blob: f2fff5799ddf0fe1b41909b64d606d747f9081f7 [file] [log] [blame]
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001/*
2 * blkfront.c
3 *
4 * XenLinux virtual block device driver.
5 *
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
18 *
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
25 *
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35 * IN THE SOFTWARE.
36 */
37
38#include <linux/interrupt.h>
39#include <linux/blkdev.h>
Ian Campbell597592d2008-02-21 13:03:45 -080040#include <linux/hdreg.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070041#include <linux/module.h>
42
43#include <xen/xenbus.h>
44#include <xen/grant_table.h>
45#include <xen/events.h>
46#include <xen/page.h>
47
48#include <xen/interface/grant_table.h>
49#include <xen/interface/io/blkif.h>
Markus Armbruster3e334232008-04-02 10:54:02 -070050#include <xen/interface/io/protocols.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070051
52#include <asm/xen/hypervisor.h>
53
54enum blkif_state {
55 BLKIF_STATE_DISCONNECTED,
56 BLKIF_STATE_CONNECTED,
57 BLKIF_STATE_SUSPENDED,
58};
59
60struct blk_shadow {
61 struct blkif_request req;
62 unsigned long request;
63 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
64};
65
66static struct block_device_operations xlvbd_block_fops;
67
68#define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
69
70/*
71 * We have one of these per vbd, whether ide, scsi or 'other'. They
72 * hang in private_data off the gendisk structure. We may end up
73 * putting all kinds of interesting stuff here :-)
74 */
75struct blkfront_info
76{
77 struct xenbus_device *xbdev;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070078 struct gendisk *gd;
79 int vdevice;
80 blkif_vdev_t handle;
81 enum blkif_state connected;
82 int ring_ref;
83 struct blkif_front_ring ring;
84 unsigned int evtchn, irq;
85 struct request_queue *rq;
86 struct work_struct work;
87 struct gnttab_free_callback callback;
88 struct blk_shadow shadow[BLK_RING_SIZE];
89 unsigned long shadow_free;
90 int feature_barrier;
Christian Limpach1d78d702008-04-02 10:54:04 -070091 int is_ready;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070092
93 /**
94 * The number of people holding this device open. We won't allow a
95 * hot-unplug unless this is 0.
96 */
97 int users;
98};
99
100static DEFINE_SPINLOCK(blkif_io_lock);
101
102#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
103 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
104#define GRANT_INVALID_REF 0
105
106#define PARTS_PER_DISK 16
107
108#define BLKIF_MAJOR(dev) ((dev)>>8)
109#define BLKIF_MINOR(dev) ((dev) & 0xff)
110
111#define DEV_NAME "xvd" /* name in /dev */
112
113/* Information about our VBDs. */
114#define MAX_VBDS 64
115static LIST_HEAD(vbds_list);
116
117static int get_id_from_freelist(struct blkfront_info *info)
118{
119 unsigned long free = info->shadow_free;
120 BUG_ON(free > BLK_RING_SIZE);
121 info->shadow_free = info->shadow[free].req.id;
122 info->shadow[free].req.id = 0x0fffffee; /* debug */
123 return free;
124}
125
126static void add_id_to_freelist(struct blkfront_info *info,
127 unsigned long id)
128{
129 info->shadow[id].req.id = info->shadow_free;
130 info->shadow[id].request = 0;
131 info->shadow_free = id;
132}
133
134static void blkif_restart_queue_callback(void *arg)
135{
136 struct blkfront_info *info = (struct blkfront_info *)arg;
137 schedule_work(&info->work);
138}
139
Harvey Harrisonafe42d72008-04-29 00:59:47 -0700140static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
Ian Campbell597592d2008-02-21 13:03:45 -0800141{
142 /* We don't have real geometry info, but let's at least return
143 values consistent with the size of the device */
144 sector_t nsect = get_capacity(bd->bd_disk);
145 sector_t cylinders = nsect;
146
147 hg->heads = 0xff;
148 hg->sectors = 0x3f;
149 sector_div(cylinders, hg->heads * hg->sectors);
150 hg->cylinders = cylinders;
151 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
152 hg->cylinders = 0xffff;
153 return 0;
154}
155
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700156/*
157 * blkif_queue_request
158 *
159 * request block io
160 *
161 * id: for guest use only.
162 * operation: BLKIF_OP_{READ,WRITE,PROBE}
163 * buffer: buffer to read/write into. this should be a
164 * virtual address in the guest os.
165 */
166static int blkif_queue_request(struct request *req)
167{
168 struct blkfront_info *info = req->rq_disk->private_data;
169 unsigned long buffer_mfn;
170 struct blkif_request *ring_req;
NeilBrown5705f702007-09-25 12:35:59 +0200171 struct req_iterator iter;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700172 struct bio_vec *bvec;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700173 unsigned long id;
174 unsigned int fsect, lsect;
175 int ref;
176 grant_ref_t gref_head;
177
178 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
179 return 1;
180
181 if (gnttab_alloc_grant_references(
182 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
183 gnttab_request_free_callback(
184 &info->callback,
185 blkif_restart_queue_callback,
186 info,
187 BLKIF_MAX_SEGMENTS_PER_REQUEST);
188 return 1;
189 }
190
191 /* Fill out a communications ring structure. */
192 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
193 id = get_id_from_freelist(info);
194 info->shadow[id].request = (unsigned long)req;
195
196 ring_req->id = id;
197 ring_req->sector_number = (blkif_sector_t)req->sector;
198 ring_req->handle = info->handle;
199
200 ring_req->operation = rq_data_dir(req) ?
201 BLKIF_OP_WRITE : BLKIF_OP_READ;
202 if (blk_barrier_rq(req))
203 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
204
205 ring_req->nr_segments = 0;
NeilBrown5705f702007-09-25 12:35:59 +0200206 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200207 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
208 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
209 fsect = bvec->bv_offset >> 9;
210 lsect = fsect + (bvec->bv_len >> 9) - 1;
211 /* install a grant reference. */
212 ref = gnttab_claim_grant_reference(&gref_head);
213 BUG_ON(ref == -ENOSPC);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700214
Jens Axboe6c92e692007-08-16 13:43:12 +0200215 gnttab_grant_foreign_access_ref(
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700216 ref,
217 info->xbdev->otherend_id,
218 buffer_mfn,
219 rq_data_dir(req) );
220
Jens Axboe6c92e692007-08-16 13:43:12 +0200221 info->shadow[id].frame[ring_req->nr_segments] =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700222 mfn_to_pfn(buffer_mfn);
223
Jens Axboe6c92e692007-08-16 13:43:12 +0200224 ring_req->seg[ring_req->nr_segments] =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700225 (struct blkif_request_segment) {
226 .gref = ref,
227 .first_sect = fsect,
228 .last_sect = lsect };
229
Jens Axboe6c92e692007-08-16 13:43:12 +0200230 ring_req->nr_segments++;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700231 }
232
233 info->ring.req_prod_pvt++;
234
235 /* Keep a private copy so we can reissue requests when recovering. */
236 info->shadow[id].req = *ring_req;
237
238 gnttab_free_grant_references(gref_head);
239
240 return 0;
241}
242
243
244static inline void flush_requests(struct blkfront_info *info)
245{
246 int notify;
247
248 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
249
250 if (notify)
251 notify_remote_via_irq(info->irq);
252}
253
254/*
255 * do_blkif_request
256 * read a block; request is in a request queue
257 */
Jens Axboe165125e2007-07-24 09:28:11 +0200258static void do_blkif_request(struct request_queue *rq)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700259{
260 struct blkfront_info *info = NULL;
261 struct request *req;
262 int queued;
263
264 pr_debug("Entered do_blkif_request\n");
265
266 queued = 0;
267
268 while ((req = elv_next_request(rq)) != NULL) {
269 info = req->rq_disk->private_data;
270 if (!blk_fs_request(req)) {
271 end_request(req, 0);
272 continue;
273 }
274
275 if (RING_FULL(&info->ring))
276 goto wait;
277
278 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
279 "(%u/%li) buffer:%p [%s]\n",
280 req, req->cmd, (unsigned long)req->sector,
281 req->current_nr_sectors,
282 req->nr_sectors, req->buffer,
283 rq_data_dir(req) ? "write" : "read");
284
285
286 blkdev_dequeue_request(req);
287 if (blkif_queue_request(req)) {
288 blk_requeue_request(rq, req);
289wait:
290 /* Avoid pointless unplugs. */
291 blk_stop_queue(rq);
292 break;
293 }
294
295 queued++;
296 }
297
298 if (queued != 0)
299 flush_requests(info);
300}
301
302static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
303{
Jens Axboe165125e2007-07-24 09:28:11 +0200304 struct request_queue *rq;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700305
306 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
307 if (rq == NULL)
308 return -1;
309
310 elevator_init(rq, "noop");
311
312 /* Hard sector size and max sectors impersonate the equiv. hardware. */
313 blk_queue_hardsect_size(rq, sector_size);
314 blk_queue_max_sectors(rq, 512);
315
316 /* Each segment in a request is up to an aligned page in size. */
317 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
318 blk_queue_max_segment_size(rq, PAGE_SIZE);
319
320 /* Ensure a merged request will fit in a single I/O ring slot. */
321 blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
322 blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
323
324 /* Make sure buffer addresses are sector-aligned. */
325 blk_queue_dma_alignment(rq, 511);
326
327 gd->queue = rq;
328
329 return 0;
330}
331
332
333static int xlvbd_barrier(struct blkfront_info *info)
334{
335 int err;
336
337 err = blk_queue_ordered(info->rq,
338 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
339 NULL);
340
341 if (err)
342 return err;
343
344 printk(KERN_INFO "blkfront: %s: barriers %s\n",
345 info->gd->disk_name,
346 info->feature_barrier ? "enabled" : "disabled");
347 return 0;
348}
349
350
351static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
352 int vdevice, u16 vdisk_info, u16 sector_size,
353 struct blkfront_info *info)
354{
355 struct gendisk *gd;
356 int nr_minors = 1;
357 int err = -ENODEV;
358
359 BUG_ON(info->gd != NULL);
360 BUG_ON(info->rq != NULL);
361
362 if ((minor % PARTS_PER_DISK) == 0)
363 nr_minors = PARTS_PER_DISK;
364
365 gd = alloc_disk(nr_minors);
366 if (gd == NULL)
367 goto out;
368
369 if (nr_minors > 1)
370 sprintf(gd->disk_name, "%s%c", DEV_NAME,
371 'a' + minor / PARTS_PER_DISK);
372 else
373 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
374 'a' + minor / PARTS_PER_DISK,
375 minor % PARTS_PER_DISK);
376
377 gd->major = XENVBD_MAJOR;
378 gd->first_minor = minor;
379 gd->fops = &xlvbd_block_fops;
380 gd->private_data = info;
381 gd->driverfs_dev = &(info->xbdev->dev);
382 set_capacity(gd, capacity);
383
384 if (xlvbd_init_blk_queue(gd, sector_size)) {
385 del_gendisk(gd);
386 goto out;
387 }
388
389 info->rq = gd->queue;
390 info->gd = gd;
391
392 if (info->feature_barrier)
393 xlvbd_barrier(info);
394
395 if (vdisk_info & VDISK_READONLY)
396 set_disk_ro(gd, 1);
397
398 if (vdisk_info & VDISK_REMOVABLE)
399 gd->flags |= GENHD_FL_REMOVABLE;
400
401 if (vdisk_info & VDISK_CDROM)
402 gd->flags |= GENHD_FL_CD;
403
404 return 0;
405
406 out:
407 return err;
408}
409
410static void kick_pending_request_queues(struct blkfront_info *info)
411{
412 if (!RING_FULL(&info->ring)) {
413 /* Re-enable calldowns. */
414 blk_start_queue(info->rq);
415 /* Kick things off immediately. */
416 do_blkif_request(info->rq);
417 }
418}
419
420static void blkif_restart_queue(struct work_struct *work)
421{
422 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
423
424 spin_lock_irq(&blkif_io_lock);
425 if (info->connected == BLKIF_STATE_CONNECTED)
426 kick_pending_request_queues(info);
427 spin_unlock_irq(&blkif_io_lock);
428}
429
430static void blkif_free(struct blkfront_info *info, int suspend)
431{
432 /* Prevent new requests being issued until we fix things up. */
433 spin_lock_irq(&blkif_io_lock);
434 info->connected = suspend ?
435 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
436 /* No more blkif_request(). */
437 if (info->rq)
438 blk_stop_queue(info->rq);
439 /* No more gnttab callback work. */
440 gnttab_cancel_free_callback(&info->callback);
441 spin_unlock_irq(&blkif_io_lock);
442
443 /* Flush gnttab callback work. Must be done with no locks held. */
444 flush_scheduled_work();
445
446 /* Free resources associated with old device channel. */
447 if (info->ring_ref != GRANT_INVALID_REF) {
448 gnttab_end_foreign_access(info->ring_ref, 0,
449 (unsigned long)info->ring.sring);
450 info->ring_ref = GRANT_INVALID_REF;
451 info->ring.sring = NULL;
452 }
453 if (info->irq)
454 unbind_from_irqhandler(info->irq, info);
455 info->evtchn = info->irq = 0;
456
457}
458
459static void blkif_completion(struct blk_shadow *s)
460{
461 int i;
462 for (i = 0; i < s->req.nr_segments; i++)
463 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
464}
465
466static irqreturn_t blkif_interrupt(int irq, void *dev_id)
467{
468 struct request *req;
469 struct blkif_response *bret;
470 RING_IDX i, rp;
471 unsigned long flags;
472 struct blkfront_info *info = (struct blkfront_info *)dev_id;
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500473 int error;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700474
475 spin_lock_irqsave(&blkif_io_lock, flags);
476
477 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
478 spin_unlock_irqrestore(&blkif_io_lock, flags);
479 return IRQ_HANDLED;
480 }
481
482 again:
483 rp = info->ring.sring->rsp_prod;
484 rmb(); /* Ensure we see queued responses up to 'rp'. */
485
486 for (i = info->ring.rsp_cons; i != rp; i++) {
487 unsigned long id;
488 int ret;
489
490 bret = RING_GET_RESPONSE(&info->ring, i);
491 id = bret->id;
492 req = (struct request *)info->shadow[id].request;
493
494 blkif_completion(&info->shadow[id]);
495
496 add_id_to_freelist(info, id);
497
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500498 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700499 switch (bret->operation) {
500 case BLKIF_OP_WRITE_BARRIER:
501 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
502 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
503 info->gd->disk_name);
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500504 error = -EOPNOTSUPP;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700505 info->feature_barrier = 0;
506 xlvbd_barrier(info);
507 }
508 /* fall through */
509 case BLKIF_OP_READ:
510 case BLKIF_OP_WRITE:
511 if (unlikely(bret->status != BLKIF_RSP_OKAY))
512 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
513 "request: %x\n", bret->status);
514
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500515 ret = __blk_end_request(req, error, blk_rq_bytes(req));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700516 BUG_ON(ret);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700517 break;
518 default:
519 BUG();
520 }
521 }
522
523 info->ring.rsp_cons = i;
524
525 if (i != info->ring.req_prod_pvt) {
526 int more_to_do;
527 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
528 if (more_to_do)
529 goto again;
530 } else
531 info->ring.sring->rsp_event = i + 1;
532
533 kick_pending_request_queues(info);
534
535 spin_unlock_irqrestore(&blkif_io_lock, flags);
536
537 return IRQ_HANDLED;
538}
539
540
541static int setup_blkring(struct xenbus_device *dev,
542 struct blkfront_info *info)
543{
544 struct blkif_sring *sring;
545 int err;
546
547 info->ring_ref = GRANT_INVALID_REF;
548
549 sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL);
550 if (!sring) {
551 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
552 return -ENOMEM;
553 }
554 SHARED_RING_INIT(sring);
555 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
556
557 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
558 if (err < 0) {
559 free_page((unsigned long)sring);
560 info->ring.sring = NULL;
561 goto fail;
562 }
563 info->ring_ref = err;
564
565 err = xenbus_alloc_evtchn(dev, &info->evtchn);
566 if (err)
567 goto fail;
568
569 err = bind_evtchn_to_irqhandler(info->evtchn,
570 blkif_interrupt,
571 IRQF_SAMPLE_RANDOM, "blkif", info);
572 if (err <= 0) {
573 xenbus_dev_fatal(dev, err,
574 "bind_evtchn_to_irqhandler failed");
575 goto fail;
576 }
577 info->irq = err;
578
579 return 0;
580fail:
581 blkif_free(info, 0);
582 return err;
583}
584
585
586/* Common code used when first setting up, and when resuming. */
587static int talk_to_backend(struct xenbus_device *dev,
588 struct blkfront_info *info)
589{
590 const char *message = NULL;
591 struct xenbus_transaction xbt;
592 int err;
593
594 /* Create shared ring, alloc event channel. */
595 err = setup_blkring(dev, info);
596 if (err)
597 goto out;
598
599again:
600 err = xenbus_transaction_start(&xbt);
601 if (err) {
602 xenbus_dev_fatal(dev, err, "starting transaction");
603 goto destroy_blkring;
604 }
605
606 err = xenbus_printf(xbt, dev->nodename,
607 "ring-ref", "%u", info->ring_ref);
608 if (err) {
609 message = "writing ring-ref";
610 goto abort_transaction;
611 }
612 err = xenbus_printf(xbt, dev->nodename,
613 "event-channel", "%u", info->evtchn);
614 if (err) {
615 message = "writing event-channel";
616 goto abort_transaction;
617 }
Markus Armbruster3e334232008-04-02 10:54:02 -0700618 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
619 XEN_IO_PROTO_ABI_NATIVE);
620 if (err) {
621 message = "writing protocol";
622 goto abort_transaction;
623 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700624
625 err = xenbus_transaction_end(xbt, 0);
626 if (err) {
627 if (err == -EAGAIN)
628 goto again;
629 xenbus_dev_fatal(dev, err, "completing transaction");
630 goto destroy_blkring;
631 }
632
633 xenbus_switch_state(dev, XenbusStateInitialised);
634
635 return 0;
636
637 abort_transaction:
638 xenbus_transaction_end(xbt, 1);
639 if (message)
640 xenbus_dev_fatal(dev, err, "%s", message);
641 destroy_blkring:
642 blkif_free(info, 0);
643 out:
644 return err;
645}
646
647
648/**
649 * Entry point to this code when a new device is created. Allocate the basic
650 * structures and the ring buffer for communication with the backend, and
651 * inform the backend of the appropriate details for those. Switch to
652 * Initialised state.
653 */
654static int blkfront_probe(struct xenbus_device *dev,
655 const struct xenbus_device_id *id)
656{
657 int err, vdevice, i;
658 struct blkfront_info *info;
659
660 /* FIXME: Use dynamic device id if this is not set. */
661 err = xenbus_scanf(XBT_NIL, dev->nodename,
662 "virtual-device", "%i", &vdevice);
663 if (err != 1) {
664 xenbus_dev_fatal(dev, err, "reading virtual-device");
665 return err;
666 }
667
668 info = kzalloc(sizeof(*info), GFP_KERNEL);
669 if (!info) {
670 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
671 return -ENOMEM;
672 }
673
674 info->xbdev = dev;
675 info->vdevice = vdevice;
676 info->connected = BLKIF_STATE_DISCONNECTED;
677 INIT_WORK(&info->work, blkif_restart_queue);
678
679 for (i = 0; i < BLK_RING_SIZE; i++)
680 info->shadow[i].req.id = i+1;
681 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
682
683 /* Front end dir is a number, which is used as the id. */
684 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
685 dev->dev.driver_data = info;
686
687 err = talk_to_backend(dev, info);
688 if (err) {
689 kfree(info);
690 dev->dev.driver_data = NULL;
691 return err;
692 }
693
694 return 0;
695}
696
697
698static int blkif_recover(struct blkfront_info *info)
699{
700 int i;
701 struct blkif_request *req;
702 struct blk_shadow *copy;
703 int j;
704
705 /* Stage 1: Make a safe copy of the shadow state. */
706 copy = kmalloc(sizeof(info->shadow), GFP_KERNEL);
707 if (!copy)
708 return -ENOMEM;
709 memcpy(copy, info->shadow, sizeof(info->shadow));
710
711 /* Stage 2: Set up free list. */
712 memset(&info->shadow, 0, sizeof(info->shadow));
713 for (i = 0; i < BLK_RING_SIZE; i++)
714 info->shadow[i].req.id = i+1;
715 info->shadow_free = info->ring.req_prod_pvt;
716 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
717
718 /* Stage 3: Find pending requests and requeue them. */
719 for (i = 0; i < BLK_RING_SIZE; i++) {
720 /* Not in use? */
721 if (copy[i].request == 0)
722 continue;
723
724 /* Grab a request slot and copy shadow state into it. */
725 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
726 *req = copy[i].req;
727
728 /* We get a new request id, and must reset the shadow state. */
729 req->id = get_id_from_freelist(info);
730 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
731
732 /* Rewrite any grant references invalidated by susp/resume. */
733 for (j = 0; j < req->nr_segments; j++)
734 gnttab_grant_foreign_access_ref(
735 req->seg[j].gref,
736 info->xbdev->otherend_id,
737 pfn_to_mfn(info->shadow[req->id].frame[j]),
738 rq_data_dir(
739 (struct request *)
740 info->shadow[req->id].request));
741 info->shadow[req->id].req = *req;
742
743 info->ring.req_prod_pvt++;
744 }
745
746 kfree(copy);
747
748 xenbus_switch_state(info->xbdev, XenbusStateConnected);
749
750 spin_lock_irq(&blkif_io_lock);
751
752 /* Now safe for us to use the shared ring */
753 info->connected = BLKIF_STATE_CONNECTED;
754
755 /* Send off requeued requests */
756 flush_requests(info);
757
758 /* Kick any other new requests queued since we resumed */
759 kick_pending_request_queues(info);
760
761 spin_unlock_irq(&blkif_io_lock);
762
763 return 0;
764}
765
766/**
767 * We are reconnecting to the backend, due to a suspend/resume, or a backend
768 * driver restart. We tear down our blkif structure and recreate it, but
769 * leave the device-layer structures intact so that this is transparent to the
770 * rest of the kernel.
771 */
772static int blkfront_resume(struct xenbus_device *dev)
773{
774 struct blkfront_info *info = dev->dev.driver_data;
775 int err;
776
777 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
778
779 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
780
781 err = talk_to_backend(dev, info);
782 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
783 err = blkif_recover(info);
784
785 return err;
786}
787
788
789/*
790 * Invoked when the backend is finally 'ready' (and has told produced
791 * the details about the physical device - #sectors, size, etc).
792 */
793static void blkfront_connect(struct blkfront_info *info)
794{
795 unsigned long long sectors;
796 unsigned long sector_size;
797 unsigned int binfo;
798 int err;
799
800 if ((info->connected == BLKIF_STATE_CONNECTED) ||
801 (info->connected == BLKIF_STATE_SUSPENDED) )
802 return;
803
804 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
805 __func__, info->xbdev->otherend);
806
807 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
808 "sectors", "%llu", &sectors,
809 "info", "%u", &binfo,
810 "sector-size", "%lu", &sector_size,
811 NULL);
812 if (err) {
813 xenbus_dev_fatal(info->xbdev, err,
814 "reading backend fields at %s",
815 info->xbdev->otherend);
816 return;
817 }
818
819 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
820 "feature-barrier", "%lu", &info->feature_barrier,
821 NULL);
822 if (err)
823 info->feature_barrier = 0;
824
825 err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
826 sectors, info->vdevice,
827 binfo, sector_size, info);
828 if (err) {
829 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
830 info->xbdev->otherend);
831 return;
832 }
833
834 xenbus_switch_state(info->xbdev, XenbusStateConnected);
835
836 /* Kick pending requests. */
837 spin_lock_irq(&blkif_io_lock);
838 info->connected = BLKIF_STATE_CONNECTED;
839 kick_pending_request_queues(info);
840 spin_unlock_irq(&blkif_io_lock);
841
842 add_disk(info->gd);
Christian Limpach1d78d702008-04-02 10:54:04 -0700843
844 info->is_ready = 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700845}
846
847/**
848 * Handle the change of state of the backend to Closing. We must delete our
849 * device-layer structures now, to ensure that writes are flushed through to
850 * the backend. Once is this done, we can switch to Closed in
851 * acknowledgement.
852 */
853static void blkfront_closing(struct xenbus_device *dev)
854{
855 struct blkfront_info *info = dev->dev.driver_data;
856 unsigned long flags;
857
858 dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
859
860 if (info->rq == NULL)
861 goto out;
862
863 spin_lock_irqsave(&blkif_io_lock, flags);
864
865 del_gendisk(info->gd);
866
867 /* No more blkif_request(). */
868 blk_stop_queue(info->rq);
869
870 /* No more gnttab callback work. */
871 gnttab_cancel_free_callback(&info->callback);
872 spin_unlock_irqrestore(&blkif_io_lock, flags);
873
874 /* Flush gnttab callback work. Must be done with no locks held. */
875 flush_scheduled_work();
876
877 blk_cleanup_queue(info->rq);
878 info->rq = NULL;
879
880 out:
881 xenbus_frontend_closed(dev);
882}
883
884/**
885 * Callback received when the backend's state changes.
886 */
887static void backend_changed(struct xenbus_device *dev,
888 enum xenbus_state backend_state)
889{
890 struct blkfront_info *info = dev->dev.driver_data;
891 struct block_device *bd;
892
893 dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
894
895 switch (backend_state) {
896 case XenbusStateInitialising:
897 case XenbusStateInitWait:
898 case XenbusStateInitialised:
899 case XenbusStateUnknown:
900 case XenbusStateClosed:
901 break;
902
903 case XenbusStateConnected:
904 blkfront_connect(info);
905 break;
906
907 case XenbusStateClosing:
Jeremy Fitzhardinge53f0e8a2008-04-02 10:54:03 -0700908 bd = bdget_disk(info->gd, 0);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700909 if (bd == NULL)
910 xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
911
912 mutex_lock(&bd->bd_mutex);
913 if (info->users > 0)
914 xenbus_dev_error(dev, -EBUSY,
915 "Device in use; refusing to close");
916 else
917 blkfront_closing(dev);
918 mutex_unlock(&bd->bd_mutex);
919 bdput(bd);
920 break;
921 }
922}
923
924static int blkfront_remove(struct xenbus_device *dev)
925{
926 struct blkfront_info *info = dev->dev.driver_data;
927
928 dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
929
930 blkif_free(info, 0);
931
932 kfree(info);
933
934 return 0;
935}
936
Christian Limpach1d78d702008-04-02 10:54:04 -0700937static int blkfront_is_ready(struct xenbus_device *dev)
938{
939 struct blkfront_info *info = dev->dev.driver_data;
940
941 return info->is_ready;
942}
943
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700944static int blkif_open(struct inode *inode, struct file *filep)
945{
946 struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
947 info->users++;
948 return 0;
949}
950
951static int blkif_release(struct inode *inode, struct file *filep)
952{
953 struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
954 info->users--;
955 if (info->users == 0) {
956 /* Check whether we have been instructed to close. We will
957 have ignored this request initially, as the device was
958 still mounted. */
959 struct xenbus_device *dev = info->xbdev;
960 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
961
962 if (state == XenbusStateClosing)
963 blkfront_closing(dev);
964 }
965 return 0;
966}
967
968static struct block_device_operations xlvbd_block_fops =
969{
970 .owner = THIS_MODULE,
971 .open = blkif_open,
972 .release = blkif_release,
Ian Campbell597592d2008-02-21 13:03:45 -0800973 .getgeo = blkif_getgeo,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700974};
975
976
977static struct xenbus_device_id blkfront_ids[] = {
978 { "vbd" },
979 { "" }
980};
981
982static struct xenbus_driver blkfront = {
983 .name = "vbd",
984 .owner = THIS_MODULE,
985 .ids = blkfront_ids,
986 .probe = blkfront_probe,
987 .remove = blkfront_remove,
988 .resume = blkfront_resume,
989 .otherend_changed = backend_changed,
Christian Limpach1d78d702008-04-02 10:54:04 -0700990 .is_ready = blkfront_is_ready,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700991};
992
993static int __init xlblk_init(void)
994{
995 if (!is_running_on_xen())
996 return -ENODEV;
997
998 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
999 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1000 XENVBD_MAJOR, DEV_NAME);
1001 return -ENODEV;
1002 }
1003
1004 return xenbus_register_frontend(&blkfront);
1005}
1006module_init(xlblk_init);
1007
1008
1009static void xlblk_exit(void)
1010{
1011 return xenbus_unregister_driver(&blkfront);
1012}
1013module_exit(xlblk_exit);
1014
1015MODULE_DESCRIPTION("Xen virtual block device frontend");
1016MODULE_LICENSE("GPL");
1017MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
Mark McLoughlind2f0c522008-04-02 10:54:05 -07001018MODULE_ALIAS("xen:vbd");
Mark McLoughlin4f93f09b2008-04-02 10:54:06 -07001019MODULE_ALIAS("xenblk");