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