blob: a808b1530b3bd344c290e0fed84e5d35e72d8338 [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>
66#include <scsi/osd_initiator.h>
67#include <scsi/osd_attributes.h>
68#include <scsi/osd_sec.h>
Boaz Harroshbc47df02009-05-20 18:50:34 +030069#include <scsi/scsi_device.h>
Jeff Garzik2a138772009-04-10 07:50:45 -040070
71#define DRV_NAME "osdblk"
72#define PFX DRV_NAME ": "
73
74/* #define _OSDBLK_DEBUG */
75#ifdef _OSDBLK_DEBUG
76#define OSDBLK_DEBUG(fmt, a...) \
77 printk(KERN_NOTICE "osdblk @%s:%d: " fmt, __func__, __LINE__, ##a)
78#else
79#define OSDBLK_DEBUG(fmt, a...) \
80 do { if (0) printk(fmt, ##a); } while (0)
81#endif
82
83MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
84MODULE_DESCRIPTION("block device inside an OSD object osdblk.ko");
85MODULE_LICENSE("GPL");
86
87struct osdblk_device;
88
89enum {
90 OSDBLK_MINORS_PER_MAJOR = 256, /* max minors per blkdev */
91 OSDBLK_MAX_REQ = 32, /* max parallel requests */
92 OSDBLK_OP_TIMEOUT = 4 * 60, /* sync OSD req timeout */
93};
94
95struct osdblk_request {
96 struct request *rq; /* blk layer request */
97 struct bio *bio; /* cloned bio */
98 struct osdblk_device *osdev; /* associated blkdev */
99};
100
101struct osdblk_device {
102 int id; /* blkdev unique id */
103
104 int major; /* blkdev assigned major */
105 struct gendisk *disk; /* blkdev's gendisk and rq */
106 struct request_queue *q;
107
108 struct osd_dev *osd; /* associated OSD */
109
110 char name[32]; /* blkdev name, e.g. osdblk34 */
111
112 spinlock_t lock; /* queue lock */
113
114 struct osd_obj_id obj; /* OSD partition, obj id */
115 uint8_t obj_cred[OSD_CAP_LEN]; /* OSD cred */
116
117 struct osdblk_request req[OSDBLK_MAX_REQ]; /* request table */
118
119 struct list_head node;
120
121 char osd_path[0]; /* OSD device path */
122};
123
124static struct class *class_osdblk; /* /sys/class/osdblk */
125static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */
126static LIST_HEAD(osdblkdev_list);
127
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700128static const struct block_device_operations osdblk_bd_ops = {
Jeff Garzik2a138772009-04-10 07:50:45 -0400129 .owner = THIS_MODULE,
130};
131
132static const struct osd_attr g_attr_logical_length = ATTR_DEF(
133 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
134
135static void osdblk_make_credential(u8 cred_a[OSD_CAP_LEN],
136 const struct osd_obj_id *obj)
137{
138 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
139}
140
141/* copied from exofs; move to libosd? */
142/*
143 * Perform a synchronous OSD operation. copied from exofs; move to libosd?
144 */
145static int osd_sync_op(struct osd_request *or, int timeout, uint8_t *credential)
146{
147 int ret;
148
149 or->timeout = timeout;
150 ret = osd_finalize_request(or, 0, credential, NULL);
151 if (ret)
152 return ret;
153
154 ret = osd_execute_request(or);
155
156 /* osd_req_decode_sense(or, ret); */
157 return ret;
158}
159
160/*
161 * Perform an asynchronous OSD operation. copied from exofs; move to libosd?
162 */
163static int osd_async_op(struct osd_request *or, osd_req_done_fn *async_done,
164 void *caller_context, u8 *cred)
165{
166 int ret;
167
168 ret = osd_finalize_request(or, 0, cred, NULL);
169 if (ret)
170 return ret;
171
172 ret = osd_execute_request_async(or, async_done, caller_context);
173
174 return ret;
175}
176
177/* copied from exofs; move to libosd? */
178static int extract_attr_from_req(struct osd_request *or, struct osd_attr *attr)
179{
180 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
181 void *iter = NULL;
182 int nelem;
183
184 do {
185 nelem = 1;
186 osd_req_decode_get_attr_list(or, &cur_attr, &nelem, &iter);
187 if ((cur_attr.attr_page == attr->attr_page) &&
188 (cur_attr.attr_id == attr->attr_id)) {
189 attr->len = cur_attr.len;
190 attr->val_ptr = cur_attr.val_ptr;
191 return 0;
192 }
193 } while (iter);
194
195 return -EIO;
196}
197
198static int osdblk_get_obj_size(struct osdblk_device *osdev, u64 *size_out)
199{
200 struct osd_request *or;
201 struct osd_attr attr;
202 int ret;
203
204 /* start request */
205 or = osd_start_request(osdev->osd, GFP_KERNEL);
206 if (!or)
207 return -ENOMEM;
208
209 /* create a get-attributes(length) request */
210 osd_req_get_attributes(or, &osdev->obj);
211
212 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
213
214 /* execute op synchronously */
215 ret = osd_sync_op(or, OSDBLK_OP_TIMEOUT, osdev->obj_cred);
216 if (ret)
217 goto out;
218
219 /* extract length from returned attribute info */
220 attr = g_attr_logical_length;
221 ret = extract_attr_from_req(or, &attr);
222 if (ret)
223 goto out;
224
225 *size_out = get_unaligned_be64(attr.val_ptr);
226
227out:
228 osd_end_request(or);
229 return ret;
230
231}
232
233static void osdblk_osd_complete(struct osd_request *or, void *private)
234{
235 struct osdblk_request *orq = private;
236 struct osd_sense_info osi;
237 int ret = osd_req_decode_sense(or, &osi);
238
239 if (ret) {
240 ret = -EIO;
241 OSDBLK_DEBUG("osdblk_osd_complete with err=%d\n", ret);
242 }
243
244 /* complete OSD request */
245 osd_end_request(or);
246
247 /* complete request passed to osdblk by block layer */
248 __blk_end_request_all(orq->rq, ret);
249}
250
251static void bio_chain_put(struct bio *chain)
252{
253 struct bio *tmp;
254
255 while (chain) {
256 tmp = chain;
257 chain = chain->bi_next;
258
259 bio_put(tmp);
260 }
261}
262
263static struct bio *bio_chain_clone(struct bio *old_chain, gfp_t gfpmask)
264{
265 struct bio *tmp, *new_chain = NULL, *tail = NULL;
266
267 while (old_chain) {
268 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
269 if (!tmp)
270 goto err_out;
271
272 __bio_clone(tmp, old_chain);
273 tmp->bi_bdev = NULL;
274 gfpmask &= ~__GFP_WAIT;
275 tmp->bi_next = NULL;
276
277 if (!new_chain)
278 new_chain = tail = tmp;
279 else {
280 tail->bi_next = tmp;
281 tail = tmp;
282 }
283
284 old_chain = old_chain->bi_next;
285 }
286
287 return new_chain;
288
289err_out:
290 OSDBLK_DEBUG("bio_chain_clone with err\n");
291 bio_chain_put(new_chain);
292 return NULL;
293}
294
295static void osdblk_rq_fn(struct request_queue *q)
296{
297 struct osdblk_device *osdev = q->queuedata;
298
299 while (1) {
300 struct request *rq;
301 struct osdblk_request *orq;
302 struct osd_request *or;
303 struct bio *bio;
304 bool do_write, do_flush;
305
306 /* peek at request from block layer */
307 rq = blk_fetch_request(q);
308 if (!rq)
309 break;
310
311 /* filter out block requests we don't understand */
312 if (!blk_fs_request(rq) && !blk_barrier_rq(rq)) {
313 blk_end_request_all(rq, 0);
314 continue;
315 }
316
317 /* deduce our operation (read, write, flush) */
318 /* I wish the block layer simplified cmd_type/cmd_flags/cmd[]
319 * into a clearly defined set of RPC commands:
320 * read, write, flush, scsi command, power mgmt req,
321 * driver-specific, etc.
322 */
323
324 do_flush = (rq->special == (void *) 0xdeadbeefUL);
325 do_write = (rq_data_dir(rq) == WRITE);
326
327 if (!do_flush) { /* osd_flush does not use a bio */
328 /* a bio clone to be passed down to OSD request */
329 bio = bio_chain_clone(rq->bio, GFP_ATOMIC);
330 if (!bio)
331 break;
332 } else
333 bio = NULL;
334
335 /* alloc internal OSD request, for OSD command execution */
336 or = osd_start_request(osdev->osd, GFP_ATOMIC);
337 if (!or) {
338 bio_chain_put(bio);
339 OSDBLK_DEBUG("osd_start_request with err\n");
340 break;
341 }
342
343 orq = &osdev->req[rq->tag];
344 orq->rq = rq;
345 orq->bio = bio;
346 orq->osdev = osdev;
347
348 /* init OSD command: flush, write or read */
349 if (do_flush)
350 osd_req_flush_object(or, &osdev->obj,
351 OSD_CDB_FLUSH_ALL, 0, 0);
352 else if (do_write)
353 osd_req_write(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
354 bio, blk_rq_bytes(rq));
355 else
356 osd_req_read(or, &osdev->obj, blk_rq_pos(rq) * 512ULL,
357 bio, blk_rq_bytes(rq));
358
359 OSDBLK_DEBUG("%s 0x%x bytes at 0x%llx\n",
360 do_flush ? "flush" : do_write ?
361 "write" : "read", blk_rq_bytes(rq),
362 blk_rq_pos(rq) * 512ULL);
363
364 /* begin OSD command execution */
365 if (osd_async_op(or, osdblk_osd_complete, orq,
366 osdev->obj_cred)) {
367 osd_end_request(or);
368 blk_requeue_request(q, rq);
369 bio_chain_put(bio);
370 OSDBLK_DEBUG("osd_execute_request_async with err\n");
371 break;
372 }
373
374 /* remove the special 'flush' marker, now that the command
375 * is executing
376 */
377 rq->special = NULL;
378 }
379}
380
381static void osdblk_prepare_flush(struct request_queue *q, struct request *rq)
382{
383 /* add driver-specific marker, to indicate that this request
384 * is a flush command
385 */
386 rq->special = (void *) 0xdeadbeefUL;
387}
388
389static void osdblk_free_disk(struct osdblk_device *osdev)
390{
391 struct gendisk *disk = osdev->disk;
392
393 if (!disk)
394 return;
395
396 if (disk->flags & GENHD_FL_UP)
397 del_gendisk(disk);
398 if (disk->queue)
399 blk_cleanup_queue(disk->queue);
400 put_disk(disk);
401}
402
403static int osdblk_init_disk(struct osdblk_device *osdev)
404{
405 struct gendisk *disk;
406 struct request_queue *q;
407 int rc;
408 u64 obj_size = 0;
409
410 /* contact OSD, request size info about the object being mapped */
411 rc = osdblk_get_obj_size(osdev, &obj_size);
412 if (rc)
413 return rc;
414
415 /* create gendisk info */
416 disk = alloc_disk(OSDBLK_MINORS_PER_MAJOR);
417 if (!disk)
418 return -ENOMEM;
419
420 sprintf(disk->disk_name, DRV_NAME "%d", osdev->id);
421 disk->major = osdev->major;
422 disk->first_minor = 0;
423 disk->fops = &osdblk_bd_ops;
424 disk->private_data = osdev;
425
426 /* init rq */
427 q = blk_init_queue(osdblk_rq_fn, &osdev->lock);
428 if (!q) {
429 put_disk(disk);
430 return -ENOMEM;
431 }
432
433 /* switch queue to TCQ mode; allocate tag map */
434 rc = blk_queue_init_tags(q, OSDBLK_MAX_REQ, NULL);
435 if (rc) {
436 blk_cleanup_queue(q);
437 put_disk(disk);
438 return rc;
439 }
440
Boaz Harroshbc47df02009-05-20 18:50:34 +0300441 /* Set our limits to the lower device limits, because osdblk cannot
442 * sleep when allocating a lower-request and therefore cannot be
443 * bouncing.
444 */
445 blk_queue_stack_limits(q, osd_request_queue(osdev->osd));
446
Jeff Garzik2a138772009-04-10 07:50:45 -0400447 blk_queue_prep_rq(q, blk_queue_start_tag);
448 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, osdblk_prepare_flush);
449
450 disk->queue = q;
451
452 q->queuedata = osdev;
453
454 osdev->disk = disk;
455 osdev->q = q;
456
457 /* finally, announce the disk to the world */
458 set_capacity(disk, obj_size / 512ULL);
459 add_disk(disk);
460
461 printk(KERN_INFO "%s: Added of size 0x%llx\n",
462 disk->disk_name, (unsigned long long)obj_size);
463
464 return 0;
465}
466
467/********************************************************************
468 * /sys/class/osdblk/
469 * add map OSD object to blkdev
470 * remove unmap OSD object
471 * list show mappings
472 *******************************************************************/
473
474static void class_osdblk_release(struct class *cls)
475{
476 kfree(cls);
477}
478
479static ssize_t class_osdblk_list(struct class *c, char *data)
480{
481 int n = 0;
482 struct list_head *tmp;
483
484 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
485
486 list_for_each(tmp, &osdblkdev_list) {
487 struct osdblk_device *osdev;
488
489 osdev = list_entry(tmp, struct osdblk_device, node);
490
491 n += sprintf(data+n, "%d %d %llu %llu %s\n",
492 osdev->id,
493 osdev->major,
494 osdev->obj.partition,
495 osdev->obj.id,
496 osdev->osd_path);
497 }
498
499 mutex_unlock(&ctl_mutex);
500 return n;
501}
502
503static ssize_t class_osdblk_add(struct class *c, const char *buf, size_t count)
504{
505 struct osdblk_device *osdev;
506 ssize_t rc;
507 int irc, new_id = 0;
508 struct list_head *tmp;
509
510 if (!try_module_get(THIS_MODULE))
511 return -ENODEV;
512
513 /* new osdblk_device object */
514 osdev = kzalloc(sizeof(*osdev) + strlen(buf) + 1, GFP_KERNEL);
515 if (!osdev) {
516 rc = -ENOMEM;
517 goto err_out_mod;
518 }
519
520 /* static osdblk_device initialization */
521 spin_lock_init(&osdev->lock);
522 INIT_LIST_HEAD(&osdev->node);
523
524 /* generate unique id: find highest unique id, add one */
525
526 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
527
528 list_for_each(tmp, &osdblkdev_list) {
529 struct osdblk_device *osdev;
530
531 osdev = list_entry(tmp, struct osdblk_device, node);
532 if (osdev->id > new_id)
533 new_id = osdev->id + 1;
534 }
535
536 osdev->id = new_id;
537
538 /* add to global list */
539 list_add_tail(&osdev->node, &osdblkdev_list);
540
541 mutex_unlock(&ctl_mutex);
542
543 /* parse add command */
544 if (sscanf(buf, "%llu %llu %s", &osdev->obj.partition, &osdev->obj.id,
545 osdev->osd_path) != 3) {
546 rc = -EINVAL;
547 goto err_out_slot;
548 }
549
550 /* initialize rest of new object */
551 sprintf(osdev->name, DRV_NAME "%d", osdev->id);
552
553 /* contact requested OSD */
554 osdev->osd = osduld_path_lookup(osdev->osd_path);
555 if (IS_ERR(osdev->osd)) {
556 rc = PTR_ERR(osdev->osd);
557 goto err_out_slot;
558 }
559
560 /* build OSD credential */
561 osdblk_make_credential(osdev->obj_cred, &osdev->obj);
562
563 /* register our block device */
564 irc = register_blkdev(0, osdev->name);
565 if (irc < 0) {
566 rc = irc;
567 goto err_out_osd;
568 }
569
570 osdev->major = irc;
571
572 /* set up and announce blkdev mapping */
573 rc = osdblk_init_disk(osdev);
574 if (rc)
575 goto err_out_blkdev;
576
577 return count;
578
579err_out_blkdev:
580 unregister_blkdev(osdev->major, osdev->name);
581err_out_osd:
582 osduld_put_device(osdev->osd);
583err_out_slot:
584 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
585 list_del_init(&osdev->node);
586 mutex_unlock(&ctl_mutex);
587
588 kfree(osdev);
589err_out_mod:
590 OSDBLK_DEBUG("Error adding device %s\n", buf);
591 module_put(THIS_MODULE);
592 return rc;
593}
594
595static ssize_t class_osdblk_remove(struct class *c, const char *buf,
596 size_t count)
597{
598 struct osdblk_device *osdev = NULL;
599 int target_id, rc;
600 unsigned long ul;
601 struct list_head *tmp;
602
603 rc = strict_strtoul(buf, 10, &ul);
604 if (rc)
605 return rc;
606
607 /* convert to int; abort if we lost anything in the conversion */
608 target_id = (int) ul;
609 if (target_id != ul)
610 return -EINVAL;
611
612 /* remove object from list immediately */
613 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
614
615 list_for_each(tmp, &osdblkdev_list) {
616 osdev = list_entry(tmp, struct osdblk_device, node);
617 if (osdev->id == target_id) {
618 list_del_init(&osdev->node);
619 break;
620 }
621 osdev = NULL;
622 }
623
624 mutex_unlock(&ctl_mutex);
625
626 if (!osdev)
627 return -ENOENT;
628
629 /* clean up and free blkdev and associated OSD connection */
630 osdblk_free_disk(osdev);
631 unregister_blkdev(osdev->major, osdev->name);
632 osduld_put_device(osdev->osd);
633 kfree(osdev);
634
635 /* release module ref */
636 module_put(THIS_MODULE);
637
638 return count;
639}
640
641static struct class_attribute class_osdblk_attrs[] = {
642 __ATTR(add, 0200, NULL, class_osdblk_add),
643 __ATTR(remove, 0200, NULL, class_osdblk_remove),
644 __ATTR(list, 0444, class_osdblk_list, NULL),
645 __ATTR_NULL
646};
647
648static int osdblk_sysfs_init(void)
649{
650 int ret = 0;
651
652 /*
653 * create control files in sysfs
654 * /sys/class/osdblk/...
655 */
656 class_osdblk = kzalloc(sizeof(*class_osdblk), GFP_KERNEL);
657 if (!class_osdblk)
658 return -ENOMEM;
659
660 class_osdblk->name = DRV_NAME;
661 class_osdblk->owner = THIS_MODULE;
662 class_osdblk->class_release = class_osdblk_release;
663 class_osdblk->class_attrs = class_osdblk_attrs;
664
665 ret = class_register(class_osdblk);
666 if (ret) {
667 kfree(class_osdblk);
668 class_osdblk = NULL;
669 printk(PFX "failed to create class osdblk\n");
670 return ret;
671 }
672
673 return 0;
674}
675
676static void osdblk_sysfs_cleanup(void)
677{
678 if (class_osdblk)
679 class_destroy(class_osdblk);
680 class_osdblk = NULL;
681}
682
683static int __init osdblk_init(void)
684{
685 int rc;
686
687 rc = osdblk_sysfs_init();
688 if (rc)
689 return rc;
690
691 return 0;
692}
693
694static void __exit osdblk_exit(void)
695{
696 osdblk_sysfs_cleanup();
697}
698
699module_init(osdblk_init);
700module_exit(osdblk_exit);
701