blob: 8afce67c0aa56293f815f55fb949983082ba2a9d [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>
40#include <linux/module.h>
41
42#include <xen/xenbus.h>
43#include <xen/grant_table.h>
44#include <xen/events.h>
45#include <xen/page.h>
46
47#include <xen/interface/grant_table.h>
48#include <xen/interface/io/blkif.h>
49
50#include <asm/xen/hypervisor.h>
51
52enum blkif_state {
53 BLKIF_STATE_DISCONNECTED,
54 BLKIF_STATE_CONNECTED,
55 BLKIF_STATE_SUSPENDED,
56};
57
58struct blk_shadow {
59 struct blkif_request req;
60 unsigned long request;
61 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
62};
63
64static struct block_device_operations xlvbd_block_fops;
65
66#define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
67
68/*
69 * We have one of these per vbd, whether ide, scsi or 'other'. They
70 * hang in private_data off the gendisk structure. We may end up
71 * putting all kinds of interesting stuff here :-)
72 */
73struct blkfront_info
74{
75 struct xenbus_device *xbdev;
76 dev_t dev;
77 struct gendisk *gd;
78 int vdevice;
79 blkif_vdev_t handle;
80 enum blkif_state connected;
81 int ring_ref;
82 struct blkif_front_ring ring;
83 unsigned int evtchn, irq;
84 struct request_queue *rq;
85 struct work_struct work;
86 struct gnttab_free_callback callback;
87 struct blk_shadow shadow[BLK_RING_SIZE];
88 unsigned long shadow_free;
89 int feature_barrier;
90
91 /**
92 * The number of people holding this device open. We won't allow a
93 * hot-unplug unless this is 0.
94 */
95 int users;
96};
97
98static DEFINE_SPINLOCK(blkif_io_lock);
99
100#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
101 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
102#define GRANT_INVALID_REF 0
103
104#define PARTS_PER_DISK 16
105
106#define BLKIF_MAJOR(dev) ((dev)>>8)
107#define BLKIF_MINOR(dev) ((dev) & 0xff)
108
109#define DEV_NAME "xvd" /* name in /dev */
110
111/* Information about our VBDs. */
112#define MAX_VBDS 64
113static LIST_HEAD(vbds_list);
114
115static int get_id_from_freelist(struct blkfront_info *info)
116{
117 unsigned long free = info->shadow_free;
118 BUG_ON(free > BLK_RING_SIZE);
119 info->shadow_free = info->shadow[free].req.id;
120 info->shadow[free].req.id = 0x0fffffee; /* debug */
121 return free;
122}
123
124static void add_id_to_freelist(struct blkfront_info *info,
125 unsigned long id)
126{
127 info->shadow[id].req.id = info->shadow_free;
128 info->shadow[id].request = 0;
129 info->shadow_free = id;
130}
131
132static void blkif_restart_queue_callback(void *arg)
133{
134 struct blkfront_info *info = (struct blkfront_info *)arg;
135 schedule_work(&info->work);
136}
137
138/*
139 * blkif_queue_request
140 *
141 * request block io
142 *
143 * id: for guest use only.
144 * operation: BLKIF_OP_{READ,WRITE,PROBE}
145 * buffer: buffer to read/write into. this should be a
146 * virtual address in the guest os.
147 */
148static int blkif_queue_request(struct request *req)
149{
150 struct blkfront_info *info = req->rq_disk->private_data;
151 unsigned long buffer_mfn;
152 struct blkif_request *ring_req;
NeilBrown5705f702007-09-25 12:35:59 +0200153 struct req_iterator iter;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700154 struct bio_vec *bvec;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700155 unsigned long id;
156 unsigned int fsect, lsect;
157 int ref;
158 grant_ref_t gref_head;
159
160 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
161 return 1;
162
163 if (gnttab_alloc_grant_references(
164 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
165 gnttab_request_free_callback(
166 &info->callback,
167 blkif_restart_queue_callback,
168 info,
169 BLKIF_MAX_SEGMENTS_PER_REQUEST);
170 return 1;
171 }
172
173 /* Fill out a communications ring structure. */
174 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
175 id = get_id_from_freelist(info);
176 info->shadow[id].request = (unsigned long)req;
177
178 ring_req->id = id;
179 ring_req->sector_number = (blkif_sector_t)req->sector;
180 ring_req->handle = info->handle;
181
182 ring_req->operation = rq_data_dir(req) ?
183 BLKIF_OP_WRITE : BLKIF_OP_READ;
184 if (blk_barrier_rq(req))
185 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
186
187 ring_req->nr_segments = 0;
NeilBrown5705f702007-09-25 12:35:59 +0200188 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200189 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
190 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
191 fsect = bvec->bv_offset >> 9;
192 lsect = fsect + (bvec->bv_len >> 9) - 1;
193 /* install a grant reference. */
194 ref = gnttab_claim_grant_reference(&gref_head);
195 BUG_ON(ref == -ENOSPC);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700196
Jens Axboe6c92e692007-08-16 13:43:12 +0200197 gnttab_grant_foreign_access_ref(
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700198 ref,
199 info->xbdev->otherend_id,
200 buffer_mfn,
201 rq_data_dir(req) );
202
Jens Axboe6c92e692007-08-16 13:43:12 +0200203 info->shadow[id].frame[ring_req->nr_segments] =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700204 mfn_to_pfn(buffer_mfn);
205
Jens Axboe6c92e692007-08-16 13:43:12 +0200206 ring_req->seg[ring_req->nr_segments] =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700207 (struct blkif_request_segment) {
208 .gref = ref,
209 .first_sect = fsect,
210 .last_sect = lsect };
211
Jens Axboe6c92e692007-08-16 13:43:12 +0200212 ring_req->nr_segments++;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700213 }
214
215 info->ring.req_prod_pvt++;
216
217 /* Keep a private copy so we can reissue requests when recovering. */
218 info->shadow[id].req = *ring_req;
219
220 gnttab_free_grant_references(gref_head);
221
222 return 0;
223}
224
225
226static inline void flush_requests(struct blkfront_info *info)
227{
228 int notify;
229
230 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
231
232 if (notify)
233 notify_remote_via_irq(info->irq);
234}
235
236/*
237 * do_blkif_request
238 * read a block; request is in a request queue
239 */
Jens Axboe165125e2007-07-24 09:28:11 +0200240static void do_blkif_request(struct request_queue *rq)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700241{
242 struct blkfront_info *info = NULL;
243 struct request *req;
244 int queued;
245
246 pr_debug("Entered do_blkif_request\n");
247
248 queued = 0;
249
250 while ((req = elv_next_request(rq)) != NULL) {
251 info = req->rq_disk->private_data;
252 if (!blk_fs_request(req)) {
253 end_request(req, 0);
254 continue;
255 }
256
257 if (RING_FULL(&info->ring))
258 goto wait;
259
260 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
261 "(%u/%li) buffer:%p [%s]\n",
262 req, req->cmd, (unsigned long)req->sector,
263 req->current_nr_sectors,
264 req->nr_sectors, req->buffer,
265 rq_data_dir(req) ? "write" : "read");
266
267
268 blkdev_dequeue_request(req);
269 if (blkif_queue_request(req)) {
270 blk_requeue_request(rq, req);
271wait:
272 /* Avoid pointless unplugs. */
273 blk_stop_queue(rq);
274 break;
275 }
276
277 queued++;
278 }
279
280 if (queued != 0)
281 flush_requests(info);
282}
283
284static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
285{
Jens Axboe165125e2007-07-24 09:28:11 +0200286 struct request_queue *rq;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700287
288 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
289 if (rq == NULL)
290 return -1;
291
292 elevator_init(rq, "noop");
293
294 /* Hard sector size and max sectors impersonate the equiv. hardware. */
295 blk_queue_hardsect_size(rq, sector_size);
296 blk_queue_max_sectors(rq, 512);
297
298 /* Each segment in a request is up to an aligned page in size. */
299 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
300 blk_queue_max_segment_size(rq, PAGE_SIZE);
301
302 /* Ensure a merged request will fit in a single I/O ring slot. */
303 blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
304 blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
305
306 /* Make sure buffer addresses are sector-aligned. */
307 blk_queue_dma_alignment(rq, 511);
308
309 gd->queue = rq;
310
311 return 0;
312}
313
314
315static int xlvbd_barrier(struct blkfront_info *info)
316{
317 int err;
318
319 err = blk_queue_ordered(info->rq,
320 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
321 NULL);
322
323 if (err)
324 return err;
325
326 printk(KERN_INFO "blkfront: %s: barriers %s\n",
327 info->gd->disk_name,
328 info->feature_barrier ? "enabled" : "disabled");
329 return 0;
330}
331
332
333static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
334 int vdevice, u16 vdisk_info, u16 sector_size,
335 struct blkfront_info *info)
336{
337 struct gendisk *gd;
338 int nr_minors = 1;
339 int err = -ENODEV;
340
341 BUG_ON(info->gd != NULL);
342 BUG_ON(info->rq != NULL);
343
344 if ((minor % PARTS_PER_DISK) == 0)
345 nr_minors = PARTS_PER_DISK;
346
347 gd = alloc_disk(nr_minors);
348 if (gd == NULL)
349 goto out;
350
351 if (nr_minors > 1)
352 sprintf(gd->disk_name, "%s%c", DEV_NAME,
353 'a' + minor / PARTS_PER_DISK);
354 else
355 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
356 'a' + minor / PARTS_PER_DISK,
357 minor % PARTS_PER_DISK);
358
359 gd->major = XENVBD_MAJOR;
360 gd->first_minor = minor;
361 gd->fops = &xlvbd_block_fops;
362 gd->private_data = info;
363 gd->driverfs_dev = &(info->xbdev->dev);
364 set_capacity(gd, capacity);
365
366 if (xlvbd_init_blk_queue(gd, sector_size)) {
367 del_gendisk(gd);
368 goto out;
369 }
370
371 info->rq = gd->queue;
372 info->gd = gd;
373
374 if (info->feature_barrier)
375 xlvbd_barrier(info);
376
377 if (vdisk_info & VDISK_READONLY)
378 set_disk_ro(gd, 1);
379
380 if (vdisk_info & VDISK_REMOVABLE)
381 gd->flags |= GENHD_FL_REMOVABLE;
382
383 if (vdisk_info & VDISK_CDROM)
384 gd->flags |= GENHD_FL_CD;
385
386 return 0;
387
388 out:
389 return err;
390}
391
392static void kick_pending_request_queues(struct blkfront_info *info)
393{
394 if (!RING_FULL(&info->ring)) {
395 /* Re-enable calldowns. */
396 blk_start_queue(info->rq);
397 /* Kick things off immediately. */
398 do_blkif_request(info->rq);
399 }
400}
401
402static void blkif_restart_queue(struct work_struct *work)
403{
404 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
405
406 spin_lock_irq(&blkif_io_lock);
407 if (info->connected == BLKIF_STATE_CONNECTED)
408 kick_pending_request_queues(info);
409 spin_unlock_irq(&blkif_io_lock);
410}
411
412static void blkif_free(struct blkfront_info *info, int suspend)
413{
414 /* Prevent new requests being issued until we fix things up. */
415 spin_lock_irq(&blkif_io_lock);
416 info->connected = suspend ?
417 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
418 /* No more blkif_request(). */
419 if (info->rq)
420 blk_stop_queue(info->rq);
421 /* No more gnttab callback work. */
422 gnttab_cancel_free_callback(&info->callback);
423 spin_unlock_irq(&blkif_io_lock);
424
425 /* Flush gnttab callback work. Must be done with no locks held. */
426 flush_scheduled_work();
427
428 /* Free resources associated with old device channel. */
429 if (info->ring_ref != GRANT_INVALID_REF) {
430 gnttab_end_foreign_access(info->ring_ref, 0,
431 (unsigned long)info->ring.sring);
432 info->ring_ref = GRANT_INVALID_REF;
433 info->ring.sring = NULL;
434 }
435 if (info->irq)
436 unbind_from_irqhandler(info->irq, info);
437 info->evtchn = info->irq = 0;
438
439}
440
441static void blkif_completion(struct blk_shadow *s)
442{
443 int i;
444 for (i = 0; i < s->req.nr_segments; i++)
445 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
446}
447
448static irqreturn_t blkif_interrupt(int irq, void *dev_id)
449{
450 struct request *req;
451 struct blkif_response *bret;
452 RING_IDX i, rp;
453 unsigned long flags;
454 struct blkfront_info *info = (struct blkfront_info *)dev_id;
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500455 int error;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700456
457 spin_lock_irqsave(&blkif_io_lock, flags);
458
459 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
460 spin_unlock_irqrestore(&blkif_io_lock, flags);
461 return IRQ_HANDLED;
462 }
463
464 again:
465 rp = info->ring.sring->rsp_prod;
466 rmb(); /* Ensure we see queued responses up to 'rp'. */
467
468 for (i = info->ring.rsp_cons; i != rp; i++) {
469 unsigned long id;
470 int ret;
471
472 bret = RING_GET_RESPONSE(&info->ring, i);
473 id = bret->id;
474 req = (struct request *)info->shadow[id].request;
475
476 blkif_completion(&info->shadow[id]);
477
478 add_id_to_freelist(info, id);
479
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500480 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700481 switch (bret->operation) {
482 case BLKIF_OP_WRITE_BARRIER:
483 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
484 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
485 info->gd->disk_name);
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500486 error = -EOPNOTSUPP;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700487 info->feature_barrier = 0;
488 xlvbd_barrier(info);
489 }
490 /* fall through */
491 case BLKIF_OP_READ:
492 case BLKIF_OP_WRITE:
493 if (unlikely(bret->status != BLKIF_RSP_OKAY))
494 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
495 "request: %x\n", bret->status);
496
Kiyoshi Uedaf530f0362007-12-11 17:47:36 -0500497 ret = __blk_end_request(req, error, blk_rq_bytes(req));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700498 BUG_ON(ret);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700499 break;
500 default:
501 BUG();
502 }
503 }
504
505 info->ring.rsp_cons = i;
506
507 if (i != info->ring.req_prod_pvt) {
508 int more_to_do;
509 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
510 if (more_to_do)
511 goto again;
512 } else
513 info->ring.sring->rsp_event = i + 1;
514
515 kick_pending_request_queues(info);
516
517 spin_unlock_irqrestore(&blkif_io_lock, flags);
518
519 return IRQ_HANDLED;
520}
521
522
523static int setup_blkring(struct xenbus_device *dev,
524 struct blkfront_info *info)
525{
526 struct blkif_sring *sring;
527 int err;
528
529 info->ring_ref = GRANT_INVALID_REF;
530
531 sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL);
532 if (!sring) {
533 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
534 return -ENOMEM;
535 }
536 SHARED_RING_INIT(sring);
537 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
538
539 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
540 if (err < 0) {
541 free_page((unsigned long)sring);
542 info->ring.sring = NULL;
543 goto fail;
544 }
545 info->ring_ref = err;
546
547 err = xenbus_alloc_evtchn(dev, &info->evtchn);
548 if (err)
549 goto fail;
550
551 err = bind_evtchn_to_irqhandler(info->evtchn,
552 blkif_interrupt,
553 IRQF_SAMPLE_RANDOM, "blkif", info);
554 if (err <= 0) {
555 xenbus_dev_fatal(dev, err,
556 "bind_evtchn_to_irqhandler failed");
557 goto fail;
558 }
559 info->irq = err;
560
561 return 0;
562fail:
563 blkif_free(info, 0);
564 return err;
565}
566
567
568/* Common code used when first setting up, and when resuming. */
569static int talk_to_backend(struct xenbus_device *dev,
570 struct blkfront_info *info)
571{
572 const char *message = NULL;
573 struct xenbus_transaction xbt;
574 int err;
575
576 /* Create shared ring, alloc event channel. */
577 err = setup_blkring(dev, info);
578 if (err)
579 goto out;
580
581again:
582 err = xenbus_transaction_start(&xbt);
583 if (err) {
584 xenbus_dev_fatal(dev, err, "starting transaction");
585 goto destroy_blkring;
586 }
587
588 err = xenbus_printf(xbt, dev->nodename,
589 "ring-ref", "%u", info->ring_ref);
590 if (err) {
591 message = "writing ring-ref";
592 goto abort_transaction;
593 }
594 err = xenbus_printf(xbt, dev->nodename,
595 "event-channel", "%u", info->evtchn);
596 if (err) {
597 message = "writing event-channel";
598 goto abort_transaction;
599 }
600
601 err = xenbus_transaction_end(xbt, 0);
602 if (err) {
603 if (err == -EAGAIN)
604 goto again;
605 xenbus_dev_fatal(dev, err, "completing transaction");
606 goto destroy_blkring;
607 }
608
609 xenbus_switch_state(dev, XenbusStateInitialised);
610
611 return 0;
612
613 abort_transaction:
614 xenbus_transaction_end(xbt, 1);
615 if (message)
616 xenbus_dev_fatal(dev, err, "%s", message);
617 destroy_blkring:
618 blkif_free(info, 0);
619 out:
620 return err;
621}
622
623
624/**
625 * Entry point to this code when a new device is created. Allocate the basic
626 * structures and the ring buffer for communication with the backend, and
627 * inform the backend of the appropriate details for those. Switch to
628 * Initialised state.
629 */
630static int blkfront_probe(struct xenbus_device *dev,
631 const struct xenbus_device_id *id)
632{
633 int err, vdevice, i;
634 struct blkfront_info *info;
635
636 /* FIXME: Use dynamic device id if this is not set. */
637 err = xenbus_scanf(XBT_NIL, dev->nodename,
638 "virtual-device", "%i", &vdevice);
639 if (err != 1) {
640 xenbus_dev_fatal(dev, err, "reading virtual-device");
641 return err;
642 }
643
644 info = kzalloc(sizeof(*info), GFP_KERNEL);
645 if (!info) {
646 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
647 return -ENOMEM;
648 }
649
650 info->xbdev = dev;
651 info->vdevice = vdevice;
652 info->connected = BLKIF_STATE_DISCONNECTED;
653 INIT_WORK(&info->work, blkif_restart_queue);
654
655 for (i = 0; i < BLK_RING_SIZE; i++)
656 info->shadow[i].req.id = i+1;
657 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
658
659 /* Front end dir is a number, which is used as the id. */
660 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
661 dev->dev.driver_data = info;
662
663 err = talk_to_backend(dev, info);
664 if (err) {
665 kfree(info);
666 dev->dev.driver_data = NULL;
667 return err;
668 }
669
670 return 0;
671}
672
673
674static int blkif_recover(struct blkfront_info *info)
675{
676 int i;
677 struct blkif_request *req;
678 struct blk_shadow *copy;
679 int j;
680
681 /* Stage 1: Make a safe copy of the shadow state. */
682 copy = kmalloc(sizeof(info->shadow), GFP_KERNEL);
683 if (!copy)
684 return -ENOMEM;
685 memcpy(copy, info->shadow, sizeof(info->shadow));
686
687 /* Stage 2: Set up free list. */
688 memset(&info->shadow, 0, sizeof(info->shadow));
689 for (i = 0; i < BLK_RING_SIZE; i++)
690 info->shadow[i].req.id = i+1;
691 info->shadow_free = info->ring.req_prod_pvt;
692 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
693
694 /* Stage 3: Find pending requests and requeue them. */
695 for (i = 0; i < BLK_RING_SIZE; i++) {
696 /* Not in use? */
697 if (copy[i].request == 0)
698 continue;
699
700 /* Grab a request slot and copy shadow state into it. */
701 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
702 *req = copy[i].req;
703
704 /* We get a new request id, and must reset the shadow state. */
705 req->id = get_id_from_freelist(info);
706 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
707
708 /* Rewrite any grant references invalidated by susp/resume. */
709 for (j = 0; j < req->nr_segments; j++)
710 gnttab_grant_foreign_access_ref(
711 req->seg[j].gref,
712 info->xbdev->otherend_id,
713 pfn_to_mfn(info->shadow[req->id].frame[j]),
714 rq_data_dir(
715 (struct request *)
716 info->shadow[req->id].request));
717 info->shadow[req->id].req = *req;
718
719 info->ring.req_prod_pvt++;
720 }
721
722 kfree(copy);
723
724 xenbus_switch_state(info->xbdev, XenbusStateConnected);
725
726 spin_lock_irq(&blkif_io_lock);
727
728 /* Now safe for us to use the shared ring */
729 info->connected = BLKIF_STATE_CONNECTED;
730
731 /* Send off requeued requests */
732 flush_requests(info);
733
734 /* Kick any other new requests queued since we resumed */
735 kick_pending_request_queues(info);
736
737 spin_unlock_irq(&blkif_io_lock);
738
739 return 0;
740}
741
742/**
743 * We are reconnecting to the backend, due to a suspend/resume, or a backend
744 * driver restart. We tear down our blkif structure and recreate it, but
745 * leave the device-layer structures intact so that this is transparent to the
746 * rest of the kernel.
747 */
748static int blkfront_resume(struct xenbus_device *dev)
749{
750 struct blkfront_info *info = dev->dev.driver_data;
751 int err;
752
753 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
754
755 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
756
757 err = talk_to_backend(dev, info);
758 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
759 err = blkif_recover(info);
760
761 return err;
762}
763
764
765/*
766 * Invoked when the backend is finally 'ready' (and has told produced
767 * the details about the physical device - #sectors, size, etc).
768 */
769static void blkfront_connect(struct blkfront_info *info)
770{
771 unsigned long long sectors;
772 unsigned long sector_size;
773 unsigned int binfo;
774 int err;
775
776 if ((info->connected == BLKIF_STATE_CONNECTED) ||
777 (info->connected == BLKIF_STATE_SUSPENDED) )
778 return;
779
780 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
781 __func__, info->xbdev->otherend);
782
783 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
784 "sectors", "%llu", &sectors,
785 "info", "%u", &binfo,
786 "sector-size", "%lu", &sector_size,
787 NULL);
788 if (err) {
789 xenbus_dev_fatal(info->xbdev, err,
790 "reading backend fields at %s",
791 info->xbdev->otherend);
792 return;
793 }
794
795 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
796 "feature-barrier", "%lu", &info->feature_barrier,
797 NULL);
798 if (err)
799 info->feature_barrier = 0;
800
801 err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
802 sectors, info->vdevice,
803 binfo, sector_size, info);
804 if (err) {
805 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
806 info->xbdev->otherend);
807 return;
808 }
809
810 xenbus_switch_state(info->xbdev, XenbusStateConnected);
811
812 /* Kick pending requests. */
813 spin_lock_irq(&blkif_io_lock);
814 info->connected = BLKIF_STATE_CONNECTED;
815 kick_pending_request_queues(info);
816 spin_unlock_irq(&blkif_io_lock);
817
818 add_disk(info->gd);
819}
820
821/**
822 * Handle the change of state of the backend to Closing. We must delete our
823 * device-layer structures now, to ensure that writes are flushed through to
824 * the backend. Once is this done, we can switch to Closed in
825 * acknowledgement.
826 */
827static void blkfront_closing(struct xenbus_device *dev)
828{
829 struct blkfront_info *info = dev->dev.driver_data;
830 unsigned long flags;
831
832 dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
833
834 if (info->rq == NULL)
835 goto out;
836
837 spin_lock_irqsave(&blkif_io_lock, flags);
838
839 del_gendisk(info->gd);
840
841 /* No more blkif_request(). */
842 blk_stop_queue(info->rq);
843
844 /* No more gnttab callback work. */
845 gnttab_cancel_free_callback(&info->callback);
846 spin_unlock_irqrestore(&blkif_io_lock, flags);
847
848 /* Flush gnttab callback work. Must be done with no locks held. */
849 flush_scheduled_work();
850
851 blk_cleanup_queue(info->rq);
852 info->rq = NULL;
853
854 out:
855 xenbus_frontend_closed(dev);
856}
857
858/**
859 * Callback received when the backend's state changes.
860 */
861static void backend_changed(struct xenbus_device *dev,
862 enum xenbus_state backend_state)
863{
864 struct blkfront_info *info = dev->dev.driver_data;
865 struct block_device *bd;
866
867 dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
868
869 switch (backend_state) {
870 case XenbusStateInitialising:
871 case XenbusStateInitWait:
872 case XenbusStateInitialised:
873 case XenbusStateUnknown:
874 case XenbusStateClosed:
875 break;
876
877 case XenbusStateConnected:
878 blkfront_connect(info);
879 break;
880
881 case XenbusStateClosing:
882 bd = bdget(info->dev);
883 if (bd == NULL)
884 xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
885
886 mutex_lock(&bd->bd_mutex);
887 if (info->users > 0)
888 xenbus_dev_error(dev, -EBUSY,
889 "Device in use; refusing to close");
890 else
891 blkfront_closing(dev);
892 mutex_unlock(&bd->bd_mutex);
893 bdput(bd);
894 break;
895 }
896}
897
898static int blkfront_remove(struct xenbus_device *dev)
899{
900 struct blkfront_info *info = dev->dev.driver_data;
901
902 dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
903
904 blkif_free(info, 0);
905
906 kfree(info);
907
908 return 0;
909}
910
911static int blkif_open(struct inode *inode, struct file *filep)
912{
913 struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
914 info->users++;
915 return 0;
916}
917
918static int blkif_release(struct inode *inode, struct file *filep)
919{
920 struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
921 info->users--;
922 if (info->users == 0) {
923 /* Check whether we have been instructed to close. We will
924 have ignored this request initially, as the device was
925 still mounted. */
926 struct xenbus_device *dev = info->xbdev;
927 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
928
929 if (state == XenbusStateClosing)
930 blkfront_closing(dev);
931 }
932 return 0;
933}
934
935static struct block_device_operations xlvbd_block_fops =
936{
937 .owner = THIS_MODULE,
938 .open = blkif_open,
939 .release = blkif_release,
940};
941
942
943static struct xenbus_device_id blkfront_ids[] = {
944 { "vbd" },
945 { "" }
946};
947
948static struct xenbus_driver blkfront = {
949 .name = "vbd",
950 .owner = THIS_MODULE,
951 .ids = blkfront_ids,
952 .probe = blkfront_probe,
953 .remove = blkfront_remove,
954 .resume = blkfront_resume,
955 .otherend_changed = backend_changed,
956};
957
958static int __init xlblk_init(void)
959{
960 if (!is_running_on_xen())
961 return -ENODEV;
962
963 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
964 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
965 XENVBD_MAJOR, DEV_NAME);
966 return -ENODEV;
967 }
968
969 return xenbus_register_frontend(&blkfront);
970}
971module_init(xlblk_init);
972
973
974static void xlblk_exit(void)
975{
976 return xenbus_unregister_driver(&blkfront);
977}
978module_exit(xlblk_exit);
979
980MODULE_DESCRIPTION("Xen virtual block device frontend");
981MODULE_LICENSE("GPL");
982MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);