3 osdblk.c -- Export a single SCSI OSD object as a Linux block device
6 Copyright 2009 Red Hat, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; see the file COPYING. If not, write to
19 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 1) Map a Linux block device to an existing OSD object.
27 In this example, we will use partition id 1234, object id 5678,
30 $ echo "1234 5678 /dev/osd1" > /sys/class/osdblk/add
33 2) List all active blkdev<->object mappings.
35 In this example, we have performed step #1 twice, creating two blkdevs,
36 mapped to two separate OSD objects.
38 $ cat /sys/class/osdblk/list
39 0 174 1234 5678 /dev/osd1
40 1 179 1994 897123 /dev/osd0
42 The columns, in order, are:
44 - blkdev assigned major
45 - OSD object partition id
50 3) Remove an active blkdev<->object mapping.
52 In this example, we remove the mapping with blkdev unique id 1.
54 $ echo 1 > /sys/class/osdblk/remove
57 NOTE: The actual creation and deletion of OSD objects is outside the scope
62 #include <linux/kernel.h>
63 #include <linux/device.h>
64 #include <linux/module.h>
66 #include <linux/slab.h>
67 #include <scsi/osd_initiator.h>
68 #include <scsi/osd_attributes.h>
69 #include <scsi/osd_sec.h>
70 #include <scsi/scsi_device.h>
72 #define DRV_NAME "osdblk"
73 #define PFX DRV_NAME ": "
75 /* #define _OSDBLK_DEBUG */
77 #define OSDBLK_DEBUG(fmt, a...) \
78 printk(KERN_NOTICE "osdblk @%s:%d: " fmt, __func__, __LINE__, ##a)
80 #define OSDBLK_DEBUG(fmt, a...) \
81 do { if (0) printk(fmt, ##a); } while (0)
84 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
85 MODULE_DESCRIPTION("block device inside an OSD object osdblk.ko");
86 MODULE_LICENSE("GPL");
91 OSDBLK_MINORS_PER_MAJOR
= 256, /* max minors per blkdev */
92 OSDBLK_MAX_REQ
= 32, /* max parallel requests */
93 OSDBLK_OP_TIMEOUT
= 4 * 60, /* sync OSD req timeout */
96 struct osdblk_request
{
97 struct request
*rq
; /* blk layer request */
98 struct bio
*bio
; /* cloned bio */
99 struct osdblk_device
*osdev
; /* associated blkdev */
102 struct osdblk_device
{
103 int id
; /* blkdev unique id */
105 int major
; /* blkdev assigned major */
106 struct gendisk
*disk
; /* blkdev's gendisk and rq */
107 struct request_queue
*q
;
109 struct osd_dev
*osd
; /* associated OSD */
111 char name
[32]; /* blkdev name, e.g. osdblk34 */
113 spinlock_t lock
; /* queue lock */
115 struct osd_obj_id obj
; /* OSD partition, obj id */
116 uint8_t obj_cred
[OSD_CAP_LEN
]; /* OSD cred */
118 struct osdblk_request req
[OSDBLK_MAX_REQ
]; /* request table */
120 struct list_head node
;
122 char osd_path
[0]; /* OSD device path */
125 static struct class *class_osdblk
; /* /sys/class/osdblk */
126 static DEFINE_MUTEX(ctl_mutex
); /* Serialize open/close/setup/teardown */
127 static LIST_HEAD(osdblkdev_list
);
129 static const struct block_device_operations osdblk_bd_ops
= {
130 .owner
= THIS_MODULE
,
133 static const struct osd_attr g_attr_logical_length
= ATTR_DEF(
134 OSD_APAGE_OBJECT_INFORMATION
, OSD_ATTR_OI_LOGICAL_LENGTH
, 8);
136 static void osdblk_make_credential(u8 cred_a
[OSD_CAP_LEN
],
137 const struct osd_obj_id
*obj
)
139 osd_sec_init_nosec_doall_caps(cred_a
, obj
, false, true);
142 /* copied from exofs; move to libosd? */
144 * Perform a synchronous OSD operation. copied from exofs; move to libosd?
146 static int osd_sync_op(struct osd_request
*or, int timeout
, uint8_t *credential
)
150 or->timeout
= timeout
;
151 ret
= osd_finalize_request(or, 0, credential
, NULL
);
155 ret
= osd_execute_request(or);
157 /* osd_req_decode_sense(or, ret); */
162 * Perform an asynchronous OSD operation. copied from exofs; move to libosd?
164 static int osd_async_op(struct osd_request
*or, osd_req_done_fn
*async_done
,
165 void *caller_context
, u8
*cred
)
169 ret
= osd_finalize_request(or, 0, cred
, NULL
);
173 ret
= osd_execute_request_async(or, async_done
, caller_context
);
178 /* copied from exofs; move to libosd? */
179 static int extract_attr_from_req(struct osd_request
*or, struct osd_attr
*attr
)
181 struct osd_attr cur_attr
= {.attr_page
= 0}; /* start with zeros */
187 osd_req_decode_get_attr_list(or, &cur_attr
, &nelem
, &iter
);
188 if ((cur_attr
.attr_page
== attr
->attr_page
) &&
189 (cur_attr
.attr_id
== attr
->attr_id
)) {
190 attr
->len
= cur_attr
.len
;
191 attr
->val_ptr
= cur_attr
.val_ptr
;
199 static int osdblk_get_obj_size(struct osdblk_device
*osdev
, u64
*size_out
)
201 struct osd_request
*or;
202 struct osd_attr attr
;
206 or = osd_start_request(osdev
->osd
, GFP_KERNEL
);
210 /* create a get-attributes(length) request */
211 osd_req_get_attributes(or, &osdev
->obj
);
213 osd_req_add_get_attr_list(or, &g_attr_logical_length
, 1);
215 /* execute op synchronously */
216 ret
= osd_sync_op(or, OSDBLK_OP_TIMEOUT
, osdev
->obj_cred
);
220 /* extract length from returned attribute info */
221 attr
= g_attr_logical_length
;
222 ret
= extract_attr_from_req(or, &attr
);
226 *size_out
= get_unaligned_be64(attr
.val_ptr
);
234 static void osdblk_osd_complete(struct osd_request
*or, void *private)
236 struct osdblk_request
*orq
= private;
237 struct osd_sense_info osi
;
238 int ret
= osd_req_decode_sense(or, &osi
);
242 OSDBLK_DEBUG("osdblk_osd_complete with err=%d\n", ret
);
245 /* complete OSD request */
248 /* complete request passed to osdblk by block layer */
249 __blk_end_request_all(orq
->rq
, ret
);
252 static void bio_chain_put(struct bio
*chain
)
258 chain
= chain
->bi_next
;
264 static struct bio
*bio_chain_clone(struct bio
*old_chain
, gfp_t gfpmask
)
266 struct bio
*tmp
, *new_chain
= NULL
, *tail
= NULL
;
269 tmp
= bio_kmalloc(gfpmask
, old_chain
->bi_max_vecs
);
273 __bio_clone(tmp
, old_chain
);
275 gfpmask
&= ~__GFP_WAIT
;
279 new_chain
= tail
= tmp
;
285 old_chain
= old_chain
->bi_next
;
291 OSDBLK_DEBUG("bio_chain_clone with err\n");
292 bio_chain_put(new_chain
);
296 static void osdblk_rq_fn(struct request_queue
*q
)
298 struct osdblk_device
*osdev
= q
->queuedata
;
302 struct osdblk_request
*orq
;
303 struct osd_request
*or;
305 bool do_write
, do_flush
;
307 /* peek at request from block layer */
308 rq
= blk_fetch_request(q
);
312 /* filter out block requests we don't understand */
313 if (rq
->cmd_type
!= REQ_TYPE_FS
) {
314 blk_end_request_all(rq
, 0);
318 /* deduce our operation (read, write, flush) */
319 /* I wish the block layer simplified cmd_type/cmd_flags/cmd[]
320 * into a clearly defined set of RPC commands:
321 * read, write, flush, scsi command, power mgmt req,
322 * driver-specific, etc.
325 do_flush
= rq
->cmd_flags
& REQ_FLUSH
;
326 do_write
= (rq_data_dir(rq
) == WRITE
);
328 if (!do_flush
) { /* osd_flush does not use a bio */
329 /* a bio clone to be passed down to OSD request */
330 bio
= bio_chain_clone(rq
->bio
, GFP_ATOMIC
);
336 /* alloc internal OSD request, for OSD command execution */
337 or = osd_start_request(osdev
->osd
, GFP_ATOMIC
);
340 OSDBLK_DEBUG("osd_start_request with err\n");
344 orq
= &osdev
->req
[rq
->tag
];
349 /* init OSD command: flush, write or read */
351 osd_req_flush_object(or, &osdev
->obj
,
352 OSD_CDB_FLUSH_ALL
, 0, 0);
354 osd_req_write(or, &osdev
->obj
, blk_rq_pos(rq
) * 512ULL,
355 bio
, blk_rq_bytes(rq
));
357 osd_req_read(or, &osdev
->obj
, blk_rq_pos(rq
) * 512ULL,
358 bio
, blk_rq_bytes(rq
));
360 OSDBLK_DEBUG("%s 0x%x bytes at 0x%llx\n",
361 do_flush
? "flush" : do_write
?
362 "write" : "read", blk_rq_bytes(rq
),
363 blk_rq_pos(rq
) * 512ULL);
365 /* begin OSD command execution */
366 if (osd_async_op(or, osdblk_osd_complete
, orq
,
369 blk_requeue_request(q
, rq
);
371 OSDBLK_DEBUG("osd_execute_request_async with err\n");
375 /* remove the special 'flush' marker, now that the command
382 static void osdblk_free_disk(struct osdblk_device
*osdev
)
384 struct gendisk
*disk
= osdev
->disk
;
389 if (disk
->flags
& GENHD_FL_UP
)
392 blk_cleanup_queue(disk
->queue
);
396 static int osdblk_init_disk(struct osdblk_device
*osdev
)
398 struct gendisk
*disk
;
399 struct request_queue
*q
;
403 /* contact OSD, request size info about the object being mapped */
404 rc
= osdblk_get_obj_size(osdev
, &obj_size
);
408 /* create gendisk info */
409 disk
= alloc_disk(OSDBLK_MINORS_PER_MAJOR
);
413 sprintf(disk
->disk_name
, DRV_NAME
"%d", osdev
->id
);
414 disk
->major
= osdev
->major
;
415 disk
->first_minor
= 0;
416 disk
->fops
= &osdblk_bd_ops
;
417 disk
->private_data
= osdev
;
420 q
= blk_init_queue(osdblk_rq_fn
, &osdev
->lock
);
426 /* switch queue to TCQ mode; allocate tag map */
427 rc
= blk_queue_init_tags(q
, OSDBLK_MAX_REQ
, NULL
);
429 blk_cleanup_queue(q
);
434 /* Set our limits to the lower device limits, because osdblk cannot
435 * sleep when allocating a lower-request and therefore cannot be
438 blk_queue_stack_limits(q
, osd_request_queue(osdev
->osd
));
440 blk_queue_prep_rq(q
, blk_queue_start_tag
);
441 blk_queue_flush(q
, REQ_FLUSH
);
445 q
->queuedata
= osdev
;
450 /* finally, announce the disk to the world */
451 set_capacity(disk
, obj_size
/ 512ULL);
454 printk(KERN_INFO
"%s: Added of size 0x%llx\n",
455 disk
->disk_name
, (unsigned long long)obj_size
);
460 /********************************************************************
462 * add map OSD object to blkdev
463 * remove unmap OSD object
465 *******************************************************************/
467 static void class_osdblk_release(struct class *cls
)
472 static ssize_t
class_osdblk_list(struct class *c
,
473 struct class_attribute
*attr
,
477 struct list_head
*tmp
;
479 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
481 list_for_each(tmp
, &osdblkdev_list
) {
482 struct osdblk_device
*osdev
;
484 osdev
= list_entry(tmp
, struct osdblk_device
, node
);
486 n
+= sprintf(data
+n
, "%d %d %llu %llu %s\n",
489 osdev
->obj
.partition
,
494 mutex_unlock(&ctl_mutex
);
498 static ssize_t
class_osdblk_add(struct class *c
,
499 struct class_attribute
*attr
,
500 const char *buf
, size_t count
)
502 struct osdblk_device
*osdev
;
505 struct list_head
*tmp
;
507 if (!try_module_get(THIS_MODULE
))
510 /* new osdblk_device object */
511 osdev
= kzalloc(sizeof(*osdev
) + strlen(buf
) + 1, GFP_KERNEL
);
517 /* static osdblk_device initialization */
518 spin_lock_init(&osdev
->lock
);
519 INIT_LIST_HEAD(&osdev
->node
);
521 /* generate unique id: find highest unique id, add one */
523 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
525 list_for_each(tmp
, &osdblkdev_list
) {
526 struct osdblk_device
*osdev
;
528 osdev
= list_entry(tmp
, struct osdblk_device
, node
);
529 if (osdev
->id
> new_id
)
530 new_id
= osdev
->id
+ 1;
535 /* add to global list */
536 list_add_tail(&osdev
->node
, &osdblkdev_list
);
538 mutex_unlock(&ctl_mutex
);
540 /* parse add command */
541 if (sscanf(buf
, "%llu %llu %s", &osdev
->obj
.partition
, &osdev
->obj
.id
,
542 osdev
->osd_path
) != 3) {
547 /* initialize rest of new object */
548 sprintf(osdev
->name
, DRV_NAME
"%d", osdev
->id
);
550 /* contact requested OSD */
551 osdev
->osd
= osduld_path_lookup(osdev
->osd_path
);
552 if (IS_ERR(osdev
->osd
)) {
553 rc
= PTR_ERR(osdev
->osd
);
557 /* build OSD credential */
558 osdblk_make_credential(osdev
->obj_cred
, &osdev
->obj
);
560 /* register our block device */
561 irc
= register_blkdev(0, osdev
->name
);
569 /* set up and announce blkdev mapping */
570 rc
= osdblk_init_disk(osdev
);
577 unregister_blkdev(osdev
->major
, osdev
->name
);
579 osduld_put_device(osdev
->osd
);
581 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
582 list_del_init(&osdev
->node
);
583 mutex_unlock(&ctl_mutex
);
587 OSDBLK_DEBUG("Error adding device %s\n", buf
);
588 module_put(THIS_MODULE
);
592 static ssize_t
class_osdblk_remove(struct class *c
,
593 struct class_attribute
*attr
,
597 struct osdblk_device
*osdev
= NULL
;
600 struct list_head
*tmp
;
602 rc
= strict_strtoul(buf
, 10, &ul
);
606 /* convert to int; abort if we lost anything in the conversion */
607 target_id
= (int) ul
;
611 /* remove object from list immediately */
612 mutex_lock_nested(&ctl_mutex
, SINGLE_DEPTH_NESTING
);
614 list_for_each(tmp
, &osdblkdev_list
) {
615 osdev
= list_entry(tmp
, struct osdblk_device
, node
);
616 if (osdev
->id
== target_id
) {
617 list_del_init(&osdev
->node
);
623 mutex_unlock(&ctl_mutex
);
628 /* clean up and free blkdev and associated OSD connection */
629 osdblk_free_disk(osdev
);
630 unregister_blkdev(osdev
->major
, osdev
->name
);
631 osduld_put_device(osdev
->osd
);
634 /* release module ref */
635 module_put(THIS_MODULE
);
640 static struct class_attribute class_osdblk_attrs
[] = {
641 __ATTR(add
, 0200, NULL
, class_osdblk_add
),
642 __ATTR(remove
, 0200, NULL
, class_osdblk_remove
),
643 __ATTR(list
, 0444, class_osdblk_list
, NULL
),
647 static int osdblk_sysfs_init(void)
652 * create control files in sysfs
653 * /sys/class/osdblk/...
655 class_osdblk
= kzalloc(sizeof(*class_osdblk
), GFP_KERNEL
);
659 class_osdblk
->name
= DRV_NAME
;
660 class_osdblk
->owner
= THIS_MODULE
;
661 class_osdblk
->class_release
= class_osdblk_release
;
662 class_osdblk
->class_attrs
= class_osdblk_attrs
;
664 ret
= class_register(class_osdblk
);
668 printk(PFX
"failed to create class osdblk\n");
675 static void osdblk_sysfs_cleanup(void)
678 class_destroy(class_osdblk
);
682 static int __init
osdblk_init(void)
686 rc
= osdblk_sysfs_init();
693 static void __exit
osdblk_exit(void)
695 osdblk_sysfs_cleanup();
698 module_init(osdblk_init
);
699 module_exit(osdblk_exit
);