xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision c62d3cd0ddd629606a3830aa22e9dcc6c2a0d3bf)
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 blkif_req {
114 	blk_status_t	error;
115 };
116 
117 static inline struct blkif_req *blkif_req(struct request *rq)
118 {
119 	return blk_mq_rq_to_pdu(rq);
120 }
121 
122 static DEFINE_MUTEX(blkfront_mutex);
123 static const struct block_device_operations xlvbd_block_fops;
124 
125 /*
126  * Maximum number of segments in indirect requests, the actual value used by
127  * the frontend driver is the minimum of this value and the value provided
128  * by the backend driver.
129  */
130 
131 static unsigned int xen_blkif_max_segments = 32;
132 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
133 MODULE_PARM_DESC(max_indirect_segments,
134 		 "Maximum amount of segments in indirect requests (default is 32)");
135 
136 static unsigned int xen_blkif_max_queues = 4;
137 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
138 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
139 
140 /*
141  * Maximum order of pages to be used for the shared ring between front and
142  * backend, 4KB page granularity is used.
143  */
144 static unsigned int xen_blkif_max_ring_order;
145 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
146 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
147 
148 #define BLK_RING_SIZE(info)	\
149 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
150 
151 #define BLK_MAX_RING_SIZE	\
152 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
153 
154 /*
155  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
156  * characters are enough. Define to 20 to keep consistent with backend.
157  */
158 #define RINGREF_NAME_LEN (20)
159 /*
160  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
161  */
162 #define QUEUE_NAME_LEN (17)
163 
164 /*
165  *  Per-ring info.
166  *  Every blkfront device can associate with one or more blkfront_ring_info,
167  *  depending on how many hardware queues/rings to be used.
168  */
169 struct blkfront_ring_info {
170 	/* Lock to protect data in every ring buffer. */
171 	spinlock_t ring_lock;
172 	struct blkif_front_ring ring;
173 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
174 	unsigned int evtchn, irq;
175 	struct work_struct work;
176 	struct gnttab_free_callback callback;
177 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
178 	struct list_head indirect_pages;
179 	struct list_head grants;
180 	unsigned int persistent_gnts_c;
181 	unsigned long shadow_free;
182 	struct blkfront_info *dev_info;
183 };
184 
185 /*
186  * We have one of these per vbd, whether ide, scsi or 'other'.  They
187  * hang in private_data off the gendisk structure. We may end up
188  * putting all kinds of interesting stuff here :-)
189  */
190 struct blkfront_info
191 {
192 	struct mutex mutex;
193 	struct xenbus_device *xbdev;
194 	struct gendisk *gd;
195 	u16 sector_size;
196 	unsigned int physical_sector_size;
197 	int vdevice;
198 	blkif_vdev_t handle;
199 	enum blkif_state connected;
200 	/* Number of pages per ring buffer. */
201 	unsigned int nr_ring_pages;
202 	struct request_queue *rq;
203 	unsigned int feature_flush:1;
204 	unsigned int feature_fua:1;
205 	unsigned int feature_discard:1;
206 	unsigned int feature_secdiscard:1;
207 	unsigned int feature_persistent:1;
208 	unsigned int discard_granularity;
209 	unsigned int discard_alignment;
210 	/* Number of 4KB segments handled */
211 	unsigned int max_indirect_segments;
212 	int is_ready;
213 	struct blk_mq_tag_set tag_set;
214 	struct blkfront_ring_info *rinfo;
215 	unsigned int nr_rings;
216 	/* Save uncomplete reqs and bios for migration. */
217 	struct list_head requests;
218 	struct bio_list bio_list;
219 };
220 
221 static unsigned int nr_minors;
222 static unsigned long *minors;
223 static DEFINE_SPINLOCK(minor_lock);
224 
225 #define GRANT_INVALID_REF	0
226 
227 #define PARTS_PER_DISK		16
228 #define PARTS_PER_EXT_DISK      256
229 
230 #define BLKIF_MAJOR(dev) ((dev)>>8)
231 #define BLKIF_MINOR(dev) ((dev) & 0xff)
232 
233 #define EXT_SHIFT 28
234 #define EXTENDED (1<<EXT_SHIFT)
235 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
236 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
237 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
238 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
239 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
240 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
241 
242 #define DEV_NAME	"xvd"	/* name in /dev */
243 
244 /*
245  * Grants are always the same size as a Xen page (i.e 4KB).
246  * A physical segment is always the same size as a Linux page.
247  * Number of grants per physical segment
248  */
249 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
250 
251 #define GRANTS_PER_INDIRECT_FRAME \
252 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
253 
254 #define INDIRECT_GREFS(_grants)		\
255 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
256 
257 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
258 static void blkfront_gather_backend_features(struct blkfront_info *info);
259 static int negotiate_mq(struct blkfront_info *info);
260 
261 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
262 {
263 	unsigned long free = rinfo->shadow_free;
264 
265 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
266 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
267 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
268 	return free;
269 }
270 
271 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
272 			      unsigned long id)
273 {
274 	if (rinfo->shadow[id].req.u.rw.id != id)
275 		return -EINVAL;
276 	if (rinfo->shadow[id].request == NULL)
277 		return -EINVAL;
278 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
279 	rinfo->shadow[id].request = NULL;
280 	rinfo->shadow_free = id;
281 	return 0;
282 }
283 
284 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
285 {
286 	struct blkfront_info *info = rinfo->dev_info;
287 	struct page *granted_page;
288 	struct grant *gnt_list_entry, *n;
289 	int i = 0;
290 
291 	while (i < num) {
292 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
293 		if (!gnt_list_entry)
294 			goto out_of_memory;
295 
296 		if (info->feature_persistent) {
297 			granted_page = alloc_page(GFP_NOIO);
298 			if (!granted_page) {
299 				kfree(gnt_list_entry);
300 				goto out_of_memory;
301 			}
302 			gnt_list_entry->page = granted_page;
303 		}
304 
305 		gnt_list_entry->gref = GRANT_INVALID_REF;
306 		list_add(&gnt_list_entry->node, &rinfo->grants);
307 		i++;
308 	}
309 
310 	return 0;
311 
312 out_of_memory:
313 	list_for_each_entry_safe(gnt_list_entry, n,
314 	                         &rinfo->grants, node) {
315 		list_del(&gnt_list_entry->node);
316 		if (info->feature_persistent)
317 			__free_page(gnt_list_entry->page);
318 		kfree(gnt_list_entry);
319 		i--;
320 	}
321 	BUG_ON(i != 0);
322 	return -ENOMEM;
323 }
324 
325 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
326 {
327 	struct grant *gnt_list_entry;
328 
329 	BUG_ON(list_empty(&rinfo->grants));
330 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
331 					  node);
332 	list_del(&gnt_list_entry->node);
333 
334 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
335 		rinfo->persistent_gnts_c--;
336 
337 	return gnt_list_entry;
338 }
339 
340 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
341 					const struct blkfront_info *info)
342 {
343 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
344 						 info->xbdev->otherend_id,
345 						 gnt_list_entry->page,
346 						 0);
347 }
348 
349 static struct grant *get_grant(grant_ref_t *gref_head,
350 			       unsigned long gfn,
351 			       struct blkfront_ring_info *rinfo)
352 {
353 	struct grant *gnt_list_entry = get_free_grant(rinfo);
354 	struct blkfront_info *info = rinfo->dev_info;
355 
356 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
357 		return gnt_list_entry;
358 
359 	/* Assign a gref to this page */
360 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
361 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
362 	if (info->feature_persistent)
363 		grant_foreign_access(gnt_list_entry, info);
364 	else {
365 		/* Grant access to the GFN passed by the caller */
366 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
367 						info->xbdev->otherend_id,
368 						gfn, 0);
369 	}
370 
371 	return gnt_list_entry;
372 }
373 
374 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
375 					struct blkfront_ring_info *rinfo)
376 {
377 	struct grant *gnt_list_entry = get_free_grant(rinfo);
378 	struct blkfront_info *info = rinfo->dev_info;
379 
380 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
381 		return gnt_list_entry;
382 
383 	/* Assign a gref to this page */
384 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
385 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
386 	if (!info->feature_persistent) {
387 		struct page *indirect_page;
388 
389 		/* Fetch a pre-allocated page to use for indirect grefs */
390 		BUG_ON(list_empty(&rinfo->indirect_pages));
391 		indirect_page = list_first_entry(&rinfo->indirect_pages,
392 						 struct page, lru);
393 		list_del(&indirect_page->lru);
394 		gnt_list_entry->page = indirect_page;
395 	}
396 	grant_foreign_access(gnt_list_entry, info);
397 
398 	return gnt_list_entry;
399 }
400 
401 static const char *op_name(int op)
402 {
403 	static const char *const names[] = {
404 		[BLKIF_OP_READ] = "read",
405 		[BLKIF_OP_WRITE] = "write",
406 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
407 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
408 		[BLKIF_OP_DISCARD] = "discard" };
409 
410 	if (op < 0 || op >= ARRAY_SIZE(names))
411 		return "unknown";
412 
413 	if (!names[op])
414 		return "reserved";
415 
416 	return names[op];
417 }
418 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
419 {
420 	unsigned int end = minor + nr;
421 	int rc;
422 
423 	if (end > nr_minors) {
424 		unsigned long *bitmap, *old;
425 
426 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
427 				 GFP_KERNEL);
428 		if (bitmap == NULL)
429 			return -ENOMEM;
430 
431 		spin_lock(&minor_lock);
432 		if (end > nr_minors) {
433 			old = minors;
434 			memcpy(bitmap, minors,
435 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
436 			minors = bitmap;
437 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
438 		} else
439 			old = bitmap;
440 		spin_unlock(&minor_lock);
441 		kfree(old);
442 	}
443 
444 	spin_lock(&minor_lock);
445 	if (find_next_bit(minors, end, minor) >= end) {
446 		bitmap_set(minors, minor, nr);
447 		rc = 0;
448 	} else
449 		rc = -EBUSY;
450 	spin_unlock(&minor_lock);
451 
452 	return rc;
453 }
454 
455 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
456 {
457 	unsigned int end = minor + nr;
458 
459 	BUG_ON(end > nr_minors);
460 	spin_lock(&minor_lock);
461 	bitmap_clear(minors,  minor, nr);
462 	spin_unlock(&minor_lock);
463 }
464 
465 static void blkif_restart_queue_callback(void *arg)
466 {
467 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
468 	schedule_work(&rinfo->work);
469 }
470 
471 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
472 {
473 	/* We don't have real geometry info, but let's at least return
474 	   values consistent with the size of the device */
475 	sector_t nsect = get_capacity(bd->bd_disk);
476 	sector_t cylinders = nsect;
477 
478 	hg->heads = 0xff;
479 	hg->sectors = 0x3f;
480 	sector_div(cylinders, hg->heads * hg->sectors);
481 	hg->cylinders = cylinders;
482 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
483 		hg->cylinders = 0xffff;
484 	return 0;
485 }
486 
487 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
488 		       unsigned command, unsigned long argument)
489 {
490 	struct blkfront_info *info = bdev->bd_disk->private_data;
491 	int i;
492 
493 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
494 		command, (long)argument);
495 
496 	switch (command) {
497 	case CDROMMULTISESSION:
498 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
499 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
500 			if (put_user(0, (char __user *)(argument + i)))
501 				return -EFAULT;
502 		return 0;
503 
504 	case CDROM_GET_CAPABILITY: {
505 		struct gendisk *gd = info->gd;
506 		if (gd->flags & GENHD_FL_CD)
507 			return 0;
508 		return -EINVAL;
509 	}
510 
511 	default:
512 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
513 		  command);*/
514 		return -EINVAL; /* same return as native Linux */
515 	}
516 
517 	return 0;
518 }
519 
520 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
521 					    struct request *req,
522 					    struct blkif_request **ring_req)
523 {
524 	unsigned long id;
525 
526 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
527 	rinfo->ring.req_prod_pvt++;
528 
529 	id = get_id_from_freelist(rinfo);
530 	rinfo->shadow[id].request = req;
531 	rinfo->shadow[id].status = REQ_WAITING;
532 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
533 
534 	(*ring_req)->u.rw.id = id;
535 
536 	return id;
537 }
538 
539 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
540 {
541 	struct blkfront_info *info = rinfo->dev_info;
542 	struct blkif_request *ring_req;
543 	unsigned long id;
544 
545 	/* Fill out a communications ring structure. */
546 	id = blkif_ring_get_request(rinfo, req, &ring_req);
547 
548 	ring_req->operation = BLKIF_OP_DISCARD;
549 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
550 	ring_req->u.discard.id = id;
551 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
552 	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
553 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
554 	else
555 		ring_req->u.discard.flag = 0;
556 
557 	/* Keep a private copy so we can reissue requests when recovering. */
558 	rinfo->shadow[id].req = *ring_req;
559 
560 	return 0;
561 }
562 
563 struct setup_rw_req {
564 	unsigned int grant_idx;
565 	struct blkif_request_segment *segments;
566 	struct blkfront_ring_info *rinfo;
567 	struct blkif_request *ring_req;
568 	grant_ref_t gref_head;
569 	unsigned int id;
570 	/* Only used when persistent grant is used and it's a read request */
571 	bool need_copy;
572 	unsigned int bvec_off;
573 	char *bvec_data;
574 
575 	bool require_extra_req;
576 	struct blkif_request *extra_ring_req;
577 };
578 
579 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
580 				     unsigned int len, void *data)
581 {
582 	struct setup_rw_req *setup = data;
583 	int n, ref;
584 	struct grant *gnt_list_entry;
585 	unsigned int fsect, lsect;
586 	/* Convenient aliases */
587 	unsigned int grant_idx = setup->grant_idx;
588 	struct blkif_request *ring_req = setup->ring_req;
589 	struct blkfront_ring_info *rinfo = setup->rinfo;
590 	/*
591 	 * We always use the shadow of the first request to store the list
592 	 * of grant associated to the block I/O request. This made the
593 	 * completion more easy to handle even if the block I/O request is
594 	 * split.
595 	 */
596 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
597 
598 	if (unlikely(setup->require_extra_req &&
599 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
600 		/*
601 		 * We are using the second request, setup grant_idx
602 		 * to be the index of the segment array.
603 		 */
604 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
605 		ring_req = setup->extra_ring_req;
606 	}
607 
608 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
609 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
610 		if (setup->segments)
611 			kunmap_atomic(setup->segments);
612 
613 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
614 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
615 		shadow->indirect_grants[n] = gnt_list_entry;
616 		setup->segments = kmap_atomic(gnt_list_entry->page);
617 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
618 	}
619 
620 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
621 	ref = gnt_list_entry->gref;
622 	/*
623 	 * All the grants are stored in the shadow of the first
624 	 * request. Therefore we have to use the global index.
625 	 */
626 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
627 
628 	if (setup->need_copy) {
629 		void *shared_data;
630 
631 		shared_data = kmap_atomic(gnt_list_entry->page);
632 		/*
633 		 * this does not wipe data stored outside the
634 		 * range sg->offset..sg->offset+sg->length.
635 		 * Therefore, blkback *could* see data from
636 		 * previous requests. This is OK as long as
637 		 * persistent grants are shared with just one
638 		 * domain. It may need refactoring if this
639 		 * changes
640 		 */
641 		memcpy(shared_data + offset,
642 		       setup->bvec_data + setup->bvec_off,
643 		       len);
644 
645 		kunmap_atomic(shared_data);
646 		setup->bvec_off += len;
647 	}
648 
649 	fsect = offset >> 9;
650 	lsect = fsect + (len >> 9) - 1;
651 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
652 		ring_req->u.rw.seg[grant_idx] =
653 			(struct blkif_request_segment) {
654 				.gref       = ref,
655 				.first_sect = fsect,
656 				.last_sect  = lsect };
657 	} else {
658 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
659 			(struct blkif_request_segment) {
660 				.gref       = ref,
661 				.first_sect = fsect,
662 				.last_sect  = lsect };
663 	}
664 
665 	(setup->grant_idx)++;
666 }
667 
668 static void blkif_setup_extra_req(struct blkif_request *first,
669 				  struct blkif_request *second)
670 {
671 	uint16_t nr_segments = first->u.rw.nr_segments;
672 
673 	/*
674 	 * The second request is only present when the first request uses
675 	 * all its segments. It's always the continuity of the first one.
676 	 */
677 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
678 
679 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
680 	second->u.rw.sector_number = first->u.rw.sector_number +
681 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
682 
683 	second->u.rw.handle = first->u.rw.handle;
684 	second->operation = first->operation;
685 }
686 
687 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
688 {
689 	struct blkfront_info *info = rinfo->dev_info;
690 	struct blkif_request *ring_req, *extra_ring_req = NULL;
691 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
692 	bool require_extra_req = false;
693 	int i;
694 	struct setup_rw_req setup = {
695 		.grant_idx = 0,
696 		.segments = NULL,
697 		.rinfo = rinfo,
698 		.need_copy = rq_data_dir(req) && info->feature_persistent,
699 	};
700 
701 	/*
702 	 * Used to store if we are able to queue the request by just using
703 	 * existing persistent grants, or if we have to get new grants,
704 	 * as there are not sufficiently many free.
705 	 */
706 	bool new_persistent_gnts = false;
707 	struct scatterlist *sg;
708 	int num_sg, max_grefs, num_grant;
709 
710 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
711 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
712 		/*
713 		 * If we are using indirect segments we need to account
714 		 * for the indirect grefs used in the request.
715 		 */
716 		max_grefs += INDIRECT_GREFS(max_grefs);
717 
718 	/* Check if we have enough persistent grants to allocate a requests */
719 	if (rinfo->persistent_gnts_c < max_grefs) {
720 		new_persistent_gnts = true;
721 
722 		if (gnttab_alloc_grant_references(
723 		    max_grefs - rinfo->persistent_gnts_c,
724 		    &setup.gref_head) < 0) {
725 			gnttab_request_free_callback(
726 				&rinfo->callback,
727 				blkif_restart_queue_callback,
728 				rinfo,
729 				max_grefs - rinfo->persistent_gnts_c);
730 			return 1;
731 		}
732 	}
733 
734 	/* Fill out a communications ring structure. */
735 	id = blkif_ring_get_request(rinfo, req, &ring_req);
736 
737 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
738 	num_grant = 0;
739 	/* Calculate the number of grant used */
740 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
741 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
742 
743 	require_extra_req = info->max_indirect_segments == 0 &&
744 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
745 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
746 
747 	rinfo->shadow[id].num_sg = num_sg;
748 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
749 	    likely(!require_extra_req)) {
750 		/*
751 		 * The indirect operation can only be a BLKIF_OP_READ or
752 		 * BLKIF_OP_WRITE
753 		 */
754 		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
755 		ring_req->operation = BLKIF_OP_INDIRECT;
756 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
757 			BLKIF_OP_WRITE : BLKIF_OP_READ;
758 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
759 		ring_req->u.indirect.handle = info->handle;
760 		ring_req->u.indirect.nr_segments = num_grant;
761 	} else {
762 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
763 		ring_req->u.rw.handle = info->handle;
764 		ring_req->operation = rq_data_dir(req) ?
765 			BLKIF_OP_WRITE : BLKIF_OP_READ;
766 		if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
767 			/*
768 			 * Ideally we can do an unordered flush-to-disk.
769 			 * In case the backend onlysupports barriers, use that.
770 			 * A barrier request a superset of FUA, so we can
771 			 * implement it the same way.  (It's also a FLUSH+FUA,
772 			 * since it is guaranteed ordered WRT previous writes.)
773 			 */
774 			if (info->feature_flush && info->feature_fua)
775 				ring_req->operation =
776 					BLKIF_OP_WRITE_BARRIER;
777 			else if (info->feature_flush)
778 				ring_req->operation =
779 					BLKIF_OP_FLUSH_DISKCACHE;
780 			else
781 				ring_req->operation = 0;
782 		}
783 		ring_req->u.rw.nr_segments = num_grant;
784 		if (unlikely(require_extra_req)) {
785 			extra_id = blkif_ring_get_request(rinfo, req,
786 							  &extra_ring_req);
787 			/*
788 			 * Only the first request contains the scatter-gather
789 			 * list.
790 			 */
791 			rinfo->shadow[extra_id].num_sg = 0;
792 
793 			blkif_setup_extra_req(ring_req, extra_ring_req);
794 
795 			/* Link the 2 requests together */
796 			rinfo->shadow[extra_id].associated_id = id;
797 			rinfo->shadow[id].associated_id = extra_id;
798 		}
799 	}
800 
801 	setup.ring_req = ring_req;
802 	setup.id = id;
803 
804 	setup.require_extra_req = require_extra_req;
805 	if (unlikely(require_extra_req))
806 		setup.extra_ring_req = extra_ring_req;
807 
808 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
809 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
810 
811 		if (setup.need_copy) {
812 			setup.bvec_off = sg->offset;
813 			setup.bvec_data = kmap_atomic(sg_page(sg));
814 		}
815 
816 		gnttab_foreach_grant_in_range(sg_page(sg),
817 					      sg->offset,
818 					      sg->length,
819 					      blkif_setup_rw_req_grant,
820 					      &setup);
821 
822 		if (setup.need_copy)
823 			kunmap_atomic(setup.bvec_data);
824 	}
825 	if (setup.segments)
826 		kunmap_atomic(setup.segments);
827 
828 	/* Keep a private copy so we can reissue requests when recovering. */
829 	rinfo->shadow[id].req = *ring_req;
830 	if (unlikely(require_extra_req))
831 		rinfo->shadow[extra_id].req = *extra_ring_req;
832 
833 	if (new_persistent_gnts)
834 		gnttab_free_grant_references(setup.gref_head);
835 
836 	return 0;
837 }
838 
839 /*
840  * Generate a Xen blkfront IO request from a blk layer request.  Reads
841  * and writes are handled as expected.
842  *
843  * @req: a request struct
844  */
845 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
846 {
847 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
848 		return 1;
849 
850 	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
851 		     req_op(req) == REQ_OP_SECURE_ERASE))
852 		return blkif_queue_discard_req(req, rinfo);
853 	else
854 		return blkif_queue_rw_req(req, rinfo);
855 }
856 
857 static inline void flush_requests(struct blkfront_ring_info *rinfo)
858 {
859 	int notify;
860 
861 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
862 
863 	if (notify)
864 		notify_remote_via_irq(rinfo->irq);
865 }
866 
867 static inline bool blkif_request_flush_invalid(struct request *req,
868 					       struct blkfront_info *info)
869 {
870 	return (blk_rq_is_passthrough(req) ||
871 		((req_op(req) == REQ_OP_FLUSH) &&
872 		 !info->feature_flush) ||
873 		((req->cmd_flags & REQ_FUA) &&
874 		 !info->feature_fua));
875 }
876 
877 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
878 			  const struct blk_mq_queue_data *qd)
879 {
880 	unsigned long flags;
881 	int qid = hctx->queue_num;
882 	struct blkfront_info *info = hctx->queue->queuedata;
883 	struct blkfront_ring_info *rinfo = NULL;
884 
885 	BUG_ON(info->nr_rings <= qid);
886 	rinfo = &info->rinfo[qid];
887 	blk_mq_start_request(qd->rq);
888 	spin_lock_irqsave(&rinfo->ring_lock, flags);
889 	if (RING_FULL(&rinfo->ring))
890 		goto out_busy;
891 
892 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
893 		goto out_err;
894 
895 	if (blkif_queue_request(qd->rq, rinfo))
896 		goto out_busy;
897 
898 	flush_requests(rinfo);
899 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
900 	return BLK_STS_OK;
901 
902 out_err:
903 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
904 	return BLK_STS_IOERR;
905 
906 out_busy:
907 	blk_mq_stop_hw_queue(hctx);
908 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
909 	return BLK_STS_DEV_RESOURCE;
910 }
911 
912 static void blkif_complete_rq(struct request *rq)
913 {
914 	blk_mq_end_request(rq, blkif_req(rq)->error);
915 }
916 
917 static const struct blk_mq_ops blkfront_mq_ops = {
918 	.queue_rq = blkif_queue_rq,
919 	.complete = blkif_complete_rq,
920 };
921 
922 static void blkif_set_queue_limits(struct blkfront_info *info)
923 {
924 	struct request_queue *rq = info->rq;
925 	struct gendisk *gd = info->gd;
926 	unsigned int segments = info->max_indirect_segments ? :
927 				BLKIF_MAX_SEGMENTS_PER_REQUEST;
928 
929 	blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
930 
931 	if (info->feature_discard) {
932 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
933 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
934 		rq->limits.discard_granularity = info->discard_granularity;
935 		rq->limits.discard_alignment = info->discard_alignment;
936 		if (info->feature_secdiscard)
937 			blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
938 	}
939 
940 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
941 	blk_queue_logical_block_size(rq, info->sector_size);
942 	blk_queue_physical_block_size(rq, info->physical_sector_size);
943 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
944 
945 	/* Each segment in a request is up to an aligned page in size. */
946 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
947 	blk_queue_max_segment_size(rq, PAGE_SIZE);
948 
949 	/* Ensure a merged request will fit in a single I/O ring slot. */
950 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
951 
952 	/* Make sure buffer addresses are sector-aligned. */
953 	blk_queue_dma_alignment(rq, 511);
954 }
955 
956 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
957 				unsigned int physical_sector_size)
958 {
959 	struct request_queue *rq;
960 	struct blkfront_info *info = gd->private_data;
961 
962 	memset(&info->tag_set, 0, sizeof(info->tag_set));
963 	info->tag_set.ops = &blkfront_mq_ops;
964 	info->tag_set.nr_hw_queues = info->nr_rings;
965 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
966 		/*
967 		 * When indirect descriptior is not supported, the I/O request
968 		 * will be split between multiple request in the ring.
969 		 * To avoid problems when sending the request, divide by
970 		 * 2 the depth of the queue.
971 		 */
972 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
973 	} else
974 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
975 	info->tag_set.numa_node = NUMA_NO_NODE;
976 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
977 	info->tag_set.cmd_size = sizeof(struct blkif_req);
978 	info->tag_set.driver_data = info;
979 
980 	if (blk_mq_alloc_tag_set(&info->tag_set))
981 		return -EINVAL;
982 	rq = blk_mq_init_queue(&info->tag_set);
983 	if (IS_ERR(rq)) {
984 		blk_mq_free_tag_set(&info->tag_set);
985 		return PTR_ERR(rq);
986 	}
987 
988 	rq->queuedata = info;
989 	info->rq = gd->queue = rq;
990 	info->gd = gd;
991 	info->sector_size = sector_size;
992 	info->physical_sector_size = physical_sector_size;
993 	blkif_set_queue_limits(info);
994 
995 	return 0;
996 }
997 
998 static const char *flush_info(struct blkfront_info *info)
999 {
1000 	if (info->feature_flush && info->feature_fua)
1001 		return "barrier: enabled;";
1002 	else if (info->feature_flush)
1003 		return "flush diskcache: enabled;";
1004 	else
1005 		return "barrier or flush: disabled;";
1006 }
1007 
1008 static void xlvbd_flush(struct blkfront_info *info)
1009 {
1010 	blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1011 			      info->feature_fua ? true : false);
1012 	pr_info("blkfront: %s: %s %s %s %s %s\n",
1013 		info->gd->disk_name, flush_info(info),
1014 		"persistent grants:", info->feature_persistent ?
1015 		"enabled;" : "disabled;", "indirect descriptors:",
1016 		info->max_indirect_segments ? "enabled;" : "disabled;");
1017 }
1018 
1019 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1020 {
1021 	int major;
1022 	major = BLKIF_MAJOR(vdevice);
1023 	*minor = BLKIF_MINOR(vdevice);
1024 	switch (major) {
1025 		case XEN_IDE0_MAJOR:
1026 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1027 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1028 				EMULATED_HD_DISK_MINOR_OFFSET;
1029 			break;
1030 		case XEN_IDE1_MAJOR:
1031 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1032 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1033 				EMULATED_HD_DISK_MINOR_OFFSET;
1034 			break;
1035 		case XEN_SCSI_DISK0_MAJOR:
1036 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1037 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1038 			break;
1039 		case XEN_SCSI_DISK1_MAJOR:
1040 		case XEN_SCSI_DISK2_MAJOR:
1041 		case XEN_SCSI_DISK3_MAJOR:
1042 		case XEN_SCSI_DISK4_MAJOR:
1043 		case XEN_SCSI_DISK5_MAJOR:
1044 		case XEN_SCSI_DISK6_MAJOR:
1045 		case XEN_SCSI_DISK7_MAJOR:
1046 			*offset = (*minor / PARTS_PER_DISK) +
1047 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1048 				EMULATED_SD_DISK_NAME_OFFSET;
1049 			*minor = *minor +
1050 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1051 				EMULATED_SD_DISK_MINOR_OFFSET;
1052 			break;
1053 		case XEN_SCSI_DISK8_MAJOR:
1054 		case XEN_SCSI_DISK9_MAJOR:
1055 		case XEN_SCSI_DISK10_MAJOR:
1056 		case XEN_SCSI_DISK11_MAJOR:
1057 		case XEN_SCSI_DISK12_MAJOR:
1058 		case XEN_SCSI_DISK13_MAJOR:
1059 		case XEN_SCSI_DISK14_MAJOR:
1060 		case XEN_SCSI_DISK15_MAJOR:
1061 			*offset = (*minor / PARTS_PER_DISK) +
1062 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1063 				EMULATED_SD_DISK_NAME_OFFSET;
1064 			*minor = *minor +
1065 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1066 				EMULATED_SD_DISK_MINOR_OFFSET;
1067 			break;
1068 		case XENVBD_MAJOR:
1069 			*offset = *minor / PARTS_PER_DISK;
1070 			break;
1071 		default:
1072 			printk(KERN_WARNING "blkfront: your disk configuration is "
1073 					"incorrect, please use an xvd device instead\n");
1074 			return -ENODEV;
1075 	}
1076 	return 0;
1077 }
1078 
1079 static char *encode_disk_name(char *ptr, unsigned int n)
1080 {
1081 	if (n >= 26)
1082 		ptr = encode_disk_name(ptr, n / 26 - 1);
1083 	*ptr = 'a' + n % 26;
1084 	return ptr + 1;
1085 }
1086 
1087 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1088 			       struct blkfront_info *info,
1089 			       u16 vdisk_info, u16 sector_size,
1090 			       unsigned int physical_sector_size)
1091 {
1092 	struct gendisk *gd;
1093 	int nr_minors = 1;
1094 	int err;
1095 	unsigned int offset;
1096 	int minor;
1097 	int nr_parts;
1098 	char *ptr;
1099 
1100 	BUG_ON(info->gd != NULL);
1101 	BUG_ON(info->rq != NULL);
1102 
1103 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1104 		/* this is above the extended range; something is wrong */
1105 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1106 		return -ENODEV;
1107 	}
1108 
1109 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1110 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1111 		if (err)
1112 			return err;
1113  		nr_parts = PARTS_PER_DISK;
1114 	} else {
1115 		minor = BLKIF_MINOR_EXT(info->vdevice);
1116 		nr_parts = PARTS_PER_EXT_DISK;
1117 		offset = minor / nr_parts;
1118 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1119 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1120 					"emulated IDE disks,\n\t choose an xvd device name"
1121 					"from xvde on\n", info->vdevice);
1122 	}
1123 	if (minor >> MINORBITS) {
1124 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1125 			info->vdevice, minor);
1126 		return -ENODEV;
1127 	}
1128 
1129 	if ((minor % nr_parts) == 0)
1130 		nr_minors = nr_parts;
1131 
1132 	err = xlbd_reserve_minors(minor, nr_minors);
1133 	if (err)
1134 		goto out;
1135 	err = -ENODEV;
1136 
1137 	gd = alloc_disk(nr_minors);
1138 	if (gd == NULL)
1139 		goto release;
1140 
1141 	strcpy(gd->disk_name, DEV_NAME);
1142 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1143 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1144 	if (nr_minors > 1)
1145 		*ptr = 0;
1146 	else
1147 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1148 			 "%d", minor & (nr_parts - 1));
1149 
1150 	gd->major = XENVBD_MAJOR;
1151 	gd->first_minor = minor;
1152 	gd->fops = &xlvbd_block_fops;
1153 	gd->private_data = info;
1154 	set_capacity(gd, capacity);
1155 
1156 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1157 		del_gendisk(gd);
1158 		goto release;
1159 	}
1160 
1161 	xlvbd_flush(info);
1162 
1163 	if (vdisk_info & VDISK_READONLY)
1164 		set_disk_ro(gd, 1);
1165 
1166 	if (vdisk_info & VDISK_REMOVABLE)
1167 		gd->flags |= GENHD_FL_REMOVABLE;
1168 
1169 	if (vdisk_info & VDISK_CDROM)
1170 		gd->flags |= GENHD_FL_CD;
1171 
1172 	return 0;
1173 
1174  release:
1175 	xlbd_release_minors(minor, nr_minors);
1176  out:
1177 	return err;
1178 }
1179 
1180 static void xlvbd_release_gendisk(struct blkfront_info *info)
1181 {
1182 	unsigned int minor, nr_minors, i;
1183 
1184 	if (info->rq == NULL)
1185 		return;
1186 
1187 	/* No more blkif_request(). */
1188 	blk_mq_stop_hw_queues(info->rq);
1189 
1190 	for (i = 0; i < info->nr_rings; i++) {
1191 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1192 
1193 		/* No more gnttab callback work. */
1194 		gnttab_cancel_free_callback(&rinfo->callback);
1195 
1196 		/* Flush gnttab callback work. Must be done with no locks held. */
1197 		flush_work(&rinfo->work);
1198 	}
1199 
1200 	del_gendisk(info->gd);
1201 
1202 	minor = info->gd->first_minor;
1203 	nr_minors = info->gd->minors;
1204 	xlbd_release_minors(minor, nr_minors);
1205 
1206 	blk_cleanup_queue(info->rq);
1207 	blk_mq_free_tag_set(&info->tag_set);
1208 	info->rq = NULL;
1209 
1210 	put_disk(info->gd);
1211 	info->gd = NULL;
1212 }
1213 
1214 /* Already hold rinfo->ring_lock. */
1215 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1216 {
1217 	if (!RING_FULL(&rinfo->ring))
1218 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1219 }
1220 
1221 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1222 {
1223 	unsigned long flags;
1224 
1225 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1226 	kick_pending_request_queues_locked(rinfo);
1227 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1228 }
1229 
1230 static void blkif_restart_queue(struct work_struct *work)
1231 {
1232 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1233 
1234 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1235 		kick_pending_request_queues(rinfo);
1236 }
1237 
1238 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1239 {
1240 	struct grant *persistent_gnt, *n;
1241 	struct blkfront_info *info = rinfo->dev_info;
1242 	int i, j, segs;
1243 
1244 	/*
1245 	 * Remove indirect pages, this only happens when using indirect
1246 	 * descriptors but not persistent grants
1247 	 */
1248 	if (!list_empty(&rinfo->indirect_pages)) {
1249 		struct page *indirect_page, *n;
1250 
1251 		BUG_ON(info->feature_persistent);
1252 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1253 			list_del(&indirect_page->lru);
1254 			__free_page(indirect_page);
1255 		}
1256 	}
1257 
1258 	/* Remove all persistent grants. */
1259 	if (!list_empty(&rinfo->grants)) {
1260 		list_for_each_entry_safe(persistent_gnt, n,
1261 					 &rinfo->grants, node) {
1262 			list_del(&persistent_gnt->node);
1263 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1264 				gnttab_end_foreign_access(persistent_gnt->gref,
1265 							  0, 0UL);
1266 				rinfo->persistent_gnts_c--;
1267 			}
1268 			if (info->feature_persistent)
1269 				__free_page(persistent_gnt->page);
1270 			kfree(persistent_gnt);
1271 		}
1272 	}
1273 	BUG_ON(rinfo->persistent_gnts_c != 0);
1274 
1275 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1276 		/*
1277 		 * Clear persistent grants present in requests already
1278 		 * on the shared ring
1279 		 */
1280 		if (!rinfo->shadow[i].request)
1281 			goto free_shadow;
1282 
1283 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1284 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1285 		       rinfo->shadow[i].req.u.rw.nr_segments;
1286 		for (j = 0; j < segs; j++) {
1287 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1288 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1289 			if (info->feature_persistent)
1290 				__free_page(persistent_gnt->page);
1291 			kfree(persistent_gnt);
1292 		}
1293 
1294 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1295 			/*
1296 			 * If this is not an indirect operation don't try to
1297 			 * free indirect segments
1298 			 */
1299 			goto free_shadow;
1300 
1301 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1302 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1303 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1304 			__free_page(persistent_gnt->page);
1305 			kfree(persistent_gnt);
1306 		}
1307 
1308 free_shadow:
1309 		kfree(rinfo->shadow[i].grants_used);
1310 		rinfo->shadow[i].grants_used = NULL;
1311 		kfree(rinfo->shadow[i].indirect_grants);
1312 		rinfo->shadow[i].indirect_grants = NULL;
1313 		kfree(rinfo->shadow[i].sg);
1314 		rinfo->shadow[i].sg = NULL;
1315 	}
1316 
1317 	/* No more gnttab callback work. */
1318 	gnttab_cancel_free_callback(&rinfo->callback);
1319 
1320 	/* Flush gnttab callback work. Must be done with no locks held. */
1321 	flush_work(&rinfo->work);
1322 
1323 	/* Free resources associated with old device channel. */
1324 	for (i = 0; i < info->nr_ring_pages; i++) {
1325 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1326 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1327 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1328 		}
1329 	}
1330 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1331 	rinfo->ring.sring = NULL;
1332 
1333 	if (rinfo->irq)
1334 		unbind_from_irqhandler(rinfo->irq, rinfo);
1335 	rinfo->evtchn = rinfo->irq = 0;
1336 }
1337 
1338 static void blkif_free(struct blkfront_info *info, int suspend)
1339 {
1340 	unsigned int i;
1341 
1342 	/* Prevent new requests being issued until we fix things up. */
1343 	info->connected = suspend ?
1344 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1345 	/* No more blkif_request(). */
1346 	if (info->rq)
1347 		blk_mq_stop_hw_queues(info->rq);
1348 
1349 	for (i = 0; i < info->nr_rings; i++)
1350 		blkif_free_ring(&info->rinfo[i]);
1351 
1352 	kfree(info->rinfo);
1353 	info->rinfo = NULL;
1354 	info->nr_rings = 0;
1355 }
1356 
1357 struct copy_from_grant {
1358 	const struct blk_shadow *s;
1359 	unsigned int grant_idx;
1360 	unsigned int bvec_offset;
1361 	char *bvec_data;
1362 };
1363 
1364 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1365 				  unsigned int len, void *data)
1366 {
1367 	struct copy_from_grant *info = data;
1368 	char *shared_data;
1369 	/* Convenient aliases */
1370 	const struct blk_shadow *s = info->s;
1371 
1372 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1373 
1374 	memcpy(info->bvec_data + info->bvec_offset,
1375 	       shared_data + offset, len);
1376 
1377 	info->bvec_offset += len;
1378 	info->grant_idx++;
1379 
1380 	kunmap_atomic(shared_data);
1381 }
1382 
1383 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1384 {
1385 	switch (rsp)
1386 	{
1387 	case BLKIF_RSP_OKAY:
1388 		return REQ_DONE;
1389 	case BLKIF_RSP_EOPNOTSUPP:
1390 		return REQ_EOPNOTSUPP;
1391 	case BLKIF_RSP_ERROR:
1392 		/* Fallthrough. */
1393 	default:
1394 		return REQ_ERROR;
1395 	}
1396 }
1397 
1398 /*
1399  * Get the final status of the block request based on two ring response
1400  */
1401 static int blkif_get_final_status(enum blk_req_status s1,
1402 				  enum blk_req_status s2)
1403 {
1404 	BUG_ON(s1 == REQ_WAITING);
1405 	BUG_ON(s2 == REQ_WAITING);
1406 
1407 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1408 		return BLKIF_RSP_ERROR;
1409 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1410 		return BLKIF_RSP_EOPNOTSUPP;
1411 	return BLKIF_RSP_OKAY;
1412 }
1413 
1414 static bool blkif_completion(unsigned long *id,
1415 			     struct blkfront_ring_info *rinfo,
1416 			     struct blkif_response *bret)
1417 {
1418 	int i = 0;
1419 	struct scatterlist *sg;
1420 	int num_sg, num_grant;
1421 	struct blkfront_info *info = rinfo->dev_info;
1422 	struct blk_shadow *s = &rinfo->shadow[*id];
1423 	struct copy_from_grant data = {
1424 		.grant_idx = 0,
1425 	};
1426 
1427 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1428 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1429 
1430 	/* The I/O request may be split in two. */
1431 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1432 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1433 
1434 		/* Keep the status of the current response in shadow. */
1435 		s->status = blkif_rsp_to_req_status(bret->status);
1436 
1437 		/* Wait the second response if not yet here. */
1438 		if (s2->status == REQ_WAITING)
1439 			return false;
1440 
1441 		bret->status = blkif_get_final_status(s->status,
1442 						      s2->status);
1443 
1444 		/*
1445 		 * All the grants is stored in the first shadow in order
1446 		 * to make the completion code simpler.
1447 		 */
1448 		num_grant += s2->req.u.rw.nr_segments;
1449 
1450 		/*
1451 		 * The two responses may not come in order. Only the
1452 		 * first request will store the scatter-gather list.
1453 		 */
1454 		if (s2->num_sg != 0) {
1455 			/* Update "id" with the ID of the first response. */
1456 			*id = s->associated_id;
1457 			s = s2;
1458 		}
1459 
1460 		/*
1461 		 * We don't need anymore the second request, so recycling
1462 		 * it now.
1463 		 */
1464 		if (add_id_to_freelist(rinfo, s->associated_id))
1465 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1466 			     info->gd->disk_name, s->associated_id);
1467 	}
1468 
1469 	data.s = s;
1470 	num_sg = s->num_sg;
1471 
1472 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1473 		for_each_sg(s->sg, sg, num_sg, i) {
1474 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1475 
1476 			data.bvec_offset = sg->offset;
1477 			data.bvec_data = kmap_atomic(sg_page(sg));
1478 
1479 			gnttab_foreach_grant_in_range(sg_page(sg),
1480 						      sg->offset,
1481 						      sg->length,
1482 						      blkif_copy_from_grant,
1483 						      &data);
1484 
1485 			kunmap_atomic(data.bvec_data);
1486 		}
1487 	}
1488 	/* Add the persistent grant into the list of free grants */
1489 	for (i = 0; i < num_grant; i++) {
1490 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1491 			/*
1492 			 * If the grant is still mapped by the backend (the
1493 			 * backend has chosen to make this grant persistent)
1494 			 * we add it at the head of the list, so it will be
1495 			 * reused first.
1496 			 */
1497 			if (!info->feature_persistent)
1498 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1499 						     s->grants_used[i]->gref);
1500 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1501 			rinfo->persistent_gnts_c++;
1502 		} else {
1503 			/*
1504 			 * If the grant is not mapped by the backend we end the
1505 			 * foreign access and add it to the tail of the list,
1506 			 * so it will not be picked again unless we run out of
1507 			 * persistent grants.
1508 			 */
1509 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1510 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1511 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1512 		}
1513 	}
1514 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1515 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1516 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1517 				if (!info->feature_persistent)
1518 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1519 							     s->indirect_grants[i]->gref);
1520 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1521 				rinfo->persistent_gnts_c++;
1522 			} else {
1523 				struct page *indirect_page;
1524 
1525 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1526 				/*
1527 				 * Add the used indirect page back to the list of
1528 				 * available pages for indirect grefs.
1529 				 */
1530 				if (!info->feature_persistent) {
1531 					indirect_page = s->indirect_grants[i]->page;
1532 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1533 				}
1534 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1535 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1536 			}
1537 		}
1538 	}
1539 
1540 	return true;
1541 }
1542 
1543 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1544 {
1545 	struct request *req;
1546 	struct blkif_response *bret;
1547 	RING_IDX i, rp;
1548 	unsigned long flags;
1549 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1550 	struct blkfront_info *info = rinfo->dev_info;
1551 
1552 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1553 		return IRQ_HANDLED;
1554 
1555 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1556  again:
1557 	rp = rinfo->ring.sring->rsp_prod;
1558 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1559 
1560 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1561 		unsigned long id;
1562 
1563 		bret = RING_GET_RESPONSE(&rinfo->ring, i);
1564 		id   = bret->id;
1565 		/*
1566 		 * The backend has messed up and given us an id that we would
1567 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1568 		 * look in get_id_from_freelist.
1569 		 */
1570 		if (id >= BLK_RING_SIZE(info)) {
1571 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1572 			     info->gd->disk_name, op_name(bret->operation), id);
1573 			/* We can't safely get the 'struct request' as
1574 			 * the id is busted. */
1575 			continue;
1576 		}
1577 		req  = rinfo->shadow[id].request;
1578 
1579 		if (bret->operation != BLKIF_OP_DISCARD) {
1580 			/*
1581 			 * We may need to wait for an extra response if the
1582 			 * I/O request is split in 2
1583 			 */
1584 			if (!blkif_completion(&id, rinfo, bret))
1585 				continue;
1586 		}
1587 
1588 		if (add_id_to_freelist(rinfo, id)) {
1589 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1590 			     info->gd->disk_name, op_name(bret->operation), id);
1591 			continue;
1592 		}
1593 
1594 		if (bret->status == BLKIF_RSP_OKAY)
1595 			blkif_req(req)->error = BLK_STS_OK;
1596 		else
1597 			blkif_req(req)->error = BLK_STS_IOERR;
1598 
1599 		switch (bret->operation) {
1600 		case BLKIF_OP_DISCARD:
1601 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1602 				struct request_queue *rq = info->rq;
1603 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1604 					   info->gd->disk_name, op_name(bret->operation));
1605 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1606 				info->feature_discard = 0;
1607 				info->feature_secdiscard = 0;
1608 				blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1609 				blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1610 			}
1611 			break;
1612 		case BLKIF_OP_FLUSH_DISKCACHE:
1613 		case BLKIF_OP_WRITE_BARRIER:
1614 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1615 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1616 				       info->gd->disk_name, op_name(bret->operation));
1617 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1618 			}
1619 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1620 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1621 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1622 				       info->gd->disk_name, op_name(bret->operation));
1623 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1624 			}
1625 			if (unlikely(blkif_req(req)->error)) {
1626 				if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1627 					blkif_req(req)->error = BLK_STS_OK;
1628 				info->feature_fua = 0;
1629 				info->feature_flush = 0;
1630 				xlvbd_flush(info);
1631 			}
1632 			/* fall through */
1633 		case BLKIF_OP_READ:
1634 		case BLKIF_OP_WRITE:
1635 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1636 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1637 					"request: %x\n", bret->status);
1638 
1639 			break;
1640 		default:
1641 			BUG();
1642 		}
1643 
1644 		blk_mq_complete_request(req);
1645 	}
1646 
1647 	rinfo->ring.rsp_cons = i;
1648 
1649 	if (i != rinfo->ring.req_prod_pvt) {
1650 		int more_to_do;
1651 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1652 		if (more_to_do)
1653 			goto again;
1654 	} else
1655 		rinfo->ring.sring->rsp_event = i + 1;
1656 
1657 	kick_pending_request_queues_locked(rinfo);
1658 
1659 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1660 
1661 	return IRQ_HANDLED;
1662 }
1663 
1664 
1665 static int setup_blkring(struct xenbus_device *dev,
1666 			 struct blkfront_ring_info *rinfo)
1667 {
1668 	struct blkif_sring *sring;
1669 	int err, i;
1670 	struct blkfront_info *info = rinfo->dev_info;
1671 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1672 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1673 
1674 	for (i = 0; i < info->nr_ring_pages; i++)
1675 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1676 
1677 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1678 						       get_order(ring_size));
1679 	if (!sring) {
1680 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1681 		return -ENOMEM;
1682 	}
1683 	SHARED_RING_INIT(sring);
1684 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1685 
1686 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1687 	if (err < 0) {
1688 		free_pages((unsigned long)sring, get_order(ring_size));
1689 		rinfo->ring.sring = NULL;
1690 		goto fail;
1691 	}
1692 	for (i = 0; i < info->nr_ring_pages; i++)
1693 		rinfo->ring_ref[i] = gref[i];
1694 
1695 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1696 	if (err)
1697 		goto fail;
1698 
1699 	err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1700 					"blkif", rinfo);
1701 	if (err <= 0) {
1702 		xenbus_dev_fatal(dev, err,
1703 				 "bind_evtchn_to_irqhandler failed");
1704 		goto fail;
1705 	}
1706 	rinfo->irq = err;
1707 
1708 	return 0;
1709 fail:
1710 	blkif_free(info, 0);
1711 	return err;
1712 }
1713 
1714 /*
1715  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1716  * ring buffer may have multi pages depending on ->nr_ring_pages.
1717  */
1718 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1719 				struct blkfront_ring_info *rinfo, const char *dir)
1720 {
1721 	int err;
1722 	unsigned int i;
1723 	const char *message = NULL;
1724 	struct blkfront_info *info = rinfo->dev_info;
1725 
1726 	if (info->nr_ring_pages == 1) {
1727 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1728 		if (err) {
1729 			message = "writing ring-ref";
1730 			goto abort_transaction;
1731 		}
1732 	} else {
1733 		for (i = 0; i < info->nr_ring_pages; i++) {
1734 			char ring_ref_name[RINGREF_NAME_LEN];
1735 
1736 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1737 			err = xenbus_printf(xbt, dir, ring_ref_name,
1738 					    "%u", rinfo->ring_ref[i]);
1739 			if (err) {
1740 				message = "writing ring-ref";
1741 				goto abort_transaction;
1742 			}
1743 		}
1744 	}
1745 
1746 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1747 	if (err) {
1748 		message = "writing event-channel";
1749 		goto abort_transaction;
1750 	}
1751 
1752 	return 0;
1753 
1754 abort_transaction:
1755 	xenbus_transaction_end(xbt, 1);
1756 	if (message)
1757 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1758 
1759 	return err;
1760 }
1761 
1762 /* Common code used when first setting up, and when resuming. */
1763 static int talk_to_blkback(struct xenbus_device *dev,
1764 			   struct blkfront_info *info)
1765 {
1766 	const char *message = NULL;
1767 	struct xenbus_transaction xbt;
1768 	int err;
1769 	unsigned int i, max_page_order;
1770 	unsigned int ring_page_order;
1771 
1772 	if (!info)
1773 		return -ENODEV;
1774 
1775 	max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1776 					      "max-ring-page-order", 0);
1777 	ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1778 	info->nr_ring_pages = 1 << ring_page_order;
1779 
1780 	err = negotiate_mq(info);
1781 	if (err)
1782 		goto destroy_blkring;
1783 
1784 	for (i = 0; i < info->nr_rings; i++) {
1785 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1786 
1787 		/* Create shared ring, alloc event channel. */
1788 		err = setup_blkring(dev, rinfo);
1789 		if (err)
1790 			goto destroy_blkring;
1791 	}
1792 
1793 again:
1794 	err = xenbus_transaction_start(&xbt);
1795 	if (err) {
1796 		xenbus_dev_fatal(dev, err, "starting transaction");
1797 		goto destroy_blkring;
1798 	}
1799 
1800 	if (info->nr_ring_pages > 1) {
1801 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1802 				    ring_page_order);
1803 		if (err) {
1804 			message = "writing ring-page-order";
1805 			goto abort_transaction;
1806 		}
1807 	}
1808 
1809 	/* We already got the number of queues/rings in _probe */
1810 	if (info->nr_rings == 1) {
1811 		err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1812 		if (err)
1813 			goto destroy_blkring;
1814 	} else {
1815 		char *path;
1816 		size_t pathsize;
1817 
1818 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1819 				    info->nr_rings);
1820 		if (err) {
1821 			message = "writing multi-queue-num-queues";
1822 			goto abort_transaction;
1823 		}
1824 
1825 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1826 		path = kmalloc(pathsize, GFP_KERNEL);
1827 		if (!path) {
1828 			err = -ENOMEM;
1829 			message = "ENOMEM while writing ring references";
1830 			goto abort_transaction;
1831 		}
1832 
1833 		for (i = 0; i < info->nr_rings; i++) {
1834 			memset(path, 0, pathsize);
1835 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1836 			err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1837 			if (err) {
1838 				kfree(path);
1839 				goto destroy_blkring;
1840 			}
1841 		}
1842 		kfree(path);
1843 	}
1844 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1845 			    XEN_IO_PROTO_ABI_NATIVE);
1846 	if (err) {
1847 		message = "writing protocol";
1848 		goto abort_transaction;
1849 	}
1850 	err = xenbus_printf(xbt, dev->nodename,
1851 			    "feature-persistent", "%u", 1);
1852 	if (err)
1853 		dev_warn(&dev->dev,
1854 			 "writing persistent grants feature to xenbus");
1855 
1856 	err = xenbus_transaction_end(xbt, 0);
1857 	if (err) {
1858 		if (err == -EAGAIN)
1859 			goto again;
1860 		xenbus_dev_fatal(dev, err, "completing transaction");
1861 		goto destroy_blkring;
1862 	}
1863 
1864 	for (i = 0; i < info->nr_rings; i++) {
1865 		unsigned int j;
1866 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1867 
1868 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1869 			rinfo->shadow[j].req.u.rw.id = j + 1;
1870 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1871 	}
1872 	xenbus_switch_state(dev, XenbusStateInitialised);
1873 
1874 	return 0;
1875 
1876  abort_transaction:
1877 	xenbus_transaction_end(xbt, 1);
1878 	if (message)
1879 		xenbus_dev_fatal(dev, err, "%s", message);
1880  destroy_blkring:
1881 	blkif_free(info, 0);
1882 
1883 	kfree(info);
1884 	dev_set_drvdata(&dev->dev, NULL);
1885 
1886 	return err;
1887 }
1888 
1889 static int negotiate_mq(struct blkfront_info *info)
1890 {
1891 	unsigned int backend_max_queues;
1892 	unsigned int i;
1893 
1894 	BUG_ON(info->nr_rings);
1895 
1896 	/* Check if backend supports multiple queues. */
1897 	backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1898 						  "multi-queue-max-queues", 1);
1899 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1900 	/* We need at least one ring. */
1901 	if (!info->nr_rings)
1902 		info->nr_rings = 1;
1903 
1904 	info->rinfo = kcalloc(info->nr_rings,
1905 			      sizeof(struct blkfront_ring_info),
1906 			      GFP_KERNEL);
1907 	if (!info->rinfo) {
1908 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1909 		return -ENOMEM;
1910 	}
1911 
1912 	for (i = 0; i < info->nr_rings; i++) {
1913 		struct blkfront_ring_info *rinfo;
1914 
1915 		rinfo = &info->rinfo[i];
1916 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1917 		INIT_LIST_HEAD(&rinfo->grants);
1918 		rinfo->dev_info = info;
1919 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1920 		spin_lock_init(&rinfo->ring_lock);
1921 	}
1922 	return 0;
1923 }
1924 /**
1925  * Entry point to this code when a new device is created.  Allocate the basic
1926  * structures and the ring buffer for communication with the backend, and
1927  * inform the backend of the appropriate details for those.  Switch to
1928  * Initialised state.
1929  */
1930 static int blkfront_probe(struct xenbus_device *dev,
1931 			  const struct xenbus_device_id *id)
1932 {
1933 	int err, vdevice;
1934 	struct blkfront_info *info;
1935 
1936 	/* FIXME: Use dynamic device id if this is not set. */
1937 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1938 			   "virtual-device", "%i", &vdevice);
1939 	if (err != 1) {
1940 		/* go looking in the extended area instead */
1941 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1942 				   "%i", &vdevice);
1943 		if (err != 1) {
1944 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1945 			return err;
1946 		}
1947 	}
1948 
1949 	if (xen_hvm_domain()) {
1950 		char *type;
1951 		int len;
1952 		/* no unplug has been done: do not hook devices != xen vbds */
1953 		if (xen_has_pv_and_legacy_disk_devices()) {
1954 			int major;
1955 
1956 			if (!VDEV_IS_EXTENDED(vdevice))
1957 				major = BLKIF_MAJOR(vdevice);
1958 			else
1959 				major = XENVBD_MAJOR;
1960 
1961 			if (major != XENVBD_MAJOR) {
1962 				printk(KERN_INFO
1963 						"%s: HVM does not support vbd %d as xen block device\n",
1964 						__func__, vdevice);
1965 				return -ENODEV;
1966 			}
1967 		}
1968 		/* do not create a PV cdrom device if we are an HVM guest */
1969 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1970 		if (IS_ERR(type))
1971 			return -ENODEV;
1972 		if (strncmp(type, "cdrom", 5) == 0) {
1973 			kfree(type);
1974 			return -ENODEV;
1975 		}
1976 		kfree(type);
1977 	}
1978 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1979 	if (!info) {
1980 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1981 		return -ENOMEM;
1982 	}
1983 
1984 	info->xbdev = dev;
1985 
1986 	mutex_init(&info->mutex);
1987 	info->vdevice = vdevice;
1988 	info->connected = BLKIF_STATE_DISCONNECTED;
1989 
1990 	/* Front end dir is a number, which is used as the id. */
1991 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1992 	dev_set_drvdata(&dev->dev, info);
1993 
1994 	return 0;
1995 }
1996 
1997 static int blkif_recover(struct blkfront_info *info)
1998 {
1999 	unsigned int r_index;
2000 	struct request *req, *n;
2001 	int rc;
2002 	struct bio *bio;
2003 	unsigned int segs;
2004 
2005 	blkfront_gather_backend_features(info);
2006 	/* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2007 	blkif_set_queue_limits(info);
2008 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2009 	blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2010 
2011 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2012 		struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2013 
2014 		rc = blkfront_setup_indirect(rinfo);
2015 		if (rc)
2016 			return rc;
2017 	}
2018 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2019 
2020 	/* Now safe for us to use the shared ring */
2021 	info->connected = BLKIF_STATE_CONNECTED;
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 		/* Kick any other new requests queued since we resumed */
2028 		kick_pending_request_queues(rinfo);
2029 	}
2030 
2031 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2032 		/* Requeue pending requests (flush or discard) */
2033 		list_del_init(&req->queuelist);
2034 		BUG_ON(req->nr_phys_segments > segs);
2035 		blk_mq_requeue_request(req, false);
2036 	}
2037 	blk_mq_start_stopped_hw_queues(info->rq, true);
2038 	blk_mq_kick_requeue_list(info->rq);
2039 
2040 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2041 		/* Traverse the list of pending bios and re-queue them */
2042 		submit_bio(bio);
2043 	}
2044 
2045 	return 0;
2046 }
2047 
2048 /**
2049  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2050  * driver restart.  We tear down our blkif structure and recreate it, but
2051  * leave the device-layer structures intact so that this is transparent to the
2052  * rest of the kernel.
2053  */
2054 static int blkfront_resume(struct xenbus_device *dev)
2055 {
2056 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2057 	int err = 0;
2058 	unsigned int i, j;
2059 
2060 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2061 
2062 	bio_list_init(&info->bio_list);
2063 	INIT_LIST_HEAD(&info->requests);
2064 	for (i = 0; i < info->nr_rings; i++) {
2065 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
2066 		struct bio_list merge_bio;
2067 		struct blk_shadow *shadow = rinfo->shadow;
2068 
2069 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2070 			/* Not in use? */
2071 			if (!shadow[j].request)
2072 				continue;
2073 
2074 			/*
2075 			 * Get the bios in the request so we can re-queue them.
2076 			 */
2077 			if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2078 			    req_op(shadow[j].request) == REQ_OP_DISCARD ||
2079 			    req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2080 			    shadow[j].request->cmd_flags & REQ_FUA) {
2081 				/*
2082 				 * Flush operations don't contain bios, so
2083 				 * we need to requeue the whole request
2084 				 *
2085 				 * XXX: but this doesn't make any sense for a
2086 				 * write with the FUA flag set..
2087 				 */
2088 				list_add(&shadow[j].request->queuelist, &info->requests);
2089 				continue;
2090 			}
2091 			merge_bio.head = shadow[j].request->bio;
2092 			merge_bio.tail = shadow[j].request->biotail;
2093 			bio_list_merge(&info->bio_list, &merge_bio);
2094 			shadow[j].request->bio = NULL;
2095 			blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2096 		}
2097 	}
2098 
2099 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2100 
2101 	err = talk_to_blkback(dev, info);
2102 	if (!err)
2103 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2104 
2105 	/*
2106 	 * We have to wait for the backend to switch to
2107 	 * connected state, since we want to read which
2108 	 * features it supports.
2109 	 */
2110 
2111 	return err;
2112 }
2113 
2114 static void blkfront_closing(struct blkfront_info *info)
2115 {
2116 	struct xenbus_device *xbdev = info->xbdev;
2117 	struct block_device *bdev = NULL;
2118 
2119 	mutex_lock(&info->mutex);
2120 
2121 	if (xbdev->state == XenbusStateClosing) {
2122 		mutex_unlock(&info->mutex);
2123 		return;
2124 	}
2125 
2126 	if (info->gd)
2127 		bdev = bdget_disk(info->gd, 0);
2128 
2129 	mutex_unlock(&info->mutex);
2130 
2131 	if (!bdev) {
2132 		xenbus_frontend_closed(xbdev);
2133 		return;
2134 	}
2135 
2136 	mutex_lock(&bdev->bd_mutex);
2137 
2138 	if (bdev->bd_openers) {
2139 		xenbus_dev_error(xbdev, -EBUSY,
2140 				 "Device in use; refusing to close");
2141 		xenbus_switch_state(xbdev, XenbusStateClosing);
2142 	} else {
2143 		xlvbd_release_gendisk(info);
2144 		xenbus_frontend_closed(xbdev);
2145 	}
2146 
2147 	mutex_unlock(&bdev->bd_mutex);
2148 	bdput(bdev);
2149 }
2150 
2151 static void blkfront_setup_discard(struct blkfront_info *info)
2152 {
2153 	int err;
2154 	unsigned int discard_granularity;
2155 	unsigned int discard_alignment;
2156 
2157 	info->feature_discard = 1;
2158 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2159 		"discard-granularity", "%u", &discard_granularity,
2160 		"discard-alignment", "%u", &discard_alignment,
2161 		NULL);
2162 	if (!err) {
2163 		info->discard_granularity = discard_granularity;
2164 		info->discard_alignment = discard_alignment;
2165 	}
2166 	info->feature_secdiscard =
2167 		!!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2168 				       0);
2169 }
2170 
2171 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2172 {
2173 	unsigned int psegs, grants;
2174 	int err, i;
2175 	struct blkfront_info *info = rinfo->dev_info;
2176 
2177 	if (info->max_indirect_segments == 0) {
2178 		if (!HAS_EXTRA_REQ)
2179 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2180 		else {
2181 			/*
2182 			 * When an extra req is required, the maximum
2183 			 * grants supported is related to the size of the
2184 			 * Linux block segment.
2185 			 */
2186 			grants = GRANTS_PER_PSEG;
2187 		}
2188 	}
2189 	else
2190 		grants = info->max_indirect_segments;
2191 	psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2192 
2193 	err = fill_grant_buffer(rinfo,
2194 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2195 	if (err)
2196 		goto out_of_memory;
2197 
2198 	if (!info->feature_persistent && info->max_indirect_segments) {
2199 		/*
2200 		 * We are using indirect descriptors but not persistent
2201 		 * grants, we need to allocate a set of pages that can be
2202 		 * used for mapping indirect grefs
2203 		 */
2204 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2205 
2206 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2207 		for (i = 0; i < num; i++) {
2208 			struct page *indirect_page = alloc_page(GFP_NOIO);
2209 			if (!indirect_page)
2210 				goto out_of_memory;
2211 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2212 		}
2213 	}
2214 
2215 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2216 		rinfo->shadow[i].grants_used =
2217 			kcalloc(grants,
2218 				sizeof(rinfo->shadow[i].grants_used[0]),
2219 				GFP_NOIO);
2220 		rinfo->shadow[i].sg = kcalloc(psegs,
2221 					      sizeof(rinfo->shadow[i].sg[0]),
2222 					      GFP_NOIO);
2223 		if (info->max_indirect_segments)
2224 			rinfo->shadow[i].indirect_grants =
2225 				kcalloc(INDIRECT_GREFS(grants),
2226 					sizeof(rinfo->shadow[i].indirect_grants[0]),
2227 					GFP_NOIO);
2228 		if ((rinfo->shadow[i].grants_used == NULL) ||
2229 			(rinfo->shadow[i].sg == NULL) ||
2230 		     (info->max_indirect_segments &&
2231 		     (rinfo->shadow[i].indirect_grants == NULL)))
2232 			goto out_of_memory;
2233 		sg_init_table(rinfo->shadow[i].sg, psegs);
2234 	}
2235 
2236 
2237 	return 0;
2238 
2239 out_of_memory:
2240 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2241 		kfree(rinfo->shadow[i].grants_used);
2242 		rinfo->shadow[i].grants_used = NULL;
2243 		kfree(rinfo->shadow[i].sg);
2244 		rinfo->shadow[i].sg = NULL;
2245 		kfree(rinfo->shadow[i].indirect_grants);
2246 		rinfo->shadow[i].indirect_grants = NULL;
2247 	}
2248 	if (!list_empty(&rinfo->indirect_pages)) {
2249 		struct page *indirect_page, *n;
2250 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2251 			list_del(&indirect_page->lru);
2252 			__free_page(indirect_page);
2253 		}
2254 	}
2255 	return -ENOMEM;
2256 }
2257 
2258 /*
2259  * Gather all backend feature-*
2260  */
2261 static void blkfront_gather_backend_features(struct blkfront_info *info)
2262 {
2263 	unsigned int indirect_segments;
2264 
2265 	info->feature_flush = 0;
2266 	info->feature_fua = 0;
2267 
2268 	/*
2269 	 * If there's no "feature-barrier" defined, then it means
2270 	 * we're dealing with a very old backend which writes
2271 	 * synchronously; nothing to do.
2272 	 *
2273 	 * If there are barriers, then we use flush.
2274 	 */
2275 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2276 		info->feature_flush = 1;
2277 		info->feature_fua = 1;
2278 	}
2279 
2280 	/*
2281 	 * And if there is "feature-flush-cache" use that above
2282 	 * barriers.
2283 	 */
2284 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2285 				 0)) {
2286 		info->feature_flush = 1;
2287 		info->feature_fua = 0;
2288 	}
2289 
2290 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2291 		blkfront_setup_discard(info);
2292 
2293 	info->feature_persistent =
2294 		!!xenbus_read_unsigned(info->xbdev->otherend,
2295 				       "feature-persistent", 0);
2296 
2297 	indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2298 					"feature-max-indirect-segments", 0);
2299 	if (indirect_segments > xen_blkif_max_segments)
2300 		indirect_segments = xen_blkif_max_segments;
2301 	if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2302 		indirect_segments = 0;
2303 	info->max_indirect_segments = indirect_segments;
2304 }
2305 
2306 /*
2307  * Invoked when the backend is finally 'ready' (and has told produced
2308  * the details about the physical device - #sectors, size, etc).
2309  */
2310 static void blkfront_connect(struct blkfront_info *info)
2311 {
2312 	unsigned long long sectors;
2313 	unsigned long sector_size;
2314 	unsigned int physical_sector_size;
2315 	unsigned int binfo;
2316 	char *envp[] = { "RESIZE=1", NULL };
2317 	int err, i;
2318 
2319 	switch (info->connected) {
2320 	case BLKIF_STATE_CONNECTED:
2321 		/*
2322 		 * Potentially, the back-end may be signalling
2323 		 * a capacity change; update the capacity.
2324 		 */
2325 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2326 				   "sectors", "%Lu", &sectors);
2327 		if (XENBUS_EXIST_ERR(err))
2328 			return;
2329 		printk(KERN_INFO "Setting capacity to %Lu\n",
2330 		       sectors);
2331 		set_capacity(info->gd, sectors);
2332 		revalidate_disk(info->gd);
2333 		kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2334 				   KOBJ_CHANGE, envp);
2335 
2336 		return;
2337 	case BLKIF_STATE_SUSPENDED:
2338 		/*
2339 		 * If we are recovering from suspension, we need to wait
2340 		 * for the backend to announce it's features before
2341 		 * reconnecting, at least we need to know if the backend
2342 		 * supports indirect descriptors, and how many.
2343 		 */
2344 		blkif_recover(info);
2345 		return;
2346 
2347 	default:
2348 		break;
2349 	}
2350 
2351 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2352 		__func__, info->xbdev->otherend);
2353 
2354 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2355 			    "sectors", "%llu", &sectors,
2356 			    "info", "%u", &binfo,
2357 			    "sector-size", "%lu", &sector_size,
2358 			    NULL);
2359 	if (err) {
2360 		xenbus_dev_fatal(info->xbdev, err,
2361 				 "reading backend fields at %s",
2362 				 info->xbdev->otherend);
2363 		return;
2364 	}
2365 
2366 	/*
2367 	 * physcial-sector-size is a newer field, so old backends may not
2368 	 * provide this. Assume physical sector size to be the same as
2369 	 * sector_size in that case.
2370 	 */
2371 	physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2372 						    "physical-sector-size",
2373 						    sector_size);
2374 	blkfront_gather_backend_features(info);
2375 	for (i = 0; i < info->nr_rings; i++) {
2376 		err = blkfront_setup_indirect(&info->rinfo[i]);
2377 		if (err) {
2378 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2379 					 info->xbdev->otherend);
2380 			blkif_free(info, 0);
2381 			break;
2382 		}
2383 	}
2384 
2385 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2386 				  physical_sector_size);
2387 	if (err) {
2388 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2389 				 info->xbdev->otherend);
2390 		goto fail;
2391 	}
2392 
2393 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2394 
2395 	/* Kick pending requests. */
2396 	info->connected = BLKIF_STATE_CONNECTED;
2397 	for (i = 0; i < info->nr_rings; i++)
2398 		kick_pending_request_queues(&info->rinfo[i]);
2399 
2400 	device_add_disk(&info->xbdev->dev, info->gd);
2401 
2402 	info->is_ready = 1;
2403 	return;
2404 
2405 fail:
2406 	blkif_free(info, 0);
2407 	return;
2408 }
2409 
2410 /**
2411  * Callback received when the backend's state changes.
2412  */
2413 static void blkback_changed(struct xenbus_device *dev,
2414 			    enum xenbus_state backend_state)
2415 {
2416 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2417 
2418 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2419 
2420 	switch (backend_state) {
2421 	case XenbusStateInitWait:
2422 		if (dev->state != XenbusStateInitialising)
2423 			break;
2424 		if (talk_to_blkback(dev, info))
2425 			break;
2426 	case XenbusStateInitialising:
2427 	case XenbusStateInitialised:
2428 	case XenbusStateReconfiguring:
2429 	case XenbusStateReconfigured:
2430 	case XenbusStateUnknown:
2431 		break;
2432 
2433 	case XenbusStateConnected:
2434 		/*
2435 		 * talk_to_blkback sets state to XenbusStateInitialised
2436 		 * and blkfront_connect sets it to XenbusStateConnected
2437 		 * (if connection went OK).
2438 		 *
2439 		 * If the backend (or toolstack) decides to poke at backend
2440 		 * state (and re-trigger the watch by setting the state repeatedly
2441 		 * to XenbusStateConnected (4)) we need to deal with this.
2442 		 * This is allowed as this is used to communicate to the guest
2443 		 * that the size of disk has changed!
2444 		 */
2445 		if ((dev->state != XenbusStateInitialised) &&
2446 		    (dev->state != XenbusStateConnected)) {
2447 			if (talk_to_blkback(dev, info))
2448 				break;
2449 		}
2450 
2451 		blkfront_connect(info);
2452 		break;
2453 
2454 	case XenbusStateClosed:
2455 		if (dev->state == XenbusStateClosed)
2456 			break;
2457 		/* fall through */
2458 	case XenbusStateClosing:
2459 		if (info)
2460 			blkfront_closing(info);
2461 		break;
2462 	}
2463 }
2464 
2465 static int blkfront_remove(struct xenbus_device *xbdev)
2466 {
2467 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2468 	struct block_device *bdev = NULL;
2469 	struct gendisk *disk;
2470 
2471 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2472 
2473 	blkif_free(info, 0);
2474 
2475 	mutex_lock(&info->mutex);
2476 
2477 	disk = info->gd;
2478 	if (disk)
2479 		bdev = bdget_disk(disk, 0);
2480 
2481 	info->xbdev = NULL;
2482 	mutex_unlock(&info->mutex);
2483 
2484 	if (!bdev) {
2485 		kfree(info);
2486 		return 0;
2487 	}
2488 
2489 	/*
2490 	 * The xbdev was removed before we reached the Closed
2491 	 * state. See if it's safe to remove the disk. If the bdev
2492 	 * isn't closed yet, we let release take care of it.
2493 	 */
2494 
2495 	mutex_lock(&bdev->bd_mutex);
2496 	info = disk->private_data;
2497 
2498 	dev_warn(disk_to_dev(disk),
2499 		 "%s was hot-unplugged, %d stale handles\n",
2500 		 xbdev->nodename, bdev->bd_openers);
2501 
2502 	if (info && !bdev->bd_openers) {
2503 		xlvbd_release_gendisk(info);
2504 		disk->private_data = NULL;
2505 		kfree(info);
2506 	}
2507 
2508 	mutex_unlock(&bdev->bd_mutex);
2509 	bdput(bdev);
2510 
2511 	return 0;
2512 }
2513 
2514 static int blkfront_is_ready(struct xenbus_device *dev)
2515 {
2516 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2517 
2518 	return info->is_ready && info->xbdev;
2519 }
2520 
2521 static int blkif_open(struct block_device *bdev, fmode_t mode)
2522 {
2523 	struct gendisk *disk = bdev->bd_disk;
2524 	struct blkfront_info *info;
2525 	int err = 0;
2526 
2527 	mutex_lock(&blkfront_mutex);
2528 
2529 	info = disk->private_data;
2530 	if (!info) {
2531 		/* xbdev gone */
2532 		err = -ERESTARTSYS;
2533 		goto out;
2534 	}
2535 
2536 	mutex_lock(&info->mutex);
2537 
2538 	if (!info->gd)
2539 		/* xbdev is closed */
2540 		err = -ERESTARTSYS;
2541 
2542 	mutex_unlock(&info->mutex);
2543 
2544 out:
2545 	mutex_unlock(&blkfront_mutex);
2546 	return err;
2547 }
2548 
2549 static void blkif_release(struct gendisk *disk, fmode_t mode)
2550 {
2551 	struct blkfront_info *info = disk->private_data;
2552 	struct block_device *bdev;
2553 	struct xenbus_device *xbdev;
2554 
2555 	mutex_lock(&blkfront_mutex);
2556 
2557 	bdev = bdget_disk(disk, 0);
2558 
2559 	if (!bdev) {
2560 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2561 		goto out_mutex;
2562 	}
2563 	if (bdev->bd_openers)
2564 		goto out;
2565 
2566 	/*
2567 	 * Check if we have been instructed to close. We will have
2568 	 * deferred this request, because the bdev was still open.
2569 	 */
2570 
2571 	mutex_lock(&info->mutex);
2572 	xbdev = info->xbdev;
2573 
2574 	if (xbdev && xbdev->state == XenbusStateClosing) {
2575 		/* pending switch to state closed */
2576 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2577 		xlvbd_release_gendisk(info);
2578 		xenbus_frontend_closed(info->xbdev);
2579  	}
2580 
2581 	mutex_unlock(&info->mutex);
2582 
2583 	if (!xbdev) {
2584 		/* sudden device removal */
2585 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2586 		xlvbd_release_gendisk(info);
2587 		disk->private_data = NULL;
2588 		kfree(info);
2589 	}
2590 
2591 out:
2592 	bdput(bdev);
2593 out_mutex:
2594 	mutex_unlock(&blkfront_mutex);
2595 }
2596 
2597 static const struct block_device_operations xlvbd_block_fops =
2598 {
2599 	.owner = THIS_MODULE,
2600 	.open = blkif_open,
2601 	.release = blkif_release,
2602 	.getgeo = blkif_getgeo,
2603 	.ioctl = blkif_ioctl,
2604 };
2605 
2606 
2607 static const struct xenbus_device_id blkfront_ids[] = {
2608 	{ "vbd" },
2609 	{ "" }
2610 };
2611 
2612 static struct xenbus_driver blkfront_driver = {
2613 	.ids  = blkfront_ids,
2614 	.probe = blkfront_probe,
2615 	.remove = blkfront_remove,
2616 	.resume = blkfront_resume,
2617 	.otherend_changed = blkback_changed,
2618 	.is_ready = blkfront_is_ready,
2619 };
2620 
2621 static int __init xlblk_init(void)
2622 {
2623 	int ret;
2624 	int nr_cpus = num_online_cpus();
2625 
2626 	if (!xen_domain())
2627 		return -ENODEV;
2628 
2629 	if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2630 		xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2631 
2632 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2633 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2634 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2635 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2636 	}
2637 
2638 	if (xen_blkif_max_queues > nr_cpus) {
2639 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2640 			xen_blkif_max_queues, nr_cpus);
2641 		xen_blkif_max_queues = nr_cpus;
2642 	}
2643 
2644 	if (!xen_has_pv_disk_devices())
2645 		return -ENODEV;
2646 
2647 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2648 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2649 		       XENVBD_MAJOR, DEV_NAME);
2650 		return -ENODEV;
2651 	}
2652 
2653 	ret = xenbus_register_frontend(&blkfront_driver);
2654 	if (ret) {
2655 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2656 		return ret;
2657 	}
2658 
2659 	return 0;
2660 }
2661 module_init(xlblk_init);
2662 
2663 
2664 static void __exit xlblk_exit(void)
2665 {
2666 	xenbus_unregister_driver(&blkfront_driver);
2667 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2668 	kfree(minors);
2669 }
2670 module_exit(xlblk_exit);
2671 
2672 MODULE_DESCRIPTION("Xen virtual block device frontend");
2673 MODULE_LICENSE("GPL");
2674 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2675 MODULE_ALIAS("xen:vbd");
2676 MODULE_ALIAS("xenblk");
2677