xref: /linux/drivers/block/xen-blkfront.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
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/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 
50 #include <xen/xen.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56 
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
60 
61 #include <asm/xen/hypervisor.h>
62 
63 /*
64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
65  * When Linux is using a different page size than Xen, it may not be possible
66  * to put all the data in a single segment.
67  * This can happen when the backend doesn't support indirect descriptor and
68  * therefore the maximum amount of data that a request can carry is
69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70  *
71  * Note that we only support one extra request. So the Linux page size
72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73  * 88KB.
74  */
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76 
77 enum blkif_state {
78 	BLKIF_STATE_DISCONNECTED,
79 	BLKIF_STATE_CONNECTED,
80 	BLKIF_STATE_SUSPENDED,
81 };
82 
83 struct grant {
84 	grant_ref_t gref;
85 	struct page *page;
86 	struct list_head node;
87 };
88 
89 enum blk_req_status {
90 	REQ_WAITING,
91 	REQ_DONE,
92 	REQ_ERROR,
93 	REQ_EOPNOTSUPP,
94 };
95 
96 struct blk_shadow {
97 	struct blkif_request req;
98 	struct request *request;
99 	struct grant **grants_used;
100 	struct grant **indirect_grants;
101 	struct scatterlist *sg;
102 	unsigned int num_sg;
103 	enum blk_req_status status;
104 
105 	#define NO_ASSOCIATED_ID ~0UL
106 	/*
107 	 * Id of the sibling if we ever need 2 requests when handling a
108 	 * block I/O request
109 	 */
110 	unsigned long associated_id;
111 };
112 
113 struct split_bio {
114 	struct bio *bio;
115 	atomic_t pending;
116 };
117 
118 static DEFINE_MUTEX(blkfront_mutex);
119 static const struct block_device_operations xlvbd_block_fops;
120 
121 /*
122  * Maximum number of segments in indirect requests, the actual value used by
123  * the frontend driver is the minimum of this value and the value provided
124  * by the backend driver.
125  */
126 
127 static unsigned int xen_blkif_max_segments = 32;
128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
129 		   S_IRUGO);
130 MODULE_PARM_DESC(max_indirect_segments,
131 		 "Maximum amount of segments in indirect requests (default is 32)");
132 
133 static unsigned int xen_blkif_max_queues = 4;
134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
136 
137 /*
138  * Maximum order of pages to be used for the shared ring between front and
139  * backend, 4KB page granularity is used.
140  */
141 static unsigned int xen_blkif_max_ring_order;
142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
144 
145 #define BLK_RING_SIZE(info)	\
146 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
147 
148 #define BLK_MAX_RING_SIZE	\
149 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
150 
151 /*
152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
153  * characters are enough. Define to 20 to keep consistent with backend.
154  */
155 #define RINGREF_NAME_LEN (20)
156 /*
157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
158  */
159 #define QUEUE_NAME_LEN (17)
160 
161 /*
162  *  Per-ring info.
163  *  Every blkfront device can associate with one or more blkfront_ring_info,
164  *  depending on how many hardware queues/rings to be used.
165  */
166 struct blkfront_ring_info {
167 	/* Lock to protect data in every ring buffer. */
168 	spinlock_t ring_lock;
169 	struct blkif_front_ring ring;
170 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
171 	unsigned int evtchn, irq;
172 	struct work_struct work;
173 	struct gnttab_free_callback callback;
174 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
175 	struct list_head indirect_pages;
176 	struct list_head grants;
177 	unsigned int persistent_gnts_c;
178 	unsigned long shadow_free;
179 	struct blkfront_info *dev_info;
180 };
181 
182 /*
183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
184  * hang in private_data off the gendisk structure. We may end up
185  * putting all kinds of interesting stuff here :-)
186  */
187 struct blkfront_info
188 {
189 	struct mutex mutex;
190 	struct xenbus_device *xbdev;
191 	struct gendisk *gd;
192 	int vdevice;
193 	blkif_vdev_t handle;
194 	enum blkif_state connected;
195 	/* Number of pages per ring buffer. */
196 	unsigned int nr_ring_pages;
197 	struct request_queue *rq;
198 	unsigned int feature_flush;
199 	unsigned int feature_fua;
200 	unsigned int feature_discard:1;
201 	unsigned int feature_secdiscard:1;
202 	unsigned int discard_granularity;
203 	unsigned int discard_alignment;
204 	unsigned int feature_persistent:1;
205 	/* Number of 4KB segments handled */
206 	unsigned int max_indirect_segments;
207 	int is_ready;
208 	struct blk_mq_tag_set tag_set;
209 	struct blkfront_ring_info *rinfo;
210 	unsigned int nr_rings;
211 	/* Save uncomplete reqs and bios for migration. */
212 	struct list_head requests;
213 	struct bio_list bio_list;
214 };
215 
216 static unsigned int nr_minors;
217 static unsigned long *minors;
218 static DEFINE_SPINLOCK(minor_lock);
219 
220 #define GRANT_INVALID_REF	0
221 
222 #define PARTS_PER_DISK		16
223 #define PARTS_PER_EXT_DISK      256
224 
225 #define BLKIF_MAJOR(dev) ((dev)>>8)
226 #define BLKIF_MINOR(dev) ((dev) & 0xff)
227 
228 #define EXT_SHIFT 28
229 #define EXTENDED (1<<EXT_SHIFT)
230 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
231 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
232 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
233 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
234 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
235 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
236 
237 #define DEV_NAME	"xvd"	/* name in /dev */
238 
239 /*
240  * Grants are always the same size as a Xen page (i.e 4KB).
241  * A physical segment is always the same size as a Linux page.
242  * Number of grants per physical segment
243  */
244 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
245 
246 #define GRANTS_PER_INDIRECT_FRAME \
247 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
248 
249 #define PSEGS_PER_INDIRECT_FRAME	\
250 	(GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
251 
252 #define INDIRECT_GREFS(_grants)		\
253 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
254 
255 #define GREFS(_psegs)	((_psegs) * GRANTS_PER_PSEG)
256 
257 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
258 static void blkfront_gather_backend_features(struct blkfront_info *info);
259 
260 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
261 {
262 	unsigned long free = rinfo->shadow_free;
263 
264 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
265 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
266 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
267 	return free;
268 }
269 
270 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
271 			      unsigned long id)
272 {
273 	if (rinfo->shadow[id].req.u.rw.id != id)
274 		return -EINVAL;
275 	if (rinfo->shadow[id].request == NULL)
276 		return -EINVAL;
277 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
278 	rinfo->shadow[id].request = NULL;
279 	rinfo->shadow_free = id;
280 	return 0;
281 }
282 
283 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
284 {
285 	struct blkfront_info *info = rinfo->dev_info;
286 	struct page *granted_page;
287 	struct grant *gnt_list_entry, *n;
288 	int i = 0;
289 
290 	while (i < num) {
291 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
292 		if (!gnt_list_entry)
293 			goto out_of_memory;
294 
295 		if (info->feature_persistent) {
296 			granted_page = alloc_page(GFP_NOIO);
297 			if (!granted_page) {
298 				kfree(gnt_list_entry);
299 				goto out_of_memory;
300 			}
301 			gnt_list_entry->page = granted_page;
302 		}
303 
304 		gnt_list_entry->gref = GRANT_INVALID_REF;
305 		list_add(&gnt_list_entry->node, &rinfo->grants);
306 		i++;
307 	}
308 
309 	return 0;
310 
311 out_of_memory:
312 	list_for_each_entry_safe(gnt_list_entry, n,
313 	                         &rinfo->grants, node) {
314 		list_del(&gnt_list_entry->node);
315 		if (info->feature_persistent)
316 			__free_page(gnt_list_entry->page);
317 		kfree(gnt_list_entry);
318 		i--;
319 	}
320 	BUG_ON(i != 0);
321 	return -ENOMEM;
322 }
323 
324 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
325 {
326 	struct grant *gnt_list_entry;
327 
328 	BUG_ON(list_empty(&rinfo->grants));
329 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
330 					  node);
331 	list_del(&gnt_list_entry->node);
332 
333 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
334 		rinfo->persistent_gnts_c--;
335 
336 	return gnt_list_entry;
337 }
338 
339 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
340 					const struct blkfront_info *info)
341 {
342 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
343 						 info->xbdev->otherend_id,
344 						 gnt_list_entry->page,
345 						 0);
346 }
347 
348 static struct grant *get_grant(grant_ref_t *gref_head,
349 			       unsigned long gfn,
350 			       struct blkfront_ring_info *rinfo)
351 {
352 	struct grant *gnt_list_entry = get_free_grant(rinfo);
353 	struct blkfront_info *info = rinfo->dev_info;
354 
355 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
356 		return gnt_list_entry;
357 
358 	/* Assign a gref to this page */
359 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
360 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
361 	if (info->feature_persistent)
362 		grant_foreign_access(gnt_list_entry, info);
363 	else {
364 		/* Grant access to the GFN passed by the caller */
365 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
366 						info->xbdev->otherend_id,
367 						gfn, 0);
368 	}
369 
370 	return gnt_list_entry;
371 }
372 
373 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
374 					struct blkfront_ring_info *rinfo)
375 {
376 	struct grant *gnt_list_entry = get_free_grant(rinfo);
377 	struct blkfront_info *info = rinfo->dev_info;
378 
379 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
380 		return gnt_list_entry;
381 
382 	/* Assign a gref to this page */
383 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
384 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
385 	if (!info->feature_persistent) {
386 		struct page *indirect_page;
387 
388 		/* Fetch a pre-allocated page to use for indirect grefs */
389 		BUG_ON(list_empty(&rinfo->indirect_pages));
390 		indirect_page = list_first_entry(&rinfo->indirect_pages,
391 						 struct page, lru);
392 		list_del(&indirect_page->lru);
393 		gnt_list_entry->page = indirect_page;
394 	}
395 	grant_foreign_access(gnt_list_entry, info);
396 
397 	return gnt_list_entry;
398 }
399 
400 static const char *op_name(int op)
401 {
402 	static const char *const names[] = {
403 		[BLKIF_OP_READ] = "read",
404 		[BLKIF_OP_WRITE] = "write",
405 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
406 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
407 		[BLKIF_OP_DISCARD] = "discard" };
408 
409 	if (op < 0 || op >= ARRAY_SIZE(names))
410 		return "unknown";
411 
412 	if (!names[op])
413 		return "reserved";
414 
415 	return names[op];
416 }
417 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
418 {
419 	unsigned int end = minor + nr;
420 	int rc;
421 
422 	if (end > nr_minors) {
423 		unsigned long *bitmap, *old;
424 
425 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
426 				 GFP_KERNEL);
427 		if (bitmap == NULL)
428 			return -ENOMEM;
429 
430 		spin_lock(&minor_lock);
431 		if (end > nr_minors) {
432 			old = minors;
433 			memcpy(bitmap, minors,
434 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
435 			minors = bitmap;
436 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
437 		} else
438 			old = bitmap;
439 		spin_unlock(&minor_lock);
440 		kfree(old);
441 	}
442 
443 	spin_lock(&minor_lock);
444 	if (find_next_bit(minors, end, minor) >= end) {
445 		bitmap_set(minors, minor, nr);
446 		rc = 0;
447 	} else
448 		rc = -EBUSY;
449 	spin_unlock(&minor_lock);
450 
451 	return rc;
452 }
453 
454 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
455 {
456 	unsigned int end = minor + nr;
457 
458 	BUG_ON(end > nr_minors);
459 	spin_lock(&minor_lock);
460 	bitmap_clear(minors,  minor, nr);
461 	spin_unlock(&minor_lock);
462 }
463 
464 static void blkif_restart_queue_callback(void *arg)
465 {
466 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
467 	schedule_work(&rinfo->work);
468 }
469 
470 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
471 {
472 	/* We don't have real geometry info, but let's at least return
473 	   values consistent with the size of the device */
474 	sector_t nsect = get_capacity(bd->bd_disk);
475 	sector_t cylinders = nsect;
476 
477 	hg->heads = 0xff;
478 	hg->sectors = 0x3f;
479 	sector_div(cylinders, hg->heads * hg->sectors);
480 	hg->cylinders = cylinders;
481 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
482 		hg->cylinders = 0xffff;
483 	return 0;
484 }
485 
486 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
487 		       unsigned command, unsigned long argument)
488 {
489 	struct blkfront_info *info = bdev->bd_disk->private_data;
490 	int i;
491 
492 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
493 		command, (long)argument);
494 
495 	switch (command) {
496 	case CDROMMULTISESSION:
497 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
498 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
499 			if (put_user(0, (char __user *)(argument + i)))
500 				return -EFAULT;
501 		return 0;
502 
503 	case CDROM_GET_CAPABILITY: {
504 		struct gendisk *gd = info->gd;
505 		if (gd->flags & GENHD_FL_CD)
506 			return 0;
507 		return -EINVAL;
508 	}
509 
510 	default:
511 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
512 		  command);*/
513 		return -EINVAL; /* same return as native Linux */
514 	}
515 
516 	return 0;
517 }
518 
519 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
520 					    struct request *req,
521 					    struct blkif_request **ring_req)
522 {
523 	unsigned long id;
524 
525 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
526 	rinfo->ring.req_prod_pvt++;
527 
528 	id = get_id_from_freelist(rinfo);
529 	rinfo->shadow[id].request = req;
530 	rinfo->shadow[id].status = REQ_WAITING;
531 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
532 
533 	(*ring_req)->u.rw.id = id;
534 
535 	return id;
536 }
537 
538 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
539 {
540 	struct blkfront_info *info = rinfo->dev_info;
541 	struct blkif_request *ring_req;
542 	unsigned long id;
543 
544 	/* Fill out a communications ring structure. */
545 	id = blkif_ring_get_request(rinfo, req, &ring_req);
546 
547 	ring_req->operation = BLKIF_OP_DISCARD;
548 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
549 	ring_req->u.discard.id = id;
550 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
551 	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
552 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
553 	else
554 		ring_req->u.discard.flag = 0;
555 
556 	/* Keep a private copy so we can reissue requests when recovering. */
557 	rinfo->shadow[id].req = *ring_req;
558 
559 	return 0;
560 }
561 
562 struct setup_rw_req {
563 	unsigned int grant_idx;
564 	struct blkif_request_segment *segments;
565 	struct blkfront_ring_info *rinfo;
566 	struct blkif_request *ring_req;
567 	grant_ref_t gref_head;
568 	unsigned int id;
569 	/* Only used when persistent grant is used and it's a read request */
570 	bool need_copy;
571 	unsigned int bvec_off;
572 	char *bvec_data;
573 
574 	bool require_extra_req;
575 	struct blkif_request *extra_ring_req;
576 };
577 
578 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
579 				     unsigned int len, void *data)
580 {
581 	struct setup_rw_req *setup = data;
582 	int n, ref;
583 	struct grant *gnt_list_entry;
584 	unsigned int fsect, lsect;
585 	/* Convenient aliases */
586 	unsigned int grant_idx = setup->grant_idx;
587 	struct blkif_request *ring_req = setup->ring_req;
588 	struct blkfront_ring_info *rinfo = setup->rinfo;
589 	/*
590 	 * We always use the shadow of the first request to store the list
591 	 * of grant associated to the block I/O request. This made the
592 	 * completion more easy to handle even if the block I/O request is
593 	 * split.
594 	 */
595 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
596 
597 	if (unlikely(setup->require_extra_req &&
598 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
599 		/*
600 		 * We are using the second request, setup grant_idx
601 		 * to be the index of the segment array.
602 		 */
603 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
604 		ring_req = setup->extra_ring_req;
605 	}
606 
607 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
608 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
609 		if (setup->segments)
610 			kunmap_atomic(setup->segments);
611 
612 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
613 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
614 		shadow->indirect_grants[n] = gnt_list_entry;
615 		setup->segments = kmap_atomic(gnt_list_entry->page);
616 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
617 	}
618 
619 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
620 	ref = gnt_list_entry->gref;
621 	/*
622 	 * All the grants are stored in the shadow of the first
623 	 * request. Therefore we have to use the global index.
624 	 */
625 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
626 
627 	if (setup->need_copy) {
628 		void *shared_data;
629 
630 		shared_data = kmap_atomic(gnt_list_entry->page);
631 		/*
632 		 * this does not wipe data stored outside the
633 		 * range sg->offset..sg->offset+sg->length.
634 		 * Therefore, blkback *could* see data from
635 		 * previous requests. This is OK as long as
636 		 * persistent grants are shared with just one
637 		 * domain. It may need refactoring if this
638 		 * changes
639 		 */
640 		memcpy(shared_data + offset,
641 		       setup->bvec_data + setup->bvec_off,
642 		       len);
643 
644 		kunmap_atomic(shared_data);
645 		setup->bvec_off += len;
646 	}
647 
648 	fsect = offset >> 9;
649 	lsect = fsect + (len >> 9) - 1;
650 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
651 		ring_req->u.rw.seg[grant_idx] =
652 			(struct blkif_request_segment) {
653 				.gref       = ref,
654 				.first_sect = fsect,
655 				.last_sect  = lsect };
656 	} else {
657 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
658 			(struct blkif_request_segment) {
659 				.gref       = ref,
660 				.first_sect = fsect,
661 				.last_sect  = lsect };
662 	}
663 
664 	(setup->grant_idx)++;
665 }
666 
667 static void blkif_setup_extra_req(struct blkif_request *first,
668 				  struct blkif_request *second)
669 {
670 	uint16_t nr_segments = first->u.rw.nr_segments;
671 
672 	/*
673 	 * The second request is only present when the first request uses
674 	 * all its segments. It's always the continuity of the first one.
675 	 */
676 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
677 
678 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
679 	second->u.rw.sector_number = first->u.rw.sector_number +
680 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
681 
682 	second->u.rw.handle = first->u.rw.handle;
683 	second->operation = first->operation;
684 }
685 
686 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
687 {
688 	struct blkfront_info *info = rinfo->dev_info;
689 	struct blkif_request *ring_req, *extra_ring_req = NULL;
690 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
691 	bool require_extra_req = false;
692 	int i;
693 	struct setup_rw_req setup = {
694 		.grant_idx = 0,
695 		.segments = NULL,
696 		.rinfo = rinfo,
697 		.need_copy = rq_data_dir(req) && info->feature_persistent,
698 	};
699 
700 	/*
701 	 * Used to store if we are able to queue the request by just using
702 	 * existing persistent grants, or if we have to get new grants,
703 	 * as there are not sufficiently many free.
704 	 */
705 	struct scatterlist *sg;
706 	int num_sg, max_grefs, num_grant;
707 
708 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
709 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
710 		/*
711 		 * If we are using indirect segments we need to account
712 		 * for the indirect grefs used in the request.
713 		 */
714 		max_grefs += INDIRECT_GREFS(max_grefs);
715 
716 	/*
717 	 * We have to reserve 'max_grefs' grants because persistent
718 	 * grants are shared by all rings.
719 	 */
720 	if (max_grefs > 0)
721 		if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
722 			gnttab_request_free_callback(
723 				&rinfo->callback,
724 				blkif_restart_queue_callback,
725 				rinfo,
726 				max_grefs);
727 			return 1;
728 		}
729 
730 	/* Fill out a communications ring structure. */
731 	id = blkif_ring_get_request(rinfo, req, &ring_req);
732 
733 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
734 	num_grant = 0;
735 	/* Calculate the number of grant used */
736 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
737 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
738 
739 	require_extra_req = info->max_indirect_segments == 0 &&
740 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
741 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
742 
743 	rinfo->shadow[id].num_sg = num_sg;
744 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
745 	    likely(!require_extra_req)) {
746 		/*
747 		 * The indirect operation can only be a BLKIF_OP_READ or
748 		 * BLKIF_OP_WRITE
749 		 */
750 		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
751 		ring_req->operation = BLKIF_OP_INDIRECT;
752 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
753 			BLKIF_OP_WRITE : BLKIF_OP_READ;
754 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
755 		ring_req->u.indirect.handle = info->handle;
756 		ring_req->u.indirect.nr_segments = num_grant;
757 	} else {
758 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
759 		ring_req->u.rw.handle = info->handle;
760 		ring_req->operation = rq_data_dir(req) ?
761 			BLKIF_OP_WRITE : BLKIF_OP_READ;
762 		if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
763 			/*
764 			 * Ideally we can do an unordered flush-to-disk.
765 			 * In case the backend onlysupports barriers, use that.
766 			 * A barrier request a superset of FUA, so we can
767 			 * implement it the same way.  (It's also a FLUSH+FUA,
768 			 * since it is guaranteed ordered WRT previous writes.)
769 			 */
770 			if (info->feature_flush && info->feature_fua)
771 				ring_req->operation =
772 					BLKIF_OP_WRITE_BARRIER;
773 			else if (info->feature_flush)
774 				ring_req->operation =
775 					BLKIF_OP_FLUSH_DISKCACHE;
776 			else
777 				ring_req->operation = 0;
778 		}
779 		ring_req->u.rw.nr_segments = num_grant;
780 		if (unlikely(require_extra_req)) {
781 			extra_id = blkif_ring_get_request(rinfo, req,
782 							  &extra_ring_req);
783 			/*
784 			 * Only the first request contains the scatter-gather
785 			 * list.
786 			 */
787 			rinfo->shadow[extra_id].num_sg = 0;
788 
789 			blkif_setup_extra_req(ring_req, extra_ring_req);
790 
791 			/* Link the 2 requests together */
792 			rinfo->shadow[extra_id].associated_id = id;
793 			rinfo->shadow[id].associated_id = extra_id;
794 		}
795 	}
796 
797 	setup.ring_req = ring_req;
798 	setup.id = id;
799 
800 	setup.require_extra_req = require_extra_req;
801 	if (unlikely(require_extra_req))
802 		setup.extra_ring_req = extra_ring_req;
803 
804 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
805 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
806 
807 		if (setup.need_copy) {
808 			setup.bvec_off = sg->offset;
809 			setup.bvec_data = kmap_atomic(sg_page(sg));
810 		}
811 
812 		gnttab_foreach_grant_in_range(sg_page(sg),
813 					      sg->offset,
814 					      sg->length,
815 					      blkif_setup_rw_req_grant,
816 					      &setup);
817 
818 		if (setup.need_copy)
819 			kunmap_atomic(setup.bvec_data);
820 	}
821 	if (setup.segments)
822 		kunmap_atomic(setup.segments);
823 
824 	/* Keep a private copy so we can reissue requests when recovering. */
825 	rinfo->shadow[id].req = *ring_req;
826 	if (unlikely(require_extra_req))
827 		rinfo->shadow[extra_id].req = *extra_ring_req;
828 
829 	if (max_grefs > 0)
830 		gnttab_free_grant_references(setup.gref_head);
831 
832 	return 0;
833 }
834 
835 /*
836  * Generate a Xen blkfront IO request from a blk layer request.  Reads
837  * and writes are handled as expected.
838  *
839  * @req: a request struct
840  */
841 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
842 {
843 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
844 		return 1;
845 
846 	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
847 		     req_op(req) == REQ_OP_SECURE_ERASE))
848 		return blkif_queue_discard_req(req, rinfo);
849 	else
850 		return blkif_queue_rw_req(req, rinfo);
851 }
852 
853 static inline void flush_requests(struct blkfront_ring_info *rinfo)
854 {
855 	int notify;
856 
857 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
858 
859 	if (notify)
860 		notify_remote_via_irq(rinfo->irq);
861 }
862 
863 static inline bool blkif_request_flush_invalid(struct request *req,
864 					       struct blkfront_info *info)
865 {
866 	return ((req->cmd_type != REQ_TYPE_FS) ||
867 		((req_op(req) == REQ_OP_FLUSH) &&
868 		 !info->feature_flush) ||
869 		((req->cmd_flags & REQ_FUA) &&
870 		 !info->feature_fua));
871 }
872 
873 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
874 			  const struct blk_mq_queue_data *qd)
875 {
876 	unsigned long flags;
877 	int qid = hctx->queue_num;
878 	struct blkfront_info *info = hctx->queue->queuedata;
879 	struct blkfront_ring_info *rinfo = NULL;
880 
881 	BUG_ON(info->nr_rings <= qid);
882 	rinfo = &info->rinfo[qid];
883 	blk_mq_start_request(qd->rq);
884 	spin_lock_irqsave(&rinfo->ring_lock, flags);
885 	if (RING_FULL(&rinfo->ring))
886 		goto out_busy;
887 
888 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
889 		goto out_err;
890 
891 	if (blkif_queue_request(qd->rq, rinfo))
892 		goto out_busy;
893 
894 	flush_requests(rinfo);
895 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
896 	return BLK_MQ_RQ_QUEUE_OK;
897 
898 out_err:
899 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
900 	return BLK_MQ_RQ_QUEUE_ERROR;
901 
902 out_busy:
903 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
904 	blk_mq_stop_hw_queue(hctx);
905 	return BLK_MQ_RQ_QUEUE_BUSY;
906 }
907 
908 static struct blk_mq_ops blkfront_mq_ops = {
909 	.queue_rq = blkif_queue_rq,
910 	.map_queue = blk_mq_map_queue,
911 };
912 
913 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
914 				unsigned int physical_sector_size,
915 				unsigned int segments)
916 {
917 	struct request_queue *rq;
918 	struct blkfront_info *info = gd->private_data;
919 
920 	memset(&info->tag_set, 0, sizeof(info->tag_set));
921 	info->tag_set.ops = &blkfront_mq_ops;
922 	info->tag_set.nr_hw_queues = info->nr_rings;
923 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
924 		/*
925 		 * When indirect descriptior is not supported, the I/O request
926 		 * will be split between multiple request in the ring.
927 		 * To avoid problems when sending the request, divide by
928 		 * 2 the depth of the queue.
929 		 */
930 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
931 	} else
932 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
933 	info->tag_set.numa_node = NUMA_NO_NODE;
934 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
935 	info->tag_set.cmd_size = 0;
936 	info->tag_set.driver_data = info;
937 
938 	if (blk_mq_alloc_tag_set(&info->tag_set))
939 		return -EINVAL;
940 	rq = blk_mq_init_queue(&info->tag_set);
941 	if (IS_ERR(rq)) {
942 		blk_mq_free_tag_set(&info->tag_set);
943 		return PTR_ERR(rq);
944 	}
945 
946 	rq->queuedata = info;
947 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
948 
949 	if (info->feature_discard) {
950 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
951 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
952 		rq->limits.discard_granularity = info->discard_granularity;
953 		rq->limits.discard_alignment = info->discard_alignment;
954 		if (info->feature_secdiscard)
955 			queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
956 	}
957 
958 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
959 	blk_queue_logical_block_size(rq, sector_size);
960 	blk_queue_physical_block_size(rq, physical_sector_size);
961 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
962 
963 	/* Each segment in a request is up to an aligned page in size. */
964 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
965 	blk_queue_max_segment_size(rq, PAGE_SIZE);
966 
967 	/* Ensure a merged request will fit in a single I/O ring slot. */
968 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
969 
970 	/* Make sure buffer addresses are sector-aligned. */
971 	blk_queue_dma_alignment(rq, 511);
972 
973 	/* Make sure we don't use bounce buffers. */
974 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
975 
976 	gd->queue = rq;
977 
978 	return 0;
979 }
980 
981 static const char *flush_info(struct blkfront_info *info)
982 {
983 	if (info->feature_flush && info->feature_fua)
984 		return "barrier: enabled;";
985 	else if (info->feature_flush)
986 		return "flush diskcache: enabled;";
987 	else
988 		return "barrier or flush: disabled;";
989 }
990 
991 static void xlvbd_flush(struct blkfront_info *info)
992 {
993 	blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
994 			      info->feature_fua ? true : false);
995 	pr_info("blkfront: %s: %s %s %s %s %s\n",
996 		info->gd->disk_name, flush_info(info),
997 		"persistent grants:", info->feature_persistent ?
998 		"enabled;" : "disabled;", "indirect descriptors:",
999 		info->max_indirect_segments ? "enabled;" : "disabled;");
1000 }
1001 
1002 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1003 {
1004 	int major;
1005 	major = BLKIF_MAJOR(vdevice);
1006 	*minor = BLKIF_MINOR(vdevice);
1007 	switch (major) {
1008 		case XEN_IDE0_MAJOR:
1009 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1010 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1011 				EMULATED_HD_DISK_MINOR_OFFSET;
1012 			break;
1013 		case XEN_IDE1_MAJOR:
1014 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1015 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1016 				EMULATED_HD_DISK_MINOR_OFFSET;
1017 			break;
1018 		case XEN_SCSI_DISK0_MAJOR:
1019 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1020 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1021 			break;
1022 		case XEN_SCSI_DISK1_MAJOR:
1023 		case XEN_SCSI_DISK2_MAJOR:
1024 		case XEN_SCSI_DISK3_MAJOR:
1025 		case XEN_SCSI_DISK4_MAJOR:
1026 		case XEN_SCSI_DISK5_MAJOR:
1027 		case XEN_SCSI_DISK6_MAJOR:
1028 		case XEN_SCSI_DISK7_MAJOR:
1029 			*offset = (*minor / PARTS_PER_DISK) +
1030 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1031 				EMULATED_SD_DISK_NAME_OFFSET;
1032 			*minor = *minor +
1033 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1034 				EMULATED_SD_DISK_MINOR_OFFSET;
1035 			break;
1036 		case XEN_SCSI_DISK8_MAJOR:
1037 		case XEN_SCSI_DISK9_MAJOR:
1038 		case XEN_SCSI_DISK10_MAJOR:
1039 		case XEN_SCSI_DISK11_MAJOR:
1040 		case XEN_SCSI_DISK12_MAJOR:
1041 		case XEN_SCSI_DISK13_MAJOR:
1042 		case XEN_SCSI_DISK14_MAJOR:
1043 		case XEN_SCSI_DISK15_MAJOR:
1044 			*offset = (*minor / PARTS_PER_DISK) +
1045 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1046 				EMULATED_SD_DISK_NAME_OFFSET;
1047 			*minor = *minor +
1048 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1049 				EMULATED_SD_DISK_MINOR_OFFSET;
1050 			break;
1051 		case XENVBD_MAJOR:
1052 			*offset = *minor / PARTS_PER_DISK;
1053 			break;
1054 		default:
1055 			printk(KERN_WARNING "blkfront: your disk configuration is "
1056 					"incorrect, please use an xvd device instead\n");
1057 			return -ENODEV;
1058 	}
1059 	return 0;
1060 }
1061 
1062 static char *encode_disk_name(char *ptr, unsigned int n)
1063 {
1064 	if (n >= 26)
1065 		ptr = encode_disk_name(ptr, n / 26 - 1);
1066 	*ptr = 'a' + n % 26;
1067 	return ptr + 1;
1068 }
1069 
1070 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1071 			       struct blkfront_info *info,
1072 			       u16 vdisk_info, u16 sector_size,
1073 			       unsigned int physical_sector_size)
1074 {
1075 	struct gendisk *gd;
1076 	int nr_minors = 1;
1077 	int err;
1078 	unsigned int offset;
1079 	int minor;
1080 	int nr_parts;
1081 	char *ptr;
1082 
1083 	BUG_ON(info->gd != NULL);
1084 	BUG_ON(info->rq != NULL);
1085 
1086 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1087 		/* this is above the extended range; something is wrong */
1088 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1089 		return -ENODEV;
1090 	}
1091 
1092 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1093 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1094 		if (err)
1095 			return err;
1096  		nr_parts = PARTS_PER_DISK;
1097 	} else {
1098 		minor = BLKIF_MINOR_EXT(info->vdevice);
1099 		nr_parts = PARTS_PER_EXT_DISK;
1100 		offset = minor / nr_parts;
1101 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1102 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1103 					"emulated IDE disks,\n\t choose an xvd device name"
1104 					"from xvde on\n", info->vdevice);
1105 	}
1106 	if (minor >> MINORBITS) {
1107 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1108 			info->vdevice, minor);
1109 		return -ENODEV;
1110 	}
1111 
1112 	if ((minor % nr_parts) == 0)
1113 		nr_minors = nr_parts;
1114 
1115 	err = xlbd_reserve_minors(minor, nr_minors);
1116 	if (err)
1117 		goto out;
1118 	err = -ENODEV;
1119 
1120 	gd = alloc_disk(nr_minors);
1121 	if (gd == NULL)
1122 		goto release;
1123 
1124 	strcpy(gd->disk_name, DEV_NAME);
1125 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1126 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1127 	if (nr_minors > 1)
1128 		*ptr = 0;
1129 	else
1130 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1131 			 "%d", minor & (nr_parts - 1));
1132 
1133 	gd->major = XENVBD_MAJOR;
1134 	gd->first_minor = minor;
1135 	gd->fops = &xlvbd_block_fops;
1136 	gd->private_data = info;
1137 	set_capacity(gd, capacity);
1138 
1139 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1140 				 info->max_indirect_segments ? :
1141 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1142 		del_gendisk(gd);
1143 		goto release;
1144 	}
1145 
1146 	info->rq = gd->queue;
1147 	info->gd = gd;
1148 
1149 	xlvbd_flush(info);
1150 
1151 	if (vdisk_info & VDISK_READONLY)
1152 		set_disk_ro(gd, 1);
1153 
1154 	if (vdisk_info & VDISK_REMOVABLE)
1155 		gd->flags |= GENHD_FL_REMOVABLE;
1156 
1157 	if (vdisk_info & VDISK_CDROM)
1158 		gd->flags |= GENHD_FL_CD;
1159 
1160 	return 0;
1161 
1162  release:
1163 	xlbd_release_minors(minor, nr_minors);
1164  out:
1165 	return err;
1166 }
1167 
1168 static void xlvbd_release_gendisk(struct blkfront_info *info)
1169 {
1170 	unsigned int minor, nr_minors, i;
1171 
1172 	if (info->rq == NULL)
1173 		return;
1174 
1175 	/* No more blkif_request(). */
1176 	blk_mq_stop_hw_queues(info->rq);
1177 
1178 	for (i = 0; i < info->nr_rings; i++) {
1179 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1180 
1181 		/* No more gnttab callback work. */
1182 		gnttab_cancel_free_callback(&rinfo->callback);
1183 
1184 		/* Flush gnttab callback work. Must be done with no locks held. */
1185 		flush_work(&rinfo->work);
1186 	}
1187 
1188 	del_gendisk(info->gd);
1189 
1190 	minor = info->gd->first_minor;
1191 	nr_minors = info->gd->minors;
1192 	xlbd_release_minors(minor, nr_minors);
1193 
1194 	blk_cleanup_queue(info->rq);
1195 	blk_mq_free_tag_set(&info->tag_set);
1196 	info->rq = NULL;
1197 
1198 	put_disk(info->gd);
1199 	info->gd = NULL;
1200 }
1201 
1202 /* Already hold rinfo->ring_lock. */
1203 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1204 {
1205 	if (!RING_FULL(&rinfo->ring))
1206 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1207 }
1208 
1209 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1210 {
1211 	unsigned long flags;
1212 
1213 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1214 	kick_pending_request_queues_locked(rinfo);
1215 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1216 }
1217 
1218 static void blkif_restart_queue(struct work_struct *work)
1219 {
1220 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1221 
1222 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1223 		kick_pending_request_queues(rinfo);
1224 }
1225 
1226 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1227 {
1228 	struct grant *persistent_gnt, *n;
1229 	struct blkfront_info *info = rinfo->dev_info;
1230 	int i, j, segs;
1231 
1232 	/*
1233 	 * Remove indirect pages, this only happens when using indirect
1234 	 * descriptors but not persistent grants
1235 	 */
1236 	if (!list_empty(&rinfo->indirect_pages)) {
1237 		struct page *indirect_page, *n;
1238 
1239 		BUG_ON(info->feature_persistent);
1240 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1241 			list_del(&indirect_page->lru);
1242 			__free_page(indirect_page);
1243 		}
1244 	}
1245 
1246 	/* Remove all persistent grants. */
1247 	if (!list_empty(&rinfo->grants)) {
1248 		list_for_each_entry_safe(persistent_gnt, n,
1249 					 &rinfo->grants, node) {
1250 			list_del(&persistent_gnt->node);
1251 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1252 				gnttab_end_foreign_access(persistent_gnt->gref,
1253 							  0, 0UL);
1254 				rinfo->persistent_gnts_c--;
1255 			}
1256 			if (info->feature_persistent)
1257 				__free_page(persistent_gnt->page);
1258 			kfree(persistent_gnt);
1259 		}
1260 	}
1261 	BUG_ON(rinfo->persistent_gnts_c != 0);
1262 
1263 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1264 		/*
1265 		 * Clear persistent grants present in requests already
1266 		 * on the shared ring
1267 		 */
1268 		if (!rinfo->shadow[i].request)
1269 			goto free_shadow;
1270 
1271 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1272 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1273 		       rinfo->shadow[i].req.u.rw.nr_segments;
1274 		for (j = 0; j < segs; j++) {
1275 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1276 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1277 			if (info->feature_persistent)
1278 				__free_page(persistent_gnt->page);
1279 			kfree(persistent_gnt);
1280 		}
1281 
1282 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1283 			/*
1284 			 * If this is not an indirect operation don't try to
1285 			 * free indirect segments
1286 			 */
1287 			goto free_shadow;
1288 
1289 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1290 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1291 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1292 			__free_page(persistent_gnt->page);
1293 			kfree(persistent_gnt);
1294 		}
1295 
1296 free_shadow:
1297 		kfree(rinfo->shadow[i].grants_used);
1298 		rinfo->shadow[i].grants_used = NULL;
1299 		kfree(rinfo->shadow[i].indirect_grants);
1300 		rinfo->shadow[i].indirect_grants = NULL;
1301 		kfree(rinfo->shadow[i].sg);
1302 		rinfo->shadow[i].sg = NULL;
1303 	}
1304 
1305 	/* No more gnttab callback work. */
1306 	gnttab_cancel_free_callback(&rinfo->callback);
1307 
1308 	/* Flush gnttab callback work. Must be done with no locks held. */
1309 	flush_work(&rinfo->work);
1310 
1311 	/* Free resources associated with old device channel. */
1312 	for (i = 0; i < info->nr_ring_pages; i++) {
1313 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1314 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1315 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1316 		}
1317 	}
1318 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1319 	rinfo->ring.sring = NULL;
1320 
1321 	if (rinfo->irq)
1322 		unbind_from_irqhandler(rinfo->irq, rinfo);
1323 	rinfo->evtchn = rinfo->irq = 0;
1324 }
1325 
1326 static void blkif_free(struct blkfront_info *info, int suspend)
1327 {
1328 	unsigned int i;
1329 
1330 	/* Prevent new requests being issued until we fix things up. */
1331 	info->connected = suspend ?
1332 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1333 	/* No more blkif_request(). */
1334 	if (info->rq)
1335 		blk_mq_stop_hw_queues(info->rq);
1336 
1337 	for (i = 0; i < info->nr_rings; i++)
1338 		blkif_free_ring(&info->rinfo[i]);
1339 
1340 	kfree(info->rinfo);
1341 	info->rinfo = NULL;
1342 	info->nr_rings = 0;
1343 }
1344 
1345 struct copy_from_grant {
1346 	const struct blk_shadow *s;
1347 	unsigned int grant_idx;
1348 	unsigned int bvec_offset;
1349 	char *bvec_data;
1350 };
1351 
1352 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1353 				  unsigned int len, void *data)
1354 {
1355 	struct copy_from_grant *info = data;
1356 	char *shared_data;
1357 	/* Convenient aliases */
1358 	const struct blk_shadow *s = info->s;
1359 
1360 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1361 
1362 	memcpy(info->bvec_data + info->bvec_offset,
1363 	       shared_data + offset, len);
1364 
1365 	info->bvec_offset += len;
1366 	info->grant_idx++;
1367 
1368 	kunmap_atomic(shared_data);
1369 }
1370 
1371 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1372 {
1373 	switch (rsp)
1374 	{
1375 	case BLKIF_RSP_OKAY:
1376 		return REQ_DONE;
1377 	case BLKIF_RSP_EOPNOTSUPP:
1378 		return REQ_EOPNOTSUPP;
1379 	case BLKIF_RSP_ERROR:
1380 		/* Fallthrough. */
1381 	default:
1382 		return REQ_ERROR;
1383 	}
1384 }
1385 
1386 /*
1387  * Get the final status of the block request based on two ring response
1388  */
1389 static int blkif_get_final_status(enum blk_req_status s1,
1390 				  enum blk_req_status s2)
1391 {
1392 	BUG_ON(s1 == REQ_WAITING);
1393 	BUG_ON(s2 == REQ_WAITING);
1394 
1395 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1396 		return BLKIF_RSP_ERROR;
1397 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1398 		return BLKIF_RSP_EOPNOTSUPP;
1399 	return BLKIF_RSP_OKAY;
1400 }
1401 
1402 static bool blkif_completion(unsigned long *id,
1403 			     struct blkfront_ring_info *rinfo,
1404 			     struct blkif_response *bret)
1405 {
1406 	int i = 0;
1407 	struct scatterlist *sg;
1408 	int num_sg, num_grant;
1409 	struct blkfront_info *info = rinfo->dev_info;
1410 	struct blk_shadow *s = &rinfo->shadow[*id];
1411 	struct copy_from_grant data = {
1412 		.grant_idx = 0,
1413 	};
1414 
1415 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1416 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1417 
1418 	/* The I/O request may be split in two. */
1419 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1420 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1421 
1422 		/* Keep the status of the current response in shadow. */
1423 		s->status = blkif_rsp_to_req_status(bret->status);
1424 
1425 		/* Wait the second response if not yet here. */
1426 		if (s2->status == REQ_WAITING)
1427 			return 0;
1428 
1429 		bret->status = blkif_get_final_status(s->status,
1430 						      s2->status);
1431 
1432 		/*
1433 		 * All the grants is stored in the first shadow in order
1434 		 * to make the completion code simpler.
1435 		 */
1436 		num_grant += s2->req.u.rw.nr_segments;
1437 
1438 		/*
1439 		 * The two responses may not come in order. Only the
1440 		 * first request will store the scatter-gather list.
1441 		 */
1442 		if (s2->num_sg != 0) {
1443 			/* Update "id" with the ID of the first response. */
1444 			*id = s->associated_id;
1445 			s = s2;
1446 		}
1447 
1448 		/*
1449 		 * We don't need anymore the second request, so recycling
1450 		 * it now.
1451 		 */
1452 		if (add_id_to_freelist(rinfo, s->associated_id))
1453 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1454 			     info->gd->disk_name, s->associated_id);
1455 	}
1456 
1457 	data.s = s;
1458 	num_sg = s->num_sg;
1459 
1460 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1461 		for_each_sg(s->sg, sg, num_sg, i) {
1462 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1463 
1464 			data.bvec_offset = sg->offset;
1465 			data.bvec_data = kmap_atomic(sg_page(sg));
1466 
1467 			gnttab_foreach_grant_in_range(sg_page(sg),
1468 						      sg->offset,
1469 						      sg->length,
1470 						      blkif_copy_from_grant,
1471 						      &data);
1472 
1473 			kunmap_atomic(data.bvec_data);
1474 		}
1475 	}
1476 	/* Add the persistent grant into the list of free grants */
1477 	for (i = 0; i < num_grant; i++) {
1478 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1479 			/*
1480 			 * If the grant is still mapped by the backend (the
1481 			 * backend has chosen to make this grant persistent)
1482 			 * we add it at the head of the list, so it will be
1483 			 * reused first.
1484 			 */
1485 			if (!info->feature_persistent)
1486 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1487 						     s->grants_used[i]->gref);
1488 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1489 			rinfo->persistent_gnts_c++;
1490 		} else {
1491 			/*
1492 			 * If the grant is not mapped by the backend we end the
1493 			 * foreign access and add it to the tail of the list,
1494 			 * so it will not be picked again unless we run out of
1495 			 * persistent grants.
1496 			 */
1497 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1498 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1499 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1500 		}
1501 	}
1502 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1503 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1504 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1505 				if (!info->feature_persistent)
1506 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1507 							     s->indirect_grants[i]->gref);
1508 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1509 				rinfo->persistent_gnts_c++;
1510 			} else {
1511 				struct page *indirect_page;
1512 
1513 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1514 				/*
1515 				 * Add the used indirect page back to the list of
1516 				 * available pages for indirect grefs.
1517 				 */
1518 				if (!info->feature_persistent) {
1519 					indirect_page = s->indirect_grants[i]->page;
1520 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1521 				}
1522 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1523 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1524 			}
1525 		}
1526 	}
1527 
1528 	return 1;
1529 }
1530 
1531 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1532 {
1533 	struct request *req;
1534 	struct blkif_response *bret;
1535 	RING_IDX i, rp;
1536 	unsigned long flags;
1537 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1538 	struct blkfront_info *info = rinfo->dev_info;
1539 	int error;
1540 
1541 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1542 		return IRQ_HANDLED;
1543 
1544 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1545  again:
1546 	rp = rinfo->ring.sring->rsp_prod;
1547 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1548 
1549 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1550 		unsigned long id;
1551 
1552 		bret = RING_GET_RESPONSE(&rinfo->ring, i);
1553 		id   = bret->id;
1554 		/*
1555 		 * The backend has messed up and given us an id that we would
1556 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1557 		 * look in get_id_from_freelist.
1558 		 */
1559 		if (id >= BLK_RING_SIZE(info)) {
1560 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1561 			     info->gd->disk_name, op_name(bret->operation), id);
1562 			/* We can't safely get the 'struct request' as
1563 			 * the id is busted. */
1564 			continue;
1565 		}
1566 		req  = rinfo->shadow[id].request;
1567 
1568 		if (bret->operation != BLKIF_OP_DISCARD) {
1569 			/*
1570 			 * We may need to wait for an extra response if the
1571 			 * I/O request is split in 2
1572 			 */
1573 			if (!blkif_completion(&id, rinfo, bret))
1574 				continue;
1575 		}
1576 
1577 		if (add_id_to_freelist(rinfo, id)) {
1578 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1579 			     info->gd->disk_name, op_name(bret->operation), id);
1580 			continue;
1581 		}
1582 
1583 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1584 		switch (bret->operation) {
1585 		case BLKIF_OP_DISCARD:
1586 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1587 				struct request_queue *rq = info->rq;
1588 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1589 					   info->gd->disk_name, op_name(bret->operation));
1590 				error = -EOPNOTSUPP;
1591 				info->feature_discard = 0;
1592 				info->feature_secdiscard = 0;
1593 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1594 				queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1595 			}
1596 			blk_mq_complete_request(req, error);
1597 			break;
1598 		case BLKIF_OP_FLUSH_DISKCACHE:
1599 		case BLKIF_OP_WRITE_BARRIER:
1600 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1601 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1602 				       info->gd->disk_name, op_name(bret->operation));
1603 				error = -EOPNOTSUPP;
1604 			}
1605 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1606 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1607 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1608 				       info->gd->disk_name, op_name(bret->operation));
1609 				error = -EOPNOTSUPP;
1610 			}
1611 			if (unlikely(error)) {
1612 				if (error == -EOPNOTSUPP)
1613 					error = 0;
1614 				info->feature_fua = 0;
1615 				info->feature_flush = 0;
1616 				xlvbd_flush(info);
1617 			}
1618 			/* fall through */
1619 		case BLKIF_OP_READ:
1620 		case BLKIF_OP_WRITE:
1621 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1622 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1623 					"request: %x\n", bret->status);
1624 
1625 			blk_mq_complete_request(req, error);
1626 			break;
1627 		default:
1628 			BUG();
1629 		}
1630 	}
1631 
1632 	rinfo->ring.rsp_cons = i;
1633 
1634 	if (i != rinfo->ring.req_prod_pvt) {
1635 		int more_to_do;
1636 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1637 		if (more_to_do)
1638 			goto again;
1639 	} else
1640 		rinfo->ring.sring->rsp_event = i + 1;
1641 
1642 	kick_pending_request_queues_locked(rinfo);
1643 
1644 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1645 
1646 	return IRQ_HANDLED;
1647 }
1648 
1649 
1650 static int setup_blkring(struct xenbus_device *dev,
1651 			 struct blkfront_ring_info *rinfo)
1652 {
1653 	struct blkif_sring *sring;
1654 	int err, i;
1655 	struct blkfront_info *info = rinfo->dev_info;
1656 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1657 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1658 
1659 	for (i = 0; i < info->nr_ring_pages; i++)
1660 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1661 
1662 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1663 						       get_order(ring_size));
1664 	if (!sring) {
1665 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1666 		return -ENOMEM;
1667 	}
1668 	SHARED_RING_INIT(sring);
1669 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1670 
1671 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1672 	if (err < 0) {
1673 		free_pages((unsigned long)sring, get_order(ring_size));
1674 		rinfo->ring.sring = NULL;
1675 		goto fail;
1676 	}
1677 	for (i = 0; i < info->nr_ring_pages; i++)
1678 		rinfo->ring_ref[i] = gref[i];
1679 
1680 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1681 	if (err)
1682 		goto fail;
1683 
1684 	err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1685 					"blkif", rinfo);
1686 	if (err <= 0) {
1687 		xenbus_dev_fatal(dev, err,
1688 				 "bind_evtchn_to_irqhandler failed");
1689 		goto fail;
1690 	}
1691 	rinfo->irq = err;
1692 
1693 	return 0;
1694 fail:
1695 	blkif_free(info, 0);
1696 	return err;
1697 }
1698 
1699 /*
1700  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1701  * ring buffer may have multi pages depending on ->nr_ring_pages.
1702  */
1703 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1704 				struct blkfront_ring_info *rinfo, const char *dir)
1705 {
1706 	int err;
1707 	unsigned int i;
1708 	const char *message = NULL;
1709 	struct blkfront_info *info = rinfo->dev_info;
1710 
1711 	if (info->nr_ring_pages == 1) {
1712 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1713 		if (err) {
1714 			message = "writing ring-ref";
1715 			goto abort_transaction;
1716 		}
1717 	} else {
1718 		for (i = 0; i < info->nr_ring_pages; i++) {
1719 			char ring_ref_name[RINGREF_NAME_LEN];
1720 
1721 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1722 			err = xenbus_printf(xbt, dir, ring_ref_name,
1723 					    "%u", rinfo->ring_ref[i]);
1724 			if (err) {
1725 				message = "writing ring-ref";
1726 				goto abort_transaction;
1727 			}
1728 		}
1729 	}
1730 
1731 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1732 	if (err) {
1733 		message = "writing event-channel";
1734 		goto abort_transaction;
1735 	}
1736 
1737 	return 0;
1738 
1739 abort_transaction:
1740 	xenbus_transaction_end(xbt, 1);
1741 	if (message)
1742 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1743 
1744 	return err;
1745 }
1746 
1747 /* Common code used when first setting up, and when resuming. */
1748 static int talk_to_blkback(struct xenbus_device *dev,
1749 			   struct blkfront_info *info)
1750 {
1751 	const char *message = NULL;
1752 	struct xenbus_transaction xbt;
1753 	int err;
1754 	unsigned int i, max_page_order = 0;
1755 	unsigned int ring_page_order = 0;
1756 
1757 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1758 			   "max-ring-page-order", "%u", &max_page_order);
1759 	if (err != 1)
1760 		info->nr_ring_pages = 1;
1761 	else {
1762 		ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1763 		info->nr_ring_pages = 1 << ring_page_order;
1764 	}
1765 
1766 	for (i = 0; i < info->nr_rings; i++) {
1767 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1768 
1769 		/* Create shared ring, alloc event channel. */
1770 		err = setup_blkring(dev, rinfo);
1771 		if (err)
1772 			goto destroy_blkring;
1773 	}
1774 
1775 again:
1776 	err = xenbus_transaction_start(&xbt);
1777 	if (err) {
1778 		xenbus_dev_fatal(dev, err, "starting transaction");
1779 		goto destroy_blkring;
1780 	}
1781 
1782 	if (info->nr_ring_pages > 1) {
1783 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1784 				    ring_page_order);
1785 		if (err) {
1786 			message = "writing ring-page-order";
1787 			goto abort_transaction;
1788 		}
1789 	}
1790 
1791 	/* We already got the number of queues/rings in _probe */
1792 	if (info->nr_rings == 1) {
1793 		err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1794 		if (err)
1795 			goto destroy_blkring;
1796 	} else {
1797 		char *path;
1798 		size_t pathsize;
1799 
1800 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1801 				    info->nr_rings);
1802 		if (err) {
1803 			message = "writing multi-queue-num-queues";
1804 			goto abort_transaction;
1805 		}
1806 
1807 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1808 		path = kmalloc(pathsize, GFP_KERNEL);
1809 		if (!path) {
1810 			err = -ENOMEM;
1811 			message = "ENOMEM while writing ring references";
1812 			goto abort_transaction;
1813 		}
1814 
1815 		for (i = 0; i < info->nr_rings; i++) {
1816 			memset(path, 0, pathsize);
1817 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1818 			err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1819 			if (err) {
1820 				kfree(path);
1821 				goto destroy_blkring;
1822 			}
1823 		}
1824 		kfree(path);
1825 	}
1826 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1827 			    XEN_IO_PROTO_ABI_NATIVE);
1828 	if (err) {
1829 		message = "writing protocol";
1830 		goto abort_transaction;
1831 	}
1832 	err = xenbus_printf(xbt, dev->nodename,
1833 			    "feature-persistent", "%u", 1);
1834 	if (err)
1835 		dev_warn(&dev->dev,
1836 			 "writing persistent grants feature to xenbus");
1837 
1838 	err = xenbus_transaction_end(xbt, 0);
1839 	if (err) {
1840 		if (err == -EAGAIN)
1841 			goto again;
1842 		xenbus_dev_fatal(dev, err, "completing transaction");
1843 		goto destroy_blkring;
1844 	}
1845 
1846 	for (i = 0; i < info->nr_rings; i++) {
1847 		unsigned int j;
1848 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1849 
1850 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1851 			rinfo->shadow[j].req.u.rw.id = j + 1;
1852 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1853 	}
1854 	xenbus_switch_state(dev, XenbusStateInitialised);
1855 
1856 	return 0;
1857 
1858  abort_transaction:
1859 	xenbus_transaction_end(xbt, 1);
1860 	if (message)
1861 		xenbus_dev_fatal(dev, err, "%s", message);
1862  destroy_blkring:
1863 	blkif_free(info, 0);
1864 
1865 	kfree(info);
1866 	dev_set_drvdata(&dev->dev, NULL);
1867 
1868 	return err;
1869 }
1870 
1871 static int negotiate_mq(struct blkfront_info *info)
1872 {
1873 	unsigned int backend_max_queues = 0;
1874 	int err;
1875 	unsigned int i;
1876 
1877 	BUG_ON(info->nr_rings);
1878 
1879 	/* Check if backend supports multiple queues. */
1880 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1881 			   "multi-queue-max-queues", "%u", &backend_max_queues);
1882 	if (err < 0)
1883 		backend_max_queues = 1;
1884 
1885 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1886 	/* We need at least one ring. */
1887 	if (!info->nr_rings)
1888 		info->nr_rings = 1;
1889 
1890 	info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1891 	if (!info->rinfo) {
1892 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1893 		return -ENOMEM;
1894 	}
1895 
1896 	for (i = 0; i < info->nr_rings; i++) {
1897 		struct blkfront_ring_info *rinfo;
1898 
1899 		rinfo = &info->rinfo[i];
1900 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1901 		INIT_LIST_HEAD(&rinfo->grants);
1902 		rinfo->dev_info = info;
1903 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1904 		spin_lock_init(&rinfo->ring_lock);
1905 	}
1906 	return 0;
1907 }
1908 /**
1909  * Entry point to this code when a new device is created.  Allocate the basic
1910  * structures and the ring buffer for communication with the backend, and
1911  * inform the backend of the appropriate details for those.  Switch to
1912  * Initialised state.
1913  */
1914 static int blkfront_probe(struct xenbus_device *dev,
1915 			  const struct xenbus_device_id *id)
1916 {
1917 	int err, vdevice;
1918 	struct blkfront_info *info;
1919 
1920 	/* FIXME: Use dynamic device id if this is not set. */
1921 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1922 			   "virtual-device", "%i", &vdevice);
1923 	if (err != 1) {
1924 		/* go looking in the extended area instead */
1925 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1926 				   "%i", &vdevice);
1927 		if (err != 1) {
1928 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1929 			return err;
1930 		}
1931 	}
1932 
1933 	if (xen_hvm_domain()) {
1934 		char *type;
1935 		int len;
1936 		/* no unplug has been done: do not hook devices != xen vbds */
1937 		if (xen_has_pv_and_legacy_disk_devices()) {
1938 			int major;
1939 
1940 			if (!VDEV_IS_EXTENDED(vdevice))
1941 				major = BLKIF_MAJOR(vdevice);
1942 			else
1943 				major = XENVBD_MAJOR;
1944 
1945 			if (major != XENVBD_MAJOR) {
1946 				printk(KERN_INFO
1947 						"%s: HVM does not support vbd %d as xen block device\n",
1948 						__func__, vdevice);
1949 				return -ENODEV;
1950 			}
1951 		}
1952 		/* do not create a PV cdrom device if we are an HVM guest */
1953 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1954 		if (IS_ERR(type))
1955 			return -ENODEV;
1956 		if (strncmp(type, "cdrom", 5) == 0) {
1957 			kfree(type);
1958 			return -ENODEV;
1959 		}
1960 		kfree(type);
1961 	}
1962 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1963 	if (!info) {
1964 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1965 		return -ENOMEM;
1966 	}
1967 
1968 	info->xbdev = dev;
1969 	err = negotiate_mq(info);
1970 	if (err) {
1971 		kfree(info);
1972 		return err;
1973 	}
1974 
1975 	mutex_init(&info->mutex);
1976 	info->vdevice = vdevice;
1977 	info->connected = BLKIF_STATE_DISCONNECTED;
1978 
1979 	/* Front end dir is a number, which is used as the id. */
1980 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1981 	dev_set_drvdata(&dev->dev, info);
1982 
1983 	return 0;
1984 }
1985 
1986 static void split_bio_end(struct bio *bio)
1987 {
1988 	struct split_bio *split_bio = bio->bi_private;
1989 
1990 	if (atomic_dec_and_test(&split_bio->pending)) {
1991 		split_bio->bio->bi_phys_segments = 0;
1992 		split_bio->bio->bi_error = bio->bi_error;
1993 		bio_endio(split_bio->bio);
1994 		kfree(split_bio);
1995 	}
1996 	bio_put(bio);
1997 }
1998 
1999 static int blkif_recover(struct blkfront_info *info)
2000 {
2001 	unsigned int i, r_index;
2002 	struct request *req, *n;
2003 	int rc;
2004 	struct bio *bio, *cloned_bio;
2005 	unsigned int segs, offset;
2006 	int pending, size;
2007 	struct split_bio *split_bio;
2008 
2009 	blkfront_gather_backend_features(info);
2010 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2011 	blk_queue_max_segments(info->rq, segs);
2012 
2013 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2014 		struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2015 
2016 		rc = blkfront_setup_indirect(rinfo);
2017 		if (rc)
2018 			return rc;
2019 	}
2020 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2021 
2022 	/* Now safe for us to use the shared ring */
2023 	info->connected = BLKIF_STATE_CONNECTED;
2024 
2025 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2026 		struct blkfront_ring_info *rinfo;
2027 
2028 		rinfo = &info->rinfo[r_index];
2029 		/* Kick any other new requests queued since we resumed */
2030 		kick_pending_request_queues(rinfo);
2031 	}
2032 
2033 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2034 		/* Requeue pending requests (flush or discard) */
2035 		list_del_init(&req->queuelist);
2036 		BUG_ON(req->nr_phys_segments > segs);
2037 		blk_mq_requeue_request(req);
2038 	}
2039 	blk_mq_kick_requeue_list(info->rq);
2040 
2041 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2042 		/* Traverse the list of pending bios and re-queue them */
2043 		if (bio_segments(bio) > segs) {
2044 			/*
2045 			 * This bio has more segments than what we can
2046 			 * handle, we have to split it.
2047 			 */
2048 			pending = (bio_segments(bio) + segs - 1) / segs;
2049 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2050 			BUG_ON(split_bio == NULL);
2051 			atomic_set(&split_bio->pending, pending);
2052 			split_bio->bio = bio;
2053 			for (i = 0; i < pending; i++) {
2054 				offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2055 				size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2056 					   (unsigned int)bio_sectors(bio) - offset);
2057 				cloned_bio = bio_clone(bio, GFP_NOIO);
2058 				BUG_ON(cloned_bio == NULL);
2059 				bio_trim(cloned_bio, offset, size);
2060 				cloned_bio->bi_private = split_bio;
2061 				cloned_bio->bi_end_io = split_bio_end;
2062 				submit_bio(cloned_bio);
2063 			}
2064 			/*
2065 			 * Now we have to wait for all those smaller bios to
2066 			 * end, so we can also end the "parent" bio.
2067 			 */
2068 			continue;
2069 		}
2070 		/* We don't need to split this bio */
2071 		submit_bio(bio);
2072 	}
2073 
2074 	return 0;
2075 }
2076 
2077 /**
2078  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2079  * driver restart.  We tear down our blkif structure and recreate it, but
2080  * leave the device-layer structures intact so that this is transparent to the
2081  * rest of the kernel.
2082  */
2083 static int blkfront_resume(struct xenbus_device *dev)
2084 {
2085 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2086 	int err = 0;
2087 	unsigned int i, j;
2088 
2089 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2090 
2091 	bio_list_init(&info->bio_list);
2092 	INIT_LIST_HEAD(&info->requests);
2093 	for (i = 0; i < info->nr_rings; i++) {
2094 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
2095 		struct bio_list merge_bio;
2096 		struct blk_shadow *shadow = rinfo->shadow;
2097 
2098 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2099 			/* Not in use? */
2100 			if (!shadow[j].request)
2101 				continue;
2102 
2103 			/*
2104 			 * Get the bios in the request so we can re-queue them.
2105 			 */
2106 			if (req_op(shadow[i].request) == REQ_OP_FLUSH ||
2107 			    req_op(shadow[i].request) == REQ_OP_DISCARD ||
2108 			    req_op(shadow[i].request) == REQ_OP_SECURE_ERASE ||
2109 			    shadow[j].request->cmd_flags & REQ_FUA) {
2110 				/*
2111 				 * Flush operations don't contain bios, so
2112 				 * we need to requeue the whole request
2113 				 *
2114 				 * XXX: but this doesn't make any sense for a
2115 				 * write with the FUA flag set..
2116 				 */
2117 				list_add(&shadow[j].request->queuelist, &info->requests);
2118 				continue;
2119 			}
2120 			merge_bio.head = shadow[j].request->bio;
2121 			merge_bio.tail = shadow[j].request->biotail;
2122 			bio_list_merge(&info->bio_list, &merge_bio);
2123 			shadow[j].request->bio = NULL;
2124 			blk_mq_end_request(shadow[j].request, 0);
2125 		}
2126 	}
2127 
2128 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2129 
2130 	err = negotiate_mq(info);
2131 	if (err)
2132 		return err;
2133 
2134 	err = talk_to_blkback(dev, info);
2135 	if (!err)
2136 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2137 
2138 	/*
2139 	 * We have to wait for the backend to switch to
2140 	 * connected state, since we want to read which
2141 	 * features it supports.
2142 	 */
2143 
2144 	return err;
2145 }
2146 
2147 static void blkfront_closing(struct blkfront_info *info)
2148 {
2149 	struct xenbus_device *xbdev = info->xbdev;
2150 	struct block_device *bdev = NULL;
2151 
2152 	mutex_lock(&info->mutex);
2153 
2154 	if (xbdev->state == XenbusStateClosing) {
2155 		mutex_unlock(&info->mutex);
2156 		return;
2157 	}
2158 
2159 	if (info->gd)
2160 		bdev = bdget_disk(info->gd, 0);
2161 
2162 	mutex_unlock(&info->mutex);
2163 
2164 	if (!bdev) {
2165 		xenbus_frontend_closed(xbdev);
2166 		return;
2167 	}
2168 
2169 	mutex_lock(&bdev->bd_mutex);
2170 
2171 	if (bdev->bd_openers) {
2172 		xenbus_dev_error(xbdev, -EBUSY,
2173 				 "Device in use; refusing to close");
2174 		xenbus_switch_state(xbdev, XenbusStateClosing);
2175 	} else {
2176 		xlvbd_release_gendisk(info);
2177 		xenbus_frontend_closed(xbdev);
2178 	}
2179 
2180 	mutex_unlock(&bdev->bd_mutex);
2181 	bdput(bdev);
2182 }
2183 
2184 static void blkfront_setup_discard(struct blkfront_info *info)
2185 {
2186 	int err;
2187 	unsigned int discard_granularity;
2188 	unsigned int discard_alignment;
2189 	unsigned int discard_secure;
2190 
2191 	info->feature_discard = 1;
2192 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2193 		"discard-granularity", "%u", &discard_granularity,
2194 		"discard-alignment", "%u", &discard_alignment,
2195 		NULL);
2196 	if (!err) {
2197 		info->discard_granularity = discard_granularity;
2198 		info->discard_alignment = discard_alignment;
2199 	}
2200 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2201 			   "discard-secure", "%u", &discard_secure);
2202 	if (err > 0)
2203 		info->feature_secdiscard = !!discard_secure;
2204 }
2205 
2206 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2207 {
2208 	unsigned int psegs, grants;
2209 	int err, i;
2210 	struct blkfront_info *info = rinfo->dev_info;
2211 
2212 	if (info->max_indirect_segments == 0) {
2213 		if (!HAS_EXTRA_REQ)
2214 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2215 		else {
2216 			/*
2217 			 * When an extra req is required, the maximum
2218 			 * grants supported is related to the size of the
2219 			 * Linux block segment.
2220 			 */
2221 			grants = GRANTS_PER_PSEG;
2222 		}
2223 	}
2224 	else
2225 		grants = info->max_indirect_segments;
2226 	psegs = grants / GRANTS_PER_PSEG;
2227 
2228 	err = fill_grant_buffer(rinfo,
2229 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2230 	if (err)
2231 		goto out_of_memory;
2232 
2233 	if (!info->feature_persistent && info->max_indirect_segments) {
2234 		/*
2235 		 * We are using indirect descriptors but not persistent
2236 		 * grants, we need to allocate a set of pages that can be
2237 		 * used for mapping indirect grefs
2238 		 */
2239 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2240 
2241 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2242 		for (i = 0; i < num; i++) {
2243 			struct page *indirect_page = alloc_page(GFP_NOIO);
2244 			if (!indirect_page)
2245 				goto out_of_memory;
2246 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2247 		}
2248 	}
2249 
2250 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2251 		rinfo->shadow[i].grants_used = kzalloc(
2252 			sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2253 			GFP_NOIO);
2254 		rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2255 		if (info->max_indirect_segments)
2256 			rinfo->shadow[i].indirect_grants = kzalloc(
2257 				sizeof(rinfo->shadow[i].indirect_grants[0]) *
2258 				INDIRECT_GREFS(grants),
2259 				GFP_NOIO);
2260 		if ((rinfo->shadow[i].grants_used == NULL) ||
2261 			(rinfo->shadow[i].sg == NULL) ||
2262 		     (info->max_indirect_segments &&
2263 		     (rinfo->shadow[i].indirect_grants == NULL)))
2264 			goto out_of_memory;
2265 		sg_init_table(rinfo->shadow[i].sg, psegs);
2266 	}
2267 
2268 
2269 	return 0;
2270 
2271 out_of_memory:
2272 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2273 		kfree(rinfo->shadow[i].grants_used);
2274 		rinfo->shadow[i].grants_used = NULL;
2275 		kfree(rinfo->shadow[i].sg);
2276 		rinfo->shadow[i].sg = NULL;
2277 		kfree(rinfo->shadow[i].indirect_grants);
2278 		rinfo->shadow[i].indirect_grants = NULL;
2279 	}
2280 	if (!list_empty(&rinfo->indirect_pages)) {
2281 		struct page *indirect_page, *n;
2282 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2283 			list_del(&indirect_page->lru);
2284 			__free_page(indirect_page);
2285 		}
2286 	}
2287 	return -ENOMEM;
2288 }
2289 
2290 /*
2291  * Gather all backend feature-*
2292  */
2293 static void blkfront_gather_backend_features(struct blkfront_info *info)
2294 {
2295 	int err;
2296 	int barrier, flush, discard, persistent;
2297 	unsigned int indirect_segments;
2298 
2299 	info->feature_flush = 0;
2300 	info->feature_fua = 0;
2301 
2302 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2303 			   "feature-barrier", "%d", &barrier);
2304 
2305 	/*
2306 	 * If there's no "feature-barrier" defined, then it means
2307 	 * we're dealing with a very old backend which writes
2308 	 * synchronously; nothing to do.
2309 	 *
2310 	 * If there are barriers, then we use flush.
2311 	 */
2312 	if (err > 0 && barrier) {
2313 		info->feature_flush = 1;
2314 		info->feature_fua = 1;
2315 	}
2316 
2317 	/*
2318 	 * And if there is "feature-flush-cache" use that above
2319 	 * barriers.
2320 	 */
2321 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2322 			   "feature-flush-cache", "%d", &flush);
2323 
2324 	if (err > 0 && flush) {
2325 		info->feature_flush = 1;
2326 		info->feature_fua = 0;
2327 	}
2328 
2329 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2330 			   "feature-discard", "%d", &discard);
2331 
2332 	if (err > 0 && discard)
2333 		blkfront_setup_discard(info);
2334 
2335 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2336 			   "feature-persistent", "%d", &persistent);
2337 	if (err <= 0)
2338 		info->feature_persistent = 0;
2339 	else
2340 		info->feature_persistent = persistent;
2341 
2342 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2343 			   "feature-max-indirect-segments", "%u",
2344 			   &indirect_segments);
2345 	if (err <= 0)
2346 		info->max_indirect_segments = 0;
2347 	else
2348 		info->max_indirect_segments = min(indirect_segments,
2349 						  xen_blkif_max_segments);
2350 }
2351 
2352 /*
2353  * Invoked when the backend is finally 'ready' (and has told produced
2354  * the details about the physical device - #sectors, size, etc).
2355  */
2356 static void blkfront_connect(struct blkfront_info *info)
2357 {
2358 	unsigned long long sectors;
2359 	unsigned long sector_size;
2360 	unsigned int physical_sector_size;
2361 	unsigned int binfo;
2362 	int err, i;
2363 
2364 	switch (info->connected) {
2365 	case BLKIF_STATE_CONNECTED:
2366 		/*
2367 		 * Potentially, the back-end may be signalling
2368 		 * a capacity change; update the capacity.
2369 		 */
2370 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2371 				   "sectors", "%Lu", &sectors);
2372 		if (XENBUS_EXIST_ERR(err))
2373 			return;
2374 		printk(KERN_INFO "Setting capacity to %Lu\n",
2375 		       sectors);
2376 		set_capacity(info->gd, sectors);
2377 		revalidate_disk(info->gd);
2378 
2379 		return;
2380 	case BLKIF_STATE_SUSPENDED:
2381 		/*
2382 		 * If we are recovering from suspension, we need to wait
2383 		 * for the backend to announce it's features before
2384 		 * reconnecting, at least we need to know if the backend
2385 		 * supports indirect descriptors, and how many.
2386 		 */
2387 		blkif_recover(info);
2388 		return;
2389 
2390 	default:
2391 		break;
2392 	}
2393 
2394 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2395 		__func__, info->xbdev->otherend);
2396 
2397 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2398 			    "sectors", "%llu", &sectors,
2399 			    "info", "%u", &binfo,
2400 			    "sector-size", "%lu", &sector_size,
2401 			    NULL);
2402 	if (err) {
2403 		xenbus_dev_fatal(info->xbdev, err,
2404 				 "reading backend fields at %s",
2405 				 info->xbdev->otherend);
2406 		return;
2407 	}
2408 
2409 	/*
2410 	 * physcial-sector-size is a newer field, so old backends may not
2411 	 * provide this. Assume physical sector size to be the same as
2412 	 * sector_size in that case.
2413 	 */
2414 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2415 			   "physical-sector-size", "%u", &physical_sector_size);
2416 	if (err != 1)
2417 		physical_sector_size = sector_size;
2418 
2419 	blkfront_gather_backend_features(info);
2420 	for (i = 0; i < info->nr_rings; i++) {
2421 		err = blkfront_setup_indirect(&info->rinfo[i]);
2422 		if (err) {
2423 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2424 					 info->xbdev->otherend);
2425 			blkif_free(info, 0);
2426 			break;
2427 		}
2428 	}
2429 
2430 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2431 				  physical_sector_size);
2432 	if (err) {
2433 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2434 				 info->xbdev->otherend);
2435 		return;
2436 	}
2437 
2438 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2439 
2440 	/* Kick pending requests. */
2441 	info->connected = BLKIF_STATE_CONNECTED;
2442 	for (i = 0; i < info->nr_rings; i++)
2443 		kick_pending_request_queues(&info->rinfo[i]);
2444 
2445 	device_add_disk(&info->xbdev->dev, info->gd);
2446 
2447 	info->is_ready = 1;
2448 }
2449 
2450 /**
2451  * Callback received when the backend's state changes.
2452  */
2453 static void blkback_changed(struct xenbus_device *dev,
2454 			    enum xenbus_state backend_state)
2455 {
2456 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2457 
2458 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2459 
2460 	switch (backend_state) {
2461 	case XenbusStateInitWait:
2462 		if (dev->state != XenbusStateInitialising)
2463 			break;
2464 		if (talk_to_blkback(dev, info))
2465 			break;
2466 	case XenbusStateInitialising:
2467 	case XenbusStateInitialised:
2468 	case XenbusStateReconfiguring:
2469 	case XenbusStateReconfigured:
2470 	case XenbusStateUnknown:
2471 		break;
2472 
2473 	case XenbusStateConnected:
2474 		/*
2475 		 * talk_to_blkback sets state to XenbusStateInitialised
2476 		 * and blkfront_connect sets it to XenbusStateConnected
2477 		 * (if connection went OK).
2478 		 *
2479 		 * If the backend (or toolstack) decides to poke at backend
2480 		 * state (and re-trigger the watch by setting the state repeatedly
2481 		 * to XenbusStateConnected (4)) we need to deal with this.
2482 		 * This is allowed as this is used to communicate to the guest
2483 		 * that the size of disk has changed!
2484 		 */
2485 		if ((dev->state != XenbusStateInitialised) &&
2486 		    (dev->state != XenbusStateConnected)) {
2487 			if (talk_to_blkback(dev, info))
2488 				break;
2489 		}
2490 
2491 		blkfront_connect(info);
2492 		break;
2493 
2494 	case XenbusStateClosed:
2495 		if (dev->state == XenbusStateClosed)
2496 			break;
2497 		/* Missed the backend's Closing state -- fallthrough */
2498 	case XenbusStateClosing:
2499 		if (info)
2500 			blkfront_closing(info);
2501 		break;
2502 	}
2503 }
2504 
2505 static int blkfront_remove(struct xenbus_device *xbdev)
2506 {
2507 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2508 	struct block_device *bdev = NULL;
2509 	struct gendisk *disk;
2510 
2511 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2512 
2513 	blkif_free(info, 0);
2514 
2515 	mutex_lock(&info->mutex);
2516 
2517 	disk = info->gd;
2518 	if (disk)
2519 		bdev = bdget_disk(disk, 0);
2520 
2521 	info->xbdev = NULL;
2522 	mutex_unlock(&info->mutex);
2523 
2524 	if (!bdev) {
2525 		kfree(info);
2526 		return 0;
2527 	}
2528 
2529 	/*
2530 	 * The xbdev was removed before we reached the Closed
2531 	 * state. See if it's safe to remove the disk. If the bdev
2532 	 * isn't closed yet, we let release take care of it.
2533 	 */
2534 
2535 	mutex_lock(&bdev->bd_mutex);
2536 	info = disk->private_data;
2537 
2538 	dev_warn(disk_to_dev(disk),
2539 		 "%s was hot-unplugged, %d stale handles\n",
2540 		 xbdev->nodename, bdev->bd_openers);
2541 
2542 	if (info && !bdev->bd_openers) {
2543 		xlvbd_release_gendisk(info);
2544 		disk->private_data = NULL;
2545 		kfree(info);
2546 	}
2547 
2548 	mutex_unlock(&bdev->bd_mutex);
2549 	bdput(bdev);
2550 
2551 	return 0;
2552 }
2553 
2554 static int blkfront_is_ready(struct xenbus_device *dev)
2555 {
2556 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2557 
2558 	return info->is_ready && info->xbdev;
2559 }
2560 
2561 static int blkif_open(struct block_device *bdev, fmode_t mode)
2562 {
2563 	struct gendisk *disk = bdev->bd_disk;
2564 	struct blkfront_info *info;
2565 	int err = 0;
2566 
2567 	mutex_lock(&blkfront_mutex);
2568 
2569 	info = disk->private_data;
2570 	if (!info) {
2571 		/* xbdev gone */
2572 		err = -ERESTARTSYS;
2573 		goto out;
2574 	}
2575 
2576 	mutex_lock(&info->mutex);
2577 
2578 	if (!info->gd)
2579 		/* xbdev is closed */
2580 		err = -ERESTARTSYS;
2581 
2582 	mutex_unlock(&info->mutex);
2583 
2584 out:
2585 	mutex_unlock(&blkfront_mutex);
2586 	return err;
2587 }
2588 
2589 static void blkif_release(struct gendisk *disk, fmode_t mode)
2590 {
2591 	struct blkfront_info *info = disk->private_data;
2592 	struct block_device *bdev;
2593 	struct xenbus_device *xbdev;
2594 
2595 	mutex_lock(&blkfront_mutex);
2596 
2597 	bdev = bdget_disk(disk, 0);
2598 
2599 	if (!bdev) {
2600 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2601 		goto out_mutex;
2602 	}
2603 	if (bdev->bd_openers)
2604 		goto out;
2605 
2606 	/*
2607 	 * Check if we have been instructed to close. We will have
2608 	 * deferred this request, because the bdev was still open.
2609 	 */
2610 
2611 	mutex_lock(&info->mutex);
2612 	xbdev = info->xbdev;
2613 
2614 	if (xbdev && xbdev->state == XenbusStateClosing) {
2615 		/* pending switch to state closed */
2616 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2617 		xlvbd_release_gendisk(info);
2618 		xenbus_frontend_closed(info->xbdev);
2619  	}
2620 
2621 	mutex_unlock(&info->mutex);
2622 
2623 	if (!xbdev) {
2624 		/* sudden device removal */
2625 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2626 		xlvbd_release_gendisk(info);
2627 		disk->private_data = NULL;
2628 		kfree(info);
2629 	}
2630 
2631 out:
2632 	bdput(bdev);
2633 out_mutex:
2634 	mutex_unlock(&blkfront_mutex);
2635 }
2636 
2637 static const struct block_device_operations xlvbd_block_fops =
2638 {
2639 	.owner = THIS_MODULE,
2640 	.open = blkif_open,
2641 	.release = blkif_release,
2642 	.getgeo = blkif_getgeo,
2643 	.ioctl = blkif_ioctl,
2644 };
2645 
2646 
2647 static const struct xenbus_device_id blkfront_ids[] = {
2648 	{ "vbd" },
2649 	{ "" }
2650 };
2651 
2652 static struct xenbus_driver blkfront_driver = {
2653 	.ids  = blkfront_ids,
2654 	.probe = blkfront_probe,
2655 	.remove = blkfront_remove,
2656 	.resume = blkfront_resume,
2657 	.otherend_changed = blkback_changed,
2658 	.is_ready = blkfront_is_ready,
2659 };
2660 
2661 static int __init xlblk_init(void)
2662 {
2663 	int ret;
2664 	int nr_cpus = num_online_cpus();
2665 
2666 	if (!xen_domain())
2667 		return -ENODEV;
2668 
2669 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2670 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2671 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2672 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2673 	}
2674 
2675 	if (xen_blkif_max_queues > nr_cpus) {
2676 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2677 			xen_blkif_max_queues, nr_cpus);
2678 		xen_blkif_max_queues = nr_cpus;
2679 	}
2680 
2681 	if (!xen_has_pv_disk_devices())
2682 		return -ENODEV;
2683 
2684 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2685 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2686 		       XENVBD_MAJOR, DEV_NAME);
2687 		return -ENODEV;
2688 	}
2689 
2690 	ret = xenbus_register_frontend(&blkfront_driver);
2691 	if (ret) {
2692 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2693 		return ret;
2694 	}
2695 
2696 	return 0;
2697 }
2698 module_init(xlblk_init);
2699 
2700 
2701 static void __exit xlblk_exit(void)
2702 {
2703 	xenbus_unregister_driver(&blkfront_driver);
2704 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2705 	kfree(minors);
2706 }
2707 module_exit(xlblk_exit);
2708 
2709 MODULE_DESCRIPTION("Xen virtual block device frontend");
2710 MODULE_LICENSE("GPL");
2711 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2712 MODULE_ALIAS("xen:vbd");
2713 MODULE_ALIAS("xenblk");
2714