blob: 6cd8b705b11be399bca98b3ca44f503c68fbbf98 [file] [log] [blame]
Jeff Garzik2a138772009-04-10 07:50:45 -04001
2/*
3 osdblk.c -- Export a single SCSI OSD object as a Linux block device
4
5
6 Copyright 2009 Red Hat, Inc.
7
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.
11
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.
16
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.
20
21
22 Instructions for use
23 --------------------
24
25 1) Map a Linux block device to an existing OSD object.
26
27 In this example, we will use partition id 1234, object id 5678,
28 OSD device /dev/osd1.
29
30 $ echo "1234 5678 /dev/osd1" > /sys/class/osdblk/add
31
32
33 2) List all active blkdev<->object mappings.
34
35 In this example, we have performed step #1 twice, creating two blkdevs,
36 mapped to two separate OSD objects.
37
38 $ cat /sys/class/osdblk/list
39 0 174 1234 5678 /dev/osd1
40 1 179 1994 897123 /dev/osd0
41
42 The columns, in order, are:
43 - blkdev unique id
44 - blkdev assigned major
45 - OSD object partition id
46 - OSD object id
47 - OSD device
48
49
50 3) Remove an active blkdev<->object mapping.
51
52 In this example, we remove the mapping with blkdev unique id 1.
53
54 $ echo 1 > /sys/class/osdblk/remove
55
56
57 NOTE: The actual creation and deletion of OSD objects is outside the scope
58 of this driver.
59
60 */
61
62#include <linux/kernel.h>
63#include <linux/device.h>
64#include <linux/module.h>
65#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090066#include <linux/slab.h>
Jeff Garzik2a138772009-04-10 07:50:45 -040067#include <scsi/osd_initiator.h>
68#include <scsi/osd_attributes.h>
69#include <scsi/osd_sec.h>
Boaz Harroshbc47df02009-05-20 18:50:34 +030070#include <scsi/scsi_device.h>
Jeff Garzik2a138772009-04-10 07:50:45 -040071
72#define DRV_NAME "osdblk"
73#define PFX DRV_NAME ": "
74
75/* #define _OSDBLK_DEBUG */
76#ifdef _OSDBLK_DEBUG
77#define OSDBLK_DEBUG(fmt, a...) \
78 printk(KERN_NOTICE "osdblk @%s:%d: " fmt, __func__, __LINE__, ##a)
79#else
80#define OSDBLK_DEBUG(fmt, a...) \
81 do { if (0) printk(fmt, ##a); } while (0)
82#endif
83
84MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
85MODULE_DESCRIPTION("block device inside an OSD object osdblk.ko");
86MODULE_LICENSE("GPL");
87
88struct osdblk_device;
89
90enum {
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 */
94};
95
96struct osdblk_request {
97 struct request *rq; /* blk layer request */
98 struct bio *bio; /* cloned bio */
99 struct osdblk_device *osdev; /* associated blkdev */
100};
101
102struct osdblk_device {
103 int id; /* blkdev unique id */
104
105 int major; /* blkdev assigned major */
106 struct gendisk *disk; /* blkdev's gendisk and rq */
107 struct request_queue *q;
108
109 struct osd_dev *osd; /* associated OSD */
110
111 char name[32]; /* blkdev name, e.g. osdblk34 */
112
113 spinlock_t lock; /* queue lock */
114
115 struct osd_obj_id obj; /* OSD partition, obj id */
116 uint8_t obj_cred[OSD_CAP_LEN]; /* OSD cred */
117
118 struct osdblk_request req[OSDBLK_MAX_REQ]; /* request table */
119
120 struct list_head node;
121
122 char osd_path[0]; /* OSD device path */
123};
124
125static struct class *class_osdblk; /* /sys/class/osdblk */
126static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
127static LIST_HEAD(osdblkdev_list);
128
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700129static const struct block_device_operations osdblk_bd_ops = {
Jeff Garzik2a138772009-04-10 07:50:45 -0400130 .owner = THIS_MODULE,
131};
132
133static const struct osd_attr g_attr_logical_length = ATTR_DEF(
134 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
135
136static void osdblk_make_credential(u8 cred_a[OSD_CAP_LEN],
137 const struct osd_obj_id *obj)
138{
139 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
140}
141
142/* copied from exofs; move to libosd? */
143/*
144 * Perform a synchronous OSD operation. copied from exofs; move to libosd?
145 */
146static int osd_sync_op(struct osd_request *or, int timeout, uint8_t *credential)
147{
148 int ret;
149
150 or->timeout = timeout;
151 ret = osd_finalize_request(or, 0, credential, NULL);
152 if (ret)
153 return ret;
154
155 ret = osd_execute_request(or);
156
157 /* osd_req_decode_sense(or, ret); */
158 return ret;
159}
160
161/*
162 * Perform an asynchronous OSD operation. copied from exofs; move to libosd?
163 */
164static int osd_async_op(struct osd_request *or, osd_req_done_fn *async_done,
165 void *caller_context, u8 *cred)
166{
167 int ret;
168
169 ret = osd_finalize_request(or, 0, cred, NULL);
170 if (ret)
171 return ret;
172
173 ret = osd_execute_request_async(or, async_done, caller_context);
174
175 return ret;
176}
177
178/* copied from exofs; move to libosd? */
179static int extract_attr_from_req(struct osd_request *or, struct osd_attr *attr)
180{
181 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
182 void *iter = NULL;
183 int nelem;
184
185 do {
186 nelem = 1;
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;
192 return 0;
193 }
194 } while (iter);
195
196 return -EIO;
197}
198
199static int osdblk_get_obj_size(struct osdblk_device *osdev, u64 *size_out)
200{
201 struct osd_request *or;
202 struct osd_attr attr;
203 int ret;
204
205 /* start request */
206 or = osd_start_request(osdev->osd, GFP_KERNEL);
207 if (!or)
208 return -ENOMEM;
209
210 /* create a get-attributes(length) request */
211 osd_req_get_attributes(or, &osdev->obj);
212
213 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
214
215 /* execute op synchronously */
216 ret = osd_sync_op(or, OSDBLK_OP_TIMEOUT, osdev->obj_cred);
217 if (ret)
218 goto out;
219
220 /* extract length from returned attribute info */
221 attr = g_attr_logical_length;
222 ret = extract_attr_from_req(or, &attr);
223 if (ret)
224 goto out;
225
226 *size_out = get_unaligned_be64(attr.val_ptr);
227
228out:
229 osd_end_request(or);
230 return ret;
231
232}
233
234static void osdblk_osd_complete(struct osd_request *or, void *private)
235{
236 struct osdblk_request *orq = private;
237 struct osd_sense_info osi;
238 int ret = osd_req_decode_sense(or, &osi);
239
240 if (ret) {
241 ret = -EIO;
242 OSDBLK_DEBUG("osdblk_osd_complete with err=%d\n", ret);
243 }
244
245 /* complete OSD request */
246 osd_end_request(or);
247
248 /* complete request passed to osdblk by block layer */
249 __blk_end_request_all(orq->rq, ret);
250}
251
252static void bio_chain_put(struct bio *chain)
253{
254 struct bio *tmp;
255
256 while (chain) {
257 tmp = chain;
258 chain = chain->bi_next;
259
260 bio_put(tmp);
261 }
262}
263
264static struct bio *bio_chain_clone(struct bio *old_chain, gfp_t gfpmask)
265{
266 struct bio *tmp, *new_chain = NULL, *tail = NULL;
267
268 while (old_chain) {
269 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
270 if (!tmp)
271 goto err_out;
272
273 __bio_clone(tmp, old_chain);
274 tmp->bi_bdev = NULL;
275 gfpmask &= ~__GFP_WAIT;
276 tmp->bi_next = NULL;
277
278 if (!new_chain)
279 new_chain = tail = tmp;
280 else {
281 tail->bi_next = tmp;
282 tail = tmp;
283 }
284
285 old_chain = old_chain->bi_next;
286 }
287
288 return new_chain;
289
290err_out:
291 OSDBLK_DEBUG("bio_chain_clone with err\n");
292 bio_chain_put(new_chain);
293 return NULL;
294}
295
296static void osdblk_rq_fn(struct request_queue *q)
297{
298 struct osdblk_device *osdev = q->queuedata;
299
300 while (1) {
301 struct request *rq;
302 struct osdblk_request *orq;
303 struct osd_request *or;
304 struct bio *bio;
305 bool do_write, do_flush;
306
307 /* peek at request from block layer */
308 rq = blk_fetch_request(q);
309 if (!rq)
310 break;
311
312 /* filter out block requests we don't understand */
313 if (!blk_fs_request(rq) && !blk_barrier_rq(rq)) {
314 blk_end_request_all(rq, 0);
315 continue;
316 }
317
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.
323 */
324
325 do_flush = (rq->special == (void *) 0xdeadbeefUL);
326 do_write = (rq_data_dir(rq) == WRITE);
327
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);
331 if (!bio)
332 break;
333 } else
334 bio = NULL;
335
336 /* alloc internal OSD request, for OSD command execution */
337 or = osd_start_request(osdev->osd, GFP_ATOMIC);
338 if (!or) {
339 bio_chain_put(bio);
340 OSDBLK_DEBUG("osd_start_request with err\n");
341 break;
342 }
343
344 orq = &osdev->req[rq->tag];
345 orq->rq = rq;
346 orq->bio = bio;
347 orq->osdev = osdev;
348
349 /* init OSD command: flush, write or read */
350 if (do_flush)
351 osd_req_flush_object(or, &osdev->obj,
352 OSD_CDB_FLUSH_ALL, 0, 0);
353 else if (do_write)
354 osd_req_write(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
355 bio, blk_rq_bytes(rq));
356 else
357 osd_req_read(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
358 bio, blk_rq_bytes(rq));
359
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);
364
365 /* begin OSD command execution */
366 if (osd_async_op(or, osdblk_osd_complete, orq,
367 osdev->obj_cred)) {
368 osd_end_request(or);
369 blk_requeue_request(q, rq);
370 bio_chain_put(bio);
371 OSDBLK_DEBUG("osd_execute_request_async with err\n");
372 break;
373 }
374
375 /* remove the special 'flush' marker, now that the command
376 * is executing
377 */
378 rq->special = NULL;
379 }
380}
381
382static void osdblk_prepare_flush(struct request_queue *q, struct request *rq)
383{
384 /* add driver-specific marker, to indicate that this request
385 * is a flush command
386 */
387 rq->special = (void *) 0xdeadbeefUL;
388}
389
390static void osdblk_free_disk(struct osdblk_device *osdev)
391{
392 struct gendisk *disk = osdev->disk;
393
394 if (!disk)
395 return;
396
397 if (disk->flags & GENHD_FL_UP)
398 del_gendisk(disk);
399 if (disk->queue)
400 blk_cleanup_queue(disk->queue);
401 put_disk(disk);
402}
403
404static int osdblk_init_disk(struct osdblk_device *osdev)
405{
406 struct gendisk *disk;
407 struct request_queue *q;
408 int rc;
409 u64 obj_size = 0;
410
411 /* contact OSD, request size info about the object being mapped */
412 rc = osdblk_get_obj_size(osdev, &obj_size);
413 if (rc)
414 return rc;
415
416 /* create gendisk info */
417 disk = alloc_disk(OSDBLK_MINORS_PER_MAJOR);
418 if (!disk)
419 return -ENOMEM;
420
421 sprintf(disk->disk_name, DRV_NAME "%d", osdev->id);
422 disk->major = osdev->major;
423 disk->first_minor = 0;
424 disk->fops = &osdblk_bd_ops;
425 disk->private_data = osdev;
426
427 /* init rq */
428 q = blk_init_queue(osdblk_rq_fn, &osdev->lock);
429 if (!q) {
430 put_disk(disk);
431 return -ENOMEM;
432 }
433
434 /* switch queue to TCQ mode; allocate tag map */
435 rc = blk_queue_init_tags(q, OSDBLK_MAX_REQ, NULL);
436 if (rc) {
437 blk_cleanup_queue(q);
438 put_disk(disk);
439 return rc;
440 }
441
Boaz Harroshbc47df02009-05-20 18:50:34 +0300442 /* Set our limits to the lower device limits, because osdblk cannot
443 * sleep when allocating a lower-request and therefore cannot be
444 * bouncing.
445 */
446 blk_queue_stack_limits(q, osd_request_queue(osdev->osd));
447
Jeff Garzik2a138772009-04-10 07:50:45 -0400448 blk_queue_prep_rq(q, blk_queue_start_tag);
449 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, osdblk_prepare_flush);
450
451 disk->queue = q;
452
453 q->queuedata = osdev;
454
455 osdev->disk = disk;
456 osdev->q = q;
457
458 /* finally, announce the disk to the world */
459 set_capacity(disk, obj_size / 512ULL);
460 add_disk(disk);
461
462 printk(KERN_INFO "%s: Added of size 0x%llx\n",
463 disk->disk_name, (unsigned long long)obj_size);
464
465 return 0;
466}
467
468/********************************************************************
469 * /sys/class/osdblk/
470 * add map OSD object to blkdev
471 * remove unmap OSD object
472 * list show mappings
473 *******************************************************************/
474
475static void class_osdblk_release(struct class *cls)
476{
477 kfree(cls);
478}
479
Andi Kleen28812fe2010-01-05 12:48:07 +0100480static ssize_t class_osdblk_list(struct class *c,
481 struct class_attribute *attr,
482 char *data)
Jeff Garzik2a138772009-04-10 07:50:45 -0400483{
484 int n = 0;
485 struct list_head *tmp;
486
487 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
488
489 list_for_each(tmp, &osdblkdev_list) {
490 struct osdblk_device *osdev;
491
492 osdev = list_entry(tmp, struct osdblk_device, node);
493
494 n += sprintf(data+n, "%d %d %llu %llu %s\n",
495 osdev->id,
496 osdev->major,
497 osdev->obj.partition,
498 osdev->obj.id,
499 osdev->osd_path);
500 }
501
502 mutex_unlock(&ctl_mutex);
503 return n;
504}
505
Andi Kleen28812fe2010-01-05 12:48:07 +0100506static ssize_t class_osdblk_add(struct class *c,
507 struct class_attribute *attr,
508 const char *buf, size_t count)
Jeff Garzik2a138772009-04-10 07:50:45 -0400509{
510 struct osdblk_device *osdev;
511 ssize_t rc;
512 int irc, new_id = 0;
513 struct list_head *tmp;
514
515 if (!try_module_get(THIS_MODULE))
516 return -ENODEV;
517
518 /* new osdblk_device object */
519 osdev = kzalloc(sizeof(*osdev) + strlen(buf) + 1, GFP_KERNEL);
520 if (!osdev) {
521 rc = -ENOMEM;
522 goto err_out_mod;
523 }
524
525 /* static osdblk_device initialization */
526 spin_lock_init(&osdev->lock);
527 INIT_LIST_HEAD(&osdev->node);
528
529 /* generate unique id: find highest unique id, add one */
530
531 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
532
533 list_for_each(tmp, &osdblkdev_list) {
534 struct osdblk_device *osdev;
535
536 osdev = list_entry(tmp, struct osdblk_device, node);
537 if (osdev->id > new_id)
538 new_id = osdev->id + 1;
539 }
540
541 osdev->id = new_id;
542
543 /* add to global list */
544 list_add_tail(&osdev->node, &osdblkdev_list);
545
546 mutex_unlock(&ctl_mutex);
547
548 /* parse add command */
549 if (sscanf(buf, "%llu %llu %s", &osdev->obj.partition, &osdev->obj.id,
550 osdev->osd_path) != 3) {
551 rc = -EINVAL;
552 goto err_out_slot;
553 }
554
555 /* initialize rest of new object */
556 sprintf(osdev->name, DRV_NAME "%d", osdev->id);
557
558 /* contact requested OSD */
559 osdev->osd = osduld_path_lookup(osdev->osd_path);
560 if (IS_ERR(osdev->osd)) {
561 rc = PTR_ERR(osdev->osd);
562 goto err_out_slot;
563 }
564
565 /* build OSD credential */
566 osdblk_make_credential(osdev->obj_cred, &osdev->obj);
567
568 /* register our block device */
569 irc = register_blkdev(0, osdev->name);
570 if (irc < 0) {
571 rc = irc;
572 goto err_out_osd;
573 }
574
575 osdev->major = irc;
576
577 /* set up and announce blkdev mapping */
578 rc = osdblk_init_disk(osdev);
579 if (rc)
580 goto err_out_blkdev;
581
582 return count;
583
584err_out_blkdev:
585 unregister_blkdev(osdev->major, osdev->name);
586err_out_osd:
587 osduld_put_device(osdev->osd);
588err_out_slot:
589 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
590 list_del_init(&osdev->node);
591 mutex_unlock(&ctl_mutex);
592
593 kfree(osdev);
594err_out_mod:
595 OSDBLK_DEBUG("Error adding device %s\n", buf);
596 module_put(THIS_MODULE);
597 return rc;
598}
599
Andi Kleen28812fe2010-01-05 12:48:07 +0100600static ssize_t class_osdblk_remove(struct class *c,
601 struct class_attribute *attr,
602 const char *buf,
Jeff Garzik2a138772009-04-10 07:50:45 -0400603 size_t count)
604{
605 struct osdblk_device *osdev = NULL;
606 int target_id, rc;
607 unsigned long ul;
608 struct list_head *tmp;
609
610 rc = strict_strtoul(buf, 10, &ul);
611 if (rc)
612 return rc;
613
614 /* convert to int; abort if we lost anything in the conversion */
615 target_id = (int) ul;
616 if (target_id != ul)
617 return -EINVAL;
618
619 /* remove object from list immediately */
620 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
621
622 list_for_each(tmp, &osdblkdev_list) {
623 osdev = list_entry(tmp, struct osdblk_device, node);
624 if (osdev->id == target_id) {
625 list_del_init(&osdev->node);
626 break;
627 }
628 osdev = NULL;
629 }
630
631 mutex_unlock(&ctl_mutex);
632
633 if (!osdev)
634 return -ENOENT;
635
636 /* clean up and free blkdev and associated OSD connection */
637 osdblk_free_disk(osdev);
638 unregister_blkdev(osdev->major, osdev->name);
639 osduld_put_device(osdev->osd);
640 kfree(osdev);
641
642 /* release module ref */
643 module_put(THIS_MODULE);
644
645 return count;
646}
647
648static struct class_attribute class_osdblk_attrs[] = {
649 __ATTR(add, 0200, NULL, class_osdblk_add),
650 __ATTR(remove, 0200, NULL, class_osdblk_remove),
651 __ATTR(list, 0444, class_osdblk_list, NULL),
652 __ATTR_NULL
653};
654
655static int osdblk_sysfs_init(void)
656{
657 int ret = 0;
658
659 /*
660 * create control files in sysfs
661 * /sys/class/osdblk/...
662 */
663 class_osdblk = kzalloc(sizeof(*class_osdblk), GFP_KERNEL);
664 if (!class_osdblk)
665 return -ENOMEM;
666
667 class_osdblk->name = DRV_NAME;
668 class_osdblk->owner = THIS_MODULE;
669 class_osdblk->class_release = class_osdblk_release;
670 class_osdblk->class_attrs = class_osdblk_attrs;
671
672 ret = class_register(class_osdblk);
673 if (ret) {
674 kfree(class_osdblk);
675 class_osdblk = NULL;
676 printk(PFX "failed to create class osdblk\n");
677 return ret;
678 }
679
680 return 0;
681}
682
683static void osdblk_sysfs_cleanup(void)
684{
685 if (class_osdblk)
686 class_destroy(class_osdblk);
687 class_osdblk = NULL;
688}
689
690static int __init osdblk_init(void)
691{
692 int rc;
693
694 rc = osdblk_sysfs_init();
695 if (rc)
696 return rc;
697
698 return 0;
699}
700
701static void __exit osdblk_exit(void)
702{
703 osdblk_sysfs_cleanup();
704}
705
706module_init(osdblk_init);
707module_exit(osdblk_exit);
708