xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision c819e2cf)
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 grants;
125 	struct list_head indirect_pages;
126 	unsigned int persistent_gnts_c;
127 	unsigned long shadow_free;
128 	unsigned int feature_flush;
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))
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 		if (info->feature_persistent) {
204 			granted_page = alloc_page(GFP_NOIO);
205 			if (!granted_page) {
206 				kfree(gnt_list_entry);
207 				goto out_of_memory;
208 			}
209 			gnt_list_entry->pfn = page_to_pfn(granted_page);
210 		}
211 
212 		gnt_list_entry->gref = GRANT_INVALID_REF;
213 		list_add(&gnt_list_entry->node, &info->grants);
214 		i++;
215 	}
216 
217 	return 0;
218 
219 out_of_memory:
220 	list_for_each_entry_safe(gnt_list_entry, n,
221 	                         &info->grants, node) {
222 		list_del(&gnt_list_entry->node);
223 		if (info->feature_persistent)
224 			__free_page(pfn_to_page(gnt_list_entry->pfn));
225 		kfree(gnt_list_entry);
226 		i--;
227 	}
228 	BUG_ON(i != 0);
229 	return -ENOMEM;
230 }
231 
232 static struct grant *get_grant(grant_ref_t *gref_head,
233                                unsigned long pfn,
234                                struct blkfront_info *info)
235 {
236 	struct grant *gnt_list_entry;
237 	unsigned long buffer_mfn;
238 
239 	BUG_ON(list_empty(&info->grants));
240 	gnt_list_entry = list_first_entry(&info->grants, struct grant,
241 	                                  node);
242 	list_del(&gnt_list_entry->node);
243 
244 	if (gnt_list_entry->gref != GRANT_INVALID_REF) {
245 		info->persistent_gnts_c--;
246 		return gnt_list_entry;
247 	}
248 
249 	/* Assign a gref to this page */
250 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
251 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
252 	if (!info->feature_persistent) {
253 		BUG_ON(!pfn);
254 		gnt_list_entry->pfn = pfn;
255 	}
256 	buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
257 	gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
258 	                                info->xbdev->otherend_id,
259 	                                buffer_mfn, 0);
260 	return gnt_list_entry;
261 }
262 
263 static const char *op_name(int op)
264 {
265 	static const char *const names[] = {
266 		[BLKIF_OP_READ] = "read",
267 		[BLKIF_OP_WRITE] = "write",
268 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
269 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
270 		[BLKIF_OP_DISCARD] = "discard" };
271 
272 	if (op < 0 || op >= ARRAY_SIZE(names))
273 		return "unknown";
274 
275 	if (!names[op])
276 		return "reserved";
277 
278 	return names[op];
279 }
280 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
281 {
282 	unsigned int end = minor + nr;
283 	int rc;
284 
285 	if (end > nr_minors) {
286 		unsigned long *bitmap, *old;
287 
288 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
289 				 GFP_KERNEL);
290 		if (bitmap == NULL)
291 			return -ENOMEM;
292 
293 		spin_lock(&minor_lock);
294 		if (end > nr_minors) {
295 			old = minors;
296 			memcpy(bitmap, minors,
297 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
298 			minors = bitmap;
299 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
300 		} else
301 			old = bitmap;
302 		spin_unlock(&minor_lock);
303 		kfree(old);
304 	}
305 
306 	spin_lock(&minor_lock);
307 	if (find_next_bit(minors, end, minor) >= end) {
308 		bitmap_set(minors, minor, nr);
309 		rc = 0;
310 	} else
311 		rc = -EBUSY;
312 	spin_unlock(&minor_lock);
313 
314 	return rc;
315 }
316 
317 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
318 {
319 	unsigned int end = minor + nr;
320 
321 	BUG_ON(end > nr_minors);
322 	spin_lock(&minor_lock);
323 	bitmap_clear(minors,  minor, nr);
324 	spin_unlock(&minor_lock);
325 }
326 
327 static void blkif_restart_queue_callback(void *arg)
328 {
329 	struct blkfront_info *info = (struct blkfront_info *)arg;
330 	schedule_work(&info->work);
331 }
332 
333 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
334 {
335 	/* We don't have real geometry info, but let's at least return
336 	   values consistent with the size of the device */
337 	sector_t nsect = get_capacity(bd->bd_disk);
338 	sector_t cylinders = nsect;
339 
340 	hg->heads = 0xff;
341 	hg->sectors = 0x3f;
342 	sector_div(cylinders, hg->heads * hg->sectors);
343 	hg->cylinders = cylinders;
344 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
345 		hg->cylinders = 0xffff;
346 	return 0;
347 }
348 
349 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
350 		       unsigned command, unsigned long argument)
351 {
352 	struct blkfront_info *info = bdev->bd_disk->private_data;
353 	int i;
354 
355 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
356 		command, (long)argument);
357 
358 	switch (command) {
359 	case CDROMMULTISESSION:
360 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
361 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
362 			if (put_user(0, (char __user *)(argument + i)))
363 				return -EFAULT;
364 		return 0;
365 
366 	case CDROM_GET_CAPABILITY: {
367 		struct gendisk *gd = info->gd;
368 		if (gd->flags & GENHD_FL_CD)
369 			return 0;
370 		return -EINVAL;
371 	}
372 
373 	default:
374 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
375 		  command);*/
376 		return -EINVAL; /* same return as native Linux */
377 	}
378 
379 	return 0;
380 }
381 
382 /*
383  * Generate a Xen blkfront IO request from a blk layer request.  Reads
384  * and writes are handled as expected.
385  *
386  * @req: a request struct
387  */
388 static int blkif_queue_request(struct request *req)
389 {
390 	struct blkfront_info *info = req->rq_disk->private_data;
391 	struct blkif_request *ring_req;
392 	unsigned long id;
393 	unsigned int fsect, lsect;
394 	int i, ref, n;
395 	struct blkif_request_segment *segments = NULL;
396 
397 	/*
398 	 * Used to store if we are able to queue the request by just using
399 	 * existing persistent grants, or if we have to get new grants,
400 	 * as there are not sufficiently many free.
401 	 */
402 	bool new_persistent_gnts;
403 	grant_ref_t gref_head;
404 	struct grant *gnt_list_entry = NULL;
405 	struct scatterlist *sg;
406 	int nseg, max_grefs;
407 
408 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
409 		return 1;
410 
411 	max_grefs = req->nr_phys_segments;
412 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
413 		/*
414 		 * If we are using indirect segments we need to account
415 		 * for the indirect grefs used in the request.
416 		 */
417 		max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
418 
419 	/* Check if we have enough grants to allocate a requests */
420 	if (info->persistent_gnts_c < max_grefs) {
421 		new_persistent_gnts = 1;
422 		if (gnttab_alloc_grant_references(
423 		    max_grefs - info->persistent_gnts_c,
424 		    &gref_head) < 0) {
425 			gnttab_request_free_callback(
426 				&info->callback,
427 				blkif_restart_queue_callback,
428 				info,
429 				max_grefs);
430 			return 1;
431 		}
432 	} else
433 		new_persistent_gnts = 0;
434 
435 	/* Fill out a communications ring structure. */
436 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
437 	id = get_id_from_freelist(info);
438 	info->shadow[id].request = req;
439 
440 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
441 		ring_req->operation = BLKIF_OP_DISCARD;
442 		ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
443 		ring_req->u.discard.id = id;
444 		ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
445 		if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
446 			ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
447 		else
448 			ring_req->u.discard.flag = 0;
449 	} else {
450 		BUG_ON(info->max_indirect_segments == 0 &&
451 		       req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
452 		BUG_ON(info->max_indirect_segments &&
453 		       req->nr_phys_segments > info->max_indirect_segments);
454 		nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
455 		ring_req->u.rw.id = id;
456 		if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
457 			/*
458 			 * The indirect operation can only be a BLKIF_OP_READ or
459 			 * BLKIF_OP_WRITE
460 			 */
461 			BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
462 			ring_req->operation = BLKIF_OP_INDIRECT;
463 			ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
464 				BLKIF_OP_WRITE : BLKIF_OP_READ;
465 			ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
466 			ring_req->u.indirect.handle = info->handle;
467 			ring_req->u.indirect.nr_segments = nseg;
468 		} else {
469 			ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
470 			ring_req->u.rw.handle = info->handle;
471 			ring_req->operation = rq_data_dir(req) ?
472 				BLKIF_OP_WRITE : BLKIF_OP_READ;
473 			if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
474 				/*
475 				 * Ideally we can do an unordered flush-to-disk. In case the
476 				 * backend onlysupports barriers, use that. A barrier request
477 				 * a superset of FUA, so we can implement it the same
478 				 * way.  (It's also a FLUSH+FUA, since it is
479 				 * guaranteed ordered WRT previous writes.)
480 				 */
481 				switch (info->feature_flush &
482 					((REQ_FLUSH|REQ_FUA))) {
483 				case REQ_FLUSH|REQ_FUA:
484 					ring_req->operation =
485 						BLKIF_OP_WRITE_BARRIER;
486 					break;
487 				case REQ_FLUSH:
488 					ring_req->operation =
489 						BLKIF_OP_FLUSH_DISKCACHE;
490 					break;
491 				default:
492 					ring_req->operation = 0;
493 				}
494 			}
495 			ring_req->u.rw.nr_segments = nseg;
496 		}
497 		for_each_sg(info->shadow[id].sg, sg, nseg, i) {
498 			fsect = sg->offset >> 9;
499 			lsect = fsect + (sg->length >> 9) - 1;
500 
501 			if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
502 			    (i % SEGS_PER_INDIRECT_FRAME == 0)) {
503 				unsigned long uninitialized_var(pfn);
504 
505 				if (segments)
506 					kunmap_atomic(segments);
507 
508 				n = i / SEGS_PER_INDIRECT_FRAME;
509 				if (!info->feature_persistent) {
510 					struct page *indirect_page;
511 
512 					/* Fetch a pre-allocated page to use for indirect grefs */
513 					BUG_ON(list_empty(&info->indirect_pages));
514 					indirect_page = list_first_entry(&info->indirect_pages,
515 					                                 struct page, lru);
516 					list_del(&indirect_page->lru);
517 					pfn = page_to_pfn(indirect_page);
518 				}
519 				gnt_list_entry = get_grant(&gref_head, pfn, info);
520 				info->shadow[id].indirect_grants[n] = gnt_list_entry;
521 				segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
522 				ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
523 			}
524 
525 			gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
526 			ref = gnt_list_entry->gref;
527 
528 			info->shadow[id].grants_used[i] = gnt_list_entry;
529 
530 			if (rq_data_dir(req) && info->feature_persistent) {
531 				char *bvec_data;
532 				void *shared_data;
533 
534 				BUG_ON(sg->offset + sg->length > PAGE_SIZE);
535 
536 				shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
537 				bvec_data = kmap_atomic(sg_page(sg));
538 
539 				/*
540 				 * this does not wipe data stored outside the
541 				 * range sg->offset..sg->offset+sg->length.
542 				 * Therefore, blkback *could* see data from
543 				 * previous requests. This is OK as long as
544 				 * persistent grants are shared with just one
545 				 * domain. It may need refactoring if this
546 				 * changes
547 				 */
548 				memcpy(shared_data + sg->offset,
549 				       bvec_data   + sg->offset,
550 				       sg->length);
551 
552 				kunmap_atomic(bvec_data);
553 				kunmap_atomic(shared_data);
554 			}
555 			if (ring_req->operation != BLKIF_OP_INDIRECT) {
556 				ring_req->u.rw.seg[i] =
557 						(struct blkif_request_segment) {
558 							.gref       = ref,
559 							.first_sect = fsect,
560 							.last_sect  = lsect };
561 			} else {
562 				n = i % SEGS_PER_INDIRECT_FRAME;
563 				segments[n] =
564 					(struct blkif_request_segment) {
565 							.gref       = ref,
566 							.first_sect = fsect,
567 							.last_sect  = lsect };
568 			}
569 		}
570 		if (segments)
571 			kunmap_atomic(segments);
572 	}
573 
574 	info->ring.req_prod_pvt++;
575 
576 	/* Keep a private copy so we can reissue requests when recovering. */
577 	info->shadow[id].req = *ring_req;
578 
579 	if (new_persistent_gnts)
580 		gnttab_free_grant_references(gref_head);
581 
582 	return 0;
583 }
584 
585 
586 static inline void flush_requests(struct blkfront_info *info)
587 {
588 	int notify;
589 
590 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
591 
592 	if (notify)
593 		notify_remote_via_irq(info->irq);
594 }
595 
596 static inline bool blkif_request_flush_invalid(struct request *req,
597 					       struct blkfront_info *info)
598 {
599 	return ((req->cmd_type != REQ_TYPE_FS) ||
600 		((req->cmd_flags & REQ_FLUSH) &&
601 		 !(info->feature_flush & REQ_FLUSH)) ||
602 		((req->cmd_flags & REQ_FUA) &&
603 		 !(info->feature_flush & REQ_FUA)));
604 }
605 
606 /*
607  * do_blkif_request
608  *  read a block; request is in a request queue
609  */
610 static void do_blkif_request(struct request_queue *rq)
611 {
612 	struct blkfront_info *info = NULL;
613 	struct request *req;
614 	int queued;
615 
616 	pr_debug("Entered do_blkif_request\n");
617 
618 	queued = 0;
619 
620 	while ((req = blk_peek_request(rq)) != NULL) {
621 		info = req->rq_disk->private_data;
622 
623 		if (RING_FULL(&info->ring))
624 			goto wait;
625 
626 		blk_start_request(req);
627 
628 		if (blkif_request_flush_invalid(req, info)) {
629 			__blk_end_request_all(req, -EOPNOTSUPP);
630 			continue;
631 		}
632 
633 		pr_debug("do_blk_req %p: cmd %p, sec %lx, "
634 			 "(%u/%u) [%s]\n",
635 			 req, req->cmd, (unsigned long)blk_rq_pos(req),
636 			 blk_rq_cur_sectors(req), blk_rq_sectors(req),
637 			 rq_data_dir(req) ? "write" : "read");
638 
639 		if (blkif_queue_request(req)) {
640 			blk_requeue_request(rq, req);
641 wait:
642 			/* Avoid pointless unplugs. */
643 			blk_stop_queue(rq);
644 			break;
645 		}
646 
647 		queued++;
648 	}
649 
650 	if (queued != 0)
651 		flush_requests(info);
652 }
653 
654 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
655 				unsigned int physical_sector_size,
656 				unsigned int segments)
657 {
658 	struct request_queue *rq;
659 	struct blkfront_info *info = gd->private_data;
660 
661 	rq = blk_init_queue(do_blkif_request, &info->io_lock);
662 	if (rq == NULL)
663 		return -1;
664 
665 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
666 
667 	if (info->feature_discard) {
668 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
669 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
670 		rq->limits.discard_granularity = info->discard_granularity;
671 		rq->limits.discard_alignment = info->discard_alignment;
672 		if (info->feature_secdiscard)
673 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
674 	}
675 
676 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
677 	blk_queue_logical_block_size(rq, sector_size);
678 	blk_queue_physical_block_size(rq, physical_sector_size);
679 	blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
680 
681 	/* Each segment in a request is up to an aligned page in size. */
682 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
683 	blk_queue_max_segment_size(rq, PAGE_SIZE);
684 
685 	/* Ensure a merged request will fit in a single I/O ring slot. */
686 	blk_queue_max_segments(rq, segments);
687 
688 	/* Make sure buffer addresses are sector-aligned. */
689 	blk_queue_dma_alignment(rq, 511);
690 
691 	/* Make sure we don't use bounce buffers. */
692 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
693 
694 	gd->queue = rq;
695 
696 	return 0;
697 }
698 
699 static const char *flush_info(unsigned int feature_flush)
700 {
701 	switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
702 	case REQ_FLUSH|REQ_FUA:
703 		return "barrier: enabled;";
704 	case REQ_FLUSH:
705 		return "flush diskcache: enabled;";
706 	default:
707 		return "barrier or flush: disabled;";
708 	}
709 }
710 
711 static void xlvbd_flush(struct blkfront_info *info)
712 {
713 	blk_queue_flush(info->rq, info->feature_flush);
714 	pr_info("blkfront: %s: %s %s %s %s %s\n",
715 		info->gd->disk_name, flush_info(info->feature_flush),
716 		"persistent grants:", info->feature_persistent ?
717 		"enabled;" : "disabled;", "indirect descriptors:",
718 		info->max_indirect_segments ? "enabled;" : "disabled;");
719 }
720 
721 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
722 {
723 	int major;
724 	major = BLKIF_MAJOR(vdevice);
725 	*minor = BLKIF_MINOR(vdevice);
726 	switch (major) {
727 		case XEN_IDE0_MAJOR:
728 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
729 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
730 				EMULATED_HD_DISK_MINOR_OFFSET;
731 			break;
732 		case XEN_IDE1_MAJOR:
733 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
734 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
735 				EMULATED_HD_DISK_MINOR_OFFSET;
736 			break;
737 		case XEN_SCSI_DISK0_MAJOR:
738 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
739 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
740 			break;
741 		case XEN_SCSI_DISK1_MAJOR:
742 		case XEN_SCSI_DISK2_MAJOR:
743 		case XEN_SCSI_DISK3_MAJOR:
744 		case XEN_SCSI_DISK4_MAJOR:
745 		case XEN_SCSI_DISK5_MAJOR:
746 		case XEN_SCSI_DISK6_MAJOR:
747 		case XEN_SCSI_DISK7_MAJOR:
748 			*offset = (*minor / PARTS_PER_DISK) +
749 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
750 				EMULATED_SD_DISK_NAME_OFFSET;
751 			*minor = *minor +
752 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
753 				EMULATED_SD_DISK_MINOR_OFFSET;
754 			break;
755 		case XEN_SCSI_DISK8_MAJOR:
756 		case XEN_SCSI_DISK9_MAJOR:
757 		case XEN_SCSI_DISK10_MAJOR:
758 		case XEN_SCSI_DISK11_MAJOR:
759 		case XEN_SCSI_DISK12_MAJOR:
760 		case XEN_SCSI_DISK13_MAJOR:
761 		case XEN_SCSI_DISK14_MAJOR:
762 		case XEN_SCSI_DISK15_MAJOR:
763 			*offset = (*minor / PARTS_PER_DISK) +
764 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
765 				EMULATED_SD_DISK_NAME_OFFSET;
766 			*minor = *minor +
767 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
768 				EMULATED_SD_DISK_MINOR_OFFSET;
769 			break;
770 		case XENVBD_MAJOR:
771 			*offset = *minor / PARTS_PER_DISK;
772 			break;
773 		default:
774 			printk(KERN_WARNING "blkfront: your disk configuration is "
775 					"incorrect, please use an xvd device instead\n");
776 			return -ENODEV;
777 	}
778 	return 0;
779 }
780 
781 static char *encode_disk_name(char *ptr, unsigned int n)
782 {
783 	if (n >= 26)
784 		ptr = encode_disk_name(ptr, n / 26 - 1);
785 	*ptr = 'a' + n % 26;
786 	return ptr + 1;
787 }
788 
789 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
790 			       struct blkfront_info *info,
791 			       u16 vdisk_info, u16 sector_size,
792 			       unsigned int physical_sector_size)
793 {
794 	struct gendisk *gd;
795 	int nr_minors = 1;
796 	int err;
797 	unsigned int offset;
798 	int minor;
799 	int nr_parts;
800 	char *ptr;
801 
802 	BUG_ON(info->gd != NULL);
803 	BUG_ON(info->rq != NULL);
804 
805 	if ((info->vdevice>>EXT_SHIFT) > 1) {
806 		/* this is above the extended range; something is wrong */
807 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
808 		return -ENODEV;
809 	}
810 
811 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
812 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
813 		if (err)
814 			return err;
815  		nr_parts = PARTS_PER_DISK;
816 	} else {
817 		minor = BLKIF_MINOR_EXT(info->vdevice);
818 		nr_parts = PARTS_PER_EXT_DISK;
819 		offset = minor / nr_parts;
820 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
821 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
822 					"emulated IDE disks,\n\t choose an xvd device name"
823 					"from xvde on\n", info->vdevice);
824 	}
825 	if (minor >> MINORBITS) {
826 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
827 			info->vdevice, minor);
828 		return -ENODEV;
829 	}
830 
831 	if ((minor % nr_parts) == 0)
832 		nr_minors = nr_parts;
833 
834 	err = xlbd_reserve_minors(minor, nr_minors);
835 	if (err)
836 		goto out;
837 	err = -ENODEV;
838 
839 	gd = alloc_disk(nr_minors);
840 	if (gd == NULL)
841 		goto release;
842 
843 	strcpy(gd->disk_name, DEV_NAME);
844 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
845 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
846 	if (nr_minors > 1)
847 		*ptr = 0;
848 	else
849 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
850 			 "%d", minor & (nr_parts - 1));
851 
852 	gd->major = XENVBD_MAJOR;
853 	gd->first_minor = minor;
854 	gd->fops = &xlvbd_block_fops;
855 	gd->private_data = info;
856 	gd->driverfs_dev = &(info->xbdev->dev);
857 	set_capacity(gd, capacity);
858 
859 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
860 				 info->max_indirect_segments ? :
861 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
862 		del_gendisk(gd);
863 		goto release;
864 	}
865 
866 	info->rq = gd->queue;
867 	info->gd = gd;
868 
869 	xlvbd_flush(info);
870 
871 	if (vdisk_info & VDISK_READONLY)
872 		set_disk_ro(gd, 1);
873 
874 	if (vdisk_info & VDISK_REMOVABLE)
875 		gd->flags |= GENHD_FL_REMOVABLE;
876 
877 	if (vdisk_info & VDISK_CDROM)
878 		gd->flags |= GENHD_FL_CD;
879 
880 	return 0;
881 
882  release:
883 	xlbd_release_minors(minor, nr_minors);
884  out:
885 	return err;
886 }
887 
888 static void xlvbd_release_gendisk(struct blkfront_info *info)
889 {
890 	unsigned int minor, nr_minors;
891 	unsigned long flags;
892 
893 	if (info->rq == NULL)
894 		return;
895 
896 	spin_lock_irqsave(&info->io_lock, flags);
897 
898 	/* No more blkif_request(). */
899 	blk_stop_queue(info->rq);
900 
901 	/* No more gnttab callback work. */
902 	gnttab_cancel_free_callback(&info->callback);
903 	spin_unlock_irqrestore(&info->io_lock, flags);
904 
905 	/* Flush gnttab callback work. Must be done with no locks held. */
906 	flush_work(&info->work);
907 
908 	del_gendisk(info->gd);
909 
910 	minor = info->gd->first_minor;
911 	nr_minors = info->gd->minors;
912 	xlbd_release_minors(minor, nr_minors);
913 
914 	blk_cleanup_queue(info->rq);
915 	info->rq = NULL;
916 
917 	put_disk(info->gd);
918 	info->gd = NULL;
919 }
920 
921 static void kick_pending_request_queues(struct blkfront_info *info)
922 {
923 	if (!RING_FULL(&info->ring)) {
924 		/* Re-enable calldowns. */
925 		blk_start_queue(info->rq);
926 		/* Kick things off immediately. */
927 		do_blkif_request(info->rq);
928 	}
929 }
930 
931 static void blkif_restart_queue(struct work_struct *work)
932 {
933 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
934 
935 	spin_lock_irq(&info->io_lock);
936 	if (info->connected == BLKIF_STATE_CONNECTED)
937 		kick_pending_request_queues(info);
938 	spin_unlock_irq(&info->io_lock);
939 }
940 
941 static void blkif_free(struct blkfront_info *info, int suspend)
942 {
943 	struct grant *persistent_gnt;
944 	struct grant *n;
945 	int i, j, segs;
946 
947 	/* Prevent new requests being issued until we fix things up. */
948 	spin_lock_irq(&info->io_lock);
949 	info->connected = suspend ?
950 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
951 	/* No more blkif_request(). */
952 	if (info->rq)
953 		blk_stop_queue(info->rq);
954 
955 	/* Remove all persistent grants */
956 	if (!list_empty(&info->grants)) {
957 		list_for_each_entry_safe(persistent_gnt, n,
958 		                         &info->grants, node) {
959 			list_del(&persistent_gnt->node);
960 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
961 				gnttab_end_foreign_access(persistent_gnt->gref,
962 				                          0, 0UL);
963 				info->persistent_gnts_c--;
964 			}
965 			if (info->feature_persistent)
966 				__free_page(pfn_to_page(persistent_gnt->pfn));
967 			kfree(persistent_gnt);
968 		}
969 	}
970 	BUG_ON(info->persistent_gnts_c != 0);
971 
972 	/*
973 	 * Remove indirect pages, this only happens when using indirect
974 	 * descriptors but not persistent grants
975 	 */
976 	if (!list_empty(&info->indirect_pages)) {
977 		struct page *indirect_page, *n;
978 
979 		BUG_ON(info->feature_persistent);
980 		list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
981 			list_del(&indirect_page->lru);
982 			__free_page(indirect_page);
983 		}
984 	}
985 
986 	for (i = 0; i < BLK_RING_SIZE; i++) {
987 		/*
988 		 * Clear persistent grants present in requests already
989 		 * on the shared ring
990 		 */
991 		if (!info->shadow[i].request)
992 			goto free_shadow;
993 
994 		segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
995 		       info->shadow[i].req.u.indirect.nr_segments :
996 		       info->shadow[i].req.u.rw.nr_segments;
997 		for (j = 0; j < segs; j++) {
998 			persistent_gnt = info->shadow[i].grants_used[j];
999 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1000 			if (info->feature_persistent)
1001 				__free_page(pfn_to_page(persistent_gnt->pfn));
1002 			kfree(persistent_gnt);
1003 		}
1004 
1005 		if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1006 			/*
1007 			 * If this is not an indirect operation don't try to
1008 			 * free indirect segments
1009 			 */
1010 			goto free_shadow;
1011 
1012 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1013 			persistent_gnt = info->shadow[i].indirect_grants[j];
1014 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1015 			__free_page(pfn_to_page(persistent_gnt->pfn));
1016 			kfree(persistent_gnt);
1017 		}
1018 
1019 free_shadow:
1020 		kfree(info->shadow[i].grants_used);
1021 		info->shadow[i].grants_used = NULL;
1022 		kfree(info->shadow[i].indirect_grants);
1023 		info->shadow[i].indirect_grants = NULL;
1024 		kfree(info->shadow[i].sg);
1025 		info->shadow[i].sg = NULL;
1026 	}
1027 
1028 	/* No more gnttab callback work. */
1029 	gnttab_cancel_free_callback(&info->callback);
1030 	spin_unlock_irq(&info->io_lock);
1031 
1032 	/* Flush gnttab callback work. Must be done with no locks held. */
1033 	flush_work(&info->work);
1034 
1035 	/* Free resources associated with old device channel. */
1036 	if (info->ring_ref != GRANT_INVALID_REF) {
1037 		gnttab_end_foreign_access(info->ring_ref, 0,
1038 					  (unsigned long)info->ring.sring);
1039 		info->ring_ref = GRANT_INVALID_REF;
1040 		info->ring.sring = NULL;
1041 	}
1042 	if (info->irq)
1043 		unbind_from_irqhandler(info->irq, info);
1044 	info->evtchn = info->irq = 0;
1045 
1046 }
1047 
1048 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1049 			     struct blkif_response *bret)
1050 {
1051 	int i = 0;
1052 	struct scatterlist *sg;
1053 	char *bvec_data;
1054 	void *shared_data;
1055 	int nseg;
1056 
1057 	nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1058 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1059 
1060 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1061 		/*
1062 		 * Copy the data received from the backend into the bvec.
1063 		 * Since bv_offset can be different than 0, and bv_len different
1064 		 * than PAGE_SIZE, we have to keep track of the current offset,
1065 		 * to be sure we are copying the data from the right shared page.
1066 		 */
1067 		for_each_sg(s->sg, sg, nseg, i) {
1068 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1069 			shared_data = kmap_atomic(
1070 				pfn_to_page(s->grants_used[i]->pfn));
1071 			bvec_data = kmap_atomic(sg_page(sg));
1072 			memcpy(bvec_data   + sg->offset,
1073 			       shared_data + sg->offset,
1074 			       sg->length);
1075 			kunmap_atomic(bvec_data);
1076 			kunmap_atomic(shared_data);
1077 		}
1078 	}
1079 	/* Add the persistent grant into the list of free grants */
1080 	for (i = 0; i < nseg; i++) {
1081 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1082 			/*
1083 			 * If the grant is still mapped by the backend (the
1084 			 * backend has chosen to make this grant persistent)
1085 			 * we add it at the head of the list, so it will be
1086 			 * reused first.
1087 			 */
1088 			if (!info->feature_persistent)
1089 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1090 						     s->grants_used[i]->gref);
1091 			list_add(&s->grants_used[i]->node, &info->grants);
1092 			info->persistent_gnts_c++;
1093 		} else {
1094 			/*
1095 			 * If the grant is not mapped by the backend we end the
1096 			 * foreign access and add it to the tail of the list,
1097 			 * so it will not be picked again unless we run out of
1098 			 * persistent grants.
1099 			 */
1100 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1101 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1102 			list_add_tail(&s->grants_used[i]->node, &info->grants);
1103 		}
1104 	}
1105 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1106 		for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1107 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1108 				if (!info->feature_persistent)
1109 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1110 							     s->indirect_grants[i]->gref);
1111 				list_add(&s->indirect_grants[i]->node, &info->grants);
1112 				info->persistent_gnts_c++;
1113 			} else {
1114 				struct page *indirect_page;
1115 
1116 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1117 				/*
1118 				 * Add the used indirect page back to the list of
1119 				 * available pages for indirect grefs.
1120 				 */
1121 				indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1122 				list_add(&indirect_page->lru, &info->indirect_pages);
1123 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1124 				list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1125 			}
1126 		}
1127 	}
1128 }
1129 
1130 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1131 {
1132 	struct request *req;
1133 	struct blkif_response *bret;
1134 	RING_IDX i, rp;
1135 	unsigned long flags;
1136 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
1137 	int error;
1138 
1139 	spin_lock_irqsave(&info->io_lock, flags);
1140 
1141 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1142 		spin_unlock_irqrestore(&info->io_lock, flags);
1143 		return IRQ_HANDLED;
1144 	}
1145 
1146  again:
1147 	rp = info->ring.sring->rsp_prod;
1148 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1149 
1150 	for (i = info->ring.rsp_cons; i != rp; i++) {
1151 		unsigned long id;
1152 
1153 		bret = RING_GET_RESPONSE(&info->ring, i);
1154 		id   = bret->id;
1155 		/*
1156 		 * The backend has messed up and given us an id that we would
1157 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1158 		 * look in get_id_from_freelist.
1159 		 */
1160 		if (id >= BLK_RING_SIZE) {
1161 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1162 			     info->gd->disk_name, op_name(bret->operation), id);
1163 			/* We can't safely get the 'struct request' as
1164 			 * the id is busted. */
1165 			continue;
1166 		}
1167 		req  = info->shadow[id].request;
1168 
1169 		if (bret->operation != BLKIF_OP_DISCARD)
1170 			blkif_completion(&info->shadow[id], info, bret);
1171 
1172 		if (add_id_to_freelist(info, id)) {
1173 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1174 			     info->gd->disk_name, op_name(bret->operation), id);
1175 			continue;
1176 		}
1177 
1178 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1179 		switch (bret->operation) {
1180 		case BLKIF_OP_DISCARD:
1181 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1182 				struct request_queue *rq = info->rq;
1183 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1184 					   info->gd->disk_name, op_name(bret->operation));
1185 				error = -EOPNOTSUPP;
1186 				info->feature_discard = 0;
1187 				info->feature_secdiscard = 0;
1188 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1189 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1190 			}
1191 			__blk_end_request_all(req, error);
1192 			break;
1193 		case BLKIF_OP_FLUSH_DISKCACHE:
1194 		case BLKIF_OP_WRITE_BARRIER:
1195 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1196 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1197 				       info->gd->disk_name, op_name(bret->operation));
1198 				error = -EOPNOTSUPP;
1199 			}
1200 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1201 				     info->shadow[id].req.u.rw.nr_segments == 0)) {
1202 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1203 				       info->gd->disk_name, op_name(bret->operation));
1204 				error = -EOPNOTSUPP;
1205 			}
1206 			if (unlikely(error)) {
1207 				if (error == -EOPNOTSUPP)
1208 					error = 0;
1209 				info->feature_flush = 0;
1210 				xlvbd_flush(info);
1211 			}
1212 			/* fall through */
1213 		case BLKIF_OP_READ:
1214 		case BLKIF_OP_WRITE:
1215 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1216 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1217 					"request: %x\n", bret->status);
1218 
1219 			__blk_end_request_all(req, error);
1220 			break;
1221 		default:
1222 			BUG();
1223 		}
1224 	}
1225 
1226 	info->ring.rsp_cons = i;
1227 
1228 	if (i != info->ring.req_prod_pvt) {
1229 		int more_to_do;
1230 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1231 		if (more_to_do)
1232 			goto again;
1233 	} else
1234 		info->ring.sring->rsp_event = i + 1;
1235 
1236 	kick_pending_request_queues(info);
1237 
1238 	spin_unlock_irqrestore(&info->io_lock, flags);
1239 
1240 	return IRQ_HANDLED;
1241 }
1242 
1243 
1244 static int setup_blkring(struct xenbus_device *dev,
1245 			 struct blkfront_info *info)
1246 {
1247 	struct blkif_sring *sring;
1248 	int err;
1249 
1250 	info->ring_ref = GRANT_INVALID_REF;
1251 
1252 	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1253 	if (!sring) {
1254 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1255 		return -ENOMEM;
1256 	}
1257 	SHARED_RING_INIT(sring);
1258 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1259 
1260 	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1261 	if (err < 0) {
1262 		free_page((unsigned long)sring);
1263 		info->ring.sring = NULL;
1264 		goto fail;
1265 	}
1266 	info->ring_ref = err;
1267 
1268 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
1269 	if (err)
1270 		goto fail;
1271 
1272 	err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1273 					"blkif", info);
1274 	if (err <= 0) {
1275 		xenbus_dev_fatal(dev, err,
1276 				 "bind_evtchn_to_irqhandler failed");
1277 		goto fail;
1278 	}
1279 	info->irq = err;
1280 
1281 	return 0;
1282 fail:
1283 	blkif_free(info, 0);
1284 	return err;
1285 }
1286 
1287 
1288 /* Common code used when first setting up, and when resuming. */
1289 static int talk_to_blkback(struct xenbus_device *dev,
1290 			   struct blkfront_info *info)
1291 {
1292 	const char *message = NULL;
1293 	struct xenbus_transaction xbt;
1294 	int err;
1295 
1296 	/* Create shared ring, alloc event channel. */
1297 	err = setup_blkring(dev, info);
1298 	if (err)
1299 		goto out;
1300 
1301 again:
1302 	err = xenbus_transaction_start(&xbt);
1303 	if (err) {
1304 		xenbus_dev_fatal(dev, err, "starting transaction");
1305 		goto destroy_blkring;
1306 	}
1307 
1308 	err = xenbus_printf(xbt, dev->nodename,
1309 			    "ring-ref", "%u", info->ring_ref);
1310 	if (err) {
1311 		message = "writing ring-ref";
1312 		goto abort_transaction;
1313 	}
1314 	err = xenbus_printf(xbt, dev->nodename,
1315 			    "event-channel", "%u", info->evtchn);
1316 	if (err) {
1317 		message = "writing event-channel";
1318 		goto abort_transaction;
1319 	}
1320 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1321 			    XEN_IO_PROTO_ABI_NATIVE);
1322 	if (err) {
1323 		message = "writing protocol";
1324 		goto abort_transaction;
1325 	}
1326 	err = xenbus_printf(xbt, dev->nodename,
1327 			    "feature-persistent", "%u", 1);
1328 	if (err)
1329 		dev_warn(&dev->dev,
1330 			 "writing persistent grants feature to xenbus");
1331 
1332 	err = xenbus_transaction_end(xbt, 0);
1333 	if (err) {
1334 		if (err == -EAGAIN)
1335 			goto again;
1336 		xenbus_dev_fatal(dev, err, "completing transaction");
1337 		goto destroy_blkring;
1338 	}
1339 
1340 	xenbus_switch_state(dev, XenbusStateInitialised);
1341 
1342 	return 0;
1343 
1344  abort_transaction:
1345 	xenbus_transaction_end(xbt, 1);
1346 	if (message)
1347 		xenbus_dev_fatal(dev, err, "%s", message);
1348  destroy_blkring:
1349 	blkif_free(info, 0);
1350  out:
1351 	return err;
1352 }
1353 
1354 /**
1355  * Entry point to this code when a new device is created.  Allocate the basic
1356  * structures and the ring buffer for communication with the backend, and
1357  * inform the backend of the appropriate details for those.  Switch to
1358  * Initialised state.
1359  */
1360 static int blkfront_probe(struct xenbus_device *dev,
1361 			  const struct xenbus_device_id *id)
1362 {
1363 	int err, vdevice, i;
1364 	struct blkfront_info *info;
1365 
1366 	/* FIXME: Use dynamic device id if this is not set. */
1367 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1368 			   "virtual-device", "%i", &vdevice);
1369 	if (err != 1) {
1370 		/* go looking in the extended area instead */
1371 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1372 				   "%i", &vdevice);
1373 		if (err != 1) {
1374 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1375 			return err;
1376 		}
1377 	}
1378 
1379 	if (xen_hvm_domain()) {
1380 		char *type;
1381 		int len;
1382 		/* no unplug has been done: do not hook devices != xen vbds */
1383 		if (xen_has_pv_and_legacy_disk_devices()) {
1384 			int major;
1385 
1386 			if (!VDEV_IS_EXTENDED(vdevice))
1387 				major = BLKIF_MAJOR(vdevice);
1388 			else
1389 				major = XENVBD_MAJOR;
1390 
1391 			if (major != XENVBD_MAJOR) {
1392 				printk(KERN_INFO
1393 						"%s: HVM does not support vbd %d as xen block device\n",
1394 						__FUNCTION__, vdevice);
1395 				return -ENODEV;
1396 			}
1397 		}
1398 		/* do not create a PV cdrom device if we are an HVM guest */
1399 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1400 		if (IS_ERR(type))
1401 			return -ENODEV;
1402 		if (strncmp(type, "cdrom", 5) == 0) {
1403 			kfree(type);
1404 			return -ENODEV;
1405 		}
1406 		kfree(type);
1407 	}
1408 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1409 	if (!info) {
1410 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1411 		return -ENOMEM;
1412 	}
1413 
1414 	mutex_init(&info->mutex);
1415 	spin_lock_init(&info->io_lock);
1416 	info->xbdev = dev;
1417 	info->vdevice = vdevice;
1418 	INIT_LIST_HEAD(&info->grants);
1419 	INIT_LIST_HEAD(&info->indirect_pages);
1420 	info->persistent_gnts_c = 0;
1421 	info->connected = BLKIF_STATE_DISCONNECTED;
1422 	INIT_WORK(&info->work, blkif_restart_queue);
1423 
1424 	for (i = 0; i < BLK_RING_SIZE; i++)
1425 		info->shadow[i].req.u.rw.id = i+1;
1426 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1427 
1428 	/* Front end dir is a number, which is used as the id. */
1429 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1430 	dev_set_drvdata(&dev->dev, info);
1431 
1432 	err = talk_to_blkback(dev, info);
1433 	if (err) {
1434 		kfree(info);
1435 		dev_set_drvdata(&dev->dev, NULL);
1436 		return err;
1437 	}
1438 
1439 	return 0;
1440 }
1441 
1442 static void split_bio_end(struct bio *bio, int error)
1443 {
1444 	struct split_bio *split_bio = bio->bi_private;
1445 
1446 	if (error)
1447 		split_bio->err = error;
1448 
1449 	if (atomic_dec_and_test(&split_bio->pending)) {
1450 		split_bio->bio->bi_phys_segments = 0;
1451 		bio_endio(split_bio->bio, split_bio->err);
1452 		kfree(split_bio);
1453 	}
1454 	bio_put(bio);
1455 }
1456 
1457 static int blkif_recover(struct blkfront_info *info)
1458 {
1459 	int i;
1460 	struct request *req, *n;
1461 	struct blk_shadow *copy;
1462 	int rc;
1463 	struct bio *bio, *cloned_bio;
1464 	struct bio_list bio_list, merge_bio;
1465 	unsigned int segs, offset;
1466 	int pending, size;
1467 	struct split_bio *split_bio;
1468 	struct list_head requests;
1469 
1470 	/* Stage 1: Make a safe copy of the shadow state. */
1471 	copy = kmemdup(info->shadow, sizeof(info->shadow),
1472 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1473 	if (!copy)
1474 		return -ENOMEM;
1475 
1476 	/* Stage 2: Set up free list. */
1477 	memset(&info->shadow, 0, sizeof(info->shadow));
1478 	for (i = 0; i < BLK_RING_SIZE; i++)
1479 		info->shadow[i].req.u.rw.id = i+1;
1480 	info->shadow_free = info->ring.req_prod_pvt;
1481 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1482 
1483 	rc = blkfront_setup_indirect(info);
1484 	if (rc) {
1485 		kfree(copy);
1486 		return rc;
1487 	}
1488 
1489 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1490 	blk_queue_max_segments(info->rq, segs);
1491 	bio_list_init(&bio_list);
1492 	INIT_LIST_HEAD(&requests);
1493 	for (i = 0; i < BLK_RING_SIZE; i++) {
1494 		/* Not in use? */
1495 		if (!copy[i].request)
1496 			continue;
1497 
1498 		/*
1499 		 * Get the bios in the request so we can re-queue them.
1500 		 */
1501 		if (copy[i].request->cmd_flags &
1502 		    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1503 			/*
1504 			 * Flush operations don't contain bios, so
1505 			 * we need to requeue the whole request
1506 			 */
1507 			list_add(&copy[i].request->queuelist, &requests);
1508 			continue;
1509 		}
1510 		merge_bio.head = copy[i].request->bio;
1511 		merge_bio.tail = copy[i].request->biotail;
1512 		bio_list_merge(&bio_list, &merge_bio);
1513 		copy[i].request->bio = NULL;
1514 		blk_put_request(copy[i].request);
1515 	}
1516 
1517 	kfree(copy);
1518 
1519 	/*
1520 	 * Empty the queue, this is important because we might have
1521 	 * requests in the queue with more segments than what we
1522 	 * can handle now.
1523 	 */
1524 	spin_lock_irq(&info->io_lock);
1525 	while ((req = blk_fetch_request(info->rq)) != NULL) {
1526 		if (req->cmd_flags &
1527 		    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1528 			list_add(&req->queuelist, &requests);
1529 			continue;
1530 		}
1531 		merge_bio.head = req->bio;
1532 		merge_bio.tail = req->biotail;
1533 		bio_list_merge(&bio_list, &merge_bio);
1534 		req->bio = NULL;
1535 		if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1536 			pr_alert("diskcache flush request found!\n");
1537 		__blk_put_request(info->rq, req);
1538 	}
1539 	spin_unlock_irq(&info->io_lock);
1540 
1541 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1542 
1543 	spin_lock_irq(&info->io_lock);
1544 
1545 	/* Now safe for us to use the shared ring */
1546 	info->connected = BLKIF_STATE_CONNECTED;
1547 
1548 	/* Kick any other new requests queued since we resumed */
1549 	kick_pending_request_queues(info);
1550 
1551 	list_for_each_entry_safe(req, n, &requests, queuelist) {
1552 		/* Requeue pending requests (flush or discard) */
1553 		list_del_init(&req->queuelist);
1554 		BUG_ON(req->nr_phys_segments > segs);
1555 		blk_requeue_request(info->rq, req);
1556 	}
1557 	spin_unlock_irq(&info->io_lock);
1558 
1559 	while ((bio = bio_list_pop(&bio_list)) != NULL) {
1560 		/* Traverse the list of pending bios and re-queue them */
1561 		if (bio_segments(bio) > segs) {
1562 			/*
1563 			 * This bio has more segments than what we can
1564 			 * handle, we have to split it.
1565 			 */
1566 			pending = (bio_segments(bio) + segs - 1) / segs;
1567 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1568 			BUG_ON(split_bio == NULL);
1569 			atomic_set(&split_bio->pending, pending);
1570 			split_bio->bio = bio;
1571 			for (i = 0; i < pending; i++) {
1572 				offset = (i * segs * PAGE_SIZE) >> 9;
1573 				size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1574 					   (unsigned int)bio_sectors(bio) - offset);
1575 				cloned_bio = bio_clone(bio, GFP_NOIO);
1576 				BUG_ON(cloned_bio == NULL);
1577 				bio_trim(cloned_bio, offset, size);
1578 				cloned_bio->bi_private = split_bio;
1579 				cloned_bio->bi_end_io = split_bio_end;
1580 				submit_bio(cloned_bio->bi_rw, cloned_bio);
1581 			}
1582 			/*
1583 			 * Now we have to wait for all those smaller bios to
1584 			 * end, so we can also end the "parent" bio.
1585 			 */
1586 			continue;
1587 		}
1588 		/* We don't need to split this bio */
1589 		submit_bio(bio->bi_rw, bio);
1590 	}
1591 
1592 	return 0;
1593 }
1594 
1595 /**
1596  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1597  * driver restart.  We tear down our blkif structure and recreate it, but
1598  * leave the device-layer structures intact so that this is transparent to the
1599  * rest of the kernel.
1600  */
1601 static int blkfront_resume(struct xenbus_device *dev)
1602 {
1603 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1604 	int err;
1605 
1606 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1607 
1608 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1609 
1610 	err = talk_to_blkback(dev, info);
1611 
1612 	/*
1613 	 * We have to wait for the backend to switch to
1614 	 * connected state, since we want to read which
1615 	 * features it supports.
1616 	 */
1617 
1618 	return err;
1619 }
1620 
1621 static void
1622 blkfront_closing(struct blkfront_info *info)
1623 {
1624 	struct xenbus_device *xbdev = info->xbdev;
1625 	struct block_device *bdev = NULL;
1626 
1627 	mutex_lock(&info->mutex);
1628 
1629 	if (xbdev->state == XenbusStateClosing) {
1630 		mutex_unlock(&info->mutex);
1631 		return;
1632 	}
1633 
1634 	if (info->gd)
1635 		bdev = bdget_disk(info->gd, 0);
1636 
1637 	mutex_unlock(&info->mutex);
1638 
1639 	if (!bdev) {
1640 		xenbus_frontend_closed(xbdev);
1641 		return;
1642 	}
1643 
1644 	mutex_lock(&bdev->bd_mutex);
1645 
1646 	if (bdev->bd_openers) {
1647 		xenbus_dev_error(xbdev, -EBUSY,
1648 				 "Device in use; refusing to close");
1649 		xenbus_switch_state(xbdev, XenbusStateClosing);
1650 	} else {
1651 		xlvbd_release_gendisk(info);
1652 		xenbus_frontend_closed(xbdev);
1653 	}
1654 
1655 	mutex_unlock(&bdev->bd_mutex);
1656 	bdput(bdev);
1657 }
1658 
1659 static void blkfront_setup_discard(struct blkfront_info *info)
1660 {
1661 	int err;
1662 	unsigned int discard_granularity;
1663 	unsigned int discard_alignment;
1664 	unsigned int discard_secure;
1665 
1666 	info->feature_discard = 1;
1667 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1668 		"discard-granularity", "%u", &discard_granularity,
1669 		"discard-alignment", "%u", &discard_alignment,
1670 		NULL);
1671 	if (!err) {
1672 		info->discard_granularity = discard_granularity;
1673 		info->discard_alignment = discard_alignment;
1674 	}
1675 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1676 		    "discard-secure", "%d", &discard_secure,
1677 		    NULL);
1678 	if (!err)
1679 		info->feature_secdiscard = !!discard_secure;
1680 }
1681 
1682 static int blkfront_setup_indirect(struct blkfront_info *info)
1683 {
1684 	unsigned int indirect_segments, segs;
1685 	int err, i;
1686 
1687 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1688 			    "feature-max-indirect-segments", "%u", &indirect_segments,
1689 			    NULL);
1690 	if (err) {
1691 		info->max_indirect_segments = 0;
1692 		segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1693 	} else {
1694 		info->max_indirect_segments = min(indirect_segments,
1695 						  xen_blkif_max_segments);
1696 		segs = info->max_indirect_segments;
1697 	}
1698 
1699 	err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1700 	if (err)
1701 		goto out_of_memory;
1702 
1703 	if (!info->feature_persistent && info->max_indirect_segments) {
1704 		/*
1705 		 * We are using indirect descriptors but not persistent
1706 		 * grants, we need to allocate a set of pages that can be
1707 		 * used for mapping indirect grefs
1708 		 */
1709 		int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1710 
1711 		BUG_ON(!list_empty(&info->indirect_pages));
1712 		for (i = 0; i < num; i++) {
1713 			struct page *indirect_page = alloc_page(GFP_NOIO);
1714 			if (!indirect_page)
1715 				goto out_of_memory;
1716 			list_add(&indirect_page->lru, &info->indirect_pages);
1717 		}
1718 	}
1719 
1720 	for (i = 0; i < BLK_RING_SIZE; i++) {
1721 		info->shadow[i].grants_used = kzalloc(
1722 			sizeof(info->shadow[i].grants_used[0]) * segs,
1723 			GFP_NOIO);
1724 		info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1725 		if (info->max_indirect_segments)
1726 			info->shadow[i].indirect_grants = kzalloc(
1727 				sizeof(info->shadow[i].indirect_grants[0]) *
1728 				INDIRECT_GREFS(segs),
1729 				GFP_NOIO);
1730 		if ((info->shadow[i].grants_used == NULL) ||
1731 			(info->shadow[i].sg == NULL) ||
1732 		     (info->max_indirect_segments &&
1733 		     (info->shadow[i].indirect_grants == NULL)))
1734 			goto out_of_memory;
1735 		sg_init_table(info->shadow[i].sg, segs);
1736 	}
1737 
1738 
1739 	return 0;
1740 
1741 out_of_memory:
1742 	for (i = 0; i < BLK_RING_SIZE; i++) {
1743 		kfree(info->shadow[i].grants_used);
1744 		info->shadow[i].grants_used = NULL;
1745 		kfree(info->shadow[i].sg);
1746 		info->shadow[i].sg = NULL;
1747 		kfree(info->shadow[i].indirect_grants);
1748 		info->shadow[i].indirect_grants = NULL;
1749 	}
1750 	if (!list_empty(&info->indirect_pages)) {
1751 		struct page *indirect_page, *n;
1752 		list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1753 			list_del(&indirect_page->lru);
1754 			__free_page(indirect_page);
1755 		}
1756 	}
1757 	return -ENOMEM;
1758 }
1759 
1760 /*
1761  * Invoked when the backend is finally 'ready' (and has told produced
1762  * the details about the physical device - #sectors, size, etc).
1763  */
1764 static void blkfront_connect(struct blkfront_info *info)
1765 {
1766 	unsigned long long sectors;
1767 	unsigned long sector_size;
1768 	unsigned int physical_sector_size;
1769 	unsigned int binfo;
1770 	int err;
1771 	int barrier, flush, discard, persistent;
1772 
1773 	switch (info->connected) {
1774 	case BLKIF_STATE_CONNECTED:
1775 		/*
1776 		 * Potentially, the back-end may be signalling
1777 		 * a capacity change; update the capacity.
1778 		 */
1779 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1780 				   "sectors", "%Lu", &sectors);
1781 		if (XENBUS_EXIST_ERR(err))
1782 			return;
1783 		printk(KERN_INFO "Setting capacity to %Lu\n",
1784 		       sectors);
1785 		set_capacity(info->gd, sectors);
1786 		revalidate_disk(info->gd);
1787 
1788 		return;
1789 	case BLKIF_STATE_SUSPENDED:
1790 		/*
1791 		 * If we are recovering from suspension, we need to wait
1792 		 * for the backend to announce it's features before
1793 		 * reconnecting, at least we need to know if the backend
1794 		 * supports indirect descriptors, and how many.
1795 		 */
1796 		blkif_recover(info);
1797 		return;
1798 
1799 	default:
1800 		break;
1801 	}
1802 
1803 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1804 		__func__, info->xbdev->otherend);
1805 
1806 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1807 			    "sectors", "%llu", &sectors,
1808 			    "info", "%u", &binfo,
1809 			    "sector-size", "%lu", &sector_size,
1810 			    NULL);
1811 	if (err) {
1812 		xenbus_dev_fatal(info->xbdev, err,
1813 				 "reading backend fields at %s",
1814 				 info->xbdev->otherend);
1815 		return;
1816 	}
1817 
1818 	/*
1819 	 * physcial-sector-size is a newer field, so old backends may not
1820 	 * provide this. Assume physical sector size to be the same as
1821 	 * sector_size in that case.
1822 	 */
1823 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1824 			   "physical-sector-size", "%u", &physical_sector_size);
1825 	if (err != 1)
1826 		physical_sector_size = sector_size;
1827 
1828 	info->feature_flush = 0;
1829 
1830 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1831 			    "feature-barrier", "%d", &barrier,
1832 			    NULL);
1833 
1834 	/*
1835 	 * If there's no "feature-barrier" defined, then it means
1836 	 * we're dealing with a very old backend which writes
1837 	 * synchronously; nothing to do.
1838 	 *
1839 	 * If there are barriers, then we use flush.
1840 	 */
1841 	if (!err && barrier)
1842 		info->feature_flush = REQ_FLUSH | REQ_FUA;
1843 	/*
1844 	 * And if there is "feature-flush-cache" use that above
1845 	 * barriers.
1846 	 */
1847 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1848 			    "feature-flush-cache", "%d", &flush,
1849 			    NULL);
1850 
1851 	if (!err && flush)
1852 		info->feature_flush = REQ_FLUSH;
1853 
1854 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1855 			    "feature-discard", "%d", &discard,
1856 			    NULL);
1857 
1858 	if (!err && discard)
1859 		blkfront_setup_discard(info);
1860 
1861 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1862 			    "feature-persistent", "%u", &persistent,
1863 			    NULL);
1864 	if (err)
1865 		info->feature_persistent = 0;
1866 	else
1867 		info->feature_persistent = persistent;
1868 
1869 	err = blkfront_setup_indirect(info);
1870 	if (err) {
1871 		xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1872 				 info->xbdev->otherend);
1873 		return;
1874 	}
1875 
1876 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1877 				  physical_sector_size);
1878 	if (err) {
1879 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1880 				 info->xbdev->otherend);
1881 		return;
1882 	}
1883 
1884 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1885 
1886 	/* Kick pending requests. */
1887 	spin_lock_irq(&info->io_lock);
1888 	info->connected = BLKIF_STATE_CONNECTED;
1889 	kick_pending_request_queues(info);
1890 	spin_unlock_irq(&info->io_lock);
1891 
1892 	add_disk(info->gd);
1893 
1894 	info->is_ready = 1;
1895 }
1896 
1897 /**
1898  * Callback received when the backend's state changes.
1899  */
1900 static void blkback_changed(struct xenbus_device *dev,
1901 			    enum xenbus_state backend_state)
1902 {
1903 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1904 
1905 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1906 
1907 	switch (backend_state) {
1908 	case XenbusStateInitialising:
1909 	case XenbusStateInitWait:
1910 	case XenbusStateInitialised:
1911 	case XenbusStateReconfiguring:
1912 	case XenbusStateReconfigured:
1913 	case XenbusStateUnknown:
1914 		break;
1915 
1916 	case XenbusStateConnected:
1917 		blkfront_connect(info);
1918 		break;
1919 
1920 	case XenbusStateClosed:
1921 		if (dev->state == XenbusStateClosed)
1922 			break;
1923 		/* Missed the backend's Closing state -- fallthrough */
1924 	case XenbusStateClosing:
1925 		blkfront_closing(info);
1926 		break;
1927 	}
1928 }
1929 
1930 static int blkfront_remove(struct xenbus_device *xbdev)
1931 {
1932 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1933 	struct block_device *bdev = NULL;
1934 	struct gendisk *disk;
1935 
1936 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1937 
1938 	blkif_free(info, 0);
1939 
1940 	mutex_lock(&info->mutex);
1941 
1942 	disk = info->gd;
1943 	if (disk)
1944 		bdev = bdget_disk(disk, 0);
1945 
1946 	info->xbdev = NULL;
1947 	mutex_unlock(&info->mutex);
1948 
1949 	if (!bdev) {
1950 		kfree(info);
1951 		return 0;
1952 	}
1953 
1954 	/*
1955 	 * The xbdev was removed before we reached the Closed
1956 	 * state. See if it's safe to remove the disk. If the bdev
1957 	 * isn't closed yet, we let release take care of it.
1958 	 */
1959 
1960 	mutex_lock(&bdev->bd_mutex);
1961 	info = disk->private_data;
1962 
1963 	dev_warn(disk_to_dev(disk),
1964 		 "%s was hot-unplugged, %d stale handles\n",
1965 		 xbdev->nodename, bdev->bd_openers);
1966 
1967 	if (info && !bdev->bd_openers) {
1968 		xlvbd_release_gendisk(info);
1969 		disk->private_data = NULL;
1970 		kfree(info);
1971 	}
1972 
1973 	mutex_unlock(&bdev->bd_mutex);
1974 	bdput(bdev);
1975 
1976 	return 0;
1977 }
1978 
1979 static int blkfront_is_ready(struct xenbus_device *dev)
1980 {
1981 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1982 
1983 	return info->is_ready && info->xbdev;
1984 }
1985 
1986 static int blkif_open(struct block_device *bdev, fmode_t mode)
1987 {
1988 	struct gendisk *disk = bdev->bd_disk;
1989 	struct blkfront_info *info;
1990 	int err = 0;
1991 
1992 	mutex_lock(&blkfront_mutex);
1993 
1994 	info = disk->private_data;
1995 	if (!info) {
1996 		/* xbdev gone */
1997 		err = -ERESTARTSYS;
1998 		goto out;
1999 	}
2000 
2001 	mutex_lock(&info->mutex);
2002 
2003 	if (!info->gd)
2004 		/* xbdev is closed */
2005 		err = -ERESTARTSYS;
2006 
2007 	mutex_unlock(&info->mutex);
2008 
2009 out:
2010 	mutex_unlock(&blkfront_mutex);
2011 	return err;
2012 }
2013 
2014 static void blkif_release(struct gendisk *disk, fmode_t mode)
2015 {
2016 	struct blkfront_info *info = disk->private_data;
2017 	struct block_device *bdev;
2018 	struct xenbus_device *xbdev;
2019 
2020 	mutex_lock(&blkfront_mutex);
2021 
2022 	bdev = bdget_disk(disk, 0);
2023 
2024 	if (!bdev) {
2025 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2026 		goto out_mutex;
2027 	}
2028 	if (bdev->bd_openers)
2029 		goto out;
2030 
2031 	/*
2032 	 * Check if we have been instructed to close. We will have
2033 	 * deferred this request, because the bdev was still open.
2034 	 */
2035 
2036 	mutex_lock(&info->mutex);
2037 	xbdev = info->xbdev;
2038 
2039 	if (xbdev && xbdev->state == XenbusStateClosing) {
2040 		/* pending switch to state closed */
2041 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2042 		xlvbd_release_gendisk(info);
2043 		xenbus_frontend_closed(info->xbdev);
2044  	}
2045 
2046 	mutex_unlock(&info->mutex);
2047 
2048 	if (!xbdev) {
2049 		/* sudden device removal */
2050 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2051 		xlvbd_release_gendisk(info);
2052 		disk->private_data = NULL;
2053 		kfree(info);
2054 	}
2055 
2056 out:
2057 	bdput(bdev);
2058 out_mutex:
2059 	mutex_unlock(&blkfront_mutex);
2060 }
2061 
2062 static const struct block_device_operations xlvbd_block_fops =
2063 {
2064 	.owner = THIS_MODULE,
2065 	.open = blkif_open,
2066 	.release = blkif_release,
2067 	.getgeo = blkif_getgeo,
2068 	.ioctl = blkif_ioctl,
2069 };
2070 
2071 
2072 static const struct xenbus_device_id blkfront_ids[] = {
2073 	{ "vbd" },
2074 	{ "" }
2075 };
2076 
2077 static struct xenbus_driver blkfront_driver = {
2078 	.ids  = blkfront_ids,
2079 	.probe = blkfront_probe,
2080 	.remove = blkfront_remove,
2081 	.resume = blkfront_resume,
2082 	.otherend_changed = blkback_changed,
2083 	.is_ready = blkfront_is_ready,
2084 };
2085 
2086 static int __init xlblk_init(void)
2087 {
2088 	int ret;
2089 
2090 	if (!xen_domain())
2091 		return -ENODEV;
2092 
2093 	if (!xen_has_pv_disk_devices())
2094 		return -ENODEV;
2095 
2096 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2097 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2098 		       XENVBD_MAJOR, DEV_NAME);
2099 		return -ENODEV;
2100 	}
2101 
2102 	ret = xenbus_register_frontend(&blkfront_driver);
2103 	if (ret) {
2104 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2105 		return ret;
2106 	}
2107 
2108 	return 0;
2109 }
2110 module_init(xlblk_init);
2111 
2112 
2113 static void __exit xlblk_exit(void)
2114 {
2115 	xenbus_unregister_driver(&blkfront_driver);
2116 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2117 	kfree(minors);
2118 }
2119 module_exit(xlblk_exit);
2120 
2121 MODULE_DESCRIPTION("Xen virtual block device frontend");
2122 MODULE_LICENSE("GPL");
2123 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2124 MODULE_ALIAS("xen:vbd");
2125 MODULE_ALIAS("xenblk");
2126