Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[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 #include <linux/scatterlist.h>
44
45 #include <xen/xen.h>
46 #include <xen/xenbus.h>
47 #include <xen/grant_table.h>
48 #include <xen/events.h>
49 #include <xen/page.h>
50
51 #include <xen/interface/grant_table.h>
52 #include <xen/interface/io/blkif.h>
53 #include <xen/interface/io/protocols.h>
54
55 #include <asm/xen/hypervisor.h>
56
57 enum blkif_state {
58         BLKIF_STATE_DISCONNECTED,
59         BLKIF_STATE_CONNECTED,
60         BLKIF_STATE_SUSPENDED,
61 };
62
63 struct blk_shadow {
64         struct blkif_request req;
65         unsigned long request;
66         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
67 };
68
69 static const struct block_device_operations xlvbd_block_fops;
70
71 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
72
73 /*
74  * We have one of these per vbd, whether ide, scsi or 'other'.  They
75  * hang in private_data off the gendisk structure. We may end up
76  * putting all kinds of interesting stuff here :-)
77  */
78 struct blkfront_info
79 {
80         struct xenbus_device *xbdev;
81         struct gendisk *gd;
82         int vdevice;
83         blkif_vdev_t handle;
84         enum blkif_state connected;
85         int ring_ref;
86         struct blkif_front_ring ring;
87         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
88         unsigned int evtchn, irq;
89         struct request_queue *rq;
90         struct work_struct work;
91         struct gnttab_free_callback callback;
92         struct blk_shadow shadow[BLK_RING_SIZE];
93         unsigned long shadow_free;
94         int feature_barrier;
95         int is_ready;
96
97         /**
98          * The number of people holding this device open.  We won't allow a
99          * hot-unplug unless this is 0.
100          */
101         int users;
102 };
103
104 static DEFINE_SPINLOCK(blkif_io_lock);
105
106 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
107         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
108 #define GRANT_INVALID_REF       0
109
110 #define PARTS_PER_DISK          16
111 #define PARTS_PER_EXT_DISK      256
112
113 #define BLKIF_MAJOR(dev) ((dev)>>8)
114 #define BLKIF_MINOR(dev) ((dev) & 0xff)
115
116 #define EXT_SHIFT 28
117 #define EXTENDED (1<<EXT_SHIFT)
118 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
119 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
120
121 #define DEV_NAME        "xvd"   /* name in /dev */
122
123 static int get_id_from_freelist(struct blkfront_info *info)
124 {
125         unsigned long free = info->shadow_free;
126         BUG_ON(free >= BLK_RING_SIZE);
127         info->shadow_free = info->shadow[free].req.id;
128         info->shadow[free].req.id = 0x0fffffee; /* debug */
129         return free;
130 }
131
132 static void add_id_to_freelist(struct blkfront_info *info,
133                                unsigned long id)
134 {
135         info->shadow[id].req.id  = info->shadow_free;
136         info->shadow[id].request = 0;
137         info->shadow_free = id;
138 }
139
140 static void blkif_restart_queue_callback(void *arg)
141 {
142         struct blkfront_info *info = (struct blkfront_info *)arg;
143         schedule_work(&info->work);
144 }
145
146 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
147 {
148         /* We don't have real geometry info, but let's at least return
149            values consistent with the size of the device */
150         sector_t nsect = get_capacity(bd->bd_disk);
151         sector_t cylinders = nsect;
152
153         hg->heads = 0xff;
154         hg->sectors = 0x3f;
155         sector_div(cylinders, hg->heads * hg->sectors);
156         hg->cylinders = cylinders;
157         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
158                 hg->cylinders = 0xffff;
159         return 0;
160 }
161
162 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
163                        unsigned command, unsigned long argument)
164 {
165         struct blkfront_info *info = bdev->bd_disk->private_data;
166         int i;
167
168         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
169                 command, (long)argument);
170
171         switch (command) {
172         case CDROMMULTISESSION:
173                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
174                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
175                         if (put_user(0, (char __user *)(argument + i)))
176                                 return -EFAULT;
177                 return 0;
178
179         case CDROM_GET_CAPABILITY: {
180                 struct gendisk *gd = info->gd;
181                 if (gd->flags & GENHD_FL_CD)
182                         return 0;
183                 return -EINVAL;
184         }
185
186         default:
187                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
188                   command);*/
189                 return -EINVAL; /* same return as native Linux */
190         }
191
192         return 0;
193 }
194
195 /*
196  * blkif_queue_request
197  *
198  * request block io
199  *
200  * id: for guest use only.
201  * operation: BLKIF_OP_{READ,WRITE,PROBE}
202  * buffer: buffer to read/write into. this should be a
203  *   virtual address in the guest os.
204  */
205 static int blkif_queue_request(struct request *req)
206 {
207         struct blkfront_info *info = req->rq_disk->private_data;
208         unsigned long buffer_mfn;
209         struct blkif_request *ring_req;
210         unsigned long id;
211         unsigned int fsect, lsect;
212         int i, ref;
213         grant_ref_t gref_head;
214         struct scatterlist *sg;
215
216         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
217                 return 1;
218
219         if (gnttab_alloc_grant_references(
220                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
221                 gnttab_request_free_callback(
222                         &info->callback,
223                         blkif_restart_queue_callback,
224                         info,
225                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
226                 return 1;
227         }
228
229         /* Fill out a communications ring structure. */
230         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
231         id = get_id_from_freelist(info);
232         info->shadow[id].request = (unsigned long)req;
233
234         ring_req->id = id;
235         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
236         ring_req->handle = info->handle;
237
238         ring_req->operation = rq_data_dir(req) ?
239                 BLKIF_OP_WRITE : BLKIF_OP_READ;
240         if (blk_barrier_rq(req))
241                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
242
243         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
244         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
245
246         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
247                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
248                 fsect = sg->offset >> 9;
249                 lsect = fsect + (sg->length >> 9) - 1;
250                 /* install a grant reference. */
251                 ref = gnttab_claim_grant_reference(&gref_head);
252                 BUG_ON(ref == -ENOSPC);
253
254                 gnttab_grant_foreign_access_ref(
255                                 ref,
256                                 info->xbdev->otherend_id,
257                                 buffer_mfn,
258                                 rq_data_dir(req) );
259
260                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
261                 ring_req->seg[i] =
262                                 (struct blkif_request_segment) {
263                                         .gref       = ref,
264                                         .first_sect = fsect,
265                                         .last_sect  = lsect };
266         }
267
268         info->ring.req_prod_pvt++;
269
270         /* Keep a private copy so we can reissue requests when recovering. */
271         info->shadow[id].req = *ring_req;
272
273         gnttab_free_grant_references(gref_head);
274
275         return 0;
276 }
277
278
279 static inline void flush_requests(struct blkfront_info *info)
280 {
281         int notify;
282
283         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
284
285         if (notify)
286                 notify_remote_via_irq(info->irq);
287 }
288
289 /*
290  * do_blkif_request
291  *  read a block; request is in a request queue
292  */
293 static void do_blkif_request(struct request_queue *rq)
294 {
295         struct blkfront_info *info = NULL;
296         struct request *req;
297         int queued;
298
299         pr_debug("Entered do_blkif_request\n");
300
301         queued = 0;
302
303         while ((req = blk_peek_request(rq)) != NULL) {
304                 info = req->rq_disk->private_data;
305
306                 if (RING_FULL(&info->ring))
307                         goto wait;
308
309                 blk_start_request(req);
310
311                 if (!blk_fs_request(req)) {
312                         __blk_end_request_all(req, -EIO);
313                         continue;
314                 }
315
316                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
317                          "(%u/%u) buffer:%p [%s]\n",
318                          req, req->cmd, (unsigned long)blk_rq_pos(req),
319                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
320                          req->buffer, rq_data_dir(req) ? "write" : "read");
321
322                 if (blkif_queue_request(req)) {
323                         blk_requeue_request(rq, req);
324 wait:
325                         /* Avoid pointless unplugs. */
326                         blk_stop_queue(rq);
327                         break;
328                 }
329
330                 queued++;
331         }
332
333         if (queued != 0)
334                 flush_requests(info);
335 }
336
337 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
338 {
339         struct request_queue *rq;
340
341         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
342         if (rq == NULL)
343                 return -1;
344
345         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
346
347         /* Hard sector size and max sectors impersonate the equiv. hardware. */
348         blk_queue_logical_block_size(rq, sector_size);
349         blk_queue_max_sectors(rq, 512);
350
351         /* Each segment in a request is up to an aligned page in size. */
352         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
353         blk_queue_max_segment_size(rq, PAGE_SIZE);
354
355         /* Ensure a merged request will fit in a single I/O ring slot. */
356         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
357         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
358
359         /* Make sure buffer addresses are sector-aligned. */
360         blk_queue_dma_alignment(rq, 511);
361
362         /* Make sure we don't use bounce buffers. */
363         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
364
365         gd->queue = rq;
366
367         return 0;
368 }
369
370
371 static int xlvbd_barrier(struct blkfront_info *info)
372 {
373         int err;
374
375         err = blk_queue_ordered(info->rq,
376                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
377                                 NULL);
378
379         if (err)
380                 return err;
381
382         printk(KERN_INFO "blkfront: %s: barriers %s\n",
383                info->gd->disk_name,
384                info->feature_barrier ? "enabled" : "disabled");
385         return 0;
386 }
387
388
389 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
390                                struct blkfront_info *info,
391                                u16 vdisk_info, u16 sector_size)
392 {
393         struct gendisk *gd;
394         int nr_minors = 1;
395         int err = -ENODEV;
396         unsigned int offset;
397         int minor;
398         int nr_parts;
399
400         BUG_ON(info->gd != NULL);
401         BUG_ON(info->rq != NULL);
402
403         if ((info->vdevice>>EXT_SHIFT) > 1) {
404                 /* this is above the extended range; something is wrong */
405                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
406                 return -ENODEV;
407         }
408
409         if (!VDEV_IS_EXTENDED(info->vdevice)) {
410                 minor = BLKIF_MINOR(info->vdevice);
411                 nr_parts = PARTS_PER_DISK;
412         } else {
413                 minor = BLKIF_MINOR_EXT(info->vdevice);
414                 nr_parts = PARTS_PER_EXT_DISK;
415         }
416
417         if ((minor % nr_parts) == 0)
418                 nr_minors = nr_parts;
419
420         gd = alloc_disk(nr_minors);
421         if (gd == NULL)
422                 goto out;
423
424         offset = minor / nr_parts;
425
426         if (nr_minors > 1) {
427                 if (offset < 26)
428                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
429                 else
430                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
431                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
432         } else {
433                 if (offset < 26)
434                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
435                                 'a' + offset,
436                                 minor & (nr_parts - 1));
437                 else
438                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
439                                 'a' + ((offset / 26) - 1),
440                                 'a' + (offset % 26),
441                                 minor & (nr_parts - 1));
442         }
443
444         gd->major = XENVBD_MAJOR;
445         gd->first_minor = minor;
446         gd->fops = &xlvbd_block_fops;
447         gd->private_data = info;
448         gd->driverfs_dev = &(info->xbdev->dev);
449         set_capacity(gd, capacity);
450
451         if (xlvbd_init_blk_queue(gd, sector_size)) {
452                 del_gendisk(gd);
453                 goto out;
454         }
455
456         info->rq = gd->queue;
457         info->gd = gd;
458
459         if (info->feature_barrier)
460                 xlvbd_barrier(info);
461
462         if (vdisk_info & VDISK_READONLY)
463                 set_disk_ro(gd, 1);
464
465         if (vdisk_info & VDISK_REMOVABLE)
466                 gd->flags |= GENHD_FL_REMOVABLE;
467
468         if (vdisk_info & VDISK_CDROM)
469                 gd->flags |= GENHD_FL_CD;
470
471         return 0;
472
473  out:
474         return err;
475 }
476
477 static void kick_pending_request_queues(struct blkfront_info *info)
478 {
479         if (!RING_FULL(&info->ring)) {
480                 /* Re-enable calldowns. */
481                 blk_start_queue(info->rq);
482                 /* Kick things off immediately. */
483                 do_blkif_request(info->rq);
484         }
485 }
486
487 static void blkif_restart_queue(struct work_struct *work)
488 {
489         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
490
491         spin_lock_irq(&blkif_io_lock);
492         if (info->connected == BLKIF_STATE_CONNECTED)
493                 kick_pending_request_queues(info);
494         spin_unlock_irq(&blkif_io_lock);
495 }
496
497 static void blkif_free(struct blkfront_info *info, int suspend)
498 {
499         /* Prevent new requests being issued until we fix things up. */
500         spin_lock_irq(&blkif_io_lock);
501         info->connected = suspend ?
502                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
503         /* No more blkif_request(). */
504         if (info->rq)
505                 blk_stop_queue(info->rq);
506         /* No more gnttab callback work. */
507         gnttab_cancel_free_callback(&info->callback);
508         spin_unlock_irq(&blkif_io_lock);
509
510         /* Flush gnttab callback work. Must be done with no locks held. */
511         flush_scheduled_work();
512
513         /* Free resources associated with old device channel. */
514         if (info->ring_ref != GRANT_INVALID_REF) {
515                 gnttab_end_foreign_access(info->ring_ref, 0,
516                                           (unsigned long)info->ring.sring);
517                 info->ring_ref = GRANT_INVALID_REF;
518                 info->ring.sring = NULL;
519         }
520         if (info->irq)
521                 unbind_from_irqhandler(info->irq, info);
522         info->evtchn = info->irq = 0;
523
524 }
525
526 static void blkif_completion(struct blk_shadow *s)
527 {
528         int i;
529         for (i = 0; i < s->req.nr_segments; i++)
530                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
531 }
532
533 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
534 {
535         struct request *req;
536         struct blkif_response *bret;
537         RING_IDX i, rp;
538         unsigned long flags;
539         struct blkfront_info *info = (struct blkfront_info *)dev_id;
540         int error;
541
542         spin_lock_irqsave(&blkif_io_lock, flags);
543
544         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
545                 spin_unlock_irqrestore(&blkif_io_lock, flags);
546                 return IRQ_HANDLED;
547         }
548
549  again:
550         rp = info->ring.sring->rsp_prod;
551         rmb(); /* Ensure we see queued responses up to 'rp'. */
552
553         for (i = info->ring.rsp_cons; i != rp; i++) {
554                 unsigned long id;
555
556                 bret = RING_GET_RESPONSE(&info->ring, i);
557                 id   = bret->id;
558                 req  = (struct request *)info->shadow[id].request;
559
560                 blkif_completion(&info->shadow[id]);
561
562                 add_id_to_freelist(info, id);
563
564                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
565                 switch (bret->operation) {
566                 case BLKIF_OP_WRITE_BARRIER:
567                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
568                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
569                                        info->gd->disk_name);
570                                 error = -EOPNOTSUPP;
571                                 info->feature_barrier = 0;
572                                 xlvbd_barrier(info);
573                         }
574                         /* fall through */
575                 case BLKIF_OP_READ:
576                 case BLKIF_OP_WRITE:
577                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
578                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
579                                         "request: %x\n", bret->status);
580
581                         __blk_end_request_all(req, error);
582                         break;
583                 default:
584                         BUG();
585                 }
586         }
587
588         info->ring.rsp_cons = i;
589
590         if (i != info->ring.req_prod_pvt) {
591                 int more_to_do;
592                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
593                 if (more_to_do)
594                         goto again;
595         } else
596                 info->ring.sring->rsp_event = i + 1;
597
598         kick_pending_request_queues(info);
599
600         spin_unlock_irqrestore(&blkif_io_lock, flags);
601
602         return IRQ_HANDLED;
603 }
604
605
606 static int setup_blkring(struct xenbus_device *dev,
607                          struct blkfront_info *info)
608 {
609         struct blkif_sring *sring;
610         int err;
611
612         info->ring_ref = GRANT_INVALID_REF;
613
614         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
615         if (!sring) {
616                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
617                 return -ENOMEM;
618         }
619         SHARED_RING_INIT(sring);
620         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
621
622         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
623
624         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
625         if (err < 0) {
626                 free_page((unsigned long)sring);
627                 info->ring.sring = NULL;
628                 goto fail;
629         }
630         info->ring_ref = err;
631
632         err = xenbus_alloc_evtchn(dev, &info->evtchn);
633         if (err)
634                 goto fail;
635
636         err = bind_evtchn_to_irqhandler(info->evtchn,
637                                         blkif_interrupt,
638                                         IRQF_SAMPLE_RANDOM, "blkif", info);
639         if (err <= 0) {
640                 xenbus_dev_fatal(dev, err,
641                                  "bind_evtchn_to_irqhandler failed");
642                 goto fail;
643         }
644         info->irq = err;
645
646         return 0;
647 fail:
648         blkif_free(info, 0);
649         return err;
650 }
651
652
653 /* Common code used when first setting up, and when resuming. */
654 static int talk_to_backend(struct xenbus_device *dev,
655                            struct blkfront_info *info)
656 {
657         const char *message = NULL;
658         struct xenbus_transaction xbt;
659         int err;
660
661         /* Create shared ring, alloc event channel. */
662         err = setup_blkring(dev, info);
663         if (err)
664                 goto out;
665
666 again:
667         err = xenbus_transaction_start(&xbt);
668         if (err) {
669                 xenbus_dev_fatal(dev, err, "starting transaction");
670                 goto destroy_blkring;
671         }
672
673         err = xenbus_printf(xbt, dev->nodename,
674                             "ring-ref", "%u", info->ring_ref);
675         if (err) {
676                 message = "writing ring-ref";
677                 goto abort_transaction;
678         }
679         err = xenbus_printf(xbt, dev->nodename,
680                             "event-channel", "%u", info->evtchn);
681         if (err) {
682                 message = "writing event-channel";
683                 goto abort_transaction;
684         }
685         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
686                             XEN_IO_PROTO_ABI_NATIVE);
687         if (err) {
688                 message = "writing protocol";
689                 goto abort_transaction;
690         }
691
692         err = xenbus_transaction_end(xbt, 0);
693         if (err) {
694                 if (err == -EAGAIN)
695                         goto again;
696                 xenbus_dev_fatal(dev, err, "completing transaction");
697                 goto destroy_blkring;
698         }
699
700         xenbus_switch_state(dev, XenbusStateInitialised);
701
702         return 0;
703
704  abort_transaction:
705         xenbus_transaction_end(xbt, 1);
706         if (message)
707                 xenbus_dev_fatal(dev, err, "%s", message);
708  destroy_blkring:
709         blkif_free(info, 0);
710  out:
711         return err;
712 }
713
714
715 /**
716  * Entry point to this code when a new device is created.  Allocate the basic
717  * structures and the ring buffer for communication with the backend, and
718  * inform the backend of the appropriate details for those.  Switch to
719  * Initialised state.
720  */
721 static int blkfront_probe(struct xenbus_device *dev,
722                           const struct xenbus_device_id *id)
723 {
724         int err, vdevice, i;
725         struct blkfront_info *info;
726
727         /* FIXME: Use dynamic device id if this is not set. */
728         err = xenbus_scanf(XBT_NIL, dev->nodename,
729                            "virtual-device", "%i", &vdevice);
730         if (err != 1) {
731                 /* go looking in the extended area instead */
732                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
733                                    "%i", &vdevice);
734                 if (err != 1) {
735                         xenbus_dev_fatal(dev, err, "reading virtual-device");
736                         return err;
737                 }
738         }
739
740         info = kzalloc(sizeof(*info), GFP_KERNEL);
741         if (!info) {
742                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
743                 return -ENOMEM;
744         }
745
746         info->xbdev = dev;
747         info->vdevice = vdevice;
748         info->connected = BLKIF_STATE_DISCONNECTED;
749         INIT_WORK(&info->work, blkif_restart_queue);
750
751         for (i = 0; i < BLK_RING_SIZE; i++)
752                 info->shadow[i].req.id = i+1;
753         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
754
755         /* Front end dir is a number, which is used as the id. */
756         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
757         dev_set_drvdata(&dev->dev, info);
758
759         err = talk_to_backend(dev, info);
760         if (err) {
761                 kfree(info);
762                 dev_set_drvdata(&dev->dev, NULL);
763                 return err;
764         }
765
766         return 0;
767 }
768
769
770 static int blkif_recover(struct blkfront_info *info)
771 {
772         int i;
773         struct blkif_request *req;
774         struct blk_shadow *copy;
775         int j;
776
777         /* Stage 1: Make a safe copy of the shadow state. */
778         copy = kmalloc(sizeof(info->shadow),
779                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
780         if (!copy)
781                 return -ENOMEM;
782         memcpy(copy, info->shadow, sizeof(info->shadow));
783
784         /* Stage 2: Set up free list. */
785         memset(&info->shadow, 0, sizeof(info->shadow));
786         for (i = 0; i < BLK_RING_SIZE; i++)
787                 info->shadow[i].req.id = i+1;
788         info->shadow_free = info->ring.req_prod_pvt;
789         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
790
791         /* Stage 3: Find pending requests and requeue them. */
792         for (i = 0; i < BLK_RING_SIZE; i++) {
793                 /* Not in use? */
794                 if (copy[i].request == 0)
795                         continue;
796
797                 /* Grab a request slot and copy shadow state into it. */
798                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
799                 *req = copy[i].req;
800
801                 /* We get a new request id, and must reset the shadow state. */
802                 req->id = get_id_from_freelist(info);
803                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
804
805                 /* Rewrite any grant references invalidated by susp/resume. */
806                 for (j = 0; j < req->nr_segments; j++)
807                         gnttab_grant_foreign_access_ref(
808                                 req->seg[j].gref,
809                                 info->xbdev->otherend_id,
810                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
811                                 rq_data_dir(
812                                         (struct request *)
813                                         info->shadow[req->id].request));
814                 info->shadow[req->id].req = *req;
815
816                 info->ring.req_prod_pvt++;
817         }
818
819         kfree(copy);
820
821         xenbus_switch_state(info->xbdev, XenbusStateConnected);
822
823         spin_lock_irq(&blkif_io_lock);
824
825         /* Now safe for us to use the shared ring */
826         info->connected = BLKIF_STATE_CONNECTED;
827
828         /* Send off requeued requests */
829         flush_requests(info);
830
831         /* Kick any other new requests queued since we resumed */
832         kick_pending_request_queues(info);
833
834         spin_unlock_irq(&blkif_io_lock);
835
836         return 0;
837 }
838
839 /**
840  * We are reconnecting to the backend, due to a suspend/resume, or a backend
841  * driver restart.  We tear down our blkif structure and recreate it, but
842  * leave the device-layer structures intact so that this is transparent to the
843  * rest of the kernel.
844  */
845 static int blkfront_resume(struct xenbus_device *dev)
846 {
847         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
848         int err;
849
850         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
851
852         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
853
854         err = talk_to_backend(dev, info);
855         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
856                 err = blkif_recover(info);
857
858         return err;
859 }
860
861
862 /*
863  * Invoked when the backend is finally 'ready' (and has told produced
864  * the details about the physical device - #sectors, size, etc).
865  */
866 static void blkfront_connect(struct blkfront_info *info)
867 {
868         unsigned long long sectors;
869         unsigned long sector_size;
870         unsigned int binfo;
871         int err;
872
873         if ((info->connected == BLKIF_STATE_CONNECTED) ||
874             (info->connected == BLKIF_STATE_SUSPENDED) )
875                 return;
876
877         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
878                 __func__, info->xbdev->otherend);
879
880         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
881                             "sectors", "%llu", &sectors,
882                             "info", "%u", &binfo,
883                             "sector-size", "%lu", &sector_size,
884                             NULL);
885         if (err) {
886                 xenbus_dev_fatal(info->xbdev, err,
887                                  "reading backend fields at %s",
888                                  info->xbdev->otherend);
889                 return;
890         }
891
892         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
893                             "feature-barrier", "%lu", &info->feature_barrier,
894                             NULL);
895         if (err)
896                 info->feature_barrier = 0;
897
898         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
899         if (err) {
900                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
901                                  info->xbdev->otherend);
902                 return;
903         }
904
905         xenbus_switch_state(info->xbdev, XenbusStateConnected);
906
907         /* Kick pending requests. */
908         spin_lock_irq(&blkif_io_lock);
909         info->connected = BLKIF_STATE_CONNECTED;
910         kick_pending_request_queues(info);
911         spin_unlock_irq(&blkif_io_lock);
912
913         add_disk(info->gd);
914
915         info->is_ready = 1;
916 }
917
918 /**
919  * Handle the change of state of the backend to Closing.  We must delete our
920  * device-layer structures now, to ensure that writes are flushed through to
921  * the backend.  Once is this done, we can switch to Closed in
922  * acknowledgement.
923  */
924 static void blkfront_closing(struct xenbus_device *dev)
925 {
926         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
927         unsigned long flags;
928
929         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
930
931         if (info->rq == NULL)
932                 goto out;
933
934         spin_lock_irqsave(&blkif_io_lock, flags);
935
936         /* No more blkif_request(). */
937         blk_stop_queue(info->rq);
938
939         /* No more gnttab callback work. */
940         gnttab_cancel_free_callback(&info->callback);
941         spin_unlock_irqrestore(&blkif_io_lock, flags);
942
943         /* Flush gnttab callback work. Must be done with no locks held. */
944         flush_scheduled_work();
945
946         blk_cleanup_queue(info->rq);
947         info->rq = NULL;
948
949         del_gendisk(info->gd);
950
951  out:
952         xenbus_frontend_closed(dev);
953 }
954
955 /**
956  * Callback received when the backend's state changes.
957  */
958 static void backend_changed(struct xenbus_device *dev,
959                             enum xenbus_state backend_state)
960 {
961         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
962         struct block_device *bd;
963
964         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
965
966         switch (backend_state) {
967         case XenbusStateInitialising:
968         case XenbusStateInitWait:
969         case XenbusStateInitialised:
970         case XenbusStateUnknown:
971         case XenbusStateClosed:
972                 break;
973
974         case XenbusStateConnected:
975                 blkfront_connect(info);
976                 break;
977
978         case XenbusStateClosing:
979                 if (info->gd == NULL) {
980                         xenbus_frontend_closed(dev);
981                         break;
982                 }
983                 bd = bdget_disk(info->gd, 0);
984                 if (bd == NULL)
985                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
986
987                 mutex_lock(&bd->bd_mutex);
988                 if (info->users > 0)
989                         xenbus_dev_error(dev, -EBUSY,
990                                          "Device in use; refusing to close");
991                 else
992                         blkfront_closing(dev);
993                 mutex_unlock(&bd->bd_mutex);
994                 bdput(bd);
995                 break;
996         }
997 }
998
999 static int blkfront_remove(struct xenbus_device *dev)
1000 {
1001         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1002
1003         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1004
1005         blkif_free(info, 0);
1006
1007         kfree(info);
1008
1009         return 0;
1010 }
1011
1012 static int blkfront_is_ready(struct xenbus_device *dev)
1013 {
1014         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1015
1016         return info->is_ready;
1017 }
1018
1019 static int blkif_open(struct block_device *bdev, fmode_t mode)
1020 {
1021         struct blkfront_info *info = bdev->bd_disk->private_data;
1022         info->users++;
1023         return 0;
1024 }
1025
1026 static int blkif_release(struct gendisk *disk, fmode_t mode)
1027 {
1028         struct blkfront_info *info = disk->private_data;
1029         info->users--;
1030         if (info->users == 0) {
1031                 /* Check whether we have been instructed to close.  We will
1032                    have ignored this request initially, as the device was
1033                    still mounted. */
1034                 struct xenbus_device *dev = info->xbdev;
1035                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1036
1037                 if (state == XenbusStateClosing && info->is_ready)
1038                         blkfront_closing(dev);
1039         }
1040         return 0;
1041 }
1042
1043 static const struct block_device_operations xlvbd_block_fops =
1044 {
1045         .owner = THIS_MODULE,
1046         .open = blkif_open,
1047         .release = blkif_release,
1048         .getgeo = blkif_getgeo,
1049         .locked_ioctl = blkif_ioctl,
1050 };
1051
1052
1053 static struct xenbus_device_id blkfront_ids[] = {
1054         { "vbd" },
1055         { "" }
1056 };
1057
1058 static struct xenbus_driver blkfront = {
1059         .name = "vbd",
1060         .owner = THIS_MODULE,
1061         .ids = blkfront_ids,
1062         .probe = blkfront_probe,
1063         .remove = blkfront_remove,
1064         .resume = blkfront_resume,
1065         .otherend_changed = backend_changed,
1066         .is_ready = blkfront_is_ready,
1067 };
1068
1069 static int __init xlblk_init(void)
1070 {
1071         if (!xen_domain())
1072                 return -ENODEV;
1073
1074         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1075                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1076                        XENVBD_MAJOR, DEV_NAME);
1077                 return -ENODEV;
1078         }
1079
1080         return xenbus_register_frontend(&blkfront);
1081 }
1082 module_init(xlblk_init);
1083
1084
1085 static void __exit xlblk_exit(void)
1086 {
1087         return xenbus_unregister_driver(&blkfront);
1088 }
1089 module_exit(xlblk_exit);
1090
1091 MODULE_DESCRIPTION("Xen virtual block device frontend");
1092 MODULE_LICENSE("GPL");
1093 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1094 MODULE_ALIAS("xen:vbd");
1095 MODULE_ALIAS("xenblk");