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