xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 8bd1369b)
1 /*
2  * blkfront.c
3  *
4  * XenLinux virtual block device driver.
5  *
6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8  * Copyright (c) 2004, Christian Limpach
9  * Copyright (c) 2004, Andrew Warfield
10  * Copyright (c) 2005, Christopher Clark
11  * Copyright (c) 2005, XenSource Ltd
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License version 2
15  * as published by the Free Software Foundation; or, when distributed
16  * separately from the Linux kernel or incorporated into other
17  * software packages, subject to the following license:
18  *
19  * Permission is hereby granted, free of charge, to any person obtaining a copy
20  * of this source file (the "Software"), to deal in the Software without
21  * restriction, including without limitation the rights to use, copy, modify,
22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23  * and to permit persons to whom the Software is furnished to do so, subject to
24  * the following conditions:
25  *
26  * The above copyright notice and this permission notice shall be included in
27  * all copies or substantial portions of the Software.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35  * IN THE SOFTWARE.
36  */
37 
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 
50 #include <xen/xen.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56 
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
60 
61 #include <asm/xen/hypervisor.h>
62 
63 /*
64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
65  * When Linux is using a different page size than Xen, it may not be possible
66  * to put all the data in a single segment.
67  * This can happen when the backend doesn't support indirect descriptor and
68  * therefore the maximum amount of data that a request can carry is
69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70  *
71  * Note that we only support one extra request. So the Linux page size
72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73  * 88KB.
74  */
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76 
77 enum blkif_state {
78 	BLKIF_STATE_DISCONNECTED,
79 	BLKIF_STATE_CONNECTED,
80 	BLKIF_STATE_SUSPENDED,
81 };
82 
83 struct grant {
84 	grant_ref_t gref;
85 	struct page *page;
86 	struct list_head node;
87 };
88 
89 enum blk_req_status {
90 	REQ_WAITING,
91 	REQ_DONE,
92 	REQ_ERROR,
93 	REQ_EOPNOTSUPP,
94 };
95 
96 struct blk_shadow {
97 	struct blkif_request req;
98 	struct request *request;
99 	struct grant **grants_used;
100 	struct grant **indirect_grants;
101 	struct scatterlist *sg;
102 	unsigned int num_sg;
103 	enum blk_req_status status;
104 
105 	#define NO_ASSOCIATED_ID ~0UL
106 	/*
107 	 * Id of the sibling if we ever need 2 requests when handling a
108 	 * block I/O request
109 	 */
110 	unsigned long associated_id;
111 };
112 
113 struct blkif_req {
114 	blk_status_t	error;
115 };
116 
117 static inline struct blkif_req *blkif_req(struct request *rq)
118 {
119 	return blk_mq_rq_to_pdu(rq);
120 }
121 
122 static DEFINE_MUTEX(blkfront_mutex);
123 static const struct block_device_operations xlvbd_block_fops;
124 
125 /*
126  * Maximum number of segments in indirect requests, the actual value used by
127  * the frontend driver is the minimum of this value and the value provided
128  * by the backend driver.
129  */
130 
131 static unsigned int xen_blkif_max_segments = 32;
132 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
133 MODULE_PARM_DESC(max_indirect_segments,
134 		 "Maximum amount of segments in indirect requests (default is 32)");
135 
136 static unsigned int xen_blkif_max_queues = 4;
137 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
138 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
139 
140 /*
141  * Maximum order of pages to be used for the shared ring between front and
142  * backend, 4KB page granularity is used.
143  */
144 static unsigned int xen_blkif_max_ring_order;
145 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
146 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
147 
148 #define BLK_RING_SIZE(info)	\
149 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
150 
151 #define BLK_MAX_RING_SIZE	\
152 	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
153 
154 /*
155  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
156  * characters are enough. Define to 20 to keep consistent with backend.
157  */
158 #define RINGREF_NAME_LEN (20)
159 /*
160  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
161  */
162 #define QUEUE_NAME_LEN (17)
163 
164 /*
165  *  Per-ring info.
166  *  Every blkfront device can associate with one or more blkfront_ring_info,
167  *  depending on how many hardware queues/rings to be used.
168  */
169 struct blkfront_ring_info {
170 	/* Lock to protect data in every ring buffer. */
171 	spinlock_t ring_lock;
172 	struct blkif_front_ring ring;
173 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
174 	unsigned int evtchn, irq;
175 	struct work_struct work;
176 	struct gnttab_free_callback callback;
177 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
178 	struct list_head indirect_pages;
179 	struct list_head grants;
180 	unsigned int persistent_gnts_c;
181 	unsigned long shadow_free;
182 	struct blkfront_info *dev_info;
183 };
184 
185 /*
186  * We have one of these per vbd, whether ide, scsi or 'other'.  They
187  * hang in private_data off the gendisk structure. We may end up
188  * putting all kinds of interesting stuff here :-)
189  */
190 struct blkfront_info
191 {
192 	struct mutex mutex;
193 	struct xenbus_device *xbdev;
194 	struct gendisk *gd;
195 	u16 sector_size;
196 	unsigned int physical_sector_size;
197 	int vdevice;
198 	blkif_vdev_t handle;
199 	enum blkif_state connected;
200 	/* Number of pages per ring buffer. */
201 	unsigned int nr_ring_pages;
202 	struct request_queue *rq;
203 	unsigned int feature_flush:1;
204 	unsigned int feature_fua:1;
205 	unsigned int feature_discard:1;
206 	unsigned int feature_secdiscard:1;
207 	unsigned int feature_persistent:1;
208 	unsigned int discard_granularity;
209 	unsigned int discard_alignment;
210 	/* Number of 4KB segments handled */
211 	unsigned int max_indirect_segments;
212 	int is_ready;
213 	struct blk_mq_tag_set tag_set;
214 	struct blkfront_ring_info *rinfo;
215 	unsigned int nr_rings;
216 	/* Save uncomplete reqs and bios for migration. */
217 	struct list_head requests;
218 	struct bio_list bio_list;
219 };
220 
221 static unsigned int nr_minors;
222 static unsigned long *minors;
223 static DEFINE_SPINLOCK(minor_lock);
224 
225 #define GRANT_INVALID_REF	0
226 
227 #define PARTS_PER_DISK		16
228 #define PARTS_PER_EXT_DISK      256
229 
230 #define BLKIF_MAJOR(dev) ((dev)>>8)
231 #define BLKIF_MINOR(dev) ((dev) & 0xff)
232 
233 #define EXT_SHIFT 28
234 #define EXTENDED (1<<EXT_SHIFT)
235 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
236 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
237 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
238 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
239 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
240 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
241 
242 #define DEV_NAME	"xvd"	/* name in /dev */
243 
244 /*
245  * Grants are always the same size as a Xen page (i.e 4KB).
246  * A physical segment is always the same size as a Linux page.
247  * Number of grants per physical segment
248  */
249 #define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
250 
251 #define GRANTS_PER_INDIRECT_FRAME \
252 	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
253 
254 #define PSEGS_PER_INDIRECT_FRAME	\
255 	(GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
256 
257 #define INDIRECT_GREFS(_grants)		\
258 	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
259 
260 #define GREFS(_psegs)	((_psegs) * GRANTS_PER_PSEG)
261 
262 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
263 static void blkfront_gather_backend_features(struct blkfront_info *info);
264 static int negotiate_mq(struct blkfront_info *info);
265 
266 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
267 {
268 	unsigned long free = rinfo->shadow_free;
269 
270 	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
271 	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
272 	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
273 	return free;
274 }
275 
276 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
277 			      unsigned long id)
278 {
279 	if (rinfo->shadow[id].req.u.rw.id != id)
280 		return -EINVAL;
281 	if (rinfo->shadow[id].request == NULL)
282 		return -EINVAL;
283 	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
284 	rinfo->shadow[id].request = NULL;
285 	rinfo->shadow_free = id;
286 	return 0;
287 }
288 
289 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
290 {
291 	struct blkfront_info *info = rinfo->dev_info;
292 	struct page *granted_page;
293 	struct grant *gnt_list_entry, *n;
294 	int i = 0;
295 
296 	while (i < num) {
297 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
298 		if (!gnt_list_entry)
299 			goto out_of_memory;
300 
301 		if (info->feature_persistent) {
302 			granted_page = alloc_page(GFP_NOIO);
303 			if (!granted_page) {
304 				kfree(gnt_list_entry);
305 				goto out_of_memory;
306 			}
307 			gnt_list_entry->page = granted_page;
308 		}
309 
310 		gnt_list_entry->gref = GRANT_INVALID_REF;
311 		list_add(&gnt_list_entry->node, &rinfo->grants);
312 		i++;
313 	}
314 
315 	return 0;
316 
317 out_of_memory:
318 	list_for_each_entry_safe(gnt_list_entry, n,
319 	                         &rinfo->grants, node) {
320 		list_del(&gnt_list_entry->node);
321 		if (info->feature_persistent)
322 			__free_page(gnt_list_entry->page);
323 		kfree(gnt_list_entry);
324 		i--;
325 	}
326 	BUG_ON(i != 0);
327 	return -ENOMEM;
328 }
329 
330 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
331 {
332 	struct grant *gnt_list_entry;
333 
334 	BUG_ON(list_empty(&rinfo->grants));
335 	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
336 					  node);
337 	list_del(&gnt_list_entry->node);
338 
339 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
340 		rinfo->persistent_gnts_c--;
341 
342 	return gnt_list_entry;
343 }
344 
345 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
346 					const struct blkfront_info *info)
347 {
348 	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
349 						 info->xbdev->otherend_id,
350 						 gnt_list_entry->page,
351 						 0);
352 }
353 
354 static struct grant *get_grant(grant_ref_t *gref_head,
355 			       unsigned long gfn,
356 			       struct blkfront_ring_info *rinfo)
357 {
358 	struct grant *gnt_list_entry = get_free_grant(rinfo);
359 	struct blkfront_info *info = rinfo->dev_info;
360 
361 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
362 		return gnt_list_entry;
363 
364 	/* Assign a gref to this page */
365 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
366 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
367 	if (info->feature_persistent)
368 		grant_foreign_access(gnt_list_entry, info);
369 	else {
370 		/* Grant access to the GFN passed by the caller */
371 		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
372 						info->xbdev->otherend_id,
373 						gfn, 0);
374 	}
375 
376 	return gnt_list_entry;
377 }
378 
379 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
380 					struct blkfront_ring_info *rinfo)
381 {
382 	struct grant *gnt_list_entry = get_free_grant(rinfo);
383 	struct blkfront_info *info = rinfo->dev_info;
384 
385 	if (gnt_list_entry->gref != GRANT_INVALID_REF)
386 		return gnt_list_entry;
387 
388 	/* Assign a gref to this page */
389 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
390 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
391 	if (!info->feature_persistent) {
392 		struct page *indirect_page;
393 
394 		/* Fetch a pre-allocated page to use for indirect grefs */
395 		BUG_ON(list_empty(&rinfo->indirect_pages));
396 		indirect_page = list_first_entry(&rinfo->indirect_pages,
397 						 struct page, lru);
398 		list_del(&indirect_page->lru);
399 		gnt_list_entry->page = indirect_page;
400 	}
401 	grant_foreign_access(gnt_list_entry, info);
402 
403 	return gnt_list_entry;
404 }
405 
406 static const char *op_name(int op)
407 {
408 	static const char *const names[] = {
409 		[BLKIF_OP_READ] = "read",
410 		[BLKIF_OP_WRITE] = "write",
411 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
412 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
413 		[BLKIF_OP_DISCARD] = "discard" };
414 
415 	if (op < 0 || op >= ARRAY_SIZE(names))
416 		return "unknown";
417 
418 	if (!names[op])
419 		return "reserved";
420 
421 	return names[op];
422 }
423 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
424 {
425 	unsigned int end = minor + nr;
426 	int rc;
427 
428 	if (end > nr_minors) {
429 		unsigned long *bitmap, *old;
430 
431 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
432 				 GFP_KERNEL);
433 		if (bitmap == NULL)
434 			return -ENOMEM;
435 
436 		spin_lock(&minor_lock);
437 		if (end > nr_minors) {
438 			old = minors;
439 			memcpy(bitmap, minors,
440 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
441 			minors = bitmap;
442 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
443 		} else
444 			old = bitmap;
445 		spin_unlock(&minor_lock);
446 		kfree(old);
447 	}
448 
449 	spin_lock(&minor_lock);
450 	if (find_next_bit(minors, end, minor) >= end) {
451 		bitmap_set(minors, minor, nr);
452 		rc = 0;
453 	} else
454 		rc = -EBUSY;
455 	spin_unlock(&minor_lock);
456 
457 	return rc;
458 }
459 
460 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
461 {
462 	unsigned int end = minor + nr;
463 
464 	BUG_ON(end > nr_minors);
465 	spin_lock(&minor_lock);
466 	bitmap_clear(minors,  minor, nr);
467 	spin_unlock(&minor_lock);
468 }
469 
470 static void blkif_restart_queue_callback(void *arg)
471 {
472 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
473 	schedule_work(&rinfo->work);
474 }
475 
476 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
477 {
478 	/* We don't have real geometry info, but let's at least return
479 	   values consistent with the size of the device */
480 	sector_t nsect = get_capacity(bd->bd_disk);
481 	sector_t cylinders = nsect;
482 
483 	hg->heads = 0xff;
484 	hg->sectors = 0x3f;
485 	sector_div(cylinders, hg->heads * hg->sectors);
486 	hg->cylinders = cylinders;
487 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
488 		hg->cylinders = 0xffff;
489 	return 0;
490 }
491 
492 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
493 		       unsigned command, unsigned long argument)
494 {
495 	struct blkfront_info *info = bdev->bd_disk->private_data;
496 	int i;
497 
498 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
499 		command, (long)argument);
500 
501 	switch (command) {
502 	case CDROMMULTISESSION:
503 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
504 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
505 			if (put_user(0, (char __user *)(argument + i)))
506 				return -EFAULT;
507 		return 0;
508 
509 	case CDROM_GET_CAPABILITY: {
510 		struct gendisk *gd = info->gd;
511 		if (gd->flags & GENHD_FL_CD)
512 			return 0;
513 		return -EINVAL;
514 	}
515 
516 	default:
517 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
518 		  command);*/
519 		return -EINVAL; /* same return as native Linux */
520 	}
521 
522 	return 0;
523 }
524 
525 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
526 					    struct request *req,
527 					    struct blkif_request **ring_req)
528 {
529 	unsigned long id;
530 
531 	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
532 	rinfo->ring.req_prod_pvt++;
533 
534 	id = get_id_from_freelist(rinfo);
535 	rinfo->shadow[id].request = req;
536 	rinfo->shadow[id].status = REQ_WAITING;
537 	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
538 
539 	(*ring_req)->u.rw.id = id;
540 
541 	return id;
542 }
543 
544 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
545 {
546 	struct blkfront_info *info = rinfo->dev_info;
547 	struct blkif_request *ring_req;
548 	unsigned long id;
549 
550 	/* Fill out a communications ring structure. */
551 	id = blkif_ring_get_request(rinfo, req, &ring_req);
552 
553 	ring_req->operation = BLKIF_OP_DISCARD;
554 	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
555 	ring_req->u.discard.id = id;
556 	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
557 	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
558 		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
559 	else
560 		ring_req->u.discard.flag = 0;
561 
562 	/* Keep a private copy so we can reissue requests when recovering. */
563 	rinfo->shadow[id].req = *ring_req;
564 
565 	return 0;
566 }
567 
568 struct setup_rw_req {
569 	unsigned int grant_idx;
570 	struct blkif_request_segment *segments;
571 	struct blkfront_ring_info *rinfo;
572 	struct blkif_request *ring_req;
573 	grant_ref_t gref_head;
574 	unsigned int id;
575 	/* Only used when persistent grant is used and it's a read request */
576 	bool need_copy;
577 	unsigned int bvec_off;
578 	char *bvec_data;
579 
580 	bool require_extra_req;
581 	struct blkif_request *extra_ring_req;
582 };
583 
584 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
585 				     unsigned int len, void *data)
586 {
587 	struct setup_rw_req *setup = data;
588 	int n, ref;
589 	struct grant *gnt_list_entry;
590 	unsigned int fsect, lsect;
591 	/* Convenient aliases */
592 	unsigned int grant_idx = setup->grant_idx;
593 	struct blkif_request *ring_req = setup->ring_req;
594 	struct blkfront_ring_info *rinfo = setup->rinfo;
595 	/*
596 	 * We always use the shadow of the first request to store the list
597 	 * of grant associated to the block I/O request. This made the
598 	 * completion more easy to handle even if the block I/O request is
599 	 * split.
600 	 */
601 	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
602 
603 	if (unlikely(setup->require_extra_req &&
604 		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
605 		/*
606 		 * We are using the second request, setup grant_idx
607 		 * to be the index of the segment array.
608 		 */
609 		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
610 		ring_req = setup->extra_ring_req;
611 	}
612 
613 	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
614 	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
615 		if (setup->segments)
616 			kunmap_atomic(setup->segments);
617 
618 		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
619 		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
620 		shadow->indirect_grants[n] = gnt_list_entry;
621 		setup->segments = kmap_atomic(gnt_list_entry->page);
622 		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
623 	}
624 
625 	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
626 	ref = gnt_list_entry->gref;
627 	/*
628 	 * All the grants are stored in the shadow of the first
629 	 * request. Therefore we have to use the global index.
630 	 */
631 	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
632 
633 	if (setup->need_copy) {
634 		void *shared_data;
635 
636 		shared_data = kmap_atomic(gnt_list_entry->page);
637 		/*
638 		 * this does not wipe data stored outside the
639 		 * range sg->offset..sg->offset+sg->length.
640 		 * Therefore, blkback *could* see data from
641 		 * previous requests. This is OK as long as
642 		 * persistent grants are shared with just one
643 		 * domain. It may need refactoring if this
644 		 * changes
645 		 */
646 		memcpy(shared_data + offset,
647 		       setup->bvec_data + setup->bvec_off,
648 		       len);
649 
650 		kunmap_atomic(shared_data);
651 		setup->bvec_off += len;
652 	}
653 
654 	fsect = offset >> 9;
655 	lsect = fsect + (len >> 9) - 1;
656 	if (ring_req->operation != BLKIF_OP_INDIRECT) {
657 		ring_req->u.rw.seg[grant_idx] =
658 			(struct blkif_request_segment) {
659 				.gref       = ref,
660 				.first_sect = fsect,
661 				.last_sect  = lsect };
662 	} else {
663 		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
664 			(struct blkif_request_segment) {
665 				.gref       = ref,
666 				.first_sect = fsect,
667 				.last_sect  = lsect };
668 	}
669 
670 	(setup->grant_idx)++;
671 }
672 
673 static void blkif_setup_extra_req(struct blkif_request *first,
674 				  struct blkif_request *second)
675 {
676 	uint16_t nr_segments = first->u.rw.nr_segments;
677 
678 	/*
679 	 * The second request is only present when the first request uses
680 	 * all its segments. It's always the continuity of the first one.
681 	 */
682 	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
683 
684 	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
685 	second->u.rw.sector_number = first->u.rw.sector_number +
686 		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
687 
688 	second->u.rw.handle = first->u.rw.handle;
689 	second->operation = first->operation;
690 }
691 
692 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
693 {
694 	struct blkfront_info *info = rinfo->dev_info;
695 	struct blkif_request *ring_req, *extra_ring_req = NULL;
696 	unsigned long id, extra_id = NO_ASSOCIATED_ID;
697 	bool require_extra_req = false;
698 	int i;
699 	struct setup_rw_req setup = {
700 		.grant_idx = 0,
701 		.segments = NULL,
702 		.rinfo = rinfo,
703 		.need_copy = rq_data_dir(req) && info->feature_persistent,
704 	};
705 
706 	/*
707 	 * Used to store if we are able to queue the request by just using
708 	 * existing persistent grants, or if we have to get new grants,
709 	 * as there are not sufficiently many free.
710 	 */
711 	bool new_persistent_gnts = false;
712 	struct scatterlist *sg;
713 	int num_sg, max_grefs, num_grant;
714 
715 	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
716 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
717 		/*
718 		 * If we are using indirect segments we need to account
719 		 * for the indirect grefs used in the request.
720 		 */
721 		max_grefs += INDIRECT_GREFS(max_grefs);
722 
723 	/* Check if we have enough persistent grants to allocate a requests */
724 	if (rinfo->persistent_gnts_c < max_grefs) {
725 		new_persistent_gnts = true;
726 
727 		if (gnttab_alloc_grant_references(
728 		    max_grefs - rinfo->persistent_gnts_c,
729 		    &setup.gref_head) < 0) {
730 			gnttab_request_free_callback(
731 				&rinfo->callback,
732 				blkif_restart_queue_callback,
733 				rinfo,
734 				max_grefs - rinfo->persistent_gnts_c);
735 			return 1;
736 		}
737 	}
738 
739 	/* Fill out a communications ring structure. */
740 	id = blkif_ring_get_request(rinfo, req, &ring_req);
741 
742 	num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
743 	num_grant = 0;
744 	/* Calculate the number of grant used */
745 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
746 	       num_grant += gnttab_count_grant(sg->offset, sg->length);
747 
748 	require_extra_req = info->max_indirect_segments == 0 &&
749 		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
750 	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
751 
752 	rinfo->shadow[id].num_sg = num_sg;
753 	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
754 	    likely(!require_extra_req)) {
755 		/*
756 		 * The indirect operation can only be a BLKIF_OP_READ or
757 		 * BLKIF_OP_WRITE
758 		 */
759 		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
760 		ring_req->operation = BLKIF_OP_INDIRECT;
761 		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
762 			BLKIF_OP_WRITE : BLKIF_OP_READ;
763 		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
764 		ring_req->u.indirect.handle = info->handle;
765 		ring_req->u.indirect.nr_segments = num_grant;
766 	} else {
767 		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
768 		ring_req->u.rw.handle = info->handle;
769 		ring_req->operation = rq_data_dir(req) ?
770 			BLKIF_OP_WRITE : BLKIF_OP_READ;
771 		if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
772 			/*
773 			 * Ideally we can do an unordered flush-to-disk.
774 			 * In case the backend onlysupports barriers, use that.
775 			 * A barrier request a superset of FUA, so we can
776 			 * implement it the same way.  (It's also a FLUSH+FUA,
777 			 * since it is guaranteed ordered WRT previous writes.)
778 			 */
779 			if (info->feature_flush && info->feature_fua)
780 				ring_req->operation =
781 					BLKIF_OP_WRITE_BARRIER;
782 			else if (info->feature_flush)
783 				ring_req->operation =
784 					BLKIF_OP_FLUSH_DISKCACHE;
785 			else
786 				ring_req->operation = 0;
787 		}
788 		ring_req->u.rw.nr_segments = num_grant;
789 		if (unlikely(require_extra_req)) {
790 			extra_id = blkif_ring_get_request(rinfo, req,
791 							  &extra_ring_req);
792 			/*
793 			 * Only the first request contains the scatter-gather
794 			 * list.
795 			 */
796 			rinfo->shadow[extra_id].num_sg = 0;
797 
798 			blkif_setup_extra_req(ring_req, extra_ring_req);
799 
800 			/* Link the 2 requests together */
801 			rinfo->shadow[extra_id].associated_id = id;
802 			rinfo->shadow[id].associated_id = extra_id;
803 		}
804 	}
805 
806 	setup.ring_req = ring_req;
807 	setup.id = id;
808 
809 	setup.require_extra_req = require_extra_req;
810 	if (unlikely(require_extra_req))
811 		setup.extra_ring_req = extra_ring_req;
812 
813 	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
814 		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
815 
816 		if (setup.need_copy) {
817 			setup.bvec_off = sg->offset;
818 			setup.bvec_data = kmap_atomic(sg_page(sg));
819 		}
820 
821 		gnttab_foreach_grant_in_range(sg_page(sg),
822 					      sg->offset,
823 					      sg->length,
824 					      blkif_setup_rw_req_grant,
825 					      &setup);
826 
827 		if (setup.need_copy)
828 			kunmap_atomic(setup.bvec_data);
829 	}
830 	if (setup.segments)
831 		kunmap_atomic(setup.segments);
832 
833 	/* Keep a private copy so we can reissue requests when recovering. */
834 	rinfo->shadow[id].req = *ring_req;
835 	if (unlikely(require_extra_req))
836 		rinfo->shadow[extra_id].req = *extra_ring_req;
837 
838 	if (new_persistent_gnts)
839 		gnttab_free_grant_references(setup.gref_head);
840 
841 	return 0;
842 }
843 
844 /*
845  * Generate a Xen blkfront IO request from a blk layer request.  Reads
846  * and writes are handled as expected.
847  *
848  * @req: a request struct
849  */
850 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
851 {
852 	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
853 		return 1;
854 
855 	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
856 		     req_op(req) == REQ_OP_SECURE_ERASE))
857 		return blkif_queue_discard_req(req, rinfo);
858 	else
859 		return blkif_queue_rw_req(req, rinfo);
860 }
861 
862 static inline void flush_requests(struct blkfront_ring_info *rinfo)
863 {
864 	int notify;
865 
866 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
867 
868 	if (notify)
869 		notify_remote_via_irq(rinfo->irq);
870 }
871 
872 static inline bool blkif_request_flush_invalid(struct request *req,
873 					       struct blkfront_info *info)
874 {
875 	return (blk_rq_is_passthrough(req) ||
876 		((req_op(req) == REQ_OP_FLUSH) &&
877 		 !info->feature_flush) ||
878 		((req->cmd_flags & REQ_FUA) &&
879 		 !info->feature_fua));
880 }
881 
882 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
883 			  const struct blk_mq_queue_data *qd)
884 {
885 	unsigned long flags;
886 	int qid = hctx->queue_num;
887 	struct blkfront_info *info = hctx->queue->queuedata;
888 	struct blkfront_ring_info *rinfo = NULL;
889 
890 	BUG_ON(info->nr_rings <= qid);
891 	rinfo = &info->rinfo[qid];
892 	blk_mq_start_request(qd->rq);
893 	spin_lock_irqsave(&rinfo->ring_lock, flags);
894 	if (RING_FULL(&rinfo->ring))
895 		goto out_busy;
896 
897 	if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
898 		goto out_err;
899 
900 	if (blkif_queue_request(qd->rq, rinfo))
901 		goto out_busy;
902 
903 	flush_requests(rinfo);
904 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
905 	return BLK_STS_OK;
906 
907 out_err:
908 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
909 	return BLK_STS_IOERR;
910 
911 out_busy:
912 	blk_mq_stop_hw_queue(hctx);
913 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
914 	return BLK_STS_DEV_RESOURCE;
915 }
916 
917 static void blkif_complete_rq(struct request *rq)
918 {
919 	blk_mq_end_request(rq, blkif_req(rq)->error);
920 }
921 
922 static const struct blk_mq_ops blkfront_mq_ops = {
923 	.queue_rq = blkif_queue_rq,
924 	.complete = blkif_complete_rq,
925 };
926 
927 static void blkif_set_queue_limits(struct blkfront_info *info)
928 {
929 	struct request_queue *rq = info->rq;
930 	struct gendisk *gd = info->gd;
931 	unsigned int segments = info->max_indirect_segments ? :
932 				BLKIF_MAX_SEGMENTS_PER_REQUEST;
933 
934 	blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
935 
936 	if (info->feature_discard) {
937 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
938 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
939 		rq->limits.discard_granularity = info->discard_granularity;
940 		rq->limits.discard_alignment = info->discard_alignment;
941 		if (info->feature_secdiscard)
942 			blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
943 	}
944 
945 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
946 	blk_queue_logical_block_size(rq, info->sector_size);
947 	blk_queue_physical_block_size(rq, info->physical_sector_size);
948 	blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
949 
950 	/* Each segment in a request is up to an aligned page in size. */
951 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
952 	blk_queue_max_segment_size(rq, PAGE_SIZE);
953 
954 	/* Ensure a merged request will fit in a single I/O ring slot. */
955 	blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
956 
957 	/* Make sure buffer addresses are sector-aligned. */
958 	blk_queue_dma_alignment(rq, 511);
959 }
960 
961 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
962 				unsigned int physical_sector_size)
963 {
964 	struct request_queue *rq;
965 	struct blkfront_info *info = gd->private_data;
966 
967 	memset(&info->tag_set, 0, sizeof(info->tag_set));
968 	info->tag_set.ops = &blkfront_mq_ops;
969 	info->tag_set.nr_hw_queues = info->nr_rings;
970 	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
971 		/*
972 		 * When indirect descriptior is not supported, the I/O request
973 		 * will be split between multiple request in the ring.
974 		 * To avoid problems when sending the request, divide by
975 		 * 2 the depth of the queue.
976 		 */
977 		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
978 	} else
979 		info->tag_set.queue_depth = BLK_RING_SIZE(info);
980 	info->tag_set.numa_node = NUMA_NO_NODE;
981 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
982 	info->tag_set.cmd_size = sizeof(struct blkif_req);
983 	info->tag_set.driver_data = info;
984 
985 	if (blk_mq_alloc_tag_set(&info->tag_set))
986 		return -EINVAL;
987 	rq = blk_mq_init_queue(&info->tag_set);
988 	if (IS_ERR(rq)) {
989 		blk_mq_free_tag_set(&info->tag_set);
990 		return PTR_ERR(rq);
991 	}
992 
993 	rq->queuedata = info;
994 	info->rq = gd->queue = rq;
995 	info->gd = gd;
996 	info->sector_size = sector_size;
997 	info->physical_sector_size = physical_sector_size;
998 	blkif_set_queue_limits(info);
999 
1000 	return 0;
1001 }
1002 
1003 static const char *flush_info(struct blkfront_info *info)
1004 {
1005 	if (info->feature_flush && info->feature_fua)
1006 		return "barrier: enabled;";
1007 	else if (info->feature_flush)
1008 		return "flush diskcache: enabled;";
1009 	else
1010 		return "barrier or flush: disabled;";
1011 }
1012 
1013 static void xlvbd_flush(struct blkfront_info *info)
1014 {
1015 	blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1016 			      info->feature_fua ? true : false);
1017 	pr_info("blkfront: %s: %s %s %s %s %s\n",
1018 		info->gd->disk_name, flush_info(info),
1019 		"persistent grants:", info->feature_persistent ?
1020 		"enabled;" : "disabled;", "indirect descriptors:",
1021 		info->max_indirect_segments ? "enabled;" : "disabled;");
1022 }
1023 
1024 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1025 {
1026 	int major;
1027 	major = BLKIF_MAJOR(vdevice);
1028 	*minor = BLKIF_MINOR(vdevice);
1029 	switch (major) {
1030 		case XEN_IDE0_MAJOR:
1031 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1032 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1033 				EMULATED_HD_DISK_MINOR_OFFSET;
1034 			break;
1035 		case XEN_IDE1_MAJOR:
1036 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1037 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1038 				EMULATED_HD_DISK_MINOR_OFFSET;
1039 			break;
1040 		case XEN_SCSI_DISK0_MAJOR:
1041 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1042 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1043 			break;
1044 		case XEN_SCSI_DISK1_MAJOR:
1045 		case XEN_SCSI_DISK2_MAJOR:
1046 		case XEN_SCSI_DISK3_MAJOR:
1047 		case XEN_SCSI_DISK4_MAJOR:
1048 		case XEN_SCSI_DISK5_MAJOR:
1049 		case XEN_SCSI_DISK6_MAJOR:
1050 		case XEN_SCSI_DISK7_MAJOR:
1051 			*offset = (*minor / PARTS_PER_DISK) +
1052 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1053 				EMULATED_SD_DISK_NAME_OFFSET;
1054 			*minor = *minor +
1055 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1056 				EMULATED_SD_DISK_MINOR_OFFSET;
1057 			break;
1058 		case XEN_SCSI_DISK8_MAJOR:
1059 		case XEN_SCSI_DISK9_MAJOR:
1060 		case XEN_SCSI_DISK10_MAJOR:
1061 		case XEN_SCSI_DISK11_MAJOR:
1062 		case XEN_SCSI_DISK12_MAJOR:
1063 		case XEN_SCSI_DISK13_MAJOR:
1064 		case XEN_SCSI_DISK14_MAJOR:
1065 		case XEN_SCSI_DISK15_MAJOR:
1066 			*offset = (*minor / PARTS_PER_DISK) +
1067 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1068 				EMULATED_SD_DISK_NAME_OFFSET;
1069 			*minor = *minor +
1070 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1071 				EMULATED_SD_DISK_MINOR_OFFSET;
1072 			break;
1073 		case XENVBD_MAJOR:
1074 			*offset = *minor / PARTS_PER_DISK;
1075 			break;
1076 		default:
1077 			printk(KERN_WARNING "blkfront: your disk configuration is "
1078 					"incorrect, please use an xvd device instead\n");
1079 			return -ENODEV;
1080 	}
1081 	return 0;
1082 }
1083 
1084 static char *encode_disk_name(char *ptr, unsigned int n)
1085 {
1086 	if (n >= 26)
1087 		ptr = encode_disk_name(ptr, n / 26 - 1);
1088 	*ptr = 'a' + n % 26;
1089 	return ptr + 1;
1090 }
1091 
1092 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1093 			       struct blkfront_info *info,
1094 			       u16 vdisk_info, u16 sector_size,
1095 			       unsigned int physical_sector_size)
1096 {
1097 	struct gendisk *gd;
1098 	int nr_minors = 1;
1099 	int err;
1100 	unsigned int offset;
1101 	int minor;
1102 	int nr_parts;
1103 	char *ptr;
1104 
1105 	BUG_ON(info->gd != NULL);
1106 	BUG_ON(info->rq != NULL);
1107 
1108 	if ((info->vdevice>>EXT_SHIFT) > 1) {
1109 		/* this is above the extended range; something is wrong */
1110 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1111 		return -ENODEV;
1112 	}
1113 
1114 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1115 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1116 		if (err)
1117 			return err;
1118  		nr_parts = PARTS_PER_DISK;
1119 	} else {
1120 		minor = BLKIF_MINOR_EXT(info->vdevice);
1121 		nr_parts = PARTS_PER_EXT_DISK;
1122 		offset = minor / nr_parts;
1123 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1124 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1125 					"emulated IDE disks,\n\t choose an xvd device name"
1126 					"from xvde on\n", info->vdevice);
1127 	}
1128 	if (minor >> MINORBITS) {
1129 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1130 			info->vdevice, minor);
1131 		return -ENODEV;
1132 	}
1133 
1134 	if ((minor % nr_parts) == 0)
1135 		nr_minors = nr_parts;
1136 
1137 	err = xlbd_reserve_minors(minor, nr_minors);
1138 	if (err)
1139 		goto out;
1140 	err = -ENODEV;
1141 
1142 	gd = alloc_disk(nr_minors);
1143 	if (gd == NULL)
1144 		goto release;
1145 
1146 	strcpy(gd->disk_name, DEV_NAME);
1147 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1148 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1149 	if (nr_minors > 1)
1150 		*ptr = 0;
1151 	else
1152 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1153 			 "%d", minor & (nr_parts - 1));
1154 
1155 	gd->major = XENVBD_MAJOR;
1156 	gd->first_minor = minor;
1157 	gd->fops = &xlvbd_block_fops;
1158 	gd->private_data = info;
1159 	set_capacity(gd, capacity);
1160 
1161 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1162 		del_gendisk(gd);
1163 		goto release;
1164 	}
1165 
1166 	xlvbd_flush(info);
1167 
1168 	if (vdisk_info & VDISK_READONLY)
1169 		set_disk_ro(gd, 1);
1170 
1171 	if (vdisk_info & VDISK_REMOVABLE)
1172 		gd->flags |= GENHD_FL_REMOVABLE;
1173 
1174 	if (vdisk_info & VDISK_CDROM)
1175 		gd->flags |= GENHD_FL_CD;
1176 
1177 	return 0;
1178 
1179  release:
1180 	xlbd_release_minors(minor, nr_minors);
1181  out:
1182 	return err;
1183 }
1184 
1185 static void xlvbd_release_gendisk(struct blkfront_info *info)
1186 {
1187 	unsigned int minor, nr_minors, i;
1188 
1189 	if (info->rq == NULL)
1190 		return;
1191 
1192 	/* No more blkif_request(). */
1193 	blk_mq_stop_hw_queues(info->rq);
1194 
1195 	for (i = 0; i < info->nr_rings; i++) {
1196 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1197 
1198 		/* No more gnttab callback work. */
1199 		gnttab_cancel_free_callback(&rinfo->callback);
1200 
1201 		/* Flush gnttab callback work. Must be done with no locks held. */
1202 		flush_work(&rinfo->work);
1203 	}
1204 
1205 	del_gendisk(info->gd);
1206 
1207 	minor = info->gd->first_minor;
1208 	nr_minors = info->gd->minors;
1209 	xlbd_release_minors(minor, nr_minors);
1210 
1211 	blk_cleanup_queue(info->rq);
1212 	blk_mq_free_tag_set(&info->tag_set);
1213 	info->rq = NULL;
1214 
1215 	put_disk(info->gd);
1216 	info->gd = NULL;
1217 }
1218 
1219 /* Already hold rinfo->ring_lock. */
1220 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1221 {
1222 	if (!RING_FULL(&rinfo->ring))
1223 		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1224 }
1225 
1226 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1227 {
1228 	unsigned long flags;
1229 
1230 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1231 	kick_pending_request_queues_locked(rinfo);
1232 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1233 }
1234 
1235 static void blkif_restart_queue(struct work_struct *work)
1236 {
1237 	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1238 
1239 	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1240 		kick_pending_request_queues(rinfo);
1241 }
1242 
1243 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1244 {
1245 	struct grant *persistent_gnt, *n;
1246 	struct blkfront_info *info = rinfo->dev_info;
1247 	int i, j, segs;
1248 
1249 	/*
1250 	 * Remove indirect pages, this only happens when using indirect
1251 	 * descriptors but not persistent grants
1252 	 */
1253 	if (!list_empty(&rinfo->indirect_pages)) {
1254 		struct page *indirect_page, *n;
1255 
1256 		BUG_ON(info->feature_persistent);
1257 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1258 			list_del(&indirect_page->lru);
1259 			__free_page(indirect_page);
1260 		}
1261 	}
1262 
1263 	/* Remove all persistent grants. */
1264 	if (!list_empty(&rinfo->grants)) {
1265 		list_for_each_entry_safe(persistent_gnt, n,
1266 					 &rinfo->grants, node) {
1267 			list_del(&persistent_gnt->node);
1268 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
1269 				gnttab_end_foreign_access(persistent_gnt->gref,
1270 							  0, 0UL);
1271 				rinfo->persistent_gnts_c--;
1272 			}
1273 			if (info->feature_persistent)
1274 				__free_page(persistent_gnt->page);
1275 			kfree(persistent_gnt);
1276 		}
1277 	}
1278 	BUG_ON(rinfo->persistent_gnts_c != 0);
1279 
1280 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1281 		/*
1282 		 * Clear persistent grants present in requests already
1283 		 * on the shared ring
1284 		 */
1285 		if (!rinfo->shadow[i].request)
1286 			goto free_shadow;
1287 
1288 		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1289 		       rinfo->shadow[i].req.u.indirect.nr_segments :
1290 		       rinfo->shadow[i].req.u.rw.nr_segments;
1291 		for (j = 0; j < segs; j++) {
1292 			persistent_gnt = rinfo->shadow[i].grants_used[j];
1293 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1294 			if (info->feature_persistent)
1295 				__free_page(persistent_gnt->page);
1296 			kfree(persistent_gnt);
1297 		}
1298 
1299 		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1300 			/*
1301 			 * If this is not an indirect operation don't try to
1302 			 * free indirect segments
1303 			 */
1304 			goto free_shadow;
1305 
1306 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1307 			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1308 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1309 			__free_page(persistent_gnt->page);
1310 			kfree(persistent_gnt);
1311 		}
1312 
1313 free_shadow:
1314 		kfree(rinfo->shadow[i].grants_used);
1315 		rinfo->shadow[i].grants_used = NULL;
1316 		kfree(rinfo->shadow[i].indirect_grants);
1317 		rinfo->shadow[i].indirect_grants = NULL;
1318 		kfree(rinfo->shadow[i].sg);
1319 		rinfo->shadow[i].sg = NULL;
1320 	}
1321 
1322 	/* No more gnttab callback work. */
1323 	gnttab_cancel_free_callback(&rinfo->callback);
1324 
1325 	/* Flush gnttab callback work. Must be done with no locks held. */
1326 	flush_work(&rinfo->work);
1327 
1328 	/* Free resources associated with old device channel. */
1329 	for (i = 0; i < info->nr_ring_pages; i++) {
1330 		if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1331 			gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1332 			rinfo->ring_ref[i] = GRANT_INVALID_REF;
1333 		}
1334 	}
1335 	free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1336 	rinfo->ring.sring = NULL;
1337 
1338 	if (rinfo->irq)
1339 		unbind_from_irqhandler(rinfo->irq, rinfo);
1340 	rinfo->evtchn = rinfo->irq = 0;
1341 }
1342 
1343 static void blkif_free(struct blkfront_info *info, int suspend)
1344 {
1345 	unsigned int i;
1346 
1347 	/* Prevent new requests being issued until we fix things up. */
1348 	info->connected = suspend ?
1349 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1350 	/* No more blkif_request(). */
1351 	if (info->rq)
1352 		blk_mq_stop_hw_queues(info->rq);
1353 
1354 	for (i = 0; i < info->nr_rings; i++)
1355 		blkif_free_ring(&info->rinfo[i]);
1356 
1357 	kfree(info->rinfo);
1358 	info->rinfo = NULL;
1359 	info->nr_rings = 0;
1360 }
1361 
1362 struct copy_from_grant {
1363 	const struct blk_shadow *s;
1364 	unsigned int grant_idx;
1365 	unsigned int bvec_offset;
1366 	char *bvec_data;
1367 };
1368 
1369 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1370 				  unsigned int len, void *data)
1371 {
1372 	struct copy_from_grant *info = data;
1373 	char *shared_data;
1374 	/* Convenient aliases */
1375 	const struct blk_shadow *s = info->s;
1376 
1377 	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1378 
1379 	memcpy(info->bvec_data + info->bvec_offset,
1380 	       shared_data + offset, len);
1381 
1382 	info->bvec_offset += len;
1383 	info->grant_idx++;
1384 
1385 	kunmap_atomic(shared_data);
1386 }
1387 
1388 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1389 {
1390 	switch (rsp)
1391 	{
1392 	case BLKIF_RSP_OKAY:
1393 		return REQ_DONE;
1394 	case BLKIF_RSP_EOPNOTSUPP:
1395 		return REQ_EOPNOTSUPP;
1396 	case BLKIF_RSP_ERROR:
1397 		/* Fallthrough. */
1398 	default:
1399 		return REQ_ERROR;
1400 	}
1401 }
1402 
1403 /*
1404  * Get the final status of the block request based on two ring response
1405  */
1406 static int blkif_get_final_status(enum blk_req_status s1,
1407 				  enum blk_req_status s2)
1408 {
1409 	BUG_ON(s1 == REQ_WAITING);
1410 	BUG_ON(s2 == REQ_WAITING);
1411 
1412 	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1413 		return BLKIF_RSP_ERROR;
1414 	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1415 		return BLKIF_RSP_EOPNOTSUPP;
1416 	return BLKIF_RSP_OKAY;
1417 }
1418 
1419 static bool blkif_completion(unsigned long *id,
1420 			     struct blkfront_ring_info *rinfo,
1421 			     struct blkif_response *bret)
1422 {
1423 	int i = 0;
1424 	struct scatterlist *sg;
1425 	int num_sg, num_grant;
1426 	struct blkfront_info *info = rinfo->dev_info;
1427 	struct blk_shadow *s = &rinfo->shadow[*id];
1428 	struct copy_from_grant data = {
1429 		.grant_idx = 0,
1430 	};
1431 
1432 	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1433 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1434 
1435 	/* The I/O request may be split in two. */
1436 	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1437 		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1438 
1439 		/* Keep the status of the current response in shadow. */
1440 		s->status = blkif_rsp_to_req_status(bret->status);
1441 
1442 		/* Wait the second response if not yet here. */
1443 		if (s2->status == REQ_WAITING)
1444 			return 0;
1445 
1446 		bret->status = blkif_get_final_status(s->status,
1447 						      s2->status);
1448 
1449 		/*
1450 		 * All the grants is stored in the first shadow in order
1451 		 * to make the completion code simpler.
1452 		 */
1453 		num_grant += s2->req.u.rw.nr_segments;
1454 
1455 		/*
1456 		 * The two responses may not come in order. Only the
1457 		 * first request will store the scatter-gather list.
1458 		 */
1459 		if (s2->num_sg != 0) {
1460 			/* Update "id" with the ID of the first response. */
1461 			*id = s->associated_id;
1462 			s = s2;
1463 		}
1464 
1465 		/*
1466 		 * We don't need anymore the second request, so recycling
1467 		 * it now.
1468 		 */
1469 		if (add_id_to_freelist(rinfo, s->associated_id))
1470 			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1471 			     info->gd->disk_name, s->associated_id);
1472 	}
1473 
1474 	data.s = s;
1475 	num_sg = s->num_sg;
1476 
1477 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1478 		for_each_sg(s->sg, sg, num_sg, i) {
1479 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1480 
1481 			data.bvec_offset = sg->offset;
1482 			data.bvec_data = kmap_atomic(sg_page(sg));
1483 
1484 			gnttab_foreach_grant_in_range(sg_page(sg),
1485 						      sg->offset,
1486 						      sg->length,
1487 						      blkif_copy_from_grant,
1488 						      &data);
1489 
1490 			kunmap_atomic(data.bvec_data);
1491 		}
1492 	}
1493 	/* Add the persistent grant into the list of free grants */
1494 	for (i = 0; i < num_grant; i++) {
1495 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1496 			/*
1497 			 * If the grant is still mapped by the backend (the
1498 			 * backend has chosen to make this grant persistent)
1499 			 * we add it at the head of the list, so it will be
1500 			 * reused first.
1501 			 */
1502 			if (!info->feature_persistent)
1503 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1504 						     s->grants_used[i]->gref);
1505 			list_add(&s->grants_used[i]->node, &rinfo->grants);
1506 			rinfo->persistent_gnts_c++;
1507 		} else {
1508 			/*
1509 			 * If the grant is not mapped by the backend we end the
1510 			 * foreign access and add it to the tail of the list,
1511 			 * so it will not be picked again unless we run out of
1512 			 * persistent grants.
1513 			 */
1514 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1515 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1516 			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1517 		}
1518 	}
1519 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1520 		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1521 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1522 				if (!info->feature_persistent)
1523 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1524 							     s->indirect_grants[i]->gref);
1525 				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1526 				rinfo->persistent_gnts_c++;
1527 			} else {
1528 				struct page *indirect_page;
1529 
1530 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1531 				/*
1532 				 * Add the used indirect page back to the list of
1533 				 * available pages for indirect grefs.
1534 				 */
1535 				if (!info->feature_persistent) {
1536 					indirect_page = s->indirect_grants[i]->page;
1537 					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1538 				}
1539 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1540 				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1541 			}
1542 		}
1543 	}
1544 
1545 	return 1;
1546 }
1547 
1548 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1549 {
1550 	struct request *req;
1551 	struct blkif_response *bret;
1552 	RING_IDX i, rp;
1553 	unsigned long flags;
1554 	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1555 	struct blkfront_info *info = rinfo->dev_info;
1556 
1557 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1558 		return IRQ_HANDLED;
1559 
1560 	spin_lock_irqsave(&rinfo->ring_lock, flags);
1561  again:
1562 	rp = rinfo->ring.sring->rsp_prod;
1563 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1564 
1565 	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1566 		unsigned long id;
1567 
1568 		bret = RING_GET_RESPONSE(&rinfo->ring, i);
1569 		id   = bret->id;
1570 		/*
1571 		 * The backend has messed up and given us an id that we would
1572 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1573 		 * look in get_id_from_freelist.
1574 		 */
1575 		if (id >= BLK_RING_SIZE(info)) {
1576 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1577 			     info->gd->disk_name, op_name(bret->operation), id);
1578 			/* We can't safely get the 'struct request' as
1579 			 * the id is busted. */
1580 			continue;
1581 		}
1582 		req  = rinfo->shadow[id].request;
1583 
1584 		if (bret->operation != BLKIF_OP_DISCARD) {
1585 			/*
1586 			 * We may need to wait for an extra response if the
1587 			 * I/O request is split in 2
1588 			 */
1589 			if (!blkif_completion(&id, rinfo, bret))
1590 				continue;
1591 		}
1592 
1593 		if (add_id_to_freelist(rinfo, id)) {
1594 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1595 			     info->gd->disk_name, op_name(bret->operation), id);
1596 			continue;
1597 		}
1598 
1599 		if (bret->status == BLKIF_RSP_OKAY)
1600 			blkif_req(req)->error = BLK_STS_OK;
1601 		else
1602 			blkif_req(req)->error = BLK_STS_IOERR;
1603 
1604 		switch (bret->operation) {
1605 		case BLKIF_OP_DISCARD:
1606 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1607 				struct request_queue *rq = info->rq;
1608 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1609 					   info->gd->disk_name, op_name(bret->operation));
1610 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1611 				info->feature_discard = 0;
1612 				info->feature_secdiscard = 0;
1613 				blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1614 				blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1615 			}
1616 			break;
1617 		case BLKIF_OP_FLUSH_DISKCACHE:
1618 		case BLKIF_OP_WRITE_BARRIER:
1619 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1620 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1621 				       info->gd->disk_name, op_name(bret->operation));
1622 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1623 			}
1624 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1625 				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1626 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1627 				       info->gd->disk_name, op_name(bret->operation));
1628 				blkif_req(req)->error = BLK_STS_NOTSUPP;
1629 			}
1630 			if (unlikely(blkif_req(req)->error)) {
1631 				if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1632 					blkif_req(req)->error = BLK_STS_OK;
1633 				info->feature_fua = 0;
1634 				info->feature_flush = 0;
1635 				xlvbd_flush(info);
1636 			}
1637 			/* fall through */
1638 		case BLKIF_OP_READ:
1639 		case BLKIF_OP_WRITE:
1640 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1641 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1642 					"request: %x\n", bret->status);
1643 
1644 			break;
1645 		default:
1646 			BUG();
1647 		}
1648 
1649 		blk_mq_complete_request(req);
1650 	}
1651 
1652 	rinfo->ring.rsp_cons = i;
1653 
1654 	if (i != rinfo->ring.req_prod_pvt) {
1655 		int more_to_do;
1656 		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1657 		if (more_to_do)
1658 			goto again;
1659 	} else
1660 		rinfo->ring.sring->rsp_event = i + 1;
1661 
1662 	kick_pending_request_queues_locked(rinfo);
1663 
1664 	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1665 
1666 	return IRQ_HANDLED;
1667 }
1668 
1669 
1670 static int setup_blkring(struct xenbus_device *dev,
1671 			 struct blkfront_ring_info *rinfo)
1672 {
1673 	struct blkif_sring *sring;
1674 	int err, i;
1675 	struct blkfront_info *info = rinfo->dev_info;
1676 	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1677 	grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1678 
1679 	for (i = 0; i < info->nr_ring_pages; i++)
1680 		rinfo->ring_ref[i] = GRANT_INVALID_REF;
1681 
1682 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1683 						       get_order(ring_size));
1684 	if (!sring) {
1685 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1686 		return -ENOMEM;
1687 	}
1688 	SHARED_RING_INIT(sring);
1689 	FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1690 
1691 	err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1692 	if (err < 0) {
1693 		free_pages((unsigned long)sring, get_order(ring_size));
1694 		rinfo->ring.sring = NULL;
1695 		goto fail;
1696 	}
1697 	for (i = 0; i < info->nr_ring_pages; i++)
1698 		rinfo->ring_ref[i] = gref[i];
1699 
1700 	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1701 	if (err)
1702 		goto fail;
1703 
1704 	err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1705 					"blkif", rinfo);
1706 	if (err <= 0) {
1707 		xenbus_dev_fatal(dev, err,
1708 				 "bind_evtchn_to_irqhandler failed");
1709 		goto fail;
1710 	}
1711 	rinfo->irq = err;
1712 
1713 	return 0;
1714 fail:
1715 	blkif_free(info, 0);
1716 	return err;
1717 }
1718 
1719 /*
1720  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1721  * ring buffer may have multi pages depending on ->nr_ring_pages.
1722  */
1723 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1724 				struct blkfront_ring_info *rinfo, const char *dir)
1725 {
1726 	int err;
1727 	unsigned int i;
1728 	const char *message = NULL;
1729 	struct blkfront_info *info = rinfo->dev_info;
1730 
1731 	if (info->nr_ring_pages == 1) {
1732 		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1733 		if (err) {
1734 			message = "writing ring-ref";
1735 			goto abort_transaction;
1736 		}
1737 	} else {
1738 		for (i = 0; i < info->nr_ring_pages; i++) {
1739 			char ring_ref_name[RINGREF_NAME_LEN];
1740 
1741 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1742 			err = xenbus_printf(xbt, dir, ring_ref_name,
1743 					    "%u", rinfo->ring_ref[i]);
1744 			if (err) {
1745 				message = "writing ring-ref";
1746 				goto abort_transaction;
1747 			}
1748 		}
1749 	}
1750 
1751 	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1752 	if (err) {
1753 		message = "writing event-channel";
1754 		goto abort_transaction;
1755 	}
1756 
1757 	return 0;
1758 
1759 abort_transaction:
1760 	xenbus_transaction_end(xbt, 1);
1761 	if (message)
1762 		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1763 
1764 	return err;
1765 }
1766 
1767 /* Common code used when first setting up, and when resuming. */
1768 static int talk_to_blkback(struct xenbus_device *dev,
1769 			   struct blkfront_info *info)
1770 {
1771 	const char *message = NULL;
1772 	struct xenbus_transaction xbt;
1773 	int err;
1774 	unsigned int i, max_page_order;
1775 	unsigned int ring_page_order;
1776 
1777 	if (!info)
1778 		return -ENODEV;
1779 
1780 	max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1781 					      "max-ring-page-order", 0);
1782 	ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1783 	info->nr_ring_pages = 1 << ring_page_order;
1784 
1785 	err = negotiate_mq(info);
1786 	if (err)
1787 		goto destroy_blkring;
1788 
1789 	for (i = 0; i < info->nr_rings; i++) {
1790 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1791 
1792 		/* Create shared ring, alloc event channel. */
1793 		err = setup_blkring(dev, rinfo);
1794 		if (err)
1795 			goto destroy_blkring;
1796 	}
1797 
1798 again:
1799 	err = xenbus_transaction_start(&xbt);
1800 	if (err) {
1801 		xenbus_dev_fatal(dev, err, "starting transaction");
1802 		goto destroy_blkring;
1803 	}
1804 
1805 	if (info->nr_ring_pages > 1) {
1806 		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1807 				    ring_page_order);
1808 		if (err) {
1809 			message = "writing ring-page-order";
1810 			goto abort_transaction;
1811 		}
1812 	}
1813 
1814 	/* We already got the number of queues/rings in _probe */
1815 	if (info->nr_rings == 1) {
1816 		err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1817 		if (err)
1818 			goto destroy_blkring;
1819 	} else {
1820 		char *path;
1821 		size_t pathsize;
1822 
1823 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1824 				    info->nr_rings);
1825 		if (err) {
1826 			message = "writing multi-queue-num-queues";
1827 			goto abort_transaction;
1828 		}
1829 
1830 		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1831 		path = kmalloc(pathsize, GFP_KERNEL);
1832 		if (!path) {
1833 			err = -ENOMEM;
1834 			message = "ENOMEM while writing ring references";
1835 			goto abort_transaction;
1836 		}
1837 
1838 		for (i = 0; i < info->nr_rings; i++) {
1839 			memset(path, 0, pathsize);
1840 			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1841 			err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1842 			if (err) {
1843 				kfree(path);
1844 				goto destroy_blkring;
1845 			}
1846 		}
1847 		kfree(path);
1848 	}
1849 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1850 			    XEN_IO_PROTO_ABI_NATIVE);
1851 	if (err) {
1852 		message = "writing protocol";
1853 		goto abort_transaction;
1854 	}
1855 	err = xenbus_printf(xbt, dev->nodename,
1856 			    "feature-persistent", "%u", 1);
1857 	if (err)
1858 		dev_warn(&dev->dev,
1859 			 "writing persistent grants feature to xenbus");
1860 
1861 	err = xenbus_transaction_end(xbt, 0);
1862 	if (err) {
1863 		if (err == -EAGAIN)
1864 			goto again;
1865 		xenbus_dev_fatal(dev, err, "completing transaction");
1866 		goto destroy_blkring;
1867 	}
1868 
1869 	for (i = 0; i < info->nr_rings; i++) {
1870 		unsigned int j;
1871 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
1872 
1873 		for (j = 0; j < BLK_RING_SIZE(info); j++)
1874 			rinfo->shadow[j].req.u.rw.id = j + 1;
1875 		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1876 	}
1877 	xenbus_switch_state(dev, XenbusStateInitialised);
1878 
1879 	return 0;
1880 
1881  abort_transaction:
1882 	xenbus_transaction_end(xbt, 1);
1883 	if (message)
1884 		xenbus_dev_fatal(dev, err, "%s", message);
1885  destroy_blkring:
1886 	blkif_free(info, 0);
1887 
1888 	kfree(info);
1889 	dev_set_drvdata(&dev->dev, NULL);
1890 
1891 	return err;
1892 }
1893 
1894 static int negotiate_mq(struct blkfront_info *info)
1895 {
1896 	unsigned int backend_max_queues;
1897 	unsigned int i;
1898 
1899 	BUG_ON(info->nr_rings);
1900 
1901 	/* Check if backend supports multiple queues. */
1902 	backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1903 						  "multi-queue-max-queues", 1);
1904 	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1905 	/* We need at least one ring. */
1906 	if (!info->nr_rings)
1907 		info->nr_rings = 1;
1908 
1909 	info->rinfo = kcalloc(info->nr_rings,
1910 			      sizeof(struct blkfront_ring_info),
1911 			      GFP_KERNEL);
1912 	if (!info->rinfo) {
1913 		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1914 		return -ENOMEM;
1915 	}
1916 
1917 	for (i = 0; i < info->nr_rings; i++) {
1918 		struct blkfront_ring_info *rinfo;
1919 
1920 		rinfo = &info->rinfo[i];
1921 		INIT_LIST_HEAD(&rinfo->indirect_pages);
1922 		INIT_LIST_HEAD(&rinfo->grants);
1923 		rinfo->dev_info = info;
1924 		INIT_WORK(&rinfo->work, blkif_restart_queue);
1925 		spin_lock_init(&rinfo->ring_lock);
1926 	}
1927 	return 0;
1928 }
1929 /**
1930  * Entry point to this code when a new device is created.  Allocate the basic
1931  * structures and the ring buffer for communication with the backend, and
1932  * inform the backend of the appropriate details for those.  Switch to
1933  * Initialised state.
1934  */
1935 static int blkfront_probe(struct xenbus_device *dev,
1936 			  const struct xenbus_device_id *id)
1937 {
1938 	int err, vdevice;
1939 	struct blkfront_info *info;
1940 
1941 	/* FIXME: Use dynamic device id if this is not set. */
1942 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1943 			   "virtual-device", "%i", &vdevice);
1944 	if (err != 1) {
1945 		/* go looking in the extended area instead */
1946 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1947 				   "%i", &vdevice);
1948 		if (err != 1) {
1949 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1950 			return err;
1951 		}
1952 	}
1953 
1954 	if (xen_hvm_domain()) {
1955 		char *type;
1956 		int len;
1957 		/* no unplug has been done: do not hook devices != xen vbds */
1958 		if (xen_has_pv_and_legacy_disk_devices()) {
1959 			int major;
1960 
1961 			if (!VDEV_IS_EXTENDED(vdevice))
1962 				major = BLKIF_MAJOR(vdevice);
1963 			else
1964 				major = XENVBD_MAJOR;
1965 
1966 			if (major != XENVBD_MAJOR) {
1967 				printk(KERN_INFO
1968 						"%s: HVM does not support vbd %d as xen block device\n",
1969 						__func__, vdevice);
1970 				return -ENODEV;
1971 			}
1972 		}
1973 		/* do not create a PV cdrom device if we are an HVM guest */
1974 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1975 		if (IS_ERR(type))
1976 			return -ENODEV;
1977 		if (strncmp(type, "cdrom", 5) == 0) {
1978 			kfree(type);
1979 			return -ENODEV;
1980 		}
1981 		kfree(type);
1982 	}
1983 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1984 	if (!info) {
1985 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1986 		return -ENOMEM;
1987 	}
1988 
1989 	info->xbdev = dev;
1990 
1991 	mutex_init(&info->mutex);
1992 	info->vdevice = vdevice;
1993 	info->connected = BLKIF_STATE_DISCONNECTED;
1994 
1995 	/* Front end dir is a number, which is used as the id. */
1996 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1997 	dev_set_drvdata(&dev->dev, info);
1998 
1999 	return 0;
2000 }
2001 
2002 static int blkif_recover(struct blkfront_info *info)
2003 {
2004 	unsigned int r_index;
2005 	struct request *req, *n;
2006 	int rc;
2007 	struct bio *bio;
2008 	unsigned int segs;
2009 
2010 	blkfront_gather_backend_features(info);
2011 	/* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2012 	blkif_set_queue_limits(info);
2013 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2014 	blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2015 
2016 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2017 		struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2018 
2019 		rc = blkfront_setup_indirect(rinfo);
2020 		if (rc)
2021 			return rc;
2022 	}
2023 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2024 
2025 	/* Now safe for us to use the shared ring */
2026 	info->connected = BLKIF_STATE_CONNECTED;
2027 
2028 	for (r_index = 0; r_index < info->nr_rings; r_index++) {
2029 		struct blkfront_ring_info *rinfo;
2030 
2031 		rinfo = &info->rinfo[r_index];
2032 		/* Kick any other new requests queued since we resumed */
2033 		kick_pending_request_queues(rinfo);
2034 	}
2035 
2036 	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2037 		/* Requeue pending requests (flush or discard) */
2038 		list_del_init(&req->queuelist);
2039 		BUG_ON(req->nr_phys_segments > segs);
2040 		blk_mq_requeue_request(req, false);
2041 	}
2042 	blk_mq_start_stopped_hw_queues(info->rq, true);
2043 	blk_mq_kick_requeue_list(info->rq);
2044 
2045 	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2046 		/* Traverse the list of pending bios and re-queue them */
2047 		submit_bio(bio);
2048 	}
2049 
2050 	return 0;
2051 }
2052 
2053 /**
2054  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2055  * driver restart.  We tear down our blkif structure and recreate it, but
2056  * leave the device-layer structures intact so that this is transparent to the
2057  * rest of the kernel.
2058  */
2059 static int blkfront_resume(struct xenbus_device *dev)
2060 {
2061 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2062 	int err = 0;
2063 	unsigned int i, j;
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 (i = 0; i < info->nr_rings; i++) {
2070 		struct blkfront_ring_info *rinfo = &info->rinfo[i];
2071 		struct bio_list merge_bio;
2072 		struct blk_shadow *shadow = rinfo->shadow;
2073 
2074 		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2075 			/* Not in use? */
2076 			if (!shadow[j].request)
2077 				continue;
2078 
2079 			/*
2080 			 * Get the bios in the request so we can re-queue them.
2081 			 */
2082 			if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2083 			    req_op(shadow[j].request) == REQ_OP_DISCARD ||
2084 			    req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2085 			    shadow[j].request->cmd_flags & REQ_FUA) {
2086 				/*
2087 				 * Flush operations don't contain bios, so
2088 				 * we need to requeue the whole request
2089 				 *
2090 				 * XXX: but this doesn't make any sense for a
2091 				 * write with the FUA flag set..
2092 				 */
2093 				list_add(&shadow[j].request->queuelist, &info->requests);
2094 				continue;
2095 			}
2096 			merge_bio.head = shadow[j].request->bio;
2097 			merge_bio.tail = shadow[j].request->biotail;
2098 			bio_list_merge(&info->bio_list, &merge_bio);
2099 			shadow[j].request->bio = NULL;
2100 			blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2101 		}
2102 	}
2103 
2104 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2105 
2106 	err = talk_to_blkback(dev, info);
2107 	if (!err)
2108 		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2109 
2110 	/*
2111 	 * We have to wait for the backend to switch to
2112 	 * connected state, since we want to read which
2113 	 * features it supports.
2114 	 */
2115 
2116 	return err;
2117 }
2118 
2119 static void blkfront_closing(struct blkfront_info *info)
2120 {
2121 	struct xenbus_device *xbdev = info->xbdev;
2122 	struct block_device *bdev = NULL;
2123 
2124 	mutex_lock(&info->mutex);
2125 
2126 	if (xbdev->state == XenbusStateClosing) {
2127 		mutex_unlock(&info->mutex);
2128 		return;
2129 	}
2130 
2131 	if (info->gd)
2132 		bdev = bdget_disk(info->gd, 0);
2133 
2134 	mutex_unlock(&info->mutex);
2135 
2136 	if (!bdev) {
2137 		xenbus_frontend_closed(xbdev);
2138 		return;
2139 	}
2140 
2141 	mutex_lock(&bdev->bd_mutex);
2142 
2143 	if (bdev->bd_openers) {
2144 		xenbus_dev_error(xbdev, -EBUSY,
2145 				 "Device in use; refusing to close");
2146 		xenbus_switch_state(xbdev, XenbusStateClosing);
2147 	} else {
2148 		xlvbd_release_gendisk(info);
2149 		xenbus_frontend_closed(xbdev);
2150 	}
2151 
2152 	mutex_unlock(&bdev->bd_mutex);
2153 	bdput(bdev);
2154 }
2155 
2156 static void blkfront_setup_discard(struct blkfront_info *info)
2157 {
2158 	int err;
2159 	unsigned int discard_granularity;
2160 	unsigned int discard_alignment;
2161 
2162 	info->feature_discard = 1;
2163 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2164 		"discard-granularity", "%u", &discard_granularity,
2165 		"discard-alignment", "%u", &discard_alignment,
2166 		NULL);
2167 	if (!err) {
2168 		info->discard_granularity = discard_granularity;
2169 		info->discard_alignment = discard_alignment;
2170 	}
2171 	info->feature_secdiscard =
2172 		!!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2173 				       0);
2174 }
2175 
2176 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2177 {
2178 	unsigned int psegs, grants;
2179 	int err, i;
2180 	struct blkfront_info *info = rinfo->dev_info;
2181 
2182 	if (info->max_indirect_segments == 0) {
2183 		if (!HAS_EXTRA_REQ)
2184 			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2185 		else {
2186 			/*
2187 			 * When an extra req is required, the maximum
2188 			 * grants supported is related to the size of the
2189 			 * Linux block segment.
2190 			 */
2191 			grants = GRANTS_PER_PSEG;
2192 		}
2193 	}
2194 	else
2195 		grants = info->max_indirect_segments;
2196 	psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2197 
2198 	err = fill_grant_buffer(rinfo,
2199 				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2200 	if (err)
2201 		goto out_of_memory;
2202 
2203 	if (!info->feature_persistent && info->max_indirect_segments) {
2204 		/*
2205 		 * We are using indirect descriptors but not persistent
2206 		 * grants, we need to allocate a set of pages that can be
2207 		 * used for mapping indirect grefs
2208 		 */
2209 		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2210 
2211 		BUG_ON(!list_empty(&rinfo->indirect_pages));
2212 		for (i = 0; i < num; i++) {
2213 			struct page *indirect_page = alloc_page(GFP_NOIO);
2214 			if (!indirect_page)
2215 				goto out_of_memory;
2216 			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2217 		}
2218 	}
2219 
2220 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2221 		rinfo->shadow[i].grants_used =
2222 			kcalloc(grants,
2223 				sizeof(rinfo->shadow[i].grants_used[0]),
2224 				GFP_NOIO);
2225 		rinfo->shadow[i].sg = kcalloc(psegs,
2226 					      sizeof(rinfo->shadow[i].sg[0]),
2227 					      GFP_NOIO);
2228 		if (info->max_indirect_segments)
2229 			rinfo->shadow[i].indirect_grants =
2230 				kcalloc(INDIRECT_GREFS(grants),
2231 					sizeof(rinfo->shadow[i].indirect_grants[0]),
2232 					GFP_NOIO);
2233 		if ((rinfo->shadow[i].grants_used == NULL) ||
2234 			(rinfo->shadow[i].sg == NULL) ||
2235 		     (info->max_indirect_segments &&
2236 		     (rinfo->shadow[i].indirect_grants == NULL)))
2237 			goto out_of_memory;
2238 		sg_init_table(rinfo->shadow[i].sg, psegs);
2239 	}
2240 
2241 
2242 	return 0;
2243 
2244 out_of_memory:
2245 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2246 		kfree(rinfo->shadow[i].grants_used);
2247 		rinfo->shadow[i].grants_used = NULL;
2248 		kfree(rinfo->shadow[i].sg);
2249 		rinfo->shadow[i].sg = NULL;
2250 		kfree(rinfo->shadow[i].indirect_grants);
2251 		rinfo->shadow[i].indirect_grants = NULL;
2252 	}
2253 	if (!list_empty(&rinfo->indirect_pages)) {
2254 		struct page *indirect_page, *n;
2255 		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2256 			list_del(&indirect_page->lru);
2257 			__free_page(indirect_page);
2258 		}
2259 	}
2260 	return -ENOMEM;
2261 }
2262 
2263 /*
2264  * Gather all backend feature-*
2265  */
2266 static void blkfront_gather_backend_features(struct blkfront_info *info)
2267 {
2268 	unsigned int indirect_segments;
2269 
2270 	info->feature_flush = 0;
2271 	info->feature_fua = 0;
2272 
2273 	/*
2274 	 * If there's no "feature-barrier" defined, then it means
2275 	 * we're dealing with a very old backend which writes
2276 	 * synchronously; nothing to do.
2277 	 *
2278 	 * If there are barriers, then we use flush.
2279 	 */
2280 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2281 		info->feature_flush = 1;
2282 		info->feature_fua = 1;
2283 	}
2284 
2285 	/*
2286 	 * And if there is "feature-flush-cache" use that above
2287 	 * barriers.
2288 	 */
2289 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2290 				 0)) {
2291 		info->feature_flush = 1;
2292 		info->feature_fua = 0;
2293 	}
2294 
2295 	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2296 		blkfront_setup_discard(info);
2297 
2298 	info->feature_persistent =
2299 		!!xenbus_read_unsigned(info->xbdev->otherend,
2300 				       "feature-persistent", 0);
2301 
2302 	indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2303 					"feature-max-indirect-segments", 0);
2304 	if (indirect_segments > xen_blkif_max_segments)
2305 		indirect_segments = xen_blkif_max_segments;
2306 	if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2307 		indirect_segments = 0;
2308 	info->max_indirect_segments = indirect_segments;
2309 }
2310 
2311 /*
2312  * Invoked when the backend is finally 'ready' (and has told produced
2313  * the details about the physical device - #sectors, size, etc).
2314  */
2315 static void blkfront_connect(struct blkfront_info *info)
2316 {
2317 	unsigned long long sectors;
2318 	unsigned long sector_size;
2319 	unsigned int physical_sector_size;
2320 	unsigned int binfo;
2321 	char *envp[] = { "RESIZE=1", NULL };
2322 	int err, i;
2323 
2324 	switch (info->connected) {
2325 	case BLKIF_STATE_CONNECTED:
2326 		/*
2327 		 * Potentially, the back-end may be signalling
2328 		 * a capacity change; update the capacity.
2329 		 */
2330 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2331 				   "sectors", "%Lu", &sectors);
2332 		if (XENBUS_EXIST_ERR(err))
2333 			return;
2334 		printk(KERN_INFO "Setting capacity to %Lu\n",
2335 		       sectors);
2336 		set_capacity(info->gd, sectors);
2337 		revalidate_disk(info->gd);
2338 		kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2339 				   KOBJ_CHANGE, envp);
2340 
2341 		return;
2342 	case BLKIF_STATE_SUSPENDED:
2343 		/*
2344 		 * If we are recovering from suspension, we need to wait
2345 		 * for the backend to announce it's features before
2346 		 * reconnecting, at least we need to know if the backend
2347 		 * supports indirect descriptors, and how many.
2348 		 */
2349 		blkif_recover(info);
2350 		return;
2351 
2352 	default:
2353 		break;
2354 	}
2355 
2356 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2357 		__func__, info->xbdev->otherend);
2358 
2359 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2360 			    "sectors", "%llu", &sectors,
2361 			    "info", "%u", &binfo,
2362 			    "sector-size", "%lu", &sector_size,
2363 			    NULL);
2364 	if (err) {
2365 		xenbus_dev_fatal(info->xbdev, err,
2366 				 "reading backend fields at %s",
2367 				 info->xbdev->otherend);
2368 		return;
2369 	}
2370 
2371 	/*
2372 	 * physcial-sector-size is a newer field, so old backends may not
2373 	 * provide this. Assume physical sector size to be the same as
2374 	 * sector_size in that case.
2375 	 */
2376 	physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2377 						    "physical-sector-size",
2378 						    sector_size);
2379 	blkfront_gather_backend_features(info);
2380 	for (i = 0; i < info->nr_rings; i++) {
2381 		err = blkfront_setup_indirect(&info->rinfo[i]);
2382 		if (err) {
2383 			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2384 					 info->xbdev->otherend);
2385 			blkif_free(info, 0);
2386 			break;
2387 		}
2388 	}
2389 
2390 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2391 				  physical_sector_size);
2392 	if (err) {
2393 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2394 				 info->xbdev->otherend);
2395 		goto fail;
2396 	}
2397 
2398 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2399 
2400 	/* Kick pending requests. */
2401 	info->connected = BLKIF_STATE_CONNECTED;
2402 	for (i = 0; i < info->nr_rings; i++)
2403 		kick_pending_request_queues(&info->rinfo[i]);
2404 
2405 	device_add_disk(&info->xbdev->dev, info->gd);
2406 
2407 	info->is_ready = 1;
2408 	return;
2409 
2410 fail:
2411 	blkif_free(info, 0);
2412 	return;
2413 }
2414 
2415 /**
2416  * Callback received when the backend's state changes.
2417  */
2418 static void blkback_changed(struct xenbus_device *dev,
2419 			    enum xenbus_state backend_state)
2420 {
2421 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2422 
2423 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2424 
2425 	switch (backend_state) {
2426 	case XenbusStateInitWait:
2427 		if (dev->state != XenbusStateInitialising)
2428 			break;
2429 		if (talk_to_blkback(dev, info))
2430 			break;
2431 	case XenbusStateInitialising:
2432 	case XenbusStateInitialised:
2433 	case XenbusStateReconfiguring:
2434 	case XenbusStateReconfigured:
2435 	case XenbusStateUnknown:
2436 		break;
2437 
2438 	case XenbusStateConnected:
2439 		/*
2440 		 * talk_to_blkback sets state to XenbusStateInitialised
2441 		 * and blkfront_connect sets it to XenbusStateConnected
2442 		 * (if connection went OK).
2443 		 *
2444 		 * If the backend (or toolstack) decides to poke at backend
2445 		 * state (and re-trigger the watch by setting the state repeatedly
2446 		 * to XenbusStateConnected (4)) we need to deal with this.
2447 		 * This is allowed as this is used to communicate to the guest
2448 		 * that the size of disk has changed!
2449 		 */
2450 		if ((dev->state != XenbusStateInitialised) &&
2451 		    (dev->state != XenbusStateConnected)) {
2452 			if (talk_to_blkback(dev, info))
2453 				break;
2454 		}
2455 
2456 		blkfront_connect(info);
2457 		break;
2458 
2459 	case XenbusStateClosed:
2460 		if (dev->state == XenbusStateClosed)
2461 			break;
2462 		/* fall through */
2463 	case XenbusStateClosing:
2464 		if (info)
2465 			blkfront_closing(info);
2466 		break;
2467 	}
2468 }
2469 
2470 static int blkfront_remove(struct xenbus_device *xbdev)
2471 {
2472 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2473 	struct block_device *bdev = NULL;
2474 	struct gendisk *disk;
2475 
2476 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2477 
2478 	blkif_free(info, 0);
2479 
2480 	mutex_lock(&info->mutex);
2481 
2482 	disk = info->gd;
2483 	if (disk)
2484 		bdev = bdget_disk(disk, 0);
2485 
2486 	info->xbdev = NULL;
2487 	mutex_unlock(&info->mutex);
2488 
2489 	if (!bdev) {
2490 		kfree(info);
2491 		return 0;
2492 	}
2493 
2494 	/*
2495 	 * The xbdev was removed before we reached the Closed
2496 	 * state. See if it's safe to remove the disk. If the bdev
2497 	 * isn't closed yet, we let release take care of it.
2498 	 */
2499 
2500 	mutex_lock(&bdev->bd_mutex);
2501 	info = disk->private_data;
2502 
2503 	dev_warn(disk_to_dev(disk),
2504 		 "%s was hot-unplugged, %d stale handles\n",
2505 		 xbdev->nodename, bdev->bd_openers);
2506 
2507 	if (info && !bdev->bd_openers) {
2508 		xlvbd_release_gendisk(info);
2509 		disk->private_data = NULL;
2510 		kfree(info);
2511 	}
2512 
2513 	mutex_unlock(&bdev->bd_mutex);
2514 	bdput(bdev);
2515 
2516 	return 0;
2517 }
2518 
2519 static int blkfront_is_ready(struct xenbus_device *dev)
2520 {
2521 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2522 
2523 	return info->is_ready && info->xbdev;
2524 }
2525 
2526 static int blkif_open(struct block_device *bdev, fmode_t mode)
2527 {
2528 	struct gendisk *disk = bdev->bd_disk;
2529 	struct blkfront_info *info;
2530 	int err = 0;
2531 
2532 	mutex_lock(&blkfront_mutex);
2533 
2534 	info = disk->private_data;
2535 	if (!info) {
2536 		/* xbdev gone */
2537 		err = -ERESTARTSYS;
2538 		goto out;
2539 	}
2540 
2541 	mutex_lock(&info->mutex);
2542 
2543 	if (!info->gd)
2544 		/* xbdev is closed */
2545 		err = -ERESTARTSYS;
2546 
2547 	mutex_unlock(&info->mutex);
2548 
2549 out:
2550 	mutex_unlock(&blkfront_mutex);
2551 	return err;
2552 }
2553 
2554 static void blkif_release(struct gendisk *disk, fmode_t mode)
2555 {
2556 	struct blkfront_info *info = disk->private_data;
2557 	struct block_device *bdev;
2558 	struct xenbus_device *xbdev;
2559 
2560 	mutex_lock(&blkfront_mutex);
2561 
2562 	bdev = bdget_disk(disk, 0);
2563 
2564 	if (!bdev) {
2565 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2566 		goto out_mutex;
2567 	}
2568 	if (bdev->bd_openers)
2569 		goto out;
2570 
2571 	/*
2572 	 * Check if we have been instructed to close. We will have
2573 	 * deferred this request, because the bdev was still open.
2574 	 */
2575 
2576 	mutex_lock(&info->mutex);
2577 	xbdev = info->xbdev;
2578 
2579 	if (xbdev && xbdev->state == XenbusStateClosing) {
2580 		/* pending switch to state closed */
2581 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2582 		xlvbd_release_gendisk(info);
2583 		xenbus_frontend_closed(info->xbdev);
2584  	}
2585 
2586 	mutex_unlock(&info->mutex);
2587 
2588 	if (!xbdev) {
2589 		/* sudden device removal */
2590 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2591 		xlvbd_release_gendisk(info);
2592 		disk->private_data = NULL;
2593 		kfree(info);
2594 	}
2595 
2596 out:
2597 	bdput(bdev);
2598 out_mutex:
2599 	mutex_unlock(&blkfront_mutex);
2600 }
2601 
2602 static const struct block_device_operations xlvbd_block_fops =
2603 {
2604 	.owner = THIS_MODULE,
2605 	.open = blkif_open,
2606 	.release = blkif_release,
2607 	.getgeo = blkif_getgeo,
2608 	.ioctl = blkif_ioctl,
2609 };
2610 
2611 
2612 static const struct xenbus_device_id blkfront_ids[] = {
2613 	{ "vbd" },
2614 	{ "" }
2615 };
2616 
2617 static struct xenbus_driver blkfront_driver = {
2618 	.ids  = blkfront_ids,
2619 	.probe = blkfront_probe,
2620 	.remove = blkfront_remove,
2621 	.resume = blkfront_resume,
2622 	.otherend_changed = blkback_changed,
2623 	.is_ready = blkfront_is_ready,
2624 };
2625 
2626 static int __init xlblk_init(void)
2627 {
2628 	int ret;
2629 	int nr_cpus = num_online_cpus();
2630 
2631 	if (!xen_domain())
2632 		return -ENODEV;
2633 
2634 	if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2635 		xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2636 
2637 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2638 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2639 			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2640 		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2641 	}
2642 
2643 	if (xen_blkif_max_queues > nr_cpus) {
2644 		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2645 			xen_blkif_max_queues, nr_cpus);
2646 		xen_blkif_max_queues = nr_cpus;
2647 	}
2648 
2649 	if (!xen_has_pv_disk_devices())
2650 		return -ENODEV;
2651 
2652 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2653 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2654 		       XENVBD_MAJOR, DEV_NAME);
2655 		return -ENODEV;
2656 	}
2657 
2658 	ret = xenbus_register_frontend(&blkfront_driver);
2659 	if (ret) {
2660 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2661 		return ret;
2662 	}
2663 
2664 	return 0;
2665 }
2666 module_init(xlblk_init);
2667 
2668 
2669 static void __exit xlblk_exit(void)
2670 {
2671 	xenbus_unregister_driver(&blkfront_driver);
2672 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2673 	kfree(minors);
2674 }
2675 module_exit(xlblk_exit);
2676 
2677 MODULE_DESCRIPTION("Xen virtual block device frontend");
2678 MODULE_LICENSE("GPL");
2679 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2680 MODULE_ALIAS("xen:vbd");
2681 MODULE_ALIAS("xenblk");
2682