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