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