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 #include <asm/maddr.h>
  50
  51 #define BLKIF_STATE_DISCONNECTED 0
  52 #define BLKIF_STATE_CONNECTED    1
  53 #define BLKIF_STATE_SUSPENDED    2
  54
  55 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
  56     (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
  57 #define GRANT_INVALID_REF       0
  58
  59 static void connect(struct blkfront_info *);
  60 static void blkfront_closing(struct xenbus_device *);
  61 static int blkfront_remove(struct xenbus_device *);
  62 static int talk_to_backend(struct xenbus_device *, struct blkfront_info *);
  63 static int setup_blkring(struct xenbus_device *, struct blkfront_info *);
  64
  65 static void kick_pending_request_queues(struct blkfront_info *);
  66
  67 static irqreturn_t blkif_int(int irq, void *dev_id);
  68 static void blkif_restart_queue(struct work_struct *work);
  69 static void blkif_recover(struct blkfront_info *);
  70 static void blkif_completion(struct blk_shadow *);
  71 static void blkif_free(struct blkfront_info *, int);
  72
  73
  74 /**
  75  * Entry point to this code when a new device is created.  Allocate the basic
  76  * structures and the ring buffer for communication with the backend, and
  77  * inform the backend of the appropriate details for those.  Switch to
  78  * Initialised state.
  79  */
  80 static int blkfront_probe(struct xenbus_device *dev,
  81                           const struct xenbus_device_id *id)
  82 {
  83         int err, vdevice, i;
  84         struct blkfront_info *info;
  85
  86         /* FIXME: Use dynamic device id if this is not set. */
  87         err = xenbus_scanf(XBT_NIL, dev->nodename,
  88                            "virtual-device", "%i", &vdevice);
  89         if (err != 1) {
  90                 xenbus_dev_fatal(dev, err, "reading virtual-device");
  91                 return err;
  92         }
  93
  94         info = kzalloc(sizeof(*info), GFP_KERNEL);
  95         if (!info) {
  96                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
  97                 return -ENOMEM;
  98         }
  99
 100         info->xbdev = dev;
 101         info->vdevice = vdevice;
 102         info->connected = BLKIF_STATE_DISCONNECTED;
 103         INIT_WORK(&info->work, blkif_restart_queue);
 104
 105         for (i = 0; i < BLK_RING_SIZE; i++)
 106                 info->shadow[i].req.id = i+1;
 107         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 108
 109         /* Front end dir is a number, which is used as the id. */
 110         info->handle = simple_strtoul(strrchr(dev->nodename,'/')+1, NULL, 0);
 111         dev->dev.driver_data = info;
 112
 113         err = talk_to_backend(dev, info);
 114         if (err) {
 115                 kfree(info);
 116                 dev->dev.driver_data = NULL;
 117                 return err;
 118         }
 119
 120         return 0;
 121 }
 122
 123
 124 /**
 125  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 126  * driver restart.  We tear down our blkif structure and recreate it, but
 127  * leave the device-layer structures intact so that this is transparent to the
 128  * rest of the kernel.
 129  */
 130 static int blkfront_resume(struct xenbus_device *dev)
 131 {
 132         struct blkfront_info *info = dev->dev.driver_data;
 133         int err;
 134
 135         DPRINTK("blkfront_resume: %s\n", dev->nodename);
 136
 137         blkif_free(info, 1);
 138
 139         err = talk_to_backend(dev, info);
 140         if (!err)
 141                 blkif_recover(info);
 142
 143         return err;
 144 }
 145
 146
 147 /* Common code used when first setting up, and when resuming. */
 148 static int talk_to_backend(struct xenbus_device *dev,
 149                            struct blkfront_info *info)
 150 {
 151         const char *message = NULL;
 152         struct xenbus_transaction xbt;
 153         int err;
 154
 155         /* Create shared ring, alloc event channel. */
 156         err = setup_blkring(dev, info);
 157         if (err)
 158                 goto out;
 159
 160 again:
 161         err = xenbus_transaction_start(&xbt);
 162         if (err) {
 163                 xenbus_dev_fatal(dev, err, "starting transaction");
 164                 goto destroy_blkring;
 165         }
 166
 167         err = xenbus_printf(xbt, dev->nodename,
 168                             "ring-ref","%u", info->ring_ref);
 169         if (err) {
 170                 message = "writing ring-ref";
 171                 goto abort_transaction;
 172         }
 173         err = xenbus_printf(xbt, dev->nodename,
 174                             "event-channel", "%u", info->evtchn);
 175         if (err) {
 176                 message = "writing event-channel";
 177                 goto abort_transaction;
 178         }
 179
 180         err = xenbus_transaction_end(xbt, 0);
 181         if (err) {
 182                 if (err == -EAGAIN)
 183                         goto again;
 184                 xenbus_dev_fatal(dev, err, "completing transaction");
 185                 goto destroy_blkring;
 186         }
 187
 188         xenbus_switch_state(dev, XenbusStateInitialised);
 189
 190         return 0;
 191
 192  abort_transaction:
 193         xenbus_transaction_end(xbt, 1);
 194         if (message)
 195                 xenbus_dev_fatal(dev, err, "%s", message);
 196  destroy_blkring:
 197         blkif_free(info, 0);
 198  out:
 199         return err;
 200 }
 201
 202
 203 static int setup_blkring(struct xenbus_device *dev,
 204                          struct blkfront_info *info)
 205 {
 206         blkif_sring_t *sring;
 207         int err;
 208
 209         info->ring_ref = GRANT_INVALID_REF;
 210
 211         sring = (blkif_sring_t *)__get_free_page(GFP_KERNEL);
 212         if (!sring) {
 213                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 214                 return -ENOMEM;
 215         }
 216         SHARED_RING_INIT(sring);
 217         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 218
 219         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 220         if (err < 0) {
 221                 free_page((unsigned long)sring);
 222                 info->ring.sring = NULL;
 223                 goto fail;
 224         }
 225         info->ring_ref = err;
 226
 227         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 228         if (err)
 229                 goto fail;
 230
 231         err = bind_evtchn_to_irqhandler(
 232                 info->evtchn, blkif_int, SA_SAMPLE_RANDOM, "blkif", info);
 233         if (err <= 0) {
 234                 xenbus_dev_fatal(dev, err,
 235                                  "bind_evtchn_to_irqhandler failed");
 236                 goto fail;
 237         }
 238         info->irq = err;
 239
 240         return 0;
 241 fail:
 242         blkif_free(info, 0);
 243         return err;
 244 }
 245
 246
 247 /**
 248  * Callback received when the backend's state changes.
 249  */
 250 static void backend_changed(struct xenbus_device *dev,
 251                             enum xenbus_state backend_state)
 252 {
 253         struct blkfront_info *info = dev->dev.driver_data;
 254         struct block_device *bd;
 255
 256         DPRINTK("blkfront:backend_changed.\n");
 257
 258         switch (backend_state) {
 259         case XenbusStateInitialising:
 260         case XenbusStateInitWait:
 261         case XenbusStateInitialised:
 262         case XenbusStateUnknown:
 263         case XenbusStateClosed:
 264                 break;
 265
 266         case XenbusStateConnected:
 267                 connect(info);
 268                 break;
 269
 270         case XenbusStateClosing:
 271                 bd = bdget(info->dev);
 272                 if (bd == NULL)
 273                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 274
 275                 mutex_lock(&bd->bd_mutex);
 276                 if (info->users > 0)
 277                         xenbus_dev_error(dev, -EBUSY,
 278                                          "Device in use; refusing to close");
 279                 else
 280                         blkfront_closing(dev);
 281                 mutex_unlock(&bd->bd_mutex);
 282                 bdput(bd);
 283                 break;
 284         }
 285 }
 286
 287
 288 /* ** Connection ** */
 289
 290
 291 /*
 292  * Invoked when the backend is finally 'ready' (and has told produced
 293  * the details about the physical device - #sectors, size, etc).
 294  */
 295 static void connect(struct blkfront_info *info)
 296 {
 297         unsigned long long sectors;
 298         unsigned long sector_size;
 299         unsigned int binfo;
 300         int err;
 301
 302         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 303             (info->connected == BLKIF_STATE_SUSPENDED) )
 304                 return;
 305
 306         DPRINTK("blkfront.c:connect:%s.\n", info->xbdev->otherend);
 307
 308         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 309                             "sectors", "%llu", &sectors,
 310                             "info", "%u", &binfo,
 311                             "sector-size", "%lu", &sector_size,
 312                             NULL);
 313         if (err) {
 314                 xenbus_dev_fatal(info->xbdev, err,
 315                                  "reading backend fields at %s",
 316                                  info->xbdev->otherend);
 317                 return;
 318         }
 319
 320         err = xlvbd_add(sectors, info->vdevice, binfo, sector_size, info);
 321         if (err) {
 322                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 323                                  info->xbdev->otherend);
 324                 return;
 325         }
 326
 327         (void)xenbus_switch_state(info->xbdev, XenbusStateConnected);
 328
 329         /* Kick pending requests. */
 330         spin_lock_irq(&blkif_io_lock);
 331         info->connected = BLKIF_STATE_CONNECTED;
 332         kick_pending_request_queues(info);
 333         spin_unlock_irq(&blkif_io_lock);
 334
 335         add_disk(info->gd);
 336 }
 337
 338 /**
 339  * Handle the change of state of the backend to Closing.  We must delete our
 340  * device-layer structures now, to ensure that writes are flushed through to
 341  * the backend.  Once is this done, we can switch to Closed in
 342  * acknowledgement.
 343  */
 344 static void blkfront_closing(struct xenbus_device *dev)
 345 {
 346         struct blkfront_info *info = dev->dev.driver_data;
 347         unsigned long flags;
 348
 349         DPRINTK("blkfront_closing: %s removed\n", dev->nodename);
 350
 351         if (info->rq == NULL)
 352                 goto out;
 353
 354         spin_lock_irqsave(&blkif_io_lock, flags);
 355         /* No more blkif_request(). */
 356         blk_stop_queue(info->rq);
 357         /* No more gnttab callback work. */
 358         gnttab_cancel_free_callback(&info->callback);
 359         spin_unlock_irqrestore(&blkif_io_lock, flags);
 360
 361         /* Flush gnttab callback work. Must be done with no locks held. */
 362         flush_scheduled_work();
 363
 364         xlvbd_del(info);
 365
 366 out:
 367         xenbus_frontend_closed(dev);
 368 }
 369
 370
 371 static int blkfront_remove(struct xenbus_device *dev)
 372 {
 373         struct blkfront_info *info = dev->dev.driver_data;
 374
 375         DPRINTK("blkfront_remove: %s removed\n", dev->nodename);
 376
 377         blkif_free(info, 0);
 378
 379         kfree(info);
 380
 381         return 0;
 382 }
 383
 384
 385 static inline int GET_ID_FROM_FREELIST(
 386         struct blkfront_info *info)
 387 {
 388         unsigned long free = info->shadow_free;
 389         BUG_ON(free > BLK_RING_SIZE);
 390         info->shadow_free = info->shadow[free].req.id;
 391         info->shadow[free].req.id = 0x0fffffee; /* debug */
 392         return free;
 393 }
 394
 395 static inline void ADD_ID_TO_FREELIST(
 396         struct blkfront_info *info, unsigned long id)
 397 {
 398         info->shadow[id].req.id  = info->shadow_free;
 399         info->shadow[id].request = 0;
 400         info->shadow_free = id;
 401 }
 402
 403 static inline void flush_requests(struct blkfront_info *info)
 404 {
 405         int notify;
 406
 407         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 408
 409         if (notify)
 410                 notify_remote_via_irq(info->irq);
 411 }
 412
 413 static void kick_pending_request_queues(struct blkfront_info *info)
 414 {
 415         if (!RING_FULL(&info->ring)) {
 416                 /* Re-enable calldowns. */
 417                 blk_start_queue(info->rq);
 418                 /* Kick things off immediately. */
 419                 do_blkif_request(info->rq);
 420         }
 421 }
 422
 423 static void blkif_restart_queue(struct work_struct *work)
 424 {
 425         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 426
 427         spin_lock_irq(&blkif_io_lock);
 428         if (info->connected == BLKIF_STATE_CONNECTED)
 429                 kick_pending_request_queues(info);
 430         spin_unlock_irq(&blkif_io_lock);
 431 }
 432
 433 static void blkif_restart_queue_callback(void *arg)
 434 {
 435         struct blkfront_info *info = (struct blkfront_info *)arg;
 436         schedule_work(&info->work);
 437 }
 438
 439 int blkif_open(struct inode *inode, struct file *filep)
 440 {
 441         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 442         info->users++;
 443         return 0;
 444 }
 445
 446
 447 int blkif_release(struct inode *inode, struct file *filep)
 448 {
 449         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 450         info->users--;
 451         if (info->users == 0) {
 452                 /* Check whether we have been instructed to close.  We will
 453                    have ignored this request initially, as the device was
 454                    still mounted. */
 455                 struct xenbus_device * dev = info->xbdev;
 456                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
 457
 458                 if (state == XenbusStateClosing)
 459                         blkfront_closing(dev);
 460         }
 461         return 0;
 462 }
 463
 464
 465 int blkif_ioctl(struct inode *inode, struct file *filep,
 466                 unsigned command, unsigned long argument)
 467 {
 468         int i;
 469
 470         DPRINTK_IOCTL("command: 0x%x, argument: 0x%lx, dev: 0x%04x\n",
 471                       command, (long)argument, inode->i_rdev);
 472
 473         switch (command) {
 474         case CDROMMULTISESSION:
 475                 DPRINTK("FIXME: support multisession CDs later\n");
 476                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 477                         if (put_user(0, (char __user *)(argument + i)))
 478                                 return -EFAULT;
 479                 return 0;
 480
 481         default:
 482                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 483                   command);*/
 484                 return -EINVAL; /* same return as native Linux */
 485         }
 486
 487         return 0;
 488 }
 489
 490
 491 int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 492 {
 493         /* We don't have real geometry info, but let's at least return
 494            values consistent with the size of the device */
 495         sector_t nsect = get_capacity(bd->bd_disk);
 496         sector_t cylinders = nsect;
 497
 498         hg->heads = 0xff;
 499         hg->sectors = 0x3f;
 500         sector_div(cylinders, hg->heads * hg->sectors);
 501         hg->cylinders = cylinders;
 502         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 503                 hg->cylinders = 0xffff;
 504         return 0;
 505 }
 506
 507
 508 /*
 509  * blkif_queue_request
 510  *
 511  * request block io
 512  *
 513  * id: for guest use only.
 514  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 515  * buffer: buffer to read/write into. this should be a
 516  *   virtual address in the guest os.
 517  */
 518 static int blkif_queue_request(struct request *req)
 519 {
 520         struct blkfront_info *info = req->rq_disk->private_data;
 521         unsigned long buffer_mfn;
 522         blkif_request_t *ring_req;
 523         struct bio *bio;
 524         struct bio_vec *bvec;
 525         int idx;
 526         unsigned long id;
 527         unsigned int fsect, lsect;
 528         int ref;
 529         grant_ref_t gref_head;
 530
 531         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 532                 return 1;
 533
 534         if (gnttab_alloc_grant_references(
 535                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 536                 gnttab_request_free_callback(
 537                         &info->callback,
 538                         blkif_restart_queue_callback,
 539                         info,
 540                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 541                 return 1;
 542         }
 543
 544         /* Fill out a communications ring structure. */
 545         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 546         id = GET_ID_FROM_FREELIST(info);
 547         info->shadow[id].request = (unsigned long)req;
 548
 549         ring_req->id = id;
 550         ring_req->operation = rq_data_dir(req) ?
 551                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 552         ring_req->sector_number = (blkif_sector_t)req->sector;
 553         ring_req->handle = info->handle;
 554
 555         ring_req->nr_segments = 0;
 556         rq_for_each_bio (bio, req) {
 557                 bio_for_each_segment (bvec, bio, idx) {
 558                         BUG_ON(ring_req->nr_segments
 559                                == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 560                         buffer_mfn = page_to_phys(bvec->bv_page) >> PAGE_SHIFT;
 561                         fsect = bvec->bv_offset >> 9;
 562                         lsect = fsect + (bvec->bv_len >> 9) - 1;
 563                         /* install a grant reference. */
 564                         ref = gnttab_claim_grant_reference(&gref_head);
 565                         BUG_ON(ref == -ENOSPC);
 566
 567                         gnttab_grant_foreign_access_ref(
 568                                 ref,
 569                                 info->xbdev->otherend_id,
 570                                 buffer_mfn,
 571                                 rq_data_dir(req) );
 572
 573                         info->shadow[id].frame[ring_req->nr_segments] =
 574                                 mfn_to_pfn(buffer_mfn);
 575
 576                         ring_req->seg[ring_req->nr_segments] =
 577                                 (struct blkif_request_segment) {
 578                                         .gref       = ref,
 579                                         .first_sect = fsect,
 580                                         .last_sect  = lsect };
 581
 582                         ring_req->nr_segments++;
 583                 }
 584         }
 585
 586         info->ring.req_prod_pvt++;
 587
 588         /* Keep a private copy so we can reissue requests when recovering. */
 589         info->shadow[id].req = *ring_req;
 590
 591         gnttab_free_grant_references(gref_head);
 592
 593         return 0;
 594 }
 595
 596 /*
 597  * do_blkif_request
 598  *  read a block; request is in a request queue
 599  */
 600 void do_blkif_request(request_queue_t *rq)
 601 {
 602         struct blkfront_info *info = NULL;
 603         struct request *req;
 604         int queued;
 605
 606         DPRINTK("Entered do_blkif_request\n");
 607
 608         queued = 0;
 609
 610         while ((req = elv_next_request(rq)) != NULL) {
 611                 info = req->rq_disk->private_data;
 612                 if (!blk_fs_request(req)) {
 613                         end_request(req, 0);
 614                         continue;
 615                 }
 616
 617                 if (RING_FULL(&info->ring))
 618                         goto wait;
 619
 620                 DPRINTK("do_blk_req %p: cmd %p, sec %llx, "
 621                         "(%u/%li) buffer:%p [%s]\n",
 622                         req, req->cmd, (u64)req->sector, req->current_nr_sectors,
 623                         req->nr_sectors, req->buffer,
 624                         rq_data_dir(req) ? "write" : "read");
 625
 626
 627                 blkdev_dequeue_request(req);
 628                 if (blkif_queue_request(req)) {
 629                         blk_requeue_request(rq, req);
 630                 wait:
 631                         /* Avoid pointless unplugs. */
 632                         blk_stop_queue(rq);
 633                         break;
 634                 }
 635
 636                 queued++;
 637         }
 638
 639         if (queued != 0)
 640                 flush_requests(info);
 641 }
 642
 643
 644 static irqreturn_t blkif_int(int irq, void *dev_id)
 645 {
 646         struct request *req;
 647         blkif_response_t *bret;
 648         RING_IDX i, rp;
 649         unsigned long flags;
 650         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 651
 652         spin_lock_irqsave(&blkif_io_lock, flags);
 653
 654         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 655                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 656                 return IRQ_HANDLED;
 657         }
 658
 659  again:
 660         rp = info->ring.sring->rsp_prod;
 661         rmb(); /* Ensure we see queued responses up to 'rp'. */
 662
 663         for (i = info->ring.rsp_cons; i != rp; i++) {
 664                 unsigned long id;
 665                 int ret;
 666
 667                 bret = RING_GET_RESPONSE(&info->ring, i);
 668                 id   = bret->id;
 669                 req  = (struct request *)info->shadow[id].request;
 670
 671                 blkif_completion(&info->shadow[id]);
 672
 673                 ADD_ID_TO_FREELIST(info, id);
 674
 675                 switch (bret->operation) {
 676                 case BLKIF_OP_READ:
 677                 case BLKIF_OP_WRITE:
 678                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 679                                 DPRINTK("Bad return from blkdev data "
 680                                         "request: %x\n", bret->status);
 681
 682                         ret = end_that_request_first(
 683                                 req, (bret->status == BLKIF_RSP_OKAY),
 684                                 req->hard_nr_sectors);
 685                         BUG_ON(ret);
 686                         end_that_request_last(
 687                                 req, (bret->status == BLKIF_RSP_OKAY));
 688                         break;
 689                 default:
 690                         BUG();
 691                 }
 692         }
 693
 694         info->ring.rsp_cons = i;
 695
 696         if (i != info->ring.req_prod_pvt) {
 697                 int more_to_do;
 698                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 699                 if (more_to_do)
 700                         goto again;
 701         } else
 702                 info->ring.sring->rsp_event = i + 1;
 703
 704         kick_pending_request_queues(info);
 705
 706         spin_unlock_irqrestore(&blkif_io_lock, flags);
 707
 708         return IRQ_HANDLED;
 709 }
 710
 711 static void blkif_free(struct blkfront_info *info, int suspend)
 712 {
 713         /* Prevent new requests being issued until we fix things up. */
 714         spin_lock_irq(&blkif_io_lock);
 715         info->connected = suspend ?
 716                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 717         /* No more blkif_request(). */
 718         if (info->rq)
 719                 blk_stop_queue(info->rq);
 720         /* No more gnttab callback work. */
 721         gnttab_cancel_free_callback(&info->callback);
 722         spin_unlock_irq(&blkif_io_lock);
 723
 724         /* Flush gnttab callback work. Must be done with no locks held. */
 725         flush_scheduled_work();
 726
 727         /* Free resources associated with old device channel. */
 728         if (info->ring_ref != GRANT_INVALID_REF) {
 729                 gnttab_end_foreign_access(info->ring_ref, 0,
 730                                           (unsigned long)info->ring.sring);
 731                 info->ring_ref = GRANT_INVALID_REF;
 732                 info->ring.sring = NULL;
 733         }
 734         if (info->irq)
 735                 unbind_from_irqhandler(info->irq, info);
 736         info->evtchn = info->irq = 0;
 737
 738 }
 739
 740 static void blkif_completion(struct blk_shadow *s)
 741 {
 742         int i;
 743         for (i = 0; i < s->req.nr_segments; i++)
 744                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 745 }
 746
 747 static void blkif_recover(struct blkfront_info *info)
 748 {
 749         int i;
 750         blkif_request_t *req;
 751         struct blk_shadow *copy;
 752         int j;
 753
 754         /* Stage 1: Make a safe copy of the shadow state. */
 755         copy = kmalloc(sizeof(info->shadow), GFP_KERNEL | __GFP_NOFAIL);
 756         memcpy(copy, info->shadow, sizeof(info->shadow));
 757
 758         /* Stage 2: Set up free list. */
 759         memset(&info->shadow, 0, sizeof(info->shadow));
 760         for (i = 0; i < BLK_RING_SIZE; i++)
 761                 info->shadow[i].req.id = i+1;
 762         info->shadow_free = info->ring.req_prod_pvt;
 763         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 764
 765         /* Stage 3: Find pending requests and requeue them. */
 766         for (i = 0; i < BLK_RING_SIZE; i++) {
 767                 /* Not in use? */
 768                 if (copy[i].request == 0)
 769                         continue;
 770
 771                 /* Grab a request slot and copy shadow state into it. */
 772                 req = RING_GET_REQUEST(
 773                         &info->ring, info->ring.req_prod_pvt);
 774                 *req = copy[i].req;
 775
 776                 /* We get a new request id, and must reset the shadow state. */
 777                 req->id = GET_ID_FROM_FREELIST(info);
 778                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 779
 780                 /* Rewrite any grant references invalidated by susp/resume. */
 781                 for (j = 0; j < req->nr_segments; j++)
 782                         gnttab_grant_foreign_access_ref(
 783                                 req->seg[j].gref,
 784                                 info->xbdev->otherend_id,
 785                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 786                                 rq_data_dir(
 787                                         (struct request *)
 788                                         info->shadow[req->id].request));
 789                 info->shadow[req->id].req = *req;
 790
 791                 info->ring.req_prod_pvt++;
 792         }
 793
 794         kfree(copy);
 795
 796         (void)xenbus_switch_state(info->xbdev, XenbusStateConnected);
 797
 798         spin_lock_irq(&blkif_io_lock);
 799
 800         /* Now safe for us to use the shared ring */
 801         info->connected = BLKIF_STATE_CONNECTED;
 802
 803         /* Send off requeued requests */
 804         flush_requests(info);
 805
 806         /* Kick any other new requests queued since we resumed */
 807         kick_pending_request_queues(info);
 808
 809         spin_unlock_irq(&blkif_io_lock);
 810 }
 811
 812
 813 /* ** Driver Registration ** */
 814
 815
 816 static struct xenbus_device_id blkfront_ids[] = {
 817         { "vbd" },
 818         { "" }
 819 };
 820
 821
 822 static struct xenbus_driver blkfront = {
 823         .name = "vbd",
 824         .owner = THIS_MODULE,
 825         .ids = blkfront_ids,
 826         .probe = blkfront_probe,
 827         .remove = blkfront_remove,
 828         .resume = blkfront_resume,
 829         .otherend_changed = backend_changed,
 830 };
 831
 832
 833 static int __init xlblk_init(void)
 834 {
 835         if (!is_running_on_xen())
 836                 return -ENODEV;
 837
 838         return xenbus_register_frontend(&blkfront);
 839 }
 840 module_init(xlblk_init);
 841
 842
 843 static void xlblk_exit(void)
 844 {
 845         return xenbus_unregister_driver(&blkfront);
 846 }
 847 module_exit(xlblk_exit);
 848
 849 MODULE_LICENSE("Dual BSD/GPL");