xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 2f0754f2)
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 #define BLK_RING_SIZE(info)	\
156 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
157 
158 /*
159  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
160  * characters are enough. Define to 20 to keep consistent with backend.
161  */
162 #define RINGREF_NAME_LEN (20)
163 /*
164  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
165  */
166 #define QUEUE_NAME_LEN (17)
167 
168 /*
169  *  Per-ring info.
170  *  Every blkfront device can associate with one or more blkfront_ring_info,
171  *  depending on how many hardware queues/rings to be used.
172  */
173 struct blkfront_ring_info {
174 	/* Lock to protect data in every ring buffer. */
175 	spinlock_t ring_lock;
176 	struct blkif_front_ring ring;
177 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
178 	unsigned int evtchn, irq;
179 	struct work_struct work;
180 	struct gnttab_free_callback callback;
181 	struct list_head indirect_pages;
182 	struct list_head grants;
183 	unsigned int persistent_gnts_c;
184 	unsigned long shadow_free;
185 	struct blkfront_info *dev_info;
186 	struct blk_shadow shadow[];
187 };
188 
189 /*
190  * We have one of these per vbd, whether ide, scsi or 'other'.  They
191  * hang in private_data off the gendisk structure. We may end up
192  * putting all kinds of interesting stuff here :-)
193  */
194 struct blkfront_info
195 {
196 	struct mutex mutex;
197 	struct xenbus_device *xbdev;
198 	struct gendisk *gd;
199 	u16 sector_size;
200 	unsigned int physical_sector_size;
201 	unsigned long vdisk_info;
202 	int vdevice;
203 	blkif_vdev_t handle;
204 	enum blkif_state connected;
205 	/* Number of pages per ring buffer. */
206 	unsigned int nr_ring_pages;
207 	struct request_queue *rq;
208 	unsigned int feature_flush:1;
209 	unsigned int feature_fua:1;
210 	unsigned int feature_discard:1;
211 	unsigned int feature_secdiscard:1;
212 	unsigned int feature_persistent:1;
213 	unsigned int discard_granularity;
214 	unsigned int discard_alignment;
215 	/* Number of 4KB segments handled */
216 	unsigned int max_indirect_segments;
217 	int is_ready;
218 	struct blk_mq_tag_set tag_set;
219 	struct blkfront_ring_info *rinfo;
220 	unsigned int nr_rings;
221 	unsigned int rinfo_size;
222 	/* Save uncomplete reqs and bios for migration. */
223 	struct list_head requests;
224 	struct bio_list bio_list;
225 	struct list_head info_list;
226 };
227 
228 static unsigned int nr_minors;
229 static unsigned long *minors;
230 static DEFINE_SPINLOCK(minor_lock);
231 
232 #define GRANT_INVALID_REF	0
233 
234 #define PARTS_PER_DISK		16
235 #define PARTS_PER_EXT_DISK      256
236 
237 #define BLKIF_MAJOR(dev) ((dev)>>8)
238 #define BLKIF_MINOR(dev) ((dev) & 0xff)
239 
240 #define EXT_SHIFT 28
241 #define EXTENDED (1<<EXT_SHIFT)
242 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
243 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
244 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
245 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
246 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
247 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
248 
249 #define DEV_NAME	"xvd"	/* name in /dev */
250 
251 /*
252  * Grants are always the same size as a Xen page (i.e 4KB).
253  * A physical segment is always the same size as a Linux page.
254  * Number of grants per physical segment
255  */
256 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
257 
258 #define GRANTS_PER_INDIRECT_FRAME \
259 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
260 
261 #define INDIRECT_GREFS(_grants)		\
262 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
263 
264 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
265 static void blkfront_gather_backend_features(struct blkfront_info *info);
266 static int negotiate_mq(struct blkfront_info *info);
267 
268 #define for_each_rinfo(info, ptr, idx)				\
269 	for ((ptr) = (info)->rinfo, (idx) = 0;			\
270 	     (idx) < (info)->nr_rings;				\
271 	     (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
272 
273 static inline struct blkfront_ring_info *
274 get_rinfo(const struct blkfront_info *info, unsigned int i)
275 {
276 	BUG_ON(i >= info->nr_rings);
277 	return (void *)info->rinfo + i * info->rinfo_size;
278 }
279 
280 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
281 {
282 	unsigned long free = rinfo->shadow_free;
283 
284 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
285 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
286 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
287 	return free;
288 }
289 
290 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
291 			      unsigned long id)
292 {
293 	if (rinfo->shadow[id].req.u.rw.id != id)
294 		return -EINVAL;
295 	if (rinfo->shadow[id].request == NULL)
296 		return -EINVAL;
297 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
298 	rinfo->shadow[id].request = NULL;
299 	rinfo->shadow_free = id;
300 	return 0;
301 }
302 
303 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
304 {
305 	struct blkfront_info *info = rinfo->dev_info;
306 	struct page *granted_page;
307 	struct grant *gnt_list_entry, *n;
308 	int i = 0;
309 
310 	while (i < num) {
311 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
312 		if (!gnt_list_entry)
313 			goto out_of_memory;
314 
315 		if (info->feature_persistent) {
316 			granted_page = alloc_page(GFP_NOIO);
317 			if (!granted_page) {
318 				kfree(gnt_list_entry);
319 				goto out_of_memory;
320 			}
321 			gnt_list_entry->page = granted_page;
322 		}
323 
324 		gnt_list_entry->gref = GRANT_INVALID_REF;
325 		list_add(&gnt_list_entry->node, &rinfo->grants);
326 		i++;
327 	}
328 
329 	return 0;
330 
331 out_of_memory:
332 	list_for_each_entry_safe(gnt_list_entry, n,
333 	                         &rinfo->grants, node) {
334 		list_del(&gnt_list_entry->node);
335 		if (info->feature_persistent)
336 			__free_page(gnt_list_entry->page);
337 		kfree(gnt_list_entry);
338 		i--;
339 	}
340 	BUG_ON(i != 0);
341 	return -ENOMEM;
342 }
343 
344 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
345 {
346 	struct grant *gnt_list_entry;
347 
348 	BUG_ON(list_empty(&rinfo->grants));
349 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
350 					  node);
351 	list_del(&gnt_list_entry->node);
352 
353 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
354 		rinfo->persistent_gnts_c--;
355 
356 	return gnt_list_entry;
357 }
358 
359 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
360 					const struct blkfront_info *info)
361 {
362 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
363 						 info->xbdev->otherend_id,
364 						 gnt_list_entry->page,
365 						 0);
366 }
367 
368 static struct grant *get_grant(grant_ref_t *gref_head,
369 			       unsigned long gfn,
370 			       struct blkfront_ring_info *rinfo)
371 {
372 	struct grant *gnt_list_entry = get_free_grant(rinfo);
373 	struct blkfront_info *info = rinfo->dev_info;
374 
375 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
376 		return gnt_list_entry;
377 
378 	/* Assign a gref to this page */
379 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
380 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
381 	if (info->feature_persistent)
382 		grant_foreign_access(gnt_list_entry, info);
383 	else {
384 		/* Grant access to the GFN passed by the caller */
385 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
386 						info->xbdev->otherend_id,
387 						gfn, 0);
388 	}
389 
390 	return gnt_list_entry;
391 }
392 
393 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
394 					struct blkfront_ring_info *rinfo)
395 {
396 	struct grant *gnt_list_entry = get_free_grant(rinfo);
397 	struct blkfront_info *info = rinfo->dev_info;
398 
399 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
400 		return gnt_list_entry;
401 
402 	/* Assign a gref to this page */
403 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
404 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
405 	if (!info->feature_persistent) {
406 		struct page *indirect_page;
407 
408 		/* Fetch a pre-allocated page to use for indirect grefs */
409 		BUG_ON(list_empty(&rinfo->indirect_pages));
410 		indirect_page = list_first_entry(&rinfo->indirect_pages,
411 						 struct page, lru);
412 		list_del(&indirect_page->lru);
413 		gnt_list_entry->page = indirect_page;
414 	}
415 	grant_foreign_access(gnt_list_entry, info);
416 
417 	return gnt_list_entry;
418 }
419 
420 static const char *op_name(int op)
421 {
422 	static const char *const names[] = {
423 		[BLKIF_OP_READ] = "read",
424 		[BLKIF_OP_WRITE] = "write",
425 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
426 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
427 		[BLKIF_OP_DISCARD] = "discard" };
428 
429 	if (op < 0 || op >= ARRAY_SIZE(names))
430 		return "unknown";
431 
432 	if (!names[op])
433 		return "reserved";
434 
435 	return names[op];
436 }
437 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
438 {
439 	unsigned int end = minor + nr;
440 	int rc;
441 
442 	if (end > nr_minors) {
443 		unsigned long *bitmap, *old;
444 
445 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
446 				 GFP_KERNEL);
447 		if (bitmap == NULL)
448 			return -ENOMEM;
449 
450 		spin_lock(&minor_lock);
451 		if (end > nr_minors) {
452 			old = minors;
453 			memcpy(bitmap, minors,
454 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
455 			minors = bitmap;
456 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
457 		} else
458 			old = bitmap;
459 		spin_unlock(&minor_lock);
460 		kfree(old);
461 	}
462 
463 	spin_lock(&minor_lock);
464 	if (find_next_bit(minors, end, minor) >= end) {
465 		bitmap_set(minors, minor, nr);
466 		rc = 0;
467 	} else
468 		rc = -EBUSY;
469 	spin_unlock(&minor_lock);
470 
471 	return rc;
472 }
473 
474 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
475 {
476 	unsigned int end = minor + nr;
477 
478 	BUG_ON(end > nr_minors);
479 	spin_lock(&minor_lock);
480 	bitmap_clear(minors,  minor, nr);
481 	spin_unlock(&minor_lock);
482 }
483 
484 static void blkif_restart_queue_callback(void *arg)
485 {
486 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
487 	schedule_work(&rinfo->work);
488 }
489 
490 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
491 {
492 	/* We don't have real geometry info, but let's at least return
493 	   values consistent with the size of the device */
494 	sector_t nsect = get_capacity(bd->bd_disk);
495 	sector_t cylinders = nsect;
496 
497 	hg->heads = 0xff;
498 	hg->sectors = 0x3f;
499 	sector_div(cylinders, hg->heads * hg->sectors);
500 	hg->cylinders = cylinders;
501 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
502 		hg->cylinders = 0xffff;
503 	return 0;
504 }
505 
506 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
507 		       unsigned command, unsigned long argument)
508 {
509 	struct blkfront_info *info = bdev->bd_disk->private_data;
510 	int i;
511 
512 	switch (command) {
513 	case CDROMMULTISESSION:
514 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
515 			if (put_user(0, (char __user *)(argument + i)))
516 				return -EFAULT;
517 		return 0;
518 	case CDROM_GET_CAPABILITY:
519 		if (!(info->vdisk_info & VDISK_CDROM))
520 			return -EINVAL;
521 		return 0;
522 	default:
523 		return -EINVAL;
524 	}
525 }
526 
527 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
528 					    struct request *req,
529 					    struct blkif_request **ring_req)
530 {
531 	unsigned long id;
532 
533 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
534 	rinfo->ring.req_prod_pvt++;
535 
536 	id = get_id_from_freelist(rinfo);
537 	rinfo->shadow[id].request = req;
538 	rinfo->shadow[id].status = REQ_PROCESSING;
539 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
540 
541 	rinfo->shadow[id].req.u.rw.id = id;
542 
543 	return id;
544 }
545 
546 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
547 {
548 	struct blkfront_info *info = rinfo->dev_info;
549 	struct blkif_request *ring_req, *final_ring_req;
550 	unsigned long id;
551 
552 	/* Fill out a communications ring structure. */
553 	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
554 	ring_req = &rinfo->shadow[id].req;
555 
556 	ring_req->operation = BLKIF_OP_DISCARD;
557 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
558 	ring_req->u.discard.id = id;
559 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
560 	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
561 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
562 	else
563 		ring_req->u.discard.flag = 0;
564 
565 	/* Copy the request to the ring page. */
566 	*final_ring_req = *ring_req;
567 	rinfo->shadow[id].status = REQ_WAITING;
568 
569 	return 0;
570 }
571 
572 struct setup_rw_req {
573 	unsigned int grant_idx;
574 	struct blkif_request_segment *segments;
575 	struct blkfront_ring_info *rinfo;
576 	struct blkif_request *ring_req;
577 	grant_ref_t gref_head;
578 	unsigned int id;
579 	/* Only used when persistent grant is used and it's a read request */
580 	bool need_copy;
581 	unsigned int bvec_off;
582 	char *bvec_data;
583 
584 	bool require_extra_req;
585 	struct blkif_request *extra_ring_req;
586 };
587 
588 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
589 				     unsigned int len, void *data)
590 {
591 	struct setup_rw_req *setup = data;
592 	int n, ref;
593 	struct grant *gnt_list_entry;
594 	unsigned int fsect, lsect;
595 	/* Convenient aliases */
596 	unsigned int grant_idx = setup->grant_idx;
597 	struct blkif_request *ring_req = setup->ring_req;
598 	struct blkfront_ring_info *rinfo = setup->rinfo;
599 	/*
600 	 * We always use the shadow of the first request to store the list
601 	 * of grant associated to the block I/O request. This made the
602 	 * completion more easy to handle even if the block I/O request is
603 	 * split.
604 	 */
605 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
606 
607 	if (unlikely(setup->require_extra_req &&
608 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
609 		/*
610 		 * We are using the second request, setup grant_idx
611 		 * to be the index of the segment array.
612 		 */
613 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
614 		ring_req = setup->extra_ring_req;
615 	}
616 
617 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
618 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
619 		if (setup->segments)
620 			kunmap_atomic(setup->segments);
621 
622 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
623 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
624 		shadow->indirect_grants[n] = gnt_list_entry;
625 		setup->segments = kmap_atomic(gnt_list_entry->page);
626 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
627 	}
628 
629 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
630 	ref = gnt_list_entry->gref;
631 	/*
632 	 * All the grants are stored in the shadow of the first
633 	 * request. Therefore we have to use the global index.
634 	 */
635 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
636 
637 	if (setup->need_copy) {
638 		void *shared_data;
639 
640 		shared_data = kmap_atomic(gnt_list_entry->page);
641 		/*
642 		 * this does not wipe data stored outside the
643 		 * range sg->offset..sg->offset+sg->length.
644 		 * Therefore, blkback *could* see data from
645 		 * previous requests. This is OK as long as
646 		 * persistent grants are shared with just one
647 		 * domain. It may need refactoring if this
648 		 * changes
649 		 */
650 		memcpy(shared_data + offset,
651 		       setup->bvec_data + setup->bvec_off,
652 		       len);
653 
654 		kunmap_atomic(shared_data);
655 		setup->bvec_off += len;
656 	}
657 
658 	fsect = offset >> 9;
659 	lsect = fsect + (len >> 9) - 1;
660 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
661 		ring_req->u.rw.seg[grant_idx] =
662 			(struct blkif_request_segment) {
663 				.gref       = ref,
664 				.first_sect = fsect,
665 				.last_sect  = lsect };
666 	} else {
667 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
668 			(struct blkif_request_segment) {
669 				.gref       = ref,
670 				.first_sect = fsect,
671 				.last_sect  = lsect };
672 	}
673 
674 	(setup->grant_idx)++;
675 }
676 
677 static void blkif_setup_extra_req(struct blkif_request *first,
678 				  struct blkif_request *second)
679 {
680 	uint16_t nr_segments = first->u.rw.nr_segments;
681 
682 	/*
683 	 * The second request is only present when the first request uses
684 	 * all its segments. It's always the continuity of the first one.
685 	 */
686 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
687 
688 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
689 	second->u.rw.sector_number = first->u.rw.sector_number +
690 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
691 
692 	second->u.rw.handle = first->u.rw.handle;
693 	second->operation = first->operation;
694 }
695 
696 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
697 {
698 	struct blkfront_info *info = rinfo->dev_info;
699 	struct blkif_request *ring_req, *extra_ring_req = NULL;
700 	struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
701 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
702 	bool require_extra_req = false;
703 	int i;
704 	struct setup_rw_req setup = {
705 		.grant_idx = 0,
706 		.segments = NULL,
707 		.rinfo = rinfo,
708 		.need_copy = rq_data_dir(req) && info->feature_persistent,
709 	};
710 
711 	/*
712 	 * Used to store if we are able to queue the request by just using
713 	 * existing persistent grants, or if we have to get new grants,
714 	 * as there are not sufficiently many free.
715 	 */
716 	bool new_persistent_gnts = false;
717 	struct scatterlist *sg;
718 	int num_sg, max_grefs, num_grant;
719 
720 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
721 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
722 		/*
723 		 * If we are using indirect segments we need to account
724 		 * for the indirect grefs used in the request.
725 		 */
726 		max_grefs += INDIRECT_GREFS(max_grefs);
727 
728 	/* Check if we have enough persistent grants to allocate a requests */
729 	if (rinfo->persistent_gnts_c < max_grefs) {
730 		new_persistent_gnts = true;
731 
732 		if (gnttab_alloc_grant_references(
733 		    max_grefs - rinfo->persistent_gnts_c,
734 		    &setup.gref_head) < 0) {
735 			gnttab_request_free_callback(
736 				&rinfo->callback,
737 				blkif_restart_queue_callback,
738 				rinfo,
739 				max_grefs - rinfo->persistent_gnts_c);
740 			return 1;
741 		}
742 	}
743 
744 	/* Fill out a communications ring structure. */
745 	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
746 	ring_req = &rinfo->shadow[id].req;
747 
748 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
749 	num_grant = 0;
750 	/* Calculate the number of grant used */
751 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
752 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
753 
754 	require_extra_req = info->max_indirect_segments == 0 &&
755 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
756 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
757 
758 	rinfo->shadow[id].num_sg = num_sg;
759 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
760 	    likely(!require_extra_req)) {
761 		/*
762 		 * The indirect operation can only be a BLKIF_OP_READ or
763 		 * BLKIF_OP_WRITE
764 		 */
765 		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
766 		ring_req->operation = BLKIF_OP_INDIRECT;
767 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
768 			BLKIF_OP_WRITE : BLKIF_OP_READ;
769 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
770 		ring_req->u.indirect.handle = info->handle;
771 		ring_req->u.indirect.nr_segments = num_grant;
772 	} else {
773 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
774 		ring_req->u.rw.handle = info->handle;
775 		ring_req->operation = rq_data_dir(req) ?
776 			BLKIF_OP_WRITE : BLKIF_OP_READ;
777 		if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
778 			/*
779 			 * Ideally we can do an unordered flush-to-disk.
780 			 * In case the backend onlysupports barriers, use that.
781 			 * A barrier request a superset of FUA, so we can
782 			 * implement it the same way.  (It's also a FLUSH+FUA,
783 			 * since it is guaranteed ordered WRT previous writes.)
784 			 */
785 			if (info->feature_flush && info->feature_fua)
786 				ring_req->operation =
787 					BLKIF_OP_WRITE_BARRIER;
788 			else if (info->feature_flush)
789 				ring_req->operation =
790 					BLKIF_OP_FLUSH_DISKCACHE;
791 			else
792 				ring_req->operation = 0;
793 		}
794 		ring_req->u.rw.nr_segments = num_grant;
795 		if (unlikely(require_extra_req)) {
796 			extra_id = blkif_ring_get_request(rinfo, req,
797 							  &final_extra_ring_req);
798 			extra_ring_req = &rinfo->shadow[extra_id].req;
799 
800 			/*
801 			 * Only the first request contains the scatter-gather
802 			 * list.
803 			 */
804 			rinfo->shadow[extra_id].num_sg = 0;
805 
806 			blkif_setup_extra_req(ring_req, extra_ring_req);
807 
808 			/* Link the 2 requests together */
809 			rinfo->shadow[extra_id].associated_id = id;
810 			rinfo->shadow[id].associated_id = extra_id;
811 		}
812 	}
813 
814 	setup.ring_req = ring_req;
815 	setup.id = id;
816 
817 	setup.require_extra_req = require_extra_req;
818 	if (unlikely(require_extra_req))
819 		setup.extra_ring_req = extra_ring_req;
820 
821 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
822 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
823 
824 		if (setup.need_copy) {
825 			setup.bvec_off = sg->offset;
826 			setup.bvec_data = kmap_atomic(sg_page(sg));
827 		}
828 
829 		gnttab_foreach_grant_in_range(sg_page(sg),
830 					      sg->offset,
831 					      sg->length,
832 					      blkif_setup_rw_req_grant,
833 					      &setup);
834 
835 		if (setup.need_copy)
836 			kunmap_atomic(setup.bvec_data);
837 	}
838 	if (setup.segments)
839 		kunmap_atomic(setup.segments);
840 
841 	/* Copy request(s) to the ring page. */
842 	*final_ring_req = *ring_req;
843 	rinfo->shadow[id].status = REQ_WAITING;
844 	if (unlikely(require_extra_req)) {
845 		*final_extra_ring_req = *extra_ring_req;
846 		rinfo->shadow[extra_id].status = REQ_WAITING;
847 	}
848 
849 	if (new_persistent_gnts)
850 		gnttab_free_grant_references(setup.gref_head);
851 
852 	return 0;
853 }
854 
855 /*
856  * Generate a Xen blkfront IO request from a blk layer request.  Reads
857  * and writes are handled as expected.
858  *
859  * @req: a request struct
860  */
861 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
862 {
863 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
864 		return 1;
865 
866 	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
867 		     req_op(req) == REQ_OP_SECURE_ERASE))
868 		return blkif_queue_discard_req(req, rinfo);
869 	else
870 		return blkif_queue_rw_req(req, rinfo);
871 }
872 
873 static inline void flush_requests(struct blkfront_ring_info *rinfo)
874 {
875 	int notify;
876 
877 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
878 
879 	if (notify)
880 		notify_remote_via_irq(rinfo->irq);
881 }
882 
883 static inline bool blkif_request_flush_invalid(struct request *req,
884 					       struct blkfront_info *info)
885 {
886 	return (blk_rq_is_passthrough(req) ||
887 		((req_op(req) == REQ_OP_FLUSH) &&
888 		 !info->feature_flush) ||
889 		((req->cmd_flags & REQ_FUA) &&
890 		 !info->feature_fua));
891 }
892 
893 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
894 			  const struct blk_mq_queue_data *qd)
895 {
896 	unsigned long flags;
897 	int qid = hctx->queue_num;
898 	struct blkfront_info *info = hctx->queue->queuedata;
899 	struct blkfront_ring_info *rinfo = NULL;
900 
901 	rinfo = get_rinfo(info, qid);
902 	blk_mq_start_request(qd->rq);
903 	spin_lock_irqsave(&rinfo->ring_lock, flags);
904 	if (RING_FULL(&rinfo->ring))
905 		goto out_busy;
906 
907 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
908 		goto out_err;
909 
910 	if (blkif_queue_request(qd->rq, rinfo))
911 		goto out_busy;
912 
913 	flush_requests(rinfo);
914 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
915 	return BLK_STS_OK;
916 
917 out_err:
918 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
919 	return BLK_STS_IOERR;
920 
921 out_busy:
922 	blk_mq_stop_hw_queue(hctx);
923 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
924 	return BLK_STS_DEV_RESOURCE;
925 }
926 
927 static void blkif_complete_rq(struct request *rq)
928 {
929 	blk_mq_end_request(rq, blkif_req(rq)->error);
930 }
931 
932 static const struct blk_mq_ops blkfront_mq_ops = {
933 	.queue_rq = blkif_queue_rq,
934 	.complete = blkif_complete_rq,
935 };
936 
937 static void blkif_set_queue_limits(struct blkfront_info *info)
938 {
939 	struct request_queue *rq = info->rq;
940 	struct gendisk *gd = info->gd;
941 	unsigned int segments = info->max_indirect_segments ? :
942 				BLKIF_MAX_SEGMENTS_PER_REQUEST;
943 
944 	blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
945 
946 	if (info->feature_discard) {
947 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
948 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
949 		rq->limits.discard_granularity = info->discard_granularity ?:
950 						 info->physical_sector_size;
951 		rq->limits.discard_alignment = info->discard_alignment;
952 		if (info->feature_secdiscard)
953 			blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
954 	}
955 
956 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
957 	blk_queue_logical_block_size(rq, info->sector_size);
958 	blk_queue_physical_block_size(rq, info->physical_sector_size);
959 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
960 
961 	/* Each segment in a request is up to an aligned page in size. */
962 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
963 	blk_queue_max_segment_size(rq, PAGE_SIZE);
964 
965 	/* Ensure a merged request will fit in a single I/O ring slot. */
966 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
967 
968 	/* Make sure buffer addresses are sector-aligned. */
969 	blk_queue_dma_alignment(rq, 511);
970 }
971 
972 static const char *flush_info(struct blkfront_info *info)
973 {
974 	if (info->feature_flush && info->feature_fua)
975 		return "barrier: enabled;";
976 	else if (info->feature_flush)
977 		return "flush diskcache: enabled;";
978 	else
979 		return "barrier or flush: disabled;";
980 }
981 
982 static void xlvbd_flush(struct blkfront_info *info)
983 {
984 	blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
985 			      info->feature_fua ? true : false);
986 	pr_info("blkfront: %s: %s %s %s %s %s\n",
987 		info->gd->disk_name, flush_info(info),
988 		"persistent grants:", info->feature_persistent ?
989 		"enabled;" : "disabled;", "indirect descriptors:",
990 		info->max_indirect_segments ? "enabled;" : "disabled;");
991 }
992 
993 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
994 {
995 	int major;
996 	major = BLKIF_MAJOR(vdevice);
997 	*minor = BLKIF_MINOR(vdevice);
998 	switch (major) {
999 		case XEN_IDE0_MAJOR:
1000 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1001 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1002 				EMULATED_HD_DISK_MINOR_OFFSET;
1003 			break;
1004 		case XEN_IDE1_MAJOR:
1005 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1006 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1007 				EMULATED_HD_DISK_MINOR_OFFSET;
1008 			break;
1009 		case XEN_SCSI_DISK0_MAJOR:
1010 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1011 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1012 			break;
1013 		case XEN_SCSI_DISK1_MAJOR:
1014 		case XEN_SCSI_DISK2_MAJOR:
1015 		case XEN_SCSI_DISK3_MAJOR:
1016 		case XEN_SCSI_DISK4_MAJOR:
1017 		case XEN_SCSI_DISK5_MAJOR:
1018 		case XEN_SCSI_DISK6_MAJOR:
1019 		case XEN_SCSI_DISK7_MAJOR:
1020 			*offset = (*minor / PARTS_PER_DISK) +
1021 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1022 				EMULATED_SD_DISK_NAME_OFFSET;
1023 			*minor = *minor +
1024 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1025 				EMULATED_SD_DISK_MINOR_OFFSET;
1026 			break;
1027 		case XEN_SCSI_DISK8_MAJOR:
1028 		case XEN_SCSI_DISK9_MAJOR:
1029 		case XEN_SCSI_DISK10_MAJOR:
1030 		case XEN_SCSI_DISK11_MAJOR:
1031 		case XEN_SCSI_DISK12_MAJOR:
1032 		case XEN_SCSI_DISK13_MAJOR:
1033 		case XEN_SCSI_DISK14_MAJOR:
1034 		case XEN_SCSI_DISK15_MAJOR:
1035 			*offset = (*minor / PARTS_PER_DISK) +
1036 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1037 				EMULATED_SD_DISK_NAME_OFFSET;
1038 			*minor = *minor +
1039 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1040 				EMULATED_SD_DISK_MINOR_OFFSET;
1041 			break;
1042 		case XENVBD_MAJOR:
1043 			*offset = *minor / PARTS_PER_DISK;
1044 			break;
1045 		default:
1046 			printk(KERN_WARNING "blkfront: your disk configuration is "
1047 					"incorrect, please use an xvd device instead\n");
1048 			return -ENODEV;
1049 	}
1050 	return 0;
1051 }
1052 
1053 static char *encode_disk_name(char *ptr, unsigned int n)
1054 {
1055 	if (n >= 26)
1056 		ptr = encode_disk_name(ptr, n / 26 - 1);
1057 	*ptr = 'a' + n % 26;
1058 	return ptr + 1;
1059 }
1060 
1061 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1062 		struct blkfront_info *info, u16 sector_size,
1063 		unsigned int physical_sector_size)
1064 {
1065 	struct gendisk *gd;
1066 	int nr_minors = 1;
1067 	int err;
1068 	unsigned int offset;
1069 	int minor;
1070 	int nr_parts;
1071 	char *ptr;
1072 
1073 	BUG_ON(info->gd != NULL);
1074 	BUG_ON(info->rq != NULL);
1075 
1076 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1077 		/* this is above the extended range; something is wrong */
1078 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1079 		return -ENODEV;
1080 	}
1081 
1082 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1083 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1084 		if (err)
1085 			return err;
1086 		nr_parts = PARTS_PER_DISK;
1087 	} else {
1088 		minor = BLKIF_MINOR_EXT(info->vdevice);
1089 		nr_parts = PARTS_PER_EXT_DISK;
1090 		offset = minor / nr_parts;
1091 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1092 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1093 					"emulated IDE disks,\n\t choose an xvd device name"
1094 					"from xvde on\n", info->vdevice);
1095 	}
1096 	if (minor >> MINORBITS) {
1097 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1098 			info->vdevice, minor);
1099 		return -ENODEV;
1100 	}
1101 
1102 	if ((minor % nr_parts) == 0)
1103 		nr_minors = nr_parts;
1104 
1105 	err = xlbd_reserve_minors(minor, nr_minors);
1106 	if (err)
1107 		return err;
1108 
1109 	memset(&info->tag_set, 0, sizeof(info->tag_set));
1110 	info->tag_set.ops = &blkfront_mq_ops;
1111 	info->tag_set.nr_hw_queues = info->nr_rings;
1112 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1113 		/*
1114 		 * When indirect descriptior is not supported, the I/O request
1115 		 * will be split between multiple request in the ring.
1116 		 * To avoid problems when sending the request, divide by
1117 		 * 2 the depth of the queue.
1118 		 */
1119 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1120 	} else
1121 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
1122 	info->tag_set.numa_node = NUMA_NO_NODE;
1123 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1124 	info->tag_set.cmd_size = sizeof(struct blkif_req);
1125 	info->tag_set.driver_data = info;
1126 
1127 	err = blk_mq_alloc_tag_set(&info->tag_set);
1128 	if (err)
1129 		goto out_release_minors;
1130 
1131 	gd = blk_mq_alloc_disk(&info->tag_set, info);
1132 	if (IS_ERR(gd)) {
1133 		err = PTR_ERR(gd);
1134 		goto out_free_tag_set;
1135 	}
1136 
1137 	strcpy(gd->disk_name, DEV_NAME);
1138 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1139 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1140 	if (nr_minors > 1)
1141 		*ptr = 0;
1142 	else
1143 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1144 			 "%d", minor & (nr_parts - 1));
1145 
1146 	gd->major = XENVBD_MAJOR;
1147 	gd->first_minor = minor;
1148 	gd->minors = nr_minors;
1149 	gd->fops = &xlvbd_block_fops;
1150 	gd->private_data = info;
1151 	set_capacity(gd, capacity);
1152 
1153 	info->rq = gd->queue;
1154 	info->gd = gd;
1155 	info->sector_size = sector_size;
1156 	info->physical_sector_size = physical_sector_size;
1157 	blkif_set_queue_limits(info);
1158 
1159 	xlvbd_flush(info);
1160 
1161 	if (info->vdisk_info & VDISK_READONLY)
1162 		set_disk_ro(gd, 1);
1163 	if (info->vdisk_info & VDISK_REMOVABLE)
1164 		gd->flags |= GENHD_FL_REMOVABLE;
1165 
1166 	return 0;
1167 
1168 out_free_tag_set:
1169 	blk_mq_free_tag_set(&info->tag_set);
1170 out_release_minors:
1171 	xlbd_release_minors(minor, nr_minors);
1172 	return err;
1173 }
1174 
1175 /* Already hold rinfo->ring_lock. */
1176 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1177 {
1178 	if (!RING_FULL(&rinfo->ring))
1179 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1180 }
1181 
1182 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1183 {
1184 	unsigned long flags;
1185 
1186 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1187 	kick_pending_request_queues_locked(rinfo);
1188 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1189 }
1190 
1191 static void blkif_restart_queue(struct work_struct *work)
1192 {
1193 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1194 
1195 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1196 		kick_pending_request_queues(rinfo);
1197 }
1198 
1199 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1200 {
1201 	struct grant *persistent_gnt, *n;
1202 	struct blkfront_info *info = rinfo->dev_info;
1203 	int i, j, segs;
1204 
1205 	/*
1206 	 * Remove indirect pages, this only happens when using indirect
1207 	 * descriptors but not persistent grants
1208 	 */
1209 	if (!list_empty(&rinfo->indirect_pages)) {
1210 		struct page *indirect_page, *n;
1211 
1212 		BUG_ON(info->feature_persistent);
1213 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1214 			list_del(&indirect_page->lru);
1215 			__free_page(indirect_page);
1216 		}
1217 	}
1218 
1219 	/* Remove all persistent grants. */
1220 	if (!list_empty(&rinfo->grants)) {
1221 		list_for_each_entry_safe(persistent_gnt, n,
1222 					 &rinfo->grants, node) {
1223 			list_del(&persistent_gnt->node);
1224 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1225 				gnttab_end_foreign_access(persistent_gnt->gref,
1226 							  0, 0UL);
1227 				rinfo->persistent_gnts_c--;
1228 			}
1229 			if (info->feature_persistent)
1230 				__free_page(persistent_gnt->page);
1231 			kfree(persistent_gnt);
1232 		}
1233 	}
1234 	BUG_ON(rinfo->persistent_gnts_c != 0);
1235 
1236 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1237 		/*
1238 		 * Clear persistent grants present in requests already
1239 		 * on the shared ring
1240 		 */
1241 		if (!rinfo->shadow[i].request)
1242 			goto free_shadow;
1243 
1244 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1245 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1246 		       rinfo->shadow[i].req.u.rw.nr_segments;
1247 		for (j = 0; j < segs; j++) {
1248 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1249 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1250 			if (info->feature_persistent)
1251 				__free_page(persistent_gnt->page);
1252 			kfree(persistent_gnt);
1253 		}
1254 
1255 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1256 			/*
1257 			 * If this is not an indirect operation don't try to
1258 			 * free indirect segments
1259 			 */
1260 			goto free_shadow;
1261 
1262 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1263 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1264 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1265 			__free_page(persistent_gnt->page);
1266 			kfree(persistent_gnt);
1267 		}
1268 
1269 free_shadow:
1270 		kvfree(rinfo->shadow[i].grants_used);
1271 		rinfo->shadow[i].grants_used = NULL;
1272 		kvfree(rinfo->shadow[i].indirect_grants);
1273 		rinfo->shadow[i].indirect_grants = NULL;
1274 		kvfree(rinfo->shadow[i].sg);
1275 		rinfo->shadow[i].sg = NULL;
1276 	}
1277 
1278 	/* No more gnttab callback work. */
1279 	gnttab_cancel_free_callback(&rinfo->callback);
1280 
1281 	/* Flush gnttab callback work. Must be done with no locks held. */
1282 	flush_work(&rinfo->work);
1283 
1284 	/* Free resources associated with old device channel. */
1285 	for (i = 0; i < info->nr_ring_pages; i++) {
1286 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1287 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1288 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1289 		}
1290 	}
1291 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1292 	rinfo->ring.sring = NULL;
1293 
1294 	if (rinfo->irq)
1295 		unbind_from_irqhandler(rinfo->irq, rinfo);
1296 	rinfo->evtchn = rinfo->irq = 0;
1297 }
1298 
1299 static void blkif_free(struct blkfront_info *info, int suspend)
1300 {
1301 	unsigned int i;
1302 	struct blkfront_ring_info *rinfo;
1303 
1304 	/* Prevent new requests being issued until we fix things up. */
1305 	info->connected = suspend ?
1306 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1307 	/* No more blkif_request(). */
1308 	if (info->rq)
1309 		blk_mq_stop_hw_queues(info->rq);
1310 
1311 	for_each_rinfo(info, rinfo, i)
1312 		blkif_free_ring(rinfo);
1313 
1314 	kvfree(info->rinfo);
1315 	info->rinfo = NULL;
1316 	info->nr_rings = 0;
1317 }
1318 
1319 struct copy_from_grant {
1320 	const struct blk_shadow *s;
1321 	unsigned int grant_idx;
1322 	unsigned int bvec_offset;
1323 	char *bvec_data;
1324 };
1325 
1326 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1327 				  unsigned int len, void *data)
1328 {
1329 	struct copy_from_grant *info = data;
1330 	char *shared_data;
1331 	/* Convenient aliases */
1332 	const struct blk_shadow *s = info->s;
1333 
1334 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1335 
1336 	memcpy(info->bvec_data + info->bvec_offset,
1337 	       shared_data + offset, len);
1338 
1339 	info->bvec_offset += len;
1340 	info->grant_idx++;
1341 
1342 	kunmap_atomic(shared_data);
1343 }
1344 
1345 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1346 {
1347 	switch (rsp)
1348 	{
1349 	case BLKIF_RSP_OKAY:
1350 		return REQ_DONE;
1351 	case BLKIF_RSP_EOPNOTSUPP:
1352 		return REQ_EOPNOTSUPP;
1353 	case BLKIF_RSP_ERROR:
1354 	default:
1355 		return REQ_ERROR;
1356 	}
1357 }
1358 
1359 /*
1360  * Get the final status of the block request based on two ring response
1361  */
1362 static int blkif_get_final_status(enum blk_req_status s1,
1363 				  enum blk_req_status s2)
1364 {
1365 	BUG_ON(s1 < REQ_DONE);
1366 	BUG_ON(s2 < REQ_DONE);
1367 
1368 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1369 		return BLKIF_RSP_ERROR;
1370 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1371 		return BLKIF_RSP_EOPNOTSUPP;
1372 	return BLKIF_RSP_OKAY;
1373 }
1374 
1375 static bool blkif_completion(unsigned long *id,
1376 			     struct blkfront_ring_info *rinfo,
1377 			     struct blkif_response *bret)
1378 {
1379 	int i = 0;
1380 	struct scatterlist *sg;
1381 	int num_sg, num_grant;
1382 	struct blkfront_info *info = rinfo->dev_info;
1383 	struct blk_shadow *s = &rinfo->shadow[*id];
1384 	struct copy_from_grant data = {
1385 		.grant_idx = 0,
1386 	};
1387 
1388 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1389 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1390 
1391 	/* The I/O request may be split in two. */
1392 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1393 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1394 
1395 		/* Keep the status of the current response in shadow. */
1396 		s->status = blkif_rsp_to_req_status(bret->status);
1397 
1398 		/* Wait the second response if not yet here. */
1399 		if (s2->status < REQ_DONE)
1400 			return false;
1401 
1402 		bret->status = blkif_get_final_status(s->status,
1403 						      s2->status);
1404 
1405 		/*
1406 		 * All the grants is stored in the first shadow in order
1407 		 * to make the completion code simpler.
1408 		 */
1409 		num_grant += s2->req.u.rw.nr_segments;
1410 
1411 		/*
1412 		 * The two responses may not come in order. Only the
1413 		 * first request will store the scatter-gather list.
1414 		 */
1415 		if (s2->num_sg != 0) {
1416 			/* Update "id" with the ID of the first response. */
1417 			*id = s->associated_id;
1418 			s = s2;
1419 		}
1420 
1421 		/*
1422 		 * We don't need anymore the second request, so recycling
1423 		 * it now.
1424 		 */
1425 		if (add_id_to_freelist(rinfo, s->associated_id))
1426 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1427 			     info->gd->disk_name, s->associated_id);
1428 	}
1429 
1430 	data.s = s;
1431 	num_sg = s->num_sg;
1432 
1433 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1434 		for_each_sg(s->sg, sg, num_sg, i) {
1435 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1436 
1437 			data.bvec_offset = sg->offset;
1438 			data.bvec_data = kmap_atomic(sg_page(sg));
1439 
1440 			gnttab_foreach_grant_in_range(sg_page(sg),
1441 						      sg->offset,
1442 						      sg->length,
1443 						      blkif_copy_from_grant,
1444 						      &data);
1445 
1446 			kunmap_atomic(data.bvec_data);
1447 		}
1448 	}
1449 	/* Add the persistent grant into the list of free grants */
1450 	for (i = 0; i < num_grant; i++) {
1451 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1452 			/*
1453 			 * If the grant is still mapped by the backend (the
1454 			 * backend has chosen to make this grant persistent)
1455 			 * we add it at the head of the list, so it will be
1456 			 * reused first.
1457 			 */
1458 			if (!info->feature_persistent)
1459 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1460 						     s->grants_used[i]->gref);
1461 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1462 			rinfo->persistent_gnts_c++;
1463 		} else {
1464 			/*
1465 			 * If the grant is not mapped by the backend we end the
1466 			 * foreign access and add it to the tail of the list,
1467 			 * so it will not be picked again unless we run out of
1468 			 * persistent grants.
1469 			 */
1470 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1471 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1472 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1473 		}
1474 	}
1475 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1476 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1477 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1478 				if (!info->feature_persistent)
1479 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1480 							     s->indirect_grants[i]->gref);
1481 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1482 				rinfo->persistent_gnts_c++;
1483 			} else {
1484 				struct page *indirect_page;
1485 
1486 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1487 				/*
1488 				 * Add the used indirect page back to the list of
1489 				 * available pages for indirect grefs.
1490 				 */
1491 				if (!info->feature_persistent) {
1492 					indirect_page = s->indirect_grants[i]->page;
1493 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1494 				}
1495 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1496 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1497 			}
1498 		}
1499 	}
1500 
1501 	return true;
1502 }
1503 
1504 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1505 {
1506 	struct request *req;
1507 	struct blkif_response bret;
1508 	RING_IDX i, rp;
1509 	unsigned long flags;
1510 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1511 	struct blkfront_info *info = rinfo->dev_info;
1512 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1513 
1514 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1515 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1516 		return IRQ_HANDLED;
1517 	}
1518 
1519 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1520  again:
1521 	rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1522 	virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1523 	if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1524 		pr_alert("%s: illegal number of responses %u\n",
1525 			 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1526 		goto err;
1527 	}
1528 
1529 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1530 		unsigned long id;
1531 		unsigned int op;
1532 
1533 		eoiflag = 0;
1534 
1535 		RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1536 		id = bret.id;
1537 
1538 		/*
1539 		 * The backend has messed up and given us an id that we would
1540 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1541 		 * look in get_id_from_freelist.
1542 		 */
1543 		if (id >= BLK_RING_SIZE(info)) {
1544 			pr_alert("%s: response has incorrect id (%ld)\n",
1545 				 info->gd->disk_name, id);
1546 			goto err;
1547 		}
1548 		if (rinfo->shadow[id].status != REQ_WAITING) {
1549 			pr_alert("%s: response references no pending request\n",
1550 				 info->gd->disk_name);
1551 			goto err;
1552 		}
1553 
1554 		rinfo->shadow[id].status = REQ_PROCESSING;
1555 		req  = rinfo->shadow[id].request;
1556 
1557 		op = rinfo->shadow[id].req.operation;
1558 		if (op == BLKIF_OP_INDIRECT)
1559 			op = rinfo->shadow[id].req.u.indirect.indirect_op;
1560 		if (bret.operation != op) {
1561 			pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1562 				 info->gd->disk_name, bret.operation, op);
1563 			goto err;
1564 		}
1565 
1566 		if (bret.operation != BLKIF_OP_DISCARD) {
1567 			/*
1568 			 * We may need to wait for an extra response if the
1569 			 * I/O request is split in 2
1570 			 */
1571 			if (!blkif_completion(&id, rinfo, &bret))
1572 				continue;
1573 		}
1574 
1575 		if (add_id_to_freelist(rinfo, id)) {
1576 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1577 			     info->gd->disk_name, op_name(bret.operation), id);
1578 			continue;
1579 		}
1580 
1581 		if (bret.status == BLKIF_RSP_OKAY)
1582 			blkif_req(req)->error = BLK_STS_OK;
1583 		else
1584 			blkif_req(req)->error = BLK_STS_IOERR;
1585 
1586 		switch (bret.operation) {
1587 		case BLKIF_OP_DISCARD:
1588 			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1589 				struct request_queue *rq = info->rq;
1590 
1591 				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1592 					   info->gd->disk_name, op_name(bret.operation));
1593 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1594 				info->feature_discard = 0;
1595 				info->feature_secdiscard = 0;
1596 				blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1597 				blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1598 			}
1599 			break;
1600 		case BLKIF_OP_FLUSH_DISKCACHE:
1601 		case BLKIF_OP_WRITE_BARRIER:
1602 			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1603 				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1604 				       info->gd->disk_name, op_name(bret.operation));
1605 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1606 			}
1607 			if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1608 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1609 				pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1610 				       info->gd->disk_name, op_name(bret.operation));
1611 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1612 			}
1613 			if (unlikely(blkif_req(req)->error)) {
1614 				if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1615 					blkif_req(req)->error = BLK_STS_OK;
1616 				info->feature_fua = 0;
1617 				info->feature_flush = 0;
1618 				xlvbd_flush(info);
1619 			}
1620 			fallthrough;
1621 		case BLKIF_OP_READ:
1622 		case BLKIF_OP_WRITE:
1623 			if (unlikely(bret.status != BLKIF_RSP_OKAY))
1624 				dev_dbg_ratelimited(&info->xbdev->dev,
1625 					"Bad return from blkdev data request: %#x\n",
1626 					bret.status);
1627 
1628 			break;
1629 		default:
1630 			BUG();
1631 		}
1632 
1633 		if (likely(!blk_should_fake_timeout(req->q)))
1634 			blk_mq_complete_request(req);
1635 	}
1636 
1637 	rinfo->ring.rsp_cons = i;
1638 
1639 	if (i != rinfo->ring.req_prod_pvt) {
1640 		int more_to_do;
1641 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1642 		if (more_to_do)
1643 			goto again;
1644 	} else
1645 		rinfo->ring.sring->rsp_event = i + 1;
1646 
1647 	kick_pending_request_queues_locked(rinfo);
1648 
1649 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1650 
1651 	xen_irq_lateeoi(irq, eoiflag);
1652 
1653 	return IRQ_HANDLED;
1654 
1655  err:
1656 	info->connected = BLKIF_STATE_ERROR;
1657 
1658 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1659 
1660 	/* No EOI in order to avoid further interrupts. */
1661 
1662 	pr_alert("%s disabled for further use\n", info->gd->disk_name);
1663 	return IRQ_HANDLED;
1664 }
1665 
1666 
1667 static int setup_blkring(struct xenbus_device *dev,
1668 			 struct blkfront_ring_info *rinfo)
1669 {
1670 	struct blkif_sring *sring;
1671 	int err, i;
1672 	struct blkfront_info *info = rinfo->dev_info;
1673 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1674 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1675 
1676 	for (i = 0; i < info->nr_ring_pages; i++)
1677 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1678 
1679 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1680 						       get_order(ring_size));
1681 	if (!sring) {
1682 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1683 		return -ENOMEM;
1684 	}
1685 	SHARED_RING_INIT(sring);
1686 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1687 
1688 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1689 	if (err < 0) {
1690 		free_pages((unsigned long)sring, get_order(ring_size));
1691 		rinfo->ring.sring = NULL;
1692 		goto fail;
1693 	}
1694 	for (i = 0; i < info->nr_ring_pages; i++)
1695 		rinfo->ring_ref[i] = gref[i];
1696 
1697 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1698 	if (err)
1699 		goto fail;
1700 
1701 	err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1702 						0, "blkif", rinfo);
1703 	if (err <= 0) {
1704 		xenbus_dev_fatal(dev, err,
1705 				 "bind_evtchn_to_irqhandler failed");
1706 		goto fail;
1707 	}
1708 	rinfo->irq = err;
1709 
1710 	return 0;
1711 fail:
1712 	blkif_free(info, 0);
1713 	return err;
1714 }
1715 
1716 /*
1717  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1718  * ring buffer may have multi pages depending on ->nr_ring_pages.
1719  */
1720 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1721 				struct blkfront_ring_info *rinfo, const char *dir)
1722 {
1723 	int err;
1724 	unsigned int i;
1725 	const char *message = NULL;
1726 	struct blkfront_info *info = rinfo->dev_info;
1727 
1728 	if (info->nr_ring_pages == 1) {
1729 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1730 		if (err) {
1731 			message = "writing ring-ref";
1732 			goto abort_transaction;
1733 		}
1734 	} else {
1735 		for (i = 0; i < info->nr_ring_pages; i++) {
1736 			char ring_ref_name[RINGREF_NAME_LEN];
1737 
1738 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1739 			err = xenbus_printf(xbt, dir, ring_ref_name,
1740 					    "%u", rinfo->ring_ref[i]);
1741 			if (err) {
1742 				message = "writing ring-ref";
1743 				goto abort_transaction;
1744 			}
1745 		}
1746 	}
1747 
1748 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1749 	if (err) {
1750 		message = "writing event-channel";
1751 		goto abort_transaction;
1752 	}
1753 
1754 	return 0;
1755 
1756 abort_transaction:
1757 	xenbus_transaction_end(xbt, 1);
1758 	if (message)
1759 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1760 
1761 	return err;
1762 }
1763 
1764 /* Common code used when first setting up, and when resuming. */
1765 static int talk_to_blkback(struct xenbus_device *dev,
1766 			   struct blkfront_info *info)
1767 {
1768 	const char *message = NULL;
1769 	struct xenbus_transaction xbt;
1770 	int err;
1771 	unsigned int i, max_page_order;
1772 	unsigned int ring_page_order;
1773 	struct blkfront_ring_info *rinfo;
1774 
1775 	if (!info)
1776 		return -ENODEV;
1777 
1778 	max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1779 					      "max-ring-page-order", 0);
1780 	ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1781 	info->nr_ring_pages = 1 << ring_page_order;
1782 
1783 	err = negotiate_mq(info);
1784 	if (err)
1785 		goto destroy_blkring;
1786 
1787 	for_each_rinfo(info, rinfo, i) {
1788 		/* Create shared ring, alloc event channel. */
1789 		err = setup_blkring(dev, rinfo);
1790 		if (err)
1791 			goto destroy_blkring;
1792 	}
1793 
1794 again:
1795 	err = xenbus_transaction_start(&xbt);
1796 	if (err) {
1797 		xenbus_dev_fatal(dev, err, "starting transaction");
1798 		goto destroy_blkring;
1799 	}
1800 
1801 	if (info->nr_ring_pages > 1) {
1802 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1803 				    ring_page_order);
1804 		if (err) {
1805 			message = "writing ring-page-order";
1806 			goto abort_transaction;
1807 		}
1808 	}
1809 
1810 	/* We already got the number of queues/rings in _probe */
1811 	if (info->nr_rings == 1) {
1812 		err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1813 		if (err)
1814 			goto destroy_blkring;
1815 	} else {
1816 		char *path;
1817 		size_t pathsize;
1818 
1819 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1820 				    info->nr_rings);
1821 		if (err) {
1822 			message = "writing multi-queue-num-queues";
1823 			goto abort_transaction;
1824 		}
1825 
1826 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1827 		path = kmalloc(pathsize, GFP_KERNEL);
1828 		if (!path) {
1829 			err = -ENOMEM;
1830 			message = "ENOMEM while writing ring references";
1831 			goto abort_transaction;
1832 		}
1833 
1834 		for_each_rinfo(info, rinfo, i) {
1835 			memset(path, 0, pathsize);
1836 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1837 			err = write_per_ring_nodes(xbt, rinfo, path);
1838 			if (err) {
1839 				kfree(path);
1840 				goto destroy_blkring;
1841 			}
1842 		}
1843 		kfree(path);
1844 	}
1845 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1846 			    XEN_IO_PROTO_ABI_NATIVE);
1847 	if (err) {
1848 		message = "writing protocol";
1849 		goto abort_transaction;
1850 	}
1851 	err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1852 			info->feature_persistent);
1853 	if (err)
1854 		dev_warn(&dev->dev,
1855 			 "writing persistent grants feature to xenbus");
1856 
1857 	err = xenbus_transaction_end(xbt, 0);
1858 	if (err) {
1859 		if (err == -EAGAIN)
1860 			goto again;
1861 		xenbus_dev_fatal(dev, err, "completing transaction");
1862 		goto destroy_blkring;
1863 	}
1864 
1865 	for_each_rinfo(info, rinfo, i) {
1866 		unsigned int j;
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 	return err;
1883 }
1884 
1885 static int negotiate_mq(struct blkfront_info *info)
1886 {
1887 	unsigned int backend_max_queues;
1888 	unsigned int i;
1889 	struct blkfront_ring_info *rinfo;
1890 
1891 	BUG_ON(info->nr_rings);
1892 
1893 	/* Check if backend supports multiple queues. */
1894 	backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1895 						  "multi-queue-max-queues", 1);
1896 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1897 	/* We need at least one ring. */
1898 	if (!info->nr_rings)
1899 		info->nr_rings = 1;
1900 
1901 	info->rinfo_size = struct_size(info->rinfo, shadow,
1902 				       BLK_RING_SIZE(info));
1903 	info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1904 	if (!info->rinfo) {
1905 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1906 		info->nr_rings = 0;
1907 		return -ENOMEM;
1908 	}
1909 
1910 	for_each_rinfo(info, rinfo, i) {
1911 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1912 		INIT_LIST_HEAD(&rinfo->grants);
1913 		rinfo->dev_info = info;
1914 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1915 		spin_lock_init(&rinfo->ring_lock);
1916 	}
1917 	return 0;
1918 }
1919 
1920 /* Enable the persistent grants feature. */
1921 static bool feature_persistent = true;
1922 module_param(feature_persistent, bool, 0644);
1923 MODULE_PARM_DESC(feature_persistent,
1924 		"Enables the persistent grants feature");
1925 
1926 /*
1927  * Entry point to this code when a new device is created.  Allocate the basic
1928  * structures and the ring buffer for communication with the backend, and
1929  * inform the backend of the appropriate details for those.  Switch to
1930  * Initialised state.
1931  */
1932 static int blkfront_probe(struct xenbus_device *dev,
1933 			  const struct xenbus_device_id *id)
1934 {
1935 	int err, vdevice;
1936 	struct blkfront_info *info;
1937 
1938 	/* FIXME: Use dynamic device id if this is not set. */
1939 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1940 			   "virtual-device", "%i", &vdevice);
1941 	if (err != 1) {
1942 		/* go looking in the extended area instead */
1943 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1944 				   "%i", &vdevice);
1945 		if (err != 1) {
1946 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1947 			return err;
1948 		}
1949 	}
1950 
1951 	if (xen_hvm_domain()) {
1952 		char *type;
1953 		int len;
1954 		/* no unplug has been done: do not hook devices != xen vbds */
1955 		if (xen_has_pv_and_legacy_disk_devices()) {
1956 			int major;
1957 
1958 			if (!VDEV_IS_EXTENDED(vdevice))
1959 				major = BLKIF_MAJOR(vdevice);
1960 			else
1961 				major = XENVBD_MAJOR;
1962 
1963 			if (major != XENVBD_MAJOR) {
1964 				printk(KERN_INFO
1965 						"%s: HVM does not support vbd %d as xen block device\n",
1966 						__func__, vdevice);
1967 				return -ENODEV;
1968 			}
1969 		}
1970 		/* do not create a PV cdrom device if we are an HVM guest */
1971 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1972 		if (IS_ERR(type))
1973 			return -ENODEV;
1974 		if (strncmp(type, "cdrom", 5) == 0) {
1975 			kfree(type);
1976 			return -ENODEV;
1977 		}
1978 		kfree(type);
1979 	}
1980 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1981 	if (!info) {
1982 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1983 		return -ENOMEM;
1984 	}
1985 
1986 	info->xbdev = dev;
1987 
1988 	mutex_init(&info->mutex);
1989 	info->vdevice = vdevice;
1990 	info->connected = BLKIF_STATE_DISCONNECTED;
1991 
1992 	info->feature_persistent = feature_persistent;
1993 
1994 	/* Front end dir is a number, which is used as the id. */
1995 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1996 	dev_set_drvdata(&dev->dev, info);
1997 
1998 	mutex_lock(&blkfront_mutex);
1999 	list_add(&info->info_list, &info_list);
2000 	mutex_unlock(&blkfront_mutex);
2001 
2002 	return 0;
2003 }
2004 
2005 static int blkif_recover(struct blkfront_info *info)
2006 {
2007 	unsigned int r_index;
2008 	struct request *req, *n;
2009 	int rc;
2010 	struct bio *bio;
2011 	unsigned int segs;
2012 	struct blkfront_ring_info *rinfo;
2013 
2014 	blkfront_gather_backend_features(info);
2015 	/* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2016 	blkif_set_queue_limits(info);
2017 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2018 	blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2019 
2020 	for_each_rinfo(info, rinfo, r_index) {
2021 		rc = blkfront_setup_indirect(rinfo);
2022 		if (rc)
2023 			return rc;
2024 	}
2025 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2026 
2027 	/* Now safe for us to use the shared ring */
2028 	info->connected = BLKIF_STATE_CONNECTED;
2029 
2030 	for_each_rinfo(info, rinfo, r_index) {
2031 		/* Kick any other new requests queued since we resumed */
2032 		kick_pending_request_queues(rinfo);
2033 	}
2034 
2035 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2036 		/* Requeue pending requests (flush or discard) */
2037 		list_del_init(&req->queuelist);
2038 		BUG_ON(req->nr_phys_segments > segs);
2039 		blk_mq_requeue_request(req, false);
2040 	}
2041 	blk_mq_start_stopped_hw_queues(info->rq, true);
2042 	blk_mq_kick_requeue_list(info->rq);
2043 
2044 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2045 		/* Traverse the list of pending bios and re-queue them */
2046 		submit_bio(bio);
2047 	}
2048 
2049 	return 0;
2050 }
2051 
2052 /*
2053  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2054  * driver restart.  We tear down our blkif structure and recreate it, but
2055  * leave the device-layer structures intact so that this is transparent to the
2056  * rest of the kernel.
2057  */
2058 static int blkfront_resume(struct xenbus_device *dev)
2059 {
2060 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2061 	int err = 0;
2062 	unsigned int i, j;
2063 	struct blkfront_ring_info *rinfo;
2064 
2065 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2066 
2067 	bio_list_init(&info->bio_list);
2068 	INIT_LIST_HEAD(&info->requests);
2069 	for_each_rinfo(info, rinfo, i) {
2070 		struct bio_list merge_bio;
2071 		struct blk_shadow *shadow = rinfo->shadow;
2072 
2073 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2074 			/* Not in use? */
2075 			if (!shadow[j].request)
2076 				continue;
2077 
2078 			/*
2079 			 * Get the bios in the request so we can re-queue them.
2080 			 */
2081 			if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2082 			    req_op(shadow[j].request) == REQ_OP_DISCARD ||
2083 			    req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2084 			    shadow[j].request->cmd_flags & REQ_FUA) {
2085 				/*
2086 				 * Flush operations don't contain bios, so
2087 				 * we need to requeue the whole request
2088 				 *
2089 				 * XXX: but this doesn't make any sense for a
2090 				 * write with the FUA flag set..
2091 				 */
2092 				list_add(&shadow[j].request->queuelist, &info->requests);
2093 				continue;
2094 			}
2095 			merge_bio.head = shadow[j].request->bio;
2096 			merge_bio.tail = shadow[j].request->biotail;
2097 			bio_list_merge(&info->bio_list, &merge_bio);
2098 			shadow[j].request->bio = NULL;
2099 			blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2100 		}
2101 	}
2102 
2103 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2104 
2105 	err = talk_to_blkback(dev, info);
2106 	if (!err)
2107 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2108 
2109 	/*
2110 	 * We have to wait for the backend to switch to
2111 	 * connected state, since we want to read which
2112 	 * features it supports.
2113 	 */
2114 
2115 	return err;
2116 }
2117 
2118 static void blkfront_closing(struct blkfront_info *info)
2119 {
2120 	struct xenbus_device *xbdev = info->xbdev;
2121 	struct blkfront_ring_info *rinfo;
2122 	unsigned int i;
2123 
2124 	if (xbdev->state == XenbusStateClosing)
2125 		return;
2126 
2127 	/* No more blkif_request(). */
2128 	blk_mq_stop_hw_queues(info->rq);
2129 	blk_set_queue_dying(info->rq);
2130 	set_capacity(info->gd, 0);
2131 
2132 	for_each_rinfo(info, rinfo, i) {
2133 		/* No more gnttab callback work. */
2134 		gnttab_cancel_free_callback(&rinfo->callback);
2135 
2136 		/* Flush gnttab callback work. Must be done with no locks held. */
2137 		flush_work(&rinfo->work);
2138 	}
2139 
2140 	xenbus_frontend_closed(xbdev);
2141 }
2142 
2143 static void blkfront_setup_discard(struct blkfront_info *info)
2144 {
2145 	info->feature_discard = 1;
2146 	info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2147 							 "discard-granularity",
2148 							 0);
2149 	info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2150 						       "discard-alignment", 0);
2151 	info->feature_secdiscard =
2152 		!!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2153 				       0);
2154 }
2155 
2156 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2157 {
2158 	unsigned int psegs, grants, memflags;
2159 	int err, i;
2160 	struct blkfront_info *info = rinfo->dev_info;
2161 
2162 	memflags = memalloc_noio_save();
2163 
2164 	if (info->max_indirect_segments == 0) {
2165 		if (!HAS_EXTRA_REQ)
2166 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2167 		else {
2168 			/*
2169 			 * When an extra req is required, the maximum
2170 			 * grants supported is related to the size of the
2171 			 * Linux block segment.
2172 			 */
2173 			grants = GRANTS_PER_PSEG;
2174 		}
2175 	}
2176 	else
2177 		grants = info->max_indirect_segments;
2178 	psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2179 
2180 	err = fill_grant_buffer(rinfo,
2181 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2182 	if (err)
2183 		goto out_of_memory;
2184 
2185 	if (!info->feature_persistent && info->max_indirect_segments) {
2186 		/*
2187 		 * We are using indirect descriptors but not persistent
2188 		 * grants, we need to allocate a set of pages that can be
2189 		 * used for mapping indirect grefs
2190 		 */
2191 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2192 
2193 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2194 		for (i = 0; i < num; i++) {
2195 			struct page *indirect_page = alloc_page(GFP_KERNEL);
2196 			if (!indirect_page)
2197 				goto out_of_memory;
2198 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2199 		}
2200 	}
2201 
2202 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2203 		rinfo->shadow[i].grants_used =
2204 			kvcalloc(grants,
2205 				 sizeof(rinfo->shadow[i].grants_used[0]),
2206 				 GFP_KERNEL);
2207 		rinfo->shadow[i].sg = kvcalloc(psegs,
2208 					       sizeof(rinfo->shadow[i].sg[0]),
2209 					       GFP_KERNEL);
2210 		if (info->max_indirect_segments)
2211 			rinfo->shadow[i].indirect_grants =
2212 				kvcalloc(INDIRECT_GREFS(grants),
2213 					 sizeof(rinfo->shadow[i].indirect_grants[0]),
2214 					 GFP_KERNEL);
2215 		if ((rinfo->shadow[i].grants_used == NULL) ||
2216 			(rinfo->shadow[i].sg == NULL) ||
2217 		     (info->max_indirect_segments &&
2218 		     (rinfo->shadow[i].indirect_grants == NULL)))
2219 			goto out_of_memory;
2220 		sg_init_table(rinfo->shadow[i].sg, psegs);
2221 	}
2222 
2223 	memalloc_noio_restore(memflags);
2224 
2225 	return 0;
2226 
2227 out_of_memory:
2228 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2229 		kvfree(rinfo->shadow[i].grants_used);
2230 		rinfo->shadow[i].grants_used = NULL;
2231 		kvfree(rinfo->shadow[i].sg);
2232 		rinfo->shadow[i].sg = NULL;
2233 		kvfree(rinfo->shadow[i].indirect_grants);
2234 		rinfo->shadow[i].indirect_grants = NULL;
2235 	}
2236 	if (!list_empty(&rinfo->indirect_pages)) {
2237 		struct page *indirect_page, *n;
2238 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2239 			list_del(&indirect_page->lru);
2240 			__free_page(indirect_page);
2241 		}
2242 	}
2243 
2244 	memalloc_noio_restore(memflags);
2245 
2246 	return -ENOMEM;
2247 }
2248 
2249 /*
2250  * Gather all backend feature-*
2251  */
2252 static void blkfront_gather_backend_features(struct blkfront_info *info)
2253 {
2254 	unsigned int indirect_segments;
2255 
2256 	info->feature_flush = 0;
2257 	info->feature_fua = 0;
2258 
2259 	/*
2260 	 * If there's no "feature-barrier" defined, then it means
2261 	 * we're dealing with a very old backend which writes
2262 	 * synchronously; nothing to do.
2263 	 *
2264 	 * If there are barriers, then we use flush.
2265 	 */
2266 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2267 		info->feature_flush = 1;
2268 		info->feature_fua = 1;
2269 	}
2270 
2271 	/*
2272 	 * And if there is "feature-flush-cache" use that above
2273 	 * barriers.
2274 	 */
2275 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2276 				 0)) {
2277 		info->feature_flush = 1;
2278 		info->feature_fua = 0;
2279 	}
2280 
2281 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2282 		blkfront_setup_discard(info);
2283 
2284 	if (info->feature_persistent)
2285 		info->feature_persistent =
2286 			!!xenbus_read_unsigned(info->xbdev->otherend,
2287 					       "feature-persistent", 0);
2288 
2289 	indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2290 					"feature-max-indirect-segments", 0);
2291 	if (indirect_segments > xen_blkif_max_segments)
2292 		indirect_segments = xen_blkif_max_segments;
2293 	if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2294 		indirect_segments = 0;
2295 	info->max_indirect_segments = indirect_segments;
2296 
2297 	if (info->feature_persistent) {
2298 		mutex_lock(&blkfront_mutex);
2299 		schedule_delayed_work(&blkfront_work, HZ * 10);
2300 		mutex_unlock(&blkfront_mutex);
2301 	}
2302 }
2303 
2304 /*
2305  * Invoked when the backend is finally 'ready' (and has told produced
2306  * the details about the physical device - #sectors, size, etc).
2307  */
2308 static void blkfront_connect(struct blkfront_info *info)
2309 {
2310 	unsigned long long sectors;
2311 	unsigned long sector_size;
2312 	unsigned int physical_sector_size;
2313 	int err, i;
2314 	struct blkfront_ring_info *rinfo;
2315 
2316 	switch (info->connected) {
2317 	case BLKIF_STATE_CONNECTED:
2318 		/*
2319 		 * Potentially, the back-end may be signalling
2320 		 * a capacity change; update the capacity.
2321 		 */
2322 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2323 				   "sectors", "%Lu", &sectors);
2324 		if (XENBUS_EXIST_ERR(err))
2325 			return;
2326 		printk(KERN_INFO "Setting capacity to %Lu\n",
2327 		       sectors);
2328 		set_capacity_and_notify(info->gd, sectors);
2329 
2330 		return;
2331 	case BLKIF_STATE_SUSPENDED:
2332 		/*
2333 		 * If we are recovering from suspension, we need to wait
2334 		 * for the backend to announce it's features before
2335 		 * reconnecting, at least we need to know if the backend
2336 		 * supports indirect descriptors, and how many.
2337 		 */
2338 		blkif_recover(info);
2339 		return;
2340 
2341 	default:
2342 		break;
2343 	}
2344 
2345 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2346 		__func__, info->xbdev->otherend);
2347 
2348 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2349 			    "sectors", "%llu", &sectors,
2350 			    "info", "%u", &info->vdisk_info,
2351 			    "sector-size", "%lu", &sector_size,
2352 			    NULL);
2353 	if (err) {
2354 		xenbus_dev_fatal(info->xbdev, err,
2355 				 "reading backend fields at %s",
2356 				 info->xbdev->otherend);
2357 		return;
2358 	}
2359 
2360 	/*
2361 	 * physical-sector-size is a newer field, so old backends may not
2362 	 * provide this. Assume physical sector size to be the same as
2363 	 * sector_size in that case.
2364 	 */
2365 	physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2366 						    "physical-sector-size",
2367 						    sector_size);
2368 	blkfront_gather_backend_features(info);
2369 	for_each_rinfo(info, rinfo, i) {
2370 		err = blkfront_setup_indirect(rinfo);
2371 		if (err) {
2372 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2373 					 info->xbdev->otherend);
2374 			blkif_free(info, 0);
2375 			break;
2376 		}
2377 	}
2378 
2379 	err = xlvbd_alloc_gendisk(sectors, info, sector_size,
2380 				  physical_sector_size);
2381 	if (err) {
2382 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2383 				 info->xbdev->otherend);
2384 		goto fail;
2385 	}
2386 
2387 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2388 
2389 	/* Kick pending requests. */
2390 	info->connected = BLKIF_STATE_CONNECTED;
2391 	for_each_rinfo(info, rinfo, i)
2392 		kick_pending_request_queues(rinfo);
2393 
2394 	err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2395 	if (err) {
2396 		blk_cleanup_disk(info->gd);
2397 		blk_mq_free_tag_set(&info->tag_set);
2398 		info->rq = NULL;
2399 		goto fail;
2400 	}
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 		break;
2427 	case XenbusStateInitialising:
2428 	case XenbusStateInitialised:
2429 	case XenbusStateReconfiguring:
2430 	case XenbusStateReconfigured:
2431 	case XenbusStateUnknown:
2432 		break;
2433 
2434 	case XenbusStateConnected:
2435 		/*
2436 		 * talk_to_blkback sets state to XenbusStateInitialised
2437 		 * and blkfront_connect sets it to XenbusStateConnected
2438 		 * (if connection went OK).
2439 		 *
2440 		 * If the backend (or toolstack) decides to poke at backend
2441 		 * state (and re-trigger the watch by setting the state repeatedly
2442 		 * to XenbusStateConnected (4)) we need to deal with this.
2443 		 * This is allowed as this is used to communicate to the guest
2444 		 * that the size of disk has changed!
2445 		 */
2446 		if ((dev->state != XenbusStateInitialised) &&
2447 		    (dev->state != XenbusStateConnected)) {
2448 			if (talk_to_blkback(dev, info))
2449 				break;
2450 		}
2451 
2452 		blkfront_connect(info);
2453 		break;
2454 
2455 	case XenbusStateClosed:
2456 		if (dev->state == XenbusStateClosed)
2457 			break;
2458 		fallthrough;
2459 	case XenbusStateClosing:
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 
2469 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2470 
2471 	del_gendisk(info->gd);
2472 
2473 	mutex_lock(&blkfront_mutex);
2474 	list_del(&info->info_list);
2475 	mutex_unlock(&blkfront_mutex);
2476 
2477 	blkif_free(info, 0);
2478 	xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2479 	blk_cleanup_disk(info->gd);
2480 	blk_mq_free_tag_set(&info->tag_set);
2481 
2482 	kfree(info);
2483 	return 0;
2484 }
2485 
2486 static int blkfront_is_ready(struct xenbus_device *dev)
2487 {
2488 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2489 
2490 	return info->is_ready && info->xbdev;
2491 }
2492 
2493 static const struct block_device_operations xlvbd_block_fops =
2494 {
2495 	.owner = THIS_MODULE,
2496 	.getgeo = blkif_getgeo,
2497 	.ioctl = blkif_ioctl,
2498 	.compat_ioctl = blkdev_compat_ptr_ioctl,
2499 };
2500 
2501 
2502 static const struct xenbus_device_id blkfront_ids[] = {
2503 	{ "vbd" },
2504 	{ "" }
2505 };
2506 
2507 static struct xenbus_driver blkfront_driver = {
2508 	.ids  = blkfront_ids,
2509 	.probe = blkfront_probe,
2510 	.remove = blkfront_remove,
2511 	.resume = blkfront_resume,
2512 	.otherend_changed = blkback_changed,
2513 	.is_ready = blkfront_is_ready,
2514 };
2515 
2516 static void purge_persistent_grants(struct blkfront_info *info)
2517 {
2518 	unsigned int i;
2519 	unsigned long flags;
2520 	struct blkfront_ring_info *rinfo;
2521 
2522 	for_each_rinfo(info, rinfo, i) {
2523 		struct grant *gnt_list_entry, *tmp;
2524 
2525 		spin_lock_irqsave(&rinfo->ring_lock, flags);
2526 
2527 		if (rinfo->persistent_gnts_c == 0) {
2528 			spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2529 			continue;
2530 		}
2531 
2532 		list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2533 					 node) {
2534 			if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2535 			    gnttab_query_foreign_access(gnt_list_entry->gref))
2536 				continue;
2537 
2538 			list_del(&gnt_list_entry->node);
2539 			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2540 			rinfo->persistent_gnts_c--;
2541 			gnt_list_entry->gref = GRANT_INVALID_REF;
2542 			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2543 		}
2544 
2545 		spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2546 	}
2547 }
2548 
2549 static void blkfront_delay_work(struct work_struct *work)
2550 {
2551 	struct blkfront_info *info;
2552 	bool need_schedule_work = false;
2553 
2554 	mutex_lock(&blkfront_mutex);
2555 
2556 	list_for_each_entry(info, &info_list, info_list) {
2557 		if (info->feature_persistent) {
2558 			need_schedule_work = true;
2559 			mutex_lock(&info->mutex);
2560 			purge_persistent_grants(info);
2561 			mutex_unlock(&info->mutex);
2562 		}
2563 	}
2564 
2565 	if (need_schedule_work)
2566 		schedule_delayed_work(&blkfront_work, HZ * 10);
2567 
2568 	mutex_unlock(&blkfront_mutex);
2569 }
2570 
2571 static int __init xlblk_init(void)
2572 {
2573 	int ret;
2574 	int nr_cpus = num_online_cpus();
2575 
2576 	if (!xen_domain())
2577 		return -ENODEV;
2578 
2579 	if (!xen_has_pv_disk_devices())
2580 		return -ENODEV;
2581 
2582 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2583 		pr_warn("xen_blk: can't get major %d with name %s\n",
2584 			XENVBD_MAJOR, DEV_NAME);
2585 		return -ENODEV;
2586 	}
2587 
2588 	if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2589 		xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2590 
2591 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2592 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2593 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2594 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2595 	}
2596 
2597 	if (xen_blkif_max_queues > nr_cpus) {
2598 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2599 			xen_blkif_max_queues, nr_cpus);
2600 		xen_blkif_max_queues = nr_cpus;
2601 	}
2602 
2603 	INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2604 
2605 	ret = xenbus_register_frontend(&blkfront_driver);
2606 	if (ret) {
2607 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2608 		return ret;
2609 	}
2610 
2611 	return 0;
2612 }
2613 module_init(xlblk_init);
2614 
2615 
2616 static void __exit xlblk_exit(void)
2617 {
2618 	cancel_delayed_work_sync(&blkfront_work);
2619 
2620 	xenbus_unregister_driver(&blkfront_driver);
2621 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2622 	kfree(minors);
2623 }
2624 module_exit(xlblk_exit);
2625 
2626 MODULE_DESCRIPTION("Xen virtual block device frontend");
2627 MODULE_LICENSE("GPL");
2628 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2629 MODULE_ALIAS("xen:vbd");
2630 MODULE_ALIAS("xenblk");
2631