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 <scsi/osd_initiator.h>
67 #include <scsi/osd_attributes.h>
68 #include <scsi/osd_sec.h>
70 #define DRV_NAME "osdblk"
71 #define PFX DRV_NAME ": "
73 /* #define _OSDBLK_DEBUG */
75 #define OSDBLK_DEBUG(fmt, a...) \
76 printk(KERN_NOTICE "osdblk @%s:%d: " fmt, __func__, __LINE__, ##a)
78 #define OSDBLK_DEBUG(fmt, a...) \
79 do { if (0) printk(fmt, ##a); } while (0)
82 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
83 MODULE_DESCRIPTION("block device inside an OSD object osdblk.ko");
84 MODULE_LICENSE("GPL");
89 OSDBLK_MINORS_PER_MAJOR = 256, /* max minors per blkdev */
90 OSDBLK_MAX_REQ = 32, /* max parallel requests */
91 OSDBLK_OP_TIMEOUT = 4 * 60, /* sync OSD req timeout */
94 struct osdblk_request {
95 struct request *rq; /* blk layer request */
96 struct bio *bio; /* cloned bio */
97 struct osdblk_device *osdev; /* associated blkdev */
100 struct osdblk_device {
101 int id; /* blkdev unique id */
103 int major; /* blkdev assigned major */
104 struct gendisk *disk; /* blkdev's gendisk and rq */
105 struct request_queue *q;
107 struct osd_dev *osd; /* associated OSD */
109 char name[32]; /* blkdev name, e.g. osdblk34 */
111 spinlock_t lock; /* queue lock */
113 struct osd_obj_id obj; /* OSD partition, obj id */
114 uint8_t obj_cred[OSD_CAP_LEN]; /* OSD cred */
116 struct osdblk_request req[OSDBLK_MAX_REQ]; /* request table */
118 struct list_head node;
120 char osd_path[0]; /* OSD device path */
123 static struct class *class_osdblk; /* /sys/class/osdblk */
124 static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
125 static LIST_HEAD(osdblkdev_list);
127 static struct block_device_operations osdblk_bd_ops = {
128 .owner = THIS_MODULE,
131 static const struct osd_attr g_attr_logical_length = ATTR_DEF(
132 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
134 static void osdblk_make_credential(u8 cred_a[OSD_CAP_LEN],
135 const struct osd_obj_id *obj)
137 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
140 /* copied from exofs; move to libosd? */
142 * Perform a synchronous OSD operation. copied from exofs; move to libosd?
144 static int osd_sync_op(struct osd_request *or, int timeout, uint8_t *credential)
148 or->timeout = timeout;
149 ret = osd_finalize_request(or, 0, credential, NULL);
153 ret = osd_execute_request(or);
155 /* osd_req_decode_sense(or, ret); */
160 * Perform an asynchronous OSD operation. copied from exofs; move to libosd?
162 static int osd_async_op(struct osd_request *or, osd_req_done_fn *async_done,
163 void *caller_context, u8 *cred)
167 ret = osd_finalize_request(or, 0, cred, NULL);
171 ret = osd_execute_request_async(or, async_done, caller_context);
176 /* copied from exofs; move to libosd? */
177 static int extract_attr_from_req(struct osd_request *or, struct osd_attr *attr)
179 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
185 osd_req_decode_get_attr_list(or, &cur_attr, &nelem, &iter);
186 if ((cur_attr.attr_page == attr->attr_page) &&
187 (cur_attr.attr_id == attr->attr_id)) {
188 attr->len = cur_attr.len;
189 attr->val_ptr = cur_attr.val_ptr;
197 static int osdblk_get_obj_size(struct osdblk_device *osdev, u64 *size_out)
199 struct osd_request *or;
200 struct osd_attr attr;
204 or = osd_start_request(osdev->osd, GFP_KERNEL);
208 /* create a get-attributes(length) request */
209 osd_req_get_attributes(or, &osdev->obj);
211 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
213 /* execute op synchronously */
214 ret = osd_sync_op(or, OSDBLK_OP_TIMEOUT, osdev->obj_cred);
218 /* extract length from returned attribute info */
219 attr = g_attr_logical_length;
220 ret = extract_attr_from_req(or, &attr);
224 *size_out = get_unaligned_be64(attr.val_ptr);
232 static void osdblk_osd_complete(struct osd_request *or, void *private)
234 struct osdblk_request *orq = private;
235 struct osd_sense_info osi;
236 int ret = osd_req_decode_sense(or, &osi);
240 OSDBLK_DEBUG("osdblk_osd_complete with err=%d\n", ret);
243 /* complete OSD request */
246 /* complete request passed to osdblk by block layer */
247 __blk_end_request_all(orq->rq, ret);
250 static void bio_chain_put(struct bio *chain)
256 chain = chain->bi_next;
262 static struct bio *bio_chain_clone(struct bio *old_chain, gfp_t gfpmask)
264 struct bio *tmp, *new_chain = NULL, *tail = NULL;
267 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
271 __bio_clone(tmp, old_chain);
273 gfpmask &= ~__GFP_WAIT;
277 new_chain = tail = tmp;
283 old_chain = old_chain->bi_next;
289 OSDBLK_DEBUG("bio_chain_clone with err\n");
290 bio_chain_put(new_chain);
294 static void osdblk_rq_fn(struct request_queue *q)
296 struct osdblk_device *osdev = q->queuedata;
300 struct osdblk_request *orq;
301 struct osd_request *or;
303 bool do_write, do_flush;
305 /* peek at request from block layer */
306 rq = blk_fetch_request(q);
310 /* filter out block requests we don't understand */
311 if (!blk_fs_request(rq) && !blk_barrier_rq(rq)) {
312 blk_end_request_all(rq, 0);
316 /* deduce our operation (read, write, flush) */
317 /* I wish the block layer simplified cmd_type/cmd_flags/cmd[]
318 * into a clearly defined set of RPC commands:
319 * read, write, flush, scsi command, power mgmt req,
320 * driver-specific, etc.
323 do_flush = (rq->special == (void *) 0xdeadbeefUL);
324 do_write = (rq_data_dir(rq) == WRITE);
326 if (!do_flush) { /* osd_flush does not use a bio */
327 /* a bio clone to be passed down to OSD request */
328 bio = bio_chain_clone(rq->bio, GFP_ATOMIC);
334 /* alloc internal OSD request, for OSD command execution */
335 or = osd_start_request(osdev->osd, GFP_ATOMIC);
338 OSDBLK_DEBUG("osd_start_request with err\n");
342 orq = &osdev->req[rq->tag];
347 /* init OSD command: flush, write or read */
349 osd_req_flush_object(or, &osdev->obj,
350 OSD_CDB_FLUSH_ALL, 0, 0);
352 osd_req_write(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
353 bio, blk_rq_bytes(rq));
355 osd_req_read(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
356 bio, blk_rq_bytes(rq));
358 OSDBLK_DEBUG("%s 0x%x bytes at 0x%llx\n",
359 do_flush ? "flush" : do_write ?
360 "write" : "read", blk_rq_bytes(rq),
361 blk_rq_pos(rq) * 512ULL);
363 /* begin OSD command execution */
364 if (osd_async_op(or, osdblk_osd_complete, orq,
367 blk_requeue_request(q, rq);
369 OSDBLK_DEBUG("osd_execute_request_async with err\n");
373 /* remove the special 'flush' marker, now that the command
380 static void osdblk_prepare_flush(struct request_queue *q, struct request *rq)
382 /* add driver-specific marker, to indicate that this request
385 rq->special = (void *) 0xdeadbeefUL;
388 static void osdblk_free_disk(struct osdblk_device *osdev)
390 struct gendisk *disk = osdev->disk;
395 if (disk->flags & GENHD_FL_UP)
398 blk_cleanup_queue(disk->queue);
402 static int osdblk_init_disk(struct osdblk_device *osdev)
404 struct gendisk *disk;
405 struct request_queue *q;
409 /* contact OSD, request size info about the object being mapped */
410 rc = osdblk_get_obj_size(osdev, &obj_size);
414 /* create gendisk info */
415 disk = alloc_disk(OSDBLK_MINORS_PER_MAJOR);
419 sprintf(disk->disk_name, DRV_NAME "%d", osdev->id);
420 disk->major = osdev->major;
421 disk->first_minor = 0;
422 disk->fops = &osdblk_bd_ops;
423 disk->private_data = osdev;
426 q = blk_init_queue(osdblk_rq_fn, &osdev->lock);
432 /* switch queue to TCQ mode; allocate tag map */
433 rc = blk_queue_init_tags(q, OSDBLK_MAX_REQ, NULL);
435 blk_cleanup_queue(q);
440 blk_queue_prep_rq(q, blk_queue_start_tag);
441 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, osdblk_prepare_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, char *data)
475 struct list_head *tmp;
477 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
479 list_for_each(tmp, &osdblkdev_list) {
480 struct osdblk_device *osdev;
482 osdev = list_entry(tmp, struct osdblk_device, node);
484 n += sprintf(data+n, "%d %d %llu %llu %s\n",
487 osdev->obj.partition,
492 mutex_unlock(&ctl_mutex);
496 static ssize_t class_osdblk_add(struct class *c, const char *buf, size_t count)
498 struct osdblk_device *osdev;
501 struct list_head *tmp;
503 if (!try_module_get(THIS_MODULE))
506 /* new osdblk_device object */
507 osdev = kzalloc(sizeof(*osdev) + strlen(buf) + 1, GFP_KERNEL);
513 /* static osdblk_device initialization */
514 spin_lock_init(&osdev->lock);
515 INIT_LIST_HEAD(&osdev->node);
517 /* generate unique id: find highest unique id, add one */
519 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
521 list_for_each(tmp, &osdblkdev_list) {
522 struct osdblk_device *osdev;
524 osdev = list_entry(tmp, struct osdblk_device, node);
525 if (osdev->id > new_id)
526 new_id = osdev->id + 1;
531 /* add to global list */
532 list_add_tail(&osdev->node, &osdblkdev_list);
534 mutex_unlock(&ctl_mutex);
536 /* parse add command */
537 if (sscanf(buf, "%llu %llu %s", &osdev->obj.partition, &osdev->obj.id,
538 osdev->osd_path) != 3) {
543 /* initialize rest of new object */
544 sprintf(osdev->name, DRV_NAME "%d", osdev->id);
546 /* contact requested OSD */
547 osdev->osd = osduld_path_lookup(osdev->osd_path);
548 if (IS_ERR(osdev->osd)) {
549 rc = PTR_ERR(osdev->osd);
553 /* build OSD credential */
554 osdblk_make_credential(osdev->obj_cred, &osdev->obj);
556 /* register our block device */
557 irc = register_blkdev(0, osdev->name);
565 /* set up and announce blkdev mapping */
566 rc = osdblk_init_disk(osdev);
573 unregister_blkdev(osdev->major, osdev->name);
575 osduld_put_device(osdev->osd);
577 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
578 list_del_init(&osdev->node);
579 mutex_unlock(&ctl_mutex);
583 OSDBLK_DEBUG("Error adding device %s\n", buf);
584 module_put(THIS_MODULE);
588 static ssize_t class_osdblk_remove(struct class *c, const char *buf,
591 struct osdblk_device *osdev = NULL;
594 struct list_head *tmp;
596 rc = strict_strtoul(buf, 10, &ul);
600 /* convert to int; abort if we lost anything in the conversion */
601 target_id = (int) ul;
605 /* remove object from list immediately */
606 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
608 list_for_each(tmp, &osdblkdev_list) {
609 osdev = list_entry(tmp, struct osdblk_device, node);
610 if (osdev->id == target_id) {
611 list_del_init(&osdev->node);
617 mutex_unlock(&ctl_mutex);
622 /* clean up and free blkdev and associated OSD connection */
623 osdblk_free_disk(osdev);
624 unregister_blkdev(osdev->major, osdev->name);
625 osduld_put_device(osdev->osd);
628 /* release module ref */
629 module_put(THIS_MODULE);
634 static struct class_attribute class_osdblk_attrs[] = {
635 __ATTR(add, 0200, NULL, class_osdblk_add),
636 __ATTR(remove, 0200, NULL, class_osdblk_remove),
637 __ATTR(list, 0444, class_osdblk_list, NULL),
641 static int osdblk_sysfs_init(void)
646 * create control files in sysfs
647 * /sys/class/osdblk/...
649 class_osdblk = kzalloc(sizeof(*class_osdblk), GFP_KERNEL);
653 class_osdblk->name = DRV_NAME;
654 class_osdblk->owner = THIS_MODULE;
655 class_osdblk->class_release = class_osdblk_release;
656 class_osdblk->class_attrs = class_osdblk_attrs;
658 ret = class_register(class_osdblk);
662 printk(PFX "failed to create class osdblk\n");
669 static void osdblk_sysfs_cleanup(void)
672 class_destroy(class_osdblk);
676 static int __init osdblk_init(void)
680 rc = osdblk_sysfs_init();
687 static void __exit osdblk_exit(void)
689 osdblk_sysfs_cleanup();
692 module_init(osdblk_init);
693 module_exit(osdblk_exit);