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 !(rq->cmd_flags & REQ_HARDBARRIER)) {
315 blk_end_request_all(rq, 0);
319 /* deduce our operation (read, write, flush) */
320 /* I wish the block layer simplified cmd_type/cmd_flags/cmd[]
321 * into a clearly defined set of RPC commands:
322 * read, write, flush, scsi command, power mgmt req,
323 * driver-specific, etc.
326 do_flush = (rq->special == (void *) 0xdeadbeefUL);
327 do_write = (rq_data_dir(rq) == WRITE);
329 if (!do_flush) { /* osd_flush does not use a bio */
330 /* a bio clone to be passed down to OSD request */
331 bio = bio_chain_clone(rq->bio, GFP_ATOMIC);
337 /* alloc internal OSD request, for OSD command execution */
338 or = osd_start_request(osdev->osd, GFP_ATOMIC);
341 OSDBLK_DEBUG("osd_start_request with err\n");
345 orq = &osdev->req[rq->tag];
350 /* init OSD command: flush, write or read */
352 osd_req_flush_object(or, &osdev->obj,
353 OSD_CDB_FLUSH_ALL, 0, 0);
355 osd_req_write(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
356 bio, blk_rq_bytes(rq));
358 osd_req_read(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
359 bio, blk_rq_bytes(rq));
361 OSDBLK_DEBUG("%s 0x%x bytes at 0x%llx\n",
362 do_flush ? "flush" : do_write ?
363 "write" : "read", blk_rq_bytes(rq),
364 blk_rq_pos(rq) * 512ULL);
366 /* begin OSD command execution */
367 if (osd_async_op(or, osdblk_osd_complete, orq,
370 blk_requeue_request(q, rq);
372 OSDBLK_DEBUG("osd_execute_request_async with err\n");
376 /* remove the special 'flush' marker, now that the command
383 static void osdblk_prepare_flush(struct request_queue *q, struct request *rq)
385 /* add driver-specific marker, to indicate that this request
388 rq->special = (void *) 0xdeadbeefUL;
391 static void osdblk_free_disk(struct osdblk_device *osdev)
393 struct gendisk *disk = osdev->disk;
398 if (disk->flags & GENHD_FL_UP)
401 blk_cleanup_queue(disk->queue);
405 static int osdblk_init_disk(struct osdblk_device *osdev)
407 struct gendisk *disk;
408 struct request_queue *q;
412 /* contact OSD, request size info about the object being mapped */
413 rc = osdblk_get_obj_size(osdev, &obj_size);
417 /* create gendisk info */
418 disk = alloc_disk(OSDBLK_MINORS_PER_MAJOR);
422 sprintf(disk->disk_name, DRV_NAME "%d", osdev->id);
423 disk->major = osdev->major;
424 disk->first_minor = 0;
425 disk->fops = &osdblk_bd_ops;
426 disk->private_data = osdev;
429 q = blk_init_queue(osdblk_rq_fn, &osdev->lock);
435 /* switch queue to TCQ mode; allocate tag map */
436 rc = blk_queue_init_tags(q, OSDBLK_MAX_REQ, NULL);
438 blk_cleanup_queue(q);
443 /* Set our limits to the lower device limits, because osdblk cannot
444 * sleep when allocating a lower-request and therefore cannot be
447 blk_queue_stack_limits(q, osd_request_queue(osdev->osd));
449 blk_queue_prep_rq(q, blk_queue_start_tag);
450 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, osdblk_prepare_flush);
454 q->queuedata = osdev;
459 /* finally, announce the disk to the world */
460 set_capacity(disk, obj_size / 512ULL);
463 printk(KERN_INFO "%s: Added of size 0x%llx\n",
464 disk->disk_name, (unsigned long long)obj_size);
469 /********************************************************************
471 * add map OSD object to blkdev
472 * remove unmap OSD object
474 *******************************************************************/
476 static void class_osdblk_release(struct class *cls)
481 static ssize_t class_osdblk_list(struct class *c,
482 struct class_attribute *attr,
486 struct list_head *tmp;
488 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
490 list_for_each(tmp, &osdblkdev_list) {
491 struct osdblk_device *osdev;
493 osdev = list_entry(tmp, struct osdblk_device, node);
495 n += sprintf(data+n, "%d %d %llu %llu %s\n",
498 osdev->obj.partition,
503 mutex_unlock(&ctl_mutex);
507 static ssize_t class_osdblk_add(struct class *c,
508 struct class_attribute *attr,
509 const char *buf, size_t count)
511 struct osdblk_device *osdev;
514 struct list_head *tmp;
516 if (!try_module_get(THIS_MODULE))
519 /* new osdblk_device object */
520 osdev = kzalloc(sizeof(*osdev) + strlen(buf) + 1, GFP_KERNEL);
526 /* static osdblk_device initialization */
527 spin_lock_init(&osdev->lock);
528 INIT_LIST_HEAD(&osdev->node);
530 /* generate unique id: find highest unique id, add one */
532 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
534 list_for_each(tmp, &osdblkdev_list) {
535 struct osdblk_device *osdev;
537 osdev = list_entry(tmp, struct osdblk_device, node);
538 if (osdev->id > new_id)
539 new_id = osdev->id + 1;
544 /* add to global list */
545 list_add_tail(&osdev->node, &osdblkdev_list);
547 mutex_unlock(&ctl_mutex);
549 /* parse add command */
550 if (sscanf(buf, "%llu %llu %s", &osdev->obj.partition, &osdev->obj.id,
551 osdev->osd_path) != 3) {
556 /* initialize rest of new object */
557 sprintf(osdev->name, DRV_NAME "%d", osdev->id);
559 /* contact requested OSD */
560 osdev->osd = osduld_path_lookup(osdev->osd_path);
561 if (IS_ERR(osdev->osd)) {
562 rc = PTR_ERR(osdev->osd);
566 /* build OSD credential */
567 osdblk_make_credential(osdev->obj_cred, &osdev->obj);
569 /* register our block device */
570 irc = register_blkdev(0, osdev->name);
578 /* set up and announce blkdev mapping */
579 rc = osdblk_init_disk(osdev);
586 unregister_blkdev(osdev->major, osdev->name);
588 osduld_put_device(osdev->osd);
590 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
591 list_del_init(&osdev->node);
592 mutex_unlock(&ctl_mutex);
596 OSDBLK_DEBUG("Error adding device %s\n", buf);
597 module_put(THIS_MODULE);
601 static ssize_t class_osdblk_remove(struct class *c,
602 struct class_attribute *attr,
606 struct osdblk_device *osdev = NULL;
609 struct list_head *tmp;
611 rc = strict_strtoul(buf, 10, &ul);
615 /* convert to int; abort if we lost anything in the conversion */
616 target_id = (int) ul;
620 /* remove object from list immediately */
621 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
623 list_for_each(tmp, &osdblkdev_list) {
624 osdev = list_entry(tmp, struct osdblk_device, node);
625 if (osdev->id == target_id) {
626 list_del_init(&osdev->node);
632 mutex_unlock(&ctl_mutex);
637 /* clean up and free blkdev and associated OSD connection */
638 osdblk_free_disk(osdev);
639 unregister_blkdev(osdev->major, osdev->name);
640 osduld_put_device(osdev->osd);
643 /* release module ref */
644 module_put(THIS_MODULE);
649 static struct class_attribute class_osdblk_attrs[] = {
650 __ATTR(add, 0200, NULL, class_osdblk_add),
651 __ATTR(remove, 0200, NULL, class_osdblk_remove),
652 __ATTR(list, 0444, class_osdblk_list, NULL),
656 static int osdblk_sysfs_init(void)
661 * create control files in sysfs
662 * /sys/class/osdblk/...
664 class_osdblk = kzalloc(sizeof(*class_osdblk), GFP_KERNEL);
668 class_osdblk->name = DRV_NAME;
669 class_osdblk->owner = THIS_MODULE;
670 class_osdblk->class_release = class_osdblk_release;
671 class_osdblk->class_attrs = class_osdblk_attrs;
673 ret = class_register(class_osdblk);
677 printk(PFX "failed to create class osdblk\n");
684 static void osdblk_sysfs_cleanup(void)
687 class_destroy(class_osdblk);
691 static int __init osdblk_init(void)
695 rc = osdblk_sysfs_init();
702 static void __exit osdblk_exit(void)
704 osdblk_sysfs_cleanup();
707 module_init(osdblk_init);
708 module_exit(osdblk_exit);