xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 9ac8d3fb)
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 block_device *bdev, fmode_t mode,
160 		       unsigned command, unsigned long argument)
161 {
162 	struct blkfront_info *info = bdev->bd_disk->private_data;
163 	int i;
164 
165 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
166 		command, (long)argument);
167 
168 	switch (command) {
169 	case CDROMMULTISESSION:
170 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
171 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
172 			if (put_user(0, (char __user *)(argument + i)))
173 				return -EFAULT;
174 		return 0;
175 
176 	case CDROM_GET_CAPABILITY: {
177 		struct gendisk *gd = info->gd;
178 		if (gd->flags & GENHD_FL_CD)
179 			return 0;
180 		return -EINVAL;
181 	}
182 
183 	default:
184 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
185 		  command);*/
186 		return -EINVAL; /* same return as native Linux */
187 	}
188 
189 	return 0;
190 }
191 
192 /*
193  * blkif_queue_request
194  *
195  * request block io
196  *
197  * id: for guest use only.
198  * operation: BLKIF_OP_{READ,WRITE,PROBE}
199  * buffer: buffer to read/write into. this should be a
200  *   virtual address in the guest os.
201  */
202 static int blkif_queue_request(struct request *req)
203 {
204 	struct blkfront_info *info = req->rq_disk->private_data;
205 	unsigned long buffer_mfn;
206 	struct blkif_request *ring_req;
207 	struct req_iterator iter;
208 	struct bio_vec *bvec;
209 	unsigned long id;
210 	unsigned int fsect, lsect;
211 	int ref;
212 	grant_ref_t gref_head;
213 
214 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
215 		return 1;
216 
217 	if (gnttab_alloc_grant_references(
218 		BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
219 		gnttab_request_free_callback(
220 			&info->callback,
221 			blkif_restart_queue_callback,
222 			info,
223 			BLKIF_MAX_SEGMENTS_PER_REQUEST);
224 		return 1;
225 	}
226 
227 	/* Fill out a communications ring structure. */
228 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
229 	id = get_id_from_freelist(info);
230 	info->shadow[id].request = (unsigned long)req;
231 
232 	ring_req->id = id;
233 	ring_req->sector_number = (blkif_sector_t)req->sector;
234 	ring_req->handle = info->handle;
235 
236 	ring_req->operation = rq_data_dir(req) ?
237 		BLKIF_OP_WRITE : BLKIF_OP_READ;
238 	if (blk_barrier_rq(req))
239 		ring_req->operation = BLKIF_OP_WRITE_BARRIER;
240 
241 	ring_req->nr_segments = 0;
242 	rq_for_each_segment(bvec, req, iter) {
243 		BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
244 		buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
245 		fsect = bvec->bv_offset >> 9;
246 		lsect = fsect + (bvec->bv_len >> 9) - 1;
247 		/* install a grant reference. */
248 		ref = gnttab_claim_grant_reference(&gref_head);
249 		BUG_ON(ref == -ENOSPC);
250 
251 		gnttab_grant_foreign_access_ref(
252 				ref,
253 				info->xbdev->otherend_id,
254 				buffer_mfn,
255 				rq_data_dir(req) );
256 
257 		info->shadow[id].frame[ring_req->nr_segments] =
258 				mfn_to_pfn(buffer_mfn);
259 
260 		ring_req->seg[ring_req->nr_segments] =
261 				(struct blkif_request_segment) {
262 					.gref       = ref,
263 					.first_sect = fsect,
264 					.last_sect  = lsect };
265 
266 		ring_req->nr_segments++;
267 	}
268 
269 	info->ring.req_prod_pvt++;
270 
271 	/* Keep a private copy so we can reissue requests when recovering. */
272 	info->shadow[id].req = *ring_req;
273 
274 	gnttab_free_grant_references(gref_head);
275 
276 	return 0;
277 }
278 
279 
280 static inline void flush_requests(struct blkfront_info *info)
281 {
282 	int notify;
283 
284 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
285 
286 	if (notify)
287 		notify_remote_via_irq(info->irq);
288 }
289 
290 /*
291  * do_blkif_request
292  *  read a block; request is in a request queue
293  */
294 static void do_blkif_request(struct request_queue *rq)
295 {
296 	struct blkfront_info *info = NULL;
297 	struct request *req;
298 	int queued;
299 
300 	pr_debug("Entered do_blkif_request\n");
301 
302 	queued = 0;
303 
304 	while ((req = elv_next_request(rq)) != NULL) {
305 		info = req->rq_disk->private_data;
306 		if (!blk_fs_request(req)) {
307 			end_request(req, 0);
308 			continue;
309 		}
310 
311 		if (RING_FULL(&info->ring))
312 			goto wait;
313 
314 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
315 			 "(%u/%li) buffer:%p [%s]\n",
316 			 req, req->cmd, (unsigned long)req->sector,
317 			 req->current_nr_sectors,
318 			 req->nr_sectors, req->buffer,
319 			 rq_data_dir(req) ? "write" : "read");
320 
321 
322 		blkdev_dequeue_request(req);
323 		if (blkif_queue_request(req)) {
324 			blk_requeue_request(rq, req);
325 wait:
326 			/* Avoid pointless unplugs. */
327 			blk_stop_queue(rq);
328 			break;
329 		}
330 
331 		queued++;
332 	}
333 
334 	if (queued != 0)
335 		flush_requests(info);
336 }
337 
338 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
339 {
340 	struct request_queue *rq;
341 
342 	rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
343 	if (rq == NULL)
344 		return -1;
345 
346 	elevator_init(rq, "noop");
347 
348 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
349 	blk_queue_hardsect_size(rq, sector_size);
350 	blk_queue_max_sectors(rq, 512);
351 
352 	/* Each segment in a request is up to an aligned page in size. */
353 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
354 	blk_queue_max_segment_size(rq, PAGE_SIZE);
355 
356 	/* Ensure a merged request will fit in a single I/O ring slot. */
357 	blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
358 	blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
359 
360 	/* Make sure buffer addresses are sector-aligned. */
361 	blk_queue_dma_alignment(rq, 511);
362 
363 	/* Make sure we don't use bounce buffers. */
364 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
365 
366 	gd->queue = rq;
367 
368 	return 0;
369 }
370 
371 
372 static int xlvbd_barrier(struct blkfront_info *info)
373 {
374 	int err;
375 
376 	err = blk_queue_ordered(info->rq,
377 				info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
378 				NULL);
379 
380 	if (err)
381 		return err;
382 
383 	printk(KERN_INFO "blkfront: %s: barriers %s\n",
384 	       info->gd->disk_name,
385 	       info->feature_barrier ? "enabled" : "disabled");
386 	return 0;
387 }
388 
389 
390 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
391 			       struct blkfront_info *info,
392 			       u16 vdisk_info, u16 sector_size)
393 {
394 	struct gendisk *gd;
395 	int nr_minors = 1;
396 	int err = -ENODEV;
397 	unsigned int offset;
398 	int minor;
399 	int nr_parts;
400 
401 	BUG_ON(info->gd != NULL);
402 	BUG_ON(info->rq != NULL);
403 
404 	if ((info->vdevice>>EXT_SHIFT) > 1) {
405 		/* this is above the extended range; something is wrong */
406 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
407 		return -ENODEV;
408 	}
409 
410 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
411 		minor = BLKIF_MINOR(info->vdevice);
412 		nr_parts = PARTS_PER_DISK;
413 	} else {
414 		minor = BLKIF_MINOR_EXT(info->vdevice);
415 		nr_parts = PARTS_PER_EXT_DISK;
416 	}
417 
418 	if ((minor % nr_parts) == 0)
419 		nr_minors = nr_parts;
420 
421 	gd = alloc_disk(nr_minors);
422 	if (gd == NULL)
423 		goto out;
424 
425 	offset = minor / nr_parts;
426 
427 	if (nr_minors > 1) {
428 		if (offset < 26)
429 			sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
430 		else
431 			sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
432 				'a' + ((offset / 26)-1), 'a' + (offset % 26));
433 	} else {
434 		if (offset < 26)
435 			sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
436 				'a' + offset,
437 				minor & (nr_parts - 1));
438 		else
439 			sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
440 				'a' + ((offset / 26) - 1),
441 				'a' + (offset % 26),
442 				minor & (nr_parts - 1));
443 	}
444 
445 	gd->major = XENVBD_MAJOR;
446 	gd->first_minor = minor;
447 	gd->fops = &xlvbd_block_fops;
448 	gd->private_data = info;
449 	gd->driverfs_dev = &(info->xbdev->dev);
450 	set_capacity(gd, capacity);
451 
452 	if (xlvbd_init_blk_queue(gd, sector_size)) {
453 		del_gendisk(gd);
454 		goto out;
455 	}
456 
457 	info->rq = gd->queue;
458 	info->gd = gd;
459 
460 	if (info->feature_barrier)
461 		xlvbd_barrier(info);
462 
463 	if (vdisk_info & VDISK_READONLY)
464 		set_disk_ro(gd, 1);
465 
466 	if (vdisk_info & VDISK_REMOVABLE)
467 		gd->flags |= GENHD_FL_REMOVABLE;
468 
469 	if (vdisk_info & VDISK_CDROM)
470 		gd->flags |= GENHD_FL_CD;
471 
472 	return 0;
473 
474  out:
475 	return err;
476 }
477 
478 static void kick_pending_request_queues(struct blkfront_info *info)
479 {
480 	if (!RING_FULL(&info->ring)) {
481 		/* Re-enable calldowns. */
482 		blk_start_queue(info->rq);
483 		/* Kick things off immediately. */
484 		do_blkif_request(info->rq);
485 	}
486 }
487 
488 static void blkif_restart_queue(struct work_struct *work)
489 {
490 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
491 
492 	spin_lock_irq(&blkif_io_lock);
493 	if (info->connected == BLKIF_STATE_CONNECTED)
494 		kick_pending_request_queues(info);
495 	spin_unlock_irq(&blkif_io_lock);
496 }
497 
498 static void blkif_free(struct blkfront_info *info, int suspend)
499 {
500 	/* Prevent new requests being issued until we fix things up. */
501 	spin_lock_irq(&blkif_io_lock);
502 	info->connected = suspend ?
503 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
504 	/* No more blkif_request(). */
505 	if (info->rq)
506 		blk_stop_queue(info->rq);
507 	/* No more gnttab callback work. */
508 	gnttab_cancel_free_callback(&info->callback);
509 	spin_unlock_irq(&blkif_io_lock);
510 
511 	/* Flush gnttab callback work. Must be done with no locks held. */
512 	flush_scheduled_work();
513 
514 	/* Free resources associated with old device channel. */
515 	if (info->ring_ref != GRANT_INVALID_REF) {
516 		gnttab_end_foreign_access(info->ring_ref, 0,
517 					  (unsigned long)info->ring.sring);
518 		info->ring_ref = GRANT_INVALID_REF;
519 		info->ring.sring = NULL;
520 	}
521 	if (info->irq)
522 		unbind_from_irqhandler(info->irq, info);
523 	info->evtchn = info->irq = 0;
524 
525 }
526 
527 static void blkif_completion(struct blk_shadow *s)
528 {
529 	int i;
530 	for (i = 0; i < s->req.nr_segments; i++)
531 		gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
532 }
533 
534 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
535 {
536 	struct request *req;
537 	struct blkif_response *bret;
538 	RING_IDX i, rp;
539 	unsigned long flags;
540 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
541 	int error;
542 
543 	spin_lock_irqsave(&blkif_io_lock, flags);
544 
545 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
546 		spin_unlock_irqrestore(&blkif_io_lock, flags);
547 		return IRQ_HANDLED;
548 	}
549 
550  again:
551 	rp = info->ring.sring->rsp_prod;
552 	rmb(); /* Ensure we see queued responses up to 'rp'. */
553 
554 	for (i = info->ring.rsp_cons; i != rp; i++) {
555 		unsigned long id;
556 		int ret;
557 
558 		bret = RING_GET_RESPONSE(&info->ring, i);
559 		id   = bret->id;
560 		req  = (struct request *)info->shadow[id].request;
561 
562 		blkif_completion(&info->shadow[id]);
563 
564 		add_id_to_freelist(info, id);
565 
566 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
567 		switch (bret->operation) {
568 		case BLKIF_OP_WRITE_BARRIER:
569 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
570 				printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
571 				       info->gd->disk_name);
572 				error = -EOPNOTSUPP;
573 				info->feature_barrier = 0;
574 				xlvbd_barrier(info);
575 			}
576 			/* fall through */
577 		case BLKIF_OP_READ:
578 		case BLKIF_OP_WRITE:
579 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
580 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
581 					"request: %x\n", bret->status);
582 
583 			ret = __blk_end_request(req, error, blk_rq_bytes(req));
584 			BUG_ON(ret);
585 			break;
586 		default:
587 			BUG();
588 		}
589 	}
590 
591 	info->ring.rsp_cons = i;
592 
593 	if (i != info->ring.req_prod_pvt) {
594 		int more_to_do;
595 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
596 		if (more_to_do)
597 			goto again;
598 	} else
599 		info->ring.sring->rsp_event = i + 1;
600 
601 	kick_pending_request_queues(info);
602 
603 	spin_unlock_irqrestore(&blkif_io_lock, flags);
604 
605 	return IRQ_HANDLED;
606 }
607 
608 
609 static int setup_blkring(struct xenbus_device *dev,
610 			 struct blkfront_info *info)
611 {
612 	struct blkif_sring *sring;
613 	int err;
614 
615 	info->ring_ref = GRANT_INVALID_REF;
616 
617 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
618 	if (!sring) {
619 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
620 		return -ENOMEM;
621 	}
622 	SHARED_RING_INIT(sring);
623 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
624 
625 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
626 	if (err < 0) {
627 		free_page((unsigned long)sring);
628 		info->ring.sring = NULL;
629 		goto fail;
630 	}
631 	info->ring_ref = err;
632 
633 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
634 	if (err)
635 		goto fail;
636 
637 	err = bind_evtchn_to_irqhandler(info->evtchn,
638 					blkif_interrupt,
639 					IRQF_SAMPLE_RANDOM, "blkif", info);
640 	if (err <= 0) {
641 		xenbus_dev_fatal(dev, err,
642 				 "bind_evtchn_to_irqhandler failed");
643 		goto fail;
644 	}
645 	info->irq = err;
646 
647 	return 0;
648 fail:
649 	blkif_free(info, 0);
650 	return err;
651 }
652 
653 
654 /* Common code used when first setting up, and when resuming. */
655 static int talk_to_backend(struct xenbus_device *dev,
656 			   struct blkfront_info *info)
657 {
658 	const char *message = NULL;
659 	struct xenbus_transaction xbt;
660 	int err;
661 
662 	/* Create shared ring, alloc event channel. */
663 	err = setup_blkring(dev, info);
664 	if (err)
665 		goto out;
666 
667 again:
668 	err = xenbus_transaction_start(&xbt);
669 	if (err) {
670 		xenbus_dev_fatal(dev, err, "starting transaction");
671 		goto destroy_blkring;
672 	}
673 
674 	err = xenbus_printf(xbt, dev->nodename,
675 			    "ring-ref", "%u", info->ring_ref);
676 	if (err) {
677 		message = "writing ring-ref";
678 		goto abort_transaction;
679 	}
680 	err = xenbus_printf(xbt, dev->nodename,
681 			    "event-channel", "%u", info->evtchn);
682 	if (err) {
683 		message = "writing event-channel";
684 		goto abort_transaction;
685 	}
686 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
687 			    XEN_IO_PROTO_ABI_NATIVE);
688 	if (err) {
689 		message = "writing protocol";
690 		goto abort_transaction;
691 	}
692 
693 	err = xenbus_transaction_end(xbt, 0);
694 	if (err) {
695 		if (err == -EAGAIN)
696 			goto again;
697 		xenbus_dev_fatal(dev, err, "completing transaction");
698 		goto destroy_blkring;
699 	}
700 
701 	xenbus_switch_state(dev, XenbusStateInitialised);
702 
703 	return 0;
704 
705  abort_transaction:
706 	xenbus_transaction_end(xbt, 1);
707 	if (message)
708 		xenbus_dev_fatal(dev, err, "%s", message);
709  destroy_blkring:
710 	blkif_free(info, 0);
711  out:
712 	return err;
713 }
714 
715 
716 /**
717  * Entry point to this code when a new device is created.  Allocate the basic
718  * structures and the ring buffer for communication with the backend, and
719  * inform the backend of the appropriate details for those.  Switch to
720  * Initialised state.
721  */
722 static int blkfront_probe(struct xenbus_device *dev,
723 			  const struct xenbus_device_id *id)
724 {
725 	int err, vdevice, i;
726 	struct blkfront_info *info;
727 
728 	/* FIXME: Use dynamic device id if this is not set. */
729 	err = xenbus_scanf(XBT_NIL, dev->nodename,
730 			   "virtual-device", "%i", &vdevice);
731 	if (err != 1) {
732 		/* go looking in the extended area instead */
733 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
734 				   "%i", &vdevice);
735 		if (err != 1) {
736 			xenbus_dev_fatal(dev, err, "reading virtual-device");
737 			return err;
738 		}
739 	}
740 
741 	info = kzalloc(sizeof(*info), GFP_KERNEL);
742 	if (!info) {
743 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
744 		return -ENOMEM;
745 	}
746 
747 	info->xbdev = dev;
748 	info->vdevice = vdevice;
749 	info->connected = BLKIF_STATE_DISCONNECTED;
750 	INIT_WORK(&info->work, blkif_restart_queue);
751 
752 	for (i = 0; i < BLK_RING_SIZE; i++)
753 		info->shadow[i].req.id = i+1;
754 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
755 
756 	/* Front end dir is a number, which is used as the id. */
757 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
758 	dev->dev.driver_data = info;
759 
760 	err = talk_to_backend(dev, info);
761 	if (err) {
762 		kfree(info);
763 		dev->dev.driver_data = NULL;
764 		return err;
765 	}
766 
767 	return 0;
768 }
769 
770 
771 static int blkif_recover(struct blkfront_info *info)
772 {
773 	int i;
774 	struct blkif_request *req;
775 	struct blk_shadow *copy;
776 	int j;
777 
778 	/* Stage 1: Make a safe copy of the shadow state. */
779 	copy = kmalloc(sizeof(info->shadow),
780 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
781 	if (!copy)
782 		return -ENOMEM;
783 	memcpy(copy, info->shadow, sizeof(info->shadow));
784 
785 	/* Stage 2: Set up free list. */
786 	memset(&info->shadow, 0, sizeof(info->shadow));
787 	for (i = 0; i < BLK_RING_SIZE; i++)
788 		info->shadow[i].req.id = i+1;
789 	info->shadow_free = info->ring.req_prod_pvt;
790 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
791 
792 	/* Stage 3: Find pending requests and requeue them. */
793 	for (i = 0; i < BLK_RING_SIZE; i++) {
794 		/* Not in use? */
795 		if (copy[i].request == 0)
796 			continue;
797 
798 		/* Grab a request slot and copy shadow state into it. */
799 		req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
800 		*req = copy[i].req;
801 
802 		/* We get a new request id, and must reset the shadow state. */
803 		req->id = get_id_from_freelist(info);
804 		memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
805 
806 		/* Rewrite any grant references invalidated by susp/resume. */
807 		for (j = 0; j < req->nr_segments; j++)
808 			gnttab_grant_foreign_access_ref(
809 				req->seg[j].gref,
810 				info->xbdev->otherend_id,
811 				pfn_to_mfn(info->shadow[req->id].frame[j]),
812 				rq_data_dir(
813 					(struct request *)
814 					info->shadow[req->id].request));
815 		info->shadow[req->id].req = *req;
816 
817 		info->ring.req_prod_pvt++;
818 	}
819 
820 	kfree(copy);
821 
822 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
823 
824 	spin_lock_irq(&blkif_io_lock);
825 
826 	/* Now safe for us to use the shared ring */
827 	info->connected = BLKIF_STATE_CONNECTED;
828 
829 	/* Send off requeued requests */
830 	flush_requests(info);
831 
832 	/* Kick any other new requests queued since we resumed */
833 	kick_pending_request_queues(info);
834 
835 	spin_unlock_irq(&blkif_io_lock);
836 
837 	return 0;
838 }
839 
840 /**
841  * We are reconnecting to the backend, due to a suspend/resume, or a backend
842  * driver restart.  We tear down our blkif structure and recreate it, but
843  * leave the device-layer structures intact so that this is transparent to the
844  * rest of the kernel.
845  */
846 static int blkfront_resume(struct xenbus_device *dev)
847 {
848 	struct blkfront_info *info = dev->dev.driver_data;
849 	int err;
850 
851 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
852 
853 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
854 
855 	err = talk_to_backend(dev, info);
856 	if (info->connected == BLKIF_STATE_SUSPENDED && !err)
857 		err = blkif_recover(info);
858 
859 	return err;
860 }
861 
862 
863 /*
864  * Invoked when the backend is finally 'ready' (and has told produced
865  * the details about the physical device - #sectors, size, etc).
866  */
867 static void blkfront_connect(struct blkfront_info *info)
868 {
869 	unsigned long long sectors;
870 	unsigned long sector_size;
871 	unsigned int binfo;
872 	int err;
873 
874 	if ((info->connected == BLKIF_STATE_CONNECTED) ||
875 	    (info->connected == BLKIF_STATE_SUSPENDED) )
876 		return;
877 
878 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
879 		__func__, info->xbdev->otherend);
880 
881 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
882 			    "sectors", "%llu", &sectors,
883 			    "info", "%u", &binfo,
884 			    "sector-size", "%lu", &sector_size,
885 			    NULL);
886 	if (err) {
887 		xenbus_dev_fatal(info->xbdev, err,
888 				 "reading backend fields at %s",
889 				 info->xbdev->otherend);
890 		return;
891 	}
892 
893 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
894 			    "feature-barrier", "%lu", &info->feature_barrier,
895 			    NULL);
896 	if (err)
897 		info->feature_barrier = 0;
898 
899 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
900 	if (err) {
901 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
902 				 info->xbdev->otherend);
903 		return;
904 	}
905 
906 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
907 
908 	/* Kick pending requests. */
909 	spin_lock_irq(&blkif_io_lock);
910 	info->connected = BLKIF_STATE_CONNECTED;
911 	kick_pending_request_queues(info);
912 	spin_unlock_irq(&blkif_io_lock);
913 
914 	add_disk(info->gd);
915 
916 	info->is_ready = 1;
917 }
918 
919 /**
920  * Handle the change of state of the backend to Closing.  We must delete our
921  * device-layer structures now, to ensure that writes are flushed through to
922  * the backend.  Once is this done, we can switch to Closed in
923  * acknowledgement.
924  */
925 static void blkfront_closing(struct xenbus_device *dev)
926 {
927 	struct blkfront_info *info = dev->dev.driver_data;
928 	unsigned long flags;
929 
930 	dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
931 
932 	if (info->rq == NULL)
933 		goto out;
934 
935 	spin_lock_irqsave(&blkif_io_lock, flags);
936 
937 	del_gendisk(info->gd);
938 
939 	/* No more blkif_request(). */
940 	blk_stop_queue(info->rq);
941 
942 	/* No more gnttab callback work. */
943 	gnttab_cancel_free_callback(&info->callback);
944 	spin_unlock_irqrestore(&blkif_io_lock, flags);
945 
946 	/* Flush gnttab callback work. Must be done with no locks held. */
947 	flush_scheduled_work();
948 
949 	blk_cleanup_queue(info->rq);
950 	info->rq = NULL;
951 
952  out:
953 	xenbus_frontend_closed(dev);
954 }
955 
956 /**
957  * Callback received when the backend's state changes.
958  */
959 static void backend_changed(struct xenbus_device *dev,
960 			    enum xenbus_state backend_state)
961 {
962 	struct blkfront_info *info = dev->dev.driver_data;
963 	struct block_device *bd;
964 
965 	dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
966 
967 	switch (backend_state) {
968 	case XenbusStateInitialising:
969 	case XenbusStateInitWait:
970 	case XenbusStateInitialised:
971 	case XenbusStateUnknown:
972 	case XenbusStateClosed:
973 		break;
974 
975 	case XenbusStateConnected:
976 		blkfront_connect(info);
977 		break;
978 
979 	case XenbusStateClosing:
980 		bd = bdget_disk(info->gd, 0);
981 		if (bd == NULL)
982 			xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
983 
984 		mutex_lock(&bd->bd_mutex);
985 		if (info->users > 0)
986 			xenbus_dev_error(dev, -EBUSY,
987 					 "Device in use; refusing to close");
988 		else
989 			blkfront_closing(dev);
990 		mutex_unlock(&bd->bd_mutex);
991 		bdput(bd);
992 		break;
993 	}
994 }
995 
996 static int blkfront_remove(struct xenbus_device *dev)
997 {
998 	struct blkfront_info *info = dev->dev.driver_data;
999 
1000 	dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1001 
1002 	blkif_free(info, 0);
1003 
1004 	kfree(info);
1005 
1006 	return 0;
1007 }
1008 
1009 static int blkfront_is_ready(struct xenbus_device *dev)
1010 {
1011 	struct blkfront_info *info = dev->dev.driver_data;
1012 
1013 	return info->is_ready;
1014 }
1015 
1016 static int blkif_open(struct block_device *bdev, fmode_t mode)
1017 {
1018 	struct blkfront_info *info = bdev->bd_disk->private_data;
1019 	info->users++;
1020 	return 0;
1021 }
1022 
1023 static int blkif_release(struct gendisk *disk, fmode_t mode)
1024 {
1025 	struct blkfront_info *info = disk->private_data;
1026 	info->users--;
1027 	if (info->users == 0) {
1028 		/* Check whether we have been instructed to close.  We will
1029 		   have ignored this request initially, as the device was
1030 		   still mounted. */
1031 		struct xenbus_device *dev = info->xbdev;
1032 		enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1033 
1034 		if (state == XenbusStateClosing && info->is_ready)
1035 			blkfront_closing(dev);
1036 	}
1037 	return 0;
1038 }
1039 
1040 static struct block_device_operations xlvbd_block_fops =
1041 {
1042 	.owner = THIS_MODULE,
1043 	.open = blkif_open,
1044 	.release = blkif_release,
1045 	.getgeo = blkif_getgeo,
1046 	.locked_ioctl = blkif_ioctl,
1047 };
1048 
1049 
1050 static struct xenbus_device_id blkfront_ids[] = {
1051 	{ "vbd" },
1052 	{ "" }
1053 };
1054 
1055 static struct xenbus_driver blkfront = {
1056 	.name = "vbd",
1057 	.owner = THIS_MODULE,
1058 	.ids = blkfront_ids,
1059 	.probe = blkfront_probe,
1060 	.remove = blkfront_remove,
1061 	.resume = blkfront_resume,
1062 	.otherend_changed = backend_changed,
1063 	.is_ready = blkfront_is_ready,
1064 };
1065 
1066 static int __init xlblk_init(void)
1067 {
1068 	if (!xen_domain())
1069 		return -ENODEV;
1070 
1071 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1072 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1073 		       XENVBD_MAJOR, DEV_NAME);
1074 		return -ENODEV;
1075 	}
1076 
1077 	return xenbus_register_frontend(&blkfront);
1078 }
1079 module_init(xlblk_init);
1080 
1081 
1082 static void __exit xlblk_exit(void)
1083 {
1084 	return xenbus_unregister_driver(&blkfront);
1085 }
1086 module_exit(xlblk_exit);
1087 
1088 MODULE_DESCRIPTION("Xen virtual block device frontend");
1089 MODULE_LICENSE("GPL");
1090 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1091 MODULE_ALIAS("xen:vbd");
1092 MODULE_ALIAS("xenblk");
1093