Merge branches 'x86/cleanups', 'x86/cpu', 'x86/debug', 'x86/mce2', 'x86/mm', 'x86...
[linux-2.6] / 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/xenbus.h>
46 #include <xen/grant_table.h>
47 #include <xen/events.h>
48 #include <xen/page.h>
49
50 #include <xen/interface/grant_table.h>
51 #include <xen/interface/io/blkif.h>
52 #include <xen/interface/io/protocols.h>
53
54 #include <asm/xen/hypervisor.h>
55
56 enum blkif_state {
57         BLKIF_STATE_DISCONNECTED,
58         BLKIF_STATE_CONNECTED,
59         BLKIF_STATE_SUSPENDED,
60 };
61
62 struct blk_shadow {
63         struct blkif_request req;
64         unsigned long request;
65         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
66 };
67
68 static struct block_device_operations xlvbd_block_fops;
69
70 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
71
72 /*
73  * We have one of these per vbd, whether ide, scsi or 'other'.  They
74  * hang in private_data off the gendisk structure. We may end up
75  * putting all kinds of interesting stuff here :-)
76  */
77 struct blkfront_info
78 {
79         struct xenbus_device *xbdev;
80         struct gendisk *gd;
81         int vdevice;
82         blkif_vdev_t handle;
83         enum blkif_state connected;
84         int ring_ref;
85         struct blkif_front_ring ring;
86         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
87         unsigned int evtchn, irq;
88         struct request_queue *rq;
89         struct work_struct work;
90         struct gnttab_free_callback callback;
91         struct blk_shadow shadow[BLK_RING_SIZE];
92         unsigned long shadow_free;
93         int feature_barrier;
94         int is_ready;
95
96         /**
97          * The number of people holding this device open.  We won't allow a
98          * hot-unplug unless this is 0.
99          */
100         int users;
101 };
102
103 static DEFINE_SPINLOCK(blkif_io_lock);
104
105 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
106         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
107 #define GRANT_INVALID_REF       0
108
109 #define PARTS_PER_DISK          16
110 #define PARTS_PER_EXT_DISK      256
111
112 #define BLKIF_MAJOR(dev) ((dev)>>8)
113 #define BLKIF_MINOR(dev) ((dev) & 0xff)
114
115 #define EXT_SHIFT 28
116 #define EXTENDED (1<<EXT_SHIFT)
117 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
118 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
119
120 #define DEV_NAME        "xvd"   /* name in /dev */
121
122 static int get_id_from_freelist(struct blkfront_info *info)
123 {
124         unsigned long free = info->shadow_free;
125         BUG_ON(free > BLK_RING_SIZE);
126         info->shadow_free = info->shadow[free].req.id;
127         info->shadow[free].req.id = 0x0fffffee; /* debug */
128         return free;
129 }
130
131 static void add_id_to_freelist(struct blkfront_info *info,
132                                unsigned long id)
133 {
134         info->shadow[id].req.id  = info->shadow_free;
135         info->shadow[id].request = 0;
136         info->shadow_free = id;
137 }
138
139 static void blkif_restart_queue_callback(void *arg)
140 {
141         struct blkfront_info *info = (struct blkfront_info *)arg;
142         schedule_work(&info->work);
143 }
144
145 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
146 {
147         /* We don't have real geometry info, but let's at least return
148            values consistent with the size of the device */
149         sector_t nsect = get_capacity(bd->bd_disk);
150         sector_t cylinders = nsect;
151
152         hg->heads = 0xff;
153         hg->sectors = 0x3f;
154         sector_div(cylinders, hg->heads * hg->sectors);
155         hg->cylinders = cylinders;
156         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
157                 hg->cylinders = 0xffff;
158         return 0;
159 }
160
161 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
162                        unsigned command, unsigned long argument)
163 {
164         struct blkfront_info *info = bdev->bd_disk->private_data;
165         int i;
166
167         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
168                 command, (long)argument);
169
170         switch (command) {
171         case CDROMMULTISESSION:
172                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
173                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
174                         if (put_user(0, (char __user *)(argument + i)))
175                                 return -EFAULT;
176                 return 0;
177
178         case CDROM_GET_CAPABILITY: {
179                 struct gendisk *gd = info->gd;
180                 if (gd->flags & GENHD_FL_CD)
181                         return 0;
182                 return -EINVAL;
183         }
184
185         default:
186                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
187                   command);*/
188                 return -EINVAL; /* same return as native Linux */
189         }
190
191         return 0;
192 }
193
194 /*
195  * blkif_queue_request
196  *
197  * request block io
198  *
199  * id: for guest use only.
200  * operation: BLKIF_OP_{READ,WRITE,PROBE}
201  * buffer: buffer to read/write into. this should be a
202  *   virtual address in the guest os.
203  */
204 static int blkif_queue_request(struct request *req)
205 {
206         struct blkfront_info *info = req->rq_disk->private_data;
207         unsigned long buffer_mfn;
208         struct blkif_request *ring_req;
209         unsigned long id;
210         unsigned int fsect, lsect;
211         int i, ref;
212         grant_ref_t gref_head;
213         struct scatterlist *sg;
214
215         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
216                 return 1;
217
218         if (gnttab_alloc_grant_references(
219                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
220                 gnttab_request_free_callback(
221                         &info->callback,
222                         blkif_restart_queue_callback,
223                         info,
224                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
225                 return 1;
226         }
227
228         /* Fill out a communications ring structure. */
229         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
230         id = get_id_from_freelist(info);
231         info->shadow[id].request = (unsigned long)req;
232
233         ring_req->id = id;
234         ring_req->sector_number = (blkif_sector_t)req->sector;
235         ring_req->handle = info->handle;
236
237         ring_req->operation = rq_data_dir(req) ?
238                 BLKIF_OP_WRITE : BLKIF_OP_READ;
239         if (blk_barrier_rq(req))
240                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
241
242         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
243         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
244
245         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
246                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
247                 fsect = sg->offset >> 9;
248                 lsect = fsect + (sg->length >> 9) - 1;
249                 /* install a grant reference. */
250                 ref = gnttab_claim_grant_reference(&gref_head);
251                 BUG_ON(ref == -ENOSPC);
252
253                 gnttab_grant_foreign_access_ref(
254                                 ref,
255                                 info->xbdev->otherend_id,
256                                 buffer_mfn,
257                                 rq_data_dir(req) );
258
259                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
260                 ring_req->seg[i] =
261                                 (struct blkif_request_segment) {
262                                         .gref       = ref,
263                                         .first_sect = fsect,
264                                         .last_sect  = lsect };
265         }
266
267         info->ring.req_prod_pvt++;
268
269         /* Keep a private copy so we can reissue requests when recovering. */
270         info->shadow[id].req = *ring_req;
271
272         gnttab_free_grant_references(gref_head);
273
274         return 0;
275 }
276
277
278 static inline void flush_requests(struct blkfront_info *info)
279 {
280         int notify;
281
282         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
283
284         if (notify)
285                 notify_remote_via_irq(info->irq);
286 }
287
288 /*
289  * do_blkif_request
290  *  read a block; request is in a request queue
291  */
292 static void do_blkif_request(struct request_queue *rq)
293 {
294         struct blkfront_info *info = NULL;
295         struct request *req;
296         int queued;
297
298         pr_debug("Entered do_blkif_request\n");
299
300         queued = 0;
301
302         while ((req = elv_next_request(rq)) != NULL) {
303                 info = req->rq_disk->private_data;
304                 if (!blk_fs_request(req)) {
305                         end_request(req, 0);
306                         continue;
307                 }
308
309                 if (RING_FULL(&info->ring))
310                         goto wait;
311
312                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
313                          "(%u/%li) buffer:%p [%s]\n",
314                          req, req->cmd, (unsigned long)req->sector,
315                          req->current_nr_sectors,
316                          req->nr_sectors, req->buffer,
317                          rq_data_dir(req) ? "write" : "read");
318
319
320                 blkdev_dequeue_request(req);
321                 if (blkif_queue_request(req)) {
322                         blk_requeue_request(rq, req);
323 wait:
324                         /* Avoid pointless unplugs. */
325                         blk_stop_queue(rq);
326                         break;
327                 }
328
329                 queued++;
330         }
331
332         if (queued != 0)
333                 flush_requests(info);
334 }
335
336 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
337 {
338         struct request_queue *rq;
339
340         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
341         if (rq == NULL)
342                 return -1;
343
344         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
345
346         /* Hard sector size and max sectors impersonate the equiv. hardware. */
347         blk_queue_hardsect_size(rq, sector_size);
348         blk_queue_max_sectors(rq, 512);
349
350         /* Each segment in a request is up to an aligned page in size. */
351         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
352         blk_queue_max_segment_size(rq, PAGE_SIZE);
353
354         /* Ensure a merged request will fit in a single I/O ring slot. */
355         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
356         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
357
358         /* Make sure buffer addresses are sector-aligned. */
359         blk_queue_dma_alignment(rq, 511);
360
361         /* Make sure we don't use bounce buffers. */
362         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
363
364         gd->queue = rq;
365
366         return 0;
367 }
368
369
370 static int xlvbd_barrier(struct blkfront_info *info)
371 {
372         int err;
373
374         err = blk_queue_ordered(info->rq,
375                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
376                                 NULL);
377
378         if (err)
379                 return err;
380
381         printk(KERN_INFO "blkfront: %s: barriers %s\n",
382                info->gd->disk_name,
383                info->feature_barrier ? "enabled" : "disabled");
384         return 0;
385 }
386
387
388 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
389                                struct blkfront_info *info,
390                                u16 vdisk_info, u16 sector_size)
391 {
392         struct gendisk *gd;
393         int nr_minors = 1;
394         int err = -ENODEV;
395         unsigned int offset;
396         int minor;
397         int nr_parts;
398
399         BUG_ON(info->gd != NULL);
400         BUG_ON(info->rq != NULL);
401
402         if ((info->vdevice>>EXT_SHIFT) > 1) {
403                 /* this is above the extended range; something is wrong */
404                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
405                 return -ENODEV;
406         }
407
408         if (!VDEV_IS_EXTENDED(info->vdevice)) {
409                 minor = BLKIF_MINOR(info->vdevice);
410                 nr_parts = PARTS_PER_DISK;
411         } else {
412                 minor = BLKIF_MINOR_EXT(info->vdevice);
413                 nr_parts = PARTS_PER_EXT_DISK;
414         }
415
416         if ((minor % nr_parts) == 0)
417                 nr_minors = nr_parts;
418
419         gd = alloc_disk(nr_minors);
420         if (gd == NULL)
421                 goto out;
422
423         offset = minor / nr_parts;
424
425         if (nr_minors > 1) {
426                 if (offset < 26)
427                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
428                 else
429                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
430                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
431         } else {
432                 if (offset < 26)
433                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
434                                 'a' + offset,
435                                 minor & (nr_parts - 1));
436                 else
437                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
438                                 'a' + ((offset / 26) - 1),
439                                 'a' + (offset % 26),
440                                 minor & (nr_parts - 1));
441         }
442
443         gd->major = XENVBD_MAJOR;
444         gd->first_minor = minor;
445         gd->fops = &xlvbd_block_fops;
446         gd->private_data = info;
447         gd->driverfs_dev = &(info->xbdev->dev);
448         set_capacity(gd, capacity);
449
450         if (xlvbd_init_blk_queue(gd, sector_size)) {
451                 del_gendisk(gd);
452                 goto out;
453         }
454
455         info->rq = gd->queue;
456         info->gd = gd;
457
458         if (info->feature_barrier)
459                 xlvbd_barrier(info);
460
461         if (vdisk_info & VDISK_READONLY)
462                 set_disk_ro(gd, 1);
463
464         if (vdisk_info & VDISK_REMOVABLE)
465                 gd->flags |= GENHD_FL_REMOVABLE;
466
467         if (vdisk_info & VDISK_CDROM)
468                 gd->flags |= GENHD_FL_CD;
469
470         return 0;
471
472  out:
473         return err;
474 }
475
476 static void kick_pending_request_queues(struct blkfront_info *info)
477 {
478         if (!RING_FULL(&info->ring)) {
479                 /* Re-enable calldowns. */
480                 blk_start_queue(info->rq);
481                 /* Kick things off immediately. */
482                 do_blkif_request(info->rq);
483         }
484 }
485
486 static void blkif_restart_queue(struct work_struct *work)
487 {
488         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
489
490         spin_lock_irq(&blkif_io_lock);
491         if (info->connected == BLKIF_STATE_CONNECTED)
492                 kick_pending_request_queues(info);
493         spin_unlock_irq(&blkif_io_lock);
494 }
495
496 static void blkif_free(struct blkfront_info *info, int suspend)
497 {
498         /* Prevent new requests being issued until we fix things up. */
499         spin_lock_irq(&blkif_io_lock);
500         info->connected = suspend ?
501                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
502         /* No more blkif_request(). */
503         if (info->rq)
504                 blk_stop_queue(info->rq);
505         /* No more gnttab callback work. */
506         gnttab_cancel_free_callback(&info->callback);
507         spin_unlock_irq(&blkif_io_lock);
508
509         /* Flush gnttab callback work. Must be done with no locks held. */
510         flush_scheduled_work();
511
512         /* Free resources associated with old device channel. */
513         if (info->ring_ref != GRANT_INVALID_REF) {
514                 gnttab_end_foreign_access(info->ring_ref, 0,
515                                           (unsigned long)info->ring.sring);
516                 info->ring_ref = GRANT_INVALID_REF;
517                 info->ring.sring = NULL;
518         }
519         if (info->irq)
520                 unbind_from_irqhandler(info->irq, info);
521         info->evtchn = info->irq = 0;
522
523 }
524
525 static void blkif_completion(struct blk_shadow *s)
526 {
527         int i;
528         for (i = 0; i < s->req.nr_segments; i++)
529                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
530 }
531
532 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
533 {
534         struct request *req;
535         struct blkif_response *bret;
536         RING_IDX i, rp;
537         unsigned long flags;
538         struct blkfront_info *info = (struct blkfront_info *)dev_id;
539         int error;
540
541         spin_lock_irqsave(&blkif_io_lock, flags);
542
543         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
544                 spin_unlock_irqrestore(&blkif_io_lock, flags);
545                 return IRQ_HANDLED;
546         }
547
548  again:
549         rp = info->ring.sring->rsp_prod;
550         rmb(); /* Ensure we see queued responses up to 'rp'. */
551
552         for (i = info->ring.rsp_cons; i != rp; i++) {
553                 unsigned long id;
554                 int ret;
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                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
582                         BUG_ON(ret);
583                         break;
584                 default:
585                         BUG();
586                 }
587         }
588
589         info->ring.rsp_cons = i;
590
591         if (i != info->ring.req_prod_pvt) {
592                 int more_to_do;
593                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
594                 if (more_to_do)
595                         goto again;
596         } else
597                 info->ring.sring->rsp_event = i + 1;
598
599         kick_pending_request_queues(info);
600
601         spin_unlock_irqrestore(&blkif_io_lock, flags);
602
603         return IRQ_HANDLED;
604 }
605
606
607 static int setup_blkring(struct xenbus_device *dev,
608                          struct blkfront_info *info)
609 {
610         struct blkif_sring *sring;
611         int err;
612
613         info->ring_ref = GRANT_INVALID_REF;
614
615         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
616         if (!sring) {
617                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
618                 return -ENOMEM;
619         }
620         SHARED_RING_INIT(sring);
621         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
622
623         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
624
625         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
626         if (err < 0) {
627                 free_page((unsigned long)sring);
628                 info->ring.sring = NULL;
629                 goto fail;
630         }
631         info->ring_ref = err;
632
633         err = xenbus_alloc_evtchn(dev, &info->evtchn);
634         if (err)
635                 goto fail;
636
637         err = bind_evtchn_to_irqhandler(info->evtchn,
638                                         blkif_interrupt,
639                                         IRQF_SAMPLE_RANDOM, "blkif", info);
640         if (err <= 0) {
641                 xenbus_dev_fatal(dev, err,
642                                  "bind_evtchn_to_irqhandler failed");
643                 goto fail;
644         }
645         info->irq = err;
646
647         return 0;
648 fail:
649         blkif_free(info, 0);
650         return err;
651 }
652
653
654 /* Common code used when first setting up, and when resuming. */
655 static int talk_to_backend(struct xenbus_device *dev,
656                            struct blkfront_info *info)
657 {
658         const char *message = NULL;
659         struct xenbus_transaction xbt;
660         int err;
661
662         /* Create shared ring, alloc event channel. */
663         err = setup_blkring(dev, info);
664         if (err)
665                 goto out;
666
667 again:
668         err = xenbus_transaction_start(&xbt);
669         if (err) {
670                 xenbus_dev_fatal(dev, err, "starting transaction");
671                 goto destroy_blkring;
672         }
673
674         err = xenbus_printf(xbt, dev->nodename,
675                             "ring-ref", "%u", info->ring_ref);
676         if (err) {
677                 message = "writing ring-ref";
678                 goto abort_transaction;
679         }
680         err = xenbus_printf(xbt, dev->nodename,
681                             "event-channel", "%u", info->evtchn);
682         if (err) {
683                 message = "writing event-channel";
684                 goto abort_transaction;
685         }
686         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
687                             XEN_IO_PROTO_ABI_NATIVE);
688         if (err) {
689                 message = "writing protocol";
690                 goto abort_transaction;
691         }
692
693         err = xenbus_transaction_end(xbt, 0);
694         if (err) {
695                 if (err == -EAGAIN)
696                         goto again;
697                 xenbus_dev_fatal(dev, err, "completing transaction");
698                 goto destroy_blkring;
699         }
700
701         xenbus_switch_state(dev, XenbusStateInitialised);
702
703         return 0;
704
705  abort_transaction:
706         xenbus_transaction_end(xbt, 1);
707         if (message)
708                 xenbus_dev_fatal(dev, err, "%s", message);
709  destroy_blkring:
710         blkif_free(info, 0);
711  out:
712         return err;
713 }
714
715
716 /**
717  * Entry point to this code when a new device is created.  Allocate the basic
718  * structures and the ring buffer for communication with the backend, and
719  * inform the backend of the appropriate details for those.  Switch to
720  * Initialised state.
721  */
722 static int blkfront_probe(struct xenbus_device *dev,
723                           const struct xenbus_device_id *id)
724 {
725         int err, vdevice, i;
726         struct blkfront_info *info;
727
728         /* FIXME: Use dynamic device id if this is not set. */
729         err = xenbus_scanf(XBT_NIL, dev->nodename,
730                            "virtual-device", "%i", &vdevice);
731         if (err != 1) {
732                 /* go looking in the extended area instead */
733                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
734                                    "%i", &vdevice);
735                 if (err != 1) {
736                         xenbus_dev_fatal(dev, err, "reading virtual-device");
737                         return err;
738                 }
739         }
740
741         info = kzalloc(sizeof(*info), GFP_KERNEL);
742         if (!info) {
743                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
744                 return -ENOMEM;
745         }
746
747         info->xbdev = dev;
748         info->vdevice = vdevice;
749         info->connected = BLKIF_STATE_DISCONNECTED;
750         INIT_WORK(&info->work, blkif_restart_queue);
751
752         for (i = 0; i < BLK_RING_SIZE; i++)
753                 info->shadow[i].req.id = i+1;
754         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
755
756         /* Front end dir is a number, which is used as the id. */
757         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
758         dev->dev.driver_data = info;
759
760         err = talk_to_backend(dev, info);
761         if (err) {
762                 kfree(info);
763                 dev->dev.driver_data = NULL;
764                 return err;
765         }
766
767         return 0;
768 }
769
770
771 static int blkif_recover(struct blkfront_info *info)
772 {
773         int i;
774         struct blkif_request *req;
775         struct blk_shadow *copy;
776         int j;
777
778         /* Stage 1: Make a safe copy of the shadow state. */
779         copy = kmalloc(sizeof(info->shadow),
780                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
781         if (!copy)
782                 return -ENOMEM;
783         memcpy(copy, info->shadow, sizeof(info->shadow));
784
785         /* Stage 2: Set up free list. */
786         memset(&info->shadow, 0, sizeof(info->shadow));
787         for (i = 0; i < BLK_RING_SIZE; i++)
788                 info->shadow[i].req.id = i+1;
789         info->shadow_free = info->ring.req_prod_pvt;
790         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
791
792         /* Stage 3: Find pending requests and requeue them. */
793         for (i = 0; i < BLK_RING_SIZE; i++) {
794                 /* Not in use? */
795                 if (copy[i].request == 0)
796                         continue;
797
798                 /* Grab a request slot and copy shadow state into it. */
799                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
800                 *req = copy[i].req;
801
802                 /* We get a new request id, and must reset the shadow state. */
803                 req->id = get_id_from_freelist(info);
804                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
805
806                 /* Rewrite any grant references invalidated by susp/resume. */
807                 for (j = 0; j < req->nr_segments; j++)
808                         gnttab_grant_foreign_access_ref(
809                                 req->seg[j].gref,
810                                 info->xbdev->otherend_id,
811                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
812                                 rq_data_dir(
813                                         (struct request *)
814                                         info->shadow[req->id].request));
815                 info->shadow[req->id].req = *req;
816
817                 info->ring.req_prod_pvt++;
818         }
819
820         kfree(copy);
821
822         xenbus_switch_state(info->xbdev, XenbusStateConnected);
823
824         spin_lock_irq(&blkif_io_lock);
825
826         /* Now safe for us to use the shared ring */
827         info->connected = BLKIF_STATE_CONNECTED;
828
829         /* Send off requeued requests */
830         flush_requests(info);
831
832         /* Kick any other new requests queued since we resumed */
833         kick_pending_request_queues(info);
834
835         spin_unlock_irq(&blkif_io_lock);
836
837         return 0;
838 }
839
840 /**
841  * We are reconnecting to the backend, due to a suspend/resume, or a backend
842  * driver restart.  We tear down our blkif structure and recreate it, but
843  * leave the device-layer structures intact so that this is transparent to the
844  * rest of the kernel.
845  */
846 static int blkfront_resume(struct xenbus_device *dev)
847 {
848         struct blkfront_info *info = dev->dev.driver_data;
849         int err;
850
851         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
852
853         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
854
855         err = talk_to_backend(dev, info);
856         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
857                 err = blkif_recover(info);
858
859         return err;
860 }
861
862
863 /*
864  * Invoked when the backend is finally 'ready' (and has told produced
865  * the details about the physical device - #sectors, size, etc).
866  */
867 static void blkfront_connect(struct blkfront_info *info)
868 {
869         unsigned long long sectors;
870         unsigned long sector_size;
871         unsigned int binfo;
872         int err;
873
874         if ((info->connected == BLKIF_STATE_CONNECTED) ||
875             (info->connected == BLKIF_STATE_SUSPENDED) )
876                 return;
877
878         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
879                 __func__, info->xbdev->otherend);
880
881         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
882                             "sectors", "%llu", &sectors,
883                             "info", "%u", &binfo,
884                             "sector-size", "%lu", &sector_size,
885                             NULL);
886         if (err) {
887                 xenbus_dev_fatal(info->xbdev, err,
888                                  "reading backend fields at %s",
889                                  info->xbdev->otherend);
890                 return;
891         }
892
893         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
894                             "feature-barrier", "%lu", &info->feature_barrier,
895                             NULL);
896         if (err)
897                 info->feature_barrier = 0;
898
899         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
900         if (err) {
901                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
902                                  info->xbdev->otherend);
903                 return;
904         }
905
906         xenbus_switch_state(info->xbdev, XenbusStateConnected);
907
908         /* Kick pending requests. */
909         spin_lock_irq(&blkif_io_lock);
910         info->connected = BLKIF_STATE_CONNECTED;
911         kick_pending_request_queues(info);
912         spin_unlock_irq(&blkif_io_lock);
913
914         add_disk(info->gd);
915
916         info->is_ready = 1;
917 }
918
919 /**
920  * Handle the change of state of the backend to Closing.  We must delete our
921  * device-layer structures now, to ensure that writes are flushed through to
922  * the backend.  Once is this done, we can switch to Closed in
923  * acknowledgement.
924  */
925 static void blkfront_closing(struct xenbus_device *dev)
926 {
927         struct blkfront_info *info = dev->dev.driver_data;
928         unsigned long flags;
929
930         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
931
932         if (info->rq == NULL)
933                 goto out;
934
935         spin_lock_irqsave(&blkif_io_lock, flags);
936
937         del_gendisk(info->gd);
938
939         /* No more blkif_request(). */
940         blk_stop_queue(info->rq);
941
942         /* No more gnttab callback work. */
943         gnttab_cancel_free_callback(&info->callback);
944         spin_unlock_irqrestore(&blkif_io_lock, flags);
945
946         /* Flush gnttab callback work. Must be done with no locks held. */
947         flush_scheduled_work();
948
949         blk_cleanup_queue(info->rq);
950         info->rq = NULL;
951
952  out:
953         xenbus_frontend_closed(dev);
954 }
955
956 /**
957  * Callback received when the backend's state changes.
958  */
959 static void backend_changed(struct xenbus_device *dev,
960                             enum xenbus_state backend_state)
961 {
962         struct blkfront_info *info = dev->dev.driver_data;
963         struct block_device *bd;
964
965         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
966
967         switch (backend_state) {
968         case XenbusStateInitialising:
969         case XenbusStateInitWait:
970         case XenbusStateInitialised:
971         case XenbusStateUnknown:
972         case XenbusStateClosed:
973                 break;
974
975         case XenbusStateConnected:
976                 blkfront_connect(info);
977                 break;
978
979         case XenbusStateClosing:
980                 if (info->gd == NULL)
981                         xenbus_dev_fatal(dev, -ENODEV, "gd is NULL");
982                 bd = bdget_disk(info->gd, 0);
983                 if (bd == NULL)
984                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
985
986                 mutex_lock(&bd->bd_mutex);
987                 if (info->users > 0)
988                         xenbus_dev_error(dev, -EBUSY,
989                                          "Device in use; refusing to close");
990                 else
991                         blkfront_closing(dev);
992                 mutex_unlock(&bd->bd_mutex);
993                 bdput(bd);
994                 break;
995         }
996 }
997
998 static int blkfront_remove(struct xenbus_device *dev)
999 {
1000         struct blkfront_info *info = dev->dev.driver_data;
1001
1002         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1003
1004         blkif_free(info, 0);
1005
1006         kfree(info);
1007
1008         return 0;
1009 }
1010
1011 static int blkfront_is_ready(struct xenbus_device *dev)
1012 {
1013         struct blkfront_info *info = dev->dev.driver_data;
1014
1015         return info->is_ready;
1016 }
1017
1018 static int blkif_open(struct block_device *bdev, fmode_t mode)
1019 {
1020         struct blkfront_info *info = bdev->bd_disk->private_data;
1021         info->users++;
1022         return 0;
1023 }
1024
1025 static int blkif_release(struct gendisk *disk, fmode_t mode)
1026 {
1027         struct blkfront_info *info = disk->private_data;
1028         info->users--;
1029         if (info->users == 0) {
1030                 /* Check whether we have been instructed to close.  We will
1031                    have ignored this request initially, as the device was
1032                    still mounted. */
1033                 struct xenbus_device *dev = info->xbdev;
1034                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1035
1036                 if (state == XenbusStateClosing && info->is_ready)
1037                         blkfront_closing(dev);
1038         }
1039         return 0;
1040 }
1041
1042 static struct block_device_operations xlvbd_block_fops =
1043 {
1044         .owner = THIS_MODULE,
1045         .open = blkif_open,
1046         .release = blkif_release,
1047         .getgeo = blkif_getgeo,
1048         .locked_ioctl = blkif_ioctl,
1049 };
1050
1051
1052 static struct xenbus_device_id blkfront_ids[] = {
1053         { "vbd" },
1054         { "" }
1055 };
1056
1057 static struct xenbus_driver blkfront = {
1058         .name = "vbd",
1059         .owner = THIS_MODULE,
1060         .ids = blkfront_ids,
1061         .probe = blkfront_probe,
1062         .remove = blkfront_remove,
1063         .resume = blkfront_resume,
1064         .otherend_changed = backend_changed,
1065         .is_ready = blkfront_is_ready,
1066 };
1067
1068 static int __init xlblk_init(void)
1069 {
1070         if (!xen_domain())
1071                 return -ENODEV;
1072
1073         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1074                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1075                        XENVBD_MAJOR, DEV_NAME);
1076                 return -ENODEV;
1077         }
1078
1079         return xenbus_register_frontend(&blkfront);
1080 }
1081 module_init(xlblk_init);
1082
1083
1084 static void __exit xlblk_exit(void)
1085 {
1086         return xenbus_unregister_driver(&blkfront);
1087 }
1088 module_exit(xlblk_exit);
1089
1090 MODULE_DESCRIPTION("Xen virtual block device frontend");
1091 MODULE_LICENSE("GPL");
1092 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1093 MODULE_ALIAS("xen:vbd");
1094 MODULE_ALIAS("xenblk");