]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - drivers/block/xen-blkfront.c
blkfront: fixes for 'xm block-detach ... --force'
[~shefty/rdma-dev.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/slab.h>
44 #include <linux/smp_lock.h>
45 #include <linux/scatterlist.h>
46
47 #include <xen/xen.h>
48 #include <xen/xenbus.h>
49 #include <xen/grant_table.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52
53 #include <xen/interface/grant_table.h>
54 #include <xen/interface/io/blkif.h>
55 #include <xen/interface/io/protocols.h>
56
57 #include <asm/xen/hypervisor.h>
58
59 enum blkif_state {
60         BLKIF_STATE_DISCONNECTED,
61         BLKIF_STATE_CONNECTED,
62         BLKIF_STATE_SUSPENDED,
63 };
64
65 struct blk_shadow {
66         struct blkif_request req;
67         unsigned long request;
68         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
69 };
70
71 static const struct block_device_operations xlvbd_block_fops;
72
73 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
74
75 /*
76  * We have one of these per vbd, whether ide, scsi or 'other'.  They
77  * hang in private_data off the gendisk structure. We may end up
78  * putting all kinds of interesting stuff here :-)
79  */
80 struct blkfront_info
81 {
82         struct xenbus_device *xbdev;
83         struct gendisk *gd;
84         int vdevice;
85         blkif_vdev_t handle;
86         enum blkif_state connected;
87         int ring_ref;
88         struct blkif_front_ring ring;
89         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
90         unsigned int evtchn, irq;
91         struct request_queue *rq;
92         struct work_struct work;
93         struct gnttab_free_callback callback;
94         struct blk_shadow shadow[BLK_RING_SIZE];
95         unsigned long shadow_free;
96         int feature_barrier;
97         int is_ready;
98
99         /**
100          * The number of people holding this device open.  We won't allow a
101          * hot-unplug unless this is 0.
102          */
103         int users;
104 };
105
106 static DEFINE_SPINLOCK(blkif_io_lock);
107
108 static unsigned int nr_minors;
109 static unsigned long *minors;
110 static DEFINE_SPINLOCK(minor_lock);
111
112 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
113         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
114 #define GRANT_INVALID_REF       0
115
116 #define PARTS_PER_DISK          16
117 #define PARTS_PER_EXT_DISK      256
118
119 #define BLKIF_MAJOR(dev) ((dev)>>8)
120 #define BLKIF_MINOR(dev) ((dev) & 0xff)
121
122 #define EXT_SHIFT 28
123 #define EXTENDED (1<<EXT_SHIFT)
124 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
125 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
126
127 #define DEV_NAME        "xvd"   /* name in /dev */
128
129 static int get_id_from_freelist(struct blkfront_info *info)
130 {
131         unsigned long free = info->shadow_free;
132         BUG_ON(free >= BLK_RING_SIZE);
133         info->shadow_free = info->shadow[free].req.id;
134         info->shadow[free].req.id = 0x0fffffee; /* debug */
135         return free;
136 }
137
138 static void add_id_to_freelist(struct blkfront_info *info,
139                                unsigned long id)
140 {
141         info->shadow[id].req.id  = info->shadow_free;
142         info->shadow[id].request = 0;
143         info->shadow_free = id;
144 }
145
146 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
147 {
148         unsigned int end = minor + nr;
149         int rc;
150
151         if (end > nr_minors) {
152                 unsigned long *bitmap, *old;
153
154                 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
155                                  GFP_KERNEL);
156                 if (bitmap == NULL)
157                         return -ENOMEM;
158
159                 spin_lock(&minor_lock);
160                 if (end > nr_minors) {
161                         old = minors;
162                         memcpy(bitmap, minors,
163                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
164                         minors = bitmap;
165                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
166                 } else
167                         old = bitmap;
168                 spin_unlock(&minor_lock);
169                 kfree(old);
170         }
171
172         spin_lock(&minor_lock);
173         if (find_next_bit(minors, end, minor) >= end) {
174                 for (; minor < end; ++minor)
175                         __set_bit(minor, minors);
176                 rc = 0;
177         } else
178                 rc = -EBUSY;
179         spin_unlock(&minor_lock);
180
181         return rc;
182 }
183
184 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
185 {
186         unsigned int end = minor + nr;
187
188         BUG_ON(end > nr_minors);
189         spin_lock(&minor_lock);
190         for (; minor < end; ++minor)
191                 __clear_bit(minor, minors);
192         spin_unlock(&minor_lock);
193 }
194
195 static void blkif_restart_queue_callback(void *arg)
196 {
197         struct blkfront_info *info = (struct blkfront_info *)arg;
198         schedule_work(&info->work);
199 }
200
201 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
202 {
203         /* We don't have real geometry info, but let's at least return
204            values consistent with the size of the device */
205         sector_t nsect = get_capacity(bd->bd_disk);
206         sector_t cylinders = nsect;
207
208         hg->heads = 0xff;
209         hg->sectors = 0x3f;
210         sector_div(cylinders, hg->heads * hg->sectors);
211         hg->cylinders = cylinders;
212         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
213                 hg->cylinders = 0xffff;
214         return 0;
215 }
216
217 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
218                        unsigned command, unsigned long argument)
219 {
220         struct blkfront_info *info = bdev->bd_disk->private_data;
221         int i;
222
223         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
224                 command, (long)argument);
225
226         switch (command) {
227         case CDROMMULTISESSION:
228                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
229                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
230                         if (put_user(0, (char __user *)(argument + i)))
231                                 return -EFAULT;
232                 return 0;
233
234         case CDROM_GET_CAPABILITY: {
235                 struct gendisk *gd = info->gd;
236                 if (gd->flags & GENHD_FL_CD)
237                         return 0;
238                 return -EINVAL;
239         }
240
241         default:
242                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
243                   command);*/
244                 return -EINVAL; /* same return as native Linux */
245         }
246
247         return 0;
248 }
249
250 /*
251  * blkif_queue_request
252  *
253  * request block io
254  *
255  * id: for guest use only.
256  * operation: BLKIF_OP_{READ,WRITE,PROBE}
257  * buffer: buffer to read/write into. this should be a
258  *   virtual address in the guest os.
259  */
260 static int blkif_queue_request(struct request *req)
261 {
262         struct blkfront_info *info = req->rq_disk->private_data;
263         unsigned long buffer_mfn;
264         struct blkif_request *ring_req;
265         unsigned long id;
266         unsigned int fsect, lsect;
267         int i, ref;
268         grant_ref_t gref_head;
269         struct scatterlist *sg;
270
271         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
272                 return 1;
273
274         if (gnttab_alloc_grant_references(
275                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
276                 gnttab_request_free_callback(
277                         &info->callback,
278                         blkif_restart_queue_callback,
279                         info,
280                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
281                 return 1;
282         }
283
284         /* Fill out a communications ring structure. */
285         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
286         id = get_id_from_freelist(info);
287         info->shadow[id].request = (unsigned long)req;
288
289         ring_req->id = id;
290         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
291         ring_req->handle = info->handle;
292
293         ring_req->operation = rq_data_dir(req) ?
294                 BLKIF_OP_WRITE : BLKIF_OP_READ;
295         if (req->cmd_flags & REQ_HARDBARRIER)
296                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
297
298         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
299         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
300
301         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
302                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
303                 fsect = sg->offset >> 9;
304                 lsect = fsect + (sg->length >> 9) - 1;
305                 /* install a grant reference. */
306                 ref = gnttab_claim_grant_reference(&gref_head);
307                 BUG_ON(ref == -ENOSPC);
308
309                 gnttab_grant_foreign_access_ref(
310                                 ref,
311                                 info->xbdev->otherend_id,
312                                 buffer_mfn,
313                                 rq_data_dir(req) );
314
315                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
316                 ring_req->seg[i] =
317                                 (struct blkif_request_segment) {
318                                         .gref       = ref,
319                                         .first_sect = fsect,
320                                         .last_sect  = lsect };
321         }
322
323         info->ring.req_prod_pvt++;
324
325         /* Keep a private copy so we can reissue requests when recovering. */
326         info->shadow[id].req = *ring_req;
327
328         gnttab_free_grant_references(gref_head);
329
330         return 0;
331 }
332
333
334 static inline void flush_requests(struct blkfront_info *info)
335 {
336         int notify;
337
338         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
339
340         if (notify)
341                 notify_remote_via_irq(info->irq);
342 }
343
344 /*
345  * do_blkif_request
346  *  read a block; request is in a request queue
347  */
348 static void do_blkif_request(struct request_queue *rq)
349 {
350         struct blkfront_info *info = NULL;
351         struct request *req;
352         int queued;
353
354         pr_debug("Entered do_blkif_request\n");
355
356         queued = 0;
357
358         while ((req = blk_peek_request(rq)) != NULL) {
359                 info = req->rq_disk->private_data;
360
361                 if (RING_FULL(&info->ring))
362                         goto wait;
363
364                 blk_start_request(req);
365
366                 if (req->cmd_type != REQ_TYPE_FS) {
367                         __blk_end_request_all(req, -EIO);
368                         continue;
369                 }
370
371                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
372                          "(%u/%u) buffer:%p [%s]\n",
373                          req, req->cmd, (unsigned long)blk_rq_pos(req),
374                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
375                          req->buffer, rq_data_dir(req) ? "write" : "read");
376
377                 if (blkif_queue_request(req)) {
378                         blk_requeue_request(rq, req);
379 wait:
380                         /* Avoid pointless unplugs. */
381                         blk_stop_queue(rq);
382                         break;
383                 }
384
385                 queued++;
386         }
387
388         if (queued != 0)
389                 flush_requests(info);
390 }
391
392 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
393 {
394         struct request_queue *rq;
395
396         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
397         if (rq == NULL)
398                 return -1;
399
400         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
401
402         /* Hard sector size and max sectors impersonate the equiv. hardware. */
403         blk_queue_logical_block_size(rq, sector_size);
404         blk_queue_max_hw_sectors(rq, 512);
405
406         /* Each segment in a request is up to an aligned page in size. */
407         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
408         blk_queue_max_segment_size(rq, PAGE_SIZE);
409
410         /* Ensure a merged request will fit in a single I/O ring slot. */
411         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
412
413         /* Make sure buffer addresses are sector-aligned. */
414         blk_queue_dma_alignment(rq, 511);
415
416         /* Make sure we don't use bounce buffers. */
417         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
418
419         gd->queue = rq;
420
421         return 0;
422 }
423
424
425 static int xlvbd_barrier(struct blkfront_info *info)
426 {
427         int err;
428
429         err = blk_queue_ordered(info->rq,
430                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE);
431
432         if (err)
433                 return err;
434
435         printk(KERN_INFO "blkfront: %s: barriers %s\n",
436                info->gd->disk_name,
437                info->feature_barrier ? "enabled" : "disabled");
438         return 0;
439 }
440
441
442 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
443                                struct blkfront_info *info,
444                                u16 vdisk_info, u16 sector_size)
445 {
446         struct gendisk *gd;
447         int nr_minors = 1;
448         int err = -ENODEV;
449         unsigned int offset;
450         int minor;
451         int nr_parts;
452
453         BUG_ON(info->gd != NULL);
454         BUG_ON(info->rq != NULL);
455
456         if ((info->vdevice>>EXT_SHIFT) > 1) {
457                 /* this is above the extended range; something is wrong */
458                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
459                 return -ENODEV;
460         }
461
462         if (!VDEV_IS_EXTENDED(info->vdevice)) {
463                 minor = BLKIF_MINOR(info->vdevice);
464                 nr_parts = PARTS_PER_DISK;
465         } else {
466                 minor = BLKIF_MINOR_EXT(info->vdevice);
467                 nr_parts = PARTS_PER_EXT_DISK;
468         }
469
470         if ((minor % nr_parts) == 0)
471                 nr_minors = nr_parts;
472
473         err = xlbd_reserve_minors(minor, nr_minors);
474         if (err)
475                 goto out;
476         err = -ENODEV;
477
478         gd = alloc_disk(nr_minors);
479         if (gd == NULL)
480                 goto release;
481
482         offset = minor / nr_parts;
483
484         if (nr_minors > 1) {
485                 if (offset < 26)
486                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
487                 else
488                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
489                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
490         } else {
491                 if (offset < 26)
492                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
493                                 'a' + offset,
494                                 minor & (nr_parts - 1));
495                 else
496                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
497                                 'a' + ((offset / 26) - 1),
498                                 'a' + (offset % 26),
499                                 minor & (nr_parts - 1));
500         }
501
502         gd->major = XENVBD_MAJOR;
503         gd->first_minor = minor;
504         gd->fops = &xlvbd_block_fops;
505         gd->private_data = info;
506         gd->driverfs_dev = &(info->xbdev->dev);
507         set_capacity(gd, capacity);
508
509         if (xlvbd_init_blk_queue(gd, sector_size)) {
510                 del_gendisk(gd);
511                 goto release;
512         }
513
514         info->rq = gd->queue;
515         info->gd = gd;
516
517         if (info->feature_barrier)
518                 xlvbd_barrier(info);
519
520         if (vdisk_info & VDISK_READONLY)
521                 set_disk_ro(gd, 1);
522
523         if (vdisk_info & VDISK_REMOVABLE)
524                 gd->flags |= GENHD_FL_REMOVABLE;
525
526         if (vdisk_info & VDISK_CDROM)
527                 gd->flags |= GENHD_FL_CD;
528
529         return 0;
530
531  release:
532         xlbd_release_minors(minor, nr_minors);
533  out:
534         return err;
535 }
536
537 static void kick_pending_request_queues(struct blkfront_info *info)
538 {
539         if (!RING_FULL(&info->ring)) {
540                 /* Re-enable calldowns. */
541                 blk_start_queue(info->rq);
542                 /* Kick things off immediately. */
543                 do_blkif_request(info->rq);
544         }
545 }
546
547 static void blkif_restart_queue(struct work_struct *work)
548 {
549         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
550
551         spin_lock_irq(&blkif_io_lock);
552         if (info->connected == BLKIF_STATE_CONNECTED)
553                 kick_pending_request_queues(info);
554         spin_unlock_irq(&blkif_io_lock);
555 }
556
557 static void blkif_free(struct blkfront_info *info, int suspend)
558 {
559         /* Prevent new requests being issued until we fix things up. */
560         spin_lock_irq(&blkif_io_lock);
561         info->connected = suspend ?
562                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
563         /* No more blkif_request(). */
564         if (info->rq)
565                 blk_stop_queue(info->rq);
566         /* No more gnttab callback work. */
567         gnttab_cancel_free_callback(&info->callback);
568         spin_unlock_irq(&blkif_io_lock);
569
570         /* Flush gnttab callback work. Must be done with no locks held. */
571         flush_scheduled_work();
572
573         /* Free resources associated with old device channel. */
574         if (info->ring_ref != GRANT_INVALID_REF) {
575                 gnttab_end_foreign_access(info->ring_ref, 0,
576                                           (unsigned long)info->ring.sring);
577                 info->ring_ref = GRANT_INVALID_REF;
578                 info->ring.sring = NULL;
579         }
580         if (info->irq)
581                 unbind_from_irqhandler(info->irq, info);
582         info->evtchn = info->irq = 0;
583
584 }
585
586 static void blkif_completion(struct blk_shadow *s)
587 {
588         int i;
589         for (i = 0; i < s->req.nr_segments; i++)
590                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
591 }
592
593 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
594 {
595         struct request *req;
596         struct blkif_response *bret;
597         RING_IDX i, rp;
598         unsigned long flags;
599         struct blkfront_info *info = (struct blkfront_info *)dev_id;
600         int error;
601
602         spin_lock_irqsave(&blkif_io_lock, flags);
603
604         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
605                 spin_unlock_irqrestore(&blkif_io_lock, flags);
606                 return IRQ_HANDLED;
607         }
608
609  again:
610         rp = info->ring.sring->rsp_prod;
611         rmb(); /* Ensure we see queued responses up to 'rp'. */
612
613         for (i = info->ring.rsp_cons; i != rp; i++) {
614                 unsigned long id;
615
616                 bret = RING_GET_RESPONSE(&info->ring, i);
617                 id   = bret->id;
618                 req  = (struct request *)info->shadow[id].request;
619
620                 blkif_completion(&info->shadow[id]);
621
622                 add_id_to_freelist(info, id);
623
624                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
625                 switch (bret->operation) {
626                 case BLKIF_OP_WRITE_BARRIER:
627                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
628                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
629                                        info->gd->disk_name);
630                                 error = -EOPNOTSUPP;
631                                 info->feature_barrier = 0;
632                                 xlvbd_barrier(info);
633                         }
634                         /* fall through */
635                 case BLKIF_OP_READ:
636                 case BLKIF_OP_WRITE:
637                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
638                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
639                                         "request: %x\n", bret->status);
640
641                         __blk_end_request_all(req, error);
642                         break;
643                 default:
644                         BUG();
645                 }
646         }
647
648         info->ring.rsp_cons = i;
649
650         if (i != info->ring.req_prod_pvt) {
651                 int more_to_do;
652                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
653                 if (more_to_do)
654                         goto again;
655         } else
656                 info->ring.sring->rsp_event = i + 1;
657
658         kick_pending_request_queues(info);
659
660         spin_unlock_irqrestore(&blkif_io_lock, flags);
661
662         return IRQ_HANDLED;
663 }
664
665
666 static int setup_blkring(struct xenbus_device *dev,
667                          struct blkfront_info *info)
668 {
669         struct blkif_sring *sring;
670         int err;
671
672         info->ring_ref = GRANT_INVALID_REF;
673
674         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
675         if (!sring) {
676                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
677                 return -ENOMEM;
678         }
679         SHARED_RING_INIT(sring);
680         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
681
682         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
683
684         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
685         if (err < 0) {
686                 free_page((unsigned long)sring);
687                 info->ring.sring = NULL;
688                 goto fail;
689         }
690         info->ring_ref = err;
691
692         err = xenbus_alloc_evtchn(dev, &info->evtchn);
693         if (err)
694                 goto fail;
695
696         err = bind_evtchn_to_irqhandler(info->evtchn,
697                                         blkif_interrupt,
698                                         IRQF_SAMPLE_RANDOM, "blkif", info);
699         if (err <= 0) {
700                 xenbus_dev_fatal(dev, err,
701                                  "bind_evtchn_to_irqhandler failed");
702                 goto fail;
703         }
704         info->irq = err;
705
706         return 0;
707 fail:
708         blkif_free(info, 0);
709         return err;
710 }
711
712
713 /* Common code used when first setting up, and when resuming. */
714 static int talk_to_blkback(struct xenbus_device *dev,
715                            struct blkfront_info *info)
716 {
717         const char *message = NULL;
718         struct xenbus_transaction xbt;
719         int err;
720
721         /* Create shared ring, alloc event channel. */
722         err = setup_blkring(dev, info);
723         if (err)
724                 goto out;
725
726 again:
727         err = xenbus_transaction_start(&xbt);
728         if (err) {
729                 xenbus_dev_fatal(dev, err, "starting transaction");
730                 goto destroy_blkring;
731         }
732
733         err = xenbus_printf(xbt, dev->nodename,
734                             "ring-ref", "%u", info->ring_ref);
735         if (err) {
736                 message = "writing ring-ref";
737                 goto abort_transaction;
738         }
739         err = xenbus_printf(xbt, dev->nodename,
740                             "event-channel", "%u", info->evtchn);
741         if (err) {
742                 message = "writing event-channel";
743                 goto abort_transaction;
744         }
745         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
746                             XEN_IO_PROTO_ABI_NATIVE);
747         if (err) {
748                 message = "writing protocol";
749                 goto abort_transaction;
750         }
751
752         err = xenbus_transaction_end(xbt, 0);
753         if (err) {
754                 if (err == -EAGAIN)
755                         goto again;
756                 xenbus_dev_fatal(dev, err, "completing transaction");
757                 goto destroy_blkring;
758         }
759
760         xenbus_switch_state(dev, XenbusStateInitialised);
761
762         return 0;
763
764  abort_transaction:
765         xenbus_transaction_end(xbt, 1);
766         if (message)
767                 xenbus_dev_fatal(dev, err, "%s", message);
768  destroy_blkring:
769         blkif_free(info, 0);
770  out:
771         return err;
772 }
773
774
775 /**
776  * Entry point to this code when a new device is created.  Allocate the basic
777  * structures and the ring buffer for communication with the backend, and
778  * inform the backend of the appropriate details for those.  Switch to
779  * Initialised state.
780  */
781 static int blkfront_probe(struct xenbus_device *dev,
782                           const struct xenbus_device_id *id)
783 {
784         int err, vdevice, i;
785         struct blkfront_info *info;
786
787         /* FIXME: Use dynamic device id if this is not set. */
788         err = xenbus_scanf(XBT_NIL, dev->nodename,
789                            "virtual-device", "%i", &vdevice);
790         if (err != 1) {
791                 /* go looking in the extended area instead */
792                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
793                                    "%i", &vdevice);
794                 if (err != 1) {
795                         xenbus_dev_fatal(dev, err, "reading virtual-device");
796                         return err;
797                 }
798         }
799
800         info = kzalloc(sizeof(*info), GFP_KERNEL);
801         if (!info) {
802                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
803                 return -ENOMEM;
804         }
805
806         info->xbdev = dev;
807         info->vdevice = vdevice;
808         info->connected = BLKIF_STATE_DISCONNECTED;
809         INIT_WORK(&info->work, blkif_restart_queue);
810
811         for (i = 0; i < BLK_RING_SIZE; i++)
812                 info->shadow[i].req.id = i+1;
813         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
814
815         /* Front end dir is a number, which is used as the id. */
816         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
817         dev_set_drvdata(&dev->dev, info);
818
819         err = talk_to_blkback(dev, info);
820         if (err) {
821                 kfree(info);
822                 dev_set_drvdata(&dev->dev, NULL);
823                 return err;
824         }
825
826         return 0;
827 }
828
829
830 static int blkif_recover(struct blkfront_info *info)
831 {
832         int i;
833         struct blkif_request *req;
834         struct blk_shadow *copy;
835         int j;
836
837         /* Stage 1: Make a safe copy of the shadow state. */
838         copy = kmalloc(sizeof(info->shadow),
839                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
840         if (!copy)
841                 return -ENOMEM;
842         memcpy(copy, info->shadow, sizeof(info->shadow));
843
844         /* Stage 2: Set up free list. */
845         memset(&info->shadow, 0, sizeof(info->shadow));
846         for (i = 0; i < BLK_RING_SIZE; i++)
847                 info->shadow[i].req.id = i+1;
848         info->shadow_free = info->ring.req_prod_pvt;
849         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
850
851         /* Stage 3: Find pending requests and requeue them. */
852         for (i = 0; i < BLK_RING_SIZE; i++) {
853                 /* Not in use? */
854                 if (copy[i].request == 0)
855                         continue;
856
857                 /* Grab a request slot and copy shadow state into it. */
858                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
859                 *req = copy[i].req;
860
861                 /* We get a new request id, and must reset the shadow state. */
862                 req->id = get_id_from_freelist(info);
863                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
864
865                 /* Rewrite any grant references invalidated by susp/resume. */
866                 for (j = 0; j < req->nr_segments; j++)
867                         gnttab_grant_foreign_access_ref(
868                                 req->seg[j].gref,
869                                 info->xbdev->otherend_id,
870                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
871                                 rq_data_dir(
872                                         (struct request *)
873                                         info->shadow[req->id].request));
874                 info->shadow[req->id].req = *req;
875
876                 info->ring.req_prod_pvt++;
877         }
878
879         kfree(copy);
880
881         xenbus_switch_state(info->xbdev, XenbusStateConnected);
882
883         spin_lock_irq(&blkif_io_lock);
884
885         /* Now safe for us to use the shared ring */
886         info->connected = BLKIF_STATE_CONNECTED;
887
888         /* Send off requeued requests */
889         flush_requests(info);
890
891         /* Kick any other new requests queued since we resumed */
892         kick_pending_request_queues(info);
893
894         spin_unlock_irq(&blkif_io_lock);
895
896         return 0;
897 }
898
899 /**
900  * We are reconnecting to the backend, due to a suspend/resume, or a backend
901  * driver restart.  We tear down our blkif structure and recreate it, but
902  * leave the device-layer structures intact so that this is transparent to the
903  * rest of the kernel.
904  */
905 static int blkfront_resume(struct xenbus_device *dev)
906 {
907         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
908         int err;
909
910         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
911
912         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
913
914         err = talk_to_blkback(dev, info);
915         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
916                 err = blkif_recover(info);
917
918         return err;
919 }
920
921
922 /*
923  * Invoked when the backend is finally 'ready' (and has told produced
924  * the details about the physical device - #sectors, size, etc).
925  */
926 static void blkfront_connect(struct blkfront_info *info)
927 {
928         unsigned long long sectors;
929         unsigned long sector_size;
930         unsigned int binfo;
931         int err;
932
933         if ((info->connected == BLKIF_STATE_CONNECTED) ||
934             (info->connected == BLKIF_STATE_SUSPENDED) )
935                 return;
936
937         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
938                 __func__, info->xbdev->otherend);
939
940         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
941                             "sectors", "%llu", &sectors,
942                             "info", "%u", &binfo,
943                             "sector-size", "%lu", &sector_size,
944                             NULL);
945         if (err) {
946                 xenbus_dev_fatal(info->xbdev, err,
947                                  "reading backend fields at %s",
948                                  info->xbdev->otherend);
949                 return;
950         }
951
952         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
953                             "feature-barrier", "%lu", &info->feature_barrier,
954                             NULL);
955         if (err)
956                 info->feature_barrier = 0;
957
958         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
959         if (err) {
960                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
961                                  info->xbdev->otherend);
962                 return;
963         }
964
965         xenbus_switch_state(info->xbdev, XenbusStateConnected);
966
967         /* Kick pending requests. */
968         spin_lock_irq(&blkif_io_lock);
969         info->connected = BLKIF_STATE_CONNECTED;
970         kick_pending_request_queues(info);
971         spin_unlock_irq(&blkif_io_lock);
972
973         add_disk(info->gd);
974
975         info->is_ready = 1;
976 }
977
978 /**
979  * Handle the change of state of the backend to Closing.  We must delete our
980  * device-layer structures now, to ensure that writes are flushed through to
981  * the backend.  Once is this done, we can switch to Closed in
982  * acknowledgement.
983  */
984 static void blkfront_closing(struct xenbus_device *dev)
985 {
986         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
987         unsigned int minor, nr_minors;
988         unsigned long flags;
989
990         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
991
992         if (info->rq == NULL)
993                 goto out;
994
995         spin_lock_irqsave(&blkif_io_lock, flags);
996
997         /* No more blkif_request(). */
998         blk_stop_queue(info->rq);
999
1000         /* No more gnttab callback work. */
1001         gnttab_cancel_free_callback(&info->callback);
1002         spin_unlock_irqrestore(&blkif_io_lock, flags);
1003
1004         /* Flush gnttab callback work. Must be done with no locks held. */
1005         flush_scheduled_work();
1006
1007         blk_cleanup_queue(info->rq);
1008         info->rq = NULL;
1009
1010         minor = info->gd->first_minor;
1011         nr_minors = info->gd->minors;
1012         del_gendisk(info->gd);
1013         xlbd_release_minors(minor, nr_minors);
1014
1015  out:
1016         xenbus_frontend_closed(dev);
1017 }
1018
1019 /**
1020  * Callback received when the backend's state changes.
1021  */
1022 static void blkback_changed(struct xenbus_device *dev,
1023                             enum xenbus_state backend_state)
1024 {
1025         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1026         struct block_device *bd;
1027
1028         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1029
1030         switch (backend_state) {
1031         case XenbusStateInitialising:
1032         case XenbusStateInitWait:
1033         case XenbusStateInitialised:
1034         case XenbusStateUnknown:
1035         case XenbusStateClosed:
1036                 break;
1037
1038         case XenbusStateConnected:
1039                 blkfront_connect(info);
1040                 break;
1041
1042         case XenbusStateClosing:
1043                 if (info->gd == NULL) {
1044                         xenbus_frontend_closed(dev);
1045                         break;
1046                 }
1047                 bd = bdget_disk(info->gd, 0);
1048                 if (bd == NULL)
1049                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
1050
1051                 mutex_lock(&bd->bd_mutex);
1052                 if (info->users > 0)
1053                         xenbus_dev_error(dev, -EBUSY,
1054                                          "Device in use; refusing to close");
1055                 else
1056                         blkfront_closing(dev);
1057                 mutex_unlock(&bd->bd_mutex);
1058                 bdput(bd);
1059                 break;
1060         }
1061 }
1062
1063 static int blkfront_remove(struct xenbus_device *dev)
1064 {
1065         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1066
1067         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1068
1069         blkif_free(info, 0);
1070
1071         if(info->users == 0)
1072                 kfree(info);
1073         else
1074                 info->is_ready = -1;
1075
1076         return 0;
1077 }
1078
1079 static int blkfront_is_ready(struct xenbus_device *dev)
1080 {
1081         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1082
1083         return info->is_ready > 0;
1084 }
1085
1086 static int blkif_open(struct block_device *bdev, fmode_t mode)
1087 {
1088         struct blkfront_info *info = bdev->bd_disk->private_data;
1089         int ret = 0;
1090
1091         lock_kernel();
1092         if (info->is_ready < 0)
1093                 ret = -ENODEV;
1094         else
1095                 info->users++;
1096         unlock_kernel();
1097
1098         return ret;
1099 }
1100
1101 static int blkif_release(struct gendisk *disk, fmode_t mode)
1102 {
1103         struct blkfront_info *info = disk->private_data;
1104         lock_kernel();
1105         info->users--;
1106         if (info->users == 0) {
1107                 /* Check whether we have been instructed to close.  We will
1108                    have ignored this request initially, as the device was
1109                    still mounted. */
1110                 struct xenbus_device *dev = info->xbdev;
1111                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1112
1113                 if(info->is_ready < 0) {
1114                         blkfront_closing(dev);
1115                         kfree(info);
1116                 } else if (state == XenbusStateClosing && info->is_ready)
1117                         blkfront_closing(dev);
1118         }
1119         unlock_kernel();
1120         return 0;
1121 }
1122
1123 static const struct block_device_operations xlvbd_block_fops =
1124 {
1125         .owner = THIS_MODULE,
1126         .open = blkif_open,
1127         .release = blkif_release,
1128         .getgeo = blkif_getgeo,
1129         .ioctl = blkif_ioctl,
1130 };
1131
1132
1133 static const struct xenbus_device_id blkfront_ids[] = {
1134         { "vbd" },
1135         { "" }
1136 };
1137
1138 static struct xenbus_driver blkfront = {
1139         .name = "vbd",
1140         .owner = THIS_MODULE,
1141         .ids = blkfront_ids,
1142         .probe = blkfront_probe,
1143         .remove = blkfront_remove,
1144         .resume = blkfront_resume,
1145         .otherend_changed = blkback_changed,
1146         .is_ready = blkfront_is_ready,
1147 };
1148
1149 static int __init xlblk_init(void)
1150 {
1151         if (!xen_domain())
1152                 return -ENODEV;
1153
1154         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1155                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1156                        XENVBD_MAJOR, DEV_NAME);
1157                 return -ENODEV;
1158         }
1159
1160         return xenbus_register_frontend(&blkfront);
1161 }
1162 module_init(xlblk_init);
1163
1164
1165 static void __exit xlblk_exit(void)
1166 {
1167         return xenbus_unregister_driver(&blkfront);
1168 }
1169 module_exit(xlblk_exit);
1170
1171 MODULE_DESCRIPTION("Xen virtual block device frontend");
1172 MODULE_LICENSE("GPL");
1173 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1174 MODULE_ALIAS("xen:vbd");
1175 MODULE_ALIAS("xenblk");