xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 2596e07a)
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, xen_blkif_max_segments, int, S_IRUGO);
129 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
130 
131 static unsigned int xen_blkif_max_queues = 4;
132 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
133 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
134 
135 /*
136  * Maximum order of pages to be used for the shared ring between front and
137  * backend, 4KB page granularity is used.
138  */
139 static unsigned int xen_blkif_max_ring_order;
140 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
141 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
142 
143 #define BLK_RING_SIZE(info)	\
144 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
145 
146 #define BLK_MAX_RING_SIZE	\
147 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
148 
149 /*
150  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
151  * characters are enough. Define to 20 to keep consistent with backend.
152  */
153 #define RINGREF_NAME_LEN (20)
154 /*
155  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
156  */
157 #define QUEUE_NAME_LEN (17)
158 
159 /*
160  *  Per-ring info.
161  *  Every blkfront device can associate with one or more blkfront_ring_info,
162  *  depending on how many hardware queues/rings to be used.
163  */
164 struct blkfront_ring_info {
165 	/* Lock to protect data in every ring buffer. */
166 	spinlock_t ring_lock;
167 	struct blkif_front_ring ring;
168 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
169 	unsigned int evtchn, irq;
170 	struct work_struct work;
171 	struct gnttab_free_callback callback;
172 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
173 	struct list_head indirect_pages;
174 	struct list_head grants;
175 	unsigned int persistent_gnts_c;
176 	unsigned long shadow_free;
177 	struct blkfront_info *dev_info;
178 };
179 
180 /*
181  * We have one of these per vbd, whether ide, scsi or 'other'.  They
182  * hang in private_data off the gendisk structure. We may end up
183  * putting all kinds of interesting stuff here :-)
184  */
185 struct blkfront_info
186 {
187 	struct mutex mutex;
188 	struct xenbus_device *xbdev;
189 	struct gendisk *gd;
190 	int vdevice;
191 	blkif_vdev_t handle;
192 	enum blkif_state connected;
193 	/* Number of pages per ring buffer. */
194 	unsigned int nr_ring_pages;
195 	struct request_queue *rq;
196 	unsigned int feature_flush;
197 	unsigned int feature_discard:1;
198 	unsigned int feature_secdiscard:1;
199 	unsigned int discard_granularity;
200 	unsigned int discard_alignment;
201 	unsigned int feature_persistent:1;
202 	/* Number of 4KB segments handled */
203 	unsigned int max_indirect_segments;
204 	int is_ready;
205 	struct blk_mq_tag_set tag_set;
206 	struct blkfront_ring_info *rinfo;
207 	unsigned int nr_rings;
208 };
209 
210 static unsigned int nr_minors;
211 static unsigned long *minors;
212 static DEFINE_SPINLOCK(minor_lock);
213 
214 #define GRANT_INVALID_REF	0
215 
216 #define PARTS_PER_DISK		16
217 #define PARTS_PER_EXT_DISK      256
218 
219 #define BLKIF_MAJOR(dev) ((dev)>>8)
220 #define BLKIF_MINOR(dev) ((dev) & 0xff)
221 
222 #define EXT_SHIFT 28
223 #define EXTENDED (1<<EXT_SHIFT)
224 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
225 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
226 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
227 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
228 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
229 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
230 
231 #define DEV_NAME	"xvd"	/* name in /dev */
232 
233 /*
234  * Grants are always the same size as a Xen page (i.e 4KB).
235  * A physical segment is always the same size as a Linux page.
236  * Number of grants per physical segment
237  */
238 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
239 
240 #define GRANTS_PER_INDIRECT_FRAME \
241 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
242 
243 #define PSEGS_PER_INDIRECT_FRAME	\
244 	(GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
245 
246 #define INDIRECT_GREFS(_grants)		\
247 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
248 
249 #define GREFS(_psegs)	((_psegs) * GRANTS_PER_PSEG)
250 
251 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
252 static void blkfront_gather_backend_features(struct blkfront_info *info);
253 
254 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
255 {
256 	unsigned long free = rinfo->shadow_free;
257 
258 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
259 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
260 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
261 	return free;
262 }
263 
264 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
265 			      unsigned long id)
266 {
267 	if (rinfo->shadow[id].req.u.rw.id != id)
268 		return -EINVAL;
269 	if (rinfo->shadow[id].request == NULL)
270 		return -EINVAL;
271 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
272 	rinfo->shadow[id].request = NULL;
273 	rinfo->shadow_free = id;
274 	return 0;
275 }
276 
277 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
278 {
279 	struct blkfront_info *info = rinfo->dev_info;
280 	struct page *granted_page;
281 	struct grant *gnt_list_entry, *n;
282 	int i = 0;
283 
284 	while (i < num) {
285 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
286 		if (!gnt_list_entry)
287 			goto out_of_memory;
288 
289 		if (info->feature_persistent) {
290 			granted_page = alloc_page(GFP_NOIO);
291 			if (!granted_page) {
292 				kfree(gnt_list_entry);
293 				goto out_of_memory;
294 			}
295 			gnt_list_entry->page = granted_page;
296 		}
297 
298 		gnt_list_entry->gref = GRANT_INVALID_REF;
299 		list_add(&gnt_list_entry->node, &rinfo->grants);
300 		i++;
301 	}
302 
303 	return 0;
304 
305 out_of_memory:
306 	list_for_each_entry_safe(gnt_list_entry, n,
307 	                         &rinfo->grants, node) {
308 		list_del(&gnt_list_entry->node);
309 		if (info->feature_persistent)
310 			__free_page(gnt_list_entry->page);
311 		kfree(gnt_list_entry);
312 		i--;
313 	}
314 	BUG_ON(i != 0);
315 	return -ENOMEM;
316 }
317 
318 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
319 {
320 	struct grant *gnt_list_entry;
321 
322 	BUG_ON(list_empty(&rinfo->grants));
323 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
324 					  node);
325 	list_del(&gnt_list_entry->node);
326 
327 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
328 		rinfo->persistent_gnts_c--;
329 
330 	return gnt_list_entry;
331 }
332 
333 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
334 					const struct blkfront_info *info)
335 {
336 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
337 						 info->xbdev->otherend_id,
338 						 gnt_list_entry->page,
339 						 0);
340 }
341 
342 static struct grant *get_grant(grant_ref_t *gref_head,
343 			       unsigned long gfn,
344 			       struct blkfront_ring_info *rinfo)
345 {
346 	struct grant *gnt_list_entry = get_free_grant(rinfo);
347 	struct blkfront_info *info = rinfo->dev_info;
348 
349 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
350 		return gnt_list_entry;
351 
352 	/* Assign a gref to this page */
353 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
354 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
355 	if (info->feature_persistent)
356 		grant_foreign_access(gnt_list_entry, info);
357 	else {
358 		/* Grant access to the GFN passed by the caller */
359 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
360 						info->xbdev->otherend_id,
361 						gfn, 0);
362 	}
363 
364 	return gnt_list_entry;
365 }
366 
367 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
368 					struct blkfront_ring_info *rinfo)
369 {
370 	struct grant *gnt_list_entry = get_free_grant(rinfo);
371 	struct blkfront_info *info = rinfo->dev_info;
372 
373 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
374 		return gnt_list_entry;
375 
376 	/* Assign a gref to this page */
377 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
378 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
379 	if (!info->feature_persistent) {
380 		struct page *indirect_page;
381 
382 		/* Fetch a pre-allocated page to use for indirect grefs */
383 		BUG_ON(list_empty(&rinfo->indirect_pages));
384 		indirect_page = list_first_entry(&rinfo->indirect_pages,
385 						 struct page, lru);
386 		list_del(&indirect_page->lru);
387 		gnt_list_entry->page = indirect_page;
388 	}
389 	grant_foreign_access(gnt_list_entry, info);
390 
391 	return gnt_list_entry;
392 }
393 
394 static const char *op_name(int op)
395 {
396 	static const char *const names[] = {
397 		[BLKIF_OP_READ] = "read",
398 		[BLKIF_OP_WRITE] = "write",
399 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
400 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
401 		[BLKIF_OP_DISCARD] = "discard" };
402 
403 	if (op < 0 || op >= ARRAY_SIZE(names))
404 		return "unknown";
405 
406 	if (!names[op])
407 		return "reserved";
408 
409 	return names[op];
410 }
411 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
412 {
413 	unsigned int end = minor + nr;
414 	int rc;
415 
416 	if (end > nr_minors) {
417 		unsigned long *bitmap, *old;
418 
419 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
420 				 GFP_KERNEL);
421 		if (bitmap == NULL)
422 			return -ENOMEM;
423 
424 		spin_lock(&minor_lock);
425 		if (end > nr_minors) {
426 			old = minors;
427 			memcpy(bitmap, minors,
428 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
429 			minors = bitmap;
430 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
431 		} else
432 			old = bitmap;
433 		spin_unlock(&minor_lock);
434 		kfree(old);
435 	}
436 
437 	spin_lock(&minor_lock);
438 	if (find_next_bit(minors, end, minor) >= end) {
439 		bitmap_set(minors, minor, nr);
440 		rc = 0;
441 	} else
442 		rc = -EBUSY;
443 	spin_unlock(&minor_lock);
444 
445 	return rc;
446 }
447 
448 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
449 {
450 	unsigned int end = minor + nr;
451 
452 	BUG_ON(end > nr_minors);
453 	spin_lock(&minor_lock);
454 	bitmap_clear(minors,  minor, nr);
455 	spin_unlock(&minor_lock);
456 }
457 
458 static void blkif_restart_queue_callback(void *arg)
459 {
460 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
461 	schedule_work(&rinfo->work);
462 }
463 
464 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
465 {
466 	/* We don't have real geometry info, but let's at least return
467 	   values consistent with the size of the device */
468 	sector_t nsect = get_capacity(bd->bd_disk);
469 	sector_t cylinders = nsect;
470 
471 	hg->heads = 0xff;
472 	hg->sectors = 0x3f;
473 	sector_div(cylinders, hg->heads * hg->sectors);
474 	hg->cylinders = cylinders;
475 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
476 		hg->cylinders = 0xffff;
477 	return 0;
478 }
479 
480 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
481 		       unsigned command, unsigned long argument)
482 {
483 	struct blkfront_info *info = bdev->bd_disk->private_data;
484 	int i;
485 
486 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
487 		command, (long)argument);
488 
489 	switch (command) {
490 	case CDROMMULTISESSION:
491 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
492 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
493 			if (put_user(0, (char __user *)(argument + i)))
494 				return -EFAULT;
495 		return 0;
496 
497 	case CDROM_GET_CAPABILITY: {
498 		struct gendisk *gd = info->gd;
499 		if (gd->flags & GENHD_FL_CD)
500 			return 0;
501 		return -EINVAL;
502 	}
503 
504 	default:
505 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
506 		  command);*/
507 		return -EINVAL; /* same return as native Linux */
508 	}
509 
510 	return 0;
511 }
512 
513 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
514 					    struct request *req,
515 					    struct blkif_request **ring_req)
516 {
517 	unsigned long id;
518 
519 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
520 	rinfo->ring.req_prod_pvt++;
521 
522 	id = get_id_from_freelist(rinfo);
523 	rinfo->shadow[id].request = req;
524 	rinfo->shadow[id].status = REQ_WAITING;
525 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
526 
527 	(*ring_req)->u.rw.id = id;
528 
529 	return id;
530 }
531 
532 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
533 {
534 	struct blkfront_info *info = rinfo->dev_info;
535 	struct blkif_request *ring_req;
536 	unsigned long id;
537 
538 	/* Fill out a communications ring structure. */
539 	id = blkif_ring_get_request(rinfo, req, &ring_req);
540 
541 	ring_req->operation = BLKIF_OP_DISCARD;
542 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
543 	ring_req->u.discard.id = id;
544 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
545 	if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
546 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
547 	else
548 		ring_req->u.discard.flag = 0;
549 
550 	/* Keep a private copy so we can reissue requests when recovering. */
551 	rinfo->shadow[id].req = *ring_req;
552 
553 	return 0;
554 }
555 
556 struct setup_rw_req {
557 	unsigned int grant_idx;
558 	struct blkif_request_segment *segments;
559 	struct blkfront_ring_info *rinfo;
560 	struct blkif_request *ring_req;
561 	grant_ref_t gref_head;
562 	unsigned int id;
563 	/* Only used when persistent grant is used and it's a read request */
564 	bool need_copy;
565 	unsigned int bvec_off;
566 	char *bvec_data;
567 
568 	bool require_extra_req;
569 	struct blkif_request *extra_ring_req;
570 };
571 
572 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
573 				     unsigned int len, void *data)
574 {
575 	struct setup_rw_req *setup = data;
576 	int n, ref;
577 	struct grant *gnt_list_entry;
578 	unsigned int fsect, lsect;
579 	/* Convenient aliases */
580 	unsigned int grant_idx = setup->grant_idx;
581 	struct blkif_request *ring_req = setup->ring_req;
582 	struct blkfront_ring_info *rinfo = setup->rinfo;
583 	/*
584 	 * We always use the shadow of the first request to store the list
585 	 * of grant associated to the block I/O request. This made the
586 	 * completion more easy to handle even if the block I/O request is
587 	 * split.
588 	 */
589 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
590 
591 	if (unlikely(setup->require_extra_req &&
592 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
593 		/*
594 		 * We are using the second request, setup grant_idx
595 		 * to be the index of the segment array.
596 		 */
597 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
598 		ring_req = setup->extra_ring_req;
599 	}
600 
601 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
602 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
603 		if (setup->segments)
604 			kunmap_atomic(setup->segments);
605 
606 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
607 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
608 		shadow->indirect_grants[n] = gnt_list_entry;
609 		setup->segments = kmap_atomic(gnt_list_entry->page);
610 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
611 	}
612 
613 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
614 	ref = gnt_list_entry->gref;
615 	/*
616 	 * All the grants are stored in the shadow of the first
617 	 * request. Therefore we have to use the global index.
618 	 */
619 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
620 
621 	if (setup->need_copy) {
622 		void *shared_data;
623 
624 		shared_data = kmap_atomic(gnt_list_entry->page);
625 		/*
626 		 * this does not wipe data stored outside the
627 		 * range sg->offset..sg->offset+sg->length.
628 		 * Therefore, blkback *could* see data from
629 		 * previous requests. This is OK as long as
630 		 * persistent grants are shared with just one
631 		 * domain. It may need refactoring if this
632 		 * changes
633 		 */
634 		memcpy(shared_data + offset,
635 		       setup->bvec_data + setup->bvec_off,
636 		       len);
637 
638 		kunmap_atomic(shared_data);
639 		setup->bvec_off += len;
640 	}
641 
642 	fsect = offset >> 9;
643 	lsect = fsect + (len >> 9) - 1;
644 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
645 		ring_req->u.rw.seg[grant_idx] =
646 			(struct blkif_request_segment) {
647 				.gref       = ref,
648 				.first_sect = fsect,
649 				.last_sect  = lsect };
650 	} else {
651 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
652 			(struct blkif_request_segment) {
653 				.gref       = ref,
654 				.first_sect = fsect,
655 				.last_sect  = lsect };
656 	}
657 
658 	(setup->grant_idx)++;
659 }
660 
661 static void blkif_setup_extra_req(struct blkif_request *first,
662 				  struct blkif_request *second)
663 {
664 	uint16_t nr_segments = first->u.rw.nr_segments;
665 
666 	/*
667 	 * The second request is only present when the first request uses
668 	 * all its segments. It's always the continuity of the first one.
669 	 */
670 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
671 
672 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
673 	second->u.rw.sector_number = first->u.rw.sector_number +
674 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
675 
676 	second->u.rw.handle = first->u.rw.handle;
677 	second->operation = first->operation;
678 }
679 
680 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
681 {
682 	struct blkfront_info *info = rinfo->dev_info;
683 	struct blkif_request *ring_req, *extra_ring_req = NULL;
684 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
685 	bool require_extra_req = false;
686 	int i;
687 	struct setup_rw_req setup = {
688 		.grant_idx = 0,
689 		.segments = NULL,
690 		.rinfo = rinfo,
691 		.need_copy = rq_data_dir(req) && info->feature_persistent,
692 	};
693 
694 	/*
695 	 * Used to store if we are able to queue the request by just using
696 	 * existing persistent grants, or if we have to get new grants,
697 	 * as there are not sufficiently many free.
698 	 */
699 	struct scatterlist *sg;
700 	int num_sg, max_grefs, num_grant;
701 
702 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
703 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
704 		/*
705 		 * If we are using indirect segments we need to account
706 		 * for the indirect grefs used in the request.
707 		 */
708 		max_grefs += INDIRECT_GREFS(max_grefs);
709 
710 	/*
711 	 * We have to reserve 'max_grefs' grants because persistent
712 	 * grants are shared by all rings.
713 	 */
714 	if (max_grefs > 0)
715 		if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
716 			gnttab_request_free_callback(
717 				&rinfo->callback,
718 				blkif_restart_queue_callback,
719 				rinfo,
720 				max_grefs);
721 			return 1;
722 		}
723 
724 	/* Fill out a communications ring structure. */
725 	id = blkif_ring_get_request(rinfo, req, &ring_req);
726 
727 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
728 	num_grant = 0;
729 	/* Calculate the number of grant used */
730 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
731 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
732 
733 	require_extra_req = info->max_indirect_segments == 0 &&
734 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
735 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
736 
737 	rinfo->shadow[id].num_sg = num_sg;
738 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
739 	    likely(!require_extra_req)) {
740 		/*
741 		 * The indirect operation can only be a BLKIF_OP_READ or
742 		 * BLKIF_OP_WRITE
743 		 */
744 		BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
745 		ring_req->operation = BLKIF_OP_INDIRECT;
746 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
747 			BLKIF_OP_WRITE : BLKIF_OP_READ;
748 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
749 		ring_req->u.indirect.handle = info->handle;
750 		ring_req->u.indirect.nr_segments = num_grant;
751 	} else {
752 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
753 		ring_req->u.rw.handle = info->handle;
754 		ring_req->operation = rq_data_dir(req) ?
755 			BLKIF_OP_WRITE : BLKIF_OP_READ;
756 		if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
757 			/*
758 			 * Ideally we can do an unordered flush-to-disk.
759 			 * In case the backend onlysupports barriers, use that.
760 			 * A barrier request a superset of FUA, so we can
761 			 * implement it the same way.  (It's also a FLUSH+FUA,
762 			 * since it is guaranteed ordered WRT previous writes.)
763 			 */
764 			switch (info->feature_flush &
765 				((REQ_FLUSH|REQ_FUA))) {
766 			case REQ_FLUSH|REQ_FUA:
767 				ring_req->operation =
768 					BLKIF_OP_WRITE_BARRIER;
769 				break;
770 			case REQ_FLUSH:
771 				ring_req->operation =
772 					BLKIF_OP_FLUSH_DISKCACHE;
773 				break;
774 			default:
775 				ring_req->operation = 0;
776 			}
777 		}
778 		ring_req->u.rw.nr_segments = num_grant;
779 		if (unlikely(require_extra_req)) {
780 			extra_id = blkif_ring_get_request(rinfo, req,
781 							  &extra_ring_req);
782 			/*
783 			 * Only the first request contains the scatter-gather
784 			 * list.
785 			 */
786 			rinfo->shadow[extra_id].num_sg = 0;
787 
788 			blkif_setup_extra_req(ring_req, extra_ring_req);
789 
790 			/* Link the 2 requests together */
791 			rinfo->shadow[extra_id].associated_id = id;
792 			rinfo->shadow[id].associated_id = extra_id;
793 		}
794 	}
795 
796 	setup.ring_req = ring_req;
797 	setup.id = id;
798 
799 	setup.require_extra_req = require_extra_req;
800 	if (unlikely(require_extra_req))
801 		setup.extra_ring_req = extra_ring_req;
802 
803 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
804 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
805 
806 		if (setup.need_copy) {
807 			setup.bvec_off = sg->offset;
808 			setup.bvec_data = kmap_atomic(sg_page(sg));
809 		}
810 
811 		gnttab_foreach_grant_in_range(sg_page(sg),
812 					      sg->offset,
813 					      sg->length,
814 					      blkif_setup_rw_req_grant,
815 					      &setup);
816 
817 		if (setup.need_copy)
818 			kunmap_atomic(setup.bvec_data);
819 	}
820 	if (setup.segments)
821 		kunmap_atomic(setup.segments);
822 
823 	/* Keep a private copy so we can reissue requests when recovering. */
824 	rinfo->shadow[id].req = *ring_req;
825 	if (unlikely(require_extra_req))
826 		rinfo->shadow[extra_id].req = *extra_ring_req;
827 
828 	if (max_grefs > 0)
829 		gnttab_free_grant_references(setup.gref_head);
830 
831 	return 0;
832 }
833 
834 /*
835  * Generate a Xen blkfront IO request from a blk layer request.  Reads
836  * and writes are handled as expected.
837  *
838  * @req: a request struct
839  */
840 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
841 {
842 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
843 		return 1;
844 
845 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
846 		return blkif_queue_discard_req(req, rinfo);
847 	else
848 		return blkif_queue_rw_req(req, rinfo);
849 }
850 
851 static inline void flush_requests(struct blkfront_ring_info *rinfo)
852 {
853 	int notify;
854 
855 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
856 
857 	if (notify)
858 		notify_remote_via_irq(rinfo->irq);
859 }
860 
861 static inline bool blkif_request_flush_invalid(struct request *req,
862 					       struct blkfront_info *info)
863 {
864 	return ((req->cmd_type != REQ_TYPE_FS) ||
865 		((req->cmd_flags & REQ_FLUSH) &&
866 		 !(info->feature_flush & REQ_FLUSH)) ||
867 		((req->cmd_flags & REQ_FUA) &&
868 		 !(info->feature_flush & REQ_FUA)));
869 }
870 
871 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
872 			  const struct blk_mq_queue_data *qd)
873 {
874 	unsigned long flags;
875 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)hctx->driver_data;
876 
877 	blk_mq_start_request(qd->rq);
878 	spin_lock_irqsave(&rinfo->ring_lock, flags);
879 	if (RING_FULL(&rinfo->ring))
880 		goto out_busy;
881 
882 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
883 		goto out_err;
884 
885 	if (blkif_queue_request(qd->rq, rinfo))
886 		goto out_busy;
887 
888 	flush_requests(rinfo);
889 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
890 	return BLK_MQ_RQ_QUEUE_OK;
891 
892 out_err:
893 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
894 	return BLK_MQ_RQ_QUEUE_ERROR;
895 
896 out_busy:
897 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
898 	blk_mq_stop_hw_queue(hctx);
899 	return BLK_MQ_RQ_QUEUE_BUSY;
900 }
901 
902 static int blk_mq_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
903 			    unsigned int index)
904 {
905 	struct blkfront_info *info = (struct blkfront_info *)data;
906 
907 	BUG_ON(info->nr_rings <= index);
908 	hctx->driver_data = &info->rinfo[index];
909 	return 0;
910 }
911 
912 static struct blk_mq_ops blkfront_mq_ops = {
913 	.queue_rq = blkif_queue_rq,
914 	.map_queue = blk_mq_map_queue,
915 	.init_hctx = blk_mq_init_hctx,
916 };
917 
918 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
919 				unsigned int physical_sector_size,
920 				unsigned int segments)
921 {
922 	struct request_queue *rq;
923 	struct blkfront_info *info = gd->private_data;
924 
925 	memset(&info->tag_set, 0, sizeof(info->tag_set));
926 	info->tag_set.ops = &blkfront_mq_ops;
927 	info->tag_set.nr_hw_queues = info->nr_rings;
928 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
929 		/*
930 		 * When indirect descriptior is not supported, the I/O request
931 		 * will be split between multiple request in the ring.
932 		 * To avoid problems when sending the request, divide by
933 		 * 2 the depth of the queue.
934 		 */
935 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
936 	} else
937 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
938 	info->tag_set.numa_node = NUMA_NO_NODE;
939 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
940 	info->tag_set.cmd_size = 0;
941 	info->tag_set.driver_data = info;
942 
943 	if (blk_mq_alloc_tag_set(&info->tag_set))
944 		return -EINVAL;
945 	rq = blk_mq_init_queue(&info->tag_set);
946 	if (IS_ERR(rq)) {
947 		blk_mq_free_tag_set(&info->tag_set);
948 		return PTR_ERR(rq);
949 	}
950 
951 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
952 
953 	if (info->feature_discard) {
954 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
955 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
956 		rq->limits.discard_granularity = info->discard_granularity;
957 		rq->limits.discard_alignment = info->discard_alignment;
958 		if (info->feature_secdiscard)
959 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
960 	}
961 
962 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
963 	blk_queue_logical_block_size(rq, sector_size);
964 	blk_queue_physical_block_size(rq, physical_sector_size);
965 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
966 
967 	/* Each segment in a request is up to an aligned page in size. */
968 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
969 	blk_queue_max_segment_size(rq, PAGE_SIZE);
970 
971 	/* Ensure a merged request will fit in a single I/O ring slot. */
972 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
973 
974 	/* Make sure buffer addresses are sector-aligned. */
975 	blk_queue_dma_alignment(rq, 511);
976 
977 	/* Make sure we don't use bounce buffers. */
978 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
979 
980 	gd->queue = rq;
981 
982 	return 0;
983 }
984 
985 static const char *flush_info(unsigned int feature_flush)
986 {
987 	switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
988 	case REQ_FLUSH|REQ_FUA:
989 		return "barrier: enabled;";
990 	case REQ_FLUSH:
991 		return "flush diskcache: enabled;";
992 	default:
993 		return "barrier or flush: disabled;";
994 	}
995 }
996 
997 static void xlvbd_flush(struct blkfront_info *info)
998 {
999 	blk_queue_flush(info->rq, info->feature_flush);
1000 	pr_info("blkfront: %s: %s %s %s %s %s\n",
1001 		info->gd->disk_name, flush_info(info->feature_flush),
1002 		"persistent grants:", info->feature_persistent ?
1003 		"enabled;" : "disabled;", "indirect descriptors:",
1004 		info->max_indirect_segments ? "enabled;" : "disabled;");
1005 }
1006 
1007 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1008 {
1009 	int major;
1010 	major = BLKIF_MAJOR(vdevice);
1011 	*minor = BLKIF_MINOR(vdevice);
1012 	switch (major) {
1013 		case XEN_IDE0_MAJOR:
1014 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1015 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1016 				EMULATED_HD_DISK_MINOR_OFFSET;
1017 			break;
1018 		case XEN_IDE1_MAJOR:
1019 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1020 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1021 				EMULATED_HD_DISK_MINOR_OFFSET;
1022 			break;
1023 		case XEN_SCSI_DISK0_MAJOR:
1024 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1025 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1026 			break;
1027 		case XEN_SCSI_DISK1_MAJOR:
1028 		case XEN_SCSI_DISK2_MAJOR:
1029 		case XEN_SCSI_DISK3_MAJOR:
1030 		case XEN_SCSI_DISK4_MAJOR:
1031 		case XEN_SCSI_DISK5_MAJOR:
1032 		case XEN_SCSI_DISK6_MAJOR:
1033 		case XEN_SCSI_DISK7_MAJOR:
1034 			*offset = (*minor / PARTS_PER_DISK) +
1035 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1036 				EMULATED_SD_DISK_NAME_OFFSET;
1037 			*minor = *minor +
1038 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1039 				EMULATED_SD_DISK_MINOR_OFFSET;
1040 			break;
1041 		case XEN_SCSI_DISK8_MAJOR:
1042 		case XEN_SCSI_DISK9_MAJOR:
1043 		case XEN_SCSI_DISK10_MAJOR:
1044 		case XEN_SCSI_DISK11_MAJOR:
1045 		case XEN_SCSI_DISK12_MAJOR:
1046 		case XEN_SCSI_DISK13_MAJOR:
1047 		case XEN_SCSI_DISK14_MAJOR:
1048 		case XEN_SCSI_DISK15_MAJOR:
1049 			*offset = (*minor / PARTS_PER_DISK) +
1050 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1051 				EMULATED_SD_DISK_NAME_OFFSET;
1052 			*minor = *minor +
1053 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1054 				EMULATED_SD_DISK_MINOR_OFFSET;
1055 			break;
1056 		case XENVBD_MAJOR:
1057 			*offset = *minor / PARTS_PER_DISK;
1058 			break;
1059 		default:
1060 			printk(KERN_WARNING "blkfront: your disk configuration is "
1061 					"incorrect, please use an xvd device instead\n");
1062 			return -ENODEV;
1063 	}
1064 	return 0;
1065 }
1066 
1067 static char *encode_disk_name(char *ptr, unsigned int n)
1068 {
1069 	if (n >= 26)
1070 		ptr = encode_disk_name(ptr, n / 26 - 1);
1071 	*ptr = 'a' + n % 26;
1072 	return ptr + 1;
1073 }
1074 
1075 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1076 			       struct blkfront_info *info,
1077 			       u16 vdisk_info, u16 sector_size,
1078 			       unsigned int physical_sector_size)
1079 {
1080 	struct gendisk *gd;
1081 	int nr_minors = 1;
1082 	int err;
1083 	unsigned int offset;
1084 	int minor;
1085 	int nr_parts;
1086 	char *ptr;
1087 
1088 	BUG_ON(info->gd != NULL);
1089 	BUG_ON(info->rq != NULL);
1090 
1091 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1092 		/* this is above the extended range; something is wrong */
1093 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1094 		return -ENODEV;
1095 	}
1096 
1097 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1098 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1099 		if (err)
1100 			return err;
1101  		nr_parts = PARTS_PER_DISK;
1102 	} else {
1103 		minor = BLKIF_MINOR_EXT(info->vdevice);
1104 		nr_parts = PARTS_PER_EXT_DISK;
1105 		offset = minor / nr_parts;
1106 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1107 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1108 					"emulated IDE disks,\n\t choose an xvd device name"
1109 					"from xvde on\n", info->vdevice);
1110 	}
1111 	if (minor >> MINORBITS) {
1112 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1113 			info->vdevice, minor);
1114 		return -ENODEV;
1115 	}
1116 
1117 	if ((minor % nr_parts) == 0)
1118 		nr_minors = nr_parts;
1119 
1120 	err = xlbd_reserve_minors(minor, nr_minors);
1121 	if (err)
1122 		goto out;
1123 	err = -ENODEV;
1124 
1125 	gd = alloc_disk(nr_minors);
1126 	if (gd == NULL)
1127 		goto release;
1128 
1129 	strcpy(gd->disk_name, DEV_NAME);
1130 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1131 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1132 	if (nr_minors > 1)
1133 		*ptr = 0;
1134 	else
1135 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1136 			 "%d", minor & (nr_parts - 1));
1137 
1138 	gd->major = XENVBD_MAJOR;
1139 	gd->first_minor = minor;
1140 	gd->fops = &xlvbd_block_fops;
1141 	gd->private_data = info;
1142 	gd->driverfs_dev = &(info->xbdev->dev);
1143 	set_capacity(gd, capacity);
1144 
1145 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1146 				 info->max_indirect_segments ? :
1147 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1148 		del_gendisk(gd);
1149 		goto release;
1150 	}
1151 
1152 	info->rq = gd->queue;
1153 	info->gd = gd;
1154 
1155 	xlvbd_flush(info);
1156 
1157 	if (vdisk_info & VDISK_READONLY)
1158 		set_disk_ro(gd, 1);
1159 
1160 	if (vdisk_info & VDISK_REMOVABLE)
1161 		gd->flags |= GENHD_FL_REMOVABLE;
1162 
1163 	if (vdisk_info & VDISK_CDROM)
1164 		gd->flags |= GENHD_FL_CD;
1165 
1166 	return 0;
1167 
1168  release:
1169 	xlbd_release_minors(minor, nr_minors);
1170  out:
1171 	return err;
1172 }
1173 
1174 static void xlvbd_release_gendisk(struct blkfront_info *info)
1175 {
1176 	unsigned int minor, nr_minors, i;
1177 
1178 	if (info->rq == NULL)
1179 		return;
1180 
1181 	/* No more blkif_request(). */
1182 	blk_mq_stop_hw_queues(info->rq);
1183 
1184 	for (i = 0; i < info->nr_rings; i++) {
1185 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1186 
1187 		/* No more gnttab callback work. */
1188 		gnttab_cancel_free_callback(&rinfo->callback);
1189 
1190 		/* Flush gnttab callback work. Must be done with no locks held. */
1191 		flush_work(&rinfo->work);
1192 	}
1193 
1194 	del_gendisk(info->gd);
1195 
1196 	minor = info->gd->first_minor;
1197 	nr_minors = info->gd->minors;
1198 	xlbd_release_minors(minor, nr_minors);
1199 
1200 	blk_cleanup_queue(info->rq);
1201 	blk_mq_free_tag_set(&info->tag_set);
1202 	info->rq = NULL;
1203 
1204 	put_disk(info->gd);
1205 	info->gd = NULL;
1206 }
1207 
1208 /* Already hold rinfo->ring_lock. */
1209 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1210 {
1211 	if (!RING_FULL(&rinfo->ring))
1212 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1213 }
1214 
1215 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1216 {
1217 	unsigned long flags;
1218 
1219 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1220 	kick_pending_request_queues_locked(rinfo);
1221 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1222 }
1223 
1224 static void blkif_restart_queue(struct work_struct *work)
1225 {
1226 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1227 
1228 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1229 		kick_pending_request_queues(rinfo);
1230 }
1231 
1232 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1233 {
1234 	struct grant *persistent_gnt, *n;
1235 	struct blkfront_info *info = rinfo->dev_info;
1236 	int i, j, segs;
1237 
1238 	/*
1239 	 * Remove indirect pages, this only happens when using indirect
1240 	 * descriptors but not persistent grants
1241 	 */
1242 	if (!list_empty(&rinfo->indirect_pages)) {
1243 		struct page *indirect_page, *n;
1244 
1245 		BUG_ON(info->feature_persistent);
1246 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1247 			list_del(&indirect_page->lru);
1248 			__free_page(indirect_page);
1249 		}
1250 	}
1251 
1252 	/* Remove all persistent grants. */
1253 	if (!list_empty(&rinfo->grants)) {
1254 		list_for_each_entry_safe(persistent_gnt, n,
1255 					 &rinfo->grants, node) {
1256 			list_del(&persistent_gnt->node);
1257 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1258 				gnttab_end_foreign_access(persistent_gnt->gref,
1259 							  0, 0UL);
1260 				rinfo->persistent_gnts_c--;
1261 			}
1262 			if (info->feature_persistent)
1263 				__free_page(persistent_gnt->page);
1264 			kfree(persistent_gnt);
1265 		}
1266 	}
1267 	BUG_ON(rinfo->persistent_gnts_c != 0);
1268 
1269 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1270 		/*
1271 		 * Clear persistent grants present in requests already
1272 		 * on the shared ring
1273 		 */
1274 		if (!rinfo->shadow[i].request)
1275 			goto free_shadow;
1276 
1277 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1278 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1279 		       rinfo->shadow[i].req.u.rw.nr_segments;
1280 		for (j = 0; j < segs; j++) {
1281 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1282 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1283 			if (info->feature_persistent)
1284 				__free_page(persistent_gnt->page);
1285 			kfree(persistent_gnt);
1286 		}
1287 
1288 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1289 			/*
1290 			 * If this is not an indirect operation don't try to
1291 			 * free indirect segments
1292 			 */
1293 			goto free_shadow;
1294 
1295 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1296 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1297 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1298 			__free_page(persistent_gnt->page);
1299 			kfree(persistent_gnt);
1300 		}
1301 
1302 free_shadow:
1303 		kfree(rinfo->shadow[i].grants_used);
1304 		rinfo->shadow[i].grants_used = NULL;
1305 		kfree(rinfo->shadow[i].indirect_grants);
1306 		rinfo->shadow[i].indirect_grants = NULL;
1307 		kfree(rinfo->shadow[i].sg);
1308 		rinfo->shadow[i].sg = NULL;
1309 	}
1310 
1311 	/* No more gnttab callback work. */
1312 	gnttab_cancel_free_callback(&rinfo->callback);
1313 
1314 	/* Flush gnttab callback work. Must be done with no locks held. */
1315 	flush_work(&rinfo->work);
1316 
1317 	/* Free resources associated with old device channel. */
1318 	for (i = 0; i < info->nr_ring_pages; i++) {
1319 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1320 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1321 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1322 		}
1323 	}
1324 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1325 	rinfo->ring.sring = NULL;
1326 
1327 	if (rinfo->irq)
1328 		unbind_from_irqhandler(rinfo->irq, rinfo);
1329 	rinfo->evtchn = rinfo->irq = 0;
1330 }
1331 
1332 static void blkif_free(struct blkfront_info *info, int suspend)
1333 {
1334 	unsigned int i;
1335 
1336 	/* Prevent new requests being issued until we fix things up. */
1337 	info->connected = suspend ?
1338 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1339 	/* No more blkif_request(). */
1340 	if (info->rq)
1341 		blk_mq_stop_hw_queues(info->rq);
1342 
1343 	for (i = 0; i < info->nr_rings; i++)
1344 		blkif_free_ring(&info->rinfo[i]);
1345 
1346 	kfree(info->rinfo);
1347 	info->rinfo = NULL;
1348 	info->nr_rings = 0;
1349 }
1350 
1351 struct copy_from_grant {
1352 	const struct blk_shadow *s;
1353 	unsigned int grant_idx;
1354 	unsigned int bvec_offset;
1355 	char *bvec_data;
1356 };
1357 
1358 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1359 				  unsigned int len, void *data)
1360 {
1361 	struct copy_from_grant *info = data;
1362 	char *shared_data;
1363 	/* Convenient aliases */
1364 	const struct blk_shadow *s = info->s;
1365 
1366 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1367 
1368 	memcpy(info->bvec_data + info->bvec_offset,
1369 	       shared_data + offset, len);
1370 
1371 	info->bvec_offset += len;
1372 	info->grant_idx++;
1373 
1374 	kunmap_atomic(shared_data);
1375 }
1376 
1377 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1378 {
1379 	switch (rsp)
1380 	{
1381 	case BLKIF_RSP_OKAY:
1382 		return REQ_DONE;
1383 	case BLKIF_RSP_EOPNOTSUPP:
1384 		return REQ_EOPNOTSUPP;
1385 	case BLKIF_RSP_ERROR:
1386 		/* Fallthrough. */
1387 	default:
1388 		return REQ_ERROR;
1389 	}
1390 }
1391 
1392 /*
1393  * Get the final status of the block request based on two ring response
1394  */
1395 static int blkif_get_final_status(enum blk_req_status s1,
1396 				  enum blk_req_status s2)
1397 {
1398 	BUG_ON(s1 == REQ_WAITING);
1399 	BUG_ON(s2 == REQ_WAITING);
1400 
1401 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1402 		return BLKIF_RSP_ERROR;
1403 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1404 		return BLKIF_RSP_EOPNOTSUPP;
1405 	return BLKIF_RSP_OKAY;
1406 }
1407 
1408 static bool blkif_completion(unsigned long *id,
1409 			     struct blkfront_ring_info *rinfo,
1410 			     struct blkif_response *bret)
1411 {
1412 	int i = 0;
1413 	struct scatterlist *sg;
1414 	int num_sg, num_grant;
1415 	struct blkfront_info *info = rinfo->dev_info;
1416 	struct blk_shadow *s = &rinfo->shadow[*id];
1417 	struct copy_from_grant data = {
1418 		.grant_idx = 0,
1419 	};
1420 
1421 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1422 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1423 
1424 	/* The I/O request may be split in two. */
1425 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1426 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1427 
1428 		/* Keep the status of the current response in shadow. */
1429 		s->status = blkif_rsp_to_req_status(bret->status);
1430 
1431 		/* Wait the second response if not yet here. */
1432 		if (s2->status == REQ_WAITING)
1433 			return 0;
1434 
1435 		bret->status = blkif_get_final_status(s->status,
1436 						      s2->status);
1437 
1438 		/*
1439 		 * All the grants is stored in the first shadow in order
1440 		 * to make the completion code simpler.
1441 		 */
1442 		num_grant += s2->req.u.rw.nr_segments;
1443 
1444 		/*
1445 		 * The two responses may not come in order. Only the
1446 		 * first request will store the scatter-gather list.
1447 		 */
1448 		if (s2->num_sg != 0) {
1449 			/* Update "id" with the ID of the first response. */
1450 			*id = s->associated_id;
1451 			s = s2;
1452 		}
1453 
1454 		/*
1455 		 * We don't need anymore the second request, so recycling
1456 		 * it now.
1457 		 */
1458 		if (add_id_to_freelist(rinfo, s->associated_id))
1459 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1460 			     info->gd->disk_name, s->associated_id);
1461 	}
1462 
1463 	data.s = s;
1464 	num_sg = s->num_sg;
1465 
1466 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1467 		for_each_sg(s->sg, sg, num_sg, i) {
1468 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1469 
1470 			data.bvec_offset = sg->offset;
1471 			data.bvec_data = kmap_atomic(sg_page(sg));
1472 
1473 			gnttab_foreach_grant_in_range(sg_page(sg),
1474 						      sg->offset,
1475 						      sg->length,
1476 						      blkif_copy_from_grant,
1477 						      &data);
1478 
1479 			kunmap_atomic(data.bvec_data);
1480 		}
1481 	}
1482 	/* Add the persistent grant into the list of free grants */
1483 	for (i = 0; i < num_grant; i++) {
1484 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1485 			/*
1486 			 * If the grant is still mapped by the backend (the
1487 			 * backend has chosen to make this grant persistent)
1488 			 * we add it at the head of the list, so it will be
1489 			 * reused first.
1490 			 */
1491 			if (!info->feature_persistent)
1492 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1493 						     s->grants_used[i]->gref);
1494 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1495 			rinfo->persistent_gnts_c++;
1496 		} else {
1497 			/*
1498 			 * If the grant is not mapped by the backend we end the
1499 			 * foreign access and add it to the tail of the list,
1500 			 * so it will not be picked again unless we run out of
1501 			 * persistent grants.
1502 			 */
1503 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1504 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1505 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1506 		}
1507 	}
1508 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1509 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1510 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1511 				if (!info->feature_persistent)
1512 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1513 							     s->indirect_grants[i]->gref);
1514 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1515 				rinfo->persistent_gnts_c++;
1516 			} else {
1517 				struct page *indirect_page;
1518 
1519 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1520 				/*
1521 				 * Add the used indirect page back to the list of
1522 				 * available pages for indirect grefs.
1523 				 */
1524 				if (!info->feature_persistent) {
1525 					indirect_page = s->indirect_grants[i]->page;
1526 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1527 				}
1528 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1529 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1530 			}
1531 		}
1532 	}
1533 
1534 	return 1;
1535 }
1536 
1537 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1538 {
1539 	struct request *req;
1540 	struct blkif_response *bret;
1541 	RING_IDX i, rp;
1542 	unsigned long flags;
1543 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1544 	struct blkfront_info *info = rinfo->dev_info;
1545 	int error;
1546 
1547 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1548 		return IRQ_HANDLED;
1549 
1550 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1551  again:
1552 	rp = rinfo->ring.sring->rsp_prod;
1553 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1554 
1555 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1556 		unsigned long id;
1557 
1558 		bret = RING_GET_RESPONSE(&rinfo->ring, i);
1559 		id   = bret->id;
1560 		/*
1561 		 * The backend has messed up and given us an id that we would
1562 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1563 		 * look in get_id_from_freelist.
1564 		 */
1565 		if (id >= BLK_RING_SIZE(info)) {
1566 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1567 			     info->gd->disk_name, op_name(bret->operation), id);
1568 			/* We can't safely get the 'struct request' as
1569 			 * the id is busted. */
1570 			continue;
1571 		}
1572 		req  = rinfo->shadow[id].request;
1573 
1574 		if (bret->operation != BLKIF_OP_DISCARD) {
1575 			/*
1576 			 * We may need to wait for an extra response if the
1577 			 * I/O request is split in 2
1578 			 */
1579 			if (!blkif_completion(&id, rinfo, bret))
1580 				continue;
1581 		}
1582 
1583 		if (add_id_to_freelist(rinfo, id)) {
1584 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1585 			     info->gd->disk_name, op_name(bret->operation), id);
1586 			continue;
1587 		}
1588 
1589 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1590 		switch (bret->operation) {
1591 		case BLKIF_OP_DISCARD:
1592 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1593 				struct request_queue *rq = info->rq;
1594 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1595 					   info->gd->disk_name, op_name(bret->operation));
1596 				error = -EOPNOTSUPP;
1597 				info->feature_discard = 0;
1598 				info->feature_secdiscard = 0;
1599 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1600 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1601 			}
1602 			blk_mq_complete_request(req, error);
1603 			break;
1604 		case BLKIF_OP_FLUSH_DISKCACHE:
1605 		case BLKIF_OP_WRITE_BARRIER:
1606 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1607 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1608 				       info->gd->disk_name, op_name(bret->operation));
1609 				error = -EOPNOTSUPP;
1610 			}
1611 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1612 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1613 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1614 				       info->gd->disk_name, op_name(bret->operation));
1615 				error = -EOPNOTSUPP;
1616 			}
1617 			if (unlikely(error)) {
1618 				if (error == -EOPNOTSUPP)
1619 					error = 0;
1620 				info->feature_flush = 0;
1621 				xlvbd_flush(info);
1622 			}
1623 			/* fall through */
1624 		case BLKIF_OP_READ:
1625 		case BLKIF_OP_WRITE:
1626 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1627 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1628 					"request: %x\n", bret->status);
1629 
1630 			blk_mq_complete_request(req, error);
1631 			break;
1632 		default:
1633 			BUG();
1634 		}
1635 	}
1636 
1637 	rinfo->ring.rsp_cons = i;
1638 
1639 	if (i != rinfo->ring.req_prod_pvt) {
1640 		int more_to_do;
1641 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1642 		if (more_to_do)
1643 			goto again;
1644 	} else
1645 		rinfo->ring.sring->rsp_event = i + 1;
1646 
1647 	kick_pending_request_queues_locked(rinfo);
1648 
1649 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1650 
1651 	return IRQ_HANDLED;
1652 }
1653 
1654 
1655 static int setup_blkring(struct xenbus_device *dev,
1656 			 struct blkfront_ring_info *rinfo)
1657 {
1658 	struct blkif_sring *sring;
1659 	int err, i;
1660 	struct blkfront_info *info = rinfo->dev_info;
1661 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1662 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1663 
1664 	for (i = 0; i < info->nr_ring_pages; i++)
1665 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1666 
1667 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1668 						       get_order(ring_size));
1669 	if (!sring) {
1670 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1671 		return -ENOMEM;
1672 	}
1673 	SHARED_RING_INIT(sring);
1674 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1675 
1676 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1677 	if (err < 0) {
1678 		free_pages((unsigned long)sring, get_order(ring_size));
1679 		rinfo->ring.sring = NULL;
1680 		goto fail;
1681 	}
1682 	for (i = 0; i < info->nr_ring_pages; i++)
1683 		rinfo->ring_ref[i] = gref[i];
1684 
1685 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1686 	if (err)
1687 		goto fail;
1688 
1689 	err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1690 					"blkif", rinfo);
1691 	if (err <= 0) {
1692 		xenbus_dev_fatal(dev, err,
1693 				 "bind_evtchn_to_irqhandler failed");
1694 		goto fail;
1695 	}
1696 	rinfo->irq = err;
1697 
1698 	return 0;
1699 fail:
1700 	blkif_free(info, 0);
1701 	return err;
1702 }
1703 
1704 /*
1705  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1706  * ring buffer may have multi pages depending on ->nr_ring_pages.
1707  */
1708 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1709 				struct blkfront_ring_info *rinfo, const char *dir)
1710 {
1711 	int err;
1712 	unsigned int i;
1713 	const char *message = NULL;
1714 	struct blkfront_info *info = rinfo->dev_info;
1715 
1716 	if (info->nr_ring_pages == 1) {
1717 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1718 		if (err) {
1719 			message = "writing ring-ref";
1720 			goto abort_transaction;
1721 		}
1722 	} else {
1723 		for (i = 0; i < info->nr_ring_pages; i++) {
1724 			char ring_ref_name[RINGREF_NAME_LEN];
1725 
1726 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1727 			err = xenbus_printf(xbt, dir, ring_ref_name,
1728 					    "%u", rinfo->ring_ref[i]);
1729 			if (err) {
1730 				message = "writing ring-ref";
1731 				goto abort_transaction;
1732 			}
1733 		}
1734 	}
1735 
1736 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1737 	if (err) {
1738 		message = "writing event-channel";
1739 		goto abort_transaction;
1740 	}
1741 
1742 	return 0;
1743 
1744 abort_transaction:
1745 	xenbus_transaction_end(xbt, 1);
1746 	if (message)
1747 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1748 
1749 	return err;
1750 }
1751 
1752 /* Common code used when first setting up, and when resuming. */
1753 static int talk_to_blkback(struct xenbus_device *dev,
1754 			   struct blkfront_info *info)
1755 {
1756 	const char *message = NULL;
1757 	struct xenbus_transaction xbt;
1758 	int err;
1759 	unsigned int i, max_page_order = 0;
1760 	unsigned int ring_page_order = 0;
1761 
1762 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1763 			   "max-ring-page-order", "%u", &max_page_order);
1764 	if (err != 1)
1765 		info->nr_ring_pages = 1;
1766 	else {
1767 		ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1768 		info->nr_ring_pages = 1 << ring_page_order;
1769 	}
1770 
1771 	for (i = 0; i < info->nr_rings; i++) {
1772 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1773 
1774 		/* Create shared ring, alloc event channel. */
1775 		err = setup_blkring(dev, rinfo);
1776 		if (err)
1777 			goto destroy_blkring;
1778 	}
1779 
1780 again:
1781 	err = xenbus_transaction_start(&xbt);
1782 	if (err) {
1783 		xenbus_dev_fatal(dev, err, "starting transaction");
1784 		goto destroy_blkring;
1785 	}
1786 
1787 	if (info->nr_ring_pages > 1) {
1788 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1789 				    ring_page_order);
1790 		if (err) {
1791 			message = "writing ring-page-order";
1792 			goto abort_transaction;
1793 		}
1794 	}
1795 
1796 	/* We already got the number of queues/rings in _probe */
1797 	if (info->nr_rings == 1) {
1798 		err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1799 		if (err)
1800 			goto destroy_blkring;
1801 	} else {
1802 		char *path;
1803 		size_t pathsize;
1804 
1805 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1806 				    info->nr_rings);
1807 		if (err) {
1808 			message = "writing multi-queue-num-queues";
1809 			goto abort_transaction;
1810 		}
1811 
1812 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1813 		path = kmalloc(pathsize, GFP_KERNEL);
1814 		if (!path) {
1815 			err = -ENOMEM;
1816 			message = "ENOMEM while writing ring references";
1817 			goto abort_transaction;
1818 		}
1819 
1820 		for (i = 0; i < info->nr_rings; i++) {
1821 			memset(path, 0, pathsize);
1822 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1823 			err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1824 			if (err) {
1825 				kfree(path);
1826 				goto destroy_blkring;
1827 			}
1828 		}
1829 		kfree(path);
1830 	}
1831 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1832 			    XEN_IO_PROTO_ABI_NATIVE);
1833 	if (err) {
1834 		message = "writing protocol";
1835 		goto abort_transaction;
1836 	}
1837 	err = xenbus_printf(xbt, dev->nodename,
1838 			    "feature-persistent", "%u", 1);
1839 	if (err)
1840 		dev_warn(&dev->dev,
1841 			 "writing persistent grants feature to xenbus");
1842 
1843 	err = xenbus_transaction_end(xbt, 0);
1844 	if (err) {
1845 		if (err == -EAGAIN)
1846 			goto again;
1847 		xenbus_dev_fatal(dev, err, "completing transaction");
1848 		goto destroy_blkring;
1849 	}
1850 
1851 	for (i = 0; i < info->nr_rings; i++) {
1852 		unsigned int j;
1853 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1854 
1855 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1856 			rinfo->shadow[j].req.u.rw.id = j + 1;
1857 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1858 	}
1859 	xenbus_switch_state(dev, XenbusStateInitialised);
1860 
1861 	return 0;
1862 
1863  abort_transaction:
1864 	xenbus_transaction_end(xbt, 1);
1865 	if (message)
1866 		xenbus_dev_fatal(dev, err, "%s", message);
1867  destroy_blkring:
1868 	blkif_free(info, 0);
1869 
1870 	kfree(info);
1871 	dev_set_drvdata(&dev->dev, NULL);
1872 
1873 	return err;
1874 }
1875 
1876 static int negotiate_mq(struct blkfront_info *info)
1877 {
1878 	unsigned int backend_max_queues = 0;
1879 	int err;
1880 	unsigned int i;
1881 
1882 	BUG_ON(info->nr_rings);
1883 
1884 	/* Check if backend supports multiple queues. */
1885 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1886 			   "multi-queue-max-queues", "%u", &backend_max_queues);
1887 	if (err < 0)
1888 		backend_max_queues = 1;
1889 
1890 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1891 	/* We need at least one ring. */
1892 	if (!info->nr_rings)
1893 		info->nr_rings = 1;
1894 
1895 	info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1896 	if (!info->rinfo) {
1897 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1898 		return -ENOMEM;
1899 	}
1900 
1901 	for (i = 0; i < info->nr_rings; i++) {
1902 		struct blkfront_ring_info *rinfo;
1903 
1904 		rinfo = &info->rinfo[i];
1905 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1906 		INIT_LIST_HEAD(&rinfo->grants);
1907 		rinfo->dev_info = info;
1908 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1909 		spin_lock_init(&rinfo->ring_lock);
1910 	}
1911 	return 0;
1912 }
1913 /**
1914  * Entry point to this code when a new device is created.  Allocate the basic
1915  * structures and the ring buffer for communication with the backend, and
1916  * inform the backend of the appropriate details for those.  Switch to
1917  * Initialised state.
1918  */
1919 static int blkfront_probe(struct xenbus_device *dev,
1920 			  const struct xenbus_device_id *id)
1921 {
1922 	int err, vdevice;
1923 	struct blkfront_info *info;
1924 
1925 	/* FIXME: Use dynamic device id if this is not set. */
1926 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1927 			   "virtual-device", "%i", &vdevice);
1928 	if (err != 1) {
1929 		/* go looking in the extended area instead */
1930 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1931 				   "%i", &vdevice);
1932 		if (err != 1) {
1933 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1934 			return err;
1935 		}
1936 	}
1937 
1938 	if (xen_hvm_domain()) {
1939 		char *type;
1940 		int len;
1941 		/* no unplug has been done: do not hook devices != xen vbds */
1942 		if (xen_has_pv_and_legacy_disk_devices()) {
1943 			int major;
1944 
1945 			if (!VDEV_IS_EXTENDED(vdevice))
1946 				major = BLKIF_MAJOR(vdevice);
1947 			else
1948 				major = XENVBD_MAJOR;
1949 
1950 			if (major != XENVBD_MAJOR) {
1951 				printk(KERN_INFO
1952 						"%s: HVM does not support vbd %d as xen block device\n",
1953 						__func__, vdevice);
1954 				return -ENODEV;
1955 			}
1956 		}
1957 		/* do not create a PV cdrom device if we are an HVM guest */
1958 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1959 		if (IS_ERR(type))
1960 			return -ENODEV;
1961 		if (strncmp(type, "cdrom", 5) == 0) {
1962 			kfree(type);
1963 			return -ENODEV;
1964 		}
1965 		kfree(type);
1966 	}
1967 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1968 	if (!info) {
1969 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1970 		return -ENOMEM;
1971 	}
1972 
1973 	info->xbdev = dev;
1974 	err = negotiate_mq(info);
1975 	if (err) {
1976 		kfree(info);
1977 		return err;
1978 	}
1979 
1980 	mutex_init(&info->mutex);
1981 	info->vdevice = vdevice;
1982 	info->connected = BLKIF_STATE_DISCONNECTED;
1983 
1984 	/* Front end dir is a number, which is used as the id. */
1985 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1986 	dev_set_drvdata(&dev->dev, info);
1987 
1988 	return 0;
1989 }
1990 
1991 static void split_bio_end(struct bio *bio)
1992 {
1993 	struct split_bio *split_bio = bio->bi_private;
1994 
1995 	if (atomic_dec_and_test(&split_bio->pending)) {
1996 		split_bio->bio->bi_phys_segments = 0;
1997 		split_bio->bio->bi_error = bio->bi_error;
1998 		bio_endio(split_bio->bio);
1999 		kfree(split_bio);
2000 	}
2001 	bio_put(bio);
2002 }
2003 
2004 static int blkif_recover(struct blkfront_info *info)
2005 {
2006 	unsigned int i, r_index;
2007 	struct request *req, *n;
2008 	struct blk_shadow *copy;
2009 	int rc;
2010 	struct bio *bio, *cloned_bio;
2011 	struct bio_list bio_list, merge_bio;
2012 	unsigned int segs, offset;
2013 	int pending, size;
2014 	struct split_bio *split_bio;
2015 	struct list_head requests;
2016 
2017 	blkfront_gather_backend_features(info);
2018 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2019 	blk_queue_max_segments(info->rq, segs);
2020 	bio_list_init(&bio_list);
2021 	INIT_LIST_HEAD(&requests);
2022 
2023 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2024 		struct blkfront_ring_info *rinfo;
2025 
2026 		rinfo = &info->rinfo[r_index];
2027 		/* Stage 1: Make a safe copy of the shadow state. */
2028 		copy = kmemdup(rinfo->shadow, sizeof(rinfo->shadow),
2029 			       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
2030 		if (!copy)
2031 			return -ENOMEM;
2032 
2033 		/* Stage 2: Set up free list. */
2034 		memset(&rinfo->shadow, 0, sizeof(rinfo->shadow));
2035 		for (i = 0; i < BLK_RING_SIZE(info); i++)
2036 			rinfo->shadow[i].req.u.rw.id = i+1;
2037 		rinfo->shadow_free = rinfo->ring.req_prod_pvt;
2038 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
2039 
2040 		rc = blkfront_setup_indirect(rinfo);
2041 		if (rc) {
2042 			kfree(copy);
2043 			return rc;
2044 		}
2045 
2046 		for (i = 0; i < BLK_RING_SIZE(info); i++) {
2047 			/* Not in use? */
2048 			if (!copy[i].request)
2049 				continue;
2050 
2051 			/*
2052 			 * Get the bios in the request so we can re-queue them.
2053 			 */
2054 			if (copy[i].request->cmd_flags &
2055 			    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
2056 				/*
2057 				 * Flush operations don't contain bios, so
2058 				 * we need to requeue the whole request
2059 				 */
2060 				list_add(&copy[i].request->queuelist, &requests);
2061 				continue;
2062 			}
2063 			merge_bio.head = copy[i].request->bio;
2064 			merge_bio.tail = copy[i].request->biotail;
2065 			bio_list_merge(&bio_list, &merge_bio);
2066 			copy[i].request->bio = NULL;
2067 			blk_end_request_all(copy[i].request, 0);
2068 		}
2069 
2070 		kfree(copy);
2071 	}
2072 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2073 
2074 	/* Now safe for us to use the shared ring */
2075 	info->connected = BLKIF_STATE_CONNECTED;
2076 
2077 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2078 		struct blkfront_ring_info *rinfo;
2079 
2080 		rinfo = &info->rinfo[r_index];
2081 		/* Kick any other new requests queued since we resumed */
2082 		kick_pending_request_queues(rinfo);
2083 	}
2084 
2085 	list_for_each_entry_safe(req, n, &requests, queuelist) {
2086 		/* Requeue pending requests (flush or discard) */
2087 		list_del_init(&req->queuelist);
2088 		BUG_ON(req->nr_phys_segments > segs);
2089 		blk_mq_requeue_request(req);
2090 	}
2091 	blk_mq_kick_requeue_list(info->rq);
2092 
2093 	while ((bio = bio_list_pop(&bio_list)) != NULL) {
2094 		/* Traverse the list of pending bios and re-queue them */
2095 		if (bio_segments(bio) > segs) {
2096 			/*
2097 			 * This bio has more segments than what we can
2098 			 * handle, we have to split it.
2099 			 */
2100 			pending = (bio_segments(bio) + segs - 1) / segs;
2101 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2102 			BUG_ON(split_bio == NULL);
2103 			atomic_set(&split_bio->pending, pending);
2104 			split_bio->bio = bio;
2105 			for (i = 0; i < pending; i++) {
2106 				offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2107 				size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2108 					   (unsigned int)bio_sectors(bio) - offset);
2109 				cloned_bio = bio_clone(bio, GFP_NOIO);
2110 				BUG_ON(cloned_bio == NULL);
2111 				bio_trim(cloned_bio, offset, size);
2112 				cloned_bio->bi_private = split_bio;
2113 				cloned_bio->bi_end_io = split_bio_end;
2114 				submit_bio(cloned_bio->bi_rw, cloned_bio);
2115 			}
2116 			/*
2117 			 * Now we have to wait for all those smaller bios to
2118 			 * end, so we can also end the "parent" bio.
2119 			 */
2120 			continue;
2121 		}
2122 		/* We don't need to split this bio */
2123 		submit_bio(bio->bi_rw, bio);
2124 	}
2125 
2126 	return 0;
2127 }
2128 
2129 /**
2130  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2131  * driver restart.  We tear down our blkif structure and recreate it, but
2132  * leave the device-layer structures intact so that this is transparent to the
2133  * rest of the kernel.
2134  */
2135 static int blkfront_resume(struct xenbus_device *dev)
2136 {
2137 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2138 	int err = 0;
2139 
2140 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2141 
2142 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2143 
2144 	err = negotiate_mq(info);
2145 	if (err)
2146 		return err;
2147 
2148 	err = talk_to_blkback(dev, info);
2149 
2150 	/*
2151 	 * We have to wait for the backend to switch to
2152 	 * connected state, since we want to read which
2153 	 * features it supports.
2154 	 */
2155 
2156 	return err;
2157 }
2158 
2159 static void blkfront_closing(struct blkfront_info *info)
2160 {
2161 	struct xenbus_device *xbdev = info->xbdev;
2162 	struct block_device *bdev = NULL;
2163 
2164 	mutex_lock(&info->mutex);
2165 
2166 	if (xbdev->state == XenbusStateClosing) {
2167 		mutex_unlock(&info->mutex);
2168 		return;
2169 	}
2170 
2171 	if (info->gd)
2172 		bdev = bdget_disk(info->gd, 0);
2173 
2174 	mutex_unlock(&info->mutex);
2175 
2176 	if (!bdev) {
2177 		xenbus_frontend_closed(xbdev);
2178 		return;
2179 	}
2180 
2181 	mutex_lock(&bdev->bd_mutex);
2182 
2183 	if (bdev->bd_openers) {
2184 		xenbus_dev_error(xbdev, -EBUSY,
2185 				 "Device in use; refusing to close");
2186 		xenbus_switch_state(xbdev, XenbusStateClosing);
2187 	} else {
2188 		xlvbd_release_gendisk(info);
2189 		xenbus_frontend_closed(xbdev);
2190 	}
2191 
2192 	mutex_unlock(&bdev->bd_mutex);
2193 	bdput(bdev);
2194 }
2195 
2196 static void blkfront_setup_discard(struct blkfront_info *info)
2197 {
2198 	int err;
2199 	unsigned int discard_granularity;
2200 	unsigned int discard_alignment;
2201 	unsigned int discard_secure;
2202 
2203 	info->feature_discard = 1;
2204 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2205 		"discard-granularity", "%u", &discard_granularity,
2206 		"discard-alignment", "%u", &discard_alignment,
2207 		NULL);
2208 	if (!err) {
2209 		info->discard_granularity = discard_granularity;
2210 		info->discard_alignment = discard_alignment;
2211 	}
2212 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2213 		    "discard-secure", "%d", &discard_secure,
2214 		    NULL);
2215 	if (!err)
2216 		info->feature_secdiscard = !!discard_secure;
2217 }
2218 
2219 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2220 {
2221 	unsigned int psegs, grants;
2222 	int err, i;
2223 	struct blkfront_info *info = rinfo->dev_info;
2224 
2225 	if (info->max_indirect_segments == 0) {
2226 		if (!HAS_EXTRA_REQ)
2227 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2228 		else {
2229 			/*
2230 			 * When an extra req is required, the maximum
2231 			 * grants supported is related to the size of the
2232 			 * Linux block segment.
2233 			 */
2234 			grants = GRANTS_PER_PSEG;
2235 		}
2236 	}
2237 	else
2238 		grants = info->max_indirect_segments;
2239 	psegs = grants / GRANTS_PER_PSEG;
2240 
2241 	err = fill_grant_buffer(rinfo,
2242 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2243 	if (err)
2244 		goto out_of_memory;
2245 
2246 	if (!info->feature_persistent && info->max_indirect_segments) {
2247 		/*
2248 		 * We are using indirect descriptors but not persistent
2249 		 * grants, we need to allocate a set of pages that can be
2250 		 * used for mapping indirect grefs
2251 		 */
2252 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2253 
2254 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2255 		for (i = 0; i < num; i++) {
2256 			struct page *indirect_page = alloc_page(GFP_NOIO);
2257 			if (!indirect_page)
2258 				goto out_of_memory;
2259 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2260 		}
2261 	}
2262 
2263 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2264 		rinfo->shadow[i].grants_used = kzalloc(
2265 			sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2266 			GFP_NOIO);
2267 		rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2268 		if (info->max_indirect_segments)
2269 			rinfo->shadow[i].indirect_grants = kzalloc(
2270 				sizeof(rinfo->shadow[i].indirect_grants[0]) *
2271 				INDIRECT_GREFS(grants),
2272 				GFP_NOIO);
2273 		if ((rinfo->shadow[i].grants_used == NULL) ||
2274 			(rinfo->shadow[i].sg == NULL) ||
2275 		     (info->max_indirect_segments &&
2276 		     (rinfo->shadow[i].indirect_grants == NULL)))
2277 			goto out_of_memory;
2278 		sg_init_table(rinfo->shadow[i].sg, psegs);
2279 	}
2280 
2281 
2282 	return 0;
2283 
2284 out_of_memory:
2285 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2286 		kfree(rinfo->shadow[i].grants_used);
2287 		rinfo->shadow[i].grants_used = NULL;
2288 		kfree(rinfo->shadow[i].sg);
2289 		rinfo->shadow[i].sg = NULL;
2290 		kfree(rinfo->shadow[i].indirect_grants);
2291 		rinfo->shadow[i].indirect_grants = NULL;
2292 	}
2293 	if (!list_empty(&rinfo->indirect_pages)) {
2294 		struct page *indirect_page, *n;
2295 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2296 			list_del(&indirect_page->lru);
2297 			__free_page(indirect_page);
2298 		}
2299 	}
2300 	return -ENOMEM;
2301 }
2302 
2303 /*
2304  * Gather all backend feature-*
2305  */
2306 static void blkfront_gather_backend_features(struct blkfront_info *info)
2307 {
2308 	int err;
2309 	int barrier, flush, discard, persistent;
2310 	unsigned int indirect_segments;
2311 
2312 	info->feature_flush = 0;
2313 
2314 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2315 			"feature-barrier", "%d", &barrier,
2316 			NULL);
2317 
2318 	/*
2319 	 * If there's no "feature-barrier" defined, then it means
2320 	 * we're dealing with a very old backend which writes
2321 	 * synchronously; nothing to do.
2322 	 *
2323 	 * If there are barriers, then we use flush.
2324 	 */
2325 	if (!err && barrier)
2326 		info->feature_flush = REQ_FLUSH | REQ_FUA;
2327 	/*
2328 	 * And if there is "feature-flush-cache" use that above
2329 	 * barriers.
2330 	 */
2331 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2332 			"feature-flush-cache", "%d", &flush,
2333 			NULL);
2334 
2335 	if (!err && flush)
2336 		info->feature_flush = REQ_FLUSH;
2337 
2338 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2339 			"feature-discard", "%d", &discard,
2340 			NULL);
2341 
2342 	if (!err && discard)
2343 		blkfront_setup_discard(info);
2344 
2345 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2346 			"feature-persistent", "%u", &persistent,
2347 			NULL);
2348 	if (err)
2349 		info->feature_persistent = 0;
2350 	else
2351 		info->feature_persistent = persistent;
2352 
2353 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2354 			    "feature-max-indirect-segments", "%u", &indirect_segments,
2355 			    NULL);
2356 	if (err)
2357 		info->max_indirect_segments = 0;
2358 	else
2359 		info->max_indirect_segments = min(indirect_segments,
2360 						  xen_blkif_max_segments);
2361 }
2362 
2363 /*
2364  * Invoked when the backend is finally 'ready' (and has told produced
2365  * the details about the physical device - #sectors, size, etc).
2366  */
2367 static void blkfront_connect(struct blkfront_info *info)
2368 {
2369 	unsigned long long sectors;
2370 	unsigned long sector_size;
2371 	unsigned int physical_sector_size;
2372 	unsigned int binfo;
2373 	int err, i;
2374 
2375 	switch (info->connected) {
2376 	case BLKIF_STATE_CONNECTED:
2377 		/*
2378 		 * Potentially, the back-end may be signalling
2379 		 * a capacity change; update the capacity.
2380 		 */
2381 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2382 				   "sectors", "%Lu", &sectors);
2383 		if (XENBUS_EXIST_ERR(err))
2384 			return;
2385 		printk(KERN_INFO "Setting capacity to %Lu\n",
2386 		       sectors);
2387 		set_capacity(info->gd, sectors);
2388 		revalidate_disk(info->gd);
2389 
2390 		return;
2391 	case BLKIF_STATE_SUSPENDED:
2392 		/*
2393 		 * If we are recovering from suspension, we need to wait
2394 		 * for the backend to announce it's features before
2395 		 * reconnecting, at least we need to know if the backend
2396 		 * supports indirect descriptors, and how many.
2397 		 */
2398 		blkif_recover(info);
2399 		return;
2400 
2401 	default:
2402 		break;
2403 	}
2404 
2405 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2406 		__func__, info->xbdev->otherend);
2407 
2408 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2409 			    "sectors", "%llu", &sectors,
2410 			    "info", "%u", &binfo,
2411 			    "sector-size", "%lu", &sector_size,
2412 			    NULL);
2413 	if (err) {
2414 		xenbus_dev_fatal(info->xbdev, err,
2415 				 "reading backend fields at %s",
2416 				 info->xbdev->otherend);
2417 		return;
2418 	}
2419 
2420 	/*
2421 	 * physcial-sector-size is a newer field, so old backends may not
2422 	 * provide this. Assume physical sector size to be the same as
2423 	 * sector_size in that case.
2424 	 */
2425 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2426 			   "physical-sector-size", "%u", &physical_sector_size);
2427 	if (err != 1)
2428 		physical_sector_size = sector_size;
2429 
2430 	blkfront_gather_backend_features(info);
2431 	for (i = 0; i < info->nr_rings; i++) {
2432 		err = blkfront_setup_indirect(&info->rinfo[i]);
2433 		if (err) {
2434 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2435 					 info->xbdev->otherend);
2436 			blkif_free(info, 0);
2437 			break;
2438 		}
2439 	}
2440 
2441 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2442 				  physical_sector_size);
2443 	if (err) {
2444 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2445 				 info->xbdev->otherend);
2446 		return;
2447 	}
2448 
2449 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2450 
2451 	/* Kick pending requests. */
2452 	info->connected = BLKIF_STATE_CONNECTED;
2453 	for (i = 0; i < info->nr_rings; i++)
2454 		kick_pending_request_queues(&info->rinfo[i]);
2455 
2456 	add_disk(info->gd);
2457 
2458 	info->is_ready = 1;
2459 }
2460 
2461 /**
2462  * Callback received when the backend's state changes.
2463  */
2464 static void blkback_changed(struct xenbus_device *dev,
2465 			    enum xenbus_state backend_state)
2466 {
2467 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2468 
2469 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2470 
2471 	switch (backend_state) {
2472 	case XenbusStateInitWait:
2473 		if (dev->state != XenbusStateInitialising)
2474 			break;
2475 		if (talk_to_blkback(dev, info))
2476 			break;
2477 	case XenbusStateInitialising:
2478 	case XenbusStateInitialised:
2479 	case XenbusStateReconfiguring:
2480 	case XenbusStateReconfigured:
2481 	case XenbusStateUnknown:
2482 		break;
2483 
2484 	case XenbusStateConnected:
2485 		if (dev->state != XenbusStateInitialised) {
2486 			if (talk_to_blkback(dev, info))
2487 				break;
2488 		}
2489 		blkfront_connect(info);
2490 		break;
2491 
2492 	case XenbusStateClosed:
2493 		if (dev->state == XenbusStateClosed)
2494 			break;
2495 		/* Missed the backend's Closing state -- fallthrough */
2496 	case XenbusStateClosing:
2497 		if (info)
2498 			blkfront_closing(info);
2499 		break;
2500 	}
2501 }
2502 
2503 static int blkfront_remove(struct xenbus_device *xbdev)
2504 {
2505 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2506 	struct block_device *bdev = NULL;
2507 	struct gendisk *disk;
2508 
2509 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2510 
2511 	blkif_free(info, 0);
2512 
2513 	mutex_lock(&info->mutex);
2514 
2515 	disk = info->gd;
2516 	if (disk)
2517 		bdev = bdget_disk(disk, 0);
2518 
2519 	info->xbdev = NULL;
2520 	mutex_unlock(&info->mutex);
2521 
2522 	if (!bdev) {
2523 		kfree(info);
2524 		return 0;
2525 	}
2526 
2527 	/*
2528 	 * The xbdev was removed before we reached the Closed
2529 	 * state. See if it's safe to remove the disk. If the bdev
2530 	 * isn't closed yet, we let release take care of it.
2531 	 */
2532 
2533 	mutex_lock(&bdev->bd_mutex);
2534 	info = disk->private_data;
2535 
2536 	dev_warn(disk_to_dev(disk),
2537 		 "%s was hot-unplugged, %d stale handles\n",
2538 		 xbdev->nodename, bdev->bd_openers);
2539 
2540 	if (info && !bdev->bd_openers) {
2541 		xlvbd_release_gendisk(info);
2542 		disk->private_data = NULL;
2543 		kfree(info);
2544 	}
2545 
2546 	mutex_unlock(&bdev->bd_mutex);
2547 	bdput(bdev);
2548 
2549 	return 0;
2550 }
2551 
2552 static int blkfront_is_ready(struct xenbus_device *dev)
2553 {
2554 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2555 
2556 	return info->is_ready && info->xbdev;
2557 }
2558 
2559 static int blkif_open(struct block_device *bdev, fmode_t mode)
2560 {
2561 	struct gendisk *disk = bdev->bd_disk;
2562 	struct blkfront_info *info;
2563 	int err = 0;
2564 
2565 	mutex_lock(&blkfront_mutex);
2566 
2567 	info = disk->private_data;
2568 	if (!info) {
2569 		/* xbdev gone */
2570 		err = -ERESTARTSYS;
2571 		goto out;
2572 	}
2573 
2574 	mutex_lock(&info->mutex);
2575 
2576 	if (!info->gd)
2577 		/* xbdev is closed */
2578 		err = -ERESTARTSYS;
2579 
2580 	mutex_unlock(&info->mutex);
2581 
2582 out:
2583 	mutex_unlock(&blkfront_mutex);
2584 	return err;
2585 }
2586 
2587 static void blkif_release(struct gendisk *disk, fmode_t mode)
2588 {
2589 	struct blkfront_info *info = disk->private_data;
2590 	struct block_device *bdev;
2591 	struct xenbus_device *xbdev;
2592 
2593 	mutex_lock(&blkfront_mutex);
2594 
2595 	bdev = bdget_disk(disk, 0);
2596 
2597 	if (!bdev) {
2598 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2599 		goto out_mutex;
2600 	}
2601 	if (bdev->bd_openers)
2602 		goto out;
2603 
2604 	/*
2605 	 * Check if we have been instructed to close. We will have
2606 	 * deferred this request, because the bdev was still open.
2607 	 */
2608 
2609 	mutex_lock(&info->mutex);
2610 	xbdev = info->xbdev;
2611 
2612 	if (xbdev && xbdev->state == XenbusStateClosing) {
2613 		/* pending switch to state closed */
2614 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2615 		xlvbd_release_gendisk(info);
2616 		xenbus_frontend_closed(info->xbdev);
2617  	}
2618 
2619 	mutex_unlock(&info->mutex);
2620 
2621 	if (!xbdev) {
2622 		/* sudden device removal */
2623 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2624 		xlvbd_release_gendisk(info);
2625 		disk->private_data = NULL;
2626 		kfree(info);
2627 	}
2628 
2629 out:
2630 	bdput(bdev);
2631 out_mutex:
2632 	mutex_unlock(&blkfront_mutex);
2633 }
2634 
2635 static const struct block_device_operations xlvbd_block_fops =
2636 {
2637 	.owner = THIS_MODULE,
2638 	.open = blkif_open,
2639 	.release = blkif_release,
2640 	.getgeo = blkif_getgeo,
2641 	.ioctl = blkif_ioctl,
2642 };
2643 
2644 
2645 static const struct xenbus_device_id blkfront_ids[] = {
2646 	{ "vbd" },
2647 	{ "" }
2648 };
2649 
2650 static struct xenbus_driver blkfront_driver = {
2651 	.ids  = blkfront_ids,
2652 	.probe = blkfront_probe,
2653 	.remove = blkfront_remove,
2654 	.resume = blkfront_resume,
2655 	.otherend_changed = blkback_changed,
2656 	.is_ready = blkfront_is_ready,
2657 };
2658 
2659 static int __init xlblk_init(void)
2660 {
2661 	int ret;
2662 	int nr_cpus = num_online_cpus();
2663 
2664 	if (!xen_domain())
2665 		return -ENODEV;
2666 
2667 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2668 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2669 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2670 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2671 	}
2672 
2673 	if (xen_blkif_max_queues > nr_cpus) {
2674 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2675 			xen_blkif_max_queues, nr_cpus);
2676 		xen_blkif_max_queues = nr_cpus;
2677 	}
2678 
2679 	if (!xen_has_pv_disk_devices())
2680 		return -ENODEV;
2681 
2682 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2683 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2684 		       XENVBD_MAJOR, DEV_NAME);
2685 		return -ENODEV;
2686 	}
2687 
2688 	ret = xenbus_register_frontend(&blkfront_driver);
2689 	if (ret) {
2690 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2691 		return ret;
2692 	}
2693 
2694 	return 0;
2695 }
2696 module_init(xlblk_init);
2697 
2698 
2699 static void __exit xlblk_exit(void)
2700 {
2701 	xenbus_unregister_driver(&blkfront_driver);
2702 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2703 	kfree(minors);
2704 }
2705 module_exit(xlblk_exit);
2706 
2707 MODULE_DESCRIPTION("Xen virtual block device frontend");
2708 MODULE_LICENSE("GPL");
2709 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2710 MODULE_ALIAS("xen:vbd");
2711 MODULE_ALIAS("xenblk");
2712