blob: 06eaa11cbc2fdb6f298ef7cd88708adb2f059898 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 prepare_write and/or commit_write are not available on the
44 * backing filesystem.
45 * Anton Altaparmakov, 16 Feb 2005
46 *
47 * Still To Fix:
48 * - Advisory locking is ignored here.
49 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
50 *
51 */
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/module.h>
54#include <linux/moduleparam.h>
55#include <linux/sched.h>
56#include <linux/fs.h>
57#include <linux/file.h>
58#include <linux/stat.h>
59#include <linux/errno.h>
60#include <linux/major.h>
61#include <linux/wait.h>
62#include <linux/blkdev.h>
63#include <linux/blkpg.h>
64#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/smp_lock.h>
66#include <linux/swap.h>
67#include <linux/slab.h>
68#include <linux/loop.h>
David Howells863d5b822006-08-29 19:06:14 +010069#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070071#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/writeback.h>
73#include <linux/buffer_head.h> /* for invalidate_bdev() */
74#include <linux/completion.h>
75#include <linux/highmem.h>
76#include <linux/gfp.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070077#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020078#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80#include <asm/uaccess.h>
81
Ken Chen73285082007-05-08 00:28:20 -070082static LIST_HEAD(loop_devices);
83static DEFINE_MUTEX(loop_devices_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85/*
86 * Transfer functions
87 */
88static int transfer_none(struct loop_device *lo, int cmd,
89 struct page *raw_page, unsigned raw_off,
90 struct page *loop_page, unsigned loop_off,
91 int size, sector_t real_block)
92{
93 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
94 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
95
96 if (cmd == READ)
97 memcpy(loop_buf, raw_buf, size);
98 else
99 memcpy(raw_buf, loop_buf, size);
100
101 kunmap_atomic(raw_buf, KM_USER0);
102 kunmap_atomic(loop_buf, KM_USER1);
103 cond_resched();
104 return 0;
105}
106
107static int transfer_xor(struct loop_device *lo, int cmd,
108 struct page *raw_page, unsigned raw_off,
109 struct page *loop_page, unsigned loop_off,
110 int size, sector_t real_block)
111{
112 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
113 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
114 char *in, *out, *key;
115 int i, keysize;
116
117 if (cmd == READ) {
118 in = raw_buf;
119 out = loop_buf;
120 } else {
121 in = loop_buf;
122 out = raw_buf;
123 }
124
125 key = lo->lo_encrypt_key;
126 keysize = lo->lo_encrypt_key_size;
127 for (i = 0; i < size; i++)
128 *out++ = *in++ ^ key[(i & 511) % keysize];
129
130 kunmap_atomic(raw_buf, KM_USER0);
131 kunmap_atomic(loop_buf, KM_USER1);
132 cond_resched();
133 return 0;
134}
135
136static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
137{
138 if (unlikely(info->lo_encrypt_key_size <= 0))
139 return -EINVAL;
140 return 0;
141}
142
143static struct loop_func_table none_funcs = {
144 .number = LO_CRYPT_NONE,
145 .transfer = transfer_none,
146};
147
148static struct loop_func_table xor_funcs = {
149 .number = LO_CRYPT_XOR,
150 .transfer = transfer_xor,
151 .init = xor_init
152};
153
154/* xfer_funcs[0] is special - its release function is never called */
155static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
156 &none_funcs,
157 &xor_funcs
158};
159
160static loff_t get_loop_size(struct loop_device *lo, struct file *file)
161{
162 loff_t size, offset, loopsize;
163
164 /* Compute loopsize in bytes */
165 size = i_size_read(file->f_mapping->host);
166 offset = lo->lo_offset;
167 loopsize = size - offset;
168 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
169 loopsize = lo->lo_sizelimit;
170
171 /*
172 * Unfortunately, if we want to do I/O on the device,
173 * the number of 512-byte sectors has to fit into a sector_t.
174 */
175 return loopsize >> 9;
176}
177
178static int
179figure_loop_size(struct loop_device *lo)
180{
181 loff_t size = get_loop_size(lo, lo->lo_backing_file);
182 sector_t x = (sector_t)size;
183
184 if (unlikely((loff_t)x != size))
185 return -EFBIG;
186
Ken Chen73285082007-05-08 00:28:20 -0700187 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 return 0;
189}
190
191static inline int
192lo_do_transfer(struct loop_device *lo, int cmd,
193 struct page *rpage, unsigned roffs,
194 struct page *lpage, unsigned loffs,
195 int size, sector_t rblock)
196{
197 if (unlikely(!lo->transfer))
198 return 0;
199
200 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
201}
202
203/**
204 * do_lo_send_aops - helper for writing data to a loop device
205 *
206 * This is the fast version for backing filesystems which implement the address
207 * space operations prepare_write and commit_write.
208 */
209static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
210 int bsize, loff_t pos, struct page *page)
211{
212 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
213 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700214 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 pgoff_t index;
216 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800217 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800219 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 index = pos >> PAGE_CACHE_SHIFT;
221 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
222 bv_offs = bvec->bv_offset;
223 len = bvec->bv_len;
224 while (len > 0) {
225 sector_t IV;
226 unsigned size;
227 int transfer_result;
228
229 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
230 size = PAGE_CACHE_SIZE - offset;
231 if (size > len)
232 size = len;
233 page = grab_cache_page(mapping, index);
234 if (unlikely(!page))
235 goto fail;
Zach Brown994fc28c2005-12-15 14:28:17 -0800236 ret = aops->prepare_write(file, page, offset,
237 offset + size);
238 if (unlikely(ret)) {
239 if (ret == AOP_TRUNCATED_PAGE) {
240 page_cache_release(page);
241 continue;
242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 goto unlock;
Zach Brown994fc28c2005-12-15 14:28:17 -0800244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
246 bvec->bv_page, bv_offs, size, IV);
247 if (unlikely(transfer_result)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 /*
249 * The transfer failed, but we still write the data to
250 * keep prepare/commit calls balanced.
251 */
252 printk(KERN_ERR "loop: transfer error block %llu\n",
253 (unsigned long long)index);
Nate Diller01f2705d2007-05-09 02:35:07 -0700254 zero_user_page(page, offset, size, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 }
256 flush_dcache_page(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800257 ret = aops->commit_write(file, page, offset,
258 offset + size);
259 if (unlikely(ret)) {
260 if (ret == AOP_TRUNCATED_PAGE) {
261 page_cache_release(page);
262 continue;
263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 goto unlock;
Zach Brown994fc28c2005-12-15 14:28:17 -0800265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 if (unlikely(transfer_result))
267 goto unlock;
268 bv_offs += size;
269 len -= size;
270 offset = 0;
271 index++;
272 pos += size;
273 unlock_page(page);
274 page_cache_release(page);
275 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800276 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800278 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return ret;
280unlock:
281 unlock_page(page);
282 page_cache_release(page);
283fail:
284 ret = -1;
285 goto out;
286}
287
288/**
289 * __do_lo_send_write - helper for writing data to a loop device
290 *
291 * This helper just factors out common code between do_lo_send_direct_write()
292 * and do_lo_send_write().
293 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800294static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100295 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 ssize_t bw;
298 mm_segment_t old_fs = get_fs();
299
300 set_fs(get_ds());
301 bw = file->f_op->write(file, buf, len, &pos);
302 set_fs(old_fs);
303 if (likely(bw == len))
304 return 0;
305 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
306 (unsigned long long)pos, len);
307 if (bw >= 0)
308 bw = -EIO;
309 return bw;
310}
311
312/**
313 * do_lo_send_direct_write - helper for writing data to a loop device
314 *
315 * This is the fast, non-transforming version for backing filesystems which do
316 * not implement the address space operations prepare_write and commit_write.
317 * It uses the write file operation which should be present on all writeable
318 * filesystems.
319 */
320static int do_lo_send_direct_write(struct loop_device *lo,
321 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
322{
323 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100324 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 bvec->bv_len, pos);
326 kunmap(bvec->bv_page);
327 cond_resched();
328 return bw;
329}
330
331/**
332 * do_lo_send_write - helper for writing data to a loop device
333 *
334 * This is the slow, transforming version for filesystems which do not
335 * implement the address space operations prepare_write and commit_write. It
336 * uses the write file operation which should be present on all writeable
337 * filesystems.
338 *
339 * Using fops->write is slower than using aops->{prepare,commit}_write in the
340 * transforming case because we need to double buffer the data as we cannot do
341 * the transformations in place as we do not have direct access to the
342 * destination pages of the backing file.
343 */
344static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
345 int bsize, loff_t pos, struct page *page)
346{
347 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
348 bvec->bv_offset, bvec->bv_len, pos >> 9);
349 if (likely(!ret))
350 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100351 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 pos);
353 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
354 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
355 if (ret > 0)
356 ret = -EIO;
357 return ret;
358}
359
360static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
361 loff_t pos)
362{
363 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
364 struct page *page);
365 struct bio_vec *bvec;
366 struct page *page = NULL;
367 int i, ret = 0;
368
369 do_lo_send = do_lo_send_aops;
370 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
371 do_lo_send = do_lo_send_direct_write;
372 if (lo->transfer != transfer_none) {
373 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
374 if (unlikely(!page))
375 goto fail;
376 kmap(page);
377 do_lo_send = do_lo_send_write;
378 }
379 }
380 bio_for_each_segment(bvec, bio, i) {
381 ret = do_lo_send(lo, bvec, bsize, pos, page);
382 if (ret < 0)
383 break;
384 pos += bvec->bv_len;
385 }
386 if (page) {
387 kunmap(page);
388 __free_page(page);
389 }
390out:
391 return ret;
392fail:
393 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
394 ret = -ENOMEM;
395 goto out;
396}
397
398struct lo_read_data {
399 struct loop_device *lo;
400 struct page *page;
401 unsigned offset;
402 int bsize;
403};
404
405static int
Jens Axboefd582142007-06-12 21:20:37 +0200406lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
407 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
Jens Axboefd582142007-06-12 21:20:37 +0200409 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200411 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 sector_t IV;
Jens Axboefd582142007-06-12 21:20:37 +0200413 size_t size;
414 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Jens Axboecac36bb02007-06-14 13:10:48 +0200416 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200417 if (unlikely(ret))
418 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Jens Axboefd582142007-06-12 21:20:37 +0200420 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
421 (buf->offset >> 9);
422 size = sd->len;
423 if (size > p->bsize)
424 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Jens Axboefd582142007-06-12 21:20:37 +0200426 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 printk(KERN_ERR "loop: transfer error block %ld\n",
428 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200429 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
431
432 flush_dcache_page(p->page);
433
Jens Axboefd582142007-06-12 21:20:37 +0200434 if (size > 0)
435 p->offset += size;
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return size;
438}
439
440static int
Jens Axboefd582142007-06-12 21:20:37 +0200441lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
442{
443 return __splice_from_pipe(pipe, sd, lo_splice_actor);
444}
445
446static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447do_lo_receive(struct loop_device *lo,
448 struct bio_vec *bvec, int bsize, loff_t pos)
449{
450 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200451 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200453 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 cookie.lo = lo;
456 cookie.page = bvec->bv_page;
457 cookie.offset = bvec->bv_offset;
458 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200459
460 sd.len = 0;
461 sd.total_len = bvec->bv_len;
462 sd.flags = 0;
463 sd.pos = pos;
464 sd.u.data = &cookie;
465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200467 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
468
469 if (retval < 0)
470 return retval;
471
472 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475static int
476lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
477{
478 struct bio_vec *bvec;
479 int i, ret = 0;
480
481 bio_for_each_segment(bvec, bio, i) {
482 ret = do_lo_receive(lo, bvec, bsize, pos);
483 if (ret < 0)
484 break;
485 pos += bvec->bv_len;
486 }
487 return ret;
488}
489
490static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
491{
492 loff_t pos;
493 int ret;
494
495 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
496 if (bio_rw(bio) == WRITE)
497 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
498 else
499 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
500 return ret;
501}
502
503/*
504 * Add bio to back of pending list
505 */
506static void loop_add_bio(struct loop_device *lo, struct bio *bio)
507{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (lo->lo_biotail) {
509 lo->lo_biotail->bi_next = bio;
510 lo->lo_biotail = bio;
511 } else
512 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515/*
516 * Grab first pending buffer
517 */
518static struct bio *loop_get_bio(struct loop_device *lo)
519{
520 struct bio *bio;
521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if ((bio = lo->lo_bio)) {
523 if (bio == lo->lo_biotail)
524 lo->lo_biotail = NULL;
525 lo->lo_bio = bio->bi_next;
526 bio->bi_next = NULL;
527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
529 return bio;
530}
531
532static int loop_make_request(request_queue_t *q, struct bio *old_bio)
533{
534 struct loop_device *lo = q->queuedata;
535 int rw = bio_rw(old_bio);
536
Nick Piggin35a82d12005-06-23 00:09:06 -0700537 if (rw == READA)
538 rw = READ;
539
540 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 spin_lock_irq(&lo->lo_lock);
543 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700544 goto out;
545 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
546 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700548 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700549 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700553 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 bio_io_error(old_bio, old_bio->bi_size);
555 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
558/*
559 * kick off io on the underlying address space
560 */
561static void loop_unplug(request_queue_t *q)
562{
563 struct loop_device *lo = q->queuedata;
564
565 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
566 blk_run_address_space(lo->lo_backing_file->f_mapping);
567}
568
569struct switch_request {
570 struct file *file;
571 struct completion wait;
572};
573
574static void do_loop_switch(struct loop_device *, struct switch_request *);
575
576static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
577{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (unlikely(!bio->bi_bdev)) {
579 do_loop_switch(lo, bio->bi_private);
580 bio_put(bio);
581 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700582 int ret = do_bio_filebacked(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 bio_endio(bio, bio->bi_size, ret);
584 }
585}
586
587/*
588 * worker thread that handles reads/writes to file backed loop devices,
589 * to avoid blocking in our make_request_fn. it also does loop decrypting
590 * on reads for block backed loop, as that is too heavy to do from
591 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700592 *
593 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
594 * calling kthread_stop(). Therefore once kthread_should_stop() is
595 * true, make_request will not place any more requests. Therefore
596 * once kthread_should_stop() is true and lo_bio is NULL, we are
597 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 */
599static int loop_thread(void *data)
600{
601 struct loop_device *lo = data;
602 struct bio *bio;
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 set_user_nice(current, -20);
605
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700606 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700608 wait_event_interruptible(lo->lo_event,
609 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700610
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700611 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700612 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700613 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700615 spin_unlock_irq(&lo->lo_lock);
616
617 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return 0;
622}
623
624/*
625 * loop_switch performs the hard work of switching a backing store.
626 * First it needs to flush existing IO, it does this by sending a magic
627 * BIO down the pipe. The completion of this BIO does the actual switch.
628 */
629static int loop_switch(struct loop_device *lo, struct file *file)
630{
631 struct switch_request w;
632 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
633 if (!bio)
634 return -ENOMEM;
635 init_completion(&w.wait);
636 w.file = file;
637 bio->bi_private = &w;
638 bio->bi_bdev = NULL;
639 loop_make_request(lo->lo_queue, bio);
640 wait_for_completion(&w.wait);
641 return 0;
642}
643
644/*
645 * Do the actual switch; called from the BIO completion routine
646 */
647static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
648{
649 struct file *file = p->file;
650 struct file *old_file = lo->lo_backing_file;
651 struct address_space *mapping = file->f_mapping;
652
653 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
654 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700655 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
656 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 lo->old_gfp_mask = mapping_gfp_mask(mapping);
658 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
659 complete(&p->wait);
660}
661
662
663/*
664 * loop_change_fd switched the backing store of a loopback device to
665 * a new file. This is useful for operating system installers to free up
666 * the original file and in High Availability environments to switch to
667 * an alternative location for the content in case of server meltdown.
668 * This can only work if the loop device is used read-only, and if the
669 * new backing store is the same size and type as the old backing store.
670 */
671static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
672 struct block_device *bdev, unsigned int arg)
673{
674 struct file *file, *old_file;
675 struct inode *inode;
676 int error;
677
678 error = -ENXIO;
679 if (lo->lo_state != Lo_bound)
680 goto out;
681
682 /* the loop device has to be read-only */
683 error = -EINVAL;
684 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
685 goto out;
686
687 error = -EBADF;
688 file = fget(arg);
689 if (!file)
690 goto out;
691
692 inode = file->f_mapping->host;
693 old_file = lo->lo_backing_file;
694
695 error = -EINVAL;
696
697 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
698 goto out_putf;
699
Jens Axboefd582142007-06-12 21:20:37 +0200700 /* new backing store needs to support loop (eg splice_read) */
701 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 goto out_putf;
703
704 /* size of the new backing store needs to be the same */
705 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
706 goto out_putf;
707
708 /* and ... switch */
709 error = loop_switch(lo, file);
710 if (error)
711 goto out_putf;
712
713 fput(old_file);
714 return 0;
715
716 out_putf:
717 fput(file);
718 out:
719 return error;
720}
721
722static inline int is_loop_device(struct file *file)
723{
724 struct inode *i = file->f_mapping->host;
725
726 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
727}
728
729static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
730 struct block_device *bdev, unsigned int arg)
731{
732 struct file *file, *f;
733 struct inode *inode;
734 struct address_space *mapping;
735 unsigned lo_blocksize;
736 int lo_flags = 0;
737 int error;
738 loff_t size;
739
740 /* This is safe, since we have a reference from open(). */
741 __module_get(THIS_MODULE);
742
743 error = -EBADF;
744 file = fget(arg);
745 if (!file)
746 goto out;
747
748 error = -EBUSY;
749 if (lo->lo_state != Lo_unbound)
750 goto out_putf;
751
752 /* Avoid recursion */
753 f = file;
754 while (is_loop_device(f)) {
755 struct loop_device *l;
756
757 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
758 goto out_putf;
759
760 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
761 if (l->lo_state == Lo_unbound) {
762 error = -EINVAL;
763 goto out_putf;
764 }
765 f = l->lo_backing_file;
766 }
767
768 mapping = file->f_mapping;
769 inode = mapping->host;
770
771 if (!(file->f_mode & FMODE_WRITE))
772 lo_flags |= LO_FLAGS_READ_ONLY;
773
774 error = -EINVAL;
775 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700776 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 /*
778 * If we can't read - sorry. If we only can't write - well,
779 * it's going to be read-only.
780 */
Jens Axboefd582142007-06-12 21:20:37 +0200781 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 goto out_putf;
783 if (aops->prepare_write && aops->commit_write)
784 lo_flags |= LO_FLAGS_USE_AOPS;
785 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
786 lo_flags |= LO_FLAGS_READ_ONLY;
787
Theodore Ts'oba52de12006-09-27 01:50:49 -0700788 lo_blocksize = S_ISBLK(inode->i_mode) ?
789 inode->i_bdev->bd_block_size : PAGE_SIZE;
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 error = 0;
792 } else {
793 goto out_putf;
794 }
795
796 size = get_loop_size(lo, file);
797
798 if ((loff_t)(sector_t)size != size) {
799 error = -EFBIG;
800 goto out_putf;
801 }
802
803 if (!(lo_file->f_mode & FMODE_WRITE))
804 lo_flags |= LO_FLAGS_READ_ONLY;
805
806 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
807
808 lo->lo_blocksize = lo_blocksize;
809 lo->lo_device = bdev;
810 lo->lo_flags = lo_flags;
811 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700812 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 lo->ioctl = NULL;
814 lo->lo_sizelimit = 0;
815 lo->old_gfp_mask = mapping_gfp_mask(mapping);
816 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
817
818 lo->lo_bio = lo->lo_biotail = NULL;
819
820 /*
821 * set queue make_request_fn, and add limits based on lower level
822 * device
823 */
824 blk_queue_make_request(lo->lo_queue, loop_make_request);
825 lo->lo_queue->queuedata = lo;
826 lo->lo_queue->unplug_fn = loop_unplug;
827
Ken Chen73285082007-05-08 00:28:20 -0700828 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 bd_set_size(bdev, size << 9);
830
831 set_blocksize(bdev, lo_blocksize);
832
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700833 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
834 lo->lo_number);
835 if (IS_ERR(lo->lo_thread)) {
836 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700837 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700838 }
839 lo->lo_state = Lo_bound;
840 wake_up_process(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 return 0;
842
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700843out_clr:
844 lo->lo_thread = NULL;
845 lo->lo_device = NULL;
846 lo->lo_backing_file = NULL;
847 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700848 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700849 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700850 bd_set_size(bdev, 0);
851 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
852 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 out_putf:
854 fput(file);
855 out:
856 /* This is safe: open() is still holding a reference. */
857 module_put(THIS_MODULE);
858 return error;
859}
860
861static int
862loop_release_xfer(struct loop_device *lo)
863{
864 int err = 0;
865 struct loop_func_table *xfer = lo->lo_encryption;
866
867 if (xfer) {
868 if (xfer->release)
869 err = xfer->release(lo);
870 lo->transfer = NULL;
871 lo->lo_encryption = NULL;
872 module_put(xfer->owner);
873 }
874 return err;
875}
876
877static int
878loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
879 const struct loop_info64 *i)
880{
881 int err = 0;
882
883 if (xfer) {
884 struct module *owner = xfer->owner;
885
886 if (!try_module_get(owner))
887 return -EINVAL;
888 if (xfer->init)
889 err = xfer->init(lo, i);
890 if (err)
891 module_put(owner);
892 else
893 lo->lo_encryption = xfer;
894 }
895 return err;
896}
897
898static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
899{
900 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400901 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 if (lo->lo_state != Lo_bound)
904 return -ENXIO;
905
906 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
907 return -EBUSY;
908
909 if (filp == NULL)
910 return -EINVAL;
911
912 spin_lock_irq(&lo->lo_lock);
913 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 spin_unlock_irq(&lo->lo_lock);
915
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700916 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
918 lo->lo_backing_file = NULL;
919
920 loop_release_xfer(lo);
921 lo->transfer = NULL;
922 lo->ioctl = NULL;
923 lo->lo_device = NULL;
924 lo->lo_encryption = NULL;
925 lo->lo_offset = 0;
926 lo->lo_sizelimit = 0;
927 lo->lo_encrypt_key_size = 0;
928 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700929 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
931 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
932 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700933 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700934 set_capacity(lo->lo_disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 bd_set_size(bdev, 0);
936 mapping_set_gfp_mask(filp->f_mapping, gfp);
937 lo->lo_state = Lo_unbound;
938 fput(filp);
939 /* This is safe: open() is still holding a reference. */
940 module_put(THIS_MODULE);
941 return 0;
942}
943
944static int
945loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
946{
947 int err;
948 struct loop_func_table *xfer;
949
950 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
951 !capable(CAP_SYS_ADMIN))
952 return -EPERM;
953 if (lo->lo_state != Lo_bound)
954 return -ENXIO;
955 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
956 return -EINVAL;
957
958 err = loop_release_xfer(lo);
959 if (err)
960 return err;
961
962 if (info->lo_encrypt_type) {
963 unsigned int type = info->lo_encrypt_type;
964
965 if (type >= MAX_LO_CRYPT)
966 return -EINVAL;
967 xfer = xfer_funcs[type];
968 if (xfer == NULL)
969 return -EINVAL;
970 } else
971 xfer = NULL;
972
973 err = loop_init_xfer(lo, xfer, info);
974 if (err)
975 return err;
976
977 if (lo->lo_offset != info->lo_offset ||
978 lo->lo_sizelimit != info->lo_sizelimit) {
979 lo->lo_offset = info->lo_offset;
980 lo->lo_sizelimit = info->lo_sizelimit;
981 if (figure_loop_size(lo))
982 return -EFBIG;
983 }
984
985 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
986 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
987 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
988 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
989
990 if (!xfer)
991 xfer = &none_funcs;
992 lo->transfer = xfer->transfer;
993 lo->ioctl = xfer->ioctl;
994
995 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
996 lo->lo_init[0] = info->lo_init[0];
997 lo->lo_init[1] = info->lo_init[1];
998 if (info->lo_encrypt_key_size) {
999 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1000 info->lo_encrypt_key_size);
1001 lo->lo_key_owner = current->uid;
1002 }
1003
1004 return 0;
1005}
1006
1007static int
1008loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1009{
1010 struct file *file = lo->lo_backing_file;
1011 struct kstat stat;
1012 int error;
1013
1014 if (lo->lo_state != Lo_bound)
1015 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001016 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 if (error)
1018 return error;
1019 memset(info, 0, sizeof(*info));
1020 info->lo_number = lo->lo_number;
1021 info->lo_device = huge_encode_dev(stat.dev);
1022 info->lo_inode = stat.ino;
1023 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1024 info->lo_offset = lo->lo_offset;
1025 info->lo_sizelimit = lo->lo_sizelimit;
1026 info->lo_flags = lo->lo_flags;
1027 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1028 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1029 info->lo_encrypt_type =
1030 lo->lo_encryption ? lo->lo_encryption->number : 0;
1031 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1032 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1033 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1034 lo->lo_encrypt_key_size);
1035 }
1036 return 0;
1037}
1038
1039static void
1040loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1041{
1042 memset(info64, 0, sizeof(*info64));
1043 info64->lo_number = info->lo_number;
1044 info64->lo_device = info->lo_device;
1045 info64->lo_inode = info->lo_inode;
1046 info64->lo_rdevice = info->lo_rdevice;
1047 info64->lo_offset = info->lo_offset;
1048 info64->lo_sizelimit = 0;
1049 info64->lo_encrypt_type = info->lo_encrypt_type;
1050 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1051 info64->lo_flags = info->lo_flags;
1052 info64->lo_init[0] = info->lo_init[0];
1053 info64->lo_init[1] = info->lo_init[1];
1054 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1055 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1056 else
1057 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1058 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1059}
1060
1061static int
1062loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1063{
1064 memset(info, 0, sizeof(*info));
1065 info->lo_number = info64->lo_number;
1066 info->lo_device = info64->lo_device;
1067 info->lo_inode = info64->lo_inode;
1068 info->lo_rdevice = info64->lo_rdevice;
1069 info->lo_offset = info64->lo_offset;
1070 info->lo_encrypt_type = info64->lo_encrypt_type;
1071 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1072 info->lo_flags = info64->lo_flags;
1073 info->lo_init[0] = info64->lo_init[0];
1074 info->lo_init[1] = info64->lo_init[1];
1075 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1076 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1077 else
1078 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1079 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1080
1081 /* error in case values were truncated */
1082 if (info->lo_device != info64->lo_device ||
1083 info->lo_rdevice != info64->lo_rdevice ||
1084 info->lo_inode != info64->lo_inode ||
1085 info->lo_offset != info64->lo_offset)
1086 return -EOVERFLOW;
1087
1088 return 0;
1089}
1090
1091static int
1092loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1093{
1094 struct loop_info info;
1095 struct loop_info64 info64;
1096
1097 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1098 return -EFAULT;
1099 loop_info64_from_old(&info, &info64);
1100 return loop_set_status(lo, &info64);
1101}
1102
1103static int
1104loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1105{
1106 struct loop_info64 info64;
1107
1108 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1109 return -EFAULT;
1110 return loop_set_status(lo, &info64);
1111}
1112
1113static int
1114loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1115 struct loop_info info;
1116 struct loop_info64 info64;
1117 int err = 0;
1118
1119 if (!arg)
1120 err = -EINVAL;
1121 if (!err)
1122 err = loop_get_status(lo, &info64);
1123 if (!err)
1124 err = loop_info64_to_old(&info64, &info);
1125 if (!err && copy_to_user(arg, &info, sizeof(info)))
1126 err = -EFAULT;
1127
1128 return err;
1129}
1130
1131static int
1132loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1133 struct loop_info64 info64;
1134 int err = 0;
1135
1136 if (!arg)
1137 err = -EINVAL;
1138 if (!err)
1139 err = loop_get_status(lo, &info64);
1140 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1141 err = -EFAULT;
1142
1143 return err;
1144}
1145
1146static int lo_ioctl(struct inode * inode, struct file * file,
1147 unsigned int cmd, unsigned long arg)
1148{
1149 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1150 int err;
1151
Ingo Molnarf85221d2006-03-23 03:00:38 -08001152 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 switch (cmd) {
1154 case LOOP_SET_FD:
1155 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1156 break;
1157 case LOOP_CHANGE_FD:
1158 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1159 break;
1160 case LOOP_CLR_FD:
1161 err = loop_clr_fd(lo, inode->i_bdev);
1162 break;
1163 case LOOP_SET_STATUS:
1164 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1165 break;
1166 case LOOP_GET_STATUS:
1167 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1168 break;
1169 case LOOP_SET_STATUS64:
1170 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1171 break;
1172 case LOOP_GET_STATUS64:
1173 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1174 break;
1175 default:
1176 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1177 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001178 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return err;
1180}
1181
David Howells863d5b822006-08-29 19:06:14 +01001182#ifdef CONFIG_COMPAT
1183struct compat_loop_info {
1184 compat_int_t lo_number; /* ioctl r/o */
1185 compat_dev_t lo_device; /* ioctl r/o */
1186 compat_ulong_t lo_inode; /* ioctl r/o */
1187 compat_dev_t lo_rdevice; /* ioctl r/o */
1188 compat_int_t lo_offset;
1189 compat_int_t lo_encrypt_type;
1190 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1191 compat_int_t lo_flags; /* ioctl r/o */
1192 char lo_name[LO_NAME_SIZE];
1193 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1194 compat_ulong_t lo_init[2];
1195 char reserved[4];
1196};
1197
1198/*
1199 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1200 * - noinlined to reduce stack space usage in main part of driver
1201 */
1202static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001203loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001204 struct loop_info64 *info64)
1205{
1206 struct compat_loop_info info;
1207
1208 if (copy_from_user(&info, arg, sizeof(info)))
1209 return -EFAULT;
1210
1211 memset(info64, 0, sizeof(*info64));
1212 info64->lo_number = info.lo_number;
1213 info64->lo_device = info.lo_device;
1214 info64->lo_inode = info.lo_inode;
1215 info64->lo_rdevice = info.lo_rdevice;
1216 info64->lo_offset = info.lo_offset;
1217 info64->lo_sizelimit = 0;
1218 info64->lo_encrypt_type = info.lo_encrypt_type;
1219 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1220 info64->lo_flags = info.lo_flags;
1221 info64->lo_init[0] = info.lo_init[0];
1222 info64->lo_init[1] = info.lo_init[1];
1223 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1224 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1225 else
1226 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1227 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1228 return 0;
1229}
1230
1231/*
1232 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1233 * - noinlined to reduce stack space usage in main part of driver
1234 */
1235static noinline int
1236loop_info64_to_compat(const struct loop_info64 *info64,
1237 struct compat_loop_info __user *arg)
1238{
1239 struct compat_loop_info info;
1240
1241 memset(&info, 0, sizeof(info));
1242 info.lo_number = info64->lo_number;
1243 info.lo_device = info64->lo_device;
1244 info.lo_inode = info64->lo_inode;
1245 info.lo_rdevice = info64->lo_rdevice;
1246 info.lo_offset = info64->lo_offset;
1247 info.lo_encrypt_type = info64->lo_encrypt_type;
1248 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1249 info.lo_flags = info64->lo_flags;
1250 info.lo_init[0] = info64->lo_init[0];
1251 info.lo_init[1] = info64->lo_init[1];
1252 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1253 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1254 else
1255 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1256 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1257
1258 /* error in case values were truncated */
1259 if (info.lo_device != info64->lo_device ||
1260 info.lo_rdevice != info64->lo_rdevice ||
1261 info.lo_inode != info64->lo_inode ||
1262 info.lo_offset != info64->lo_offset ||
1263 info.lo_init[0] != info64->lo_init[0] ||
1264 info.lo_init[1] != info64->lo_init[1])
1265 return -EOVERFLOW;
1266
1267 if (copy_to_user(arg, &info, sizeof(info)))
1268 return -EFAULT;
1269 return 0;
1270}
1271
1272static int
1273loop_set_status_compat(struct loop_device *lo,
1274 const struct compat_loop_info __user *arg)
1275{
1276 struct loop_info64 info64;
1277 int ret;
1278
1279 ret = loop_info64_from_compat(arg, &info64);
1280 if (ret < 0)
1281 return ret;
1282 return loop_set_status(lo, &info64);
1283}
1284
1285static int
1286loop_get_status_compat(struct loop_device *lo,
1287 struct compat_loop_info __user *arg)
1288{
1289 struct loop_info64 info64;
1290 int err = 0;
1291
1292 if (!arg)
1293 err = -EINVAL;
1294 if (!err)
1295 err = loop_get_status(lo, &info64);
1296 if (!err)
1297 err = loop_info64_to_compat(&info64, arg);
1298 return err;
1299}
1300
1301static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1302{
Josef Sipek6c648be2006-12-08 02:36:55 -08001303 struct inode *inode = file->f_path.dentry->d_inode;
David Howells863d5b822006-08-29 19:06:14 +01001304 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1305 int err;
1306
1307 lock_kernel();
1308 switch(cmd) {
1309 case LOOP_SET_STATUS:
1310 mutex_lock(&lo->lo_ctl_mutex);
1311 err = loop_set_status_compat(
1312 lo, (const struct compat_loop_info __user *) arg);
1313 mutex_unlock(&lo->lo_ctl_mutex);
1314 break;
1315 case LOOP_GET_STATUS:
1316 mutex_lock(&lo->lo_ctl_mutex);
1317 err = loop_get_status_compat(
1318 lo, (struct compat_loop_info __user *) arg);
1319 mutex_unlock(&lo->lo_ctl_mutex);
1320 break;
1321 case LOOP_CLR_FD:
1322 case LOOP_GET_STATUS64:
1323 case LOOP_SET_STATUS64:
1324 arg = (unsigned long) compat_ptr(arg);
1325 case LOOP_SET_FD:
1326 case LOOP_CHANGE_FD:
1327 err = lo_ioctl(inode, file, cmd, arg);
1328 break;
1329 default:
1330 err = -ENOIOCTLCMD;
1331 break;
1332 }
1333 unlock_kernel();
1334 return err;
1335}
1336#endif
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338static int lo_open(struct inode *inode, struct file *file)
1339{
1340 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1341
Ingo Molnarf85221d2006-03-23 03:00:38 -08001342 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001344 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 return 0;
1347}
1348
1349static int lo_release(struct inode *inode, struct file *file)
1350{
1351 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1352
Ingo Molnarf85221d2006-03-23 03:00:38 -08001353 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 --lo->lo_refcnt;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001355 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
1357 return 0;
1358}
1359
1360static struct block_device_operations lo_fops = {
1361 .owner = THIS_MODULE,
1362 .open = lo_open,
1363 .release = lo_release,
1364 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001365#ifdef CONFIG_COMPAT
1366 .compat_ioctl = lo_compat_ioctl,
1367#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368};
1369
1370/*
1371 * And now the modules code and kernel interface.
1372 */
Ken Chen73285082007-05-08 00:28:20 -07001373static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001375MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376MODULE_LICENSE("GPL");
1377MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1378
1379int loop_register_transfer(struct loop_func_table *funcs)
1380{
1381 unsigned int n = funcs->number;
1382
1383 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1384 return -EINVAL;
1385 xfer_funcs[n] = funcs;
1386 return 0;
1387}
1388
1389int loop_unregister_transfer(int number)
1390{
1391 unsigned int n = number;
1392 struct loop_device *lo;
1393 struct loop_func_table *xfer;
1394
1395 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1396 return -EINVAL;
1397
1398 xfer_funcs[n] = NULL;
1399
Ken Chen73285082007-05-08 00:28:20 -07001400 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001401 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 if (lo->lo_encryption == xfer)
1404 loop_release_xfer(lo);
1405
Ingo Molnarf85221d2006-03-23 03:00:38 -08001406 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
1408
1409 return 0;
1410}
1411
1412EXPORT_SYMBOL(loop_register_transfer);
1413EXPORT_SYMBOL(loop_unregister_transfer);
1414
Ken Chena47653f2007-06-08 13:46:44 -07001415static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001416{
1417 struct loop_device *lo;
1418 struct gendisk *disk;
1419
1420 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1421 if (!lo)
1422 goto out;
1423
1424 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1425 if (!lo->lo_queue)
1426 goto out_free_dev;
1427
1428 disk = lo->lo_disk = alloc_disk(1);
1429 if (!disk)
1430 goto out_free_queue;
1431
1432 mutex_init(&lo->lo_ctl_mutex);
1433 lo->lo_number = i;
1434 lo->lo_thread = NULL;
1435 init_waitqueue_head(&lo->lo_event);
1436 spin_lock_init(&lo->lo_lock);
1437 disk->major = LOOP_MAJOR;
1438 disk->first_minor = i;
1439 disk->fops = &lo_fops;
1440 disk->private_data = lo;
1441 disk->queue = lo->lo_queue;
1442 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001443 return lo;
1444
1445out_free_queue:
1446 blk_cleanup_queue(lo->lo_queue);
1447out_free_dev:
1448 kfree(lo);
1449out:
Al Viro07002e92007-05-12 16:23:15 -04001450 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001451}
1452
Ken Chena47653f2007-06-08 13:46:44 -07001453static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001454{
Ken Chen73285082007-05-08 00:28:20 -07001455 blk_cleanup_queue(lo->lo_queue);
1456 put_disk(lo->lo_disk);
1457 list_del(&lo->lo_list);
1458 kfree(lo);
1459}
1460
Ken Chena47653f2007-06-08 13:46:44 -07001461static struct loop_device *loop_init_one(int i)
1462{
1463 struct loop_device *lo;
1464
1465 list_for_each_entry(lo, &loop_devices, lo_list) {
1466 if (lo->lo_number == i)
1467 return lo;
1468 }
1469
1470 lo = loop_alloc(i);
1471 if (lo) {
1472 add_disk(lo->lo_disk);
1473 list_add_tail(&lo->lo_list, &loop_devices);
1474 }
1475 return lo;
1476}
1477
1478static void loop_del_one(struct loop_device *lo)
1479{
1480 del_gendisk(lo->lo_disk);
1481 loop_free(lo);
1482}
1483
Ken Chen73285082007-05-08 00:28:20 -07001484static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1485{
Al Viro705962c2007-05-13 05:52:32 -04001486 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001487 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001488
Al Viro705962c2007-05-13 05:52:32 -04001489 mutex_lock(&loop_devices_mutex);
1490 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001491 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001492 mutex_unlock(&loop_devices_mutex);
1493
1494 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001495 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001496}
1497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498static int __init loop_init(void)
1499{
Ken Chena47653f2007-06-08 13:46:44 -07001500 int i, nr;
1501 unsigned long range;
1502 struct loop_device *lo, *next;
1503
1504 /*
1505 * loop module now has a feature to instantiate underlying device
1506 * structure on-demand, provided that there is an access dev node.
1507 * However, this will not work well with user space tool that doesn't
1508 * know about such "feature". In order to not break any existing
1509 * tool, we do the following:
1510 *
1511 * (1) if max_loop is specified, create that many upfront, and this
1512 * also becomes a hard limit.
1513 * (2) if max_loop is not specified, create 8 loop device on module
1514 * load, user can further extend loop device by create dev node
1515 * themselves and have kernel automatically instantiate actual
1516 * device on-demand.
1517 */
1518 if (max_loop > 1UL << MINORBITS)
1519 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
Ken Chen73285082007-05-08 00:28:20 -07001521 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001522 nr = max_loop;
1523 range = max_loop;
1524 } else {
1525 nr = 8;
1526 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
Ken Chena47653f2007-06-08 13:46:44 -07001528
1529 if (register_blkdev(LOOP_MAJOR, "loop"))
1530 return -EIO;
1531
1532 for (i = 0; i < nr; i++) {
1533 lo = loop_alloc(i);
1534 if (!lo)
1535 goto Enomem;
1536 list_add_tail(&lo->lo_list, &loop_devices);
1537 }
1538
1539 /* point of no return */
1540
1541 list_for_each_entry(lo, &loop_devices, lo_list)
1542 add_disk(lo->lo_disk);
1543
1544 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1545 THIS_MODULE, loop_probe, NULL, NULL);
1546
Ken Chen73285082007-05-08 00:28:20 -07001547 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001549
1550Enomem:
1551 printk(KERN_INFO "loop: out of memory\n");
1552
1553 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1554 loop_free(lo);
1555
1556 unregister_blkdev(LOOP_MAJOR, "loop");
1557 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
1559
Ken Chen73285082007-05-08 00:28:20 -07001560static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Ken Chena47653f2007-06-08 13:46:44 -07001562 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001563 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Ken Chena47653f2007-06-08 13:46:44 -07001565 range = max_loop ? max_loop : 1UL << MINORBITS;
1566
Ken Chen73285082007-05-08 00:28:20 -07001567 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1568 loop_del_one(lo);
1569
Ken Chena47653f2007-06-08 13:46:44 -07001570 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1572 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573}
1574
1575module_init(loop_init);
1576module_exit(loop_exit);
1577
1578#ifndef MODULE
1579static int __init max_loop_setup(char *str)
1580{
1581 max_loop = simple_strtol(str, NULL, 0);
1582 return 1;
1583}
1584
1585__setup("max_loop=", max_loop_setup);
1586#endif