xref: /linux/drivers/block/xen-blkfront.c (revision a234ca0faa65dcd5cc473915bd925130ebb7b74b)
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 #include <xen/platform_pci.h>
53 
54 #include <xen/interface/grant_table.h>
55 #include <xen/interface/io/blkif.h>
56 #include <xen/interface/io/protocols.h>
57 
58 #include <asm/xen/hypervisor.h>
59 
60 enum blkif_state {
61 	BLKIF_STATE_DISCONNECTED,
62 	BLKIF_STATE_CONNECTED,
63 	BLKIF_STATE_SUSPENDED,
64 };
65 
66 struct blk_shadow {
67 	struct blkif_request req;
68 	unsigned long request;
69 	unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
70 };
71 
72 static const struct block_device_operations xlvbd_block_fops;
73 
74 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
75 
76 /*
77  * We have one of these per vbd, whether ide, scsi or 'other'.  They
78  * hang in private_data off the gendisk structure. We may end up
79  * putting all kinds of interesting stuff here :-)
80  */
81 struct blkfront_info
82 {
83 	struct mutex mutex;
84 	struct xenbus_device *xbdev;
85 	struct gendisk *gd;
86 	int vdevice;
87 	blkif_vdev_t handle;
88 	enum blkif_state connected;
89 	int ring_ref;
90 	struct blkif_front_ring ring;
91 	struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
92 	unsigned int evtchn, irq;
93 	struct request_queue *rq;
94 	struct work_struct work;
95 	struct gnttab_free_callback callback;
96 	struct blk_shadow shadow[BLK_RING_SIZE];
97 	unsigned long shadow_free;
98 	int feature_barrier;
99 	int is_ready;
100 };
101 
102 static DEFINE_SPINLOCK(blkif_io_lock);
103 
104 static unsigned int nr_minors;
105 static unsigned long *minors;
106 static DEFINE_SPINLOCK(minor_lock);
107 
108 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
109 	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
110 #define GRANT_INVALID_REF	0
111 
112 #define PARTS_PER_DISK		16
113 #define PARTS_PER_EXT_DISK      256
114 
115 #define BLKIF_MAJOR(dev) ((dev)>>8)
116 #define BLKIF_MINOR(dev) ((dev) & 0xff)
117 
118 #define EXT_SHIFT 28
119 #define EXTENDED (1<<EXT_SHIFT)
120 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
121 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
122 
123 #define DEV_NAME	"xvd"	/* name in /dev */
124 
125 static int get_id_from_freelist(struct blkfront_info *info)
126 {
127 	unsigned long free = info->shadow_free;
128 	BUG_ON(free >= BLK_RING_SIZE);
129 	info->shadow_free = info->shadow[free].req.id;
130 	info->shadow[free].req.id = 0x0fffffee; /* debug */
131 	return free;
132 }
133 
134 static void add_id_to_freelist(struct blkfront_info *info,
135 			       unsigned long id)
136 {
137 	info->shadow[id].req.id  = info->shadow_free;
138 	info->shadow[id].request = 0;
139 	info->shadow_free = id;
140 }
141 
142 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
143 {
144 	unsigned int end = minor + nr;
145 	int rc;
146 
147 	if (end > nr_minors) {
148 		unsigned long *bitmap, *old;
149 
150 		bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
151 				 GFP_KERNEL);
152 		if (bitmap == NULL)
153 			return -ENOMEM;
154 
155 		spin_lock(&minor_lock);
156 		if (end > nr_minors) {
157 			old = minors;
158 			memcpy(bitmap, minors,
159 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
160 			minors = bitmap;
161 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
162 		} else
163 			old = bitmap;
164 		spin_unlock(&minor_lock);
165 		kfree(old);
166 	}
167 
168 	spin_lock(&minor_lock);
169 	if (find_next_bit(minors, end, minor) >= end) {
170 		for (; minor < end; ++minor)
171 			__set_bit(minor, minors);
172 		rc = 0;
173 	} else
174 		rc = -EBUSY;
175 	spin_unlock(&minor_lock);
176 
177 	return rc;
178 }
179 
180 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
181 {
182 	unsigned int end = minor + nr;
183 
184 	BUG_ON(end > nr_minors);
185 	spin_lock(&minor_lock);
186 	for (; minor < end; ++minor)
187 		__clear_bit(minor, minors);
188 	spin_unlock(&minor_lock);
189 }
190 
191 static void blkif_restart_queue_callback(void *arg)
192 {
193 	struct blkfront_info *info = (struct blkfront_info *)arg;
194 	schedule_work(&info->work);
195 }
196 
197 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
198 {
199 	/* We don't have real geometry info, but let's at least return
200 	   values consistent with the size of the device */
201 	sector_t nsect = get_capacity(bd->bd_disk);
202 	sector_t cylinders = nsect;
203 
204 	hg->heads = 0xff;
205 	hg->sectors = 0x3f;
206 	sector_div(cylinders, hg->heads * hg->sectors);
207 	hg->cylinders = cylinders;
208 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
209 		hg->cylinders = 0xffff;
210 	return 0;
211 }
212 
213 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
214 		       unsigned command, unsigned long argument)
215 {
216 	struct blkfront_info *info = bdev->bd_disk->private_data;
217 	int i;
218 
219 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
220 		command, (long)argument);
221 
222 	switch (command) {
223 	case CDROMMULTISESSION:
224 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
225 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
226 			if (put_user(0, (char __user *)(argument + i)))
227 				return -EFAULT;
228 		return 0;
229 
230 	case CDROM_GET_CAPABILITY: {
231 		struct gendisk *gd = info->gd;
232 		if (gd->flags & GENHD_FL_CD)
233 			return 0;
234 		return -EINVAL;
235 	}
236 
237 	default:
238 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
239 		  command);*/
240 		return -EINVAL; /* same return as native Linux */
241 	}
242 
243 	return 0;
244 }
245 
246 /*
247  * blkif_queue_request
248  *
249  * request block io
250  *
251  * id: for guest use only.
252  * operation: BLKIF_OP_{READ,WRITE,PROBE}
253  * buffer: buffer to read/write into. this should be a
254  *   virtual address in the guest os.
255  */
256 static int blkif_queue_request(struct request *req)
257 {
258 	struct blkfront_info *info = req->rq_disk->private_data;
259 	unsigned long buffer_mfn;
260 	struct blkif_request *ring_req;
261 	unsigned long id;
262 	unsigned int fsect, lsect;
263 	int i, ref;
264 	grant_ref_t gref_head;
265 	struct scatterlist *sg;
266 
267 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
268 		return 1;
269 
270 	if (gnttab_alloc_grant_references(
271 		BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
272 		gnttab_request_free_callback(
273 			&info->callback,
274 			blkif_restart_queue_callback,
275 			info,
276 			BLKIF_MAX_SEGMENTS_PER_REQUEST);
277 		return 1;
278 	}
279 
280 	/* Fill out a communications ring structure. */
281 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
282 	id = get_id_from_freelist(info);
283 	info->shadow[id].request = (unsigned long)req;
284 
285 	ring_req->id = id;
286 	ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
287 	ring_req->handle = info->handle;
288 
289 	ring_req->operation = rq_data_dir(req) ?
290 		BLKIF_OP_WRITE : BLKIF_OP_READ;
291 	if (req->cmd_flags & REQ_HARDBARRIER)
292 		ring_req->operation = BLKIF_OP_WRITE_BARRIER;
293 
294 	ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
295 	BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
296 
297 	for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
298 		buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
299 		fsect = sg->offset >> 9;
300 		lsect = fsect + (sg->length >> 9) - 1;
301 		/* install a grant reference. */
302 		ref = gnttab_claim_grant_reference(&gref_head);
303 		BUG_ON(ref == -ENOSPC);
304 
305 		gnttab_grant_foreign_access_ref(
306 				ref,
307 				info->xbdev->otherend_id,
308 				buffer_mfn,
309 				rq_data_dir(req) );
310 
311 		info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
312 		ring_req->seg[i] =
313 				(struct blkif_request_segment) {
314 					.gref       = ref,
315 					.first_sect = fsect,
316 					.last_sect  = lsect };
317 	}
318 
319 	info->ring.req_prod_pvt++;
320 
321 	/* Keep a private copy so we can reissue requests when recovering. */
322 	info->shadow[id].req = *ring_req;
323 
324 	gnttab_free_grant_references(gref_head);
325 
326 	return 0;
327 }
328 
329 
330 static inline void flush_requests(struct blkfront_info *info)
331 {
332 	int notify;
333 
334 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
335 
336 	if (notify)
337 		notify_remote_via_irq(info->irq);
338 }
339 
340 /*
341  * do_blkif_request
342  *  read a block; request is in a request queue
343  */
344 static void do_blkif_request(struct request_queue *rq)
345 {
346 	struct blkfront_info *info = NULL;
347 	struct request *req;
348 	int queued;
349 
350 	pr_debug("Entered do_blkif_request\n");
351 
352 	queued = 0;
353 
354 	while ((req = blk_peek_request(rq)) != NULL) {
355 		info = req->rq_disk->private_data;
356 
357 		if (RING_FULL(&info->ring))
358 			goto wait;
359 
360 		blk_start_request(req);
361 
362 		if (req->cmd_type != REQ_TYPE_FS) {
363 			__blk_end_request_all(req, -EIO);
364 			continue;
365 		}
366 
367 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
368 			 "(%u/%u) buffer:%p [%s]\n",
369 			 req, req->cmd, (unsigned long)blk_rq_pos(req),
370 			 blk_rq_cur_sectors(req), blk_rq_sectors(req),
371 			 req->buffer, rq_data_dir(req) ? "write" : "read");
372 
373 		if (blkif_queue_request(req)) {
374 			blk_requeue_request(rq, req);
375 wait:
376 			/* Avoid pointless unplugs. */
377 			blk_stop_queue(rq);
378 			break;
379 		}
380 
381 		queued++;
382 	}
383 
384 	if (queued != 0)
385 		flush_requests(info);
386 }
387 
388 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
389 {
390 	struct request_queue *rq;
391 
392 	rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
393 	if (rq == NULL)
394 		return -1;
395 
396 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
397 
398 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
399 	blk_queue_logical_block_size(rq, sector_size);
400 	blk_queue_max_hw_sectors(rq, 512);
401 
402 	/* Each segment in a request is up to an aligned page in size. */
403 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
404 	blk_queue_max_segment_size(rq, PAGE_SIZE);
405 
406 	/* Ensure a merged request will fit in a single I/O ring slot. */
407 	blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
408 
409 	/* Make sure buffer addresses are sector-aligned. */
410 	blk_queue_dma_alignment(rq, 511);
411 
412 	/* Make sure we don't use bounce buffers. */
413 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
414 
415 	gd->queue = rq;
416 
417 	return 0;
418 }
419 
420 
421 static int xlvbd_barrier(struct blkfront_info *info)
422 {
423 	int err;
424 	const char *barrier;
425 
426 	switch (info->feature_barrier) {
427 	case QUEUE_ORDERED_DRAIN:	barrier = "enabled (drain)"; break;
428 	case QUEUE_ORDERED_TAG:		barrier = "enabled (tag)"; break;
429 	case QUEUE_ORDERED_NONE:	barrier = "disabled"; break;
430 	default:			return -EINVAL;
431 	}
432 
433 	err = blk_queue_ordered(info->rq, info->feature_barrier);
434 
435 	if (err)
436 		return err;
437 
438 	printk(KERN_INFO "blkfront: %s: barriers %s\n",
439 	       info->gd->disk_name, barrier);
440 	return 0;
441 }
442 
443 
444 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
445 			       struct blkfront_info *info,
446 			       u16 vdisk_info, u16 sector_size)
447 {
448 	struct gendisk *gd;
449 	int nr_minors = 1;
450 	int err = -ENODEV;
451 	unsigned int offset;
452 	int minor;
453 	int nr_parts;
454 
455 	BUG_ON(info->gd != NULL);
456 	BUG_ON(info->rq != NULL);
457 
458 	if ((info->vdevice>>EXT_SHIFT) > 1) {
459 		/* this is above the extended range; something is wrong */
460 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
461 		return -ENODEV;
462 	}
463 
464 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
465 		minor = BLKIF_MINOR(info->vdevice);
466 		nr_parts = PARTS_PER_DISK;
467 	} else {
468 		minor = BLKIF_MINOR_EXT(info->vdevice);
469 		nr_parts = PARTS_PER_EXT_DISK;
470 	}
471 
472 	if ((minor % nr_parts) == 0)
473 		nr_minors = nr_parts;
474 
475 	err = xlbd_reserve_minors(minor, nr_minors);
476 	if (err)
477 		goto out;
478 	err = -ENODEV;
479 
480 	gd = alloc_disk(nr_minors);
481 	if (gd == NULL)
482 		goto release;
483 
484 	offset = minor / nr_parts;
485 
486 	if (nr_minors > 1) {
487 		if (offset < 26)
488 			sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
489 		else
490 			sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
491 				'a' + ((offset / 26)-1), 'a' + (offset % 26));
492 	} else {
493 		if (offset < 26)
494 			sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
495 				'a' + offset,
496 				minor & (nr_parts - 1));
497 		else
498 			sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
499 				'a' + ((offset / 26) - 1),
500 				'a' + (offset % 26),
501 				minor & (nr_parts - 1));
502 	}
503 
504 	gd->major = XENVBD_MAJOR;
505 	gd->first_minor = minor;
506 	gd->fops = &xlvbd_block_fops;
507 	gd->private_data = info;
508 	gd->driverfs_dev = &(info->xbdev->dev);
509 	set_capacity(gd, capacity);
510 
511 	if (xlvbd_init_blk_queue(gd, sector_size)) {
512 		del_gendisk(gd);
513 		goto release;
514 	}
515 
516 	info->rq = gd->queue;
517 	info->gd = gd;
518 
519 	xlvbd_barrier(info);
520 
521 	if (vdisk_info & VDISK_READONLY)
522 		set_disk_ro(gd, 1);
523 
524 	if (vdisk_info & VDISK_REMOVABLE)
525 		gd->flags |= GENHD_FL_REMOVABLE;
526 
527 	if (vdisk_info & VDISK_CDROM)
528 		gd->flags |= GENHD_FL_CD;
529 
530 	return 0;
531 
532  release:
533 	xlbd_release_minors(minor, nr_minors);
534  out:
535 	return err;
536 }
537 
538 static void xlvbd_release_gendisk(struct blkfront_info *info)
539 {
540 	unsigned int minor, nr_minors;
541 	unsigned long flags;
542 
543 	if (info->rq == NULL)
544 		return;
545 
546 	spin_lock_irqsave(&blkif_io_lock, flags);
547 
548 	/* No more blkif_request(). */
549 	blk_stop_queue(info->rq);
550 
551 	/* No more gnttab callback work. */
552 	gnttab_cancel_free_callback(&info->callback);
553 	spin_unlock_irqrestore(&blkif_io_lock, flags);
554 
555 	/* Flush gnttab callback work. Must be done with no locks held. */
556 	flush_scheduled_work();
557 
558 	del_gendisk(info->gd);
559 
560 	minor = info->gd->first_minor;
561 	nr_minors = info->gd->minors;
562 	xlbd_release_minors(minor, nr_minors);
563 
564 	blk_cleanup_queue(info->rq);
565 	info->rq = NULL;
566 
567 	put_disk(info->gd);
568 	info->gd = NULL;
569 }
570 
571 static void kick_pending_request_queues(struct blkfront_info *info)
572 {
573 	if (!RING_FULL(&info->ring)) {
574 		/* Re-enable calldowns. */
575 		blk_start_queue(info->rq);
576 		/* Kick things off immediately. */
577 		do_blkif_request(info->rq);
578 	}
579 }
580 
581 static void blkif_restart_queue(struct work_struct *work)
582 {
583 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
584 
585 	spin_lock_irq(&blkif_io_lock);
586 	if (info->connected == BLKIF_STATE_CONNECTED)
587 		kick_pending_request_queues(info);
588 	spin_unlock_irq(&blkif_io_lock);
589 }
590 
591 static void blkif_free(struct blkfront_info *info, int suspend)
592 {
593 	/* Prevent new requests being issued until we fix things up. */
594 	spin_lock_irq(&blkif_io_lock);
595 	info->connected = suspend ?
596 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
597 	/* No more blkif_request(). */
598 	if (info->rq)
599 		blk_stop_queue(info->rq);
600 	/* No more gnttab callback work. */
601 	gnttab_cancel_free_callback(&info->callback);
602 	spin_unlock_irq(&blkif_io_lock);
603 
604 	/* Flush gnttab callback work. Must be done with no locks held. */
605 	flush_scheduled_work();
606 
607 	/* Free resources associated with old device channel. */
608 	if (info->ring_ref != GRANT_INVALID_REF) {
609 		gnttab_end_foreign_access(info->ring_ref, 0,
610 					  (unsigned long)info->ring.sring);
611 		info->ring_ref = GRANT_INVALID_REF;
612 		info->ring.sring = NULL;
613 	}
614 	if (info->irq)
615 		unbind_from_irqhandler(info->irq, info);
616 	info->evtchn = info->irq = 0;
617 
618 }
619 
620 static void blkif_completion(struct blk_shadow *s)
621 {
622 	int i;
623 	for (i = 0; i < s->req.nr_segments; i++)
624 		gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
625 }
626 
627 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
628 {
629 	struct request *req;
630 	struct blkif_response *bret;
631 	RING_IDX i, rp;
632 	unsigned long flags;
633 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
634 	int error;
635 
636 	spin_lock_irqsave(&blkif_io_lock, flags);
637 
638 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
639 		spin_unlock_irqrestore(&blkif_io_lock, flags);
640 		return IRQ_HANDLED;
641 	}
642 
643  again:
644 	rp = info->ring.sring->rsp_prod;
645 	rmb(); /* Ensure we see queued responses up to 'rp'. */
646 
647 	for (i = info->ring.rsp_cons; i != rp; i++) {
648 		unsigned long id;
649 
650 		bret = RING_GET_RESPONSE(&info->ring, i);
651 		id   = bret->id;
652 		req  = (struct request *)info->shadow[id].request;
653 
654 		blkif_completion(&info->shadow[id]);
655 
656 		add_id_to_freelist(info, id);
657 
658 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
659 		switch (bret->operation) {
660 		case BLKIF_OP_WRITE_BARRIER:
661 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
662 				printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
663 				       info->gd->disk_name);
664 				error = -EOPNOTSUPP;
665 				info->feature_barrier = QUEUE_ORDERED_NONE;
666 				xlvbd_barrier(info);
667 			}
668 			/* fall through */
669 		case BLKIF_OP_READ:
670 		case BLKIF_OP_WRITE:
671 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
672 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
673 					"request: %x\n", bret->status);
674 
675 			__blk_end_request_all(req, error);
676 			break;
677 		default:
678 			BUG();
679 		}
680 	}
681 
682 	info->ring.rsp_cons = i;
683 
684 	if (i != info->ring.req_prod_pvt) {
685 		int more_to_do;
686 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
687 		if (more_to_do)
688 			goto again;
689 	} else
690 		info->ring.sring->rsp_event = i + 1;
691 
692 	kick_pending_request_queues(info);
693 
694 	spin_unlock_irqrestore(&blkif_io_lock, flags);
695 
696 	return IRQ_HANDLED;
697 }
698 
699 
700 static int setup_blkring(struct xenbus_device *dev,
701 			 struct blkfront_info *info)
702 {
703 	struct blkif_sring *sring;
704 	int err;
705 
706 	info->ring_ref = GRANT_INVALID_REF;
707 
708 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
709 	if (!sring) {
710 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
711 		return -ENOMEM;
712 	}
713 	SHARED_RING_INIT(sring);
714 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
715 
716 	sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
717 
718 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
719 	if (err < 0) {
720 		free_page((unsigned long)sring);
721 		info->ring.sring = NULL;
722 		goto fail;
723 	}
724 	info->ring_ref = err;
725 
726 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
727 	if (err)
728 		goto fail;
729 
730 	err = bind_evtchn_to_irqhandler(info->evtchn,
731 					blkif_interrupt,
732 					IRQF_SAMPLE_RANDOM, "blkif", info);
733 	if (err <= 0) {
734 		xenbus_dev_fatal(dev, err,
735 				 "bind_evtchn_to_irqhandler failed");
736 		goto fail;
737 	}
738 	info->irq = err;
739 
740 	return 0;
741 fail:
742 	blkif_free(info, 0);
743 	return err;
744 }
745 
746 
747 /* Common code used when first setting up, and when resuming. */
748 static int talk_to_blkback(struct xenbus_device *dev,
749 			   struct blkfront_info *info)
750 {
751 	const char *message = NULL;
752 	struct xenbus_transaction xbt;
753 	int err;
754 
755 	/* Create shared ring, alloc event channel. */
756 	err = setup_blkring(dev, info);
757 	if (err)
758 		goto out;
759 
760 again:
761 	err = xenbus_transaction_start(&xbt);
762 	if (err) {
763 		xenbus_dev_fatal(dev, err, "starting transaction");
764 		goto destroy_blkring;
765 	}
766 
767 	err = xenbus_printf(xbt, dev->nodename,
768 			    "ring-ref", "%u", info->ring_ref);
769 	if (err) {
770 		message = "writing ring-ref";
771 		goto abort_transaction;
772 	}
773 	err = xenbus_printf(xbt, dev->nodename,
774 			    "event-channel", "%u", info->evtchn);
775 	if (err) {
776 		message = "writing event-channel";
777 		goto abort_transaction;
778 	}
779 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
780 			    XEN_IO_PROTO_ABI_NATIVE);
781 	if (err) {
782 		message = "writing protocol";
783 		goto abort_transaction;
784 	}
785 
786 	err = xenbus_transaction_end(xbt, 0);
787 	if (err) {
788 		if (err == -EAGAIN)
789 			goto again;
790 		xenbus_dev_fatal(dev, err, "completing transaction");
791 		goto destroy_blkring;
792 	}
793 
794 	xenbus_switch_state(dev, XenbusStateInitialised);
795 
796 	return 0;
797 
798  abort_transaction:
799 	xenbus_transaction_end(xbt, 1);
800 	if (message)
801 		xenbus_dev_fatal(dev, err, "%s", message);
802  destroy_blkring:
803 	blkif_free(info, 0);
804  out:
805 	return err;
806 }
807 
808 /**
809  * Entry point to this code when a new device is created.  Allocate the basic
810  * structures and the ring buffer for communication with the backend, and
811  * inform the backend of the appropriate details for those.  Switch to
812  * Initialised state.
813  */
814 static int blkfront_probe(struct xenbus_device *dev,
815 			  const struct xenbus_device_id *id)
816 {
817 	int err, vdevice, i;
818 	struct blkfront_info *info;
819 
820 	/* FIXME: Use dynamic device id if this is not set. */
821 	err = xenbus_scanf(XBT_NIL, dev->nodename,
822 			   "virtual-device", "%i", &vdevice);
823 	if (err != 1) {
824 		/* go looking in the extended area instead */
825 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
826 				   "%i", &vdevice);
827 		if (err != 1) {
828 			xenbus_dev_fatal(dev, err, "reading virtual-device");
829 			return err;
830 		}
831 	}
832 
833 	if (xen_hvm_domain()) {
834 		char *type;
835 		int len;
836 		/* no unplug has been done: do not hook devices != xen vbds */
837 		if (xen_platform_pci_unplug & XEN_UNPLUG_IGNORE) {
838 			int major;
839 
840 			if (!VDEV_IS_EXTENDED(vdevice))
841 				major = BLKIF_MAJOR(vdevice);
842 			else
843 				major = XENVBD_MAJOR;
844 
845 			if (major != XENVBD_MAJOR) {
846 				printk(KERN_INFO
847 						"%s: HVM does not support vbd %d as xen block device\n",
848 						__FUNCTION__, vdevice);
849 				return -ENODEV;
850 			}
851 		}
852 		/* do not create a PV cdrom device if we are an HVM guest */
853 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
854 		if (IS_ERR(type))
855 			return -ENODEV;
856 		if (strncmp(type, "cdrom", 5) == 0) {
857 			kfree(type);
858 			return -ENODEV;
859 		}
860 		kfree(type);
861 	}
862 	info = kzalloc(sizeof(*info), GFP_KERNEL);
863 	if (!info) {
864 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
865 		return -ENOMEM;
866 	}
867 
868 	mutex_init(&info->mutex);
869 	info->xbdev = dev;
870 	info->vdevice = vdevice;
871 	info->connected = BLKIF_STATE_DISCONNECTED;
872 	INIT_WORK(&info->work, blkif_restart_queue);
873 
874 	for (i = 0; i < BLK_RING_SIZE; i++)
875 		info->shadow[i].req.id = i+1;
876 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
877 
878 	/* Front end dir is a number, which is used as the id. */
879 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
880 	dev_set_drvdata(&dev->dev, info);
881 
882 	err = talk_to_blkback(dev, info);
883 	if (err) {
884 		kfree(info);
885 		dev_set_drvdata(&dev->dev, NULL);
886 		return err;
887 	}
888 
889 	return 0;
890 }
891 
892 
893 static int blkif_recover(struct blkfront_info *info)
894 {
895 	int i;
896 	struct blkif_request *req;
897 	struct blk_shadow *copy;
898 	int j;
899 
900 	/* Stage 1: Make a safe copy of the shadow state. */
901 	copy = kmalloc(sizeof(info->shadow),
902 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
903 	if (!copy)
904 		return -ENOMEM;
905 	memcpy(copy, info->shadow, sizeof(info->shadow));
906 
907 	/* Stage 2: Set up free list. */
908 	memset(&info->shadow, 0, sizeof(info->shadow));
909 	for (i = 0; i < BLK_RING_SIZE; i++)
910 		info->shadow[i].req.id = i+1;
911 	info->shadow_free = info->ring.req_prod_pvt;
912 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
913 
914 	/* Stage 3: Find pending requests and requeue them. */
915 	for (i = 0; i < BLK_RING_SIZE; i++) {
916 		/* Not in use? */
917 		if (copy[i].request == 0)
918 			continue;
919 
920 		/* Grab a request slot and copy shadow state into it. */
921 		req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
922 		*req = copy[i].req;
923 
924 		/* We get a new request id, and must reset the shadow state. */
925 		req->id = get_id_from_freelist(info);
926 		memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
927 
928 		/* Rewrite any grant references invalidated by susp/resume. */
929 		for (j = 0; j < req->nr_segments; j++)
930 			gnttab_grant_foreign_access_ref(
931 				req->seg[j].gref,
932 				info->xbdev->otherend_id,
933 				pfn_to_mfn(info->shadow[req->id].frame[j]),
934 				rq_data_dir(
935 					(struct request *)
936 					info->shadow[req->id].request));
937 		info->shadow[req->id].req = *req;
938 
939 		info->ring.req_prod_pvt++;
940 	}
941 
942 	kfree(copy);
943 
944 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
945 
946 	spin_lock_irq(&blkif_io_lock);
947 
948 	/* Now safe for us to use the shared ring */
949 	info->connected = BLKIF_STATE_CONNECTED;
950 
951 	/* Send off requeued requests */
952 	flush_requests(info);
953 
954 	/* Kick any other new requests queued since we resumed */
955 	kick_pending_request_queues(info);
956 
957 	spin_unlock_irq(&blkif_io_lock);
958 
959 	return 0;
960 }
961 
962 /**
963  * We are reconnecting to the backend, due to a suspend/resume, or a backend
964  * driver restart.  We tear down our blkif structure and recreate it, but
965  * leave the device-layer structures intact so that this is transparent to the
966  * rest of the kernel.
967  */
968 static int blkfront_resume(struct xenbus_device *dev)
969 {
970 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
971 	int err;
972 
973 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
974 
975 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
976 
977 	err = talk_to_blkback(dev, info);
978 	if (info->connected == BLKIF_STATE_SUSPENDED && !err)
979 		err = blkif_recover(info);
980 
981 	return err;
982 }
983 
984 static void
985 blkfront_closing(struct blkfront_info *info)
986 {
987 	struct xenbus_device *xbdev = info->xbdev;
988 	struct block_device *bdev = NULL;
989 
990 	mutex_lock(&info->mutex);
991 
992 	if (xbdev->state == XenbusStateClosing) {
993 		mutex_unlock(&info->mutex);
994 		return;
995 	}
996 
997 	if (info->gd)
998 		bdev = bdget_disk(info->gd, 0);
999 
1000 	mutex_unlock(&info->mutex);
1001 
1002 	if (!bdev) {
1003 		xenbus_frontend_closed(xbdev);
1004 		return;
1005 	}
1006 
1007 	mutex_lock(&bdev->bd_mutex);
1008 
1009 	if (bdev->bd_openers) {
1010 		xenbus_dev_error(xbdev, -EBUSY,
1011 				 "Device in use; refusing to close");
1012 		xenbus_switch_state(xbdev, XenbusStateClosing);
1013 	} else {
1014 		xlvbd_release_gendisk(info);
1015 		xenbus_frontend_closed(xbdev);
1016 	}
1017 
1018 	mutex_unlock(&bdev->bd_mutex);
1019 	bdput(bdev);
1020 }
1021 
1022 /*
1023  * Invoked when the backend is finally 'ready' (and has told produced
1024  * the details about the physical device - #sectors, size, etc).
1025  */
1026 static void blkfront_connect(struct blkfront_info *info)
1027 {
1028 	unsigned long long sectors;
1029 	unsigned long sector_size;
1030 	unsigned int binfo;
1031 	int err;
1032 	int barrier;
1033 
1034 	switch (info->connected) {
1035 	case BLKIF_STATE_CONNECTED:
1036 		/*
1037 		 * Potentially, the back-end may be signalling
1038 		 * a capacity change; update the capacity.
1039 		 */
1040 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1041 				   "sectors", "%Lu", &sectors);
1042 		if (XENBUS_EXIST_ERR(err))
1043 			return;
1044 		printk(KERN_INFO "Setting capacity to %Lu\n",
1045 		       sectors);
1046 		set_capacity(info->gd, sectors);
1047 		revalidate_disk(info->gd);
1048 
1049 		/* fall through */
1050 	case BLKIF_STATE_SUSPENDED:
1051 		return;
1052 
1053 	default:
1054 		break;
1055 	}
1056 
1057 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1058 		__func__, info->xbdev->otherend);
1059 
1060 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1061 			    "sectors", "%llu", &sectors,
1062 			    "info", "%u", &binfo,
1063 			    "sector-size", "%lu", &sector_size,
1064 			    NULL);
1065 	if (err) {
1066 		xenbus_dev_fatal(info->xbdev, err,
1067 				 "reading backend fields at %s",
1068 				 info->xbdev->otherend);
1069 		return;
1070 	}
1071 
1072 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1073 			    "feature-barrier", "%lu", &barrier,
1074 			    NULL);
1075 
1076 	/*
1077 	 * If there's no "feature-barrier" defined, then it means
1078 	 * we're dealing with a very old backend which writes
1079 	 * synchronously; draining will do what needs to get done.
1080 	 *
1081 	 * If there are barriers, then we can do full queued writes
1082 	 * with tagged barriers.
1083 	 *
1084 	 * If barriers are not supported, then there's no much we can
1085 	 * do, so just set ordering to NONE.
1086 	 */
1087 	if (err)
1088 		info->feature_barrier = QUEUE_ORDERED_DRAIN;
1089 	else if (barrier)
1090 		info->feature_barrier = QUEUE_ORDERED_TAG;
1091 	else
1092 		info->feature_barrier = QUEUE_ORDERED_NONE;
1093 
1094 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1095 	if (err) {
1096 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1097 				 info->xbdev->otherend);
1098 		return;
1099 	}
1100 
1101 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1102 
1103 	/* Kick pending requests. */
1104 	spin_lock_irq(&blkif_io_lock);
1105 	info->connected = BLKIF_STATE_CONNECTED;
1106 	kick_pending_request_queues(info);
1107 	spin_unlock_irq(&blkif_io_lock);
1108 
1109 	add_disk(info->gd);
1110 
1111 	info->is_ready = 1;
1112 }
1113 
1114 /**
1115  * Callback received when the backend's state changes.
1116  */
1117 static void blkback_changed(struct xenbus_device *dev,
1118 			    enum xenbus_state backend_state)
1119 {
1120 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1121 
1122 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1123 
1124 	switch (backend_state) {
1125 	case XenbusStateInitialising:
1126 	case XenbusStateInitWait:
1127 	case XenbusStateInitialised:
1128 	case XenbusStateUnknown:
1129 	case XenbusStateClosed:
1130 		break;
1131 
1132 	case XenbusStateConnected:
1133 		blkfront_connect(info);
1134 		break;
1135 
1136 	case XenbusStateClosing:
1137 		blkfront_closing(info);
1138 		break;
1139 	}
1140 }
1141 
1142 static int blkfront_remove(struct xenbus_device *xbdev)
1143 {
1144 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1145 	struct block_device *bdev = NULL;
1146 	struct gendisk *disk;
1147 
1148 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1149 
1150 	blkif_free(info, 0);
1151 
1152 	mutex_lock(&info->mutex);
1153 
1154 	disk = info->gd;
1155 	if (disk)
1156 		bdev = bdget_disk(disk, 0);
1157 
1158 	info->xbdev = NULL;
1159 	mutex_unlock(&info->mutex);
1160 
1161 	if (!bdev) {
1162 		kfree(info);
1163 		return 0;
1164 	}
1165 
1166 	/*
1167 	 * The xbdev was removed before we reached the Closed
1168 	 * state. See if it's safe to remove the disk. If the bdev
1169 	 * isn't closed yet, we let release take care of it.
1170 	 */
1171 
1172 	mutex_lock(&bdev->bd_mutex);
1173 	info = disk->private_data;
1174 
1175 	dev_warn(disk_to_dev(disk),
1176 		 "%s was hot-unplugged, %d stale handles\n",
1177 		 xbdev->nodename, bdev->bd_openers);
1178 
1179 	if (info && !bdev->bd_openers) {
1180 		xlvbd_release_gendisk(info);
1181 		disk->private_data = NULL;
1182 		kfree(info);
1183 	}
1184 
1185 	mutex_unlock(&bdev->bd_mutex);
1186 	bdput(bdev);
1187 
1188 	return 0;
1189 }
1190 
1191 static int blkfront_is_ready(struct xenbus_device *dev)
1192 {
1193 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1194 
1195 	return info->is_ready && info->xbdev;
1196 }
1197 
1198 static int blkif_open(struct block_device *bdev, fmode_t mode)
1199 {
1200 	struct gendisk *disk = bdev->bd_disk;
1201 	struct blkfront_info *info;
1202 	int err = 0;
1203 
1204 	lock_kernel();
1205 
1206 	info = disk->private_data;
1207 	if (!info) {
1208 		/* xbdev gone */
1209 		err = -ERESTARTSYS;
1210 		goto out;
1211 	}
1212 
1213 	mutex_lock(&info->mutex);
1214 
1215 	if (!info->gd)
1216 		/* xbdev is closed */
1217 		err = -ERESTARTSYS;
1218 
1219 	mutex_unlock(&info->mutex);
1220 
1221 out:
1222 	unlock_kernel();
1223 	return err;
1224 }
1225 
1226 static int blkif_release(struct gendisk *disk, fmode_t mode)
1227 {
1228 	struct blkfront_info *info = disk->private_data;
1229 	struct block_device *bdev;
1230 	struct xenbus_device *xbdev;
1231 
1232 	lock_kernel();
1233 
1234 	bdev = bdget_disk(disk, 0);
1235 	bdput(bdev);
1236 
1237 	if (bdev->bd_openers)
1238 		goto out;
1239 
1240 	/*
1241 	 * Check if we have been instructed to close. We will have
1242 	 * deferred this request, because the bdev was still open.
1243 	 */
1244 
1245 	mutex_lock(&info->mutex);
1246 	xbdev = info->xbdev;
1247 
1248 	if (xbdev && xbdev->state == XenbusStateClosing) {
1249 		/* pending switch to state closed */
1250 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1251 		xlvbd_release_gendisk(info);
1252 		xenbus_frontend_closed(info->xbdev);
1253  	}
1254 
1255 	mutex_unlock(&info->mutex);
1256 
1257 	if (!xbdev) {
1258 		/* sudden device removal */
1259 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1260 		xlvbd_release_gendisk(info);
1261 		disk->private_data = NULL;
1262 		kfree(info);
1263 	}
1264 
1265 out:
1266 	unlock_kernel();
1267 	return 0;
1268 }
1269 
1270 static const struct block_device_operations xlvbd_block_fops =
1271 {
1272 	.owner = THIS_MODULE,
1273 	.open = blkif_open,
1274 	.release = blkif_release,
1275 	.getgeo = blkif_getgeo,
1276 	.ioctl = blkif_ioctl,
1277 };
1278 
1279 
1280 static const struct xenbus_device_id blkfront_ids[] = {
1281 	{ "vbd" },
1282 	{ "" }
1283 };
1284 
1285 static struct xenbus_driver blkfront = {
1286 	.name = "vbd",
1287 	.owner = THIS_MODULE,
1288 	.ids = blkfront_ids,
1289 	.probe = blkfront_probe,
1290 	.remove = blkfront_remove,
1291 	.resume = blkfront_resume,
1292 	.otherend_changed = blkback_changed,
1293 	.is_ready = blkfront_is_ready,
1294 };
1295 
1296 static int __init xlblk_init(void)
1297 {
1298 	if (!xen_domain())
1299 		return -ENODEV;
1300 
1301 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1302 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1303 		       XENVBD_MAJOR, DEV_NAME);
1304 		return -ENODEV;
1305 	}
1306 
1307 	return xenbus_register_frontend(&blkfront);
1308 }
1309 module_init(xlblk_init);
1310 
1311 
1312 static void __exit xlblk_exit(void)
1313 {
1314 	return xenbus_unregister_driver(&blkfront);
1315 }
1316 module_exit(xlblk_exit);
1317 
1318 MODULE_DESCRIPTION("Xen virtual block device frontend");
1319 MODULE_LICENSE("GPL");
1320 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1321 MODULE_ALIAS("xen:vbd");
1322 MODULE_ALIAS("xenblk");
1323