blob: b8af22e610dfe4f90ac1b637f0670ec5f9e6bfee [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 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
Nick Pigginafddba42007-10-16 01:25:01 -0700207 * space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
209static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
Nick Pigginafddba42007-10-16 01:25:01 -0700210 int bsize, loff_t pos, struct page *unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
212 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
213 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 pgoff_t index;
215 unsigned offset, bv_offs;
Zach Brown994fc28c2005-12-15 14:28:17 -0800216 int len, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800218 mutex_lock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 index = pos >> PAGE_CACHE_SHIFT;
220 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
221 bv_offs = bvec->bv_offset;
222 len = bvec->bv_len;
223 while (len > 0) {
224 sector_t IV;
Nick Pigginafddba42007-10-16 01:25:01 -0700225 unsigned size, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 int transfer_result;
Nick Pigginafddba42007-10-16 01:25:01 -0700227 struct page *page;
228 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
231 size = PAGE_CACHE_SIZE - offset;
232 if (size > len)
233 size = len;
Nick Pigginafddba42007-10-16 01:25:01 -0700234
235 ret = pagecache_write_begin(file, mapping, pos, size, 0,
236 &page, &fsdata);
237 if (ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
241 bvec->bv_page, bv_offs, size, IV);
Nick Pigginafddba42007-10-16 01:25:01 -0700242 copied = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (unlikely(transfer_result))
Nick Pigginafddba42007-10-16 01:25:01 -0700244 copied = 0;
245
246 ret = pagecache_write_end(file, mapping, pos, size, copied,
247 page, fsdata);
Dmitry Monakhov8268f5a2007-10-16 01:25:02 -0700248 if (ret < 0 || ret != copied)
Nick Pigginafddba42007-10-16 01:25:01 -0700249 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700250
251 if (unlikely(transfer_result))
252 goto fail;
253
254 bv_offs += copied;
255 len -= copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 offset = 0;
257 index++;
Nick Pigginafddba42007-10-16 01:25:01 -0700258 pos += copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800260 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800262 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264fail:
265 ret = -1;
266 goto out;
267}
268
269/**
270 * __do_lo_send_write - helper for writing data to a loop device
271 *
272 * This helper just factors out common code between do_lo_send_direct_write()
273 * and do_lo_send_write().
274 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800275static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100276 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
278 ssize_t bw;
279 mm_segment_t old_fs = get_fs();
280
281 set_fs(get_ds());
282 bw = file->f_op->write(file, buf, len, &pos);
283 set_fs(old_fs);
284 if (likely(bw == len))
285 return 0;
286 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
287 (unsigned long long)pos, len);
288 if (bw >= 0)
289 bw = -EIO;
290 return bw;
291}
292
293/**
294 * do_lo_send_direct_write - helper for writing data to a loop device
295 *
296 * This is the fast, non-transforming version for backing filesystems which do
Nick Pigginafddba42007-10-16 01:25:01 -0700297 * not implement the address space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 * It uses the write file operation which should be present on all writeable
299 * filesystems.
300 */
301static int do_lo_send_direct_write(struct loop_device *lo,
302 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
303{
304 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100305 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 bvec->bv_len, pos);
307 kunmap(bvec->bv_page);
308 cond_resched();
309 return bw;
310}
311
312/**
313 * do_lo_send_write - helper for writing data to a loop device
314 *
315 * This is the slow, transforming version for filesystems which do not
Nick Pigginafddba42007-10-16 01:25:01 -0700316 * implement the address space operations write_begin and write_end. It
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 * uses the write file operation which should be present on all writeable
318 * filesystems.
319 *
320 * Using fops->write is slower than using aops->{prepare,commit}_write in the
321 * transforming case because we need to double buffer the data as we cannot do
322 * the transformations in place as we do not have direct access to the
323 * destination pages of the backing file.
324 */
325static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
326 int bsize, loff_t pos, struct page *page)
327{
328 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
329 bvec->bv_offset, bvec->bv_len, pos >> 9);
330 if (likely(!ret))
331 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100332 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 pos);
334 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
335 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
336 if (ret > 0)
337 ret = -EIO;
338 return ret;
339}
340
341static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
342 loff_t pos)
343{
344 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
345 struct page *page);
346 struct bio_vec *bvec;
347 struct page *page = NULL;
348 int i, ret = 0;
349
350 do_lo_send = do_lo_send_aops;
351 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
352 do_lo_send = do_lo_send_direct_write;
353 if (lo->transfer != transfer_none) {
354 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
355 if (unlikely(!page))
356 goto fail;
357 kmap(page);
358 do_lo_send = do_lo_send_write;
359 }
360 }
361 bio_for_each_segment(bvec, bio, i) {
362 ret = do_lo_send(lo, bvec, bsize, pos, page);
363 if (ret < 0)
364 break;
365 pos += bvec->bv_len;
366 }
367 if (page) {
368 kunmap(page);
369 __free_page(page);
370 }
371out:
372 return ret;
373fail:
374 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
375 ret = -ENOMEM;
376 goto out;
377}
378
379struct lo_read_data {
380 struct loop_device *lo;
381 struct page *page;
382 unsigned offset;
383 int bsize;
384};
385
386static int
Jens Axboefd582142007-06-12 21:20:37 +0200387lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
388 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Jens Axboefd582142007-06-12 21:20:37 +0200390 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200392 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 sector_t IV;
Jens Axboefd582142007-06-12 21:20:37 +0200394 size_t size;
395 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Jens Axboecac36bb02007-06-14 13:10:48 +0200397 ret = buf->ops->confirm(pipe, buf);
Jens Axboefd582142007-06-12 21:20:37 +0200398 if (unlikely(ret))
399 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Jens Axboefd582142007-06-12 21:20:37 +0200401 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
402 (buf->offset >> 9);
403 size = sd->len;
404 if (size > p->bsize)
405 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Jens Axboefd582142007-06-12 21:20:37 +0200407 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 printk(KERN_ERR "loop: transfer error block %ld\n",
409 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200410 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
412
413 flush_dcache_page(p->page);
414
Jens Axboefd582142007-06-12 21:20:37 +0200415 if (size > 0)
416 p->offset += size;
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return size;
419}
420
421static int
Jens Axboefd582142007-06-12 21:20:37 +0200422lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
423{
424 return __splice_from_pipe(pipe, sd, lo_splice_actor);
425}
426
427static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428do_lo_receive(struct loop_device *lo,
429 struct bio_vec *bvec, int bsize, loff_t pos)
430{
431 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200432 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200434 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 cookie.lo = lo;
437 cookie.page = bvec->bv_page;
438 cookie.offset = bvec->bv_offset;
439 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200440
441 sd.len = 0;
442 sd.total_len = bvec->bv_len;
443 sd.flags = 0;
444 sd.pos = pos;
445 sd.u.data = &cookie;
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200448 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
449
450 if (retval < 0)
451 return retval;
452
453 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456static int
457lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
458{
459 struct bio_vec *bvec;
460 int i, ret = 0;
461
462 bio_for_each_segment(bvec, bio, i) {
463 ret = do_lo_receive(lo, bvec, bsize, pos);
464 if (ret < 0)
465 break;
466 pos += bvec->bv_len;
467 }
468 return ret;
469}
470
471static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
472{
473 loff_t pos;
474 int ret;
475
476 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
477 if (bio_rw(bio) == WRITE)
478 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
479 else
480 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
481 return ret;
482}
483
484/*
485 * Add bio to back of pending list
486 */
487static void loop_add_bio(struct loop_device *lo, struct bio *bio)
488{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (lo->lo_biotail) {
490 lo->lo_biotail->bi_next = bio;
491 lo->lo_biotail = bio;
492 } else
493 lo->lo_bio = lo->lo_biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
496/*
497 * Grab first pending buffer
498 */
499static struct bio *loop_get_bio(struct loop_device *lo)
500{
501 struct bio *bio;
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if ((bio = lo->lo_bio)) {
504 if (bio == lo->lo_biotail)
505 lo->lo_biotail = NULL;
506 lo->lo_bio = bio->bi_next;
507 bio->bi_next = NULL;
508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 return bio;
511}
512
Jens Axboe165125e2007-07-24 09:28:11 +0200513static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
515 struct loop_device *lo = q->queuedata;
516 int rw = bio_rw(old_bio);
517
Nick Piggin35a82d12005-06-23 00:09:06 -0700518 if (rw == READA)
519 rw = READ;
520
521 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 spin_lock_irq(&lo->lo_lock);
524 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700525 goto out;
526 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
527 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700529 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700530 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700534 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200535 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
539/*
540 * kick off io on the underlying address space
541 */
Jens Axboe165125e2007-07-24 09:28:11 +0200542static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
544 struct loop_device *lo = q->queuedata;
545
546 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
547 blk_run_address_space(lo->lo_backing_file->f_mapping);
548}
549
550struct switch_request {
551 struct file *file;
552 struct completion wait;
553};
554
555static void do_loop_switch(struct loop_device *, struct switch_request *);
556
557static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
558{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (unlikely(!bio->bi_bdev)) {
560 do_loop_switch(lo, bio->bi_private);
561 bio_put(bio);
562 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700563 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200564 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566}
567
568/*
569 * worker thread that handles reads/writes to file backed loop devices,
570 * to avoid blocking in our make_request_fn. it also does loop decrypting
571 * on reads for block backed loop, as that is too heavy to do from
572 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700573 *
574 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
575 * calling kthread_stop(). Therefore once kthread_should_stop() is
576 * true, make_request will not place any more requests. Therefore
577 * once kthread_should_stop() is true and lo_bio is NULL, we are
578 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 */
580static int loop_thread(void *data)
581{
582 struct loop_device *lo = data;
583 struct bio *bio;
584
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 set_user_nice(current, -20);
586
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700587 while (!kthread_should_stop() || lo->lo_bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700589 wait_event_interruptible(lo->lo_event,
590 lo->lo_bio || kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700591
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700592 if (!lo->lo_bio)
Nick Piggin35a82d12005-06-23 00:09:06 -0700593 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700594 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700596 spin_unlock_irq(&lo->lo_lock);
597
598 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 }
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return 0;
603}
604
605/*
606 * loop_switch performs the hard work of switching a backing store.
607 * First it needs to flush existing IO, it does this by sending a magic
608 * BIO down the pipe. The completion of this BIO does the actual switch.
609 */
610static int loop_switch(struct loop_device *lo, struct file *file)
611{
612 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100613 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (!bio)
615 return -ENOMEM;
616 init_completion(&w.wait);
617 w.file = file;
618 bio->bi_private = &w;
619 bio->bi_bdev = NULL;
620 loop_make_request(lo->lo_queue, bio);
621 wait_for_completion(&w.wait);
622 return 0;
623}
624
625/*
626 * Do the actual switch; called from the BIO completion routine
627 */
628static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
629{
630 struct file *file = p->file;
631 struct file *old_file = lo->lo_backing_file;
632 struct address_space *mapping = file->f_mapping;
633
634 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
635 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700636 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
637 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 lo->old_gfp_mask = mapping_gfp_mask(mapping);
639 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
640 complete(&p->wait);
641}
642
643
644/*
645 * loop_change_fd switched the backing store of a loopback device to
646 * a new file. This is useful for operating system installers to free up
647 * the original file and in High Availability environments to switch to
648 * an alternative location for the content in case of server meltdown.
649 * This can only work if the loop device is used read-only, and if the
650 * new backing store is the same size and type as the old backing store.
651 */
652static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
653 struct block_device *bdev, unsigned int arg)
654{
655 struct file *file, *old_file;
656 struct inode *inode;
657 int error;
658
659 error = -ENXIO;
660 if (lo->lo_state != Lo_bound)
661 goto out;
662
663 /* the loop device has to be read-only */
664 error = -EINVAL;
665 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
666 goto out;
667
668 error = -EBADF;
669 file = fget(arg);
670 if (!file)
671 goto out;
672
673 inode = file->f_mapping->host;
674 old_file = lo->lo_backing_file;
675
676 error = -EINVAL;
677
678 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
679 goto out_putf;
680
Jens Axboefd582142007-06-12 21:20:37 +0200681 /* new backing store needs to support loop (eg splice_read) */
682 if (!inode->i_fop->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 goto out_putf;
684
685 /* size of the new backing store needs to be the same */
686 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
687 goto out_putf;
688
689 /* and ... switch */
690 error = loop_switch(lo, file);
691 if (error)
692 goto out_putf;
693
694 fput(old_file);
695 return 0;
696
697 out_putf:
698 fput(file);
699 out:
700 return error;
701}
702
703static inline int is_loop_device(struct file *file)
704{
705 struct inode *i = file->f_mapping->host;
706
707 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
708}
709
710static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
711 struct block_device *bdev, unsigned int arg)
712{
713 struct file *file, *f;
714 struct inode *inode;
715 struct address_space *mapping;
716 unsigned lo_blocksize;
717 int lo_flags = 0;
718 int error;
719 loff_t size;
720
721 /* This is safe, since we have a reference from open(). */
722 __module_get(THIS_MODULE);
723
724 error = -EBADF;
725 file = fget(arg);
726 if (!file)
727 goto out;
728
729 error = -EBUSY;
730 if (lo->lo_state != Lo_unbound)
731 goto out_putf;
732
733 /* Avoid recursion */
734 f = file;
735 while (is_loop_device(f)) {
736 struct loop_device *l;
737
738 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
739 goto out_putf;
740
741 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
742 if (l->lo_state == Lo_unbound) {
743 error = -EINVAL;
744 goto out_putf;
745 }
746 f = l->lo_backing_file;
747 }
748
749 mapping = file->f_mapping;
750 inode = mapping->host;
751
752 if (!(file->f_mode & FMODE_WRITE))
753 lo_flags |= LO_FLAGS_READ_ONLY;
754
755 error = -EINVAL;
756 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700757 const struct address_space_operations *aops = mapping->a_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 /*
759 * If we can't read - sorry. If we only can't write - well,
760 * it's going to be read-only.
761 */
Jens Axboefd582142007-06-12 21:20:37 +0200762 if (!file->f_op->splice_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 goto out_putf;
Nick Pigginafddba42007-10-16 01:25:01 -0700764 if (aops->prepare_write || aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 lo_flags |= LO_FLAGS_USE_AOPS;
766 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
767 lo_flags |= LO_FLAGS_READ_ONLY;
768
Theodore Ts'oba52de12006-09-27 01:50:49 -0700769 lo_blocksize = S_ISBLK(inode->i_mode) ?
770 inode->i_bdev->bd_block_size : PAGE_SIZE;
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 error = 0;
773 } else {
774 goto out_putf;
775 }
776
777 size = get_loop_size(lo, file);
778
779 if ((loff_t)(sector_t)size != size) {
780 error = -EFBIG;
781 goto out_putf;
782 }
783
784 if (!(lo_file->f_mode & FMODE_WRITE))
785 lo_flags |= LO_FLAGS_READ_ONLY;
786
787 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
788
789 lo->lo_blocksize = lo_blocksize;
790 lo->lo_device = bdev;
791 lo->lo_flags = lo_flags;
792 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700793 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 lo->ioctl = NULL;
795 lo->lo_sizelimit = 0;
796 lo->old_gfp_mask = mapping_gfp_mask(mapping);
797 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
798
799 lo->lo_bio = lo->lo_biotail = NULL;
800
801 /*
802 * set queue make_request_fn, and add limits based on lower level
803 * device
804 */
805 blk_queue_make_request(lo->lo_queue, loop_make_request);
806 lo->lo_queue->queuedata = lo;
807 lo->lo_queue->unplug_fn = loop_unplug;
808
Ken Chen73285082007-05-08 00:28:20 -0700809 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 bd_set_size(bdev, size << 9);
811
812 set_blocksize(bdev, lo_blocksize);
813
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700814 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
815 lo->lo_number);
816 if (IS_ERR(lo->lo_thread)) {
817 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700818 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700819 }
820 lo->lo_state = Lo_bound;
821 wake_up_process(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 return 0;
823
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700824out_clr:
825 lo->lo_thread = NULL;
826 lo->lo_device = NULL;
827 lo->lo_backing_file = NULL;
828 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700829 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700830 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700831 bd_set_size(bdev, 0);
832 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
833 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 out_putf:
835 fput(file);
836 out:
837 /* This is safe: open() is still holding a reference. */
838 module_put(THIS_MODULE);
839 return error;
840}
841
842static int
843loop_release_xfer(struct loop_device *lo)
844{
845 int err = 0;
846 struct loop_func_table *xfer = lo->lo_encryption;
847
848 if (xfer) {
849 if (xfer->release)
850 err = xfer->release(lo);
851 lo->transfer = NULL;
852 lo->lo_encryption = NULL;
853 module_put(xfer->owner);
854 }
855 return err;
856}
857
858static int
859loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
860 const struct loop_info64 *i)
861{
862 int err = 0;
863
864 if (xfer) {
865 struct module *owner = xfer->owner;
866
867 if (!try_module_get(owner))
868 return -EINVAL;
869 if (xfer->init)
870 err = xfer->init(lo, i);
871 if (err)
872 module_put(owner);
873 else
874 lo->lo_encryption = xfer;
875 }
876 return err;
877}
878
879static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
880{
881 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -0400882 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 if (lo->lo_state != Lo_bound)
885 return -ENXIO;
886
887 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
888 return -EBUSY;
889
890 if (filp == NULL)
891 return -EINVAL;
892
893 spin_lock_irq(&lo->lo_lock);
894 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 spin_unlock_irq(&lo->lo_lock);
896
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700897 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 lo->lo_backing_file = NULL;
900
901 loop_release_xfer(lo);
902 lo->transfer = NULL;
903 lo->ioctl = NULL;
904 lo->lo_device = NULL;
905 lo->lo_encryption = NULL;
906 lo->lo_offset = 0;
907 lo->lo_sizelimit = 0;
908 lo->lo_encrypt_key_size = 0;
909 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700910 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
912 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
913 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700914 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -0700915 set_capacity(lo->lo_disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 bd_set_size(bdev, 0);
917 mapping_set_gfp_mask(filp->f_mapping, gfp);
918 lo->lo_state = Lo_unbound;
919 fput(filp);
920 /* This is safe: open() is still holding a reference. */
921 module_put(THIS_MODULE);
922 return 0;
923}
924
925static int
926loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
927{
928 int err;
929 struct loop_func_table *xfer;
930
931 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
932 !capable(CAP_SYS_ADMIN))
933 return -EPERM;
934 if (lo->lo_state != Lo_bound)
935 return -ENXIO;
936 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
937 return -EINVAL;
938
939 err = loop_release_xfer(lo);
940 if (err)
941 return err;
942
943 if (info->lo_encrypt_type) {
944 unsigned int type = info->lo_encrypt_type;
945
946 if (type >= MAX_LO_CRYPT)
947 return -EINVAL;
948 xfer = xfer_funcs[type];
949 if (xfer == NULL)
950 return -EINVAL;
951 } else
952 xfer = NULL;
953
954 err = loop_init_xfer(lo, xfer, info);
955 if (err)
956 return err;
957
958 if (lo->lo_offset != info->lo_offset ||
959 lo->lo_sizelimit != info->lo_sizelimit) {
960 lo->lo_offset = info->lo_offset;
961 lo->lo_sizelimit = info->lo_sizelimit;
962 if (figure_loop_size(lo))
963 return -EFBIG;
964 }
965
966 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
967 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
968 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
969 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
970
971 if (!xfer)
972 xfer = &none_funcs;
973 lo->transfer = xfer->transfer;
974 lo->ioctl = xfer->ioctl;
975
976 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
977 lo->lo_init[0] = info->lo_init[0];
978 lo->lo_init[1] = info->lo_init[1];
979 if (info->lo_encrypt_key_size) {
980 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
981 info->lo_encrypt_key_size);
982 lo->lo_key_owner = current->uid;
983 }
984
985 return 0;
986}
987
988static int
989loop_get_status(struct loop_device *lo, struct loop_info64 *info)
990{
991 struct file *file = lo->lo_backing_file;
992 struct kstat stat;
993 int error;
994
995 if (lo->lo_state != Lo_bound)
996 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -0800997 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 if (error)
999 return error;
1000 memset(info, 0, sizeof(*info));
1001 info->lo_number = lo->lo_number;
1002 info->lo_device = huge_encode_dev(stat.dev);
1003 info->lo_inode = stat.ino;
1004 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1005 info->lo_offset = lo->lo_offset;
1006 info->lo_sizelimit = lo->lo_sizelimit;
1007 info->lo_flags = lo->lo_flags;
1008 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1009 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1010 info->lo_encrypt_type =
1011 lo->lo_encryption ? lo->lo_encryption->number : 0;
1012 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1013 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1014 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1015 lo->lo_encrypt_key_size);
1016 }
1017 return 0;
1018}
1019
1020static void
1021loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1022{
1023 memset(info64, 0, sizeof(*info64));
1024 info64->lo_number = info->lo_number;
1025 info64->lo_device = info->lo_device;
1026 info64->lo_inode = info->lo_inode;
1027 info64->lo_rdevice = info->lo_rdevice;
1028 info64->lo_offset = info->lo_offset;
1029 info64->lo_sizelimit = 0;
1030 info64->lo_encrypt_type = info->lo_encrypt_type;
1031 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1032 info64->lo_flags = info->lo_flags;
1033 info64->lo_init[0] = info->lo_init[0];
1034 info64->lo_init[1] = info->lo_init[1];
1035 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1036 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1037 else
1038 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1039 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1040}
1041
1042static int
1043loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1044{
1045 memset(info, 0, sizeof(*info));
1046 info->lo_number = info64->lo_number;
1047 info->lo_device = info64->lo_device;
1048 info->lo_inode = info64->lo_inode;
1049 info->lo_rdevice = info64->lo_rdevice;
1050 info->lo_offset = info64->lo_offset;
1051 info->lo_encrypt_type = info64->lo_encrypt_type;
1052 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1053 info->lo_flags = info64->lo_flags;
1054 info->lo_init[0] = info64->lo_init[0];
1055 info->lo_init[1] = info64->lo_init[1];
1056 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1057 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1058 else
1059 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1060 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1061
1062 /* error in case values were truncated */
1063 if (info->lo_device != info64->lo_device ||
1064 info->lo_rdevice != info64->lo_rdevice ||
1065 info->lo_inode != info64->lo_inode ||
1066 info->lo_offset != info64->lo_offset)
1067 return -EOVERFLOW;
1068
1069 return 0;
1070}
1071
1072static int
1073loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1074{
1075 struct loop_info info;
1076 struct loop_info64 info64;
1077
1078 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1079 return -EFAULT;
1080 loop_info64_from_old(&info, &info64);
1081 return loop_set_status(lo, &info64);
1082}
1083
1084static int
1085loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1086{
1087 struct loop_info64 info64;
1088
1089 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1090 return -EFAULT;
1091 return loop_set_status(lo, &info64);
1092}
1093
1094static int
1095loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1096 struct loop_info info;
1097 struct loop_info64 info64;
1098 int err = 0;
1099
1100 if (!arg)
1101 err = -EINVAL;
1102 if (!err)
1103 err = loop_get_status(lo, &info64);
1104 if (!err)
1105 err = loop_info64_to_old(&info64, &info);
1106 if (!err && copy_to_user(arg, &info, sizeof(info)))
1107 err = -EFAULT;
1108
1109 return err;
1110}
1111
1112static int
1113loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1114 struct loop_info64 info64;
1115 int err = 0;
1116
1117 if (!arg)
1118 err = -EINVAL;
1119 if (!err)
1120 err = loop_get_status(lo, &info64);
1121 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1122 err = -EFAULT;
1123
1124 return err;
1125}
1126
1127static int lo_ioctl(struct inode * inode, struct file * file,
1128 unsigned int cmd, unsigned long arg)
1129{
1130 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1131 int err;
1132
Ingo Molnarf85221d2006-03-23 03:00:38 -08001133 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 switch (cmd) {
1135 case LOOP_SET_FD:
1136 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1137 break;
1138 case LOOP_CHANGE_FD:
1139 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1140 break;
1141 case LOOP_CLR_FD:
1142 err = loop_clr_fd(lo, inode->i_bdev);
1143 break;
1144 case LOOP_SET_STATUS:
1145 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1146 break;
1147 case LOOP_GET_STATUS:
1148 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1149 break;
1150 case LOOP_SET_STATUS64:
1151 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1152 break;
1153 case LOOP_GET_STATUS64:
1154 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1155 break;
1156 default:
1157 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1158 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001159 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 return err;
1161}
1162
David Howells863d5b822006-08-29 19:06:14 +01001163#ifdef CONFIG_COMPAT
1164struct compat_loop_info {
1165 compat_int_t lo_number; /* ioctl r/o */
1166 compat_dev_t lo_device; /* ioctl r/o */
1167 compat_ulong_t lo_inode; /* ioctl r/o */
1168 compat_dev_t lo_rdevice; /* ioctl r/o */
1169 compat_int_t lo_offset;
1170 compat_int_t lo_encrypt_type;
1171 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1172 compat_int_t lo_flags; /* ioctl r/o */
1173 char lo_name[LO_NAME_SIZE];
1174 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1175 compat_ulong_t lo_init[2];
1176 char reserved[4];
1177};
1178
1179/*
1180 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1181 * - noinlined to reduce stack space usage in main part of driver
1182 */
1183static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001184loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001185 struct loop_info64 *info64)
1186{
1187 struct compat_loop_info info;
1188
1189 if (copy_from_user(&info, arg, sizeof(info)))
1190 return -EFAULT;
1191
1192 memset(info64, 0, sizeof(*info64));
1193 info64->lo_number = info.lo_number;
1194 info64->lo_device = info.lo_device;
1195 info64->lo_inode = info.lo_inode;
1196 info64->lo_rdevice = info.lo_rdevice;
1197 info64->lo_offset = info.lo_offset;
1198 info64->lo_sizelimit = 0;
1199 info64->lo_encrypt_type = info.lo_encrypt_type;
1200 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1201 info64->lo_flags = info.lo_flags;
1202 info64->lo_init[0] = info.lo_init[0];
1203 info64->lo_init[1] = info.lo_init[1];
1204 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1205 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1206 else
1207 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1208 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1209 return 0;
1210}
1211
1212/*
1213 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1214 * - noinlined to reduce stack space usage in main part of driver
1215 */
1216static noinline int
1217loop_info64_to_compat(const struct loop_info64 *info64,
1218 struct compat_loop_info __user *arg)
1219{
1220 struct compat_loop_info info;
1221
1222 memset(&info, 0, sizeof(info));
1223 info.lo_number = info64->lo_number;
1224 info.lo_device = info64->lo_device;
1225 info.lo_inode = info64->lo_inode;
1226 info.lo_rdevice = info64->lo_rdevice;
1227 info.lo_offset = info64->lo_offset;
1228 info.lo_encrypt_type = info64->lo_encrypt_type;
1229 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1230 info.lo_flags = info64->lo_flags;
1231 info.lo_init[0] = info64->lo_init[0];
1232 info.lo_init[1] = info64->lo_init[1];
1233 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1234 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1235 else
1236 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1237 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1238
1239 /* error in case values were truncated */
1240 if (info.lo_device != info64->lo_device ||
1241 info.lo_rdevice != info64->lo_rdevice ||
1242 info.lo_inode != info64->lo_inode ||
1243 info.lo_offset != info64->lo_offset ||
1244 info.lo_init[0] != info64->lo_init[0] ||
1245 info.lo_init[1] != info64->lo_init[1])
1246 return -EOVERFLOW;
1247
1248 if (copy_to_user(arg, &info, sizeof(info)))
1249 return -EFAULT;
1250 return 0;
1251}
1252
1253static int
1254loop_set_status_compat(struct loop_device *lo,
1255 const struct compat_loop_info __user *arg)
1256{
1257 struct loop_info64 info64;
1258 int ret;
1259
1260 ret = loop_info64_from_compat(arg, &info64);
1261 if (ret < 0)
1262 return ret;
1263 return loop_set_status(lo, &info64);
1264}
1265
1266static int
1267loop_get_status_compat(struct loop_device *lo,
1268 struct compat_loop_info __user *arg)
1269{
1270 struct loop_info64 info64;
1271 int err = 0;
1272
1273 if (!arg)
1274 err = -EINVAL;
1275 if (!err)
1276 err = loop_get_status(lo, &info64);
1277 if (!err)
1278 err = loop_info64_to_compat(&info64, arg);
1279 return err;
1280}
1281
1282static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1283{
Josef Sipek6c648be2006-12-08 02:36:55 -08001284 struct inode *inode = file->f_path.dentry->d_inode;
David Howells863d5b822006-08-29 19:06:14 +01001285 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1286 int err;
1287
David Howells863d5b822006-08-29 19:06:14 +01001288 switch(cmd) {
1289 case LOOP_SET_STATUS:
1290 mutex_lock(&lo->lo_ctl_mutex);
1291 err = loop_set_status_compat(
1292 lo, (const struct compat_loop_info __user *) arg);
1293 mutex_unlock(&lo->lo_ctl_mutex);
1294 break;
1295 case LOOP_GET_STATUS:
1296 mutex_lock(&lo->lo_ctl_mutex);
1297 err = loop_get_status_compat(
1298 lo, (struct compat_loop_info __user *) arg);
1299 mutex_unlock(&lo->lo_ctl_mutex);
1300 break;
1301 case LOOP_CLR_FD:
1302 case LOOP_GET_STATUS64:
1303 case LOOP_SET_STATUS64:
1304 arg = (unsigned long) compat_ptr(arg);
1305 case LOOP_SET_FD:
1306 case LOOP_CHANGE_FD:
1307 err = lo_ioctl(inode, file, cmd, arg);
1308 break;
1309 default:
1310 err = -ENOIOCTLCMD;
1311 break;
1312 }
David Howells863d5b822006-08-29 19:06:14 +01001313 return err;
1314}
1315#endif
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317static int lo_open(struct inode *inode, struct file *file)
1318{
1319 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1320
Ingo Molnarf85221d2006-03-23 03:00:38 -08001321 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001323 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 return 0;
1326}
1327
1328static int lo_release(struct inode *inode, struct file *file)
1329{
1330 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1331
Ingo Molnarf85221d2006-03-23 03:00:38 -08001332 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 --lo->lo_refcnt;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001334 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 return 0;
1337}
1338
1339static struct block_device_operations lo_fops = {
1340 .owner = THIS_MODULE,
1341 .open = lo_open,
1342 .release = lo_release,
1343 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001344#ifdef CONFIG_COMPAT
1345 .compat_ioctl = lo_compat_ioctl,
1346#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347};
1348
1349/*
1350 * And now the modules code and kernel interface.
1351 */
Ken Chen73285082007-05-08 00:28:20 -07001352static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001354MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355MODULE_LICENSE("GPL");
1356MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1357
1358int loop_register_transfer(struct loop_func_table *funcs)
1359{
1360 unsigned int n = funcs->number;
1361
1362 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1363 return -EINVAL;
1364 xfer_funcs[n] = funcs;
1365 return 0;
1366}
1367
1368int loop_unregister_transfer(int number)
1369{
1370 unsigned int n = number;
1371 struct loop_device *lo;
1372 struct loop_func_table *xfer;
1373
1374 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1375 return -EINVAL;
1376
1377 xfer_funcs[n] = NULL;
1378
Ken Chen73285082007-05-08 00:28:20 -07001379 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001380 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 if (lo->lo_encryption == xfer)
1383 loop_release_xfer(lo);
1384
Ingo Molnarf85221d2006-03-23 03:00:38 -08001385 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 }
1387
1388 return 0;
1389}
1390
1391EXPORT_SYMBOL(loop_register_transfer);
1392EXPORT_SYMBOL(loop_unregister_transfer);
1393
Ken Chena47653f2007-06-08 13:46:44 -07001394static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001395{
1396 struct loop_device *lo;
1397 struct gendisk *disk;
1398
1399 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1400 if (!lo)
1401 goto out;
1402
1403 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1404 if (!lo->lo_queue)
1405 goto out_free_dev;
1406
1407 disk = lo->lo_disk = alloc_disk(1);
1408 if (!disk)
1409 goto out_free_queue;
1410
1411 mutex_init(&lo->lo_ctl_mutex);
1412 lo->lo_number = i;
1413 lo->lo_thread = NULL;
1414 init_waitqueue_head(&lo->lo_event);
1415 spin_lock_init(&lo->lo_lock);
1416 disk->major = LOOP_MAJOR;
1417 disk->first_minor = i;
1418 disk->fops = &lo_fops;
1419 disk->private_data = lo;
1420 disk->queue = lo->lo_queue;
1421 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001422 return lo;
1423
1424out_free_queue:
1425 blk_cleanup_queue(lo->lo_queue);
1426out_free_dev:
1427 kfree(lo);
1428out:
Al Viro07002e92007-05-12 16:23:15 -04001429 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001430}
1431
Ken Chena47653f2007-06-08 13:46:44 -07001432static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001433{
Ken Chen73285082007-05-08 00:28:20 -07001434 blk_cleanup_queue(lo->lo_queue);
1435 put_disk(lo->lo_disk);
1436 list_del(&lo->lo_list);
1437 kfree(lo);
1438}
1439
Ken Chena47653f2007-06-08 13:46:44 -07001440static struct loop_device *loop_init_one(int i)
1441{
1442 struct loop_device *lo;
1443
1444 list_for_each_entry(lo, &loop_devices, lo_list) {
1445 if (lo->lo_number == i)
1446 return lo;
1447 }
1448
1449 lo = loop_alloc(i);
1450 if (lo) {
1451 add_disk(lo->lo_disk);
1452 list_add_tail(&lo->lo_list, &loop_devices);
1453 }
1454 return lo;
1455}
1456
1457static void loop_del_one(struct loop_device *lo)
1458{
1459 del_gendisk(lo->lo_disk);
1460 loop_free(lo);
1461}
1462
Ken Chen73285082007-05-08 00:28:20 -07001463static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1464{
Al Viro705962c2007-05-13 05:52:32 -04001465 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001466 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001467
Al Viro705962c2007-05-13 05:52:32 -04001468 mutex_lock(&loop_devices_mutex);
1469 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001470 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001471 mutex_unlock(&loop_devices_mutex);
1472
1473 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001474 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001475}
1476
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477static int __init loop_init(void)
1478{
Ken Chena47653f2007-06-08 13:46:44 -07001479 int i, nr;
1480 unsigned long range;
1481 struct loop_device *lo, *next;
1482
1483 /*
1484 * loop module now has a feature to instantiate underlying device
1485 * structure on-demand, provided that there is an access dev node.
1486 * However, this will not work well with user space tool that doesn't
1487 * know about such "feature". In order to not break any existing
1488 * tool, we do the following:
1489 *
1490 * (1) if max_loop is specified, create that many upfront, and this
1491 * also becomes a hard limit.
1492 * (2) if max_loop is not specified, create 8 loop device on module
1493 * load, user can further extend loop device by create dev node
1494 * themselves and have kernel automatically instantiate actual
1495 * device on-demand.
1496 */
1497 if (max_loop > 1UL << MINORBITS)
1498 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Ken Chen73285082007-05-08 00:28:20 -07001500 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001501 nr = max_loop;
1502 range = max_loop;
1503 } else {
1504 nr = 8;
1505 range = 1UL << MINORBITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
Ken Chena47653f2007-06-08 13:46:44 -07001507
1508 if (register_blkdev(LOOP_MAJOR, "loop"))
1509 return -EIO;
1510
1511 for (i = 0; i < nr; i++) {
1512 lo = loop_alloc(i);
1513 if (!lo)
1514 goto Enomem;
1515 list_add_tail(&lo->lo_list, &loop_devices);
1516 }
1517
1518 /* point of no return */
1519
1520 list_for_each_entry(lo, &loop_devices, lo_list)
1521 add_disk(lo->lo_disk);
1522
1523 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1524 THIS_MODULE, loop_probe, NULL, NULL);
1525
Ken Chen73285082007-05-08 00:28:20 -07001526 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001528
1529Enomem:
1530 printk(KERN_INFO "loop: out of memory\n");
1531
1532 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1533 loop_free(lo);
1534
1535 unregister_blkdev(LOOP_MAJOR, "loop");
1536 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538
Ken Chen73285082007-05-08 00:28:20 -07001539static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Ken Chena47653f2007-06-08 13:46:44 -07001541 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001542 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Ken Chena47653f2007-06-08 13:46:44 -07001544 range = max_loop ? max_loop : 1UL << MINORBITS;
1545
Ken Chen73285082007-05-08 00:28:20 -07001546 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1547 loop_del_one(lo);
1548
Ken Chena47653f2007-06-08 13:46:44 -07001549 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001550 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
1553module_init(loop_init);
1554module_exit(loop_exit);
1555
1556#ifndef MODULE
1557static int __init max_loop_setup(char *str)
1558{
1559 max_loop = simple_strtol(str, NULL, 0);
1560 return 1;
1561}
1562
1563__setup("max_loop=", max_loop_setup);
1564#endif