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/module.h>
  41
  42 #include <xen/xenbus.h>
  43 #include <xen/grant_table.h>
  44 #include <xen/events.h>
  45 #include <xen/page.h>
  46
  47 #include <xen/interface/grant_table.h>
  48 #include <xen/interface/io/blkif.h>
  49
  50 #include <asm/xen/hypervisor.h>
  51
  52 enum blkif_state {
  53         BLKIF_STATE_DISCONNECTED,
  54         BLKIF_STATE_CONNECTED,
  55         BLKIF_STATE_SUSPENDED,
  56 };
  57
  58 struct blk_shadow {
  59         struct blkif_request req;
  60         unsigned long request;
  61         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  62 };
  63
  64 static struct block_device_operations xlvbd_block_fops;
  65
  66 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  67
  68 /*
  69  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  70  * hang in private_data off the gendisk structure. We may end up
  71  * putting all kinds of interesting stuff here :-)
  72  */
  73 struct blkfront_info
  74 {
  75         struct xenbus_device *xbdev;
  76         dev_t dev;
  77         struct gendisk *gd;
  78         int vdevice;
  79         blkif_vdev_t handle;
  80         enum blkif_state connected;
  81         int ring_ref;
  82         struct blkif_front_ring ring;
  83         unsigned int evtchn, irq;
  84         struct request_queue *rq;
  85         struct work_struct work;
  86         struct gnttab_free_callback callback;
  87         struct blk_shadow shadow[BLK_RING_SIZE];
  88         unsigned long shadow_free;
  89         int feature_barrier;
  90
  91         /**
  92          * The number of people holding this device open.  We won't allow a
  93          * hot-unplug unless this is 0.
  94          */
  95         int users;
  96 };
  97
  98 static DEFINE_SPINLOCK(blkif_io_lock);
  99
 100 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 101         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 102 #define GRANT_INVALID_REF       0
 103
 104 #define PARTS_PER_DISK          16
 105
 106 #define BLKIF_MAJOR(dev) ((dev)>>8)
 107 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 108
 109 #define DEV_NAME        "xvd"   /* name in /dev */
 110
 111 /* Information about our VBDs. */
 112 #define MAX_VBDS 64
 113 static LIST_HEAD(vbds_list);
 114
 115 static int get_id_from_freelist(struct blkfront_info *info)
 116 {
 117         unsigned long free = info->shadow_free;
 118         BUG_ON(free > BLK_RING_SIZE);
 119         info->shadow_free = info->shadow[free].req.id;
 120         info->shadow[free].req.id = 0x0fffffee; /* debug */
 121         return free;
 122 }
 123
 124 static void add_id_to_freelist(struct blkfront_info *info,
 125                                unsigned long id)
 126 {
 127         info->shadow[id].req.id  = info->shadow_free;
 128         info->shadow[id].request = 0;
 129         info->shadow_free = id;
 130 }
 131
 132 static void blkif_restart_queue_callback(void *arg)
 133 {
 134         struct blkfront_info *info = (struct blkfront_info *)arg;
 135         schedule_work(&info->work);
 136 }
 137
 138 /*
 139  * blkif_queue_request
 140  *
 141  * request block io
 142  *
 143  * id: for guest use only.
 144  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 145  * buffer: buffer to read/write into. this should be a
 146  *   virtual address in the guest os.
 147  */
 148 static int blkif_queue_request(struct request *req)
 149 {
 150         struct blkfront_info *info = req->rq_disk->private_data;
 151         unsigned long buffer_mfn;
 152         struct blkif_request *ring_req;
 153         struct bio *bio;
 154         struct bio_vec *bvec;
 155         int idx;
 156         unsigned long id;
 157         unsigned int fsect, lsect;
 158         int ref;
 159         grant_ref_t gref_head;
 160
 161         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 162                 return 1;
 163
 164         if (gnttab_alloc_grant_references(
 165                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 166                 gnttab_request_free_callback(
 167                         &info->callback,
 168                         blkif_restart_queue_callback,
 169                         info,
 170                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 171                 return 1;
 172         }
 173
 174         /* Fill out a communications ring structure. */
 175         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 176         id = get_id_from_freelist(info);
 177         info->shadow[id].request = (unsigned long)req;
 178
 179         ring_req->id = id;
 180         ring_req->sector_number = (blkif_sector_t)req->sector;
 181         ring_req->handle = info->handle;
 182
 183         ring_req->operation = rq_data_dir(req) ?
 184                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 185         if (blk_barrier_rq(req))
 186                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 187
 188         ring_req->nr_segments = 0;
 189         rq_for_each_bio (bio, req) {
 190                 bio_for_each_segment (bvec, bio, idx) {
 191                         BUG_ON(ring_req->nr_segments
 192                                == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 193                         buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
 194                         fsect = bvec->bv_offset >> 9;
 195                         lsect = fsect + (bvec->bv_len >> 9) - 1;
 196                         /* install a grant reference. */
 197                         ref = gnttab_claim_grant_reference(&gref_head);
 198                         BUG_ON(ref == -ENOSPC);
 199
 200                         gnttab_grant_foreign_access_ref(
 201                                 ref,
 202                                 info->xbdev->otherend_id,
 203                                 buffer_mfn,
 204                                 rq_data_dir(req) );
 205
 206                         info->shadow[id].frame[ring_req->nr_segments] =
 207                                 mfn_to_pfn(buffer_mfn);
 208
 209                         ring_req->seg[ring_req->nr_segments] =
 210                                 (struct blkif_request_segment) {
 211                                         .gref       = ref,
 212                                         .first_sect = fsect,
 213                                         .last_sect  = lsect };
 214
 215                         ring_req->nr_segments++;
 216                 }
 217         }
 218
 219         info->ring.req_prod_pvt++;
 220
 221         /* Keep a private copy so we can reissue requests when recovering. */
 222         info->shadow[id].req = *ring_req;
 223
 224         gnttab_free_grant_references(gref_head);
 225
 226         return 0;
 227 }
 228
 229
 230 static inline void flush_requests(struct blkfront_info *info)
 231 {
 232         int notify;
 233
 234         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 235
 236         if (notify)
 237                 notify_remote_via_irq(info->irq);
 238 }
 239
 240 /*
 241  * do_blkif_request
 242  *  read a block; request is in a request queue
 243  */
 244 static void do_blkif_request(struct request_queue *rq)
 245 {
 246         struct blkfront_info *info = NULL;
 247         struct request *req;
 248         int queued;
 249
 250         pr_debug("Entered do_blkif_request\n");
 251
 252         queued = 0;
 253
 254         while ((req = elv_next_request(rq)) != NULL) {
 255                 info = req->rq_disk->private_data;
 256                 if (!blk_fs_request(req)) {
 257                         end_request(req, 0);
 258                         continue;
 259                 }
 260
 261                 if (RING_FULL(&info->ring))
 262                         goto wait;
 263
 264                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 265                          "(%u/%li) buffer:%p [%s]\n",
 266                          req, req->cmd, (unsigned long)req->sector,
 267                          req->current_nr_sectors,
 268                          req->nr_sectors, req->buffer,
 269                          rq_data_dir(req) ? "write" : "read");
 270
 271
 272                 blkdev_dequeue_request(req);
 273                 if (blkif_queue_request(req)) {
 274                         blk_requeue_request(rq, req);
 275 wait:
 276                         /* Avoid pointless unplugs. */
 277                         blk_stop_queue(rq);
 278                         break;
 279                 }
 280
 281                 queued++;
 282         }
 283
 284         if (queued != 0)
 285                 flush_requests(info);
 286 }
 287
 288 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 289 {
 290         struct request_queue *rq;
 291
 292         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 293         if (rq == NULL)
 294                 return -1;
 295
 296         elevator_init(rq, "noop");
 297
 298         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 299         blk_queue_hardsect_size(rq, sector_size);
 300         blk_queue_max_sectors(rq, 512);
 301
 302         /* Each segment in a request is up to an aligned page in size. */
 303         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 304         blk_queue_max_segment_size(rq, PAGE_SIZE);
 305
 306         /* Ensure a merged request will fit in a single I/O ring slot. */
 307         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 308         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 309
 310         /* Make sure buffer addresses are sector-aligned. */
 311         blk_queue_dma_alignment(rq, 511);
 312
 313         gd->queue = rq;
 314
 315         return 0;
 316 }
 317
 318
 319 static int xlvbd_barrier(struct blkfront_info *info)
 320 {
 321         int err;
 322
 323         err = blk_queue_ordered(info->rq,
 324                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 325                                 NULL);
 326
 327         if (err)
 328                 return err;
 329
 330         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 331                info->gd->disk_name,
 332                info->feature_barrier ? "enabled" : "disabled");
 333         return 0;
 334 }
 335
 336
 337 static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
 338                                int vdevice, u16 vdisk_info, u16 sector_size,
 339                                struct blkfront_info *info)
 340 {
 341         struct gendisk *gd;
 342         int nr_minors = 1;
 343         int err = -ENODEV;
 344
 345         BUG_ON(info->gd != NULL);
 346         BUG_ON(info->rq != NULL);
 347
 348         if ((minor % PARTS_PER_DISK) == 0)
 349                 nr_minors = PARTS_PER_DISK;
 350
 351         gd = alloc_disk(nr_minors);
 352         if (gd == NULL)
 353                 goto out;
 354
 355         if (nr_minors > 1)
 356                 sprintf(gd->disk_name, "%s%c", DEV_NAME,
 357                         'a' + minor / PARTS_PER_DISK);
 358         else
 359                 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 360                         'a' + minor / PARTS_PER_DISK,
 361                         minor % PARTS_PER_DISK);
 362
 363         gd->major = XENVBD_MAJOR;
 364         gd->first_minor = minor;
 365         gd->fops = &xlvbd_block_fops;
 366         gd->private_data = info;
 367         gd->driverfs_dev = &(info->xbdev->dev);
 368         set_capacity(gd, capacity);
 369
 370         if (xlvbd_init_blk_queue(gd, sector_size)) {
 371                 del_gendisk(gd);
 372                 goto out;
 373         }
 374
 375         info->rq = gd->queue;
 376         info->gd = gd;
 377
 378         if (info->feature_barrier)
 379                 xlvbd_barrier(info);
 380
 381         if (vdisk_info & VDISK_READONLY)
 382                 set_disk_ro(gd, 1);
 383
 384         if (vdisk_info & VDISK_REMOVABLE)
 385                 gd->flags |= GENHD_FL_REMOVABLE;
 386
 387         if (vdisk_info & VDISK_CDROM)
 388                 gd->flags |= GENHD_FL_CD;
 389
 390         return 0;
 391
 392  out:
 393         return err;
 394 }
 395
 396 static void kick_pending_request_queues(struct blkfront_info *info)
 397 {
 398         if (!RING_FULL(&info->ring)) {
 399                 /* Re-enable calldowns. */
 400                 blk_start_queue(info->rq);
 401                 /* Kick things off immediately. */
 402                 do_blkif_request(info->rq);
 403         }
 404 }
 405
 406 static void blkif_restart_queue(struct work_struct *work)
 407 {
 408         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 409
 410         spin_lock_irq(&blkif_io_lock);
 411         if (info->connected == BLKIF_STATE_CONNECTED)
 412                 kick_pending_request_queues(info);
 413         spin_unlock_irq(&blkif_io_lock);
 414 }
 415
 416 static void blkif_free(struct blkfront_info *info, int suspend)
 417 {
 418         /* Prevent new requests being issued until we fix things up. */
 419         spin_lock_irq(&blkif_io_lock);
 420         info->connected = suspend ?
 421                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 422         /* No more blkif_request(). */
 423         if (info->rq)
 424                 blk_stop_queue(info->rq);
 425         /* No more gnttab callback work. */
 426         gnttab_cancel_free_callback(&info->callback);
 427         spin_unlock_irq(&blkif_io_lock);
 428
 429         /* Flush gnttab callback work. Must be done with no locks held. */
 430         flush_scheduled_work();
 431
 432         /* Free resources associated with old device channel. */
 433         if (info->ring_ref != GRANT_INVALID_REF) {
 434                 gnttab_end_foreign_access(info->ring_ref, 0,
 435                                           (unsigned long)info->ring.sring);
 436                 info->ring_ref = GRANT_INVALID_REF;
 437                 info->ring.sring = NULL;
 438         }
 439         if (info->irq)
 440                 unbind_from_irqhandler(info->irq, info);
 441         info->evtchn = info->irq = 0;
 442
 443 }
 444
 445 static void blkif_completion(struct blk_shadow *s)
 446 {
 447         int i;
 448         for (i = 0; i < s->req.nr_segments; i++)
 449                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 450 }
 451
 452 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 453 {
 454         struct request *req;
 455         struct blkif_response *bret;
 456         RING_IDX i, rp;
 457         unsigned long flags;
 458         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 459         int uptodate;
 460
 461         spin_lock_irqsave(&blkif_io_lock, flags);
 462
 463         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 464                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 465                 return IRQ_HANDLED;
 466         }
 467
 468  again:
 469         rp = info->ring.sring->rsp_prod;
 470         rmb(); /* Ensure we see queued responses up to 'rp'. */
 471
 472         for (i = info->ring.rsp_cons; i != rp; i++) {
 473                 unsigned long id;
 474                 int ret;
 475
 476                 bret = RING_GET_RESPONSE(&info->ring, i);
 477                 id   = bret->id;
 478                 req  = (struct request *)info->shadow[id].request;
 479
 480                 blkif_completion(&info->shadow[id]);
 481
 482                 add_id_to_freelist(info, id);
 483
 484                 uptodate = (bret->status == BLKIF_RSP_OKAY);
 485                 switch (bret->operation) {
 486                 case BLKIF_OP_WRITE_BARRIER:
 487                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 488                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 489                                        info->gd->disk_name);
 490                                 uptodate = -EOPNOTSUPP;
 491                                 info->feature_barrier = 0;
 492                                 xlvbd_barrier(info);
 493                         }
 494                         /* fall through */
 495                 case BLKIF_OP_READ:
 496                 case BLKIF_OP_WRITE:
 497                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 498                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 499                                         "request: %x\n", bret->status);
 500
 501                         ret = end_that_request_first(req, uptodate,
 502                                 req->hard_nr_sectors);
 503                         BUG_ON(ret);
 504                         end_that_request_last(req, uptodate);
 505                         break;
 506                 default:
 507                         BUG();
 508                 }
 509         }
 510
 511         info->ring.rsp_cons = i;
 512
 513         if (i != info->ring.req_prod_pvt) {
 514                 int more_to_do;
 515                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 516                 if (more_to_do)
 517                         goto again;
 518         } else
 519                 info->ring.sring->rsp_event = i + 1;
 520
 521         kick_pending_request_queues(info);
 522
 523         spin_unlock_irqrestore(&blkif_io_lock, flags);
 524
 525         return IRQ_HANDLED;
 526 }
 527
 528
 529 static int setup_blkring(struct xenbus_device *dev,
 530                          struct blkfront_info *info)
 531 {
 532         struct blkif_sring *sring;
 533         int err;
 534
 535         info->ring_ref = GRANT_INVALID_REF;
 536
 537         sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL);
 538         if (!sring) {
 539                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 540                 return -ENOMEM;
 541         }
 542         SHARED_RING_INIT(sring);
 543         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 544
 545         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 546         if (err < 0) {
 547                 free_page((unsigned long)sring);
 548                 info->ring.sring = NULL;
 549                 goto fail;
 550         }
 551         info->ring_ref = err;
 552
 553         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 554         if (err)
 555                 goto fail;
 556
 557         err = bind_evtchn_to_irqhandler(info->evtchn,
 558                                         blkif_interrupt,
 559                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 560         if (err <= 0) {
 561                 xenbus_dev_fatal(dev, err,
 562                                  "bind_evtchn_to_irqhandler failed");
 563                 goto fail;
 564         }
 565         info->irq = err;
 566
 567         return 0;
 568 fail:
 569         blkif_free(info, 0);
 570         return err;
 571 }
 572
 573
 574 /* Common code used when first setting up, and when resuming. */
 575 static int talk_to_backend(struct xenbus_device *dev,
 576                            struct blkfront_info *info)
 577 {
 578         const char *message = NULL;
 579         struct xenbus_transaction xbt;
 580         int err;
 581
 582         /* Create shared ring, alloc event channel. */
 583         err = setup_blkring(dev, info);
 584         if (err)
 585                 goto out;
 586
 587 again:
 588         err = xenbus_transaction_start(&xbt);
 589         if (err) {
 590                 xenbus_dev_fatal(dev, err, "starting transaction");
 591                 goto destroy_blkring;
 592         }
 593
 594         err = xenbus_printf(xbt, dev->nodename,
 595                             "ring-ref", "%u", info->ring_ref);
 596         if (err) {
 597                 message = "writing ring-ref";
 598                 goto abort_transaction;
 599         }
 600         err = xenbus_printf(xbt, dev->nodename,
 601                             "event-channel", "%u", info->evtchn);
 602         if (err) {
 603                 message = "writing event-channel";
 604                 goto abort_transaction;
 605         }
 606
 607         err = xenbus_transaction_end(xbt, 0);
 608         if (err) {
 609                 if (err == -EAGAIN)
 610                         goto again;
 611                 xenbus_dev_fatal(dev, err, "completing transaction");
 612                 goto destroy_blkring;
 613         }
 614
 615         xenbus_switch_state(dev, XenbusStateInitialised);
 616
 617         return 0;
 618
 619  abort_transaction:
 620         xenbus_transaction_end(xbt, 1);
 621         if (message)
 622                 xenbus_dev_fatal(dev, err, "%s", message);
 623  destroy_blkring:
 624         blkif_free(info, 0);
 625  out:
 626         return err;
 627 }
 628
 629
 630 /**
 631  * Entry point to this code when a new device is created.  Allocate the basic
 632  * structures and the ring buffer for communication with the backend, and
 633  * inform the backend of the appropriate details for those.  Switch to
 634  * Initialised state.
 635  */
 636 static int blkfront_probe(struct xenbus_device *dev,
 637                           const struct xenbus_device_id *id)
 638 {
 639         int err, vdevice, i;
 640         struct blkfront_info *info;
 641
 642         /* FIXME: Use dynamic device id if this is not set. */
 643         err = xenbus_scanf(XBT_NIL, dev->nodename,
 644                            "virtual-device", "%i", &vdevice);
 645         if (err != 1) {
 646                 xenbus_dev_fatal(dev, err, "reading virtual-device");
 647                 return err;
 648         }
 649
 650         info = kzalloc(sizeof(*info), GFP_KERNEL);
 651         if (!info) {
 652                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 653                 return -ENOMEM;
 654         }
 655
 656         info->xbdev = dev;
 657         info->vdevice = vdevice;
 658         info->connected = BLKIF_STATE_DISCONNECTED;
 659         INIT_WORK(&info->work, blkif_restart_queue);
 660
 661         for (i = 0; i < BLK_RING_SIZE; i++)
 662                 info->shadow[i].req.id = i+1;
 663         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 664
 665         /* Front end dir is a number, which is used as the id. */
 666         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 667         dev->dev.driver_data = info;
 668
 669         err = talk_to_backend(dev, info);
 670         if (err) {
 671                 kfree(info);
 672                 dev->dev.driver_data = NULL;
 673                 return err;
 674         }
 675
 676         return 0;
 677 }
 678
 679
 680 static int blkif_recover(struct blkfront_info *info)
 681 {
 682         int i;
 683         struct blkif_request *req;
 684         struct blk_shadow *copy;
 685         int j;
 686
 687         /* Stage 1: Make a safe copy of the shadow state. */
 688         copy = kmalloc(sizeof(info->shadow), GFP_KERNEL);
 689         if (!copy)
 690                 return -ENOMEM;
 691         memcpy(copy, info->shadow, sizeof(info->shadow));
 692
 693         /* Stage 2: Set up free list. */
 694         memset(&info->shadow, 0, sizeof(info->shadow));
 695         for (i = 0; i < BLK_RING_SIZE; i++)
 696                 info->shadow[i].req.id = i+1;
 697         info->shadow_free = info->ring.req_prod_pvt;
 698         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 699
 700         /* Stage 3: Find pending requests and requeue them. */
 701         for (i = 0; i < BLK_RING_SIZE; i++) {
 702                 /* Not in use? */
 703                 if (copy[i].request == 0)
 704                         continue;
 705
 706                 /* Grab a request slot and copy shadow state into it. */
 707                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 708                 *req = copy[i].req;
 709
 710                 /* We get a new request id, and must reset the shadow state. */
 711                 req->id = get_id_from_freelist(info);
 712                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 713
 714                 /* Rewrite any grant references invalidated by susp/resume. */
 715                 for (j = 0; j < req->nr_segments; j++)
 716                         gnttab_grant_foreign_access_ref(
 717                                 req->seg[j].gref,
 718                                 info->xbdev->otherend_id,
 719                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 720                                 rq_data_dir(
 721                                         (struct request *)
 722                                         info->shadow[req->id].request));
 723                 info->shadow[req->id].req = *req;
 724
 725                 info->ring.req_prod_pvt++;
 726         }
 727
 728         kfree(copy);
 729
 730         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 731
 732         spin_lock_irq(&blkif_io_lock);
 733
 734         /* Now safe for us to use the shared ring */
 735         info->connected = BLKIF_STATE_CONNECTED;
 736
 737         /* Send off requeued requests */
 738         flush_requests(info);
 739
 740         /* Kick any other new requests queued since we resumed */
 741         kick_pending_request_queues(info);
 742
 743         spin_unlock_irq(&blkif_io_lock);
 744
 745         return 0;
 746 }
 747
 748 /**
 749  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 750  * driver restart.  We tear down our blkif structure and recreate it, but
 751  * leave the device-layer structures intact so that this is transparent to the
 752  * rest of the kernel.
 753  */
 754 static int blkfront_resume(struct xenbus_device *dev)
 755 {
 756         struct blkfront_info *info = dev->dev.driver_data;
 757         int err;
 758
 759         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 760
 761         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 762
 763         err = talk_to_backend(dev, info);
 764         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 765                 err = blkif_recover(info);
 766
 767         return err;
 768 }
 769
 770
 771 /*
 772  * Invoked when the backend is finally 'ready' (and has told produced
 773  * the details about the physical device - #sectors, size, etc).
 774  */
 775 static void blkfront_connect(struct blkfront_info *info)
 776 {
 777         unsigned long long sectors;
 778         unsigned long sector_size;
 779         unsigned int binfo;
 780         int err;
 781
 782         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 783             (info->connected == BLKIF_STATE_SUSPENDED) )
 784                 return;
 785
 786         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 787                 __func__, info->xbdev->otherend);
 788
 789         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 790                             "sectors", "%llu", &sectors,
 791                             "info", "%u", &binfo,
 792                             "sector-size", "%lu", &sector_size,
 793                             NULL);
 794         if (err) {
 795                 xenbus_dev_fatal(info->xbdev, err,
 796                                  "reading backend fields at %s",
 797                                  info->xbdev->otherend);
 798                 return;
 799         }
 800
 801         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 802                             "feature-barrier", "%lu", &info->feature_barrier,
 803                             NULL);
 804         if (err)
 805                 info->feature_barrier = 0;
 806
 807         err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
 808                                   sectors, info->vdevice,
 809                                   binfo, sector_size, info);
 810         if (err) {
 811                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 812                                  info->xbdev->otherend);
 813                 return;
 814         }
 815
 816         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 817
 818         /* Kick pending requests. */
 819         spin_lock_irq(&blkif_io_lock);
 820         info->connected = BLKIF_STATE_CONNECTED;
 821         kick_pending_request_queues(info);
 822         spin_unlock_irq(&blkif_io_lock);
 823
 824         add_disk(info->gd);
 825 }
 826
 827 /**
 828  * Handle the change of state of the backend to Closing.  We must delete our
 829  * device-layer structures now, to ensure that writes are flushed through to
 830  * the backend.  Once is this done, we can switch to Closed in
 831  * acknowledgement.
 832  */
 833 static void blkfront_closing(struct xenbus_device *dev)
 834 {
 835         struct blkfront_info *info = dev->dev.driver_data;
 836         unsigned long flags;
 837
 838         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 839
 840         if (info->rq == NULL)
 841                 goto out;
 842
 843         spin_lock_irqsave(&blkif_io_lock, flags);
 844
 845         del_gendisk(info->gd);
 846
 847         /* No more blkif_request(). */
 848         blk_stop_queue(info->rq);
 849
 850         /* No more gnttab callback work. */
 851         gnttab_cancel_free_callback(&info->callback);
 852         spin_unlock_irqrestore(&blkif_io_lock, flags);
 853
 854         /* Flush gnttab callback work. Must be done with no locks held. */
 855         flush_scheduled_work();
 856
 857         blk_cleanup_queue(info->rq);
 858         info->rq = NULL;
 859
 860  out:
 861         xenbus_frontend_closed(dev);
 862 }
 863
 864 /**
 865  * Callback received when the backend's state changes.
 866  */
 867 static void backend_changed(struct xenbus_device *dev,
 868                             enum xenbus_state backend_state)
 869 {
 870         struct blkfront_info *info = dev->dev.driver_data;
 871         struct block_device *bd;
 872
 873         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 874
 875         switch (backend_state) {
 876         case XenbusStateInitialising:
 877         case XenbusStateInitWait:
 878         case XenbusStateInitialised:
 879         case XenbusStateUnknown:
 880         case XenbusStateClosed:
 881                 break;
 882
 883         case XenbusStateConnected:
 884                 blkfront_connect(info);
 885                 break;
 886
 887         case XenbusStateClosing:
 888                 bd = bdget(info->dev);
 889                 if (bd == NULL)
 890                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 891
 892                 mutex_lock(&bd->bd_mutex);
 893                 if (info->users > 0)
 894                         xenbus_dev_error(dev, -EBUSY,
 895                                          "Device in use; refusing to close");
 896                 else
 897                         blkfront_closing(dev);
 898                 mutex_unlock(&bd->bd_mutex);
 899                 bdput(bd);
 900                 break;
 901         }
 902 }
 903
 904 static int blkfront_remove(struct xenbus_device *dev)
 905 {
 906         struct blkfront_info *info = dev->dev.driver_data;
 907
 908         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
 909
 910         blkif_free(info, 0);
 911
 912         kfree(info);
 913
 914         return 0;
 915 }
 916
 917 static int blkif_open(struct inode *inode, struct file *filep)
 918 {
 919         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 920         info->users++;
 921         return 0;
 922 }
 923
 924 static int blkif_release(struct inode *inode, struct file *filep)
 925 {
 926         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 927         info->users--;
 928         if (info->users == 0) {
 929                 /* Check whether we have been instructed to close.  We will
 930                    have ignored this request initially, as the device was
 931                    still mounted. */
 932                 struct xenbus_device *dev = info->xbdev;
 933                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
 934
 935                 if (state == XenbusStateClosing)
 936                         blkfront_closing(dev);
 937         }
 938         return 0;
 939 }
 940
 941 static struct block_device_operations xlvbd_block_fops =
 942 {
 943         .owner = THIS_MODULE,
 944         .open = blkif_open,
 945         .release = blkif_release,
 946 };
 947
 948
 949 static struct xenbus_device_id blkfront_ids[] = {
 950         { "vbd" },
 951         { "" }
 952 };
 953
 954 static struct xenbus_driver blkfront = {
 955         .name = "vbd",
 956         .owner = THIS_MODULE,
 957         .ids = blkfront_ids,
 958         .probe = blkfront_probe,
 959         .remove = blkfront_remove,
 960         .resume = blkfront_resume,
 961         .otherend_changed = backend_changed,
 962 };
 963
 964 static int __init xlblk_init(void)
 965 {
 966         if (!is_running_on_xen())
 967                 return -ENODEV;
 968
 969         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
 970                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
 971                        XENVBD_MAJOR, DEV_NAME);
 972                 return -ENODEV;
 973         }
 974
 975         return xenbus_register_frontend(&blkfront);
 976 }
 977 module_init(xlblk_init);
 978
 979
 980 static void xlblk_exit(void)
 981 {
 982         return xenbus_unregister_driver(&blkfront);
 983 }
 984 module_exit(xlblk_exit);
 985
 986 MODULE_DESCRIPTION("Xen virtual block device frontend");
 987 MODULE_LICENSE("GPL");
 988 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);