drivers/xen/blkfront/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/version.h>
  39 #include "block.h"
  40 #include <linux/cdrom.h>
  41 #include <linux/sched.h>
  42 #include <linux/interrupt.h>
  43 #include <scsi/scsi.h>
  44 #include <xen/evtchn.h>
  45 #include <xen/xenbus.h>
  46 #include <xen/interface/grant_table.h>
  47 #include <xen/gnttab.h>
  48 #include <asm/hypervisor.h>
  49
  50 #define BLKIF_STATE_DISCONNECTED 0
  51 #define BLKIF_STATE_CONNECTED    1
  52 #define BLKIF_STATE_SUSPENDED    2
  53
  54 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
  55     (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
  56 #define GRANT_INVALID_REF       0
  57
  58 static void connect(struct blkfront_info *);
  59 static void blkfront_closing(struct xenbus_device *);
  60 static int blkfront_remove(struct xenbus_device *);
  61 static int talk_to_backend(struct xenbus_device *, struct blkfront_info *);
  62 static int setup_blkring(struct xenbus_device *, struct blkfront_info *);
  63
  64 static void kick_pending_request_queues(struct blkfront_info *);
  65
  66 static irqreturn_t blkif_int(int irq, void *dev_id, struct pt_regs *ptregs);
  67 static void blkif_restart_queue(void *arg);
  68 static void blkif_recover(struct blkfront_info *);
  69 static void blkif_completion(struct blk_shadow *);
  70 static void blkif_free(struct blkfront_info *, int);
  71
  72
  73 /**
  74  * Entry point to this code when a new device is created.  Allocate the basic
  75  * structures and the ring buffer for communication with the backend, and
  76  * inform the backend of the appropriate details for those.  Switch to
  77  * Initialised state.
  78  */
  79 static int blkfront_probe(struct xenbus_device *dev,
  80                           const struct xenbus_device_id *id)
  81 {
  82         int err, vdevice, i;
  83         struct blkfront_info *info;
  84
  85         /* FIXME: Use dynamic device id if this is not set. */
  86         err = xenbus_scanf(XBT_NIL, dev->nodename,
  87                            "virtual-device", "%i", &vdevice);
  88         if (err != 1) {
  89                 xenbus_dev_fatal(dev, err, "reading virtual-device");
  90                 return err;
  91         }
  92
  93         info = kzalloc(sizeof(*info), GFP_KERNEL);
  94         if (!info) {
  95                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
  96                 return -ENOMEM;
  97         }
  98
  99         info->xbdev = dev;
 100         info->vdevice = vdevice;
 101         info->connected = BLKIF_STATE_DISCONNECTED;
 102         INIT_WORK(&info->work, blkif_restart_queue, (void *)info);
 103
 104         for (i = 0; i < BLK_RING_SIZE; i++)
 105                 info->shadow[i].req.id = i+1;
 106         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 107
 108         /* Front end dir is a number, which is used as the id. */
 109         info->handle = simple_strtoul(strrchr(dev->nodename,'/')+1, NULL, 0);
 110         dev->dev.driver_data = info;
 111
 112         err = talk_to_backend(dev, info);
 113         if (err) {
 114                 kfree(info);
 115                 dev->dev.driver_data = NULL;
 116                 return err;
 117         }
 118
 119         return 0;
 120 }
 121
 122
 123 /**
 124  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 125  * driver restart.  We tear down our blkif structure and recreate it, but
 126  * leave the device-layer structures intact so that this is transparent to the
 127  * rest of the kernel.
 128  */
 129 static int blkfront_resume(struct xenbus_device *dev)
 130 {
 131         struct blkfront_info *info = dev->dev.driver_data;
 132         int err;
 133
 134         DPRINTK("blkfront_resume: %s\n", dev->nodename);
 135
 136         blkif_free(info, 1);
 137
 138         err = talk_to_backend(dev, info);
 139         if (!err)
 140                 blkif_recover(info);
 141
 142         return err;
 143 }
 144
 145
 146 /* Common code used when first setting up, and when resuming. */
 147 static int talk_to_backend(struct xenbus_device *dev,
 148                            struct blkfront_info *info)
 149 {
 150         const char *message = NULL;
 151         struct xenbus_transaction xbt;
 152         int err;
 153
 154         /* Create shared ring, alloc event channel. */
 155         err = setup_blkring(dev, info);
 156         if (err)
 157                 goto out;
 158
 159 again:
 160         err = xenbus_transaction_start(&xbt);
 161         if (err) {
 162                 xenbus_dev_fatal(dev, err, "starting transaction");
 163                 goto destroy_blkring;
 164         }
 165
 166         err = xenbus_printf(xbt, dev->nodename,
 167                             "ring-ref","%u", info->ring_ref);
 168         if (err) {
 169                 message = "writing ring-ref";
 170                 goto abort_transaction;
 171         }
 172         err = xenbus_printf(xbt, dev->nodename,
 173                             "event-channel", "%u", info->evtchn);
 174         if (err) {
 175                 message = "writing event-channel";
 176                 goto abort_transaction;
 177         }
 178
 179         err = xenbus_transaction_end(xbt, 0);
 180         if (err) {
 181                 if (err == -EAGAIN)
 182                         goto again;
 183                 xenbus_dev_fatal(dev, err, "completing transaction");
 184                 goto destroy_blkring;
 185         }
 186
 187         xenbus_switch_state(dev, XenbusStateInitialised);
 188
 189         return 0;
 190
 191  abort_transaction:
 192         xenbus_transaction_end(xbt, 1);
 193         if (message)
 194                 xenbus_dev_fatal(dev, err, "%s", message);
 195  destroy_blkring:
 196         blkif_free(info, 0);
 197  out:
 198         return err;
 199 }
 200
 201
 202 static int setup_blkring(struct xenbus_device *dev,
 203                          struct blkfront_info *info)
 204 {
 205         blkif_sring_t *sring;
 206         int err;
 207
 208         info->ring_ref = GRANT_INVALID_REF;
 209
 210         sring = (blkif_sring_t *)__get_free_page(GFP_KERNEL);
 211         if (!sring) {
 212                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 213                 return -ENOMEM;
 214         }
 215         SHARED_RING_INIT(sring);
 216         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 217
 218         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 219         if (err < 0) {
 220                 free_page((unsigned long)sring);
 221                 info->ring.sring = NULL;
 222                 goto fail;
 223         }
 224         info->ring_ref = err;
 225
 226         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 227         if (err)
 228                 goto fail;
 229
 230         err = bind_evtchn_to_irqhandler(
 231                 info->evtchn, blkif_int, SA_SAMPLE_RANDOM, "blkif", info);
 232         if (err <= 0) {
 233                 xenbus_dev_fatal(dev, err,
 234                                  "bind_evtchn_to_irqhandler failed");
 235                 goto fail;
 236         }
 237         info->irq = err;
 238
 239         return 0;
 240 fail:
 241         blkif_free(info, 0);
 242         return err;
 243 }
 244
 245
 246 /**
 247  * Callback received when the backend's state changes.
 248  */
 249 static void backend_changed(struct xenbus_device *dev,
 250                             enum xenbus_state backend_state)
 251 {
 252         struct blkfront_info *info = dev->dev.driver_data;
 253         struct block_device *bd;
 254
 255         DPRINTK("blkfront:backend_changed.\n");
 256
 257         switch (backend_state) {
 258         case XenbusStateUnknown:
 259         case XenbusStateInitialising:
 260         case XenbusStateInitWait:
 261         case XenbusStateInitialised:
 262         case XenbusStateClosed:
 263                 break;
 264
 265         case XenbusStateConnected:
 266                 connect(info);
 267                 break;
 268
 269         case XenbusStateClosing:
 270                 bd = bdget(info->dev);
 271                 if (bd == NULL)
 272                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 273
 274                 mutex_lock(&bd->bd_mutex);
 275                 if (info->users > 0)
 276                         xenbus_dev_error(dev, -EBUSY,
 277                                          "Device in use; refusing to close");
 278                 else
 279                         blkfront_closing(dev);
 280                 mutex_unlock(&bd->bd_mutex);
 281                 bdput(bd);
 282                 break;
 283         }
 284 }
 285
 286
 287 /* ** Connection ** */
 288
 289
 290 /*
 291  * Invoked when the backend is finally 'ready' (and has told produced
 292  * the details about the physical device - #sectors, size, etc).
 293  */
 294 static void connect(struct blkfront_info *info)
 295 {
 296         unsigned long sectors, sector_size;
 297         unsigned int binfo;
 298         int err;
 299
 300         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 301             (info->connected == BLKIF_STATE_SUSPENDED) )
 302                 return;
 303
 304         DPRINTK("blkfront.c:connect:%s.\n", info->xbdev->otherend);
 305
 306         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 307                             "sectors", "%lu", &sectors,
 308                             "info", "%u", &binfo,
 309                             "sector-size", "%lu", &sector_size,
 310                             NULL);
 311         if (err) {
 312                 xenbus_dev_fatal(info->xbdev, err,
 313                                  "reading backend fields at %s",
 314                                  info->xbdev->otherend);
 315                 return;
 316         }
 317
 318         err = xlvbd_add(sectors, info->vdevice, binfo, sector_size, info);
 319         if (err) {
 320                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 321                                  info->xbdev->otherend);
 322                 return;
 323         }
 324
 325         (void)xenbus_switch_state(info->xbdev, XenbusStateConnected);
 326
 327         /* Kick pending requests. */
 328         spin_lock_irq(&blkif_io_lock);
 329         info->connected = BLKIF_STATE_CONNECTED;
 330         kick_pending_request_queues(info);
 331         spin_unlock_irq(&blkif_io_lock);
 332
 333         add_disk(info->gd);
 334 }
 335
 336 /**
 337  * Handle the change of state of the backend to Closing.  We must delete our
 338  * device-layer structures now, to ensure that writes are flushed through to
 339  * the backend.  Once is this done, we can switch to Closed in
 340  * acknowledgement.
 341  */
 342 static void blkfront_closing(struct xenbus_device *dev)
 343 {
 344         struct blkfront_info *info = dev->dev.driver_data;
 345         unsigned long flags;
 346
 347         DPRINTK("blkfront_closing: %s removed\n", dev->nodename);
 348
 349         if (info->rq == NULL)
 350                 return;
 351
 352         spin_lock_irqsave(&blkif_io_lock, flags);
 353         /* No more blkif_request(). */
 354         blk_stop_queue(info->rq);
 355         /* No more gnttab callback work. */
 356         gnttab_cancel_free_callback(&info->callback);
 357         flush_scheduled_work();
 358         spin_unlock_irqrestore(&blkif_io_lock, flags);
 359
 360         xlvbd_del(info);
 361
 362         xenbus_switch_state(dev, XenbusStateClosed);
 363 }
 364
 365
 366 static int blkfront_remove(struct xenbus_device *dev)
 367 {
 368         struct blkfront_info *info = dev->dev.driver_data;
 369
 370         DPRINTK("blkfront_remove: %s removed\n", dev->nodename);
 371
 372         blkif_free(info, 0);
 373
 374         kfree(info);
 375
 376         return 0;
 377 }
 378
 379
 380 static inline int GET_ID_FROM_FREELIST(
 381         struct blkfront_info *info)
 382 {
 383         unsigned long free = info->shadow_free;
 384         BUG_ON(free > BLK_RING_SIZE);
 385         info->shadow_free = info->shadow[free].req.id;
 386         info->shadow[free].req.id = 0x0fffffee; /* debug */
 387         return free;
 388 }
 389
 390 static inline void ADD_ID_TO_FREELIST(
 391         struct blkfront_info *info, unsigned long id)
 392 {
 393         info->shadow[id].req.id  = info->shadow_free;
 394         info->shadow[id].request = 0;
 395         info->shadow_free = id;
 396 }
 397
 398 static inline void flush_requests(struct blkfront_info *info)
 399 {
 400         int notify;
 401
 402         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 403
 404         if (notify)
 405                 notify_remote_via_irq(info->irq);
 406 }
 407
 408 static void kick_pending_request_queues(struct blkfront_info *info)
 409 {
 410         if (!RING_FULL(&info->ring)) {
 411                 /* Re-enable calldowns. */
 412                 blk_start_queue(info->rq);
 413                 /* Kick things off immediately. */
 414                 do_blkif_request(info->rq);
 415         }
 416 }
 417
 418 static void blkif_restart_queue(void *arg)
 419 {
 420         struct blkfront_info *info = (struct blkfront_info *)arg;
 421         spin_lock_irq(&blkif_io_lock);
 422         if (info->connected == BLKIF_STATE_CONNECTED)
 423                 kick_pending_request_queues(info);
 424         spin_unlock_irq(&blkif_io_lock);
 425 }
 426
 427 static void blkif_restart_queue_callback(void *arg)
 428 {
 429         struct blkfront_info *info = (struct blkfront_info *)arg;
 430         schedule_work(&info->work);
 431 }
 432
 433 int blkif_open(struct inode *inode, struct file *filep)
 434 {
 435         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 436         info->users++;
 437         return 0;
 438 }
 439
 440
 441 int blkif_release(struct inode *inode, struct file *filep)
 442 {
 443         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 444         info->users--;
 445         if (info->users == 0) {
 446                 /* Check whether we have been instructed to close.  We will
 447                    have ignored this request initially, as the device was
 448                    still mounted. */
 449                 struct xenbus_device * dev = info->xbdev;
 450                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
 451
 452                 if (state == XenbusStateClosing)
 453                         blkfront_closing(dev);
 454         }
 455         return 0;
 456 }
 457
 458
 459 int blkif_ioctl(struct inode *inode, struct file *filep,
 460                 unsigned command, unsigned long argument)
 461 {
 462         int i;
 463
 464         DPRINTK_IOCTL("command: 0x%x, argument: 0x%lx, dev: 0x%04x\n",
 465                       command, (long)argument, inode->i_rdev);
 466
 467         switch (command) {
 468         case CDROMMULTISESSION:
 469                 DPRINTK("FIXME: support multisession CDs later\n");
 470                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 471                         if (put_user(0, (char __user *)(argument + i)))
 472                                 return -EFAULT;
 473                 return 0;
 474
 475         default:
 476                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 477                   command);*/
 478                 return -EINVAL; /* same return as native Linux */
 479         }
 480
 481         return 0;
 482 }
 483
 484
 485 int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 486 {
 487         /* We don't have real geometry info, but let's at least return
 488            values consistent with the size of the device */
 489         sector_t nsect = get_capacity(bd->bd_disk);
 490         sector_t cylinders = nsect;
 491
 492         hg->heads = 0xff;
 493         hg->sectors = 0x3f;
 494         sector_div(cylinders, hg->heads * hg->sectors);
 495         hg->cylinders = cylinders;
 496         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 497                 hg->cylinders = 0xffff;
 498         return 0;
 499 }
 500
 501
 502 /*
 503  * blkif_queue_request
 504  *
 505  * request block io
 506  *
 507  * id: for guest use only.
 508  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 509  * buffer: buffer to read/write into. this should be a
 510  *   virtual address in the guest os.
 511  */
 512 static int blkif_queue_request(struct request *req)
 513 {
 514         struct blkfront_info *info = req->rq_disk->private_data;
 515         unsigned long buffer_mfn;
 516         blkif_request_t *ring_req;
 517         struct bio *bio;
 518         struct bio_vec *bvec;
 519         int idx;
 520         unsigned long id;
 521         unsigned int fsect, lsect;
 522         int ref;
 523         grant_ref_t gref_head;
 524
 525         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 526                 return 1;
 527
 528         if (gnttab_alloc_grant_references(
 529                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 530                 gnttab_request_free_callback(
 531                         &info->callback,
 532                         blkif_restart_queue_callback,
 533                         info,
 534                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 535                 return 1;
 536         }
 537
 538         /* Fill out a communications ring structure. */
 539         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 540         id = GET_ID_FROM_FREELIST(info);
 541         info->shadow[id].request = (unsigned long)req;
 542
 543         ring_req->id = id;
 544         ring_req->operation = rq_data_dir(req) ?
 545                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 546         ring_req->sector_number = (blkif_sector_t)req->sector;
 547         ring_req->handle = info->handle;
 548
 549         ring_req->nr_segments = 0;
 550         rq_for_each_bio (bio, req) {
 551                 bio_for_each_segment (bvec, bio, idx) {
 552                         BUG_ON(ring_req->nr_segments
 553                                == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 554                         buffer_mfn = page_to_phys(bvec->bv_page) >> PAGE_SHIFT;
 555                         fsect = bvec->bv_offset >> 9;
 556                         lsect = fsect + (bvec->bv_len >> 9) - 1;
 557                         /* install a grant reference. */
 558                         ref = gnttab_claim_grant_reference(&gref_head);
 559                         BUG_ON(ref == -ENOSPC);
 560
 561                         gnttab_grant_foreign_access_ref(
 562                                 ref,
 563                                 info->xbdev->otherend_id,
 564                                 buffer_mfn,
 565                                 rq_data_dir(req) );
 566
 567                         info->shadow[id].frame[ring_req->nr_segments] =
 568                                 mfn_to_pfn(buffer_mfn);
 569
 570                         ring_req->seg[ring_req->nr_segments] =
 571                                 (struct blkif_request_segment) {
 572                                         .gref       = ref,
 573                                         .first_sect = fsect,
 574                                         .last_sect  = lsect };
 575
 576                         ring_req->nr_segments++;
 577                 }
 578         }
 579
 580         info->ring.req_prod_pvt++;
 581
 582         /* Keep a private copy so we can reissue requests when recovering. */
 583         info->shadow[id].req = *ring_req;
 584
 585         gnttab_free_grant_references(gref_head);
 586
 587         return 0;
 588 }
 589
 590 /*
 591  * do_blkif_request
 592  *  read a block; request is in a request queue
 593  */
 594 void do_blkif_request(request_queue_t *rq)
 595 {
 596         struct blkfront_info *info = NULL;
 597         struct request *req;
 598         int queued;
 599
 600         DPRINTK("Entered do_blkif_request\n");
 601
 602         queued = 0;
 603
 604         while ((req = elv_next_request(rq)) != NULL) {
 605                 info = req->rq_disk->private_data;
 606                 if (!blk_fs_request(req)) {
 607                         end_request(req, 0);
 608                         continue;
 609                 }
 610
 611                 if (RING_FULL(&info->ring))
 612                         goto wait;
 613
 614                 DPRINTK("do_blk_req %p: cmd %p, sec %lx, "
 615                         "(%u/%li) buffer:%p [%s]\n",
 616                         req, req->cmd, req->sector, req->current_nr_sectors,
 617                         req->nr_sectors, req->buffer,
 618                         rq_data_dir(req) ? "write" : "read");
 619
 620
 621                 blkdev_dequeue_request(req);
 622                 if (blkif_queue_request(req)) {
 623                         blk_requeue_request(rq, req);
 624                 wait:
 625                         /* Avoid pointless unplugs. */
 626                         blk_stop_queue(rq);
 627                         break;
 628                 }
 629
 630                 queued++;
 631         }
 632
 633         if (queued != 0)
 634                 flush_requests(info);
 635 }
 636
 637
 638 static irqreturn_t blkif_int(int irq, void *dev_id, struct pt_regs *ptregs)
 639 {
 640         struct request *req;
 641         blkif_response_t *bret;
 642         RING_IDX i, rp;
 643         unsigned long flags;
 644         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 645
 646         spin_lock_irqsave(&blkif_io_lock, flags);
 647
 648         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 649                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 650                 return IRQ_HANDLED;
 651         }
 652
 653  again:
 654         rp = info->ring.sring->rsp_prod;
 655         rmb(); /* Ensure we see queued responses up to 'rp'. */
 656
 657         for (i = info->ring.rsp_cons; i != rp; i++) {
 658                 unsigned long id;
 659                 int ret;
 660
 661                 bret = RING_GET_RESPONSE(&info->ring, i);
 662                 id   = bret->id;
 663                 req  = (struct request *)info->shadow[id].request;
 664
 665                 blkif_completion(&info->shadow[id]);
 666
 667                 ADD_ID_TO_FREELIST(info, id);
 668
 669                 switch (bret->operation) {
 670                 case BLKIF_OP_READ:
 671                 case BLKIF_OP_WRITE:
 672                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 673                                 DPRINTK("Bad return from blkdev data "
 674                                         "request: %x\n", bret->status);
 675
 676                         ret = end_that_request_first(
 677                                 req, (bret->status == BLKIF_RSP_OKAY),
 678                                 req->hard_nr_sectors);
 679                         BUG_ON(ret);
 680                         end_that_request_last(
 681                                 req, (bret->status == BLKIF_RSP_OKAY));
 682                         break;
 683                 default:
 684                         BUG();
 685                 }
 686         }
 687
 688         info->ring.rsp_cons = i;
 689
 690         if (i != info->ring.req_prod_pvt) {
 691                 int more_to_do;
 692                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 693                 if (more_to_do)
 694                         goto again;
 695         } else
 696                 info->ring.sring->rsp_event = i + 1;
 697
 698         kick_pending_request_queues(info);
 699
 700         spin_unlock_irqrestore(&blkif_io_lock, flags);
 701
 702         return IRQ_HANDLED;
 703 }
 704
 705 static void blkif_free(struct blkfront_info *info, int suspend)
 706 {
 707         /* Prevent new requests being issued until we fix things up. */
 708         spin_lock_irq(&blkif_io_lock);
 709         info->connected = suspend ?
 710                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 711         /* No more blkif_request(). */
 712         if (info->rq)
 713                 blk_stop_queue(info->rq);
 714         /* No more gnttab callback work. */
 715         gnttab_cancel_free_callback(&info->callback);
 716         flush_scheduled_work();
 717         spin_unlock_irq(&blkif_io_lock);
 718
 719         /* Free resources associated with old device channel. */
 720         if (info->ring_ref != GRANT_INVALID_REF) {
 721                 gnttab_end_foreign_access(info->ring_ref, 0,
 722                                           (unsigned long)info->ring.sring);
 723                 info->ring_ref = GRANT_INVALID_REF;
 724                 info->ring.sring = NULL;
 725         }
 726         if (info->irq)
 727                 unbind_from_irqhandler(info->irq, info);
 728         info->evtchn = info->irq = 0;
 729
 730 }
 731
 732 static void blkif_completion(struct blk_shadow *s)
 733 {
 734         int i;
 735         for (i = 0; i < s->req.nr_segments; i++)
 736                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 737 }
 738
 739 static void blkif_recover(struct blkfront_info *info)
 740 {
 741         int i;
 742         blkif_request_t *req;
 743         struct blk_shadow *copy;
 744         int j;
 745
 746         /* Stage 1: Make a safe copy of the shadow state. */
 747         copy = kmalloc(sizeof(info->shadow), GFP_KERNEL | __GFP_NOFAIL);
 748         memcpy(copy, info->shadow, sizeof(info->shadow));
 749
 750         /* Stage 2: Set up free list. */
 751         memset(&info->shadow, 0, sizeof(info->shadow));
 752         for (i = 0; i < BLK_RING_SIZE; i++)
 753                 info->shadow[i].req.id = i+1;
 754         info->shadow_free = info->ring.req_prod_pvt;
 755         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 756
 757         /* Stage 3: Find pending requests and requeue them. */
 758         for (i = 0; i < BLK_RING_SIZE; i++) {
 759                 /* Not in use? */
 760                 if (copy[i].request == 0)
 761                         continue;
 762
 763                 /* Grab a request slot and copy shadow state into it. */
 764                 req = RING_GET_REQUEST(
 765                         &info->ring, info->ring.req_prod_pvt);
 766                 *req = copy[i].req;
 767
 768                 /* We get a new request id, and must reset the shadow state. */
 769                 req->id = GET_ID_FROM_FREELIST(info);
 770                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 771
 772                 /* Rewrite any grant references invalidated by susp/resume. */
 773                 for (j = 0; j < req->nr_segments; j++)
 774                         gnttab_grant_foreign_access_ref(
 775                                 req->seg[j].gref,
 776                                 info->xbdev->otherend_id,
 777                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 778                                 rq_data_dir(
 779                                         (struct request *)
 780                                         info->shadow[req->id].request));
 781                 info->shadow[req->id].req = *req;
 782
 783                 info->ring.req_prod_pvt++;
 784         }
 785
 786         kfree(copy);
 787
 788         (void)xenbus_switch_state(info->xbdev, XenbusStateConnected);
 789
 790         spin_lock_irq(&blkif_io_lock);
 791
 792         /* Now safe for us to use the shared ring */
 793         info->connected = BLKIF_STATE_CONNECTED;
 794
 795         /* Send off requeued requests */
 796         flush_requests(info);
 797
 798         /* Kick any other new requests queued since we resumed */
 799         kick_pending_request_queues(info);
 800
 801         spin_unlock_irq(&blkif_io_lock);
 802 }
 803
 804
 805 /* ** Driver Registration ** */
 806
 807
 808 static struct xenbus_device_id blkfront_ids[] = {
 809         { "vbd" },
 810         { "" }
 811 };
 812
 813
 814 static struct xenbus_driver blkfront = {
 815         .name = "vbd",
 816         .owner = THIS_MODULE,
 817         .ids = blkfront_ids,
 818         .probe = blkfront_probe,
 819         .remove = blkfront_remove,
 820         .resume = blkfront_resume,
 821         .otherend_changed = backend_changed,
 822 };
 823
 824
 825 static int __init xlblk_init(void)
 826 {
 827         if (!is_running_on_xen())
 828                 return -ENODEV;
 829
 830         return xenbus_register_frontend(&blkfront);
 831 }
 832 module_init(xlblk_init);
 833
 834
 835 static void xlblk_exit(void)
 836 {
 837         return xenbus_unregister_driver(&blkfront);
 838 }
 839 module_exit(xlblk_exit);
 840
 841 MODULE_LICENSE("Dual BSD/GPL");