drivers/block/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/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/sched/mm.h>
  51
  52 #include <xen/xen.h>
  53 #include <xen/xenbus.h>
  54 #include <xen/grant_table.h>
  55 #include <xen/events.h>
  56 #include <xen/page.h>
  57 #include <xen/platform_pci.h>
  58
  59 #include <xen/interface/grant_table.h>
  60 #include <xen/interface/io/blkif.h>
  61 #include <xen/interface/io/protocols.h>
  62
  63 #include <asm/xen/hypervisor.h>
  64
  65 /*
  66  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  67  * When Linux is using a different page size than Xen, it may not be possible
  68  * to put all the data in a single segment.
  69  * This can happen when the backend doesn't support indirect descriptor and
  70  * therefore the maximum amount of data that a request can carry is
  71  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  72  *
  73  * Note that we only support one extra request. So the Linux page size
  74  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  75  * 88KB.
  76  */
  77 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  78
  79 enum blkif_state {
  80         BLKIF_STATE_DISCONNECTED,
  81         BLKIF_STATE_CONNECTED,
  82         BLKIF_STATE_SUSPENDED,
  83 };
  84
  85 struct grant {
  86         grant_ref_t gref;
  87         struct page *page;
  88         struct list_head node;
  89 };
  90
  91 enum blk_req_status {
  92         REQ_WAITING,
  93         REQ_DONE,
  94         REQ_ERROR,
  95         REQ_EOPNOTSUPP,
  96 };
  97
  98 struct blk_shadow {
  99         struct blkif_request req;
 100         struct request *request;
 101         struct grant **grants_used;
 102         struct grant **indirect_grants;
 103         struct scatterlist *sg;
 104         unsigned int num_sg;
 105         enum blk_req_status status;
 106
 107         #define NO_ASSOCIATED_ID ~0UL
 108         /*
 109          * Id of the sibling if we ever need 2 requests when handling a
 110          * block I/O request
 111          */
 112         unsigned long associated_id;
 113 };
 114
 115 struct blkif_req {
 116         blk_status_t    error;
 117 };
 118
 119 static inline struct blkif_req *blkif_req(struct request *rq)
 120 {
 121         return blk_mq_rq_to_pdu(rq);
 122 }
 123
 124 static DEFINE_MUTEX(blkfront_mutex);
 125 static const struct block_device_operations xlvbd_block_fops;
 126 static struct delayed_work blkfront_work;
 127 static LIST_HEAD(info_list);
 128
 129 /*
 130  * Maximum number of segments in indirect requests, the actual value used by
 131  * the frontend driver is the minimum of this value and the value provided
 132  * by the backend driver.
 133  */
 134
 135 static unsigned int xen_blkif_max_segments = 32;
 136 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 137 MODULE_PARM_DESC(max_indirect_segments,
 138                  "Maximum amount of segments in indirect requests (default is 32)");
 139
 140 static unsigned int xen_blkif_max_queues = 4;
 141 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 142 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 143
 144 /*
 145  * Maximum order of pages to be used for the shared ring between front and
 146  * backend, 4KB page granularity is used.
 147  */
 148 static unsigned int xen_blkif_max_ring_order;
 149 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 150 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 151
 152 #define BLK_RING_SIZE(info)     \
 153         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 154
 155 /*
 156  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 157  * characters are enough. Define to 20 to keep consistent with backend.
 158  */
 159 #define RINGREF_NAME_LEN (20)
 160 /*
 161  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 162  */
 163 #define QUEUE_NAME_LEN (17)
 164
 165 /*
 166  *  Per-ring info.
 167  *  Every blkfront device can associate with one or more blkfront_ring_info,
 168  *  depending on how many hardware queues/rings to be used.
 169  */
 170 struct blkfront_ring_info {
 171         /* Lock to protect data in every ring buffer. */
 172         spinlock_t ring_lock;
 173         struct blkif_front_ring ring;
 174         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 175         unsigned int evtchn, irq;
 176         struct work_struct work;
 177         struct gnttab_free_callback callback;
 178         struct list_head indirect_pages;
 179         struct list_head grants;
 180         unsigned int persistent_gnts_c;
 181         unsigned long shadow_free;
 182         struct blkfront_info *dev_info;
 183         struct blk_shadow shadow[];
 184 };
 185
 186 /*
 187  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 188  * hang in private_data off the gendisk structure. We may end up
 189  * putting all kinds of interesting stuff here :-)
 190  */
 191 struct blkfront_info
 192 {
 193         struct mutex mutex;
 194         struct xenbus_device *xbdev;
 195         struct gendisk *gd;
 196         u16 sector_size;
 197         unsigned int physical_sector_size;
 198         int vdevice;
 199         blkif_vdev_t handle;
 200         enum blkif_state connected;
 201         /* Number of pages per ring buffer. */
 202         unsigned int nr_ring_pages;
 203         struct request_queue *rq;
 204         unsigned int feature_flush:1;
 205         unsigned int feature_fua:1;
 206         unsigned int feature_discard:1;
 207         unsigned int feature_secdiscard:1;
 208         unsigned int feature_persistent:1;
 209         unsigned int discard_granularity;
 210         unsigned int discard_alignment;
 211         /* Number of 4KB segments handled */
 212         unsigned int max_indirect_segments;
 213         int is_ready;
 214         struct blk_mq_tag_set tag_set;
 215         struct blkfront_ring_info *rinfo;
 216         unsigned int nr_rings;
 217         unsigned int rinfo_size;
 218         /* Save uncomplete reqs and bios for migration. */
 219         struct list_head requests;
 220         struct bio_list bio_list;
 221         struct list_head info_list;
 222 };
 223
 224 static unsigned int nr_minors;
 225 static unsigned long *minors;
 226 static DEFINE_SPINLOCK(minor_lock);
 227
 228 #define GRANT_INVALID_REF       0
 229
 230 #define PARTS_PER_DISK          16
 231 #define PARTS_PER_EXT_DISK      256
 232
 233 #define BLKIF_MAJOR(dev) ((dev)>>8)
 234 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 235
 236 #define EXT_SHIFT 28
 237 #define EXTENDED (1<<EXT_SHIFT)
 238 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 239 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 240 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 241 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 242 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 243 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 244
 245 #define DEV_NAME        "xvd"   /* name in /dev */
 246
 247 /*
 248  * Grants are always the same size as a Xen page (i.e 4KB).
 249  * A physical segment is always the same size as a Linux page.
 250  * Number of grants per physical segment
 251  */
 252 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 253
 254 #define GRANTS_PER_INDIRECT_FRAME \
 255         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 256
 257 #define INDIRECT_GREFS(_grants)         \
 258         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 259
 260 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 261 static void blkfront_gather_backend_features(struct blkfront_info *info);
 262 static int negotiate_mq(struct blkfront_info *info);
 263
 264 #define for_each_rinfo(info, ptr, idx)                          \
 265         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 266              (idx) < (info)->nr_rings;                          \
 267              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 268
 269 static inline struct blkfront_ring_info *
 270 get_rinfo(const struct blkfront_info *info, unsigned int i)
 271 {
 272         BUG_ON(i >= info->nr_rings);
 273         return (void *)info->rinfo + i * info->rinfo_size;
 274 }
 275
 276 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 277 {
 278         unsigned long free = rinfo->shadow_free;
 279
 280         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 281         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 282         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 283         return free;
 284 }
 285
 286 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 287                               unsigned long id)
 288 {
 289         if (rinfo->shadow[id].req.u.rw.id != id)
 290                 return -EINVAL;
 291         if (rinfo->shadow[id].request == NULL)
 292                 return -EINVAL;
 293         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 294         rinfo->shadow[id].request = NULL;
 295         rinfo->shadow_free = id;
 296         return 0;
 297 }
 298
 299 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 300 {
 301         struct blkfront_info *info = rinfo->dev_info;
 302         struct page *granted_page;
 303         struct grant *gnt_list_entry, *n;
 304         int i = 0;
 305
 306         while (i < num) {
 307                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 308                 if (!gnt_list_entry)
 309                         goto out_of_memory;
 310
 311                 if (info->feature_persistent) {
 312                         granted_page = alloc_page(GFP_NOIO);
 313                         if (!granted_page) {
 314                                 kfree(gnt_list_entry);
 315                                 goto out_of_memory;
 316                         }
 317                         gnt_list_entry->page = granted_page;
 318                 }
 319
 320                 gnt_list_entry->gref = GRANT_INVALID_REF;
 321                 list_add(&gnt_list_entry->node, &rinfo->grants);
 322                 i++;
 323         }
 324
 325         return 0;
 326
 327 out_of_memory:
 328         list_for_each_entry_safe(gnt_list_entry, n,
 329                                  &rinfo->grants, node) {
 330                 list_del(&gnt_list_entry->node);
 331                 if (info->feature_persistent)
 332                         __free_page(gnt_list_entry->page);
 333                 kfree(gnt_list_entry);
 334                 i--;
 335         }
 336         BUG_ON(i != 0);
 337         return -ENOMEM;
 338 }
 339
 340 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 341 {
 342         struct grant *gnt_list_entry;
 343
 344         BUG_ON(list_empty(&rinfo->grants));
 345         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 346                                           node);
 347         list_del(&gnt_list_entry->node);
 348
 349         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 350                 rinfo->persistent_gnts_c--;
 351
 352         return gnt_list_entry;
 353 }
 354
 355 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 356                                         const struct blkfront_info *info)
 357 {
 358         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 359                                                  info->xbdev->otherend_id,
 360                                                  gnt_list_entry->page,
 361                                                  0);
 362 }
 363
 364 static struct grant *get_grant(grant_ref_t *gref_head,
 365                                unsigned long gfn,
 366                                struct blkfront_ring_info *rinfo)
 367 {
 368         struct grant *gnt_list_entry = get_free_grant(rinfo);
 369         struct blkfront_info *info = rinfo->dev_info;
 370
 371         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 372                 return gnt_list_entry;
 373
 374         /* Assign a gref to this page */
 375         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 376         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 377         if (info->feature_persistent)
 378                 grant_foreign_access(gnt_list_entry, info);
 379         else {
 380                 /* Grant access to the GFN passed by the caller */
 381                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 382                                                 info->xbdev->otherend_id,
 383                                                 gfn, 0);
 384         }
 385
 386         return gnt_list_entry;
 387 }
 388
 389 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 390                                         struct blkfront_ring_info *rinfo)
 391 {
 392         struct grant *gnt_list_entry = get_free_grant(rinfo);
 393         struct blkfront_info *info = rinfo->dev_info;
 394
 395         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 396                 return gnt_list_entry;
 397
 398         /* Assign a gref to this page */
 399         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 400         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 401         if (!info->feature_persistent) {
 402                 struct page *indirect_page;
 403
 404                 /* Fetch a pre-allocated page to use for indirect grefs */
 405                 BUG_ON(list_empty(&rinfo->indirect_pages));
 406                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 407                                                  struct page, lru);
 408                 list_del(&indirect_page->lru);
 409                 gnt_list_entry->page = indirect_page;
 410         }
 411         grant_foreign_access(gnt_list_entry, info);
 412
 413         return gnt_list_entry;
 414 }
 415
 416 static const char *op_name(int op)
 417 {
 418         static const char *const names[] = {
 419                 [BLKIF_OP_READ] = "read",
 420                 [BLKIF_OP_WRITE] = "write",
 421                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 422                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 423                 [BLKIF_OP_DISCARD] = "discard" };
 424
 425         if (op < 0 || op >= ARRAY_SIZE(names))
 426                 return "unknown";
 427
 428         if (!names[op])
 429                 return "reserved";
 430
 431         return names[op];
 432 }
 433 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 434 {
 435         unsigned int end = minor + nr;
 436         int rc;
 437
 438         if (end > nr_minors) {
 439                 unsigned long *bitmap, *old;
 440
 441                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 442                                  GFP_KERNEL);
 443                 if (bitmap == NULL)
 444                         return -ENOMEM;
 445
 446                 spin_lock(&minor_lock);
 447                 if (end > nr_minors) {
 448                         old = minors;
 449                         memcpy(bitmap, minors,
 450                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 451                         minors = bitmap;
 452                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 453                 } else
 454                         old = bitmap;
 455                 spin_unlock(&minor_lock);
 456                 kfree(old);
 457         }
 458
 459         spin_lock(&minor_lock);
 460         if (find_next_bit(minors, end, minor) >= end) {
 461                 bitmap_set(minors, minor, nr);
 462                 rc = 0;
 463         } else
 464                 rc = -EBUSY;
 465         spin_unlock(&minor_lock);
 466
 467         return rc;
 468 }
 469
 470 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 471 {
 472         unsigned int end = minor + nr;
 473
 474         BUG_ON(end > nr_minors);
 475         spin_lock(&minor_lock);
 476         bitmap_clear(minors,  minor, nr);
 477         spin_unlock(&minor_lock);
 478 }
 479
 480 static void blkif_restart_queue_callback(void *arg)
 481 {
 482         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 483         schedule_work(&rinfo->work);
 484 }
 485
 486 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 487 {
 488         /* We don't have real geometry info, but let's at least return
 489            values consistent with the size of the device */
 490         sector_t nsect = get_capacity(bd->bd_disk);
 491         sector_t cylinders = nsect;
 492
 493         hg->heads = 0xff;
 494         hg->sectors = 0x3f;
 495         sector_div(cylinders, hg->heads * hg->sectors);
 496         hg->cylinders = cylinders;
 497         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 498                 hg->cylinders = 0xffff;
 499         return 0;
 500 }
 501
 502 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 503                        unsigned command, unsigned long argument)
 504 {
 505         int i;
 506
 507         switch (command) {
 508         case CDROMMULTISESSION:
 509                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 510                         if (put_user(0, (char __user *)(argument + i)))
 511                                 return -EFAULT;
 512                 return 0;
 513         case CDROM_GET_CAPABILITY:
 514                 if (bdev->bd_disk->flags & GENHD_FL_CD)
 515                         return 0;
 516                 return -EINVAL;
 517         default:
 518                 return -EINVAL;
 519         }
 520 }
 521
 522 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 523                                             struct request *req,
 524                                             struct blkif_request **ring_req)
 525 {
 526         unsigned long id;
 527
 528         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 529         rinfo->ring.req_prod_pvt++;
 530
 531         id = get_id_from_freelist(rinfo);
 532         rinfo->shadow[id].request = req;
 533         rinfo->shadow[id].status = REQ_WAITING;
 534         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 535
 536         (*ring_req)->u.rw.id = id;
 537
 538         return id;
 539 }
 540
 541 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 542 {
 543         struct blkfront_info *info = rinfo->dev_info;
 544         struct blkif_request *ring_req;
 545         unsigned long id;
 546
 547         /* Fill out a communications ring structure. */
 548         id = blkif_ring_get_request(rinfo, req, &ring_req);
 549
 550         ring_req->operation = BLKIF_OP_DISCARD;
 551         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 552         ring_req->u.discard.id = id;
 553         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 554         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 555                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 556         else
 557                 ring_req->u.discard.flag = 0;
 558
 559         /* Keep a private copy so we can reissue requests when recovering. */
 560         rinfo->shadow[id].req = *ring_req;
 561
 562         return 0;
 563 }
 564
 565 struct setup_rw_req {
 566         unsigned int grant_idx;
 567         struct blkif_request_segment *segments;
 568         struct blkfront_ring_info *rinfo;
 569         struct blkif_request *ring_req;
 570         grant_ref_t gref_head;
 571         unsigned int id;
 572         /* Only used when persistent grant is used and it's a read request */
 573         bool need_copy;
 574         unsigned int bvec_off;
 575         char *bvec_data;
 576
 577         bool require_extra_req;
 578         struct blkif_request *extra_ring_req;
 579 };
 580
 581 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 582                                      unsigned int len, void *data)
 583 {
 584         struct setup_rw_req *setup = data;
 585         int n, ref;
 586         struct grant *gnt_list_entry;
 587         unsigned int fsect, lsect;
 588         /* Convenient aliases */
 589         unsigned int grant_idx = setup->grant_idx;
 590         struct blkif_request *ring_req = setup->ring_req;
 591         struct blkfront_ring_info *rinfo = setup->rinfo;
 592         /*
 593          * We always use the shadow of the first request to store the list
 594          * of grant associated to the block I/O request. This made the
 595          * completion more easy to handle even if the block I/O request is
 596          * split.
 597          */
 598         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 599
 600         if (unlikely(setup->require_extra_req &&
 601                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 602                 /*
 603                  * We are using the second request, setup grant_idx
 604                  * to be the index of the segment array.
 605                  */
 606                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 607                 ring_req = setup->extra_ring_req;
 608         }
 609
 610         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 611             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 612                 if (setup->segments)
 613                         kunmap_atomic(setup->segments);
 614
 615                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 616                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 617                 shadow->indirect_grants[n] = gnt_list_entry;
 618                 setup->segments = kmap_atomic(gnt_list_entry->page);
 619                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 620         }
 621
 622         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 623         ref = gnt_list_entry->gref;
 624         /*
 625          * All the grants are stored in the shadow of the first
 626          * request. Therefore we have to use the global index.
 627          */
 628         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 629
 630         if (setup->need_copy) {
 631                 void *shared_data;
 632
 633                 shared_data = kmap_atomic(gnt_list_entry->page);
 634                 /*
 635                  * this does not wipe data stored outside the
 636                  * range sg->offset..sg->offset+sg->length.
 637                  * Therefore, blkback *could* see data from
 638                  * previous requests. This is OK as long as
 639                  * persistent grants are shared with just one
 640                  * domain. It may need refactoring if this
 641                  * changes
 642                  */
 643                 memcpy(shared_data + offset,
 644                        setup->bvec_data + setup->bvec_off,
 645                        len);
 646
 647                 kunmap_atomic(shared_data);
 648                 setup->bvec_off += len;
 649         }
 650
 651         fsect = offset >> 9;
 652         lsect = fsect + (len >> 9) - 1;
 653         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 654                 ring_req->u.rw.seg[grant_idx] =
 655                         (struct blkif_request_segment) {
 656                                 .gref       = ref,
 657                                 .first_sect = fsect,
 658                                 .last_sect  = lsect };
 659         } else {
 660                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 661                         (struct blkif_request_segment) {
 662                                 .gref       = ref,
 663                                 .first_sect = fsect,
 664                                 .last_sect  = lsect };
 665         }
 666
 667         (setup->grant_idx)++;
 668 }
 669
 670 static void blkif_setup_extra_req(struct blkif_request *first,
 671                                   struct blkif_request *second)
 672 {
 673         uint16_t nr_segments = first->u.rw.nr_segments;
 674
 675         /*
 676          * The second request is only present when the first request uses
 677          * all its segments. It's always the continuity of the first one.
 678          */
 679         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 680
 681         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 682         second->u.rw.sector_number = first->u.rw.sector_number +
 683                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 684
 685         second->u.rw.handle = first->u.rw.handle;
 686         second->operation = first->operation;
 687 }
 688
 689 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 690 {
 691         struct blkfront_info *info = rinfo->dev_info;
 692         struct blkif_request *ring_req, *extra_ring_req = NULL;
 693         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 694         bool require_extra_req = false;
 695         int i;
 696         struct setup_rw_req setup = {
 697                 .grant_idx = 0,
 698                 .segments = NULL,
 699                 .rinfo = rinfo,
 700                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 701         };
 702
 703         /*
 704          * Used to store if we are able to queue the request by just using
 705          * existing persistent grants, or if we have to get new grants,
 706          * as there are not sufficiently many free.
 707          */
 708         bool new_persistent_gnts = false;
 709         struct scatterlist *sg;
 710         int num_sg, max_grefs, num_grant;
 711
 712         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 713         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 714                 /*
 715                  * If we are using indirect segments we need to account
 716                  * for the indirect grefs used in the request.
 717                  */
 718                 max_grefs += INDIRECT_GREFS(max_grefs);
 719
 720         /* Check if we have enough persistent grants to allocate a requests */
 721         if (rinfo->persistent_gnts_c < max_grefs) {
 722                 new_persistent_gnts = true;
 723
 724                 if (gnttab_alloc_grant_references(
 725                     max_grefs - rinfo->persistent_gnts_c,
 726                     &setup.gref_head) < 0) {
 727                         gnttab_request_free_callback(
 728                                 &rinfo->callback,
 729                                 blkif_restart_queue_callback,
 730                                 rinfo,
 731                                 max_grefs - rinfo->persistent_gnts_c);
 732                         return 1;
 733                 }
 734         }
 735
 736         /* Fill out a communications ring structure. */
 737         id = blkif_ring_get_request(rinfo, req, &ring_req);
 738
 739         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 740         num_grant = 0;
 741         /* Calculate the number of grant used */
 742         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 743                num_grant += gnttab_count_grant(sg->offset, sg->length);
 744
 745         require_extra_req = info->max_indirect_segments == 0 &&
 746                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 747         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 748
 749         rinfo->shadow[id].num_sg = num_sg;
 750         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 751             likely(!require_extra_req)) {
 752                 /*
 753                  * The indirect operation can only be a BLKIF_OP_READ or
 754                  * BLKIF_OP_WRITE
 755                  */
 756                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 757                 ring_req->operation = BLKIF_OP_INDIRECT;
 758                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 759                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 760                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 761                 ring_req->u.indirect.handle = info->handle;
 762                 ring_req->u.indirect.nr_segments = num_grant;
 763         } else {
 764                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 765                 ring_req->u.rw.handle = info->handle;
 766                 ring_req->operation = rq_data_dir(req) ?
 767                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 768                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 769                         /*
 770                          * Ideally we can do an unordered flush-to-disk.
 771                          * In case the backend onlysupports barriers, use that.
 772                          * A barrier request a superset of FUA, so we can
 773                          * implement it the same way.  (It's also a FLUSH+FUA,
 774                          * since it is guaranteed ordered WRT previous writes.)
 775                          */
 776                         if (info->feature_flush && info->feature_fua)
 777                                 ring_req->operation =
 778                                         BLKIF_OP_WRITE_BARRIER;
 779                         else if (info->feature_flush)
 780                                 ring_req->operation =
 781                                         BLKIF_OP_FLUSH_DISKCACHE;
 782                         else
 783                                 ring_req->operation = 0;
 784                 }
 785                 ring_req->u.rw.nr_segments = num_grant;
 786                 if (unlikely(require_extra_req)) {
 787                         extra_id = blkif_ring_get_request(rinfo, req,
 788                                                           &extra_ring_req);
 789                         /*
 790                          * Only the first request contains the scatter-gather
 791                          * list.
 792                          */
 793                         rinfo->shadow[extra_id].num_sg = 0;
 794
 795                         blkif_setup_extra_req(ring_req, extra_ring_req);
 796
 797                         /* Link the 2 requests together */
 798                         rinfo->shadow[extra_id].associated_id = id;
 799                         rinfo->shadow[id].associated_id = extra_id;
 800                 }
 801         }
 802
 803         setup.ring_req = ring_req;
 804         setup.id = id;
 805
 806         setup.require_extra_req = require_extra_req;
 807         if (unlikely(require_extra_req))
 808                 setup.extra_ring_req = extra_ring_req;
 809
 810         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 811                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 812
 813                 if (setup.need_copy) {
 814                         setup.bvec_off = sg->offset;
 815                         setup.bvec_data = kmap_atomic(sg_page(sg));
 816                 }
 817
 818                 gnttab_foreach_grant_in_range(sg_page(sg),
 819                                               sg->offset,
 820                                               sg->length,
 821                                               blkif_setup_rw_req_grant,
 822                                               &setup);
 823
 824                 if (setup.need_copy)
 825                         kunmap_atomic(setup.bvec_data);
 826         }
 827         if (setup.segments)
 828                 kunmap_atomic(setup.segments);
 829
 830         /* Keep a private copy so we can reissue requests when recovering. */
 831         rinfo->shadow[id].req = *ring_req;
 832         if (unlikely(require_extra_req))
 833                 rinfo->shadow[extra_id].req = *extra_ring_req;
 834
 835         if (new_persistent_gnts)
 836                 gnttab_free_grant_references(setup.gref_head);
 837
 838         return 0;
 839 }
 840
 841 /*
 842  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 843  * and writes are handled as expected.
 844  *
 845  * @req: a request struct
 846  */
 847 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 848 {
 849         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 850                 return 1;
 851
 852         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 853                      req_op(req) == REQ_OP_SECURE_ERASE))
 854                 return blkif_queue_discard_req(req, rinfo);
 855         else
 856                 return blkif_queue_rw_req(req, rinfo);
 857 }
 858
 859 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 860 {
 861         int notify;
 862
 863         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 864
 865         if (notify)
 866                 notify_remote_via_irq(rinfo->irq);
 867 }
 868
 869 static inline bool blkif_request_flush_invalid(struct request *req,
 870                                                struct blkfront_info *info)
 871 {
 872         return (blk_rq_is_passthrough(req) ||
 873                 ((req_op(req) == REQ_OP_FLUSH) &&
 874                  !info->feature_flush) ||
 875                 ((req->cmd_flags & REQ_FUA) &&
 876                  !info->feature_fua));
 877 }
 878
 879 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 880                           const struct blk_mq_queue_data *qd)
 881 {
 882         unsigned long flags;
 883         int qid = hctx->queue_num;
 884         struct blkfront_info *info = hctx->queue->queuedata;
 885         struct blkfront_ring_info *rinfo = NULL;
 886
 887         rinfo = get_rinfo(info, qid);
 888         blk_mq_start_request(qd->rq);
 889         spin_lock_irqsave(&rinfo->ring_lock, flags);
 890         if (RING_FULL(&rinfo->ring))
 891                 goto out_busy;
 892
 893         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 894                 goto out_err;
 895
 896         if (blkif_queue_request(qd->rq, rinfo))
 897                 goto out_busy;
 898
 899         flush_requests(rinfo);
 900         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 901         return BLK_STS_OK;
 902
 903 out_err:
 904         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 905         return BLK_STS_IOERR;
 906
 907 out_busy:
 908         blk_mq_stop_hw_queue(hctx);
 909         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 910         return BLK_STS_DEV_RESOURCE;
 911 }
 912
 913 static void blkif_complete_rq(struct request *rq)
 914 {
 915         blk_mq_end_request(rq, blkif_req(rq)->error);
 916 }
 917
 918 static const struct blk_mq_ops blkfront_mq_ops = {
 919         .queue_rq = blkif_queue_rq,
 920         .complete = blkif_complete_rq,
 921 };
 922
 923 static void blkif_set_queue_limits(struct blkfront_info *info)
 924 {
 925         struct request_queue *rq = info->rq;
 926         struct gendisk *gd = info->gd;
 927         unsigned int segments = info->max_indirect_segments ? :
 928                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 929
 930         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 931
 932         if (info->feature_discard) {
 933                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 934                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 935                 rq->limits.discard_granularity = info->discard_granularity ?:
 936                                                  info->physical_sector_size;
 937                 rq->limits.discard_alignment = info->discard_alignment;
 938                 if (info->feature_secdiscard)
 939                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 940         }
 941
 942         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 943         blk_queue_logical_block_size(rq, info->sector_size);
 944         blk_queue_physical_block_size(rq, info->physical_sector_size);
 945         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 946
 947         /* Each segment in a request is up to an aligned page in size. */
 948         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 949         blk_queue_max_segment_size(rq, PAGE_SIZE);
 950
 951         /* Ensure a merged request will fit in a single I/O ring slot. */
 952         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 953
 954         /* Make sure buffer addresses are sector-aligned. */
 955         blk_queue_dma_alignment(rq, 511);
 956 }
 957
 958 static const char *flush_info(struct blkfront_info *info)
 959 {
 960         if (info->feature_flush && info->feature_fua)
 961                 return "barrier: enabled;";
 962         else if (info->feature_flush)
 963                 return "flush diskcache: enabled;";
 964         else
 965                 return "barrier or flush: disabled;";
 966 }
 967
 968 static void xlvbd_flush(struct blkfront_info *info)
 969 {
 970         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 971                               info->feature_fua ? true : false);
 972         pr_info("blkfront: %s: %s %s %s %s %s\n",
 973                 info->gd->disk_name, flush_info(info),
 974                 "persistent grants:", info->feature_persistent ?
 975                 "enabled;" : "disabled;", "indirect descriptors:",
 976                 info->max_indirect_segments ? "enabled;" : "disabled;");
 977 }
 978
 979 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 980 {
 981         int major;
 982         major = BLKIF_MAJOR(vdevice);
 983         *minor = BLKIF_MINOR(vdevice);
 984         switch (major) {
 985                 case XEN_IDE0_MAJOR:
 986                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 987                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 988                                 EMULATED_HD_DISK_MINOR_OFFSET;
 989                         break;
 990                 case XEN_IDE1_MAJOR:
 991                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 992                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 993                                 EMULATED_HD_DISK_MINOR_OFFSET;
 994                         break;
 995                 case XEN_SCSI_DISK0_MAJOR:
 996                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 997                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 998                         break;
 999                 case XEN_SCSI_DISK1_MAJOR:
1000                 case XEN_SCSI_DISK2_MAJOR:
1001                 case XEN_SCSI_DISK3_MAJOR:
1002                 case XEN_SCSI_DISK4_MAJOR:
1003                 case XEN_SCSI_DISK5_MAJOR:
1004                 case XEN_SCSI_DISK6_MAJOR:
1005                 case XEN_SCSI_DISK7_MAJOR:
1006                         *offset = (*minor / PARTS_PER_DISK) +
1007                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1008                                 EMULATED_SD_DISK_NAME_OFFSET;
1009                         *minor = *minor +
1010                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1011                                 EMULATED_SD_DISK_MINOR_OFFSET;
1012                         break;
1013                 case XEN_SCSI_DISK8_MAJOR:
1014                 case XEN_SCSI_DISK9_MAJOR:
1015                 case XEN_SCSI_DISK10_MAJOR:
1016                 case XEN_SCSI_DISK11_MAJOR:
1017                 case XEN_SCSI_DISK12_MAJOR:
1018                 case XEN_SCSI_DISK13_MAJOR:
1019                 case XEN_SCSI_DISK14_MAJOR:
1020                 case XEN_SCSI_DISK15_MAJOR:
1021                         *offset = (*minor / PARTS_PER_DISK) +
1022                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1023                                 EMULATED_SD_DISK_NAME_OFFSET;
1024                         *minor = *minor +
1025                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1026                                 EMULATED_SD_DISK_MINOR_OFFSET;
1027                         break;
1028                 case XENVBD_MAJOR:
1029                         *offset = *minor / PARTS_PER_DISK;
1030                         break;
1031                 default:
1032                         printk(KERN_WARNING "blkfront: your disk configuration is "
1033                                         "incorrect, please use an xvd device instead\n");
1034                         return -ENODEV;
1035         }
1036         return 0;
1037 }
1038
1039 static char *encode_disk_name(char *ptr, unsigned int n)
1040 {
1041         if (n >= 26)
1042                 ptr = encode_disk_name(ptr, n / 26 - 1);
1043         *ptr = 'a' + n % 26;
1044         return ptr + 1;
1045 }
1046
1047 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1048                                struct blkfront_info *info,
1049                                u16 vdisk_info, u16 sector_size,
1050                                unsigned int physical_sector_size)
1051 {
1052         struct gendisk *gd;
1053         int nr_minors = 1;
1054         int err;
1055         unsigned int offset;
1056         int minor;
1057         int nr_parts;
1058         char *ptr;
1059
1060         BUG_ON(info->gd != NULL);
1061         BUG_ON(info->rq != NULL);
1062
1063         if ((info->vdevice>>EXT_SHIFT) > 1) {
1064                 /* this is above the extended range; something is wrong */
1065                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1066                 return -ENODEV;
1067         }
1068
1069         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1070                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1071                 if (err)
1072                         return err;
1073                 nr_parts = PARTS_PER_DISK;
1074         } else {
1075                 minor = BLKIF_MINOR_EXT(info->vdevice);
1076                 nr_parts = PARTS_PER_EXT_DISK;
1077                 offset = minor / nr_parts;
1078                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1079                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1080                                         "emulated IDE disks,\n\t choose an xvd device name"
1081                                         "from xvde on\n", info->vdevice);
1082         }
1083         if (minor >> MINORBITS) {
1084                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1085                         info->vdevice, minor);
1086                 return -ENODEV;
1087         }
1088
1089         if ((minor % nr_parts) == 0)
1090                 nr_minors = nr_parts;
1091
1092         err = xlbd_reserve_minors(minor, nr_minors);
1093         if (err)
1094                 return err;
1095         err = -ENODEV;
1096
1097         memset(&info->tag_set, 0, sizeof(info->tag_set));
1098         info->tag_set.ops = &blkfront_mq_ops;
1099         info->tag_set.nr_hw_queues = info->nr_rings;
1100         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1101                 /*
1102                  * When indirect descriptior is not supported, the I/O request
1103                  * will be split between multiple request in the ring.
1104                  * To avoid problems when sending the request, divide by
1105                  * 2 the depth of the queue.
1106                  */
1107                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1108         } else
1109                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1110         info->tag_set.numa_node = NUMA_NO_NODE;
1111         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1112         info->tag_set.cmd_size = sizeof(struct blkif_req);
1113         info->tag_set.driver_data = info;
1114
1115         err = blk_mq_alloc_tag_set(&info->tag_set);
1116         if (err)
1117                 goto out_release_minors;
1118
1119         gd = blk_mq_alloc_disk(&info->tag_set, info);
1120         if (IS_ERR(gd)) {
1121                 err = PTR_ERR(gd);
1122                 goto out_free_tag_set;
1123         }
1124
1125         strcpy(gd->disk_name, DEV_NAME);
1126         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1127         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1128         if (nr_minors > 1)
1129                 *ptr = 0;
1130         else
1131                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1132                          "%d", minor & (nr_parts - 1));
1133
1134         gd->major = XENVBD_MAJOR;
1135         gd->first_minor = minor;
1136         gd->minors = nr_minors;
1137         gd->fops = &xlvbd_block_fops;
1138         gd->private_data = info;
1139         set_capacity(gd, capacity);
1140
1141         info->rq = gd->queue;
1142         info->gd = gd;
1143         info->sector_size = sector_size;
1144         info->physical_sector_size = physical_sector_size;
1145         blkif_set_queue_limits(info);
1146
1147         xlvbd_flush(info);
1148
1149         if (vdisk_info & VDISK_READONLY)
1150                 set_disk_ro(gd, 1);
1151
1152         if (vdisk_info & VDISK_REMOVABLE)
1153                 gd->flags |= GENHD_FL_REMOVABLE;
1154
1155         if (vdisk_info & VDISK_CDROM)
1156                 gd->flags |= GENHD_FL_CD;
1157
1158         return 0;
1159
1160 out_free_tag_set:
1161         blk_mq_free_tag_set(&info->tag_set);
1162 out_release_minors:
1163         xlbd_release_minors(minor, nr_minors);
1164         return err;
1165 }
1166
1167 /* Already hold rinfo->ring_lock. */
1168 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1169 {
1170         if (!RING_FULL(&rinfo->ring))
1171                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1172 }
1173
1174 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1175 {
1176         unsigned long flags;
1177
1178         spin_lock_irqsave(&rinfo->ring_lock, flags);
1179         kick_pending_request_queues_locked(rinfo);
1180         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1181 }
1182
1183 static void blkif_restart_queue(struct work_struct *work)
1184 {
1185         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1186
1187         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1188                 kick_pending_request_queues(rinfo);
1189 }
1190
1191 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1192 {
1193         struct grant *persistent_gnt, *n;
1194         struct blkfront_info *info = rinfo->dev_info;
1195         int i, j, segs;
1196
1197         /*
1198          * Remove indirect pages, this only happens when using indirect
1199          * descriptors but not persistent grants
1200          */
1201         if (!list_empty(&rinfo->indirect_pages)) {
1202                 struct page *indirect_page, *n;
1203
1204                 BUG_ON(info->feature_persistent);
1205                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1206                         list_del(&indirect_page->lru);
1207                         __free_page(indirect_page);
1208                 }
1209         }
1210
1211         /* Remove all persistent grants. */
1212         if (!list_empty(&rinfo->grants)) {
1213                 list_for_each_entry_safe(persistent_gnt, n,
1214                                          &rinfo->grants, node) {
1215                         list_del(&persistent_gnt->node);
1216                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1217                                 gnttab_end_foreign_access(persistent_gnt->gref,
1218                                                           0, 0UL);
1219                                 rinfo->persistent_gnts_c--;
1220                         }
1221                         if (info->feature_persistent)
1222                                 __free_page(persistent_gnt->page);
1223                         kfree(persistent_gnt);
1224                 }
1225         }
1226         BUG_ON(rinfo->persistent_gnts_c != 0);
1227
1228         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1229                 /*
1230                  * Clear persistent grants present in requests already
1231                  * on the shared ring
1232                  */
1233                 if (!rinfo->shadow[i].request)
1234                         goto free_shadow;
1235
1236                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1237                        rinfo->shadow[i].req.u.indirect.nr_segments :
1238                        rinfo->shadow[i].req.u.rw.nr_segments;
1239                 for (j = 0; j < segs; j++) {
1240                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1241                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1242                         if (info->feature_persistent)
1243                                 __free_page(persistent_gnt->page);
1244                         kfree(persistent_gnt);
1245                 }
1246
1247                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1248                         /*
1249                          * If this is not an indirect operation don't try to
1250                          * free indirect segments
1251                          */
1252                         goto free_shadow;
1253
1254                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1255                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1256                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1257                         __free_page(persistent_gnt->page);
1258                         kfree(persistent_gnt);
1259                 }
1260
1261 free_shadow:
1262                 kvfree(rinfo->shadow[i].grants_used);
1263                 rinfo->shadow[i].grants_used = NULL;
1264                 kvfree(rinfo->shadow[i].indirect_grants);
1265                 rinfo->shadow[i].indirect_grants = NULL;
1266                 kvfree(rinfo->shadow[i].sg);
1267                 rinfo->shadow[i].sg = NULL;
1268         }
1269
1270         /* No more gnttab callback work. */
1271         gnttab_cancel_free_callback(&rinfo->callback);
1272
1273         /* Flush gnttab callback work. Must be done with no locks held. */
1274         flush_work(&rinfo->work);
1275
1276         /* Free resources associated with old device channel. */
1277         for (i = 0; i < info->nr_ring_pages; i++) {
1278                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1279                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1280                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1281                 }
1282         }
1283         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1284         rinfo->ring.sring = NULL;
1285
1286         if (rinfo->irq)
1287                 unbind_from_irqhandler(rinfo->irq, rinfo);
1288         rinfo->evtchn = rinfo->irq = 0;
1289 }
1290
1291 static void blkif_free(struct blkfront_info *info, int suspend)
1292 {
1293         unsigned int i;
1294         struct blkfront_ring_info *rinfo;
1295
1296         /* Prevent new requests being issued until we fix things up. */
1297         info->connected = suspend ?
1298                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1299         /* No more blkif_request(). */
1300         if (info->rq)
1301                 blk_mq_stop_hw_queues(info->rq);
1302
1303         for_each_rinfo(info, rinfo, i)
1304                 blkif_free_ring(rinfo);
1305
1306         kvfree(info->rinfo);
1307         info->rinfo = NULL;
1308         info->nr_rings = 0;
1309 }
1310
1311 struct copy_from_grant {
1312         const struct blk_shadow *s;
1313         unsigned int grant_idx;
1314         unsigned int bvec_offset;
1315         char *bvec_data;
1316 };
1317
1318 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1319                                   unsigned int len, void *data)
1320 {
1321         struct copy_from_grant *info = data;
1322         char *shared_data;
1323         /* Convenient aliases */
1324         const struct blk_shadow *s = info->s;
1325
1326         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1327
1328         memcpy(info->bvec_data + info->bvec_offset,
1329                shared_data + offset, len);
1330
1331         info->bvec_offset += len;
1332         info->grant_idx++;
1333
1334         kunmap_atomic(shared_data);
1335 }
1336
1337 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1338 {
1339         switch (rsp)
1340         {
1341         case BLKIF_RSP_OKAY:
1342                 return REQ_DONE;
1343         case BLKIF_RSP_EOPNOTSUPP:
1344                 return REQ_EOPNOTSUPP;
1345         case BLKIF_RSP_ERROR:
1346         default:
1347                 return REQ_ERROR;
1348         }
1349 }
1350
1351 /*
1352  * Get the final status of the block request based on two ring response
1353  */
1354 static int blkif_get_final_status(enum blk_req_status s1,
1355                                   enum blk_req_status s2)
1356 {
1357         BUG_ON(s1 == REQ_WAITING);
1358         BUG_ON(s2 == REQ_WAITING);
1359
1360         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1361                 return BLKIF_RSP_ERROR;
1362         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1363                 return BLKIF_RSP_EOPNOTSUPP;
1364         return BLKIF_RSP_OKAY;
1365 }
1366
1367 static bool blkif_completion(unsigned long *id,
1368                              struct blkfront_ring_info *rinfo,
1369                              struct blkif_response *bret)
1370 {
1371         int i = 0;
1372         struct scatterlist *sg;
1373         int num_sg, num_grant;
1374         struct blkfront_info *info = rinfo->dev_info;
1375         struct blk_shadow *s = &rinfo->shadow[*id];
1376         struct copy_from_grant data = {
1377                 .grant_idx = 0,
1378         };
1379
1380         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1381                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1382
1383         /* The I/O request may be split in two. */
1384         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1385                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1386
1387                 /* Keep the status of the current response in shadow. */
1388                 s->status = blkif_rsp_to_req_status(bret->status);
1389
1390                 /* Wait the second response if not yet here. */
1391                 if (s2->status == REQ_WAITING)
1392                         return false;
1393
1394                 bret->status = blkif_get_final_status(s->status,
1395                                                       s2->status);
1396
1397                 /*
1398                  * All the grants is stored in the first shadow in order
1399                  * to make the completion code simpler.
1400                  */
1401                 num_grant += s2->req.u.rw.nr_segments;
1402
1403                 /*
1404                  * The two responses may not come in order. Only the
1405                  * first request will store the scatter-gather list.
1406                  */
1407                 if (s2->num_sg != 0) {
1408                         /* Update "id" with the ID of the first response. */
1409                         *id = s->associated_id;
1410                         s = s2;
1411                 }
1412
1413                 /*
1414                  * We don't need anymore the second request, so recycling
1415                  * it now.
1416                  */
1417                 if (add_id_to_freelist(rinfo, s->associated_id))
1418                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1419                              info->gd->disk_name, s->associated_id);
1420         }
1421
1422         data.s = s;
1423         num_sg = s->num_sg;
1424
1425         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1426                 for_each_sg(s->sg, sg, num_sg, i) {
1427                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1428
1429                         data.bvec_offset = sg->offset;
1430                         data.bvec_data = kmap_atomic(sg_page(sg));
1431
1432                         gnttab_foreach_grant_in_range(sg_page(sg),
1433                                                       sg->offset,
1434                                                       sg->length,
1435                                                       blkif_copy_from_grant,
1436                                                       &data);
1437
1438                         kunmap_atomic(data.bvec_data);
1439                 }
1440         }
1441         /* Add the persistent grant into the list of free grants */
1442         for (i = 0; i < num_grant; i++) {
1443                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1444                         /*
1445                          * If the grant is still mapped by the backend (the
1446                          * backend has chosen to make this grant persistent)
1447                          * we add it at the head of the list, so it will be
1448                          * reused first.
1449                          */
1450                         if (!info->feature_persistent)
1451                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1452                                                      s->grants_used[i]->gref);
1453                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1454                         rinfo->persistent_gnts_c++;
1455                 } else {
1456                         /*
1457                          * If the grant is not mapped by the backend we end the
1458                          * foreign access and add it to the tail of the list,
1459                          * so it will not be picked again unless we run out of
1460                          * persistent grants.
1461                          */
1462                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1463                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1464                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1465                 }
1466         }
1467         if (s->req.operation == BLKIF_OP_INDIRECT) {
1468                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1469                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1470                                 if (!info->feature_persistent)
1471                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1472                                                              s->indirect_grants[i]->gref);
1473                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1474                                 rinfo->persistent_gnts_c++;
1475                         } else {
1476                                 struct page *indirect_page;
1477
1478                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1479                                 /*
1480                                  * Add the used indirect page back to the list of
1481                                  * available pages for indirect grefs.
1482                                  */
1483                                 if (!info->feature_persistent) {
1484                                         indirect_page = s->indirect_grants[i]->page;
1485                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1486                                 }
1487                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1488                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1489                         }
1490                 }
1491         }
1492
1493         return true;
1494 }
1495
1496 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1497 {
1498         struct request *req;
1499         struct blkif_response *bret;
1500         RING_IDX i, rp;
1501         unsigned long flags;
1502         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1503         struct blkfront_info *info = rinfo->dev_info;
1504
1505         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1506                 return IRQ_HANDLED;
1507
1508         spin_lock_irqsave(&rinfo->ring_lock, flags);
1509  again:
1510         rp = rinfo->ring.sring->rsp_prod;
1511         rmb(); /* Ensure we see queued responses up to 'rp'. */
1512
1513         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1514                 unsigned long id;
1515
1516                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1517                 id   = bret->id;
1518                 /*
1519                  * The backend has messed up and given us an id that we would
1520                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1521                  * look in get_id_from_freelist.
1522                  */
1523                 if (id >= BLK_RING_SIZE(info)) {
1524                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1525                              info->gd->disk_name, op_name(bret->operation), id);
1526                         /* We can't safely get the 'struct request' as
1527                          * the id is busted. */
1528                         continue;
1529                 }
1530                 req  = rinfo->shadow[id].request;
1531
1532                 if (bret->operation != BLKIF_OP_DISCARD) {
1533                         /*
1534                          * We may need to wait for an extra response if the
1535                          * I/O request is split in 2
1536                          */
1537                         if (!blkif_completion(&id, rinfo, bret))
1538                                 continue;
1539                 }
1540
1541                 if (add_id_to_freelist(rinfo, id)) {
1542                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1543                              info->gd->disk_name, op_name(bret->operation), id);
1544                         continue;
1545                 }
1546
1547                 if (bret->status == BLKIF_RSP_OKAY)
1548                         blkif_req(req)->error = BLK_STS_OK;
1549                 else
1550                         blkif_req(req)->error = BLK_STS_IOERR;
1551
1552                 switch (bret->operation) {
1553                 case BLKIF_OP_DISCARD:
1554                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1555                                 struct request_queue *rq = info->rq;
1556                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1557                                            info->gd->disk_name, op_name(bret->operation));
1558                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1559                                 info->feature_discard = 0;
1560                                 info->feature_secdiscard = 0;
1561                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1562                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1563                         }
1564                         break;
1565                 case BLKIF_OP_FLUSH_DISKCACHE:
1566                 case BLKIF_OP_WRITE_BARRIER:
1567                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1568                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1569                                        info->gd->disk_name, op_name(bret->operation));
1570                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1571                         }
1572                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1573                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1574                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1575                                        info->gd->disk_name, op_name(bret->operation));
1576                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1577                         }
1578                         if (unlikely(blkif_req(req)->error)) {
1579                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1580                                         blkif_req(req)->error = BLK_STS_OK;
1581                                 info->feature_fua = 0;
1582                                 info->feature_flush = 0;
1583                                 xlvbd_flush(info);
1584                         }
1585                         fallthrough;
1586                 case BLKIF_OP_READ:
1587                 case BLKIF_OP_WRITE:
1588                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1589                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1590                                         "request: %x\n", bret->status);
1591
1592                         break;
1593                 default:
1594                         BUG();
1595                 }
1596
1597                 if (likely(!blk_should_fake_timeout(req->q)))
1598                         blk_mq_complete_request(req);
1599         }
1600
1601         rinfo->ring.rsp_cons = i;
1602
1603         if (i != rinfo->ring.req_prod_pvt) {
1604                 int more_to_do;
1605                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1606                 if (more_to_do)
1607                         goto again;
1608         } else
1609                 rinfo->ring.sring->rsp_event = i + 1;
1610
1611         kick_pending_request_queues_locked(rinfo);
1612
1613         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1614
1615         return IRQ_HANDLED;
1616 }
1617
1618
1619 static int setup_blkring(struct xenbus_device *dev,
1620                          struct blkfront_ring_info *rinfo)
1621 {
1622         struct blkif_sring *sring;
1623         int err, i;
1624         struct blkfront_info *info = rinfo->dev_info;
1625         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1626         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1627
1628         for (i = 0; i < info->nr_ring_pages; i++)
1629                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1630
1631         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1632                                                        get_order(ring_size));
1633         if (!sring) {
1634                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1635                 return -ENOMEM;
1636         }
1637         SHARED_RING_INIT(sring);
1638         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1639
1640         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1641         if (err < 0) {
1642                 free_pages((unsigned long)sring, get_order(ring_size));
1643                 rinfo->ring.sring = NULL;
1644                 goto fail;
1645         }
1646         for (i = 0; i < info->nr_ring_pages; i++)
1647                 rinfo->ring_ref[i] = gref[i];
1648
1649         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1650         if (err)
1651                 goto fail;
1652
1653         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1654                                         "blkif", rinfo);
1655         if (err <= 0) {
1656                 xenbus_dev_fatal(dev, err,
1657                                  "bind_evtchn_to_irqhandler failed");
1658                 goto fail;
1659         }
1660         rinfo->irq = err;
1661
1662         return 0;
1663 fail:
1664         blkif_free(info, 0);
1665         return err;
1666 }
1667
1668 /*
1669  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1670  * ring buffer may have multi pages depending on ->nr_ring_pages.
1671  */
1672 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1673                                 struct blkfront_ring_info *rinfo, const char *dir)
1674 {
1675         int err;
1676         unsigned int i;
1677         const char *message = NULL;
1678         struct blkfront_info *info = rinfo->dev_info;
1679
1680         if (info->nr_ring_pages == 1) {
1681                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1682                 if (err) {
1683                         message = "writing ring-ref";
1684                         goto abort_transaction;
1685                 }
1686         } else {
1687                 for (i = 0; i < info->nr_ring_pages; i++) {
1688                         char ring_ref_name[RINGREF_NAME_LEN];
1689
1690                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1691                         err = xenbus_printf(xbt, dir, ring_ref_name,
1692                                             "%u", rinfo->ring_ref[i]);
1693                         if (err) {
1694                                 message = "writing ring-ref";
1695                                 goto abort_transaction;
1696                         }
1697                 }
1698         }
1699
1700         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1701         if (err) {
1702                 message = "writing event-channel";
1703                 goto abort_transaction;
1704         }
1705
1706         return 0;
1707
1708 abort_transaction:
1709         xenbus_transaction_end(xbt, 1);
1710         if (message)
1711                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1712
1713         return err;
1714 }
1715
1716 /* Common code used when first setting up, and when resuming. */
1717 static int talk_to_blkback(struct xenbus_device *dev,
1718                            struct blkfront_info *info)
1719 {
1720         const char *message = NULL;
1721         struct xenbus_transaction xbt;
1722         int err;
1723         unsigned int i, max_page_order;
1724         unsigned int ring_page_order;
1725         struct blkfront_ring_info *rinfo;
1726
1727         if (!info)
1728                 return -ENODEV;
1729
1730         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1731                                               "max-ring-page-order", 0);
1732         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1733         info->nr_ring_pages = 1 << ring_page_order;
1734
1735         err = negotiate_mq(info);
1736         if (err)
1737                 goto destroy_blkring;
1738
1739         for_each_rinfo(info, rinfo, i) {
1740                 /* Create shared ring, alloc event channel. */
1741                 err = setup_blkring(dev, rinfo);
1742                 if (err)
1743                         goto destroy_blkring;
1744         }
1745
1746 again:
1747         err = xenbus_transaction_start(&xbt);
1748         if (err) {
1749                 xenbus_dev_fatal(dev, err, "starting transaction");
1750                 goto destroy_blkring;
1751         }
1752
1753         if (info->nr_ring_pages > 1) {
1754                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1755                                     ring_page_order);
1756                 if (err) {
1757                         message = "writing ring-page-order";
1758                         goto abort_transaction;
1759                 }
1760         }
1761
1762         /* We already got the number of queues/rings in _probe */
1763         if (info->nr_rings == 1) {
1764                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1765                 if (err)
1766                         goto destroy_blkring;
1767         } else {
1768                 char *path;
1769                 size_t pathsize;
1770
1771                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1772                                     info->nr_rings);
1773                 if (err) {
1774                         message = "writing multi-queue-num-queues";
1775                         goto abort_transaction;
1776                 }
1777
1778                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1779                 path = kmalloc(pathsize, GFP_KERNEL);
1780                 if (!path) {
1781                         err = -ENOMEM;
1782                         message = "ENOMEM while writing ring references";
1783                         goto abort_transaction;
1784                 }
1785
1786                 for_each_rinfo(info, rinfo, i) {
1787                         memset(path, 0, pathsize);
1788                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1789                         err = write_per_ring_nodes(xbt, rinfo, path);
1790                         if (err) {
1791                                 kfree(path);
1792                                 goto destroy_blkring;
1793                         }
1794                 }
1795                 kfree(path);
1796         }
1797         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1798                             XEN_IO_PROTO_ABI_NATIVE);
1799         if (err) {
1800                 message = "writing protocol";
1801                 goto abort_transaction;
1802         }
1803         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1804                         info->feature_persistent);
1805         if (err)
1806                 dev_warn(&dev->dev,
1807                          "writing persistent grants feature to xenbus");
1808
1809         err = xenbus_transaction_end(xbt, 0);
1810         if (err) {
1811                 if (err == -EAGAIN)
1812                         goto again;
1813                 xenbus_dev_fatal(dev, err, "completing transaction");
1814                 goto destroy_blkring;
1815         }
1816
1817         for_each_rinfo(info, rinfo, i) {
1818                 unsigned int j;
1819
1820                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1821                         rinfo->shadow[j].req.u.rw.id = j + 1;
1822                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1823         }
1824         xenbus_switch_state(dev, XenbusStateInitialised);
1825
1826         return 0;
1827
1828  abort_transaction:
1829         xenbus_transaction_end(xbt, 1);
1830         if (message)
1831                 xenbus_dev_fatal(dev, err, "%s", message);
1832  destroy_blkring:
1833         blkif_free(info, 0);
1834         return err;
1835 }
1836
1837 static int negotiate_mq(struct blkfront_info *info)
1838 {
1839         unsigned int backend_max_queues;
1840         unsigned int i;
1841         struct blkfront_ring_info *rinfo;
1842
1843         BUG_ON(info->nr_rings);
1844
1845         /* Check if backend supports multiple queues. */
1846         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1847                                                   "multi-queue-max-queues", 1);
1848         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1849         /* We need at least one ring. */
1850         if (!info->nr_rings)
1851                 info->nr_rings = 1;
1852
1853         info->rinfo_size = struct_size(info->rinfo, shadow,
1854                                        BLK_RING_SIZE(info));
1855         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1856         if (!info->rinfo) {
1857                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1858                 info->nr_rings = 0;
1859                 return -ENOMEM;
1860         }
1861
1862         for_each_rinfo(info, rinfo, i) {
1863                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1864                 INIT_LIST_HEAD(&rinfo->grants);
1865                 rinfo->dev_info = info;
1866                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1867                 spin_lock_init(&rinfo->ring_lock);
1868         }
1869         return 0;
1870 }
1871
1872 /* Enable the persistent grants feature. */
1873 static bool feature_persistent = true;
1874 module_param(feature_persistent, bool, 0644);
1875 MODULE_PARM_DESC(feature_persistent,
1876                 "Enables the persistent grants feature");
1877
1878 /*
1879  * Entry point to this code when a new device is created.  Allocate the basic
1880  * structures and the ring buffer for communication with the backend, and
1881  * inform the backend of the appropriate details for those.  Switch to
1882  * Initialised state.
1883  */
1884 static int blkfront_probe(struct xenbus_device *dev,
1885                           const struct xenbus_device_id *id)
1886 {
1887         int err, vdevice;
1888         struct blkfront_info *info;
1889
1890         /* FIXME: Use dynamic device id if this is not set. */
1891         err = xenbus_scanf(XBT_NIL, dev->nodename,
1892                            "virtual-device", "%i", &vdevice);
1893         if (err != 1) {
1894                 /* go looking in the extended area instead */
1895                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1896                                    "%i", &vdevice);
1897                 if (err != 1) {
1898                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1899                         return err;
1900                 }
1901         }
1902
1903         if (xen_hvm_domain()) {
1904                 char *type;
1905                 int len;
1906                 /* no unplug has been done: do not hook devices != xen vbds */
1907                 if (xen_has_pv_and_legacy_disk_devices()) {
1908                         int major;
1909
1910                         if (!VDEV_IS_EXTENDED(vdevice))
1911                                 major = BLKIF_MAJOR(vdevice);
1912                         else
1913                                 major = XENVBD_MAJOR;
1914
1915                         if (major != XENVBD_MAJOR) {
1916                                 printk(KERN_INFO
1917                                                 "%s: HVM does not support vbd %d as xen block device\n",
1918                                                 __func__, vdevice);
1919                                 return -ENODEV;
1920                         }
1921                 }
1922                 /* do not create a PV cdrom device if we are an HVM guest */
1923                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1924                 if (IS_ERR(type))
1925                         return -ENODEV;
1926                 if (strncmp(type, "cdrom", 5) == 0) {
1927                         kfree(type);
1928                         return -ENODEV;
1929                 }
1930                 kfree(type);
1931         }
1932         info = kzalloc(sizeof(*info), GFP_KERNEL);
1933         if (!info) {
1934                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1935                 return -ENOMEM;
1936         }
1937
1938         info->xbdev = dev;
1939
1940         mutex_init(&info->mutex);
1941         info->vdevice = vdevice;
1942         info->connected = BLKIF_STATE_DISCONNECTED;
1943
1944         info->feature_persistent = feature_persistent;
1945
1946         /* Front end dir is a number, which is used as the id. */
1947         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1948         dev_set_drvdata(&dev->dev, info);
1949
1950         mutex_lock(&blkfront_mutex);
1951         list_add(&info->info_list, &info_list);
1952         mutex_unlock(&blkfront_mutex);
1953
1954         return 0;
1955 }
1956
1957 static int blkif_recover(struct blkfront_info *info)
1958 {
1959         unsigned int r_index;
1960         struct request *req, *n;
1961         int rc;
1962         struct bio *bio;
1963         unsigned int segs;
1964         struct blkfront_ring_info *rinfo;
1965
1966         blkfront_gather_backend_features(info);
1967         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
1968         blkif_set_queue_limits(info);
1969         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1970         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
1971
1972         for_each_rinfo(info, rinfo, r_index) {
1973                 rc = blkfront_setup_indirect(rinfo);
1974                 if (rc)
1975                         return rc;
1976         }
1977         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1978
1979         /* Now safe for us to use the shared ring */
1980         info->connected = BLKIF_STATE_CONNECTED;
1981
1982         for_each_rinfo(info, rinfo, r_index) {
1983                 /* Kick any other new requests queued since we resumed */
1984                 kick_pending_request_queues(rinfo);
1985         }
1986
1987         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
1988                 /* Requeue pending requests (flush or discard) */
1989                 list_del_init(&req->queuelist);
1990                 BUG_ON(req->nr_phys_segments > segs);
1991                 blk_mq_requeue_request(req, false);
1992         }
1993         blk_mq_start_stopped_hw_queues(info->rq, true);
1994         blk_mq_kick_requeue_list(info->rq);
1995
1996         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
1997                 /* Traverse the list of pending bios and re-queue them */
1998                 submit_bio(bio);
1999         }
2000
2001         return 0;
2002 }
2003
2004 /*
2005  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2006  * driver restart.  We tear down our blkif structure and recreate it, but
2007  * leave the device-layer structures intact so that this is transparent to the
2008  * rest of the kernel.
2009  */
2010 static int blkfront_resume(struct xenbus_device *dev)
2011 {
2012         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2013         int err = 0;
2014         unsigned int i, j;
2015         struct blkfront_ring_info *rinfo;
2016
2017         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2018
2019         bio_list_init(&info->bio_list);
2020         INIT_LIST_HEAD(&info->requests);
2021         for_each_rinfo(info, rinfo, i) {
2022                 struct bio_list merge_bio;
2023                 struct blk_shadow *shadow = rinfo->shadow;
2024
2025                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2026                         /* Not in use? */
2027                         if (!shadow[j].request)
2028                                 continue;
2029
2030                         /*
2031                          * Get the bios in the request so we can re-queue them.
2032                          */
2033                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2034                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2035                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2036                             shadow[j].request->cmd_flags & REQ_FUA) {
2037                                 /*
2038                                  * Flush operations don't contain bios, so
2039                                  * we need to requeue the whole request
2040                                  *
2041                                  * XXX: but this doesn't make any sense for a
2042                                  * write with the FUA flag set..
2043                                  */
2044                                 list_add(&shadow[j].request->queuelist, &info->requests);
2045                                 continue;
2046                         }
2047                         merge_bio.head = shadow[j].request->bio;
2048                         merge_bio.tail = shadow[j].request->biotail;
2049                         bio_list_merge(&info->bio_list, &merge_bio);
2050                         shadow[j].request->bio = NULL;
2051                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2052                 }
2053         }
2054
2055         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2056
2057         err = talk_to_blkback(dev, info);
2058         if (!err)
2059                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2060
2061         /*
2062          * We have to wait for the backend to switch to
2063          * connected state, since we want to read which
2064          * features it supports.
2065          */
2066
2067         return err;
2068 }
2069
2070 static void blkfront_closing(struct blkfront_info *info)
2071 {
2072         struct xenbus_device *xbdev = info->xbdev;
2073         struct blkfront_ring_info *rinfo;
2074         unsigned int i;
2075
2076         if (xbdev->state == XenbusStateClosing)
2077                 return;
2078
2079         /* No more blkif_request(). */
2080         blk_mq_stop_hw_queues(info->rq);
2081         blk_set_queue_dying(info->rq);
2082         set_capacity(info->gd, 0);
2083
2084         for_each_rinfo(info, rinfo, i) {
2085                 /* No more gnttab callback work. */
2086                 gnttab_cancel_free_callback(&rinfo->callback);
2087
2088                 /* Flush gnttab callback work. Must be done with no locks held. */
2089                 flush_work(&rinfo->work);
2090         }
2091
2092         xenbus_frontend_closed(xbdev);
2093 }
2094
2095 static void blkfront_setup_discard(struct blkfront_info *info)
2096 {
2097         info->feature_discard = 1;
2098         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2099                                                          "discard-granularity",
2100                                                          0);
2101         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2102                                                        "discard-alignment", 0);
2103         info->feature_secdiscard =
2104                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2105                                        0);
2106 }
2107
2108 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2109 {
2110         unsigned int psegs, grants, memflags;
2111         int err, i;
2112         struct blkfront_info *info = rinfo->dev_info;
2113
2114         memflags = memalloc_noio_save();
2115
2116         if (info->max_indirect_segments == 0) {
2117                 if (!HAS_EXTRA_REQ)
2118                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2119                 else {
2120                         /*
2121                          * When an extra req is required, the maximum
2122                          * grants supported is related to the size of the
2123                          * Linux block segment.
2124                          */
2125                         grants = GRANTS_PER_PSEG;
2126                 }
2127         }
2128         else
2129                 grants = info->max_indirect_segments;
2130         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2131
2132         err = fill_grant_buffer(rinfo,
2133                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2134         if (err)
2135                 goto out_of_memory;
2136
2137         if (!info->feature_persistent && info->max_indirect_segments) {
2138                 /*
2139                  * We are using indirect descriptors but not persistent
2140                  * grants, we need to allocate a set of pages that can be
2141                  * used for mapping indirect grefs
2142                  */
2143                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2144
2145                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2146                 for (i = 0; i < num; i++) {
2147                         struct page *indirect_page = alloc_page(GFP_KERNEL);
2148                         if (!indirect_page)
2149                                 goto out_of_memory;
2150                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2151                 }
2152         }
2153
2154         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2155                 rinfo->shadow[i].grants_used =
2156                         kvcalloc(grants,
2157                                  sizeof(rinfo->shadow[i].grants_used[0]),
2158                                  GFP_KERNEL);
2159                 rinfo->shadow[i].sg = kvcalloc(psegs,
2160                                                sizeof(rinfo->shadow[i].sg[0]),
2161                                                GFP_KERNEL);
2162                 if (info->max_indirect_segments)
2163                         rinfo->shadow[i].indirect_grants =
2164                                 kvcalloc(INDIRECT_GREFS(grants),
2165                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2166                                          GFP_KERNEL);
2167                 if ((rinfo->shadow[i].grants_used == NULL) ||
2168                         (rinfo->shadow[i].sg == NULL) ||
2169                      (info->max_indirect_segments &&
2170                      (rinfo->shadow[i].indirect_grants == NULL)))
2171                         goto out_of_memory;
2172                 sg_init_table(rinfo->shadow[i].sg, psegs);
2173         }
2174
2175         memalloc_noio_restore(memflags);
2176
2177         return 0;
2178
2179 out_of_memory:
2180         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2181                 kvfree(rinfo->shadow[i].grants_used);
2182                 rinfo->shadow[i].grants_used = NULL;
2183                 kvfree(rinfo->shadow[i].sg);
2184                 rinfo->shadow[i].sg = NULL;
2185                 kvfree(rinfo->shadow[i].indirect_grants);
2186                 rinfo->shadow[i].indirect_grants = NULL;
2187         }
2188         if (!list_empty(&rinfo->indirect_pages)) {
2189                 struct page *indirect_page, *n;
2190                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2191                         list_del(&indirect_page->lru);
2192                         __free_page(indirect_page);
2193                 }
2194         }
2195
2196         memalloc_noio_restore(memflags);
2197
2198         return -ENOMEM;
2199 }
2200
2201 /*
2202  * Gather all backend feature-*
2203  */
2204 static void blkfront_gather_backend_features(struct blkfront_info *info)
2205 {
2206         unsigned int indirect_segments;
2207
2208         info->feature_flush = 0;
2209         info->feature_fua = 0;
2210
2211         /*
2212          * If there's no "feature-barrier" defined, then it means
2213          * we're dealing with a very old backend which writes
2214          * synchronously; nothing to do.
2215          *
2216          * If there are barriers, then we use flush.
2217          */
2218         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2219                 info->feature_flush = 1;
2220                 info->feature_fua = 1;
2221         }
2222
2223         /*
2224          * And if there is "feature-flush-cache" use that above
2225          * barriers.
2226          */
2227         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2228                                  0)) {
2229                 info->feature_flush = 1;
2230                 info->feature_fua = 0;
2231         }
2232
2233         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2234                 blkfront_setup_discard(info);
2235
2236         if (info->feature_persistent)
2237                 info->feature_persistent =
2238                         !!xenbus_read_unsigned(info->xbdev->otherend,
2239                                                "feature-persistent", 0);
2240
2241         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2242                                         "feature-max-indirect-segments", 0);
2243         if (indirect_segments > xen_blkif_max_segments)
2244                 indirect_segments = xen_blkif_max_segments;
2245         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2246                 indirect_segments = 0;
2247         info->max_indirect_segments = indirect_segments;
2248
2249         if (info->feature_persistent) {
2250                 mutex_lock(&blkfront_mutex);
2251                 schedule_delayed_work(&blkfront_work, HZ * 10);
2252                 mutex_unlock(&blkfront_mutex);
2253         }
2254 }
2255
2256 /*
2257  * Invoked when the backend is finally 'ready' (and has told produced
2258  * the details about the physical device - #sectors, size, etc).
2259  */
2260 static void blkfront_connect(struct blkfront_info *info)
2261 {
2262         unsigned long long sectors;
2263         unsigned long sector_size;
2264         unsigned int physical_sector_size;
2265         unsigned int binfo;
2266         int err, i;
2267         struct blkfront_ring_info *rinfo;
2268
2269         switch (info->connected) {
2270         case BLKIF_STATE_CONNECTED:
2271                 /*
2272                  * Potentially, the back-end may be signalling
2273                  * a capacity change; update the capacity.
2274                  */
2275                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2276                                    "sectors", "%Lu", &sectors);
2277                 if (XENBUS_EXIST_ERR(err))
2278                         return;
2279                 printk(KERN_INFO "Setting capacity to %Lu\n",
2280                        sectors);
2281                 set_capacity_and_notify(info->gd, sectors);
2282
2283                 return;
2284         case BLKIF_STATE_SUSPENDED:
2285                 /*
2286                  * If we are recovering from suspension, we need to wait
2287                  * for the backend to announce it's features before
2288                  * reconnecting, at least we need to know if the backend
2289                  * supports indirect descriptors, and how many.
2290                  */
2291                 blkif_recover(info);
2292                 return;
2293
2294         default:
2295                 break;
2296         }
2297
2298         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2299                 __func__, info->xbdev->otherend);
2300
2301         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2302                             "sectors", "%llu", &sectors,
2303                             "info", "%u", &binfo,
2304                             "sector-size", "%lu", &sector_size,
2305                             NULL);
2306         if (err) {
2307                 xenbus_dev_fatal(info->xbdev, err,
2308                                  "reading backend fields at %s",
2309                                  info->xbdev->otherend);
2310                 return;
2311         }
2312
2313         /*
2314          * physical-sector-size is a newer field, so old backends may not
2315          * provide this. Assume physical sector size to be the same as
2316          * sector_size in that case.
2317          */
2318         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2319                                                     "physical-sector-size",
2320                                                     sector_size);
2321         blkfront_gather_backend_features(info);
2322         for_each_rinfo(info, rinfo, i) {
2323                 err = blkfront_setup_indirect(rinfo);
2324                 if (err) {
2325                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2326                                          info->xbdev->otherend);
2327                         blkif_free(info, 0);
2328                         break;
2329                 }
2330         }
2331
2332         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2333                                   physical_sector_size);
2334         if (err) {
2335                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2336                                  info->xbdev->otherend);
2337                 goto fail;
2338         }
2339
2340         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2341
2342         /* Kick pending requests. */
2343         info->connected = BLKIF_STATE_CONNECTED;
2344         for_each_rinfo(info, rinfo, i)
2345                 kick_pending_request_queues(rinfo);
2346
2347         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2348
2349         info->is_ready = 1;
2350         return;
2351
2352 fail:
2353         blkif_free(info, 0);
2354         return;
2355 }
2356
2357 /*
2358  * Callback received when the backend's state changes.
2359  */
2360 static void blkback_changed(struct xenbus_device *dev,
2361                             enum xenbus_state backend_state)
2362 {
2363         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2364
2365         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2366
2367         switch (backend_state) {
2368         case XenbusStateInitWait:
2369                 if (dev->state != XenbusStateInitialising)
2370                         break;
2371                 if (talk_to_blkback(dev, info))
2372                         break;
2373                 break;
2374         case XenbusStateInitialising:
2375         case XenbusStateInitialised:
2376         case XenbusStateReconfiguring:
2377         case XenbusStateReconfigured:
2378         case XenbusStateUnknown:
2379                 break;
2380
2381         case XenbusStateConnected:
2382                 /*
2383                  * talk_to_blkback sets state to XenbusStateInitialised
2384                  * and blkfront_connect sets it to XenbusStateConnected
2385                  * (if connection went OK).
2386                  *
2387                  * If the backend (or toolstack) decides to poke at backend
2388                  * state (and re-trigger the watch by setting the state repeatedly
2389                  * to XenbusStateConnected (4)) we need to deal with this.
2390                  * This is allowed as this is used to communicate to the guest
2391                  * that the size of disk has changed!
2392                  */
2393                 if ((dev->state != XenbusStateInitialised) &&
2394                     (dev->state != XenbusStateConnected)) {
2395                         if (talk_to_blkback(dev, info))
2396                                 break;
2397                 }
2398
2399                 blkfront_connect(info);
2400                 break;
2401
2402         case XenbusStateClosed:
2403                 if (dev->state == XenbusStateClosed)
2404                         break;
2405                 fallthrough;
2406         case XenbusStateClosing:
2407                 blkfront_closing(info);
2408                 break;
2409         }
2410 }
2411
2412 static int blkfront_remove(struct xenbus_device *xbdev)
2413 {
2414         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2415
2416         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2417
2418         del_gendisk(info->gd);
2419
2420         mutex_lock(&blkfront_mutex);
2421         list_del(&info->info_list);
2422         mutex_unlock(&blkfront_mutex);
2423
2424         blkif_free(info, 0);
2425         xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2426         blk_cleanup_disk(info->gd);
2427         blk_mq_free_tag_set(&info->tag_set);
2428
2429         kfree(info);
2430         return 0;
2431 }
2432
2433 static int blkfront_is_ready(struct xenbus_device *dev)
2434 {
2435         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2436
2437         return info->is_ready && info->xbdev;
2438 }
2439
2440 static const struct block_device_operations xlvbd_block_fops =
2441 {
2442         .owner = THIS_MODULE,
2443         .getgeo = blkif_getgeo,
2444         .ioctl = blkif_ioctl,
2445         .compat_ioctl = blkdev_compat_ptr_ioctl,
2446 };
2447
2448
2449 static const struct xenbus_device_id blkfront_ids[] = {
2450         { "vbd" },
2451         { "" }
2452 };
2453
2454 static struct xenbus_driver blkfront_driver = {
2455         .ids  = blkfront_ids,
2456         .probe = blkfront_probe,
2457         .remove = blkfront_remove,
2458         .resume = blkfront_resume,
2459         .otherend_changed = blkback_changed,
2460         .is_ready = blkfront_is_ready,
2461 };
2462
2463 static void purge_persistent_grants(struct blkfront_info *info)
2464 {
2465         unsigned int i;
2466         unsigned long flags;
2467         struct blkfront_ring_info *rinfo;
2468
2469         for_each_rinfo(info, rinfo, i) {
2470                 struct grant *gnt_list_entry, *tmp;
2471
2472                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2473
2474                 if (rinfo->persistent_gnts_c == 0) {
2475                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2476                         continue;
2477                 }
2478
2479                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2480                                          node) {
2481                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2482                             gnttab_query_foreign_access(gnt_list_entry->gref))
2483                                 continue;
2484
2485                         list_del(&gnt_list_entry->node);
2486                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2487                         rinfo->persistent_gnts_c--;
2488                         gnt_list_entry->gref = GRANT_INVALID_REF;
2489                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2490                 }
2491
2492                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2493         }
2494 }
2495
2496 static void blkfront_delay_work(struct work_struct *work)
2497 {
2498         struct blkfront_info *info;
2499         bool need_schedule_work = false;
2500
2501         mutex_lock(&blkfront_mutex);
2502
2503         list_for_each_entry(info, &info_list, info_list) {
2504                 if (info->feature_persistent) {
2505                         need_schedule_work = true;
2506                         mutex_lock(&info->mutex);
2507                         purge_persistent_grants(info);
2508                         mutex_unlock(&info->mutex);
2509                 }
2510         }
2511
2512         if (need_schedule_work)
2513                 schedule_delayed_work(&blkfront_work, HZ * 10);
2514
2515         mutex_unlock(&blkfront_mutex);
2516 }
2517
2518 static int __init xlblk_init(void)
2519 {
2520         int ret;
2521         int nr_cpus = num_online_cpus();
2522
2523         if (!xen_domain())
2524                 return -ENODEV;
2525
2526         if (!xen_has_pv_disk_devices())
2527                 return -ENODEV;
2528
2529         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2530                 pr_warn("xen_blk: can't get major %d with name %s\n",
2531                         XENVBD_MAJOR, DEV_NAME);
2532                 return -ENODEV;
2533         }
2534
2535         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2536                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2537
2538         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2539                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2540                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2541                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2542         }
2543
2544         if (xen_blkif_max_queues > nr_cpus) {
2545                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2546                         xen_blkif_max_queues, nr_cpus);
2547                 xen_blkif_max_queues = nr_cpus;
2548         }
2549
2550         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2551
2552         ret = xenbus_register_frontend(&blkfront_driver);
2553         if (ret) {
2554                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2555                 return ret;
2556         }
2557
2558         return 0;
2559 }
2560 module_init(xlblk_init);
2561
2562
2563 static void __exit xlblk_exit(void)
2564 {
2565         cancel_delayed_work_sync(&blkfront_work);
2566
2567         xenbus_unregister_driver(&blkfront_driver);
2568         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2569         kfree(minors);
2570 }
2571 module_exit(xlblk_exit);
2572
2573 MODULE_DESCRIPTION("Xen virtual block device frontend");
2574 MODULE_LICENSE("GPL");
2575 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2576 MODULE_ALIAS("xen:vbd");
2577 MODULE_ALIAS("xenblk");