2d19f0cc47f272c5ae4f96ee97ea34a1bce5a407
[linux-2.6.git] / drivers / block / xen-blkfront.c
1 /*
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/hdreg.h>
41 #include <linux/cdrom.h>
42 #include <linux/module.h>
43
44 #include <xen/xenbus.h>
45 #include <xen/grant_table.h>
46 #include <xen/events.h>
47 #include <xen/page.h>
48
49 #include <xen/interface/grant_table.h>
50 #include <xen/interface/io/blkif.h>
51 #include <xen/interface/io/protocols.h>
52
53 #include <asm/xen/hypervisor.h>
54
55 enum blkif_state {
56         BLKIF_STATE_DISCONNECTED,
57         BLKIF_STATE_CONNECTED,
58         BLKIF_STATE_SUSPENDED,
59 };
60
61 struct blk_shadow {
62         struct blkif_request req;
63         unsigned long request;
64         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
65 };
66
67 static struct block_device_operations xlvbd_block_fops;
68
69 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
70
71 /*
72  * We have one of these per vbd, whether ide, scsi or 'other'.  They
73  * hang in private_data off the gendisk structure. We may end up
74  * putting all kinds of interesting stuff here :-)
75  */
76 struct blkfront_info
77 {
78         struct xenbus_device *xbdev;
79         struct gendisk *gd;
80         int vdevice;
81         blkif_vdev_t handle;
82         enum blkif_state connected;
83         int ring_ref;
84         struct blkif_front_ring ring;
85         unsigned int evtchn, irq;
86         struct request_queue *rq;
87         struct work_struct work;
88         struct gnttab_free_callback callback;
89         struct blk_shadow shadow[BLK_RING_SIZE];
90         unsigned long shadow_free;
91         int feature_barrier;
92         int is_ready;
93
94         /**
95          * The number of people holding this device open.  We won't allow a
96          * hot-unplug unless this is 0.
97          */
98         int users;
99 };
100
101 static DEFINE_SPINLOCK(blkif_io_lock);
102
103 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
104         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
105 #define GRANT_INVALID_REF       0
106
107 #define PARTS_PER_DISK          16
108 #define PARTS_PER_EXT_DISK      256
109
110 #define BLKIF_MAJOR(dev) ((dev)>>8)
111 #define BLKIF_MINOR(dev) ((dev) & 0xff)
112
113 #define EXT_SHIFT 28
114 #define EXTENDED (1<<EXT_SHIFT)
115 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
116 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
117
118 #define DEV_NAME        "xvd"   /* name in /dev */
119
120 static int get_id_from_freelist(struct blkfront_info *info)
121 {
122         unsigned long free = info->shadow_free;
123         BUG_ON(free > BLK_RING_SIZE);
124         info->shadow_free = info->shadow[free].req.id;
125         info->shadow[free].req.id = 0x0fffffee; /* debug */
126         return free;
127 }
128
129 static void add_id_to_freelist(struct blkfront_info *info,
130                                unsigned long id)
131 {
132         info->shadow[id].req.id  = info->shadow_free;
133         info->shadow[id].request = 0;
134         info->shadow_free = id;
135 }
136
137 static void blkif_restart_queue_callback(void *arg)
138 {
139         struct blkfront_info *info = (struct blkfront_info *)arg;
140         schedule_work(&info->work);
141 }
142
143 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
144 {
145         /* We don't have real geometry info, but let's at least return
146            values consistent with the size of the device */
147         sector_t nsect = get_capacity(bd->bd_disk);
148         sector_t cylinders = nsect;
149
150         hg->heads = 0xff;
151         hg->sectors = 0x3f;
152         sector_div(cylinders, hg->heads * hg->sectors);
153         hg->cylinders = cylinders;
154         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
155                 hg->cylinders = 0xffff;
156         return 0;
157 }
158
159 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
160                        unsigned command, unsigned long argument)
161 {
162         struct blkfront_info *info = bdev->bd_disk->private_data;
163         int i;
164
165         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
166                 command, (long)argument);
167
168         switch (command) {
169         case CDROMMULTISESSION:
170                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
171                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
172                         if (put_user(0, (char __user *)(argument + i)))
173                                 return -EFAULT;
174                 return 0;
175
176         case CDROM_GET_CAPABILITY: {
177                 struct gendisk *gd = info->gd;
178                 if (gd->flags & GENHD_FL_CD)
179                         return 0;
180                 return -EINVAL;
181         }
182
183         default:
184                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
185                   command);*/
186                 return -EINVAL; /* same return as native Linux */
187         }
188
189         return 0;
190 }
191
192 /*
193  * blkif_queue_request
194  *
195  * request block io
196  *
197  * id: for guest use only.
198  * operation: BLKIF_OP_{READ,WRITE,PROBE}
199  * buffer: buffer to read/write into. this should be a
200  *   virtual address in the guest os.
201  */
202 static int blkif_queue_request(struct request *req)
203 {
204         struct blkfront_info *info = req->rq_disk->private_data;
205         unsigned long buffer_mfn;
206         struct blkif_request *ring_req;
207         struct req_iterator iter;
208         struct bio_vec *bvec;
209         unsigned long id;
210         unsigned int fsect, lsect;
211         int ref;
212         grant_ref_t gref_head;
213
214         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
215                 return 1;
216
217         if (gnttab_alloc_grant_references(
218                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
219                 gnttab_request_free_callback(
220                         &info->callback,
221                         blkif_restart_queue_callback,
222                         info,
223                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
224                 return 1;
225         }
226
227         /* Fill out a communications ring structure. */
228         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
229         id = get_id_from_freelist(info);
230         info->shadow[id].request = (unsigned long)req;
231
232         ring_req->id = id;
233         ring_req->sector_number = (blkif_sector_t)req->sector;
234         ring_req->handle = info->handle;
235
236         ring_req->operation = rq_data_dir(req) ?
237                 BLKIF_OP_WRITE : BLKIF_OP_READ;
238         if (blk_barrier_rq(req))
239                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
240
241         ring_req->nr_segments = 0;
242         rq_for_each_segment(bvec, req, iter) {
243                 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
244                 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
245                 fsect = bvec->bv_offset >> 9;
246                 lsect = fsect + (bvec->bv_len >> 9) - 1;
247                 /* install a grant reference. */
248                 ref = gnttab_claim_grant_reference(&gref_head);
249                 BUG_ON(ref == -ENOSPC);
250
251                 gnttab_grant_foreign_access_ref(
252                                 ref,
253                                 info->xbdev->otherend_id,
254                                 buffer_mfn,
255                                 rq_data_dir(req) );
256
257                 info->shadow[id].frame[ring_req->nr_segments] =
258                                 mfn_to_pfn(buffer_mfn);
259
260                 ring_req->seg[ring_req->nr_segments] =
261                                 (struct blkif_request_segment) {
262                                         .gref       = ref,
263                                         .first_sect = fsect,
264                                         .last_sect  = lsect };
265
266                 ring_req->nr_segments++;
267         }
268
269         info->ring.req_prod_pvt++;
270
271         /* Keep a private copy so we can reissue requests when recovering. */
272         info->shadow[id].req = *ring_req;
273
274         gnttab_free_grant_references(gref_head);
275
276         return 0;
277 }
278
279
280 static inline void flush_requests(struct blkfront_info *info)
281 {
282         int notify;
283
284         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
285
286         if (notify)
287                 notify_remote_via_irq(info->irq);
288 }
289
290 /*
291  * do_blkif_request
292  *  read a block; request is in a request queue
293  */
294 static void do_blkif_request(struct request_queue *rq)
295 {
296         struct blkfront_info *info = NULL;
297         struct request *req;
298         int queued;
299
300         pr_debug("Entered do_blkif_request\n");
301
302         queued = 0;
303
304         while ((req = elv_next_request(rq)) != NULL) {
305                 info = req->rq_disk->private_data;
306                 if (!blk_fs_request(req)) {
307                         end_request(req, 0);
308                         continue;
309                 }
310
311                 if (RING_FULL(&info->ring))
312                         goto wait;
313
314                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
315                          "(%u/%li) buffer:%p [%s]\n",
316                          req, req->cmd, (unsigned long)req->sector,
317                          req->current_nr_sectors,
318                          req->nr_sectors, req->buffer,
319                          rq_data_dir(req) ? "write" : "read");
320
321
322                 blkdev_dequeue_request(req);
323                 if (blkif_queue_request(req)) {
324                         blk_requeue_request(rq, req);
325 wait:
326                         /* Avoid pointless unplugs. */
327                         blk_stop_queue(rq);
328                         break;
329                 }
330
331                 queued++;
332         }
333
334         if (queued != 0)
335                 flush_requests(info);
336 }
337
338 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
339 {
340         struct request_queue *rq;
341         elevator_t *old_e;
342
343         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
344         if (rq == NULL)
345                 return -1;
346
347         old_e = rq->elevator;
348         if (IS_ERR_VALUE(elevator_init(rq, "noop")))
349                 printk(KERN_WARNING
350                         "blkfront: Switch elevator failed, use default\n");
351         else
352                 elevator_exit(old_e);
353
354         /* Hard sector size and max sectors impersonate the equiv. hardware. */
355         blk_queue_hardsect_size(rq, sector_size);
356         blk_queue_max_sectors(rq, 512);
357
358         /* Each segment in a request is up to an aligned page in size. */
359         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
360         blk_queue_max_segment_size(rq, PAGE_SIZE);
361
362         /* Ensure a merged request will fit in a single I/O ring slot. */
363         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
364         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
365
366         /* Make sure buffer addresses are sector-aligned. */
367         blk_queue_dma_alignment(rq, 511);
368
369         /* Make sure we don't use bounce buffers. */
370         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
371
372         gd->queue = rq;
373
374         return 0;
375 }
376
377
378 static int xlvbd_barrier(struct blkfront_info *info)
379 {
380         int err;
381
382         err = blk_queue_ordered(info->rq,
383                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
384                                 NULL);
385
386         if (err)
387                 return err;
388
389         printk(KERN_INFO "blkfront: %s: barriers %s\n",
390                info->gd->disk_name,
391                info->feature_barrier ? "enabled" : "disabled");
392         return 0;
393 }
394
395
396 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
397                                struct blkfront_info *info,
398                                u16 vdisk_info, u16 sector_size)
399 {
400         struct gendisk *gd;
401         int nr_minors = 1;
402         int err = -ENODEV;
403         unsigned int offset;
404         int minor;
405         int nr_parts;
406
407         BUG_ON(info->gd != NULL);
408         BUG_ON(info->rq != NULL);
409
410         if ((info->vdevice>>EXT_SHIFT) > 1) {
411                 /* this is above the extended range; something is wrong */
412                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
413                 return -ENODEV;
414         }
415
416         if (!VDEV_IS_EXTENDED(info->vdevice)) {
417                 minor = BLKIF_MINOR(info->vdevice);
418                 nr_parts = PARTS_PER_DISK;
419         } else {
420                 minor = BLKIF_MINOR_EXT(info->vdevice);
421                 nr_parts = PARTS_PER_EXT_DISK;
422         }
423
424         if ((minor % nr_parts) == 0)
425                 nr_minors = nr_parts;
426
427         gd = alloc_disk(nr_minors);
428         if (gd == NULL)
429                 goto out;
430
431         offset = minor / nr_parts;
432
433         if (nr_minors > 1) {
434                 if (offset < 26)
435                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
436                 else
437                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
438                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
439         } else {
440                 if (offset < 26)
441                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
442                                 'a' + offset,
443                                 minor & (nr_parts - 1));
444                 else
445                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
446                                 'a' + ((offset / 26) - 1),
447                                 'a' + (offset % 26),
448                                 minor & (nr_parts - 1));
449         }
450
451         gd->major = XENVBD_MAJOR;
452         gd->first_minor = minor;
453         gd->fops = &xlvbd_block_fops;
454         gd->private_data = info;
455         gd->driverfs_dev = &(info->xbdev->dev);
456         set_capacity(gd, capacity);
457
458         if (xlvbd_init_blk_queue(gd, sector_size)) {
459                 del_gendisk(gd);
460                 goto out;
461         }
462
463         info->rq = gd->queue;
464         info->gd = gd;
465
466         if (info->feature_barrier)
467                 xlvbd_barrier(info);
468
469         if (vdisk_info & VDISK_READONLY)
470                 set_disk_ro(gd, 1);
471
472         if (vdisk_info & VDISK_REMOVABLE)
473                 gd->flags |= GENHD_FL_REMOVABLE;
474
475         if (vdisk_info & VDISK_CDROM)
476                 gd->flags |= GENHD_FL_CD;
477
478         return 0;
479
480  out:
481         return err;
482 }
483
484 static void kick_pending_request_queues(struct blkfront_info *info)
485 {
486         if (!RING_FULL(&info->ring)) {
487                 /* Re-enable calldowns. */
488                 blk_start_queue(info->rq);
489                 /* Kick things off immediately. */
490                 do_blkif_request(info->rq);
491         }
492 }
493
494 static void blkif_restart_queue(struct work_struct *work)
495 {
496         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
497
498         spin_lock_irq(&blkif_io_lock);
499         if (info->connected == BLKIF_STATE_CONNECTED)
500                 kick_pending_request_queues(info);
501         spin_unlock_irq(&blkif_io_lock);
502 }
503
504 static void blkif_free(struct blkfront_info *info, int suspend)
505 {
506         /* Prevent new requests being issued until we fix things up. */
507         spin_lock_irq(&blkif_io_lock);
508         info->connected = suspend ?
509                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
510         /* No more blkif_request(). */
511         if (info->rq)
512                 blk_stop_queue(info->rq);
513         /* No more gnttab callback work. */
514         gnttab_cancel_free_callback(&info->callback);
515         spin_unlock_irq(&blkif_io_lock);
516
517         /* Flush gnttab callback work. Must be done with no locks held. */
518         flush_scheduled_work();
519
520         /* Free resources associated with old device channel. */
521         if (info->ring_ref != GRANT_INVALID_REF) {
522                 gnttab_end_foreign_access(info->ring_ref, 0,
523                                           (unsigned long)info->ring.sring);
524                 info->ring_ref = GRANT_INVALID_REF;
525                 info->ring.sring = NULL;
526         }
527         if (info->irq)
528                 unbind_from_irqhandler(info->irq, info);
529         info->evtchn = info->irq = 0;
530
531 }
532
533 static void blkif_completion(struct blk_shadow *s)
534 {
535         int i;
536         for (i = 0; i < s->req.nr_segments; i++)
537                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
538 }
539
540 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
541 {
542         struct request *req;
543         struct blkif_response *bret;
544         RING_IDX i, rp;
545         unsigned long flags;
546         struct blkfront_info *info = (struct blkfront_info *)dev_id;
547         int error;
548
549         spin_lock_irqsave(&blkif_io_lock, flags);
550
551         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
552                 spin_unlock_irqrestore(&blkif_io_lock, flags);
553                 return IRQ_HANDLED;
554         }
555
556  again:
557         rp = info->ring.sring->rsp_prod;
558         rmb(); /* Ensure we see queued responses up to 'rp'. */
559
560         for (i = info->ring.rsp_cons; i != rp; i++) {
561                 unsigned long id;
562                 int ret;
563
564                 bret = RING_GET_RESPONSE(&info->ring, i);
565                 id   = bret->id;
566                 req  = (struct request *)info->shadow[id].request;
567
568                 blkif_completion(&info->shadow[id]);
569
570                 add_id_to_freelist(info, id);
571
572                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
573                 switch (bret->operation) {
574                 case BLKIF_OP_WRITE_BARRIER:
575                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
576                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
577                                        info->gd->disk_name);
578                                 error = -EOPNOTSUPP;
579                                 info->feature_barrier = 0;
580                                 xlvbd_barrier(info);
581                         }
582                         /* fall through */
583                 case BLKIF_OP_READ:
584                 case BLKIF_OP_WRITE:
585                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
586                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
587                                         "request: %x\n", bret->status);
588
589                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
590                         BUG_ON(ret);
591                         break;
592                 default:
593                         BUG();
594                 }
595         }
596
597         info->ring.rsp_cons = i;
598
599         if (i != info->ring.req_prod_pvt) {
600                 int more_to_do;
601                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
602                 if (more_to_do)
603                         goto again;
604         } else
605                 info->ring.sring->rsp_event = i + 1;
606
607         kick_pending_request_queues(info);
608
609         spin_unlock_irqrestore(&blkif_io_lock, flags);
610
611         return IRQ_HANDLED;
612 }
613
614
615 static int setup_blkring(struct xenbus_device *dev,
616                          struct blkfront_info *info)
617 {
618         struct blkif_sring *sring;
619         int err;
620
621         info->ring_ref = GRANT_INVALID_REF;
622
623         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
624         if (!sring) {
625                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
626                 return -ENOMEM;
627         }
628         SHARED_RING_INIT(sring);
629         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
630
631         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
632         if (err < 0) {
633                 free_page((unsigned long)sring);
634                 info->ring.sring = NULL;
635                 goto fail;
636         }
637         info->ring_ref = err;
638
639         err = xenbus_alloc_evtchn(dev, &info->evtchn);
640         if (err)
641                 goto fail;
642
643         err = bind_evtchn_to_irqhandler(info->evtchn,
644                                         blkif_interrupt,
645                                         IRQF_SAMPLE_RANDOM, "blkif", info);
646         if (err <= 0) {
647                 xenbus_dev_fatal(dev, err,
648                                  "bind_evtchn_to_irqhandler failed");
649                 goto fail;
650         }
651         info->irq = err;
652
653         return 0;
654 fail:
655         blkif_free(info, 0);
656         return err;
657 }
658
659
660 /* Common code used when first setting up, and when resuming. */
661 static int talk_to_backend(struct xenbus_device *dev,
662                            struct blkfront_info *info)
663 {
664         const char *message = NULL;
665         struct xenbus_transaction xbt;
666         int err;
667
668         /* Create shared ring, alloc event channel. */
669         err = setup_blkring(dev, info);
670         if (err)
671                 goto out;
672
673 again:
674         err = xenbus_transaction_start(&xbt);
675         if (err) {
676                 xenbus_dev_fatal(dev, err, "starting transaction");
677                 goto destroy_blkring;
678         }
679
680         err = xenbus_printf(xbt, dev->nodename,
681                             "ring-ref", "%u", info->ring_ref);
682         if (err) {
683                 message = "writing ring-ref";
684                 goto abort_transaction;
685         }
686         err = xenbus_printf(xbt, dev->nodename,
687                             "event-channel", "%u", info->evtchn);
688         if (err) {
689                 message = "writing event-channel";
690                 goto abort_transaction;
691         }
692         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
693                             XEN_IO_PROTO_ABI_NATIVE);
694         if (err) {
695                 message = "writing protocol";
696                 goto abort_transaction;
697         }
698
699         err = xenbus_transaction_end(xbt, 0);
700         if (err) {
701                 if (err == -EAGAIN)
702                         goto again;
703                 xenbus_dev_fatal(dev, err, "completing transaction");
704                 goto destroy_blkring;
705         }
706
707         xenbus_switch_state(dev, XenbusStateInitialised);
708
709         return 0;
710
711  abort_transaction:
712         xenbus_transaction_end(xbt, 1);
713         if (message)
714                 xenbus_dev_fatal(dev, err, "%s", message);
715  destroy_blkring:
716         blkif_free(info, 0);
717  out:
718         return err;
719 }
720
721
722 /**
723  * Entry point to this code when a new device is created.  Allocate the basic
724  * structures and the ring buffer for communication with the backend, and
725  * inform the backend of the appropriate details for those.  Switch to
726  * Initialised state.
727  */
728 static int blkfront_probe(struct xenbus_device *dev,
729                           const struct xenbus_device_id *id)
730 {
731         int err, vdevice, i;
732         struct blkfront_info *info;
733
734         /* FIXME: Use dynamic device id if this is not set. */
735         err = xenbus_scanf(XBT_NIL, dev->nodename,
736                            "virtual-device", "%i", &vdevice);
737         if (err != 1) {
738                 /* go looking in the extended area instead */
739                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
740                                    "%i", &vdevice);
741                 if (err != 1) {
742                         xenbus_dev_fatal(dev, err, "reading virtual-device");
743                         return err;
744                 }
745         }
746
747         info = kzalloc(sizeof(*info), GFP_KERNEL);
748         if (!info) {
749                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
750                 return -ENOMEM;
751         }
752
753         info->xbdev = dev;
754         info->vdevice = vdevice;
755         info->connected = BLKIF_STATE_DISCONNECTED;
756         INIT_WORK(&info->work, blkif_restart_queue);
757
758         for (i = 0; i < BLK_RING_SIZE; i++)
759                 info->shadow[i].req.id = i+1;
760         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
761
762         /* Front end dir is a number, which is used as the id. */
763         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
764         dev->dev.driver_data = info;
765
766         err = talk_to_backend(dev, info);
767         if (err) {
768                 kfree(info);
769                 dev->dev.driver_data = NULL;
770                 return err;
771         }
772
773         return 0;
774 }
775
776
777 static int blkif_recover(struct blkfront_info *info)
778 {
779         int i;
780         struct blkif_request *req;
781         struct blk_shadow *copy;
782         int j;
783
784         /* Stage 1: Make a safe copy of the shadow state. */
785         copy = kmalloc(sizeof(info->shadow),
786                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
787         if (!copy)
788                 return -ENOMEM;
789         memcpy(copy, info->shadow, sizeof(info->shadow));
790
791         /* Stage 2: Set up free list. */
792         memset(&info->shadow, 0, sizeof(info->shadow));
793         for (i = 0; i < BLK_RING_SIZE; i++)
794                 info->shadow[i].req.id = i+1;
795         info->shadow_free = info->ring.req_prod_pvt;
796         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
797
798         /* Stage 3: Find pending requests and requeue them. */
799         for (i = 0; i < BLK_RING_SIZE; i++) {
800                 /* Not in use? */
801                 if (copy[i].request == 0)
802                         continue;
803
804                 /* Grab a request slot and copy shadow state into it. */
805                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
806                 *req = copy[i].req;
807
808                 /* We get a new request id, and must reset the shadow state. */
809                 req->id = get_id_from_freelist(info);
810                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
811
812                 /* Rewrite any grant references invalidated by susp/resume. */
813                 for (j = 0; j < req->nr_segments; j++)
814                         gnttab_grant_foreign_access_ref(
815                                 req->seg[j].gref,
816                                 info->xbdev->otherend_id,
817                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
818                                 rq_data_dir(
819                                         (struct request *)
820                                         info->shadow[req->id].request));
821                 info->shadow[req->id].req = *req;
822
823                 info->ring.req_prod_pvt++;
824         }
825
826         kfree(copy);
827
828         xenbus_switch_state(info->xbdev, XenbusStateConnected);
829
830         spin_lock_irq(&blkif_io_lock);
831
832         /* Now safe for us to use the shared ring */
833         info->connected = BLKIF_STATE_CONNECTED;
834
835         /* Send off requeued requests */
836         flush_requests(info);
837
838         /* Kick any other new requests queued since we resumed */
839         kick_pending_request_queues(info);
840
841         spin_unlock_irq(&blkif_io_lock);
842
843         return 0;
844 }
845
846 /**
847  * We are reconnecting to the backend, due to a suspend/resume, or a backend
848  * driver restart.  We tear down our blkif structure and recreate it, but
849  * leave the device-layer structures intact so that this is transparent to the
850  * rest of the kernel.
851  */
852 static int blkfront_resume(struct xenbus_device *dev)
853 {
854         struct blkfront_info *info = dev->dev.driver_data;
855         int err;
856
857         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
858
859         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
860
861         err = talk_to_backend(dev, info);
862         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
863                 err = blkif_recover(info);
864
865         return err;
866 }
867
868
869 /*
870  * Invoked when the backend is finally 'ready' (and has told produced
871  * the details about the physical device - #sectors, size, etc).
872  */
873 static void blkfront_connect(struct blkfront_info *info)
874 {
875         unsigned long long sectors;
876         unsigned long sector_size;
877         unsigned int binfo;
878         int err;
879
880         if ((info->connected == BLKIF_STATE_CONNECTED) ||
881             (info->connected == BLKIF_STATE_SUSPENDED) )
882                 return;
883
884         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
885                 __func__, info->xbdev->otherend);
886
887         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
888                             "sectors", "%llu", &sectors,
889                             "info", "%u", &binfo,
890                             "sector-size", "%lu", &sector_size,
891                             NULL);
892         if (err) {
893                 xenbus_dev_fatal(info->xbdev, err,
894                                  "reading backend fields at %s",
895                                  info->xbdev->otherend);
896                 return;
897         }
898
899         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
900                             "feature-barrier", "%lu", &info->feature_barrier,
901                             NULL);
902         if (err)
903                 info->feature_barrier = 0;
904
905         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
906         if (err) {
907                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
908                                  info->xbdev->otherend);
909                 return;
910         }
911
912         xenbus_switch_state(info->xbdev, XenbusStateConnected);
913
914         /* Kick pending requests. */
915         spin_lock_irq(&blkif_io_lock);
916         info->connected = BLKIF_STATE_CONNECTED;
917         kick_pending_request_queues(info);
918         spin_unlock_irq(&blkif_io_lock);
919
920         add_disk(info->gd);
921
922         info->is_ready = 1;
923 }
924
925 /**
926  * Handle the change of state of the backend to Closing.  We must delete our
927  * device-layer structures now, to ensure that writes are flushed through to
928  * the backend.  Once is this done, we can switch to Closed in
929  * acknowledgement.
930  */
931 static void blkfront_closing(struct xenbus_device *dev)
932 {
933         struct blkfront_info *info = dev->dev.driver_data;
934         unsigned long flags;
935
936         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
937
938         if (info->rq == NULL)
939                 goto out;
940
941         spin_lock_irqsave(&blkif_io_lock, flags);
942
943         del_gendisk(info->gd);
944
945         /* No more blkif_request(). */
946         blk_stop_queue(info->rq);
947
948         /* No more gnttab callback work. */
949         gnttab_cancel_free_callback(&info->callback);
950         spin_unlock_irqrestore(&blkif_io_lock, flags);
951
952         /* Flush gnttab callback work. Must be done with no locks held. */
953         flush_scheduled_work();
954
955         blk_cleanup_queue(info->rq);
956         info->rq = NULL;
957
958  out:
959         xenbus_frontend_closed(dev);
960 }
961
962 /**
963  * Callback received when the backend's state changes.
964  */
965 static void backend_changed(struct xenbus_device *dev,
966                             enum xenbus_state backend_state)
967 {
968         struct blkfront_info *info = dev->dev.driver_data;
969         struct block_device *bd;
970
971         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
972
973         switch (backend_state) {
974         case XenbusStateInitialising:
975         case XenbusStateInitWait:
976         case XenbusStateInitialised:
977         case XenbusStateUnknown:
978         case XenbusStateClosed:
979                 break;
980
981         case XenbusStateConnected:
982                 blkfront_connect(info);
983                 break;
984
985         case XenbusStateClosing:
986                 bd = bdget_disk(info->gd, 0);
987                 if (bd == NULL)
988                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
989
990                 mutex_lock(&bd->bd_mutex);
991                 if (info->users > 0)
992                         xenbus_dev_error(dev, -EBUSY,
993                                          "Device in use; refusing to close");
994                 else
995                         blkfront_closing(dev);
996                 mutex_unlock(&bd->bd_mutex);
997                 bdput(bd);
998                 break;
999         }
1000 }
1001
1002 static int blkfront_remove(struct xenbus_device *dev)
1003 {
1004         struct blkfront_info *info = dev->dev.driver_data;
1005
1006         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1007
1008         blkif_free(info, 0);
1009
1010         kfree(info);
1011
1012         return 0;
1013 }
1014
1015 static int blkfront_is_ready(struct xenbus_device *dev)
1016 {
1017         struct blkfront_info *info = dev->dev.driver_data;
1018
1019         return info->is_ready;
1020 }
1021
1022 static int blkif_open(struct block_device *bdev, fmode_t mode)
1023 {
1024         struct blkfront_info *info = bdev->bd_disk->private_data;
1025         info->users++;
1026         return 0;
1027 }
1028
1029 static int blkif_release(struct gendisk *disk, fmode_t mode)
1030 {
1031         struct blkfront_info *info = disk->private_data;
1032         info->users--;
1033         if (info->users == 0) {
1034                 /* Check whether we have been instructed to close.  We will
1035                    have ignored this request initially, as the device was
1036                    still mounted. */
1037                 struct xenbus_device *dev = info->xbdev;
1038                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1039
1040                 if (state == XenbusStateClosing && info->is_ready)
1041                         blkfront_closing(dev);
1042         }
1043         return 0;
1044 }
1045
1046 static struct block_device_operations xlvbd_block_fops =
1047 {
1048         .owner = THIS_MODULE,
1049         .open = blkif_open,
1050         .release = blkif_release,
1051         .getgeo = blkif_getgeo,
1052         .locked_ioctl = blkif_ioctl,
1053 };
1054
1055
1056 static struct xenbus_device_id blkfront_ids[] = {
1057         { "vbd" },
1058         { "" }
1059 };
1060
1061 static struct xenbus_driver blkfront = {
1062         .name = "vbd",
1063         .owner = THIS_MODULE,
1064         .ids = blkfront_ids,
1065         .probe = blkfront_probe,
1066         .remove = blkfront_remove,
1067         .resume = blkfront_resume,
1068         .otherend_changed = backend_changed,
1069         .is_ready = blkfront_is_ready,
1070 };
1071
1072 static int __init xlblk_init(void)
1073 {
1074         if (!xen_domain())
1075                 return -ENODEV;
1076
1077         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1078                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1079                        XENVBD_MAJOR, DEV_NAME);
1080                 return -ENODEV;
1081         }
1082
1083         return xenbus_register_frontend(&blkfront);
1084 }
1085 module_init(xlblk_init);
1086
1087
1088 static void __exit xlblk_exit(void)
1089 {
1090         return xenbus_unregister_driver(&blkfront);
1091 }
1092 module_exit(xlblk_exit);
1093
1094 MODULE_DESCRIPTION("Xen virtual block device frontend");
1095 MODULE_LICENSE("GPL");
1096 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1097 MODULE_ALIAS("xen:vbd");
1098 MODULE_ALIAS("xenblk");