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