4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * linux/drivers/block/loop.c
40 * Written by Theodore Ts'o, 3/29/93
42 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
43 * permitted under the GNU General Public License.
45 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
46 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
48 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
50 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
52 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
54 * Loadable modules and other fixes by AK, 1998
56 * Maximum number of loop devices now dynamic via max_loop module parameter.
57 * Russell Kroll <rkroll@exploits.org> 19990701
59 * Maximum number of loop devices when compiled-in now selectable by passing
60 * max_loop=<1-255> to the kernel on boot.
61 * Erik I. Bols?, <eriki@himolde.no>, Oct 31, 1999
63 * Completely rewrite request handling to be make_request_fn style and
64 * non blocking, pushing work to a helper thread. Lots of fixes from
66 * Jens Axboe <axboe@suse.de>, Nov 2000
68 * Support up to 256 loop devices
69 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
71 * Support for falling back on the write file operation when the address space
72 * operations prepare_write and/or commit_write are not available on the
74 * Anton Altaparmakov, 16 Feb 2005
77 * - Advisory locking is ignored here.
78 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
82 #include <linux/module.h>
84 #include <linux/sched.h>
86 #include <linux/file.h>
87 #include <linux/stat.h>
88 #include <linux/errno.h>
89 #include <linux/major.h>
90 #include <linux/wait.h>
91 #include <linux/blkdev.h>
92 #include <linux/blkpg.h>
93 #include <linux/init.h>
94 #include <linux/swap.h>
95 #include <linux/slab.h>
96 #include <linux/suspend.h>
97 #include <linux/writeback.h>
98 #include <linux/buffer_head.h> /* for invalidate_bdev() */
99 #include <linux/completion.h>
100 #include <linux/highmem.h>
101 #include <linux/gfp.h>
102 #include <linux/pagevec.h>
103 #include <linux/uaccess.h>
105 #include "../include/lustre_lib.h"
106 #include "../include/lustre_lite.h"
107 #include "llite_internal.h"
109 #define LLOOP_MAX_SEGMENTS LNET_MAX_IOV
111 /* Possible states of device */
118 struct lloop_device
{
124 struct file
*lo_backing_file
;
125 struct block_device
*lo_device
;
126 unsigned lo_blocksize
;
132 struct bio
*lo_biotail
;
134 struct semaphore lo_sem
;
135 struct mutex lo_ctl_mutex
;
137 wait_queue_head_t lo_bh_wait
;
139 struct request_queue
*lo_queue
;
141 const struct lu_env
*lo_env
;
143 struct ll_dio_pages lo_pvec
;
145 /* data to handle bio for lustre. */
146 struct lo_request_data
{
147 struct page
*lrd_pages
[LLOOP_MAX_SEGMENTS
];
148 loff_t lrd_offsets
[LLOOP_MAX_SEGMENTS
];
156 LO_FLAGS_READ_ONLY
= 1,
159 static int lloop_major
;
160 #define MAX_LOOP_DEFAULT 16
161 static int max_loop
= MAX_LOOP_DEFAULT
;
162 static struct lloop_device
*loop_dev
;
163 static struct gendisk
**disks
;
164 static struct mutex lloop_mutex
;
165 static void *ll_iocontrol_magic
;
167 static loff_t
get_loop_size(struct lloop_device
*lo
, struct file
*file
)
169 loff_t size
, offset
, loopsize
;
171 /* Compute loopsize in bytes */
172 size
= i_size_read(file
->f_mapping
->host
);
173 offset
= lo
->lo_offset
;
174 loopsize
= size
- offset
;
175 if (lo
->lo_sizelimit
> 0 && lo
->lo_sizelimit
< loopsize
)
176 loopsize
= lo
->lo_sizelimit
;
179 * Unfortunately, if we want to do I/O on the device,
180 * the number of 512-byte sectors has to fit into a sector_t.
182 return loopsize
>> 9;
185 static int do_bio_lustrebacked(struct lloop_device
*lo
, struct bio
*head
)
187 const struct lu_env
*env
= lo
->lo_env
;
188 struct cl_io
*io
= &lo
->lo_io
;
189 struct inode
*inode
= file_inode(lo
->lo_backing_file
);
190 struct cl_object
*obj
= ll_i2info(inode
)->lli_clob
;
196 struct bvec_iter iter
;
200 struct ll_dio_pages
*pvec
= &lo
->lo_pvec
;
201 struct page
**pages
= pvec
->ldp_pages
;
202 loff_t
*offsets
= pvec
->ldp_offsets
;
204 truncate_inode_pages(inode
->i_mapping
, 0);
206 /* initialize the IO */
207 memset(io
, 0, sizeof(*io
));
209 ret
= cl_io_init(env
, io
, CIT_MISC
, obj
);
211 return io
->ci_result
;
212 io
->ci_lockreq
= CILR_NEVER
;
214 rw
= bio_data_dir(head
);
215 for (bio
= head
; bio
; bio
= bio
->bi_next
) {
216 LASSERT(rw
== bio_data_dir(bio
));
218 offset
= (pgoff_t
)(bio
->bi_iter
.bi_sector
<< 9) + lo
->lo_offset
;
219 bio_for_each_segment(bvec
, bio
, iter
) {
220 BUG_ON(bvec
.bv_offset
!= 0);
221 BUG_ON(bvec
.bv_len
!= PAGE_SIZE
);
223 pages
[page_count
] = bvec
.bv_page
;
224 offsets
[page_count
] = offset
;
226 offset
+= bvec
.bv_len
;
228 LASSERT(page_count
<= LLOOP_MAX_SEGMENTS
);
231 ll_stats_ops_tally(ll_i2sbi(inode
),
232 (rw
== WRITE
) ? LPROC_LL_BRW_WRITE
: LPROC_LL_BRW_READ
,
235 pvec
->ldp_size
= page_count
<< PAGE_SHIFT
;
236 pvec
->ldp_nr
= page_count
;
238 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
239 * write those pages into OST. Even worse case is that more pages
240 * would be asked to write out to swap space, and then finally get here
242 * Unfortunately this is NOT easy to fix.
243 * Thoughts on solution:
244 * 0. Define a reserved pool for cl_pages, which could be a list of
245 * pre-allocated cl_pages;
246 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
247 * which measures how many layers for this lustre object. Generally
248 * speaking, the depth would be 2, one for llite, and one for lovsub.
249 * However, for SNS, there will be more since we need additional page
251 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
252 * pool. Afterwards, the clio would allocate the pages from reserved
253 * pool, this guarantees we needn't allocate the cl_pages from
254 * generic cl_page slab cache.
255 * Of course, if there is NOT enough pages in the pool, we might
256 * be asked to write less pages once, this purely depends on
257 * implementation. Anyway, we should be careful to avoid deadlocking.
260 bytes
= ll_direct_rw_pages(env
, io
, rw
, inode
, pvec
);
263 return (bytes
== pvec
->ldp_size
) ? 0 : (int)bytes
;
267 * Add bio to back of pending list
269 static void loop_add_bio(struct lloop_device
*lo
, struct bio
*bio
)
273 spin_lock_irqsave(&lo
->lo_lock
, flags
);
274 if (lo
->lo_biotail
) {
275 lo
->lo_biotail
->bi_next
= bio
;
276 lo
->lo_biotail
= bio
;
278 lo
->lo_bio
= lo
->lo_biotail
= bio
;
280 spin_unlock_irqrestore(&lo
->lo_lock
, flags
);
282 atomic_inc(&lo
->lo_pending
);
283 if (waitqueue_active(&lo
->lo_bh_wait
))
284 wake_up(&lo
->lo_bh_wait
);
288 * Grab first pending buffer
290 static unsigned int loop_get_bio(struct lloop_device
*lo
, struct bio
**req
)
294 unsigned int count
= 0;
295 unsigned int page_count
= 0;
298 spin_lock_irq(&lo
->lo_lock
);
300 if (unlikely(!first
)) {
301 spin_unlock_irq(&lo
->lo_lock
);
305 /* TODO: need to split the bio, too bad. */
306 LASSERT(first
->bi_vcnt
<= LLOOP_MAX_SEGMENTS
);
308 rw
= bio_data_dir(first
);
310 while (*bio
&& (*bio
)->bi_rw
== rw
) {
311 CDEBUG(D_INFO
, "bio sector %llu size %u count %u vcnt%u\n",
312 (unsigned long long)(*bio
)->bi_iter
.bi_sector
,
313 (*bio
)->bi_iter
.bi_size
,
314 page_count
, (*bio
)->bi_vcnt
);
315 if (page_count
+ (*bio
)->bi_vcnt
> LLOOP_MAX_SEGMENTS
)
318 page_count
+= (*bio
)->bi_vcnt
;
320 bio
= &(*bio
)->bi_next
;
323 /* Some of bios can't be mergeable. */
327 /* Hit the end of queue */
328 lo
->lo_biotail
= NULL
;
332 spin_unlock_irq(&lo
->lo_lock
);
336 static blk_qc_t
loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
338 struct lloop_device
*lo
= q
->queuedata
;
339 int rw
= bio_rw(old_bio
);
342 blk_queue_split(q
, &old_bio
, q
->bio_split
);
347 CDEBUG(D_INFO
, "submit bio sector %llu size %u\n",
348 (unsigned long long)old_bio
->bi_iter
.bi_sector
,
349 old_bio
->bi_iter
.bi_size
);
351 spin_lock_irq(&lo
->lo_lock
);
352 inactive
= lo
->lo_state
!= LLOOP_BOUND
;
353 spin_unlock_irq(&lo
->lo_lock
);
358 if (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)
360 } else if (rw
== READA
) {
362 } else if (rw
!= READ
) {
363 CERROR("lloop: unknown command (%x)\n", rw
);
366 loop_add_bio(lo
, old_bio
);
367 return BLK_QC_T_NONE
;
369 bio_io_error(old_bio
);
370 return BLK_QC_T_NONE
;
373 static inline void loop_handle_bio(struct lloop_device
*lo
, struct bio
*bio
)
377 ret
= do_bio_lustrebacked(lo
, bio
);
379 struct bio
*tmp
= bio
->bi_next
;
388 static inline int loop_active(struct lloop_device
*lo
)
390 return atomic_read(&lo
->lo_pending
) ||
391 (lo
->lo_state
== LLOOP_RUNDOWN
);
395 * worker thread that handles reads/writes to file backed loop devices,
396 * to avoid blocking in our make_request_fn.
398 static int loop_thread(void *data
)
400 struct lloop_device
*lo
= data
;
403 unsigned long times
= 0;
404 unsigned long total_count
= 0;
410 set_user_nice(current
, MIN_NICE
);
412 lo
->lo_state
= LLOOP_BOUND
;
414 env
= cl_env_get(&refcheck
);
421 memset(&lo
->lo_pvec
, 0, sizeof(lo
->lo_pvec
));
422 lo
->lo_pvec
.ldp_pages
= lo
->lo_requests
[0].lrd_pages
;
423 lo
->lo_pvec
.ldp_offsets
= lo
->lo_requests
[0].lrd_offsets
;
426 * up sem, we are running
431 wait_event(lo
->lo_bh_wait
, loop_active(lo
));
432 if (!atomic_read(&lo
->lo_pending
)) {
435 spin_lock_irq(&lo
->lo_lock
);
436 exiting
= (lo
->lo_state
== LLOOP_RUNDOWN
);
437 spin_unlock_irq(&lo
->lo_lock
);
443 count
= loop_get_bio(lo
, &bio
);
445 CWARN("lloop(minor: %d): missing bio\n", lo
->lo_number
);
449 total_count
+= count
;
450 if (total_count
< count
) { /* overflow */
456 if ((times
& 127) == 0) {
457 CDEBUG(D_INFO
, "total: %lu, count: %lu, avg: %lu\n",
458 total_count
, times
, total_count
/ times
);
462 LASSERT(count
<= atomic_read(&lo
->lo_pending
));
463 loop_handle_bio(lo
, bio
);
464 atomic_sub(count
, &lo
->lo_pending
);
466 cl_env_put(env
, &refcheck
);
473 static int loop_set_fd(struct lloop_device
*lo
, struct file
*unused
,
474 struct block_device
*bdev
, struct file
*file
)
477 struct address_space
*mapping
;
482 if (!try_module_get(THIS_MODULE
))
486 if (lo
->lo_state
!= LLOOP_UNBOUND
)
489 mapping
= file
->f_mapping
;
490 inode
= mapping
->host
;
493 if (!S_ISREG(inode
->i_mode
) || inode
->i_sb
->s_magic
!= LL_SUPER_MAGIC
)
496 if (!(file
->f_mode
& FMODE_WRITE
))
497 lo_flags
|= LO_FLAGS_READ_ONLY
;
499 size
= get_loop_size(lo
, file
);
501 if ((loff_t
)(sector_t
)size
!= size
) {
506 /* remove all pages in cache so as dirty pages not to be existent. */
507 truncate_inode_pages(mapping
, 0);
509 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
511 lo
->lo_blocksize
= PAGE_SIZE
;
512 lo
->lo_device
= bdev
;
513 lo
->lo_flags
= lo_flags
;
514 lo
->lo_backing_file
= file
;
515 lo
->lo_sizelimit
= 0;
516 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
517 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
519 lo
->lo_bio
= lo
->lo_biotail
= NULL
;
522 * set queue make_request_fn, and add limits based on lower level
525 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
526 lo
->lo_queue
->queuedata
= lo
;
528 /* queue parameters */
529 CLASSERT(PAGE_SIZE
< (1 << (sizeof(unsigned short) * 8)));
530 blk_queue_logical_block_size(lo
->lo_queue
,
531 (unsigned short)PAGE_SIZE
);
532 blk_queue_max_hw_sectors(lo
->lo_queue
,
533 LLOOP_MAX_SEGMENTS
<< (PAGE_SHIFT
- 9));
534 blk_queue_max_segments(lo
->lo_queue
, LLOOP_MAX_SEGMENTS
);
536 set_capacity(disks
[lo
->lo_number
], size
);
537 bd_set_size(bdev
, size
<< 9);
539 set_blocksize(bdev
, lo
->lo_blocksize
);
541 kthread_run(loop_thread
, lo
, "lloop%d", lo
->lo_number
);
546 /* This is safe: open() is still holding a reference. */
547 module_put(THIS_MODULE
);
551 static int loop_clr_fd(struct lloop_device
*lo
, struct block_device
*bdev
,
554 struct file
*filp
= lo
->lo_backing_file
;
555 gfp_t gfp
= lo
->old_gfp_mask
;
557 if (lo
->lo_state
!= LLOOP_BOUND
)
560 if (lo
->lo_refcnt
> count
) /* we needed one fd for the ioctl */
566 spin_lock_irq(&lo
->lo_lock
);
567 lo
->lo_state
= LLOOP_RUNDOWN
;
568 spin_unlock_irq(&lo
->lo_lock
);
569 wake_up(&lo
->lo_bh_wait
);
572 lo
->lo_backing_file
= NULL
;
573 lo
->lo_device
= NULL
;
575 lo
->lo_sizelimit
= 0;
577 invalidate_bdev(bdev
);
578 set_capacity(disks
[lo
->lo_number
], 0);
579 bd_set_size(bdev
, 0);
580 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
581 lo
->lo_state
= LLOOP_UNBOUND
;
583 /* This is safe: open() is still holding a reference. */
584 module_put(THIS_MODULE
);
588 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
590 struct lloop_device
*lo
= bdev
->bd_disk
->private_data
;
592 mutex_lock(&lo
->lo_ctl_mutex
);
594 mutex_unlock(&lo
->lo_ctl_mutex
);
599 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
601 struct lloop_device
*lo
= disk
->private_data
;
603 mutex_lock(&lo
->lo_ctl_mutex
);
605 mutex_unlock(&lo
->lo_ctl_mutex
);
608 /* lloop device node's ioctl function. */
609 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
610 unsigned int cmd
, unsigned long arg
)
612 struct lloop_device
*lo
= bdev
->bd_disk
->private_data
;
613 struct inode
*inode
= NULL
;
616 mutex_lock(&lloop_mutex
);
618 case LL_IOC_LLOOP_DETACH
: {
619 err
= loop_clr_fd(lo
, bdev
, 2);
621 blkdev_put(bdev
, 0); /* grabbed in LLOOP_ATTACH */
625 case LL_IOC_LLOOP_INFO
: {
628 if (!lo
->lo_backing_file
) {
633 inode
= file_inode(lo
->lo_backing_file
);
634 if (lo
->lo_state
== LLOOP_BOUND
)
635 fid
= ll_i2info(inode
)->lli_fid
;
639 if (copy_to_user((void __user
*)arg
, &fid
, sizeof(fid
)))
648 mutex_unlock(&lloop_mutex
);
653 static struct block_device_operations lo_fops
= {
654 .owner
= THIS_MODULE
,
656 .release
= lo_release
,
660 /* dynamic iocontrol callback.
661 * This callback is registered in lloop_init and will be called by
664 * This is a llite regular file ioctl function. It takes the responsibility
665 * of attaching or detaching a file by a lloop's device number.
667 static enum llioc_iter
lloop_ioctl(struct inode
*unused
, struct file
*file
,
668 unsigned int cmd
, unsigned long arg
,
669 void *magic
, int *rcp
)
671 struct lloop_device
*lo
= NULL
;
672 struct block_device
*bdev
= NULL
;
676 if (magic
!= ll_iocontrol_magic
)
684 CWARN("Enter llop_ioctl\n");
686 mutex_lock(&lloop_mutex
);
688 case LL_IOC_LLOOP_ATTACH
: {
689 struct lloop_device
*lo_free
= NULL
;
692 for (i
= 0; i
< max_loop
; i
++, lo
= NULL
) {
694 if (lo
->lo_state
== LLOOP_UNBOUND
) {
699 if (file_inode(lo
->lo_backing_file
) == file_inode(file
))
702 if (lo
|| !lo_free
) {
708 dev
= MKDEV(lloop_major
, lo
->lo_number
);
710 /* quit if the used pointer is writable */
711 if (put_user((long)old_encode_dev(dev
), (long __user
*)arg
)) {
716 bdev
= blkdev_get_by_dev(dev
, file
->f_mode
, NULL
);
723 err
= loop_set_fd(lo
, NULL
, bdev
, file
);
732 case LL_IOC_LLOOP_DETACH_BYDEV
: {
735 dev
= old_decode_dev(arg
);
736 if (MAJOR(dev
) != lloop_major
) {
742 if (minor
> max_loop
- 1) {
747 lo
= &loop_dev
[minor
];
748 if (lo
->lo_state
!= LLOOP_BOUND
) {
753 bdev
= lo
->lo_device
;
754 err
= loop_clr_fd(lo
, bdev
, 1);
756 blkdev_put(bdev
, 0); /* grabbed in LLOOP_ATTACH */
767 mutex_unlock(&lloop_mutex
);
774 static int __init
lloop_init(void)
777 unsigned int cmdlist
[] = {
779 LL_IOC_LLOOP_DETACH_BYDEV
,
782 if (max_loop
< 1 || max_loop
> 256) {
783 max_loop
= MAX_LOOP_DEFAULT
;
784 CWARN("lloop: invalid max_loop (must be between 1 and 256), using default (%u)\n",
788 lloop_major
= register_blkdev(0, "lloop");
792 CDEBUG(D_CONFIG
, "registered lloop major %d with %u minors\n",
793 lloop_major
, max_loop
);
795 ll_iocontrol_magic
= ll_iocontrol_register(lloop_ioctl
, 2, cmdlist
);
796 if (!ll_iocontrol_magic
)
799 loop_dev
= kcalloc(max_loop
, sizeof(*loop_dev
), GFP_KERNEL
);
803 disks
= kcalloc(max_loop
, sizeof(*disks
), GFP_KERNEL
);
807 for (i
= 0; i
< max_loop
; i
++) {
808 disks
[i
] = alloc_disk(1);
813 mutex_init(&lloop_mutex
);
815 for (i
= 0; i
< max_loop
; i
++) {
816 struct lloop_device
*lo
= &loop_dev
[i
];
817 struct gendisk
*disk
= disks
[i
];
819 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
823 mutex_init(&lo
->lo_ctl_mutex
);
824 sema_init(&lo
->lo_sem
, 0);
825 init_waitqueue_head(&lo
->lo_bh_wait
);
827 spin_lock_init(&lo
->lo_lock
);
828 disk
->major
= lloop_major
;
829 disk
->first_minor
= i
;
830 disk
->fops
= &lo_fops
;
831 sprintf(disk
->disk_name
, "lloop%d", i
);
832 disk
->private_data
= lo
;
833 disk
->queue
= lo
->lo_queue
;
836 /* We cannot fail after we call this, so another loop!*/
837 for (i
= 0; i
< max_loop
; i
++)
843 blk_cleanup_queue(loop_dev
[i
].lo_queue
);
852 unregister_blkdev(lloop_major
, "lloop");
853 ll_iocontrol_unregister(ll_iocontrol_magic
);
854 CERROR("lloop: ran out of memory\n");
858 static void lloop_exit(void)
862 ll_iocontrol_unregister(ll_iocontrol_magic
);
863 for (i
= 0; i
< max_loop
; i
++) {
864 del_gendisk(disks
[i
]);
865 blk_cleanup_queue(loop_dev
[i
].lo_queue
);
869 unregister_blkdev(lloop_major
, "lloop");
875 module_param(max_loop
, int, 0444);
876 MODULE_PARM_DESC(max_loop
, "maximum of lloop_device");
877 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
878 MODULE_DESCRIPTION("Lustre virtual block device");
879 MODULE_VERSION(LUSTRE_VERSION_STRING
);
880 MODULE_LICENSE("GPL");
882 module_init(lloop_init
);
883 module_exit(lloop_exit
);