xen-blkfront.c

   1/*
   2 * blkfront.c
   3 *
   4 * XenLinux virtual block device driver.
   5 *
   6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8 * Copyright (c) 2004, Christian Limpach
   9 * Copyright (c) 2004, Andrew Warfield
  10 * Copyright (c) 2005, Christopher Clark
  11 * Copyright (c) 2005, XenSource Ltd
  12 *
  13 * This program is free software; you can redistribute it and/or
  14 * modify it under the terms of the GNU General Public License version 2
  15 * as published by the Free Software Foundation; or, when distributed
  16 * separately from the Linux kernel or incorporated into other
  17 * software packages, subject to the following license:
  18 *
  19 * Permission is hereby granted, free of charge, to any person obtaining a copy
  20 * of this source file (the "Software"), to deal in the Software without
  21 * restriction, including without limitation the rights to use, copy, modify,
  22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23 * and to permit persons to whom the Software is furnished to do so, subject to
  24 * the following conditions:
  25 *
  26 * The above copyright notice and this permission notice shall be included in
  27 * all copies or substantial portions of the Software.
  28 *
  29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35 * IN THE SOFTWARE.
  36 */
  37
  38#include <linux/interrupt.h>
  39#include <linux/blkdev.h>
  40#include <linux/blk-mq.h>
  41#include <linux/hdreg.h>
  42#include <linux/cdrom.h>
  43#include <linux/module.h>
  44#include <linux/slab.h>
  45#include <linux/major.h>
  46#include <linux/mutex.h>
  47#include <linux/scatterlist.h>
  48#include <linux/bitmap.h>
  49#include <linux/list.h>
  50#include <linux/workqueue.h>
  51#include <linux/sched/mm.h>
  52
  53#include <xen/xen.h>
  54#include <xen/xenbus.h>
  55#include <xen/grant_table.h>
  56#include <xen/events.h>
  57#include <xen/page.h>
  58#include <xen/platform_pci.h>
  59
  60#include <xen/interface/grant_table.h>
  61#include <xen/interface/io/blkif.h>
  62#include <xen/interface/io/protocols.h>
  63
  64#include <asm/xen/hypervisor.h>
  65
  66/*
  67 * The minimal size of segment supported by the block framework is PAGE_SIZE.
  68 * When Linux is using a different page size than Xen, it may not be possible
  69 * to put all the data in a single segment.
  70 * This can happen when the backend doesn't support indirect descriptor and
  71 * therefore the maximum amount of data that a request can carry is
  72 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  73 *
  74 * Note that we only support one extra request. So the Linux page size
  75 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  76 * 88KB.
  77 */
  78#define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  79
  80enum blkif_state {
  81	BLKIF_STATE_DISCONNECTED,
  82	BLKIF_STATE_CONNECTED,
  83	BLKIF_STATE_SUSPENDED,
  84	BLKIF_STATE_ERROR,
  85};
  86
  87struct grant {
  88	grant_ref_t gref;
  89	struct page *page;
  90	struct list_head node;
  91};
  92
  93enum blk_req_status {
  94	REQ_PROCESSING,
  95	REQ_WAITING,
  96	REQ_DONE,
  97	REQ_ERROR,
  98	REQ_EOPNOTSUPP,
  99};
 100
 101struct blk_shadow {
 102	struct blkif_request req;
 103	struct request *request;
 104	struct grant **grants_used;
 105	struct grant **indirect_grants;
 106	struct scatterlist *sg;
 107	unsigned int num_sg;
 108	enum blk_req_status status;
 109
 110	#define NO_ASSOCIATED_ID ~0UL
 111	/*
 112	 * Id of the sibling if we ever need 2 requests when handling a
 113	 * block I/O request
 114	 */
 115	unsigned long associated_id;
 116};
 117
 118struct blkif_req {
 119	blk_status_t	error;
 120};
 121
 122static inline struct blkif_req *blkif_req(struct request *rq)
 123{
 124	return blk_mq_rq_to_pdu(rq);
 125}
 126
 127static DEFINE_MUTEX(blkfront_mutex);
 128static const struct block_device_operations xlvbd_block_fops;
 129static struct delayed_work blkfront_work;
 130static LIST_HEAD(info_list);
 131
 132/*
 133 * Maximum number of segments in indirect requests, the actual value used by
 134 * the frontend driver is the minimum of this value and the value provided
 135 * by the backend driver.
 136 */
 137
 138static unsigned int xen_blkif_max_segments = 32;
 139module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 140MODULE_PARM_DESC(max_indirect_segments,
 141		 "Maximum amount of segments in indirect requests (default is 32)");
 142
 143static unsigned int xen_blkif_max_queues = 4;
 144module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 145MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 146
 147/*
 148 * Maximum order of pages to be used for the shared ring between front and
 149 * backend, 4KB page granularity is used.
 150 */
 151static unsigned int xen_blkif_max_ring_order;
 152module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 153MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 154
 155static bool __read_mostly xen_blkif_trusted = true;
 156module_param_named(trusted, xen_blkif_trusted, bool, 0644);
 157MODULE_PARM_DESC(trusted, "Is the backend trusted");
 158
 159#define BLK_RING_SIZE(info)	\
 160	__CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 161
 162/*
 163 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 164 * characters are enough. Define to 20 to keep consistent with backend.
 165 */
 166#define RINGREF_NAME_LEN (20)
 167/*
 168 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 169 */
 170#define QUEUE_NAME_LEN (17)
 171
 172/*
 173 *  Per-ring info.
 174 *  Every blkfront device can associate with one or more blkfront_ring_info,
 175 *  depending on how many hardware queues/rings to be used.
 176 */
 177struct blkfront_ring_info {
 178	/* Lock to protect data in every ring buffer. */
 179	spinlock_t ring_lock;
 180	struct blkif_front_ring ring;
 181	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 182	unsigned int evtchn, irq;
 183	struct work_struct work;
 184	struct gnttab_free_callback callback;
 185	struct list_head indirect_pages;
 186	struct list_head grants;
 187	unsigned int persistent_gnts_c;
 188	unsigned long shadow_free;
 189	struct blkfront_info *dev_info;
 190	struct blk_shadow shadow[];
 191};
 192
 193/*
 194 * We have one of these per vbd, whether ide, scsi or 'other'.  They
 195 * hang in private_data off the gendisk structure. We may end up
 196 * putting all kinds of interesting stuff here :-)
 197 */
 198struct blkfront_info
 199{
 200	struct mutex mutex;
 201	struct xenbus_device *xbdev;
 202	struct gendisk *gd;
 203	u16 sector_size;
 204	unsigned int physical_sector_size;
 205	unsigned long vdisk_info;
 206	int vdevice;
 207	blkif_vdev_t handle;
 208	enum blkif_state connected;
 209	/* Number of pages per ring buffer. */
 210	unsigned int nr_ring_pages;
 211	struct request_queue *rq;
 212	unsigned int feature_flush:1;
 213	unsigned int feature_fua:1;
 214	unsigned int feature_discard:1;
 215	unsigned int feature_secdiscard:1;
 216	/* Connect-time cached feature_persistent parameter */
 217	unsigned int feature_persistent_parm:1;
 218	/* Persistent grants feature negotiation result */
 219	unsigned int feature_persistent:1;
 220	unsigned int bounce:1;
 221	unsigned int discard_granularity;
 222	unsigned int discard_alignment;
 223	/* Number of 4KB segments handled */
 224	unsigned int max_indirect_segments;
 225	int is_ready;
 226	struct blk_mq_tag_set tag_set;
 227	struct blkfront_ring_info *rinfo;
 228	unsigned int nr_rings;
 229	unsigned int rinfo_size;
 230	/* Save uncomplete reqs and bios for migration. */
 231	struct list_head requests;
 232	struct bio_list bio_list;
 233	struct list_head info_list;
 234};
 235
 236static unsigned int nr_minors;
 237static unsigned long *minors;
 238static DEFINE_SPINLOCK(minor_lock);
 239
 240#define PARTS_PER_DISK		16
 241#define PARTS_PER_EXT_DISK      256
 242
 243#define BLKIF_MAJOR(dev) ((dev)>>8)
 244#define BLKIF_MINOR(dev) ((dev) & 0xff)
 245
 246#define EXT_SHIFT 28
 247#define EXTENDED (1<<EXT_SHIFT)
 248#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 249#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 250#define EMULATED_HD_DISK_MINOR_OFFSET (0)
 251#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 252#define EMULATED_SD_DISK_MINOR_OFFSET (0)
 253#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 254
 255#define DEV_NAME	"xvd"	/* name in /dev */
 256
 257/*
 258 * Grants are always the same size as a Xen page (i.e 4KB).
 259 * A physical segment is always the same size as a Linux page.
 260 * Number of grants per physical segment
 261 */
 262#define GRANTS_PER_PSEG	(PAGE_SIZE / XEN_PAGE_SIZE)
 263
 264#define GRANTS_PER_INDIRECT_FRAME \
 265	(XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 266
 267#define INDIRECT_GREFS(_grants)		\
 268	DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 269
 270static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 271static void blkfront_gather_backend_features(struct blkfront_info *info);
 272static int negotiate_mq(struct blkfront_info *info);
 273
 274#define for_each_rinfo(info, ptr, idx)				\
 275	for ((ptr) = (info)->rinfo, (idx) = 0;			\
 276	     (idx) < (info)->nr_rings;				\
 277	     (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 278
 279static inline struct blkfront_ring_info *
 280get_rinfo(const struct blkfront_info *info, unsigned int i)
 281{
 282	BUG_ON(i >= info->nr_rings);
 283	return (void *)info->rinfo + i * info->rinfo_size;
 284}
 285
 286static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 287{
 288	unsigned long free = rinfo->shadow_free;
 289
 290	BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 291	rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 292	rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 293	return free;
 294}
 295
 296static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 297			      unsigned long id)
 298{
 299	if (rinfo->shadow[id].req.u.rw.id != id)
 300		return -EINVAL;
 301	if (rinfo->shadow[id].request == NULL)
 302		return -EINVAL;
 303	rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 304	rinfo->shadow[id].request = NULL;
 305	rinfo->shadow_free = id;
 306	return 0;
 307}
 308
 309static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 310{
 311	struct blkfront_info *info = rinfo->dev_info;
 312	struct page *granted_page;
 313	struct grant *gnt_list_entry, *n;
 314	int i = 0;
 315
 316	while (i < num) {
 317		gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 318		if (!gnt_list_entry)
 319			goto out_of_memory;
 320
 321		if (info->bounce) {
 322			granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
 323			if (!granted_page) {
 324				kfree(gnt_list_entry);
 325				goto out_of_memory;
 326			}
 327			gnt_list_entry->page = granted_page;
 328		}
 329
 330		gnt_list_entry->gref = INVALID_GRANT_REF;
 331		list_add(&gnt_list_entry->node, &rinfo->grants);
 332		i++;
 333	}
 334
 335	return 0;
 336
 337out_of_memory:
 338	list_for_each_entry_safe(gnt_list_entry, n,
 339	                         &rinfo->grants, node) {
 340		list_del(&gnt_list_entry->node);
 341		if (info->bounce)
 342			__free_page(gnt_list_entry->page);
 343		kfree(gnt_list_entry);
 344		i--;
 345	}
 346	BUG_ON(i != 0);
 347	return -ENOMEM;
 348}
 349
 350static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 351{
 352	struct grant *gnt_list_entry;
 353
 354	BUG_ON(list_empty(&rinfo->grants));
 355	gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 356					  node);
 357	list_del(&gnt_list_entry->node);
 358
 359	if (gnt_list_entry->gref != INVALID_GRANT_REF)
 360		rinfo->persistent_gnts_c--;
 361
 362	return gnt_list_entry;
 363}
 364
 365static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 366					const struct blkfront_info *info)
 367{
 368	gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 369						 info->xbdev->otherend_id,
 370						 gnt_list_entry->page,
 371						 0);
 372}
 373
 374static struct grant *get_grant(grant_ref_t *gref_head,
 375			       unsigned long gfn,
 376			       struct blkfront_ring_info *rinfo)
 377{
 378	struct grant *gnt_list_entry = get_free_grant(rinfo);
 379	struct blkfront_info *info = rinfo->dev_info;
 380
 381	if (gnt_list_entry->gref != INVALID_GRANT_REF)
 382		return gnt_list_entry;
 383
 384	/* Assign a gref to this page */
 385	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 386	BUG_ON(gnt_list_entry->gref == -ENOSPC);
 387	if (info->bounce)
 388		grant_foreign_access(gnt_list_entry, info);
 389	else {
 390		/* Grant access to the GFN passed by the caller */
 391		gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 392						info->xbdev->otherend_id,
 393						gfn, 0);
 394	}
 395
 396	return gnt_list_entry;
 397}
 398
 399static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 400					struct blkfront_ring_info *rinfo)
 401{
 402	struct grant *gnt_list_entry = get_free_grant(rinfo);
 403	struct blkfront_info *info = rinfo->dev_info;
 404
 405	if (gnt_list_entry->gref != INVALID_GRANT_REF)
 406		return gnt_list_entry;
 407
 408	/* Assign a gref to this page */
 409	gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 410	BUG_ON(gnt_list_entry->gref == -ENOSPC);
 411	if (!info->bounce) {
 412		struct page *indirect_page;
 413
 414		/* Fetch a pre-allocated page to use for indirect grefs */
 415		BUG_ON(list_empty(&rinfo->indirect_pages));
 416		indirect_page = list_first_entry(&rinfo->indirect_pages,
 417						 struct page, lru);
 418		list_del(&indirect_page->lru);
 419		gnt_list_entry->page = indirect_page;
 420	}
 421	grant_foreign_access(gnt_list_entry, info);
 422
 423	return gnt_list_entry;
 424}
 425
 426static const char *op_name(int op)
 427{
 428	static const char *const names[] = {
 429		[BLKIF_OP_READ] = "read",
 430		[BLKIF_OP_WRITE] = "write",
 431		[BLKIF_OP_WRITE_BARRIER] = "barrier",
 432		[BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 433		[BLKIF_OP_DISCARD] = "discard" };
 434
 435	if (op < 0 || op >= ARRAY_SIZE(names))
 436		return "unknown";
 437
 438	if (!names[op])
 439		return "reserved";
 440
 441	return names[op];
 442}
 443static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 444{
 445	unsigned int end = minor + nr;
 446	int rc;
 447
 448	if (end > nr_minors) {
 449		unsigned long *bitmap, *old;
 450
 451		bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 452				 GFP_KERNEL);
 453		if (bitmap == NULL)
 454			return -ENOMEM;
 455
 456		spin_lock(&minor_lock);
 457		if (end > nr_minors) {
 458			old = minors;
 459			memcpy(bitmap, minors,
 460			       BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 461			minors = bitmap;
 462			nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 463		} else
 464			old = bitmap;
 465		spin_unlock(&minor_lock);
 466		kfree(old);
 467	}
 468
 469	spin_lock(&minor_lock);
 470	if (find_next_bit(minors, end, minor) >= end) {
 471		bitmap_set(minors, minor, nr);
 472		rc = 0;
 473	} else
 474		rc = -EBUSY;
 475	spin_unlock(&minor_lock);
 476
 477	return rc;
 478}
 479
 480static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 481{
 482	unsigned int end = minor + nr;
 483
 484	BUG_ON(end > nr_minors);
 485	spin_lock(&minor_lock);
 486	bitmap_clear(minors,  minor, nr);
 487	spin_unlock(&minor_lock);
 488}
 489
 490static void blkif_restart_queue_callback(void *arg)
 491{
 492	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 493	schedule_work(&rinfo->work);
 494}
 495
 496static int blkif_getgeo(struct gendisk *disk, struct hd_geometry *hg)
 497{
 498	/* We don't have real geometry info, but let's at least return
 499	   values consistent with the size of the device */
 500	sector_t nsect = get_capacity(disk);
 501	sector_t cylinders = nsect;
 502
 503	hg->heads = 0xff;
 504	hg->sectors = 0x3f;
 505	sector_div(cylinders, hg->heads * hg->sectors);
 506	hg->cylinders = cylinders;
 507	if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 508		hg->cylinders = 0xffff;
 509	return 0;
 510}
 511
 512static int blkif_ioctl(struct block_device *bdev, blk_mode_t mode,
 513		       unsigned command, unsigned long argument)
 514{
 515	struct blkfront_info *info = bdev->bd_disk->private_data;
 516	int i;
 517
 518	switch (command) {
 519	case CDROMMULTISESSION:
 520		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 521			if (put_user(0, (char __user *)(argument + i)))
 522				return -EFAULT;
 523		return 0;
 524	case CDROM_GET_CAPABILITY:
 525		if (!(info->vdisk_info & VDISK_CDROM))
 526			return -EINVAL;
 527		return 0;
 528	default:
 529		return -EINVAL;
 530	}
 531}
 532
 533static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 534					    struct request *req,
 535					    struct blkif_request **ring_req)
 536{
 537	unsigned long id;
 538
 539	*ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 540	rinfo->ring.req_prod_pvt++;
 541
 542	id = get_id_from_freelist(rinfo);
 543	rinfo->shadow[id].request = req;
 544	rinfo->shadow[id].status = REQ_PROCESSING;
 545	rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 546
 547	rinfo->shadow[id].req.u.rw.id = id;
 548
 549	return id;
 550}
 551
 552static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 553{
 554	struct blkfront_info *info = rinfo->dev_info;
 555	struct blkif_request *ring_req, *final_ring_req;
 556	unsigned long id;
 557
 558	/* Fill out a communications ring structure. */
 559	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 560	ring_req = &rinfo->shadow[id].req;
 561
 562	ring_req->operation = BLKIF_OP_DISCARD;
 563	ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 564	ring_req->u.discard.id = id;
 565	ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 566	if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 567		ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 568	else
 569		ring_req->u.discard.flag = 0;
 570
 571	/* Copy the request to the ring page. */
 572	*final_ring_req = *ring_req;
 573	rinfo->shadow[id].status = REQ_WAITING;
 574
 575	return 0;
 576}
 577
 578struct setup_rw_req {
 579	unsigned int grant_idx;
 580	struct blkif_request_segment *segments;
 581	struct blkfront_ring_info *rinfo;
 582	struct blkif_request *ring_req;
 583	grant_ref_t gref_head;
 584	unsigned int id;
 585	/* Only used when persistent grant is used and it's a write request */
 586	bool need_copy;
 587	unsigned int bvec_off;
 588	char *bvec_data;
 589
 590	bool require_extra_req;
 591	struct blkif_request *extra_ring_req;
 592};
 593
 594static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 595				     unsigned int len, void *data)
 596{
 597	struct setup_rw_req *setup = data;
 598	int n, ref;
 599	struct grant *gnt_list_entry;
 600	unsigned int fsect, lsect;
 601	/* Convenient aliases */
 602	unsigned int grant_idx = setup->grant_idx;
 603	struct blkif_request *ring_req = setup->ring_req;
 604	struct blkfront_ring_info *rinfo = setup->rinfo;
 605	/*
 606	 * We always use the shadow of the first request to store the list
 607	 * of grant associated to the block I/O request. This made the
 608	 * completion more easy to handle even if the block I/O request is
 609	 * split.
 610	 */
 611	struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 612
 613	if (unlikely(setup->require_extra_req &&
 614		     grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 615		/*
 616		 * We are using the second request, setup grant_idx
 617		 * to be the index of the segment array.
 618		 */
 619		grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 620		ring_req = setup->extra_ring_req;
 621	}
 622
 623	if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 624	    (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 625		if (setup->segments)
 626			kunmap_atomic(setup->segments);
 627
 628		n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 629		gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 630		shadow->indirect_grants[n] = gnt_list_entry;
 631		setup->segments = kmap_atomic(gnt_list_entry->page);
 632		ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 633	}
 634
 635	gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 636	ref = gnt_list_entry->gref;
 637	/*
 638	 * All the grants are stored in the shadow of the first
 639	 * request. Therefore we have to use the global index.
 640	 */
 641	shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 642
 643	if (setup->need_copy) {
 644		void *shared_data;
 645
 646		shared_data = kmap_atomic(gnt_list_entry->page);
 647		/*
 648		 * this does not wipe data stored outside the
 649		 * range sg->offset..sg->offset+sg->length.
 650		 * Therefore, blkback *could* see data from
 651		 * previous requests. This is OK as long as
 652		 * persistent grants are shared with just one
 653		 * domain. It may need refactoring if this
 654		 * changes
 655		 */
 656		memcpy(shared_data + offset,
 657		       setup->bvec_data + setup->bvec_off,
 658		       len);
 659
 660		kunmap_atomic(shared_data);
 661		setup->bvec_off += len;
 662	}
 663
 664	fsect = offset >> 9;
 665	lsect = fsect + (len >> 9) - 1;
 666	if (ring_req->operation != BLKIF_OP_INDIRECT) {
 667		ring_req->u.rw.seg[grant_idx] =
 668			(struct blkif_request_segment) {
 669				.gref       = ref,
 670				.first_sect = fsect,
 671				.last_sect  = lsect };
 672	} else {
 673		setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 674			(struct blkif_request_segment) {
 675				.gref       = ref,
 676				.first_sect = fsect,
 677				.last_sect  = lsect };
 678	}
 679
 680	(setup->grant_idx)++;
 681}
 682
 683static void blkif_setup_extra_req(struct blkif_request *first,
 684				  struct blkif_request *second)
 685{
 686	uint16_t nr_segments = first->u.rw.nr_segments;
 687
 688	/*
 689	 * The second request is only present when the first request uses
 690	 * all its segments. It's always the continuity of the first one.
 691	 */
 692	first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 693
 694	second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 695	second->u.rw.sector_number = first->u.rw.sector_number +
 696		(BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 697
 698	second->u.rw.handle = first->u.rw.handle;
 699	second->operation = first->operation;
 700}
 701
 702static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 703{
 704	struct blkfront_info *info = rinfo->dev_info;
 705	struct blkif_request *ring_req, *extra_ring_req = NULL;
 706	struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 707	unsigned long id, extra_id = NO_ASSOCIATED_ID;
 708	bool require_extra_req = false;
 709	int i;
 710	struct setup_rw_req setup = {
 711		.grant_idx = 0,
 712		.segments = NULL,
 713		.rinfo = rinfo,
 714		.need_copy = rq_data_dir(req) && info->bounce,
 715	};
 716
 717	/*
 718	 * Used to store if we are able to queue the request by just using
 719	 * existing persistent grants, or if we have to get new grants,
 720	 * as there are not sufficiently many free.
 721	 */
 722	bool new_persistent_gnts = false;
 723	struct scatterlist *sg;
 724	int num_sg, max_grefs, num_grant;
 725
 726	max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 727	if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 728		/*
 729		 * If we are using indirect segments we need to account
 730		 * for the indirect grefs used in the request.
 731		 */
 732		max_grefs += INDIRECT_GREFS(max_grefs);
 733
 734	/* Check if we have enough persistent grants to allocate a requests */
 735	if (rinfo->persistent_gnts_c < max_grefs) {
 736		new_persistent_gnts = true;
 737
 738		if (gnttab_alloc_grant_references(
 739		    max_grefs - rinfo->persistent_gnts_c,
 740		    &setup.gref_head) < 0) {
 741			gnttab_request_free_callback(
 742				&rinfo->callback,
 743				blkif_restart_queue_callback,
 744				rinfo,
 745				max_grefs - rinfo->persistent_gnts_c);
 746			return 1;
 747		}
 748	}
 749
 750	/* Fill out a communications ring structure. */
 751	id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 752	ring_req = &rinfo->shadow[id].req;
 753
 754	num_sg = blk_rq_map_sg(req, rinfo->shadow[id].sg);
 755	num_grant = 0;
 756	/* Calculate the number of grant used */
 757	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 758	       num_grant += gnttab_count_grant(sg->offset, sg->length);
 759
 760	require_extra_req = info->max_indirect_segments == 0 &&
 761		num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 762	BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 763
 764	rinfo->shadow[id].num_sg = num_sg;
 765	if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 766	    likely(!require_extra_req)) {
 767		/*
 768		 * The indirect operation can only be a BLKIF_OP_READ or
 769		 * BLKIF_OP_WRITE
 770		 */
 771		BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 772		ring_req->operation = BLKIF_OP_INDIRECT;
 773		ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 774			BLKIF_OP_WRITE : BLKIF_OP_READ;
 775		ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 776		ring_req->u.indirect.handle = info->handle;
 777		ring_req->u.indirect.nr_segments = num_grant;
 778	} else {
 779		ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 780		ring_req->u.rw.handle = info->handle;
 781		ring_req->operation = rq_data_dir(req) ?
 782			BLKIF_OP_WRITE : BLKIF_OP_READ;
 783		if (req_op(req) == REQ_OP_FLUSH ||
 784		    (req_op(req) == REQ_OP_WRITE && (req->cmd_flags & REQ_FUA))) {
 785			/*
 786			 * Ideally we can do an unordered flush-to-disk.
 787			 * In case the backend onlysupports barriers, use that.
 788			 * A barrier request a superset of FUA, so we can
 789			 * implement it the same way.  (It's also a FLUSH+FUA,
 790			 * since it is guaranteed ordered WRT previous writes.)
 791			 *
 792			 * Note that can end up here with a FUA write and the
 793			 * flags cleared.  This happens when the flag was
 794			 * run-time disabled after a failing I/O, and we'll
 795			 * simplify submit it as a normal write.
 796			 */
 797			if (info->feature_flush && info->feature_fua)
 798				ring_req->operation =
 799					BLKIF_OP_WRITE_BARRIER;
 800			else if (info->feature_flush)
 801				ring_req->operation =
 802					BLKIF_OP_FLUSH_DISKCACHE;
 803		}
 804		ring_req->u.rw.nr_segments = num_grant;
 805		if (unlikely(require_extra_req)) {
 806			extra_id = blkif_ring_get_request(rinfo, req,
 807							  &final_extra_ring_req);
 808			extra_ring_req = &rinfo->shadow[extra_id].req;
 809
 810			/*
 811			 * Only the first request contains the scatter-gather
 812			 * list.
 813			 */
 814			rinfo->shadow[extra_id].num_sg = 0;
 815
 816			blkif_setup_extra_req(ring_req, extra_ring_req);
 817
 818			/* Link the 2 requests together */
 819			rinfo->shadow[extra_id].associated_id = id;
 820			rinfo->shadow[id].associated_id = extra_id;
 821		}
 822	}
 823
 824	setup.ring_req = ring_req;
 825	setup.id = id;
 826
 827	setup.require_extra_req = require_extra_req;
 828	if (unlikely(require_extra_req))
 829		setup.extra_ring_req = extra_ring_req;
 830
 831	for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 832		BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 833
 834		if (setup.need_copy) {
 835			setup.bvec_off = sg->offset;
 836			setup.bvec_data = kmap_atomic(sg_page(sg));
 837		}
 838
 839		gnttab_foreach_grant_in_range(sg_page(sg),
 840					      sg->offset,
 841					      sg->length,
 842					      blkif_setup_rw_req_grant,
 843					      &setup);
 844
 845		if (setup.need_copy)
 846			kunmap_atomic(setup.bvec_data);
 847	}
 848	if (setup.segments)
 849		kunmap_atomic(setup.segments);
 850
 851	/* Copy request(s) to the ring page. */
 852	*final_ring_req = *ring_req;
 853	rinfo->shadow[id].status = REQ_WAITING;
 854	if (unlikely(require_extra_req)) {
 855		*final_extra_ring_req = *extra_ring_req;
 856		rinfo->shadow[extra_id].status = REQ_WAITING;
 857	}
 858
 859	if (new_persistent_gnts)
 860		gnttab_free_grant_references(setup.gref_head);
 861
 862	return 0;
 863}
 864
 865/*
 866 * Generate a Xen blkfront IO request from a blk layer request.  Reads
 867 * and writes are handled as expected.
 868 *
 869 * @req: a request struct
 870 */
 871static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 872{
 873	if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 874		return 1;
 875
 876	if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 877		     req_op(req) == REQ_OP_SECURE_ERASE))
 878		return blkif_queue_discard_req(req, rinfo);
 879	else
 880		return blkif_queue_rw_req(req, rinfo);
 881}
 882
 883static inline void flush_requests(struct blkfront_ring_info *rinfo)
 884{
 885	int notify;
 886
 887	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 888
 889	if (notify)
 890		notify_remote_via_irq(rinfo->irq);
 891}
 892
 893static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 894			  const struct blk_mq_queue_data *qd)
 895{
 896	unsigned long flags;
 897	int qid = hctx->queue_num;
 898	struct blkfront_info *info = hctx->queue->queuedata;
 899	struct blkfront_ring_info *rinfo = NULL;
 900
 901	rinfo = get_rinfo(info, qid);
 902	blk_mq_start_request(qd->rq);
 903	spin_lock_irqsave(&rinfo->ring_lock, flags);
 904
 905	/*
 906	 * Check if the backend actually supports flushes.
 907	 *
 908	 * While the block layer won't send us flushes if we don't claim to
 909	 * support them, the Xen protocol allows the backend to revoke support
 910	 * at any time.  That is of course a really bad idea and dangerous, but
 911	 * has been allowed for 10+ years.  In that case we simply clear the
 912	 * flags, and directly return here for an empty flush and ignore the
 913	 * FUA flag later on.
 914	 */
 915	if (unlikely(req_op(qd->rq) == REQ_OP_FLUSH && !info->feature_flush))
 916		goto complete;
 917
 918	if (RING_FULL(&rinfo->ring))
 919		goto out_busy;
 920	if (blkif_queue_request(qd->rq, rinfo))
 921		goto out_busy;
 922
 923	flush_requests(rinfo);
 924	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 925	return BLK_STS_OK;
 926
 927out_busy:
 928	blk_mq_stop_hw_queue(hctx);
 929	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 930	return BLK_STS_DEV_RESOURCE;
 931complete:
 932	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 933	blk_mq_end_request(qd->rq, BLK_STS_OK);
 934	return BLK_STS_OK;
 935}
 936
 937static void blkif_complete_rq(struct request *rq)
 938{
 939	blk_mq_end_request(rq, blkif_req(rq)->error);
 940}
 941
 942static const struct blk_mq_ops blkfront_mq_ops = {
 943	.queue_rq = blkif_queue_rq,
 944	.complete = blkif_complete_rq,
 945};
 946
 947static void blkif_set_queue_limits(const struct blkfront_info *info,
 948		struct queue_limits *lim)
 949{
 950	unsigned int segments = info->max_indirect_segments ? :
 951				BLKIF_MAX_SEGMENTS_PER_REQUEST;
 952
 953	if (info->feature_discard) {
 954		lim->max_hw_discard_sectors = UINT_MAX;
 955		if (info->discard_granularity)
 956			lim->discard_granularity = info->discard_granularity;
 957		lim->discard_alignment = info->discard_alignment;
 958		if (info->feature_secdiscard)
 959			lim->max_secure_erase_sectors = UINT_MAX;
 960	}
 961
 962	if (info->feature_flush) {
 963		lim->features |= BLK_FEAT_WRITE_CACHE;
 964		if (info->feature_fua)
 965			lim->features |= BLK_FEAT_FUA;
 966	}
 967
 968	/* Hard sector size and max sectors impersonate the equiv. hardware. */
 969	lim->logical_block_size = info->sector_size;
 970	lim->physical_block_size = info->physical_sector_size;
 971	lim->max_hw_sectors = (segments * XEN_PAGE_SIZE) / 512;
 972
 973	/* Each segment in a request is up to an aligned page in size. */
 974	lim->seg_boundary_mask = PAGE_SIZE - 1;
 975	lim->max_segment_size = PAGE_SIZE;
 976
 977	/* Ensure a merged request will fit in a single I/O ring slot. */
 978	lim->max_segments = segments / GRANTS_PER_PSEG;
 979
 980	/* Make sure buffer addresses are sector-aligned. */
 981	lim->dma_alignment = 511;
 982}
 983
 984static const char *flush_info(struct blkfront_info *info)
 985{
 986	if (info->feature_flush && info->feature_fua)
 987		return "barrier: enabled;";
 988	else if (info->feature_flush)
 989		return "flush diskcache: enabled;";
 990	else
 991		return "barrier or flush: disabled;";
 992}
 993
 994static void xlvbd_flush(struct blkfront_info *info)
 995{
 996	pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
 997		info->gd->disk_name, flush_info(info),
 998		"persistent grants:", info->feature_persistent ?
 999		"enabled;" : "disabled;", "indirect descriptors:",
1000		info->max_indirect_segments ? "enabled;" : "disabled;",
1001		"bounce buffer:", info->bounce ? "enabled" : "disabled;");
1002}
1003
1004static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1005{
1006	int major;
1007	major = BLKIF_MAJOR(vdevice);
1008	*minor = BLKIF_MINOR(vdevice);
1009	switch (major) {
1010		case XEN_IDE0_MAJOR:
1011			*offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1012			*minor = ((*minor / 64) * PARTS_PER_DISK) +
1013				EMULATED_HD_DISK_MINOR_OFFSET;
1014			break;
1015		case XEN_IDE1_MAJOR:
1016			*offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1017			*minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1018				EMULATED_HD_DISK_MINOR_OFFSET;
1019			break;
1020		case XEN_SCSI_DISK0_MAJOR:
1021			*offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1022			*minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1023			break;
1024		case XEN_SCSI_DISK1_MAJOR:
1025		case XEN_SCSI_DISK2_MAJOR:
1026		case XEN_SCSI_DISK3_MAJOR:
1027		case XEN_SCSI_DISK4_MAJOR:
1028		case XEN_SCSI_DISK5_MAJOR:
1029		case XEN_SCSI_DISK6_MAJOR:
1030		case XEN_SCSI_DISK7_MAJOR:
1031			*offset = (*minor / PARTS_PER_DISK) + 
1032				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1033				EMULATED_SD_DISK_NAME_OFFSET;
1034			*minor = *minor +
1035				((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1036				EMULATED_SD_DISK_MINOR_OFFSET;
1037			break;
1038		case XEN_SCSI_DISK8_MAJOR:
1039		case XEN_SCSI_DISK9_MAJOR:
1040		case XEN_SCSI_DISK10_MAJOR:
1041		case XEN_SCSI_DISK11_MAJOR:
1042		case XEN_SCSI_DISK12_MAJOR:
1043		case XEN_SCSI_DISK13_MAJOR:
1044		case XEN_SCSI_DISK14_MAJOR:
1045		case XEN_SCSI_DISK15_MAJOR:
1046			*offset = (*minor / PARTS_PER_DISK) + 
1047				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1048				EMULATED_SD_DISK_NAME_OFFSET;
1049			*minor = *minor +
1050				((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1051				EMULATED_SD_DISK_MINOR_OFFSET;
1052			break;
1053		case XENVBD_MAJOR:
1054			*offset = *minor / PARTS_PER_DISK;
1055			break;
1056		default:
1057			printk(KERN_WARNING "blkfront: your disk configuration is "
1058					"incorrect, please use an xvd device instead\n");
1059			return -ENODEV;
1060	}
1061	return 0;
1062}
1063
1064static char *encode_disk_name(char *ptr, unsigned int n)
1065{
1066	if (n >= 26)
1067		ptr = encode_disk_name(ptr, n / 26 - 1);
1068	*ptr = 'a' + n % 26;
1069	return ptr + 1;
1070}
1071
1072static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1073		struct blkfront_info *info)
1074{
1075	struct queue_limits lim = {};
1076	struct gendisk *gd;
1077	int nr_minors = 1;
1078	int err;
1079	unsigned int offset;
1080	int minor;
1081	int nr_parts;
1082	char *ptr;
1083
1084	BUG_ON(info->gd != NULL);
1085	BUG_ON(info->rq != NULL);
1086
1087	if ((info->vdevice>>EXT_SHIFT) > 1) {
1088		/* this is above the extended range; something is wrong */
1089		printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1090		return -ENODEV;
1091	}
1092
1093	if (!VDEV_IS_EXTENDED(info->vdevice)) {
1094		err = xen_translate_vdev(info->vdevice, &minor, &offset);
1095		if (err)
1096			return err;
1097		nr_parts = PARTS_PER_DISK;
1098	} else {
1099		minor = BLKIF_MINOR_EXT(info->vdevice);
1100		nr_parts = PARTS_PER_EXT_DISK;
1101		offset = minor / nr_parts;
1102		if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1103			printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1104					"emulated IDE disks,\n\t choose an xvd device name"
1105					"from xvde on\n", info->vdevice);
1106	}
1107	if (minor >> MINORBITS) {
1108		pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1109			info->vdevice, minor);
1110		return -ENODEV;
1111	}
1112
1113	if ((minor % nr_parts) == 0)
1114		nr_minors = nr_parts;
1115
1116	err = xlbd_reserve_minors(minor, nr_minors);
1117	if (err)
1118		return err;
1119
1120	memset(&info->tag_set, 0, sizeof(info->tag_set));
1121	info->tag_set.ops = &blkfront_mq_ops;
1122	info->tag_set.nr_hw_queues = info->nr_rings;
1123	if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1124		/*
1125		 * When indirect descriptior is not supported, the I/O request
1126		 * will be split between multiple request in the ring.
1127		 * To avoid problems when sending the request, divide by
1128		 * 2 the depth of the queue.
1129		 */
1130		info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1131	} else
1132		info->tag_set.queue_depth = BLK_RING_SIZE(info);
1133	info->tag_set.numa_node = NUMA_NO_NODE;
1134	info->tag_set.cmd_size = sizeof(struct blkif_req);
1135	info->tag_set.driver_data = info;
1136
1137	err = blk_mq_alloc_tag_set(&info->tag_set);
1138	if (err)
1139		goto out_release_minors;
1140
1141	blkif_set_queue_limits(info, &lim);
1142	gd = blk_mq_alloc_disk(&info->tag_set, &lim, info);
1143	if (IS_ERR(gd)) {
1144		err = PTR_ERR(gd);
1145		goto out_free_tag_set;
1146	}
1147
1148	strcpy(gd->disk_name, DEV_NAME);
1149	ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1150	BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1151	if (nr_minors > 1)
1152		*ptr = 0;
1153	else
1154		snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1155			 "%d", minor & (nr_parts - 1));
1156
1157	gd->major = XENVBD_MAJOR;
1158	gd->first_minor = minor;
1159	gd->minors = nr_minors;
1160	gd->fops = &xlvbd_block_fops;
1161	gd->private_data = info;
1162	set_capacity(gd, capacity);
1163
1164	info->rq = gd->queue;
1165	info->gd = gd;
1166
1167	xlvbd_flush(info);
1168
1169	if (info->vdisk_info & VDISK_READONLY)
1170		set_disk_ro(gd, 1);
1171	if (info->vdisk_info & VDISK_REMOVABLE)
1172		gd->flags |= GENHD_FL_REMOVABLE;
1173
1174	return 0;
1175
1176out_free_tag_set:
1177	blk_mq_free_tag_set(&info->tag_set);
1178out_release_minors:
1179	xlbd_release_minors(minor, nr_minors);
1180	return err;
1181}
1182
1183/* Already hold rinfo->ring_lock. */
1184static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1185{
1186	if (!RING_FULL(&rinfo->ring))
1187		blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1188}
1189
1190static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1191{
1192	unsigned long flags;
1193
1194	spin_lock_irqsave(&rinfo->ring_lock, flags);
1195	kick_pending_request_queues_locked(rinfo);
1196	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1197}
1198
1199static void blkif_restart_queue(struct work_struct *work)
1200{
1201	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1202
1203	if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1204		kick_pending_request_queues(rinfo);
1205}
1206
1207static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1208{
1209	struct grant *persistent_gnt, *n;
1210	struct blkfront_info *info = rinfo->dev_info;
1211	int i, j, segs;
1212
1213	/*
1214	 * Remove indirect pages, this only happens when using indirect
1215	 * descriptors but not persistent grants
1216	 */
1217	if (!list_empty(&rinfo->indirect_pages)) {
1218		struct page *indirect_page, *n;
1219
1220		BUG_ON(info->bounce);
1221		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1222			list_del(&indirect_page->lru);
1223			__free_page(indirect_page);
1224		}
1225	}
1226
1227	/* Remove all persistent grants. */
1228	if (!list_empty(&rinfo->grants)) {
1229		list_for_each_entry_safe(persistent_gnt, n,
1230					 &rinfo->grants, node) {
1231			list_del(&persistent_gnt->node);
1232			if (persistent_gnt->gref != INVALID_GRANT_REF) {
1233				gnttab_end_foreign_access(persistent_gnt->gref,
1234							  NULL);
1235				rinfo->persistent_gnts_c--;
1236			}
1237			if (info->bounce)
1238				__free_page(persistent_gnt->page);
1239			kfree(persistent_gnt);
1240		}
1241	}
1242	BUG_ON(rinfo->persistent_gnts_c != 0);
1243
1244	for (i = 0; i < BLK_RING_SIZE(info); i++) {
1245		/*
1246		 * Clear persistent grants present in requests already
1247		 * on the shared ring
1248		 */
1249		if (!rinfo->shadow[i].request)
1250			goto free_shadow;
1251
1252		segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1253		       rinfo->shadow[i].req.u.indirect.nr_segments :
1254		       rinfo->shadow[i].req.u.rw.nr_segments;
1255		for (j = 0; j < segs; j++) {
1256			persistent_gnt = rinfo->shadow[i].grants_used[j];
1257			gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1258			if (info->bounce)
1259				__free_page(persistent_gnt->page);
1260			kfree(persistent_gnt);
1261		}
1262
1263		if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1264			/*
1265			 * If this is not an indirect operation don't try to
1266			 * free indirect segments
1267			 */
1268			goto free_shadow;
1269
1270		for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1271			persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1272			gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1273			__free_page(persistent_gnt->page);
1274			kfree(persistent_gnt);
1275		}
1276
1277free_shadow:
1278		kvfree(rinfo->shadow[i].grants_used);
1279		rinfo->shadow[i].grants_used = NULL;
1280		kvfree(rinfo->shadow[i].indirect_grants);
1281		rinfo->shadow[i].indirect_grants = NULL;
1282		kvfree(rinfo->shadow[i].sg);
1283		rinfo->shadow[i].sg = NULL;
1284	}
1285
1286	/* No more gnttab callback work. */
1287	gnttab_cancel_free_callback(&rinfo->callback);
1288
1289	/* Flush gnttab callback work. Must be done with no locks held. */
1290	flush_work(&rinfo->work);
1291
1292	/* Free resources associated with old device channel. */
1293	xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages,
1294			     rinfo->ring_ref);
1295
1296	if (rinfo->irq)
1297		unbind_from_irqhandler(rinfo->irq, rinfo);
1298	rinfo->evtchn = rinfo->irq = 0;
1299}
1300
1301static void blkif_free(struct blkfront_info *info, int suspend)
1302{
1303	unsigned int i;
1304	struct blkfront_ring_info *rinfo;
1305
1306	/* Prevent new requests being issued until we fix things up. */
1307	info->connected = suspend ?
1308		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1309	/* No more blkif_request(). */
1310	if (info->rq)
1311		blk_mq_stop_hw_queues(info->rq);
1312
1313	for_each_rinfo(info, rinfo, i)
1314		blkif_free_ring(rinfo);
1315
1316	kvfree(info->rinfo);
1317	info->rinfo = NULL;
1318	info->nr_rings = 0;
1319}
1320
1321struct copy_from_grant {
1322	const struct blk_shadow *s;
1323	unsigned int grant_idx;
1324	unsigned int bvec_offset;
1325	char *bvec_data;
1326};
1327
1328static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1329				  unsigned int len, void *data)
1330{
1331	struct copy_from_grant *info = data;
1332	char *shared_data;
1333	/* Convenient aliases */
1334	const struct blk_shadow *s = info->s;
1335
1336	shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1337
1338	memcpy(info->bvec_data + info->bvec_offset,
1339	       shared_data + offset, len);
1340
1341	info->bvec_offset += len;
1342	info->grant_idx++;
1343
1344	kunmap_atomic(shared_data);
1345}
1346
1347static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1348{
1349	switch (rsp)
1350	{
1351	case BLKIF_RSP_OKAY:
1352		return REQ_DONE;
1353	case BLKIF_RSP_EOPNOTSUPP:
1354		return REQ_EOPNOTSUPP;
1355	case BLKIF_RSP_ERROR:
1356	default:
1357		return REQ_ERROR;
1358	}
1359}
1360
1361/*
1362 * Get the final status of the block request based on two ring response
1363 */
1364static int blkif_get_final_status(enum blk_req_status s1,
1365				  enum blk_req_status s2)
1366{
1367	BUG_ON(s1 < REQ_DONE);
1368	BUG_ON(s2 < REQ_DONE);
1369
1370	if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1371		return BLKIF_RSP_ERROR;
1372	else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1373		return BLKIF_RSP_EOPNOTSUPP;
1374	return BLKIF_RSP_OKAY;
1375}
1376
1377/*
1378 * Return values:
1379 *  1 response processed.
1380 *  0 missing further responses.
1381 * -1 error while processing.
1382 */
1383static int blkif_completion(unsigned long *id,
1384			    struct blkfront_ring_info *rinfo,
1385			    struct blkif_response *bret)
1386{
1387	int i = 0;
1388	struct scatterlist *sg;
1389	int num_sg, num_grant;
1390	struct blkfront_info *info = rinfo->dev_info;
1391	struct blk_shadow *s = &rinfo->shadow[*id];
1392	struct copy_from_grant data = {
1393		.grant_idx = 0,
1394	};
1395
1396	num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1397		s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1398
1399	/* The I/O request may be split in two. */
1400	if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1401		struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1402
1403		/* Keep the status of the current response in shadow. */
1404		s->status = blkif_rsp_to_req_status(bret->status);
1405
1406		/* Wait the second response if not yet here. */
1407		if (s2->status < REQ_DONE)
1408			return 0;
1409
1410		bret->status = blkif_get_final_status(s->status,
1411						      s2->status);
1412
1413		/*
1414		 * All the grants is stored in the first shadow in order
1415		 * to make the completion code simpler.
1416		 */
1417		num_grant += s2->req.u.rw.nr_segments;
1418
1419		/*
1420		 * The two responses may not come in order. Only the
1421		 * first request will store the scatter-gather list.
1422		 */
1423		if (s2->num_sg != 0) {
1424			/* Update "id" with the ID of the first response. */
1425			*id = s->associated_id;
1426			s = s2;
1427		}
1428
1429		/*
1430		 * We don't need anymore the second request, so recycling
1431		 * it now.
1432		 */
1433		if (add_id_to_freelist(rinfo, s->associated_id))
1434			WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1435			     info->gd->disk_name, s->associated_id);
1436	}
1437
1438	data.s = s;
1439	num_sg = s->num_sg;
1440
1441	if (bret->operation == BLKIF_OP_READ && info->bounce) {
1442		for_each_sg(s->sg, sg, num_sg, i) {
1443			BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1444
1445			data.bvec_offset = sg->offset;
1446			data.bvec_data = kmap_atomic(sg_page(sg));
1447
1448			gnttab_foreach_grant_in_range(sg_page(sg),
1449						      sg->offset,
1450						      sg->length,
1451						      blkif_copy_from_grant,
1452						      &data);
1453
1454			kunmap_atomic(data.bvec_data);
1455		}
1456	}
1457	/* Add the persistent grant into the list of free grants */
1458	for (i = 0; i < num_grant; i++) {
1459		if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1460			/*
1461			 * If the grant is still mapped by the backend (the
1462			 * backend has chosen to make this grant persistent)
1463			 * we add it at the head of the list, so it will be
1464			 * reused first.
1465			 */
1466			if (!info->feature_persistent) {
1467				pr_alert("backed has not unmapped grant: %u\n",
1468					 s->grants_used[i]->gref);
1469				return -1;
1470			}
1471			list_add(&s->grants_used[i]->node, &rinfo->grants);
1472			rinfo->persistent_gnts_c++;
1473		} else {
1474			/*
1475			 * If the grant is not mapped by the backend we add it
1476			 * to the tail of the list, so it will not be picked
1477			 * again unless we run out of persistent grants.
1478			 */
1479			s->grants_used[i]->gref = INVALID_GRANT_REF;
1480			list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1481		}
1482	}
1483	if (s->req.operation == BLKIF_OP_INDIRECT) {
1484		for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1485			if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1486				if (!info->feature_persistent) {
1487					pr_alert("backed has not unmapped grant: %u\n",
1488						 s->indirect_grants[i]->gref);
1489					return -1;
1490				}
1491				list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1492				rinfo->persistent_gnts_c++;
1493			} else {
1494				struct page *indirect_page;
1495
1496				/*
1497				 * Add the used indirect page back to the list of
1498				 * available pages for indirect grefs.
1499				 */
1500				if (!info->bounce) {
1501					indirect_page = s->indirect_grants[i]->page;
1502					list_add(&indirect_page->lru, &rinfo->indirect_pages);
1503				}
1504				s->indirect_grants[i]->gref = INVALID_GRANT_REF;
1505				list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1506			}
1507		}
1508	}
1509
1510	return 1;
1511}
1512
1513static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1514{
1515	struct request *req;
1516	struct blkif_response bret;
1517	RING_IDX i, rp;
1518	unsigned long flags;
1519	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1520	struct blkfront_info *info = rinfo->dev_info;
1521	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1522
1523	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1524		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1525		return IRQ_HANDLED;
1526	}
1527
1528	spin_lock_irqsave(&rinfo->ring_lock, flags);
1529 again:
1530	rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1531	virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1532	if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1533		pr_alert("%s: illegal number of responses %u\n",
1534			 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1535		goto err;
1536	}
1537
1538	for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1539		unsigned long id;
1540		unsigned int op;
1541
1542		eoiflag = 0;
1543
1544		RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1545		id = bret.id;
1546
1547		/*
1548		 * The backend has messed up and given us an id that we would
1549		 * never have given to it (we stamp it up to BLK_RING_SIZE -
1550		 * look in get_id_from_freelist.
1551		 */
1552		if (id >= BLK_RING_SIZE(info)) {
1553			pr_alert("%s: response has incorrect id (%ld)\n",
1554				 info->gd->disk_name, id);
1555			goto err;
1556		}
1557		if (rinfo->shadow[id].status != REQ_WAITING) {
1558			pr_alert("%s: response references no pending request\n",
1559				 info->gd->disk_name);
1560			goto err;
1561		}
1562
1563		rinfo->shadow[id].status = REQ_PROCESSING;
1564		req  = rinfo->shadow[id].request;
1565
1566		op = rinfo->shadow[id].req.operation;
1567		if (op == BLKIF_OP_INDIRECT)
1568			op = rinfo->shadow[id].req.u.indirect.indirect_op;
1569		if (bret.operation != op) {
1570			pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1571				 info->gd->disk_name, bret.operation, op);
1572			goto err;
1573		}
1574
1575		if (bret.operation != BLKIF_OP_DISCARD) {
1576			int ret;
1577
1578			/*
1579			 * We may need to wait for an extra response if the
1580			 * I/O request is split in 2
1581			 */
1582			ret = blkif_completion(&id, rinfo, &bret);
1583			if (!ret)
1584				continue;
1585			if (unlikely(ret < 0))
1586				goto err;
1587		}
1588
1589		if (add_id_to_freelist(rinfo, id)) {
1590			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1591			     info->gd->disk_name, op_name(bret.operation), id);
1592			continue;
1593		}
1594
1595		if (bret.status == BLKIF_RSP_OKAY)
1596			blkif_req(req)->error = BLK_STS_OK;
1597		else
1598			blkif_req(req)->error = BLK_STS_IOERR;
1599
1600		switch (bret.operation) {
1601		case BLKIF_OP_DISCARD:
1602			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1603				struct request_queue *rq = info->rq;
1604
1605				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1606					   info->gd->disk_name, op_name(bret.operation));
1607				blkif_req(req)->error = BLK_STS_NOTSUPP;
1608				info->feature_discard = 0;
1609				info->feature_secdiscard = 0;
1610				blk_queue_disable_discard(rq);
1611				blk_queue_disable_secure_erase(rq);
1612			}
1613			break;
1614		case BLKIF_OP_FLUSH_DISKCACHE:
1615		case BLKIF_OP_WRITE_BARRIER:
1616			if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1617				pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1618				       info->gd->disk_name, op_name(bret.operation));
1619				blkif_req(req)->error = BLK_STS_NOTSUPP;
1620			}
1621			if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1622				     rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1623				pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1624				       info->gd->disk_name, op_name(bret.operation));
1625				blkif_req(req)->error = BLK_STS_NOTSUPP;
1626			}
1627			if (unlikely(blkif_req(req)->error)) {
1628				if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1629					blkif_req(req)->error = BLK_STS_OK;
1630				info->feature_fua = 0;
1631				info->feature_flush = 0;
1632			}
1633			fallthrough;
1634		case BLKIF_OP_READ:
1635		case BLKIF_OP_WRITE:
1636			if (unlikely(bret.status != BLKIF_RSP_OKAY))
1637				dev_dbg_ratelimited(&info->xbdev->dev,
1638					"Bad return from blkdev data request: %#x\n",
1639					bret.status);
1640
1641			break;
1642		default:
1643			BUG();
1644		}
1645
1646		if (likely(!blk_should_fake_timeout(req->q)))
1647			blk_mq_complete_request(req);
1648	}
1649
1650	rinfo->ring.rsp_cons = i;
1651
1652	if (i != rinfo->ring.req_prod_pvt) {
1653		int more_to_do;
1654		RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1655		if (more_to_do)
1656			goto again;
1657	} else
1658		rinfo->ring.sring->rsp_event = i + 1;
1659
1660	kick_pending_request_queues_locked(rinfo);
1661
1662	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1663
1664	xen_irq_lateeoi(irq, eoiflag);
1665
1666	return IRQ_HANDLED;
1667
1668 err:
1669	info->connected = BLKIF_STATE_ERROR;
1670
1671	spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1672
1673	/* No EOI in order to avoid further interrupts. */
1674
1675	pr_alert("%s disabled for further use\n", info->gd->disk_name);
1676	return IRQ_HANDLED;
1677}
1678
1679
1680static int setup_blkring(struct xenbus_device *dev,
1681			 struct blkfront_ring_info *rinfo)
1682{
1683	struct blkif_sring *sring;
1684	int err;
1685	struct blkfront_info *info = rinfo->dev_info;
1686	unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1687
1688	err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring,
1689				info->nr_ring_pages, rinfo->ring_ref);
1690	if (err)
1691		goto fail;
1692
1693	XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1694
1695	err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1696	if (err)
1697		goto fail;
1698
1699	err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1700						0, "blkif", rinfo);
1701	if (err <= 0) {
1702		xenbus_dev_fatal(dev, err,
1703				 "bind_evtchn_to_irqhandler failed");
1704		goto fail;
1705	}
1706	rinfo->irq = err;
1707
1708	return 0;
1709fail:
1710	blkif_free(info, 0);
1711	return err;
1712}
1713
1714/*
1715 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1716 * ring buffer may have multi pages depending on ->nr_ring_pages.
1717 */
1718static int write_per_ring_nodes(struct xenbus_transaction xbt,
1719				struct blkfront_ring_info *rinfo, const char *dir)
1720{
1721	int err;
1722	unsigned int i;
1723	const char *message = NULL;
1724	struct blkfront_info *info = rinfo->dev_info;
1725
1726	if (info->nr_ring_pages == 1) {
1727		err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1728		if (err) {
1729			message = "writing ring-ref";
1730			goto abort_transaction;
1731		}
1732	} else {
1733		for (i = 0; i < info->nr_ring_pages; i++) {
1734			char ring_ref_name[RINGREF_NAME_LEN];
1735
1736			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1737			err = xenbus_printf(xbt, dir, ring_ref_name,
1738					    "%u", rinfo->ring_ref[i]);
1739			if (err) {
1740				message = "writing ring-ref";
1741				goto abort_transaction;
1742			}
1743		}
1744	}
1745
1746	err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1747	if (err) {
1748		message = "writing event-channel";
1749		goto abort_transaction;
1750	}
1751
1752	return 0;
1753
1754abort_transaction:
1755	xenbus_transaction_end(xbt, 1);
1756	if (message)
1757		xenbus_dev_fatal(info->xbdev, err, "%s", message);
1758
1759	return err;
1760}
1761
1762/* Enable the persistent grants feature. */
1763static bool feature_persistent = true;
1764module_param(feature_persistent, bool, 0644);
1765MODULE_PARM_DESC(feature_persistent,
1766		"Enables the persistent grants feature");
1767
1768/* Common code used when first setting up, and when resuming. */
1769static int talk_to_blkback(struct xenbus_device *dev,
1770			   struct blkfront_info *info)
1771{
1772	const char *message = NULL;
1773	struct xenbus_transaction xbt;
1774	int err;
1775	unsigned int i, max_page_order;
1776	unsigned int ring_page_order;
1777	struct blkfront_ring_info *rinfo;
1778
1779	if (!info)
1780		return -ENODEV;
1781
1782	/* Check if backend is trusted. */
1783	info->bounce = !xen_blkif_trusted ||
1784		       !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1785
1786	max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1787					      "max-ring-page-order", 0);
1788	ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1789	info->nr_ring_pages = 1 << ring_page_order;
1790
1791	err = negotiate_mq(info);
1792	if (err)
1793		goto destroy_blkring;
1794
1795	for_each_rinfo(info, rinfo, i) {
1796		/* Create shared ring, alloc event channel. */
1797		err = setup_blkring(dev, rinfo);
1798		if (err)
1799			goto destroy_blkring;
1800	}
1801
1802again:
1803	err = xenbus_transaction_start(&xbt);
1804	if (err) {
1805		xenbus_dev_fatal(dev, err, "starting transaction");
1806		goto destroy_blkring;
1807	}
1808
1809	if (info->nr_ring_pages > 1) {
1810		err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1811				    ring_page_order);
1812		if (err) {
1813			message = "writing ring-page-order";
1814			goto abort_transaction;
1815		}
1816	}
1817
1818	/* We already got the number of queues/rings in _probe */
1819	if (info->nr_rings == 1) {
1820		err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1821		if (err)
1822			goto destroy_blkring;
1823	} else {
1824		char *path;
1825		size_t pathsize;
1826
1827		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1828				    info->nr_rings);
1829		if (err) {
1830			message = "writing multi-queue-num-queues";
1831			goto abort_transaction;
1832		}
1833
1834		pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1835		path = kmalloc(pathsize, GFP_KERNEL);
1836		if (!path) {
1837			err = -ENOMEM;
1838			message = "ENOMEM while writing ring references";
1839			goto abort_transaction;
1840		}
1841
1842		for_each_rinfo(info, rinfo, i) {
1843			memset(path, 0, pathsize);
1844			snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1845			err = write_per_ring_nodes(xbt, rinfo, path);
1846			if (err) {
1847				kfree(path);
1848				goto destroy_blkring;
1849			}
1850		}
1851		kfree(path);
1852	}
1853	err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1854			    XEN_IO_PROTO_ABI_NATIVE);
1855	if (err) {
1856		message = "writing protocol";
1857		goto abort_transaction;
1858	}
1859	info->feature_persistent_parm = feature_persistent;
1860	err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1861			info->feature_persistent_parm);
1862	if (err)
1863		dev_warn(&dev->dev,
1864			 "writing persistent grants feature to xenbus");
1865
1866	err = xenbus_transaction_end(xbt, 0);
1867	if (err) {
1868		if (err == -EAGAIN)
1869			goto again;
1870		xenbus_dev_fatal(dev, err, "completing transaction");
1871		goto destroy_blkring;
1872	}
1873
1874	for_each_rinfo(info, rinfo, i) {
1875		unsigned int j;
1876
1877		for (j = 0; j < BLK_RING_SIZE(info); j++)
1878			rinfo->shadow[j].req.u.rw.id = j + 1;
1879		rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1880	}
1881	xenbus_switch_state(dev, XenbusStateInitialised);
1882
1883	return 0;
1884
1885 abort_transaction:
1886	xenbus_transaction_end(xbt, 1);
1887	if (message)
1888		xenbus_dev_fatal(dev, err, "%s", message);
1889 destroy_blkring:
1890	blkif_free(info, 0);
1891	return err;
1892}
1893
1894static int negotiate_mq(struct blkfront_info *info)
1895{
1896	unsigned int backend_max_queues;
1897	unsigned int i;
1898	struct blkfront_ring_info *rinfo;
1899
1900	BUG_ON(info->nr_rings);
1901
1902	/* Check if backend supports multiple queues. */
1903	backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1904						  "multi-queue-max-queues", 1);
1905	info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1906	/* We need at least one ring. */
1907	if (!info->nr_rings)
1908		info->nr_rings = 1;
1909
1910	info->rinfo_size = struct_size(info->rinfo, shadow,
1911				       BLK_RING_SIZE(info));
1912	info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1913	if (!info->rinfo) {
1914		xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1915		info->nr_rings = 0;
1916		return -ENOMEM;
1917	}
1918
1919	for_each_rinfo(info, rinfo, i) {
1920		INIT_LIST_HEAD(&rinfo->indirect_pages);
1921		INIT_LIST_HEAD(&rinfo->grants);
1922		rinfo->dev_info = info;
1923		INIT_WORK(&rinfo->work, blkif_restart_queue);
1924		spin_lock_init(&rinfo->ring_lock);
1925	}
1926	return 0;
1927}
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 */
1935static 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	mutex_lock(&blkfront_mutex);
2000	list_add(&info->info_list, &info_list);
2001	mutex_unlock(&blkfront_mutex);
2002
2003	return 0;
2004}
2005
2006static int blkif_recover(struct blkfront_info *info)
2007{
2008	struct queue_limits lim;
2009	unsigned int r_index;
2010	struct request *req, *n;
2011	int rc;
2012	struct bio *bio;
2013	struct blkfront_ring_info *rinfo;
2014
2015	lim = queue_limits_start_update(info->rq);
2016	blkfront_gather_backend_features(info);
2017	blkif_set_queue_limits(info, &lim);
2018	rc = queue_limits_commit_update(info->rq, &lim);
2019	if (rc)
2020		return rc;
2021
2022	for_each_rinfo(info, rinfo, r_index) {
2023		rc = blkfront_setup_indirect(rinfo);
2024		if (rc)
2025			return rc;
2026	}
2027	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2028
2029	/* Now safe for us to use the shared ring */
2030	info->connected = BLKIF_STATE_CONNECTED;
2031
2032	for_each_rinfo(info, rinfo, r_index) {
2033		/* Kick any other new requests queued since we resumed */
2034		kick_pending_request_queues(rinfo);
2035	}
2036
2037	list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2038		/* Requeue pending requests (flush or discard) */
2039		list_del_init(&req->queuelist);
2040		BUG_ON(req->nr_phys_segments >
2041		       (info->max_indirect_segments ? :
2042			BLKIF_MAX_SEGMENTS_PER_REQUEST));
2043		blk_mq_requeue_request(req, false);
2044	}
2045	blk_mq_start_stopped_hw_queues(info->rq, true);
2046	blk_mq_kick_requeue_list(info->rq);
2047
2048	while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2049		/* Traverse the list of pending bios and re-queue them */
2050		submit_bio(bio);
2051	}
2052
2053	return 0;
2054}
2055
2056/*
2057 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2058 * driver restart.  We tear down our blkif structure and recreate it, but
2059 * leave the device-layer structures intact so that this is transparent to the
2060 * rest of the kernel.
2061 */
2062static int blkfront_resume(struct xenbus_device *dev)
2063{
2064	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2065	int err = 0;
2066	unsigned int i, j;
2067	struct blkfront_ring_info *rinfo;
2068
2069	dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2070
2071	bio_list_init(&info->bio_list);
2072	INIT_LIST_HEAD(&info->requests);
2073	for_each_rinfo(info, rinfo, i) {
2074		struct bio_list merge_bio;
2075		struct blk_shadow *shadow = rinfo->shadow;
2076
2077		for (j = 0; j < BLK_RING_SIZE(info); j++) {
2078			/* Not in use? */
2079			if (!shadow[j].request)
2080				continue;
2081
2082			/*
2083			 * Get the bios in the request so we can re-queue them.
2084			 */
2085			if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2086			    req_op(shadow[j].request) == REQ_OP_DISCARD ||
2087			    req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2088			    shadow[j].request->cmd_flags & REQ_FUA) {
2089				/*
2090				 * Flush operations don't contain bios, so
2091				 * we need to requeue the whole request
2092				 *
2093				 * XXX: but this doesn't make any sense for a
2094				 * write with the FUA flag set..
2095				 */
2096				list_add(&shadow[j].request->queuelist, &info->requests);
2097				continue;
2098			}
2099			merge_bio.head = shadow[j].request->bio;
2100			merge_bio.tail = shadow[j].request->biotail;
2101			bio_list_merge(&info->bio_list, &merge_bio);
2102			shadow[j].request->bio = NULL;
2103			blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2104		}
2105	}
2106
2107	blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2108
2109	err = talk_to_blkback(dev, info);
2110	if (!err)
2111		blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2112
2113	/*
2114	 * We have to wait for the backend to switch to
2115	 * connected state, since we want to read which
2116	 * features it supports.
2117	 */
2118
2119	return err;
2120}
2121
2122static void blkfront_closing(struct blkfront_info *info)
2123{
2124	struct xenbus_device *xbdev = info->xbdev;
2125	struct blkfront_ring_info *rinfo;
2126	unsigned int i;
2127
2128	if (xbdev->state == XenbusStateClosing)
2129		return;
2130
2131	/* No more blkif_request(). */
2132	if (info->rq && info->gd) {
2133		blk_mq_stop_hw_queues(info->rq);
2134		blk_mark_disk_dead(info->gd);
2135	}
2136
2137	for_each_rinfo(info, rinfo, i) {
2138		/* No more gnttab callback work. */
2139		gnttab_cancel_free_callback(&rinfo->callback);
2140
2141		/* Flush gnttab callback work. Must be done with no locks held. */
2142		flush_work(&rinfo->work);
2143	}
2144
2145	xenbus_frontend_closed(xbdev);
2146}
2147
2148static void blkfront_setup_discard(struct blkfront_info *info)
2149{
2150	info->feature_discard = 1;
2151	info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2152							 "discard-granularity",
2153							 0);
2154	info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2155						       "discard-alignment", 0);
2156	info->feature_secdiscard =
2157		!!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2158				       0);
2159}
2160
2161static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2162{
2163	unsigned int psegs, grants, memflags;
2164	int err, i;
2165	struct blkfront_info *info = rinfo->dev_info;
2166
2167	memflags = memalloc_noio_save();
2168
2169	if (info->max_indirect_segments == 0) {
2170		if (!HAS_EXTRA_REQ)
2171			grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2172		else {
2173			/*
2174			 * When an extra req is required, the maximum
2175			 * grants supported is related to the size of the
2176			 * Linux block segment.
2177			 */
2178			grants = GRANTS_PER_PSEG;
2179		}
2180	}
2181	else
2182		grants = info->max_indirect_segments;
2183	psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2184
2185	err = fill_grant_buffer(rinfo,
2186				(grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2187	if (err)
2188		goto out_of_memory;
2189
2190	if (!info->bounce && info->max_indirect_segments) {
2191		/*
2192		 * We are using indirect descriptors but don't have a bounce
2193		 * buffer, we need to allocate a set of pages that can be
2194		 * used for mapping indirect grefs
2195		 */
2196		int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2197
2198		BUG_ON(!list_empty(&rinfo->indirect_pages));
2199		for (i = 0; i < num; i++) {
2200			struct page *indirect_page = alloc_page(GFP_KERNEL |
2201								__GFP_ZERO);
2202			if (!indirect_page)
2203				goto out_of_memory;
2204			list_add(&indirect_page->lru, &rinfo->indirect_pages);
2205		}
2206	}
2207
2208	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2209		rinfo->shadow[i].grants_used =
2210			kvcalloc(grants,
2211				 sizeof(rinfo->shadow[i].grants_used[0]),
2212				 GFP_KERNEL);
2213		rinfo->shadow[i].sg = kvcalloc(psegs,
2214					       sizeof(rinfo->shadow[i].sg[0]),
2215					       GFP_KERNEL);
2216		if (info->max_indirect_segments)
2217			rinfo->shadow[i].indirect_grants =
2218				kvcalloc(INDIRECT_GREFS(grants),
2219					 sizeof(rinfo->shadow[i].indirect_grants[0]),
2220					 GFP_KERNEL);
2221		if ((rinfo->shadow[i].grants_used == NULL) ||
2222			(rinfo->shadow[i].sg == NULL) ||
2223		     (info->max_indirect_segments &&
2224		     (rinfo->shadow[i].indirect_grants == NULL)))
2225			goto out_of_memory;
2226		sg_init_table(rinfo->shadow[i].sg, psegs);
2227	}
2228
2229	memalloc_noio_restore(memflags);
2230
2231	return 0;
2232
2233out_of_memory:
2234	for (i = 0; i < BLK_RING_SIZE(info); i++) {
2235		kvfree(rinfo->shadow[i].grants_used);
2236		rinfo->shadow[i].grants_used = NULL;
2237		kvfree(rinfo->shadow[i].sg);
2238		rinfo->shadow[i].sg = NULL;
2239		kvfree(rinfo->shadow[i].indirect_grants);
2240		rinfo->shadow[i].indirect_grants = NULL;
2241	}
2242	if (!list_empty(&rinfo->indirect_pages)) {
2243		struct page *indirect_page, *n;
2244		list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2245			list_del(&indirect_page->lru);
2246			__free_page(indirect_page);
2247		}
2248	}
2249
2250	memalloc_noio_restore(memflags);
2251
2252	return -ENOMEM;
2253}
2254
2255/*
2256 * Gather all backend feature-*
2257 */
2258static void blkfront_gather_backend_features(struct blkfront_info *info)
2259{
2260	unsigned int indirect_segments;
2261
2262	info->feature_flush = 0;
2263	info->feature_fua = 0;
2264
2265	/*
2266	 * If there's no "feature-barrier" defined, then it means
2267	 * we're dealing with a very old backend which writes
2268	 * synchronously; nothing to do.
2269	 *
2270	 * If there are barriers, then we use flush.
2271	 */
2272	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2273		info->feature_flush = 1;
2274		info->feature_fua = 1;
2275	}
2276
2277	/*
2278	 * And if there is "feature-flush-cache" use that above
2279	 * barriers.
2280	 */
2281	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2282				 0)) {
2283		info->feature_flush = 1;
2284		info->feature_fua = 0;
2285	}
2286
2287	if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2288		blkfront_setup_discard(info);
2289
2290	if (info->feature_persistent_parm)
2291		info->feature_persistent =
2292			!!xenbus_read_unsigned(info->xbdev->otherend,
2293					       "feature-persistent", 0);
2294	if (info->feature_persistent)
2295		info->bounce = true;
2296
2297	indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2298					"feature-max-indirect-segments", 0);
2299	if (indirect_segments > xen_blkif_max_segments)
2300		indirect_segments = xen_blkif_max_segments;
2301	if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2302		indirect_segments = 0;
2303	info->max_indirect_segments = indirect_segments;
2304
2305	if (info->feature_persistent) {
2306		mutex_lock(&blkfront_mutex);
2307		schedule_delayed_work(&blkfront_work, HZ * 10);
2308		mutex_unlock(&blkfront_mutex);
2309	}
2310}
2311
2312/*
2313 * Invoked when the backend is finally 'ready' (and has told produced
2314 * the details about the physical device - #sectors, size, etc).
2315 */
2316static void blkfront_connect(struct blkfront_info *info)
2317{
2318	unsigned long long sectors;
2319	int err, i;
2320	struct blkfront_ring_info *rinfo;
2321
2322	switch (info->connected) {
2323	case BLKIF_STATE_CONNECTED:
2324		/*
2325		 * Potentially, the back-end may be signalling
2326		 * a capacity change; update the capacity.
2327		 */
2328		err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2329				   "sectors", "%Lu", &sectors);
2330		if (XENBUS_EXIST_ERR(err))
2331			return;
2332		printk(KERN_INFO "Setting capacity to %Lu\n",
2333		       sectors);
2334		set_capacity_and_notify(info->gd, sectors);
2335
2336		return;
2337	case BLKIF_STATE_SUSPENDED:
2338		/*
2339		 * If we are recovering from suspension, we need to wait
2340		 * for the backend to announce it's features before
2341		 * reconnecting, at least we need to know if the backend
2342		 * supports indirect descriptors, and how many.
2343		 */
2344		blkif_recover(info);
2345		return;
2346
2347	default:
2348		break;
2349	}
2350
2351	dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2352		__func__, info->xbdev->otherend);
2353
2354	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2355			    "sectors", "%llu", &sectors,
2356			    "info", "%u", &info->vdisk_info,
2357			    "sector-size", "%lu", &info->sector_size,
2358			    NULL);
2359	if (err) {
2360		xenbus_dev_fatal(info->xbdev, err,
2361				 "reading backend fields at %s",
2362				 info->xbdev->otherend);
2363		return;
2364	}
2365
2366	/*
2367	 * physical-sector-size is a newer field, so old backends may not
2368	 * provide this. Assume physical sector size to be the same as
2369	 * sector_size in that case.
2370	 */
2371	info->physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2372						    "physical-sector-size",
2373						    info->sector_size);
2374	blkfront_gather_backend_features(info);
2375	for_each_rinfo(info, rinfo, i) {
2376		err = blkfront_setup_indirect(rinfo);
2377		if (err) {
2378			xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2379					 info->xbdev->otherend);
2380			blkif_free(info, 0);
2381			break;
2382		}
2383	}
2384
2385	err = xlvbd_alloc_gendisk(sectors, info);
2386	if (err) {
2387		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2388				 info->xbdev->otherend);
2389		goto fail;
2390	}
2391
2392	xenbus_switch_state(info->xbdev, XenbusStateConnected);
2393
2394	/* Kick pending requests. */
2395	info->connected = BLKIF_STATE_CONNECTED;
2396	for_each_rinfo(info, rinfo, i)
2397		kick_pending_request_queues(rinfo);
2398
2399	err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2400	if (err) {
2401		put_disk(info->gd);
2402		blk_mq_free_tag_set(&info->tag_set);
2403		info->rq = NULL;
2404		goto fail;
2405	}
2406
2407	info->is_ready = 1;
2408	return;
2409
2410fail:
2411	blkif_free(info, 0);
2412	return;
2413}
2414
2415/*
2416 * Callback received when the backend's state changes.
2417 */
2418static 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		break;
2432	case XenbusStateInitialising:
2433	case XenbusStateInitialised:
2434	case XenbusStateReconfiguring:
2435	case XenbusStateReconfigured:
2436	case XenbusStateUnknown:
2437		break;
2438
2439	case XenbusStateConnected:
2440		/*
2441		 * talk_to_blkback sets state to XenbusStateInitialised
2442		 * and blkfront_connect sets it to XenbusStateConnected
2443		 * (if connection went OK).
2444		 *
2445		 * If the backend (or toolstack) decides to poke at backend
2446		 * state (and re-trigger the watch by setting the state repeatedly
2447		 * to XenbusStateConnected (4)) we need to deal with this.
2448		 * This is allowed as this is used to communicate to the guest
2449		 * that the size of disk has changed!
2450		 */
2451		if ((dev->state != XenbusStateInitialised) &&
2452		    (dev->state != XenbusStateConnected)) {
2453			if (talk_to_blkback(dev, info))
2454				break;
2455		}
2456
2457		blkfront_connect(info);
2458		break;
2459
2460	case XenbusStateClosed:
2461		if (dev->state == XenbusStateClosed)
2462			break;
2463		fallthrough;
2464	case XenbusStateClosing:
2465		blkfront_closing(info);
2466		break;
2467	}
2468}
2469
2470static void blkfront_remove(struct xenbus_device *xbdev)
2471{
2472	struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2473
2474	dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2475
2476	if (info->gd)
2477		del_gendisk(info->gd);
2478
2479	mutex_lock(&blkfront_mutex);
2480	list_del(&info->info_list);
2481	mutex_unlock(&blkfront_mutex);
2482
2483	blkif_free(info, 0);
2484	if (info->gd) {
2485		xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2486		put_disk(info->gd);
2487		blk_mq_free_tag_set(&info->tag_set);
2488	}
2489
2490	kfree(info);
2491}
2492
2493static int blkfront_is_ready(struct xenbus_device *dev)
2494{
2495	struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2496
2497	return info->is_ready && info->xbdev;
2498}
2499
2500static const struct block_device_operations xlvbd_block_fops =
2501{
2502	.owner = THIS_MODULE,
2503	.getgeo = blkif_getgeo,
2504	.ioctl = blkif_ioctl,
2505	.compat_ioctl = blkdev_compat_ptr_ioctl,
2506};
2507
2508
2509static const struct xenbus_device_id blkfront_ids[] = {
2510	{ "vbd" },
2511	{ "" }
2512};
2513
2514static struct xenbus_driver blkfront_driver = {
2515	.ids  = blkfront_ids,
2516	.probe = blkfront_probe,
2517	.remove = blkfront_remove,
2518	.resume = blkfront_resume,
2519	.otherend_changed = blkback_changed,
2520	.is_ready = blkfront_is_ready,
2521};
2522
2523static void purge_persistent_grants(struct blkfront_info *info)
2524{
2525	unsigned int i;
2526	unsigned long flags;
2527	struct blkfront_ring_info *rinfo;
2528
2529	for_each_rinfo(info, rinfo, i) {
2530		struct grant *gnt_list_entry, *tmp;
2531		LIST_HEAD(grants);
2532
2533		spin_lock_irqsave(&rinfo->ring_lock, flags);
2534
2535		if (rinfo->persistent_gnts_c == 0) {
2536			spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2537			continue;
2538		}
2539
2540		list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2541					 node) {
2542			if (gnt_list_entry->gref == INVALID_GRANT_REF ||
2543			    !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2544				continue;
2545
2546			list_del(&gnt_list_entry->node);
2547			rinfo->persistent_gnts_c--;
2548			gnt_list_entry->gref = INVALID_GRANT_REF;
2549			list_add_tail(&gnt_list_entry->node, &grants);
2550		}
2551
2552		list_splice_tail(&grants, &rinfo->grants);
2553
2554		spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2555	}
2556}
2557
2558static void blkfront_delay_work(struct work_struct *work)
2559{
2560	struct blkfront_info *info;
2561	bool need_schedule_work = false;
2562
2563	/*
2564	 * Note that when using bounce buffers but not persistent grants
2565	 * there's no need to run blkfront_delay_work because grants are
2566	 * revoked in blkif_completion or else an error is reported and the
2567	 * connection is closed.
2568	 */
2569
2570	mutex_lock(&blkfront_mutex);
2571
2572	list_for_each_entry(info, &info_list, info_list) {
2573		if (info->feature_persistent) {
2574			need_schedule_work = true;
2575			mutex_lock(&info->mutex);
2576			purge_persistent_grants(info);
2577			mutex_unlock(&info->mutex);
2578		}
2579	}
2580
2581	if (need_schedule_work)
2582		schedule_delayed_work(&blkfront_work, HZ * 10);
2583
2584	mutex_unlock(&blkfront_mutex);
2585}
2586
2587static int __init xlblk_init(void)
2588{
2589	int ret;
2590	int nr_cpus = num_online_cpus();
2591
2592	if (!xen_domain())
2593		return -ENODEV;
2594
2595	if (!xen_has_pv_disk_devices())
2596		return -ENODEV;
2597
2598	if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2599		pr_warn("xen_blk: can't get major %d with name %s\n",
2600			XENVBD_MAJOR, DEV_NAME);
2601		return -ENODEV;
2602	}
2603
2604	if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2605		xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2606
2607	if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2608		pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2609			xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2610		xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2611	}
2612
2613	if (xen_blkif_max_queues > nr_cpus) {
2614		pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2615			xen_blkif_max_queues, nr_cpus);
2616		xen_blkif_max_queues = nr_cpus;
2617	}
2618
2619	INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2620
2621	ret = xenbus_register_frontend(&blkfront_driver);
2622	if (ret) {
2623		unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2624		return ret;
2625	}
2626
2627	return 0;
2628}
2629module_init(xlblk_init);
2630
2631
2632static void __exit xlblk_exit(void)
2633{
2634	cancel_delayed_work_sync(&blkfront_work);
2635
2636	xenbus_unregister_driver(&blkfront_driver);
2637	unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2638	kfree(minors);
2639}
2640module_exit(xlblk_exit);
2641
2642MODULE_DESCRIPTION("Xen virtual block device frontend");
2643MODULE_LICENSE("GPL");
2644MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2645MODULE_ALIAS("xen:vbd");
2646MODULE_ALIAS("xenblk");