xref: /openbmc/linux/drivers/block/xen-blkfront.c (revision 3821a065)
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 enum blkif_state {
64 	BLKIF_STATE_DISCONNECTED,
65 	BLKIF_STATE_CONNECTED,
66 	BLKIF_STATE_SUSPENDED,
67 };
68 
69 struct grant {
70 	grant_ref_t gref;
71 	unsigned long pfn;
72 	struct list_head node;
73 };
74 
75 struct blk_shadow {
76 	struct blkif_request req;
77 	struct request *request;
78 	struct grant **grants_used;
79 	struct grant **indirect_grants;
80 	struct scatterlist *sg;
81 };
82 
83 struct split_bio {
84 	struct bio *bio;
85 	atomic_t pending;
86 };
87 
88 static DEFINE_MUTEX(blkfront_mutex);
89 static const struct block_device_operations xlvbd_block_fops;
90 
91 /*
92  * Maximum number of segments in indirect requests, the actual value used by
93  * the frontend driver is the minimum of this value and the value provided
94  * by the backend driver.
95  */
96 
97 static unsigned int xen_blkif_max_segments = 32;
98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
100 
101 /*
102  * Maximum order of pages to be used for the shared ring between front and
103  * backend, 4KB page granularity is used.
104  */
105 static unsigned int xen_blkif_max_ring_order;
106 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
107 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
108 
109 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages)
110 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
111 /*
112  * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
113  * characters are enough. Define to 20 to keep consist with backend.
114  */
115 #define RINGREF_NAME_LEN (20)
116 
117 /*
118  * We have one of these per vbd, whether ide, scsi or 'other'.  They
119  * hang in private_data off the gendisk structure. We may end up
120  * putting all kinds of interesting stuff here :-)
121  */
122 struct blkfront_info
123 {
124 	spinlock_t io_lock;
125 	struct mutex mutex;
126 	struct xenbus_device *xbdev;
127 	struct gendisk *gd;
128 	int vdevice;
129 	blkif_vdev_t handle;
130 	enum blkif_state connected;
131 	int ring_ref[XENBUS_MAX_RING_PAGES];
132 	unsigned int nr_ring_pages;
133 	struct blkif_front_ring ring;
134 	unsigned int evtchn, irq;
135 	struct request_queue *rq;
136 	struct work_struct work;
137 	struct gnttab_free_callback callback;
138 	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
139 	struct list_head grants;
140 	struct list_head indirect_pages;
141 	unsigned int persistent_gnts_c;
142 	unsigned long shadow_free;
143 	unsigned int feature_flush;
144 	unsigned int feature_discard:1;
145 	unsigned int feature_secdiscard:1;
146 	unsigned int discard_granularity;
147 	unsigned int discard_alignment;
148 	unsigned int feature_persistent:1;
149 	unsigned int max_indirect_segments;
150 	int is_ready;
151 	struct blk_mq_tag_set tag_set;
152 };
153 
154 static unsigned int nr_minors;
155 static unsigned long *minors;
156 static DEFINE_SPINLOCK(minor_lock);
157 
158 #define GRANT_INVALID_REF	0
159 
160 #define PARTS_PER_DISK		16
161 #define PARTS_PER_EXT_DISK      256
162 
163 #define BLKIF_MAJOR(dev) ((dev)>>8)
164 #define BLKIF_MINOR(dev) ((dev) & 0xff)
165 
166 #define EXT_SHIFT 28
167 #define EXTENDED (1<<EXT_SHIFT)
168 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
169 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
170 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
171 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
172 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
173 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
174 
175 #define DEV_NAME	"xvd"	/* name in /dev */
176 
177 #define SEGS_PER_INDIRECT_FRAME \
178 	(PAGE_SIZE/sizeof(struct blkif_request_segment))
179 #define INDIRECT_GREFS(_segs) \
180 	((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
181 
182 static int blkfront_setup_indirect(struct blkfront_info *info);
183 static int blkfront_gather_backend_features(struct blkfront_info *info);
184 
185 static int get_id_from_freelist(struct blkfront_info *info)
186 {
187 	unsigned long free = info->shadow_free;
188 	BUG_ON(free >= BLK_RING_SIZE(info));
189 	info->shadow_free = info->shadow[free].req.u.rw.id;
190 	info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
191 	return free;
192 }
193 
194 static int add_id_to_freelist(struct blkfront_info *info,
195 			       unsigned long id)
196 {
197 	if (info->shadow[id].req.u.rw.id != id)
198 		return -EINVAL;
199 	if (info->shadow[id].request == NULL)
200 		return -EINVAL;
201 	info->shadow[id].req.u.rw.id  = info->shadow_free;
202 	info->shadow[id].request = NULL;
203 	info->shadow_free = id;
204 	return 0;
205 }
206 
207 static int fill_grant_buffer(struct blkfront_info *info, int num)
208 {
209 	struct page *granted_page;
210 	struct grant *gnt_list_entry, *n;
211 	int i = 0;
212 
213 	while(i < num) {
214 		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
215 		if (!gnt_list_entry)
216 			goto out_of_memory;
217 
218 		if (info->feature_persistent) {
219 			granted_page = alloc_page(GFP_NOIO);
220 			if (!granted_page) {
221 				kfree(gnt_list_entry);
222 				goto out_of_memory;
223 			}
224 			gnt_list_entry->pfn = page_to_pfn(granted_page);
225 		}
226 
227 		gnt_list_entry->gref = GRANT_INVALID_REF;
228 		list_add(&gnt_list_entry->node, &info->grants);
229 		i++;
230 	}
231 
232 	return 0;
233 
234 out_of_memory:
235 	list_for_each_entry_safe(gnt_list_entry, n,
236 	                         &info->grants, node) {
237 		list_del(&gnt_list_entry->node);
238 		if (info->feature_persistent)
239 			__free_page(pfn_to_page(gnt_list_entry->pfn));
240 		kfree(gnt_list_entry);
241 		i--;
242 	}
243 	BUG_ON(i != 0);
244 	return -ENOMEM;
245 }
246 
247 static struct grant *get_grant(grant_ref_t *gref_head,
248                                unsigned long pfn,
249                                struct blkfront_info *info)
250 {
251 	struct grant *gnt_list_entry;
252 	unsigned long buffer_gfn;
253 
254 	BUG_ON(list_empty(&info->grants));
255 	gnt_list_entry = list_first_entry(&info->grants, struct grant,
256 	                                  node);
257 	list_del(&gnt_list_entry->node);
258 
259 	if (gnt_list_entry->gref != GRANT_INVALID_REF) {
260 		info->persistent_gnts_c--;
261 		return gnt_list_entry;
262 	}
263 
264 	/* Assign a gref to this page */
265 	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
266 	BUG_ON(gnt_list_entry->gref == -ENOSPC);
267 	if (!info->feature_persistent) {
268 		BUG_ON(!pfn);
269 		gnt_list_entry->pfn = pfn;
270 	}
271 	buffer_gfn = pfn_to_gfn(gnt_list_entry->pfn);
272 	gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
273 	                                info->xbdev->otherend_id,
274 	                                buffer_gfn, 0);
275 	return gnt_list_entry;
276 }
277 
278 static const char *op_name(int op)
279 {
280 	static const char *const names[] = {
281 		[BLKIF_OP_READ] = "read",
282 		[BLKIF_OP_WRITE] = "write",
283 		[BLKIF_OP_WRITE_BARRIER] = "barrier",
284 		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
285 		[BLKIF_OP_DISCARD] = "discard" };
286 
287 	if (op < 0 || op >= ARRAY_SIZE(names))
288 		return "unknown";
289 
290 	if (!names[op])
291 		return "reserved";
292 
293 	return names[op];
294 }
295 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
296 {
297 	unsigned int end = minor + nr;
298 	int rc;
299 
300 	if (end > nr_minors) {
301 		unsigned long *bitmap, *old;
302 
303 		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
304 				 GFP_KERNEL);
305 		if (bitmap == NULL)
306 			return -ENOMEM;
307 
308 		spin_lock(&minor_lock);
309 		if (end > nr_minors) {
310 			old = minors;
311 			memcpy(bitmap, minors,
312 			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
313 			minors = bitmap;
314 			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
315 		} else
316 			old = bitmap;
317 		spin_unlock(&minor_lock);
318 		kfree(old);
319 	}
320 
321 	spin_lock(&minor_lock);
322 	if (find_next_bit(minors, end, minor) >= end) {
323 		bitmap_set(minors, minor, nr);
324 		rc = 0;
325 	} else
326 		rc = -EBUSY;
327 	spin_unlock(&minor_lock);
328 
329 	return rc;
330 }
331 
332 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
333 {
334 	unsigned int end = minor + nr;
335 
336 	BUG_ON(end > nr_minors);
337 	spin_lock(&minor_lock);
338 	bitmap_clear(minors,  minor, nr);
339 	spin_unlock(&minor_lock);
340 }
341 
342 static void blkif_restart_queue_callback(void *arg)
343 {
344 	struct blkfront_info *info = (struct blkfront_info *)arg;
345 	schedule_work(&info->work);
346 }
347 
348 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
349 {
350 	/* We don't have real geometry info, but let's at least return
351 	   values consistent with the size of the device */
352 	sector_t nsect = get_capacity(bd->bd_disk);
353 	sector_t cylinders = nsect;
354 
355 	hg->heads = 0xff;
356 	hg->sectors = 0x3f;
357 	sector_div(cylinders, hg->heads * hg->sectors);
358 	hg->cylinders = cylinders;
359 	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
360 		hg->cylinders = 0xffff;
361 	return 0;
362 }
363 
364 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
365 		       unsigned command, unsigned long argument)
366 {
367 	struct blkfront_info *info = bdev->bd_disk->private_data;
368 	int i;
369 
370 	dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
371 		command, (long)argument);
372 
373 	switch (command) {
374 	case CDROMMULTISESSION:
375 		dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
376 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
377 			if (put_user(0, (char __user *)(argument + i)))
378 				return -EFAULT;
379 		return 0;
380 
381 	case CDROM_GET_CAPABILITY: {
382 		struct gendisk *gd = info->gd;
383 		if (gd->flags & GENHD_FL_CD)
384 			return 0;
385 		return -EINVAL;
386 	}
387 
388 	default:
389 		/*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
390 		  command);*/
391 		return -EINVAL; /* same return as native Linux */
392 	}
393 
394 	return 0;
395 }
396 
397 /*
398  * Generate a Xen blkfront IO request from a blk layer request.  Reads
399  * and writes are handled as expected.
400  *
401  * @req: a request struct
402  */
403 static int blkif_queue_request(struct request *req)
404 {
405 	struct blkfront_info *info = req->rq_disk->private_data;
406 	struct blkif_request *ring_req;
407 	unsigned long id;
408 	unsigned int fsect, lsect;
409 	int i, ref, n;
410 	struct blkif_request_segment *segments = NULL;
411 
412 	/*
413 	 * Used to store if we are able to queue the request by just using
414 	 * existing persistent grants, or if we have to get new grants,
415 	 * as there are not sufficiently many free.
416 	 */
417 	bool new_persistent_gnts;
418 	grant_ref_t gref_head;
419 	struct grant *gnt_list_entry = NULL;
420 	struct scatterlist *sg;
421 	int nseg, max_grefs;
422 
423 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
424 		return 1;
425 
426 	max_grefs = req->nr_phys_segments;
427 	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
428 		/*
429 		 * If we are using indirect segments we need to account
430 		 * for the indirect grefs used in the request.
431 		 */
432 		max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
433 
434 	/* Check if we have enough grants to allocate a requests */
435 	if (info->persistent_gnts_c < max_grefs) {
436 		new_persistent_gnts = 1;
437 		if (gnttab_alloc_grant_references(
438 		    max_grefs - info->persistent_gnts_c,
439 		    &gref_head) < 0) {
440 			gnttab_request_free_callback(
441 				&info->callback,
442 				blkif_restart_queue_callback,
443 				info,
444 				max_grefs);
445 			return 1;
446 		}
447 	} else
448 		new_persistent_gnts = 0;
449 
450 	/* Fill out a communications ring structure. */
451 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
452 	id = get_id_from_freelist(info);
453 	info->shadow[id].request = req;
454 
455 	if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
456 		ring_req->operation = BLKIF_OP_DISCARD;
457 		ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
458 		ring_req->u.discard.id = id;
459 		ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
460 		if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
461 			ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
462 		else
463 			ring_req->u.discard.flag = 0;
464 	} else {
465 		BUG_ON(info->max_indirect_segments == 0 &&
466 		       req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
467 		BUG_ON(info->max_indirect_segments &&
468 		       req->nr_phys_segments > info->max_indirect_segments);
469 		nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
470 		ring_req->u.rw.id = id;
471 		if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
472 			/*
473 			 * The indirect operation can only be a BLKIF_OP_READ or
474 			 * BLKIF_OP_WRITE
475 			 */
476 			BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
477 			ring_req->operation = BLKIF_OP_INDIRECT;
478 			ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
479 				BLKIF_OP_WRITE : BLKIF_OP_READ;
480 			ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
481 			ring_req->u.indirect.handle = info->handle;
482 			ring_req->u.indirect.nr_segments = nseg;
483 		} else {
484 			ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
485 			ring_req->u.rw.handle = info->handle;
486 			ring_req->operation = rq_data_dir(req) ?
487 				BLKIF_OP_WRITE : BLKIF_OP_READ;
488 			if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
489 				/*
490 				 * Ideally we can do an unordered flush-to-disk. In case the
491 				 * backend onlysupports barriers, use that. A barrier request
492 				 * a superset of FUA, so we can implement it the same
493 				 * way.  (It's also a FLUSH+FUA, since it is
494 				 * guaranteed ordered WRT previous writes.)
495 				 */
496 				switch (info->feature_flush &
497 					((REQ_FLUSH|REQ_FUA))) {
498 				case REQ_FLUSH|REQ_FUA:
499 					ring_req->operation =
500 						BLKIF_OP_WRITE_BARRIER;
501 					break;
502 				case REQ_FLUSH:
503 					ring_req->operation =
504 						BLKIF_OP_FLUSH_DISKCACHE;
505 					break;
506 				default:
507 					ring_req->operation = 0;
508 				}
509 			}
510 			ring_req->u.rw.nr_segments = nseg;
511 		}
512 		for_each_sg(info->shadow[id].sg, sg, nseg, i) {
513 			fsect = sg->offset >> 9;
514 			lsect = fsect + (sg->length >> 9) - 1;
515 
516 			if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
517 			    (i % SEGS_PER_INDIRECT_FRAME == 0)) {
518 				unsigned long uninitialized_var(pfn);
519 
520 				if (segments)
521 					kunmap_atomic(segments);
522 
523 				n = i / SEGS_PER_INDIRECT_FRAME;
524 				if (!info->feature_persistent) {
525 					struct page *indirect_page;
526 
527 					/* Fetch a pre-allocated page to use for indirect grefs */
528 					BUG_ON(list_empty(&info->indirect_pages));
529 					indirect_page = list_first_entry(&info->indirect_pages,
530 					                                 struct page, lru);
531 					list_del(&indirect_page->lru);
532 					pfn = page_to_pfn(indirect_page);
533 				}
534 				gnt_list_entry = get_grant(&gref_head, pfn, info);
535 				info->shadow[id].indirect_grants[n] = gnt_list_entry;
536 				segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
537 				ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
538 			}
539 
540 			gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
541 			ref = gnt_list_entry->gref;
542 
543 			info->shadow[id].grants_used[i] = gnt_list_entry;
544 
545 			if (rq_data_dir(req) && info->feature_persistent) {
546 				char *bvec_data;
547 				void *shared_data;
548 
549 				BUG_ON(sg->offset + sg->length > PAGE_SIZE);
550 
551 				shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
552 				bvec_data = kmap_atomic(sg_page(sg));
553 
554 				/*
555 				 * this does not wipe data stored outside the
556 				 * range sg->offset..sg->offset+sg->length.
557 				 * Therefore, blkback *could* see data from
558 				 * previous requests. This is OK as long as
559 				 * persistent grants are shared with just one
560 				 * domain. It may need refactoring if this
561 				 * changes
562 				 */
563 				memcpy(shared_data + sg->offset,
564 				       bvec_data   + sg->offset,
565 				       sg->length);
566 
567 				kunmap_atomic(bvec_data);
568 				kunmap_atomic(shared_data);
569 			}
570 			if (ring_req->operation != BLKIF_OP_INDIRECT) {
571 				ring_req->u.rw.seg[i] =
572 						(struct blkif_request_segment) {
573 							.gref       = ref,
574 							.first_sect = fsect,
575 							.last_sect  = lsect };
576 			} else {
577 				n = i % SEGS_PER_INDIRECT_FRAME;
578 				segments[n] =
579 					(struct blkif_request_segment) {
580 							.gref       = ref,
581 							.first_sect = fsect,
582 							.last_sect  = lsect };
583 			}
584 		}
585 		if (segments)
586 			kunmap_atomic(segments);
587 	}
588 
589 	info->ring.req_prod_pvt++;
590 
591 	/* Keep a private copy so we can reissue requests when recovering. */
592 	info->shadow[id].req = *ring_req;
593 
594 	if (new_persistent_gnts)
595 		gnttab_free_grant_references(gref_head);
596 
597 	return 0;
598 }
599 
600 
601 static inline void flush_requests(struct blkfront_info *info)
602 {
603 	int notify;
604 
605 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
606 
607 	if (notify)
608 		notify_remote_via_irq(info->irq);
609 }
610 
611 static inline bool blkif_request_flush_invalid(struct request *req,
612 					       struct blkfront_info *info)
613 {
614 	return ((req->cmd_type != REQ_TYPE_FS) ||
615 		((req->cmd_flags & REQ_FLUSH) &&
616 		 !(info->feature_flush & REQ_FLUSH)) ||
617 		((req->cmd_flags & REQ_FUA) &&
618 		 !(info->feature_flush & REQ_FUA)));
619 }
620 
621 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
622 			   const struct blk_mq_queue_data *qd)
623 {
624 	struct blkfront_info *info = qd->rq->rq_disk->private_data;
625 
626 	blk_mq_start_request(qd->rq);
627 	spin_lock_irq(&info->io_lock);
628 	if (RING_FULL(&info->ring))
629 		goto out_busy;
630 
631 	if (blkif_request_flush_invalid(qd->rq, info))
632 		goto out_err;
633 
634 	if (blkif_queue_request(qd->rq))
635 		goto out_busy;
636 
637 	flush_requests(info);
638 	spin_unlock_irq(&info->io_lock);
639 	return BLK_MQ_RQ_QUEUE_OK;
640 
641 out_err:
642 	spin_unlock_irq(&info->io_lock);
643 	return BLK_MQ_RQ_QUEUE_ERROR;
644 
645 out_busy:
646 	spin_unlock_irq(&info->io_lock);
647 	blk_mq_stop_hw_queue(hctx);
648 	return BLK_MQ_RQ_QUEUE_BUSY;
649 }
650 
651 static struct blk_mq_ops blkfront_mq_ops = {
652 	.queue_rq = blkif_queue_rq,
653 	.map_queue = blk_mq_map_queue,
654 };
655 
656 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
657 				unsigned int physical_sector_size,
658 				unsigned int segments)
659 {
660 	struct request_queue *rq;
661 	struct blkfront_info *info = gd->private_data;
662 
663 	memset(&info->tag_set, 0, sizeof(info->tag_set));
664 	info->tag_set.ops = &blkfront_mq_ops;
665 	info->tag_set.nr_hw_queues = 1;
666 	info->tag_set.queue_depth =  BLK_RING_SIZE(info);
667 	info->tag_set.numa_node = NUMA_NO_NODE;
668 	info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
669 	info->tag_set.cmd_size = 0;
670 	info->tag_set.driver_data = info;
671 
672 	if (blk_mq_alloc_tag_set(&info->tag_set))
673 		return -1;
674 	rq = blk_mq_init_queue(&info->tag_set);
675 	if (IS_ERR(rq)) {
676 		blk_mq_free_tag_set(&info->tag_set);
677 		return -1;
678 	}
679 
680 	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
681 
682 	if (info->feature_discard) {
683 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
684 		blk_queue_max_discard_sectors(rq, get_capacity(gd));
685 		rq->limits.discard_granularity = info->discard_granularity;
686 		rq->limits.discard_alignment = info->discard_alignment;
687 		if (info->feature_secdiscard)
688 			queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
689 	}
690 
691 	/* Hard sector size and max sectors impersonate the equiv. hardware. */
692 	blk_queue_logical_block_size(rq, sector_size);
693 	blk_queue_physical_block_size(rq, physical_sector_size);
694 	blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
695 
696 	/* Each segment in a request is up to an aligned page in size. */
697 	blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
698 	blk_queue_max_segment_size(rq, PAGE_SIZE);
699 
700 	/* Ensure a merged request will fit in a single I/O ring slot. */
701 	blk_queue_max_segments(rq, segments);
702 
703 	/* Make sure buffer addresses are sector-aligned. */
704 	blk_queue_dma_alignment(rq, 511);
705 
706 	/* Make sure we don't use bounce buffers. */
707 	blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
708 
709 	gd->queue = rq;
710 
711 	return 0;
712 }
713 
714 static const char *flush_info(unsigned int feature_flush)
715 {
716 	switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
717 	case REQ_FLUSH|REQ_FUA:
718 		return "barrier: enabled;";
719 	case REQ_FLUSH:
720 		return "flush diskcache: enabled;";
721 	default:
722 		return "barrier or flush: disabled;";
723 	}
724 }
725 
726 static void xlvbd_flush(struct blkfront_info *info)
727 {
728 	blk_queue_flush(info->rq, info->feature_flush);
729 	pr_info("blkfront: %s: %s %s %s %s %s\n",
730 		info->gd->disk_name, flush_info(info->feature_flush),
731 		"persistent grants:", info->feature_persistent ?
732 		"enabled;" : "disabled;", "indirect descriptors:",
733 		info->max_indirect_segments ? "enabled;" : "disabled;");
734 }
735 
736 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
737 {
738 	int major;
739 	major = BLKIF_MAJOR(vdevice);
740 	*minor = BLKIF_MINOR(vdevice);
741 	switch (major) {
742 		case XEN_IDE0_MAJOR:
743 			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
744 			*minor = ((*minor / 64) * PARTS_PER_DISK) +
745 				EMULATED_HD_DISK_MINOR_OFFSET;
746 			break;
747 		case XEN_IDE1_MAJOR:
748 			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
749 			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
750 				EMULATED_HD_DISK_MINOR_OFFSET;
751 			break;
752 		case XEN_SCSI_DISK0_MAJOR:
753 			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
754 			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
755 			break;
756 		case XEN_SCSI_DISK1_MAJOR:
757 		case XEN_SCSI_DISK2_MAJOR:
758 		case XEN_SCSI_DISK3_MAJOR:
759 		case XEN_SCSI_DISK4_MAJOR:
760 		case XEN_SCSI_DISK5_MAJOR:
761 		case XEN_SCSI_DISK6_MAJOR:
762 		case XEN_SCSI_DISK7_MAJOR:
763 			*offset = (*minor / PARTS_PER_DISK) +
764 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
765 				EMULATED_SD_DISK_NAME_OFFSET;
766 			*minor = *minor +
767 				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
768 				EMULATED_SD_DISK_MINOR_OFFSET;
769 			break;
770 		case XEN_SCSI_DISK8_MAJOR:
771 		case XEN_SCSI_DISK9_MAJOR:
772 		case XEN_SCSI_DISK10_MAJOR:
773 		case XEN_SCSI_DISK11_MAJOR:
774 		case XEN_SCSI_DISK12_MAJOR:
775 		case XEN_SCSI_DISK13_MAJOR:
776 		case XEN_SCSI_DISK14_MAJOR:
777 		case XEN_SCSI_DISK15_MAJOR:
778 			*offset = (*minor / PARTS_PER_DISK) +
779 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
780 				EMULATED_SD_DISK_NAME_OFFSET;
781 			*minor = *minor +
782 				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
783 				EMULATED_SD_DISK_MINOR_OFFSET;
784 			break;
785 		case XENVBD_MAJOR:
786 			*offset = *minor / PARTS_PER_DISK;
787 			break;
788 		default:
789 			printk(KERN_WARNING "blkfront: your disk configuration is "
790 					"incorrect, please use an xvd device instead\n");
791 			return -ENODEV;
792 	}
793 	return 0;
794 }
795 
796 static char *encode_disk_name(char *ptr, unsigned int n)
797 {
798 	if (n >= 26)
799 		ptr = encode_disk_name(ptr, n / 26 - 1);
800 	*ptr = 'a' + n % 26;
801 	return ptr + 1;
802 }
803 
804 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
805 			       struct blkfront_info *info,
806 			       u16 vdisk_info, u16 sector_size,
807 			       unsigned int physical_sector_size)
808 {
809 	struct gendisk *gd;
810 	int nr_minors = 1;
811 	int err;
812 	unsigned int offset;
813 	int minor;
814 	int nr_parts;
815 	char *ptr;
816 
817 	BUG_ON(info->gd != NULL);
818 	BUG_ON(info->rq != NULL);
819 
820 	if ((info->vdevice>>EXT_SHIFT) > 1) {
821 		/* this is above the extended range; something is wrong */
822 		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
823 		return -ENODEV;
824 	}
825 
826 	if (!VDEV_IS_EXTENDED(info->vdevice)) {
827 		err = xen_translate_vdev(info->vdevice, &minor, &offset);
828 		if (err)
829 			return err;
830  		nr_parts = PARTS_PER_DISK;
831 	} else {
832 		minor = BLKIF_MINOR_EXT(info->vdevice);
833 		nr_parts = PARTS_PER_EXT_DISK;
834 		offset = minor / nr_parts;
835 		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
836 			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
837 					"emulated IDE disks,\n\t choose an xvd device name"
838 					"from xvde on\n", info->vdevice);
839 	}
840 	if (minor >> MINORBITS) {
841 		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
842 			info->vdevice, minor);
843 		return -ENODEV;
844 	}
845 
846 	if ((minor % nr_parts) == 0)
847 		nr_minors = nr_parts;
848 
849 	err = xlbd_reserve_minors(minor, nr_minors);
850 	if (err)
851 		goto out;
852 	err = -ENODEV;
853 
854 	gd = alloc_disk(nr_minors);
855 	if (gd == NULL)
856 		goto release;
857 
858 	strcpy(gd->disk_name, DEV_NAME);
859 	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
860 	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
861 	if (nr_minors > 1)
862 		*ptr = 0;
863 	else
864 		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
865 			 "%d", minor & (nr_parts - 1));
866 
867 	gd->major = XENVBD_MAJOR;
868 	gd->first_minor = minor;
869 	gd->fops = &xlvbd_block_fops;
870 	gd->private_data = info;
871 	gd->driverfs_dev = &(info->xbdev->dev);
872 	set_capacity(gd, capacity);
873 
874 	if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
875 				 info->max_indirect_segments ? :
876 				 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
877 		del_gendisk(gd);
878 		goto release;
879 	}
880 
881 	info->rq = gd->queue;
882 	info->gd = gd;
883 
884 	xlvbd_flush(info);
885 
886 	if (vdisk_info & VDISK_READONLY)
887 		set_disk_ro(gd, 1);
888 
889 	if (vdisk_info & VDISK_REMOVABLE)
890 		gd->flags |= GENHD_FL_REMOVABLE;
891 
892 	if (vdisk_info & VDISK_CDROM)
893 		gd->flags |= GENHD_FL_CD;
894 
895 	return 0;
896 
897  release:
898 	xlbd_release_minors(minor, nr_minors);
899  out:
900 	return err;
901 }
902 
903 static void xlvbd_release_gendisk(struct blkfront_info *info)
904 {
905 	unsigned int minor, nr_minors;
906 
907 	if (info->rq == NULL)
908 		return;
909 
910 	/* No more blkif_request(). */
911 	blk_mq_stop_hw_queues(info->rq);
912 
913 	/* No more gnttab callback work. */
914 	gnttab_cancel_free_callback(&info->callback);
915 
916 	/* Flush gnttab callback work. Must be done with no locks held. */
917 	flush_work(&info->work);
918 
919 	del_gendisk(info->gd);
920 
921 	minor = info->gd->first_minor;
922 	nr_minors = info->gd->minors;
923 	xlbd_release_minors(minor, nr_minors);
924 
925 	blk_cleanup_queue(info->rq);
926 	blk_mq_free_tag_set(&info->tag_set);
927 	info->rq = NULL;
928 
929 	put_disk(info->gd);
930 	info->gd = NULL;
931 }
932 
933 /* Must be called with io_lock holded */
934 static void kick_pending_request_queues(struct blkfront_info *info)
935 {
936 	if (!RING_FULL(&info->ring))
937 		blk_mq_start_stopped_hw_queues(info->rq, true);
938 }
939 
940 static void blkif_restart_queue(struct work_struct *work)
941 {
942 	struct blkfront_info *info = container_of(work, struct blkfront_info, work);
943 
944 	spin_lock_irq(&info->io_lock);
945 	if (info->connected == BLKIF_STATE_CONNECTED)
946 		kick_pending_request_queues(info);
947 	spin_unlock_irq(&info->io_lock);
948 }
949 
950 static void blkif_free(struct blkfront_info *info, int suspend)
951 {
952 	struct grant *persistent_gnt;
953 	struct grant *n;
954 	int i, j, segs;
955 
956 	/* Prevent new requests being issued until we fix things up. */
957 	spin_lock_irq(&info->io_lock);
958 	info->connected = suspend ?
959 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
960 	/* No more blkif_request(). */
961 	if (info->rq)
962 		blk_mq_stop_hw_queues(info->rq);
963 
964 	/* Remove all persistent grants */
965 	if (!list_empty(&info->grants)) {
966 		list_for_each_entry_safe(persistent_gnt, n,
967 		                         &info->grants, node) {
968 			list_del(&persistent_gnt->node);
969 			if (persistent_gnt->gref != GRANT_INVALID_REF) {
970 				gnttab_end_foreign_access(persistent_gnt->gref,
971 				                          0, 0UL);
972 				info->persistent_gnts_c--;
973 			}
974 			if (info->feature_persistent)
975 				__free_page(pfn_to_page(persistent_gnt->pfn));
976 			kfree(persistent_gnt);
977 		}
978 	}
979 	BUG_ON(info->persistent_gnts_c != 0);
980 
981 	/*
982 	 * Remove indirect pages, this only happens when using indirect
983 	 * descriptors but not persistent grants
984 	 */
985 	if (!list_empty(&info->indirect_pages)) {
986 		struct page *indirect_page, *n;
987 
988 		BUG_ON(info->feature_persistent);
989 		list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
990 			list_del(&indirect_page->lru);
991 			__free_page(indirect_page);
992 		}
993 	}
994 
995 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
996 		/*
997 		 * Clear persistent grants present in requests already
998 		 * on the shared ring
999 		 */
1000 		if (!info->shadow[i].request)
1001 			goto free_shadow;
1002 
1003 		segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1004 		       info->shadow[i].req.u.indirect.nr_segments :
1005 		       info->shadow[i].req.u.rw.nr_segments;
1006 		for (j = 0; j < segs; j++) {
1007 			persistent_gnt = info->shadow[i].grants_used[j];
1008 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1009 			if (info->feature_persistent)
1010 				__free_page(pfn_to_page(persistent_gnt->pfn));
1011 			kfree(persistent_gnt);
1012 		}
1013 
1014 		if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1015 			/*
1016 			 * If this is not an indirect operation don't try to
1017 			 * free indirect segments
1018 			 */
1019 			goto free_shadow;
1020 
1021 		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1022 			persistent_gnt = info->shadow[i].indirect_grants[j];
1023 			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1024 			__free_page(pfn_to_page(persistent_gnt->pfn));
1025 			kfree(persistent_gnt);
1026 		}
1027 
1028 free_shadow:
1029 		kfree(info->shadow[i].grants_used);
1030 		info->shadow[i].grants_used = NULL;
1031 		kfree(info->shadow[i].indirect_grants);
1032 		info->shadow[i].indirect_grants = NULL;
1033 		kfree(info->shadow[i].sg);
1034 		info->shadow[i].sg = NULL;
1035 	}
1036 
1037 	/* No more gnttab callback work. */
1038 	gnttab_cancel_free_callback(&info->callback);
1039 	spin_unlock_irq(&info->io_lock);
1040 
1041 	/* Flush gnttab callback work. Must be done with no locks held. */
1042 	flush_work(&info->work);
1043 
1044 	/* Free resources associated with old device channel. */
1045 	for (i = 0; i < info->nr_ring_pages; i++) {
1046 		if (info->ring_ref[i] != GRANT_INVALID_REF) {
1047 			gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
1048 			info->ring_ref[i] = GRANT_INVALID_REF;
1049 		}
1050 	}
1051 	free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1052 	info->ring.sring = NULL;
1053 
1054 	if (info->irq)
1055 		unbind_from_irqhandler(info->irq, info);
1056 	info->evtchn = info->irq = 0;
1057 
1058 }
1059 
1060 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1061 			     struct blkif_response *bret)
1062 {
1063 	int i = 0;
1064 	struct scatterlist *sg;
1065 	char *bvec_data;
1066 	void *shared_data;
1067 	int nseg;
1068 
1069 	nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1070 		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1071 
1072 	if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1073 		for_each_sg(s->sg, sg, nseg, i) {
1074 			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1075 			shared_data = kmap_atomic(
1076 				pfn_to_page(s->grants_used[i]->pfn));
1077 			bvec_data = kmap_atomic(sg_page(sg));
1078 			memcpy(bvec_data   + sg->offset,
1079 			       shared_data + sg->offset,
1080 			       sg->length);
1081 			kunmap_atomic(bvec_data);
1082 			kunmap_atomic(shared_data);
1083 		}
1084 	}
1085 	/* Add the persistent grant into the list of free grants */
1086 	for (i = 0; i < nseg; i++) {
1087 		if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1088 			/*
1089 			 * If the grant is still mapped by the backend (the
1090 			 * backend has chosen to make this grant persistent)
1091 			 * we add it at the head of the list, so it will be
1092 			 * reused first.
1093 			 */
1094 			if (!info->feature_persistent)
1095 				pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1096 						     s->grants_used[i]->gref);
1097 			list_add(&s->grants_used[i]->node, &info->grants);
1098 			info->persistent_gnts_c++;
1099 		} else {
1100 			/*
1101 			 * If the grant is not mapped by the backend we end the
1102 			 * foreign access and add it to the tail of the list,
1103 			 * so it will not be picked again unless we run out of
1104 			 * persistent grants.
1105 			 */
1106 			gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1107 			s->grants_used[i]->gref = GRANT_INVALID_REF;
1108 			list_add_tail(&s->grants_used[i]->node, &info->grants);
1109 		}
1110 	}
1111 	if (s->req.operation == BLKIF_OP_INDIRECT) {
1112 		for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1113 			if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1114 				if (!info->feature_persistent)
1115 					pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1116 							     s->indirect_grants[i]->gref);
1117 				list_add(&s->indirect_grants[i]->node, &info->grants);
1118 				info->persistent_gnts_c++;
1119 			} else {
1120 				struct page *indirect_page;
1121 
1122 				gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1123 				/*
1124 				 * Add the used indirect page back to the list of
1125 				 * available pages for indirect grefs.
1126 				 */
1127 				if (!info->feature_persistent) {
1128 					indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1129 					list_add(&indirect_page->lru, &info->indirect_pages);
1130 				}
1131 				s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1132 				list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1133 			}
1134 		}
1135 	}
1136 }
1137 
1138 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1139 {
1140 	struct request *req;
1141 	struct blkif_response *bret;
1142 	RING_IDX i, rp;
1143 	unsigned long flags;
1144 	struct blkfront_info *info = (struct blkfront_info *)dev_id;
1145 
1146 	spin_lock_irqsave(&info->io_lock, flags);
1147 
1148 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1149 		spin_unlock_irqrestore(&info->io_lock, flags);
1150 		return IRQ_HANDLED;
1151 	}
1152 
1153  again:
1154 	rp = info->ring.sring->rsp_prod;
1155 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1156 
1157 	for (i = info->ring.rsp_cons; i != rp; i++) {
1158 		unsigned long id;
1159 
1160 		bret = RING_GET_RESPONSE(&info->ring, i);
1161 		id   = bret->id;
1162 		/*
1163 		 * The backend has messed up and given us an id that we would
1164 		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1165 		 * look in get_id_from_freelist.
1166 		 */
1167 		if (id >= BLK_RING_SIZE(info)) {
1168 			WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1169 			     info->gd->disk_name, op_name(bret->operation), id);
1170 			/* We can't safely get the 'struct request' as
1171 			 * the id is busted. */
1172 			continue;
1173 		}
1174 		req  = info->shadow[id].request;
1175 
1176 		if (bret->operation != BLKIF_OP_DISCARD)
1177 			blkif_completion(&info->shadow[id], info, bret);
1178 
1179 		if (add_id_to_freelist(info, id)) {
1180 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1181 			     info->gd->disk_name, op_name(bret->operation), id);
1182 			continue;
1183 		}
1184 
1185 		req->errors = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1186 		switch (bret->operation) {
1187 		case BLKIF_OP_DISCARD:
1188 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1189 				struct request_queue *rq = info->rq;
1190 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1191 					   info->gd->disk_name, op_name(bret->operation));
1192 				req->errors = -EOPNOTSUPP;
1193 				info->feature_discard = 0;
1194 				info->feature_secdiscard = 0;
1195 				queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1196 				queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1197 			}
1198 			blk_mq_complete_request(req);
1199 			break;
1200 		case BLKIF_OP_FLUSH_DISKCACHE:
1201 		case BLKIF_OP_WRITE_BARRIER:
1202 			if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1203 				printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1204 				       info->gd->disk_name, op_name(bret->operation));
1205 				req->errors = -EOPNOTSUPP;
1206 			}
1207 			if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1208 				     info->shadow[id].req.u.rw.nr_segments == 0)) {
1209 				printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1210 				       info->gd->disk_name, op_name(bret->operation));
1211 				req->errors = -EOPNOTSUPP;
1212 			}
1213 			if (unlikely(req->errors)) {
1214 				if (req->errors == -EOPNOTSUPP)
1215 					req->errors = 0;
1216 				info->feature_flush = 0;
1217 				xlvbd_flush(info);
1218 			}
1219 			/* fall through */
1220 		case BLKIF_OP_READ:
1221 		case BLKIF_OP_WRITE:
1222 			if (unlikely(bret->status != BLKIF_RSP_OKAY))
1223 				dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1224 					"request: %x\n", bret->status);
1225 
1226 			blk_mq_complete_request(req);
1227 			break;
1228 		default:
1229 			BUG();
1230 		}
1231 	}
1232 
1233 	info->ring.rsp_cons = i;
1234 
1235 	if (i != info->ring.req_prod_pvt) {
1236 		int more_to_do;
1237 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1238 		if (more_to_do)
1239 			goto again;
1240 	} else
1241 		info->ring.sring->rsp_event = i + 1;
1242 
1243 	kick_pending_request_queues(info);
1244 
1245 	spin_unlock_irqrestore(&info->io_lock, flags);
1246 
1247 	return IRQ_HANDLED;
1248 }
1249 
1250 
1251 static int setup_blkring(struct xenbus_device *dev,
1252 			 struct blkfront_info *info)
1253 {
1254 	struct blkif_sring *sring;
1255 	int err, i;
1256 	unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
1257 	grant_ref_t gref[XENBUS_MAX_RING_PAGES];
1258 
1259 	for (i = 0; i < info->nr_ring_pages; i++)
1260 		info->ring_ref[i] = GRANT_INVALID_REF;
1261 
1262 	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1263 						       get_order(ring_size));
1264 	if (!sring) {
1265 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1266 		return -ENOMEM;
1267 	}
1268 	SHARED_RING_INIT(sring);
1269 	FRONT_RING_INIT(&info->ring, sring, ring_size);
1270 
1271 	err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref);
1272 	if (err < 0) {
1273 		free_pages((unsigned long)sring, get_order(ring_size));
1274 		info->ring.sring = NULL;
1275 		goto fail;
1276 	}
1277 	for (i = 0; i < info->nr_ring_pages; i++)
1278 		info->ring_ref[i] = gref[i];
1279 
1280 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
1281 	if (err)
1282 		goto fail;
1283 
1284 	err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1285 					"blkif", info);
1286 	if (err <= 0) {
1287 		xenbus_dev_fatal(dev, err,
1288 				 "bind_evtchn_to_irqhandler failed");
1289 		goto fail;
1290 	}
1291 	info->irq = err;
1292 
1293 	return 0;
1294 fail:
1295 	blkif_free(info, 0);
1296 	return err;
1297 }
1298 
1299 
1300 /* Common code used when first setting up, and when resuming. */
1301 static int talk_to_blkback(struct xenbus_device *dev,
1302 			   struct blkfront_info *info)
1303 {
1304 	const char *message = NULL;
1305 	struct xenbus_transaction xbt;
1306 	int err, i;
1307 	unsigned int max_page_order = 0;
1308 	unsigned int ring_page_order = 0;
1309 
1310 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1311 			   "max-ring-page-order", "%u", &max_page_order);
1312 	if (err != 1)
1313 		info->nr_ring_pages = 1;
1314 	else {
1315 		ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1316 		info->nr_ring_pages = 1 << ring_page_order;
1317 	}
1318 
1319 	/* Create shared ring, alloc event channel. */
1320 	err = setup_blkring(dev, info);
1321 	if (err)
1322 		goto out;
1323 
1324 again:
1325 	err = xenbus_transaction_start(&xbt);
1326 	if (err) {
1327 		xenbus_dev_fatal(dev, err, "starting transaction");
1328 		goto destroy_blkring;
1329 	}
1330 
1331 	if (info->nr_ring_pages == 1) {
1332 		err = xenbus_printf(xbt, dev->nodename,
1333 				    "ring-ref", "%u", info->ring_ref[0]);
1334 		if (err) {
1335 			message = "writing ring-ref";
1336 			goto abort_transaction;
1337 		}
1338 	} else {
1339 		err = xenbus_printf(xbt, dev->nodename,
1340 				    "ring-page-order", "%u", ring_page_order);
1341 		if (err) {
1342 			message = "writing ring-page-order";
1343 			goto abort_transaction;
1344 		}
1345 
1346 		for (i = 0; i < info->nr_ring_pages; i++) {
1347 			char ring_ref_name[RINGREF_NAME_LEN];
1348 
1349 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1350 			err = xenbus_printf(xbt, dev->nodename, ring_ref_name,
1351 					    "%u", info->ring_ref[i]);
1352 			if (err) {
1353 				message = "writing ring-ref";
1354 				goto abort_transaction;
1355 			}
1356 		}
1357 	}
1358 	err = xenbus_printf(xbt, dev->nodename,
1359 			    "event-channel", "%u", info->evtchn);
1360 	if (err) {
1361 		message = "writing event-channel";
1362 		goto abort_transaction;
1363 	}
1364 	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1365 			    XEN_IO_PROTO_ABI_NATIVE);
1366 	if (err) {
1367 		message = "writing protocol";
1368 		goto abort_transaction;
1369 	}
1370 	err = xenbus_printf(xbt, dev->nodename,
1371 			    "feature-persistent", "%u", 1);
1372 	if (err)
1373 		dev_warn(&dev->dev,
1374 			 "writing persistent grants feature to xenbus");
1375 
1376 	err = xenbus_transaction_end(xbt, 0);
1377 	if (err) {
1378 		if (err == -EAGAIN)
1379 			goto again;
1380 		xenbus_dev_fatal(dev, err, "completing transaction");
1381 		goto destroy_blkring;
1382 	}
1383 
1384 	for (i = 0; i < BLK_RING_SIZE(info); i++)
1385 		info->shadow[i].req.u.rw.id = i+1;
1386 	info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1387 	xenbus_switch_state(dev, XenbusStateInitialised);
1388 
1389 	return 0;
1390 
1391  abort_transaction:
1392 	xenbus_transaction_end(xbt, 1);
1393 	if (message)
1394 		xenbus_dev_fatal(dev, err, "%s", message);
1395  destroy_blkring:
1396 	blkif_free(info, 0);
1397  out:
1398 	return err;
1399 }
1400 
1401 /**
1402  * Entry point to this code when a new device is created.  Allocate the basic
1403  * structures and the ring buffer for communication with the backend, and
1404  * inform the backend of the appropriate details for those.  Switch to
1405  * Initialised state.
1406  */
1407 static int blkfront_probe(struct xenbus_device *dev,
1408 			  const struct xenbus_device_id *id)
1409 {
1410 	int err, vdevice;
1411 	struct blkfront_info *info;
1412 
1413 	/* FIXME: Use dynamic device id if this is not set. */
1414 	err = xenbus_scanf(XBT_NIL, dev->nodename,
1415 			   "virtual-device", "%i", &vdevice);
1416 	if (err != 1) {
1417 		/* go looking in the extended area instead */
1418 		err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1419 				   "%i", &vdevice);
1420 		if (err != 1) {
1421 			xenbus_dev_fatal(dev, err, "reading virtual-device");
1422 			return err;
1423 		}
1424 	}
1425 
1426 	if (xen_hvm_domain()) {
1427 		char *type;
1428 		int len;
1429 		/* no unplug has been done: do not hook devices != xen vbds */
1430 		if (xen_has_pv_and_legacy_disk_devices()) {
1431 			int major;
1432 
1433 			if (!VDEV_IS_EXTENDED(vdevice))
1434 				major = BLKIF_MAJOR(vdevice);
1435 			else
1436 				major = XENVBD_MAJOR;
1437 
1438 			if (major != XENVBD_MAJOR) {
1439 				printk(KERN_INFO
1440 						"%s: HVM does not support vbd %d as xen block device\n",
1441 						__func__, vdevice);
1442 				return -ENODEV;
1443 			}
1444 		}
1445 		/* do not create a PV cdrom device if we are an HVM guest */
1446 		type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1447 		if (IS_ERR(type))
1448 			return -ENODEV;
1449 		if (strncmp(type, "cdrom", 5) == 0) {
1450 			kfree(type);
1451 			return -ENODEV;
1452 		}
1453 		kfree(type);
1454 	}
1455 	info = kzalloc(sizeof(*info), GFP_KERNEL);
1456 	if (!info) {
1457 		xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1458 		return -ENOMEM;
1459 	}
1460 
1461 	mutex_init(&info->mutex);
1462 	spin_lock_init(&info->io_lock);
1463 	info->xbdev = dev;
1464 	info->vdevice = vdevice;
1465 	INIT_LIST_HEAD(&info->grants);
1466 	INIT_LIST_HEAD(&info->indirect_pages);
1467 	info->persistent_gnts_c = 0;
1468 	info->connected = BLKIF_STATE_DISCONNECTED;
1469 	INIT_WORK(&info->work, blkif_restart_queue);
1470 
1471 	/* Front end dir is a number, which is used as the id. */
1472 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1473 	dev_set_drvdata(&dev->dev, info);
1474 
1475 	return 0;
1476 }
1477 
1478 static void split_bio_end(struct bio *bio)
1479 {
1480 	struct split_bio *split_bio = bio->bi_private;
1481 
1482 	if (atomic_dec_and_test(&split_bio->pending)) {
1483 		split_bio->bio->bi_phys_segments = 0;
1484 		split_bio->bio->bi_error = bio->bi_error;
1485 		bio_endio(split_bio->bio);
1486 		kfree(split_bio);
1487 	}
1488 	bio_put(bio);
1489 }
1490 
1491 static int blkif_recover(struct blkfront_info *info)
1492 {
1493 	int i;
1494 	struct request *req, *n;
1495 	struct blk_shadow *copy;
1496 	int rc;
1497 	struct bio *bio, *cloned_bio;
1498 	struct bio_list bio_list, merge_bio;
1499 	unsigned int segs, offset;
1500 	int pending, size;
1501 	struct split_bio *split_bio;
1502 	struct list_head requests;
1503 
1504 	/* Stage 1: Make a safe copy of the shadow state. */
1505 	copy = kmemdup(info->shadow, sizeof(info->shadow),
1506 		       GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1507 	if (!copy)
1508 		return -ENOMEM;
1509 
1510 	/* Stage 2: Set up free list. */
1511 	memset(&info->shadow, 0, sizeof(info->shadow));
1512 	for (i = 0; i < BLK_RING_SIZE(info); i++)
1513 		info->shadow[i].req.u.rw.id = i+1;
1514 	info->shadow_free = info->ring.req_prod_pvt;
1515 	info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1516 
1517 	rc = blkfront_gather_backend_features(info);
1518 	if (rc) {
1519 		kfree(copy);
1520 		return rc;
1521 	}
1522 
1523 	segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1524 	blk_queue_max_segments(info->rq, segs);
1525 	bio_list_init(&bio_list);
1526 	INIT_LIST_HEAD(&requests);
1527 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1528 		/* Not in use? */
1529 		if (!copy[i].request)
1530 			continue;
1531 
1532 		/*
1533 		 * Get the bios in the request so we can re-queue them.
1534 		 */
1535 		if (copy[i].request->cmd_flags &
1536 		    (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1537 			/*
1538 			 * Flush operations don't contain bios, so
1539 			 * we need to requeue the whole request
1540 			 */
1541 			list_add(&copy[i].request->queuelist, &requests);
1542 			continue;
1543 		}
1544 		merge_bio.head = copy[i].request->bio;
1545 		merge_bio.tail = copy[i].request->biotail;
1546 		bio_list_merge(&bio_list, &merge_bio);
1547 		copy[i].request->bio = NULL;
1548 		blk_end_request_all(copy[i].request, 0);
1549 	}
1550 
1551 	kfree(copy);
1552 
1553 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1554 
1555 	spin_lock_irq(&info->io_lock);
1556 
1557 	/* Now safe for us to use the shared ring */
1558 	info->connected = BLKIF_STATE_CONNECTED;
1559 
1560 	/* Kick any other new requests queued since we resumed */
1561 	kick_pending_request_queues(info);
1562 
1563 	list_for_each_entry_safe(req, n, &requests, queuelist) {
1564 		/* Requeue pending requests (flush or discard) */
1565 		list_del_init(&req->queuelist);
1566 		BUG_ON(req->nr_phys_segments > segs);
1567 		blk_mq_requeue_request(req);
1568 	}
1569 	spin_unlock_irq(&info->io_lock);
1570 	blk_mq_kick_requeue_list(info->rq);
1571 
1572 	while ((bio = bio_list_pop(&bio_list)) != NULL) {
1573 		/* Traverse the list of pending bios and re-queue them */
1574 		if (bio_segments(bio) > segs) {
1575 			/*
1576 			 * This bio has more segments than what we can
1577 			 * handle, we have to split it.
1578 			 */
1579 			pending = (bio_segments(bio) + segs - 1) / segs;
1580 			split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1581 			BUG_ON(split_bio == NULL);
1582 			atomic_set(&split_bio->pending, pending);
1583 			split_bio->bio = bio;
1584 			for (i = 0; i < pending; i++) {
1585 				offset = (i * segs * PAGE_SIZE) >> 9;
1586 				size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1587 					   (unsigned int)bio_sectors(bio) - offset);
1588 				cloned_bio = bio_clone(bio, GFP_NOIO);
1589 				BUG_ON(cloned_bio == NULL);
1590 				bio_trim(cloned_bio, offset, size);
1591 				cloned_bio->bi_private = split_bio;
1592 				cloned_bio->bi_end_io = split_bio_end;
1593 				submit_bio(cloned_bio->bi_rw, cloned_bio);
1594 			}
1595 			/*
1596 			 * Now we have to wait for all those smaller bios to
1597 			 * end, so we can also end the "parent" bio.
1598 			 */
1599 			continue;
1600 		}
1601 		/* We don't need to split this bio */
1602 		submit_bio(bio->bi_rw, bio);
1603 	}
1604 
1605 	return 0;
1606 }
1607 
1608 /**
1609  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1610  * driver restart.  We tear down our blkif structure and recreate it, but
1611  * leave the device-layer structures intact so that this is transparent to the
1612  * rest of the kernel.
1613  */
1614 static int blkfront_resume(struct xenbus_device *dev)
1615 {
1616 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1617 	int err;
1618 
1619 	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1620 
1621 	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1622 
1623 	err = talk_to_blkback(dev, info);
1624 
1625 	/*
1626 	 * We have to wait for the backend to switch to
1627 	 * connected state, since we want to read which
1628 	 * features it supports.
1629 	 */
1630 
1631 	return err;
1632 }
1633 
1634 static void
1635 blkfront_closing(struct blkfront_info *info)
1636 {
1637 	struct xenbus_device *xbdev = info->xbdev;
1638 	struct block_device *bdev = NULL;
1639 
1640 	mutex_lock(&info->mutex);
1641 
1642 	if (xbdev->state == XenbusStateClosing) {
1643 		mutex_unlock(&info->mutex);
1644 		return;
1645 	}
1646 
1647 	if (info->gd)
1648 		bdev = bdget_disk(info->gd, 0);
1649 
1650 	mutex_unlock(&info->mutex);
1651 
1652 	if (!bdev) {
1653 		xenbus_frontend_closed(xbdev);
1654 		return;
1655 	}
1656 
1657 	mutex_lock(&bdev->bd_mutex);
1658 
1659 	if (bdev->bd_openers) {
1660 		xenbus_dev_error(xbdev, -EBUSY,
1661 				 "Device in use; refusing to close");
1662 		xenbus_switch_state(xbdev, XenbusStateClosing);
1663 	} else {
1664 		xlvbd_release_gendisk(info);
1665 		xenbus_frontend_closed(xbdev);
1666 	}
1667 
1668 	mutex_unlock(&bdev->bd_mutex);
1669 	bdput(bdev);
1670 }
1671 
1672 static void blkfront_setup_discard(struct blkfront_info *info)
1673 {
1674 	int err;
1675 	unsigned int discard_granularity;
1676 	unsigned int discard_alignment;
1677 	unsigned int discard_secure;
1678 
1679 	info->feature_discard = 1;
1680 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1681 		"discard-granularity", "%u", &discard_granularity,
1682 		"discard-alignment", "%u", &discard_alignment,
1683 		NULL);
1684 	if (!err) {
1685 		info->discard_granularity = discard_granularity;
1686 		info->discard_alignment = discard_alignment;
1687 	}
1688 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1689 		    "discard-secure", "%d", &discard_secure,
1690 		    NULL);
1691 	if (!err)
1692 		info->feature_secdiscard = !!discard_secure;
1693 }
1694 
1695 static int blkfront_setup_indirect(struct blkfront_info *info)
1696 {
1697 	unsigned int segs;
1698 	int err, i;
1699 
1700 	if (info->max_indirect_segments == 0)
1701 		segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1702 	else
1703 		segs = info->max_indirect_segments;
1704 
1705 	err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE(info));
1706 	if (err)
1707 		goto out_of_memory;
1708 
1709 	if (!info->feature_persistent && info->max_indirect_segments) {
1710 		/*
1711 		 * We are using indirect descriptors but not persistent
1712 		 * grants, we need to allocate a set of pages that can be
1713 		 * used for mapping indirect grefs
1714 		 */
1715 		int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE(info);
1716 
1717 		BUG_ON(!list_empty(&info->indirect_pages));
1718 		for (i = 0; i < num; i++) {
1719 			struct page *indirect_page = alloc_page(GFP_NOIO);
1720 			if (!indirect_page)
1721 				goto out_of_memory;
1722 			list_add(&indirect_page->lru, &info->indirect_pages);
1723 		}
1724 	}
1725 
1726 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1727 		info->shadow[i].grants_used = kzalloc(
1728 			sizeof(info->shadow[i].grants_used[0]) * segs,
1729 			GFP_NOIO);
1730 		info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1731 		if (info->max_indirect_segments)
1732 			info->shadow[i].indirect_grants = kzalloc(
1733 				sizeof(info->shadow[i].indirect_grants[0]) *
1734 				INDIRECT_GREFS(segs),
1735 				GFP_NOIO);
1736 		if ((info->shadow[i].grants_used == NULL) ||
1737 			(info->shadow[i].sg == NULL) ||
1738 		     (info->max_indirect_segments &&
1739 		     (info->shadow[i].indirect_grants == NULL)))
1740 			goto out_of_memory;
1741 		sg_init_table(info->shadow[i].sg, segs);
1742 	}
1743 
1744 
1745 	return 0;
1746 
1747 out_of_memory:
1748 	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1749 		kfree(info->shadow[i].grants_used);
1750 		info->shadow[i].grants_used = NULL;
1751 		kfree(info->shadow[i].sg);
1752 		info->shadow[i].sg = NULL;
1753 		kfree(info->shadow[i].indirect_grants);
1754 		info->shadow[i].indirect_grants = NULL;
1755 	}
1756 	if (!list_empty(&info->indirect_pages)) {
1757 		struct page *indirect_page, *n;
1758 		list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1759 			list_del(&indirect_page->lru);
1760 			__free_page(indirect_page);
1761 		}
1762 	}
1763 	return -ENOMEM;
1764 }
1765 
1766 /*
1767  * Gather all backend feature-*
1768  */
1769 static int blkfront_gather_backend_features(struct blkfront_info *info)
1770 {
1771 	int err;
1772 	int barrier, flush, discard, persistent;
1773 	unsigned int indirect_segments;
1774 
1775 	info->feature_flush = 0;
1776 
1777 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1778 			"feature-barrier", "%d", &barrier,
1779 			NULL);
1780 
1781 	/*
1782 	 * If there's no "feature-barrier" defined, then it means
1783 	 * we're dealing with a very old backend which writes
1784 	 * synchronously; nothing to do.
1785 	 *
1786 	 * If there are barriers, then we use flush.
1787 	 */
1788 	if (!err && barrier)
1789 		info->feature_flush = REQ_FLUSH | REQ_FUA;
1790 	/*
1791 	 * And if there is "feature-flush-cache" use that above
1792 	 * barriers.
1793 	 */
1794 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1795 			"feature-flush-cache", "%d", &flush,
1796 			NULL);
1797 
1798 	if (!err && flush)
1799 		info->feature_flush = REQ_FLUSH;
1800 
1801 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1802 			"feature-discard", "%d", &discard,
1803 			NULL);
1804 
1805 	if (!err && discard)
1806 		blkfront_setup_discard(info);
1807 
1808 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1809 			"feature-persistent", "%u", &persistent,
1810 			NULL);
1811 	if (err)
1812 		info->feature_persistent = 0;
1813 	else
1814 		info->feature_persistent = persistent;
1815 
1816 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1817 			    "feature-max-indirect-segments", "%u", &indirect_segments,
1818 			    NULL);
1819 	if (err)
1820 		info->max_indirect_segments = 0;
1821 	else
1822 		info->max_indirect_segments = min(indirect_segments,
1823 						  xen_blkif_max_segments);
1824 
1825 	return blkfront_setup_indirect(info);
1826 }
1827 
1828 /*
1829  * Invoked when the backend is finally 'ready' (and has told produced
1830  * the details about the physical device - #sectors, size, etc).
1831  */
1832 static void blkfront_connect(struct blkfront_info *info)
1833 {
1834 	unsigned long long sectors;
1835 	unsigned long sector_size;
1836 	unsigned int physical_sector_size;
1837 	unsigned int binfo;
1838 	int err;
1839 
1840 	switch (info->connected) {
1841 	case BLKIF_STATE_CONNECTED:
1842 		/*
1843 		 * Potentially, the back-end may be signalling
1844 		 * a capacity change; update the capacity.
1845 		 */
1846 		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1847 				   "sectors", "%Lu", &sectors);
1848 		if (XENBUS_EXIST_ERR(err))
1849 			return;
1850 		printk(KERN_INFO "Setting capacity to %Lu\n",
1851 		       sectors);
1852 		set_capacity(info->gd, sectors);
1853 		revalidate_disk(info->gd);
1854 
1855 		return;
1856 	case BLKIF_STATE_SUSPENDED:
1857 		/*
1858 		 * If we are recovering from suspension, we need to wait
1859 		 * for the backend to announce it's features before
1860 		 * reconnecting, at least we need to know if the backend
1861 		 * supports indirect descriptors, and how many.
1862 		 */
1863 		blkif_recover(info);
1864 		return;
1865 
1866 	default:
1867 		break;
1868 	}
1869 
1870 	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1871 		__func__, info->xbdev->otherend);
1872 
1873 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1874 			    "sectors", "%llu", &sectors,
1875 			    "info", "%u", &binfo,
1876 			    "sector-size", "%lu", &sector_size,
1877 			    NULL);
1878 	if (err) {
1879 		xenbus_dev_fatal(info->xbdev, err,
1880 				 "reading backend fields at %s",
1881 				 info->xbdev->otherend);
1882 		return;
1883 	}
1884 
1885 	/*
1886 	 * physcial-sector-size is a newer field, so old backends may not
1887 	 * provide this. Assume physical sector size to be the same as
1888 	 * sector_size in that case.
1889 	 */
1890 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1891 			   "physical-sector-size", "%u", &physical_sector_size);
1892 	if (err != 1)
1893 		physical_sector_size = sector_size;
1894 
1895 	err = blkfront_gather_backend_features(info);
1896 	if (err) {
1897 		xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1898 				 info->xbdev->otherend);
1899 		return;
1900 	}
1901 
1902 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1903 				  physical_sector_size);
1904 	if (err) {
1905 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1906 				 info->xbdev->otherend);
1907 		return;
1908 	}
1909 
1910 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
1911 
1912 	/* Kick pending requests. */
1913 	spin_lock_irq(&info->io_lock);
1914 	info->connected = BLKIF_STATE_CONNECTED;
1915 	kick_pending_request_queues(info);
1916 	spin_unlock_irq(&info->io_lock);
1917 
1918 	add_disk(info->gd);
1919 
1920 	info->is_ready = 1;
1921 }
1922 
1923 /**
1924  * Callback received when the backend's state changes.
1925  */
1926 static void blkback_changed(struct xenbus_device *dev,
1927 			    enum xenbus_state backend_state)
1928 {
1929 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1930 
1931 	dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1932 
1933 	switch (backend_state) {
1934 	case XenbusStateInitWait:
1935 		if (dev->state != XenbusStateInitialising)
1936 			break;
1937 		if (talk_to_blkback(dev, info)) {
1938 			kfree(info);
1939 			dev_set_drvdata(&dev->dev, NULL);
1940 			break;
1941 		}
1942 	case XenbusStateInitialising:
1943 	case XenbusStateInitialised:
1944 	case XenbusStateReconfiguring:
1945 	case XenbusStateReconfigured:
1946 	case XenbusStateUnknown:
1947 		break;
1948 
1949 	case XenbusStateConnected:
1950 		blkfront_connect(info);
1951 		break;
1952 
1953 	case XenbusStateClosed:
1954 		if (dev->state == XenbusStateClosed)
1955 			break;
1956 		/* Missed the backend's Closing state -- fallthrough */
1957 	case XenbusStateClosing:
1958 		blkfront_closing(info);
1959 		break;
1960 	}
1961 }
1962 
1963 static int blkfront_remove(struct xenbus_device *xbdev)
1964 {
1965 	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1966 	struct block_device *bdev = NULL;
1967 	struct gendisk *disk;
1968 
1969 	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1970 
1971 	blkif_free(info, 0);
1972 
1973 	mutex_lock(&info->mutex);
1974 
1975 	disk = info->gd;
1976 	if (disk)
1977 		bdev = bdget_disk(disk, 0);
1978 
1979 	info->xbdev = NULL;
1980 	mutex_unlock(&info->mutex);
1981 
1982 	if (!bdev) {
1983 		kfree(info);
1984 		return 0;
1985 	}
1986 
1987 	/*
1988 	 * The xbdev was removed before we reached the Closed
1989 	 * state. See if it's safe to remove the disk. If the bdev
1990 	 * isn't closed yet, we let release take care of it.
1991 	 */
1992 
1993 	mutex_lock(&bdev->bd_mutex);
1994 	info = disk->private_data;
1995 
1996 	dev_warn(disk_to_dev(disk),
1997 		 "%s was hot-unplugged, %d stale handles\n",
1998 		 xbdev->nodename, bdev->bd_openers);
1999 
2000 	if (info && !bdev->bd_openers) {
2001 		xlvbd_release_gendisk(info);
2002 		disk->private_data = NULL;
2003 		kfree(info);
2004 	}
2005 
2006 	mutex_unlock(&bdev->bd_mutex);
2007 	bdput(bdev);
2008 
2009 	return 0;
2010 }
2011 
2012 static int blkfront_is_ready(struct xenbus_device *dev)
2013 {
2014 	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2015 
2016 	return info->is_ready && info->xbdev;
2017 }
2018 
2019 static int blkif_open(struct block_device *bdev, fmode_t mode)
2020 {
2021 	struct gendisk *disk = bdev->bd_disk;
2022 	struct blkfront_info *info;
2023 	int err = 0;
2024 
2025 	mutex_lock(&blkfront_mutex);
2026 
2027 	info = disk->private_data;
2028 	if (!info) {
2029 		/* xbdev gone */
2030 		err = -ERESTARTSYS;
2031 		goto out;
2032 	}
2033 
2034 	mutex_lock(&info->mutex);
2035 
2036 	if (!info->gd)
2037 		/* xbdev is closed */
2038 		err = -ERESTARTSYS;
2039 
2040 	mutex_unlock(&info->mutex);
2041 
2042 out:
2043 	mutex_unlock(&blkfront_mutex);
2044 	return err;
2045 }
2046 
2047 static void blkif_release(struct gendisk *disk, fmode_t mode)
2048 {
2049 	struct blkfront_info *info = disk->private_data;
2050 	struct block_device *bdev;
2051 	struct xenbus_device *xbdev;
2052 
2053 	mutex_lock(&blkfront_mutex);
2054 
2055 	bdev = bdget_disk(disk, 0);
2056 
2057 	if (!bdev) {
2058 		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2059 		goto out_mutex;
2060 	}
2061 	if (bdev->bd_openers)
2062 		goto out;
2063 
2064 	/*
2065 	 * Check if we have been instructed to close. We will have
2066 	 * deferred this request, because the bdev was still open.
2067 	 */
2068 
2069 	mutex_lock(&info->mutex);
2070 	xbdev = info->xbdev;
2071 
2072 	if (xbdev && xbdev->state == XenbusStateClosing) {
2073 		/* pending switch to state closed */
2074 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2075 		xlvbd_release_gendisk(info);
2076 		xenbus_frontend_closed(info->xbdev);
2077  	}
2078 
2079 	mutex_unlock(&info->mutex);
2080 
2081 	if (!xbdev) {
2082 		/* sudden device removal */
2083 		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2084 		xlvbd_release_gendisk(info);
2085 		disk->private_data = NULL;
2086 		kfree(info);
2087 	}
2088 
2089 out:
2090 	bdput(bdev);
2091 out_mutex:
2092 	mutex_unlock(&blkfront_mutex);
2093 }
2094 
2095 static const struct block_device_operations xlvbd_block_fops =
2096 {
2097 	.owner = THIS_MODULE,
2098 	.open = blkif_open,
2099 	.release = blkif_release,
2100 	.getgeo = blkif_getgeo,
2101 	.ioctl = blkif_ioctl,
2102 };
2103 
2104 
2105 static const struct xenbus_device_id blkfront_ids[] = {
2106 	{ "vbd" },
2107 	{ "" }
2108 };
2109 
2110 static struct xenbus_driver blkfront_driver = {
2111 	.ids  = blkfront_ids,
2112 	.probe = blkfront_probe,
2113 	.remove = blkfront_remove,
2114 	.resume = blkfront_resume,
2115 	.otherend_changed = blkback_changed,
2116 	.is_ready = blkfront_is_ready,
2117 };
2118 
2119 static int __init xlblk_init(void)
2120 {
2121 	int ret;
2122 
2123 	if (!xen_domain())
2124 		return -ENODEV;
2125 
2126 	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) {
2127 		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2128 			xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER);
2129 		xen_blkif_max_ring_order = 0;
2130 	}
2131 
2132 	if (!xen_has_pv_disk_devices())
2133 		return -ENODEV;
2134 
2135 	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2136 		printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2137 		       XENVBD_MAJOR, DEV_NAME);
2138 		return -ENODEV;
2139 	}
2140 
2141 	ret = xenbus_register_frontend(&blkfront_driver);
2142 	if (ret) {
2143 		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2144 		return ret;
2145 	}
2146 
2147 	return 0;
2148 }
2149 module_init(xlblk_init);
2150 
2151 
2152 static void __exit xlblk_exit(void)
2153 {
2154 	xenbus_unregister_driver(&blkfront_driver);
2155 	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2156 	kfree(minors);
2157 }
2158 module_exit(xlblk_exit);
2159 
2160 MODULE_DESCRIPTION("Xen virtual block device frontend");
2161 MODULE_LICENSE("GPL");
2162 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2163 MODULE_ALIAS("xen:vbd");
2164 MODULE_ALIAS("xenblk");
2165