blob: d3a25b027ff9fc5e6331372a252b5e134b6270c4 [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.
Jan Engelhardt96de0e22007-10-19 23:21:04 +020032 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 *
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 Howells863d5b82006-08-29 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
Laurent Vivier476a4812008-03-26 12:11:53 +010085static int max_part;
86static int part_shift;
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
89 * Transfer functions
90 */
91static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
95{
96 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
98
99 if (cmd == READ)
100 memcpy(loop_buf, raw_buf, size);
101 else
102 memcpy(raw_buf, loop_buf, size);
103
104 kunmap_atomic(raw_buf, KM_USER0);
105 kunmap_atomic(loop_buf, KM_USER1);
106 cond_resched();
107 return 0;
108}
109
110static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
114{
115 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
117 char *in, *out, *key;
118 int i, keysize;
119
120 if (cmd == READ) {
121 in = raw_buf;
122 out = loop_buf;
123 } else {
124 in = loop_buf;
125 out = raw_buf;
126 }
127
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
132
133 kunmap_atomic(raw_buf, KM_USER0);
134 kunmap_atomic(loop_buf, KM_USER1);
135 cond_resched();
136 return 0;
137}
138
139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140{
141 if (unlikely(info->lo_encrypt_key_size <= 0))
142 return -EINVAL;
143 return 0;
144}
145
146static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
149};
150
151static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
154 .init = xor_init
155};
156
157/* xfer_funcs[0] is special - its release function is never called */
158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
159 &none_funcs,
160 &xor_funcs
161};
162
163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164{
165 loff_t size, offset, loopsize;
166
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 offset = lo->lo_offset;
170 loopsize = size - offset;
171 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
172 loopsize = lo->lo_sizelimit;
173
174 /*
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
177 */
178 return loopsize >> 9;
179}
180
181static int
182figure_loop_size(struct loop_device *lo)
183{
184 loff_t size = get_loop_size(lo, lo->lo_backing_file);
185 sector_t x = (sector_t)size;
186
187 if (unlikely((loff_t)x != size))
188 return -EFBIG;
189
Ken Chen73285082007-05-08 00:28:20 -0700190 set_capacity(lo->lo_disk, x);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return 0;
192}
193
194static inline int
195lo_do_transfer(struct loop_device *lo, int cmd,
196 struct page *rpage, unsigned roffs,
197 struct page *lpage, unsigned loffs,
198 int size, sector_t rblock)
199{
200 if (unlikely(!lo->transfer))
201 return 0;
202
203 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
204}
205
206/**
207 * do_lo_send_aops - helper for writing data to a loop device
208 *
209 * This is the fast version for backing filesystems which implement the address
Nick Pigginafddba42007-10-16 01:25:01 -0700210 * space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 */
212static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
Nick Pigginafddba42007-10-16 01:25:01 -0700213 int bsize, loff_t pos, struct page *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
216 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 pgoff_t index;
218 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800219 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800221 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 index = pos >> PAGE_CACHE_SHIFT;
223 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
224 bv_offs = bvec->bv_offset;
225 len = bvec->bv_len;
226 while (len > 0) {
227 sector_t IV;
Nick Pigginafddba42007-10-16 01:25:01 -0700228 unsigned size, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 int transfer_result;
Nick Pigginafddba42007-10-16 01:25:01 -0700230 struct page *page;
231 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
234 size = PAGE_CACHE_SIZE - offset;
235 if (size > len)
236 size = len;
Nick Pigginafddba42007-10-16 01:25:01 -0700237
238 ret = pagecache_write_begin(file, mapping, pos, size, 0,
239 &page, &fsdata);
240 if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
244 bvec->bv_page, bv_offs, size, IV);
Nick Pigginafddba42007-10-16 01:25:01 -0700245 copied = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (unlikely(transfer_result))
Nick Pigginafddba42007-10-16 01:25:01 -0700247 copied = 0;
248
249 ret = pagecache_write_end(file, mapping, pos, size, copied,
250 page, fsdata);
Dmitry Monakhov8268f5a2007-10-16 01:25:02 -0700251 if (ret < 0 || ret != copied)
Nick Pigginafddba42007-10-16 01:25:01 -0700252 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700253
254 if (unlikely(transfer_result))
255 goto fail;
256
257 bv_offs += copied;
258 len -= copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 offset = 0;
260 index++;
Nick Pigginafddba42007-10-16 01:25:01 -0700261 pos += copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800263 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800265 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267fail:
268 ret = -1;
269 goto out;
270}
271
272/**
273 * __do_lo_send_write - helper for writing data to a loop device
274 *
275 * This helper just factors out common code between do_lo_send_direct_write()
276 * and do_lo_send_write().
277 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800278static int __do_lo_send_write(struct file *file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100279 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 ssize_t bw;
282 mm_segment_t old_fs = get_fs();
283
284 set_fs(get_ds());
285 bw = file->f_op->write(file, buf, len, &pos);
286 set_fs(old_fs);
287 if (likely(bw == len))
288 return 0;
289 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
290 (unsigned long long)pos, len);
291 if (bw >= 0)
292 bw = -EIO;
293 return bw;
294}
295
296/**
297 * do_lo_send_direct_write - helper for writing data to a loop device
298 *
299 * This is the fast, non-transforming version for backing filesystems which do
Nick Pigginafddba42007-10-16 01:25:01 -0700300 * not implement the address space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * It uses the write file operation which should be present on all writeable
302 * filesystems.
303 */
304static int do_lo_send_direct_write(struct loop_device *lo,
305 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
306{
307 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100308 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 bvec->bv_len, pos);
310 kunmap(bvec->bv_page);
311 cond_resched();
312 return bw;
313}
314
315/**
316 * do_lo_send_write - helper for writing data to a loop device
317 *
318 * This is the slow, transforming version for filesystems which do not
Nick Pigginafddba42007-10-16 01:25:01 -0700319 * implement the address space operations write_begin and write_end. It
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * uses the write file operation which should be present on all writeable
321 * filesystems.
322 *
323 * Using fops->write is slower than using aops->{prepare,commit}_write in the
324 * transforming case because we need to double buffer the data as we cannot do
325 * the transformations in place as we do not have direct access to the
326 * destination pages of the backing file.
327 */
328static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
329 int bsize, loff_t pos, struct page *page)
330{
331 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
332 bvec->bv_offset, bvec->bv_len, pos >> 9);
333 if (likely(!ret))
334 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6cc2006-10-10 22:45:07 +0100335 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 pos);
337 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
338 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
339 if (ret > 0)
340 ret = -EIO;
341 return ret;
342}
343
344static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
345 loff_t pos)
346{
347 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
348 struct page *page);
349 struct bio_vec *bvec;
350 struct page *page = NULL;
351 int i, ret = 0;
352
353 do_lo_send = do_lo_send_aops;
354 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
355 do_lo_send = do_lo_send_direct_write;
356 if (lo->transfer != transfer_none) {
357 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
358 if (unlikely(!page))
359 goto fail;
360 kmap(page);
361 do_lo_send = do_lo_send_write;
362 }
363 }
364 bio_for_each_segment(bvec, bio, i) {
365 ret = do_lo_send(lo, bvec, bsize, pos, page);
366 if (ret < 0)
367 break;
368 pos += bvec->bv_len;
369 }
370 if (page) {
371 kunmap(page);
372 __free_page(page);
373 }
374out:
375 return ret;
376fail:
377 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
378 ret = -ENOMEM;
379 goto out;
380}
381
382struct lo_read_data {
383 struct loop_device *lo;
384 struct page *page;
385 unsigned offset;
386 int bsize;
387};
388
389static int
Jens Axboefd582142007-06-12 21:20:37 +0200390lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
391 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Jens Axboefd582142007-06-12 21:20:37 +0200393 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200395 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 sector_t IV;
Jens Axboefd582142007-06-12 21:20:37 +0200397 size_t size;
398 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Jens Axboecac36bb2007-06-14 13:10:48 +0200400 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200401 if (unlikely(ret))
402 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Jens Axboefd582142007-06-12 21:20:37 +0200404 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
405 (buf->offset >> 9);
406 size = sd->len;
407 if (size > p->bsize)
408 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Jens Axboefd582142007-06-12 21:20:37 +0200410 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 printk(KERN_ERR "loop: transfer error block %ld\n",
412 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200413 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
415
416 flush_dcache_page(p->page);
417
Jens Axboefd582142007-06-12 21:20:37 +0200418 if (size > 0)
419 p->offset += size;
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return size;
422}
423
424static int
Jens Axboefd582142007-06-12 21:20:37 +0200425lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
426{
427 return __splice_from_pipe(pipe, sd, lo_splice_actor);
428}
429
430static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431do_lo_receive(struct loop_device *lo,
432 struct bio_vec *bvec, int bsize, loff_t pos)
433{
434 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200435 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200437 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 cookie.lo = lo;
440 cookie.page = bvec->bv_page;
441 cookie.offset = bvec->bv_offset;
442 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200443
444 sd.len = 0;
445 sd.total_len = bvec->bv_len;
446 sd.flags = 0;
447 sd.pos = pos;
448 sd.u.data = &cookie;
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200451 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
452
453 if (retval < 0)
454 return retval;
455
456 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
459static int
460lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
461{
462 struct bio_vec *bvec;
463 int i, ret = 0;
464
465 bio_for_each_segment(bvec, bio, i) {
466 ret = do_lo_receive(lo, bvec, bsize, pos);
467 if (ret < 0)
468 break;
469 pos += bvec->bv_len;
470 }
471 return ret;
472}
473
474static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
475{
476 loff_t pos;
477 int ret;
478
479 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
480 if (bio_rw(bio) == WRITE)
481 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
482 else
483 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
484 return ret;
485}
486
487/*
488 * Add bio to back of pending list
489 */
490static void loop_add_bio(struct loop_device *lo, struct bio *bio)
491{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (lo->lo_biotail) {
493 lo->lo_biotail->bi_next = bio;
494 lo->lo_biotail = bio;
495 } else
496 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
499/*
500 * Grab first pending buffer
501 */
502static struct bio *loop_get_bio(struct loop_device *lo)
503{
504 struct bio *bio;
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if ((bio = lo->lo_bio)) {
507 if (bio == lo->lo_biotail)
508 lo->lo_biotail = NULL;
509 lo->lo_bio = bio->bi_next;
510 bio->bi_next = NULL;
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 return bio;
514}
515
Jens Axboe165125e2007-07-24 09:28:11 +0200516static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 struct loop_device *lo = q->queuedata;
519 int rw = bio_rw(old_bio);
520
Nick Piggin35a82d12005-06-23 00:09:06 -0700521 if (rw == READA)
522 rw = READ;
523
524 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 spin_lock_irq(&lo->lo_lock);
527 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700528 goto out;
529 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
530 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700532 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700533 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700537 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200538 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
542/*
543 * kick off io on the underlying address space
544 */
Jens Axboe165125e2007-07-24 09:28:11 +0200545static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct loop_device *lo = q->queuedata;
548
Nick Piggin75ad23b2008-04-29 14:48:33 +0200549 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 blk_run_address_space(lo->lo_backing_file->f_mapping);
551}
552
553struct switch_request {
554 struct file *file;
555 struct completion wait;
556};
557
558static void do_loop_switch(struct loop_device *, struct switch_request *);
559
560static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
561{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (unlikely(!bio->bi_bdev)) {
563 do_loop_switch(lo, bio->bi_private);
564 bio_put(bio);
565 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700566 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200567 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569}
570
571/*
572 * worker thread that handles reads/writes to file backed loop devices,
573 * to avoid blocking in our make_request_fn. it also does loop decrypting
574 * on reads for block backed loop, as that is too heavy to do from
575 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700576 *
577 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
578 * calling kthread_stop(). Therefore once kthread_should_stop() is
579 * true, make_request will not place any more requests. Therefore
580 * once kthread_should_stop() is true and lo_bio is NULL, we are
581 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
583static int loop_thread(void *data)
584{
585 struct loop_device *lo = data;
586 struct bio *bio;
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 set_user_nice(current, -20);
589
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700590 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700592 wait_event_interruptible(lo->lo_event,
593 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700594
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700595 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700596 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700597 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700599 spin_unlock_irq(&lo->lo_lock);
600
601 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return 0;
606}
607
608/*
609 * loop_switch performs the hard work of switching a backing store.
610 * First it needs to flush existing IO, it does this by sending a magic
611 * BIO down the pipe. The completion of this BIO does the actual switch.
612 */
613static int loop_switch(struct loop_device *lo, struct file *file)
614{
615 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100616 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (!bio)
618 return -ENOMEM;
619 init_completion(&w.wait);
620 w.file = file;
621 bio->bi_private = &w;
622 bio->bi_bdev = NULL;
623 loop_make_request(lo->lo_queue, bio);
624 wait_for_completion(&w.wait);
625 return 0;
626}
627
628/*
629 * Do the actual switch; called from the BIO completion routine
630 */
631static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
632{
633 struct file *file = p->file;
634 struct file *old_file = lo->lo_backing_file;
635 struct address_space *mapping = file->f_mapping;
636
637 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
638 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700639 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
640 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 lo->old_gfp_mask = mapping_gfp_mask(mapping);
642 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
643 complete(&p->wait);
644}
645
646
647/*
648 * loop_change_fd switched the backing store of a loopback device to
649 * a new file. This is useful for operating system installers to free up
650 * the original file and in High Availability environments to switch to
651 * an alternative location for the content in case of server meltdown.
652 * This can only work if the loop device is used read-only, and if the
653 * new backing store is the same size and type as the old backing store.
654 */
655static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
656 struct block_device *bdev, unsigned int arg)
657{
658 struct file *file, *old_file;
659 struct inode *inode;
660 int error;
661
662 error = -ENXIO;
663 if (lo->lo_state != Lo_bound)
664 goto out;
665
666 /* the loop device has to be read-only */
667 error = -EINVAL;
668 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
669 goto out;
670
671 error = -EBADF;
672 file = fget(arg);
673 if (!file)
674 goto out;
675
676 inode = file->f_mapping->host;
677 old_file = lo->lo_backing_file;
678
679 error = -EINVAL;
680
681 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 goto out_putf;
683
Jens Axboefd582142007-06-12 21:20:37 +0200684 /* new backing store needs to support loop (eg splice_read) */
685 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 goto out_putf;
687
688 /* size of the new backing store needs to be the same */
689 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
690 goto out_putf;
691
692 /* and ... switch */
693 error = loop_switch(lo, file);
694 if (error)
695 goto out_putf;
696
697 fput(old_file);
Laurent Vivier476a4812008-03-26 12:11:53 +0100698 if (max_part > 0)
699 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701
702 out_putf:
703 fput(file);
704 out:
705 return error;
706}
707
708static inline int is_loop_device(struct file *file)
709{
710 struct inode *i = file->f_mapping->host;
711
712 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
713}
714
715static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
716 struct block_device *bdev, unsigned int arg)
717{
718 struct file *file, *f;
719 struct inode *inode;
720 struct address_space *mapping;
721 unsigned lo_blocksize;
722 int lo_flags = 0;
723 int error;
724 loff_t size;
725
726 /* This is safe, since we have a reference from open(). */
727 __module_get(THIS_MODULE);
728
729 error = -EBADF;
730 file = fget(arg);
731 if (!file)
732 goto out;
733
734 error = -EBUSY;
735 if (lo->lo_state != Lo_unbound)
736 goto out_putf;
737
738 /* Avoid recursion */
739 f = file;
740 while (is_loop_device(f)) {
741 struct loop_device *l;
742
743 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
744 goto out_putf;
745
746 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
747 if (l->lo_state == Lo_unbound) {
748 error = -EINVAL;
749 goto out_putf;
750 }
751 f = l->lo_backing_file;
752 }
753
754 mapping = file->f_mapping;
755 inode = mapping->host;
756
757 if (!(file->f_mode & FMODE_WRITE))
758 lo_flags |= LO_FLAGS_READ_ONLY;
759
760 error = -EINVAL;
761 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700762 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 /*
764 * If we can't read - sorry. If we only can't write - well,
765 * it's going to be read-only.
766 */
Jens Axboefd582142007-06-12 21:20:37 +0200767 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 goto out_putf;
Nick Pigginafddba42007-10-16 01:25:01 -0700769 if (aops->prepare_write || aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 lo_flags |= LO_FLAGS_USE_AOPS;
771 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
772 lo_flags |= LO_FLAGS_READ_ONLY;
773
Theodore Ts'oba52de12006-09-27 01:50:49 -0700774 lo_blocksize = S_ISBLK(inode->i_mode) ?
775 inode->i_bdev->bd_block_size : PAGE_SIZE;
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 error = 0;
778 } else {
779 goto out_putf;
780 }
781
782 size = get_loop_size(lo, file);
783
784 if ((loff_t)(sector_t)size != size) {
785 error = -EFBIG;
786 goto out_putf;
787 }
788
789 if (!(lo_file->f_mode & FMODE_WRITE))
790 lo_flags |= LO_FLAGS_READ_ONLY;
791
792 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
793
794 lo->lo_blocksize = lo_blocksize;
795 lo->lo_device = bdev;
796 lo->lo_flags = lo_flags;
797 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700798 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 lo->ioctl = NULL;
800 lo->lo_sizelimit = 0;
801 lo->old_gfp_mask = mapping_gfp_mask(mapping);
802 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
803
804 lo->lo_bio = lo->lo_biotail = NULL;
805
806 /*
807 * set queue make_request_fn, and add limits based on lower level
808 * device
809 */
810 blk_queue_make_request(lo->lo_queue, loop_make_request);
811 lo->lo_queue->queuedata = lo;
812 lo->lo_queue->unplug_fn = loop_unplug;
813
Ken Chen73285082007-05-08 00:28:20 -0700814 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 bd_set_size(bdev, size << 9);
816
817 set_blocksize(bdev, lo_blocksize);
818
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700819 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
820 lo->lo_number);
821 if (IS_ERR(lo->lo_thread)) {
822 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700823 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700824 }
825 lo->lo_state = Lo_bound;
826 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100827 if (max_part > 0)
828 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return 0;
830
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700831out_clr:
832 lo->lo_thread = NULL;
833 lo->lo_device = NULL;
834 lo->lo_backing_file = NULL;
835 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700836 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700837 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700838 bd_set_size(bdev, 0);
839 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
840 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 out_putf:
842 fput(file);
843 out:
844 /* This is safe: open() is still holding a reference. */
845 module_put(THIS_MODULE);
846 return error;
847}
848
849static int
850loop_release_xfer(struct loop_device *lo)
851{
852 int err = 0;
853 struct loop_func_table *xfer = lo->lo_encryption;
854
855 if (xfer) {
856 if (xfer->release)
857 err = xfer->release(lo);
858 lo->transfer = NULL;
859 lo->lo_encryption = NULL;
860 module_put(xfer->owner);
861 }
862 return err;
863}
864
865static int
866loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
867 const struct loop_info64 *i)
868{
869 int err = 0;
870
871 if (xfer) {
872 struct module *owner = xfer->owner;
873
874 if (!try_module_get(owner))
875 return -EINVAL;
876 if (xfer->init)
877 err = xfer->init(lo, i);
878 if (err)
879 module_put(owner);
880 else
881 lo->lo_encryption = xfer;
882 }
883 return err;
884}
885
886static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
887{
888 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400889 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 if (lo->lo_state != Lo_bound)
892 return -ENXIO;
893
894 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
895 return -EBUSY;
896
897 if (filp == NULL)
898 return -EINVAL;
899
900 spin_lock_irq(&lo->lo_lock);
901 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 spin_unlock_irq(&lo->lo_lock);
903
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700904 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 lo->lo_backing_file = NULL;
907
908 loop_release_xfer(lo);
909 lo->transfer = NULL;
910 lo->ioctl = NULL;
911 lo->lo_device = NULL;
912 lo->lo_encryption = NULL;
913 lo->lo_offset = 0;
914 lo->lo_sizelimit = 0;
915 lo->lo_encrypt_key_size = 0;
916 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700917 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
919 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
920 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700921 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700922 set_capacity(lo->lo_disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 bd_set_size(bdev, 0);
924 mapping_set_gfp_mask(filp->f_mapping, gfp);
925 lo->lo_state = Lo_unbound;
926 fput(filp);
927 /* This is safe: open() is still holding a reference. */
928 module_put(THIS_MODULE);
Laurent Vivier476a4812008-03-26 12:11:53 +0100929 if (max_part > 0)
930 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 return 0;
932}
933
934static int
935loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
936{
937 int err;
938 struct loop_func_table *xfer;
939
940 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
941 !capable(CAP_SYS_ADMIN))
942 return -EPERM;
943 if (lo->lo_state != Lo_bound)
944 return -ENXIO;
945 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
946 return -EINVAL;
947
948 err = loop_release_xfer(lo);
949 if (err)
950 return err;
951
952 if (info->lo_encrypt_type) {
953 unsigned int type = info->lo_encrypt_type;
954
955 if (type >= MAX_LO_CRYPT)
956 return -EINVAL;
957 xfer = xfer_funcs[type];
958 if (xfer == NULL)
959 return -EINVAL;
960 } else
961 xfer = NULL;
962
963 err = loop_init_xfer(lo, xfer, info);
964 if (err)
965 return err;
966
967 if (lo->lo_offset != info->lo_offset ||
968 lo->lo_sizelimit != info->lo_sizelimit) {
969 lo->lo_offset = info->lo_offset;
970 lo->lo_sizelimit = info->lo_sizelimit;
971 if (figure_loop_size(lo))
972 return -EFBIG;
973 }
974
975 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
976 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
977 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
978 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
979
980 if (!xfer)
981 xfer = &none_funcs;
982 lo->transfer = xfer->transfer;
983 lo->ioctl = xfer->ioctl;
984
David Woodhouse96c58652008-02-06 01:36:27 -0800985 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
986 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
987 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
990 lo->lo_init[0] = info->lo_init[0];
991 lo->lo_init[1] = info->lo_init[1];
992 if (info->lo_encrypt_key_size) {
993 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
994 info->lo_encrypt_key_size);
995 lo->lo_key_owner = current->uid;
996 }
997
998 return 0;
999}
1000
1001static int
1002loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1003{
1004 struct file *file = lo->lo_backing_file;
1005 struct kstat stat;
1006 int error;
1007
1008 if (lo->lo_state != Lo_bound)
1009 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001010 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 if (error)
1012 return error;
1013 memset(info, 0, sizeof(*info));
1014 info->lo_number = lo->lo_number;
1015 info->lo_device = huge_encode_dev(stat.dev);
1016 info->lo_inode = stat.ino;
1017 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1018 info->lo_offset = lo->lo_offset;
1019 info->lo_sizelimit = lo->lo_sizelimit;
1020 info->lo_flags = lo->lo_flags;
1021 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1022 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1023 info->lo_encrypt_type =
1024 lo->lo_encryption ? lo->lo_encryption->number : 0;
1025 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1026 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1027 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1028 lo->lo_encrypt_key_size);
1029 }
1030 return 0;
1031}
1032
1033static void
1034loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1035{
1036 memset(info64, 0, sizeof(*info64));
1037 info64->lo_number = info->lo_number;
1038 info64->lo_device = info->lo_device;
1039 info64->lo_inode = info->lo_inode;
1040 info64->lo_rdevice = info->lo_rdevice;
1041 info64->lo_offset = info->lo_offset;
1042 info64->lo_sizelimit = 0;
1043 info64->lo_encrypt_type = info->lo_encrypt_type;
1044 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1045 info64->lo_flags = info->lo_flags;
1046 info64->lo_init[0] = info->lo_init[0];
1047 info64->lo_init[1] = info->lo_init[1];
1048 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1049 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1050 else
1051 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1052 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1053}
1054
1055static int
1056loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1057{
1058 memset(info, 0, sizeof(*info));
1059 info->lo_number = info64->lo_number;
1060 info->lo_device = info64->lo_device;
1061 info->lo_inode = info64->lo_inode;
1062 info->lo_rdevice = info64->lo_rdevice;
1063 info->lo_offset = info64->lo_offset;
1064 info->lo_encrypt_type = info64->lo_encrypt_type;
1065 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1066 info->lo_flags = info64->lo_flags;
1067 info->lo_init[0] = info64->lo_init[0];
1068 info->lo_init[1] = info64->lo_init[1];
1069 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1070 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1071 else
1072 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1073 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1074
1075 /* error in case values were truncated */
1076 if (info->lo_device != info64->lo_device ||
1077 info->lo_rdevice != info64->lo_rdevice ||
1078 info->lo_inode != info64->lo_inode ||
1079 info->lo_offset != info64->lo_offset)
1080 return -EOVERFLOW;
1081
1082 return 0;
1083}
1084
1085static int
1086loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1087{
1088 struct loop_info info;
1089 struct loop_info64 info64;
1090
1091 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1092 return -EFAULT;
1093 loop_info64_from_old(&info, &info64);
1094 return loop_set_status(lo, &info64);
1095}
1096
1097static int
1098loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1099{
1100 struct loop_info64 info64;
1101
1102 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1103 return -EFAULT;
1104 return loop_set_status(lo, &info64);
1105}
1106
1107static int
1108loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1109 struct loop_info info;
1110 struct loop_info64 info64;
1111 int err = 0;
1112
1113 if (!arg)
1114 err = -EINVAL;
1115 if (!err)
1116 err = loop_get_status(lo, &info64);
1117 if (!err)
1118 err = loop_info64_to_old(&info64, &info);
1119 if (!err && copy_to_user(arg, &info, sizeof(info)))
1120 err = -EFAULT;
1121
1122 return err;
1123}
1124
1125static int
1126loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1127 struct loop_info64 info64;
1128 int err = 0;
1129
1130 if (!arg)
1131 err = -EINVAL;
1132 if (!err)
1133 err = loop_get_status(lo, &info64);
1134 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1135 err = -EFAULT;
1136
1137 return err;
1138}
1139
1140static int lo_ioctl(struct inode * inode, struct file * file,
1141 unsigned int cmd, unsigned long arg)
1142{
1143 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1144 int err;
1145
Ingo Molnarf85221d2006-03-23 03:00:38 -08001146 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 switch (cmd) {
1148 case LOOP_SET_FD:
1149 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1150 break;
1151 case LOOP_CHANGE_FD:
1152 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1153 break;
1154 case LOOP_CLR_FD:
1155 err = loop_clr_fd(lo, inode->i_bdev);
1156 break;
1157 case LOOP_SET_STATUS:
1158 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1159 break;
1160 case LOOP_GET_STATUS:
1161 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1162 break;
1163 case LOOP_SET_STATUS64:
1164 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1165 break;
1166 case LOOP_GET_STATUS64:
1167 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1168 break;
1169 default:
1170 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1171 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001172 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 return err;
1174}
1175
David Howells863d5b82006-08-29 19:06:14 +01001176#ifdef CONFIG_COMPAT
1177struct compat_loop_info {
1178 compat_int_t lo_number; /* ioctl r/o */
1179 compat_dev_t lo_device; /* ioctl r/o */
1180 compat_ulong_t lo_inode; /* ioctl r/o */
1181 compat_dev_t lo_rdevice; /* ioctl r/o */
1182 compat_int_t lo_offset;
1183 compat_int_t lo_encrypt_type;
1184 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1185 compat_int_t lo_flags; /* ioctl r/o */
1186 char lo_name[LO_NAME_SIZE];
1187 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1188 compat_ulong_t lo_init[2];
1189 char reserved[4];
1190};
1191
1192/*
1193 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1194 * - noinlined to reduce stack space usage in main part of driver
1195 */
1196static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001197loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b82006-08-29 19:06:14 +01001198 struct loop_info64 *info64)
1199{
1200 struct compat_loop_info info;
1201
1202 if (copy_from_user(&info, arg, sizeof(info)))
1203 return -EFAULT;
1204
1205 memset(info64, 0, sizeof(*info64));
1206 info64->lo_number = info.lo_number;
1207 info64->lo_device = info.lo_device;
1208 info64->lo_inode = info.lo_inode;
1209 info64->lo_rdevice = info.lo_rdevice;
1210 info64->lo_offset = info.lo_offset;
1211 info64->lo_sizelimit = 0;
1212 info64->lo_encrypt_type = info.lo_encrypt_type;
1213 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1214 info64->lo_flags = info.lo_flags;
1215 info64->lo_init[0] = info.lo_init[0];
1216 info64->lo_init[1] = info.lo_init[1];
1217 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1218 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1219 else
1220 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1221 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1222 return 0;
1223}
1224
1225/*
1226 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1227 * - noinlined to reduce stack space usage in main part of driver
1228 */
1229static noinline int
1230loop_info64_to_compat(const struct loop_info64 *info64,
1231 struct compat_loop_info __user *arg)
1232{
1233 struct compat_loop_info info;
1234
1235 memset(&info, 0, sizeof(info));
1236 info.lo_number = info64->lo_number;
1237 info.lo_device = info64->lo_device;
1238 info.lo_inode = info64->lo_inode;
1239 info.lo_rdevice = info64->lo_rdevice;
1240 info.lo_offset = info64->lo_offset;
1241 info.lo_encrypt_type = info64->lo_encrypt_type;
1242 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1243 info.lo_flags = info64->lo_flags;
1244 info.lo_init[0] = info64->lo_init[0];
1245 info.lo_init[1] = info64->lo_init[1];
1246 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1247 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1248 else
1249 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1250 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1251
1252 /* error in case values were truncated */
1253 if (info.lo_device != info64->lo_device ||
1254 info.lo_rdevice != info64->lo_rdevice ||
1255 info.lo_inode != info64->lo_inode ||
1256 info.lo_offset != info64->lo_offset ||
1257 info.lo_init[0] != info64->lo_init[0] ||
1258 info.lo_init[1] != info64->lo_init[1])
1259 return -EOVERFLOW;
1260
1261 if (copy_to_user(arg, &info, sizeof(info)))
1262 return -EFAULT;
1263 return 0;
1264}
1265
1266static int
1267loop_set_status_compat(struct loop_device *lo,
1268 const struct compat_loop_info __user *arg)
1269{
1270 struct loop_info64 info64;
1271 int ret;
1272
1273 ret = loop_info64_from_compat(arg, &info64);
1274 if (ret < 0)
1275 return ret;
1276 return loop_set_status(lo, &info64);
1277}
1278
1279static int
1280loop_get_status_compat(struct loop_device *lo,
1281 struct compat_loop_info __user *arg)
1282{
1283 struct loop_info64 info64;
1284 int err = 0;
1285
1286 if (!arg)
1287 err = -EINVAL;
1288 if (!err)
1289 err = loop_get_status(lo, &info64);
1290 if (!err)
1291 err = loop_info64_to_compat(&info64, arg);
1292 return err;
1293}
1294
1295static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1296{
Josef Sipek6c648be2006-12-08 02:36:55 -08001297 struct inode *inode = file->f_path.dentry->d_inode;
David Howells863d5b82006-08-29 19:06:14 +01001298 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1299 int err;
1300
David Howells863d5b82006-08-29 19:06:14 +01001301 switch(cmd) {
1302 case LOOP_SET_STATUS:
1303 mutex_lock(&lo->lo_ctl_mutex);
1304 err = loop_set_status_compat(
1305 lo, (const struct compat_loop_info __user *) arg);
1306 mutex_unlock(&lo->lo_ctl_mutex);
1307 break;
1308 case LOOP_GET_STATUS:
1309 mutex_lock(&lo->lo_ctl_mutex);
1310 err = loop_get_status_compat(
1311 lo, (struct compat_loop_info __user *) arg);
1312 mutex_unlock(&lo->lo_ctl_mutex);
1313 break;
1314 case LOOP_CLR_FD:
1315 case LOOP_GET_STATUS64:
1316 case LOOP_SET_STATUS64:
1317 arg = (unsigned long) compat_ptr(arg);
1318 case LOOP_SET_FD:
1319 case LOOP_CHANGE_FD:
1320 err = lo_ioctl(inode, file, cmd, arg);
1321 break;
1322 default:
1323 err = -ENOIOCTLCMD;
1324 break;
1325 }
David Howells863d5b82006-08-29 19:06:14 +01001326 return err;
1327}
1328#endif
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330static int lo_open(struct inode *inode, struct file *file)
1331{
1332 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1333
Ingo Molnarf85221d2006-03-23 03:00:38 -08001334 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001336 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 return 0;
1339}
1340
1341static int lo_release(struct inode *inode, struct file *file)
1342{
1343 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1344
Ingo Molnarf85221d2006-03-23 03:00:38 -08001345 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 --lo->lo_refcnt;
David Woodhouse96c58652008-02-06 01:36:27 -08001347
1348 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt)
1349 loop_clr_fd(lo, inode->i_bdev);
1350
Ingo Molnarf85221d2006-03-23 03:00:38 -08001351 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 return 0;
1354}
1355
1356static struct block_device_operations lo_fops = {
1357 .owner = THIS_MODULE,
1358 .open = lo_open,
1359 .release = lo_release,
1360 .ioctl = lo_ioctl,
David Howells863d5b82006-08-29 19:06:14 +01001361#ifdef CONFIG_COMPAT
1362 .compat_ioctl = lo_compat_ioctl,
1363#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364};
1365
1366/*
1367 * And now the modules code and kernel interface.
1368 */
Ken Chen73285082007-05-08 00:28:20 -07001369static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001371MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001372module_param(max_part, int, 0);
1373MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374MODULE_LICENSE("GPL");
1375MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1376
1377int loop_register_transfer(struct loop_func_table *funcs)
1378{
1379 unsigned int n = funcs->number;
1380
1381 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1382 return -EINVAL;
1383 xfer_funcs[n] = funcs;
1384 return 0;
1385}
1386
1387int loop_unregister_transfer(int number)
1388{
1389 unsigned int n = number;
1390 struct loop_device *lo;
1391 struct loop_func_table *xfer;
1392
1393 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1394 return -EINVAL;
1395
1396 xfer_funcs[n] = NULL;
1397
Ken Chen73285082007-05-08 00:28:20 -07001398 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001399 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
1401 if (lo->lo_encryption == xfer)
1402 loop_release_xfer(lo);
1403
Ingo Molnarf85221d2006-03-23 03:00:38 -08001404 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 }
1406
1407 return 0;
1408}
1409
1410EXPORT_SYMBOL(loop_register_transfer);
1411EXPORT_SYMBOL(loop_unregister_transfer);
1412
Ken Chena47653f2007-06-08 13:46:44 -07001413static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001414{
1415 struct loop_device *lo;
1416 struct gendisk *disk;
1417
1418 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1419 if (!lo)
1420 goto out;
1421
1422 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1423 if (!lo->lo_queue)
1424 goto out_free_dev;
1425
Laurent Vivier476a4812008-03-26 12:11:53 +01001426 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001427 if (!disk)
1428 goto out_free_queue;
1429
1430 mutex_init(&lo->lo_ctl_mutex);
1431 lo->lo_number = i;
1432 lo->lo_thread = NULL;
1433 init_waitqueue_head(&lo->lo_event);
1434 spin_lock_init(&lo->lo_lock);
1435 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001436 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001437 disk->fops = &lo_fops;
1438 disk->private_data = lo;
1439 disk->queue = lo->lo_queue;
1440 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001441 return lo;
1442
1443out_free_queue:
1444 blk_cleanup_queue(lo->lo_queue);
1445out_free_dev:
1446 kfree(lo);
1447out:
Al Viro07002e92007-05-12 16:23:15 -04001448 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001449}
1450
Ken Chena47653f2007-06-08 13:46:44 -07001451static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001452{
Ken Chen73285082007-05-08 00:28:20 -07001453 blk_cleanup_queue(lo->lo_queue);
1454 put_disk(lo->lo_disk);
1455 list_del(&lo->lo_list);
1456 kfree(lo);
1457}
1458
Ken Chena47653f2007-06-08 13:46:44 -07001459static struct loop_device *loop_init_one(int i)
1460{
1461 struct loop_device *lo;
1462
1463 list_for_each_entry(lo, &loop_devices, lo_list) {
1464 if (lo->lo_number == i)
1465 return lo;
1466 }
1467
1468 lo = loop_alloc(i);
1469 if (lo) {
1470 add_disk(lo->lo_disk);
1471 list_add_tail(&lo->lo_list, &loop_devices);
1472 }
1473 return lo;
1474}
1475
1476static void loop_del_one(struct loop_device *lo)
1477{
1478 del_gendisk(lo->lo_disk);
1479 loop_free(lo);
1480}
1481
Ken Chen73285082007-05-08 00:28:20 -07001482static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1483{
Al Viro705962c2007-05-13 05:52:32 -04001484 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001485 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001486
Al Viro705962c2007-05-13 05:52:32 -04001487 mutex_lock(&loop_devices_mutex);
1488 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001489 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001490 mutex_unlock(&loop_devices_mutex);
1491
1492 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001493 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001494}
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496static int __init loop_init(void)
1497{
Ken Chena47653f2007-06-08 13:46:44 -07001498 int i, nr;
1499 unsigned long range;
1500 struct loop_device *lo, *next;
1501
1502 /*
1503 * loop module now has a feature to instantiate underlying device
1504 * structure on-demand, provided that there is an access dev node.
1505 * However, this will not work well with user space tool that doesn't
1506 * know about such "feature". In order to not break any existing
1507 * tool, we do the following:
1508 *
1509 * (1) if max_loop is specified, create that many upfront, and this
1510 * also becomes a hard limit.
1511 * (2) if max_loop is not specified, create 8 loop device on module
1512 * load, user can further extend loop device by create dev node
1513 * themselves and have kernel automatically instantiate actual
1514 * device on-demand.
1515 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001516
1517 part_shift = 0;
1518 if (max_part > 0)
1519 part_shift = fls(max_part);
1520
1521 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001522 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Ken Chen73285082007-05-08 00:28:20 -07001524 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001525 nr = max_loop;
1526 range = max_loop;
1527 } else {
1528 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001529 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
Ken Chena47653f2007-06-08 13:46:44 -07001531
1532 if (register_blkdev(LOOP_MAJOR, "loop"))
1533 return -EIO;
1534
1535 for (i = 0; i < nr; i++) {
1536 lo = loop_alloc(i);
1537 if (!lo)
1538 goto Enomem;
1539 list_add_tail(&lo->lo_list, &loop_devices);
1540 }
1541
1542 /* point of no return */
1543
1544 list_for_each_entry(lo, &loop_devices, lo_list)
1545 add_disk(lo->lo_disk);
1546
1547 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1548 THIS_MODULE, loop_probe, NULL, NULL);
1549
Ken Chen73285082007-05-08 00:28:20 -07001550 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001552
1553Enomem:
1554 printk(KERN_INFO "loop: out of memory\n");
1555
1556 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1557 loop_free(lo);
1558
1559 unregister_blkdev(LOOP_MAJOR, "loop");
1560 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561}
1562
Ken Chen73285082007-05-08 00:28:20 -07001563static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564{
Ken Chena47653f2007-06-08 13:46:44 -07001565 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001566 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Laurent Vivier476a4812008-03-26 12:11:53 +01001568 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001569
Ken Chen73285082007-05-08 00:28:20 -07001570 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1571 loop_del_one(lo);
1572
Ken Chena47653f2007-06-08 13:46:44 -07001573 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001574 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575}
1576
1577module_init(loop_init);
1578module_exit(loop_exit);
1579
1580#ifndef MODULE
1581static int __init max_loop_setup(char *str)
1582{
1583 max_loop = simple_strtol(str, NULL, 0);
1584 return 1;
1585}
1586
1587__setup("max_loop=", max_loop_setup);
1588#endif