xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 9d749629)
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/mutex.h>
45 #include <linux/scatterlist.h>
46 #include <linux/bitmap.h>
47 #include <linux/llist.h>
48 
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
53 #include <xen/page.h>
54 #include <xen/platform_pci.h>
55 
56 #include <xen/interface/grant_table.h>
57 #include <xen/interface/io/blkif.h>
58 #include <xen/interface/io/protocols.h>
59 
60 #include <asm/xen/hypervisor.h>
61 
62 enum blkif_state {
63 	BLKIF_STATE_DISCONNECTED,
64 	BLKIF_STATE_CONNECTED,
65 	BLKIF_STATE_SUSPENDED,
66 };
67 
68 struct grant {
69 	grant_ref_t gref;
70 	unsigned long pfn;
71 	struct llist_node node;
72 };
73 
74 struct blk_shadow {
75 	struct blkif_request req;
76 	struct request *request;
77 	unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
78 	struct grant *grants_used[BLKIF_MAX_SEGMENTS_PER_REQUEST];
79 };
80 
81 static DEFINE_MUTEX(blkfront_mutex);
82 static const struct block_device_operations xlvbd_block_fops;
83 
84 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
85 
86 /*
87  * We have one of these per vbd, whether ide, scsi or 'other'.  They
88  * hang in private_data off the gendisk structure. We may end up
89  * putting all kinds of interesting stuff here :-)
90  */
91 struct blkfront_info
92 {
93 	spinlock_t io_lock;
94 	struct mutex mutex;
95 	struct xenbus_device *xbdev;
96 	struct gendisk *gd;
97 	int vdevice;
98 	blkif_vdev_t handle;
99 	enum blkif_state connected;
100 	int ring_ref;
101 	struct blkif_front_ring ring;
102 	struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
103 	unsigned int evtchn, irq;
104 	struct request_queue *rq;
105 	struct work_struct work;
106 	struct gnttab_free_callback callback;
107 	struct blk_shadow shadow[BLK_RING_SIZE];
108 	struct llist_head persistent_gnts;
109 	unsigned int persistent_gnts_c;
110 	unsigned long shadow_free;
111 	unsigned int feature_flush;
112 	unsigned int flush_op;
113 	unsigned int feature_discard:1;
114 	unsigned int feature_secdiscard:1;
115 	unsigned int discard_granularity;
116 	unsigned int discard_alignment;
117 	unsigned int feature_persistent:1;
118 	int is_ready;
119 };
120 
121 static unsigned int nr_minors;
122 static unsigned long *minors;
123 static DEFINE_SPINLOCK(minor_lock);
124 
125 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
126 	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
127 #define GRANT_INVALID_REF	0
128 
129 #define PARTS_PER_DISK		16
130 #define PARTS_PER_EXT_DISK      256
131 
132 #define BLKIF_MAJOR(dev) ((dev)>>8)
133 #define BLKIF_MINOR(dev) ((dev) & 0xff)
134 
135 #define EXT_SHIFT 28
136 #define EXTENDED (1<<EXT_SHIFT)
137 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
138 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
139 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
140 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
141 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
142 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
143 
144 #define DEV_NAME	"xvd"	/* name in /dev */
145 
146 static int get_id_from_freelist(struct blkfront_info *info)
147 {
148 	unsigned long free = info->shadow_free;
149 	BUG_ON(free >= BLK_RING_SIZE);
150 	info->shadow_free = info->shadow[free].req.u.rw.id;
151 	info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
152 	return free;
153 }
154 
155 static int add_id_to_freelist(struct blkfront_info *info,
156 			       unsigned long id)
157 {
158 	if (info->shadow[id].req.u.rw.id != id)
159 		return -EINVAL;
160 	if (info->shadow[id].request == NULL)
161 		return -EINVAL;
162 	info->shadow[id].req.u.rw.id  = info->shadow_free;
163 	info->shadow[id].request = NULL;
164 	info->shadow_free = id;
165 	return 0;
166 }
167 
168 static const char *op_name(int op)
169 {
170 	static const char *const names[] = {
171 		[BLKIF_OP_READ] = "read",
172 		[BLKIF_OP_WRITE] = "write",
173 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
174 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
175 		[BLKIF_OP_DISCARD] = "discard" };
176 
177 	if (op < 0 || op >= ARRAY_SIZE(names))
178 		return "unknown";
179 
180 	if (!names[op])
181 		return "reserved";
182 
183 	return names[op];
184 }
185 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
186 {
187 	unsigned int end = minor + nr;
188 	int rc;
189 
190 	if (end > nr_minors) {
191 		unsigned long *bitmap, *old;
192 
193 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
194 				 GFP_KERNEL);
195 		if (bitmap == NULL)
196 			return -ENOMEM;
197 
198 		spin_lock(&minor_lock);
199 		if (end > nr_minors) {
200 			old = minors;
201 			memcpy(bitmap, minors,
202 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
203 			minors = bitmap;
204 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
205 		} else
206 			old = bitmap;
207 		spin_unlock(&minor_lock);
208 		kfree(old);
209 	}
210 
211 	spin_lock(&minor_lock);
212 	if (find_next_bit(minors, end, minor) >= end) {
213 		bitmap_set(minors, minor, nr);
214 		rc = 0;
215 	} else
216 		rc = -EBUSY;
217 	spin_unlock(&minor_lock);
218 
219 	return rc;
220 }
221 
222 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
223 {
224 	unsigned int end = minor + nr;
225 
226 	BUG_ON(end > nr_minors);
227 	spin_lock(&minor_lock);
228 	bitmap_clear(minors,  minor, nr);
229 	spin_unlock(&minor_lock);
230 }
231 
232 static void blkif_restart_queue_callback(void *arg)
233 {
234 	struct blkfront_info *info = (struct blkfront_info *)arg;
235 	schedule_work(&info->work);
236 }
237 
238 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
239 {
240 	/* We don't have real geometry info, but let's at least return
241 	   values consistent with the size of the device */
242 	sector_t nsect = get_capacity(bd->bd_disk);
243 	sector_t cylinders = nsect;
244 
245 	hg->heads = 0xff;
246 	hg->sectors = 0x3f;
247 	sector_div(cylinders, hg->heads * hg->sectors);
248 	hg->cylinders = cylinders;
249 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
250 		hg->cylinders = 0xffff;
251 	return 0;
252 }
253 
254 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
255 		       unsigned command, unsigned long argument)
256 {
257 	struct blkfront_info *info = bdev->bd_disk->private_data;
258 	int i;
259 
260 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
261 		command, (long)argument);
262 
263 	switch (command) {
264 	case CDROMMULTISESSION:
265 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
266 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
267 			if (put_user(0, (char __user *)(argument + i)))
268 				return -EFAULT;
269 		return 0;
270 
271 	case CDROM_GET_CAPABILITY: {
272 		struct gendisk *gd = info->gd;
273 		if (gd->flags & GENHD_FL_CD)
274 			return 0;
275 		return -EINVAL;
276 	}
277 
278 	default:
279 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
280 		  command);*/
281 		return -EINVAL; /* same return as native Linux */
282 	}
283 
284 	return 0;
285 }
286 
287 /*
288  * Generate a Xen blkfront IO request from a blk layer request.  Reads
289  * and writes are handled as expected.
290  *
291  * @req: a request struct
292  */
293 static int blkif_queue_request(struct request *req)
294 {
295 	struct blkfront_info *info = req->rq_disk->private_data;
296 	unsigned long buffer_mfn;
297 	struct blkif_request *ring_req;
298 	unsigned long id;
299 	unsigned int fsect, lsect;
300 	int i, ref;
301 
302 	/*
303 	 * Used to store if we are able to queue the request by just using
304 	 * existing persistent grants, or if we have to get new grants,
305 	 * as there are not sufficiently many free.
306 	 */
307 	bool new_persistent_gnts;
308 	grant_ref_t gref_head;
309 	struct page *granted_page;
310 	struct grant *gnt_list_entry = NULL;
311 	struct scatterlist *sg;
312 
313 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
314 		return 1;
315 
316 	/* Check if we have enought grants to allocate a requests */
317 	if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
318 		new_persistent_gnts = 1;
319 		if (gnttab_alloc_grant_references(
320 		    BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
321 		    &gref_head) < 0) {
322 			gnttab_request_free_callback(
323 				&info->callback,
324 				blkif_restart_queue_callback,
325 				info,
326 				BLKIF_MAX_SEGMENTS_PER_REQUEST);
327 			return 1;
328 		}
329 	} else
330 		new_persistent_gnts = 0;
331 
332 	/* Fill out a communications ring structure. */
333 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
334 	id = get_id_from_freelist(info);
335 	info->shadow[id].request = req;
336 
337 	ring_req->u.rw.id = id;
338 	ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
339 	ring_req->u.rw.handle = info->handle;
340 
341 	ring_req->operation = rq_data_dir(req) ?
342 		BLKIF_OP_WRITE : BLKIF_OP_READ;
343 
344 	if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
345 		/*
346 		 * Ideally we can do an unordered flush-to-disk. In case the
347 		 * backend onlysupports barriers, use that. A barrier request
348 		 * a superset of FUA, so we can implement it the same
349 		 * way.  (It's also a FLUSH+FUA, since it is
350 		 * guaranteed ordered WRT previous writes.)
351 		 */
352 		ring_req->operation = info->flush_op;
353 	}
354 
355 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
356 		/* id, sector_number and handle are set above. */
357 		ring_req->operation = BLKIF_OP_DISCARD;
358 		ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
359 		if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
360 			ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
361 		else
362 			ring_req->u.discard.flag = 0;
363 	} else {
364 		ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
365 							   info->sg);
366 		BUG_ON(ring_req->u.rw.nr_segments >
367 		       BLKIF_MAX_SEGMENTS_PER_REQUEST);
368 
369 		for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
370 			fsect = sg->offset >> 9;
371 			lsect = fsect + (sg->length >> 9) - 1;
372 
373 			if (info->persistent_gnts_c) {
374 				BUG_ON(llist_empty(&info->persistent_gnts));
375 				gnt_list_entry = llist_entry(
376 					llist_del_first(&info->persistent_gnts),
377 					struct grant, node);
378 
379 				ref = gnt_list_entry->gref;
380 				buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
381 				info->persistent_gnts_c--;
382 			} else {
383 				ref = gnttab_claim_grant_reference(&gref_head);
384 				BUG_ON(ref == -ENOSPC);
385 
386 				gnt_list_entry =
387 					kmalloc(sizeof(struct grant),
388 							 GFP_ATOMIC);
389 				if (!gnt_list_entry)
390 					return -ENOMEM;
391 
392 				granted_page = alloc_page(GFP_ATOMIC);
393 				if (!granted_page) {
394 					kfree(gnt_list_entry);
395 					return -ENOMEM;
396 				}
397 
398 				gnt_list_entry->pfn =
399 					page_to_pfn(granted_page);
400 				gnt_list_entry->gref = ref;
401 
402 				buffer_mfn = pfn_to_mfn(page_to_pfn(
403 								granted_page));
404 				gnttab_grant_foreign_access_ref(ref,
405 					info->xbdev->otherend_id,
406 					buffer_mfn, 0);
407 			}
408 
409 			info->shadow[id].grants_used[i] = gnt_list_entry;
410 
411 			if (rq_data_dir(req)) {
412 				char *bvec_data;
413 				void *shared_data;
414 
415 				BUG_ON(sg->offset + sg->length > PAGE_SIZE);
416 
417 				shared_data = kmap_atomic(
418 					pfn_to_page(gnt_list_entry->pfn));
419 				bvec_data = kmap_atomic(sg_page(sg));
420 
421 				/*
422 				 * this does not wipe data stored outside the
423 				 * range sg->offset..sg->offset+sg->length.
424 				 * Therefore, blkback *could* see data from
425 				 * previous requests. This is OK as long as
426 				 * persistent grants are shared with just one
427 				 * domain. It may need refactoring if this
428 				 * changes
429 				 */
430 				memcpy(shared_data + sg->offset,
431 				       bvec_data   + sg->offset,
432 				       sg->length);
433 
434 				kunmap_atomic(bvec_data);
435 				kunmap_atomic(shared_data);
436 			}
437 
438 			info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
439 			ring_req->u.rw.seg[i] =
440 					(struct blkif_request_segment) {
441 						.gref       = ref,
442 						.first_sect = fsect,
443 						.last_sect  = lsect };
444 		}
445 	}
446 
447 	info->ring.req_prod_pvt++;
448 
449 	/* Keep a private copy so we can reissue requests when recovering. */
450 	info->shadow[id].req = *ring_req;
451 
452 	if (new_persistent_gnts)
453 		gnttab_free_grant_references(gref_head);
454 
455 	return 0;
456 }
457 
458 
459 static inline void flush_requests(struct blkfront_info *info)
460 {
461 	int notify;
462 
463 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
464 
465 	if (notify)
466 		notify_remote_via_irq(info->irq);
467 }
468 
469 /*
470  * do_blkif_request
471  *  read a block; request is in a request queue
472  */
473 static void do_blkif_request(struct request_queue *rq)
474 {
475 	struct blkfront_info *info = NULL;
476 	struct request *req;
477 	int queued;
478 
479 	pr_debug("Entered do_blkif_request\n");
480 
481 	queued = 0;
482 
483 	while ((req = blk_peek_request(rq)) != NULL) {
484 		info = req->rq_disk->private_data;
485 
486 		if (RING_FULL(&info->ring))
487 			goto wait;
488 
489 		blk_start_request(req);
490 
491 		if ((req->cmd_type != REQ_TYPE_FS) ||
492 		    ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
493 		    !info->flush_op)) {
494 			__blk_end_request_all(req, -EIO);
495 			continue;
496 		}
497 
498 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
499 			 "(%u/%u) buffer:%p [%s]\n",
500 			 req, req->cmd, (unsigned long)blk_rq_pos(req),
501 			 blk_rq_cur_sectors(req), blk_rq_sectors(req),
502 			 req->buffer, rq_data_dir(req) ? "write" : "read");
503 
504 		if (blkif_queue_request(req)) {
505 			blk_requeue_request(rq, req);
506 wait:
507 			/* Avoid pointless unplugs. */
508 			blk_stop_queue(rq);
509 			break;
510 		}
511 
512 		queued++;
513 	}
514 
515 	if (queued != 0)
516 		flush_requests(info);
517 }
518 
519 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
520 {
521 	struct request_queue *rq;
522 	struct blkfront_info *info = gd->private_data;
523 
524 	rq = blk_init_queue(do_blkif_request, &info->io_lock);
525 	if (rq == NULL)
526 		return -1;
527 
528 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
529 
530 	if (info->feature_discard) {
531 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
532 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
533 		rq->limits.discard_granularity = info->discard_granularity;
534 		rq->limits.discard_alignment = info->discard_alignment;
535 		if (info->feature_secdiscard)
536 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
537 	}
538 
539 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
540 	blk_queue_logical_block_size(rq, sector_size);
541 	blk_queue_max_hw_sectors(rq, 512);
542 
543 	/* Each segment in a request is up to an aligned page in size. */
544 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
545 	blk_queue_max_segment_size(rq, PAGE_SIZE);
546 
547 	/* Ensure a merged request will fit in a single I/O ring slot. */
548 	blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
549 
550 	/* Make sure buffer addresses are sector-aligned. */
551 	blk_queue_dma_alignment(rq, 511);
552 
553 	/* Make sure we don't use bounce buffers. */
554 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
555 
556 	gd->queue = rq;
557 
558 	return 0;
559 }
560 
561 
562 static void xlvbd_flush(struct blkfront_info *info)
563 {
564 	blk_queue_flush(info->rq, info->feature_flush);
565 	printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
566 	       info->gd->disk_name,
567 	       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
568 		"barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
569 		"flush diskcache" : "barrier or flush"),
570 	       info->feature_flush ? "enabled" : "disabled",
571 	       info->feature_persistent ? "using persistent grants" : "");
572 }
573 
574 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
575 {
576 	int major;
577 	major = BLKIF_MAJOR(vdevice);
578 	*minor = BLKIF_MINOR(vdevice);
579 	switch (major) {
580 		case XEN_IDE0_MAJOR:
581 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
582 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
583 				EMULATED_HD_DISK_MINOR_OFFSET;
584 			break;
585 		case XEN_IDE1_MAJOR:
586 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
587 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
588 				EMULATED_HD_DISK_MINOR_OFFSET;
589 			break;
590 		case XEN_SCSI_DISK0_MAJOR:
591 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
592 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
593 			break;
594 		case XEN_SCSI_DISK1_MAJOR:
595 		case XEN_SCSI_DISK2_MAJOR:
596 		case XEN_SCSI_DISK3_MAJOR:
597 		case XEN_SCSI_DISK4_MAJOR:
598 		case XEN_SCSI_DISK5_MAJOR:
599 		case XEN_SCSI_DISK6_MAJOR:
600 		case XEN_SCSI_DISK7_MAJOR:
601 			*offset = (*minor / PARTS_PER_DISK) +
602 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
603 				EMULATED_SD_DISK_NAME_OFFSET;
604 			*minor = *minor +
605 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
606 				EMULATED_SD_DISK_MINOR_OFFSET;
607 			break;
608 		case XEN_SCSI_DISK8_MAJOR:
609 		case XEN_SCSI_DISK9_MAJOR:
610 		case XEN_SCSI_DISK10_MAJOR:
611 		case XEN_SCSI_DISK11_MAJOR:
612 		case XEN_SCSI_DISK12_MAJOR:
613 		case XEN_SCSI_DISK13_MAJOR:
614 		case XEN_SCSI_DISK14_MAJOR:
615 		case XEN_SCSI_DISK15_MAJOR:
616 			*offset = (*minor / PARTS_PER_DISK) +
617 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
618 				EMULATED_SD_DISK_NAME_OFFSET;
619 			*minor = *minor +
620 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
621 				EMULATED_SD_DISK_MINOR_OFFSET;
622 			break;
623 		case XENVBD_MAJOR:
624 			*offset = *minor / PARTS_PER_DISK;
625 			break;
626 		default:
627 			printk(KERN_WARNING "blkfront: your disk configuration is "
628 					"incorrect, please use an xvd device instead\n");
629 			return -ENODEV;
630 	}
631 	return 0;
632 }
633 
634 static char *encode_disk_name(char *ptr, unsigned int n)
635 {
636 	if (n >= 26)
637 		ptr = encode_disk_name(ptr, n / 26 - 1);
638 	*ptr = 'a' + n % 26;
639 	return ptr + 1;
640 }
641 
642 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
643 			       struct blkfront_info *info,
644 			       u16 vdisk_info, u16 sector_size)
645 {
646 	struct gendisk *gd;
647 	int nr_minors = 1;
648 	int err;
649 	unsigned int offset;
650 	int minor;
651 	int nr_parts;
652 	char *ptr;
653 
654 	BUG_ON(info->gd != NULL);
655 	BUG_ON(info->rq != NULL);
656 
657 	if ((info->vdevice>>EXT_SHIFT) > 1) {
658 		/* this is above the extended range; something is wrong */
659 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
660 		return -ENODEV;
661 	}
662 
663 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
664 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
665 		if (err)
666 			return err;
667  		nr_parts = PARTS_PER_DISK;
668 	} else {
669 		minor = BLKIF_MINOR_EXT(info->vdevice);
670 		nr_parts = PARTS_PER_EXT_DISK;
671 		offset = minor / nr_parts;
672 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
673 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
674 					"emulated IDE disks,\n\t choose an xvd device name"
675 					"from xvde on\n", info->vdevice);
676 	}
677 	if (minor >> MINORBITS) {
678 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
679 			info->vdevice, minor);
680 		return -ENODEV;
681 	}
682 
683 	if ((minor % nr_parts) == 0)
684 		nr_minors = nr_parts;
685 
686 	err = xlbd_reserve_minors(minor, nr_minors);
687 	if (err)
688 		goto out;
689 	err = -ENODEV;
690 
691 	gd = alloc_disk(nr_minors);
692 	if (gd == NULL)
693 		goto release;
694 
695 	strcpy(gd->disk_name, DEV_NAME);
696 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
697 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
698 	if (nr_minors > 1)
699 		*ptr = 0;
700 	else
701 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
702 			 "%d", minor & (nr_parts - 1));
703 
704 	gd->major = XENVBD_MAJOR;
705 	gd->first_minor = minor;
706 	gd->fops = &xlvbd_block_fops;
707 	gd->private_data = info;
708 	gd->driverfs_dev = &(info->xbdev->dev);
709 	set_capacity(gd, capacity);
710 
711 	if (xlvbd_init_blk_queue(gd, sector_size)) {
712 		del_gendisk(gd);
713 		goto release;
714 	}
715 
716 	info->rq = gd->queue;
717 	info->gd = gd;
718 
719 	xlvbd_flush(info);
720 
721 	if (vdisk_info & VDISK_READONLY)
722 		set_disk_ro(gd, 1);
723 
724 	if (vdisk_info & VDISK_REMOVABLE)
725 		gd->flags |= GENHD_FL_REMOVABLE;
726 
727 	if (vdisk_info & VDISK_CDROM)
728 		gd->flags |= GENHD_FL_CD;
729 
730 	return 0;
731 
732  release:
733 	xlbd_release_minors(minor, nr_minors);
734  out:
735 	return err;
736 }
737 
738 static void xlvbd_release_gendisk(struct blkfront_info *info)
739 {
740 	unsigned int minor, nr_minors;
741 	unsigned long flags;
742 
743 	if (info->rq == NULL)
744 		return;
745 
746 	spin_lock_irqsave(&info->io_lock, flags);
747 
748 	/* No more blkif_request(). */
749 	blk_stop_queue(info->rq);
750 
751 	/* No more gnttab callback work. */
752 	gnttab_cancel_free_callback(&info->callback);
753 	spin_unlock_irqrestore(&info->io_lock, flags);
754 
755 	/* Flush gnttab callback work. Must be done with no locks held. */
756 	flush_work(&info->work);
757 
758 	del_gendisk(info->gd);
759 
760 	minor = info->gd->first_minor;
761 	nr_minors = info->gd->minors;
762 	xlbd_release_minors(minor, nr_minors);
763 
764 	blk_cleanup_queue(info->rq);
765 	info->rq = NULL;
766 
767 	put_disk(info->gd);
768 	info->gd = NULL;
769 }
770 
771 static void kick_pending_request_queues(struct blkfront_info *info)
772 {
773 	if (!RING_FULL(&info->ring)) {
774 		/* Re-enable calldowns. */
775 		blk_start_queue(info->rq);
776 		/* Kick things off immediately. */
777 		do_blkif_request(info->rq);
778 	}
779 }
780 
781 static void blkif_restart_queue(struct work_struct *work)
782 {
783 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
784 
785 	spin_lock_irq(&info->io_lock);
786 	if (info->connected == BLKIF_STATE_CONNECTED)
787 		kick_pending_request_queues(info);
788 	spin_unlock_irq(&info->io_lock);
789 }
790 
791 static void blkif_free(struct blkfront_info *info, int suspend)
792 {
793 	struct llist_node *all_gnts;
794 	struct grant *persistent_gnt;
795 	struct llist_node *n;
796 
797 	/* Prevent new requests being issued until we fix things up. */
798 	spin_lock_irq(&info->io_lock);
799 	info->connected = suspend ?
800 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
801 	/* No more blkif_request(). */
802 	if (info->rq)
803 		blk_stop_queue(info->rq);
804 
805 	/* Remove all persistent grants */
806 	if (info->persistent_gnts_c) {
807 		all_gnts = llist_del_all(&info->persistent_gnts);
808 		llist_for_each_entry_safe(persistent_gnt, n, all_gnts, node) {
809 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
810 			__free_page(pfn_to_page(persistent_gnt->pfn));
811 			kfree(persistent_gnt);
812 		}
813 		info->persistent_gnts_c = 0;
814 	}
815 
816 	/* No more gnttab callback work. */
817 	gnttab_cancel_free_callback(&info->callback);
818 	spin_unlock_irq(&info->io_lock);
819 
820 	/* Flush gnttab callback work. Must be done with no locks held. */
821 	flush_work(&info->work);
822 
823 	/* Free resources associated with old device channel. */
824 	if (info->ring_ref != GRANT_INVALID_REF) {
825 		gnttab_end_foreign_access(info->ring_ref, 0,
826 					  (unsigned long)info->ring.sring);
827 		info->ring_ref = GRANT_INVALID_REF;
828 		info->ring.sring = NULL;
829 	}
830 	if (info->irq)
831 		unbind_from_irqhandler(info->irq, info);
832 	info->evtchn = info->irq = 0;
833 
834 }
835 
836 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
837 			     struct blkif_response *bret)
838 {
839 	int i = 0;
840 	struct bio_vec *bvec;
841 	struct req_iterator iter;
842 	unsigned long flags;
843 	char *bvec_data;
844 	void *shared_data;
845 	unsigned int offset = 0;
846 
847 	if (bret->operation == BLKIF_OP_READ) {
848 		/*
849 		 * Copy the data received from the backend into the bvec.
850 		 * Since bv_offset can be different than 0, and bv_len different
851 		 * than PAGE_SIZE, we have to keep track of the current offset,
852 		 * to be sure we are copying the data from the right shared page.
853 		 */
854 		rq_for_each_segment(bvec, s->request, iter) {
855 			BUG_ON((bvec->bv_offset + bvec->bv_len) > PAGE_SIZE);
856 			if (bvec->bv_offset < offset)
857 				i++;
858 			BUG_ON(i >= s->req.u.rw.nr_segments);
859 			shared_data = kmap_atomic(
860 				pfn_to_page(s->grants_used[i]->pfn));
861 			bvec_data = bvec_kmap_irq(bvec, &flags);
862 			memcpy(bvec_data, shared_data + bvec->bv_offset,
863 				bvec->bv_len);
864 			bvec_kunmap_irq(bvec_data, &flags);
865 			kunmap_atomic(shared_data);
866 			offset = bvec->bv_offset + bvec->bv_len;
867 		}
868 	}
869 	/* Add the persistent grant into the list of free grants */
870 	for (i = 0; i < s->req.u.rw.nr_segments; i++) {
871 		llist_add(&s->grants_used[i]->node, &info->persistent_gnts);
872 		info->persistent_gnts_c++;
873 	}
874 }
875 
876 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
877 {
878 	struct request *req;
879 	struct blkif_response *bret;
880 	RING_IDX i, rp;
881 	unsigned long flags;
882 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
883 	int error;
884 
885 	spin_lock_irqsave(&info->io_lock, flags);
886 
887 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
888 		spin_unlock_irqrestore(&info->io_lock, flags);
889 		return IRQ_HANDLED;
890 	}
891 
892  again:
893 	rp = info->ring.sring->rsp_prod;
894 	rmb(); /* Ensure we see queued responses up to 'rp'. */
895 
896 	for (i = info->ring.rsp_cons; i != rp; i++) {
897 		unsigned long id;
898 
899 		bret = RING_GET_RESPONSE(&info->ring, i);
900 		id   = bret->id;
901 		/*
902 		 * The backend has messed up and given us an id that we would
903 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
904 		 * look in get_id_from_freelist.
905 		 */
906 		if (id >= BLK_RING_SIZE) {
907 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
908 			     info->gd->disk_name, op_name(bret->operation), id);
909 			/* We can't safely get the 'struct request' as
910 			 * the id is busted. */
911 			continue;
912 		}
913 		req  = info->shadow[id].request;
914 
915 		if (bret->operation != BLKIF_OP_DISCARD)
916 			blkif_completion(&info->shadow[id], info, bret);
917 
918 		if (add_id_to_freelist(info, id)) {
919 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
920 			     info->gd->disk_name, op_name(bret->operation), id);
921 			continue;
922 		}
923 
924 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
925 		switch (bret->operation) {
926 		case BLKIF_OP_DISCARD:
927 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
928 				struct request_queue *rq = info->rq;
929 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
930 					   info->gd->disk_name, op_name(bret->operation));
931 				error = -EOPNOTSUPP;
932 				info->feature_discard = 0;
933 				info->feature_secdiscard = 0;
934 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
935 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
936 			}
937 			__blk_end_request_all(req, error);
938 			break;
939 		case BLKIF_OP_FLUSH_DISKCACHE:
940 		case BLKIF_OP_WRITE_BARRIER:
941 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
942 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
943 				       info->gd->disk_name, op_name(bret->operation));
944 				error = -EOPNOTSUPP;
945 			}
946 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
947 				     info->shadow[id].req.u.rw.nr_segments == 0)) {
948 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
949 				       info->gd->disk_name, op_name(bret->operation));
950 				error = -EOPNOTSUPP;
951 			}
952 			if (unlikely(error)) {
953 				if (error == -EOPNOTSUPP)
954 					error = 0;
955 				info->feature_flush = 0;
956 				info->flush_op = 0;
957 				xlvbd_flush(info);
958 			}
959 			/* fall through */
960 		case BLKIF_OP_READ:
961 		case BLKIF_OP_WRITE:
962 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
963 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
964 					"request: %x\n", bret->status);
965 
966 			__blk_end_request_all(req, error);
967 			break;
968 		default:
969 			BUG();
970 		}
971 	}
972 
973 	info->ring.rsp_cons = i;
974 
975 	if (i != info->ring.req_prod_pvt) {
976 		int more_to_do;
977 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
978 		if (more_to_do)
979 			goto again;
980 	} else
981 		info->ring.sring->rsp_event = i + 1;
982 
983 	kick_pending_request_queues(info);
984 
985 	spin_unlock_irqrestore(&info->io_lock, flags);
986 
987 	return IRQ_HANDLED;
988 }
989 
990 
991 static int setup_blkring(struct xenbus_device *dev,
992 			 struct blkfront_info *info)
993 {
994 	struct blkif_sring *sring;
995 	int err;
996 
997 	info->ring_ref = GRANT_INVALID_REF;
998 
999 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1000 	if (!sring) {
1001 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1002 		return -ENOMEM;
1003 	}
1004 	SHARED_RING_INIT(sring);
1005 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1006 
1007 	sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
1008 
1009 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1010 	if (err < 0) {
1011 		free_page((unsigned long)sring);
1012 		info->ring.sring = NULL;
1013 		goto fail;
1014 	}
1015 	info->ring_ref = err;
1016 
1017 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
1018 	if (err)
1019 		goto fail;
1020 
1021 	err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1022 					"blkif", info);
1023 	if (err <= 0) {
1024 		xenbus_dev_fatal(dev, err,
1025 				 "bind_evtchn_to_irqhandler failed");
1026 		goto fail;
1027 	}
1028 	info->irq = err;
1029 
1030 	return 0;
1031 fail:
1032 	blkif_free(info, 0);
1033 	return err;
1034 }
1035 
1036 
1037 /* Common code used when first setting up, and when resuming. */
1038 static int talk_to_blkback(struct xenbus_device *dev,
1039 			   struct blkfront_info *info)
1040 {
1041 	const char *message = NULL;
1042 	struct xenbus_transaction xbt;
1043 	int err;
1044 
1045 	/* Create shared ring, alloc event channel. */
1046 	err = setup_blkring(dev, info);
1047 	if (err)
1048 		goto out;
1049 
1050 again:
1051 	err = xenbus_transaction_start(&xbt);
1052 	if (err) {
1053 		xenbus_dev_fatal(dev, err, "starting transaction");
1054 		goto destroy_blkring;
1055 	}
1056 
1057 	err = xenbus_printf(xbt, dev->nodename,
1058 			    "ring-ref", "%u", info->ring_ref);
1059 	if (err) {
1060 		message = "writing ring-ref";
1061 		goto abort_transaction;
1062 	}
1063 	err = xenbus_printf(xbt, dev->nodename,
1064 			    "event-channel", "%u", info->evtchn);
1065 	if (err) {
1066 		message = "writing event-channel";
1067 		goto abort_transaction;
1068 	}
1069 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1070 			    XEN_IO_PROTO_ABI_NATIVE);
1071 	if (err) {
1072 		message = "writing protocol";
1073 		goto abort_transaction;
1074 	}
1075 	err = xenbus_printf(xbt, dev->nodename,
1076 			    "feature-persistent", "%u", 1);
1077 	if (err)
1078 		dev_warn(&dev->dev,
1079 			 "writing persistent grants feature to xenbus");
1080 
1081 	err = xenbus_transaction_end(xbt, 0);
1082 	if (err) {
1083 		if (err == -EAGAIN)
1084 			goto again;
1085 		xenbus_dev_fatal(dev, err, "completing transaction");
1086 		goto destroy_blkring;
1087 	}
1088 
1089 	xenbus_switch_state(dev, XenbusStateInitialised);
1090 
1091 	return 0;
1092 
1093  abort_transaction:
1094 	xenbus_transaction_end(xbt, 1);
1095 	if (message)
1096 		xenbus_dev_fatal(dev, err, "%s", message);
1097  destroy_blkring:
1098 	blkif_free(info, 0);
1099  out:
1100 	return err;
1101 }
1102 
1103 /**
1104  * Entry point to this code when a new device is created.  Allocate the basic
1105  * structures and the ring buffer for communication with the backend, and
1106  * inform the backend of the appropriate details for those.  Switch to
1107  * Initialised state.
1108  */
1109 static int blkfront_probe(struct xenbus_device *dev,
1110 			  const struct xenbus_device_id *id)
1111 {
1112 	int err, vdevice, i;
1113 	struct blkfront_info *info;
1114 
1115 	/* FIXME: Use dynamic device id if this is not set. */
1116 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1117 			   "virtual-device", "%i", &vdevice);
1118 	if (err != 1) {
1119 		/* go looking in the extended area instead */
1120 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1121 				   "%i", &vdevice);
1122 		if (err != 1) {
1123 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1124 			return err;
1125 		}
1126 	}
1127 
1128 	if (xen_hvm_domain()) {
1129 		char *type;
1130 		int len;
1131 		/* no unplug has been done: do not hook devices != xen vbds */
1132 		if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1133 			int major;
1134 
1135 			if (!VDEV_IS_EXTENDED(vdevice))
1136 				major = BLKIF_MAJOR(vdevice);
1137 			else
1138 				major = XENVBD_MAJOR;
1139 
1140 			if (major != XENVBD_MAJOR) {
1141 				printk(KERN_INFO
1142 						"%s: HVM does not support vbd %d as xen block device\n",
1143 						__FUNCTION__, vdevice);
1144 				return -ENODEV;
1145 			}
1146 		}
1147 		/* do not create a PV cdrom device if we are an HVM guest */
1148 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1149 		if (IS_ERR(type))
1150 			return -ENODEV;
1151 		if (strncmp(type, "cdrom", 5) == 0) {
1152 			kfree(type);
1153 			return -ENODEV;
1154 		}
1155 		kfree(type);
1156 	}
1157 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1158 	if (!info) {
1159 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1160 		return -ENOMEM;
1161 	}
1162 
1163 	mutex_init(&info->mutex);
1164 	spin_lock_init(&info->io_lock);
1165 	info->xbdev = dev;
1166 	info->vdevice = vdevice;
1167 	init_llist_head(&info->persistent_gnts);
1168 	info->persistent_gnts_c = 0;
1169 	info->connected = BLKIF_STATE_DISCONNECTED;
1170 	INIT_WORK(&info->work, blkif_restart_queue);
1171 
1172 	for (i = 0; i < BLK_RING_SIZE; i++)
1173 		info->shadow[i].req.u.rw.id = i+1;
1174 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1175 
1176 	/* Front end dir is a number, which is used as the id. */
1177 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1178 	dev_set_drvdata(&dev->dev, info);
1179 
1180 	err = talk_to_blkback(dev, info);
1181 	if (err) {
1182 		kfree(info);
1183 		dev_set_drvdata(&dev->dev, NULL);
1184 		return err;
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 
1191 static int blkif_recover(struct blkfront_info *info)
1192 {
1193 	int i;
1194 	struct blkif_request *req;
1195 	struct blk_shadow *copy;
1196 	int j;
1197 
1198 	/* Stage 1: Make a safe copy of the shadow state. */
1199 	copy = kmalloc(sizeof(info->shadow),
1200 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1201 	if (!copy)
1202 		return -ENOMEM;
1203 	memcpy(copy, info->shadow, sizeof(info->shadow));
1204 
1205 	/* Stage 2: Set up free list. */
1206 	memset(&info->shadow, 0, sizeof(info->shadow));
1207 	for (i = 0; i < BLK_RING_SIZE; i++)
1208 		info->shadow[i].req.u.rw.id = i+1;
1209 	info->shadow_free = info->ring.req_prod_pvt;
1210 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1211 
1212 	/* Stage 3: Find pending requests and requeue them. */
1213 	for (i = 0; i < BLK_RING_SIZE; i++) {
1214 		/* Not in use? */
1215 		if (!copy[i].request)
1216 			continue;
1217 
1218 		/* Grab a request slot and copy shadow state into it. */
1219 		req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1220 		*req = copy[i].req;
1221 
1222 		/* We get a new request id, and must reset the shadow state. */
1223 		req->u.rw.id = get_id_from_freelist(info);
1224 		memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1225 
1226 		if (req->operation != BLKIF_OP_DISCARD) {
1227 		/* Rewrite any grant references invalidated by susp/resume. */
1228 			for (j = 0; j < req->u.rw.nr_segments; j++)
1229 				gnttab_grant_foreign_access_ref(
1230 					req->u.rw.seg[j].gref,
1231 					info->xbdev->otherend_id,
1232 					pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1233 					0);
1234 		}
1235 		info->shadow[req->u.rw.id].req = *req;
1236 
1237 		info->ring.req_prod_pvt++;
1238 	}
1239 
1240 	kfree(copy);
1241 
1242 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1243 
1244 	spin_lock_irq(&info->io_lock);
1245 
1246 	/* Now safe for us to use the shared ring */
1247 	info->connected = BLKIF_STATE_CONNECTED;
1248 
1249 	/* Send off requeued requests */
1250 	flush_requests(info);
1251 
1252 	/* Kick any other new requests queued since we resumed */
1253 	kick_pending_request_queues(info);
1254 
1255 	spin_unlock_irq(&info->io_lock);
1256 
1257 	return 0;
1258 }
1259 
1260 /**
1261  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1262  * driver restart.  We tear down our blkif structure and recreate it, but
1263  * leave the device-layer structures intact so that this is transparent to the
1264  * rest of the kernel.
1265  */
1266 static int blkfront_resume(struct xenbus_device *dev)
1267 {
1268 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1269 	int err;
1270 
1271 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1272 
1273 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1274 
1275 	err = talk_to_blkback(dev, info);
1276 	if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1277 		err = blkif_recover(info);
1278 
1279 	return err;
1280 }
1281 
1282 static void
1283 blkfront_closing(struct blkfront_info *info)
1284 {
1285 	struct xenbus_device *xbdev = info->xbdev;
1286 	struct block_device *bdev = NULL;
1287 
1288 	mutex_lock(&info->mutex);
1289 
1290 	if (xbdev->state == XenbusStateClosing) {
1291 		mutex_unlock(&info->mutex);
1292 		return;
1293 	}
1294 
1295 	if (info->gd)
1296 		bdev = bdget_disk(info->gd, 0);
1297 
1298 	mutex_unlock(&info->mutex);
1299 
1300 	if (!bdev) {
1301 		xenbus_frontend_closed(xbdev);
1302 		return;
1303 	}
1304 
1305 	mutex_lock(&bdev->bd_mutex);
1306 
1307 	if (bdev->bd_openers) {
1308 		xenbus_dev_error(xbdev, -EBUSY,
1309 				 "Device in use; refusing to close");
1310 		xenbus_switch_state(xbdev, XenbusStateClosing);
1311 	} else {
1312 		xlvbd_release_gendisk(info);
1313 		xenbus_frontend_closed(xbdev);
1314 	}
1315 
1316 	mutex_unlock(&bdev->bd_mutex);
1317 	bdput(bdev);
1318 }
1319 
1320 static void blkfront_setup_discard(struct blkfront_info *info)
1321 {
1322 	int err;
1323 	char *type;
1324 	unsigned int discard_granularity;
1325 	unsigned int discard_alignment;
1326 	unsigned int discard_secure;
1327 
1328 	type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1329 	if (IS_ERR(type))
1330 		return;
1331 
1332 	info->feature_secdiscard = 0;
1333 	if (strncmp(type, "phy", 3) == 0) {
1334 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1335 			"discard-granularity", "%u", &discard_granularity,
1336 			"discard-alignment", "%u", &discard_alignment,
1337 			NULL);
1338 		if (!err) {
1339 			info->feature_discard = 1;
1340 			info->discard_granularity = discard_granularity;
1341 			info->discard_alignment = discard_alignment;
1342 		}
1343 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1344 			    "discard-secure", "%d", &discard_secure,
1345 			    NULL);
1346 		if (!err)
1347 			info->feature_secdiscard = discard_secure;
1348 
1349 	} else if (strncmp(type, "file", 4) == 0)
1350 		info->feature_discard = 1;
1351 
1352 	kfree(type);
1353 }
1354 
1355 /*
1356  * Invoked when the backend is finally 'ready' (and has told produced
1357  * the details about the physical device - #sectors, size, etc).
1358  */
1359 static void blkfront_connect(struct blkfront_info *info)
1360 {
1361 	unsigned long long sectors;
1362 	unsigned long sector_size;
1363 	unsigned int binfo;
1364 	int err;
1365 	int barrier, flush, discard, persistent;
1366 
1367 	switch (info->connected) {
1368 	case BLKIF_STATE_CONNECTED:
1369 		/*
1370 		 * Potentially, the back-end may be signalling
1371 		 * a capacity change; update the capacity.
1372 		 */
1373 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1374 				   "sectors", "%Lu", &sectors);
1375 		if (XENBUS_EXIST_ERR(err))
1376 			return;
1377 		printk(KERN_INFO "Setting capacity to %Lu\n",
1378 		       sectors);
1379 		set_capacity(info->gd, sectors);
1380 		revalidate_disk(info->gd);
1381 
1382 		/* fall through */
1383 	case BLKIF_STATE_SUSPENDED:
1384 		return;
1385 
1386 	default:
1387 		break;
1388 	}
1389 
1390 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1391 		__func__, info->xbdev->otherend);
1392 
1393 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1394 			    "sectors", "%llu", &sectors,
1395 			    "info", "%u", &binfo,
1396 			    "sector-size", "%lu", &sector_size,
1397 			    NULL);
1398 	if (err) {
1399 		xenbus_dev_fatal(info->xbdev, err,
1400 				 "reading backend fields at %s",
1401 				 info->xbdev->otherend);
1402 		return;
1403 	}
1404 
1405 	info->feature_flush = 0;
1406 	info->flush_op = 0;
1407 
1408 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1409 			    "feature-barrier", "%d", &barrier,
1410 			    NULL);
1411 
1412 	/*
1413 	 * If there's no "feature-barrier" defined, then it means
1414 	 * we're dealing with a very old backend which writes
1415 	 * synchronously; nothing to do.
1416 	 *
1417 	 * If there are barriers, then we use flush.
1418 	 */
1419 	if (!err && barrier) {
1420 		info->feature_flush = REQ_FLUSH | REQ_FUA;
1421 		info->flush_op = BLKIF_OP_WRITE_BARRIER;
1422 	}
1423 	/*
1424 	 * And if there is "feature-flush-cache" use that above
1425 	 * barriers.
1426 	 */
1427 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1428 			    "feature-flush-cache", "%d", &flush,
1429 			    NULL);
1430 
1431 	if (!err && flush) {
1432 		info->feature_flush = REQ_FLUSH;
1433 		info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1434 	}
1435 
1436 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1437 			    "feature-discard", "%d", &discard,
1438 			    NULL);
1439 
1440 	if (!err && discard)
1441 		blkfront_setup_discard(info);
1442 
1443 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1444 			    "feature-persistent", "%u", &persistent,
1445 			    NULL);
1446 	if (err)
1447 		info->feature_persistent = 0;
1448 	else
1449 		info->feature_persistent = persistent;
1450 
1451 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1452 	if (err) {
1453 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1454 				 info->xbdev->otherend);
1455 		return;
1456 	}
1457 
1458 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1459 
1460 	/* Kick pending requests. */
1461 	spin_lock_irq(&info->io_lock);
1462 	info->connected = BLKIF_STATE_CONNECTED;
1463 	kick_pending_request_queues(info);
1464 	spin_unlock_irq(&info->io_lock);
1465 
1466 	add_disk(info->gd);
1467 
1468 	info->is_ready = 1;
1469 }
1470 
1471 /**
1472  * Callback received when the backend's state changes.
1473  */
1474 static void blkback_changed(struct xenbus_device *dev,
1475 			    enum xenbus_state backend_state)
1476 {
1477 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1478 
1479 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1480 
1481 	switch (backend_state) {
1482 	case XenbusStateInitialising:
1483 	case XenbusStateInitWait:
1484 	case XenbusStateInitialised:
1485 	case XenbusStateReconfiguring:
1486 	case XenbusStateReconfigured:
1487 	case XenbusStateUnknown:
1488 	case XenbusStateClosed:
1489 		break;
1490 
1491 	case XenbusStateConnected:
1492 		blkfront_connect(info);
1493 		break;
1494 
1495 	case XenbusStateClosing:
1496 		blkfront_closing(info);
1497 		break;
1498 	}
1499 }
1500 
1501 static int blkfront_remove(struct xenbus_device *xbdev)
1502 {
1503 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1504 	struct block_device *bdev = NULL;
1505 	struct gendisk *disk;
1506 
1507 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1508 
1509 	blkif_free(info, 0);
1510 
1511 	mutex_lock(&info->mutex);
1512 
1513 	disk = info->gd;
1514 	if (disk)
1515 		bdev = bdget_disk(disk, 0);
1516 
1517 	info->xbdev = NULL;
1518 	mutex_unlock(&info->mutex);
1519 
1520 	if (!bdev) {
1521 		kfree(info);
1522 		return 0;
1523 	}
1524 
1525 	/*
1526 	 * The xbdev was removed before we reached the Closed
1527 	 * state. See if it's safe to remove the disk. If the bdev
1528 	 * isn't closed yet, we let release take care of it.
1529 	 */
1530 
1531 	mutex_lock(&bdev->bd_mutex);
1532 	info = disk->private_data;
1533 
1534 	dev_warn(disk_to_dev(disk),
1535 		 "%s was hot-unplugged, %d stale handles\n",
1536 		 xbdev->nodename, bdev->bd_openers);
1537 
1538 	if (info && !bdev->bd_openers) {
1539 		xlvbd_release_gendisk(info);
1540 		disk->private_data = NULL;
1541 		kfree(info);
1542 	}
1543 
1544 	mutex_unlock(&bdev->bd_mutex);
1545 	bdput(bdev);
1546 
1547 	return 0;
1548 }
1549 
1550 static int blkfront_is_ready(struct xenbus_device *dev)
1551 {
1552 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1553 
1554 	return info->is_ready && info->xbdev;
1555 }
1556 
1557 static int blkif_open(struct block_device *bdev, fmode_t mode)
1558 {
1559 	struct gendisk *disk = bdev->bd_disk;
1560 	struct blkfront_info *info;
1561 	int err = 0;
1562 
1563 	mutex_lock(&blkfront_mutex);
1564 
1565 	info = disk->private_data;
1566 	if (!info) {
1567 		/* xbdev gone */
1568 		err = -ERESTARTSYS;
1569 		goto out;
1570 	}
1571 
1572 	mutex_lock(&info->mutex);
1573 
1574 	if (!info->gd)
1575 		/* xbdev is closed */
1576 		err = -ERESTARTSYS;
1577 
1578 	mutex_unlock(&info->mutex);
1579 
1580 out:
1581 	mutex_unlock(&blkfront_mutex);
1582 	return err;
1583 }
1584 
1585 static int blkif_release(struct gendisk *disk, fmode_t mode)
1586 {
1587 	struct blkfront_info *info = disk->private_data;
1588 	struct block_device *bdev;
1589 	struct xenbus_device *xbdev;
1590 
1591 	mutex_lock(&blkfront_mutex);
1592 
1593 	bdev = bdget_disk(disk, 0);
1594 
1595 	if (bdev->bd_openers)
1596 		goto out;
1597 
1598 	/*
1599 	 * Check if we have been instructed to close. We will have
1600 	 * deferred this request, because the bdev was still open.
1601 	 */
1602 
1603 	mutex_lock(&info->mutex);
1604 	xbdev = info->xbdev;
1605 
1606 	if (xbdev && xbdev->state == XenbusStateClosing) {
1607 		/* pending switch to state closed */
1608 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1609 		xlvbd_release_gendisk(info);
1610 		xenbus_frontend_closed(info->xbdev);
1611  	}
1612 
1613 	mutex_unlock(&info->mutex);
1614 
1615 	if (!xbdev) {
1616 		/* sudden device removal */
1617 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1618 		xlvbd_release_gendisk(info);
1619 		disk->private_data = NULL;
1620 		kfree(info);
1621 	}
1622 
1623 out:
1624 	bdput(bdev);
1625 	mutex_unlock(&blkfront_mutex);
1626 	return 0;
1627 }
1628 
1629 static const struct block_device_operations xlvbd_block_fops =
1630 {
1631 	.owner = THIS_MODULE,
1632 	.open = blkif_open,
1633 	.release = blkif_release,
1634 	.getgeo = blkif_getgeo,
1635 	.ioctl = blkif_ioctl,
1636 };
1637 
1638 
1639 static const struct xenbus_device_id blkfront_ids[] = {
1640 	{ "vbd" },
1641 	{ "" }
1642 };
1643 
1644 static DEFINE_XENBUS_DRIVER(blkfront, ,
1645 	.probe = blkfront_probe,
1646 	.remove = blkfront_remove,
1647 	.resume = blkfront_resume,
1648 	.otherend_changed = blkback_changed,
1649 	.is_ready = blkfront_is_ready,
1650 );
1651 
1652 static int __init xlblk_init(void)
1653 {
1654 	int ret;
1655 
1656 	if (!xen_domain())
1657 		return -ENODEV;
1658 
1659 	if (xen_hvm_domain() && !xen_platform_pci_unplug)
1660 		return -ENODEV;
1661 
1662 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1663 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1664 		       XENVBD_MAJOR, DEV_NAME);
1665 		return -ENODEV;
1666 	}
1667 
1668 	ret = xenbus_register_frontend(&blkfront_driver);
1669 	if (ret) {
1670 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1671 		return ret;
1672 	}
1673 
1674 	return 0;
1675 }
1676 module_init(xlblk_init);
1677 
1678 
1679 static void __exit xlblk_exit(void)
1680 {
1681 	xenbus_unregister_driver(&blkfront_driver);
1682 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1683 	kfree(minors);
1684 }
1685 module_exit(xlblk_exit);
1686 
1687 MODULE_DESCRIPTION("Xen virtual block device frontend");
1688 MODULE_LICENSE("GPL");
1689 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1690 MODULE_ALIAS("xen:vbd");
1691 MODULE_ALIAS("xenblk");
1692