio) |
640 | return -ENOMEM; | |
641 | init_completion(&w.wait); | |
642 | w.file = file; | |
643 | bio->bi_private = &w; | |
644 | bio->bi_bdev = NULL; | |
645 | loop_make_request(lo->lo_queue, bio); | |
646 | wait_for_completion(&w.wait); | |
647 | return 0; | |
648 | } | |
649 | ||
14f27939 MB |
650 | /* |
651 | * Helper to flush the IOs in loop, but keeping loop thread running | |
652 | */ | |
653 | static int loop_flush(struct loop_device *lo) | |
654 | { | |
655 | /* loop not yet configured, no running thread, nothing to flush */ | |
656 | if (!lo->lo_thread) | |
657 | return 0; | |
658 | ||
659 | return loop_switch(lo, NULL); | |
660 | } | |
661 | ||
1da177e4 LT |
662 | /* |
663 | * Do the actual switch; called from the BIO completion routine | |
664 | */ | |
665 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
666 | { | |
667 | struct file *file = p->file; | |
668 | struct file *old_file = lo->lo_backing_file; | |
14f27939 MB |
669 | struct address_space *mapping; |
670 | ||
671 | /* if no new file, only flush of queued bios requested */ | |
672 | if (!file) | |
673 | goto out; | |
1da177e4 | 674 | |
14f27939 | 675 | mapping = file->f_mapping; |
1da177e4 LT |
676 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); |
677 | lo->lo_backing_file = file; | |
ba52de12 TT |
678 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
679 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 LT |
680 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
681 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
14f27939 | 682 | out: |
1da177e4 LT |
683 | complete(&p->wait); |
684 | } | |
685 | ||
686 | ||
687 | /* | |
688 | * loop_change_fd switched the backing store of a loopback device to | |
689 | * a new file. This is useful for operating system installers to free up | |
690 | * the original file and in High Availability environments to switch to | |
691 | * an alternative location for the content in case of server meltdown. | |
692 | * This can only work if the loop device is used read-only, and if the | |
693 | * new backing store is the same size and type as the old backing store. | |
694 | */ | |
bb214884 AV |
695 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
696 | unsigned int arg) | |
1da177e4 LT |
697 | { |
698 | struct file *file, *old_file; | |
699 | struct inode *inode; | |
700 | int error; | |
701 | ||
702 | error = -ENXIO; | |
703 | if (lo->lo_state != Lo_bound) | |
704 | goto out; | |
705 | ||
706 | /* the loop device has to be read-only */ | |
707 | error = -EINVAL; | |
708 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
709 | goto out; | |
710 | ||
711 | error = -EBADF; | |
712 | file = fget(arg); | |
713 | if (!file) | |
714 | goto out; | |
715 | ||
716 | inode = file->f_mapping->host; | |
717 | old_file = lo->lo_backing_file; | |
718 | ||
719 | error = -EINVAL; | |
720 | ||
721 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
722 | goto out_putf; | |
723 | ||
fd582140 JA |
724 | /* new backing store needs to support loop (eg splice_read) */ |
725 | if (!inode->i_fop->splice_read) | |
1da177e4 LT |
726 | goto out_putf; |
727 | ||
728 | /* size of the new backing store needs to be the same */ | |
729 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
730 | goto out_putf; | |
731 | ||
732 | /* and ... switch */ | |
733 | error = loop_switch(lo, file); | |
734 | if (error) | |
735 | goto out_putf; | |
736 | ||
737 | fput(old_file); | |
476a4813 LV |
738 | if (max_part > 0) |
739 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
740 | return 0; |
741 | ||
742 | out_putf: | |
743 | fput(file); | |
744 | out: | |
745 | return error; | |
746 | } | |
747 | ||
748 | static inline int is_loop_device(struct file *file) | |
749 | { | |
750 | struct inode *i = file->f_mapping->host; | |
751 | ||
752 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
753 | } | |
754 | ||
bb214884 | 755 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
756 | struct block_device *bdev, unsigned int arg) |
757 | { | |
758 | struct file *file, *f; | |
759 | struct inode *inode; | |
760 | struct address_space *mapping; | |
761 | unsigned lo_blocksize; | |
762 | int lo_flags = 0; | |
763 | int error; | |
764 | loff_t size; | |
765 | ||
766 | /* This is safe, since we have a reference from open(). */ | |
767 | __module_get(THIS_MODULE); | |
768 | ||
769 | error = -EBADF; | |
770 | file = fget(arg); | |
771 | if (!file) | |
772 | goto out; | |
773 | ||
774 | error = -EBUSY; | |
775 | if (lo->lo_state != Lo_unbound) | |
776 | goto out_putf; | |
777 | ||
778 | /* Avoid recursion */ | |
779 | f = file; | |
780 | while (is_loop_device(f)) { | |
781 | struct loop_device *l; | |
782 | ||
bb214884 | 783 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
784 | goto out_putf; |
785 | ||
786 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
787 | if (l->lo_state == Lo_unbound) { | |
788 | error = -EINVAL; | |
789 | goto out_putf; | |
790 | } | |
791 | f = l->lo_backing_file; | |
792 | } | |
793 | ||
794 | mapping = file->f_mapping; | |
795 | inode = mapping->host; | |
796 | ||
797 | if (!(file->f_mode & FMODE_WRITE)) | |
798 | lo_flags |= LO_FLAGS_READ_ONLY; | |
799 | ||
800 | error = -EINVAL; | |
801 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
f5e54d6e | 802 | const struct address_space_operations *aops = mapping->a_ops; |
1da177e4 LT |
803 | /* |
804 | * If we can't read - sorry. If we only can't write - well, | |
805 | * it's going to be read-only. | |
806 | */ | |
fd582140 | 807 | if (!file->f_op->splice_read) |
1da177e4 | 808 | goto out_putf; |
4e02ed4b | 809 | if (aops->write_begin) |
1da177e4 LT |
810 | lo_flags |= LO_FLAGS_USE_AOPS; |
811 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) | |
812 | lo_flags |= LO_FLAGS_READ_ONLY; | |
813 | ||
ba52de12 TT |
814 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
815 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
816 | ||
1da177e4 LT |
817 | error = 0; |
818 | } else { | |
819 | goto out_putf; | |
820 | } | |
821 | ||
822 | size = get_loop_size(lo, file); | |
823 | ||
824 | if ((loff_t)(sector_t)size != size) { | |
825 | error = -EFBIG; | |
826 | goto out_putf; | |
827 | } | |
828 | ||
bb214884 | 829 | if (!(mode & FMODE_WRITE)) |
1da177e4 LT |
830 | lo_flags |= LO_FLAGS_READ_ONLY; |
831 | ||
832 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
833 | ||
834 | lo->lo_blocksize = lo_blocksize; | |
835 | lo->lo_device = bdev; | |
836 | lo->lo_flags = lo_flags; | |
837 | lo->lo_backing_file = file; | |
eefe85ee | 838 | lo->transfer = transfer_none; |
1da177e4 LT |
839 | lo->ioctl = NULL; |
840 | lo->lo_sizelimit = 0; | |
841 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
842 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
843 | ||
844 | lo->lo_bio = lo->lo_biotail = NULL; | |
845 | ||
846 | /* | |
847 | * set queue make_request_fn, and add limits based on lower level | |
848 | * device | |
849 | */ | |
850 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
851 | lo->lo_queue->queuedata = lo; | |
852 | lo->lo_queue->unplug_fn = loop_unplug; | |
853 | ||
68db1961 NK |
854 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
855 | blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL); | |
856 | ||
73285082 | 857 | set_capacity(lo->lo_disk, size); |
1da177e4 LT |
858 | bd_set_size(bdev, size << 9); |
859 | ||
860 | set_blocksize(bdev, lo_blocksize); | |
861 | ||
6c997918 SH |
862 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", |
863 | lo->lo_number); | |
864 | if (IS_ERR(lo->lo_thread)) { | |
865 | error = PTR_ERR(lo->lo_thread); | |
a7422bf8 | 866 | goto out_clr; |
6c997918 SH |
867 | } |
868 | lo->lo_state = Lo_bound; | |
869 | wake_up_process(lo->lo_thread); | |
476a4813 LV |
870 | if (max_part > 0) |
871 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
872 | return 0; |
873 | ||
a7422bf8 SH |
874 | out_clr: |
875 | lo->lo_thread = NULL; | |
876 | lo->lo_device = NULL; | |
877 | lo->lo_backing_file = NULL; | |
878 | lo->lo_flags = 0; | |
73285082 | 879 | set_capacity(lo->lo_disk, 0); |
f98393a6 | 880 | invalidate_bdev(bdev); |
a7422bf8 SH |
881 | bd_set_size(bdev, 0); |
882 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); | |
883 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
884 | out_putf: |
885 | fput(file); | |
886 | out: | |
887 | /* This is safe: open() is still holding a reference. */ | |
888 | module_put(THIS_MODULE); | |
889 | return error; | |
890 | } | |
891 | ||
892 | static int | |
893 | loop_release_xfer(struct loop_device *lo) | |
894 | { | |
895 | int err = 0; | |
896 | struct loop_func_table *xfer = lo->lo_encryption; | |
897 | ||
898 | if (xfer) { | |
899 | if (xfer->release) | |
900 | err = xfer->release(lo); | |
901 | lo->transfer = NULL; | |
902 | lo->lo_encryption = NULL; | |
903 | module_put(xfer->owner); | |
904 | } | |
905 | return err; | |
906 | } | |
907 | ||
908 | static int | |
909 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
910 | const struct loop_info64 *i) | |
911 | { | |
912 | int err = 0; | |
913 | ||
914 | if (xfer) { | |
915 | struct module *owner = xfer->owner; | |
916 | ||
917 | if (!try_module_get(owner)) | |
918 | return -EINVAL; | |
919 | if (xfer->init) | |
920 | err = xfer->init(lo, i); | |
921 | if (err) | |
922 | module_put(owner); | |
923 | else | |
924 | lo->lo_encryption = xfer; | |
925 | } | |
926 | return err; | |
927 | } | |
928 | ||
929 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev) | |
930 | { | |
931 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 932 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 LT |
933 | |
934 | if (lo->lo_state != Lo_bound) | |
935 | return -ENXIO; | |
936 | ||
937 | if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */ | |
938 | return -EBUSY; | |
939 | ||
940 | if (filp == NULL) | |
941 | return -EINVAL; | |
942 | ||
943 | spin_lock_irq(&lo->lo_lock); | |
944 | lo->lo_state = Lo_rundown; | |
1da177e4 LT |
945 | spin_unlock_irq(&lo->lo_lock); |
946 | ||
6c997918 | 947 | kthread_stop(lo->lo_thread); |
1da177e4 | 948 | |
8ae30b89 | 949 | lo->lo_queue->unplug_fn = NULL; |
1da177e4 LT |
950 | lo->lo_backing_file = NULL; |
951 | ||
952 | loop_release_xfer(lo); | |
953 | lo->transfer = NULL; | |
954 | lo->ioctl = NULL; | |
955 | lo->lo_device = NULL; | |
956 | lo->lo_encryption = NULL; | |
957 | lo->lo_offset = 0; | |
958 | lo->lo_sizelimit = 0; | |
959 | lo->lo_encrypt_key_size = 0; | |
960 | lo->lo_flags = 0; | |
6c997918 | 961 | lo->lo_thread = NULL; |
1da177e4 LT |
962 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
963 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
964 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
bb214884 AV |
965 | if (bdev) |
966 | invalidate_bdev(bdev); | |
73285082 | 967 | set_capacity(lo->lo_disk, 0); |
bb214884 AV |
968 | if (bdev) |
969 | bd_set_size(bdev, 0); | |
1da177e4 LT |
970 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
971 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
972 | /* This is safe: open() is still holding a reference. */ |
973 | module_put(THIS_MODULE); | |
476a4813 LV |
974 | if (max_part > 0) |
975 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
f028f3b2 NK |
976 | mutex_unlock(&lo->lo_ctl_mutex); |
977 | /* | |
978 | * Need not hold lo_ctl_mutex to fput backing file. | |
979 | * Calling fput holding lo_ctl_mutex triggers a circular | |
980 | * lock dependency possibility warning as fput can take | |
981 | * bd_mutex which is usually taken before lo_ctl_mutex. | |
982 | */ | |
983 | fput(filp); | |
1da177e4 LT |
984 | return 0; |
985 | } | |
986 | ||
987 | static int | |
988 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
989 | { | |
990 | int err; | |
991 | struct loop_func_table *xfer; | |
b0fafa81 | 992 | uid_t uid = current_uid(); |
1da177e4 | 993 | |
b0fafa81 DH |
994 | if (lo->lo_encrypt_key_size && |
995 | lo->lo_key_owner != uid && | |
1da177e4 LT |
996 | !capable(CAP_SYS_ADMIN)) |
997 | return -EPERM; | |
998 | if (lo->lo_state != Lo_bound) | |
999 | return -ENXIO; | |
1000 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1001 | return -EINVAL; | |
1002 | ||
1003 | err = loop_release_xfer(lo); | |
1004 | if (err) | |
1005 | return err; | |
1006 | ||
1007 | if (info->lo_encrypt_type) { | |
1008 | unsigned int type = info->lo_encrypt_type; | |
1009 | ||
1010 | if (type >= MAX_LO_CRYPT) | |
1011 | return -EINVAL; | |
1012 | xfer = xfer_funcs[type]; | |
1013 | if (xfer == NULL) | |
1014 | return -EINVAL; | |
1015 | } else | |
1016 | xfer = NULL; | |
1017 | ||
1018 | err = loop_init_xfer(lo, xfer, info); | |
1019 | if (err) | |
1020 | return err; | |
1021 | ||
1022 | if (lo->lo_offset != info->lo_offset || | |
1023 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1024 | lo->lo_offset = info->lo_offset; | |
1025 | lo->lo_sizelimit = info->lo_sizelimit; | |
1026 | if (figure_loop_size(lo)) | |
1027 | return -EFBIG; | |
1028 | } | |
1029 | ||
1030 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1031 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1032 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1033 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1034 | ||
1035 | if (!xfer) | |
1036 | xfer = &none_funcs; | |
1037 | lo->transfer = xfer->transfer; | |
1038 | lo->ioctl = xfer->ioctl; | |
1039 | ||
96c58655 DW |
1040 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1041 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1042 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1043 | ||
1da177e4 LT |
1044 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1045 | lo->lo_init[0] = info->lo_init[0]; | |
1046 | lo->lo_init[1] = info->lo_init[1]; | |
1047 | if (info->lo_encrypt_key_size) { | |
1048 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1049 | info->lo_encrypt_key_size); | |
b0fafa81 | 1050 | lo->lo_key_owner = uid; |
1da177e4 LT |
1051 | } |
1052 | ||
1053 | return 0; | |
1054 | } | |
1055 | ||
1056 | static int | |
1057 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1058 | { | |
1059 | struct file *file = lo->lo_backing_file; | |
1060 | struct kstat stat; | |
1061 | int error; | |
1062 | ||
1063 | if (lo->lo_state != Lo_bound) | |
1064 | return -ENXIO; | |
6c648be6 | 1065 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); |
1da177e4 LT |
1066 | if (error) |
1067 | return error; | |
1068 | memset(info, 0, sizeof(*info)); | |
1069 | info->lo_number = lo->lo_number; | |
1070 | info->lo_device = huge_encode_dev(stat.dev); | |
1071 | info->lo_inode = stat.ino; | |
1072 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1073 | info->lo_offset = lo->lo_offset; | |
1074 | info->lo_sizelimit = lo->lo_sizelimit; | |
1075 | info->lo_flags = lo->lo_flags; | |
1076 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1077 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1078 | info->lo_encrypt_type = | |
1079 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1080 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1081 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1082 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1083 | lo->lo_encrypt_key_size); | |
1084 | } | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | static void | |
1089 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1090 | { | |
1091 | memset(info64, 0, sizeof(*info64)); | |
1092 | info64->lo_number = info->lo_number; | |
1093 | info64->lo_device = info->lo_device; | |
1094 | info64->lo_inode = info->lo_inode; | |
1095 | info64->lo_rdevice = info->lo_rdevice; | |
1096 | info64->lo_offset = info->lo_offset; | |
1097 | info64->lo_sizelimit = 0; | |
1098 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1099 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1100 | info64->lo_flags = info->lo_flags; | |
1101 | info64->lo_init[0] = info->lo_init[0]; | |
1102 | info64->lo_init[1] = info->lo_init[1]; | |
1103 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1104 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1105 | else | |
1106 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1107 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1108 | } | |
1109 | ||
1110 | static int | |
1111 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1112 | { | |
1113 | memset(info, 0, sizeof(*info)); | |
1114 | info->lo_number = info64->lo_number; | |
1115 | info->lo_device = info64->lo_device; | |
1116 | info->lo_inode = info64->lo_inode; | |
1117 | info->lo_rdevice = info64->lo_rdevice; | |
1118 | info->lo_offset = info64->lo_offset; | |
1119 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1120 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1121 | info->lo_flags = info64->lo_flags; | |
1122 | info->lo_init[0] = info64->lo_init[0]; | |
1123 | info->lo_init[1] = info64->lo_init[1]; | |
1124 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1125 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1126 | else | |
1127 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1128 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1129 | ||
1130 | /* error in case values were truncated */ | |
1131 | if (info->lo_device != info64->lo_device || | |
1132 | info->lo_rdevice != info64->lo_rdevice || | |
1133 | info->lo_inode != info64->lo_inode || | |
1134 | info->lo_offset != info64->lo_offset) | |
1135 | return -EOVERFLOW; | |
1136 | ||
1137 | return 0; | |
1138 | } | |
1139 | ||
1140 | static int | |
1141 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1142 | { | |
1143 | struct loop_info info; | |
1144 | struct loop_info64 info64; | |
1145 | ||
1146 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1147 | return -EFAULT; | |
1148 | loop_info64_from_old(&info, &info64); | |
1149 | return loop_set_status(lo, &info64); | |
1150 | } | |
1151 | ||
1152 | static int | |
1153 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1154 | { | |
1155 | struct loop_info64 info64; | |
1156 | ||
1157 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1158 | return -EFAULT; | |
1159 | return loop_set_status(lo, &info64); | |
1160 | } | |
1161 | ||
1162 | static int | |
1163 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1164 | struct loop_info info; | |
1165 | struct loop_info64 info64; | |
1166 | int err = 0; | |
1167 | ||
1168 | if (!arg) | |
1169 | err = -EINVAL; | |
1170 | if (!err) | |
1171 | err = loop_get_status(lo, &info64); | |
1172 | if (!err) | |
1173 | err = loop_info64_to_old(&info64, &info); | |
1174 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1175 | err = -EFAULT; | |
1176 | ||
1177 | return err; | |
1178 | } | |
1179 | ||
1180 | static int | |
1181 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1182 | struct loop_info64 info64; | |
1183 | int err = 0; | |
1184 | ||
1185 | if (!arg) | |
1186 | err = -EINVAL; | |
1187 | if (!err) | |
1188 | err = loop_get_status(lo, &info64); | |
1189 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1190 | err = -EFAULT; | |
1191 | ||
1192 | return err; | |
1193 | } | |
1194 | ||
53d66608 O |
1195 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev) |
1196 | { | |
1197 | int err; | |
1198 | sector_t sec; | |
1199 | loff_t sz; | |
1200 | ||
1201 | err = -ENXIO; | |
1202 | if (unlikely(lo->lo_state != Lo_bound)) | |
1203 | goto out; | |
1204 | err = figure_loop_size(lo); | |
1205 | if (unlikely(err)) | |
1206 | goto out; | |
1207 | sec = get_capacity(lo->lo_disk); | |
1208 | /* the width of sector_t may be narrow for bit-shift */ | |
1209 | sz = sec; | |
1210 | sz <<= 9; | |
1211 | mutex_lock(&bdev->bd_mutex); | |
1212 | bd_set_size(bdev, sz); | |
1213 | mutex_unlock(&bdev->bd_mutex); | |
1214 | ||
1215 | out: | |
1216 | return err; | |
1217 | } | |
1218 | ||
bb214884 | 1219 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1220 | unsigned int cmd, unsigned long arg) |
1221 | { | |
bb214884 | 1222 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1223 | int err; |
1224 | ||
f028f3b2 | 1225 | mutex_lock_nested(&lo->lo_ctl_mutex, 1); |
1da177e4 LT |
1226 | switch (cmd) { |
1227 | case LOOP_SET_FD: | |
bb214884 | 1228 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1229 | break; |
1230 | case LOOP_CHANGE_FD: | |
bb214884 | 1231 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1232 | break; |
1233 | case LOOP_CLR_FD: | |
f028f3b2 | 1234 | /* loop_clr_fd would have unlocked lo_ctl_mutex on success */ |
bb214884 | 1235 | err = loop_clr_fd(lo, bdev); |
f028f3b2 NK |
1236 | if (!err) |
1237 | goto out_unlocked; | |
1da177e4 LT |
1238 | break; |
1239 | case LOOP_SET_STATUS: | |
1240 | err = loop_set_status_old(lo, (struct loop_info __user *) arg); | |
1241 | break; | |
1242 | case LOOP_GET_STATUS: | |
1243 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1244 | break; | |
1245 | case LOOP_SET_STATUS64: | |
1246 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg); | |
1247 | break; | |
1248 | case LOOP_GET_STATUS64: | |
1249 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1250 | break; | |
53d66608 O |
1251 | case LOOP_SET_CAPACITY: |
1252 | err = -EPERM; | |
1253 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) | |
1254 | err = loop_set_capacity(lo, bdev); | |
1255 | break; | |
1da177e4 LT |
1256 | default: |
1257 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1258 | } | |
f85221dd | 1259 | mutex_unlock(&lo->lo_ctl_mutex); |
f028f3b2 NK |
1260 | |
1261 | out_unlocked: | |
1da177e4 LT |
1262 | return err; |
1263 | } | |
1264 | ||
863d5b82 DH |
1265 | #ifdef CONFIG_COMPAT |
1266 | struct compat_loop_info { | |
1267 | compat_int_t lo_number; /* ioctl r/o */ | |
1268 | compat_dev_t lo_device; /* ioctl r/o */ | |
1269 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1270 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1271 | compat_int_t lo_offset; | |
1272 | compat_int_t lo_encrypt_type; | |
1273 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1274 | compat_int_t lo_flags; /* ioctl r/o */ | |
1275 | char lo_name[LO_NAME_SIZE]; | |
1276 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1277 | compat_ulong_t lo_init[2]; | |
1278 | char reserved[4]; | |
1279 | }; | |
1280 | ||
1281 | /* | |
1282 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1283 | * - noinlined to reduce stack space usage in main part of driver | |
1284 | */ | |
1285 | static noinline int | |
ba674cfc | 1286 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1287 | struct loop_info64 *info64) |
1288 | { | |
1289 | struct compat_loop_info info; | |
1290 | ||
1291 | if (copy_from_user(&info, arg, sizeof(info))) | |
1292 | return -EFAULT; | |
1293 | ||
1294 | memset(info64, 0, sizeof(*info64)); | |
1295 | info64->lo_number = info.lo_number; | |
1296 | info64->lo_device = info.lo_device; | |
1297 | info64->lo_inode = info.lo_inode; | |
1298 | info64->lo_rdevice = info.lo_rdevice; | |
1299 | info64->lo_offset = info.lo_offset; | |
1300 | info64->lo_sizelimit = 0; | |
1301 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1302 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1303 | info64->lo_flags = info.lo_flags; | |
1304 | info64->lo_init[0] = info.lo_init[0]; | |
1305 | info64->lo_init[1] = info.lo_init[1]; | |
1306 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1307 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1308 | else | |
1309 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1310 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1311 | return 0; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1316 | * - noinlined to reduce stack space usage in main part of driver | |
1317 | */ | |
1318 | static noinline int | |
1319 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1320 | struct compat_loop_info __user *arg) | |
1321 | { | |
1322 | struct compat_loop_info info; | |
1323 | ||
1324 | memset(&info, 0, sizeof(info)); | |
1325 | info.lo_number = info64->lo_number; | |
1326 | info.lo_device = info64->lo_device; | |
1327 | info.lo_inode = info64->lo_inode; | |
1328 | info.lo_rdevice = info64->lo_rdevice; | |
1329 | info.lo_offset = info64->lo_offset; | |
1330 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1331 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1332 | info.lo_flags = info64->lo_flags; | |
1333 | info.lo_init[0] = info64->lo_init[0]; | |
1334 | info.lo_init[1] = info64->lo_init[1]; | |
1335 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1336 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1337 | else | |
1338 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1339 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1340 | ||
1341 | /* error in case values were truncated */ | |
1342 | if (info.lo_device != info64->lo_device || | |
1343 | info.lo_rdevice != info64->lo_rdevice || | |
1344 | info.lo_inode != info64->lo_inode || | |
1345 | info.lo_offset != info64->lo_offset || | |
1346 | info.lo_init[0] != info64->lo_init[0] || | |
1347 | info.lo_init[1] != info64->lo_init[1]) | |
1348 | return -EOVERFLOW; | |
1349 | ||
1350 | if (copy_to_user(arg, &info, sizeof(info))) | |
1351 | return -EFAULT; | |
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static int | |
1356 | loop_set_status_compat(struct loop_device *lo, | |
1357 | const struct compat_loop_info __user *arg) | |
1358 | { | |
1359 | struct loop_info64 info64; | |
1360 | int ret; | |
1361 | ||
1362 | ret = loop_info64_from_compat(arg, &info64); | |
1363 | if (ret < 0) | |
1364 | return ret; | |
1365 | return loop_set_status(lo, &info64); | |
1366 | } | |
1367 | ||
1368 | static int | |
1369 | loop_get_status_compat(struct loop_device *lo, | |
1370 | struct compat_loop_info __user *arg) | |
1371 | { | |
1372 | struct loop_info64 info64; | |
1373 | int err = 0; | |
1374 | ||
1375 | if (!arg) | |
1376 | err = -EINVAL; | |
1377 | if (!err) | |
1378 | err = loop_get_status(lo, &info64); | |
1379 | if (!err) | |
1380 | err = loop_info64_to_compat(&info64, arg); | |
1381 | return err; | |
1382 | } | |
1383 | ||
bb214884 AV |
1384 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1385 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1386 | { |
bb214884 | 1387 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1388 | int err; |
1389 | ||
863d5b82 DH |
1390 | switch(cmd) { |
1391 | case LOOP_SET_STATUS: | |
1392 | mutex_lock(&lo->lo_ctl_mutex); | |
1393 | err = loop_set_status_compat( | |
1394 | lo, (const struct compat_loop_info __user *) arg); | |
1395 | mutex_unlock(&lo->lo_ctl_mutex); | |
1396 | break; | |
1397 | case LOOP_GET_STATUS: | |
1398 | mutex_lock(&lo->lo_ctl_mutex); | |
1399 | err = loop_get_status_compat( | |
1400 | lo, (struct compat_loop_info __user *) arg); | |
1401 | mutex_unlock(&lo->lo_ctl_mutex); | |
1402 | break; | |
53d66608 | 1403 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1404 | case LOOP_CLR_FD: |
1405 | case LOOP_GET_STATUS64: | |
1406 | case LOOP_SET_STATUS64: | |
1407 | arg = (unsigned long) compat_ptr(arg); | |
1408 | case LOOP_SET_FD: | |
1409 | case LOOP_CHANGE_FD: | |
bb214884 | 1410 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1411 | break; |
1412 | default: | |
1413 | err = -ENOIOCTLCMD; | |
1414 | break; | |
1415 | } | |
863d5b82 DH |
1416 | return err; |
1417 | } | |
1418 | #endif | |
1419 | ||
bb214884 | 1420 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1421 | { |
bb214884 | 1422 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 | 1423 | |
f85221dd | 1424 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1425 | lo->lo_refcnt++; |
f85221dd | 1426 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1427 | |
1428 | return 0; | |
1429 | } | |
1430 | ||
bb214884 | 1431 | static int lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1432 | { |
bb214884 | 1433 | struct loop_device *lo = disk->private_data; |
ffcd7dca | 1434 | int err; |
1da177e4 | 1435 | |
f85221dd | 1436 | mutex_lock(&lo->lo_ctl_mutex); |
96c58655 | 1437 | |
14f27939 MB |
1438 | if (--lo->lo_refcnt) |
1439 | goto out; | |
1440 | ||
1441 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
1442 | /* | |
1443 | * In autoclear mode, stop the loop thread | |
1444 | * and remove configuration after last close. | |
1445 | */ | |
ffcd7dca AB |
1446 | err = loop_clr_fd(lo, NULL); |
1447 | if (!err) | |
1448 | goto out_unlocked; | |
14f27939 MB |
1449 | } else { |
1450 | /* | |
1451 | * Otherwise keep thread (if running) and config, | |
1452 | * but flush possible ongoing bios in thread. | |
1453 | */ | |
1454 | loop_flush(lo); | |
1455 | } | |
96c58655 | 1456 | |
14f27939 | 1457 | out: |
f85221dd | 1458 | mutex_unlock(&lo->lo_ctl_mutex); |
ffcd7dca | 1459 | out_unlocked: |
1da177e4 LT |
1460 | return 0; |
1461 | } | |
1462 | ||
1463 | static struct block_device_operations lo_fops = { | |
1464 | .owner = THIS_MODULE, | |
bb214884 AV |
1465 | .open = lo_open, |
1466 | .release = lo_release, | |
1467 | .ioctl = lo_ioctl, | |
863d5b82 | 1468 | #ifdef CONFIG_COMPAT |
bb214884 | 1469 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1470 | #endif |
1da177e4 LT |
1471 | }; |
1472 | ||
1473 | /* | |
1474 | * And now the modules code and kernel interface. | |
1475 | */ | |
73285082 | 1476 | static int max_loop; |
1da177e4 | 1477 | module_param(max_loop, int, 0); |
a47653fc | 1478 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
476a4813 LV |
1479 | module_param(max_part, int, 0); |
1480 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); | |
1da177e4 LT |
1481 | MODULE_LICENSE("GPL"); |
1482 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1483 | ||
1484 | int loop_register_transfer(struct loop_func_table *funcs) | |
1485 | { | |
1486 | unsigned int n = funcs->number; | |
1487 | ||
1488 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1489 | return -EINVAL; | |
1490 | xfer_funcs[n] = funcs; | |
1491 | return 0; | |
1492 | } | |
1493 | ||
1494 | int loop_unregister_transfer(int number) | |
1495 | { | |
1496 | unsigned int n = number; | |
1497 | struct loop_device *lo; | |
1498 | struct loop_func_table *xfer; | |
1499 | ||
1500 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1501 | return -EINVAL; | |
1502 | ||
1503 | xfer_funcs[n] = NULL; | |
1504 | ||
73285082 | 1505 | list_for_each_entry(lo, &loop_devices, lo_list) { |
f85221dd | 1506 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1507 | |
1508 | if (lo->lo_encryption == xfer) | |
1509 | loop_release_xfer(lo); | |
1510 | ||
f85221dd | 1511 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | EXPORT_SYMBOL(loop_register_transfer); | |
1518 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1519 | ||
a47653fc | 1520 | static struct loop_device *loop_alloc(int i) |
73285082 KC |
1521 | { |
1522 | struct loop_device *lo; | |
1523 | struct gendisk *disk; | |
1524 | ||
1525 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); | |
1526 | if (!lo) | |
1527 | goto out; | |
1528 | ||
1529 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); | |
1530 | if (!lo->lo_queue) | |
1531 | goto out_free_dev; | |
1532 | ||
476a4813 | 1533 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1534 | if (!disk) |
1535 | goto out_free_queue; | |
1536 | ||
1537 | mutex_init(&lo->lo_ctl_mutex); | |
1538 | lo->lo_number = i; | |
1539 | lo->lo_thread = NULL; | |
1540 | init_waitqueue_head(&lo->lo_event); | |
1541 | spin_lock_init(&lo->lo_lock); | |
1542 | disk->major = LOOP_MAJOR; | |
476a4813 | 1543 | disk->first_minor = i << part_shift; |
73285082 KC |
1544 | disk->fops = &lo_fops; |
1545 | disk->private_data = lo; | |
1546 | disk->queue = lo->lo_queue; | |
1547 | sprintf(disk->disk_name, "loop%d", i); | |
73285082 KC |
1548 | return lo; |
1549 | ||
1550 | out_free_queue: | |
1551 | blk_cleanup_queue(lo->lo_queue); | |
1552 | out_free_dev: | |
1553 | kfree(lo); | |
1554 | out: | |
07002e99 | 1555 | return NULL; |
73285082 KC |
1556 | } |
1557 | ||
a47653fc | 1558 | static void loop_free(struct loop_device *lo) |
1da177e4 | 1559 | { |
73285082 KC |
1560 | blk_cleanup_queue(lo->lo_queue); |
1561 | put_disk(lo->lo_disk); | |
1562 | list_del(&lo->lo_list); | |
1563 | kfree(lo); | |
1564 | } | |
1da177e4 | 1565 | |
a47653fc KC |
1566 | static struct loop_device *loop_init_one(int i) |
1567 | { | |
1568 | struct loop_device *lo; | |
1569 | ||
1570 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1571 | if (lo->lo_number == i) | |
1572 | return lo; | |
1573 | } | |
1574 | ||
1575 | lo = loop_alloc(i); | |
1576 | if (lo) { | |
1577 | add_disk(lo->lo_disk); | |
1578 | list_add_tail(&lo->lo_list, &loop_devices); | |
1579 | } | |
1580 | return lo; | |
1581 | } | |
1582 | ||
1583 | static void loop_del_one(struct loop_device *lo) | |
1584 | { | |
1585 | del_gendisk(lo->lo_disk); | |
1586 | loop_free(lo); | |
1587 | } | |
1588 | ||
73285082 KC |
1589 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1590 | { | |
705962cc | 1591 | struct loop_device *lo; |
07002e99 | 1592 | struct kobject *kobj; |
73285082 | 1593 | |
705962cc AV |
1594 | mutex_lock(&loop_devices_mutex); |
1595 | lo = loop_init_one(dev & MINORMASK); | |
07002e99 | 1596 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM); |
73285082 KC |
1597 | mutex_unlock(&loop_devices_mutex); |
1598 | ||
1599 | *part = 0; | |
07002e99 | 1600 | return kobj; |
73285082 KC |
1601 | } |
1602 | ||
1603 | static int __init loop_init(void) | |
1604 | { | |
a47653fc KC |
1605 | int i, nr; |
1606 | unsigned long range; | |
1607 | struct loop_device *lo, *next; | |
1608 | ||
1609 | /* | |
1610 | * loop module now has a feature to instantiate underlying device | |
1611 | * structure on-demand, provided that there is an access dev node. | |
1612 | * However, this will not work well with user space tool that doesn't | |
1613 | * know about such "feature". In order to not break any existing | |
1614 | * tool, we do the following: | |
1615 | * | |
1616 | * (1) if max_loop is specified, create that many upfront, and this | |
1617 | * also becomes a hard limit. | |
1618 | * (2) if max_loop is not specified, create 8 loop device on module | |
1619 | * load, user can further extend loop device by create dev node | |
1620 | * themselves and have kernel automatically instantiate actual | |
1621 | * device on-demand. | |
1622 | */ | |
476a4813 LV |
1623 | |
1624 | part_shift = 0; | |
1625 | if (max_part > 0) | |
1626 | part_shift = fls(max_part); | |
1627 | ||
1628 | if (max_loop > 1UL << (MINORBITS - part_shift)) | |
a47653fc | 1629 | return -EINVAL; |
1da177e4 | 1630 | |
73285082 | 1631 | if (max_loop) { |
a47653fc KC |
1632 | nr = max_loop; |
1633 | range = max_loop; | |
1634 | } else { | |
1635 | nr = 8; | |
476a4813 | 1636 | range = 1UL << (MINORBITS - part_shift); |
a47653fc KC |
1637 | } |
1638 | ||
1639 | if (register_blkdev(LOOP_MAJOR, "loop")) | |
1640 | return -EIO; | |
1da177e4 | 1641 | |
a47653fc KC |
1642 | for (i = 0; i < nr; i++) { |
1643 | lo = loop_alloc(i); | |
1644 | if (!lo) | |
1645 | goto Enomem; | |
1646 | list_add_tail(&lo->lo_list, &loop_devices); | |
1da177e4 | 1647 | } |
a47653fc KC |
1648 | |
1649 | /* point of no return */ | |
1650 | ||
1651 | list_for_each_entry(lo, &loop_devices, lo_list) | |
1652 | add_disk(lo->lo_disk); | |
1653 | ||
1654 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, | |
1655 | THIS_MODULE, loop_probe, NULL, NULL); | |
1656 | ||
73285082 | 1657 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 1658 | return 0; |
a47653fc KC |
1659 | |
1660 | Enomem: | |
1661 | printk(KERN_INFO "loop: out of memory\n"); | |
1662 | ||
1663 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
1664 | loop_free(lo); | |
1665 | ||
1666 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
1667 | return -ENOMEM; | |
1da177e4 LT |
1668 | } |
1669 | ||
73285082 | 1670 | static void __exit loop_exit(void) |
1da177e4 | 1671 | { |
a47653fc | 1672 | unsigned long range; |
73285082 | 1673 | struct loop_device *lo, *next; |
1da177e4 | 1674 | |
476a4813 | 1675 | range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift); |
a47653fc | 1676 | |
73285082 KC |
1677 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) |
1678 | loop_del_one(lo); | |
1679 | ||
a47653fc | 1680 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 1681 | unregister_blkdev(LOOP_MAJOR, "loop"); |
1da177e4 LT |
1682 | } |
1683 | ||
1684 | module_init(loop_init); | |
1685 | module_exit(loop_exit); | |
1686 | ||
1687 | #ifndef MODULE | |
1688 | static int __init max_loop_setup(char *str) | |
1689 | { | |
1690 | max_loop = simple_strtol(str, NULL, 0); | |
1691 | return 1; | |
1692 | } | |
1693 | ||
1694 | __setup("max_loop=", max_loop_setup); | |
1695 | #endif |