2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
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
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
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
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
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
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
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.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>
64 #include <linux/smp_lock.h>
65 #include <linux/swap.h>
66 #include <linux/slab.h>
67 #include <linux/loop.h>
68 #include <linux/compat.h>
69 #include <linux/suspend.h>
70 #include <linux/freezer.h>
71 #include <linux/writeback.h>
72 #include <linux/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/gfp.h>
76 #include <linux/kthread.h>
77 #include <linux/splice.h>
79 #include <asm/uaccess.h>
81 static LIST_HEAD(loop_devices
);
82 static DEFINE_MUTEX(loop_devices_mutex
);
85 static int part_shift
;
90 static int transfer_none(struct loop_device
*lo
, int cmd
,
91 struct page
*raw_page
, unsigned raw_off
,
92 struct page
*loop_page
, unsigned loop_off
,
93 int size
, sector_t real_block
)
95 char *raw_buf
= kmap_atomic(raw_page
, KM_USER0
) + raw_off
;
96 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + loop_off
;
99 memcpy(loop_buf
, raw_buf
, size
);
101 memcpy(raw_buf
, loop_buf
, size
);
103 kunmap_atomic(raw_buf
, KM_USER0
);
104 kunmap_atomic(loop_buf
, KM_USER1
);
109 static int transfer_xor(struct loop_device
*lo
, int cmd
,
110 struct page
*raw_page
, unsigned raw_off
,
111 struct page
*loop_page
, unsigned loop_off
,
112 int size
, sector_t real_block
)
114 char *raw_buf
= kmap_atomic(raw_page
, KM_USER0
) + raw_off
;
115 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + loop_off
;
116 char *in
, *out
, *key
;
127 key
= lo
->lo_encrypt_key
;
128 keysize
= lo
->lo_encrypt_key_size
;
129 for (i
= 0; i
< size
; i
++)
130 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
132 kunmap_atomic(raw_buf
, KM_USER0
);
133 kunmap_atomic(loop_buf
, KM_USER1
);
138 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
140 if (unlikely(info
->lo_encrypt_key_size
<= 0))
145 static struct loop_func_table none_funcs
= {
146 .number
= LO_CRYPT_NONE
,
147 .transfer
= transfer_none
,
150 static struct loop_func_table xor_funcs
= {
151 .number
= LO_CRYPT_XOR
,
152 .transfer
= transfer_xor
,
156 /* xfer_funcs[0] is special - its release function is never called */
157 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
162 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
164 loff_t size
, offset
, loopsize
;
166 /* Compute loopsize in bytes */
167 size
= i_size_read(file
->f_mapping
->host
);
168 offset
= lo
->lo_offset
;
169 loopsize
= size
- offset
;
170 if (lo
->lo_sizelimit
> 0 && lo
->lo_sizelimit
< loopsize
)
171 loopsize
= lo
->lo_sizelimit
;
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
177 return loopsize
>> 9;
181 figure_loop_size(struct loop_device
*lo
)
183 loff_t size
= get_loop_size(lo
, lo
->lo_backing_file
);
184 sector_t x
= (sector_t
)size
;
186 if (unlikely((loff_t
)x
!= size
))
189 set_capacity(lo
->lo_disk
, x
);
194 lo_do_transfer(struct loop_device
*lo
, int cmd
,
195 struct page
*rpage
, unsigned roffs
,
196 struct page
*lpage
, unsigned loffs
,
197 int size
, sector_t rblock
)
199 if (unlikely(!lo
->transfer
))
202 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
206 * do_lo_send_aops - helper for writing data to a loop device
208 * This is the fast version for backing filesystems which implement the address
209 * space operations write_begin and write_end.
211 static int do_lo_send_aops(struct loop_device
*lo
, struct bio_vec
*bvec
,
212 loff_t pos
, struct page
*unused
)
214 struct file
*file
= lo
->lo_backing_file
; /* kudos to NFsckingS */
215 struct address_space
*mapping
= file
->f_mapping
;
217 unsigned offset
, bv_offs
;
220 mutex_lock(&mapping
->host
->i_mutex
);
221 index
= pos
>> PAGE_CACHE_SHIFT
;
222 offset
= pos
& ((pgoff_t
)PAGE_CACHE_SIZE
- 1);
223 bv_offs
= bvec
->bv_offset
;
227 unsigned size
, copied
;
232 IV
= ((sector_t
)index
<< (PAGE_CACHE_SHIFT
- 9))+(offset
>> 9);
233 size
= PAGE_CACHE_SIZE
- offset
;
237 ret
= pagecache_write_begin(file
, mapping
, pos
, size
, 0,
242 transfer_result
= lo_do_transfer(lo
, WRITE
, page
, offset
,
243 bvec
->bv_page
, bv_offs
, size
, IV
);
245 if (unlikely(transfer_result
))
248 ret
= pagecache_write_end(file
, mapping
, pos
, size
, copied
,
250 if (ret
< 0 || ret
!= copied
)
253 if (unlikely(transfer_result
))
264 mutex_unlock(&mapping
->host
->i_mutex
);
272 * __do_lo_send_write - helper for writing data to a loop device
274 * This helper just factors out common code between do_lo_send_direct_write()
275 * and do_lo_send_write().
277 static int __do_lo_send_write(struct file
*file
,
278 u8
*buf
, const int len
, loff_t pos
)
281 mm_segment_t old_fs
= get_fs();
284 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
286 if (likely(bw
== len
))
288 printk(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
289 (unsigned long long)pos
, len
);
296 * do_lo_send_direct_write - helper for writing data to a loop device
298 * This is the fast, non-transforming version for backing filesystems which do
299 * not implement the address space operations write_begin and write_end.
300 * It uses the write file operation which should be present on all writeable
303 static int do_lo_send_direct_write(struct loop_device
*lo
,
304 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
306 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
307 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
309 kunmap(bvec
->bv_page
);
315 * do_lo_send_write - helper for writing data to a loop device
317 * This is the slow, transforming version for filesystems which do not
318 * implement the address space operations write_begin and write_end. It
319 * uses the write file operation which should be present on all writeable
322 * Using fops->write is slower than using aops->{prepare,commit}_write in the
323 * transforming case because we need to double buffer the data as we cannot do
324 * the transformations in place as we do not have direct access to the
325 * destination pages of the backing file.
327 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
328 loff_t pos
, struct page
*page
)
330 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
331 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
333 return __do_lo_send_write(lo
->lo_backing_file
,
334 page_address(page
), bvec
->bv_len
,
336 printk(KERN_ERR
"loop: Transfer error at byte offset %llu, "
337 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
343 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
345 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
347 struct bio_vec
*bvec
;
348 struct page
*page
= NULL
;
351 do_lo_send
= do_lo_send_aops
;
352 if (!(lo
->lo_flags
& LO_FLAGS_USE_AOPS
)) {
353 do_lo_send
= do_lo_send_direct_write
;
354 if (lo
->transfer
!= transfer_none
) {
355 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
359 do_lo_send
= do_lo_send_write
;
362 bio_for_each_segment(bvec
, bio
, i
) {
363 ret
= do_lo_send(lo
, bvec
, pos
, page
);
375 printk(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
380 struct lo_read_data
{
381 struct loop_device
*lo
;
388 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
389 struct splice_desc
*sd
)
391 struct lo_read_data
*p
= sd
->u
.data
;
392 struct loop_device
*lo
= p
->lo
;
393 struct page
*page
= buf
->page
;
397 ret
= buf
->ops
->confirm(pipe
, buf
);
401 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
407 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
408 printk(KERN_ERR
"loop: transfer error block %ld\n",
413 flush_dcache_page(p
->page
);
422 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
424 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
428 do_lo_receive(struct loop_device
*lo
,
429 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
431 struct lo_read_data cookie
;
432 struct splice_desc sd
;
437 cookie
.page
= bvec
->bv_page
;
438 cookie
.offset
= bvec
->bv_offset
;
439 cookie
.bsize
= bsize
;
442 sd
.total_len
= bvec
->bv_len
;
447 file
= lo
->lo_backing_file
;
448 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
457 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
459 struct bio_vec
*bvec
;
462 bio_for_each_segment(bvec
, bio
, i
) {
463 ret
= do_lo_receive(lo
, bvec
, bsize
, pos
);
471 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
476 pos
= ((loff_t
) bio
->bi_sector
<< 9) + lo
->lo_offset
;
477 if (bio_rw(bio
) == WRITE
)
478 ret
= lo_send(lo
, bio
, pos
);
480 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
485 * Add bio to back of pending list
487 static void loop_add_bio(struct loop_device
*lo
, struct bio
*bio
)
489 if (lo
->lo_biotail
) {
490 lo
->lo_biotail
->bi_next
= bio
;
491 lo
->lo_biotail
= bio
;
493 lo
->lo_bio
= lo
->lo_biotail
= bio
;
497 * Grab first pending buffer
499 static struct bio
*loop_get_bio(struct loop_device
*lo
)
503 if ((bio
= lo
->lo_bio
)) {
504 if (bio
== lo
->lo_biotail
)
505 lo
->lo_biotail
= NULL
;
506 lo
->lo_bio
= bio
->bi_next
;
513 static int loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
515 struct loop_device
*lo
= q
->queuedata
;
516 int rw
= bio_rw(old_bio
);
521 BUG_ON(!lo
|| (rw
!= READ
&& rw
!= WRITE
));
523 spin_lock_irq(&lo
->lo_lock
);
524 if (lo
->lo_state
!= Lo_bound
)
526 if (unlikely(rw
== WRITE
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)))
528 loop_add_bio(lo
, old_bio
);
529 wake_up(&lo
->lo_event
);
530 spin_unlock_irq(&lo
->lo_lock
);
534 spin_unlock_irq(&lo
->lo_lock
);
535 bio_io_error(old_bio
);
540 * kick off io on the underlying address space
542 static void loop_unplug(struct request_queue
*q
)
544 struct loop_device
*lo
= q
->queuedata
;
546 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED
, q
);
547 blk_run_address_space(lo
->lo_backing_file
->f_mapping
);
550 struct switch_request
{
552 struct completion wait
;
555 static void do_loop_switch(struct loop_device
*, struct switch_request
*);
557 static inline void loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
559 if (unlikely(!bio
->bi_bdev
)) {
560 do_loop_switch(lo
, bio
->bi_private
);
563 int ret
= do_bio_filebacked(lo
, bio
);
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.
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.
580 static int loop_thread(void *data
)
582 struct loop_device
*lo
= data
;
585 set_user_nice(current
, -20);
587 while (!kthread_should_stop() || lo
->lo_bio
) {
589 wait_event_interruptible(lo
->lo_event
,
590 lo
->lo_bio
|| kthread_should_stop());
594 spin_lock_irq(&lo
->lo_lock
);
595 bio
= loop_get_bio(lo
);
596 spin_unlock_irq(&lo
->lo_lock
);
599 loop_handle_bio(lo
, bio
);
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.
610 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
612 struct switch_request w
;
613 struct bio
*bio
= bio_alloc(GFP_KERNEL
, 0);
616 init_completion(&w
.wait
);
618 bio
->bi_private
= &w
;
620 loop_make_request(lo
->lo_queue
, bio
);
621 wait_for_completion(&w
.wait
);
626 * Helper to flush the IOs in loop, but keeping loop thread running
628 static int loop_flush(struct loop_device
*lo
)
630 /* loop not yet configured, no running thread, nothing to flush */
634 return loop_switch(lo
, NULL
);
638 * Do the actual switch; called from the BIO completion routine
640 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
642 struct file
*file
= p
->file
;
643 struct file
*old_file
= lo
->lo_backing_file
;
644 struct address_space
*mapping
;
646 /* if no new file, only flush of queued bios requested */
650 mapping
= file
->f_mapping
;
651 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
652 lo
->lo_backing_file
= file
;
653 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
654 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
655 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
656 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
663 * loop_change_fd switched the backing store of a loopback device to
664 * a new file. This is useful for operating system installers to free up
665 * the original file and in High Availability environments to switch to
666 * an alternative location for the content in case of server meltdown.
667 * This can only work if the loop device is used read-only, and if the
668 * new backing store is the same size and type as the old backing store.
670 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
673 struct file
*file
, *old_file
;
678 if (lo
->lo_state
!= Lo_bound
)
681 /* the loop device has to be read-only */
683 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
691 inode
= file
->f_mapping
->host
;
692 old_file
= lo
->lo_backing_file
;
696 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
699 /* new backing store needs to support loop (eg splice_read) */
700 if (!inode
->i_fop
->splice_read
)
703 /* size of the new backing store needs to be the same */
704 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
708 error
= loop_switch(lo
, file
);
714 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
723 static inline int is_loop_device(struct file
*file
)
725 struct inode
*i
= file
->f_mapping
->host
;
727 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
730 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
731 struct block_device
*bdev
, unsigned int arg
)
733 struct file
*file
, *f
;
735 struct address_space
*mapping
;
736 unsigned lo_blocksize
;
741 /* This is safe, since we have a reference from open(). */
742 __module_get(THIS_MODULE
);
750 if (lo
->lo_state
!= Lo_unbound
)
753 /* Avoid recursion */
755 while (is_loop_device(f
)) {
756 struct loop_device
*l
;
758 if (f
->f_mapping
->host
->i_bdev
== bdev
)
761 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
762 if (l
->lo_state
== Lo_unbound
) {
766 f
= l
->lo_backing_file
;
769 mapping
= file
->f_mapping
;
770 inode
= mapping
->host
;
772 if (!(file
->f_mode
& FMODE_WRITE
))
773 lo_flags
|= LO_FLAGS_READ_ONLY
;
776 if (S_ISREG(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
777 const struct address_space_operations
*aops
= mapping
->a_ops
;
779 * If we can't read - sorry. If we only can't write - well,
780 * it's going to be read-only.
782 if (!file
->f_op
->splice_read
)
784 if (aops
->write_begin
)
785 lo_flags
|= LO_FLAGS_USE_AOPS
;
786 if (!(lo_flags
& LO_FLAGS_USE_AOPS
) && !file
->f_op
->write
)
787 lo_flags
|= LO_FLAGS_READ_ONLY
;
789 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
790 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
797 size
= get_loop_size(lo
, file
);
799 if ((loff_t
)(sector_t
)size
!= size
) {
804 if (!(mode
& FMODE_WRITE
))
805 lo_flags
|= LO_FLAGS_READ_ONLY
;
807 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
809 lo
->lo_blocksize
= lo_blocksize
;
810 lo
->lo_device
= bdev
;
811 lo
->lo_flags
= lo_flags
;
812 lo
->lo_backing_file
= file
;
813 lo
->transfer
= transfer_none
;
815 lo
->lo_sizelimit
= 0;
816 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
817 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
819 lo
->lo_bio
= lo
->lo_biotail
= NULL
;
822 * set queue make_request_fn, and add limits based on lower level
825 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
826 lo
->lo_queue
->queuedata
= lo
;
827 lo
->lo_queue
->unplug_fn
= loop_unplug
;
829 set_capacity(lo
->lo_disk
, size
);
830 bd_set_size(bdev
, size
<< 9);
832 set_blocksize(bdev
, lo_blocksize
);
834 lo
->lo_thread
= kthread_create(loop_thread
, lo
, "loop%d",
836 if (IS_ERR(lo
->lo_thread
)) {
837 error
= PTR_ERR(lo
->lo_thread
);
840 lo
->lo_state
= Lo_bound
;
841 wake_up_process(lo
->lo_thread
);
843 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
847 lo
->lo_thread
= NULL
;
848 lo
->lo_device
= NULL
;
849 lo
->lo_backing_file
= NULL
;
851 set_capacity(lo
->lo_disk
, 0);
852 invalidate_bdev(bdev
);
853 bd_set_size(bdev
, 0);
854 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
);
855 lo
->lo_state
= Lo_unbound
;
859 /* This is safe: open() is still holding a reference. */
860 module_put(THIS_MODULE
);
865 loop_release_xfer(struct loop_device
*lo
)
868 struct loop_func_table
*xfer
= lo
->lo_encryption
;
872 err
= xfer
->release(lo
);
874 lo
->lo_encryption
= NULL
;
875 module_put(xfer
->owner
);
881 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
882 const struct loop_info64
*i
)
887 struct module
*owner
= xfer
->owner
;
889 if (!try_module_get(owner
))
892 err
= xfer
->init(lo
, i
);
896 lo
->lo_encryption
= xfer
;
901 static int loop_clr_fd(struct loop_device
*lo
, struct block_device
*bdev
)
903 struct file
*filp
= lo
->lo_backing_file
;
904 gfp_t gfp
= lo
->old_gfp_mask
;
906 if (lo
->lo_state
!= Lo_bound
)
909 if (lo
->lo_refcnt
> 1) /* we needed one fd for the ioctl */
915 spin_lock_irq(&lo
->lo_lock
);
916 lo
->lo_state
= Lo_rundown
;
917 spin_unlock_irq(&lo
->lo_lock
);
919 kthread_stop(lo
->lo_thread
);
921 lo
->lo_queue
->unplug_fn
= NULL
;
922 lo
->lo_backing_file
= NULL
;
924 loop_release_xfer(lo
);
927 lo
->lo_device
= NULL
;
928 lo
->lo_encryption
= NULL
;
930 lo
->lo_sizelimit
= 0;
931 lo
->lo_encrypt_key_size
= 0;
933 lo
->lo_thread
= NULL
;
934 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
935 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
936 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
938 invalidate_bdev(bdev
);
939 set_capacity(lo
->lo_disk
, 0);
941 bd_set_size(bdev
, 0);
942 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
943 lo
->lo_state
= Lo_unbound
;
945 /* This is safe: open() is still holding a reference. */
946 module_put(THIS_MODULE
);
948 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
953 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
956 struct loop_func_table
*xfer
;
957 uid_t uid
= current_uid();
959 if (lo
->lo_encrypt_key_size
&&
960 lo
->lo_key_owner
!= uid
&&
961 !capable(CAP_SYS_ADMIN
))
963 if (lo
->lo_state
!= Lo_bound
)
965 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
968 err
= loop_release_xfer(lo
);
972 if (info
->lo_encrypt_type
) {
973 unsigned int type
= info
->lo_encrypt_type
;
975 if (type
>= MAX_LO_CRYPT
)
977 xfer
= xfer_funcs
[type
];
983 err
= loop_init_xfer(lo
, xfer
, info
);
987 if (lo
->lo_offset
!= info
->lo_offset
||
988 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
989 lo
->lo_offset
= info
->lo_offset
;
990 lo
->lo_sizelimit
= info
->lo_sizelimit
;
991 if (figure_loop_size(lo
))
995 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
996 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
997 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
998 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1002 lo
->transfer
= xfer
->transfer
;
1003 lo
->ioctl
= xfer
->ioctl
;
1005 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1006 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1007 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1009 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1010 lo
->lo_init
[0] = info
->lo_init
[0];
1011 lo
->lo_init
[1] = info
->lo_init
[1];
1012 if (info
->lo_encrypt_key_size
) {
1013 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1014 info
->lo_encrypt_key_size
);
1015 lo
->lo_key_owner
= uid
;
1022 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1024 struct file
*file
= lo
->lo_backing_file
;
1028 if (lo
->lo_state
!= Lo_bound
)
1030 error
= vfs_getattr(file
->f_path
.mnt
, file
->f_path
.dentry
, &stat
);
1033 memset(info
, 0, sizeof(*info
));
1034 info
->lo_number
= lo
->lo_number
;
1035 info
->lo_device
= huge_encode_dev(stat
.dev
);
1036 info
->lo_inode
= stat
.ino
;
1037 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1038 info
->lo_offset
= lo
->lo_offset
;
1039 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1040 info
->lo_flags
= lo
->lo_flags
;
1041 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1042 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1043 info
->lo_encrypt_type
=
1044 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1045 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1046 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1047 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1048 lo
->lo_encrypt_key_size
);
1054 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1056 memset(info64
, 0, sizeof(*info64
));
1057 info64
->lo_number
= info
->lo_number
;
1058 info64
->lo_device
= info
->lo_device
;
1059 info64
->lo_inode
= info
->lo_inode
;
1060 info64
->lo_rdevice
= info
->lo_rdevice
;
1061 info64
->lo_offset
= info
->lo_offset
;
1062 info64
->lo_sizelimit
= 0;
1063 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1064 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1065 info64
->lo_flags
= info
->lo_flags
;
1066 info64
->lo_init
[0] = info
->lo_init
[0];
1067 info64
->lo_init
[1] = info
->lo_init
[1];
1068 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1069 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1071 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1072 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1076 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1078 memset(info
, 0, sizeof(*info
));
1079 info
->lo_number
= info64
->lo_number
;
1080 info
->lo_device
= info64
->lo_device
;
1081 info
->lo_inode
= info64
->lo_inode
;
1082 info
->lo_rdevice
= info64
->lo_rdevice
;
1083 info
->lo_offset
= info64
->lo_offset
;
1084 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1085 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1086 info
->lo_flags
= info64
->lo_flags
;
1087 info
->lo_init
[0] = info64
->lo_init
[0];
1088 info
->lo_init
[1] = info64
->lo_init
[1];
1089 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1090 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1092 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1093 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1095 /* error in case values were truncated */
1096 if (info
->lo_device
!= info64
->lo_device
||
1097 info
->lo_rdevice
!= info64
->lo_rdevice
||
1098 info
->lo_inode
!= info64
->lo_inode
||
1099 info
->lo_offset
!= info64
->lo_offset
)
1106 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1108 struct loop_info info
;
1109 struct loop_info64 info64
;
1111 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1113 loop_info64_from_old(&info
, &info64
);
1114 return loop_set_status(lo
, &info64
);
1118 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1120 struct loop_info64 info64
;
1122 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1124 return loop_set_status(lo
, &info64
);
1128 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1129 struct loop_info info
;
1130 struct loop_info64 info64
;
1136 err
= loop_get_status(lo
, &info64
);
1138 err
= loop_info64_to_old(&info64
, &info
);
1139 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1146 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1147 struct loop_info64 info64
;
1153 err
= loop_get_status(lo
, &info64
);
1154 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1160 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1161 unsigned int cmd
, unsigned long arg
)
1163 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1166 mutex_lock(&lo
->lo_ctl_mutex
);
1169 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1171 case LOOP_CHANGE_FD
:
1172 err
= loop_change_fd(lo
, bdev
, arg
);
1175 err
= loop_clr_fd(lo
, bdev
);
1177 case LOOP_SET_STATUS
:
1178 err
= loop_set_status_old(lo
, (struct loop_info __user
*) arg
);
1180 case LOOP_GET_STATUS
:
1181 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1183 case LOOP_SET_STATUS64
:
1184 err
= loop_set_status64(lo
, (struct loop_info64 __user
*) arg
);
1186 case LOOP_GET_STATUS64
:
1187 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1190 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1192 mutex_unlock(&lo
->lo_ctl_mutex
);
1196 #ifdef CONFIG_COMPAT
1197 struct compat_loop_info
{
1198 compat_int_t lo_number
; /* ioctl r/o */
1199 compat_dev_t lo_device
; /* ioctl r/o */
1200 compat_ulong_t lo_inode
; /* ioctl r/o */
1201 compat_dev_t lo_rdevice
; /* ioctl r/o */
1202 compat_int_t lo_offset
;
1203 compat_int_t lo_encrypt_type
;
1204 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1205 compat_int_t lo_flags
; /* ioctl r/o */
1206 char lo_name
[LO_NAME_SIZE
];
1207 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1208 compat_ulong_t lo_init
[2];
1213 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1214 * - noinlined to reduce stack space usage in main part of driver
1217 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1218 struct loop_info64
*info64
)
1220 struct compat_loop_info info
;
1222 if (copy_from_user(&info
, arg
, sizeof(info
)))
1225 memset(info64
, 0, sizeof(*info64
));
1226 info64
->lo_number
= info
.lo_number
;
1227 info64
->lo_device
= info
.lo_device
;
1228 info64
->lo_inode
= info
.lo_inode
;
1229 info64
->lo_rdevice
= info
.lo_rdevice
;
1230 info64
->lo_offset
= info
.lo_offset
;
1231 info64
->lo_sizelimit
= 0;
1232 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1233 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1234 info64
->lo_flags
= info
.lo_flags
;
1235 info64
->lo_init
[0] = info
.lo_init
[0];
1236 info64
->lo_init
[1] = info
.lo_init
[1];
1237 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1238 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1240 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1241 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1246 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1247 * - noinlined to reduce stack space usage in main part of driver
1250 loop_info64_to_compat(const struct loop_info64
*info64
,
1251 struct compat_loop_info __user
*arg
)
1253 struct compat_loop_info info
;
1255 memset(&info
, 0, sizeof(info
));
1256 info
.lo_number
= info64
->lo_number
;
1257 info
.lo_device
= info64
->lo_device
;
1258 info
.lo_inode
= info64
->lo_inode
;
1259 info
.lo_rdevice
= info64
->lo_rdevice
;
1260 info
.lo_offset
= info64
->lo_offset
;
1261 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1262 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1263 info
.lo_flags
= info64
->lo_flags
;
1264 info
.lo_init
[0] = info64
->lo_init
[0];
1265 info
.lo_init
[1] = info64
->lo_init
[1];
1266 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1267 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1269 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1270 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1272 /* error in case values were truncated */
1273 if (info
.lo_device
!= info64
->lo_device
||
1274 info
.lo_rdevice
!= info64
->lo_rdevice
||
1275 info
.lo_inode
!= info64
->lo_inode
||
1276 info
.lo_offset
!= info64
->lo_offset
||
1277 info
.lo_init
[0] != info64
->lo_init
[0] ||
1278 info
.lo_init
[1] != info64
->lo_init
[1])
1281 if (copy_to_user(arg
, &info
, sizeof(info
)))
1287 loop_set_status_compat(struct loop_device
*lo
,
1288 const struct compat_loop_info __user
*arg
)
1290 struct loop_info64 info64
;
1293 ret
= loop_info64_from_compat(arg
, &info64
);
1296 return loop_set_status(lo
, &info64
);
1300 loop_get_status_compat(struct loop_device
*lo
,
1301 struct compat_loop_info __user
*arg
)
1303 struct loop_info64 info64
;
1309 err
= loop_get_status(lo
, &info64
);
1311 err
= loop_info64_to_compat(&info64
, arg
);
1315 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1316 unsigned int cmd
, unsigned long arg
)
1318 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1322 case LOOP_SET_STATUS
:
1323 mutex_lock(&lo
->lo_ctl_mutex
);
1324 err
= loop_set_status_compat(
1325 lo
, (const struct compat_loop_info __user
*) arg
);
1326 mutex_unlock(&lo
->lo_ctl_mutex
);
1328 case LOOP_GET_STATUS
:
1329 mutex_lock(&lo
->lo_ctl_mutex
);
1330 err
= loop_get_status_compat(
1331 lo
, (struct compat_loop_info __user
*) arg
);
1332 mutex_unlock(&lo
->lo_ctl_mutex
);
1335 case LOOP_GET_STATUS64
:
1336 case LOOP_SET_STATUS64
:
1337 arg
= (unsigned long) compat_ptr(arg
);
1339 case LOOP_CHANGE_FD
:
1340 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1350 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1352 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1354 mutex_lock(&lo
->lo_ctl_mutex
);
1356 mutex_unlock(&lo
->lo_ctl_mutex
);
1361 static int lo_release(struct gendisk
*disk
, fmode_t mode
)
1363 struct loop_device
*lo
= disk
->private_data
;
1365 mutex_lock(&lo
->lo_ctl_mutex
);
1367 if (--lo
->lo_refcnt
)
1370 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1372 * In autoclear mode, stop the loop thread
1373 * and remove configuration after last close.
1375 loop_clr_fd(lo
, NULL
);
1378 * Otherwise keep thread (if running) and config,
1379 * but flush possible ongoing bios in thread.
1385 mutex_unlock(&lo
->lo_ctl_mutex
);
1390 static struct block_device_operations lo_fops
= {
1391 .owner
= THIS_MODULE
,
1393 .release
= lo_release
,
1395 #ifdef CONFIG_COMPAT
1396 .compat_ioctl
= lo_compat_ioctl
,
1401 * And now the modules code and kernel interface.
1403 static int max_loop
;
1404 module_param(max_loop
, int, 0);
1405 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1406 module_param(max_part
, int, 0);
1407 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1408 MODULE_LICENSE("GPL");
1409 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1411 int loop_register_transfer(struct loop_func_table
*funcs
)
1413 unsigned int n
= funcs
->number
;
1415 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1417 xfer_funcs
[n
] = funcs
;
1421 int loop_unregister_transfer(int number
)
1423 unsigned int n
= number
;
1424 struct loop_device
*lo
;
1425 struct loop_func_table
*xfer
;
1427 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1430 xfer_funcs
[n
] = NULL
;
1432 list_for_each_entry(lo
, &loop_devices
, lo_list
) {
1433 mutex_lock(&lo
->lo_ctl_mutex
);
1435 if (lo
->lo_encryption
== xfer
)
1436 loop_release_xfer(lo
);
1438 mutex_unlock(&lo
->lo_ctl_mutex
);
1444 EXPORT_SYMBOL(loop_register_transfer
);
1445 EXPORT_SYMBOL(loop_unregister_transfer
);
1447 static struct loop_device
*loop_alloc(int i
)
1449 struct loop_device
*lo
;
1450 struct gendisk
*disk
;
1452 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1456 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
1460 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1462 goto out_free_queue
;
1464 mutex_init(&lo
->lo_ctl_mutex
);
1466 lo
->lo_thread
= NULL
;
1467 init_waitqueue_head(&lo
->lo_event
);
1468 spin_lock_init(&lo
->lo_lock
);
1469 disk
->major
= LOOP_MAJOR
;
1470 disk
->first_minor
= i
<< part_shift
;
1471 disk
->fops
= &lo_fops
;
1472 disk
->private_data
= lo
;
1473 disk
->queue
= lo
->lo_queue
;
1474 sprintf(disk
->disk_name
, "loop%d", i
);
1478 blk_cleanup_queue(lo
->lo_queue
);
1485 static void loop_free(struct loop_device
*lo
)
1487 blk_cleanup_queue(lo
->lo_queue
);
1488 put_disk(lo
->lo_disk
);
1489 list_del(&lo
->lo_list
);
1493 static struct loop_device
*loop_init_one(int i
)
1495 struct loop_device
*lo
;
1497 list_for_each_entry(lo
, &loop_devices
, lo_list
) {
1498 if (lo
->lo_number
== i
)
1504 add_disk(lo
->lo_disk
);
1505 list_add_tail(&lo
->lo_list
, &loop_devices
);
1510 static void loop_del_one(struct loop_device
*lo
)
1512 del_gendisk(lo
->lo_disk
);
1516 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1518 struct loop_device
*lo
;
1519 struct kobject
*kobj
;
1521 mutex_lock(&loop_devices_mutex
);
1522 lo
= loop_init_one(dev
& MINORMASK
);
1523 kobj
= lo
? get_disk(lo
->lo_disk
) : ERR_PTR(-ENOMEM
);
1524 mutex_unlock(&loop_devices_mutex
);
1530 static int __init
loop_init(void)
1533 unsigned long range
;
1534 struct loop_device
*lo
, *next
;
1537 * loop module now has a feature to instantiate underlying device
1538 * structure on-demand, provided that there is an access dev node.
1539 * However, this will not work well with user space tool that doesn't
1540 * know about such "feature". In order to not break any existing
1541 * tool, we do the following:
1543 * (1) if max_loop is specified, create that many upfront, and this
1544 * also becomes a hard limit.
1545 * (2) if max_loop is not specified, create 8 loop device on module
1546 * load, user can further extend loop device by create dev node
1547 * themselves and have kernel automatically instantiate actual
1553 part_shift
= fls(max_part
);
1555 if (max_loop
> 1UL << (MINORBITS
- part_shift
))
1563 range
= 1UL << (MINORBITS
- part_shift
);
1566 if (register_blkdev(LOOP_MAJOR
, "loop"))
1569 for (i
= 0; i
< nr
; i
++) {
1573 list_add_tail(&lo
->lo_list
, &loop_devices
);
1576 /* point of no return */
1578 list_for_each_entry(lo
, &loop_devices
, lo_list
)
1579 add_disk(lo
->lo_disk
);
1581 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1582 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1584 printk(KERN_INFO
"loop: module loaded\n");
1588 printk(KERN_INFO
"loop: out of memory\n");
1590 list_for_each_entry_safe(lo
, next
, &loop_devices
, lo_list
)
1593 unregister_blkdev(LOOP_MAJOR
, "loop");
1597 static void __exit
loop_exit(void)
1599 unsigned long range
;
1600 struct loop_device
*lo
, *next
;
1602 range
= max_loop
? max_loop
: 1UL << (MINORBITS
- part_shift
);
1604 list_for_each_entry_safe(lo
, next
, &loop_devices
, lo_list
)
1607 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1608 unregister_blkdev(LOOP_MAJOR
, "loop");
1611 module_init(loop_init
);
1612 module_exit(loop_exit
);
1615 static int __init
max_loop_setup(char *str
)
1617 max_loop
= simple_strtol(str
, NULL
, 0);
1621 __setup("max_loop=", max_loop_setup
);