xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 20055477)
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/list.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 list_head node;
72 };
73 
74 struct blk_shadow {
75 	struct blkif_request req;
76 	struct request *request;
77 	struct grant **grants_used;
78 	struct grant **indirect_grants;
79 	struct scatterlist *sg;
80 };
81 
82 struct split_bio {
83 	struct bio *bio;
84 	atomic_t pending;
85 	int err;
86 };
87 
88 static DEFINE_MUTEX(blkfront_mutex);
89 static const struct block_device_operations xlvbd_block_fops;
90 
91 /*
92  * Maximum number of segments in indirect requests, the actual value used by
93  * the frontend driver is the minimum of this value and the value provided
94  * by the backend driver.
95  */
96 
97 static unsigned int xen_blkif_max_segments = 32;
98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
100 
101 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
102 
103 /*
104  * We have one of these per vbd, whether ide, scsi or 'other'.  They
105  * hang in private_data off the gendisk structure. We may end up
106  * putting all kinds of interesting stuff here :-)
107  */
108 struct blkfront_info
109 {
110 	spinlock_t io_lock;
111 	struct mutex mutex;
112 	struct xenbus_device *xbdev;
113 	struct gendisk *gd;
114 	int vdevice;
115 	blkif_vdev_t handle;
116 	enum blkif_state connected;
117 	int ring_ref;
118 	struct blkif_front_ring ring;
119 	unsigned int evtchn, irq;
120 	struct request_queue *rq;
121 	struct work_struct work;
122 	struct gnttab_free_callback callback;
123 	struct blk_shadow shadow[BLK_RING_SIZE];
124 	struct list_head persistent_gnts;
125 	unsigned int persistent_gnts_c;
126 	unsigned long shadow_free;
127 	unsigned int feature_flush;
128 	unsigned int flush_op;
129 	unsigned int feature_discard:1;
130 	unsigned int feature_secdiscard:1;
131 	unsigned int discard_granularity;
132 	unsigned int discard_alignment;
133 	unsigned int feature_persistent:1;
134 	unsigned int max_indirect_segments;
135 	int is_ready;
136 };
137 
138 static unsigned int nr_minors;
139 static unsigned long *minors;
140 static DEFINE_SPINLOCK(minor_lock);
141 
142 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
143 	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
144 #define GRANT_INVALID_REF	0
145 
146 #define PARTS_PER_DISK		16
147 #define PARTS_PER_EXT_DISK      256
148 
149 #define BLKIF_MAJOR(dev) ((dev)>>8)
150 #define BLKIF_MINOR(dev) ((dev) & 0xff)
151 
152 #define EXT_SHIFT 28
153 #define EXTENDED (1<<EXT_SHIFT)
154 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
155 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
156 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
157 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
158 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
159 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
160 
161 #define DEV_NAME	"xvd"	/* name in /dev */
162 
163 #define SEGS_PER_INDIRECT_FRAME \
164 	(PAGE_SIZE/sizeof(struct blkif_request_segment_aligned))
165 #define INDIRECT_GREFS(_segs) \
166 	((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
167 
168 static int blkfront_setup_indirect(struct blkfront_info *info);
169 
170 static int get_id_from_freelist(struct blkfront_info *info)
171 {
172 	unsigned long free = info->shadow_free;
173 	BUG_ON(free >= BLK_RING_SIZE);
174 	info->shadow_free = info->shadow[free].req.u.rw.id;
175 	info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
176 	return free;
177 }
178 
179 static int add_id_to_freelist(struct blkfront_info *info,
180 			       unsigned long id)
181 {
182 	if (info->shadow[id].req.u.rw.id != id)
183 		return -EINVAL;
184 	if (info->shadow[id].request == NULL)
185 		return -EINVAL;
186 	info->shadow[id].req.u.rw.id  = info->shadow_free;
187 	info->shadow[id].request = NULL;
188 	info->shadow_free = id;
189 	return 0;
190 }
191 
192 static int fill_grant_buffer(struct blkfront_info *info, int num)
193 {
194 	struct page *granted_page;
195 	struct grant *gnt_list_entry, *n;
196 	int i = 0;
197 
198 	while(i < num) {
199 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
200 		if (!gnt_list_entry)
201 			goto out_of_memory;
202 
203 		granted_page = alloc_page(GFP_NOIO);
204 		if (!granted_page) {
205 			kfree(gnt_list_entry);
206 			goto out_of_memory;
207 		}
208 
209 		gnt_list_entry->pfn = page_to_pfn(granted_page);
210 		gnt_list_entry->gref = GRANT_INVALID_REF;
211 		list_add(&gnt_list_entry->node, &info->persistent_gnts);
212 		i++;
213 	}
214 
215 	return 0;
216 
217 out_of_memory:
218 	list_for_each_entry_safe(gnt_list_entry, n,
219 	                         &info->persistent_gnts, node) {
220 		list_del(&gnt_list_entry->node);
221 		__free_page(pfn_to_page(gnt_list_entry->pfn));
222 		kfree(gnt_list_entry);
223 		i--;
224 	}
225 	BUG_ON(i != 0);
226 	return -ENOMEM;
227 }
228 
229 static struct grant *get_grant(grant_ref_t *gref_head,
230                                struct blkfront_info *info)
231 {
232 	struct grant *gnt_list_entry;
233 	unsigned long buffer_mfn;
234 
235 	BUG_ON(list_empty(&info->persistent_gnts));
236 	gnt_list_entry = list_first_entry(&info->persistent_gnts, struct grant,
237 	                                  node);
238 	list_del(&gnt_list_entry->node);
239 
240 	if (gnt_list_entry->gref != GRANT_INVALID_REF) {
241 		info->persistent_gnts_c--;
242 		return gnt_list_entry;
243 	}
244 
245 	/* Assign a gref to this page */
246 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
247 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
248 	buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
249 	gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
250 	                                info->xbdev->otherend_id,
251 	                                buffer_mfn, 0);
252 	return gnt_list_entry;
253 }
254 
255 static const char *op_name(int op)
256 {
257 	static const char *const names[] = {
258 		[BLKIF_OP_READ] = "read",
259 		[BLKIF_OP_WRITE] = "write",
260 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
261 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
262 		[BLKIF_OP_DISCARD] = "discard" };
263 
264 	if (op < 0 || op >= ARRAY_SIZE(names))
265 		return "unknown";
266 
267 	if (!names[op])
268 		return "reserved";
269 
270 	return names[op];
271 }
272 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
273 {
274 	unsigned int end = minor + nr;
275 	int rc;
276 
277 	if (end > nr_minors) {
278 		unsigned long *bitmap, *old;
279 
280 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
281 				 GFP_KERNEL);
282 		if (bitmap == NULL)
283 			return -ENOMEM;
284 
285 		spin_lock(&minor_lock);
286 		if (end > nr_minors) {
287 			old = minors;
288 			memcpy(bitmap, minors,
289 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
290 			minors = bitmap;
291 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
292 		} else
293 			old = bitmap;
294 		spin_unlock(&minor_lock);
295 		kfree(old);
296 	}
297 
298 	spin_lock(&minor_lock);
299 	if (find_next_bit(minors, end, minor) >= end) {
300 		bitmap_set(minors, minor, nr);
301 		rc = 0;
302 	} else
303 		rc = -EBUSY;
304 	spin_unlock(&minor_lock);
305 
306 	return rc;
307 }
308 
309 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
310 {
311 	unsigned int end = minor + nr;
312 
313 	BUG_ON(end > nr_minors);
314 	spin_lock(&minor_lock);
315 	bitmap_clear(minors,  minor, nr);
316 	spin_unlock(&minor_lock);
317 }
318 
319 static void blkif_restart_queue_callback(void *arg)
320 {
321 	struct blkfront_info *info = (struct blkfront_info *)arg;
322 	schedule_work(&info->work);
323 }
324 
325 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
326 {
327 	/* We don't have real geometry info, but let's at least return
328 	   values consistent with the size of the device */
329 	sector_t nsect = get_capacity(bd->bd_disk);
330 	sector_t cylinders = nsect;
331 
332 	hg->heads = 0xff;
333 	hg->sectors = 0x3f;
334 	sector_div(cylinders, hg->heads * hg->sectors);
335 	hg->cylinders = cylinders;
336 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
337 		hg->cylinders = 0xffff;
338 	return 0;
339 }
340 
341 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
342 		       unsigned command, unsigned long argument)
343 {
344 	struct blkfront_info *info = bdev->bd_disk->private_data;
345 	int i;
346 
347 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
348 		command, (long)argument);
349 
350 	switch (command) {
351 	case CDROMMULTISESSION:
352 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
353 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
354 			if (put_user(0, (char __user *)(argument + i)))
355 				return -EFAULT;
356 		return 0;
357 
358 	case CDROM_GET_CAPABILITY: {
359 		struct gendisk *gd = info->gd;
360 		if (gd->flags & GENHD_FL_CD)
361 			return 0;
362 		return -EINVAL;
363 	}
364 
365 	default:
366 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
367 		  command);*/
368 		return -EINVAL; /* same return as native Linux */
369 	}
370 
371 	return 0;
372 }
373 
374 /*
375  * Generate a Xen blkfront IO request from a blk layer request.  Reads
376  * and writes are handled as expected.
377  *
378  * @req: a request struct
379  */
380 static int blkif_queue_request(struct request *req)
381 {
382 	struct blkfront_info *info = req->rq_disk->private_data;
383 	struct blkif_request *ring_req;
384 	unsigned long id;
385 	unsigned int fsect, lsect;
386 	int i, ref, n;
387 	struct blkif_request_segment_aligned *segments = NULL;
388 
389 	/*
390 	 * Used to store if we are able to queue the request by just using
391 	 * existing persistent grants, or if we have to get new grants,
392 	 * as there are not sufficiently many free.
393 	 */
394 	bool new_persistent_gnts;
395 	grant_ref_t gref_head;
396 	struct grant *gnt_list_entry = NULL;
397 	struct scatterlist *sg;
398 	int nseg, max_grefs;
399 
400 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
401 		return 1;
402 
403 	max_grefs = info->max_indirect_segments ?
404 		    info->max_indirect_segments +
405 		    INDIRECT_GREFS(info->max_indirect_segments) :
406 		    BLKIF_MAX_SEGMENTS_PER_REQUEST;
407 
408 	/* Check if we have enough grants to allocate a requests */
409 	if (info->persistent_gnts_c < max_grefs) {
410 		new_persistent_gnts = 1;
411 		if (gnttab_alloc_grant_references(
412 		    max_grefs - info->persistent_gnts_c,
413 		    &gref_head) < 0) {
414 			gnttab_request_free_callback(
415 				&info->callback,
416 				blkif_restart_queue_callback,
417 				info,
418 				max_grefs);
419 			return 1;
420 		}
421 	} else
422 		new_persistent_gnts = 0;
423 
424 	/* Fill out a communications ring structure. */
425 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
426 	id = get_id_from_freelist(info);
427 	info->shadow[id].request = req;
428 
429 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
430 		ring_req->operation = BLKIF_OP_DISCARD;
431 		ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
432 		ring_req->u.discard.id = id;
433 		ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
434 		if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
435 			ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
436 		else
437 			ring_req->u.discard.flag = 0;
438 	} else {
439 		BUG_ON(info->max_indirect_segments == 0 &&
440 		       req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
441 		BUG_ON(info->max_indirect_segments &&
442 		       req->nr_phys_segments > info->max_indirect_segments);
443 		nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
444 		ring_req->u.rw.id = id;
445 		if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
446 			/*
447 			 * The indirect operation can only be a BLKIF_OP_READ or
448 			 * BLKIF_OP_WRITE
449 			 */
450 			BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
451 			ring_req->operation = BLKIF_OP_INDIRECT;
452 			ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
453 				BLKIF_OP_WRITE : BLKIF_OP_READ;
454 			ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
455 			ring_req->u.indirect.handle = info->handle;
456 			ring_req->u.indirect.nr_segments = nseg;
457 		} else {
458 			ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
459 			ring_req->u.rw.handle = info->handle;
460 			ring_req->operation = rq_data_dir(req) ?
461 				BLKIF_OP_WRITE : BLKIF_OP_READ;
462 			if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
463 				/*
464 				 * Ideally we can do an unordered flush-to-disk. In case the
465 				 * backend onlysupports barriers, use that. A barrier request
466 				 * a superset of FUA, so we can implement it the same
467 				 * way.  (It's also a FLUSH+FUA, since it is
468 				 * guaranteed ordered WRT previous writes.)
469 				 */
470 				ring_req->operation = info->flush_op;
471 			}
472 			ring_req->u.rw.nr_segments = nseg;
473 		}
474 		for_each_sg(info->shadow[id].sg, sg, nseg, i) {
475 			fsect = sg->offset >> 9;
476 			lsect = fsect + (sg->length >> 9) - 1;
477 
478 			if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
479 			    (i % SEGS_PER_INDIRECT_FRAME == 0)) {
480 				if (segments)
481 					kunmap_atomic(segments);
482 
483 				n = i / SEGS_PER_INDIRECT_FRAME;
484 				gnt_list_entry = get_grant(&gref_head, info);
485 				info->shadow[id].indirect_grants[n] = gnt_list_entry;
486 				segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
487 				ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
488 			}
489 
490 			gnt_list_entry = get_grant(&gref_head, info);
491 			ref = gnt_list_entry->gref;
492 
493 			info->shadow[id].grants_used[i] = gnt_list_entry;
494 
495 			if (rq_data_dir(req)) {
496 				char *bvec_data;
497 				void *shared_data;
498 
499 				BUG_ON(sg->offset + sg->length > PAGE_SIZE);
500 
501 				shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
502 				bvec_data = kmap_atomic(sg_page(sg));
503 
504 				/*
505 				 * this does not wipe data stored outside the
506 				 * range sg->offset..sg->offset+sg->length.
507 				 * Therefore, blkback *could* see data from
508 				 * previous requests. This is OK as long as
509 				 * persistent grants are shared with just one
510 				 * domain. It may need refactoring if this
511 				 * changes
512 				 */
513 				memcpy(shared_data + sg->offset,
514 				       bvec_data   + sg->offset,
515 				       sg->length);
516 
517 				kunmap_atomic(bvec_data);
518 				kunmap_atomic(shared_data);
519 			}
520 			if (ring_req->operation != BLKIF_OP_INDIRECT) {
521 				ring_req->u.rw.seg[i] =
522 						(struct blkif_request_segment) {
523 							.gref       = ref,
524 							.first_sect = fsect,
525 							.last_sect  = lsect };
526 			} else {
527 				n = i % SEGS_PER_INDIRECT_FRAME;
528 				segments[n] =
529 					(struct blkif_request_segment_aligned) {
530 							.gref       = ref,
531 							.first_sect = fsect,
532 							.last_sect  = lsect };
533 			}
534 		}
535 		if (segments)
536 			kunmap_atomic(segments);
537 	}
538 
539 	info->ring.req_prod_pvt++;
540 
541 	/* Keep a private copy so we can reissue requests when recovering. */
542 	info->shadow[id].req = *ring_req;
543 
544 	if (new_persistent_gnts)
545 		gnttab_free_grant_references(gref_head);
546 
547 	return 0;
548 }
549 
550 
551 static inline void flush_requests(struct blkfront_info *info)
552 {
553 	int notify;
554 
555 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
556 
557 	if (notify)
558 		notify_remote_via_irq(info->irq);
559 }
560 
561 /*
562  * do_blkif_request
563  *  read a block; request is in a request queue
564  */
565 static void do_blkif_request(struct request_queue *rq)
566 {
567 	struct blkfront_info *info = NULL;
568 	struct request *req;
569 	int queued;
570 
571 	pr_debug("Entered do_blkif_request\n");
572 
573 	queued = 0;
574 
575 	while ((req = blk_peek_request(rq)) != NULL) {
576 		info = req->rq_disk->private_data;
577 
578 		if (RING_FULL(&info->ring))
579 			goto wait;
580 
581 		blk_start_request(req);
582 
583 		if ((req->cmd_type != REQ_TYPE_FS) ||
584 		    ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
585 		    !info->flush_op)) {
586 			__blk_end_request_all(req, -EIO);
587 			continue;
588 		}
589 
590 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
591 			 "(%u/%u) buffer:%p [%s]\n",
592 			 req, req->cmd, (unsigned long)blk_rq_pos(req),
593 			 blk_rq_cur_sectors(req), blk_rq_sectors(req),
594 			 req->buffer, rq_data_dir(req) ? "write" : "read");
595 
596 		if (blkif_queue_request(req)) {
597 			blk_requeue_request(rq, req);
598 wait:
599 			/* Avoid pointless unplugs. */
600 			blk_stop_queue(rq);
601 			break;
602 		}
603 
604 		queued++;
605 	}
606 
607 	if (queued != 0)
608 		flush_requests(info);
609 }
610 
611 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
612 				unsigned int physical_sector_size,
613 				unsigned int segments)
614 {
615 	struct request_queue *rq;
616 	struct blkfront_info *info = gd->private_data;
617 
618 	rq = blk_init_queue(do_blkif_request, &info->io_lock);
619 	if (rq == NULL)
620 		return -1;
621 
622 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
623 
624 	if (info->feature_discard) {
625 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
626 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
627 		rq->limits.discard_granularity = info->discard_granularity;
628 		rq->limits.discard_alignment = info->discard_alignment;
629 		if (info->feature_secdiscard)
630 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
631 	}
632 
633 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
634 	blk_queue_logical_block_size(rq, sector_size);
635 	blk_queue_physical_block_size(rq, physical_sector_size);
636 	blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
637 
638 	/* Each segment in a request is up to an aligned page in size. */
639 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
640 	blk_queue_max_segment_size(rq, PAGE_SIZE);
641 
642 	/* Ensure a merged request will fit in a single I/O ring slot. */
643 	blk_queue_max_segments(rq, segments);
644 
645 	/* Make sure buffer addresses are sector-aligned. */
646 	blk_queue_dma_alignment(rq, 511);
647 
648 	/* Make sure we don't use bounce buffers. */
649 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
650 
651 	gd->queue = rq;
652 
653 	return 0;
654 }
655 
656 
657 static void xlvbd_flush(struct blkfront_info *info)
658 {
659 	blk_queue_flush(info->rq, info->feature_flush);
660 	printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
661 	       info->gd->disk_name,
662 	       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
663 		"barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
664 		"flush diskcache" : "barrier or flush"),
665 	       info->feature_flush ? "enabled;" : "disabled;",
666 	       "persistent grants:",
667 	       info->feature_persistent ? "enabled;" : "disabled;",
668 	       "indirect descriptors:",
669 	       info->max_indirect_segments ? "enabled;" : "disabled;");
670 }
671 
672 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
673 {
674 	int major;
675 	major = BLKIF_MAJOR(vdevice);
676 	*minor = BLKIF_MINOR(vdevice);
677 	switch (major) {
678 		case XEN_IDE0_MAJOR:
679 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
680 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
681 				EMULATED_HD_DISK_MINOR_OFFSET;
682 			break;
683 		case XEN_IDE1_MAJOR:
684 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
685 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
686 				EMULATED_HD_DISK_MINOR_OFFSET;
687 			break;
688 		case XEN_SCSI_DISK0_MAJOR:
689 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
690 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
691 			break;
692 		case XEN_SCSI_DISK1_MAJOR:
693 		case XEN_SCSI_DISK2_MAJOR:
694 		case XEN_SCSI_DISK3_MAJOR:
695 		case XEN_SCSI_DISK4_MAJOR:
696 		case XEN_SCSI_DISK5_MAJOR:
697 		case XEN_SCSI_DISK6_MAJOR:
698 		case XEN_SCSI_DISK7_MAJOR:
699 			*offset = (*minor / PARTS_PER_DISK) +
700 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
701 				EMULATED_SD_DISK_NAME_OFFSET;
702 			*minor = *minor +
703 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
704 				EMULATED_SD_DISK_MINOR_OFFSET;
705 			break;
706 		case XEN_SCSI_DISK8_MAJOR:
707 		case XEN_SCSI_DISK9_MAJOR:
708 		case XEN_SCSI_DISK10_MAJOR:
709 		case XEN_SCSI_DISK11_MAJOR:
710 		case XEN_SCSI_DISK12_MAJOR:
711 		case XEN_SCSI_DISK13_MAJOR:
712 		case XEN_SCSI_DISK14_MAJOR:
713 		case XEN_SCSI_DISK15_MAJOR:
714 			*offset = (*minor / PARTS_PER_DISK) +
715 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
716 				EMULATED_SD_DISK_NAME_OFFSET;
717 			*minor = *minor +
718 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
719 				EMULATED_SD_DISK_MINOR_OFFSET;
720 			break;
721 		case XENVBD_MAJOR:
722 			*offset = *minor / PARTS_PER_DISK;
723 			break;
724 		default:
725 			printk(KERN_WARNING "blkfront: your disk configuration is "
726 					"incorrect, please use an xvd device instead\n");
727 			return -ENODEV;
728 	}
729 	return 0;
730 }
731 
732 static char *encode_disk_name(char *ptr, unsigned int n)
733 {
734 	if (n >= 26)
735 		ptr = encode_disk_name(ptr, n / 26 - 1);
736 	*ptr = 'a' + n % 26;
737 	return ptr + 1;
738 }
739 
740 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
741 			       struct blkfront_info *info,
742 			       u16 vdisk_info, u16 sector_size,
743 			       unsigned int physical_sector_size)
744 {
745 	struct gendisk *gd;
746 	int nr_minors = 1;
747 	int err;
748 	unsigned int offset;
749 	int minor;
750 	int nr_parts;
751 	char *ptr;
752 
753 	BUG_ON(info->gd != NULL);
754 	BUG_ON(info->rq != NULL);
755 
756 	if ((info->vdevice>>EXT_SHIFT) > 1) {
757 		/* this is above the extended range; something is wrong */
758 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
759 		return -ENODEV;
760 	}
761 
762 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
763 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
764 		if (err)
765 			return err;
766  		nr_parts = PARTS_PER_DISK;
767 	} else {
768 		minor = BLKIF_MINOR_EXT(info->vdevice);
769 		nr_parts = PARTS_PER_EXT_DISK;
770 		offset = minor / nr_parts;
771 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
772 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
773 					"emulated IDE disks,\n\t choose an xvd device name"
774 					"from xvde on\n", info->vdevice);
775 	}
776 	if (minor >> MINORBITS) {
777 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
778 			info->vdevice, minor);
779 		return -ENODEV;
780 	}
781 
782 	if ((minor % nr_parts) == 0)
783 		nr_minors = nr_parts;
784 
785 	err = xlbd_reserve_minors(minor, nr_minors);
786 	if (err)
787 		goto out;
788 	err = -ENODEV;
789 
790 	gd = alloc_disk(nr_minors);
791 	if (gd == NULL)
792 		goto release;
793 
794 	strcpy(gd->disk_name, DEV_NAME);
795 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
796 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
797 	if (nr_minors > 1)
798 		*ptr = 0;
799 	else
800 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
801 			 "%d", minor & (nr_parts - 1));
802 
803 	gd->major = XENVBD_MAJOR;
804 	gd->first_minor = minor;
805 	gd->fops = &xlvbd_block_fops;
806 	gd->private_data = info;
807 	gd->driverfs_dev = &(info->xbdev->dev);
808 	set_capacity(gd, capacity);
809 
810 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
811 				 info->max_indirect_segments ? :
812 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
813 		del_gendisk(gd);
814 		goto release;
815 	}
816 
817 	info->rq = gd->queue;
818 	info->gd = gd;
819 
820 	xlvbd_flush(info);
821 
822 	if (vdisk_info & VDISK_READONLY)
823 		set_disk_ro(gd, 1);
824 
825 	if (vdisk_info & VDISK_REMOVABLE)
826 		gd->flags |= GENHD_FL_REMOVABLE;
827 
828 	if (vdisk_info & VDISK_CDROM)
829 		gd->flags |= GENHD_FL_CD;
830 
831 	return 0;
832 
833  release:
834 	xlbd_release_minors(minor, nr_minors);
835  out:
836 	return err;
837 }
838 
839 static void xlvbd_release_gendisk(struct blkfront_info *info)
840 {
841 	unsigned int minor, nr_minors;
842 	unsigned long flags;
843 
844 	if (info->rq == NULL)
845 		return;
846 
847 	spin_lock_irqsave(&info->io_lock, flags);
848 
849 	/* No more blkif_request(). */
850 	blk_stop_queue(info->rq);
851 
852 	/* No more gnttab callback work. */
853 	gnttab_cancel_free_callback(&info->callback);
854 	spin_unlock_irqrestore(&info->io_lock, flags);
855 
856 	/* Flush gnttab callback work. Must be done with no locks held. */
857 	flush_work(&info->work);
858 
859 	del_gendisk(info->gd);
860 
861 	minor = info->gd->first_minor;
862 	nr_minors = info->gd->minors;
863 	xlbd_release_minors(minor, nr_minors);
864 
865 	blk_cleanup_queue(info->rq);
866 	info->rq = NULL;
867 
868 	put_disk(info->gd);
869 	info->gd = NULL;
870 }
871 
872 static void kick_pending_request_queues(struct blkfront_info *info)
873 {
874 	if (!RING_FULL(&info->ring)) {
875 		/* Re-enable calldowns. */
876 		blk_start_queue(info->rq);
877 		/* Kick things off immediately. */
878 		do_blkif_request(info->rq);
879 	}
880 }
881 
882 static void blkif_restart_queue(struct work_struct *work)
883 {
884 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
885 
886 	spin_lock_irq(&info->io_lock);
887 	if (info->connected == BLKIF_STATE_CONNECTED)
888 		kick_pending_request_queues(info);
889 	spin_unlock_irq(&info->io_lock);
890 }
891 
892 static void blkif_free(struct blkfront_info *info, int suspend)
893 {
894 	struct grant *persistent_gnt;
895 	struct grant *n;
896 	int i, j, segs;
897 
898 	/* Prevent new requests being issued until we fix things up. */
899 	spin_lock_irq(&info->io_lock);
900 	info->connected = suspend ?
901 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
902 	/* No more blkif_request(). */
903 	if (info->rq)
904 		blk_stop_queue(info->rq);
905 
906 	/* Remove all persistent grants */
907 	if (!list_empty(&info->persistent_gnts)) {
908 		list_for_each_entry_safe(persistent_gnt, n,
909 		                         &info->persistent_gnts, node) {
910 			list_del(&persistent_gnt->node);
911 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
912 				gnttab_end_foreign_access(persistent_gnt->gref,
913 				                          0, 0UL);
914 				info->persistent_gnts_c--;
915 			}
916 			__free_page(pfn_to_page(persistent_gnt->pfn));
917 			kfree(persistent_gnt);
918 		}
919 	}
920 	BUG_ON(info->persistent_gnts_c != 0);
921 
922 	for (i = 0; i < BLK_RING_SIZE; i++) {
923 		/*
924 		 * Clear persistent grants present in requests already
925 		 * on the shared ring
926 		 */
927 		if (!info->shadow[i].request)
928 			goto free_shadow;
929 
930 		segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
931 		       info->shadow[i].req.u.indirect.nr_segments :
932 		       info->shadow[i].req.u.rw.nr_segments;
933 		for (j = 0; j < segs; j++) {
934 			persistent_gnt = info->shadow[i].grants_used[j];
935 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
936 			__free_page(pfn_to_page(persistent_gnt->pfn));
937 			kfree(persistent_gnt);
938 		}
939 
940 		if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
941 			/*
942 			 * If this is not an indirect operation don't try to
943 			 * free indirect segments
944 			 */
945 			goto free_shadow;
946 
947 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
948 			persistent_gnt = info->shadow[i].indirect_grants[j];
949 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
950 			__free_page(pfn_to_page(persistent_gnt->pfn));
951 			kfree(persistent_gnt);
952 		}
953 
954 free_shadow:
955 		kfree(info->shadow[i].grants_used);
956 		info->shadow[i].grants_used = NULL;
957 		kfree(info->shadow[i].indirect_grants);
958 		info->shadow[i].indirect_grants = NULL;
959 		kfree(info->shadow[i].sg);
960 		info->shadow[i].sg = NULL;
961 	}
962 
963 	/* No more gnttab callback work. */
964 	gnttab_cancel_free_callback(&info->callback);
965 	spin_unlock_irq(&info->io_lock);
966 
967 	/* Flush gnttab callback work. Must be done with no locks held. */
968 	flush_work(&info->work);
969 
970 	/* Free resources associated with old device channel. */
971 	if (info->ring_ref != GRANT_INVALID_REF) {
972 		gnttab_end_foreign_access(info->ring_ref, 0,
973 					  (unsigned long)info->ring.sring);
974 		info->ring_ref = GRANT_INVALID_REF;
975 		info->ring.sring = NULL;
976 	}
977 	if (info->irq)
978 		unbind_from_irqhandler(info->irq, info);
979 	info->evtchn = info->irq = 0;
980 
981 }
982 
983 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
984 			     struct blkif_response *bret)
985 {
986 	int i = 0;
987 	struct scatterlist *sg;
988 	char *bvec_data;
989 	void *shared_data;
990 	int nseg;
991 
992 	nseg = s->req.operation == BLKIF_OP_INDIRECT ?
993 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
994 
995 	if (bret->operation == BLKIF_OP_READ) {
996 		/*
997 		 * Copy the data received from the backend into the bvec.
998 		 * Since bv_offset can be different than 0, and bv_len different
999 		 * than PAGE_SIZE, we have to keep track of the current offset,
1000 		 * to be sure we are copying the data from the right shared page.
1001 		 */
1002 		for_each_sg(s->sg, sg, nseg, i) {
1003 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1004 			shared_data = kmap_atomic(
1005 				pfn_to_page(s->grants_used[i]->pfn));
1006 			bvec_data = kmap_atomic(sg_page(sg));
1007 			memcpy(bvec_data   + sg->offset,
1008 			       shared_data + sg->offset,
1009 			       sg->length);
1010 			kunmap_atomic(bvec_data);
1011 			kunmap_atomic(shared_data);
1012 		}
1013 	}
1014 	/* Add the persistent grant into the list of free grants */
1015 	for (i = 0; i < nseg; i++) {
1016 		list_add(&s->grants_used[i]->node, &info->persistent_gnts);
1017 		info->persistent_gnts_c++;
1018 	}
1019 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1020 		for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1021 			list_add(&s->indirect_grants[i]->node, &info->persistent_gnts);
1022 			info->persistent_gnts_c++;
1023 		}
1024 	}
1025 }
1026 
1027 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1028 {
1029 	struct request *req;
1030 	struct blkif_response *bret;
1031 	RING_IDX i, rp;
1032 	unsigned long flags;
1033 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
1034 	int error;
1035 
1036 	spin_lock_irqsave(&info->io_lock, flags);
1037 
1038 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1039 		spin_unlock_irqrestore(&info->io_lock, flags);
1040 		return IRQ_HANDLED;
1041 	}
1042 
1043  again:
1044 	rp = info->ring.sring->rsp_prod;
1045 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1046 
1047 	for (i = info->ring.rsp_cons; i != rp; i++) {
1048 		unsigned long id;
1049 
1050 		bret = RING_GET_RESPONSE(&info->ring, i);
1051 		id   = bret->id;
1052 		/*
1053 		 * The backend has messed up and given us an id that we would
1054 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1055 		 * look in get_id_from_freelist.
1056 		 */
1057 		if (id >= BLK_RING_SIZE) {
1058 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1059 			     info->gd->disk_name, op_name(bret->operation), id);
1060 			/* We can't safely get the 'struct request' as
1061 			 * the id is busted. */
1062 			continue;
1063 		}
1064 		req  = info->shadow[id].request;
1065 
1066 		if (bret->operation != BLKIF_OP_DISCARD)
1067 			blkif_completion(&info->shadow[id], info, bret);
1068 
1069 		if (add_id_to_freelist(info, id)) {
1070 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1071 			     info->gd->disk_name, op_name(bret->operation), id);
1072 			continue;
1073 		}
1074 
1075 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1076 		switch (bret->operation) {
1077 		case BLKIF_OP_DISCARD:
1078 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1079 				struct request_queue *rq = info->rq;
1080 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1081 					   info->gd->disk_name, op_name(bret->operation));
1082 				error = -EOPNOTSUPP;
1083 				info->feature_discard = 0;
1084 				info->feature_secdiscard = 0;
1085 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1086 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1087 			}
1088 			__blk_end_request_all(req, error);
1089 			break;
1090 		case BLKIF_OP_FLUSH_DISKCACHE:
1091 		case BLKIF_OP_WRITE_BARRIER:
1092 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1093 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1094 				       info->gd->disk_name, op_name(bret->operation));
1095 				error = -EOPNOTSUPP;
1096 			}
1097 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1098 				     info->shadow[id].req.u.rw.nr_segments == 0)) {
1099 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1100 				       info->gd->disk_name, op_name(bret->operation));
1101 				error = -EOPNOTSUPP;
1102 			}
1103 			if (unlikely(error)) {
1104 				if (error == -EOPNOTSUPP)
1105 					error = 0;
1106 				info->feature_flush = 0;
1107 				info->flush_op = 0;
1108 				xlvbd_flush(info);
1109 			}
1110 			/* fall through */
1111 		case BLKIF_OP_READ:
1112 		case BLKIF_OP_WRITE:
1113 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1114 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1115 					"request: %x\n", bret->status);
1116 
1117 			__blk_end_request_all(req, error);
1118 			break;
1119 		default:
1120 			BUG();
1121 		}
1122 	}
1123 
1124 	info->ring.rsp_cons = i;
1125 
1126 	if (i != info->ring.req_prod_pvt) {
1127 		int more_to_do;
1128 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1129 		if (more_to_do)
1130 			goto again;
1131 	} else
1132 		info->ring.sring->rsp_event = i + 1;
1133 
1134 	kick_pending_request_queues(info);
1135 
1136 	spin_unlock_irqrestore(&info->io_lock, flags);
1137 
1138 	return IRQ_HANDLED;
1139 }
1140 
1141 
1142 static int setup_blkring(struct xenbus_device *dev,
1143 			 struct blkfront_info *info)
1144 {
1145 	struct blkif_sring *sring;
1146 	int err;
1147 
1148 	info->ring_ref = GRANT_INVALID_REF;
1149 
1150 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1151 	if (!sring) {
1152 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1153 		return -ENOMEM;
1154 	}
1155 	SHARED_RING_INIT(sring);
1156 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1157 
1158 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1159 	if (err < 0) {
1160 		free_page((unsigned long)sring);
1161 		info->ring.sring = NULL;
1162 		goto fail;
1163 	}
1164 	info->ring_ref = err;
1165 
1166 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
1167 	if (err)
1168 		goto fail;
1169 
1170 	err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1171 					"blkif", info);
1172 	if (err <= 0) {
1173 		xenbus_dev_fatal(dev, err,
1174 				 "bind_evtchn_to_irqhandler failed");
1175 		goto fail;
1176 	}
1177 	info->irq = err;
1178 
1179 	return 0;
1180 fail:
1181 	blkif_free(info, 0);
1182 	return err;
1183 }
1184 
1185 
1186 /* Common code used when first setting up, and when resuming. */
1187 static int talk_to_blkback(struct xenbus_device *dev,
1188 			   struct blkfront_info *info)
1189 {
1190 	const char *message = NULL;
1191 	struct xenbus_transaction xbt;
1192 	int err;
1193 
1194 	/* Create shared ring, alloc event channel. */
1195 	err = setup_blkring(dev, info);
1196 	if (err)
1197 		goto out;
1198 
1199 again:
1200 	err = xenbus_transaction_start(&xbt);
1201 	if (err) {
1202 		xenbus_dev_fatal(dev, err, "starting transaction");
1203 		goto destroy_blkring;
1204 	}
1205 
1206 	err = xenbus_printf(xbt, dev->nodename,
1207 			    "ring-ref", "%u", info->ring_ref);
1208 	if (err) {
1209 		message = "writing ring-ref";
1210 		goto abort_transaction;
1211 	}
1212 	err = xenbus_printf(xbt, dev->nodename,
1213 			    "event-channel", "%u", info->evtchn);
1214 	if (err) {
1215 		message = "writing event-channel";
1216 		goto abort_transaction;
1217 	}
1218 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1219 			    XEN_IO_PROTO_ABI_NATIVE);
1220 	if (err) {
1221 		message = "writing protocol";
1222 		goto abort_transaction;
1223 	}
1224 	err = xenbus_printf(xbt, dev->nodename,
1225 			    "feature-persistent", "%u", 1);
1226 	if (err)
1227 		dev_warn(&dev->dev,
1228 			 "writing persistent grants feature to xenbus");
1229 
1230 	err = xenbus_transaction_end(xbt, 0);
1231 	if (err) {
1232 		if (err == -EAGAIN)
1233 			goto again;
1234 		xenbus_dev_fatal(dev, err, "completing transaction");
1235 		goto destroy_blkring;
1236 	}
1237 
1238 	xenbus_switch_state(dev, XenbusStateInitialised);
1239 
1240 	return 0;
1241 
1242  abort_transaction:
1243 	xenbus_transaction_end(xbt, 1);
1244 	if (message)
1245 		xenbus_dev_fatal(dev, err, "%s", message);
1246  destroy_blkring:
1247 	blkif_free(info, 0);
1248  out:
1249 	return err;
1250 }
1251 
1252 /**
1253  * Entry point to this code when a new device is created.  Allocate the basic
1254  * structures and the ring buffer for communication with the backend, and
1255  * inform the backend of the appropriate details for those.  Switch to
1256  * Initialised state.
1257  */
1258 static int blkfront_probe(struct xenbus_device *dev,
1259 			  const struct xenbus_device_id *id)
1260 {
1261 	int err, vdevice, i;
1262 	struct blkfront_info *info;
1263 
1264 	/* FIXME: Use dynamic device id if this is not set. */
1265 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1266 			   "virtual-device", "%i", &vdevice);
1267 	if (err != 1) {
1268 		/* go looking in the extended area instead */
1269 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1270 				   "%i", &vdevice);
1271 		if (err != 1) {
1272 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1273 			return err;
1274 		}
1275 	}
1276 
1277 	if (xen_hvm_domain()) {
1278 		char *type;
1279 		int len;
1280 		/* no unplug has been done: do not hook devices != xen vbds */
1281 		if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1282 			int major;
1283 
1284 			if (!VDEV_IS_EXTENDED(vdevice))
1285 				major = BLKIF_MAJOR(vdevice);
1286 			else
1287 				major = XENVBD_MAJOR;
1288 
1289 			if (major != XENVBD_MAJOR) {
1290 				printk(KERN_INFO
1291 						"%s: HVM does not support vbd %d as xen block device\n",
1292 						__FUNCTION__, vdevice);
1293 				return -ENODEV;
1294 			}
1295 		}
1296 		/* do not create a PV cdrom device if we are an HVM guest */
1297 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1298 		if (IS_ERR(type))
1299 			return -ENODEV;
1300 		if (strncmp(type, "cdrom", 5) == 0) {
1301 			kfree(type);
1302 			return -ENODEV;
1303 		}
1304 		kfree(type);
1305 	}
1306 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1307 	if (!info) {
1308 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1309 		return -ENOMEM;
1310 	}
1311 
1312 	mutex_init(&info->mutex);
1313 	spin_lock_init(&info->io_lock);
1314 	info->xbdev = dev;
1315 	info->vdevice = vdevice;
1316 	INIT_LIST_HEAD(&info->persistent_gnts);
1317 	info->persistent_gnts_c = 0;
1318 	info->connected = BLKIF_STATE_DISCONNECTED;
1319 	INIT_WORK(&info->work, blkif_restart_queue);
1320 
1321 	for (i = 0; i < BLK_RING_SIZE; i++)
1322 		info->shadow[i].req.u.rw.id = i+1;
1323 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1324 
1325 	/* Front end dir is a number, which is used as the id. */
1326 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1327 	dev_set_drvdata(&dev->dev, info);
1328 
1329 	err = talk_to_blkback(dev, info);
1330 	if (err) {
1331 		kfree(info);
1332 		dev_set_drvdata(&dev->dev, NULL);
1333 		return err;
1334 	}
1335 
1336 	return 0;
1337 }
1338 
1339 /*
1340  * This is a clone of md_trim_bio, used to split a bio into smaller ones
1341  */
1342 static void trim_bio(struct bio *bio, int offset, int size)
1343 {
1344 	/* 'bio' is a cloned bio which we need to trim to match
1345 	 * the given offset and size.
1346 	 * This requires adjusting bi_sector, bi_size, and bi_io_vec
1347 	 */
1348 	int i;
1349 	struct bio_vec *bvec;
1350 	int sofar = 0;
1351 
1352 	size <<= 9;
1353 	if (offset == 0 && size == bio->bi_size)
1354 		return;
1355 
1356 	bio->bi_sector += offset;
1357 	bio->bi_size = size;
1358 	offset <<= 9;
1359 	clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1360 
1361 	while (bio->bi_idx < bio->bi_vcnt &&
1362 	       bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
1363 		/* remove this whole bio_vec */
1364 		offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
1365 		bio->bi_idx++;
1366 	}
1367 	if (bio->bi_idx < bio->bi_vcnt) {
1368 		bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
1369 		bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
1370 	}
1371 	/* avoid any complications with bi_idx being non-zero*/
1372 	if (bio->bi_idx) {
1373 		memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
1374 			(bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
1375 		bio->bi_vcnt -= bio->bi_idx;
1376 		bio->bi_idx = 0;
1377 	}
1378 	/* Make sure vcnt and last bv are not too big */
1379 	bio_for_each_segment(bvec, bio, i) {
1380 		if (sofar + bvec->bv_len > size)
1381 			bvec->bv_len = size - sofar;
1382 		if (bvec->bv_len == 0) {
1383 			bio->bi_vcnt = i;
1384 			break;
1385 		}
1386 		sofar += bvec->bv_len;
1387 	}
1388 }
1389 
1390 static void split_bio_end(struct bio *bio, int error)
1391 {
1392 	struct split_bio *split_bio = bio->bi_private;
1393 
1394 	if (error)
1395 		split_bio->err = error;
1396 
1397 	if (atomic_dec_and_test(&split_bio->pending)) {
1398 		split_bio->bio->bi_phys_segments = 0;
1399 		bio_endio(split_bio->bio, split_bio->err);
1400 		kfree(split_bio);
1401 	}
1402 	bio_put(bio);
1403 }
1404 
1405 static int blkif_recover(struct blkfront_info *info)
1406 {
1407 	int i;
1408 	struct request *req, *n;
1409 	struct blk_shadow *copy;
1410 	int rc;
1411 	struct bio *bio, *cloned_bio;
1412 	struct bio_list bio_list, merge_bio;
1413 	unsigned int segs, offset;
1414 	int pending, size;
1415 	struct split_bio *split_bio;
1416 	struct list_head requests;
1417 
1418 	/* Stage 1: Make a safe copy of the shadow state. */
1419 	copy = kmemdup(info->shadow, sizeof(info->shadow),
1420 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1421 	if (!copy)
1422 		return -ENOMEM;
1423 
1424 	/* Stage 2: Set up free list. */
1425 	memset(&info->shadow, 0, sizeof(info->shadow));
1426 	for (i = 0; i < BLK_RING_SIZE; i++)
1427 		info->shadow[i].req.u.rw.id = i+1;
1428 	info->shadow_free = info->ring.req_prod_pvt;
1429 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1430 
1431 	rc = blkfront_setup_indirect(info);
1432 	if (rc) {
1433 		kfree(copy);
1434 		return rc;
1435 	}
1436 
1437 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1438 	blk_queue_max_segments(info->rq, segs);
1439 	bio_list_init(&bio_list);
1440 	INIT_LIST_HEAD(&requests);
1441 	for (i = 0; i < BLK_RING_SIZE; i++) {
1442 		/* Not in use? */
1443 		if (!copy[i].request)
1444 			continue;
1445 
1446 		/*
1447 		 * Get the bios in the request so we can re-queue them.
1448 		 */
1449 		if (copy[i].request->cmd_flags &
1450 		    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1451 			/*
1452 			 * Flush operations don't contain bios, so
1453 			 * we need to requeue the whole request
1454 			 */
1455 			list_add(&copy[i].request->queuelist, &requests);
1456 			continue;
1457 		}
1458 		merge_bio.head = copy[i].request->bio;
1459 		merge_bio.tail = copy[i].request->biotail;
1460 		bio_list_merge(&bio_list, &merge_bio);
1461 		copy[i].request->bio = NULL;
1462 		blk_put_request(copy[i].request);
1463 	}
1464 
1465 	kfree(copy);
1466 
1467 	/*
1468 	 * Empty the queue, this is important because we might have
1469 	 * requests in the queue with more segments than what we
1470 	 * can handle now.
1471 	 */
1472 	spin_lock_irq(&info->io_lock);
1473 	while ((req = blk_fetch_request(info->rq)) != NULL) {
1474 		if (req->cmd_flags &
1475 		    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1476 			list_add(&req->queuelist, &requests);
1477 			continue;
1478 		}
1479 		merge_bio.head = req->bio;
1480 		merge_bio.tail = req->biotail;
1481 		bio_list_merge(&bio_list, &merge_bio);
1482 		req->bio = NULL;
1483 		if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1484 			pr_alert("diskcache flush request found!\n");
1485 		__blk_put_request(info->rq, req);
1486 	}
1487 	spin_unlock_irq(&info->io_lock);
1488 
1489 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1490 
1491 	spin_lock_irq(&info->io_lock);
1492 
1493 	/* Now safe for us to use the shared ring */
1494 	info->connected = BLKIF_STATE_CONNECTED;
1495 
1496 	/* Kick any other new requests queued since we resumed */
1497 	kick_pending_request_queues(info);
1498 
1499 	list_for_each_entry_safe(req, n, &requests, queuelist) {
1500 		/* Requeue pending requests (flush or discard) */
1501 		list_del_init(&req->queuelist);
1502 		BUG_ON(req->nr_phys_segments > segs);
1503 		blk_requeue_request(info->rq, req);
1504 	}
1505 	spin_unlock_irq(&info->io_lock);
1506 
1507 	while ((bio = bio_list_pop(&bio_list)) != NULL) {
1508 		/* Traverse the list of pending bios and re-queue them */
1509 		if (bio_segments(bio) > segs) {
1510 			/*
1511 			 * This bio has more segments than what we can
1512 			 * handle, we have to split it.
1513 			 */
1514 			pending = (bio_segments(bio) + segs - 1) / segs;
1515 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1516 			BUG_ON(split_bio == NULL);
1517 			atomic_set(&split_bio->pending, pending);
1518 			split_bio->bio = bio;
1519 			for (i = 0; i < pending; i++) {
1520 				offset = (i * segs * PAGE_SIZE) >> 9;
1521 				size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1522 					   (unsigned int)(bio->bi_size >> 9) - offset);
1523 				cloned_bio = bio_clone(bio, GFP_NOIO);
1524 				BUG_ON(cloned_bio == NULL);
1525 				trim_bio(cloned_bio, offset, size);
1526 				cloned_bio->bi_private = split_bio;
1527 				cloned_bio->bi_end_io = split_bio_end;
1528 				submit_bio(cloned_bio->bi_rw, cloned_bio);
1529 			}
1530 			/*
1531 			 * Now we have to wait for all those smaller bios to
1532 			 * end, so we can also end the "parent" bio.
1533 			 */
1534 			continue;
1535 		}
1536 		/* We don't need to split this bio */
1537 		submit_bio(bio->bi_rw, bio);
1538 	}
1539 
1540 	return 0;
1541 }
1542 
1543 /**
1544  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1545  * driver restart.  We tear down our blkif structure and recreate it, but
1546  * leave the device-layer structures intact so that this is transparent to the
1547  * rest of the kernel.
1548  */
1549 static int blkfront_resume(struct xenbus_device *dev)
1550 {
1551 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1552 	int err;
1553 
1554 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1555 
1556 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1557 
1558 	err = talk_to_blkback(dev, info);
1559 
1560 	/*
1561 	 * We have to wait for the backend to switch to
1562 	 * connected state, since we want to read which
1563 	 * features it supports.
1564 	 */
1565 
1566 	return err;
1567 }
1568 
1569 static void
1570 blkfront_closing(struct blkfront_info *info)
1571 {
1572 	struct xenbus_device *xbdev = info->xbdev;
1573 	struct block_device *bdev = NULL;
1574 
1575 	mutex_lock(&info->mutex);
1576 
1577 	if (xbdev->state == XenbusStateClosing) {
1578 		mutex_unlock(&info->mutex);
1579 		return;
1580 	}
1581 
1582 	if (info->gd)
1583 		bdev = bdget_disk(info->gd, 0);
1584 
1585 	mutex_unlock(&info->mutex);
1586 
1587 	if (!bdev) {
1588 		xenbus_frontend_closed(xbdev);
1589 		return;
1590 	}
1591 
1592 	mutex_lock(&bdev->bd_mutex);
1593 
1594 	if (bdev->bd_openers) {
1595 		xenbus_dev_error(xbdev, -EBUSY,
1596 				 "Device in use; refusing to close");
1597 		xenbus_switch_state(xbdev, XenbusStateClosing);
1598 	} else {
1599 		xlvbd_release_gendisk(info);
1600 		xenbus_frontend_closed(xbdev);
1601 	}
1602 
1603 	mutex_unlock(&bdev->bd_mutex);
1604 	bdput(bdev);
1605 }
1606 
1607 static void blkfront_setup_discard(struct blkfront_info *info)
1608 {
1609 	int err;
1610 	char *type;
1611 	unsigned int discard_granularity;
1612 	unsigned int discard_alignment;
1613 	unsigned int discard_secure;
1614 
1615 	type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1616 	if (IS_ERR(type))
1617 		return;
1618 
1619 	info->feature_secdiscard = 0;
1620 	if (strncmp(type, "phy", 3) == 0) {
1621 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1622 			"discard-granularity", "%u", &discard_granularity,
1623 			"discard-alignment", "%u", &discard_alignment,
1624 			NULL);
1625 		if (!err) {
1626 			info->feature_discard = 1;
1627 			info->discard_granularity = discard_granularity;
1628 			info->discard_alignment = discard_alignment;
1629 		}
1630 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1631 			    "discard-secure", "%d", &discard_secure,
1632 			    NULL);
1633 		if (!err)
1634 			info->feature_secdiscard = discard_secure;
1635 
1636 	} else if (strncmp(type, "file", 4) == 0)
1637 		info->feature_discard = 1;
1638 
1639 	kfree(type);
1640 }
1641 
1642 static int blkfront_setup_indirect(struct blkfront_info *info)
1643 {
1644 	unsigned int indirect_segments, segs;
1645 	int err, i;
1646 
1647 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1648 			    "feature-max-indirect-segments", "%u", &indirect_segments,
1649 			    NULL);
1650 	if (err) {
1651 		info->max_indirect_segments = 0;
1652 		segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1653 	} else {
1654 		info->max_indirect_segments = min(indirect_segments,
1655 						  xen_blkif_max_segments);
1656 		segs = info->max_indirect_segments;
1657 	}
1658 
1659 	err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1660 	if (err)
1661 		goto out_of_memory;
1662 
1663 	for (i = 0; i < BLK_RING_SIZE; i++) {
1664 		info->shadow[i].grants_used = kzalloc(
1665 			sizeof(info->shadow[i].grants_used[0]) * segs,
1666 			GFP_NOIO);
1667 		info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1668 		if (info->max_indirect_segments)
1669 			info->shadow[i].indirect_grants = kzalloc(
1670 				sizeof(info->shadow[i].indirect_grants[0]) *
1671 				INDIRECT_GREFS(segs),
1672 				GFP_NOIO);
1673 		if ((info->shadow[i].grants_used == NULL) ||
1674 			(info->shadow[i].sg == NULL) ||
1675 		     (info->max_indirect_segments &&
1676 		     (info->shadow[i].indirect_grants == NULL)))
1677 			goto out_of_memory;
1678 		sg_init_table(info->shadow[i].sg, segs);
1679 	}
1680 
1681 
1682 	return 0;
1683 
1684 out_of_memory:
1685 	for (i = 0; i < BLK_RING_SIZE; i++) {
1686 		kfree(info->shadow[i].grants_used);
1687 		info->shadow[i].grants_used = NULL;
1688 		kfree(info->shadow[i].sg);
1689 		info->shadow[i].sg = NULL;
1690 		kfree(info->shadow[i].indirect_grants);
1691 		info->shadow[i].indirect_grants = NULL;
1692 	}
1693 	return -ENOMEM;
1694 }
1695 
1696 /*
1697  * Invoked when the backend is finally 'ready' (and has told produced
1698  * the details about the physical device - #sectors, size, etc).
1699  */
1700 static void blkfront_connect(struct blkfront_info *info)
1701 {
1702 	unsigned long long sectors;
1703 	unsigned long sector_size;
1704 	unsigned int physical_sector_size;
1705 	unsigned int binfo;
1706 	int err;
1707 	int barrier, flush, discard, persistent;
1708 
1709 	switch (info->connected) {
1710 	case BLKIF_STATE_CONNECTED:
1711 		/*
1712 		 * Potentially, the back-end may be signalling
1713 		 * a capacity change; update the capacity.
1714 		 */
1715 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1716 				   "sectors", "%Lu", &sectors);
1717 		if (XENBUS_EXIST_ERR(err))
1718 			return;
1719 		printk(KERN_INFO "Setting capacity to %Lu\n",
1720 		       sectors);
1721 		set_capacity(info->gd, sectors);
1722 		revalidate_disk(info->gd);
1723 
1724 		return;
1725 	case BLKIF_STATE_SUSPENDED:
1726 		/*
1727 		 * If we are recovering from suspension, we need to wait
1728 		 * for the backend to announce it's features before
1729 		 * reconnecting, at least we need to know if the backend
1730 		 * supports indirect descriptors, and how many.
1731 		 */
1732 		blkif_recover(info);
1733 		return;
1734 
1735 	default:
1736 		break;
1737 	}
1738 
1739 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1740 		__func__, info->xbdev->otherend);
1741 
1742 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1743 			    "sectors", "%llu", &sectors,
1744 			    "info", "%u", &binfo,
1745 			    "sector-size", "%lu", &sector_size,
1746 			    NULL);
1747 	if (err) {
1748 		xenbus_dev_fatal(info->xbdev, err,
1749 				 "reading backend fields at %s",
1750 				 info->xbdev->otherend);
1751 		return;
1752 	}
1753 
1754 	/*
1755 	 * physcial-sector-size is a newer field, so old backends may not
1756 	 * provide this. Assume physical sector size to be the same as
1757 	 * sector_size in that case.
1758 	 */
1759 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1760 			   "physical-sector-size", "%u", &physical_sector_size);
1761 	if (err != 1)
1762 		physical_sector_size = sector_size;
1763 
1764 	info->feature_flush = 0;
1765 	info->flush_op = 0;
1766 
1767 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1768 			    "feature-barrier", "%d", &barrier,
1769 			    NULL);
1770 
1771 	/*
1772 	 * If there's no "feature-barrier" defined, then it means
1773 	 * we're dealing with a very old backend which writes
1774 	 * synchronously; nothing to do.
1775 	 *
1776 	 * If there are barriers, then we use flush.
1777 	 */
1778 	if (!err && barrier) {
1779 		info->feature_flush = REQ_FLUSH | REQ_FUA;
1780 		info->flush_op = BLKIF_OP_WRITE_BARRIER;
1781 	}
1782 	/*
1783 	 * And if there is "feature-flush-cache" use that above
1784 	 * barriers.
1785 	 */
1786 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1787 			    "feature-flush-cache", "%d", &flush,
1788 			    NULL);
1789 
1790 	if (!err && flush) {
1791 		info->feature_flush = REQ_FLUSH;
1792 		info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1793 	}
1794 
1795 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1796 			    "feature-discard", "%d", &discard,
1797 			    NULL);
1798 
1799 	if (!err && discard)
1800 		blkfront_setup_discard(info);
1801 
1802 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1803 			    "feature-persistent", "%u", &persistent,
1804 			    NULL);
1805 	if (err)
1806 		info->feature_persistent = 0;
1807 	else
1808 		info->feature_persistent = persistent;
1809 
1810 	err = blkfront_setup_indirect(info);
1811 	if (err) {
1812 		xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1813 				 info->xbdev->otherend);
1814 		return;
1815 	}
1816 
1817 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1818 				  physical_sector_size);
1819 	if (err) {
1820 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1821 				 info->xbdev->otherend);
1822 		return;
1823 	}
1824 
1825 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1826 
1827 	/* Kick pending requests. */
1828 	spin_lock_irq(&info->io_lock);
1829 	info->connected = BLKIF_STATE_CONNECTED;
1830 	kick_pending_request_queues(info);
1831 	spin_unlock_irq(&info->io_lock);
1832 
1833 	add_disk(info->gd);
1834 
1835 	info->is_ready = 1;
1836 }
1837 
1838 /**
1839  * Callback received when the backend's state changes.
1840  */
1841 static void blkback_changed(struct xenbus_device *dev,
1842 			    enum xenbus_state backend_state)
1843 {
1844 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1845 
1846 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1847 
1848 	switch (backend_state) {
1849 	case XenbusStateInitialising:
1850 	case XenbusStateInitWait:
1851 	case XenbusStateInitialised:
1852 	case XenbusStateReconfiguring:
1853 	case XenbusStateReconfigured:
1854 	case XenbusStateUnknown:
1855 	case XenbusStateClosed:
1856 		break;
1857 
1858 	case XenbusStateConnected:
1859 		blkfront_connect(info);
1860 		break;
1861 
1862 	case XenbusStateClosing:
1863 		blkfront_closing(info);
1864 		break;
1865 	}
1866 }
1867 
1868 static int blkfront_remove(struct xenbus_device *xbdev)
1869 {
1870 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1871 	struct block_device *bdev = NULL;
1872 	struct gendisk *disk;
1873 
1874 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1875 
1876 	blkif_free(info, 0);
1877 
1878 	mutex_lock(&info->mutex);
1879 
1880 	disk = info->gd;
1881 	if (disk)
1882 		bdev = bdget_disk(disk, 0);
1883 
1884 	info->xbdev = NULL;
1885 	mutex_unlock(&info->mutex);
1886 
1887 	if (!bdev) {
1888 		kfree(info);
1889 		return 0;
1890 	}
1891 
1892 	/*
1893 	 * The xbdev was removed before we reached the Closed
1894 	 * state. See if it's safe to remove the disk. If the bdev
1895 	 * isn't closed yet, we let release take care of it.
1896 	 */
1897 
1898 	mutex_lock(&bdev->bd_mutex);
1899 	info = disk->private_data;
1900 
1901 	dev_warn(disk_to_dev(disk),
1902 		 "%s was hot-unplugged, %d stale handles\n",
1903 		 xbdev->nodename, bdev->bd_openers);
1904 
1905 	if (info && !bdev->bd_openers) {
1906 		xlvbd_release_gendisk(info);
1907 		disk->private_data = NULL;
1908 		kfree(info);
1909 	}
1910 
1911 	mutex_unlock(&bdev->bd_mutex);
1912 	bdput(bdev);
1913 
1914 	return 0;
1915 }
1916 
1917 static int blkfront_is_ready(struct xenbus_device *dev)
1918 {
1919 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1920 
1921 	return info->is_ready && info->xbdev;
1922 }
1923 
1924 static int blkif_open(struct block_device *bdev, fmode_t mode)
1925 {
1926 	struct gendisk *disk = bdev->bd_disk;
1927 	struct blkfront_info *info;
1928 	int err = 0;
1929 
1930 	mutex_lock(&blkfront_mutex);
1931 
1932 	info = disk->private_data;
1933 	if (!info) {
1934 		/* xbdev gone */
1935 		err = -ERESTARTSYS;
1936 		goto out;
1937 	}
1938 
1939 	mutex_lock(&info->mutex);
1940 
1941 	if (!info->gd)
1942 		/* xbdev is closed */
1943 		err = -ERESTARTSYS;
1944 
1945 	mutex_unlock(&info->mutex);
1946 
1947 out:
1948 	mutex_unlock(&blkfront_mutex);
1949 	return err;
1950 }
1951 
1952 static void blkif_release(struct gendisk *disk, fmode_t mode)
1953 {
1954 	struct blkfront_info *info = disk->private_data;
1955 	struct block_device *bdev;
1956 	struct xenbus_device *xbdev;
1957 
1958 	mutex_lock(&blkfront_mutex);
1959 
1960 	bdev = bdget_disk(disk, 0);
1961 
1962 	if (bdev->bd_openers)
1963 		goto out;
1964 
1965 	/*
1966 	 * Check if we have been instructed to close. We will have
1967 	 * deferred this request, because the bdev was still open.
1968 	 */
1969 
1970 	mutex_lock(&info->mutex);
1971 	xbdev = info->xbdev;
1972 
1973 	if (xbdev && xbdev->state == XenbusStateClosing) {
1974 		/* pending switch to state closed */
1975 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1976 		xlvbd_release_gendisk(info);
1977 		xenbus_frontend_closed(info->xbdev);
1978  	}
1979 
1980 	mutex_unlock(&info->mutex);
1981 
1982 	if (!xbdev) {
1983 		/* sudden device removal */
1984 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1985 		xlvbd_release_gendisk(info);
1986 		disk->private_data = NULL;
1987 		kfree(info);
1988 	}
1989 
1990 out:
1991 	bdput(bdev);
1992 	mutex_unlock(&blkfront_mutex);
1993 }
1994 
1995 static const struct block_device_operations xlvbd_block_fops =
1996 {
1997 	.owner = THIS_MODULE,
1998 	.open = blkif_open,
1999 	.release = blkif_release,
2000 	.getgeo = blkif_getgeo,
2001 	.ioctl = blkif_ioctl,
2002 };
2003 
2004 
2005 static const struct xenbus_device_id blkfront_ids[] = {
2006 	{ "vbd" },
2007 	{ "" }
2008 };
2009 
2010 static DEFINE_XENBUS_DRIVER(blkfront, ,
2011 	.probe = blkfront_probe,
2012 	.remove = blkfront_remove,
2013 	.resume = blkfront_resume,
2014 	.otherend_changed = blkback_changed,
2015 	.is_ready = blkfront_is_ready,
2016 );
2017 
2018 static int __init xlblk_init(void)
2019 {
2020 	int ret;
2021 
2022 	if (!xen_domain())
2023 		return -ENODEV;
2024 
2025 	if (xen_hvm_domain() && !xen_platform_pci_unplug)
2026 		return -ENODEV;
2027 
2028 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2029 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2030 		       XENVBD_MAJOR, DEV_NAME);
2031 		return -ENODEV;
2032 	}
2033 
2034 	ret = xenbus_register_frontend(&blkfront_driver);
2035 	if (ret) {
2036 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2037 		return ret;
2038 	}
2039 
2040 	return 0;
2041 }
2042 module_init(xlblk_init);
2043 
2044 
2045 static void __exit xlblk_exit(void)
2046 {
2047 	xenbus_unregister_driver(&blkfront_driver);
2048 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2049 	kfree(minors);
2050 }
2051 module_exit(xlblk_exit);
2052 
2053 MODULE_DESCRIPTION("Xen virtual block device frontend");
2054 MODULE_LICENSE("GPL");
2055 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2056 MODULE_ALIAS("xen:vbd");
2057 MODULE_ALIAS("xenblk");
2058