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
  50 #include <xen/xen.h>
  51 #include <xen/xenbus.h>
  52 #include <xen/grant_table.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56
  57 #include <xen/interface/grant_table.h>
  58 #include <xen/interface/io/blkif.h>
  59 #include <xen/interface/io/protocols.h>
  60
  61 #include <asm/xen/hypervisor.h>
  62
  63 /*
  64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  65  * When Linux is using a different page size than Xen, it may not be possible
  66  * to put all the data in a single segment.
  67  * This can happen when the backend doesn't support indirect descriptor and
  68  * therefore the maximum amount of data that a request can carry is
  69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  70  *
  71  * Note that we only support one extra request. So the Linux page size
  72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  73  * 88KB.
  74  */
  75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  76
  77 enum blkif_state {
  78         BLKIF_STATE_DISCONNECTED,
  79         BLKIF_STATE_CONNECTED,
  80         BLKIF_STATE_SUSPENDED,
  81 };
  82
  83 struct grant {
  84         grant_ref_t gref;
  85         struct page *page;
  86         struct list_head node;
  87 };
  88
  89 enum blk_req_status {
  90         REQ_WAITING,
  91         REQ_DONE,
  92         REQ_ERROR,
  93         REQ_EOPNOTSUPP,
  94 };
  95
  96 struct blk_shadow {
  97         struct blkif_request req;
  98         struct request *request;
  99         struct grant **grants_used;
 100         struct grant **indirect_grants;
 101         struct scatterlist *sg;
 102         unsigned int num_sg;
 103         enum blk_req_status status;
 104
 105         #define NO_ASSOCIATED_ID ~0UL
 106         /*
 107          * Id of the sibling if we ever need 2 requests when handling a
 108          * block I/O request
 109          */
 110         unsigned long associated_id;
 111 };
 112
 113 struct split_bio {
 114         struct bio *bio;
 115         atomic_t pending;
 116 };
 117
 118 static DEFINE_MUTEX(blkfront_mutex);
 119 static const struct block_device_operations xlvbd_block_fops;
 120
 121 /*
 122  * Maximum number of segments in indirect requests, the actual value used by
 123  * the frontend driver is the minimum of this value and the value provided
 124  * by the backend driver.
 125  */
 126
 127 static unsigned int xen_blkif_max_segments = 32;
 128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
 129                    S_IRUGO);
 130 MODULE_PARM_DESC(max_indirect_segments,
 131                  "Maximum amount of segments in indirect requests (default is 32)");
 132
 133 static unsigned int xen_blkif_max_queues = 4;
 134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
 135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 136
 137 /*
 138  * Maximum order of pages to be used for the shared ring between front and
 139  * backend, 4KB page granularity is used.
 140  */
 141 static unsigned int xen_blkif_max_ring_order;
 142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 144
 145 #define BLK_RING_SIZE(info)     \
 146         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 147
 148 #define BLK_MAX_RING_SIZE       \
 149         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 150
 151 /*
 152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 153  * characters are enough. Define to 20 to keep consistent with backend.
 154  */
 155 #define RINGREF_NAME_LEN (20)
 156 /*
 157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 158  */
 159 #define QUEUE_NAME_LEN (17)
 160
 161 /*
 162  *  Per-ring info.
 163  *  Every blkfront device can associate with one or more blkfront_ring_info,
 164  *  depending on how many hardware queues/rings to be used.
 165  */
 166 struct blkfront_ring_info {
 167         /* Lock to protect data in every ring buffer. */
 168         spinlock_t ring_lock;
 169         struct blkif_front_ring ring;
 170         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 171         unsigned int evtchn, irq;
 172         struct work_struct work;
 173         struct gnttab_free_callback callback;
 174         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 175         struct list_head indirect_pages;
 176         struct list_head grants;
 177         unsigned int persistent_gnts_c;
 178         unsigned long shadow_free;
 179         struct blkfront_info *dev_info;
 180 };
 181
 182 /*
 183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 184  * hang in private_data off the gendisk structure. We may end up
 185  * putting all kinds of interesting stuff here :-)
 186  */
 187 struct blkfront_info
 188 {
 189         struct mutex mutex;
 190         struct xenbus_device *xbdev;
 191         struct gendisk *gd;
 192         int vdevice;
 193         blkif_vdev_t handle;
 194         enum blkif_state connected;
 195         /* Number of pages per ring buffer. */
 196         unsigned int nr_ring_pages;
 197         struct request_queue *rq;
 198         unsigned int feature_flush;
 199         unsigned int feature_fua;
 200         unsigned int feature_discard:1;
 201         unsigned int feature_secdiscard:1;
 202         unsigned int discard_granularity;
 203         unsigned int discard_alignment;
 204         unsigned int feature_persistent:1;
 205         /* Number of 4KB segments handled */
 206         unsigned int max_indirect_segments;
 207         int is_ready;
 208         struct blk_mq_tag_set tag_set;
 209         struct blkfront_ring_info *rinfo;
 210         unsigned int nr_rings;
 211         /* Save uncomplete reqs and bios for migration. */
 212         struct list_head requests;
 213         struct bio_list bio_list;
 214 };
 215
 216 static unsigned int nr_minors;
 217 static unsigned long *minors;
 218 static DEFINE_SPINLOCK(minor_lock);
 219
 220 #define GRANT_INVALID_REF       0
 221
 222 #define PARTS_PER_DISK          16
 223 #define PARTS_PER_EXT_DISK      256
 224
 225 #define BLKIF_MAJOR(dev) ((dev)>>8)
 226 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 227
 228 #define EXT_SHIFT 28
 229 #define EXTENDED (1<<EXT_SHIFT)
 230 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 231 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 232 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 233 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 234 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 235 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 236
 237 #define DEV_NAME        "xvd"   /* name in /dev */
 238
 239 /*
 240  * Grants are always the same size as a Xen page (i.e 4KB).
 241  * A physical segment is always the same size as a Linux page.
 242  * Number of grants per physical segment
 243  */
 244 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 245
 246 #define GRANTS_PER_INDIRECT_FRAME \
 247         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 248
 249 #define PSEGS_PER_INDIRECT_FRAME        \
 250         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
 251
 252 #define INDIRECT_GREFS(_grants)         \
 253         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 254
 255 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
 256
 257 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 258 static void blkfront_gather_backend_features(struct blkfront_info *info);
 259
 260 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 261 {
 262         unsigned long free = rinfo->shadow_free;
 263
 264         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 265         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 266         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 267         return free;
 268 }
 269
 270 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 271                               unsigned long id)
 272 {
 273         if (rinfo->shadow[id].req.u.rw.id != id)
 274                 return -EINVAL;
 275         if (rinfo->shadow[id].request == NULL)
 276                 return -EINVAL;
 277         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 278         rinfo->shadow[id].request = NULL;
 279         rinfo->shadow_free = id;
 280         return 0;
 281 }
 282
 283 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 284 {
 285         struct blkfront_info *info = rinfo->dev_info;
 286         struct page *granted_page;
 287         struct grant *gnt_list_entry, *n;
 288         int i = 0;
 289
 290         while (i < num) {
 291                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 292                 if (!gnt_list_entry)
 293                         goto out_of_memory;
 294
 295                 if (info->feature_persistent) {
 296                         granted_page = alloc_page(GFP_NOIO);
 297                         if (!granted_page) {
 298                                 kfree(gnt_list_entry);
 299                                 goto out_of_memory;
 300                         }
 301                         gnt_list_entry->page = granted_page;
 302                 }
 303
 304                 gnt_list_entry->gref = GRANT_INVALID_REF;
 305                 list_add(&gnt_list_entry->node, &rinfo->grants);
 306                 i++;
 307         }
 308
 309         return 0;
 310
 311 out_of_memory:
 312         list_for_each_entry_safe(gnt_list_entry, n,
 313                                  &rinfo->grants, node) {
 314                 list_del(&gnt_list_entry->node);
 315                 if (info->feature_persistent)
 316                         __free_page(gnt_list_entry->page);
 317                 kfree(gnt_list_entry);
 318                 i--;
 319         }
 320         BUG_ON(i != 0);
 321         return -ENOMEM;
 322 }
 323
 324 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 325 {
 326         struct grant *gnt_list_entry;
 327
 328         BUG_ON(list_empty(&rinfo->grants));
 329         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 330                                           node);
 331         list_del(&gnt_list_entry->node);
 332
 333         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 334                 rinfo->persistent_gnts_c--;
 335
 336         return gnt_list_entry;
 337 }
 338
 339 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 340                                         const struct blkfront_info *info)
 341 {
 342         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 343                                                  info->xbdev->otherend_id,
 344                                                  gnt_list_entry->page,
 345                                                  0);
 346 }
 347
 348 static struct grant *get_grant(grant_ref_t *gref_head,
 349                                unsigned long gfn,
 350                                struct blkfront_ring_info *rinfo)
 351 {
 352         struct grant *gnt_list_entry = get_free_grant(rinfo);
 353         struct blkfront_info *info = rinfo->dev_info;
 354
 355         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 356                 return gnt_list_entry;
 357
 358         /* Assign a gref to this page */
 359         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 360         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 361         if (info->feature_persistent)
 362                 grant_foreign_access(gnt_list_entry, info);
 363         else {
 364                 /* Grant access to the GFN passed by the caller */
 365                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 366                                                 info->xbdev->otherend_id,
 367                                                 gfn, 0);
 368         }
 369
 370         return gnt_list_entry;
 371 }
 372
 373 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 374                                         struct blkfront_ring_info *rinfo)
 375 {
 376         struct grant *gnt_list_entry = get_free_grant(rinfo);
 377         struct blkfront_info *info = rinfo->dev_info;
 378
 379         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 380                 return gnt_list_entry;
 381
 382         /* Assign a gref to this page */
 383         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 384         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 385         if (!info->feature_persistent) {
 386                 struct page *indirect_page;
 387
 388                 /* Fetch a pre-allocated page to use for indirect grefs */
 389                 BUG_ON(list_empty(&rinfo->indirect_pages));
 390                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 391                                                  struct page, lru);
 392                 list_del(&indirect_page->lru);
 393                 gnt_list_entry->page = indirect_page;
 394         }
 395         grant_foreign_access(gnt_list_entry, info);
 396
 397         return gnt_list_entry;
 398 }
 399
 400 static const char *op_name(int op)
 401 {
 402         static const char *const names[] = {
 403                 [BLKIF_OP_READ] = "read",
 404                 [BLKIF_OP_WRITE] = "write",
 405                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 406                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 407                 [BLKIF_OP_DISCARD] = "discard" };
 408
 409         if (op < 0 || op >= ARRAY_SIZE(names))
 410                 return "unknown";
 411
 412         if (!names[op])
 413                 return "reserved";
 414
 415         return names[op];
 416 }
 417 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 418 {
 419         unsigned int end = minor + nr;
 420         int rc;
 421
 422         if (end > nr_minors) {
 423                 unsigned long *bitmap, *old;
 424
 425                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 426                                  GFP_KERNEL);
 427                 if (bitmap == NULL)
 428                         return -ENOMEM;
 429
 430                 spin_lock(&minor_lock);
 431                 if (end > nr_minors) {
 432                         old = minors;
 433                         memcpy(bitmap, minors,
 434                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 435                         minors = bitmap;
 436                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 437                 } else
 438                         old = bitmap;
 439                 spin_unlock(&minor_lock);
 440                 kfree(old);
 441         }
 442
 443         spin_lock(&minor_lock);
 444         if (find_next_bit(minors, end, minor) >= end) {
 445                 bitmap_set(minors, minor, nr);
 446                 rc = 0;
 447         } else
 448                 rc = -EBUSY;
 449         spin_unlock(&minor_lock);
 450
 451         return rc;
 452 }
 453
 454 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 455 {
 456         unsigned int end = minor + nr;
 457
 458         BUG_ON(end > nr_minors);
 459         spin_lock(&minor_lock);
 460         bitmap_clear(minors,  minor, nr);
 461         spin_unlock(&minor_lock);
 462 }
 463
 464 static void blkif_restart_queue_callback(void *arg)
 465 {
 466         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 467         schedule_work(&rinfo->work);
 468 }
 469
 470 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 471 {
 472         /* We don't have real geometry info, but let's at least return
 473            values consistent with the size of the device */
 474         sector_t nsect = get_capacity(bd->bd_disk);
 475         sector_t cylinders = nsect;
 476
 477         hg->heads = 0xff;
 478         hg->sectors = 0x3f;
 479         sector_div(cylinders, hg->heads * hg->sectors);
 480         hg->cylinders = cylinders;
 481         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 482                 hg->cylinders = 0xffff;
 483         return 0;
 484 }
 485
 486 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 487                        unsigned command, unsigned long argument)
 488 {
 489         struct blkfront_info *info = bdev->bd_disk->private_data;
 490         int i;
 491
 492         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 493                 command, (long)argument);
 494
 495         switch (command) {
 496         case CDROMMULTISESSION:
 497                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 498                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 499                         if (put_user(0, (char __user *)(argument + i)))
 500                                 return -EFAULT;
 501                 return 0;
 502
 503         case CDROM_GET_CAPABILITY: {
 504                 struct gendisk *gd = info->gd;
 505                 if (gd->flags & GENHD_FL_CD)
 506                         return 0;
 507                 return -EINVAL;
 508         }
 509
 510         default:
 511                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 512                   command);*/
 513                 return -EINVAL; /* same return as native Linux */
 514         }
 515
 516         return 0;
 517 }
 518
 519 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 520                                             struct request *req,
 521                                             struct blkif_request **ring_req)
 522 {
 523         unsigned long id;
 524
 525         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 526         rinfo->ring.req_prod_pvt++;
 527
 528         id = get_id_from_freelist(rinfo);
 529         rinfo->shadow[id].request = req;
 530         rinfo->shadow[id].status = REQ_WAITING;
 531         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 532
 533         (*ring_req)->u.rw.id = id;
 534
 535         return id;
 536 }
 537
 538 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 539 {
 540         struct blkfront_info *info = rinfo->dev_info;
 541         struct blkif_request *ring_req;
 542         unsigned long id;
 543
 544         /* Fill out a communications ring structure. */
 545         id = blkif_ring_get_request(rinfo, req, &ring_req);
 546
 547         ring_req->operation = BLKIF_OP_DISCARD;
 548         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 549         ring_req->u.discard.id = id;
 550         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 551         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 552                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 553         else
 554                 ring_req->u.discard.flag = 0;
 555
 556         /* Keep a private copy so we can reissue requests when recovering. */
 557         rinfo->shadow[id].req = *ring_req;
 558
 559         return 0;
 560 }
 561
 562 struct setup_rw_req {
 563         unsigned int grant_idx;
 564         struct blkif_request_segment *segments;
 565         struct blkfront_ring_info *rinfo;
 566         struct blkif_request *ring_req;
 567         grant_ref_t gref_head;
 568         unsigned int id;
 569         /* Only used when persistent grant is used and it's a read request */
 570         bool need_copy;
 571         unsigned int bvec_off;
 572         char *bvec_data;
 573
 574         bool require_extra_req;
 575         struct blkif_request *extra_ring_req;
 576 };
 577
 578 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 579                                      unsigned int len, void *data)
 580 {
 581         struct setup_rw_req *setup = data;
 582         int n, ref;
 583         struct grant *gnt_list_entry;
 584         unsigned int fsect, lsect;
 585         /* Convenient aliases */
 586         unsigned int grant_idx = setup->grant_idx;
 587         struct blkif_request *ring_req = setup->ring_req;
 588         struct blkfront_ring_info *rinfo = setup->rinfo;
 589         /*
 590          * We always use the shadow of the first request to store the list
 591          * of grant associated to the block I/O request. This made the
 592          * completion more easy to handle even if the block I/O request is
 593          * split.
 594          */
 595         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 596
 597         if (unlikely(setup->require_extra_req &&
 598                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 599                 /*
 600                  * We are using the second request, setup grant_idx
 601                  * to be the index of the segment array.
 602                  */
 603                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 604                 ring_req = setup->extra_ring_req;
 605         }
 606
 607         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 608             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 609                 if (setup->segments)
 610                         kunmap_atomic(setup->segments);
 611
 612                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 613                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 614                 shadow->indirect_grants[n] = gnt_list_entry;
 615                 setup->segments = kmap_atomic(gnt_list_entry->page);
 616                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 617         }
 618
 619         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 620         ref = gnt_list_entry->gref;
 621         /*
 622          * All the grants are stored in the shadow of the first
 623          * request. Therefore we have to use the global index.
 624          */
 625         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 626
 627         if (setup->need_copy) {
 628                 void *shared_data;
 629
 630                 shared_data = kmap_atomic(gnt_list_entry->page);
 631                 /*
 632                  * this does not wipe data stored outside the
 633                  * range sg->offset..sg->offset+sg->length.
 634                  * Therefore, blkback *could* see data from
 635                  * previous requests. This is OK as long as
 636                  * persistent grants are shared with just one
 637                  * domain. It may need refactoring if this
 638                  * changes
 639                  */
 640                 memcpy(shared_data + offset,
 641                        setup->bvec_data + setup->bvec_off,
 642                        len);
 643
 644                 kunmap_atomic(shared_data);
 645                 setup->bvec_off += len;
 646         }
 647
 648         fsect = offset >> 9;
 649         lsect = fsect + (len >> 9) - 1;
 650         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 651                 ring_req->u.rw.seg[grant_idx] =
 652                         (struct blkif_request_segment) {
 653                                 .gref       = ref,
 654                                 .first_sect = fsect,
 655                                 .last_sect  = lsect };
 656         } else {
 657                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 658                         (struct blkif_request_segment) {
 659                                 .gref       = ref,
 660                                 .first_sect = fsect,
 661                                 .last_sect  = lsect };
 662         }
 663
 664         (setup->grant_idx)++;
 665 }
 666
 667 static void blkif_setup_extra_req(struct blkif_request *first,
 668                                   struct blkif_request *second)
 669 {
 670         uint16_t nr_segments = first->u.rw.nr_segments;
 671
 672         /*
 673          * The second request is only present when the first request uses
 674          * all its segments. It's always the continuity of the first one.
 675          */
 676         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 677
 678         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 679         second->u.rw.sector_number = first->u.rw.sector_number +
 680                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 681
 682         second->u.rw.handle = first->u.rw.handle;
 683         second->operation = first->operation;
 684 }
 685
 686 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 687 {
 688         struct blkfront_info *info = rinfo->dev_info;
 689         struct blkif_request *ring_req, *extra_ring_req = NULL;
 690         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 691         bool require_extra_req = false;
 692         int i;
 693         struct setup_rw_req setup = {
 694                 .grant_idx = 0,
 695                 .segments = NULL,
 696                 .rinfo = rinfo,
 697                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 698         };
 699
 700         /*
 701          * Used to store if we are able to queue the request by just using
 702          * existing persistent grants, or if we have to get new grants,
 703          * as there are not sufficiently many free.
 704          */
 705         struct scatterlist *sg;
 706         int num_sg, max_grefs, num_grant;
 707
 708         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 709         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 710                 /*
 711                  * If we are using indirect segments we need to account
 712                  * for the indirect grefs used in the request.
 713                  */
 714                 max_grefs += INDIRECT_GREFS(max_grefs);
 715
 716         /*
 717          * We have to reserve 'max_grefs' grants because persistent
 718          * grants are shared by all rings.
 719          */
 720         if (max_grefs > 0)
 721                 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
 722                         gnttab_request_free_callback(
 723                                 &rinfo->callback,
 724                                 blkif_restart_queue_callback,
 725                                 rinfo,
 726                                 max_grefs);
 727                         return 1;
 728                 }
 729
 730         /* Fill out a communications ring structure. */
 731         id = blkif_ring_get_request(rinfo, req, &ring_req);
 732
 733         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 734         num_grant = 0;
 735         /* Calculate the number of grant used */
 736         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 737                num_grant += gnttab_count_grant(sg->offset, sg->length);
 738
 739         require_extra_req = info->max_indirect_segments == 0 &&
 740                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 741         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 742
 743         rinfo->shadow[id].num_sg = num_sg;
 744         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 745             likely(!require_extra_req)) {
 746                 /*
 747                  * The indirect operation can only be a BLKIF_OP_READ or
 748                  * BLKIF_OP_WRITE
 749                  */
 750                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 751                 ring_req->operation = BLKIF_OP_INDIRECT;
 752                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 753                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 754                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 755                 ring_req->u.indirect.handle = info->handle;
 756                 ring_req->u.indirect.nr_segments = num_grant;
 757         } else {
 758                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 759                 ring_req->u.rw.handle = info->handle;
 760                 ring_req->operation = rq_data_dir(req) ?
 761                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 762                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 763                         /*
 764                          * Ideally we can do an unordered flush-to-disk.
 765                          * In case the backend onlysupports barriers, use that.
 766                          * A barrier request a superset of FUA, so we can
 767                          * implement it the same way.  (It's also a FLUSH+FUA,
 768                          * since it is guaranteed ordered WRT previous writes.)
 769                          */
 770                         if (info->feature_flush && info->feature_fua)
 771                                 ring_req->operation =
 772                                         BLKIF_OP_WRITE_BARRIER;
 773                         else if (info->feature_flush)
 774                                 ring_req->operation =
 775                                         BLKIF_OP_FLUSH_DISKCACHE;
 776                         else
 777                                 ring_req->operation = 0;
 778                 }
 779                 ring_req->u.rw.nr_segments = num_grant;
 780                 if (unlikely(require_extra_req)) {
 781                         extra_id = blkif_ring_get_request(rinfo, req,
 782                                                           &extra_ring_req);
 783                         /*
 784                          * Only the first request contains the scatter-gather
 785                          * list.
 786                          */
 787                         rinfo->shadow[extra_id].num_sg = 0;
 788
 789                         blkif_setup_extra_req(ring_req, extra_ring_req);
 790
 791                         /* Link the 2 requests together */
 792                         rinfo->shadow[extra_id].associated_id = id;
 793                         rinfo->shadow[id].associated_id = extra_id;
 794                 }
 795         }
 796
 797         setup.ring_req = ring_req;
 798         setup.id = id;
 799
 800         setup.require_extra_req = require_extra_req;
 801         if (unlikely(require_extra_req))
 802                 setup.extra_ring_req = extra_ring_req;
 803
 804         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 805                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 806
 807                 if (setup.need_copy) {
 808                         setup.bvec_off = sg->offset;
 809                         setup.bvec_data = kmap_atomic(sg_page(sg));
 810                 }
 811
 812                 gnttab_foreach_grant_in_range(sg_page(sg),
 813                                               sg->offset,
 814                                               sg->length,
 815                                               blkif_setup_rw_req_grant,
 816                                               &setup);
 817
 818                 if (setup.need_copy)
 819                         kunmap_atomic(setup.bvec_data);
 820         }
 821         if (setup.segments)
 822                 kunmap_atomic(setup.segments);
 823
 824         /* Keep a private copy so we can reissue requests when recovering. */
 825         rinfo->shadow[id].req = *ring_req;
 826         if (unlikely(require_extra_req))
 827                 rinfo->shadow[extra_id].req = *extra_ring_req;
 828
 829         if (max_grefs > 0)
 830                 gnttab_free_grant_references(setup.gref_head);
 831
 832         return 0;
 833 }
 834
 835 /*
 836  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 837  * and writes are handled as expected.
 838  *
 839  * @req: a request struct
 840  */
 841 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 842 {
 843         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 844                 return 1;
 845
 846         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 847                      req_op(req) == REQ_OP_SECURE_ERASE))
 848                 return blkif_queue_discard_req(req, rinfo);
 849         else
 850                 return blkif_queue_rw_req(req, rinfo);
 851 }
 852
 853 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 854 {
 855         int notify;
 856
 857         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 858
 859         if (notify)
 860                 notify_remote_via_irq(rinfo->irq);
 861 }
 862
 863 static inline bool blkif_request_flush_invalid(struct request *req,
 864                                                struct blkfront_info *info)
 865 {
 866         return ((req->cmd_type != REQ_TYPE_FS) ||
 867                 ((req_op(req) == REQ_OP_FLUSH) &&
 868                  !info->feature_flush) ||
 869                 ((req->cmd_flags & REQ_FUA) &&
 870                  !info->feature_fua));
 871 }
 872
 873 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 874                           const struct blk_mq_queue_data *qd)
 875 {
 876         unsigned long flags;
 877         int qid = hctx->queue_num;
 878         struct blkfront_info *info = hctx->queue->queuedata;
 879         struct blkfront_ring_info *rinfo = NULL;
 880
 881         BUG_ON(info->nr_rings <= qid);
 882         rinfo = &info->rinfo[qid];
 883         blk_mq_start_request(qd->rq);
 884         spin_lock_irqsave(&rinfo->ring_lock, flags);
 885         if (RING_FULL(&rinfo->ring))
 886                 goto out_busy;
 887
 888         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 889                 goto out_err;
 890
 891         if (blkif_queue_request(qd->rq, rinfo))
 892                 goto out_busy;
 893
 894         flush_requests(rinfo);
 895         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 896         return BLK_MQ_RQ_QUEUE_OK;
 897
 898 out_err:
 899         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 900         return BLK_MQ_RQ_QUEUE_ERROR;
 901
 902 out_busy:
 903         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 904         blk_mq_stop_hw_queue(hctx);
 905         return BLK_MQ_RQ_QUEUE_BUSY;
 906 }
 907
 908 static struct blk_mq_ops blkfront_mq_ops = {
 909         .queue_rq = blkif_queue_rq,
 910         .map_queue = blk_mq_map_queue,
 911 };
 912
 913 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 914                                 unsigned int physical_sector_size,
 915                                 unsigned int segments)
 916 {
 917         struct request_queue *rq;
 918         struct blkfront_info *info = gd->private_data;
 919
 920         memset(&info->tag_set, 0, sizeof(info->tag_set));
 921         info->tag_set.ops = &blkfront_mq_ops;
 922         info->tag_set.nr_hw_queues = info->nr_rings;
 923         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 924                 /*
 925                  * When indirect descriptior is not supported, the I/O request
 926                  * will be split between multiple request in the ring.
 927                  * To avoid problems when sending the request, divide by
 928                  * 2 the depth of the queue.
 929                  */
 930                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 931         } else
 932                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 933         info->tag_set.numa_node = NUMA_NO_NODE;
 934         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 935         info->tag_set.cmd_size = 0;
 936         info->tag_set.driver_data = info;
 937
 938         if (blk_mq_alloc_tag_set(&info->tag_set))
 939                 return -EINVAL;
 940         rq = blk_mq_init_queue(&info->tag_set);
 941         if (IS_ERR(rq)) {
 942                 blk_mq_free_tag_set(&info->tag_set);
 943                 return PTR_ERR(rq);
 944         }
 945
 946         rq->queuedata = info;
 947         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 948
 949         if (info->feature_discard) {
 950                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 951                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 952                 rq->limits.discard_granularity = info->discard_granularity;
 953                 rq->limits.discard_alignment = info->discard_alignment;
 954                 if (info->feature_secdiscard)
 955                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
 956         }
 957
 958         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 959         blk_queue_logical_block_size(rq, sector_size);
 960         blk_queue_physical_block_size(rq, physical_sector_size);
 961         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 962
 963         /* Each segment in a request is up to an aligned page in size. */
 964         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 965         blk_queue_max_segment_size(rq, PAGE_SIZE);
 966
 967         /* Ensure a merged request will fit in a single I/O ring slot. */
 968         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 969
 970         /* Make sure buffer addresses are sector-aligned. */
 971         blk_queue_dma_alignment(rq, 511);
 972
 973         /* Make sure we don't use bounce buffers. */
 974         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 975
 976         gd->queue = rq;
 977
 978         return 0;
 979 }
 980
 981 static const char *flush_info(struct blkfront_info *info)
 982 {
 983         if (info->feature_flush && info->feature_fua)
 984                 return "barrier: enabled;";
 985         else if (info->feature_flush)
 986                 return "flush diskcache: enabled;";
 987         else
 988                 return "barrier or flush: disabled;";
 989 }
 990
 991 static void xlvbd_flush(struct blkfront_info *info)
 992 {
 993         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 994                               info->feature_fua ? true : false);
 995         pr_info("blkfront: %s: %s %s %s %s %s\n",
 996                 info->gd->disk_name, flush_info(info),
 997                 "persistent grants:", info->feature_persistent ?
 998                 "enabled;" : "disabled;", "indirect descriptors:",
 999                 info->max_indirect_segments ? "enabled;" : "disabled;");
1000 }
1001
1002 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1003 {
1004         int major;
1005         major = BLKIF_MAJOR(vdevice);
1006         *minor = BLKIF_MINOR(vdevice);
1007         switch (major) {
1008                 case XEN_IDE0_MAJOR:
1009                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1010                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1011                                 EMULATED_HD_DISK_MINOR_OFFSET;
1012                         break;
1013                 case XEN_IDE1_MAJOR:
1014                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1015                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1016                                 EMULATED_HD_DISK_MINOR_OFFSET;
1017                         break;
1018                 case XEN_SCSI_DISK0_MAJOR:
1019                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1020                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1021                         break;
1022                 case XEN_SCSI_DISK1_MAJOR:
1023                 case XEN_SCSI_DISK2_MAJOR:
1024                 case XEN_SCSI_DISK3_MAJOR:
1025                 case XEN_SCSI_DISK4_MAJOR:
1026                 case XEN_SCSI_DISK5_MAJOR:
1027                 case XEN_SCSI_DISK6_MAJOR:
1028                 case XEN_SCSI_DISK7_MAJOR:
1029                         *offset = (*minor / PARTS_PER_DISK) +
1030                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1031                                 EMULATED_SD_DISK_NAME_OFFSET;
1032                         *minor = *minor +
1033                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1034                                 EMULATED_SD_DISK_MINOR_OFFSET;
1035                         break;
1036                 case XEN_SCSI_DISK8_MAJOR:
1037                 case XEN_SCSI_DISK9_MAJOR:
1038                 case XEN_SCSI_DISK10_MAJOR:
1039                 case XEN_SCSI_DISK11_MAJOR:
1040                 case XEN_SCSI_DISK12_MAJOR:
1041                 case XEN_SCSI_DISK13_MAJOR:
1042                 case XEN_SCSI_DISK14_MAJOR:
1043                 case XEN_SCSI_DISK15_MAJOR:
1044                         *offset = (*minor / PARTS_PER_DISK) +
1045                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1046                                 EMULATED_SD_DISK_NAME_OFFSET;
1047                         *minor = *minor +
1048                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1049                                 EMULATED_SD_DISK_MINOR_OFFSET;
1050                         break;
1051                 case XENVBD_MAJOR:
1052                         *offset = *minor / PARTS_PER_DISK;
1053                         break;
1054                 default:
1055                         printk(KERN_WARNING "blkfront: your disk configuration is "
1056                                         "incorrect, please use an xvd device instead\n");
1057                         return -ENODEV;
1058         }
1059         return 0;
1060 }
1061
1062 static char *encode_disk_name(char *ptr, unsigned int n)
1063 {
1064         if (n >= 26)
1065                 ptr = encode_disk_name(ptr, n / 26 - 1);
1066         *ptr = 'a' + n % 26;
1067         return ptr + 1;
1068 }
1069
1070 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1071                                struct blkfront_info *info,
1072                                u16 vdisk_info, u16 sector_size,
1073                                unsigned int physical_sector_size)
1074 {
1075         struct gendisk *gd;
1076         int nr_minors = 1;
1077         int err;
1078         unsigned int offset;
1079         int minor;
1080         int nr_parts;
1081         char *ptr;
1082
1083         BUG_ON(info->gd != NULL);
1084         BUG_ON(info->rq != NULL);
1085
1086         if ((info->vdevice>>EXT_SHIFT) > 1) {
1087                 /* this is above the extended range; something is wrong */
1088                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1089                 return -ENODEV;
1090         }
1091
1092         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1093                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1094                 if (err)
1095                         return err;
1096                 nr_parts = PARTS_PER_DISK;
1097         } else {
1098                 minor = BLKIF_MINOR_EXT(info->vdevice);
1099                 nr_parts = PARTS_PER_EXT_DISK;
1100                 offset = minor / nr_parts;
1101                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1102                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1103                                         "emulated IDE disks,\n\t choose an xvd device name"
1104                                         "from xvde on\n", info->vdevice);
1105         }
1106         if (minor >> MINORBITS) {
1107                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1108                         info->vdevice, minor);
1109                 return -ENODEV;
1110         }
1111
1112         if ((minor % nr_parts) == 0)
1113                 nr_minors = nr_parts;
1114
1115         err = xlbd_reserve_minors(minor, nr_minors);
1116         if (err)
1117                 goto out;
1118         err = -ENODEV;
1119
1120         gd = alloc_disk(nr_minors);
1121         if (gd == NULL)
1122                 goto release;
1123
1124         strcpy(gd->disk_name, DEV_NAME);
1125         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1126         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1127         if (nr_minors > 1)
1128                 *ptr = 0;
1129         else
1130                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1131                          "%d", minor & (nr_parts - 1));
1132
1133         gd->major = XENVBD_MAJOR;
1134         gd->first_minor = minor;
1135         gd->fops = &xlvbd_block_fops;
1136         gd->private_data = info;
1137         set_capacity(gd, capacity);
1138
1139         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1140                                  info->max_indirect_segments ? :
1141                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1142                 del_gendisk(gd);
1143                 goto release;
1144         }
1145
1146         info->rq = gd->queue;
1147         info->gd = gd;
1148
1149         xlvbd_flush(info);
1150
1151         if (vdisk_info & VDISK_READONLY)
1152                 set_disk_ro(gd, 1);
1153
1154         if (vdisk_info & VDISK_REMOVABLE)
1155                 gd->flags |= GENHD_FL_REMOVABLE;
1156
1157         if (vdisk_info & VDISK_CDROM)
1158                 gd->flags |= GENHD_FL_CD;
1159
1160         return 0;
1161
1162  release:
1163         xlbd_release_minors(minor, nr_minors);
1164  out:
1165         return err;
1166 }
1167
1168 static void xlvbd_release_gendisk(struct blkfront_info *info)
1169 {
1170         unsigned int minor, nr_minors, i;
1171
1172         if (info->rq == NULL)
1173                 return;
1174
1175         /* No more blkif_request(). */
1176         blk_mq_stop_hw_queues(info->rq);
1177
1178         for (i = 0; i < info->nr_rings; i++) {
1179                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1180
1181                 /* No more gnttab callback work. */
1182                 gnttab_cancel_free_callback(&rinfo->callback);
1183
1184                 /* Flush gnttab callback work. Must be done with no locks held. */
1185                 flush_work(&rinfo->work);
1186         }
1187
1188         del_gendisk(info->gd);
1189
1190         minor = info->gd->first_minor;
1191         nr_minors = info->gd->minors;
1192         xlbd_release_minors(minor, nr_minors);
1193
1194         blk_cleanup_queue(info->rq);
1195         blk_mq_free_tag_set(&info->tag_set);
1196         info->rq = NULL;
1197
1198         put_disk(info->gd);
1199         info->gd = NULL;
1200 }
1201
1202 /* Already hold rinfo->ring_lock. */
1203 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1204 {
1205         if (!RING_FULL(&rinfo->ring))
1206                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1207 }
1208
1209 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1210 {
1211         unsigned long flags;
1212
1213         spin_lock_irqsave(&rinfo->ring_lock, flags);
1214         kick_pending_request_queues_locked(rinfo);
1215         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1216 }
1217
1218 static void blkif_restart_queue(struct work_struct *work)
1219 {
1220         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1221
1222         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1223                 kick_pending_request_queues(rinfo);
1224 }
1225
1226 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1227 {
1228         struct grant *persistent_gnt, *n;
1229         struct blkfront_info *info = rinfo->dev_info;
1230         int i, j, segs;
1231
1232         /*
1233          * Remove indirect pages, this only happens when using indirect
1234          * descriptors but not persistent grants
1235          */
1236         if (!list_empty(&rinfo->indirect_pages)) {
1237                 struct page *indirect_page, *n;
1238
1239                 BUG_ON(info->feature_persistent);
1240                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1241                         list_del(&indirect_page->lru);
1242                         __free_page(indirect_page);
1243                 }
1244         }
1245
1246         /* Remove all persistent grants. */
1247         if (!list_empty(&rinfo->grants)) {
1248                 list_for_each_entry_safe(persistent_gnt, n,
1249                                          &rinfo->grants, node) {
1250                         list_del(&persistent_gnt->node);
1251                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1252                                 gnttab_end_foreign_access(persistent_gnt->gref,
1253                                                           0, 0UL);
1254                                 rinfo->persistent_gnts_c--;
1255                         }
1256                         if (info->feature_persistent)
1257                                 __free_page(persistent_gnt->page);
1258                         kfree(persistent_gnt);
1259                 }
1260         }
1261         BUG_ON(rinfo->persistent_gnts_c != 0);
1262
1263         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1264                 /*
1265                  * Clear persistent grants present in requests already
1266                  * on the shared ring
1267                  */
1268                 if (!rinfo->shadow[i].request)
1269                         goto free_shadow;
1270
1271                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1272                        rinfo->shadow[i].req.u.indirect.nr_segments :
1273                        rinfo->shadow[i].req.u.rw.nr_segments;
1274                 for (j = 0; j < segs; j++) {
1275                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1276                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1277                         if (info->feature_persistent)
1278                                 __free_page(persistent_gnt->page);
1279                         kfree(persistent_gnt);
1280                 }
1281
1282                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1283                         /*
1284                          * If this is not an indirect operation don't try to
1285                          * free indirect segments
1286                          */
1287                         goto free_shadow;
1288
1289                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1290                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1291                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1292                         __free_page(persistent_gnt->page);
1293                         kfree(persistent_gnt);
1294                 }
1295
1296 free_shadow:
1297                 kfree(rinfo->shadow[i].grants_used);
1298                 rinfo->shadow[i].grants_used = NULL;
1299                 kfree(rinfo->shadow[i].indirect_grants);
1300                 rinfo->shadow[i].indirect_grants = NULL;
1301                 kfree(rinfo->shadow[i].sg);
1302                 rinfo->shadow[i].sg = NULL;
1303         }
1304
1305         /* No more gnttab callback work. */
1306         gnttab_cancel_free_callback(&rinfo->callback);
1307
1308         /* Flush gnttab callback work. Must be done with no locks held. */
1309         flush_work(&rinfo->work);
1310
1311         /* Free resources associated with old device channel. */
1312         for (i = 0; i < info->nr_ring_pages; i++) {
1313                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1314                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1315                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1316                 }
1317         }
1318         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1319         rinfo->ring.sring = NULL;
1320
1321         if (rinfo->irq)
1322                 unbind_from_irqhandler(rinfo->irq, rinfo);
1323         rinfo->evtchn = rinfo->irq = 0;
1324 }
1325
1326 static void blkif_free(struct blkfront_info *info, int suspend)
1327 {
1328         unsigned int i;
1329
1330         /* Prevent new requests being issued until we fix things up. */
1331         info->connected = suspend ?
1332                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1333         /* No more blkif_request(). */
1334         if (info->rq)
1335                 blk_mq_stop_hw_queues(info->rq);
1336
1337         for (i = 0; i < info->nr_rings; i++)
1338                 blkif_free_ring(&info->rinfo[i]);
1339
1340         kfree(info->rinfo);
1341         info->rinfo = NULL;
1342         info->nr_rings = 0;
1343 }
1344
1345 struct copy_from_grant {
1346         const struct blk_shadow *s;
1347         unsigned int grant_idx;
1348         unsigned int bvec_offset;
1349         char *bvec_data;
1350 };
1351
1352 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1353                                   unsigned int len, void *data)
1354 {
1355         struct copy_from_grant *info = data;
1356         char *shared_data;
1357         /* Convenient aliases */
1358         const struct blk_shadow *s = info->s;
1359
1360         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1361
1362         memcpy(info->bvec_data + info->bvec_offset,
1363                shared_data + offset, len);
1364
1365         info->bvec_offset += len;
1366         info->grant_idx++;
1367
1368         kunmap_atomic(shared_data);
1369 }
1370
1371 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1372 {
1373         switch (rsp)
1374         {
1375         case BLKIF_RSP_OKAY:
1376                 return REQ_DONE;
1377         case BLKIF_RSP_EOPNOTSUPP:
1378                 return REQ_EOPNOTSUPP;
1379         case BLKIF_RSP_ERROR:
1380                 /* Fallthrough. */
1381         default:
1382                 return REQ_ERROR;
1383         }
1384 }
1385
1386 /*
1387  * Get the final status of the block request based on two ring response
1388  */
1389 static int blkif_get_final_status(enum blk_req_status s1,
1390                                   enum blk_req_status s2)
1391 {
1392         BUG_ON(s1 == REQ_WAITING);
1393         BUG_ON(s2 == REQ_WAITING);
1394
1395         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1396                 return BLKIF_RSP_ERROR;
1397         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1398                 return BLKIF_RSP_EOPNOTSUPP;
1399         return BLKIF_RSP_OKAY;
1400 }
1401
1402 static bool blkif_completion(unsigned long *id,
1403                              struct blkfront_ring_info *rinfo,
1404                              struct blkif_response *bret)
1405 {
1406         int i = 0;
1407         struct scatterlist *sg;
1408         int num_sg, num_grant;
1409         struct blkfront_info *info = rinfo->dev_info;
1410         struct blk_shadow *s = &rinfo->shadow[*id];
1411         struct copy_from_grant data = {
1412                 .grant_idx = 0,
1413         };
1414
1415         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1416                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1417
1418         /* The I/O request may be split in two. */
1419         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1420                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1421
1422                 /* Keep the status of the current response in shadow. */
1423                 s->status = blkif_rsp_to_req_status(bret->status);
1424
1425                 /* Wait the second response if not yet here. */
1426                 if (s2->status == REQ_WAITING)
1427                         return 0;
1428
1429                 bret->status = blkif_get_final_status(s->status,
1430                                                       s2->status);
1431
1432                 /*
1433                  * All the grants is stored in the first shadow in order
1434                  * to make the completion code simpler.
1435                  */
1436                 num_grant += s2->req.u.rw.nr_segments;
1437
1438                 /*
1439                  * The two responses may not come in order. Only the
1440                  * first request will store the scatter-gather list.
1441                  */
1442                 if (s2->num_sg != 0) {
1443                         /* Update "id" with the ID of the first response. */
1444                         *id = s->associated_id;
1445                         s = s2;
1446                 }
1447
1448                 /*
1449                  * We don't need anymore the second request, so recycling
1450                  * it now.
1451                  */
1452                 if (add_id_to_freelist(rinfo, s->associated_id))
1453                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1454                              info->gd->disk_name, s->associated_id);
1455         }
1456
1457         data.s = s;
1458         num_sg = s->num_sg;
1459
1460         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1461                 for_each_sg(s->sg, sg, num_sg, i) {
1462                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1463
1464                         data.bvec_offset = sg->offset;
1465                         data.bvec_data = kmap_atomic(sg_page(sg));
1466
1467                         gnttab_foreach_grant_in_range(sg_page(sg),
1468                                                       sg->offset,
1469                                                       sg->length,
1470                                                       blkif_copy_from_grant,
1471                                                       &data);
1472
1473                         kunmap_atomic(data.bvec_data);
1474                 }
1475         }
1476         /* Add the persistent grant into the list of free grants */
1477         for (i = 0; i < num_grant; i++) {
1478                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1479                         /*
1480                          * If the grant is still mapped by the backend (the
1481                          * backend has chosen to make this grant persistent)
1482                          * we add it at the head of the list, so it will be
1483                          * reused first.
1484                          */
1485                         if (!info->feature_persistent)
1486                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1487                                                      s->grants_used[i]->gref);
1488                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1489                         rinfo->persistent_gnts_c++;
1490                 } else {
1491                         /*
1492                          * If the grant is not mapped by the backend we end the
1493                          * foreign access and add it to the tail of the list,
1494                          * so it will not be picked again unless we run out of
1495                          * persistent grants.
1496                          */
1497                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1498                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1499                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1500                 }
1501         }
1502         if (s->req.operation == BLKIF_OP_INDIRECT) {
1503                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1504                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1505                                 if (!info->feature_persistent)
1506                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1507                                                              s->indirect_grants[i]->gref);
1508                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1509                                 rinfo->persistent_gnts_c++;
1510                         } else {
1511                                 struct page *indirect_page;
1512
1513                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1514                                 /*
1515                                  * Add the used indirect page back to the list of
1516                                  * available pages for indirect grefs.
1517                                  */
1518                                 if (!info->feature_persistent) {
1519                                         indirect_page = s->indirect_grants[i]->page;
1520                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1521                                 }
1522                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1523                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1524                         }
1525                 }
1526         }
1527
1528         return 1;
1529 }
1530
1531 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1532 {
1533         struct request *req;
1534         struct blkif_response *bret;
1535         RING_IDX i, rp;
1536         unsigned long flags;
1537         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1538         struct blkfront_info *info = rinfo->dev_info;
1539         int error;
1540
1541         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1542                 return IRQ_HANDLED;
1543
1544         spin_lock_irqsave(&rinfo->ring_lock, flags);
1545  again:
1546         rp = rinfo->ring.sring->rsp_prod;
1547         rmb(); /* Ensure we see queued responses up to 'rp'. */
1548
1549         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1550                 unsigned long id;
1551
1552                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1553                 id   = bret->id;
1554                 /*
1555                  * The backend has messed up and given us an id that we would
1556                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1557                  * look in get_id_from_freelist.
1558                  */
1559                 if (id >= BLK_RING_SIZE(info)) {
1560                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1561                              info->gd->disk_name, op_name(bret->operation), id);
1562                         /* We can't safely get the 'struct request' as
1563                          * the id is busted. */
1564                         continue;
1565                 }
1566                 req  = rinfo->shadow[id].request;
1567
1568                 if (bret->operation != BLKIF_OP_DISCARD) {
1569                         /*
1570                          * We may need to wait for an extra response if the
1571                          * I/O request is split in 2
1572                          */
1573                         if (!blkif_completion(&id, rinfo, bret))
1574                                 continue;
1575                 }
1576
1577                 if (add_id_to_freelist(rinfo, id)) {
1578                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1579                              info->gd->disk_name, op_name(bret->operation), id);
1580                         continue;
1581                 }
1582
1583                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1584                 switch (bret->operation) {
1585                 case BLKIF_OP_DISCARD:
1586                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1587                                 struct request_queue *rq = info->rq;
1588                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1589                                            info->gd->disk_name, op_name(bret->operation));
1590                                 error = -EOPNOTSUPP;
1591                                 info->feature_discard = 0;
1592                                 info->feature_secdiscard = 0;
1593                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1594                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1595                         }
1596                         blk_mq_complete_request(req, error);
1597                         break;
1598                 case BLKIF_OP_FLUSH_DISKCACHE:
1599                 case BLKIF_OP_WRITE_BARRIER:
1600                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1601                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1602                                        info->gd->disk_name, op_name(bret->operation));
1603                                 error = -EOPNOTSUPP;
1604                         }
1605                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1606                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1607                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1608                                        info->gd->disk_name, op_name(bret->operation));
1609                                 error = -EOPNOTSUPP;
1610                         }
1611                         if (unlikely(error)) {
1612                                 if (error == -EOPNOTSUPP)
1613                                         error = 0;
1614                                 info->feature_fua = 0;
1615                                 info->feature_flush = 0;
1616                                 xlvbd_flush(info);
1617                         }
1618                         /* fall through */
1619                 case BLKIF_OP_READ:
1620                 case BLKIF_OP_WRITE:
1621                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1622                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1623                                         "request: %x\n", bret->status);
1624
1625                         blk_mq_complete_request(req, error);
1626                         break;
1627                 default:
1628                         BUG();
1629                 }
1630         }
1631
1632         rinfo->ring.rsp_cons = i;
1633
1634         if (i != rinfo->ring.req_prod_pvt) {
1635                 int more_to_do;
1636                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1637                 if (more_to_do)
1638                         goto again;
1639         } else
1640                 rinfo->ring.sring->rsp_event = i + 1;
1641
1642         kick_pending_request_queues_locked(rinfo);
1643
1644         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1645
1646         return IRQ_HANDLED;
1647 }
1648
1649
1650 static int setup_blkring(struct xenbus_device *dev,
1651                          struct blkfront_ring_info *rinfo)
1652 {
1653         struct blkif_sring *sring;
1654         int err, i;
1655         struct blkfront_info *info = rinfo->dev_info;
1656         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1657         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1658
1659         for (i = 0; i < info->nr_ring_pages; i++)
1660                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1661
1662         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1663                                                        get_order(ring_size));
1664         if (!sring) {
1665                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1666                 return -ENOMEM;
1667         }
1668         SHARED_RING_INIT(sring);
1669         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1670
1671         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1672         if (err < 0) {
1673                 free_pages((unsigned long)sring, get_order(ring_size));
1674                 rinfo->ring.sring = NULL;
1675                 goto fail;
1676         }
1677         for (i = 0; i < info->nr_ring_pages; i++)
1678                 rinfo->ring_ref[i] = gref[i];
1679
1680         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1681         if (err)
1682                 goto fail;
1683
1684         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1685                                         "blkif", rinfo);
1686         if (err <= 0) {
1687                 xenbus_dev_fatal(dev, err,
1688                                  "bind_evtchn_to_irqhandler failed");
1689                 goto fail;
1690         }
1691         rinfo->irq = err;
1692
1693         return 0;
1694 fail:
1695         blkif_free(info, 0);
1696         return err;
1697 }
1698
1699 /*
1700  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1701  * ring buffer may have multi pages depending on ->nr_ring_pages.
1702  */
1703 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1704                                 struct blkfront_ring_info *rinfo, const char *dir)
1705 {
1706         int err;
1707         unsigned int i;
1708         const char *message = NULL;
1709         struct blkfront_info *info = rinfo->dev_info;
1710
1711         if (info->nr_ring_pages == 1) {
1712                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1713                 if (err) {
1714                         message = "writing ring-ref";
1715                         goto abort_transaction;
1716                 }
1717         } else {
1718                 for (i = 0; i < info->nr_ring_pages; i++) {
1719                         char ring_ref_name[RINGREF_NAME_LEN];
1720
1721                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1722                         err = xenbus_printf(xbt, dir, ring_ref_name,
1723                                             "%u", rinfo->ring_ref[i]);
1724                         if (err) {
1725                                 message = "writing ring-ref";
1726                                 goto abort_transaction;
1727                         }
1728                 }
1729         }
1730
1731         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1732         if (err) {
1733                 message = "writing event-channel";
1734                 goto abort_transaction;
1735         }
1736
1737         return 0;
1738
1739 abort_transaction:
1740         xenbus_transaction_end(xbt, 1);
1741         if (message)
1742                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1743
1744         return err;
1745 }
1746
1747 /* Common code used when first setting up, and when resuming. */
1748 static int talk_to_blkback(struct xenbus_device *dev,
1749                            struct blkfront_info *info)
1750 {
1751         const char *message = NULL;
1752         struct xenbus_transaction xbt;
1753         int err;
1754         unsigned int i, max_page_order = 0;
1755         unsigned int ring_page_order = 0;
1756
1757         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1758                            "max-ring-page-order", "%u", &max_page_order);
1759         if (err != 1)
1760                 info->nr_ring_pages = 1;
1761         else {
1762                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1763                 info->nr_ring_pages = 1 << ring_page_order;
1764         }
1765
1766         for (i = 0; i < info->nr_rings; i++) {
1767                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1768
1769                 /* Create shared ring, alloc event channel. */
1770                 err = setup_blkring(dev, rinfo);
1771                 if (err)
1772                         goto destroy_blkring;
1773         }
1774
1775 again:
1776         err = xenbus_transaction_start(&xbt);
1777         if (err) {
1778                 xenbus_dev_fatal(dev, err, "starting transaction");
1779                 goto destroy_blkring;
1780         }
1781
1782         if (info->nr_ring_pages > 1) {
1783                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1784                                     ring_page_order);
1785                 if (err) {
1786                         message = "writing ring-page-order";
1787                         goto abort_transaction;
1788                 }
1789         }
1790
1791         /* We already got the number of queues/rings in _probe */
1792         if (info->nr_rings == 1) {
1793                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1794                 if (err)
1795                         goto destroy_blkring;
1796         } else {
1797                 char *path;
1798                 size_t pathsize;
1799
1800                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1801                                     info->nr_rings);
1802                 if (err) {
1803                         message = "writing multi-queue-num-queues";
1804                         goto abort_transaction;
1805                 }
1806
1807                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1808                 path = kmalloc(pathsize, GFP_KERNEL);
1809                 if (!path) {
1810                         err = -ENOMEM;
1811                         message = "ENOMEM while writing ring references";
1812                         goto abort_transaction;
1813                 }
1814
1815                 for (i = 0; i < info->nr_rings; i++) {
1816                         memset(path, 0, pathsize);
1817                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1818                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1819                         if (err) {
1820                                 kfree(path);
1821                                 goto destroy_blkring;
1822                         }
1823                 }
1824                 kfree(path);
1825         }
1826         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1827                             XEN_IO_PROTO_ABI_NATIVE);
1828         if (err) {
1829                 message = "writing protocol";
1830                 goto abort_transaction;
1831         }
1832         err = xenbus_printf(xbt, dev->nodename,
1833                             "feature-persistent", "%u", 1);
1834         if (err)
1835                 dev_warn(&dev->dev,
1836                          "writing persistent grants feature to xenbus");
1837
1838         err = xenbus_transaction_end(xbt, 0);
1839         if (err) {
1840                 if (err == -EAGAIN)
1841                         goto again;
1842                 xenbus_dev_fatal(dev, err, "completing transaction");
1843                 goto destroy_blkring;
1844         }
1845
1846         for (i = 0; i < info->nr_rings; i++) {
1847                 unsigned int j;
1848                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1849
1850                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1851                         rinfo->shadow[j].req.u.rw.id = j + 1;
1852                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1853         }
1854         xenbus_switch_state(dev, XenbusStateInitialised);
1855
1856         return 0;
1857
1858  abort_transaction:
1859         xenbus_transaction_end(xbt, 1);
1860         if (message)
1861                 xenbus_dev_fatal(dev, err, "%s", message);
1862  destroy_blkring:
1863         blkif_free(info, 0);
1864
1865         kfree(info);
1866         dev_set_drvdata(&dev->dev, NULL);
1867
1868         return err;
1869 }
1870
1871 static int negotiate_mq(struct blkfront_info *info)
1872 {
1873         unsigned int backend_max_queues = 0;
1874         int err;
1875         unsigned int i;
1876
1877         BUG_ON(info->nr_rings);
1878
1879         /* Check if backend supports multiple queues. */
1880         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1881                            "multi-queue-max-queues", "%u", &backend_max_queues);
1882         if (err < 0)
1883                 backend_max_queues = 1;
1884
1885         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1886         /* We need at least one ring. */
1887         if (!info->nr_rings)
1888                 info->nr_rings = 1;
1889
1890         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1891         if (!info->rinfo) {
1892                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1893                 return -ENOMEM;
1894         }
1895
1896         for (i = 0; i < info->nr_rings; i++) {
1897                 struct blkfront_ring_info *rinfo;
1898
1899                 rinfo = &info->rinfo[i];
1900                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1901                 INIT_LIST_HEAD(&rinfo->grants);
1902                 rinfo->dev_info = info;
1903                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1904                 spin_lock_init(&rinfo->ring_lock);
1905         }
1906         return 0;
1907 }
1908 /**
1909  * Entry point to this code when a new device is created.  Allocate the basic
1910  * structures and the ring buffer for communication with the backend, and
1911  * inform the backend of the appropriate details for those.  Switch to
1912  * Initialised state.
1913  */
1914 static int blkfront_probe(struct xenbus_device *dev,
1915                           const struct xenbus_device_id *id)
1916 {
1917         int err, vdevice;
1918         struct blkfront_info *info;
1919
1920         /* FIXME: Use dynamic device id if this is not set. */
1921         err = xenbus_scanf(XBT_NIL, dev->nodename,
1922                            "virtual-device", "%i", &vdevice);
1923         if (err != 1) {
1924                 /* go looking in the extended area instead */
1925                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1926                                    "%i", &vdevice);
1927                 if (err != 1) {
1928                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1929                         return err;
1930                 }
1931         }
1932
1933         if (xen_hvm_domain()) {
1934                 char *type;
1935                 int len;
1936                 /* no unplug has been done: do not hook devices != xen vbds */
1937                 if (xen_has_pv_and_legacy_disk_devices()) {
1938                         int major;
1939
1940                         if (!VDEV_IS_EXTENDED(vdevice))
1941                                 major = BLKIF_MAJOR(vdevice);
1942                         else
1943                                 major = XENVBD_MAJOR;
1944
1945                         if (major != XENVBD_MAJOR) {
1946                                 printk(KERN_INFO
1947                                                 "%s: HVM does not support vbd %d as xen block device\n",
1948                                                 __func__, vdevice);
1949                                 return -ENODEV;
1950                         }
1951                 }
1952                 /* do not create a PV cdrom device if we are an HVM guest */
1953                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1954                 if (IS_ERR(type))
1955                         return -ENODEV;
1956                 if (strncmp(type, "cdrom", 5) == 0) {
1957                         kfree(type);
1958                         return -ENODEV;
1959                 }
1960                 kfree(type);
1961         }
1962         info = kzalloc(sizeof(*info), GFP_KERNEL);
1963         if (!info) {
1964                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1965                 return -ENOMEM;
1966         }
1967
1968         info->xbdev = dev;
1969         err = negotiate_mq(info);
1970         if (err) {
1971                 kfree(info);
1972                 return err;
1973         }
1974
1975         mutex_init(&info->mutex);
1976         info->vdevice = vdevice;
1977         info->connected = BLKIF_STATE_DISCONNECTED;
1978
1979         /* Front end dir is a number, which is used as the id. */
1980         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1981         dev_set_drvdata(&dev->dev, info);
1982
1983         return 0;
1984 }
1985
1986 static void split_bio_end(struct bio *bio)
1987 {
1988         struct split_bio *split_bio = bio->bi_private;
1989
1990         if (atomic_dec_and_test(&split_bio->pending)) {
1991                 split_bio->bio->bi_phys_segments = 0;
1992                 split_bio->bio->bi_error = bio->bi_error;
1993                 bio_endio(split_bio->bio);
1994                 kfree(split_bio);
1995         }
1996         bio_put(bio);
1997 }
1998
1999 static int blkif_recover(struct blkfront_info *info)
2000 {
2001         unsigned int i, r_index;
2002         struct request *req, *n;
2003         int rc;
2004         struct bio *bio, *cloned_bio;
2005         unsigned int segs, offset;
2006         int pending, size;
2007         struct split_bio *split_bio;
2008
2009         blkfront_gather_backend_features(info);
2010         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2011         blk_queue_max_segments(info->rq, segs);
2012
2013         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2014                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2015
2016                 rc = blkfront_setup_indirect(rinfo);
2017                 if (rc)
2018                         return rc;
2019         }
2020         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2021
2022         /* Now safe for us to use the shared ring */
2023         info->connected = BLKIF_STATE_CONNECTED;
2024
2025         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2026                 struct blkfront_ring_info *rinfo;
2027
2028                 rinfo = &info->rinfo[r_index];
2029                 /* Kick any other new requests queued since we resumed */
2030                 kick_pending_request_queues(rinfo);
2031         }
2032
2033         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2034                 /* Requeue pending requests (flush or discard) */
2035                 list_del_init(&req->queuelist);
2036                 BUG_ON(req->nr_phys_segments > segs);
2037                 blk_mq_requeue_request(req);
2038         }
2039         blk_mq_kick_requeue_list(info->rq);
2040
2041         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2042                 /* Traverse the list of pending bios and re-queue them */
2043                 if (bio_segments(bio) > segs) {
2044                         /*
2045                          * This bio has more segments than what we can
2046                          * handle, we have to split it.
2047                          */
2048                         pending = (bio_segments(bio) + segs - 1) / segs;
2049                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2050                         BUG_ON(split_bio == NULL);
2051                         atomic_set(&split_bio->pending, pending);
2052                         split_bio->bio = bio;
2053                         for (i = 0; i < pending; i++) {
2054                                 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2055                                 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2056                                            (unsigned int)bio_sectors(bio) - offset);
2057                                 cloned_bio = bio_clone(bio, GFP_NOIO);
2058                                 BUG_ON(cloned_bio == NULL);
2059                                 bio_trim(cloned_bio, offset, size);
2060                                 cloned_bio->bi_private = split_bio;
2061                                 cloned_bio->bi_end_io = split_bio_end;
2062                                 submit_bio(cloned_bio);
2063                         }
2064                         /*
2065                          * Now we have to wait for all those smaller bios to
2066                          * end, so we can also end the "parent" bio.
2067                          */
2068                         continue;
2069                 }
2070                 /* We don't need to split this bio */
2071                 submit_bio(bio);
2072         }
2073
2074         return 0;
2075 }
2076
2077 /**
2078  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2079  * driver restart.  We tear down our blkif structure and recreate it, but
2080  * leave the device-layer structures intact so that this is transparent to the
2081  * rest of the kernel.
2082  */
2083 static int blkfront_resume(struct xenbus_device *dev)
2084 {
2085         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2086         int err = 0;
2087         unsigned int i, j;
2088
2089         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2090
2091         bio_list_init(&info->bio_list);
2092         INIT_LIST_HEAD(&info->requests);
2093         for (i = 0; i < info->nr_rings; i++) {
2094                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2095                 struct bio_list merge_bio;
2096                 struct blk_shadow *shadow = rinfo->shadow;
2097
2098                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2099                         /* Not in use? */
2100                         if (!shadow[j].request)
2101                                 continue;
2102
2103                         /*
2104                          * Get the bios in the request so we can re-queue them.
2105                          */
2106                         if (req_op(shadow[i].request) == REQ_OP_FLUSH ||
2107                             req_op(shadow[i].request) == REQ_OP_DISCARD ||
2108                             req_op(shadow[i].request) == REQ_OP_SECURE_ERASE ||
2109                             shadow[j].request->cmd_flags & REQ_FUA) {
2110                                 /*
2111                                  * Flush operations don't contain bios, so
2112                                  * we need to requeue the whole request
2113                                  *
2114                                  * XXX: but this doesn't make any sense for a
2115                                  * write with the FUA flag set..
2116                                  */
2117                                 list_add(&shadow[j].request->queuelist, &info->requests);
2118                                 continue;
2119                         }
2120                         merge_bio.head = shadow[j].request->bio;
2121                         merge_bio.tail = shadow[j].request->biotail;
2122                         bio_list_merge(&info->bio_list, &merge_bio);
2123                         shadow[j].request->bio = NULL;
2124                         blk_mq_end_request(shadow[j].request, 0);
2125                 }
2126         }
2127
2128         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2129
2130         err = negotiate_mq(info);
2131         if (err)
2132                 return err;
2133
2134         err = talk_to_blkback(dev, info);
2135         if (!err)
2136                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2137
2138         /*
2139          * We have to wait for the backend to switch to
2140          * connected state, since we want to read which
2141          * features it supports.
2142          */
2143
2144         return err;
2145 }
2146
2147 static void blkfront_closing(struct blkfront_info *info)
2148 {
2149         struct xenbus_device *xbdev = info->xbdev;
2150         struct block_device *bdev = NULL;
2151
2152         mutex_lock(&info->mutex);
2153
2154         if (xbdev->state == XenbusStateClosing) {
2155                 mutex_unlock(&info->mutex);
2156                 return;
2157         }
2158
2159         if (info->gd)
2160                 bdev = bdget_disk(info->gd, 0);
2161
2162         mutex_unlock(&info->mutex);
2163
2164         if (!bdev) {
2165                 xenbus_frontend_closed(xbdev);
2166                 return;
2167         }
2168
2169         mutex_lock(&bdev->bd_mutex);
2170
2171         if (bdev->bd_openers) {
2172                 xenbus_dev_error(xbdev, -EBUSY,
2173                                  "Device in use; refusing to close");
2174                 xenbus_switch_state(xbdev, XenbusStateClosing);
2175         } else {
2176                 xlvbd_release_gendisk(info);
2177                 xenbus_frontend_closed(xbdev);
2178         }
2179
2180         mutex_unlock(&bdev->bd_mutex);
2181         bdput(bdev);
2182 }
2183
2184 static void blkfront_setup_discard(struct blkfront_info *info)
2185 {
2186         int err;
2187         unsigned int discard_granularity;
2188         unsigned int discard_alignment;
2189         unsigned int discard_secure;
2190
2191         info->feature_discard = 1;
2192         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2193                 "discard-granularity", "%u", &discard_granularity,
2194                 "discard-alignment", "%u", &discard_alignment,
2195                 NULL);
2196         if (!err) {
2197                 info->discard_granularity = discard_granularity;
2198                 info->discard_alignment = discard_alignment;
2199         }
2200         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2201                     "discard-secure", "%d", &discard_secure,
2202                     NULL);
2203         if (!err)
2204                 info->feature_secdiscard = !!discard_secure;
2205 }
2206
2207 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2208 {
2209         unsigned int psegs, grants;
2210         int err, i;
2211         struct blkfront_info *info = rinfo->dev_info;
2212
2213         if (info->max_indirect_segments == 0) {
2214                 if (!HAS_EXTRA_REQ)
2215                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2216                 else {
2217                         /*
2218                          * When an extra req is required, the maximum
2219                          * grants supported is related to the size of the
2220                          * Linux block segment.
2221                          */
2222                         grants = GRANTS_PER_PSEG;
2223                 }
2224         }
2225         else
2226                 grants = info->max_indirect_segments;
2227         psegs = grants / GRANTS_PER_PSEG;
2228
2229         err = fill_grant_buffer(rinfo,
2230                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2231         if (err)
2232                 goto out_of_memory;
2233
2234         if (!info->feature_persistent && info->max_indirect_segments) {
2235                 /*
2236                  * We are using indirect descriptors but not persistent
2237                  * grants, we need to allocate a set of pages that can be
2238                  * used for mapping indirect grefs
2239                  */
2240                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2241
2242                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2243                 for (i = 0; i < num; i++) {
2244                         struct page *indirect_page = alloc_page(GFP_NOIO);
2245                         if (!indirect_page)
2246                                 goto out_of_memory;
2247                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2248                 }
2249         }
2250
2251         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2252                 rinfo->shadow[i].grants_used = kzalloc(
2253                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2254                         GFP_NOIO);
2255                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2256                 if (info->max_indirect_segments)
2257                         rinfo->shadow[i].indirect_grants = kzalloc(
2258                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2259                                 INDIRECT_GREFS(grants),
2260                                 GFP_NOIO);
2261                 if ((rinfo->shadow[i].grants_used == NULL) ||
2262                         (rinfo->shadow[i].sg == NULL) ||
2263                      (info->max_indirect_segments &&
2264                      (rinfo->shadow[i].indirect_grants == NULL)))
2265                         goto out_of_memory;
2266                 sg_init_table(rinfo->shadow[i].sg, psegs);
2267         }
2268
2269
2270         return 0;
2271
2272 out_of_memory:
2273         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2274                 kfree(rinfo->shadow[i].grants_used);
2275                 rinfo->shadow[i].grants_used = NULL;
2276                 kfree(rinfo->shadow[i].sg);
2277                 rinfo->shadow[i].sg = NULL;
2278                 kfree(rinfo->shadow[i].indirect_grants);
2279                 rinfo->shadow[i].indirect_grants = NULL;
2280         }
2281         if (!list_empty(&rinfo->indirect_pages)) {
2282                 struct page *indirect_page, *n;
2283                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2284                         list_del(&indirect_page->lru);
2285                         __free_page(indirect_page);
2286                 }
2287         }
2288         return -ENOMEM;
2289 }
2290
2291 /*
2292  * Gather all backend feature-*
2293  */
2294 static void blkfront_gather_backend_features(struct blkfront_info *info)
2295 {
2296         int err;
2297         int barrier, flush, discard, persistent;
2298         unsigned int indirect_segments;
2299
2300         info->feature_flush = 0;
2301         info->feature_fua = 0;
2302
2303         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2304                         "feature-barrier", "%d", &barrier,
2305                         NULL);
2306
2307         /*
2308          * If there's no "feature-barrier" defined, then it means
2309          * we're dealing with a very old backend which writes
2310          * synchronously; nothing to do.
2311          *
2312          * If there are barriers, then we use flush.
2313          */
2314         if (!err && barrier) {
2315                 info->feature_flush = 1;
2316                 info->feature_fua = 1;
2317         }
2318
2319         /*
2320          * And if there is "feature-flush-cache" use that above
2321          * barriers.
2322          */
2323         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2324                         "feature-flush-cache", "%d", &flush,
2325                         NULL);
2326
2327         if (!err && flush) {
2328                 info->feature_flush = 1;
2329                 info->feature_fua = 0;
2330         }
2331
2332         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2333                         "feature-discard", "%d", &discard,
2334                         NULL);
2335
2336         if (!err && discard)
2337                 blkfront_setup_discard(info);
2338
2339         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2340                         "feature-persistent", "%u", &persistent,
2341                         NULL);
2342         if (err)
2343                 info->feature_persistent = 0;
2344         else
2345                 info->feature_persistent = persistent;
2346
2347         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2348                             "feature-max-indirect-segments", "%u", &indirect_segments,
2349                             NULL);
2350         if (err)
2351                 info->max_indirect_segments = 0;
2352         else
2353                 info->max_indirect_segments = min(indirect_segments,
2354                                                   xen_blkif_max_segments);
2355 }
2356
2357 /*
2358  * Invoked when the backend is finally 'ready' (and has told produced
2359  * the details about the physical device - #sectors, size, etc).
2360  */
2361 static void blkfront_connect(struct blkfront_info *info)
2362 {
2363         unsigned long long sectors;
2364         unsigned long sector_size;
2365         unsigned int physical_sector_size;
2366         unsigned int binfo;
2367         int err, i;
2368
2369         switch (info->connected) {
2370         case BLKIF_STATE_CONNECTED:
2371                 /*
2372                  * Potentially, the back-end may be signalling
2373                  * a capacity change; update the capacity.
2374                  */
2375                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2376                                    "sectors", "%Lu", &sectors);
2377                 if (XENBUS_EXIST_ERR(err))
2378                         return;
2379                 printk(KERN_INFO "Setting capacity to %Lu\n",
2380                        sectors);
2381                 set_capacity(info->gd, sectors);
2382                 revalidate_disk(info->gd);
2383
2384                 return;
2385         case BLKIF_STATE_SUSPENDED:
2386                 /*
2387                  * If we are recovering from suspension, we need to wait
2388                  * for the backend to announce it's features before
2389                  * reconnecting, at least we need to know if the backend
2390                  * supports indirect descriptors, and how many.
2391                  */
2392                 blkif_recover(info);
2393                 return;
2394
2395         default:
2396                 break;
2397         }
2398
2399         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2400                 __func__, info->xbdev->otherend);
2401
2402         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2403                             "sectors", "%llu", &sectors,
2404                             "info", "%u", &binfo,
2405                             "sector-size", "%lu", &sector_size,
2406                             NULL);
2407         if (err) {
2408                 xenbus_dev_fatal(info->xbdev, err,
2409                                  "reading backend fields at %s",
2410                                  info->xbdev->otherend);
2411                 return;
2412         }
2413
2414         /*
2415          * physcial-sector-size is a newer field, so old backends may not
2416          * provide this. Assume physical sector size to be the same as
2417          * sector_size in that case.
2418          */
2419         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2420                            "physical-sector-size", "%u", &physical_sector_size);
2421         if (err != 1)
2422                 physical_sector_size = sector_size;
2423
2424         blkfront_gather_backend_features(info);
2425         for (i = 0; i < info->nr_rings; i++) {
2426                 err = blkfront_setup_indirect(&info->rinfo[i]);
2427                 if (err) {
2428                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2429                                          info->xbdev->otherend);
2430                         blkif_free(info, 0);
2431                         break;
2432                 }
2433         }
2434
2435         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2436                                   physical_sector_size);
2437         if (err) {
2438                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2439                                  info->xbdev->otherend);
2440                 return;
2441         }
2442
2443         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2444
2445         /* Kick pending requests. */
2446         info->connected = BLKIF_STATE_CONNECTED;
2447         for (i = 0; i < info->nr_rings; i++)
2448                 kick_pending_request_queues(&info->rinfo[i]);
2449
2450         device_add_disk(&info->xbdev->dev, info->gd);
2451
2452         info->is_ready = 1;
2453 }
2454
2455 /**
2456  * Callback received when the backend's state changes.
2457  */
2458 static void blkback_changed(struct xenbus_device *dev,
2459                             enum xenbus_state backend_state)
2460 {
2461         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2462
2463         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2464
2465         switch (backend_state) {
2466         case XenbusStateInitWait:
2467                 if (dev->state != XenbusStateInitialising)
2468                         break;
2469                 if (talk_to_blkback(dev, info))
2470                         break;
2471         case XenbusStateInitialising:
2472         case XenbusStateInitialised:
2473         case XenbusStateReconfiguring:
2474         case XenbusStateReconfigured:
2475         case XenbusStateUnknown:
2476                 break;
2477
2478         case XenbusStateConnected:
2479                 /*
2480                  * talk_to_blkback sets state to XenbusStateInitialised
2481                  * and blkfront_connect sets it to XenbusStateConnected
2482                  * (if connection went OK).
2483                  *
2484                  * If the backend (or toolstack) decides to poke at backend
2485                  * state (and re-trigger the watch by setting the state repeatedly
2486                  * to XenbusStateConnected (4)) we need to deal with this.
2487                  * This is allowed as this is used to communicate to the guest
2488                  * that the size of disk has changed!
2489                  */
2490                 if ((dev->state != XenbusStateInitialised) &&
2491                     (dev->state != XenbusStateConnected)) {
2492                         if (talk_to_blkback(dev, info))
2493                                 break;
2494                 }
2495
2496                 blkfront_connect(info);
2497                 break;
2498
2499         case XenbusStateClosed:
2500                 if (dev->state == XenbusStateClosed)
2501                         break;
2502                 /* Missed the backend's Closing state -- fallthrough */
2503         case XenbusStateClosing:
2504                 if (info)
2505                         blkfront_closing(info);
2506                 break;
2507         }
2508 }
2509
2510 static int blkfront_remove(struct xenbus_device *xbdev)
2511 {
2512         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2513         struct block_device *bdev = NULL;
2514         struct gendisk *disk;
2515
2516         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2517
2518         blkif_free(info, 0);
2519
2520         mutex_lock(&info->mutex);
2521
2522         disk = info->gd;
2523         if (disk)
2524                 bdev = bdget_disk(disk, 0);
2525
2526         info->xbdev = NULL;
2527         mutex_unlock(&info->mutex);
2528
2529         if (!bdev) {
2530                 kfree(info);
2531                 return 0;
2532         }
2533
2534         /*
2535          * The xbdev was removed before we reached the Closed
2536          * state. See if it's safe to remove the disk. If the bdev
2537          * isn't closed yet, we let release take care of it.
2538          */
2539
2540         mutex_lock(&bdev->bd_mutex);
2541         info = disk->private_data;
2542
2543         dev_warn(disk_to_dev(disk),
2544                  "%s was hot-unplugged, %d stale handles\n",
2545                  xbdev->nodename, bdev->bd_openers);
2546
2547         if (info && !bdev->bd_openers) {
2548                 xlvbd_release_gendisk(info);
2549                 disk->private_data = NULL;
2550                 kfree(info);
2551         }
2552
2553         mutex_unlock(&bdev->bd_mutex);
2554         bdput(bdev);
2555
2556         return 0;
2557 }
2558
2559 static int blkfront_is_ready(struct xenbus_device *dev)
2560 {
2561         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2562
2563         return info->is_ready && info->xbdev;
2564 }
2565
2566 static int blkif_open(struct block_device *bdev, fmode_t mode)
2567 {
2568         struct gendisk *disk = bdev->bd_disk;
2569         struct blkfront_info *info;
2570         int err = 0;
2571
2572         mutex_lock(&blkfront_mutex);
2573
2574         info = disk->private_data;
2575         if (!info) {
2576                 /* xbdev gone */
2577                 err = -ERESTARTSYS;
2578                 goto out;
2579         }
2580
2581         mutex_lock(&info->mutex);
2582
2583         if (!info->gd)
2584                 /* xbdev is closed */
2585                 err = -ERESTARTSYS;
2586
2587         mutex_unlock(&info->mutex);
2588
2589 out:
2590         mutex_unlock(&blkfront_mutex);
2591         return err;
2592 }
2593
2594 static void blkif_release(struct gendisk *disk, fmode_t mode)
2595 {
2596         struct blkfront_info *info = disk->private_data;
2597         struct block_device *bdev;
2598         struct xenbus_device *xbdev;
2599
2600         mutex_lock(&blkfront_mutex);
2601
2602         bdev = bdget_disk(disk, 0);
2603
2604         if (!bdev) {
2605                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2606                 goto out_mutex;
2607         }
2608         if (bdev->bd_openers)
2609                 goto out;
2610
2611         /*
2612          * Check if we have been instructed to close. We will have
2613          * deferred this request, because the bdev was still open.
2614          */
2615
2616         mutex_lock(&info->mutex);
2617         xbdev = info->xbdev;
2618
2619         if (xbdev && xbdev->state == XenbusStateClosing) {
2620                 /* pending switch to state closed */
2621                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2622                 xlvbd_release_gendisk(info);
2623                 xenbus_frontend_closed(info->xbdev);
2624         }
2625
2626         mutex_unlock(&info->mutex);
2627
2628         if (!xbdev) {
2629                 /* sudden device removal */
2630                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2631                 xlvbd_release_gendisk(info);
2632                 disk->private_data = NULL;
2633                 kfree(info);
2634         }
2635
2636 out:
2637         bdput(bdev);
2638 out_mutex:
2639         mutex_unlock(&blkfront_mutex);
2640 }
2641
2642 static const struct block_device_operations xlvbd_block_fops =
2643 {
2644         .owner = THIS_MODULE,
2645         .open = blkif_open,
2646         .release = blkif_release,
2647         .getgeo = blkif_getgeo,
2648         .ioctl = blkif_ioctl,
2649 };
2650
2651
2652 static const struct xenbus_device_id blkfront_ids[] = {
2653         { "vbd" },
2654         { "" }
2655 };
2656
2657 static struct xenbus_driver blkfront_driver = {
2658         .ids  = blkfront_ids,
2659         .probe = blkfront_probe,
2660         .remove = blkfront_remove,
2661         .resume = blkfront_resume,
2662         .otherend_changed = blkback_changed,
2663         .is_ready = blkfront_is_ready,
2664 };
2665
2666 static int __init xlblk_init(void)
2667 {
2668         int ret;
2669         int nr_cpus = num_online_cpus();
2670
2671         if (!xen_domain())
2672                 return -ENODEV;
2673
2674         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2675                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2676                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2677                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2678         }
2679
2680         if (xen_blkif_max_queues > nr_cpus) {
2681                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2682                         xen_blkif_max_queues, nr_cpus);
2683                 xen_blkif_max_queues = nr_cpus;
2684         }
2685
2686         if (!xen_has_pv_disk_devices())
2687                 return -ENODEV;
2688
2689         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2690                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2691                        XENVBD_MAJOR, DEV_NAME);
2692                 return -ENODEV;
2693         }
2694
2695         ret = xenbus_register_frontend(&blkfront_driver);
2696         if (ret) {
2697                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2698                 return ret;
2699         }
2700
2701         return 0;
2702 }
2703 module_init(xlblk_init);
2704
2705
2706 static void __exit xlblk_exit(void)
2707 {
2708         xenbus_unregister_driver(&blkfront_driver);
2709         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2710         kfree(minors);
2711 }
2712 module_exit(xlblk_exit);
2713
2714 MODULE_DESCRIPTION("Xen virtual block device frontend");
2715 MODULE_LICENSE("GPL");
2716 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2717 MODULE_ALIAS("xen:vbd");
2718 MODULE_ALIAS("xenblk");