xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 53a2a90d)
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/major.h>
46 #include <linux/mutex.h>
47 #include <linux/scatterlist.h>
48 #include <linux/bitmap.h>
49 #include <linux/list.h>
50 #include <linux/workqueue.h>
51 #include <linux/sched/mm.h>
52 
53 #include <xen/xen.h>
54 #include <xen/xenbus.h>
55 #include <xen/grant_table.h>
56 #include <xen/events.h>
57 #include <xen/page.h>
58 #include <xen/platform_pci.h>
59 
60 #include <xen/interface/grant_table.h>
61 #include <xen/interface/io/blkif.h>
62 #include <xen/interface/io/protocols.h>
63 
64 #include <asm/xen/hypervisor.h>
65 
66 /*
67  * The minimal size of segment supported by the block framework is PAGE_SIZE.
68  * When Linux is using a different page size than Xen, it may not be possible
69  * to put all the data in a single segment.
70  * This can happen when the backend doesn't support indirect descriptor and
71  * therefore the maximum amount of data that a request can carry is
72  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
73  *
74  * Note that we only support one extra request. So the Linux page size
75  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
76  * 88KB.
77  */
78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
79 
80 enum blkif_state {
81 	BLKIF_STATE_DISCONNECTED,
82 	BLKIF_STATE_CONNECTED,
83 	BLKIF_STATE_SUSPENDED,
84 	BLKIF_STATE_ERROR,
85 };
86 
87 struct grant {
88 	grant_ref_t gref;
89 	struct page *page;
90 	struct list_head node;
91 };
92 
93 enum blk_req_status {
94 	REQ_PROCESSING,
95 	REQ_WAITING,
96 	REQ_DONE,
97 	REQ_ERROR,
98 	REQ_EOPNOTSUPP,
99 };
100 
101 struct blk_shadow {
102 	struct blkif_request req;
103 	struct request *request;
104 	struct grant **grants_used;
105 	struct grant **indirect_grants;
106 	struct scatterlist *sg;
107 	unsigned int num_sg;
108 	enum blk_req_status status;
109 
110 	#define NO_ASSOCIATED_ID ~0UL
111 	/*
112 	 * Id of the sibling if we ever need 2 requests when handling a
113 	 * block I/O request
114 	 */
115 	unsigned long associated_id;
116 };
117 
118 struct blkif_req {
119 	blk_status_t	error;
120 };
121 
122 static inline struct blkif_req *blkif_req(struct request *rq)
123 {
124 	return blk_mq_rq_to_pdu(rq);
125 }
126 
127 static DEFINE_MUTEX(blkfront_mutex);
128 static const struct block_device_operations xlvbd_block_fops;
129 static struct delayed_work blkfront_work;
130 static LIST_HEAD(info_list);
131 
132 /*
133  * Maximum number of segments in indirect requests, the actual value used by
134  * the frontend driver is the minimum of this value and the value provided
135  * by the backend driver.
136  */
137 
138 static unsigned int xen_blkif_max_segments = 32;
139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
140 MODULE_PARM_DESC(max_indirect_segments,
141 		 "Maximum amount of segments in indirect requests (default is 32)");
142 
143 static unsigned int xen_blkif_max_queues = 4;
144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
146 
147 /*
148  * Maximum order of pages to be used for the shared ring between front and
149  * backend, 4KB page granularity is used.
150  */
151 static unsigned int xen_blkif_max_ring_order;
152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
154 
155 static bool __read_mostly xen_blkif_trusted = true;
156 module_param_named(trusted, xen_blkif_trusted, bool, 0644);
157 MODULE_PARM_DESC(trusted, "Is the backend trusted");
158 
159 #define BLK_RING_SIZE(info)	\
160 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
161 
162 /*
163  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
164  * characters are enough. Define to 20 to keep consistent with backend.
165  */
166 #define RINGREF_NAME_LEN (20)
167 /*
168  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
169  */
170 #define QUEUE_NAME_LEN (17)
171 
172 /*
173  *  Per-ring info.
174  *  Every blkfront device can associate with one or more blkfront_ring_info,
175  *  depending on how many hardware queues/rings to be used.
176  */
177 struct blkfront_ring_info {
178 	/* Lock to protect data in every ring buffer. */
179 	spinlock_t ring_lock;
180 	struct blkif_front_ring ring;
181 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
182 	unsigned int evtchn, irq;
183 	struct work_struct work;
184 	struct gnttab_free_callback callback;
185 	struct list_head indirect_pages;
186 	struct list_head grants;
187 	unsigned int persistent_gnts_c;
188 	unsigned long shadow_free;
189 	struct blkfront_info *dev_info;
190 	struct blk_shadow shadow[];
191 };
192 
193 /*
194  * We have one of these per vbd, whether ide, scsi or 'other'.  They
195  * hang in private_data off the gendisk structure. We may end up
196  * putting all kinds of interesting stuff here :-)
197  */
198 struct blkfront_info
199 {
200 	struct mutex mutex;
201 	struct xenbus_device *xbdev;
202 	struct gendisk *gd;
203 	u16 sector_size;
204 	unsigned int physical_sector_size;
205 	unsigned long vdisk_info;
206 	int vdevice;
207 	blkif_vdev_t handle;
208 	enum blkif_state connected;
209 	/* Number of pages per ring buffer. */
210 	unsigned int nr_ring_pages;
211 	struct request_queue *rq;
212 	unsigned int feature_flush:1;
213 	unsigned int feature_fua:1;
214 	unsigned int feature_discard:1;
215 	unsigned int feature_secdiscard:1;
216 	unsigned int feature_persistent:1;
217 	unsigned int bounce:1;
218 	unsigned int discard_granularity;
219 	unsigned int discard_alignment;
220 	/* Number of 4KB segments handled */
221 	unsigned int max_indirect_segments;
222 	int is_ready;
223 	struct blk_mq_tag_set tag_set;
224 	struct blkfront_ring_info *rinfo;
225 	unsigned int nr_rings;
226 	unsigned int rinfo_size;
227 	/* Save uncomplete reqs and bios for migration. */
228 	struct list_head requests;
229 	struct bio_list bio_list;
230 	struct list_head info_list;
231 };
232 
233 static unsigned int nr_minors;
234 static unsigned long *minors;
235 static DEFINE_SPINLOCK(minor_lock);
236 
237 #define PARTS_PER_DISK		16
238 #define PARTS_PER_EXT_DISK      256
239 
240 #define BLKIF_MAJOR(dev) ((dev)>>8)
241 #define BLKIF_MINOR(dev) ((dev) & 0xff)
242 
243 #define EXT_SHIFT 28
244 #define EXTENDED (1<<EXT_SHIFT)
245 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
246 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
247 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
248 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
249 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
250 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
251 
252 #define DEV_NAME	"xvd"	/* name in /dev */
253 
254 /*
255  * Grants are always the same size as a Xen page (i.e 4KB).
256  * A physical segment is always the same size as a Linux page.
257  * Number of grants per physical segment
258  */
259 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
260 
261 #define GRANTS_PER_INDIRECT_FRAME \
262 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
263 
264 #define INDIRECT_GREFS(_grants)		\
265 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
266 
267 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
268 static void blkfront_gather_backend_features(struct blkfront_info *info);
269 static int negotiate_mq(struct blkfront_info *info);
270 
271 #define for_each_rinfo(info, ptr, idx)				\
272 	for ((ptr) = (info)->rinfo, (idx) = 0;			\
273 	     (idx) < (info)->nr_rings;				\
274 	     (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
275 
276 static inline struct blkfront_ring_info *
277 get_rinfo(const struct blkfront_info *info, unsigned int i)
278 {
279 	BUG_ON(i >= info->nr_rings);
280 	return (void *)info->rinfo + i * info->rinfo_size;
281 }
282 
283 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
284 {
285 	unsigned long free = rinfo->shadow_free;
286 
287 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
288 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
289 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
290 	return free;
291 }
292 
293 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
294 			      unsigned long id)
295 {
296 	if (rinfo->shadow[id].req.u.rw.id != id)
297 		return -EINVAL;
298 	if (rinfo->shadow[id].request == NULL)
299 		return -EINVAL;
300 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
301 	rinfo->shadow[id].request = NULL;
302 	rinfo->shadow_free = id;
303 	return 0;
304 }
305 
306 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
307 {
308 	struct blkfront_info *info = rinfo->dev_info;
309 	struct page *granted_page;
310 	struct grant *gnt_list_entry, *n;
311 	int i = 0;
312 
313 	while (i < num) {
314 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
315 		if (!gnt_list_entry)
316 			goto out_of_memory;
317 
318 		if (info->bounce) {
319 			granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
320 			if (!granted_page) {
321 				kfree(gnt_list_entry);
322 				goto out_of_memory;
323 			}
324 			gnt_list_entry->page = granted_page;
325 		}
326 
327 		gnt_list_entry->gref = INVALID_GRANT_REF;
328 		list_add(&gnt_list_entry->node, &rinfo->grants);
329 		i++;
330 	}
331 
332 	return 0;
333 
334 out_of_memory:
335 	list_for_each_entry_safe(gnt_list_entry, n,
336 	                         &rinfo->grants, node) {
337 		list_del(&gnt_list_entry->node);
338 		if (info->bounce)
339 			__free_page(gnt_list_entry->page);
340 		kfree(gnt_list_entry);
341 		i--;
342 	}
343 	BUG_ON(i != 0);
344 	return -ENOMEM;
345 }
346 
347 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
348 {
349 	struct grant *gnt_list_entry;
350 
351 	BUG_ON(list_empty(&rinfo->grants));
352 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
353 					  node);
354 	list_del(&gnt_list_entry->node);
355 
356 	if (gnt_list_entry->gref != INVALID_GRANT_REF)
357 		rinfo->persistent_gnts_c--;
358 
359 	return gnt_list_entry;
360 }
361 
362 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
363 					const struct blkfront_info *info)
364 {
365 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
366 						 info->xbdev->otherend_id,
367 						 gnt_list_entry->page,
368 						 0);
369 }
370 
371 static struct grant *get_grant(grant_ref_t *gref_head,
372 			       unsigned long gfn,
373 			       struct blkfront_ring_info *rinfo)
374 {
375 	struct grant *gnt_list_entry = get_free_grant(rinfo);
376 	struct blkfront_info *info = rinfo->dev_info;
377 
378 	if (gnt_list_entry->gref != INVALID_GRANT_REF)
379 		return gnt_list_entry;
380 
381 	/* Assign a gref to this page */
382 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
383 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
384 	if (info->bounce)
385 		grant_foreign_access(gnt_list_entry, info);
386 	else {
387 		/* Grant access to the GFN passed by the caller */
388 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
389 						info->xbdev->otherend_id,
390 						gfn, 0);
391 	}
392 
393 	return gnt_list_entry;
394 }
395 
396 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
397 					struct blkfront_ring_info *rinfo)
398 {
399 	struct grant *gnt_list_entry = get_free_grant(rinfo);
400 	struct blkfront_info *info = rinfo->dev_info;
401 
402 	if (gnt_list_entry->gref != INVALID_GRANT_REF)
403 		return gnt_list_entry;
404 
405 	/* Assign a gref to this page */
406 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
407 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
408 	if (!info->bounce) {
409 		struct page *indirect_page;
410 
411 		/* Fetch a pre-allocated page to use for indirect grefs */
412 		BUG_ON(list_empty(&rinfo->indirect_pages));
413 		indirect_page = list_first_entry(&rinfo->indirect_pages,
414 						 struct page, lru);
415 		list_del(&indirect_page->lru);
416 		gnt_list_entry->page = indirect_page;
417 	}
418 	grant_foreign_access(gnt_list_entry, info);
419 
420 	return gnt_list_entry;
421 }
422 
423 static const char *op_name(int op)
424 {
425 	static const char *const names[] = {
426 		[BLKIF_OP_READ] = "read",
427 		[BLKIF_OP_WRITE] = "write",
428 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
429 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
430 		[BLKIF_OP_DISCARD] = "discard" };
431 
432 	if (op < 0 || op >= ARRAY_SIZE(names))
433 		return "unknown";
434 
435 	if (!names[op])
436 		return "reserved";
437 
438 	return names[op];
439 }
440 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
441 {
442 	unsigned int end = minor + nr;
443 	int rc;
444 
445 	if (end > nr_minors) {
446 		unsigned long *bitmap, *old;
447 
448 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
449 				 GFP_KERNEL);
450 		if (bitmap == NULL)
451 			return -ENOMEM;
452 
453 		spin_lock(&minor_lock);
454 		if (end > nr_minors) {
455 			old = minors;
456 			memcpy(bitmap, minors,
457 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
458 			minors = bitmap;
459 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
460 		} else
461 			old = bitmap;
462 		spin_unlock(&minor_lock);
463 		kfree(old);
464 	}
465 
466 	spin_lock(&minor_lock);
467 	if (find_next_bit(minors, end, minor) >= end) {
468 		bitmap_set(minors, minor, nr);
469 		rc = 0;
470 	} else
471 		rc = -EBUSY;
472 	spin_unlock(&minor_lock);
473 
474 	return rc;
475 }
476 
477 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
478 {
479 	unsigned int end = minor + nr;
480 
481 	BUG_ON(end > nr_minors);
482 	spin_lock(&minor_lock);
483 	bitmap_clear(minors,  minor, nr);
484 	spin_unlock(&minor_lock);
485 }
486 
487 static void blkif_restart_queue_callback(void *arg)
488 {
489 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
490 	schedule_work(&rinfo->work);
491 }
492 
493 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
494 {
495 	/* We don't have real geometry info, but let's at least return
496 	   values consistent with the size of the device */
497 	sector_t nsect = get_capacity(bd->bd_disk);
498 	sector_t cylinders = nsect;
499 
500 	hg->heads = 0xff;
501 	hg->sectors = 0x3f;
502 	sector_div(cylinders, hg->heads * hg->sectors);
503 	hg->cylinders = cylinders;
504 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
505 		hg->cylinders = 0xffff;
506 	return 0;
507 }
508 
509 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
510 		       unsigned command, unsigned long argument)
511 {
512 	struct blkfront_info *info = bdev->bd_disk->private_data;
513 	int i;
514 
515 	switch (command) {
516 	case CDROMMULTISESSION:
517 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
518 			if (put_user(0, (char __user *)(argument + i)))
519 				return -EFAULT;
520 		return 0;
521 	case CDROM_GET_CAPABILITY:
522 		if (!(info->vdisk_info & VDISK_CDROM))
523 			return -EINVAL;
524 		return 0;
525 	default:
526 		return -EINVAL;
527 	}
528 }
529 
530 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
531 					    struct request *req,
532 					    struct blkif_request **ring_req)
533 {
534 	unsigned long id;
535 
536 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
537 	rinfo->ring.req_prod_pvt++;
538 
539 	id = get_id_from_freelist(rinfo);
540 	rinfo->shadow[id].request = req;
541 	rinfo->shadow[id].status = REQ_PROCESSING;
542 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
543 
544 	rinfo->shadow[id].req.u.rw.id = id;
545 
546 	return id;
547 }
548 
549 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
550 {
551 	struct blkfront_info *info = rinfo->dev_info;
552 	struct blkif_request *ring_req, *final_ring_req;
553 	unsigned long id;
554 
555 	/* Fill out a communications ring structure. */
556 	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
557 	ring_req = &rinfo->shadow[id].req;
558 
559 	ring_req->operation = BLKIF_OP_DISCARD;
560 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
561 	ring_req->u.discard.id = id;
562 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
563 	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
564 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
565 	else
566 		ring_req->u.discard.flag = 0;
567 
568 	/* Copy the request to the ring page. */
569 	*final_ring_req = *ring_req;
570 	rinfo->shadow[id].status = REQ_WAITING;
571 
572 	return 0;
573 }
574 
575 struct setup_rw_req {
576 	unsigned int grant_idx;
577 	struct blkif_request_segment *segments;
578 	struct blkfront_ring_info *rinfo;
579 	struct blkif_request *ring_req;
580 	grant_ref_t gref_head;
581 	unsigned int id;
582 	/* Only used when persistent grant is used and it's a write request */
583 	bool need_copy;
584 	unsigned int bvec_off;
585 	char *bvec_data;
586 
587 	bool require_extra_req;
588 	struct blkif_request *extra_ring_req;
589 };
590 
591 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
592 				     unsigned int len, void *data)
593 {
594 	struct setup_rw_req *setup = data;
595 	int n, ref;
596 	struct grant *gnt_list_entry;
597 	unsigned int fsect, lsect;
598 	/* Convenient aliases */
599 	unsigned int grant_idx = setup->grant_idx;
600 	struct blkif_request *ring_req = setup->ring_req;
601 	struct blkfront_ring_info *rinfo = setup->rinfo;
602 	/*
603 	 * We always use the shadow of the first request to store the list
604 	 * of grant associated to the block I/O request. This made the
605 	 * completion more easy to handle even if the block I/O request is
606 	 * split.
607 	 */
608 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
609 
610 	if (unlikely(setup->require_extra_req &&
611 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
612 		/*
613 		 * We are using the second request, setup grant_idx
614 		 * to be the index of the segment array.
615 		 */
616 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
617 		ring_req = setup->extra_ring_req;
618 	}
619 
620 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
621 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
622 		if (setup->segments)
623 			kunmap_atomic(setup->segments);
624 
625 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
626 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
627 		shadow->indirect_grants[n] = gnt_list_entry;
628 		setup->segments = kmap_atomic(gnt_list_entry->page);
629 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
630 	}
631 
632 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
633 	ref = gnt_list_entry->gref;
634 	/*
635 	 * All the grants are stored in the shadow of the first
636 	 * request. Therefore we have to use the global index.
637 	 */
638 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
639 
640 	if (setup->need_copy) {
641 		void *shared_data;
642 
643 		shared_data = kmap_atomic(gnt_list_entry->page);
644 		/*
645 		 * this does not wipe data stored outside the
646 		 * range sg->offset..sg->offset+sg->length.
647 		 * Therefore, blkback *could* see data from
648 		 * previous requests. This is OK as long as
649 		 * persistent grants are shared with just one
650 		 * domain. It may need refactoring if this
651 		 * changes
652 		 */
653 		memcpy(shared_data + offset,
654 		       setup->bvec_data + setup->bvec_off,
655 		       len);
656 
657 		kunmap_atomic(shared_data);
658 		setup->bvec_off += len;
659 	}
660 
661 	fsect = offset >> 9;
662 	lsect = fsect + (len >> 9) - 1;
663 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
664 		ring_req->u.rw.seg[grant_idx] =
665 			(struct blkif_request_segment) {
666 				.gref       = ref,
667 				.first_sect = fsect,
668 				.last_sect  = lsect };
669 	} else {
670 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
671 			(struct blkif_request_segment) {
672 				.gref       = ref,
673 				.first_sect = fsect,
674 				.last_sect  = lsect };
675 	}
676 
677 	(setup->grant_idx)++;
678 }
679 
680 static void blkif_setup_extra_req(struct blkif_request *first,
681 				  struct blkif_request *second)
682 {
683 	uint16_t nr_segments = first->u.rw.nr_segments;
684 
685 	/*
686 	 * The second request is only present when the first request uses
687 	 * all its segments. It's always the continuity of the first one.
688 	 */
689 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
690 
691 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
692 	second->u.rw.sector_number = first->u.rw.sector_number +
693 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
694 
695 	second->u.rw.handle = first->u.rw.handle;
696 	second->operation = first->operation;
697 }
698 
699 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
700 {
701 	struct blkfront_info *info = rinfo->dev_info;
702 	struct blkif_request *ring_req, *extra_ring_req = NULL;
703 	struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
704 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
705 	bool require_extra_req = false;
706 	int i;
707 	struct setup_rw_req setup = {
708 		.grant_idx = 0,
709 		.segments = NULL,
710 		.rinfo = rinfo,
711 		.need_copy = rq_data_dir(req) && info->bounce,
712 	};
713 
714 	/*
715 	 * Used to store if we are able to queue the request by just using
716 	 * existing persistent grants, or if we have to get new grants,
717 	 * as there are not sufficiently many free.
718 	 */
719 	bool new_persistent_gnts = false;
720 	struct scatterlist *sg;
721 	int num_sg, max_grefs, num_grant;
722 
723 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
724 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
725 		/*
726 		 * If we are using indirect segments we need to account
727 		 * for the indirect grefs used in the request.
728 		 */
729 		max_grefs += INDIRECT_GREFS(max_grefs);
730 
731 	/* Check if we have enough persistent grants to allocate a requests */
732 	if (rinfo->persistent_gnts_c < max_grefs) {
733 		new_persistent_gnts = true;
734 
735 		if (gnttab_alloc_grant_references(
736 		    max_grefs - rinfo->persistent_gnts_c,
737 		    &setup.gref_head) < 0) {
738 			gnttab_request_free_callback(
739 				&rinfo->callback,
740 				blkif_restart_queue_callback,
741 				rinfo,
742 				max_grefs - rinfo->persistent_gnts_c);
743 			return 1;
744 		}
745 	}
746 
747 	/* Fill out a communications ring structure. */
748 	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
749 	ring_req = &rinfo->shadow[id].req;
750 
751 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
752 	num_grant = 0;
753 	/* Calculate the number of grant used */
754 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
755 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
756 
757 	require_extra_req = info->max_indirect_segments == 0 &&
758 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
759 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
760 
761 	rinfo->shadow[id].num_sg = num_sg;
762 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
763 	    likely(!require_extra_req)) {
764 		/*
765 		 * The indirect operation can only be a BLKIF_OP_READ or
766 		 * BLKIF_OP_WRITE
767 		 */
768 		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
769 		ring_req->operation = BLKIF_OP_INDIRECT;
770 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
771 			BLKIF_OP_WRITE : BLKIF_OP_READ;
772 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
773 		ring_req->u.indirect.handle = info->handle;
774 		ring_req->u.indirect.nr_segments = num_grant;
775 	} else {
776 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
777 		ring_req->u.rw.handle = info->handle;
778 		ring_req->operation = rq_data_dir(req) ?
779 			BLKIF_OP_WRITE : BLKIF_OP_READ;
780 		if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
781 			/*
782 			 * Ideally we can do an unordered flush-to-disk.
783 			 * In case the backend onlysupports barriers, use that.
784 			 * A barrier request a superset of FUA, so we can
785 			 * implement it the same way.  (It's also a FLUSH+FUA,
786 			 * since it is guaranteed ordered WRT previous writes.)
787 			 */
788 			if (info->feature_flush && info->feature_fua)
789 				ring_req->operation =
790 					BLKIF_OP_WRITE_BARRIER;
791 			else if (info->feature_flush)
792 				ring_req->operation =
793 					BLKIF_OP_FLUSH_DISKCACHE;
794 			else
795 				ring_req->operation = 0;
796 		}
797 		ring_req->u.rw.nr_segments = num_grant;
798 		if (unlikely(require_extra_req)) {
799 			extra_id = blkif_ring_get_request(rinfo, req,
800 							  &final_extra_ring_req);
801 			extra_ring_req = &rinfo->shadow[extra_id].req;
802 
803 			/*
804 			 * Only the first request contains the scatter-gather
805 			 * list.
806 			 */
807 			rinfo->shadow[extra_id].num_sg = 0;
808 
809 			blkif_setup_extra_req(ring_req, extra_ring_req);
810 
811 			/* Link the 2 requests together */
812 			rinfo->shadow[extra_id].associated_id = id;
813 			rinfo->shadow[id].associated_id = extra_id;
814 		}
815 	}
816 
817 	setup.ring_req = ring_req;
818 	setup.id = id;
819 
820 	setup.require_extra_req = require_extra_req;
821 	if (unlikely(require_extra_req))
822 		setup.extra_ring_req = extra_ring_req;
823 
824 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
825 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
826 
827 		if (setup.need_copy) {
828 			setup.bvec_off = sg->offset;
829 			setup.bvec_data = kmap_atomic(sg_page(sg));
830 		}
831 
832 		gnttab_foreach_grant_in_range(sg_page(sg),
833 					      sg->offset,
834 					      sg->length,
835 					      blkif_setup_rw_req_grant,
836 					      &setup);
837 
838 		if (setup.need_copy)
839 			kunmap_atomic(setup.bvec_data);
840 	}
841 	if (setup.segments)
842 		kunmap_atomic(setup.segments);
843 
844 	/* Copy request(s) to the ring page. */
845 	*final_ring_req = *ring_req;
846 	rinfo->shadow[id].status = REQ_WAITING;
847 	if (unlikely(require_extra_req)) {
848 		*final_extra_ring_req = *extra_ring_req;
849 		rinfo->shadow[extra_id].status = REQ_WAITING;
850 	}
851 
852 	if (new_persistent_gnts)
853 		gnttab_free_grant_references(setup.gref_head);
854 
855 	return 0;
856 }
857 
858 /*
859  * Generate a Xen blkfront IO request from a blk layer request.  Reads
860  * and writes are handled as expected.
861  *
862  * @req: a request struct
863  */
864 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
865 {
866 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
867 		return 1;
868 
869 	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
870 		     req_op(req) == REQ_OP_SECURE_ERASE))
871 		return blkif_queue_discard_req(req, rinfo);
872 	else
873 		return blkif_queue_rw_req(req, rinfo);
874 }
875 
876 static inline void flush_requests(struct blkfront_ring_info *rinfo)
877 {
878 	int notify;
879 
880 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
881 
882 	if (notify)
883 		notify_remote_via_irq(rinfo->irq);
884 }
885 
886 static inline bool blkif_request_flush_invalid(struct request *req,
887 					       struct blkfront_info *info)
888 {
889 	return (blk_rq_is_passthrough(req) ||
890 		((req_op(req) == REQ_OP_FLUSH) &&
891 		 !info->feature_flush) ||
892 		((req->cmd_flags & REQ_FUA) &&
893 		 !info->feature_fua));
894 }
895 
896 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
897 			  const struct blk_mq_queue_data *qd)
898 {
899 	unsigned long flags;
900 	int qid = hctx->queue_num;
901 	struct blkfront_info *info = hctx->queue->queuedata;
902 	struct blkfront_ring_info *rinfo = NULL;
903 
904 	rinfo = get_rinfo(info, qid);
905 	blk_mq_start_request(qd->rq);
906 	spin_lock_irqsave(&rinfo->ring_lock, flags);
907 	if (RING_FULL(&rinfo->ring))
908 		goto out_busy;
909 
910 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
911 		goto out_err;
912 
913 	if (blkif_queue_request(qd->rq, rinfo))
914 		goto out_busy;
915 
916 	flush_requests(rinfo);
917 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
918 	return BLK_STS_OK;
919 
920 out_err:
921 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
922 	return BLK_STS_IOERR;
923 
924 out_busy:
925 	blk_mq_stop_hw_queue(hctx);
926 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
927 	return BLK_STS_DEV_RESOURCE;
928 }
929 
930 static void blkif_complete_rq(struct request *rq)
931 {
932 	blk_mq_end_request(rq, blkif_req(rq)->error);
933 }
934 
935 static const struct blk_mq_ops blkfront_mq_ops = {
936 	.queue_rq = blkif_queue_rq,
937 	.complete = blkif_complete_rq,
938 };
939 
940 static void blkif_set_queue_limits(struct blkfront_info *info)
941 {
942 	struct request_queue *rq = info->rq;
943 	struct gendisk *gd = info->gd;
944 	unsigned int segments = info->max_indirect_segments ? :
945 				BLKIF_MAX_SEGMENTS_PER_REQUEST;
946 
947 	blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
948 
949 	if (info->feature_discard) {
950 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
951 		rq->limits.discard_granularity = info->discard_granularity ?:
952 						 info->physical_sector_size;
953 		rq->limits.discard_alignment = info->discard_alignment;
954 		if (info->feature_secdiscard)
955 			blk_queue_max_secure_erase_sectors(rq,
956 							   get_capacity(gd));
957 	}
958 
959 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
960 	blk_queue_logical_block_size(rq, info->sector_size);
961 	blk_queue_physical_block_size(rq, info->physical_sector_size);
962 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
963 
964 	/* Each segment in a request is up to an aligned page in size. */
965 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
966 	blk_queue_max_segment_size(rq, PAGE_SIZE);
967 
968 	/* Ensure a merged request will fit in a single I/O ring slot. */
969 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
970 
971 	/* Make sure buffer addresses are sector-aligned. */
972 	blk_queue_dma_alignment(rq, 511);
973 }
974 
975 static const char *flush_info(struct blkfront_info *info)
976 {
977 	if (info->feature_flush && info->feature_fua)
978 		return "barrier: enabled;";
979 	else if (info->feature_flush)
980 		return "flush diskcache: enabled;";
981 	else
982 		return "barrier or flush: disabled;";
983 }
984 
985 static void xlvbd_flush(struct blkfront_info *info)
986 {
987 	blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
988 			      info->feature_fua ? true : false);
989 	pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
990 		info->gd->disk_name, flush_info(info),
991 		"persistent grants:", info->feature_persistent ?
992 		"enabled;" : "disabled;", "indirect descriptors:",
993 		info->max_indirect_segments ? "enabled;" : "disabled;",
994 		"bounce buffer:", info->bounce ? "enabled" : "disabled;");
995 }
996 
997 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
998 {
999 	int major;
1000 	major = BLKIF_MAJOR(vdevice);
1001 	*minor = BLKIF_MINOR(vdevice);
1002 	switch (major) {
1003 		case XEN_IDE0_MAJOR:
1004 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1005 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1006 				EMULATED_HD_DISK_MINOR_OFFSET;
1007 			break;
1008 		case XEN_IDE1_MAJOR:
1009 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1010 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1011 				EMULATED_HD_DISK_MINOR_OFFSET;
1012 			break;
1013 		case XEN_SCSI_DISK0_MAJOR:
1014 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1015 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1016 			break;
1017 		case XEN_SCSI_DISK1_MAJOR:
1018 		case XEN_SCSI_DISK2_MAJOR:
1019 		case XEN_SCSI_DISK3_MAJOR:
1020 		case XEN_SCSI_DISK4_MAJOR:
1021 		case XEN_SCSI_DISK5_MAJOR:
1022 		case XEN_SCSI_DISK6_MAJOR:
1023 		case XEN_SCSI_DISK7_MAJOR:
1024 			*offset = (*minor / PARTS_PER_DISK) +
1025 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1026 				EMULATED_SD_DISK_NAME_OFFSET;
1027 			*minor = *minor +
1028 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1029 				EMULATED_SD_DISK_MINOR_OFFSET;
1030 			break;
1031 		case XEN_SCSI_DISK8_MAJOR:
1032 		case XEN_SCSI_DISK9_MAJOR:
1033 		case XEN_SCSI_DISK10_MAJOR:
1034 		case XEN_SCSI_DISK11_MAJOR:
1035 		case XEN_SCSI_DISK12_MAJOR:
1036 		case XEN_SCSI_DISK13_MAJOR:
1037 		case XEN_SCSI_DISK14_MAJOR:
1038 		case XEN_SCSI_DISK15_MAJOR:
1039 			*offset = (*minor / PARTS_PER_DISK) +
1040 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1041 				EMULATED_SD_DISK_NAME_OFFSET;
1042 			*minor = *minor +
1043 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1044 				EMULATED_SD_DISK_MINOR_OFFSET;
1045 			break;
1046 		case XENVBD_MAJOR:
1047 			*offset = *minor / PARTS_PER_DISK;
1048 			break;
1049 		default:
1050 			printk(KERN_WARNING "blkfront: your disk configuration is "
1051 					"incorrect, please use an xvd device instead\n");
1052 			return -ENODEV;
1053 	}
1054 	return 0;
1055 }
1056 
1057 static char *encode_disk_name(char *ptr, unsigned int n)
1058 {
1059 	if (n >= 26)
1060 		ptr = encode_disk_name(ptr, n / 26 - 1);
1061 	*ptr = 'a' + n % 26;
1062 	return ptr + 1;
1063 }
1064 
1065 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1066 		struct blkfront_info *info, u16 sector_size,
1067 		unsigned int physical_sector_size)
1068 {
1069 	struct gendisk *gd;
1070 	int nr_minors = 1;
1071 	int err;
1072 	unsigned int offset;
1073 	int minor;
1074 	int nr_parts;
1075 	char *ptr;
1076 
1077 	BUG_ON(info->gd != NULL);
1078 	BUG_ON(info->rq != NULL);
1079 
1080 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1081 		/* this is above the extended range; something is wrong */
1082 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1083 		return -ENODEV;
1084 	}
1085 
1086 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1087 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1088 		if (err)
1089 			return err;
1090 		nr_parts = PARTS_PER_DISK;
1091 	} else {
1092 		minor = BLKIF_MINOR_EXT(info->vdevice);
1093 		nr_parts = PARTS_PER_EXT_DISK;
1094 		offset = minor / nr_parts;
1095 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1096 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1097 					"emulated IDE disks,\n\t choose an xvd device name"
1098 					"from xvde on\n", info->vdevice);
1099 	}
1100 	if (minor >> MINORBITS) {
1101 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1102 			info->vdevice, minor);
1103 		return -ENODEV;
1104 	}
1105 
1106 	if ((minor % nr_parts) == 0)
1107 		nr_minors = nr_parts;
1108 
1109 	err = xlbd_reserve_minors(minor, nr_minors);
1110 	if (err)
1111 		return err;
1112 
1113 	memset(&info->tag_set, 0, sizeof(info->tag_set));
1114 	info->tag_set.ops = &blkfront_mq_ops;
1115 	info->tag_set.nr_hw_queues = info->nr_rings;
1116 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1117 		/*
1118 		 * When indirect descriptior is not supported, the I/O request
1119 		 * will be split between multiple request in the ring.
1120 		 * To avoid problems when sending the request, divide by
1121 		 * 2 the depth of the queue.
1122 		 */
1123 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1124 	} else
1125 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
1126 	info->tag_set.numa_node = NUMA_NO_NODE;
1127 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1128 	info->tag_set.cmd_size = sizeof(struct blkif_req);
1129 	info->tag_set.driver_data = info;
1130 
1131 	err = blk_mq_alloc_tag_set(&info->tag_set);
1132 	if (err)
1133 		goto out_release_minors;
1134 
1135 	gd = blk_mq_alloc_disk(&info->tag_set, info);
1136 	if (IS_ERR(gd)) {
1137 		err = PTR_ERR(gd);
1138 		goto out_free_tag_set;
1139 	}
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->minors = nr_minors;
1153 	gd->fops = &xlvbd_block_fops;
1154 	gd->private_data = info;
1155 	set_capacity(gd, capacity);
1156 
1157 	info->rq = gd->queue;
1158 	info->gd = gd;
1159 	info->sector_size = sector_size;
1160 	info->physical_sector_size = physical_sector_size;
1161 	blkif_set_queue_limits(info);
1162 
1163 	xlvbd_flush(info);
1164 
1165 	if (info->vdisk_info & VDISK_READONLY)
1166 		set_disk_ro(gd, 1);
1167 	if (info->vdisk_info & VDISK_REMOVABLE)
1168 		gd->flags |= GENHD_FL_REMOVABLE;
1169 
1170 	return 0;
1171 
1172 out_free_tag_set:
1173 	blk_mq_free_tag_set(&info->tag_set);
1174 out_release_minors:
1175 	xlbd_release_minors(minor, nr_minors);
1176 	return err;
1177 }
1178 
1179 /* Already hold rinfo->ring_lock. */
1180 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1181 {
1182 	if (!RING_FULL(&rinfo->ring))
1183 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1184 }
1185 
1186 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1187 {
1188 	unsigned long flags;
1189 
1190 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1191 	kick_pending_request_queues_locked(rinfo);
1192 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1193 }
1194 
1195 static void blkif_restart_queue(struct work_struct *work)
1196 {
1197 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1198 
1199 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1200 		kick_pending_request_queues(rinfo);
1201 }
1202 
1203 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1204 {
1205 	struct grant *persistent_gnt, *n;
1206 	struct blkfront_info *info = rinfo->dev_info;
1207 	int i, j, segs;
1208 
1209 	/*
1210 	 * Remove indirect pages, this only happens when using indirect
1211 	 * descriptors but not persistent grants
1212 	 */
1213 	if (!list_empty(&rinfo->indirect_pages)) {
1214 		struct page *indirect_page, *n;
1215 
1216 		BUG_ON(info->bounce);
1217 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1218 			list_del(&indirect_page->lru);
1219 			__free_page(indirect_page);
1220 		}
1221 	}
1222 
1223 	/* Remove all persistent grants. */
1224 	if (!list_empty(&rinfo->grants)) {
1225 		list_for_each_entry_safe(persistent_gnt, n,
1226 					 &rinfo->grants, node) {
1227 			list_del(&persistent_gnt->node);
1228 			if (persistent_gnt->gref != INVALID_GRANT_REF) {
1229 				gnttab_end_foreign_access(persistent_gnt->gref,
1230 							  NULL);
1231 				rinfo->persistent_gnts_c--;
1232 			}
1233 			if (info->bounce)
1234 				__free_page(persistent_gnt->page);
1235 			kfree(persistent_gnt);
1236 		}
1237 	}
1238 	BUG_ON(rinfo->persistent_gnts_c != 0);
1239 
1240 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1241 		/*
1242 		 * Clear persistent grants present in requests already
1243 		 * on the shared ring
1244 		 */
1245 		if (!rinfo->shadow[i].request)
1246 			goto free_shadow;
1247 
1248 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1249 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1250 		       rinfo->shadow[i].req.u.rw.nr_segments;
1251 		for (j = 0; j < segs; j++) {
1252 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1253 			gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1254 			if (info->bounce)
1255 				__free_page(persistent_gnt->page);
1256 			kfree(persistent_gnt);
1257 		}
1258 
1259 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1260 			/*
1261 			 * If this is not an indirect operation don't try to
1262 			 * free indirect segments
1263 			 */
1264 			goto free_shadow;
1265 
1266 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1267 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1268 			gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1269 			__free_page(persistent_gnt->page);
1270 			kfree(persistent_gnt);
1271 		}
1272 
1273 free_shadow:
1274 		kvfree(rinfo->shadow[i].grants_used);
1275 		rinfo->shadow[i].grants_used = NULL;
1276 		kvfree(rinfo->shadow[i].indirect_grants);
1277 		rinfo->shadow[i].indirect_grants = NULL;
1278 		kvfree(rinfo->shadow[i].sg);
1279 		rinfo->shadow[i].sg = NULL;
1280 	}
1281 
1282 	/* No more gnttab callback work. */
1283 	gnttab_cancel_free_callback(&rinfo->callback);
1284 
1285 	/* Flush gnttab callback work. Must be done with no locks held. */
1286 	flush_work(&rinfo->work);
1287 
1288 	/* Free resources associated with old device channel. */
1289 	xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages,
1290 			     rinfo->ring_ref);
1291 
1292 	if (rinfo->irq)
1293 		unbind_from_irqhandler(rinfo->irq, rinfo);
1294 	rinfo->evtchn = rinfo->irq = 0;
1295 }
1296 
1297 static void blkif_free(struct blkfront_info *info, int suspend)
1298 {
1299 	unsigned int i;
1300 	struct blkfront_ring_info *rinfo;
1301 
1302 	/* Prevent new requests being issued until we fix things up. */
1303 	info->connected = suspend ?
1304 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1305 	/* No more blkif_request(). */
1306 	if (info->rq)
1307 		blk_mq_stop_hw_queues(info->rq);
1308 
1309 	for_each_rinfo(info, rinfo, i)
1310 		blkif_free_ring(rinfo);
1311 
1312 	kvfree(info->rinfo);
1313 	info->rinfo = NULL;
1314 	info->nr_rings = 0;
1315 }
1316 
1317 struct copy_from_grant {
1318 	const struct blk_shadow *s;
1319 	unsigned int grant_idx;
1320 	unsigned int bvec_offset;
1321 	char *bvec_data;
1322 };
1323 
1324 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1325 				  unsigned int len, void *data)
1326 {
1327 	struct copy_from_grant *info = data;
1328 	char *shared_data;
1329 	/* Convenient aliases */
1330 	const struct blk_shadow *s = info->s;
1331 
1332 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1333 
1334 	memcpy(info->bvec_data + info->bvec_offset,
1335 	       shared_data + offset, len);
1336 
1337 	info->bvec_offset += len;
1338 	info->grant_idx++;
1339 
1340 	kunmap_atomic(shared_data);
1341 }
1342 
1343 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1344 {
1345 	switch (rsp)
1346 	{
1347 	case BLKIF_RSP_OKAY:
1348 		return REQ_DONE;
1349 	case BLKIF_RSP_EOPNOTSUPP:
1350 		return REQ_EOPNOTSUPP;
1351 	case BLKIF_RSP_ERROR:
1352 	default:
1353 		return REQ_ERROR;
1354 	}
1355 }
1356 
1357 /*
1358  * Get the final status of the block request based on two ring response
1359  */
1360 static int blkif_get_final_status(enum blk_req_status s1,
1361 				  enum blk_req_status s2)
1362 {
1363 	BUG_ON(s1 < REQ_DONE);
1364 	BUG_ON(s2 < REQ_DONE);
1365 
1366 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1367 		return BLKIF_RSP_ERROR;
1368 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1369 		return BLKIF_RSP_EOPNOTSUPP;
1370 	return BLKIF_RSP_OKAY;
1371 }
1372 
1373 /*
1374  * Return values:
1375  *  1 response processed.
1376  *  0 missing further responses.
1377  * -1 error while processing.
1378  */
1379 static int blkif_completion(unsigned long *id,
1380 			    struct blkfront_ring_info *rinfo,
1381 			    struct blkif_response *bret)
1382 {
1383 	int i = 0;
1384 	struct scatterlist *sg;
1385 	int num_sg, num_grant;
1386 	struct blkfront_info *info = rinfo->dev_info;
1387 	struct blk_shadow *s = &rinfo->shadow[*id];
1388 	struct copy_from_grant data = {
1389 		.grant_idx = 0,
1390 	};
1391 
1392 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1393 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1394 
1395 	/* The I/O request may be split in two. */
1396 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1397 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1398 
1399 		/* Keep the status of the current response in shadow. */
1400 		s->status = blkif_rsp_to_req_status(bret->status);
1401 
1402 		/* Wait the second response if not yet here. */
1403 		if (s2->status < REQ_DONE)
1404 			return 0;
1405 
1406 		bret->status = blkif_get_final_status(s->status,
1407 						      s2->status);
1408 
1409 		/*
1410 		 * All the grants is stored in the first shadow in order
1411 		 * to make the completion code simpler.
1412 		 */
1413 		num_grant += s2->req.u.rw.nr_segments;
1414 
1415 		/*
1416 		 * The two responses may not come in order. Only the
1417 		 * first request will store the scatter-gather list.
1418 		 */
1419 		if (s2->num_sg != 0) {
1420 			/* Update "id" with the ID of the first response. */
1421 			*id = s->associated_id;
1422 			s = s2;
1423 		}
1424 
1425 		/*
1426 		 * We don't need anymore the second request, so recycling
1427 		 * it now.
1428 		 */
1429 		if (add_id_to_freelist(rinfo, s->associated_id))
1430 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1431 			     info->gd->disk_name, s->associated_id);
1432 	}
1433 
1434 	data.s = s;
1435 	num_sg = s->num_sg;
1436 
1437 	if (bret->operation == BLKIF_OP_READ && info->bounce) {
1438 		for_each_sg(s->sg, sg, num_sg, i) {
1439 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1440 
1441 			data.bvec_offset = sg->offset;
1442 			data.bvec_data = kmap_atomic(sg_page(sg));
1443 
1444 			gnttab_foreach_grant_in_range(sg_page(sg),
1445 						      sg->offset,
1446 						      sg->length,
1447 						      blkif_copy_from_grant,
1448 						      &data);
1449 
1450 			kunmap_atomic(data.bvec_data);
1451 		}
1452 	}
1453 	/* Add the persistent grant into the list of free grants */
1454 	for (i = 0; i < num_grant; i++) {
1455 		if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1456 			/*
1457 			 * If the grant is still mapped by the backend (the
1458 			 * backend has chosen to make this grant persistent)
1459 			 * we add it at the head of the list, so it will be
1460 			 * reused first.
1461 			 */
1462 			if (!info->feature_persistent) {
1463 				pr_alert("backed has not unmapped grant: %u\n",
1464 					 s->grants_used[i]->gref);
1465 				return -1;
1466 			}
1467 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1468 			rinfo->persistent_gnts_c++;
1469 		} else {
1470 			/*
1471 			 * If the grant is not mapped by the backend we add it
1472 			 * to the tail of the list, so it will not be picked
1473 			 * again unless we run out of persistent grants.
1474 			 */
1475 			s->grants_used[i]->gref = INVALID_GRANT_REF;
1476 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1477 		}
1478 	}
1479 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1480 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1481 			if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1482 				if (!info->feature_persistent) {
1483 					pr_alert("backed has not unmapped grant: %u\n",
1484 						 s->indirect_grants[i]->gref);
1485 					return -1;
1486 				}
1487 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1488 				rinfo->persistent_gnts_c++;
1489 			} else {
1490 				struct page *indirect_page;
1491 
1492 				/*
1493 				 * Add the used indirect page back to the list of
1494 				 * available pages for indirect grefs.
1495 				 */
1496 				if (!info->bounce) {
1497 					indirect_page = s->indirect_grants[i]->page;
1498 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1499 				}
1500 				s->indirect_grants[i]->gref = INVALID_GRANT_REF;
1501 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1502 			}
1503 		}
1504 	}
1505 
1506 	return 1;
1507 }
1508 
1509 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1510 {
1511 	struct request *req;
1512 	struct blkif_response bret;
1513 	RING_IDX i, rp;
1514 	unsigned long flags;
1515 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1516 	struct blkfront_info *info = rinfo->dev_info;
1517 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1518 
1519 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1520 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1521 		return IRQ_HANDLED;
1522 	}
1523 
1524 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1525  again:
1526 	rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1527 	virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1528 	if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1529 		pr_alert("%s: illegal number of responses %u\n",
1530 			 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1531 		goto err;
1532 	}
1533 
1534 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1535 		unsigned long id;
1536 		unsigned int op;
1537 
1538 		eoiflag = 0;
1539 
1540 		RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1541 		id = bret.id;
1542 
1543 		/*
1544 		 * The backend has messed up and given us an id that we would
1545 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1546 		 * look in get_id_from_freelist.
1547 		 */
1548 		if (id >= BLK_RING_SIZE(info)) {
1549 			pr_alert("%s: response has incorrect id (%ld)\n",
1550 				 info->gd->disk_name, id);
1551 			goto err;
1552 		}
1553 		if (rinfo->shadow[id].status != REQ_WAITING) {
1554 			pr_alert("%s: response references no pending request\n",
1555 				 info->gd->disk_name);
1556 			goto err;
1557 		}
1558 
1559 		rinfo->shadow[id].status = REQ_PROCESSING;
1560 		req  = rinfo->shadow[id].request;
1561 
1562 		op = rinfo->shadow[id].req.operation;
1563 		if (op == BLKIF_OP_INDIRECT)
1564 			op = rinfo->shadow[id].req.u.indirect.indirect_op;
1565 		if (bret.operation != op) {
1566 			pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1567 				 info->gd->disk_name, bret.operation, op);
1568 			goto err;
1569 		}
1570 
1571 		if (bret.operation != BLKIF_OP_DISCARD) {
1572 			int ret;
1573 
1574 			/*
1575 			 * We may need to wait for an extra response if the
1576 			 * I/O request is split in 2
1577 			 */
1578 			ret = blkif_completion(&id, rinfo, &bret);
1579 			if (!ret)
1580 				continue;
1581 			if (unlikely(ret < 0))
1582 				goto err;
1583 		}
1584 
1585 		if (add_id_to_freelist(rinfo, id)) {
1586 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1587 			     info->gd->disk_name, op_name(bret.operation), id);
1588 			continue;
1589 		}
1590 
1591 		if (bret.status == BLKIF_RSP_OKAY)
1592 			blkif_req(req)->error = BLK_STS_OK;
1593 		else
1594 			blkif_req(req)->error = BLK_STS_IOERR;
1595 
1596 		switch (bret.operation) {
1597 		case BLKIF_OP_DISCARD:
1598 			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1599 				struct request_queue *rq = info->rq;
1600 
1601 				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1602 					   info->gd->disk_name, op_name(bret.operation));
1603 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1604 				info->feature_discard = 0;
1605 				info->feature_secdiscard = 0;
1606 				blk_queue_max_discard_sectors(rq, 0);
1607 				blk_queue_max_secure_erase_sectors(rq, 0);
1608 			}
1609 			break;
1610 		case BLKIF_OP_FLUSH_DISKCACHE:
1611 		case BLKIF_OP_WRITE_BARRIER:
1612 			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1613 				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1614 				       info->gd->disk_name, op_name(bret.operation));
1615 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1616 			}
1617 			if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1618 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1619 				pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1620 				       info->gd->disk_name, op_name(bret.operation));
1621 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1622 			}
1623 			if (unlikely(blkif_req(req)->error)) {
1624 				if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1625 					blkif_req(req)->error = BLK_STS_OK;
1626 				info->feature_fua = 0;
1627 				info->feature_flush = 0;
1628 				xlvbd_flush(info);
1629 			}
1630 			fallthrough;
1631 		case BLKIF_OP_READ:
1632 		case BLKIF_OP_WRITE:
1633 			if (unlikely(bret.status != BLKIF_RSP_OKAY))
1634 				dev_dbg_ratelimited(&info->xbdev->dev,
1635 					"Bad return from blkdev data request: %#x\n",
1636 					bret.status);
1637 
1638 			break;
1639 		default:
1640 			BUG();
1641 		}
1642 
1643 		if (likely(!blk_should_fake_timeout(req->q)))
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 	xen_irq_lateeoi(irq, eoiflag);
1662 
1663 	return IRQ_HANDLED;
1664 
1665  err:
1666 	info->connected = BLKIF_STATE_ERROR;
1667 
1668 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1669 
1670 	/* No EOI in order to avoid further interrupts. */
1671 
1672 	pr_alert("%s disabled for further use\n", info->gd->disk_name);
1673 	return IRQ_HANDLED;
1674 }
1675 
1676 
1677 static int setup_blkring(struct xenbus_device *dev,
1678 			 struct blkfront_ring_info *rinfo)
1679 {
1680 	struct blkif_sring *sring;
1681 	int err;
1682 	struct blkfront_info *info = rinfo->dev_info;
1683 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1684 
1685 	err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring,
1686 				info->nr_ring_pages, rinfo->ring_ref);
1687 	if (err)
1688 		goto fail;
1689 
1690 	XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1691 
1692 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1693 	if (err)
1694 		goto fail;
1695 
1696 	err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1697 						0, "blkif", rinfo);
1698 	if (err <= 0) {
1699 		xenbus_dev_fatal(dev, err,
1700 				 "bind_evtchn_to_irqhandler failed");
1701 		goto fail;
1702 	}
1703 	rinfo->irq = err;
1704 
1705 	return 0;
1706 fail:
1707 	blkif_free(info, 0);
1708 	return err;
1709 }
1710 
1711 /*
1712  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1713  * ring buffer may have multi pages depending on ->nr_ring_pages.
1714  */
1715 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1716 				struct blkfront_ring_info *rinfo, const char *dir)
1717 {
1718 	int err;
1719 	unsigned int i;
1720 	const char *message = NULL;
1721 	struct blkfront_info *info = rinfo->dev_info;
1722 
1723 	if (info->nr_ring_pages == 1) {
1724 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1725 		if (err) {
1726 			message = "writing ring-ref";
1727 			goto abort_transaction;
1728 		}
1729 	} else {
1730 		for (i = 0; i < info->nr_ring_pages; i++) {
1731 			char ring_ref_name[RINGREF_NAME_LEN];
1732 
1733 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1734 			err = xenbus_printf(xbt, dir, ring_ref_name,
1735 					    "%u", rinfo->ring_ref[i]);
1736 			if (err) {
1737 				message = "writing ring-ref";
1738 				goto abort_transaction;
1739 			}
1740 		}
1741 	}
1742 
1743 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1744 	if (err) {
1745 		message = "writing event-channel";
1746 		goto abort_transaction;
1747 	}
1748 
1749 	return 0;
1750 
1751 abort_transaction:
1752 	xenbus_transaction_end(xbt, 1);
1753 	if (message)
1754 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1755 
1756 	return err;
1757 }
1758 
1759 /* Common code used when first setting up, and when resuming. */
1760 static int talk_to_blkback(struct xenbus_device *dev,
1761 			   struct blkfront_info *info)
1762 {
1763 	const char *message = NULL;
1764 	struct xenbus_transaction xbt;
1765 	int err;
1766 	unsigned int i, max_page_order;
1767 	unsigned int ring_page_order;
1768 	struct blkfront_ring_info *rinfo;
1769 
1770 	if (!info)
1771 		return -ENODEV;
1772 
1773 	/* Check if backend is trusted. */
1774 	info->bounce = !xen_blkif_trusted ||
1775 		       !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1776 
1777 	max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1778 					      "max-ring-page-order", 0);
1779 	ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1780 	info->nr_ring_pages = 1 << ring_page_order;
1781 
1782 	err = negotiate_mq(info);
1783 	if (err)
1784 		goto destroy_blkring;
1785 
1786 	for_each_rinfo(info, rinfo, i) {
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, 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_each_rinfo(info, rinfo, i) {
1834 			memset(path, 0, pathsize);
1835 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1836 			err = write_per_ring_nodes(xbt, rinfo, 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, "feature-persistent", "%u",
1851 			info->feature_persistent);
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_each_rinfo(info, rinfo, i) {
1865 		unsigned int j;
1866 
1867 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1868 			rinfo->shadow[j].req.u.rw.id = j + 1;
1869 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1870 	}
1871 	xenbus_switch_state(dev, XenbusStateInitialised);
1872 
1873 	return 0;
1874 
1875  abort_transaction:
1876 	xenbus_transaction_end(xbt, 1);
1877 	if (message)
1878 		xenbus_dev_fatal(dev, err, "%s", message);
1879  destroy_blkring:
1880 	blkif_free(info, 0);
1881 	return err;
1882 }
1883 
1884 static int negotiate_mq(struct blkfront_info *info)
1885 {
1886 	unsigned int backend_max_queues;
1887 	unsigned int i;
1888 	struct blkfront_ring_info *rinfo;
1889 
1890 	BUG_ON(info->nr_rings);
1891 
1892 	/* Check if backend supports multiple queues. */
1893 	backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1894 						  "multi-queue-max-queues", 1);
1895 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1896 	/* We need at least one ring. */
1897 	if (!info->nr_rings)
1898 		info->nr_rings = 1;
1899 
1900 	info->rinfo_size = struct_size(info->rinfo, shadow,
1901 				       BLK_RING_SIZE(info));
1902 	info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1903 	if (!info->rinfo) {
1904 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1905 		info->nr_rings = 0;
1906 		return -ENOMEM;
1907 	}
1908 
1909 	for_each_rinfo(info, rinfo, i) {
1910 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1911 		INIT_LIST_HEAD(&rinfo->grants);
1912 		rinfo->dev_info = info;
1913 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1914 		spin_lock_init(&rinfo->ring_lock);
1915 	}
1916 	return 0;
1917 }
1918 
1919 /* Enable the persistent grants feature. */
1920 static bool feature_persistent = true;
1921 module_param(feature_persistent, bool, 0644);
1922 MODULE_PARM_DESC(feature_persistent,
1923 		"Enables the persistent grants feature");
1924 
1925 /*
1926  * Entry point to this code when a new device is created.  Allocate the basic
1927  * structures and the ring buffer for communication with the backend, and
1928  * inform the backend of the appropriate details for those.  Switch to
1929  * Initialised state.
1930  */
1931 static int blkfront_probe(struct xenbus_device *dev,
1932 			  const struct xenbus_device_id *id)
1933 {
1934 	int err, vdevice;
1935 	struct blkfront_info *info;
1936 
1937 	/* FIXME: Use dynamic device id if this is not set. */
1938 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1939 			   "virtual-device", "%i", &vdevice);
1940 	if (err != 1) {
1941 		/* go looking in the extended area instead */
1942 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1943 				   "%i", &vdevice);
1944 		if (err != 1) {
1945 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1946 			return err;
1947 		}
1948 	}
1949 
1950 	if (xen_hvm_domain()) {
1951 		char *type;
1952 		int len;
1953 		/* no unplug has been done: do not hook devices != xen vbds */
1954 		if (xen_has_pv_and_legacy_disk_devices()) {
1955 			int major;
1956 
1957 			if (!VDEV_IS_EXTENDED(vdevice))
1958 				major = BLKIF_MAJOR(vdevice);
1959 			else
1960 				major = XENVBD_MAJOR;
1961 
1962 			if (major != XENVBD_MAJOR) {
1963 				printk(KERN_INFO
1964 						"%s: HVM does not support vbd %d as xen block device\n",
1965 						__func__, vdevice);
1966 				return -ENODEV;
1967 			}
1968 		}
1969 		/* do not create a PV cdrom device if we are an HVM guest */
1970 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1971 		if (IS_ERR(type))
1972 			return -ENODEV;
1973 		if (strncmp(type, "cdrom", 5) == 0) {
1974 			kfree(type);
1975 			return -ENODEV;
1976 		}
1977 		kfree(type);
1978 	}
1979 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1980 	if (!info) {
1981 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1982 		return -ENOMEM;
1983 	}
1984 
1985 	info->xbdev = dev;
1986 
1987 	mutex_init(&info->mutex);
1988 	info->vdevice = vdevice;
1989 	info->connected = BLKIF_STATE_DISCONNECTED;
1990 
1991 	info->feature_persistent = feature_persistent;
1992 
1993 	/* Front end dir is a number, which is used as the id. */
1994 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1995 	dev_set_drvdata(&dev->dev, info);
1996 
1997 	mutex_lock(&blkfront_mutex);
1998 	list_add(&info->info_list, &info_list);
1999 	mutex_unlock(&blkfront_mutex);
2000 
2001 	return 0;
2002 }
2003 
2004 static int blkif_recover(struct blkfront_info *info)
2005 {
2006 	unsigned int r_index;
2007 	struct request *req, *n;
2008 	int rc;
2009 	struct bio *bio;
2010 	unsigned int segs;
2011 	struct blkfront_ring_info *rinfo;
2012 
2013 	blkfront_gather_backend_features(info);
2014 	/* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2015 	blkif_set_queue_limits(info);
2016 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2017 	blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2018 
2019 	for_each_rinfo(info, rinfo, r_index) {
2020 		rc = blkfront_setup_indirect(rinfo);
2021 		if (rc)
2022 			return rc;
2023 	}
2024 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2025 
2026 	/* Now safe for us to use the shared ring */
2027 	info->connected = BLKIF_STATE_CONNECTED;
2028 
2029 	for_each_rinfo(info, rinfo, r_index) {
2030 		/* Kick any other new requests queued since we resumed */
2031 		kick_pending_request_queues(rinfo);
2032 	}
2033 
2034 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2035 		/* Requeue pending requests (flush or discard) */
2036 		list_del_init(&req->queuelist);
2037 		BUG_ON(req->nr_phys_segments > segs);
2038 		blk_mq_requeue_request(req, false);
2039 	}
2040 	blk_mq_start_stopped_hw_queues(info->rq, true);
2041 	blk_mq_kick_requeue_list(info->rq);
2042 
2043 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2044 		/* Traverse the list of pending bios and re-queue them */
2045 		submit_bio(bio);
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 /*
2052  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2053  * driver restart.  We tear down our blkif structure and recreate it, but
2054  * leave the device-layer structures intact so that this is transparent to the
2055  * rest of the kernel.
2056  */
2057 static int blkfront_resume(struct xenbus_device *dev)
2058 {
2059 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2060 	int err = 0;
2061 	unsigned int i, j;
2062 	struct blkfront_ring_info *rinfo;
2063 
2064 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2065 
2066 	bio_list_init(&info->bio_list);
2067 	INIT_LIST_HEAD(&info->requests);
2068 	for_each_rinfo(info, rinfo, i) {
2069 		struct bio_list merge_bio;
2070 		struct blk_shadow *shadow = rinfo->shadow;
2071 
2072 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2073 			/* Not in use? */
2074 			if (!shadow[j].request)
2075 				continue;
2076 
2077 			/*
2078 			 * Get the bios in the request so we can re-queue them.
2079 			 */
2080 			if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2081 			    req_op(shadow[j].request) == REQ_OP_DISCARD ||
2082 			    req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2083 			    shadow[j].request->cmd_flags & REQ_FUA) {
2084 				/*
2085 				 * Flush operations don't contain bios, so
2086 				 * we need to requeue the whole request
2087 				 *
2088 				 * XXX: but this doesn't make any sense for a
2089 				 * write with the FUA flag set..
2090 				 */
2091 				list_add(&shadow[j].request->queuelist, &info->requests);
2092 				continue;
2093 			}
2094 			merge_bio.head = shadow[j].request->bio;
2095 			merge_bio.tail = shadow[j].request->biotail;
2096 			bio_list_merge(&info->bio_list, &merge_bio);
2097 			shadow[j].request->bio = NULL;
2098 			blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2099 		}
2100 	}
2101 
2102 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2103 
2104 	err = talk_to_blkback(dev, info);
2105 	if (!err)
2106 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2107 
2108 	/*
2109 	 * We have to wait for the backend to switch to
2110 	 * connected state, since we want to read which
2111 	 * features it supports.
2112 	 */
2113 
2114 	return err;
2115 }
2116 
2117 static void blkfront_closing(struct blkfront_info *info)
2118 {
2119 	struct xenbus_device *xbdev = info->xbdev;
2120 	struct blkfront_ring_info *rinfo;
2121 	unsigned int i;
2122 
2123 	if (xbdev->state == XenbusStateClosing)
2124 		return;
2125 
2126 	/* No more blkif_request(). */
2127 	if (info->rq && info->gd) {
2128 		blk_mq_stop_hw_queues(info->rq);
2129 		blk_mark_disk_dead(info->gd);
2130 		set_capacity(info->gd, 0);
2131 	}
2132 
2133 	for_each_rinfo(info, rinfo, i) {
2134 		/* No more gnttab callback work. */
2135 		gnttab_cancel_free_callback(&rinfo->callback);
2136 
2137 		/* Flush gnttab callback work. Must be done with no locks held. */
2138 		flush_work(&rinfo->work);
2139 	}
2140 
2141 	xenbus_frontend_closed(xbdev);
2142 }
2143 
2144 static void blkfront_setup_discard(struct blkfront_info *info)
2145 {
2146 	info->feature_discard = 1;
2147 	info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2148 							 "discard-granularity",
2149 							 0);
2150 	info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2151 						       "discard-alignment", 0);
2152 	info->feature_secdiscard =
2153 		!!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2154 				       0);
2155 }
2156 
2157 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2158 {
2159 	unsigned int psegs, grants, memflags;
2160 	int err, i;
2161 	struct blkfront_info *info = rinfo->dev_info;
2162 
2163 	memflags = memalloc_noio_save();
2164 
2165 	if (info->max_indirect_segments == 0) {
2166 		if (!HAS_EXTRA_REQ)
2167 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2168 		else {
2169 			/*
2170 			 * When an extra req is required, the maximum
2171 			 * grants supported is related to the size of the
2172 			 * Linux block segment.
2173 			 */
2174 			grants = GRANTS_PER_PSEG;
2175 		}
2176 	}
2177 	else
2178 		grants = info->max_indirect_segments;
2179 	psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2180 
2181 	err = fill_grant_buffer(rinfo,
2182 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2183 	if (err)
2184 		goto out_of_memory;
2185 
2186 	if (!info->bounce && info->max_indirect_segments) {
2187 		/*
2188 		 * We are using indirect descriptors but don't have a bounce
2189 		 * buffer, we need to allocate a set of pages that can be
2190 		 * used for mapping indirect grefs
2191 		 */
2192 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2193 
2194 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2195 		for (i = 0; i < num; i++) {
2196 			struct page *indirect_page = alloc_page(GFP_KERNEL |
2197 								__GFP_ZERO);
2198 			if (!indirect_page)
2199 				goto out_of_memory;
2200 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2201 		}
2202 	}
2203 
2204 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2205 		rinfo->shadow[i].grants_used =
2206 			kvcalloc(grants,
2207 				 sizeof(rinfo->shadow[i].grants_used[0]),
2208 				 GFP_KERNEL);
2209 		rinfo->shadow[i].sg = kvcalloc(psegs,
2210 					       sizeof(rinfo->shadow[i].sg[0]),
2211 					       GFP_KERNEL);
2212 		if (info->max_indirect_segments)
2213 			rinfo->shadow[i].indirect_grants =
2214 				kvcalloc(INDIRECT_GREFS(grants),
2215 					 sizeof(rinfo->shadow[i].indirect_grants[0]),
2216 					 GFP_KERNEL);
2217 		if ((rinfo->shadow[i].grants_used == NULL) ||
2218 			(rinfo->shadow[i].sg == NULL) ||
2219 		     (info->max_indirect_segments &&
2220 		     (rinfo->shadow[i].indirect_grants == NULL)))
2221 			goto out_of_memory;
2222 		sg_init_table(rinfo->shadow[i].sg, psegs);
2223 	}
2224 
2225 	memalloc_noio_restore(memflags);
2226 
2227 	return 0;
2228 
2229 out_of_memory:
2230 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2231 		kvfree(rinfo->shadow[i].grants_used);
2232 		rinfo->shadow[i].grants_used = NULL;
2233 		kvfree(rinfo->shadow[i].sg);
2234 		rinfo->shadow[i].sg = NULL;
2235 		kvfree(rinfo->shadow[i].indirect_grants);
2236 		rinfo->shadow[i].indirect_grants = NULL;
2237 	}
2238 	if (!list_empty(&rinfo->indirect_pages)) {
2239 		struct page *indirect_page, *n;
2240 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2241 			list_del(&indirect_page->lru);
2242 			__free_page(indirect_page);
2243 		}
2244 	}
2245 
2246 	memalloc_noio_restore(memflags);
2247 
2248 	return -ENOMEM;
2249 }
2250 
2251 /*
2252  * Gather all backend feature-*
2253  */
2254 static void blkfront_gather_backend_features(struct blkfront_info *info)
2255 {
2256 	unsigned int indirect_segments;
2257 
2258 	info->feature_flush = 0;
2259 	info->feature_fua = 0;
2260 
2261 	/*
2262 	 * If there's no "feature-barrier" defined, then it means
2263 	 * we're dealing with a very old backend which writes
2264 	 * synchronously; nothing to do.
2265 	 *
2266 	 * If there are barriers, then we use flush.
2267 	 */
2268 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2269 		info->feature_flush = 1;
2270 		info->feature_fua = 1;
2271 	}
2272 
2273 	/*
2274 	 * And if there is "feature-flush-cache" use that above
2275 	 * barriers.
2276 	 */
2277 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2278 				 0)) {
2279 		info->feature_flush = 1;
2280 		info->feature_fua = 0;
2281 	}
2282 
2283 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2284 		blkfront_setup_discard(info);
2285 
2286 	if (info->feature_persistent)
2287 		info->feature_persistent =
2288 			!!xenbus_read_unsigned(info->xbdev->otherend,
2289 					       "feature-persistent", 0);
2290 	if (info->feature_persistent)
2291 		info->bounce = true;
2292 
2293 	indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2294 					"feature-max-indirect-segments", 0);
2295 	if (indirect_segments > xen_blkif_max_segments)
2296 		indirect_segments = xen_blkif_max_segments;
2297 	if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2298 		indirect_segments = 0;
2299 	info->max_indirect_segments = indirect_segments;
2300 
2301 	if (info->feature_persistent) {
2302 		mutex_lock(&blkfront_mutex);
2303 		schedule_delayed_work(&blkfront_work, HZ * 10);
2304 		mutex_unlock(&blkfront_mutex);
2305 	}
2306 }
2307 
2308 /*
2309  * Invoked when the backend is finally 'ready' (and has told produced
2310  * the details about the physical device - #sectors, size, etc).
2311  */
2312 static void blkfront_connect(struct blkfront_info *info)
2313 {
2314 	unsigned long long sectors;
2315 	unsigned long sector_size;
2316 	unsigned int physical_sector_size;
2317 	int err, i;
2318 	struct blkfront_ring_info *rinfo;
2319 
2320 	switch (info->connected) {
2321 	case BLKIF_STATE_CONNECTED:
2322 		/*
2323 		 * Potentially, the back-end may be signalling
2324 		 * a capacity change; update the capacity.
2325 		 */
2326 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2327 				   "sectors", "%Lu", &sectors);
2328 		if (XENBUS_EXIST_ERR(err))
2329 			return;
2330 		printk(KERN_INFO "Setting capacity to %Lu\n",
2331 		       sectors);
2332 		set_capacity_and_notify(info->gd, sectors);
2333 
2334 		return;
2335 	case BLKIF_STATE_SUSPENDED:
2336 		/*
2337 		 * If we are recovering from suspension, we need to wait
2338 		 * for the backend to announce it's features before
2339 		 * reconnecting, at least we need to know if the backend
2340 		 * supports indirect descriptors, and how many.
2341 		 */
2342 		blkif_recover(info);
2343 		return;
2344 
2345 	default:
2346 		break;
2347 	}
2348 
2349 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2350 		__func__, info->xbdev->otherend);
2351 
2352 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2353 			    "sectors", "%llu", &sectors,
2354 			    "info", "%u", &info->vdisk_info,
2355 			    "sector-size", "%lu", &sector_size,
2356 			    NULL);
2357 	if (err) {
2358 		xenbus_dev_fatal(info->xbdev, err,
2359 				 "reading backend fields at %s",
2360 				 info->xbdev->otherend);
2361 		return;
2362 	}
2363 
2364 	/*
2365 	 * physical-sector-size is a newer field, so old backends may not
2366 	 * provide this. Assume physical sector size to be the same as
2367 	 * sector_size in that case.
2368 	 */
2369 	physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2370 						    "physical-sector-size",
2371 						    sector_size);
2372 	blkfront_gather_backend_features(info);
2373 	for_each_rinfo(info, rinfo, i) {
2374 		err = blkfront_setup_indirect(rinfo);
2375 		if (err) {
2376 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2377 					 info->xbdev->otherend);
2378 			blkif_free(info, 0);
2379 			break;
2380 		}
2381 	}
2382 
2383 	err = xlvbd_alloc_gendisk(sectors, info, sector_size,
2384 				  physical_sector_size);
2385 	if (err) {
2386 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2387 				 info->xbdev->otherend);
2388 		goto fail;
2389 	}
2390 
2391 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2392 
2393 	/* Kick pending requests. */
2394 	info->connected = BLKIF_STATE_CONNECTED;
2395 	for_each_rinfo(info, rinfo, i)
2396 		kick_pending_request_queues(rinfo);
2397 
2398 	err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2399 	if (err) {
2400 		blk_cleanup_disk(info->gd);
2401 		blk_mq_free_tag_set(&info->tag_set);
2402 		info->rq = NULL;
2403 		goto fail;
2404 	}
2405 
2406 	info->is_ready = 1;
2407 	return;
2408 
2409 fail:
2410 	blkif_free(info, 0);
2411 	return;
2412 }
2413 
2414 /*
2415  * Callback received when the backend's state changes.
2416  */
2417 static void blkback_changed(struct xenbus_device *dev,
2418 			    enum xenbus_state backend_state)
2419 {
2420 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2421 
2422 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2423 
2424 	switch (backend_state) {
2425 	case XenbusStateInitWait:
2426 		if (dev->state != XenbusStateInitialising)
2427 			break;
2428 		if (talk_to_blkback(dev, info))
2429 			break;
2430 		break;
2431 	case XenbusStateInitialising:
2432 	case XenbusStateInitialised:
2433 	case XenbusStateReconfiguring:
2434 	case XenbusStateReconfigured:
2435 	case XenbusStateUnknown:
2436 		break;
2437 
2438 	case XenbusStateConnected:
2439 		/*
2440 		 * talk_to_blkback sets state to XenbusStateInitialised
2441 		 * and blkfront_connect sets it to XenbusStateConnected
2442 		 * (if connection went OK).
2443 		 *
2444 		 * If the backend (or toolstack) decides to poke at backend
2445 		 * state (and re-trigger the watch by setting the state repeatedly
2446 		 * to XenbusStateConnected (4)) we need to deal with this.
2447 		 * This is allowed as this is used to communicate to the guest
2448 		 * that the size of disk has changed!
2449 		 */
2450 		if ((dev->state != XenbusStateInitialised) &&
2451 		    (dev->state != XenbusStateConnected)) {
2452 			if (talk_to_blkback(dev, info))
2453 				break;
2454 		}
2455 
2456 		blkfront_connect(info);
2457 		break;
2458 
2459 	case XenbusStateClosed:
2460 		if (dev->state == XenbusStateClosed)
2461 			break;
2462 		fallthrough;
2463 	case XenbusStateClosing:
2464 		blkfront_closing(info);
2465 		break;
2466 	}
2467 }
2468 
2469 static int blkfront_remove(struct xenbus_device *xbdev)
2470 {
2471 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2472 
2473 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2474 
2475 	if (info->gd)
2476 		del_gendisk(info->gd);
2477 
2478 	mutex_lock(&blkfront_mutex);
2479 	list_del(&info->info_list);
2480 	mutex_unlock(&blkfront_mutex);
2481 
2482 	blkif_free(info, 0);
2483 	if (info->gd) {
2484 		xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2485 		blk_cleanup_disk(info->gd);
2486 		blk_mq_free_tag_set(&info->tag_set);
2487 	}
2488 
2489 	kfree(info);
2490 	return 0;
2491 }
2492 
2493 static int blkfront_is_ready(struct xenbus_device *dev)
2494 {
2495 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2496 
2497 	return info->is_ready && info->xbdev;
2498 }
2499 
2500 static const struct block_device_operations xlvbd_block_fops =
2501 {
2502 	.owner = THIS_MODULE,
2503 	.getgeo = blkif_getgeo,
2504 	.ioctl = blkif_ioctl,
2505 	.compat_ioctl = blkdev_compat_ptr_ioctl,
2506 };
2507 
2508 
2509 static const struct xenbus_device_id blkfront_ids[] = {
2510 	{ "vbd" },
2511 	{ "" }
2512 };
2513 
2514 static struct xenbus_driver blkfront_driver = {
2515 	.ids  = blkfront_ids,
2516 	.probe = blkfront_probe,
2517 	.remove = blkfront_remove,
2518 	.resume = blkfront_resume,
2519 	.otherend_changed = blkback_changed,
2520 	.is_ready = blkfront_is_ready,
2521 };
2522 
2523 static void purge_persistent_grants(struct blkfront_info *info)
2524 {
2525 	unsigned int i;
2526 	unsigned long flags;
2527 	struct blkfront_ring_info *rinfo;
2528 
2529 	for_each_rinfo(info, rinfo, i) {
2530 		struct grant *gnt_list_entry, *tmp;
2531 		LIST_HEAD(grants);
2532 
2533 		spin_lock_irqsave(&rinfo->ring_lock, flags);
2534 
2535 		if (rinfo->persistent_gnts_c == 0) {
2536 			spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2537 			continue;
2538 		}
2539 
2540 		list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2541 					 node) {
2542 			if (gnt_list_entry->gref == INVALID_GRANT_REF ||
2543 			    !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2544 				continue;
2545 
2546 			list_del(&gnt_list_entry->node);
2547 			rinfo->persistent_gnts_c--;
2548 			gnt_list_entry->gref = INVALID_GRANT_REF;
2549 			list_add_tail(&gnt_list_entry->node, &grants);
2550 		}
2551 
2552 		list_splice_tail(&grants, &rinfo->grants);
2553 
2554 		spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2555 	}
2556 }
2557 
2558 static void blkfront_delay_work(struct work_struct *work)
2559 {
2560 	struct blkfront_info *info;
2561 	bool need_schedule_work = false;
2562 
2563 	/*
2564 	 * Note that when using bounce buffers but not persistent grants
2565 	 * there's no need to run blkfront_delay_work because grants are
2566 	 * revoked in blkif_completion or else an error is reported and the
2567 	 * connection is closed.
2568 	 */
2569 
2570 	mutex_lock(&blkfront_mutex);
2571 
2572 	list_for_each_entry(info, &info_list, info_list) {
2573 		if (info->feature_persistent) {
2574 			need_schedule_work = true;
2575 			mutex_lock(&info->mutex);
2576 			purge_persistent_grants(info);
2577 			mutex_unlock(&info->mutex);
2578 		}
2579 	}
2580 
2581 	if (need_schedule_work)
2582 		schedule_delayed_work(&blkfront_work, HZ * 10);
2583 
2584 	mutex_unlock(&blkfront_mutex);
2585 }
2586 
2587 static int __init xlblk_init(void)
2588 {
2589 	int ret;
2590 	int nr_cpus = num_online_cpus();
2591 
2592 	if (!xen_domain())
2593 		return -ENODEV;
2594 
2595 	if (!xen_has_pv_disk_devices())
2596 		return -ENODEV;
2597 
2598 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2599 		pr_warn("xen_blk: can't get major %d with name %s\n",
2600 			XENVBD_MAJOR, DEV_NAME);
2601 		return -ENODEV;
2602 	}
2603 
2604 	if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2605 		xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2606 
2607 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2608 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2609 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2610 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2611 	}
2612 
2613 	if (xen_blkif_max_queues > nr_cpus) {
2614 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2615 			xen_blkif_max_queues, nr_cpus);
2616 		xen_blkif_max_queues = nr_cpus;
2617 	}
2618 
2619 	INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2620 
2621 	ret = xenbus_register_frontend(&blkfront_driver);
2622 	if (ret) {
2623 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2624 		return ret;
2625 	}
2626 
2627 	return 0;
2628 }
2629 module_init(xlblk_init);
2630 
2631 
2632 static void __exit xlblk_exit(void)
2633 {
2634 	cancel_delayed_work_sync(&blkfront_work);
2635 
2636 	xenbus_unregister_driver(&blkfront_driver);
2637 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2638 	kfree(minors);
2639 }
2640 module_exit(xlblk_exit);
2641 
2642 MODULE_DESCRIPTION("Xen virtual block device frontend");
2643 MODULE_LICENSE("GPL");
2644 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2645 MODULE_ALIAS("xen:vbd");
2646 MODULE_ALIAS("xenblk");
2647