xref: /linux/drivers/block/xen-blkfront.c (revision 1525e06e4414dbe053ea10cbd8cc03bb150e8e27)
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 
48 #include <xen/xen.h>
49 #include <xen/xenbus.h>
50 #include <xen/grant_table.h>
51 #include <xen/events.h>
52 #include <xen/page.h>
53 #include <xen/platform_pci.h>
54 
55 #include <xen/interface/grant_table.h>
56 #include <xen/interface/io/blkif.h>
57 #include <xen/interface/io/protocols.h>
58 
59 #include <asm/xen/hypervisor.h>
60 
61 enum blkif_state {
62 	BLKIF_STATE_DISCONNECTED,
63 	BLKIF_STATE_CONNECTED,
64 	BLKIF_STATE_SUSPENDED,
65 };
66 
67 struct blk_shadow {
68 	struct blkif_request req;
69 	struct request *request;
70 	unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
71 };
72 
73 static DEFINE_MUTEX(blkfront_mutex);
74 static const struct block_device_operations xlvbd_block_fops;
75 
76 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
77 
78 /*
79  * We have one of these per vbd, whether ide, scsi or 'other'.  They
80  * hang in private_data off the gendisk structure. We may end up
81  * putting all kinds of interesting stuff here :-)
82  */
83 struct blkfront_info
84 {
85 	spinlock_t io_lock;
86 	struct mutex mutex;
87 	struct xenbus_device *xbdev;
88 	struct gendisk *gd;
89 	int vdevice;
90 	blkif_vdev_t handle;
91 	enum blkif_state connected;
92 	int ring_ref;
93 	struct blkif_front_ring ring;
94 	struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
95 	unsigned int evtchn, irq;
96 	struct request_queue *rq;
97 	struct work_struct work;
98 	struct gnttab_free_callback callback;
99 	struct blk_shadow shadow[BLK_RING_SIZE];
100 	unsigned long shadow_free;
101 	unsigned int feature_flush;
102 	unsigned int flush_op;
103 	unsigned int feature_discard:1;
104 	unsigned int feature_secdiscard:1;
105 	unsigned int discard_granularity;
106 	unsigned int discard_alignment;
107 	int is_ready;
108 };
109 
110 static unsigned int nr_minors;
111 static unsigned long *minors;
112 static DEFINE_SPINLOCK(minor_lock);
113 
114 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
115 	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
116 #define GRANT_INVALID_REF	0
117 
118 #define PARTS_PER_DISK		16
119 #define PARTS_PER_EXT_DISK      256
120 
121 #define BLKIF_MAJOR(dev) ((dev)>>8)
122 #define BLKIF_MINOR(dev) ((dev) & 0xff)
123 
124 #define EXT_SHIFT 28
125 #define EXTENDED (1<<EXT_SHIFT)
126 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
127 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
128 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
129 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
130 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
131 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
132 
133 #define DEV_NAME	"xvd"	/* name in /dev */
134 
135 static int get_id_from_freelist(struct blkfront_info *info)
136 {
137 	unsigned long free = info->shadow_free;
138 	BUG_ON(free >= BLK_RING_SIZE);
139 	info->shadow_free = info->shadow[free].req.u.rw.id;
140 	info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
141 	return free;
142 }
143 
144 static void add_id_to_freelist(struct blkfront_info *info,
145 			       unsigned long id)
146 {
147 	info->shadow[id].req.u.rw.id  = info->shadow_free;
148 	info->shadow[id].request = NULL;
149 	info->shadow_free = id;
150 }
151 
152 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
153 {
154 	unsigned int end = minor + nr;
155 	int rc;
156 
157 	if (end > nr_minors) {
158 		unsigned long *bitmap, *old;
159 
160 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
161 				 GFP_KERNEL);
162 		if (bitmap == NULL)
163 			return -ENOMEM;
164 
165 		spin_lock(&minor_lock);
166 		if (end > nr_minors) {
167 			old = minors;
168 			memcpy(bitmap, minors,
169 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
170 			minors = bitmap;
171 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
172 		} else
173 			old = bitmap;
174 		spin_unlock(&minor_lock);
175 		kfree(old);
176 	}
177 
178 	spin_lock(&minor_lock);
179 	if (find_next_bit(minors, end, minor) >= end) {
180 		bitmap_set(minors, minor, nr);
181 		rc = 0;
182 	} else
183 		rc = -EBUSY;
184 	spin_unlock(&minor_lock);
185 
186 	return rc;
187 }
188 
189 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
190 {
191 	unsigned int end = minor + nr;
192 
193 	BUG_ON(end > nr_minors);
194 	spin_lock(&minor_lock);
195 	bitmap_clear(minors,  minor, nr);
196 	spin_unlock(&minor_lock);
197 }
198 
199 static void blkif_restart_queue_callback(void *arg)
200 {
201 	struct blkfront_info *info = (struct blkfront_info *)arg;
202 	schedule_work(&info->work);
203 }
204 
205 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
206 {
207 	/* We don't have real geometry info, but let's at least return
208 	   values consistent with the size of the device */
209 	sector_t nsect = get_capacity(bd->bd_disk);
210 	sector_t cylinders = nsect;
211 
212 	hg->heads = 0xff;
213 	hg->sectors = 0x3f;
214 	sector_div(cylinders, hg->heads * hg->sectors);
215 	hg->cylinders = cylinders;
216 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
217 		hg->cylinders = 0xffff;
218 	return 0;
219 }
220 
221 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
222 		       unsigned command, unsigned long argument)
223 {
224 	struct blkfront_info *info = bdev->bd_disk->private_data;
225 	int i;
226 
227 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
228 		command, (long)argument);
229 
230 	switch (command) {
231 	case CDROMMULTISESSION:
232 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
233 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
234 			if (put_user(0, (char __user *)(argument + i)))
235 				return -EFAULT;
236 		return 0;
237 
238 	case CDROM_GET_CAPABILITY: {
239 		struct gendisk *gd = info->gd;
240 		if (gd->flags & GENHD_FL_CD)
241 			return 0;
242 		return -EINVAL;
243 	}
244 
245 	default:
246 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
247 		  command);*/
248 		return -EINVAL; /* same return as native Linux */
249 	}
250 
251 	return 0;
252 }
253 
254 /*
255  * Generate a Xen blkfront IO request from a blk layer request.  Reads
256  * and writes are handled as expected.
257  *
258  * @req: a request struct
259  */
260 static int blkif_queue_request(struct request *req)
261 {
262 	struct blkfront_info *info = req->rq_disk->private_data;
263 	unsigned long buffer_mfn;
264 	struct blkif_request *ring_req;
265 	unsigned long id;
266 	unsigned int fsect, lsect;
267 	int i, ref;
268 	grant_ref_t gref_head;
269 	struct scatterlist *sg;
270 
271 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
272 		return 1;
273 
274 	if (gnttab_alloc_grant_references(
275 		BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
276 		gnttab_request_free_callback(
277 			&info->callback,
278 			blkif_restart_queue_callback,
279 			info,
280 			BLKIF_MAX_SEGMENTS_PER_REQUEST);
281 		return 1;
282 	}
283 
284 	/* Fill out a communications ring structure. */
285 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
286 	id = get_id_from_freelist(info);
287 	info->shadow[id].request = req;
288 
289 	ring_req->u.rw.id = id;
290 	ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
291 	ring_req->u.rw.handle = info->handle;
292 
293 	ring_req->operation = rq_data_dir(req) ?
294 		BLKIF_OP_WRITE : BLKIF_OP_READ;
295 
296 	if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
297 		/*
298 		 * Ideally we can do an unordered flush-to-disk. In case the
299 		 * backend onlysupports barriers, use that. A barrier request
300 		 * a superset of FUA, so we can implement it the same
301 		 * way.  (It's also a FLUSH+FUA, since it is
302 		 * guaranteed ordered WRT previous writes.)
303 		 */
304 		ring_req->operation = info->flush_op;
305 	}
306 
307 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
308 		/* id, sector_number and handle are set above. */
309 		ring_req->operation = BLKIF_OP_DISCARD;
310 		ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
311 		if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
312 			ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
313 		else
314 			ring_req->u.discard.flag = 0;
315 	} else {
316 		ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
317 							   info->sg);
318 		BUG_ON(ring_req->u.rw.nr_segments >
319 		       BLKIF_MAX_SEGMENTS_PER_REQUEST);
320 
321 		for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
322 			buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
323 			fsect = sg->offset >> 9;
324 			lsect = fsect + (sg->length >> 9) - 1;
325 			/* install a grant reference. */
326 			ref = gnttab_claim_grant_reference(&gref_head);
327 			BUG_ON(ref == -ENOSPC);
328 
329 			gnttab_grant_foreign_access_ref(
330 					ref,
331 					info->xbdev->otherend_id,
332 					buffer_mfn,
333 					rq_data_dir(req));
334 
335 			info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
336 			ring_req->u.rw.seg[i] =
337 					(struct blkif_request_segment) {
338 						.gref       = ref,
339 						.first_sect = fsect,
340 						.last_sect  = lsect };
341 		}
342 	}
343 
344 	info->ring.req_prod_pvt++;
345 
346 	/* Keep a private copy so we can reissue requests when recovering. */
347 	info->shadow[id].req = *ring_req;
348 
349 	gnttab_free_grant_references(gref_head);
350 
351 	return 0;
352 }
353 
354 
355 static inline void flush_requests(struct blkfront_info *info)
356 {
357 	int notify;
358 
359 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
360 
361 	if (notify)
362 		notify_remote_via_irq(info->irq);
363 }
364 
365 /*
366  * do_blkif_request
367  *  read a block; request is in a request queue
368  */
369 static void do_blkif_request(struct request_queue *rq)
370 {
371 	struct blkfront_info *info = NULL;
372 	struct request *req;
373 	int queued;
374 
375 	pr_debug("Entered do_blkif_request\n");
376 
377 	queued = 0;
378 
379 	while ((req = blk_peek_request(rq)) != NULL) {
380 		info = req->rq_disk->private_data;
381 
382 		if (RING_FULL(&info->ring))
383 			goto wait;
384 
385 		blk_start_request(req);
386 
387 		if ((req->cmd_type != REQ_TYPE_FS) ||
388 		    ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
389 		    !info->flush_op)) {
390 			__blk_end_request_all(req, -EIO);
391 			continue;
392 		}
393 
394 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
395 			 "(%u/%u) buffer:%p [%s]\n",
396 			 req, req->cmd, (unsigned long)blk_rq_pos(req),
397 			 blk_rq_cur_sectors(req), blk_rq_sectors(req),
398 			 req->buffer, rq_data_dir(req) ? "write" : "read");
399 
400 		if (blkif_queue_request(req)) {
401 			blk_requeue_request(rq, req);
402 wait:
403 			/* Avoid pointless unplugs. */
404 			blk_stop_queue(rq);
405 			break;
406 		}
407 
408 		queued++;
409 	}
410 
411 	if (queued != 0)
412 		flush_requests(info);
413 }
414 
415 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
416 {
417 	struct request_queue *rq;
418 	struct blkfront_info *info = gd->private_data;
419 
420 	rq = blk_init_queue(do_blkif_request, &info->io_lock);
421 	if (rq == NULL)
422 		return -1;
423 
424 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
425 
426 	if (info->feature_discard) {
427 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
428 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
429 		rq->limits.discard_granularity = info->discard_granularity;
430 		rq->limits.discard_alignment = info->discard_alignment;
431 		if (info->feature_secdiscard)
432 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
433 	}
434 
435 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
436 	blk_queue_logical_block_size(rq, sector_size);
437 	blk_queue_max_hw_sectors(rq, 512);
438 
439 	/* Each segment in a request is up to an aligned page in size. */
440 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
441 	blk_queue_max_segment_size(rq, PAGE_SIZE);
442 
443 	/* Ensure a merged request will fit in a single I/O ring slot. */
444 	blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
445 
446 	/* Make sure buffer addresses are sector-aligned. */
447 	blk_queue_dma_alignment(rq, 511);
448 
449 	/* Make sure we don't use bounce buffers. */
450 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
451 
452 	gd->queue = rq;
453 
454 	return 0;
455 }
456 
457 
458 static void xlvbd_flush(struct blkfront_info *info)
459 {
460 	blk_queue_flush(info->rq, info->feature_flush);
461 	printk(KERN_INFO "blkfront: %s: %s: %s\n",
462 	       info->gd->disk_name,
463 	       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
464 		"barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
465 		"flush diskcache" : "barrier or flush"),
466 	       info->feature_flush ? "enabled" : "disabled");
467 }
468 
469 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
470 {
471 	int major;
472 	major = BLKIF_MAJOR(vdevice);
473 	*minor = BLKIF_MINOR(vdevice);
474 	switch (major) {
475 		case XEN_IDE0_MAJOR:
476 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
477 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
478 				EMULATED_HD_DISK_MINOR_OFFSET;
479 			break;
480 		case XEN_IDE1_MAJOR:
481 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
482 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
483 				EMULATED_HD_DISK_MINOR_OFFSET;
484 			break;
485 		case XEN_SCSI_DISK0_MAJOR:
486 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
487 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
488 			break;
489 		case XEN_SCSI_DISK1_MAJOR:
490 		case XEN_SCSI_DISK2_MAJOR:
491 		case XEN_SCSI_DISK3_MAJOR:
492 		case XEN_SCSI_DISK4_MAJOR:
493 		case XEN_SCSI_DISK5_MAJOR:
494 		case XEN_SCSI_DISK6_MAJOR:
495 		case XEN_SCSI_DISK7_MAJOR:
496 			*offset = (*minor / PARTS_PER_DISK) +
497 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
498 				EMULATED_SD_DISK_NAME_OFFSET;
499 			*minor = *minor +
500 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
501 				EMULATED_SD_DISK_MINOR_OFFSET;
502 			break;
503 		case XEN_SCSI_DISK8_MAJOR:
504 		case XEN_SCSI_DISK9_MAJOR:
505 		case XEN_SCSI_DISK10_MAJOR:
506 		case XEN_SCSI_DISK11_MAJOR:
507 		case XEN_SCSI_DISK12_MAJOR:
508 		case XEN_SCSI_DISK13_MAJOR:
509 		case XEN_SCSI_DISK14_MAJOR:
510 		case XEN_SCSI_DISK15_MAJOR:
511 			*offset = (*minor / PARTS_PER_DISK) +
512 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
513 				EMULATED_SD_DISK_NAME_OFFSET;
514 			*minor = *minor +
515 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
516 				EMULATED_SD_DISK_MINOR_OFFSET;
517 			break;
518 		case XENVBD_MAJOR:
519 			*offset = *minor / PARTS_PER_DISK;
520 			break;
521 		default:
522 			printk(KERN_WARNING "blkfront: your disk configuration is "
523 					"incorrect, please use an xvd device instead\n");
524 			return -ENODEV;
525 	}
526 	return 0;
527 }
528 
529 static char *encode_disk_name(char *ptr, unsigned int n)
530 {
531 	if (n >= 26)
532 		ptr = encode_disk_name(ptr, n / 26 - 1);
533 	*ptr = 'a' + n % 26;
534 	return ptr + 1;
535 }
536 
537 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
538 			       struct blkfront_info *info,
539 			       u16 vdisk_info, u16 sector_size)
540 {
541 	struct gendisk *gd;
542 	int nr_minors = 1;
543 	int err;
544 	unsigned int offset;
545 	int minor;
546 	int nr_parts;
547 	char *ptr;
548 
549 	BUG_ON(info->gd != NULL);
550 	BUG_ON(info->rq != NULL);
551 
552 	if ((info->vdevice>>EXT_SHIFT) > 1) {
553 		/* this is above the extended range; something is wrong */
554 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
555 		return -ENODEV;
556 	}
557 
558 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
559 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
560 		if (err)
561 			return err;
562  		nr_parts = PARTS_PER_DISK;
563 	} else {
564 		minor = BLKIF_MINOR_EXT(info->vdevice);
565 		nr_parts = PARTS_PER_EXT_DISK;
566 		offset = minor / nr_parts;
567 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
568 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
569 					"emulated IDE disks,\n\t choose an xvd device name"
570 					"from xvde on\n", info->vdevice);
571 	}
572 	if (minor >> MINORBITS) {
573 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
574 			info->vdevice, minor);
575 		return -ENODEV;
576 	}
577 
578 	if ((minor % nr_parts) == 0)
579 		nr_minors = nr_parts;
580 
581 	err = xlbd_reserve_minors(minor, nr_minors);
582 	if (err)
583 		goto out;
584 	err = -ENODEV;
585 
586 	gd = alloc_disk(nr_minors);
587 	if (gd == NULL)
588 		goto release;
589 
590 	strcpy(gd->disk_name, DEV_NAME);
591 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
592 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
593 	if (nr_minors > 1)
594 		*ptr = 0;
595 	else
596 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
597 			 "%d", minor & (nr_parts - 1));
598 
599 	gd->major = XENVBD_MAJOR;
600 	gd->first_minor = minor;
601 	gd->fops = &xlvbd_block_fops;
602 	gd->private_data = info;
603 	gd->driverfs_dev = &(info->xbdev->dev);
604 	set_capacity(gd, capacity);
605 
606 	if (xlvbd_init_blk_queue(gd, sector_size)) {
607 		del_gendisk(gd);
608 		goto release;
609 	}
610 
611 	info->rq = gd->queue;
612 	info->gd = gd;
613 
614 	xlvbd_flush(info);
615 
616 	if (vdisk_info & VDISK_READONLY)
617 		set_disk_ro(gd, 1);
618 
619 	if (vdisk_info & VDISK_REMOVABLE)
620 		gd->flags |= GENHD_FL_REMOVABLE;
621 
622 	if (vdisk_info & VDISK_CDROM)
623 		gd->flags |= GENHD_FL_CD;
624 
625 	return 0;
626 
627  release:
628 	xlbd_release_minors(minor, nr_minors);
629  out:
630 	return err;
631 }
632 
633 static void xlvbd_release_gendisk(struct blkfront_info *info)
634 {
635 	unsigned int minor, nr_minors;
636 	unsigned long flags;
637 
638 	if (info->rq == NULL)
639 		return;
640 
641 	spin_lock_irqsave(&info->io_lock, flags);
642 
643 	/* No more blkif_request(). */
644 	blk_stop_queue(info->rq);
645 
646 	/* No more gnttab callback work. */
647 	gnttab_cancel_free_callback(&info->callback);
648 	spin_unlock_irqrestore(&info->io_lock, flags);
649 
650 	/* Flush gnttab callback work. Must be done with no locks held. */
651 	flush_work_sync(&info->work);
652 
653 	del_gendisk(info->gd);
654 
655 	minor = info->gd->first_minor;
656 	nr_minors = info->gd->minors;
657 	xlbd_release_minors(minor, nr_minors);
658 
659 	blk_cleanup_queue(info->rq);
660 	info->rq = NULL;
661 
662 	put_disk(info->gd);
663 	info->gd = NULL;
664 }
665 
666 static void kick_pending_request_queues(struct blkfront_info *info)
667 {
668 	if (!RING_FULL(&info->ring)) {
669 		/* Re-enable calldowns. */
670 		blk_start_queue(info->rq);
671 		/* Kick things off immediately. */
672 		do_blkif_request(info->rq);
673 	}
674 }
675 
676 static void blkif_restart_queue(struct work_struct *work)
677 {
678 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
679 
680 	spin_lock_irq(&info->io_lock);
681 	if (info->connected == BLKIF_STATE_CONNECTED)
682 		kick_pending_request_queues(info);
683 	spin_unlock_irq(&info->io_lock);
684 }
685 
686 static void blkif_free(struct blkfront_info *info, int suspend)
687 {
688 	/* Prevent new requests being issued until we fix things up. */
689 	spin_lock_irq(&info->io_lock);
690 	info->connected = suspend ?
691 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
692 	/* No more blkif_request(). */
693 	if (info->rq)
694 		blk_stop_queue(info->rq);
695 	/* No more gnttab callback work. */
696 	gnttab_cancel_free_callback(&info->callback);
697 	spin_unlock_irq(&info->io_lock);
698 
699 	/* Flush gnttab callback work. Must be done with no locks held. */
700 	flush_work_sync(&info->work);
701 
702 	/* Free resources associated with old device channel. */
703 	if (info->ring_ref != GRANT_INVALID_REF) {
704 		gnttab_end_foreign_access(info->ring_ref, 0,
705 					  (unsigned long)info->ring.sring);
706 		info->ring_ref = GRANT_INVALID_REF;
707 		info->ring.sring = NULL;
708 	}
709 	if (info->irq)
710 		unbind_from_irqhandler(info->irq, info);
711 	info->evtchn = info->irq = 0;
712 
713 }
714 
715 static void blkif_completion(struct blk_shadow *s)
716 {
717 	int i;
718 	/* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
719 	 * flag. */
720 	for (i = 0; i < s->req.u.rw.nr_segments; i++)
721 		gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
722 }
723 
724 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
725 {
726 	struct request *req;
727 	struct blkif_response *bret;
728 	RING_IDX i, rp;
729 	unsigned long flags;
730 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
731 	int error;
732 
733 	spin_lock_irqsave(&info->io_lock, flags);
734 
735 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
736 		spin_unlock_irqrestore(&info->io_lock, flags);
737 		return IRQ_HANDLED;
738 	}
739 
740  again:
741 	rp = info->ring.sring->rsp_prod;
742 	rmb(); /* Ensure we see queued responses up to 'rp'. */
743 
744 	for (i = info->ring.rsp_cons; i != rp; i++) {
745 		unsigned long id;
746 
747 		bret = RING_GET_RESPONSE(&info->ring, i);
748 		id   = bret->id;
749 		req  = info->shadow[id].request;
750 
751 		if (bret->operation != BLKIF_OP_DISCARD)
752 			blkif_completion(&info->shadow[id]);
753 
754 		add_id_to_freelist(info, id);
755 
756 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
757 		switch (bret->operation) {
758 		case BLKIF_OP_DISCARD:
759 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
760 				struct request_queue *rq = info->rq;
761 				printk(KERN_WARNING "blkfront: %s: discard op failed\n",
762 					   info->gd->disk_name);
763 				error = -EOPNOTSUPP;
764 				info->feature_discard = 0;
765 				info->feature_secdiscard = 0;
766 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
767 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
768 			}
769 			__blk_end_request_all(req, error);
770 			break;
771 		case BLKIF_OP_FLUSH_DISKCACHE:
772 		case BLKIF_OP_WRITE_BARRIER:
773 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
774 				printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
775 				       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
776 				       "barrier" :  "flush disk cache",
777 				       info->gd->disk_name);
778 				error = -EOPNOTSUPP;
779 			}
780 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
781 				     info->shadow[id].req.u.rw.nr_segments == 0)) {
782 				printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
783 				       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
784 				       "barrier" :  "flush disk cache",
785 				       info->gd->disk_name);
786 				error = -EOPNOTSUPP;
787 			}
788 			if (unlikely(error)) {
789 				if (error == -EOPNOTSUPP)
790 					error = 0;
791 				info->feature_flush = 0;
792 				info->flush_op = 0;
793 				xlvbd_flush(info);
794 			}
795 			/* fall through */
796 		case BLKIF_OP_READ:
797 		case BLKIF_OP_WRITE:
798 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
799 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
800 					"request: %x\n", bret->status);
801 
802 			__blk_end_request_all(req, error);
803 			break;
804 		default:
805 			BUG();
806 		}
807 	}
808 
809 	info->ring.rsp_cons = i;
810 
811 	if (i != info->ring.req_prod_pvt) {
812 		int more_to_do;
813 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
814 		if (more_to_do)
815 			goto again;
816 	} else
817 		info->ring.sring->rsp_event = i + 1;
818 
819 	kick_pending_request_queues(info);
820 
821 	spin_unlock_irqrestore(&info->io_lock, flags);
822 
823 	return IRQ_HANDLED;
824 }
825 
826 
827 static int setup_blkring(struct xenbus_device *dev,
828 			 struct blkfront_info *info)
829 {
830 	struct blkif_sring *sring;
831 	int err;
832 
833 	info->ring_ref = GRANT_INVALID_REF;
834 
835 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
836 	if (!sring) {
837 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
838 		return -ENOMEM;
839 	}
840 	SHARED_RING_INIT(sring);
841 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
842 
843 	sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
844 
845 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
846 	if (err < 0) {
847 		free_page((unsigned long)sring);
848 		info->ring.sring = NULL;
849 		goto fail;
850 	}
851 	info->ring_ref = err;
852 
853 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
854 	if (err)
855 		goto fail;
856 
857 	err = bind_evtchn_to_irqhandler(info->evtchn,
858 					blkif_interrupt,
859 					IRQF_SAMPLE_RANDOM, "blkif", info);
860 	if (err <= 0) {
861 		xenbus_dev_fatal(dev, err,
862 				 "bind_evtchn_to_irqhandler failed");
863 		goto fail;
864 	}
865 	info->irq = err;
866 
867 	return 0;
868 fail:
869 	blkif_free(info, 0);
870 	return err;
871 }
872 
873 
874 /* Common code used when first setting up, and when resuming. */
875 static int talk_to_blkback(struct xenbus_device *dev,
876 			   struct blkfront_info *info)
877 {
878 	const char *message = NULL;
879 	struct xenbus_transaction xbt;
880 	int err;
881 
882 	/* Create shared ring, alloc event channel. */
883 	err = setup_blkring(dev, info);
884 	if (err)
885 		goto out;
886 
887 again:
888 	err = xenbus_transaction_start(&xbt);
889 	if (err) {
890 		xenbus_dev_fatal(dev, err, "starting transaction");
891 		goto destroy_blkring;
892 	}
893 
894 	err = xenbus_printf(xbt, dev->nodename,
895 			    "ring-ref", "%u", info->ring_ref);
896 	if (err) {
897 		message = "writing ring-ref";
898 		goto abort_transaction;
899 	}
900 	err = xenbus_printf(xbt, dev->nodename,
901 			    "event-channel", "%u", info->evtchn);
902 	if (err) {
903 		message = "writing event-channel";
904 		goto abort_transaction;
905 	}
906 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
907 			    XEN_IO_PROTO_ABI_NATIVE);
908 	if (err) {
909 		message = "writing protocol";
910 		goto abort_transaction;
911 	}
912 
913 	err = xenbus_transaction_end(xbt, 0);
914 	if (err) {
915 		if (err == -EAGAIN)
916 			goto again;
917 		xenbus_dev_fatal(dev, err, "completing transaction");
918 		goto destroy_blkring;
919 	}
920 
921 	xenbus_switch_state(dev, XenbusStateInitialised);
922 
923 	return 0;
924 
925  abort_transaction:
926 	xenbus_transaction_end(xbt, 1);
927 	if (message)
928 		xenbus_dev_fatal(dev, err, "%s", message);
929  destroy_blkring:
930 	blkif_free(info, 0);
931  out:
932 	return err;
933 }
934 
935 /**
936  * Entry point to this code when a new device is created.  Allocate the basic
937  * structures and the ring buffer for communication with the backend, and
938  * inform the backend of the appropriate details for those.  Switch to
939  * Initialised state.
940  */
941 static int blkfront_probe(struct xenbus_device *dev,
942 			  const struct xenbus_device_id *id)
943 {
944 	int err, vdevice, i;
945 	struct blkfront_info *info;
946 
947 	/* FIXME: Use dynamic device id if this is not set. */
948 	err = xenbus_scanf(XBT_NIL, dev->nodename,
949 			   "virtual-device", "%i", &vdevice);
950 	if (err != 1) {
951 		/* go looking in the extended area instead */
952 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
953 				   "%i", &vdevice);
954 		if (err != 1) {
955 			xenbus_dev_fatal(dev, err, "reading virtual-device");
956 			return err;
957 		}
958 	}
959 
960 	if (xen_hvm_domain()) {
961 		char *type;
962 		int len;
963 		/* no unplug has been done: do not hook devices != xen vbds */
964 		if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
965 			int major;
966 
967 			if (!VDEV_IS_EXTENDED(vdevice))
968 				major = BLKIF_MAJOR(vdevice);
969 			else
970 				major = XENVBD_MAJOR;
971 
972 			if (major != XENVBD_MAJOR) {
973 				printk(KERN_INFO
974 						"%s: HVM does not support vbd %d as xen block device\n",
975 						__FUNCTION__, vdevice);
976 				return -ENODEV;
977 			}
978 		}
979 		/* do not create a PV cdrom device if we are an HVM guest */
980 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
981 		if (IS_ERR(type))
982 			return -ENODEV;
983 		if (strncmp(type, "cdrom", 5) == 0) {
984 			kfree(type);
985 			return -ENODEV;
986 		}
987 		kfree(type);
988 	}
989 	info = kzalloc(sizeof(*info), GFP_KERNEL);
990 	if (!info) {
991 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
992 		return -ENOMEM;
993 	}
994 
995 	mutex_init(&info->mutex);
996 	spin_lock_init(&info->io_lock);
997 	info->xbdev = dev;
998 	info->vdevice = vdevice;
999 	info->connected = BLKIF_STATE_DISCONNECTED;
1000 	INIT_WORK(&info->work, blkif_restart_queue);
1001 
1002 	for (i = 0; i < BLK_RING_SIZE; i++)
1003 		info->shadow[i].req.u.rw.id = i+1;
1004 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1005 
1006 	/* Front end dir is a number, which is used as the id. */
1007 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1008 	dev_set_drvdata(&dev->dev, info);
1009 
1010 	err = talk_to_blkback(dev, info);
1011 	if (err) {
1012 		kfree(info);
1013 		dev_set_drvdata(&dev->dev, NULL);
1014 		return err;
1015 	}
1016 
1017 	return 0;
1018 }
1019 
1020 
1021 static int blkif_recover(struct blkfront_info *info)
1022 {
1023 	int i;
1024 	struct blkif_request *req;
1025 	struct blk_shadow *copy;
1026 	int j;
1027 
1028 	/* Stage 1: Make a safe copy of the shadow state. */
1029 	copy = kmalloc(sizeof(info->shadow),
1030 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1031 	if (!copy)
1032 		return -ENOMEM;
1033 	memcpy(copy, info->shadow, sizeof(info->shadow));
1034 
1035 	/* Stage 2: Set up free list. */
1036 	memset(&info->shadow, 0, sizeof(info->shadow));
1037 	for (i = 0; i < BLK_RING_SIZE; i++)
1038 		info->shadow[i].req.u.rw.id = i+1;
1039 	info->shadow_free = info->ring.req_prod_pvt;
1040 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1041 
1042 	/* Stage 3: Find pending requests and requeue them. */
1043 	for (i = 0; i < BLK_RING_SIZE; i++) {
1044 		/* Not in use? */
1045 		if (!copy[i].request)
1046 			continue;
1047 
1048 		/* Grab a request slot and copy shadow state into it. */
1049 		req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1050 		*req = copy[i].req;
1051 
1052 		/* We get a new request id, and must reset the shadow state. */
1053 		req->u.rw.id = get_id_from_freelist(info);
1054 		memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1055 
1056 		if (req->operation != BLKIF_OP_DISCARD) {
1057 		/* Rewrite any grant references invalidated by susp/resume. */
1058 			for (j = 0; j < req->u.rw.nr_segments; j++)
1059 				gnttab_grant_foreign_access_ref(
1060 					req->u.rw.seg[j].gref,
1061 					info->xbdev->otherend_id,
1062 					pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1063 					rq_data_dir(info->shadow[req->u.rw.id].request));
1064 		}
1065 		info->shadow[req->u.rw.id].req = *req;
1066 
1067 		info->ring.req_prod_pvt++;
1068 	}
1069 
1070 	kfree(copy);
1071 
1072 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1073 
1074 	spin_lock_irq(&info->io_lock);
1075 
1076 	/* Now safe for us to use the shared ring */
1077 	info->connected = BLKIF_STATE_CONNECTED;
1078 
1079 	/* Send off requeued requests */
1080 	flush_requests(info);
1081 
1082 	/* Kick any other new requests queued since we resumed */
1083 	kick_pending_request_queues(info);
1084 
1085 	spin_unlock_irq(&info->io_lock);
1086 
1087 	return 0;
1088 }
1089 
1090 /**
1091  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1092  * driver restart.  We tear down our blkif structure and recreate it, but
1093  * leave the device-layer structures intact so that this is transparent to the
1094  * rest of the kernel.
1095  */
1096 static int blkfront_resume(struct xenbus_device *dev)
1097 {
1098 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1099 	int err;
1100 
1101 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1102 
1103 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1104 
1105 	err = talk_to_blkback(dev, info);
1106 	if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1107 		err = blkif_recover(info);
1108 
1109 	return err;
1110 }
1111 
1112 static void
1113 blkfront_closing(struct blkfront_info *info)
1114 {
1115 	struct xenbus_device *xbdev = info->xbdev;
1116 	struct block_device *bdev = NULL;
1117 
1118 	mutex_lock(&info->mutex);
1119 
1120 	if (xbdev->state == XenbusStateClosing) {
1121 		mutex_unlock(&info->mutex);
1122 		return;
1123 	}
1124 
1125 	if (info->gd)
1126 		bdev = bdget_disk(info->gd, 0);
1127 
1128 	mutex_unlock(&info->mutex);
1129 
1130 	if (!bdev) {
1131 		xenbus_frontend_closed(xbdev);
1132 		return;
1133 	}
1134 
1135 	mutex_lock(&bdev->bd_mutex);
1136 
1137 	if (bdev->bd_openers) {
1138 		xenbus_dev_error(xbdev, -EBUSY,
1139 				 "Device in use; refusing to close");
1140 		xenbus_switch_state(xbdev, XenbusStateClosing);
1141 	} else {
1142 		xlvbd_release_gendisk(info);
1143 		xenbus_frontend_closed(xbdev);
1144 	}
1145 
1146 	mutex_unlock(&bdev->bd_mutex);
1147 	bdput(bdev);
1148 }
1149 
1150 static void blkfront_setup_discard(struct blkfront_info *info)
1151 {
1152 	int err;
1153 	char *type;
1154 	unsigned int discard_granularity;
1155 	unsigned int discard_alignment;
1156 	unsigned int discard_secure;
1157 
1158 	type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1159 	if (IS_ERR(type))
1160 		return;
1161 
1162 	info->feature_secdiscard = 0;
1163 	if (strncmp(type, "phy", 3) == 0) {
1164 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1165 			"discard-granularity", "%u", &discard_granularity,
1166 			"discard-alignment", "%u", &discard_alignment,
1167 			NULL);
1168 		if (!err) {
1169 			info->feature_discard = 1;
1170 			info->discard_granularity = discard_granularity;
1171 			info->discard_alignment = discard_alignment;
1172 		}
1173 		err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1174 			    "discard-secure", "%d", &discard_secure,
1175 			    NULL);
1176 		if (!err)
1177 			info->feature_secdiscard = discard_secure;
1178 
1179 	} else if (strncmp(type, "file", 4) == 0)
1180 		info->feature_discard = 1;
1181 
1182 	kfree(type);
1183 }
1184 
1185 /*
1186  * Invoked when the backend is finally 'ready' (and has told produced
1187  * the details about the physical device - #sectors, size, etc).
1188  */
1189 static void blkfront_connect(struct blkfront_info *info)
1190 {
1191 	unsigned long long sectors;
1192 	unsigned long sector_size;
1193 	unsigned int binfo;
1194 	int err;
1195 	int barrier, flush, discard;
1196 
1197 	switch (info->connected) {
1198 	case BLKIF_STATE_CONNECTED:
1199 		/*
1200 		 * Potentially, the back-end may be signalling
1201 		 * a capacity change; update the capacity.
1202 		 */
1203 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1204 				   "sectors", "%Lu", &sectors);
1205 		if (XENBUS_EXIST_ERR(err))
1206 			return;
1207 		printk(KERN_INFO "Setting capacity to %Lu\n",
1208 		       sectors);
1209 		set_capacity(info->gd, sectors);
1210 		revalidate_disk(info->gd);
1211 
1212 		/* fall through */
1213 	case BLKIF_STATE_SUSPENDED:
1214 		return;
1215 
1216 	default:
1217 		break;
1218 	}
1219 
1220 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1221 		__func__, info->xbdev->otherend);
1222 
1223 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1224 			    "sectors", "%llu", &sectors,
1225 			    "info", "%u", &binfo,
1226 			    "sector-size", "%lu", &sector_size,
1227 			    NULL);
1228 	if (err) {
1229 		xenbus_dev_fatal(info->xbdev, err,
1230 				 "reading backend fields at %s",
1231 				 info->xbdev->otherend);
1232 		return;
1233 	}
1234 
1235 	info->feature_flush = 0;
1236 	info->flush_op = 0;
1237 
1238 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1239 			    "feature-barrier", "%d", &barrier,
1240 			    NULL);
1241 
1242 	/*
1243 	 * If there's no "feature-barrier" defined, then it means
1244 	 * we're dealing with a very old backend which writes
1245 	 * synchronously; nothing to do.
1246 	 *
1247 	 * If there are barriers, then we use flush.
1248 	 */
1249 	if (!err && barrier) {
1250 		info->feature_flush = REQ_FLUSH | REQ_FUA;
1251 		info->flush_op = BLKIF_OP_WRITE_BARRIER;
1252 	}
1253 	/*
1254 	 * And if there is "feature-flush-cache" use that above
1255 	 * barriers.
1256 	 */
1257 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1258 			    "feature-flush-cache", "%d", &flush,
1259 			    NULL);
1260 
1261 	if (!err && flush) {
1262 		info->feature_flush = REQ_FLUSH;
1263 		info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1264 	}
1265 
1266 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1267 			    "feature-discard", "%d", &discard,
1268 			    NULL);
1269 
1270 	if (!err && discard)
1271 		blkfront_setup_discard(info);
1272 
1273 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1274 	if (err) {
1275 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1276 				 info->xbdev->otherend);
1277 		return;
1278 	}
1279 
1280 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1281 
1282 	/* Kick pending requests. */
1283 	spin_lock_irq(&info->io_lock);
1284 	info->connected = BLKIF_STATE_CONNECTED;
1285 	kick_pending_request_queues(info);
1286 	spin_unlock_irq(&info->io_lock);
1287 
1288 	add_disk(info->gd);
1289 
1290 	info->is_ready = 1;
1291 }
1292 
1293 /**
1294  * Callback received when the backend's state changes.
1295  */
1296 static void blkback_changed(struct xenbus_device *dev,
1297 			    enum xenbus_state backend_state)
1298 {
1299 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1300 
1301 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1302 
1303 	switch (backend_state) {
1304 	case XenbusStateInitialising:
1305 	case XenbusStateInitWait:
1306 	case XenbusStateInitialised:
1307 	case XenbusStateReconfiguring:
1308 	case XenbusStateReconfigured:
1309 	case XenbusStateUnknown:
1310 	case XenbusStateClosed:
1311 		break;
1312 
1313 	case XenbusStateConnected:
1314 		blkfront_connect(info);
1315 		break;
1316 
1317 	case XenbusStateClosing:
1318 		blkfront_closing(info);
1319 		break;
1320 	}
1321 }
1322 
1323 static int blkfront_remove(struct xenbus_device *xbdev)
1324 {
1325 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1326 	struct block_device *bdev = NULL;
1327 	struct gendisk *disk;
1328 
1329 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1330 
1331 	blkif_free(info, 0);
1332 
1333 	mutex_lock(&info->mutex);
1334 
1335 	disk = info->gd;
1336 	if (disk)
1337 		bdev = bdget_disk(disk, 0);
1338 
1339 	info->xbdev = NULL;
1340 	mutex_unlock(&info->mutex);
1341 
1342 	if (!bdev) {
1343 		kfree(info);
1344 		return 0;
1345 	}
1346 
1347 	/*
1348 	 * The xbdev was removed before we reached the Closed
1349 	 * state. See if it's safe to remove the disk. If the bdev
1350 	 * isn't closed yet, we let release take care of it.
1351 	 */
1352 
1353 	mutex_lock(&bdev->bd_mutex);
1354 	info = disk->private_data;
1355 
1356 	dev_warn(disk_to_dev(disk),
1357 		 "%s was hot-unplugged, %d stale handles\n",
1358 		 xbdev->nodename, bdev->bd_openers);
1359 
1360 	if (info && !bdev->bd_openers) {
1361 		xlvbd_release_gendisk(info);
1362 		disk->private_data = NULL;
1363 		kfree(info);
1364 	}
1365 
1366 	mutex_unlock(&bdev->bd_mutex);
1367 	bdput(bdev);
1368 
1369 	return 0;
1370 }
1371 
1372 static int blkfront_is_ready(struct xenbus_device *dev)
1373 {
1374 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1375 
1376 	return info->is_ready && info->xbdev;
1377 }
1378 
1379 static int blkif_open(struct block_device *bdev, fmode_t mode)
1380 {
1381 	struct gendisk *disk = bdev->bd_disk;
1382 	struct blkfront_info *info;
1383 	int err = 0;
1384 
1385 	mutex_lock(&blkfront_mutex);
1386 
1387 	info = disk->private_data;
1388 	if (!info) {
1389 		/* xbdev gone */
1390 		err = -ERESTARTSYS;
1391 		goto out;
1392 	}
1393 
1394 	mutex_lock(&info->mutex);
1395 
1396 	if (!info->gd)
1397 		/* xbdev is closed */
1398 		err = -ERESTARTSYS;
1399 
1400 	mutex_unlock(&info->mutex);
1401 
1402 out:
1403 	mutex_unlock(&blkfront_mutex);
1404 	return err;
1405 }
1406 
1407 static int blkif_release(struct gendisk *disk, fmode_t mode)
1408 {
1409 	struct blkfront_info *info = disk->private_data;
1410 	struct block_device *bdev;
1411 	struct xenbus_device *xbdev;
1412 
1413 	mutex_lock(&blkfront_mutex);
1414 
1415 	bdev = bdget_disk(disk, 0);
1416 
1417 	if (bdev->bd_openers)
1418 		goto out;
1419 
1420 	/*
1421 	 * Check if we have been instructed to close. We will have
1422 	 * deferred this request, because the bdev was still open.
1423 	 */
1424 
1425 	mutex_lock(&info->mutex);
1426 	xbdev = info->xbdev;
1427 
1428 	if (xbdev && xbdev->state == XenbusStateClosing) {
1429 		/* pending switch to state closed */
1430 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1431 		xlvbd_release_gendisk(info);
1432 		xenbus_frontend_closed(info->xbdev);
1433  	}
1434 
1435 	mutex_unlock(&info->mutex);
1436 
1437 	if (!xbdev) {
1438 		/* sudden device removal */
1439 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1440 		xlvbd_release_gendisk(info);
1441 		disk->private_data = NULL;
1442 		kfree(info);
1443 	}
1444 
1445 out:
1446 	bdput(bdev);
1447 	mutex_unlock(&blkfront_mutex);
1448 	return 0;
1449 }
1450 
1451 static const struct block_device_operations xlvbd_block_fops =
1452 {
1453 	.owner = THIS_MODULE,
1454 	.open = blkif_open,
1455 	.release = blkif_release,
1456 	.getgeo = blkif_getgeo,
1457 	.ioctl = blkif_ioctl,
1458 };
1459 
1460 
1461 static const struct xenbus_device_id blkfront_ids[] = {
1462 	{ "vbd" },
1463 	{ "" }
1464 };
1465 
1466 static DEFINE_XENBUS_DRIVER(blkfront, ,
1467 	.probe = blkfront_probe,
1468 	.remove = blkfront_remove,
1469 	.resume = blkfront_resume,
1470 	.otherend_changed = blkback_changed,
1471 	.is_ready = blkfront_is_ready,
1472 );
1473 
1474 static int __init xlblk_init(void)
1475 {
1476 	int ret;
1477 
1478 	if (!xen_domain())
1479 		return -ENODEV;
1480 
1481 	if (xen_hvm_domain() && !xen_platform_pci_unplug)
1482 		return -ENODEV;
1483 
1484 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1485 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1486 		       XENVBD_MAJOR, DEV_NAME);
1487 		return -ENODEV;
1488 	}
1489 
1490 	ret = xenbus_register_frontend(&blkfront_driver);
1491 	if (ret) {
1492 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1493 		return ret;
1494 	}
1495 
1496 	return 0;
1497 }
1498 module_init(xlblk_init);
1499 
1500 
1501 static void __exit xlblk_exit(void)
1502 {
1503 	xenbus_unregister_driver(&blkfront_driver);
1504 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1505 	kfree(minors);
1506 }
1507 module_exit(xlblk_exit);
1508 
1509 MODULE_DESCRIPTION("Xen virtual block device frontend");
1510 MODULE_LICENSE("GPL");
1511 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1512 MODULE_ALIAS("xen:vbd");
1513 MODULE_ALIAS("xenblk");
1514