Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[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_hw_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_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
555                 bret = RING_GET_RESPONSE(&info->ring, i);
556                 id   = bret->id;
557                 req  = (struct request *)info->shadow[id].request;
558
559                 blkif_completion(&info->shadow[id]);
560
561                 add_id_to_freelist(info, id);
562
563                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
564                 switch (bret->operation) {
565                 case BLKIF_OP_WRITE_BARRIER:
566                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
567                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
568                                        info->gd->disk_name);
569                                 error = -EOPNOTSUPP;
570                                 info->feature_barrier = 0;
571                                 xlvbd_barrier(info);
572                         }
573                         /* fall through */
574                 case BLKIF_OP_READ:
575                 case BLKIF_OP_WRITE:
576                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
577                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
578                                         "request: %x\n", bret->status);
579
580                         __blk_end_request_all(req, error);
581                         break;
582                 default:
583                         BUG();
584                 }
585         }
586
587         info->ring.rsp_cons = i;
588
589         if (i != info->ring.req_prod_pvt) {
590                 int more_to_do;
591                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
592                 if (more_to_do)
593                         goto again;
594         } else
595                 info->ring.sring->rsp_event = i + 1;
596
597         kick_pending_request_queues(info);
598
599         spin_unlock_irqrestore(&blkif_io_lock, flags);
600
601         return IRQ_HANDLED;
602 }
603
604
605 static int setup_blkring(struct xenbus_device *dev,
606                          struct blkfront_info *info)
607 {
608         struct blkif_sring *sring;
609         int err;
610
611         info->ring_ref = GRANT_INVALID_REF;
612
613         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
614         if (!sring) {
615                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
616                 return -ENOMEM;
617         }
618         SHARED_RING_INIT(sring);
619         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
620
621         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
622
623         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
624         if (err < 0) {
625                 free_page((unsigned long)sring);
626                 info->ring.sring = NULL;
627                 goto fail;
628         }
629         info->ring_ref = err;
630
631         err = xenbus_alloc_evtchn(dev, &info->evtchn);
632         if (err)
633                 goto fail;
634
635         err = bind_evtchn_to_irqhandler(info->evtchn,
636                                         blkif_interrupt,
637                                         IRQF_SAMPLE_RANDOM, "blkif", info);
638         if (err <= 0) {
639                 xenbus_dev_fatal(dev, err,
640                                  "bind_evtchn_to_irqhandler failed");
641                 goto fail;
642         }
643         info->irq = err;
644
645         return 0;
646 fail:
647         blkif_free(info, 0);
648         return err;
649 }
650
651
652 /* Common code used when first setting up, and when resuming. */
653 static int talk_to_backend(struct xenbus_device *dev,
654                            struct blkfront_info *info)
655 {
656         const char *message = NULL;
657         struct xenbus_transaction xbt;
658         int err;
659
660         /* Create shared ring, alloc event channel. */
661         err = setup_blkring(dev, info);
662         if (err)
663                 goto out;
664
665 again:
666         err = xenbus_transaction_start(&xbt);
667         if (err) {
668                 xenbus_dev_fatal(dev, err, "starting transaction");
669                 goto destroy_blkring;
670         }
671
672         err = xenbus_printf(xbt, dev->nodename,
673                             "ring-ref", "%u", info->ring_ref);
674         if (err) {
675                 message = "writing ring-ref";
676                 goto abort_transaction;
677         }
678         err = xenbus_printf(xbt, dev->nodename,
679                             "event-channel", "%u", info->evtchn);
680         if (err) {
681                 message = "writing event-channel";
682                 goto abort_transaction;
683         }
684         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
685                             XEN_IO_PROTO_ABI_NATIVE);
686         if (err) {
687                 message = "writing protocol";
688                 goto abort_transaction;
689         }
690
691         err = xenbus_transaction_end(xbt, 0);
692         if (err) {
693                 if (err == -EAGAIN)
694                         goto again;
695                 xenbus_dev_fatal(dev, err, "completing transaction");
696                 goto destroy_blkring;
697         }
698
699         xenbus_switch_state(dev, XenbusStateInitialised);
700
701         return 0;
702
703  abort_transaction:
704         xenbus_transaction_end(xbt, 1);
705         if (message)
706                 xenbus_dev_fatal(dev, err, "%s", message);
707  destroy_blkring:
708         blkif_free(info, 0);
709  out:
710         return err;
711 }
712
713
714 /**
715  * Entry point to this code when a new device is created.  Allocate the basic
716  * structures and the ring buffer for communication with the backend, and
717  * inform the backend of the appropriate details for those.  Switch to
718  * Initialised state.
719  */
720 static int blkfront_probe(struct xenbus_device *dev,
721                           const struct xenbus_device_id *id)
722 {
723         int err, vdevice, i;
724         struct blkfront_info *info;
725
726         /* FIXME: Use dynamic device id if this is not set. */
727         err = xenbus_scanf(XBT_NIL, dev->nodename,
728                            "virtual-device", "%i", &vdevice);
729         if (err != 1) {
730                 /* go looking in the extended area instead */
731                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
732                                    "%i", &vdevice);
733                 if (err != 1) {
734                         xenbus_dev_fatal(dev, err, "reading virtual-device");
735                         return err;
736                 }
737         }
738
739         info = kzalloc(sizeof(*info), GFP_KERNEL);
740         if (!info) {
741                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
742                 return -ENOMEM;
743         }
744
745         info->xbdev = dev;
746         info->vdevice = vdevice;
747         info->connected = BLKIF_STATE_DISCONNECTED;
748         INIT_WORK(&info->work, blkif_restart_queue);
749
750         for (i = 0; i < BLK_RING_SIZE; i++)
751                 info->shadow[i].req.id = i+1;
752         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
753
754         /* Front end dir is a number, which is used as the id. */
755         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
756         dev_set_drvdata(&dev->dev, info);
757
758         err = talk_to_backend(dev, info);
759         if (err) {
760                 kfree(info);
761                 dev_set_drvdata(&dev->dev, NULL);
762                 return err;
763         }
764
765         return 0;
766 }
767
768
769 static int blkif_recover(struct blkfront_info *info)
770 {
771         int i;
772         struct blkif_request *req;
773         struct blk_shadow *copy;
774         int j;
775
776         /* Stage 1: Make a safe copy of the shadow state. */
777         copy = kmalloc(sizeof(info->shadow),
778                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
779         if (!copy)
780                 return -ENOMEM;
781         memcpy(copy, info->shadow, sizeof(info->shadow));
782
783         /* Stage 2: Set up free list. */
784         memset(&info->shadow, 0, sizeof(info->shadow));
785         for (i = 0; i < BLK_RING_SIZE; i++)
786                 info->shadow[i].req.id = i+1;
787         info->shadow_free = info->ring.req_prod_pvt;
788         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
789
790         /* Stage 3: Find pending requests and requeue them. */
791         for (i = 0; i < BLK_RING_SIZE; i++) {
792                 /* Not in use? */
793                 if (copy[i].request == 0)
794                         continue;
795
796                 /* Grab a request slot and copy shadow state into it. */
797                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
798                 *req = copy[i].req;
799
800                 /* We get a new request id, and must reset the shadow state. */
801                 req->id = get_id_from_freelist(info);
802                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
803
804                 /* Rewrite any grant references invalidated by susp/resume. */
805                 for (j = 0; j < req->nr_segments; j++)
806                         gnttab_grant_foreign_access_ref(
807                                 req->seg[j].gref,
808                                 info->xbdev->otherend_id,
809                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
810                                 rq_data_dir(
811                                         (struct request *)
812                                         info->shadow[req->id].request));
813                 info->shadow[req->id].req = *req;
814
815                 info->ring.req_prod_pvt++;
816         }
817
818         kfree(copy);
819
820         xenbus_switch_state(info->xbdev, XenbusStateConnected);
821
822         spin_lock_irq(&blkif_io_lock);
823
824         /* Now safe for us to use the shared ring */
825         info->connected = BLKIF_STATE_CONNECTED;
826
827         /* Send off requeued requests */
828         flush_requests(info);
829
830         /* Kick any other new requests queued since we resumed */
831         kick_pending_request_queues(info);
832
833         spin_unlock_irq(&blkif_io_lock);
834
835         return 0;
836 }
837
838 /**
839  * We are reconnecting to the backend, due to a suspend/resume, or a backend
840  * driver restart.  We tear down our blkif structure and recreate it, but
841  * leave the device-layer structures intact so that this is transparent to the
842  * rest of the kernel.
843  */
844 static int blkfront_resume(struct xenbus_device *dev)
845 {
846         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
847         int err;
848
849         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
850
851         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
852
853         err = talk_to_backend(dev, info);
854         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
855                 err = blkif_recover(info);
856
857         return err;
858 }
859
860
861 /*
862  * Invoked when the backend is finally 'ready' (and has told produced
863  * the details about the physical device - #sectors, size, etc).
864  */
865 static void blkfront_connect(struct blkfront_info *info)
866 {
867         unsigned long long sectors;
868         unsigned long sector_size;
869         unsigned int binfo;
870         int err;
871
872         if ((info->connected == BLKIF_STATE_CONNECTED) ||
873             (info->connected == BLKIF_STATE_SUSPENDED) )
874                 return;
875
876         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
877                 __func__, info->xbdev->otherend);
878
879         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
880                             "sectors", "%llu", &sectors,
881                             "info", "%u", &binfo,
882                             "sector-size", "%lu", &sector_size,
883                             NULL);
884         if (err) {
885                 xenbus_dev_fatal(info->xbdev, err,
886                                  "reading backend fields at %s",
887                                  info->xbdev->otherend);
888                 return;
889         }
890
891         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
892                             "feature-barrier", "%lu", &info->feature_barrier,
893                             NULL);
894         if (err)
895                 info->feature_barrier = 0;
896
897         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
898         if (err) {
899                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
900                                  info->xbdev->otherend);
901                 return;
902         }
903
904         xenbus_switch_state(info->xbdev, XenbusStateConnected);
905
906         /* Kick pending requests. */
907         spin_lock_irq(&blkif_io_lock);
908         info->connected = BLKIF_STATE_CONNECTED;
909         kick_pending_request_queues(info);
910         spin_unlock_irq(&blkif_io_lock);
911
912         add_disk(info->gd);
913
914         info->is_ready = 1;
915 }
916
917 /**
918  * Handle the change of state of the backend to Closing.  We must delete our
919  * device-layer structures now, to ensure that writes are flushed through to
920  * the backend.  Once is this done, we can switch to Closed in
921  * acknowledgement.
922  */
923 static void blkfront_closing(struct xenbus_device *dev)
924 {
925         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
926         unsigned long flags;
927
928         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
929
930         if (info->rq == NULL)
931                 goto out;
932
933         spin_lock_irqsave(&blkif_io_lock, flags);
934
935         /* No more blkif_request(). */
936         blk_stop_queue(info->rq);
937
938         /* No more gnttab callback work. */
939         gnttab_cancel_free_callback(&info->callback);
940         spin_unlock_irqrestore(&blkif_io_lock, flags);
941
942         /* Flush gnttab callback work. Must be done with no locks held. */
943         flush_scheduled_work();
944
945         blk_cleanup_queue(info->rq);
946         info->rq = NULL;
947
948         del_gendisk(info->gd);
949
950  out:
951         xenbus_frontend_closed(dev);
952 }
953
954 /**
955  * Callback received when the backend's state changes.
956  */
957 static void backend_changed(struct xenbus_device *dev,
958                             enum xenbus_state backend_state)
959 {
960         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
961         struct block_device *bd;
962
963         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
964
965         switch (backend_state) {
966         case XenbusStateInitialising:
967         case XenbusStateInitWait:
968         case XenbusStateInitialised:
969         case XenbusStateUnknown:
970         case XenbusStateClosed:
971                 break;
972
973         case XenbusStateConnected:
974                 blkfront_connect(info);
975                 break;
976
977         case XenbusStateClosing:
978                 if (info->gd == NULL) {
979                         xenbus_frontend_closed(dev);
980                         break;
981                 }
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_get_drvdata(&dev->dev);
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_get_drvdata(&dev->dev);
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 const 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 const 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");