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/hdreg.h>
  41 #include <linux/cdrom.h>
  42 #include <linux/module.h>
  43 #include <linux/slab.h>
  44 #include <linux/mutex.h>
  45 #include <linux/scatterlist.h>
  46 #include <linux/bitmap.h>
  47 #include <linux/list.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/xenbus.h>
  51 #include <xen/grant_table.h>
  52 #include <xen/events.h>
  53 #include <xen/page.h>
  54 #include <xen/platform_pci.h>
  55
  56 #include <xen/interface/grant_table.h>
  57 #include <xen/interface/io/blkif.h>
  58 #include <xen/interface/io/protocols.h>
  59
  60 #include <asm/xen/hypervisor.h>
  61
  62 enum blkif_state {
  63         BLKIF_STATE_DISCONNECTED,
  64         BLKIF_STATE_CONNECTED,
  65         BLKIF_STATE_SUSPENDED,
  66 };
  67
  68 struct grant {
  69         grant_ref_t gref;
  70         unsigned long pfn;
  71         struct list_head node;
  72 };
  73
  74 struct blk_shadow {
  75         struct blkif_request req;
  76         struct request *request;
  77         struct grant **grants_used;
  78         struct grant **indirect_grants;
  79         struct scatterlist *sg;
  80 };
  81
  82 struct split_bio {
  83         struct bio *bio;
  84         atomic_t pending;
  85         int err;
  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 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
 102
 103 /*
 104  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 105  * hang in private_data off the gendisk structure. We may end up
 106  * putting all kinds of interesting stuff here :-)
 107  */
 108 struct blkfront_info
 109 {
 110         spinlock_t io_lock;
 111         struct mutex mutex;
 112         struct xenbus_device *xbdev;
 113         struct gendisk *gd;
 114         int vdevice;
 115         blkif_vdev_t handle;
 116         enum blkif_state connected;
 117         int ring_ref;
 118         struct blkif_front_ring ring;
 119         unsigned int evtchn, irq;
 120         struct request_queue *rq;
 121         struct work_struct work;
 122         struct gnttab_free_callback callback;
 123         struct blk_shadow shadow[BLK_RING_SIZE];
 124         struct list_head grants;
 125         struct list_head indirect_pages;
 126         unsigned int persistent_gnts_c;
 127         unsigned long shadow_free;
 128         unsigned int feature_flush;
 129         unsigned int feature_discard:1;
 130         unsigned int feature_secdiscard:1;
 131         unsigned int discard_granularity;
 132         unsigned int discard_alignment;
 133         unsigned int feature_persistent:1;
 134         unsigned int max_indirect_segments;
 135         int is_ready;
 136 };
 137
 138 static unsigned int nr_minors;
 139 static unsigned long *minors;
 140 static DEFINE_SPINLOCK(minor_lock);
 141
 142 #define GRANT_INVALID_REF       0
 143
 144 #define PARTS_PER_DISK          16
 145 #define PARTS_PER_EXT_DISK      256
 146
 147 #define BLKIF_MAJOR(dev) ((dev)>>8)
 148 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 149
 150 #define EXT_SHIFT 28
 151 #define EXTENDED (1<<EXT_SHIFT)
 152 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 153 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 154 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 155 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 156 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 157 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 158
 159 #define DEV_NAME        "xvd"   /* name in /dev */
 160
 161 #define SEGS_PER_INDIRECT_FRAME \
 162         (PAGE_SIZE/sizeof(struct blkif_request_segment))
 163 #define INDIRECT_GREFS(_segs) \
 164         ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
 165
 166 static int blkfront_setup_indirect(struct blkfront_info *info);
 167
 168 static int get_id_from_freelist(struct blkfront_info *info)
 169 {
 170         unsigned long free = info->shadow_free;
 171         BUG_ON(free >= BLK_RING_SIZE);
 172         info->shadow_free = info->shadow[free].req.u.rw.id;
 173         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 174         return free;
 175 }
 176
 177 static int add_id_to_freelist(struct blkfront_info *info,
 178                                unsigned long id)
 179 {
 180         if (info->shadow[id].req.u.rw.id != id)
 181                 return -EINVAL;
 182         if (info->shadow[id].request == NULL)
 183                 return -EINVAL;
 184         info->shadow[id].req.u.rw.id  = info->shadow_free;
 185         info->shadow[id].request = NULL;
 186         info->shadow_free = id;
 187         return 0;
 188 }
 189
 190 static int fill_grant_buffer(struct blkfront_info *info, int num)
 191 {
 192         struct page *granted_page;
 193         struct grant *gnt_list_entry, *n;
 194         int i = 0;
 195
 196         while(i < num) {
 197                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 198                 if (!gnt_list_entry)
 199                         goto out_of_memory;
 200
 201                 if (info->feature_persistent) {
 202                         granted_page = alloc_page(GFP_NOIO);
 203                         if (!granted_page) {
 204                                 kfree(gnt_list_entry);
 205                                 goto out_of_memory;
 206                         }
 207                         gnt_list_entry->pfn = page_to_pfn(granted_page);
 208                 }
 209
 210                 gnt_list_entry->gref = GRANT_INVALID_REF;
 211                 list_add(&gnt_list_entry->node, &info->grants);
 212                 i++;
 213         }
 214
 215         return 0;
 216
 217 out_of_memory:
 218         list_for_each_entry_safe(gnt_list_entry, n,
 219                                  &info->grants, node) {
 220                 list_del(&gnt_list_entry->node);
 221                 if (info->feature_persistent)
 222                         __free_page(pfn_to_page(gnt_list_entry->pfn));
 223                 kfree(gnt_list_entry);
 224                 i--;
 225         }
 226         BUG_ON(i != 0);
 227         return -ENOMEM;
 228 }
 229
 230 static struct grant *get_grant(grant_ref_t *gref_head,
 231                                unsigned long pfn,
 232                                struct blkfront_info *info)
 233 {
 234         struct grant *gnt_list_entry;
 235         unsigned long buffer_mfn;
 236
 237         BUG_ON(list_empty(&info->grants));
 238         gnt_list_entry = list_first_entry(&info->grants, struct grant,
 239                                           node);
 240         list_del(&gnt_list_entry->node);
 241
 242         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 243                 info->persistent_gnts_c--;
 244                 return gnt_list_entry;
 245         }
 246
 247         /* Assign a gref to this page */
 248         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 249         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 250         if (!info->feature_persistent) {
 251                 BUG_ON(!pfn);
 252                 gnt_list_entry->pfn = pfn;
 253         }
 254         buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 255         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 256                                         info->xbdev->otherend_id,
 257                                         buffer_mfn, 0);
 258         return gnt_list_entry;
 259 }
 260
 261 static const char *op_name(int op)
 262 {
 263         static const char *const names[] = {
 264                 [BLKIF_OP_READ] = "read",
 265                 [BLKIF_OP_WRITE] = "write",
 266                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 267                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 268                 [BLKIF_OP_DISCARD] = "discard" };
 269
 270         if (op < 0 || op >= ARRAY_SIZE(names))
 271                 return "unknown";
 272
 273         if (!names[op])
 274                 return "reserved";
 275
 276         return names[op];
 277 }
 278 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 279 {
 280         unsigned int end = minor + nr;
 281         int rc;
 282
 283         if (end > nr_minors) {
 284                 unsigned long *bitmap, *old;
 285
 286                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 287                                  GFP_KERNEL);
 288                 if (bitmap == NULL)
 289                         return -ENOMEM;
 290
 291                 spin_lock(&minor_lock);
 292                 if (end > nr_minors) {
 293                         old = minors;
 294                         memcpy(bitmap, minors,
 295                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 296                         minors = bitmap;
 297                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 298                 } else
 299                         old = bitmap;
 300                 spin_unlock(&minor_lock);
 301                 kfree(old);
 302         }
 303
 304         spin_lock(&minor_lock);
 305         if (find_next_bit(minors, end, minor) >= end) {
 306                 bitmap_set(minors, minor, nr);
 307                 rc = 0;
 308         } else
 309                 rc = -EBUSY;
 310         spin_unlock(&minor_lock);
 311
 312         return rc;
 313 }
 314
 315 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 316 {
 317         unsigned int end = minor + nr;
 318
 319         BUG_ON(end > nr_minors);
 320         spin_lock(&minor_lock);
 321         bitmap_clear(minors,  minor, nr);
 322         spin_unlock(&minor_lock);
 323 }
 324
 325 static void blkif_restart_queue_callback(void *arg)
 326 {
 327         struct blkfront_info *info = (struct blkfront_info *)arg;
 328         schedule_work(&info->work);
 329 }
 330
 331 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 332 {
 333         /* We don't have real geometry info, but let's at least return
 334            values consistent with the size of the device */
 335         sector_t nsect = get_capacity(bd->bd_disk);
 336         sector_t cylinders = nsect;
 337
 338         hg->heads = 0xff;
 339         hg->sectors = 0x3f;
 340         sector_div(cylinders, hg->heads * hg->sectors);
 341         hg->cylinders = cylinders;
 342         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 343                 hg->cylinders = 0xffff;
 344         return 0;
 345 }
 346
 347 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 348                        unsigned command, unsigned long argument)
 349 {
 350         struct blkfront_info *info = bdev->bd_disk->private_data;
 351         int i;
 352
 353         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 354                 command, (long)argument);
 355
 356         switch (command) {
 357         case CDROMMULTISESSION:
 358                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 359                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 360                         if (put_user(0, (char __user *)(argument + i)))
 361                                 return -EFAULT;
 362                 return 0;
 363
 364         case CDROM_GET_CAPABILITY: {
 365                 struct gendisk *gd = info->gd;
 366                 if (gd->flags & GENHD_FL_CD)
 367                         return 0;
 368                 return -EINVAL;
 369         }
 370
 371         default:
 372                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 373                   command);*/
 374                 return -EINVAL; /* same return as native Linux */
 375         }
 376
 377         return 0;
 378 }
 379
 380 /*
 381  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 382  * and writes are handled as expected.
 383  *
 384  * @req: a request struct
 385  */
 386 static int blkif_queue_request(struct request *req)
 387 {
 388         struct blkfront_info *info = req->rq_disk->private_data;
 389         struct blkif_request *ring_req;
 390         unsigned long id;
 391         unsigned int fsect, lsect;
 392         int i, ref, n;
 393         struct blkif_request_segment *segments = NULL;
 394
 395         /*
 396          * Used to store if we are able to queue the request by just using
 397          * existing persistent grants, or if we have to get new grants,
 398          * as there are not sufficiently many free.
 399          */
 400         bool new_persistent_gnts;
 401         grant_ref_t gref_head;
 402         struct grant *gnt_list_entry = NULL;
 403         struct scatterlist *sg;
 404         int nseg, max_grefs;
 405
 406         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 407                 return 1;
 408
 409         max_grefs = req->nr_phys_segments;
 410         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 411                 /*
 412                  * If we are using indirect segments we need to account
 413                  * for the indirect grefs used in the request.
 414                  */
 415                 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
 416
 417         /* Check if we have enough grants to allocate a requests */
 418         if (info->persistent_gnts_c < max_grefs) {
 419                 new_persistent_gnts = 1;
 420                 if (gnttab_alloc_grant_references(
 421                     max_grefs - info->persistent_gnts_c,
 422                     &gref_head) < 0) {
 423                         gnttab_request_free_callback(
 424                                 &info->callback,
 425                                 blkif_restart_queue_callback,
 426                                 info,
 427                                 max_grefs);
 428                         return 1;
 429                 }
 430         } else
 431                 new_persistent_gnts = 0;
 432
 433         /* Fill out a communications ring structure. */
 434         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 435         id = get_id_from_freelist(info);
 436         info->shadow[id].request = req;
 437
 438         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 439                 ring_req->operation = BLKIF_OP_DISCARD;
 440                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 441                 ring_req->u.discard.id = id;
 442                 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 443                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 444                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 445                 else
 446                         ring_req->u.discard.flag = 0;
 447         } else {
 448                 BUG_ON(info->max_indirect_segments == 0 &&
 449                        req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 450                 BUG_ON(info->max_indirect_segments &&
 451                        req->nr_phys_segments > info->max_indirect_segments);
 452                 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
 453                 ring_req->u.rw.id = id;
 454                 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 455                         /*
 456                          * The indirect operation can only be a BLKIF_OP_READ or
 457                          * BLKIF_OP_WRITE
 458                          */
 459                         BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
 460                         ring_req->operation = BLKIF_OP_INDIRECT;
 461                         ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 462                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 463                         ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 464                         ring_req->u.indirect.handle = info->handle;
 465                         ring_req->u.indirect.nr_segments = nseg;
 466                 } else {
 467                         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 468                         ring_req->u.rw.handle = info->handle;
 469                         ring_req->operation = rq_data_dir(req) ?
 470                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 471                         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 472                                 /*
 473                                  * Ideally we can do an unordered flush-to-disk. In case the
 474                                  * backend onlysupports barriers, use that. A barrier request
 475                                  * a superset of FUA, so we can implement it the same
 476                                  * way.  (It's also a FLUSH+FUA, since it is
 477                                  * guaranteed ordered WRT previous writes.)
 478                                  */
 479                                 switch (info->feature_flush &
 480                                         ((REQ_FLUSH|REQ_FUA))) {
 481                                 case REQ_FLUSH|REQ_FUA:
 482                                         ring_req->operation =
 483                                                 BLKIF_OP_WRITE_BARRIER;
 484                                         break;
 485                                 case REQ_FLUSH:
 486                                         ring_req->operation =
 487                                                 BLKIF_OP_FLUSH_DISKCACHE;
 488                                         break;
 489                                 default:
 490                                         ring_req->operation = 0;
 491                                 }
 492                         }
 493                         ring_req->u.rw.nr_segments = nseg;
 494                 }
 495                 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
 496                         fsect = sg->offset >> 9;
 497                         lsect = fsect + (sg->length >> 9) - 1;
 498
 499                         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 500                             (i % SEGS_PER_INDIRECT_FRAME == 0)) {
 501                                 unsigned long uninitialized_var(pfn);
 502
 503                                 if (segments)
 504                                         kunmap_atomic(segments);
 505
 506                                 n = i / SEGS_PER_INDIRECT_FRAME;
 507                                 if (!info->feature_persistent) {
 508                                         struct page *indirect_page;
 509
 510                                         /* Fetch a pre-allocated page to use for indirect grefs */
 511                                         BUG_ON(list_empty(&info->indirect_pages));
 512                                         indirect_page = list_first_entry(&info->indirect_pages,
 513                                                                          struct page, lru);
 514                                         list_del(&indirect_page->lru);
 515                                         pfn = page_to_pfn(indirect_page);
 516                                 }
 517                                 gnt_list_entry = get_grant(&gref_head, pfn, info);
 518                                 info->shadow[id].indirect_grants[n] = gnt_list_entry;
 519                                 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 520                                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 521                         }
 522
 523                         gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
 524                         ref = gnt_list_entry->gref;
 525
 526                         info->shadow[id].grants_used[i] = gnt_list_entry;
 527
 528                         if (rq_data_dir(req) && info->feature_persistent) {
 529                                 char *bvec_data;
 530                                 void *shared_data;
 531
 532                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 533
 534                                 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 535                                 bvec_data = kmap_atomic(sg_page(sg));
 536
 537                                 /*
 538                                  * this does not wipe data stored outside the
 539                                  * range sg->offset..sg->offset+sg->length.
 540                                  * Therefore, blkback *could* see data from
 541                                  * previous requests. This is OK as long as
 542                                  * persistent grants are shared with just one
 543                                  * domain. It may need refactoring if this
 544                                  * changes
 545                                  */
 546                                 memcpy(shared_data + sg->offset,
 547                                        bvec_data   + sg->offset,
 548                                        sg->length);
 549
 550                                 kunmap_atomic(bvec_data);
 551                                 kunmap_atomic(shared_data);
 552                         }
 553                         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 554                                 ring_req->u.rw.seg[i] =
 555                                                 (struct blkif_request_segment) {
 556                                                         .gref       = ref,
 557                                                         .first_sect = fsect,
 558                                                         .last_sect  = lsect };
 559                         } else {
 560                                 n = i % SEGS_PER_INDIRECT_FRAME;
 561                                 segments[n] =
 562                                         (struct blkif_request_segment) {
 563                                                         .gref       = ref,
 564                                                         .first_sect = fsect,
 565                                                         .last_sect  = lsect };
 566                         }
 567                 }
 568                 if (segments)
 569                         kunmap_atomic(segments);
 570         }
 571
 572         info->ring.req_prod_pvt++;
 573
 574         /* Keep a private copy so we can reissue requests when recovering. */
 575         info->shadow[id].req = *ring_req;
 576
 577         if (new_persistent_gnts)
 578                 gnttab_free_grant_references(gref_head);
 579
 580         return 0;
 581 }
 582
 583
 584 static inline void flush_requests(struct blkfront_info *info)
 585 {
 586         int notify;
 587
 588         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 589
 590         if (notify)
 591                 notify_remote_via_irq(info->irq);
 592 }
 593
 594 static inline bool blkif_request_flush_invalid(struct request *req,
 595                                                struct blkfront_info *info)
 596 {
 597         return ((req->cmd_type != REQ_TYPE_FS) ||
 598                 ((req->cmd_flags & REQ_FLUSH) &&
 599                  !(info->feature_flush & REQ_FLUSH)) ||
 600                 ((req->cmd_flags & REQ_FUA) &&
 601                  !(info->feature_flush & REQ_FUA)));
 602 }
 603
 604 /*
 605  * do_blkif_request
 606  *  read a block; request is in a request queue
 607  */
 608 static void do_blkif_request(struct request_queue *rq)
 609 {
 610         struct blkfront_info *info = NULL;
 611         struct request *req;
 612         int queued;
 613
 614         pr_debug("Entered do_blkif_request\n");
 615
 616         queued = 0;
 617
 618         while ((req = blk_peek_request(rq)) != NULL) {
 619                 info = req->rq_disk->private_data;
 620
 621                 if (RING_FULL(&info->ring))
 622                         goto wait;
 623
 624                 blk_start_request(req);
 625
 626                 if (blkif_request_flush_invalid(req, info)) {
 627                         __blk_end_request_all(req, -EOPNOTSUPP);
 628                         continue;
 629                 }
 630
 631                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 632                          "(%u/%u) [%s]\n",
 633                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 634                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 635                          rq_data_dir(req) ? "write" : "read");
 636
 637                 if (blkif_queue_request(req)) {
 638                         blk_requeue_request(rq, req);
 639 wait:
 640                         /* Avoid pointless unplugs. */
 641                         blk_stop_queue(rq);
 642                         break;
 643                 }
 644
 645                 queued++;
 646         }
 647
 648         if (queued != 0)
 649                 flush_requests(info);
 650 }
 651
 652 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 653                                 unsigned int physical_sector_size,
 654                                 unsigned int segments)
 655 {
 656         struct request_queue *rq;
 657         struct blkfront_info *info = gd->private_data;
 658
 659         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 660         if (rq == NULL)
 661                 return -1;
 662
 663         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 664
 665         if (info->feature_discard) {
 666                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 667                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 668                 rq->limits.discard_granularity = info->discard_granularity;
 669                 rq->limits.discard_alignment = info->discard_alignment;
 670                 if (info->feature_secdiscard)
 671                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 672         }
 673
 674         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 675         blk_queue_logical_block_size(rq, sector_size);
 676         blk_queue_physical_block_size(rq, physical_sector_size);
 677         blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
 678
 679         /* Each segment in a request is up to an aligned page in size. */
 680         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 681         blk_queue_max_segment_size(rq, PAGE_SIZE);
 682
 683         /* Ensure a merged request will fit in a single I/O ring slot. */
 684         blk_queue_max_segments(rq, segments);
 685
 686         /* Make sure buffer addresses are sector-aligned. */
 687         blk_queue_dma_alignment(rq, 511);
 688
 689         /* Make sure we don't use bounce buffers. */
 690         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 691
 692         gd->queue = rq;
 693
 694         return 0;
 695 }
 696
 697 static const char *flush_info(unsigned int feature_flush)
 698 {
 699         switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
 700         case REQ_FLUSH|REQ_FUA:
 701                 return "barrier: enabled;";
 702         case REQ_FLUSH:
 703                 return "flush diskcache: enabled;";
 704         default:
 705                 return "barrier or flush: disabled;";
 706         }
 707 }
 708
 709 static void xlvbd_flush(struct blkfront_info *info)
 710 {
 711         blk_queue_flush(info->rq, info->feature_flush);
 712         pr_info("blkfront: %s: %s %s %s %s %s\n",
 713                 info->gd->disk_name, flush_info(info->feature_flush),
 714                 "persistent grants:", info->feature_persistent ?
 715                 "enabled;" : "disabled;", "indirect descriptors:",
 716                 info->max_indirect_segments ? "enabled;" : "disabled;");
 717 }
 718
 719 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 720 {
 721         int major;
 722         major = BLKIF_MAJOR(vdevice);
 723         *minor = BLKIF_MINOR(vdevice);
 724         switch (major) {
 725                 case XEN_IDE0_MAJOR:
 726                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 727                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 728                                 EMULATED_HD_DISK_MINOR_OFFSET;
 729                         break;
 730                 case XEN_IDE1_MAJOR:
 731                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 732                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 733                                 EMULATED_HD_DISK_MINOR_OFFSET;
 734                         break;
 735                 case XEN_SCSI_DISK0_MAJOR:
 736                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 737                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 738                         break;
 739                 case XEN_SCSI_DISK1_MAJOR:
 740                 case XEN_SCSI_DISK2_MAJOR:
 741                 case XEN_SCSI_DISK3_MAJOR:
 742                 case XEN_SCSI_DISK4_MAJOR:
 743                 case XEN_SCSI_DISK5_MAJOR:
 744                 case XEN_SCSI_DISK6_MAJOR:
 745                 case XEN_SCSI_DISK7_MAJOR:
 746                         *offset = (*minor / PARTS_PER_DISK) +
 747                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 748                                 EMULATED_SD_DISK_NAME_OFFSET;
 749                         *minor = *minor +
 750                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 751                                 EMULATED_SD_DISK_MINOR_OFFSET;
 752                         break;
 753                 case XEN_SCSI_DISK8_MAJOR:
 754                 case XEN_SCSI_DISK9_MAJOR:
 755                 case XEN_SCSI_DISK10_MAJOR:
 756                 case XEN_SCSI_DISK11_MAJOR:
 757                 case XEN_SCSI_DISK12_MAJOR:
 758                 case XEN_SCSI_DISK13_MAJOR:
 759                 case XEN_SCSI_DISK14_MAJOR:
 760                 case XEN_SCSI_DISK15_MAJOR:
 761                         *offset = (*minor / PARTS_PER_DISK) +
 762                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 763                                 EMULATED_SD_DISK_NAME_OFFSET;
 764                         *minor = *minor +
 765                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 766                                 EMULATED_SD_DISK_MINOR_OFFSET;
 767                         break;
 768                 case XENVBD_MAJOR:
 769                         *offset = *minor / PARTS_PER_DISK;
 770                         break;
 771                 default:
 772                         printk(KERN_WARNING "blkfront: your disk configuration is "
 773                                         "incorrect, please use an xvd device instead\n");
 774                         return -ENODEV;
 775         }
 776         return 0;
 777 }
 778
 779 static char *encode_disk_name(char *ptr, unsigned int n)
 780 {
 781         if (n >= 26)
 782                 ptr = encode_disk_name(ptr, n / 26 - 1);
 783         *ptr = 'a' + n % 26;
 784         return ptr + 1;
 785 }
 786
 787 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 788                                struct blkfront_info *info,
 789                                u16 vdisk_info, u16 sector_size,
 790                                unsigned int physical_sector_size)
 791 {
 792         struct gendisk *gd;
 793         int nr_minors = 1;
 794         int err;
 795         unsigned int offset;
 796         int minor;
 797         int nr_parts;
 798         char *ptr;
 799
 800         BUG_ON(info->gd != NULL);
 801         BUG_ON(info->rq != NULL);
 802
 803         if ((info->vdevice>>EXT_SHIFT) > 1) {
 804                 /* this is above the extended range; something is wrong */
 805                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 806                 return -ENODEV;
 807         }
 808
 809         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 810                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 811                 if (err)
 812                         return err;
 813                 nr_parts = PARTS_PER_DISK;
 814         } else {
 815                 minor = BLKIF_MINOR_EXT(info->vdevice);
 816                 nr_parts = PARTS_PER_EXT_DISK;
 817                 offset = minor / nr_parts;
 818                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 819                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 820                                         "emulated IDE disks,\n\t choose an xvd device name"
 821                                         "from xvde on\n", info->vdevice);
 822         }
 823         if (minor >> MINORBITS) {
 824                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 825                         info->vdevice, minor);
 826                 return -ENODEV;
 827         }
 828
 829         if ((minor % nr_parts) == 0)
 830                 nr_minors = nr_parts;
 831
 832         err = xlbd_reserve_minors(minor, nr_minors);
 833         if (err)
 834                 goto out;
 835         err = -ENODEV;
 836
 837         gd = alloc_disk(nr_minors);
 838         if (gd == NULL)
 839                 goto release;
 840
 841         strcpy(gd->disk_name, DEV_NAME);
 842         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 843         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 844         if (nr_minors > 1)
 845                 *ptr = 0;
 846         else
 847                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 848                          "%d", minor & (nr_parts - 1));
 849
 850         gd->major = XENVBD_MAJOR;
 851         gd->first_minor = minor;
 852         gd->fops = &xlvbd_block_fops;
 853         gd->private_data = info;
 854         gd->driverfs_dev = &(info->xbdev->dev);
 855         set_capacity(gd, capacity);
 856
 857         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
 858                                  info->max_indirect_segments ? :
 859                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 860                 del_gendisk(gd);
 861                 goto release;
 862         }
 863
 864         info->rq = gd->queue;
 865         info->gd = gd;
 866
 867         xlvbd_flush(info);
 868
 869         if (vdisk_info & VDISK_READONLY)
 870                 set_disk_ro(gd, 1);
 871
 872         if (vdisk_info & VDISK_REMOVABLE)
 873                 gd->flags |= GENHD_FL_REMOVABLE;
 874
 875         if (vdisk_info & VDISK_CDROM)
 876                 gd->flags |= GENHD_FL_CD;
 877
 878         return 0;
 879
 880  release:
 881         xlbd_release_minors(minor, nr_minors);
 882  out:
 883         return err;
 884 }
 885
 886 static void xlvbd_release_gendisk(struct blkfront_info *info)
 887 {
 888         unsigned int minor, nr_minors;
 889         unsigned long flags;
 890
 891         if (info->rq == NULL)
 892                 return;
 893
 894         spin_lock_irqsave(&info->io_lock, flags);
 895
 896         /* No more blkif_request(). */
 897         blk_stop_queue(info->rq);
 898
 899         /* No more gnttab callback work. */
 900         gnttab_cancel_free_callback(&info->callback);
 901         spin_unlock_irqrestore(&info->io_lock, flags);
 902
 903         /* Flush gnttab callback work. Must be done with no locks held. */
 904         flush_work(&info->work);
 905
 906         del_gendisk(info->gd);
 907
 908         minor = info->gd->first_minor;
 909         nr_minors = info->gd->minors;
 910         xlbd_release_minors(minor, nr_minors);
 911
 912         blk_cleanup_queue(info->rq);
 913         info->rq = NULL;
 914
 915         put_disk(info->gd);
 916         info->gd = NULL;
 917 }
 918
 919 static void kick_pending_request_queues(struct blkfront_info *info)
 920 {
 921         if (!RING_FULL(&info->ring)) {
 922                 /* Re-enable calldowns. */
 923                 blk_start_queue(info->rq);
 924                 /* Kick things off immediately. */
 925                 do_blkif_request(info->rq);
 926         }
 927 }
 928
 929 static void blkif_restart_queue(struct work_struct *work)
 930 {
 931         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 932
 933         spin_lock_irq(&info->io_lock);
 934         if (info->connected == BLKIF_STATE_CONNECTED)
 935                 kick_pending_request_queues(info);
 936         spin_unlock_irq(&info->io_lock);
 937 }
 938
 939 static void blkif_free(struct blkfront_info *info, int suspend)
 940 {
 941         struct grant *persistent_gnt;
 942         struct grant *n;
 943         int i, j, segs;
 944
 945         /* Prevent new requests being issued until we fix things up. */
 946         spin_lock_irq(&info->io_lock);
 947         info->connected = suspend ?
 948                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 949         /* No more blkif_request(). */
 950         if (info->rq)
 951                 blk_stop_queue(info->rq);
 952
 953         /* Remove all persistent grants */
 954         if (!list_empty(&info->grants)) {
 955                 list_for_each_entry_safe(persistent_gnt, n,
 956                                          &info->grants, node) {
 957                         list_del(&persistent_gnt->node);
 958                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 959                                 gnttab_end_foreign_access(persistent_gnt->gref,
 960                                                           0, 0UL);
 961                                 info->persistent_gnts_c--;
 962                         }
 963                         if (info->feature_persistent)
 964                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 965                         kfree(persistent_gnt);
 966                 }
 967         }
 968         BUG_ON(info->persistent_gnts_c != 0);
 969
 970         /*
 971          * Remove indirect pages, this only happens when using indirect
 972          * descriptors but not persistent grants
 973          */
 974         if (!list_empty(&info->indirect_pages)) {
 975                 struct page *indirect_page, *n;
 976
 977                 BUG_ON(info->feature_persistent);
 978                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
 979                         list_del(&indirect_page->lru);
 980                         __free_page(indirect_page);
 981                 }
 982         }
 983
 984         for (i = 0; i < BLK_RING_SIZE; i++) {
 985                 /*
 986                  * Clear persistent grants present in requests already
 987                  * on the shared ring
 988                  */
 989                 if (!info->shadow[i].request)
 990                         goto free_shadow;
 991
 992                 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
 993                        info->shadow[i].req.u.indirect.nr_segments :
 994                        info->shadow[i].req.u.rw.nr_segments;
 995                 for (j = 0; j < segs; j++) {
 996                         persistent_gnt = info->shadow[i].grants_used[j];
 997                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 998                         if (info->feature_persistent)
 999                                 __free_page(pfn_to_page(persistent_gnt->pfn));
1000                         kfree(persistent_gnt);
1001                 }
1002
1003                 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1004                         /*
1005                          * If this is not an indirect operation don't try to
1006                          * free indirect segments
1007                          */
1008                         goto free_shadow;
1009
1010                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1011                         persistent_gnt = info->shadow[i].indirect_grants[j];
1012                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1013                         __free_page(pfn_to_page(persistent_gnt->pfn));
1014                         kfree(persistent_gnt);
1015                 }
1016
1017 free_shadow:
1018                 kfree(info->shadow[i].grants_used);
1019                 info->shadow[i].grants_used = NULL;
1020                 kfree(info->shadow[i].indirect_grants);
1021                 info->shadow[i].indirect_grants = NULL;
1022                 kfree(info->shadow[i].sg);
1023                 info->shadow[i].sg = NULL;
1024         }
1025
1026         /* No more gnttab callback work. */
1027         gnttab_cancel_free_callback(&info->callback);
1028         spin_unlock_irq(&info->io_lock);
1029
1030         /* Flush gnttab callback work. Must be done with no locks held. */
1031         flush_work(&info->work);
1032
1033         /* Free resources associated with old device channel. */
1034         if (info->ring_ref != GRANT_INVALID_REF) {
1035                 gnttab_end_foreign_access(info->ring_ref, 0,
1036                                           (unsigned long)info->ring.sring);
1037                 info->ring_ref = GRANT_INVALID_REF;
1038                 info->ring.sring = NULL;
1039         }
1040         if (info->irq)
1041                 unbind_from_irqhandler(info->irq, info);
1042         info->evtchn = info->irq = 0;
1043
1044 }
1045
1046 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1047                              struct blkif_response *bret)
1048 {
1049         int i = 0;
1050         struct scatterlist *sg;
1051         char *bvec_data;
1052         void *shared_data;
1053         int nseg;
1054
1055         nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1056                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1057
1058         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1059                 /*
1060                  * Copy the data received from the backend into the bvec.
1061                  * Since bv_offset can be different than 0, and bv_len different
1062                  * than PAGE_SIZE, we have to keep track of the current offset,
1063                  * to be sure we are copying the data from the right shared page.
1064                  */
1065                 for_each_sg(s->sg, sg, nseg, i) {
1066                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1067                         shared_data = kmap_atomic(
1068                                 pfn_to_page(s->grants_used[i]->pfn));
1069                         bvec_data = kmap_atomic(sg_page(sg));
1070                         memcpy(bvec_data   + sg->offset,
1071                                shared_data + sg->offset,
1072                                sg->length);
1073                         kunmap_atomic(bvec_data);
1074                         kunmap_atomic(shared_data);
1075                 }
1076         }
1077         /* Add the persistent grant into the list of free grants */
1078         for (i = 0; i < nseg; i++) {
1079                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1080                         /*
1081                          * If the grant is still mapped by the backend (the
1082                          * backend has chosen to make this grant persistent)
1083                          * we add it at the head of the list, so it will be
1084                          * reused first.
1085                          */
1086                         if (!info->feature_persistent)
1087                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1088                                                      s->grants_used[i]->gref);
1089                         list_add(&s->grants_used[i]->node, &info->grants);
1090                         info->persistent_gnts_c++;
1091                 } else {
1092                         /*
1093                          * If the grant is not mapped by the backend we end the
1094                          * foreign access and add it to the tail of the list,
1095                          * so it will not be picked again unless we run out of
1096                          * persistent grants.
1097                          */
1098                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1099                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1100                         list_add_tail(&s->grants_used[i]->node, &info->grants);
1101                 }
1102         }
1103         if (s->req.operation == BLKIF_OP_INDIRECT) {
1104                 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1105                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1106                                 if (!info->feature_persistent)
1107                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1108                                                              s->indirect_grants[i]->gref);
1109                                 list_add(&s->indirect_grants[i]->node, &info->grants);
1110                                 info->persistent_gnts_c++;
1111                         } else {
1112                                 struct page *indirect_page;
1113
1114                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1115                                 /*
1116                                  * Add the used indirect page back to the list of
1117                                  * available pages for indirect grefs.
1118                                  */
1119                                 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1120                                 list_add(&indirect_page->lru, &info->indirect_pages);
1121                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1122                                 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1123                         }
1124                 }
1125         }
1126 }
1127
1128 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1129 {
1130         struct request *req;
1131         struct blkif_response *bret;
1132         RING_IDX i, rp;
1133         unsigned long flags;
1134         struct blkfront_info *info = (struct blkfront_info *)dev_id;
1135         int error;
1136
1137         spin_lock_irqsave(&info->io_lock, flags);
1138
1139         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1140                 spin_unlock_irqrestore(&info->io_lock, flags);
1141                 return IRQ_HANDLED;
1142         }
1143
1144  again:
1145         rp = info->ring.sring->rsp_prod;
1146         rmb(); /* Ensure we see queued responses up to 'rp'. */
1147
1148         for (i = info->ring.rsp_cons; i != rp; i++) {
1149                 unsigned long id;
1150
1151                 bret = RING_GET_RESPONSE(&info->ring, i);
1152                 id   = bret->id;
1153                 /*
1154                  * The backend has messed up and given us an id that we would
1155                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1156                  * look in get_id_from_freelist.
1157                  */
1158                 if (id >= BLK_RING_SIZE) {
1159                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1160                              info->gd->disk_name, op_name(bret->operation), id);
1161                         /* We can't safely get the 'struct request' as
1162                          * the id is busted. */
1163                         continue;
1164                 }
1165                 req  = info->shadow[id].request;
1166
1167                 if (bret->operation != BLKIF_OP_DISCARD)
1168                         blkif_completion(&info->shadow[id], info, bret);
1169
1170                 if (add_id_to_freelist(info, id)) {
1171                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1172                              info->gd->disk_name, op_name(bret->operation), id);
1173                         continue;
1174                 }
1175
1176                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1177                 switch (bret->operation) {
1178                 case BLKIF_OP_DISCARD:
1179                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1180                                 struct request_queue *rq = info->rq;
1181                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1182                                            info->gd->disk_name, op_name(bret->operation));
1183                                 error = -EOPNOTSUPP;
1184                                 info->feature_discard = 0;
1185                                 info->feature_secdiscard = 0;
1186                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1187                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1188                         }
1189                         __blk_end_request_all(req, error);
1190                         break;
1191                 case BLKIF_OP_FLUSH_DISKCACHE:
1192                 case BLKIF_OP_WRITE_BARRIER:
1193                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1194                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1195                                        info->gd->disk_name, op_name(bret->operation));
1196                                 error = -EOPNOTSUPP;
1197                         }
1198                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1199                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1200                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1201                                        info->gd->disk_name, op_name(bret->operation));
1202                                 error = -EOPNOTSUPP;
1203                         }
1204                         if (unlikely(error)) {
1205                                 if (error == -EOPNOTSUPP)
1206                                         error = 0;
1207                                 info->feature_flush = 0;
1208                                 xlvbd_flush(info);
1209                         }
1210                         /* fall through */
1211                 case BLKIF_OP_READ:
1212                 case BLKIF_OP_WRITE:
1213                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1214                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1215                                         "request: %x\n", bret->status);
1216
1217                         __blk_end_request_all(req, error);
1218                         break;
1219                 default:
1220                         BUG();
1221                 }
1222         }
1223
1224         info->ring.rsp_cons = i;
1225
1226         if (i != info->ring.req_prod_pvt) {
1227                 int more_to_do;
1228                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1229                 if (more_to_do)
1230                         goto again;
1231         } else
1232                 info->ring.sring->rsp_event = i + 1;
1233
1234         kick_pending_request_queues(info);
1235
1236         spin_unlock_irqrestore(&info->io_lock, flags);
1237
1238         return IRQ_HANDLED;
1239 }
1240
1241
1242 static int setup_blkring(struct xenbus_device *dev,
1243                          struct blkfront_info *info)
1244 {
1245         struct blkif_sring *sring;
1246         grant_ref_t gref;
1247         int err;
1248
1249         info->ring_ref = GRANT_INVALID_REF;
1250
1251         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1252         if (!sring) {
1253                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1254                 return -ENOMEM;
1255         }
1256         SHARED_RING_INIT(sring);
1257         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1258
1259         err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
1260         if (err < 0) {
1261                 free_page((unsigned long)sring);
1262                 info->ring.sring = NULL;
1263                 goto fail;
1264         }
1265         info->ring_ref = gref;
1266
1267         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1268         if (err)
1269                 goto fail;
1270
1271         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1272                                         "blkif", info);
1273         if (err <= 0) {
1274                 xenbus_dev_fatal(dev, err,
1275                                  "bind_evtchn_to_irqhandler failed");
1276                 goto fail;
1277         }
1278         info->irq = err;
1279
1280         return 0;
1281 fail:
1282         blkif_free(info, 0);
1283         return err;
1284 }
1285
1286
1287 /* Common code used when first setting up, and when resuming. */
1288 static int talk_to_blkback(struct xenbus_device *dev,
1289                            struct blkfront_info *info)
1290 {
1291         const char *message = NULL;
1292         struct xenbus_transaction xbt;
1293         int err;
1294
1295         /* Create shared ring, alloc event channel. */
1296         err = setup_blkring(dev, info);
1297         if (err)
1298                 goto out;
1299
1300 again:
1301         err = xenbus_transaction_start(&xbt);
1302         if (err) {
1303                 xenbus_dev_fatal(dev, err, "starting transaction");
1304                 goto destroy_blkring;
1305         }
1306
1307         err = xenbus_printf(xbt, dev->nodename,
1308                             "ring-ref", "%u", info->ring_ref);
1309         if (err) {
1310                 message = "writing ring-ref";
1311                 goto abort_transaction;
1312         }
1313         err = xenbus_printf(xbt, dev->nodename,
1314                             "event-channel", "%u", info->evtchn);
1315         if (err) {
1316                 message = "writing event-channel";
1317                 goto abort_transaction;
1318         }
1319         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1320                             XEN_IO_PROTO_ABI_NATIVE);
1321         if (err) {
1322                 message = "writing protocol";
1323                 goto abort_transaction;
1324         }
1325         err = xenbus_printf(xbt, dev->nodename,
1326                             "feature-persistent", "%u", 1);
1327         if (err)
1328                 dev_warn(&dev->dev,
1329                          "writing persistent grants feature to xenbus");
1330
1331         err = xenbus_transaction_end(xbt, 0);
1332         if (err) {
1333                 if (err == -EAGAIN)
1334                         goto again;
1335                 xenbus_dev_fatal(dev, err, "completing transaction");
1336                 goto destroy_blkring;
1337         }
1338
1339         xenbus_switch_state(dev, XenbusStateInitialised);
1340
1341         return 0;
1342
1343  abort_transaction:
1344         xenbus_transaction_end(xbt, 1);
1345         if (message)
1346                 xenbus_dev_fatal(dev, err, "%s", message);
1347  destroy_blkring:
1348         blkif_free(info, 0);
1349  out:
1350         return err;
1351 }
1352
1353 /**
1354  * Entry point to this code when a new device is created.  Allocate the basic
1355  * structures and the ring buffer for communication with the backend, and
1356  * inform the backend of the appropriate details for those.  Switch to
1357  * Initialised state.
1358  */
1359 static int blkfront_probe(struct xenbus_device *dev,
1360                           const struct xenbus_device_id *id)
1361 {
1362         int err, vdevice, i;
1363         struct blkfront_info *info;
1364
1365         /* FIXME: Use dynamic device id if this is not set. */
1366         err = xenbus_scanf(XBT_NIL, dev->nodename,
1367                            "virtual-device", "%i", &vdevice);
1368         if (err != 1) {
1369                 /* go looking in the extended area instead */
1370                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1371                                    "%i", &vdevice);
1372                 if (err != 1) {
1373                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1374                         return err;
1375                 }
1376         }
1377
1378         if (xen_hvm_domain()) {
1379                 char *type;
1380                 int len;
1381                 /* no unplug has been done: do not hook devices != xen vbds */
1382                 if (xen_has_pv_and_legacy_disk_devices()) {
1383                         int major;
1384
1385                         if (!VDEV_IS_EXTENDED(vdevice))
1386                                 major = BLKIF_MAJOR(vdevice);
1387                         else
1388                                 major = XENVBD_MAJOR;
1389
1390                         if (major != XENVBD_MAJOR) {
1391                                 printk(KERN_INFO
1392                                                 "%s: HVM does not support vbd %d as xen block device\n",
1393                                                 __func__, vdevice);
1394                                 return -ENODEV;
1395                         }
1396                 }
1397                 /* do not create a PV cdrom device if we are an HVM guest */
1398                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1399                 if (IS_ERR(type))
1400                         return -ENODEV;
1401                 if (strncmp(type, "cdrom", 5) == 0) {
1402                         kfree(type);
1403                         return -ENODEV;
1404                 }
1405                 kfree(type);
1406         }
1407         info = kzalloc(sizeof(*info), GFP_KERNEL);
1408         if (!info) {
1409                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1410                 return -ENOMEM;
1411         }
1412
1413         mutex_init(&info->mutex);
1414         spin_lock_init(&info->io_lock);
1415         info->xbdev = dev;
1416         info->vdevice = vdevice;
1417         INIT_LIST_HEAD(&info->grants);
1418         INIT_LIST_HEAD(&info->indirect_pages);
1419         info->persistent_gnts_c = 0;
1420         info->connected = BLKIF_STATE_DISCONNECTED;
1421         INIT_WORK(&info->work, blkif_restart_queue);
1422
1423         for (i = 0; i < BLK_RING_SIZE; i++)
1424                 info->shadow[i].req.u.rw.id = i+1;
1425         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1426
1427         /* Front end dir is a number, which is used as the id. */
1428         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1429         dev_set_drvdata(&dev->dev, info);
1430
1431         err = talk_to_blkback(dev, info);
1432         if (err) {
1433                 kfree(info);
1434                 dev_set_drvdata(&dev->dev, NULL);
1435                 return err;
1436         }
1437
1438         return 0;
1439 }
1440
1441 static void split_bio_end(struct bio *bio, int error)
1442 {
1443         struct split_bio *split_bio = bio->bi_private;
1444
1445         if (error)
1446                 split_bio->err = error;
1447
1448         if (atomic_dec_and_test(&split_bio->pending)) {
1449                 split_bio->bio->bi_phys_segments = 0;
1450                 bio_endio(split_bio->bio, split_bio->err);
1451                 kfree(split_bio);
1452         }
1453         bio_put(bio);
1454 }
1455
1456 static int blkif_recover(struct blkfront_info *info)
1457 {
1458         int i;
1459         struct request *req, *n;
1460         struct blk_shadow *copy;
1461         int rc;
1462         struct bio *bio, *cloned_bio;
1463         struct bio_list bio_list, merge_bio;
1464         unsigned int segs, offset;
1465         int pending, size;
1466         struct split_bio *split_bio;
1467         struct list_head requests;
1468
1469         /* Stage 1: Make a safe copy of the shadow state. */
1470         copy = kmemdup(info->shadow, sizeof(info->shadow),
1471                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1472         if (!copy)
1473                 return -ENOMEM;
1474
1475         /* Stage 2: Set up free list. */
1476         memset(&info->shadow, 0, sizeof(info->shadow));
1477         for (i = 0; i < BLK_RING_SIZE; i++)
1478                 info->shadow[i].req.u.rw.id = i+1;
1479         info->shadow_free = info->ring.req_prod_pvt;
1480         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1481
1482         rc = blkfront_setup_indirect(info);
1483         if (rc) {
1484                 kfree(copy);
1485                 return rc;
1486         }
1487
1488         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1489         blk_queue_max_segments(info->rq, segs);
1490         bio_list_init(&bio_list);
1491         INIT_LIST_HEAD(&requests);
1492         for (i = 0; i < BLK_RING_SIZE; i++) {
1493                 /* Not in use? */
1494                 if (!copy[i].request)
1495                         continue;
1496
1497                 /*
1498                  * Get the bios in the request so we can re-queue them.
1499                  */
1500                 if (copy[i].request->cmd_flags &
1501                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1502                         /*
1503                          * Flush operations don't contain bios, so
1504                          * we need to requeue the whole request
1505                          */
1506                         list_add(&copy[i].request->queuelist, &requests);
1507                         continue;
1508                 }
1509                 merge_bio.head = copy[i].request->bio;
1510                 merge_bio.tail = copy[i].request->biotail;
1511                 bio_list_merge(&bio_list, &merge_bio);
1512                 copy[i].request->bio = NULL;
1513                 blk_end_request_all(copy[i].request, 0);
1514         }
1515
1516         kfree(copy);
1517
1518         /*
1519          * Empty the queue, this is important because we might have
1520          * requests in the queue with more segments than what we
1521          * can handle now.
1522          */
1523         spin_lock_irq(&info->io_lock);
1524         while ((req = blk_fetch_request(info->rq)) != NULL) {
1525                 if (req->cmd_flags &
1526                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1527                         list_add(&req->queuelist, &requests);
1528                         continue;
1529                 }
1530                 merge_bio.head = req->bio;
1531                 merge_bio.tail = req->biotail;
1532                 bio_list_merge(&bio_list, &merge_bio);
1533                 req->bio = NULL;
1534                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1535                         pr_alert("diskcache flush request found!\n");
1536                 __blk_end_request_all(req, 0);
1537         }
1538         spin_unlock_irq(&info->io_lock);
1539
1540         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1541
1542         spin_lock_irq(&info->io_lock);
1543
1544         /* Now safe for us to use the shared ring */
1545         info->connected = BLKIF_STATE_CONNECTED;
1546
1547         /* Kick any other new requests queued since we resumed */
1548         kick_pending_request_queues(info);
1549
1550         list_for_each_entry_safe(req, n, &requests, queuelist) {
1551                 /* Requeue pending requests (flush or discard) */
1552                 list_del_init(&req->queuelist);
1553                 BUG_ON(req->nr_phys_segments > segs);
1554                 blk_requeue_request(info->rq, req);
1555         }
1556         spin_unlock_irq(&info->io_lock);
1557
1558         while ((bio = bio_list_pop(&bio_list)) != NULL) {
1559                 /* Traverse the list of pending bios and re-queue them */
1560                 if (bio_segments(bio) > segs) {
1561                         /*
1562                          * This bio has more segments than what we can
1563                          * handle, we have to split it.
1564                          */
1565                         pending = (bio_segments(bio) + segs - 1) / segs;
1566                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1567                         BUG_ON(split_bio == NULL);
1568                         atomic_set(&split_bio->pending, pending);
1569                         split_bio->bio = bio;
1570                         for (i = 0; i < pending; i++) {
1571                                 offset = (i * segs * PAGE_SIZE) >> 9;
1572                                 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1573                                            (unsigned int)bio_sectors(bio) - offset);
1574                                 cloned_bio = bio_clone(bio, GFP_NOIO);
1575                                 BUG_ON(cloned_bio == NULL);
1576                                 bio_trim(cloned_bio, offset, size);
1577                                 cloned_bio->bi_private = split_bio;
1578                                 cloned_bio->bi_end_io = split_bio_end;
1579                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
1580                         }
1581                         /*
1582                          * Now we have to wait for all those smaller bios to
1583                          * end, so we can also end the "parent" bio.
1584                          */
1585                         continue;
1586                 }
1587                 /* We don't need to split this bio */
1588                 submit_bio(bio->bi_rw, bio);
1589         }
1590
1591         return 0;
1592 }
1593
1594 /**
1595  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1596  * driver restart.  We tear down our blkif structure and recreate it, but
1597  * leave the device-layer structures intact so that this is transparent to the
1598  * rest of the kernel.
1599  */
1600 static int blkfront_resume(struct xenbus_device *dev)
1601 {
1602         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1603         int err;
1604
1605         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1606
1607         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1608
1609         err = talk_to_blkback(dev, info);
1610
1611         /*
1612          * We have to wait for the backend to switch to
1613          * connected state, since we want to read which
1614          * features it supports.
1615          */
1616
1617         return err;
1618 }
1619
1620 static void
1621 blkfront_closing(struct blkfront_info *info)
1622 {
1623         struct xenbus_device *xbdev = info->xbdev;
1624         struct block_device *bdev = NULL;
1625
1626         mutex_lock(&info->mutex);
1627
1628         if (xbdev->state == XenbusStateClosing) {
1629                 mutex_unlock(&info->mutex);
1630                 return;
1631         }
1632
1633         if (info->gd)
1634                 bdev = bdget_disk(info->gd, 0);
1635
1636         mutex_unlock(&info->mutex);
1637
1638         if (!bdev) {
1639                 xenbus_frontend_closed(xbdev);
1640                 return;
1641         }
1642
1643         mutex_lock(&bdev->bd_mutex);
1644
1645         if (bdev->bd_openers) {
1646                 xenbus_dev_error(xbdev, -EBUSY,
1647                                  "Device in use; refusing to close");
1648                 xenbus_switch_state(xbdev, XenbusStateClosing);
1649         } else {
1650                 xlvbd_release_gendisk(info);
1651                 xenbus_frontend_closed(xbdev);
1652         }
1653
1654         mutex_unlock(&bdev->bd_mutex);
1655         bdput(bdev);
1656 }
1657
1658 static void blkfront_setup_discard(struct blkfront_info *info)
1659 {
1660         int err;
1661         unsigned int discard_granularity;
1662         unsigned int discard_alignment;
1663         unsigned int discard_secure;
1664
1665         info->feature_discard = 1;
1666         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1667                 "discard-granularity", "%u", &discard_granularity,
1668                 "discard-alignment", "%u", &discard_alignment,
1669                 NULL);
1670         if (!err) {
1671                 info->discard_granularity = discard_granularity;
1672                 info->discard_alignment = discard_alignment;
1673         }
1674         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1675                     "discard-secure", "%d", &discard_secure,
1676                     NULL);
1677         if (!err)
1678                 info->feature_secdiscard = !!discard_secure;
1679 }
1680
1681 static int blkfront_setup_indirect(struct blkfront_info *info)
1682 {
1683         unsigned int indirect_segments, segs;
1684         int err, i;
1685
1686         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1687                             "feature-max-indirect-segments", "%u", &indirect_segments,
1688                             NULL);
1689         if (err) {
1690                 info->max_indirect_segments = 0;
1691                 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1692         } else {
1693                 info->max_indirect_segments = min(indirect_segments,
1694                                                   xen_blkif_max_segments);
1695                 segs = info->max_indirect_segments;
1696         }
1697
1698         err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1699         if (err)
1700                 goto out_of_memory;
1701
1702         if (!info->feature_persistent && info->max_indirect_segments) {
1703                 /*
1704                  * We are using indirect descriptors but not persistent
1705                  * grants, we need to allocate a set of pages that can be
1706                  * used for mapping indirect grefs
1707                  */
1708                 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1709
1710                 BUG_ON(!list_empty(&info->indirect_pages));
1711                 for (i = 0; i < num; i++) {
1712                         struct page *indirect_page = alloc_page(GFP_NOIO);
1713                         if (!indirect_page)
1714                                 goto out_of_memory;
1715                         list_add(&indirect_page->lru, &info->indirect_pages);
1716                 }
1717         }
1718
1719         for (i = 0; i < BLK_RING_SIZE; i++) {
1720                 info->shadow[i].grants_used = kzalloc(
1721                         sizeof(info->shadow[i].grants_used[0]) * segs,
1722                         GFP_NOIO);
1723                 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1724                 if (info->max_indirect_segments)
1725                         info->shadow[i].indirect_grants = kzalloc(
1726                                 sizeof(info->shadow[i].indirect_grants[0]) *
1727                                 INDIRECT_GREFS(segs),
1728                                 GFP_NOIO);
1729                 if ((info->shadow[i].grants_used == NULL) ||
1730                         (info->shadow[i].sg == NULL) ||
1731                      (info->max_indirect_segments &&
1732                      (info->shadow[i].indirect_grants == NULL)))
1733                         goto out_of_memory;
1734                 sg_init_table(info->shadow[i].sg, segs);
1735         }
1736
1737
1738         return 0;
1739
1740 out_of_memory:
1741         for (i = 0; i < BLK_RING_SIZE; i++) {
1742                 kfree(info->shadow[i].grants_used);
1743                 info->shadow[i].grants_used = NULL;
1744                 kfree(info->shadow[i].sg);
1745                 info->shadow[i].sg = NULL;
1746                 kfree(info->shadow[i].indirect_grants);
1747                 info->shadow[i].indirect_grants = NULL;
1748         }
1749         if (!list_empty(&info->indirect_pages)) {
1750                 struct page *indirect_page, *n;
1751                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1752                         list_del(&indirect_page->lru);
1753                         __free_page(indirect_page);
1754                 }
1755         }
1756         return -ENOMEM;
1757 }
1758
1759 /*
1760  * Invoked when the backend is finally 'ready' (and has told produced
1761  * the details about the physical device - #sectors, size, etc).
1762  */
1763 static void blkfront_connect(struct blkfront_info *info)
1764 {
1765         unsigned long long sectors;
1766         unsigned long sector_size;
1767         unsigned int physical_sector_size;
1768         unsigned int binfo;
1769         int err;
1770         int barrier, flush, discard, persistent;
1771
1772         switch (info->connected) {
1773         case BLKIF_STATE_CONNECTED:
1774                 /*
1775                  * Potentially, the back-end may be signalling
1776                  * a capacity change; update the capacity.
1777                  */
1778                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1779                                    "sectors", "%Lu", &sectors);
1780                 if (XENBUS_EXIST_ERR(err))
1781                         return;
1782                 printk(KERN_INFO "Setting capacity to %Lu\n",
1783                        sectors);
1784                 set_capacity(info->gd, sectors);
1785                 revalidate_disk(info->gd);
1786
1787                 return;
1788         case BLKIF_STATE_SUSPENDED:
1789                 /*
1790                  * If we are recovering from suspension, we need to wait
1791                  * for the backend to announce it's features before
1792                  * reconnecting, at least we need to know if the backend
1793                  * supports indirect descriptors, and how many.
1794                  */
1795                 blkif_recover(info);
1796                 return;
1797
1798         default:
1799                 break;
1800         }
1801
1802         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1803                 __func__, info->xbdev->otherend);
1804
1805         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1806                             "sectors", "%llu", &sectors,
1807                             "info", "%u", &binfo,
1808                             "sector-size", "%lu", &sector_size,
1809                             NULL);
1810         if (err) {
1811                 xenbus_dev_fatal(info->xbdev, err,
1812                                  "reading backend fields at %s",
1813                                  info->xbdev->otherend);
1814                 return;
1815         }
1816
1817         /*
1818          * physcial-sector-size is a newer field, so old backends may not
1819          * provide this. Assume physical sector size to be the same as
1820          * sector_size in that case.
1821          */
1822         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1823                            "physical-sector-size", "%u", &physical_sector_size);
1824         if (err != 1)
1825                 physical_sector_size = sector_size;
1826
1827         info->feature_flush = 0;
1828
1829         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1830                             "feature-barrier", "%d", &barrier,
1831                             NULL);
1832
1833         /*
1834          * If there's no "feature-barrier" defined, then it means
1835          * we're dealing with a very old backend which writes
1836          * synchronously; nothing to do.
1837          *
1838          * If there are barriers, then we use flush.
1839          */
1840         if (!err && barrier)
1841                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1842         /*
1843          * And if there is "feature-flush-cache" use that above
1844          * barriers.
1845          */
1846         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1847                             "feature-flush-cache", "%d", &flush,
1848                             NULL);
1849
1850         if (!err && flush)
1851                 info->feature_flush = REQ_FLUSH;
1852
1853         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1854                             "feature-discard", "%d", &discard,
1855                             NULL);
1856
1857         if (!err && discard)
1858                 blkfront_setup_discard(info);
1859
1860         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1861                             "feature-persistent", "%u", &persistent,
1862                             NULL);
1863         if (err)
1864                 info->feature_persistent = 0;
1865         else
1866                 info->feature_persistent = persistent;
1867
1868         err = blkfront_setup_indirect(info);
1869         if (err) {
1870                 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1871                                  info->xbdev->otherend);
1872                 return;
1873         }
1874
1875         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1876                                   physical_sector_size);
1877         if (err) {
1878                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1879                                  info->xbdev->otherend);
1880                 return;
1881         }
1882
1883         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1884
1885         /* Kick pending requests. */
1886         spin_lock_irq(&info->io_lock);
1887         info->connected = BLKIF_STATE_CONNECTED;
1888         kick_pending_request_queues(info);
1889         spin_unlock_irq(&info->io_lock);
1890
1891         add_disk(info->gd);
1892
1893         info->is_ready = 1;
1894 }
1895
1896 /**
1897  * Callback received when the backend's state changes.
1898  */
1899 static void blkback_changed(struct xenbus_device *dev,
1900                             enum xenbus_state backend_state)
1901 {
1902         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1903
1904         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1905
1906         switch (backend_state) {
1907         case XenbusStateInitialising:
1908         case XenbusStateInitWait:
1909         case XenbusStateInitialised:
1910         case XenbusStateReconfiguring:
1911         case XenbusStateReconfigured:
1912         case XenbusStateUnknown:
1913                 break;
1914
1915         case XenbusStateConnected:
1916                 blkfront_connect(info);
1917                 break;
1918
1919         case XenbusStateClosed:
1920                 if (dev->state == XenbusStateClosed)
1921                         break;
1922                 /* Missed the backend's Closing state -- fallthrough */
1923         case XenbusStateClosing:
1924                 blkfront_closing(info);
1925                 break;
1926         }
1927 }
1928
1929 static int blkfront_remove(struct xenbus_device *xbdev)
1930 {
1931         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1932         struct block_device *bdev = NULL;
1933         struct gendisk *disk;
1934
1935         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1936
1937         blkif_free(info, 0);
1938
1939         mutex_lock(&info->mutex);
1940
1941         disk = info->gd;
1942         if (disk)
1943                 bdev = bdget_disk(disk, 0);
1944
1945         info->xbdev = NULL;
1946         mutex_unlock(&info->mutex);
1947
1948         if (!bdev) {
1949                 kfree(info);
1950                 return 0;
1951         }
1952
1953         /*
1954          * The xbdev was removed before we reached the Closed
1955          * state. See if it's safe to remove the disk. If the bdev
1956          * isn't closed yet, we let release take care of it.
1957          */
1958
1959         mutex_lock(&bdev->bd_mutex);
1960         info = disk->private_data;
1961
1962         dev_warn(disk_to_dev(disk),
1963                  "%s was hot-unplugged, %d stale handles\n",
1964                  xbdev->nodename, bdev->bd_openers);
1965
1966         if (info && !bdev->bd_openers) {
1967                 xlvbd_release_gendisk(info);
1968                 disk->private_data = NULL;
1969                 kfree(info);
1970         }
1971
1972         mutex_unlock(&bdev->bd_mutex);
1973         bdput(bdev);
1974
1975         return 0;
1976 }
1977
1978 static int blkfront_is_ready(struct xenbus_device *dev)
1979 {
1980         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1981
1982         return info->is_ready && info->xbdev;
1983 }
1984
1985 static int blkif_open(struct block_device *bdev, fmode_t mode)
1986 {
1987         struct gendisk *disk = bdev->bd_disk;
1988         struct blkfront_info *info;
1989         int err = 0;
1990
1991         mutex_lock(&blkfront_mutex);
1992
1993         info = disk->private_data;
1994         if (!info) {
1995                 /* xbdev gone */
1996                 err = -ERESTARTSYS;
1997                 goto out;
1998         }
1999
2000         mutex_lock(&info->mutex);
2001
2002         if (!info->gd)
2003                 /* xbdev is closed */
2004                 err = -ERESTARTSYS;
2005
2006         mutex_unlock(&info->mutex);
2007
2008 out:
2009         mutex_unlock(&blkfront_mutex);
2010         return err;
2011 }
2012
2013 static void blkif_release(struct gendisk *disk, fmode_t mode)
2014 {
2015         struct blkfront_info *info = disk->private_data;
2016         struct block_device *bdev;
2017         struct xenbus_device *xbdev;
2018
2019         mutex_lock(&blkfront_mutex);
2020
2021         bdev = bdget_disk(disk, 0);
2022
2023         if (!bdev) {
2024                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2025                 goto out_mutex;
2026         }
2027         if (bdev->bd_openers)
2028                 goto out;
2029
2030         /*
2031          * Check if we have been instructed to close. We will have
2032          * deferred this request, because the bdev was still open.
2033          */
2034
2035         mutex_lock(&info->mutex);
2036         xbdev = info->xbdev;
2037
2038         if (xbdev && xbdev->state == XenbusStateClosing) {
2039                 /* pending switch to state closed */
2040                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2041                 xlvbd_release_gendisk(info);
2042                 xenbus_frontend_closed(info->xbdev);
2043         }
2044
2045         mutex_unlock(&info->mutex);
2046
2047         if (!xbdev) {
2048                 /* sudden device removal */
2049                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2050                 xlvbd_release_gendisk(info);
2051                 disk->private_data = NULL;
2052                 kfree(info);
2053         }
2054
2055 out:
2056         bdput(bdev);
2057 out_mutex:
2058         mutex_unlock(&blkfront_mutex);
2059 }
2060
2061 static const struct block_device_operations xlvbd_block_fops =
2062 {
2063         .owner = THIS_MODULE,
2064         .open = blkif_open,
2065         .release = blkif_release,
2066         .getgeo = blkif_getgeo,
2067         .ioctl = blkif_ioctl,
2068 };
2069
2070
2071 static const struct xenbus_device_id blkfront_ids[] = {
2072         { "vbd" },
2073         { "" }
2074 };
2075
2076 static struct xenbus_driver blkfront_driver = {
2077         .ids  = blkfront_ids,
2078         .probe = blkfront_probe,
2079         .remove = blkfront_remove,
2080         .resume = blkfront_resume,
2081         .otherend_changed = blkback_changed,
2082         .is_ready = blkfront_is_ready,
2083 };
2084
2085 static int __init xlblk_init(void)
2086 {
2087         int ret;
2088
2089         if (!xen_domain())
2090                 return -ENODEV;
2091
2092         if (!xen_has_pv_disk_devices())
2093                 return -ENODEV;
2094
2095         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2096                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2097                        XENVBD_MAJOR, DEV_NAME);
2098                 return -ENODEV;
2099         }
2100
2101         ret = xenbus_register_frontend(&blkfront_driver);
2102         if (ret) {
2103                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2104                 return ret;
2105         }
2106
2107         return 0;
2108 }
2109 module_init(xlblk_init);
2110
2111
2112 static void __exit xlblk_exit(void)
2113 {
2114         xenbus_unregister_driver(&blkfront_driver);
2115         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2116         kfree(minors);
2117 }
2118 module_exit(xlblk_exit);
2119
2120 MODULE_DESCRIPTION("Xen virtual block device frontend");
2121 MODULE_LICENSE("GPL");
2122 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2123 MODULE_ALIAS("xen:vbd");
2124 MODULE_ALIAS("xenblk");