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