xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 0984d159)
1 /*
2  * blkfront.c
3  *
4  * XenLinux virtual block device driver.
5  *
6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8  * Copyright (c) 2004, Christian Limpach
9  * Copyright (c) 2004, Andrew Warfield
10  * Copyright (c) 2005, Christopher Clark
11  * Copyright (c) 2005, XenSource Ltd
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License version 2
15  * as published by the Free Software Foundation; or, when distributed
16  * separately from the Linux kernel or incorporated into other
17  * software packages, subject to the following license:
18  *
19  * Permission is hereby granted, free of charge, to any person obtaining a copy
20  * of this source file (the "Software"), to deal in the Software without
21  * restriction, including without limitation the rights to use, copy, modify,
22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23  * and to permit persons to whom the Software is furnished to do so, subject to
24  * the following conditions:
25  *
26  * The above copyright notice and this permission notice shall be included in
27  * all copies or substantial portions of the Software.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35  * IN THE SOFTWARE.
36  */
37 
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 
50 #include <xen/xen.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56 
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
60 
61 #include <asm/xen/hypervisor.h>
62 
63 /*
64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
65  * When Linux is using a different page size than Xen, it may not be possible
66  * to put all the data in a single segment.
67  * This can happen when the backend doesn't support indirect descriptor and
68  * therefore the maximum amount of data that a request can carry is
69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70  *
71  * Note that we only support one extra request. So the Linux page size
72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73  * 88KB.
74  */
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76 
77 enum blkif_state {
78 	BLKIF_STATE_DISCONNECTED,
79 	BLKIF_STATE_CONNECTED,
80 	BLKIF_STATE_SUSPENDED,
81 };
82 
83 struct grant {
84 	grant_ref_t gref;
85 	struct page *page;
86 	struct list_head node;
87 };
88 
89 enum blk_req_status {
90 	REQ_WAITING,
91 	REQ_DONE,
92 	REQ_ERROR,
93 	REQ_EOPNOTSUPP,
94 };
95 
96 struct blk_shadow {
97 	struct blkif_request req;
98 	struct request *request;
99 	struct grant **grants_used;
100 	struct grant **indirect_grants;
101 	struct scatterlist *sg;
102 	unsigned int num_sg;
103 	enum blk_req_status status;
104 
105 	#define NO_ASSOCIATED_ID ~0UL
106 	/*
107 	 * Id of the sibling if we ever need 2 requests when handling a
108 	 * block I/O request
109 	 */
110 	unsigned long associated_id;
111 };
112 
113 struct split_bio {
114 	struct bio *bio;
115 	atomic_t pending;
116 };
117 
118 static DEFINE_MUTEX(blkfront_mutex);
119 static const struct block_device_operations xlvbd_block_fops;
120 
121 /*
122  * Maximum number of segments in indirect requests, the actual value used by
123  * the frontend driver is the minimum of this value and the value provided
124  * by the backend driver.
125  */
126 
127 static unsigned int xen_blkif_max_segments = 32;
128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
129 		   S_IRUGO);
130 MODULE_PARM_DESC(max_indirect_segments,
131 		 "Maximum amount of segments in indirect requests (default is 32)");
132 
133 static unsigned int xen_blkif_max_queues = 4;
134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
136 
137 /*
138  * Maximum order of pages to be used for the shared ring between front and
139  * backend, 4KB page granularity is used.
140  */
141 static unsigned int xen_blkif_max_ring_order;
142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
144 
145 #define BLK_RING_SIZE(info)	\
146 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
147 
148 #define BLK_MAX_RING_SIZE	\
149 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
150 
151 /*
152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
153  * characters are enough. Define to 20 to keep consistent with backend.
154  */
155 #define RINGREF_NAME_LEN (20)
156 /*
157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
158  */
159 #define QUEUE_NAME_LEN (17)
160 
161 /*
162  *  Per-ring info.
163  *  Every blkfront device can associate with one or more blkfront_ring_info,
164  *  depending on how many hardware queues/rings to be used.
165  */
166 struct blkfront_ring_info {
167 	/* Lock to protect data in every ring buffer. */
168 	spinlock_t ring_lock;
169 	struct blkif_front_ring ring;
170 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
171 	unsigned int evtchn, irq;
172 	struct work_struct work;
173 	struct gnttab_free_callback callback;
174 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
175 	struct list_head indirect_pages;
176 	struct list_head grants;
177 	unsigned int persistent_gnts_c;
178 	unsigned long shadow_free;
179 	struct blkfront_info *dev_info;
180 };
181 
182 /*
183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
184  * hang in private_data off the gendisk structure. We may end up
185  * putting all kinds of interesting stuff here :-)
186  */
187 struct blkfront_info
188 {
189 	struct mutex mutex;
190 	struct xenbus_device *xbdev;
191 	struct gendisk *gd;
192 	int vdevice;
193 	blkif_vdev_t handle;
194 	enum blkif_state connected;
195 	/* Number of pages per ring buffer. */
196 	unsigned int nr_ring_pages;
197 	struct request_queue *rq;
198 	unsigned int feature_flush;
199 	unsigned int feature_discard:1;
200 	unsigned int feature_secdiscard:1;
201 	unsigned int discard_granularity;
202 	unsigned int discard_alignment;
203 	unsigned int feature_persistent:1;
204 	/* Number of 4KB segments handled */
205 	unsigned int max_indirect_segments;
206 	int is_ready;
207 	struct blk_mq_tag_set tag_set;
208 	struct blkfront_ring_info *rinfo;
209 	unsigned int nr_rings;
210 	/* Save uncomplete reqs and bios for migration. */
211 	struct list_head requests;
212 	struct bio_list bio_list;
213 };
214 
215 static unsigned int nr_minors;
216 static unsigned long *minors;
217 static DEFINE_SPINLOCK(minor_lock);
218 
219 #define GRANT_INVALID_REF	0
220 
221 #define PARTS_PER_DISK		16
222 #define PARTS_PER_EXT_DISK      256
223 
224 #define BLKIF_MAJOR(dev) ((dev)>>8)
225 #define BLKIF_MINOR(dev) ((dev) & 0xff)
226 
227 #define EXT_SHIFT 28
228 #define EXTENDED (1<<EXT_SHIFT)
229 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
230 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
231 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
232 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
233 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
234 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
235 
236 #define DEV_NAME	"xvd"	/* name in /dev */
237 
238 /*
239  * Grants are always the same size as a Xen page (i.e 4KB).
240  * A physical segment is always the same size as a Linux page.
241  * Number of grants per physical segment
242  */
243 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
244 
245 #define GRANTS_PER_INDIRECT_FRAME \
246 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
247 
248 #define PSEGS_PER_INDIRECT_FRAME	\
249 	(GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
250 
251 #define INDIRECT_GREFS(_grants)		\
252 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
253 
254 #define GREFS(_psegs)	((_psegs) * GRANTS_PER_PSEG)
255 
256 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
257 static void blkfront_gather_backend_features(struct blkfront_info *info);
258 
259 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
260 {
261 	unsigned long free = rinfo->shadow_free;
262 
263 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
264 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
265 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
266 	return free;
267 }
268 
269 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
270 			      unsigned long id)
271 {
272 	if (rinfo->shadow[id].req.u.rw.id != id)
273 		return -EINVAL;
274 	if (rinfo->shadow[id].request == NULL)
275 		return -EINVAL;
276 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
277 	rinfo->shadow[id].request = NULL;
278 	rinfo->shadow_free = id;
279 	return 0;
280 }
281 
282 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
283 {
284 	struct blkfront_info *info = rinfo->dev_info;
285 	struct page *granted_page;
286 	struct grant *gnt_list_entry, *n;
287 	int i = 0;
288 
289 	while (i < num) {
290 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
291 		if (!gnt_list_entry)
292 			goto out_of_memory;
293 
294 		if (info->feature_persistent) {
295 			granted_page = alloc_page(GFP_NOIO);
296 			if (!granted_page) {
297 				kfree(gnt_list_entry);
298 				goto out_of_memory;
299 			}
300 			gnt_list_entry->page = granted_page;
301 		}
302 
303 		gnt_list_entry->gref = GRANT_INVALID_REF;
304 		list_add(&gnt_list_entry->node, &rinfo->grants);
305 		i++;
306 	}
307 
308 	return 0;
309 
310 out_of_memory:
311 	list_for_each_entry_safe(gnt_list_entry, n,
312 	                         &rinfo->grants, node) {
313 		list_del(&gnt_list_entry->node);
314 		if (info->feature_persistent)
315 			__free_page(gnt_list_entry->page);
316 		kfree(gnt_list_entry);
317 		i--;
318 	}
319 	BUG_ON(i != 0);
320 	return -ENOMEM;
321 }
322 
323 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
324 {
325 	struct grant *gnt_list_entry;
326 
327 	BUG_ON(list_empty(&rinfo->grants));
328 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
329 					  node);
330 	list_del(&gnt_list_entry->node);
331 
332 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
333 		rinfo->persistent_gnts_c--;
334 
335 	return gnt_list_entry;
336 }
337 
338 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
339 					const struct blkfront_info *info)
340 {
341 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
342 						 info->xbdev->otherend_id,
343 						 gnt_list_entry->page,
344 						 0);
345 }
346 
347 static struct grant *get_grant(grant_ref_t *gref_head,
348 			       unsigned long gfn,
349 			       struct blkfront_ring_info *rinfo)
350 {
351 	struct grant *gnt_list_entry = get_free_grant(rinfo);
352 	struct blkfront_info *info = rinfo->dev_info;
353 
354 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
355 		return gnt_list_entry;
356 
357 	/* Assign a gref to this page */
358 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
359 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
360 	if (info->feature_persistent)
361 		grant_foreign_access(gnt_list_entry, info);
362 	else {
363 		/* Grant access to the GFN passed by the caller */
364 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
365 						info->xbdev->otherend_id,
366 						gfn, 0);
367 	}
368 
369 	return gnt_list_entry;
370 }
371 
372 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
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 != GRANT_INVALID_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->feature_persistent) {
385 		struct page *indirect_page;
386 
387 		/* Fetch a pre-allocated page to use for indirect grefs */
388 		BUG_ON(list_empty(&rinfo->indirect_pages));
389 		indirect_page = list_first_entry(&rinfo->indirect_pages,
390 						 struct page, lru);
391 		list_del(&indirect_page->lru);
392 		gnt_list_entry->page = indirect_page;
393 	}
394 	grant_foreign_access(gnt_list_entry, info);
395 
396 	return gnt_list_entry;
397 }
398 
399 static const char *op_name(int op)
400 {
401 	static const char *const names[] = {
402 		[BLKIF_OP_READ] = "read",
403 		[BLKIF_OP_WRITE] = "write",
404 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
405 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
406 		[BLKIF_OP_DISCARD] = "discard" };
407 
408 	if (op < 0 || op >= ARRAY_SIZE(names))
409 		return "unknown";
410 
411 	if (!names[op])
412 		return "reserved";
413 
414 	return names[op];
415 }
416 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
417 {
418 	unsigned int end = minor + nr;
419 	int rc;
420 
421 	if (end > nr_minors) {
422 		unsigned long *bitmap, *old;
423 
424 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
425 				 GFP_KERNEL);
426 		if (bitmap == NULL)
427 			return -ENOMEM;
428 
429 		spin_lock(&minor_lock);
430 		if (end > nr_minors) {
431 			old = minors;
432 			memcpy(bitmap, minors,
433 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
434 			minors = bitmap;
435 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
436 		} else
437 			old = bitmap;
438 		spin_unlock(&minor_lock);
439 		kfree(old);
440 	}
441 
442 	spin_lock(&minor_lock);
443 	if (find_next_bit(minors, end, minor) >= end) {
444 		bitmap_set(minors, minor, nr);
445 		rc = 0;
446 	} else
447 		rc = -EBUSY;
448 	spin_unlock(&minor_lock);
449 
450 	return rc;
451 }
452 
453 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
454 {
455 	unsigned int end = minor + nr;
456 
457 	BUG_ON(end > nr_minors);
458 	spin_lock(&minor_lock);
459 	bitmap_clear(minors,  minor, nr);
460 	spin_unlock(&minor_lock);
461 }
462 
463 static void blkif_restart_queue_callback(void *arg)
464 {
465 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
466 	schedule_work(&rinfo->work);
467 }
468 
469 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
470 {
471 	/* We don't have real geometry info, but let's at least return
472 	   values consistent with the size of the device */
473 	sector_t nsect = get_capacity(bd->bd_disk);
474 	sector_t cylinders = nsect;
475 
476 	hg->heads = 0xff;
477 	hg->sectors = 0x3f;
478 	sector_div(cylinders, hg->heads * hg->sectors);
479 	hg->cylinders = cylinders;
480 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
481 		hg->cylinders = 0xffff;
482 	return 0;
483 }
484 
485 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
486 		       unsigned command, unsigned long argument)
487 {
488 	struct blkfront_info *info = bdev->bd_disk->private_data;
489 	int i;
490 
491 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
492 		command, (long)argument);
493 
494 	switch (command) {
495 	case CDROMMULTISESSION:
496 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
497 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
498 			if (put_user(0, (char __user *)(argument + i)))
499 				return -EFAULT;
500 		return 0;
501 
502 	case CDROM_GET_CAPABILITY: {
503 		struct gendisk *gd = info->gd;
504 		if (gd->flags & GENHD_FL_CD)
505 			return 0;
506 		return -EINVAL;
507 	}
508 
509 	default:
510 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
511 		  command);*/
512 		return -EINVAL; /* same return as native Linux */
513 	}
514 
515 	return 0;
516 }
517 
518 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
519 					    struct request *req,
520 					    struct blkif_request **ring_req)
521 {
522 	unsigned long id;
523 
524 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
525 	rinfo->ring.req_prod_pvt++;
526 
527 	id = get_id_from_freelist(rinfo);
528 	rinfo->shadow[id].request = req;
529 	rinfo->shadow[id].status = REQ_WAITING;
530 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
531 
532 	(*ring_req)->u.rw.id = id;
533 
534 	return id;
535 }
536 
537 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
538 {
539 	struct blkfront_info *info = rinfo->dev_info;
540 	struct blkif_request *ring_req;
541 	unsigned long id;
542 
543 	/* Fill out a communications ring structure. */
544 	id = blkif_ring_get_request(rinfo, req, &ring_req);
545 
546 	ring_req->operation = BLKIF_OP_DISCARD;
547 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
548 	ring_req->u.discard.id = id;
549 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
550 	if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
551 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
552 	else
553 		ring_req->u.discard.flag = 0;
554 
555 	/* Keep a private copy so we can reissue requests when recovering. */
556 	rinfo->shadow[id].req = *ring_req;
557 
558 	return 0;
559 }
560 
561 struct setup_rw_req {
562 	unsigned int grant_idx;
563 	struct blkif_request_segment *segments;
564 	struct blkfront_ring_info *rinfo;
565 	struct blkif_request *ring_req;
566 	grant_ref_t gref_head;
567 	unsigned int id;
568 	/* Only used when persistent grant is used and it's a read request */
569 	bool need_copy;
570 	unsigned int bvec_off;
571 	char *bvec_data;
572 
573 	bool require_extra_req;
574 	struct blkif_request *extra_ring_req;
575 };
576 
577 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
578 				     unsigned int len, void *data)
579 {
580 	struct setup_rw_req *setup = data;
581 	int n, ref;
582 	struct grant *gnt_list_entry;
583 	unsigned int fsect, lsect;
584 	/* Convenient aliases */
585 	unsigned int grant_idx = setup->grant_idx;
586 	struct blkif_request *ring_req = setup->ring_req;
587 	struct blkfront_ring_info *rinfo = setup->rinfo;
588 	/*
589 	 * We always use the shadow of the first request to store the list
590 	 * of grant associated to the block I/O request. This made the
591 	 * completion more easy to handle even if the block I/O request is
592 	 * split.
593 	 */
594 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
595 
596 	if (unlikely(setup->require_extra_req &&
597 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
598 		/*
599 		 * We are using the second request, setup grant_idx
600 		 * to be the index of the segment array.
601 		 */
602 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
603 		ring_req = setup->extra_ring_req;
604 	}
605 
606 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
607 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
608 		if (setup->segments)
609 			kunmap_atomic(setup->segments);
610 
611 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
612 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
613 		shadow->indirect_grants[n] = gnt_list_entry;
614 		setup->segments = kmap_atomic(gnt_list_entry->page);
615 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
616 	}
617 
618 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
619 	ref = gnt_list_entry->gref;
620 	/*
621 	 * All the grants are stored in the shadow of the first
622 	 * request. Therefore we have to use the global index.
623 	 */
624 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
625 
626 	if (setup->need_copy) {
627 		void *shared_data;
628 
629 		shared_data = kmap_atomic(gnt_list_entry->page);
630 		/*
631 		 * this does not wipe data stored outside the
632 		 * range sg->offset..sg->offset+sg->length.
633 		 * Therefore, blkback *could* see data from
634 		 * previous requests. This is OK as long as
635 		 * persistent grants are shared with just one
636 		 * domain. It may need refactoring if this
637 		 * changes
638 		 */
639 		memcpy(shared_data + offset,
640 		       setup->bvec_data + setup->bvec_off,
641 		       len);
642 
643 		kunmap_atomic(shared_data);
644 		setup->bvec_off += len;
645 	}
646 
647 	fsect = offset >> 9;
648 	lsect = fsect + (len >> 9) - 1;
649 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
650 		ring_req->u.rw.seg[grant_idx] =
651 			(struct blkif_request_segment) {
652 				.gref       = ref,
653 				.first_sect = fsect,
654 				.last_sect  = lsect };
655 	} else {
656 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
657 			(struct blkif_request_segment) {
658 				.gref       = ref,
659 				.first_sect = fsect,
660 				.last_sect  = lsect };
661 	}
662 
663 	(setup->grant_idx)++;
664 }
665 
666 static void blkif_setup_extra_req(struct blkif_request *first,
667 				  struct blkif_request *second)
668 {
669 	uint16_t nr_segments = first->u.rw.nr_segments;
670 
671 	/*
672 	 * The second request is only present when the first request uses
673 	 * all its segments. It's always the continuity of the first one.
674 	 */
675 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
676 
677 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
678 	second->u.rw.sector_number = first->u.rw.sector_number +
679 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
680 
681 	second->u.rw.handle = first->u.rw.handle;
682 	second->operation = first->operation;
683 }
684 
685 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
686 {
687 	struct blkfront_info *info = rinfo->dev_info;
688 	struct blkif_request *ring_req, *extra_ring_req = NULL;
689 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
690 	bool require_extra_req = false;
691 	int i;
692 	struct setup_rw_req setup = {
693 		.grant_idx = 0,
694 		.segments = NULL,
695 		.rinfo = rinfo,
696 		.need_copy = rq_data_dir(req) && info->feature_persistent,
697 	};
698 
699 	/*
700 	 * Used to store if we are able to queue the request by just using
701 	 * existing persistent grants, or if we have to get new grants,
702 	 * as there are not sufficiently many free.
703 	 */
704 	struct scatterlist *sg;
705 	int num_sg, max_grefs, num_grant;
706 
707 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
708 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
709 		/*
710 		 * If we are using indirect segments we need to account
711 		 * for the indirect grefs used in the request.
712 		 */
713 		max_grefs += INDIRECT_GREFS(max_grefs);
714 
715 	/*
716 	 * We have to reserve 'max_grefs' grants because persistent
717 	 * grants are shared by all rings.
718 	 */
719 	if (max_grefs > 0)
720 		if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
721 			gnttab_request_free_callback(
722 				&rinfo->callback,
723 				blkif_restart_queue_callback,
724 				rinfo,
725 				max_grefs);
726 			return 1;
727 		}
728 
729 	/* Fill out a communications ring structure. */
730 	id = blkif_ring_get_request(rinfo, req, &ring_req);
731 
732 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
733 	num_grant = 0;
734 	/* Calculate the number of grant used */
735 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
736 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
737 
738 	require_extra_req = info->max_indirect_segments == 0 &&
739 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
740 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
741 
742 	rinfo->shadow[id].num_sg = num_sg;
743 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
744 	    likely(!require_extra_req)) {
745 		/*
746 		 * The indirect operation can only be a BLKIF_OP_READ or
747 		 * BLKIF_OP_WRITE
748 		 */
749 		BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
750 		ring_req->operation = BLKIF_OP_INDIRECT;
751 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
752 			BLKIF_OP_WRITE : BLKIF_OP_READ;
753 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
754 		ring_req->u.indirect.handle = info->handle;
755 		ring_req->u.indirect.nr_segments = num_grant;
756 	} else {
757 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
758 		ring_req->u.rw.handle = info->handle;
759 		ring_req->operation = rq_data_dir(req) ?
760 			BLKIF_OP_WRITE : BLKIF_OP_READ;
761 		if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
762 			/*
763 			 * Ideally we can do an unordered flush-to-disk.
764 			 * In case the backend onlysupports barriers, use that.
765 			 * A barrier request a superset of FUA, so we can
766 			 * implement it the same way.  (It's also a FLUSH+FUA,
767 			 * since it is guaranteed ordered WRT previous writes.)
768 			 */
769 			switch (info->feature_flush &
770 				((REQ_FLUSH|REQ_FUA))) {
771 			case REQ_FLUSH|REQ_FUA:
772 				ring_req->operation =
773 					BLKIF_OP_WRITE_BARRIER;
774 				break;
775 			case REQ_FLUSH:
776 				ring_req->operation =
777 					BLKIF_OP_FLUSH_DISKCACHE;
778 				break;
779 			default:
780 				ring_req->operation = 0;
781 			}
782 		}
783 		ring_req->u.rw.nr_segments = num_grant;
784 		if (unlikely(require_extra_req)) {
785 			extra_id = blkif_ring_get_request(rinfo, req,
786 							  &extra_ring_req);
787 			/*
788 			 * Only the first request contains the scatter-gather
789 			 * list.
790 			 */
791 			rinfo->shadow[extra_id].num_sg = 0;
792 
793 			blkif_setup_extra_req(ring_req, extra_ring_req);
794 
795 			/* Link the 2 requests together */
796 			rinfo->shadow[extra_id].associated_id = id;
797 			rinfo->shadow[id].associated_id = extra_id;
798 		}
799 	}
800 
801 	setup.ring_req = ring_req;
802 	setup.id = id;
803 
804 	setup.require_extra_req = require_extra_req;
805 	if (unlikely(require_extra_req))
806 		setup.extra_ring_req = extra_ring_req;
807 
808 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
809 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
810 
811 		if (setup.need_copy) {
812 			setup.bvec_off = sg->offset;
813 			setup.bvec_data = kmap_atomic(sg_page(sg));
814 		}
815 
816 		gnttab_foreach_grant_in_range(sg_page(sg),
817 					      sg->offset,
818 					      sg->length,
819 					      blkif_setup_rw_req_grant,
820 					      &setup);
821 
822 		if (setup.need_copy)
823 			kunmap_atomic(setup.bvec_data);
824 	}
825 	if (setup.segments)
826 		kunmap_atomic(setup.segments);
827 
828 	/* Keep a private copy so we can reissue requests when recovering. */
829 	rinfo->shadow[id].req = *ring_req;
830 	if (unlikely(require_extra_req))
831 		rinfo->shadow[extra_id].req = *extra_ring_req;
832 
833 	if (max_grefs > 0)
834 		gnttab_free_grant_references(setup.gref_head);
835 
836 	return 0;
837 }
838 
839 /*
840  * Generate a Xen blkfront IO request from a blk layer request.  Reads
841  * and writes are handled as expected.
842  *
843  * @req: a request struct
844  */
845 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
846 {
847 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
848 		return 1;
849 
850 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
851 		return blkif_queue_discard_req(req, rinfo);
852 	else
853 		return blkif_queue_rw_req(req, rinfo);
854 }
855 
856 static inline void flush_requests(struct blkfront_ring_info *rinfo)
857 {
858 	int notify;
859 
860 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
861 
862 	if (notify)
863 		notify_remote_via_irq(rinfo->irq);
864 }
865 
866 static inline bool blkif_request_flush_invalid(struct request *req,
867 					       struct blkfront_info *info)
868 {
869 	return ((req->cmd_type != REQ_TYPE_FS) ||
870 		((req->cmd_flags & REQ_FLUSH) &&
871 		 !(info->feature_flush & REQ_FLUSH)) ||
872 		((req->cmd_flags & REQ_FUA) &&
873 		 !(info->feature_flush & REQ_FUA)));
874 }
875 
876 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
877 			  const struct blk_mq_queue_data *qd)
878 {
879 	unsigned long flags;
880 	int qid = hctx->queue_num;
881 	struct blkfront_info *info = hctx->queue->queuedata;
882 	struct blkfront_ring_info *rinfo = NULL;
883 
884 	BUG_ON(info->nr_rings <= qid);
885 	rinfo = &info->rinfo[qid];
886 	blk_mq_start_request(qd->rq);
887 	spin_lock_irqsave(&rinfo->ring_lock, flags);
888 	if (RING_FULL(&rinfo->ring))
889 		goto out_busy;
890 
891 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
892 		goto out_err;
893 
894 	if (blkif_queue_request(qd->rq, rinfo))
895 		goto out_busy;
896 
897 	flush_requests(rinfo);
898 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
899 	return BLK_MQ_RQ_QUEUE_OK;
900 
901 out_err:
902 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
903 	return BLK_MQ_RQ_QUEUE_ERROR;
904 
905 out_busy:
906 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
907 	blk_mq_stop_hw_queue(hctx);
908 	return BLK_MQ_RQ_QUEUE_BUSY;
909 }
910 
911 static struct blk_mq_ops blkfront_mq_ops = {
912 	.queue_rq = blkif_queue_rq,
913 	.map_queue = blk_mq_map_queue,
914 };
915 
916 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
917 				unsigned int physical_sector_size,
918 				unsigned int segments)
919 {
920 	struct request_queue *rq;
921 	struct blkfront_info *info = gd->private_data;
922 
923 	memset(&info->tag_set, 0, sizeof(info->tag_set));
924 	info->tag_set.ops = &blkfront_mq_ops;
925 	info->tag_set.nr_hw_queues = info->nr_rings;
926 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
927 		/*
928 		 * When indirect descriptior is not supported, the I/O request
929 		 * will be split between multiple request in the ring.
930 		 * To avoid problems when sending the request, divide by
931 		 * 2 the depth of the queue.
932 		 */
933 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
934 	} else
935 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
936 	info->tag_set.numa_node = NUMA_NO_NODE;
937 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
938 	info->tag_set.cmd_size = 0;
939 	info->tag_set.driver_data = info;
940 
941 	if (blk_mq_alloc_tag_set(&info->tag_set))
942 		return -EINVAL;
943 	rq = blk_mq_init_queue(&info->tag_set);
944 	if (IS_ERR(rq)) {
945 		blk_mq_free_tag_set(&info->tag_set);
946 		return PTR_ERR(rq);
947 	}
948 
949 	rq->queuedata = info;
950 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
951 
952 	if (info->feature_discard) {
953 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
954 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
955 		rq->limits.discard_granularity = info->discard_granularity;
956 		rq->limits.discard_alignment = info->discard_alignment;
957 		if (info->feature_secdiscard)
958 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
959 	}
960 
961 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
962 	blk_queue_logical_block_size(rq, sector_size);
963 	blk_queue_physical_block_size(rq, physical_sector_size);
964 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
965 
966 	/* Each segment in a request is up to an aligned page in size. */
967 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
968 	blk_queue_max_segment_size(rq, PAGE_SIZE);
969 
970 	/* Ensure a merged request will fit in a single I/O ring slot. */
971 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
972 
973 	/* Make sure buffer addresses are sector-aligned. */
974 	blk_queue_dma_alignment(rq, 511);
975 
976 	/* Make sure we don't use bounce buffers. */
977 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
978 
979 	gd->queue = rq;
980 
981 	return 0;
982 }
983 
984 static const char *flush_info(unsigned int feature_flush)
985 {
986 	switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
987 	case REQ_FLUSH|REQ_FUA:
988 		return "barrier: enabled;";
989 	case REQ_FLUSH:
990 		return "flush diskcache: enabled;";
991 	default:
992 		return "barrier or flush: disabled;";
993 	}
994 }
995 
996 static void xlvbd_flush(struct blkfront_info *info)
997 {
998 	blk_queue_write_cache(info->rq, info->feature_flush & REQ_FLUSH,
999 				info->feature_flush & REQ_FUA);
1000 	pr_info("blkfront: %s: %s %s %s %s %s\n",
1001 		info->gd->disk_name, flush_info(info->feature_flush),
1002 		"persistent grants:", info->feature_persistent ?
1003 		"enabled;" : "disabled;", "indirect descriptors:",
1004 		info->max_indirect_segments ? "enabled;" : "disabled;");
1005 }
1006 
1007 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1008 {
1009 	int major;
1010 	major = BLKIF_MAJOR(vdevice);
1011 	*minor = BLKIF_MINOR(vdevice);
1012 	switch (major) {
1013 		case XEN_IDE0_MAJOR:
1014 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1015 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1016 				EMULATED_HD_DISK_MINOR_OFFSET;
1017 			break;
1018 		case XEN_IDE1_MAJOR:
1019 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1020 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1021 				EMULATED_HD_DISK_MINOR_OFFSET;
1022 			break;
1023 		case XEN_SCSI_DISK0_MAJOR:
1024 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1025 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1026 			break;
1027 		case XEN_SCSI_DISK1_MAJOR:
1028 		case XEN_SCSI_DISK2_MAJOR:
1029 		case XEN_SCSI_DISK3_MAJOR:
1030 		case XEN_SCSI_DISK4_MAJOR:
1031 		case XEN_SCSI_DISK5_MAJOR:
1032 		case XEN_SCSI_DISK6_MAJOR:
1033 		case XEN_SCSI_DISK7_MAJOR:
1034 			*offset = (*minor / PARTS_PER_DISK) +
1035 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1036 				EMULATED_SD_DISK_NAME_OFFSET;
1037 			*minor = *minor +
1038 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1039 				EMULATED_SD_DISK_MINOR_OFFSET;
1040 			break;
1041 		case XEN_SCSI_DISK8_MAJOR:
1042 		case XEN_SCSI_DISK9_MAJOR:
1043 		case XEN_SCSI_DISK10_MAJOR:
1044 		case XEN_SCSI_DISK11_MAJOR:
1045 		case XEN_SCSI_DISK12_MAJOR:
1046 		case XEN_SCSI_DISK13_MAJOR:
1047 		case XEN_SCSI_DISK14_MAJOR:
1048 		case XEN_SCSI_DISK15_MAJOR:
1049 			*offset = (*minor / PARTS_PER_DISK) +
1050 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1051 				EMULATED_SD_DISK_NAME_OFFSET;
1052 			*minor = *minor +
1053 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1054 				EMULATED_SD_DISK_MINOR_OFFSET;
1055 			break;
1056 		case XENVBD_MAJOR:
1057 			*offset = *minor / PARTS_PER_DISK;
1058 			break;
1059 		default:
1060 			printk(KERN_WARNING "blkfront: your disk configuration is "
1061 					"incorrect, please use an xvd device instead\n");
1062 			return -ENODEV;
1063 	}
1064 	return 0;
1065 }
1066 
1067 static char *encode_disk_name(char *ptr, unsigned int n)
1068 {
1069 	if (n >= 26)
1070 		ptr = encode_disk_name(ptr, n / 26 - 1);
1071 	*ptr = 'a' + n % 26;
1072 	return ptr + 1;
1073 }
1074 
1075 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1076 			       struct blkfront_info *info,
1077 			       u16 vdisk_info, u16 sector_size,
1078 			       unsigned int physical_sector_size)
1079 {
1080 	struct gendisk *gd;
1081 	int nr_minors = 1;
1082 	int err;
1083 	unsigned int offset;
1084 	int minor;
1085 	int nr_parts;
1086 	char *ptr;
1087 
1088 	BUG_ON(info->gd != NULL);
1089 	BUG_ON(info->rq != NULL);
1090 
1091 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1092 		/* this is above the extended range; something is wrong */
1093 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1094 		return -ENODEV;
1095 	}
1096 
1097 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1098 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1099 		if (err)
1100 			return err;
1101  		nr_parts = PARTS_PER_DISK;
1102 	} else {
1103 		minor = BLKIF_MINOR_EXT(info->vdevice);
1104 		nr_parts = PARTS_PER_EXT_DISK;
1105 		offset = minor / nr_parts;
1106 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1107 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1108 					"emulated IDE disks,\n\t choose an xvd device name"
1109 					"from xvde on\n", info->vdevice);
1110 	}
1111 	if (minor >> MINORBITS) {
1112 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1113 			info->vdevice, minor);
1114 		return -ENODEV;
1115 	}
1116 
1117 	if ((minor % nr_parts) == 0)
1118 		nr_minors = nr_parts;
1119 
1120 	err = xlbd_reserve_minors(minor, nr_minors);
1121 	if (err)
1122 		goto out;
1123 	err = -ENODEV;
1124 
1125 	gd = alloc_disk(nr_minors);
1126 	if (gd == NULL)
1127 		goto release;
1128 
1129 	strcpy(gd->disk_name, DEV_NAME);
1130 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1131 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1132 	if (nr_minors > 1)
1133 		*ptr = 0;
1134 	else
1135 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1136 			 "%d", minor & (nr_parts - 1));
1137 
1138 	gd->major = XENVBD_MAJOR;
1139 	gd->first_minor = minor;
1140 	gd->fops = &xlvbd_block_fops;
1141 	gd->private_data = info;
1142 	gd->driverfs_dev = &(info->xbdev->dev);
1143 	set_capacity(gd, capacity);
1144 
1145 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1146 				 info->max_indirect_segments ? :
1147 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1148 		del_gendisk(gd);
1149 		goto release;
1150 	}
1151 
1152 	info->rq = gd->queue;
1153 	info->gd = gd;
1154 
1155 	xlvbd_flush(info);
1156 
1157 	if (vdisk_info & VDISK_READONLY)
1158 		set_disk_ro(gd, 1);
1159 
1160 	if (vdisk_info & VDISK_REMOVABLE)
1161 		gd->flags |= GENHD_FL_REMOVABLE;
1162 
1163 	if (vdisk_info & VDISK_CDROM)
1164 		gd->flags |= GENHD_FL_CD;
1165 
1166 	return 0;
1167 
1168  release:
1169 	xlbd_release_minors(minor, nr_minors);
1170  out:
1171 	return err;
1172 }
1173 
1174 static void xlvbd_release_gendisk(struct blkfront_info *info)
1175 {
1176 	unsigned int minor, nr_minors, i;
1177 
1178 	if (info->rq == NULL)
1179 		return;
1180 
1181 	/* No more blkif_request(). */
1182 	blk_mq_stop_hw_queues(info->rq);
1183 
1184 	for (i = 0; i < info->nr_rings; i++) {
1185 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1186 
1187 		/* No more gnttab callback work. */
1188 		gnttab_cancel_free_callback(&rinfo->callback);
1189 
1190 		/* Flush gnttab callback work. Must be done with no locks held. */
1191 		flush_work(&rinfo->work);
1192 	}
1193 
1194 	del_gendisk(info->gd);
1195 
1196 	minor = info->gd->first_minor;
1197 	nr_minors = info->gd->minors;
1198 	xlbd_release_minors(minor, nr_minors);
1199 
1200 	blk_cleanup_queue(info->rq);
1201 	blk_mq_free_tag_set(&info->tag_set);
1202 	info->rq = NULL;
1203 
1204 	put_disk(info->gd);
1205 	info->gd = NULL;
1206 }
1207 
1208 /* Already hold rinfo->ring_lock. */
1209 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1210 {
1211 	if (!RING_FULL(&rinfo->ring))
1212 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1213 }
1214 
1215 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1216 {
1217 	unsigned long flags;
1218 
1219 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1220 	kick_pending_request_queues_locked(rinfo);
1221 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1222 }
1223 
1224 static void blkif_restart_queue(struct work_struct *work)
1225 {
1226 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1227 
1228 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1229 		kick_pending_request_queues(rinfo);
1230 }
1231 
1232 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1233 {
1234 	struct grant *persistent_gnt, *n;
1235 	struct blkfront_info *info = rinfo->dev_info;
1236 	int i, j, segs;
1237 
1238 	/*
1239 	 * Remove indirect pages, this only happens when using indirect
1240 	 * descriptors but not persistent grants
1241 	 */
1242 	if (!list_empty(&rinfo->indirect_pages)) {
1243 		struct page *indirect_page, *n;
1244 
1245 		BUG_ON(info->feature_persistent);
1246 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1247 			list_del(&indirect_page->lru);
1248 			__free_page(indirect_page);
1249 		}
1250 	}
1251 
1252 	/* Remove all persistent grants. */
1253 	if (!list_empty(&rinfo->grants)) {
1254 		list_for_each_entry_safe(persistent_gnt, n,
1255 					 &rinfo->grants, node) {
1256 			list_del(&persistent_gnt->node);
1257 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1258 				gnttab_end_foreign_access(persistent_gnt->gref,
1259 							  0, 0UL);
1260 				rinfo->persistent_gnts_c--;
1261 			}
1262 			if (info->feature_persistent)
1263 				__free_page(persistent_gnt->page);
1264 			kfree(persistent_gnt);
1265 		}
1266 	}
1267 	BUG_ON(rinfo->persistent_gnts_c != 0);
1268 
1269 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1270 		/*
1271 		 * Clear persistent grants present in requests already
1272 		 * on the shared ring
1273 		 */
1274 		if (!rinfo->shadow[i].request)
1275 			goto free_shadow;
1276 
1277 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1278 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1279 		       rinfo->shadow[i].req.u.rw.nr_segments;
1280 		for (j = 0; j < segs; j++) {
1281 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1282 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1283 			if (info->feature_persistent)
1284 				__free_page(persistent_gnt->page);
1285 			kfree(persistent_gnt);
1286 		}
1287 
1288 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1289 			/*
1290 			 * If this is not an indirect operation don't try to
1291 			 * free indirect segments
1292 			 */
1293 			goto free_shadow;
1294 
1295 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1296 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1297 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1298 			__free_page(persistent_gnt->page);
1299 			kfree(persistent_gnt);
1300 		}
1301 
1302 free_shadow:
1303 		kfree(rinfo->shadow[i].grants_used);
1304 		rinfo->shadow[i].grants_used = NULL;
1305 		kfree(rinfo->shadow[i].indirect_grants);
1306 		rinfo->shadow[i].indirect_grants = NULL;
1307 		kfree(rinfo->shadow[i].sg);
1308 		rinfo->shadow[i].sg = NULL;
1309 	}
1310 
1311 	/* No more gnttab callback work. */
1312 	gnttab_cancel_free_callback(&rinfo->callback);
1313 
1314 	/* Flush gnttab callback work. Must be done with no locks held. */
1315 	flush_work(&rinfo->work);
1316 
1317 	/* Free resources associated with old device channel. */
1318 	for (i = 0; i < info->nr_ring_pages; i++) {
1319 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1320 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1321 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1322 		}
1323 	}
1324 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1325 	rinfo->ring.sring = NULL;
1326 
1327 	if (rinfo->irq)
1328 		unbind_from_irqhandler(rinfo->irq, rinfo);
1329 	rinfo->evtchn = rinfo->irq = 0;
1330 }
1331 
1332 static void blkif_free(struct blkfront_info *info, int suspend)
1333 {
1334 	unsigned int i;
1335 
1336 	/* Prevent new requests being issued until we fix things up. */
1337 	info->connected = suspend ?
1338 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1339 	/* No more blkif_request(). */
1340 	if (info->rq)
1341 		blk_mq_stop_hw_queues(info->rq);
1342 
1343 	for (i = 0; i < info->nr_rings; i++)
1344 		blkif_free_ring(&info->rinfo[i]);
1345 
1346 	kfree(info->rinfo);
1347 	info->rinfo = NULL;
1348 	info->nr_rings = 0;
1349 }
1350 
1351 struct copy_from_grant {
1352 	const struct blk_shadow *s;
1353 	unsigned int grant_idx;
1354 	unsigned int bvec_offset;
1355 	char *bvec_data;
1356 };
1357 
1358 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1359 				  unsigned int len, void *data)
1360 {
1361 	struct copy_from_grant *info = data;
1362 	char *shared_data;
1363 	/* Convenient aliases */
1364 	const struct blk_shadow *s = info->s;
1365 
1366 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1367 
1368 	memcpy(info->bvec_data + info->bvec_offset,
1369 	       shared_data + offset, len);
1370 
1371 	info->bvec_offset += len;
1372 	info->grant_idx++;
1373 
1374 	kunmap_atomic(shared_data);
1375 }
1376 
1377 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1378 {
1379 	switch (rsp)
1380 	{
1381 	case BLKIF_RSP_OKAY:
1382 		return REQ_DONE;
1383 	case BLKIF_RSP_EOPNOTSUPP:
1384 		return REQ_EOPNOTSUPP;
1385 	case BLKIF_RSP_ERROR:
1386 		/* Fallthrough. */
1387 	default:
1388 		return REQ_ERROR;
1389 	}
1390 }
1391 
1392 /*
1393  * Get the final status of the block request based on two ring response
1394  */
1395 static int blkif_get_final_status(enum blk_req_status s1,
1396 				  enum blk_req_status s2)
1397 {
1398 	BUG_ON(s1 == REQ_WAITING);
1399 	BUG_ON(s2 == REQ_WAITING);
1400 
1401 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1402 		return BLKIF_RSP_ERROR;
1403 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1404 		return BLKIF_RSP_EOPNOTSUPP;
1405 	return BLKIF_RSP_OKAY;
1406 }
1407 
1408 static bool blkif_completion(unsigned long *id,
1409 			     struct blkfront_ring_info *rinfo,
1410 			     struct blkif_response *bret)
1411 {
1412 	int i = 0;
1413 	struct scatterlist *sg;
1414 	int num_sg, num_grant;
1415 	struct blkfront_info *info = rinfo->dev_info;
1416 	struct blk_shadow *s = &rinfo->shadow[*id];
1417 	struct copy_from_grant data = {
1418 		.grant_idx = 0,
1419 	};
1420 
1421 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1422 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1423 
1424 	/* The I/O request may be split in two. */
1425 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1426 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1427 
1428 		/* Keep the status of the current response in shadow. */
1429 		s->status = blkif_rsp_to_req_status(bret->status);
1430 
1431 		/* Wait the second response if not yet here. */
1432 		if (s2->status == REQ_WAITING)
1433 			return 0;
1434 
1435 		bret->status = blkif_get_final_status(s->status,
1436 						      s2->status);
1437 
1438 		/*
1439 		 * All the grants is stored in the first shadow in order
1440 		 * to make the completion code simpler.
1441 		 */
1442 		num_grant += s2->req.u.rw.nr_segments;
1443 
1444 		/*
1445 		 * The two responses may not come in order. Only the
1446 		 * first request will store the scatter-gather list.
1447 		 */
1448 		if (s2->num_sg != 0) {
1449 			/* Update "id" with the ID of the first response. */
1450 			*id = s->associated_id;
1451 			s = s2;
1452 		}
1453 
1454 		/*
1455 		 * We don't need anymore the second request, so recycling
1456 		 * it now.
1457 		 */
1458 		if (add_id_to_freelist(rinfo, s->associated_id))
1459 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1460 			     info->gd->disk_name, s->associated_id);
1461 	}
1462 
1463 	data.s = s;
1464 	num_sg = s->num_sg;
1465 
1466 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1467 		for_each_sg(s->sg, sg, num_sg, i) {
1468 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1469 
1470 			data.bvec_offset = sg->offset;
1471 			data.bvec_data = kmap_atomic(sg_page(sg));
1472 
1473 			gnttab_foreach_grant_in_range(sg_page(sg),
1474 						      sg->offset,
1475 						      sg->length,
1476 						      blkif_copy_from_grant,
1477 						      &data);
1478 
1479 			kunmap_atomic(data.bvec_data);
1480 		}
1481 	}
1482 	/* Add the persistent grant into the list of free grants */
1483 	for (i = 0; i < num_grant; i++) {
1484 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1485 			/*
1486 			 * If the grant is still mapped by the backend (the
1487 			 * backend has chosen to make this grant persistent)
1488 			 * we add it at the head of the list, so it will be
1489 			 * reused first.
1490 			 */
1491 			if (!info->feature_persistent)
1492 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1493 						     s->grants_used[i]->gref);
1494 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1495 			rinfo->persistent_gnts_c++;
1496 		} else {
1497 			/*
1498 			 * If the grant is not mapped by the backend we end the
1499 			 * foreign access and add it to the tail of the list,
1500 			 * so it will not be picked again unless we run out of
1501 			 * persistent grants.
1502 			 */
1503 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1504 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1505 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1506 		}
1507 	}
1508 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1509 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1510 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1511 				if (!info->feature_persistent)
1512 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1513 							     s->indirect_grants[i]->gref);
1514 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1515 				rinfo->persistent_gnts_c++;
1516 			} else {
1517 				struct page *indirect_page;
1518 
1519 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1520 				/*
1521 				 * Add the used indirect page back to the list of
1522 				 * available pages for indirect grefs.
1523 				 */
1524 				if (!info->feature_persistent) {
1525 					indirect_page = s->indirect_grants[i]->page;
1526 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1527 				}
1528 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1529 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1530 			}
1531 		}
1532 	}
1533 
1534 	return 1;
1535 }
1536 
1537 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1538 {
1539 	struct request *req;
1540 	struct blkif_response *bret;
1541 	RING_IDX i, rp;
1542 	unsigned long flags;
1543 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1544 	struct blkfront_info *info = rinfo->dev_info;
1545 	int error;
1546 
1547 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1548 		return IRQ_HANDLED;
1549 
1550 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1551  again:
1552 	rp = rinfo->ring.sring->rsp_prod;
1553 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1554 
1555 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1556 		unsigned long id;
1557 
1558 		bret = RING_GET_RESPONSE(&rinfo->ring, i);
1559 		id   = bret->id;
1560 		/*
1561 		 * The backend has messed up and given us an id that we would
1562 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1563 		 * look in get_id_from_freelist.
1564 		 */
1565 		if (id >= BLK_RING_SIZE(info)) {
1566 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1567 			     info->gd->disk_name, op_name(bret->operation), id);
1568 			/* We can't safely get the 'struct request' as
1569 			 * the id is busted. */
1570 			continue;
1571 		}
1572 		req  = rinfo->shadow[id].request;
1573 
1574 		if (bret->operation != BLKIF_OP_DISCARD) {
1575 			/*
1576 			 * We may need to wait for an extra response if the
1577 			 * I/O request is split in 2
1578 			 */
1579 			if (!blkif_completion(&id, rinfo, bret))
1580 				continue;
1581 		}
1582 
1583 		if (add_id_to_freelist(rinfo, id)) {
1584 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1585 			     info->gd->disk_name, op_name(bret->operation), id);
1586 			continue;
1587 		}
1588 
1589 		error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1590 		switch (bret->operation) {
1591 		case BLKIF_OP_DISCARD:
1592 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1593 				struct request_queue *rq = info->rq;
1594 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1595 					   info->gd->disk_name, op_name(bret->operation));
1596 				error = -EOPNOTSUPP;
1597 				info->feature_discard = 0;
1598 				info->feature_secdiscard = 0;
1599 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1600 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1601 			}
1602 			blk_mq_complete_request(req, error);
1603 			break;
1604 		case BLKIF_OP_FLUSH_DISKCACHE:
1605 		case BLKIF_OP_WRITE_BARRIER:
1606 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1607 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1608 				       info->gd->disk_name, op_name(bret->operation));
1609 				error = -EOPNOTSUPP;
1610 			}
1611 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1612 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1613 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1614 				       info->gd->disk_name, op_name(bret->operation));
1615 				error = -EOPNOTSUPP;
1616 			}
1617 			if (unlikely(error)) {
1618 				if (error == -EOPNOTSUPP)
1619 					error = 0;
1620 				info->feature_flush = 0;
1621 				xlvbd_flush(info);
1622 			}
1623 			/* fall through */
1624 		case BLKIF_OP_READ:
1625 		case BLKIF_OP_WRITE:
1626 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1627 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1628 					"request: %x\n", bret->status);
1629 
1630 			blk_mq_complete_request(req, error);
1631 			break;
1632 		default:
1633 			BUG();
1634 		}
1635 	}
1636 
1637 	rinfo->ring.rsp_cons = i;
1638 
1639 	if (i != rinfo->ring.req_prod_pvt) {
1640 		int more_to_do;
1641 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1642 		if (more_to_do)
1643 			goto again;
1644 	} else
1645 		rinfo->ring.sring->rsp_event = i + 1;
1646 
1647 	kick_pending_request_queues_locked(rinfo);
1648 
1649 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1650 
1651 	return IRQ_HANDLED;
1652 }
1653 
1654 
1655 static int setup_blkring(struct xenbus_device *dev,
1656 			 struct blkfront_ring_info *rinfo)
1657 {
1658 	struct blkif_sring *sring;
1659 	int err, i;
1660 	struct blkfront_info *info = rinfo->dev_info;
1661 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1662 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1663 
1664 	for (i = 0; i < info->nr_ring_pages; i++)
1665 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1666 
1667 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1668 						       get_order(ring_size));
1669 	if (!sring) {
1670 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1671 		return -ENOMEM;
1672 	}
1673 	SHARED_RING_INIT(sring);
1674 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1675 
1676 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1677 	if (err < 0) {
1678 		free_pages((unsigned long)sring, get_order(ring_size));
1679 		rinfo->ring.sring = NULL;
1680 		goto fail;
1681 	}
1682 	for (i = 0; i < info->nr_ring_pages; i++)
1683 		rinfo->ring_ref[i] = gref[i];
1684 
1685 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1686 	if (err)
1687 		goto fail;
1688 
1689 	err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1690 					"blkif", rinfo);
1691 	if (err <= 0) {
1692 		xenbus_dev_fatal(dev, err,
1693 				 "bind_evtchn_to_irqhandler failed");
1694 		goto fail;
1695 	}
1696 	rinfo->irq = err;
1697 
1698 	return 0;
1699 fail:
1700 	blkif_free(info, 0);
1701 	return err;
1702 }
1703 
1704 /*
1705  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1706  * ring buffer may have multi pages depending on ->nr_ring_pages.
1707  */
1708 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1709 				struct blkfront_ring_info *rinfo, const char *dir)
1710 {
1711 	int err;
1712 	unsigned int i;
1713 	const char *message = NULL;
1714 	struct blkfront_info *info = rinfo->dev_info;
1715 
1716 	if (info->nr_ring_pages == 1) {
1717 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1718 		if (err) {
1719 			message = "writing ring-ref";
1720 			goto abort_transaction;
1721 		}
1722 	} else {
1723 		for (i = 0; i < info->nr_ring_pages; i++) {
1724 			char ring_ref_name[RINGREF_NAME_LEN];
1725 
1726 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1727 			err = xenbus_printf(xbt, dir, ring_ref_name,
1728 					    "%u", rinfo->ring_ref[i]);
1729 			if (err) {
1730 				message = "writing ring-ref";
1731 				goto abort_transaction;
1732 			}
1733 		}
1734 	}
1735 
1736 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1737 	if (err) {
1738 		message = "writing event-channel";
1739 		goto abort_transaction;
1740 	}
1741 
1742 	return 0;
1743 
1744 abort_transaction:
1745 	xenbus_transaction_end(xbt, 1);
1746 	if (message)
1747 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1748 
1749 	return err;
1750 }
1751 
1752 /* Common code used when first setting up, and when resuming. */
1753 static int talk_to_blkback(struct xenbus_device *dev,
1754 			   struct blkfront_info *info)
1755 {
1756 	const char *message = NULL;
1757 	struct xenbus_transaction xbt;
1758 	int err;
1759 	unsigned int i, max_page_order = 0;
1760 	unsigned int ring_page_order = 0;
1761 
1762 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1763 			   "max-ring-page-order", "%u", &max_page_order);
1764 	if (err != 1)
1765 		info->nr_ring_pages = 1;
1766 	else {
1767 		ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1768 		info->nr_ring_pages = 1 << ring_page_order;
1769 	}
1770 
1771 	for (i = 0; i < info->nr_rings; i++) {
1772 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1773 
1774 		/* Create shared ring, alloc event channel. */
1775 		err = setup_blkring(dev, rinfo);
1776 		if (err)
1777 			goto destroy_blkring;
1778 	}
1779 
1780 again:
1781 	err = xenbus_transaction_start(&xbt);
1782 	if (err) {
1783 		xenbus_dev_fatal(dev, err, "starting transaction");
1784 		goto destroy_blkring;
1785 	}
1786 
1787 	if (info->nr_ring_pages > 1) {
1788 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1789 				    ring_page_order);
1790 		if (err) {
1791 			message = "writing ring-page-order";
1792 			goto abort_transaction;
1793 		}
1794 	}
1795 
1796 	/* We already got the number of queues/rings in _probe */
1797 	if (info->nr_rings == 1) {
1798 		err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1799 		if (err)
1800 			goto destroy_blkring;
1801 	} else {
1802 		char *path;
1803 		size_t pathsize;
1804 
1805 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1806 				    info->nr_rings);
1807 		if (err) {
1808 			message = "writing multi-queue-num-queues";
1809 			goto abort_transaction;
1810 		}
1811 
1812 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1813 		path = kmalloc(pathsize, GFP_KERNEL);
1814 		if (!path) {
1815 			err = -ENOMEM;
1816 			message = "ENOMEM while writing ring references";
1817 			goto abort_transaction;
1818 		}
1819 
1820 		for (i = 0; i < info->nr_rings; i++) {
1821 			memset(path, 0, pathsize);
1822 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1823 			err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1824 			if (err) {
1825 				kfree(path);
1826 				goto destroy_blkring;
1827 			}
1828 		}
1829 		kfree(path);
1830 	}
1831 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1832 			    XEN_IO_PROTO_ABI_NATIVE);
1833 	if (err) {
1834 		message = "writing protocol";
1835 		goto abort_transaction;
1836 	}
1837 	err = xenbus_printf(xbt, dev->nodename,
1838 			    "feature-persistent", "%u", 1);
1839 	if (err)
1840 		dev_warn(&dev->dev,
1841 			 "writing persistent grants feature to xenbus");
1842 
1843 	err = xenbus_transaction_end(xbt, 0);
1844 	if (err) {
1845 		if (err == -EAGAIN)
1846 			goto again;
1847 		xenbus_dev_fatal(dev, err, "completing transaction");
1848 		goto destroy_blkring;
1849 	}
1850 
1851 	for (i = 0; i < info->nr_rings; i++) {
1852 		unsigned int j;
1853 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1854 
1855 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1856 			rinfo->shadow[j].req.u.rw.id = j + 1;
1857 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1858 	}
1859 	xenbus_switch_state(dev, XenbusStateInitialised);
1860 
1861 	return 0;
1862 
1863  abort_transaction:
1864 	xenbus_transaction_end(xbt, 1);
1865 	if (message)
1866 		xenbus_dev_fatal(dev, err, "%s", message);
1867  destroy_blkring:
1868 	blkif_free(info, 0);
1869 
1870 	kfree(info);
1871 	dev_set_drvdata(&dev->dev, NULL);
1872 
1873 	return err;
1874 }
1875 
1876 static int negotiate_mq(struct blkfront_info *info)
1877 {
1878 	unsigned int backend_max_queues = 0;
1879 	int err;
1880 	unsigned int i;
1881 
1882 	BUG_ON(info->nr_rings);
1883 
1884 	/* Check if backend supports multiple queues. */
1885 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1886 			   "multi-queue-max-queues", "%u", &backend_max_queues);
1887 	if (err < 0)
1888 		backend_max_queues = 1;
1889 
1890 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1891 	/* We need at least one ring. */
1892 	if (!info->nr_rings)
1893 		info->nr_rings = 1;
1894 
1895 	info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1896 	if (!info->rinfo) {
1897 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1898 		return -ENOMEM;
1899 	}
1900 
1901 	for (i = 0; i < info->nr_rings; i++) {
1902 		struct blkfront_ring_info *rinfo;
1903 
1904 		rinfo = &info->rinfo[i];
1905 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1906 		INIT_LIST_HEAD(&rinfo->grants);
1907 		rinfo->dev_info = info;
1908 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1909 		spin_lock_init(&rinfo->ring_lock);
1910 	}
1911 	return 0;
1912 }
1913 /**
1914  * Entry point to this code when a new device is created.  Allocate the basic
1915  * structures and the ring buffer for communication with the backend, and
1916  * inform the backend of the appropriate details for those.  Switch to
1917  * Initialised state.
1918  */
1919 static int blkfront_probe(struct xenbus_device *dev,
1920 			  const struct xenbus_device_id *id)
1921 {
1922 	int err, vdevice;
1923 	struct blkfront_info *info;
1924 
1925 	/* FIXME: Use dynamic device id if this is not set. */
1926 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1927 			   "virtual-device", "%i", &vdevice);
1928 	if (err != 1) {
1929 		/* go looking in the extended area instead */
1930 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1931 				   "%i", &vdevice);
1932 		if (err != 1) {
1933 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1934 			return err;
1935 		}
1936 	}
1937 
1938 	if (xen_hvm_domain()) {
1939 		char *type;
1940 		int len;
1941 		/* no unplug has been done: do not hook devices != xen vbds */
1942 		if (xen_has_pv_and_legacy_disk_devices()) {
1943 			int major;
1944 
1945 			if (!VDEV_IS_EXTENDED(vdevice))
1946 				major = BLKIF_MAJOR(vdevice);
1947 			else
1948 				major = XENVBD_MAJOR;
1949 
1950 			if (major != XENVBD_MAJOR) {
1951 				printk(KERN_INFO
1952 						"%s: HVM does not support vbd %d as xen block device\n",
1953 						__func__, vdevice);
1954 				return -ENODEV;
1955 			}
1956 		}
1957 		/* do not create a PV cdrom device if we are an HVM guest */
1958 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1959 		if (IS_ERR(type))
1960 			return -ENODEV;
1961 		if (strncmp(type, "cdrom", 5) == 0) {
1962 			kfree(type);
1963 			return -ENODEV;
1964 		}
1965 		kfree(type);
1966 	}
1967 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1968 	if (!info) {
1969 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1970 		return -ENOMEM;
1971 	}
1972 
1973 	info->xbdev = dev;
1974 	err = negotiate_mq(info);
1975 	if (err) {
1976 		kfree(info);
1977 		return err;
1978 	}
1979 
1980 	mutex_init(&info->mutex);
1981 	info->vdevice = vdevice;
1982 	info->connected = BLKIF_STATE_DISCONNECTED;
1983 
1984 	/* Front end dir is a number, which is used as the id. */
1985 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1986 	dev_set_drvdata(&dev->dev, info);
1987 
1988 	return 0;
1989 }
1990 
1991 static void split_bio_end(struct bio *bio)
1992 {
1993 	struct split_bio *split_bio = bio->bi_private;
1994 
1995 	if (atomic_dec_and_test(&split_bio->pending)) {
1996 		split_bio->bio->bi_phys_segments = 0;
1997 		split_bio->bio->bi_error = bio->bi_error;
1998 		bio_endio(split_bio->bio);
1999 		kfree(split_bio);
2000 	}
2001 	bio_put(bio);
2002 }
2003 
2004 static int blkif_recover(struct blkfront_info *info)
2005 {
2006 	unsigned int i, r_index;
2007 	struct request *req, *n;
2008 	int rc;
2009 	struct bio *bio, *cloned_bio;
2010 	unsigned int segs, offset;
2011 	int pending, size;
2012 	struct split_bio *split_bio;
2013 
2014 	blkfront_gather_backend_features(info);
2015 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2016 	blk_queue_max_segments(info->rq, segs);
2017 
2018 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2019 		struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2020 
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 (r_index = 0; r_index < info->nr_rings; r_index++) {
2031 		struct blkfront_ring_info *rinfo;
2032 
2033 		rinfo = &info->rinfo[r_index];
2034 		/* Kick any other new requests queued since we resumed */
2035 		kick_pending_request_queues(rinfo);
2036 	}
2037 
2038 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2039 		/* Requeue pending requests (flush or discard) */
2040 		list_del_init(&req->queuelist);
2041 		BUG_ON(req->nr_phys_segments > segs);
2042 		blk_mq_requeue_request(req);
2043 	}
2044 	blk_mq_kick_requeue_list(info->rq);
2045 
2046 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2047 		/* Traverse the list of pending bios and re-queue them */
2048 		if (bio_segments(bio) > segs) {
2049 			/*
2050 			 * This bio has more segments than what we can
2051 			 * handle, we have to split it.
2052 			 */
2053 			pending = (bio_segments(bio) + segs - 1) / segs;
2054 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2055 			BUG_ON(split_bio == NULL);
2056 			atomic_set(&split_bio->pending, pending);
2057 			split_bio->bio = bio;
2058 			for (i = 0; i < pending; i++) {
2059 				offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2060 				size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2061 					   (unsigned int)bio_sectors(bio) - offset);
2062 				cloned_bio = bio_clone(bio, GFP_NOIO);
2063 				BUG_ON(cloned_bio == NULL);
2064 				bio_trim(cloned_bio, offset, size);
2065 				cloned_bio->bi_private = split_bio;
2066 				cloned_bio->bi_end_io = split_bio_end;
2067 				submit_bio(cloned_bio->bi_rw, cloned_bio);
2068 			}
2069 			/*
2070 			 * Now we have to wait for all those smaller bios to
2071 			 * end, so we can also end the "parent" bio.
2072 			 */
2073 			continue;
2074 		}
2075 		/* We don't need to split this bio */
2076 		submit_bio(bio->bi_rw, bio);
2077 	}
2078 
2079 	return 0;
2080 }
2081 
2082 /**
2083  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2084  * driver restart.  We tear down our blkif structure and recreate it, but
2085  * leave the device-layer structures intact so that this is transparent to the
2086  * rest of the kernel.
2087  */
2088 static int blkfront_resume(struct xenbus_device *dev)
2089 {
2090 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2091 	int err = 0;
2092 	unsigned int i, j;
2093 
2094 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2095 
2096 	bio_list_init(&info->bio_list);
2097 	INIT_LIST_HEAD(&info->requests);
2098 	for (i = 0; i < info->nr_rings; i++) {
2099 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
2100 		struct bio_list merge_bio;
2101 		struct blk_shadow *shadow = rinfo->shadow;
2102 
2103 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2104 			/* Not in use? */
2105 			if (!shadow[j].request)
2106 				continue;
2107 
2108 			/*
2109 			 * Get the bios in the request so we can re-queue them.
2110 			 */
2111 			if (shadow[j].request->cmd_flags &
2112 					(REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
2113 				/*
2114 				 * Flush operations don't contain bios, so
2115 				 * we need to requeue the whole request
2116 				 */
2117 				list_add(&shadow[j].request->queuelist, &info->requests);
2118 				continue;
2119 			}
2120 			merge_bio.head = shadow[j].request->bio;
2121 			merge_bio.tail = shadow[j].request->biotail;
2122 			bio_list_merge(&info->bio_list, &merge_bio);
2123 			shadow[j].request->bio = NULL;
2124 			blk_mq_end_request(shadow[j].request, 0);
2125 		}
2126 	}
2127 
2128 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2129 
2130 	err = negotiate_mq(info);
2131 	if (err)
2132 		return err;
2133 
2134 	err = talk_to_blkback(dev, info);
2135 	if (!err)
2136 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2137 
2138 	/*
2139 	 * We have to wait for the backend to switch to
2140 	 * connected state, since we want to read which
2141 	 * features it supports.
2142 	 */
2143 
2144 	return err;
2145 }
2146 
2147 static void blkfront_closing(struct blkfront_info *info)
2148 {
2149 	struct xenbus_device *xbdev = info->xbdev;
2150 	struct block_device *bdev = NULL;
2151 
2152 	mutex_lock(&info->mutex);
2153 
2154 	if (xbdev->state == XenbusStateClosing) {
2155 		mutex_unlock(&info->mutex);
2156 		return;
2157 	}
2158 
2159 	if (info->gd)
2160 		bdev = bdget_disk(info->gd, 0);
2161 
2162 	mutex_unlock(&info->mutex);
2163 
2164 	if (!bdev) {
2165 		xenbus_frontend_closed(xbdev);
2166 		return;
2167 	}
2168 
2169 	mutex_lock(&bdev->bd_mutex);
2170 
2171 	if (bdev->bd_openers) {
2172 		xenbus_dev_error(xbdev, -EBUSY,
2173 				 "Device in use; refusing to close");
2174 		xenbus_switch_state(xbdev, XenbusStateClosing);
2175 	} else {
2176 		xlvbd_release_gendisk(info);
2177 		xenbus_frontend_closed(xbdev);
2178 	}
2179 
2180 	mutex_unlock(&bdev->bd_mutex);
2181 	bdput(bdev);
2182 }
2183 
2184 static void blkfront_setup_discard(struct blkfront_info *info)
2185 {
2186 	int err;
2187 	unsigned int discard_granularity;
2188 	unsigned int discard_alignment;
2189 	unsigned int discard_secure;
2190 
2191 	info->feature_discard = 1;
2192 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2193 		"discard-granularity", "%u", &discard_granularity,
2194 		"discard-alignment", "%u", &discard_alignment,
2195 		NULL);
2196 	if (!err) {
2197 		info->discard_granularity = discard_granularity;
2198 		info->discard_alignment = discard_alignment;
2199 	}
2200 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2201 		    "discard-secure", "%d", &discard_secure,
2202 		    NULL);
2203 	if (!err)
2204 		info->feature_secdiscard = !!discard_secure;
2205 }
2206 
2207 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2208 {
2209 	unsigned int psegs, grants;
2210 	int err, i;
2211 	struct blkfront_info *info = rinfo->dev_info;
2212 
2213 	if (info->max_indirect_segments == 0) {
2214 		if (!HAS_EXTRA_REQ)
2215 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2216 		else {
2217 			/*
2218 			 * When an extra req is required, the maximum
2219 			 * grants supported is related to the size of the
2220 			 * Linux block segment.
2221 			 */
2222 			grants = GRANTS_PER_PSEG;
2223 		}
2224 	}
2225 	else
2226 		grants = info->max_indirect_segments;
2227 	psegs = grants / GRANTS_PER_PSEG;
2228 
2229 	err = fill_grant_buffer(rinfo,
2230 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2231 	if (err)
2232 		goto out_of_memory;
2233 
2234 	if (!info->feature_persistent && info->max_indirect_segments) {
2235 		/*
2236 		 * We are using indirect descriptors but not persistent
2237 		 * grants, we need to allocate a set of pages that can be
2238 		 * used for mapping indirect grefs
2239 		 */
2240 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2241 
2242 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2243 		for (i = 0; i < num; i++) {
2244 			struct page *indirect_page = alloc_page(GFP_NOIO);
2245 			if (!indirect_page)
2246 				goto out_of_memory;
2247 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2248 		}
2249 	}
2250 
2251 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2252 		rinfo->shadow[i].grants_used = kzalloc(
2253 			sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2254 			GFP_NOIO);
2255 		rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2256 		if (info->max_indirect_segments)
2257 			rinfo->shadow[i].indirect_grants = kzalloc(
2258 				sizeof(rinfo->shadow[i].indirect_grants[0]) *
2259 				INDIRECT_GREFS(grants),
2260 				GFP_NOIO);
2261 		if ((rinfo->shadow[i].grants_used == NULL) ||
2262 			(rinfo->shadow[i].sg == NULL) ||
2263 		     (info->max_indirect_segments &&
2264 		     (rinfo->shadow[i].indirect_grants == NULL)))
2265 			goto out_of_memory;
2266 		sg_init_table(rinfo->shadow[i].sg, psegs);
2267 	}
2268 
2269 
2270 	return 0;
2271 
2272 out_of_memory:
2273 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2274 		kfree(rinfo->shadow[i].grants_used);
2275 		rinfo->shadow[i].grants_used = NULL;
2276 		kfree(rinfo->shadow[i].sg);
2277 		rinfo->shadow[i].sg = NULL;
2278 		kfree(rinfo->shadow[i].indirect_grants);
2279 		rinfo->shadow[i].indirect_grants = NULL;
2280 	}
2281 	if (!list_empty(&rinfo->indirect_pages)) {
2282 		struct page *indirect_page, *n;
2283 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2284 			list_del(&indirect_page->lru);
2285 			__free_page(indirect_page);
2286 		}
2287 	}
2288 	return -ENOMEM;
2289 }
2290 
2291 /*
2292  * Gather all backend feature-*
2293  */
2294 static void blkfront_gather_backend_features(struct blkfront_info *info)
2295 {
2296 	int err;
2297 	int barrier, flush, discard, persistent;
2298 	unsigned int indirect_segments;
2299 
2300 	info->feature_flush = 0;
2301 
2302 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2303 			"feature-barrier", "%d", &barrier,
2304 			NULL);
2305 
2306 	/*
2307 	 * If there's no "feature-barrier" defined, then it means
2308 	 * we're dealing with a very old backend which writes
2309 	 * synchronously; nothing to do.
2310 	 *
2311 	 * If there are barriers, then we use flush.
2312 	 */
2313 	if (!err && barrier)
2314 		info->feature_flush = REQ_FLUSH | REQ_FUA;
2315 	/*
2316 	 * And if there is "feature-flush-cache" use that above
2317 	 * barriers.
2318 	 */
2319 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2320 			"feature-flush-cache", "%d", &flush,
2321 			NULL);
2322 
2323 	if (!err && flush)
2324 		info->feature_flush = REQ_FLUSH;
2325 
2326 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2327 			"feature-discard", "%d", &discard,
2328 			NULL);
2329 
2330 	if (!err && discard)
2331 		blkfront_setup_discard(info);
2332 
2333 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2334 			"feature-persistent", "%u", &persistent,
2335 			NULL);
2336 	if (err)
2337 		info->feature_persistent = 0;
2338 	else
2339 		info->feature_persistent = persistent;
2340 
2341 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2342 			    "feature-max-indirect-segments", "%u", &indirect_segments,
2343 			    NULL);
2344 	if (err)
2345 		info->max_indirect_segments = 0;
2346 	else
2347 		info->max_indirect_segments = min(indirect_segments,
2348 						  xen_blkif_max_segments);
2349 }
2350 
2351 /*
2352  * Invoked when the backend is finally 'ready' (and has told produced
2353  * the details about the physical device - #sectors, size, etc).
2354  */
2355 static void blkfront_connect(struct blkfront_info *info)
2356 {
2357 	unsigned long long sectors;
2358 	unsigned long sector_size;
2359 	unsigned int physical_sector_size;
2360 	unsigned int binfo;
2361 	int err, i;
2362 
2363 	switch (info->connected) {
2364 	case BLKIF_STATE_CONNECTED:
2365 		/*
2366 		 * Potentially, the back-end may be signalling
2367 		 * a capacity change; update the capacity.
2368 		 */
2369 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2370 				   "sectors", "%Lu", &sectors);
2371 		if (XENBUS_EXIST_ERR(err))
2372 			return;
2373 		printk(KERN_INFO "Setting capacity to %Lu\n",
2374 		       sectors);
2375 		set_capacity(info->gd, sectors);
2376 		revalidate_disk(info->gd);
2377 
2378 		return;
2379 	case BLKIF_STATE_SUSPENDED:
2380 		/*
2381 		 * If we are recovering from suspension, we need to wait
2382 		 * for the backend to announce it's features before
2383 		 * reconnecting, at least we need to know if the backend
2384 		 * supports indirect descriptors, and how many.
2385 		 */
2386 		blkif_recover(info);
2387 		return;
2388 
2389 	default:
2390 		break;
2391 	}
2392 
2393 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2394 		__func__, info->xbdev->otherend);
2395 
2396 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2397 			    "sectors", "%llu", &sectors,
2398 			    "info", "%u", &binfo,
2399 			    "sector-size", "%lu", &sector_size,
2400 			    NULL);
2401 	if (err) {
2402 		xenbus_dev_fatal(info->xbdev, err,
2403 				 "reading backend fields at %s",
2404 				 info->xbdev->otherend);
2405 		return;
2406 	}
2407 
2408 	/*
2409 	 * physcial-sector-size is a newer field, so old backends may not
2410 	 * provide this. Assume physical sector size to be the same as
2411 	 * sector_size in that case.
2412 	 */
2413 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2414 			   "physical-sector-size", "%u", &physical_sector_size);
2415 	if (err != 1)
2416 		physical_sector_size = sector_size;
2417 
2418 	blkfront_gather_backend_features(info);
2419 	for (i = 0; i < info->nr_rings; i++) {
2420 		err = blkfront_setup_indirect(&info->rinfo[i]);
2421 		if (err) {
2422 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2423 					 info->xbdev->otherend);
2424 			blkif_free(info, 0);
2425 			break;
2426 		}
2427 	}
2428 
2429 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2430 				  physical_sector_size);
2431 	if (err) {
2432 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2433 				 info->xbdev->otherend);
2434 		return;
2435 	}
2436 
2437 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2438 
2439 	/* Kick pending requests. */
2440 	info->connected = BLKIF_STATE_CONNECTED;
2441 	for (i = 0; i < info->nr_rings; i++)
2442 		kick_pending_request_queues(&info->rinfo[i]);
2443 
2444 	add_disk(info->gd);
2445 
2446 	info->is_ready = 1;
2447 }
2448 
2449 /**
2450  * Callback received when the backend's state changes.
2451  */
2452 static void blkback_changed(struct xenbus_device *dev,
2453 			    enum xenbus_state backend_state)
2454 {
2455 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2456 
2457 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2458 
2459 	switch (backend_state) {
2460 	case XenbusStateInitWait:
2461 		if (dev->state != XenbusStateInitialising)
2462 			break;
2463 		if (talk_to_blkback(dev, info))
2464 			break;
2465 	case XenbusStateInitialising:
2466 	case XenbusStateInitialised:
2467 	case XenbusStateReconfiguring:
2468 	case XenbusStateReconfigured:
2469 	case XenbusStateUnknown:
2470 		break;
2471 
2472 	case XenbusStateConnected:
2473 		/*
2474 		 * talk_to_blkback sets state to XenbusStateInitialised
2475 		 * and blkfront_connect sets it to XenbusStateConnected
2476 		 * (if connection went OK).
2477 		 *
2478 		 * If the backend (or toolstack) decides to poke at backend
2479 		 * state (and re-trigger the watch by setting the state repeatedly
2480 		 * to XenbusStateConnected (4)) we need to deal with this.
2481 		 * This is allowed as this is used to communicate to the guest
2482 		 * that the size of disk has changed!
2483 		 */
2484 		if ((dev->state != XenbusStateInitialised) &&
2485 		    (dev->state != XenbusStateConnected)) {
2486 			if (talk_to_blkback(dev, info))
2487 				break;
2488 		}
2489 
2490 		blkfront_connect(info);
2491 		break;
2492 
2493 	case XenbusStateClosed:
2494 		if (dev->state == XenbusStateClosed)
2495 			break;
2496 		/* Missed the backend's Closing state -- fallthrough */
2497 	case XenbusStateClosing:
2498 		if (info)
2499 			blkfront_closing(info);
2500 		break;
2501 	}
2502 }
2503 
2504 static int blkfront_remove(struct xenbus_device *xbdev)
2505 {
2506 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2507 	struct block_device *bdev = NULL;
2508 	struct gendisk *disk;
2509 
2510 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2511 
2512 	blkif_free(info, 0);
2513 
2514 	mutex_lock(&info->mutex);
2515 
2516 	disk = info->gd;
2517 	if (disk)
2518 		bdev = bdget_disk(disk, 0);
2519 
2520 	info->xbdev = NULL;
2521 	mutex_unlock(&info->mutex);
2522 
2523 	if (!bdev) {
2524 		kfree(info);
2525 		return 0;
2526 	}
2527 
2528 	/*
2529 	 * The xbdev was removed before we reached the Closed
2530 	 * state. See if it's safe to remove the disk. If the bdev
2531 	 * isn't closed yet, we let release take care of it.
2532 	 */
2533 
2534 	mutex_lock(&bdev->bd_mutex);
2535 	info = disk->private_data;
2536 
2537 	dev_warn(disk_to_dev(disk),
2538 		 "%s was hot-unplugged, %d stale handles\n",
2539 		 xbdev->nodename, bdev->bd_openers);
2540 
2541 	if (info && !bdev->bd_openers) {
2542 		xlvbd_release_gendisk(info);
2543 		disk->private_data = NULL;
2544 		kfree(info);
2545 	}
2546 
2547 	mutex_unlock(&bdev->bd_mutex);
2548 	bdput(bdev);
2549 
2550 	return 0;
2551 }
2552 
2553 static int blkfront_is_ready(struct xenbus_device *dev)
2554 {
2555 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2556 
2557 	return info->is_ready && info->xbdev;
2558 }
2559 
2560 static int blkif_open(struct block_device *bdev, fmode_t mode)
2561 {
2562 	struct gendisk *disk = bdev->bd_disk;
2563 	struct blkfront_info *info;
2564 	int err = 0;
2565 
2566 	mutex_lock(&blkfront_mutex);
2567 
2568 	info = disk->private_data;
2569 	if (!info) {
2570 		/* xbdev gone */
2571 		err = -ERESTARTSYS;
2572 		goto out;
2573 	}
2574 
2575 	mutex_lock(&info->mutex);
2576 
2577 	if (!info->gd)
2578 		/* xbdev is closed */
2579 		err = -ERESTARTSYS;
2580 
2581 	mutex_unlock(&info->mutex);
2582 
2583 out:
2584 	mutex_unlock(&blkfront_mutex);
2585 	return err;
2586 }
2587 
2588 static void blkif_release(struct gendisk *disk, fmode_t mode)
2589 {
2590 	struct blkfront_info *info = disk->private_data;
2591 	struct block_device *bdev;
2592 	struct xenbus_device *xbdev;
2593 
2594 	mutex_lock(&blkfront_mutex);
2595 
2596 	bdev = bdget_disk(disk, 0);
2597 
2598 	if (!bdev) {
2599 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2600 		goto out_mutex;
2601 	}
2602 	if (bdev->bd_openers)
2603 		goto out;
2604 
2605 	/*
2606 	 * Check if we have been instructed to close. We will have
2607 	 * deferred this request, because the bdev was still open.
2608 	 */
2609 
2610 	mutex_lock(&info->mutex);
2611 	xbdev = info->xbdev;
2612 
2613 	if (xbdev && xbdev->state == XenbusStateClosing) {
2614 		/* pending switch to state closed */
2615 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2616 		xlvbd_release_gendisk(info);
2617 		xenbus_frontend_closed(info->xbdev);
2618  	}
2619 
2620 	mutex_unlock(&info->mutex);
2621 
2622 	if (!xbdev) {
2623 		/* sudden device removal */
2624 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2625 		xlvbd_release_gendisk(info);
2626 		disk->private_data = NULL;
2627 		kfree(info);
2628 	}
2629 
2630 out:
2631 	bdput(bdev);
2632 out_mutex:
2633 	mutex_unlock(&blkfront_mutex);
2634 }
2635 
2636 static const struct block_device_operations xlvbd_block_fops =
2637 {
2638 	.owner = THIS_MODULE,
2639 	.open = blkif_open,
2640 	.release = blkif_release,
2641 	.getgeo = blkif_getgeo,
2642 	.ioctl = blkif_ioctl,
2643 };
2644 
2645 
2646 static const struct xenbus_device_id blkfront_ids[] = {
2647 	{ "vbd" },
2648 	{ "" }
2649 };
2650 
2651 static struct xenbus_driver blkfront_driver = {
2652 	.ids  = blkfront_ids,
2653 	.probe = blkfront_probe,
2654 	.remove = blkfront_remove,
2655 	.resume = blkfront_resume,
2656 	.otherend_changed = blkback_changed,
2657 	.is_ready = blkfront_is_ready,
2658 };
2659 
2660 static int __init xlblk_init(void)
2661 {
2662 	int ret;
2663 	int nr_cpus = num_online_cpus();
2664 
2665 	if (!xen_domain())
2666 		return -ENODEV;
2667 
2668 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2669 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2670 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2671 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2672 	}
2673 
2674 	if (xen_blkif_max_queues > nr_cpus) {
2675 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2676 			xen_blkif_max_queues, nr_cpus);
2677 		xen_blkif_max_queues = nr_cpus;
2678 	}
2679 
2680 	if (!xen_has_pv_disk_devices())
2681 		return -ENODEV;
2682 
2683 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2684 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2685 		       XENVBD_MAJOR, DEV_NAME);
2686 		return -ENODEV;
2687 	}
2688 
2689 	ret = xenbus_register_frontend(&blkfront_driver);
2690 	if (ret) {
2691 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2692 		return ret;
2693 	}
2694 
2695 	return 0;
2696 }
2697 module_init(xlblk_init);
2698 
2699 
2700 static void __exit xlblk_exit(void)
2701 {
2702 	xenbus_unregister_driver(&blkfront_driver);
2703 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2704 	kfree(minors);
2705 }
2706 module_exit(xlblk_exit);
2707 
2708 MODULE_DESCRIPTION("Xen virtual block device frontend");
2709 MODULE_LICENSE("GPL");
2710 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2711 MODULE_ALIAS("xen:vbd");
2712 MODULE_ALIAS("xenblk");
2713