blob: c87b0840d2cb025f4e385a48e809c92174b9f456 [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
Nick Piggin4e02ed42008-10-29 14:00:55 -070043 * operations write_begin is not available on the backing filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * Anton Altaparmakov, 16 Feb 2005
45 *
46 * Still To Fix:
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
49 *
50 */
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/module.h>
53#include <linux/moduleparam.h>
54#include <linux/sched.h>
55#include <linux/fs.h>
56#include <linux/file.h>
57#include <linux/stat.h>
58#include <linux/errno.h>
59#include <linux/major.h>
60#include <linux/wait.h>
61#include <linux/blkdev.h>
62#include <linux/blkpg.h>
63#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/swap.h>
65#include <linux/slab.h>
66#include <linux/loop.h>
David Howells863d5b822006-08-29 19:06:14 +010067#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070069#include <linux/freezer.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020070#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <linux/writeback.h>
72#include <linux/buffer_head.h> /* for invalidate_bdev() */
73#include <linux/completion.h>
74#include <linux/highmem.h>
Serge E. Hallyn6c997912006-09-29 01:59:11 -070075#include <linux/kthread.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020076#include <linux/splice.h>
Milan Brozee862732010-08-23 15:16:00 +020077#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79#include <asm/uaccess.h>
80
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020081static DEFINE_MUTEX(loop_mutex);
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 kunmap_atomic(loop_buf, KM_USER1);
Peter Zijlstra61ecdb802010-10-26 14:21:47 -0700105 kunmap_atomic(raw_buf, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 kunmap_atomic(loop_buf, KM_USER1);
Peter Zijlstra61ecdb802010-10-26 14:21:47 -0700134 kunmap_atomic(raw_buf, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 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,
Al Viro511de732007-10-08 12:10:13 -0400213 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
Nikanth Karthikesan02246c42010-04-08 21:39:31 +0200243 file_update_time(file);
244
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
246 bvec->bv_page, bv_offs, size, IV);
Nick Pigginafddba42007-10-16 01:25:01 -0700247 copied = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (unlikely(transfer_result))
Nick Pigginafddba42007-10-16 01:25:01 -0700249 copied = 0;
250
251 ret = pagecache_write_end(file, mapping, pos, size, copied,
252 page, fsdata);
Dmitry Monakhov8268f5a2007-10-16 01:25:02 -0700253 if (ret < 0 || ret != copied)
Nick Pigginafddba42007-10-16 01:25:01 -0700254 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -0700255
256 if (unlikely(transfer_result))
257 goto fail;
258
259 bv_offs += copied;
260 len -= copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 offset = 0;
262 index++;
Nick Pigginafddba42007-10-16 01:25:01 -0700263 pos += copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800265 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800267 mutex_unlock(&mapping->host->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269fail:
270 ret = -1;
271 goto out;
272}
273
274/**
275 * __do_lo_send_write - helper for writing data to a loop device
276 *
277 * This helper just factors out common code between do_lo_send_direct_write()
278 * and do_lo_send_write().
279 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800280static int __do_lo_send_write(struct file *file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100281 u8 *buf, const int len, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
283 ssize_t bw;
284 mm_segment_t old_fs = get_fs();
285
286 set_fs(get_ds());
287 bw = file->f_op->write(file, buf, len, &pos);
288 set_fs(old_fs);
289 if (likely(bw == len))
290 return 0;
291 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)pos, len);
293 if (bw >= 0)
294 bw = -EIO;
295 return bw;
296}
297
298/**
299 * do_lo_send_direct_write - helper for writing data to a loop device
300 *
301 * This is the fast, non-transforming version for backing filesystems which do
Nick Pigginafddba42007-10-16 01:25:01 -0700302 * not implement the address space operations write_begin and write_end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 * It uses the write file operation which should be present on all writeable
304 * filesystems.
305 */
306static int do_lo_send_direct_write(struct loop_device *lo,
Al Viro511de732007-10-08 12:10:13 -0400307 struct bio_vec *bvec, loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100310 kmap(bvec->bv_page) + bvec->bv_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 bvec->bv_len, pos);
312 kunmap(bvec->bv_page);
313 cond_resched();
314 return bw;
315}
316
317/**
318 * do_lo_send_write - helper for writing data to a loop device
319 *
320 * This is the slow, transforming version for filesystems which do not
Nick Pigginafddba42007-10-16 01:25:01 -0700321 * implement the address space operations write_begin and write_end. It
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 * uses the write file operation which should be present on all writeable
323 * filesystems.
324 *
325 * Using fops->write is slower than using aops->{prepare,commit}_write in the
326 * transforming case because we need to double buffer the data as we cannot do
327 * the transformations in place as we do not have direct access to the
328 * destination pages of the backing file.
329 */
330static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
Al Viro511de732007-10-08 12:10:13 -0400331 loff_t pos, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
333 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
334 bvec->bv_offset, bvec->bv_len, pos >> 9);
335 if (likely(!ret))
336 return __do_lo_send_write(lo->lo_backing_file,
Al Viro98ae6ccd2006-10-10 22:45:07 +0100337 page_address(page), bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 pos);
339 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
340 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
341 if (ret > 0)
342 ret = -EIO;
343 return ret;
344}
345
Al Viro511de732007-10-08 12:10:13 -0400346static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Al Viro511de732007-10-08 12:10:13 -0400348 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 struct page *page);
350 struct bio_vec *bvec;
351 struct page *page = NULL;
352 int i, ret = 0;
353
354 do_lo_send = do_lo_send_aops;
355 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
356 do_lo_send = do_lo_send_direct_write;
357 if (lo->transfer != transfer_none) {
358 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
359 if (unlikely(!page))
360 goto fail;
361 kmap(page);
362 do_lo_send = do_lo_send_write;
363 }
364 }
365 bio_for_each_segment(bvec, bio, i) {
Al Viro511de732007-10-08 12:10:13 -0400366 ret = do_lo_send(lo, bvec, pos, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (ret < 0)
368 break;
369 pos += bvec->bv_len;
370 }
371 if (page) {
372 kunmap(page);
373 __free_page(page);
374 }
375out:
376 return ret;
377fail:
378 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
379 ret = -ENOMEM;
380 goto out;
381}
382
383struct lo_read_data {
384 struct loop_device *lo;
385 struct page *page;
386 unsigned offset;
387 int bsize;
388};
389
390static int
Jens Axboefd582142007-06-12 21:20:37 +0200391lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
392 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Jens Axboefd582142007-06-12 21:20:37 +0200394 struct lo_read_data *p = sd->u.data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 struct loop_device *lo = p->lo;
Jens Axboefd582142007-06-12 21:20:37 +0200396 struct page *page = buf->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 sector_t IV;
Roel Kluina3941ec2009-03-05 08:03:53 +0100398 int size, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Jens Axboefd582142007-06-12 21:20:37 +0200400 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
401 (buf->offset >> 9);
402 size = sd->len;
403 if (size > p->bsize)
404 size = p->bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Jens Axboefd582142007-06-12 21:20:37 +0200406 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 printk(KERN_ERR "loop: transfer error block %ld\n",
408 page->index);
Jens Axboefd582142007-06-12 21:20:37 +0200409 size = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
411
412 flush_dcache_page(p->page);
413
Jens Axboefd582142007-06-12 21:20:37 +0200414 if (size > 0)
415 p->offset += size;
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 return size;
418}
419
420static int
Jens Axboefd582142007-06-12 21:20:37 +0200421lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
422{
423 return __splice_from_pipe(pipe, sd, lo_splice_actor);
424}
425
426static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427do_lo_receive(struct loop_device *lo,
428 struct bio_vec *bvec, int bsize, loff_t pos)
429{
430 struct lo_read_data cookie;
Jens Axboefd582142007-06-12 21:20:37 +0200431 struct splice_desc sd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct file *file;
Jens Axboefd582142007-06-12 21:20:37 +0200433 long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
435 cookie.lo = lo;
436 cookie.page = bvec->bv_page;
437 cookie.offset = bvec->bv_offset;
438 cookie.bsize = bsize;
Jens Axboefd582142007-06-12 21:20:37 +0200439
440 sd.len = 0;
441 sd.total_len = bvec->bv_len;
442 sd.flags = 0;
443 sd.pos = pos;
444 sd.u.data = &cookie;
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 file = lo->lo_backing_file;
Jens Axboefd582142007-06-12 21:20:37 +0200447 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
448
449 if (retval < 0)
450 return retval;
451
452 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
455static int
456lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
457{
458 struct bio_vec *bvec;
459 int i, ret = 0;
460
461 bio_for_each_segment(bvec, bio, i) {
462 ret = do_lo_receive(lo, bvec, bsize, pos);
463 if (ret < 0)
464 break;
465 pos += bvec->bv_len;
466 }
467 return ret;
468}
469
470static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
471{
472 loff_t pos;
473 int ret;
474
475 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100476
477 if (bio_rw(bio) == WRITE) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100478 struct file *file = lo->lo_backing_file;
479
Tejun Heo6259f282010-09-03 11:56:17 +0200480 if (bio->bi_rw & REQ_FLUSH) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100481 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200482 if (unlikely(ret && ret != -EINVAL)) {
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100483 ret = -EIO;
484 goto out;
485 }
486 }
487
Al Viro511de732007-10-08 12:10:13 -0400488 ret = lo_send(lo, bio, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100489
Tejun Heo6259f282010-09-03 11:56:17 +0200490 if ((bio->bi_rw & REQ_FUA) && !ret) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100491 ret = vfs_fsync(file, 0);
Tejun Heo6259f282010-09-03 11:56:17 +0200492 if (unlikely(ret && ret != -EINVAL))
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100493 ret = -EIO;
494 }
495 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100497
498out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 return ret;
500}
501
502/*
503 * Add bio to back of pending list
504 */
505static void loop_add_bio(struct loop_device *lo, struct bio *bio)
506{
Akinobu Mitae6863072009-04-17 08:41:21 +0200507 bio_list_add(&lo->lo_bio_list, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
510/*
511 * Grab first pending buffer
512 */
513static struct bio *loop_get_bio(struct loop_device *lo)
514{
Akinobu Mitae6863072009-04-17 08:41:21 +0200515 return bio_list_pop(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
Jens Axboe165125e2007-07-24 09:28:11 +0200518static int loop_make_request(struct request_queue *q, struct bio *old_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
520 struct loop_device *lo = q->queuedata;
521 int rw = bio_rw(old_bio);
522
Nick Piggin35a82d12005-06-23 00:09:06 -0700523 if (rw == READA)
524 rw = READ;
525
526 BUG_ON(!lo || (rw != READ && rw != WRITE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 spin_lock_irq(&lo->lo_lock);
529 if (lo->lo_state != Lo_bound)
Nick Piggin35a82d12005-06-23 00:09:06 -0700530 goto out;
531 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
532 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 loop_add_bio(lo, old_bio);
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700534 wake_up(&lo->lo_event);
Nick Piggin35a82d12005-06-23 00:09:06 -0700535 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return 0;
Nick Piggin35a82d12005-06-23 00:09:06 -0700537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538out:
Nick Piggin35a82d12005-06-23 00:09:06 -0700539 spin_unlock_irq(&lo->lo_lock);
NeilBrown6712ecf2007-09-27 12:47:43 +0200540 bio_io_error(old_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
544/*
545 * kick off io on the underlying address space
546 */
Jens Axboe165125e2007-07-24 09:28:11 +0200547static void loop_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549 struct loop_device *lo = q->queuedata;
550
Nick Piggin75ad23b2008-04-29 14:48:33 +0200551 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 blk_run_address_space(lo->lo_backing_file->f_mapping);
553}
554
555struct switch_request {
556 struct file *file;
557 struct completion wait;
558};
559
560static void do_loop_switch(struct loop_device *, struct switch_request *);
561
562static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
563{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 if (unlikely(!bio->bi_bdev)) {
565 do_loop_switch(lo, bio->bi_private);
566 bio_put(bio);
567 } else {
Nick Piggin35a82d12005-06-23 00:09:06 -0700568 int ret = do_bio_filebacked(lo, bio);
NeilBrown6712ecf2007-09-27 12:47:43 +0200569 bio_endio(bio, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
571}
572
573/*
574 * worker thread that handles reads/writes to file backed loop devices,
575 * to avoid blocking in our make_request_fn. it also does loop decrypting
576 * on reads for block backed loop, as that is too heavy to do from
577 * b_end_io context where irqs may be disabled.
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700578 *
579 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
580 * calling kthread_stop(). Therefore once kthread_should_stop() is
581 * true, make_request will not place any more requests. Therefore
582 * once kthread_should_stop() is true and lo_bio is NULL, we are
583 * done with the loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 */
585static int loop_thread(void *data)
586{
587 struct loop_device *lo = data;
588 struct bio *bio;
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 set_user_nice(current, -20);
591
Akinobu Mitae6863072009-04-17 08:41:21 +0200592 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700594 wait_event_interruptible(lo->lo_event,
Akinobu Mitae6863072009-04-17 08:41:21 +0200595 !bio_list_empty(&lo->lo_bio_list) ||
596 kthread_should_stop());
Linus Torvalds09c0dc62006-06-26 11:55:42 -0700597
Akinobu Mitae6863072009-04-17 08:41:21 +0200598 if (bio_list_empty(&lo->lo_bio_list))
Nick Piggin35a82d12005-06-23 00:09:06 -0700599 continue;
Nick Piggin35a82d12005-06-23 00:09:06 -0700600 spin_lock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 bio = loop_get_bio(lo);
Nick Piggin35a82d12005-06-23 00:09:06 -0700602 spin_unlock_irq(&lo->lo_lock);
603
604 BUG_ON(!bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 loop_handle_bio(lo, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return 0;
609}
610
611/*
612 * loop_switch performs the hard work of switching a backing store.
613 * First it needs to flush existing IO, it does this by sending a magic
614 * BIO down the pipe. The completion of this BIO does the actual switch.
615 */
616static int loop_switch(struct loop_device *lo, struct file *file)
617{
618 struct switch_request w;
Jens Axboea24eab12008-01-11 10:14:40 +0100619 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (!bio)
621 return -ENOMEM;
622 init_completion(&w.wait);
623 w.file = file;
624 bio->bi_private = &w;
625 bio->bi_bdev = NULL;
626 loop_make_request(lo->lo_queue, bio);
627 wait_for_completion(&w.wait);
628 return 0;
629}
630
631/*
Milan Broz14f27932008-12-12 14:48:27 +0100632 * Helper to flush the IOs in loop, but keeping loop thread running
633 */
634static int loop_flush(struct loop_device *lo)
635{
636 /* loop not yet configured, no running thread, nothing to flush */
637 if (!lo->lo_thread)
638 return 0;
639
640 return loop_switch(lo, NULL);
641}
642
643/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * Do the actual switch; called from the BIO completion routine
645 */
646static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
647{
648 struct file *file = p->file;
649 struct file *old_file = lo->lo_backing_file;
Milan Broz14f27932008-12-12 14:48:27 +0100650 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Milan Broz14f27932008-12-12 14:48:27 +0100652 /* if no new file, only flush of queued bios requested */
653 if (!file)
654 goto out;
655
656 mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
658 lo->lo_backing_file = file;
Theodore Ts'oba52de12006-09-27 01:50:49 -0700659 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
660 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 lo->old_gfp_mask = mapping_gfp_mask(mapping);
662 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
Milan Broz14f27932008-12-12 14:48:27 +0100663out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 complete(&p->wait);
665}
666
667
668/*
669 * loop_change_fd switched the backing store of a loopback device to
670 * a new file. This is useful for operating system installers to free up
671 * the original file and in High Availability environments to switch to
672 * an alternative location for the content in case of server meltdown.
673 * This can only work if the loop device is used read-only, and if the
674 * new backing store is the same size and type as the old backing store.
675 */
Al Virobb214882008-03-02 09:29:48 -0500676static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
677 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct file *file, *old_file;
680 struct inode *inode;
681 int error;
682
683 error = -ENXIO;
684 if (lo->lo_state != Lo_bound)
685 goto out;
686
687 /* the loop device has to be read-only */
688 error = -EINVAL;
689 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
690 goto out;
691
692 error = -EBADF;
693 file = fget(arg);
694 if (!file)
695 goto out;
696
697 inode = file->f_mapping->host;
698 old_file = lo->lo_backing_file;
699
700 error = -EINVAL;
701
702 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
703 goto out_putf;
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 /* size of the new backing store needs to be the same */
706 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
707 goto out_putf;
708
709 /* and ... switch */
710 error = loop_switch(lo, file);
711 if (error)
712 goto out_putf;
713
714 fput(old_file);
Laurent Vivier476a4812008-03-26 12:11:53 +0100715 if (max_part > 0)
716 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 return 0;
718
719 out_putf:
720 fput(file);
721 out:
722 return error;
723}
724
725static inline int is_loop_device(struct file *file)
726{
727 struct inode *i = file->f_mapping->host;
728
729 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
730}
731
Milan Brozee862732010-08-23 15:16:00 +0200732/* loop sysfs attributes */
733
734static ssize_t loop_attr_show(struct device *dev, char *page,
735 ssize_t (*callback)(struct loop_device *, char *))
736{
737 struct loop_device *l, *lo = NULL;
738
739 mutex_lock(&loop_devices_mutex);
740 list_for_each_entry(l, &loop_devices, lo_list)
741 if (disk_to_dev(l->lo_disk) == dev) {
742 lo = l;
743 break;
744 }
745 mutex_unlock(&loop_devices_mutex);
746
747 return lo ? callback(lo, page) : -EIO;
748}
749
750#define LOOP_ATTR_RO(_name) \
751static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
752static ssize_t loop_attr_do_show_##_name(struct device *d, \
753 struct device_attribute *attr, char *b) \
754{ \
755 return loop_attr_show(d, b, loop_attr_##_name##_show); \
756} \
757static struct device_attribute loop_attr_##_name = \
758 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
759
760static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
761{
762 ssize_t ret;
763 char *p = NULL;
764
765 mutex_lock(&lo->lo_ctl_mutex);
766 if (lo->lo_backing_file)
767 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
768 mutex_unlock(&lo->lo_ctl_mutex);
769
770 if (IS_ERR_OR_NULL(p))
771 ret = PTR_ERR(p);
772 else {
773 ret = strlen(p);
774 memmove(buf, p, ret);
775 buf[ret++] = '\n';
776 buf[ret] = 0;
777 }
778
779 return ret;
780}
781
782static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
783{
784 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
785}
786
787static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
788{
789 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
790}
791
792static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
793{
794 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
795
796 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
797}
798
799LOOP_ATTR_RO(backing_file);
800LOOP_ATTR_RO(offset);
801LOOP_ATTR_RO(sizelimit);
802LOOP_ATTR_RO(autoclear);
803
804static struct attribute *loop_attrs[] = {
805 &loop_attr_backing_file.attr,
806 &loop_attr_offset.attr,
807 &loop_attr_sizelimit.attr,
808 &loop_attr_autoclear.attr,
809 NULL,
810};
811
812static struct attribute_group loop_attribute_group = {
813 .name = "loop",
814 .attrs= loop_attrs,
815};
816
817static int loop_sysfs_init(struct loop_device *lo)
818{
819 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
820 &loop_attribute_group);
821}
822
823static void loop_sysfs_exit(struct loop_device *lo)
824{
825 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
826 &loop_attribute_group);
827}
828
Al Virobb214882008-03-02 09:29:48 -0500829static int loop_set_fd(struct loop_device *lo, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 struct block_device *bdev, unsigned int arg)
831{
832 struct file *file, *f;
833 struct inode *inode;
834 struct address_space *mapping;
835 unsigned lo_blocksize;
836 int lo_flags = 0;
837 int error;
838 loff_t size;
839
840 /* This is safe, since we have a reference from open(). */
841 __module_get(THIS_MODULE);
842
843 error = -EBADF;
844 file = fget(arg);
845 if (!file)
846 goto out;
847
848 error = -EBUSY;
849 if (lo->lo_state != Lo_unbound)
850 goto out_putf;
851
852 /* Avoid recursion */
853 f = file;
854 while (is_loop_device(f)) {
855 struct loop_device *l;
856
Al Virobb214882008-03-02 09:29:48 -0500857 if (f->f_mapping->host->i_bdev == bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 goto out_putf;
859
860 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
861 if (l->lo_state == Lo_unbound) {
862 error = -EINVAL;
863 goto out_putf;
864 }
865 f = l->lo_backing_file;
866 }
867
868 mapping = file->f_mapping;
869 inode = mapping->host;
870
871 if (!(file->f_mode & FMODE_WRITE))
872 lo_flags |= LO_FLAGS_READ_ONLY;
873
874 error = -EINVAL;
875 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700876 const struct address_space_operations *aops = mapping->a_ops;
Miklos Szeredi68181732009-05-07 15:37:36 +0200877
Nick Piggin4e02ed42008-10-29 14:00:55 -0700878 if (aops->write_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 lo_flags |= LO_FLAGS_USE_AOPS;
880 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
881 lo_flags |= LO_FLAGS_READ_ONLY;
882
Theodore Ts'oba52de12006-09-27 01:50:49 -0700883 lo_blocksize = S_ISBLK(inode->i_mode) ?
884 inode->i_bdev->bd_block_size : PAGE_SIZE;
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 error = 0;
887 } else {
888 goto out_putf;
889 }
890
891 size = get_loop_size(lo, file);
892
893 if ((loff_t)(sector_t)size != size) {
894 error = -EFBIG;
895 goto out_putf;
896 }
897
Al Virobb214882008-03-02 09:29:48 -0500898 if (!(mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 lo_flags |= LO_FLAGS_READ_ONLY;
900
901 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
902
903 lo->lo_blocksize = lo_blocksize;
904 lo->lo_device = bdev;
905 lo->lo_flags = lo_flags;
906 lo->lo_backing_file = file;
Constantine Sapuntzakiseefe85e2006-06-23 02:06:08 -0700907 lo->transfer = transfer_none;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 lo->ioctl = NULL;
909 lo->lo_sizelimit = 0;
910 lo->old_gfp_mask = mapping_gfp_mask(mapping);
911 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
912
Akinobu Mitae6863072009-04-17 08:41:21 +0200913 bio_list_init(&lo->lo_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 /*
916 * set queue make_request_fn, and add limits based on lower level
917 * device
918 */
919 blk_queue_make_request(lo->lo_queue, loop_make_request);
920 lo->lo_queue->queuedata = lo;
921 lo->lo_queue->unplug_fn = loop_unplug;
922
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100923 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Tejun Heo4913efe2010-09-03 11:56:16 +0200924 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +0100925
Ken Chen73285082007-05-08 00:28:20 -0700926 set_capacity(lo->lo_disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 bd_set_size(bdev, size << 9);
Milan Brozee862732010-08-23 15:16:00 +0200928 loop_sysfs_init(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +0200929 /* let user-space know about the new size */
930 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 set_blocksize(bdev, lo_blocksize);
933
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700934 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
935 lo->lo_number);
936 if (IS_ERR(lo->lo_thread)) {
937 error = PTR_ERR(lo->lo_thread);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700938 goto out_clr;
Serge E. Hallyn6c997912006-09-29 01:59:11 -0700939 }
940 lo->lo_state = Lo_bound;
941 wake_up_process(lo->lo_thread);
Laurent Vivier476a4812008-03-26 12:11:53 +0100942 if (max_part > 0)
943 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 return 0;
945
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700946out_clr:
Milan Brozee862732010-08-23 15:16:00 +0200947 loop_sysfs_exit(lo);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700948 lo->lo_thread = NULL;
949 lo->lo_device = NULL;
950 lo->lo_backing_file = NULL;
951 lo->lo_flags = 0;
Ken Chen73285082007-05-08 00:28:20 -0700952 set_capacity(lo->lo_disk, 0);
Peter Zijlstraf98393a2007-05-06 14:49:54 -0700953 invalidate_bdev(bdev);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700954 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +0200955 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
Serge E. Hallyna7422bf2006-09-29 02:01:18 -0700956 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
957 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 out_putf:
959 fput(file);
960 out:
961 /* This is safe: open() is still holding a reference. */
962 module_put(THIS_MODULE);
963 return error;
964}
965
966static int
967loop_release_xfer(struct loop_device *lo)
968{
969 int err = 0;
970 struct loop_func_table *xfer = lo->lo_encryption;
971
972 if (xfer) {
973 if (xfer->release)
974 err = xfer->release(lo);
975 lo->transfer = NULL;
976 lo->lo_encryption = NULL;
977 module_put(xfer->owner);
978 }
979 return err;
980}
981
982static int
983loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
984 const struct loop_info64 *i)
985{
986 int err = 0;
987
988 if (xfer) {
989 struct module *owner = xfer->owner;
990
991 if (!try_module_get(owner))
992 return -EINVAL;
993 if (xfer->init)
994 err = xfer->init(lo, i);
995 if (err)
996 module_put(owner);
997 else
998 lo->lo_encryption = xfer;
999 }
1000 return err;
1001}
1002
1003static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
1004{
1005 struct file *filp = lo->lo_backing_file;
Al Virob4e3ca12005-10-21 03:22:34 -04001006 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 if (lo->lo_state != Lo_bound)
1009 return -ENXIO;
1010
1011 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
1012 return -EBUSY;
1013
1014 if (filp == NULL)
1015 return -EINVAL;
1016
1017 spin_lock_irq(&lo->lo_lock);
1018 lo->lo_state = Lo_rundown;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 spin_unlock_irq(&lo->lo_lock);
1020
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001021 kthread_stop(lo->lo_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Milan Broz8ae30b82008-12-12 14:50:49 +01001023 lo->lo_queue->unplug_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 lo->lo_backing_file = NULL;
1025
1026 loop_release_xfer(lo);
1027 lo->transfer = NULL;
1028 lo->ioctl = NULL;
1029 lo->lo_device = NULL;
1030 lo->lo_encryption = NULL;
1031 lo->lo_offset = 0;
1032 lo->lo_sizelimit = 0;
1033 lo->lo_encrypt_key_size = 0;
1034 lo->lo_flags = 0;
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001035 lo->lo_thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1037 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1038 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Al Virobb214882008-03-02 09:29:48 -05001039 if (bdev)
1040 invalidate_bdev(bdev);
Ken Chen73285082007-05-08 00:28:20 -07001041 set_capacity(lo->lo_disk, 0);
Milan Broz51a0bb02010-10-27 19:51:30 -06001042 loop_sysfs_exit(lo);
David Zeuthenc3473c62010-05-03 14:08:59 +02001043 if (bdev) {
Al Virobb214882008-03-02 09:29:48 -05001044 bd_set_size(bdev, 0);
David Zeuthenc3473c62010-05-03 14:08:59 +02001045 /* let user-space know about this change */
1046 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 mapping_set_gfp_mask(filp->f_mapping, gfp);
1049 lo->lo_state = Lo_unbound;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* This is safe: open() is still holding a reference. */
1051 module_put(THIS_MODULE);
Alexey Dobriyancf6e6932009-10-26 16:49:55 -07001052 if (max_part > 0 && bdev)
Laurent Vivier476a4812008-03-26 12:11:53 +01001053 ioctl_by_bdev(bdev, BLKRRPART, 0);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001054 mutex_unlock(&lo->lo_ctl_mutex);
1055 /*
1056 * Need not hold lo_ctl_mutex to fput backing file.
1057 * Calling fput holding lo_ctl_mutex triggers a circular
1058 * lock dependency possibility warning as fput can take
1059 * bd_mutex which is usually taken before lo_ctl_mutex.
1060 */
1061 fput(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 return 0;
1063}
1064
1065static int
1066loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1067{
1068 int err;
1069 struct loop_func_table *xfer;
David Howellsb0fafa82008-11-14 10:38:41 +11001070 uid_t uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
David Howellsb0fafa82008-11-14 10:38:41 +11001072 if (lo->lo_encrypt_key_size &&
1073 lo->lo_key_owner != uid &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 !capable(CAP_SYS_ADMIN))
1075 return -EPERM;
1076 if (lo->lo_state != Lo_bound)
1077 return -ENXIO;
1078 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1079 return -EINVAL;
1080
1081 err = loop_release_xfer(lo);
1082 if (err)
1083 return err;
1084
1085 if (info->lo_encrypt_type) {
1086 unsigned int type = info->lo_encrypt_type;
1087
1088 if (type >= MAX_LO_CRYPT)
1089 return -EINVAL;
1090 xfer = xfer_funcs[type];
1091 if (xfer == NULL)
1092 return -EINVAL;
1093 } else
1094 xfer = NULL;
1095
1096 err = loop_init_xfer(lo, xfer, info);
1097 if (err)
1098 return err;
1099
1100 if (lo->lo_offset != info->lo_offset ||
1101 lo->lo_sizelimit != info->lo_sizelimit) {
1102 lo->lo_offset = info->lo_offset;
1103 lo->lo_sizelimit = info->lo_sizelimit;
1104 if (figure_loop_size(lo))
1105 return -EFBIG;
1106 }
1107
1108 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1109 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1110 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1111 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1112
1113 if (!xfer)
1114 xfer = &none_funcs;
1115 lo->transfer = xfer->transfer;
1116 lo->ioctl = xfer->ioctl;
1117
David Woodhouse96c58652008-02-06 01:36:27 -08001118 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1119 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1120 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1123 lo->lo_init[0] = info->lo_init[0];
1124 lo->lo_init[1] = info->lo_init[1];
1125 if (info->lo_encrypt_key_size) {
1126 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1127 info->lo_encrypt_key_size);
David Howellsb0fafa82008-11-14 10:38:41 +11001128 lo->lo_key_owner = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130
1131 return 0;
1132}
1133
1134static int
1135loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1136{
1137 struct file *file = lo->lo_backing_file;
1138 struct kstat stat;
1139 int error;
1140
1141 if (lo->lo_state != Lo_bound)
1142 return -ENXIO;
Josef Sipek6c648be2006-12-08 02:36:55 -08001143 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 if (error)
1145 return error;
1146 memset(info, 0, sizeof(*info));
1147 info->lo_number = lo->lo_number;
1148 info->lo_device = huge_encode_dev(stat.dev);
1149 info->lo_inode = stat.ino;
1150 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1151 info->lo_offset = lo->lo_offset;
1152 info->lo_sizelimit = lo->lo_sizelimit;
1153 info->lo_flags = lo->lo_flags;
1154 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1155 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1156 info->lo_encrypt_type =
1157 lo->lo_encryption ? lo->lo_encryption->number : 0;
1158 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1159 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1160 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1161 lo->lo_encrypt_key_size);
1162 }
1163 return 0;
1164}
1165
1166static void
1167loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1168{
1169 memset(info64, 0, sizeof(*info64));
1170 info64->lo_number = info->lo_number;
1171 info64->lo_device = info->lo_device;
1172 info64->lo_inode = info->lo_inode;
1173 info64->lo_rdevice = info->lo_rdevice;
1174 info64->lo_offset = info->lo_offset;
1175 info64->lo_sizelimit = 0;
1176 info64->lo_encrypt_type = info->lo_encrypt_type;
1177 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1178 info64->lo_flags = info->lo_flags;
1179 info64->lo_init[0] = info->lo_init[0];
1180 info64->lo_init[1] = info->lo_init[1];
1181 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1182 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1183 else
1184 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1185 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1186}
1187
1188static int
1189loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1190{
1191 memset(info, 0, sizeof(*info));
1192 info->lo_number = info64->lo_number;
1193 info->lo_device = info64->lo_device;
1194 info->lo_inode = info64->lo_inode;
1195 info->lo_rdevice = info64->lo_rdevice;
1196 info->lo_offset = info64->lo_offset;
1197 info->lo_encrypt_type = info64->lo_encrypt_type;
1198 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1199 info->lo_flags = info64->lo_flags;
1200 info->lo_init[0] = info64->lo_init[0];
1201 info->lo_init[1] = info64->lo_init[1];
1202 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1203 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1204 else
1205 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1206 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1207
1208 /* error in case values were truncated */
1209 if (info->lo_device != info64->lo_device ||
1210 info->lo_rdevice != info64->lo_rdevice ||
1211 info->lo_inode != info64->lo_inode ||
1212 info->lo_offset != info64->lo_offset)
1213 return -EOVERFLOW;
1214
1215 return 0;
1216}
1217
1218static int
1219loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1220{
1221 struct loop_info info;
1222 struct loop_info64 info64;
1223
1224 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1225 return -EFAULT;
1226 loop_info64_from_old(&info, &info64);
1227 return loop_set_status(lo, &info64);
1228}
1229
1230static int
1231loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1232{
1233 struct loop_info64 info64;
1234
1235 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1236 return -EFAULT;
1237 return loop_set_status(lo, &info64);
1238}
1239
1240static int
1241loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1242 struct loop_info info;
1243 struct loop_info64 info64;
1244 int err = 0;
1245
1246 if (!arg)
1247 err = -EINVAL;
1248 if (!err)
1249 err = loop_get_status(lo, &info64);
1250 if (!err)
1251 err = loop_info64_to_old(&info64, &info);
1252 if (!err && copy_to_user(arg, &info, sizeof(info)))
1253 err = -EFAULT;
1254
1255 return err;
1256}
1257
1258static int
1259loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1260 struct loop_info64 info64;
1261 int err = 0;
1262
1263 if (!arg)
1264 err = -EINVAL;
1265 if (!err)
1266 err = loop_get_status(lo, &info64);
1267 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1268 err = -EFAULT;
1269
1270 return err;
1271}
1272
J. R. Okajima53d66602009-03-31 15:23:43 -07001273static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1274{
1275 int err;
1276 sector_t sec;
1277 loff_t sz;
1278
1279 err = -ENXIO;
1280 if (unlikely(lo->lo_state != Lo_bound))
1281 goto out;
1282 err = figure_loop_size(lo);
1283 if (unlikely(err))
1284 goto out;
1285 sec = get_capacity(lo->lo_disk);
1286 /* the width of sector_t may be narrow for bit-shift */
1287 sz = sec;
1288 sz <<= 9;
1289 mutex_lock(&bdev->bd_mutex);
1290 bd_set_size(bdev, sz);
David Zeuthenc3473c62010-05-03 14:08:59 +02001291 /* let user-space know about the new size */
1292 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
J. R. Okajima53d66602009-03-31 15:23:43 -07001293 mutex_unlock(&bdev->bd_mutex);
1294
1295 out:
1296 return err;
1297}
1298
Al Virobb214882008-03-02 09:29:48 -05001299static int lo_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 unsigned int cmd, unsigned long arg)
1301{
Al Virobb214882008-03-02 09:29:48 -05001302 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 int err;
1304
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001305 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 switch (cmd) {
1307 case LOOP_SET_FD:
Al Virobb214882008-03-02 09:29:48 -05001308 err = loop_set_fd(lo, mode, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 break;
1310 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001311 err = loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 break;
1313 case LOOP_CLR_FD:
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001314 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
Al Virobb214882008-03-02 09:29:48 -05001315 err = loop_clr_fd(lo, bdev);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001316 if (!err)
1317 goto out_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 break;
1319 case LOOP_SET_STATUS:
1320 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1321 break;
1322 case LOOP_GET_STATUS:
1323 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1324 break;
1325 case LOOP_SET_STATUS64:
1326 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1327 break;
1328 case LOOP_GET_STATUS64:
1329 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1330 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001331 case LOOP_SET_CAPACITY:
1332 err = -EPERM;
1333 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1334 err = loop_set_capacity(lo, bdev);
1335 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 default:
1337 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1338 }
Ingo Molnarf85221d2006-03-23 03:00:38 -08001339 mutex_unlock(&lo->lo_ctl_mutex);
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001340
1341out_unlocked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return err;
1343}
1344
David Howells863d5b822006-08-29 19:06:14 +01001345#ifdef CONFIG_COMPAT
1346struct compat_loop_info {
1347 compat_int_t lo_number; /* ioctl r/o */
1348 compat_dev_t lo_device; /* ioctl r/o */
1349 compat_ulong_t lo_inode; /* ioctl r/o */
1350 compat_dev_t lo_rdevice; /* ioctl r/o */
1351 compat_int_t lo_offset;
1352 compat_int_t lo_encrypt_type;
1353 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1354 compat_int_t lo_flags; /* ioctl r/o */
1355 char lo_name[LO_NAME_SIZE];
1356 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1357 compat_ulong_t lo_init[2];
1358 char reserved[4];
1359};
1360
1361/*
1362 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1363 * - noinlined to reduce stack space usage in main part of driver
1364 */
1365static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001366loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001367 struct loop_info64 *info64)
1368{
1369 struct compat_loop_info info;
1370
1371 if (copy_from_user(&info, arg, sizeof(info)))
1372 return -EFAULT;
1373
1374 memset(info64, 0, sizeof(*info64));
1375 info64->lo_number = info.lo_number;
1376 info64->lo_device = info.lo_device;
1377 info64->lo_inode = info.lo_inode;
1378 info64->lo_rdevice = info.lo_rdevice;
1379 info64->lo_offset = info.lo_offset;
1380 info64->lo_sizelimit = 0;
1381 info64->lo_encrypt_type = info.lo_encrypt_type;
1382 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1383 info64->lo_flags = info.lo_flags;
1384 info64->lo_init[0] = info.lo_init[0];
1385 info64->lo_init[1] = info.lo_init[1];
1386 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1387 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1388 else
1389 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1390 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1391 return 0;
1392}
1393
1394/*
1395 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1396 * - noinlined to reduce stack space usage in main part of driver
1397 */
1398static noinline int
1399loop_info64_to_compat(const struct loop_info64 *info64,
1400 struct compat_loop_info __user *arg)
1401{
1402 struct compat_loop_info info;
1403
1404 memset(&info, 0, sizeof(info));
1405 info.lo_number = info64->lo_number;
1406 info.lo_device = info64->lo_device;
1407 info.lo_inode = info64->lo_inode;
1408 info.lo_rdevice = info64->lo_rdevice;
1409 info.lo_offset = info64->lo_offset;
1410 info.lo_encrypt_type = info64->lo_encrypt_type;
1411 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1412 info.lo_flags = info64->lo_flags;
1413 info.lo_init[0] = info64->lo_init[0];
1414 info.lo_init[1] = info64->lo_init[1];
1415 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1416 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1417 else
1418 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1419 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1420
1421 /* error in case values were truncated */
1422 if (info.lo_device != info64->lo_device ||
1423 info.lo_rdevice != info64->lo_rdevice ||
1424 info.lo_inode != info64->lo_inode ||
1425 info.lo_offset != info64->lo_offset ||
1426 info.lo_init[0] != info64->lo_init[0] ||
1427 info.lo_init[1] != info64->lo_init[1])
1428 return -EOVERFLOW;
1429
1430 if (copy_to_user(arg, &info, sizeof(info)))
1431 return -EFAULT;
1432 return 0;
1433}
1434
1435static int
1436loop_set_status_compat(struct loop_device *lo,
1437 const struct compat_loop_info __user *arg)
1438{
1439 struct loop_info64 info64;
1440 int ret;
1441
1442 ret = loop_info64_from_compat(arg, &info64);
1443 if (ret < 0)
1444 return ret;
1445 return loop_set_status(lo, &info64);
1446}
1447
1448static int
1449loop_get_status_compat(struct loop_device *lo,
1450 struct compat_loop_info __user *arg)
1451{
1452 struct loop_info64 info64;
1453 int err = 0;
1454
1455 if (!arg)
1456 err = -EINVAL;
1457 if (!err)
1458 err = loop_get_status(lo, &info64);
1459 if (!err)
1460 err = loop_info64_to_compat(&info64, arg);
1461 return err;
1462}
1463
Al Virobb214882008-03-02 09:29:48 -05001464static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1465 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001466{
Al Virobb214882008-03-02 09:29:48 -05001467 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001468 int err;
1469
David Howells863d5b822006-08-29 19:06:14 +01001470 switch(cmd) {
1471 case LOOP_SET_STATUS:
1472 mutex_lock(&lo->lo_ctl_mutex);
1473 err = loop_set_status_compat(
1474 lo, (const struct compat_loop_info __user *) arg);
1475 mutex_unlock(&lo->lo_ctl_mutex);
1476 break;
1477 case LOOP_GET_STATUS:
1478 mutex_lock(&lo->lo_ctl_mutex);
1479 err = loop_get_status_compat(
1480 lo, (struct compat_loop_info __user *) arg);
1481 mutex_unlock(&lo->lo_ctl_mutex);
1482 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001483 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001484 case LOOP_CLR_FD:
1485 case LOOP_GET_STATUS64:
1486 case LOOP_SET_STATUS64:
1487 arg = (unsigned long) compat_ptr(arg);
1488 case LOOP_SET_FD:
1489 case LOOP_CHANGE_FD:
Al Virobb214882008-03-02 09:29:48 -05001490 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001491 break;
1492 default:
1493 err = -ENOIOCTLCMD;
1494 break;
1495 }
David Howells863d5b822006-08-29 19:06:14 +01001496 return err;
1497}
1498#endif
1499
Al Virobb214882008-03-02 09:29:48 -05001500static int lo_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
Al Virobb214882008-03-02 09:29:48 -05001502 struct loop_device *lo = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001504 mutex_lock(&loop_mutex);
Ingo Molnarf85221d2006-03-23 03:00:38 -08001505 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 lo->lo_refcnt++;
Ingo Molnarf85221d2006-03-23 03:00:38 -08001507 mutex_unlock(&lo->lo_ctl_mutex);
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001508 mutex_unlock(&loop_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 return 0;
1511}
1512
Al Virobb214882008-03-02 09:29:48 -05001513static int lo_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
Al Virobb214882008-03-02 09:29:48 -05001515 struct loop_device *lo = disk->private_data;
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001516 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001518 mutex_lock(&loop_mutex);
Ingo Molnarf85221d2006-03-23 03:00:38 -08001519 mutex_lock(&lo->lo_ctl_mutex);
David Woodhouse96c58652008-02-06 01:36:27 -08001520
Milan Broz14f27932008-12-12 14:48:27 +01001521 if (--lo->lo_refcnt)
1522 goto out;
1523
1524 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1525 /*
1526 * In autoclear mode, stop the loop thread
1527 * and remove configuration after last close.
1528 */
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001529 err = loop_clr_fd(lo, NULL);
1530 if (!err)
1531 goto out_unlocked;
Milan Broz14f27932008-12-12 14:48:27 +01001532 } else {
1533 /*
1534 * Otherwise keep thread (if running) and config,
1535 * but flush possible ongoing bios in thread.
1536 */
1537 loop_flush(lo);
1538 }
David Woodhouse96c58652008-02-06 01:36:27 -08001539
Milan Broz14f27932008-12-12 14:48:27 +01001540out:
Ingo Molnarf85221d2006-03-23 03:00:38 -08001541 mutex_unlock(&lo->lo_ctl_mutex);
Alexander Beregalovffcd7dc2009-04-07 13:48:21 +02001542out_unlocked:
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001543 mutex_unlock(&loop_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 return 0;
1545}
1546
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001547static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001549 .open = lo_open,
1550 .release = lo_release,
1551 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001552#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001553 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001554#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555};
1556
1557/*
1558 * And now the modules code and kernel interface.
1559 */
Ken Chen73285082007-05-08 00:28:20 -07001560static int max_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561module_param(max_loop, int, 0);
Ken Chena47653f2007-06-08 13:46:44 -07001562MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Laurent Vivier476a4812008-03-26 12:11:53 +01001563module_param(max_part, int, 0);
1564MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565MODULE_LICENSE("GPL");
1566MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1567
1568int loop_register_transfer(struct loop_func_table *funcs)
1569{
1570 unsigned int n = funcs->number;
1571
1572 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1573 return -EINVAL;
1574 xfer_funcs[n] = funcs;
1575 return 0;
1576}
1577
1578int loop_unregister_transfer(int number)
1579{
1580 unsigned int n = number;
1581 struct loop_device *lo;
1582 struct loop_func_table *xfer;
1583
1584 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1585 return -EINVAL;
1586
1587 xfer_funcs[n] = NULL;
1588
Ken Chen73285082007-05-08 00:28:20 -07001589 list_for_each_entry(lo, &loop_devices, lo_list) {
Ingo Molnarf85221d2006-03-23 03:00:38 -08001590 mutex_lock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
1592 if (lo->lo_encryption == xfer)
1593 loop_release_xfer(lo);
1594
Ingo Molnarf85221d2006-03-23 03:00:38 -08001595 mutex_unlock(&lo->lo_ctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
1597
1598 return 0;
1599}
1600
1601EXPORT_SYMBOL(loop_register_transfer);
1602EXPORT_SYMBOL(loop_unregister_transfer);
1603
Ken Chena47653f2007-06-08 13:46:44 -07001604static struct loop_device *loop_alloc(int i)
Ken Chen73285082007-05-08 00:28:20 -07001605{
1606 struct loop_device *lo;
1607 struct gendisk *disk;
1608
1609 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1610 if (!lo)
1611 goto out;
1612
1613 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1614 if (!lo->lo_queue)
1615 goto out_free_dev;
1616
Laurent Vivier476a4812008-03-26 12:11:53 +01001617 disk = lo->lo_disk = alloc_disk(1 << part_shift);
Ken Chen73285082007-05-08 00:28:20 -07001618 if (!disk)
1619 goto out_free_queue;
1620
1621 mutex_init(&lo->lo_ctl_mutex);
1622 lo->lo_number = i;
1623 lo->lo_thread = NULL;
1624 init_waitqueue_head(&lo->lo_event);
1625 spin_lock_init(&lo->lo_lock);
1626 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01001627 disk->first_minor = i << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07001628 disk->fops = &lo_fops;
1629 disk->private_data = lo;
1630 disk->queue = lo->lo_queue;
1631 sprintf(disk->disk_name, "loop%d", i);
Ken Chen73285082007-05-08 00:28:20 -07001632 return lo;
1633
1634out_free_queue:
1635 blk_cleanup_queue(lo->lo_queue);
1636out_free_dev:
1637 kfree(lo);
1638out:
Al Viro07002e92007-05-12 16:23:15 -04001639 return NULL;
Ken Chen73285082007-05-08 00:28:20 -07001640}
1641
Ken Chena47653f2007-06-08 13:46:44 -07001642static void loop_free(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07001643{
Ken Chen73285082007-05-08 00:28:20 -07001644 blk_cleanup_queue(lo->lo_queue);
1645 put_disk(lo->lo_disk);
1646 list_del(&lo->lo_list);
1647 kfree(lo);
1648}
1649
Ken Chena47653f2007-06-08 13:46:44 -07001650static struct loop_device *loop_init_one(int i)
1651{
1652 struct loop_device *lo;
1653
1654 list_for_each_entry(lo, &loop_devices, lo_list) {
1655 if (lo->lo_number == i)
1656 return lo;
1657 }
1658
1659 lo = loop_alloc(i);
1660 if (lo) {
1661 add_disk(lo->lo_disk);
1662 list_add_tail(&lo->lo_list, &loop_devices);
1663 }
1664 return lo;
1665}
1666
1667static void loop_del_one(struct loop_device *lo)
1668{
1669 del_gendisk(lo->lo_disk);
1670 loop_free(lo);
1671}
1672
Ken Chen73285082007-05-08 00:28:20 -07001673static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1674{
Al Viro705962c2007-05-13 05:52:32 -04001675 struct loop_device *lo;
Al Viro07002e92007-05-12 16:23:15 -04001676 struct kobject *kobj;
Ken Chen73285082007-05-08 00:28:20 -07001677
Al Viro705962c2007-05-13 05:52:32 -04001678 mutex_lock(&loop_devices_mutex);
1679 lo = loop_init_one(dev & MINORMASK);
Al Viro07002e92007-05-12 16:23:15 -04001680 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
Ken Chen73285082007-05-08 00:28:20 -07001681 mutex_unlock(&loop_devices_mutex);
1682
1683 *part = 0;
Al Viro07002e92007-05-12 16:23:15 -04001684 return kobj;
Ken Chen73285082007-05-08 00:28:20 -07001685}
1686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687static int __init loop_init(void)
1688{
Ken Chena47653f2007-06-08 13:46:44 -07001689 int i, nr;
1690 unsigned long range;
1691 struct loop_device *lo, *next;
1692
1693 /*
1694 * loop module now has a feature to instantiate underlying device
1695 * structure on-demand, provided that there is an access dev node.
1696 * However, this will not work well with user space tool that doesn't
1697 * know about such "feature". In order to not break any existing
1698 * tool, we do the following:
1699 *
1700 * (1) if max_loop is specified, create that many upfront, and this
1701 * also becomes a hard limit.
1702 * (2) if max_loop is not specified, create 8 loop device on module
1703 * load, user can further extend loop device by create dev node
1704 * themselves and have kernel automatically instantiate actual
1705 * device on-demand.
1706 */
Laurent Vivier476a4812008-03-26 12:11:53 +01001707
1708 part_shift = 0;
1709 if (max_part > 0)
1710 part_shift = fls(max_part);
1711
1712 if (max_loop > 1UL << (MINORBITS - part_shift))
Ken Chena47653f2007-06-08 13:46:44 -07001713 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Ken Chen73285082007-05-08 00:28:20 -07001715 if (max_loop) {
Ken Chena47653f2007-06-08 13:46:44 -07001716 nr = max_loop;
1717 range = max_loop;
1718 } else {
1719 nr = 8;
Laurent Vivier476a4812008-03-26 12:11:53 +01001720 range = 1UL << (MINORBITS - part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 }
Ken Chena47653f2007-06-08 13:46:44 -07001722
1723 if (register_blkdev(LOOP_MAJOR, "loop"))
1724 return -EIO;
1725
1726 for (i = 0; i < nr; i++) {
1727 lo = loop_alloc(i);
1728 if (!lo)
1729 goto Enomem;
1730 list_add_tail(&lo->lo_list, &loop_devices);
1731 }
1732
1733 /* point of no return */
1734
1735 list_for_each_entry(lo, &loop_devices, lo_list)
1736 add_disk(lo->lo_disk);
1737
1738 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1739 THIS_MODULE, loop_probe, NULL, NULL);
1740
Ken Chen73285082007-05-08 00:28:20 -07001741 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 return 0;
Ken Chena47653f2007-06-08 13:46:44 -07001743
1744Enomem:
1745 printk(KERN_INFO "loop: out of memory\n");
1746
1747 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1748 loop_free(lo);
1749
1750 unregister_blkdev(LOOP_MAJOR, "loop");
1751 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752}
1753
Ken Chen73285082007-05-08 00:28:20 -07001754static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755{
Ken Chena47653f2007-06-08 13:46:44 -07001756 unsigned long range;
Ken Chen73285082007-05-08 00:28:20 -07001757 struct loop_device *lo, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Laurent Vivier476a4812008-03-26 12:11:53 +01001759 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
Ken Chena47653f2007-06-08 13:46:44 -07001760
Ken Chen73285082007-05-08 00:28:20 -07001761 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1762 loop_del_one(lo);
1763
Ken Chena47653f2007-06-08 13:46:44 -07001764 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
Akinobu Mita00d59402007-07-17 04:03:46 -07001765 unregister_blkdev(LOOP_MAJOR, "loop");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
1768module_init(loop_init);
1769module_exit(loop_exit);
1770
1771#ifndef MODULE
1772static int __init max_loop_setup(char *str)
1773{
1774 max_loop = simple_strtol(str, NULL, 0);
1775 return 1;
1776}
1777
1778__setup("max_loop=", max_loop_setup);
1779#endif