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 <<<<<<< HEAD:drivers/block/xen-blkfront.c
  41 =======
  42 #include <linux/hdreg.h>
  43 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:drivers/block/xen-blkfront.c
  44 #include <linux/module.h>
  45
  46 #include <xen/xenbus.h>
  47 #include <xen/grant_table.h>
  48 #include <xen/events.h>
  49 #include <xen/page.h>
  50
  51 #include <xen/interface/grant_table.h>
  52 #include <xen/interface/io/blkif.h>
  53
  54 #include <asm/xen/hypervisor.h>
  55
  56 enum blkif_state {
  57         BLKIF_STATE_DISCONNECTED,
  58         BLKIF_STATE_CONNECTED,
  59         BLKIF_STATE_SUSPENDED,
  60 };
  61
  62 struct blk_shadow {
  63         struct blkif_request req;
  64         unsigned long request;
  65         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  66 };
  67
  68 static struct block_device_operations xlvbd_block_fops;
  69
  70 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  71
  72 /*
  73  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  74  * hang in private_data off the gendisk structure. We may end up
  75  * putting all kinds of interesting stuff here :-)
  76  */
  77 struct blkfront_info
  78 {
  79         struct xenbus_device *xbdev;
  80         dev_t dev;
  81         struct gendisk *gd;
  82         int vdevice;
  83         blkif_vdev_t handle;
  84         enum blkif_state connected;
  85         int ring_ref;
  86         struct blkif_front_ring ring;
  87         unsigned int evtchn, irq;
  88         struct request_queue *rq;
  89         struct work_struct work;
  90         struct gnttab_free_callback callback;
  91         struct blk_shadow shadow[BLK_RING_SIZE];
  92         unsigned long shadow_free;
  93         int feature_barrier;
  94
  95         /**
  96          * The number of people holding this device open.  We won't allow a
  97          * hot-unplug unless this is 0.
  98          */
  99         int users;
 100 };
 101
 102 static DEFINE_SPINLOCK(blkif_io_lock);
 103
 104 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 105         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 106 #define GRANT_INVALID_REF       0
 107
 108 #define PARTS_PER_DISK          16
 109
 110 #define BLKIF_MAJOR(dev) ((dev)>>8)
 111 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 112
 113 #define DEV_NAME        "xvd"   /* name in /dev */
 114
 115 /* Information about our VBDs. */
 116 #define MAX_VBDS 64
 117 static LIST_HEAD(vbds_list);
 118
 119 static int get_id_from_freelist(struct blkfront_info *info)
 120 {
 121         unsigned long free = info->shadow_free;
 122         BUG_ON(free > BLK_RING_SIZE);
 123         info->shadow_free = info->shadow[free].req.id;
 124         info->shadow[free].req.id = 0x0fffffee; /* debug */
 125         return free;
 126 }
 127
 128 static void add_id_to_freelist(struct blkfront_info *info,
 129                                unsigned long id)
 130 {
 131         info->shadow[id].req.id  = info->shadow_free;
 132         info->shadow[id].request = 0;
 133         info->shadow_free = id;
 134 }
 135
 136 static void blkif_restart_queue_callback(void *arg)
 137 {
 138         struct blkfront_info *info = (struct blkfront_info *)arg;
 139         schedule_work(&info->work);
 140 }
 141
 142 <<<<<<< HEAD:drivers/block/xen-blkfront.c
 143 =======
 144 int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 145 {
 146         /* We don't have real geometry info, but let's at least return
 147            values consistent with the size of the device */
 148         sector_t nsect = get_capacity(bd->bd_disk);
 149         sector_t cylinders = nsect;
 150
 151         hg->heads = 0xff;
 152         hg->sectors = 0x3f;
 153         sector_div(cylinders, hg->heads * hg->sectors);
 154         hg->cylinders = cylinders;
 155         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 156                 hg->cylinders = 0xffff;
 157         return 0;
 158 }
 159
 160 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:drivers/block/xen-blkfront.c
 161 /*
 162  * blkif_queue_request
 163  *
 164  * request block io
 165  *
 166  * id: for guest use only.
 167  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 168  * buffer: buffer to read/write into. this should be a
 169  *   virtual address in the guest os.
 170  */
 171 static int blkif_queue_request(struct request *req)
 172 {
 173         struct blkfront_info *info = req->rq_disk->private_data;
 174         unsigned long buffer_mfn;
 175         struct blkif_request *ring_req;
 176         struct req_iterator iter;
 177         struct bio_vec *bvec;
 178         unsigned long id;
 179         unsigned int fsect, lsect;
 180         int ref;
 181         grant_ref_t gref_head;
 182
 183         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 184                 return 1;
 185
 186         if (gnttab_alloc_grant_references(
 187                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 188                 gnttab_request_free_callback(
 189                         &info->callback,
 190                         blkif_restart_queue_callback,
 191                         info,
 192                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 193                 return 1;
 194         }
 195
 196         /* Fill out a communications ring structure. */
 197         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 198         id = get_id_from_freelist(info);
 199         info->shadow[id].request = (unsigned long)req;
 200
 201         ring_req->id = id;
 202         ring_req->sector_number = (blkif_sector_t)req->sector;
 203         ring_req->handle = info->handle;
 204
 205         ring_req->operation = rq_data_dir(req) ?
 206                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 207         if (blk_barrier_rq(req))
 208                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 209
 210         ring_req->nr_segments = 0;
 211         rq_for_each_segment(bvec, req, iter) {
 212                 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 213                 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
 214                 fsect = bvec->bv_offset >> 9;
 215                 lsect = fsect + (bvec->bv_len >> 9) - 1;
 216                 /* install a grant reference. */
 217                 ref = gnttab_claim_grant_reference(&gref_head);
 218                 BUG_ON(ref == -ENOSPC);
 219
 220                 gnttab_grant_foreign_access_ref(
 221                                 ref,
 222                                 info->xbdev->otherend_id,
 223                                 buffer_mfn,
 224                                 rq_data_dir(req) );
 225
 226                 info->shadow[id].frame[ring_req->nr_segments] =
 227                                 mfn_to_pfn(buffer_mfn);
 228
 229                 ring_req->seg[ring_req->nr_segments] =
 230                                 (struct blkif_request_segment) {
 231                                         .gref       = ref,
 232                                         .first_sect = fsect,
 233                                         .last_sect  = lsect };
 234
 235                 ring_req->nr_segments++;
 236         }
 237
 238         info->ring.req_prod_pvt++;
 239
 240         /* Keep a private copy so we can reissue requests when recovering. */
 241         info->shadow[id].req = *ring_req;
 242
 243         gnttab_free_grant_references(gref_head);
 244
 245         return 0;
 246 }
 247
 248
 249 static inline void flush_requests(struct blkfront_info *info)
 250 {
 251         int notify;
 252
 253         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 254
 255         if (notify)
 256                 notify_remote_via_irq(info->irq);
 257 }
 258
 259 /*
 260  * do_blkif_request
 261  *  read a block; request is in a request queue
 262  */
 263 static void do_blkif_request(struct request_queue *rq)
 264 {
 265         struct blkfront_info *info = NULL;
 266         struct request *req;
 267         int queued;
 268
 269         pr_debug("Entered do_blkif_request\n");
 270
 271         queued = 0;
 272
 273         while ((req = elv_next_request(rq)) != NULL) {
 274                 info = req->rq_disk->private_data;
 275                 if (!blk_fs_request(req)) {
 276                         end_request(req, 0);
 277                         continue;
 278                 }
 279
 280                 if (RING_FULL(&info->ring))
 281                         goto wait;
 282
 283                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 284                          "(%u/%li) buffer:%p [%s]\n",
 285                          req, req->cmd, (unsigned long)req->sector,
 286                          req->current_nr_sectors,
 287                          req->nr_sectors, req->buffer,
 288                          rq_data_dir(req) ? "write" : "read");
 289
 290
 291                 blkdev_dequeue_request(req);
 292                 if (blkif_queue_request(req)) {
 293                         blk_requeue_request(rq, req);
 294 wait:
 295                         /* Avoid pointless unplugs. */
 296                         blk_stop_queue(rq);
 297                         break;
 298                 }
 299
 300                 queued++;
 301         }
 302
 303         if (queued != 0)
 304                 flush_requests(info);
 305 }
 306
 307 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 308 {
 309         struct request_queue *rq;
 310
 311         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 312         if (rq == NULL)
 313                 return -1;
 314
 315         elevator_init(rq, "noop");
 316
 317         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 318         blk_queue_hardsect_size(rq, sector_size);
 319         blk_queue_max_sectors(rq, 512);
 320
 321         /* Each segment in a request is up to an aligned page in size. */
 322         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 323         blk_queue_max_segment_size(rq, PAGE_SIZE);
 324
 325         /* Ensure a merged request will fit in a single I/O ring slot. */
 326         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 327         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 328
 329         /* Make sure buffer addresses are sector-aligned. */
 330         blk_queue_dma_alignment(rq, 511);
 331
 332         gd->queue = rq;
 333
 334         return 0;
 335 }
 336
 337
 338 static int xlvbd_barrier(struct blkfront_info *info)
 339 {
 340         int err;
 341
 342         err = blk_queue_ordered(info->rq,
 343                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 344                                 NULL);
 345
 346         if (err)
 347                 return err;
 348
 349         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 350                info->gd->disk_name,
 351                info->feature_barrier ? "enabled" : "disabled");
 352         return 0;
 353 }
 354
 355
 356 static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
 357                                int vdevice, u16 vdisk_info, u16 sector_size,
 358                                struct blkfront_info *info)
 359 {
 360         struct gendisk *gd;
 361         int nr_minors = 1;
 362         int err = -ENODEV;
 363
 364         BUG_ON(info->gd != NULL);
 365         BUG_ON(info->rq != NULL);
 366
 367         if ((minor % PARTS_PER_DISK) == 0)
 368                 nr_minors = PARTS_PER_DISK;
 369
 370         gd = alloc_disk(nr_minors);
 371         if (gd == NULL)
 372                 goto out;
 373
 374         if (nr_minors > 1)
 375                 sprintf(gd->disk_name, "%s%c", DEV_NAME,
 376                         'a' + minor / PARTS_PER_DISK);
 377         else
 378                 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 379                         'a' + minor / PARTS_PER_DISK,
 380                         minor % PARTS_PER_DISK);
 381
 382         gd->major = XENVBD_MAJOR;
 383         gd->first_minor = minor;
 384         gd->fops = &xlvbd_block_fops;
 385         gd->private_data = info;
 386         gd->driverfs_dev = &(info->xbdev->dev);
 387         set_capacity(gd, capacity);
 388
 389         if (xlvbd_init_blk_queue(gd, sector_size)) {
 390                 del_gendisk(gd);
 391                 goto out;
 392         }
 393
 394         info->rq = gd->queue;
 395         info->gd = gd;
 396
 397         if (info->feature_barrier)
 398                 xlvbd_barrier(info);
 399
 400         if (vdisk_info & VDISK_READONLY)
 401                 set_disk_ro(gd, 1);
 402
 403         if (vdisk_info & VDISK_REMOVABLE)
 404                 gd->flags |= GENHD_FL_REMOVABLE;
 405
 406         if (vdisk_info & VDISK_CDROM)
 407                 gd->flags |= GENHD_FL_CD;
 408
 409         return 0;
 410
 411  out:
 412         return err;
 413 }
 414
 415 static void kick_pending_request_queues(struct blkfront_info *info)
 416 {
 417         if (!RING_FULL(&info->ring)) {
 418                 /* Re-enable calldowns. */
 419                 blk_start_queue(info->rq);
 420                 /* Kick things off immediately. */
 421                 do_blkif_request(info->rq);
 422         }
 423 }
 424
 425 static void blkif_restart_queue(struct work_struct *work)
 426 {
 427         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 428
 429         spin_lock_irq(&blkif_io_lock);
 430         if (info->connected == BLKIF_STATE_CONNECTED)
 431                 kick_pending_request_queues(info);
 432         spin_unlock_irq(&blkif_io_lock);
 433 }
 434
 435 static void blkif_free(struct blkfront_info *info, int suspend)
 436 {
 437         /* Prevent new requests being issued until we fix things up. */
 438         spin_lock_irq(&blkif_io_lock);
 439         info->connected = suspend ?
 440                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 441         /* No more blkif_request(). */
 442         if (info->rq)
 443                 blk_stop_queue(info->rq);
 444         /* No more gnttab callback work. */
 445         gnttab_cancel_free_callback(&info->callback);
 446         spin_unlock_irq(&blkif_io_lock);
 447
 448         /* Flush gnttab callback work. Must be done with no locks held. */
 449         flush_scheduled_work();
 450
 451         /* Free resources associated with old device channel. */
 452         if (info->ring_ref != GRANT_INVALID_REF) {
 453                 gnttab_end_foreign_access(info->ring_ref, 0,
 454                                           (unsigned long)info->ring.sring);
 455                 info->ring_ref = GRANT_INVALID_REF;
 456                 info->ring.sring = NULL;
 457         }
 458         if (info->irq)
 459                 unbind_from_irqhandler(info->irq, info);
 460         info->evtchn = info->irq = 0;
 461
 462 }
 463
 464 static void blkif_completion(struct blk_shadow *s)
 465 {
 466         int i;
 467         for (i = 0; i < s->req.nr_segments; i++)
 468                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 469 }
 470
 471 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 472 {
 473         struct request *req;
 474         struct blkif_response *bret;
 475         RING_IDX i, rp;
 476         unsigned long flags;
 477         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 478         int error;
 479
 480         spin_lock_irqsave(&blkif_io_lock, flags);
 481
 482         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 483                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 484                 return IRQ_HANDLED;
 485         }
 486
 487  again:
 488         rp = info->ring.sring->rsp_prod;
 489         rmb(); /* Ensure we see queued responses up to 'rp'. */
 490
 491         for (i = info->ring.rsp_cons; i != rp; i++) {
 492                 unsigned long id;
 493                 int ret;
 494
 495                 bret = RING_GET_RESPONSE(&info->ring, i);
 496                 id   = bret->id;
 497                 req  = (struct request *)info->shadow[id].request;
 498
 499                 blkif_completion(&info->shadow[id]);
 500
 501                 add_id_to_freelist(info, id);
 502
 503                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 504                 switch (bret->operation) {
 505                 case BLKIF_OP_WRITE_BARRIER:
 506                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 507                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 508                                        info->gd->disk_name);
 509                                 error = -EOPNOTSUPP;
 510                                 info->feature_barrier = 0;
 511                                 xlvbd_barrier(info);
 512                         }
 513                         /* fall through */
 514                 case BLKIF_OP_READ:
 515                 case BLKIF_OP_WRITE:
 516                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 517                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 518                                         "request: %x\n", bret->status);
 519
 520                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
 521                         BUG_ON(ret);
 522                         break;
 523                 default:
 524                         BUG();
 525                 }
 526         }
 527
 528         info->ring.rsp_cons = i;
 529
 530         if (i != info->ring.req_prod_pvt) {
 531                 int more_to_do;
 532                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 533                 if (more_to_do)
 534                         goto again;
 535         } else
 536                 info->ring.sring->rsp_event = i + 1;
 537
 538         kick_pending_request_queues(info);
 539
 540         spin_unlock_irqrestore(&blkif_io_lock, flags);
 541
 542         return IRQ_HANDLED;
 543 }
 544
 545
 546 static int setup_blkring(struct xenbus_device *dev,
 547                          struct blkfront_info *info)
 548 {
 549         struct blkif_sring *sring;
 550         int err;
 551
 552         info->ring_ref = GRANT_INVALID_REF;
 553
 554         sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL);
 555         if (!sring) {
 556                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 557                 return -ENOMEM;
 558         }
 559         SHARED_RING_INIT(sring);
 560         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 561
 562         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 563         if (err < 0) {
 564                 free_page((unsigned long)sring);
 565                 info->ring.sring = NULL;
 566                 goto fail;
 567         }
 568         info->ring_ref = err;
 569
 570         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 571         if (err)
 572                 goto fail;
 573
 574         err = bind_evtchn_to_irqhandler(info->evtchn,
 575                                         blkif_interrupt,
 576                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 577         if (err <= 0) {
 578                 xenbus_dev_fatal(dev, err,
 579                                  "bind_evtchn_to_irqhandler failed");
 580                 goto fail;
 581         }
 582         info->irq = err;
 583
 584         return 0;
 585 fail:
 586         blkif_free(info, 0);
 587         return err;
 588 }
 589
 590
 591 /* Common code used when first setting up, and when resuming. */
 592 static int talk_to_backend(struct xenbus_device *dev,
 593                            struct blkfront_info *info)
 594 {
 595         const char *message = NULL;
 596         struct xenbus_transaction xbt;
 597         int err;
 598
 599         /* Create shared ring, alloc event channel. */
 600         err = setup_blkring(dev, info);
 601         if (err)
 602                 goto out;
 603
 604 again:
 605         err = xenbus_transaction_start(&xbt);
 606         if (err) {
 607                 xenbus_dev_fatal(dev, err, "starting transaction");
 608                 goto destroy_blkring;
 609         }
 610
 611         err = xenbus_printf(xbt, dev->nodename,
 612                             "ring-ref", "%u", info->ring_ref);
 613         if (err) {
 614                 message = "writing ring-ref";
 615                 goto abort_transaction;
 616         }
 617         err = xenbus_printf(xbt, dev->nodename,
 618                             "event-channel", "%u", info->evtchn);
 619         if (err) {
 620                 message = "writing event-channel";
 621                 goto abort_transaction;
 622         }
 623
 624         err = xenbus_transaction_end(xbt, 0);
 625         if (err) {
 626                 if (err == -EAGAIN)
 627                         goto again;
 628                 xenbus_dev_fatal(dev, err, "completing transaction");
 629                 goto destroy_blkring;
 630         }
 631
 632         xenbus_switch_state(dev, XenbusStateInitialised);
 633
 634         return 0;
 635
 636  abort_transaction:
 637         xenbus_transaction_end(xbt, 1);
 638         if (message)
 639                 xenbus_dev_fatal(dev, err, "%s", message);
 640  destroy_blkring:
 641         blkif_free(info, 0);
 642  out:
 643         return err;
 644 }
 645
 646
 647 /**
 648  * Entry point to this code when a new device is created.  Allocate the basic
 649  * structures and the ring buffer for communication with the backend, and
 650  * inform the backend of the appropriate details for those.  Switch to
 651  * Initialised state.
 652  */
 653 static int blkfront_probe(struct xenbus_device *dev,
 654                           const struct xenbus_device_id *id)
 655 {
 656         int err, vdevice, i;
 657         struct blkfront_info *info;
 658
 659         /* FIXME: Use dynamic device id if this is not set. */
 660         err = xenbus_scanf(XBT_NIL, dev->nodename,
 661                            "virtual-device", "%i", &vdevice);
 662         if (err != 1) {
 663                 xenbus_dev_fatal(dev, err, "reading virtual-device");
 664                 return err;
 665         }
 666
 667         info = kzalloc(sizeof(*info), GFP_KERNEL);
 668         if (!info) {
 669                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 670                 return -ENOMEM;
 671         }
 672
 673         info->xbdev = dev;
 674         info->vdevice = vdevice;
 675         info->connected = BLKIF_STATE_DISCONNECTED;
 676         INIT_WORK(&info->work, blkif_restart_queue);
 677
 678         for (i = 0; i < BLK_RING_SIZE; i++)
 679                 info->shadow[i].req.id = i+1;
 680         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 681
 682         /* Front end dir is a number, which is used as the id. */
 683         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 684         dev->dev.driver_data = info;
 685
 686         err = talk_to_backend(dev, info);
 687         if (err) {
 688                 kfree(info);
 689                 dev->dev.driver_data = NULL;
 690                 return err;
 691         }
 692
 693         return 0;
 694 }
 695
 696
 697 static int blkif_recover(struct blkfront_info *info)
 698 {
 699         int i;
 700         struct blkif_request *req;
 701         struct blk_shadow *copy;
 702         int j;
 703
 704         /* Stage 1: Make a safe copy of the shadow state. */
 705         copy = kmalloc(sizeof(info->shadow), GFP_KERNEL);
 706         if (!copy)
 707                 return -ENOMEM;
 708         memcpy(copy, info->shadow, sizeof(info->shadow));
 709
 710         /* Stage 2: Set up free list. */
 711         memset(&info->shadow, 0, sizeof(info->shadow));
 712         for (i = 0; i < BLK_RING_SIZE; i++)
 713                 info->shadow[i].req.id = i+1;
 714         info->shadow_free = info->ring.req_prod_pvt;
 715         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 716
 717         /* Stage 3: Find pending requests and requeue them. */
 718         for (i = 0; i < BLK_RING_SIZE; i++) {
 719                 /* Not in use? */
 720                 if (copy[i].request == 0)
 721                         continue;
 722
 723                 /* Grab a request slot and copy shadow state into it. */
 724                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 725                 *req = copy[i].req;
 726
 727                 /* We get a new request id, and must reset the shadow state. */
 728                 req->id = get_id_from_freelist(info);
 729                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 730
 731                 /* Rewrite any grant references invalidated by susp/resume. */
 732                 for (j = 0; j < req->nr_segments; j++)
 733                         gnttab_grant_foreign_access_ref(
 734                                 req->seg[j].gref,
 735                                 info->xbdev->otherend_id,
 736                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 737                                 rq_data_dir(
 738                                         (struct request *)
 739                                         info->shadow[req->id].request));
 740                 info->shadow[req->id].req = *req;
 741
 742                 info->ring.req_prod_pvt++;
 743         }
 744
 745         kfree(copy);
 746
 747         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 748
 749         spin_lock_irq(&blkif_io_lock);
 750
 751         /* Now safe for us to use the shared ring */
 752         info->connected = BLKIF_STATE_CONNECTED;
 753
 754         /* Send off requeued requests */
 755         flush_requests(info);
 756
 757         /* Kick any other new requests queued since we resumed */
 758         kick_pending_request_queues(info);
 759
 760         spin_unlock_irq(&blkif_io_lock);
 761
 762         return 0;
 763 }
 764
 765 /**
 766  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 767  * driver restart.  We tear down our blkif structure and recreate it, but
 768  * leave the device-layer structures intact so that this is transparent to the
 769  * rest of the kernel.
 770  */
 771 static int blkfront_resume(struct xenbus_device *dev)
 772 {
 773         struct blkfront_info *info = dev->dev.driver_data;
 774         int err;
 775
 776         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 777
 778         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 779
 780         err = talk_to_backend(dev, info);
 781         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 782                 err = blkif_recover(info);
 783
 784         return err;
 785 }
 786
 787
 788 /*
 789  * Invoked when the backend is finally 'ready' (and has told produced
 790  * the details about the physical device - #sectors, size, etc).
 791  */
 792 static void blkfront_connect(struct blkfront_info *info)
 793 {
 794         unsigned long long sectors;
 795         unsigned long sector_size;
 796         unsigned int binfo;
 797         int err;
 798
 799         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 800             (info->connected == BLKIF_STATE_SUSPENDED) )
 801                 return;
 802
 803         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 804                 __func__, info->xbdev->otherend);
 805
 806         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 807                             "sectors", "%llu", &sectors,
 808                             "info", "%u", &binfo,
 809                             "sector-size", "%lu", &sector_size,
 810                             NULL);
 811         if (err) {
 812                 xenbus_dev_fatal(info->xbdev, err,
 813                                  "reading backend fields at %s",
 814                                  info->xbdev->otherend);
 815                 return;
 816         }
 817
 818         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 819                             "feature-barrier", "%lu", &info->feature_barrier,
 820                             NULL);
 821         if (err)
 822                 info->feature_barrier = 0;
 823
 824         err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
 825                                   sectors, info->vdevice,
 826                                   binfo, sector_size, info);
 827         if (err) {
 828                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 829                                  info->xbdev->otherend);
 830                 return;
 831         }
 832
 833         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 834
 835         /* Kick pending requests. */
 836         spin_lock_irq(&blkif_io_lock);
 837         info->connected = BLKIF_STATE_CONNECTED;
 838         kick_pending_request_queues(info);
 839         spin_unlock_irq(&blkif_io_lock);
 840
 841         add_disk(info->gd);
 842 }
 843
 844 /**
 845  * Handle the change of state of the backend to Closing.  We must delete our
 846  * device-layer structures now, to ensure that writes are flushed through to
 847  * the backend.  Once is this done, we can switch to Closed in
 848  * acknowledgement.
 849  */
 850 static void blkfront_closing(struct xenbus_device *dev)
 851 {
 852         struct blkfront_info *info = dev->dev.driver_data;
 853         unsigned long flags;
 854
 855         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 856
 857         if (info->rq == NULL)
 858                 goto out;
 859
 860         spin_lock_irqsave(&blkif_io_lock, flags);
 861
 862         del_gendisk(info->gd);
 863
 864         /* No more blkif_request(). */
 865         blk_stop_queue(info->rq);
 866
 867         /* No more gnttab callback work. */
 868         gnttab_cancel_free_callback(&info->callback);
 869         spin_unlock_irqrestore(&blkif_io_lock, flags);
 870
 871         /* Flush gnttab callback work. Must be done with no locks held. */
 872         flush_scheduled_work();
 873
 874         blk_cleanup_queue(info->rq);
 875         info->rq = NULL;
 876
 877  out:
 878         xenbus_frontend_closed(dev);
 879 }
 880
 881 /**
 882  * Callback received when the backend's state changes.
 883  */
 884 static void backend_changed(struct xenbus_device *dev,
 885                             enum xenbus_state backend_state)
 886 {
 887         struct blkfront_info *info = dev->dev.driver_data;
 888         struct block_device *bd;
 889
 890         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 891
 892         switch (backend_state) {
 893         case XenbusStateInitialising:
 894         case XenbusStateInitWait:
 895         case XenbusStateInitialised:
 896         case XenbusStateUnknown:
 897         case XenbusStateClosed:
 898                 break;
 899
 900         case XenbusStateConnected:
 901                 blkfront_connect(info);
 902                 break;
 903
 904         case XenbusStateClosing:
 905                 bd = bdget(info->dev);
 906                 if (bd == NULL)
 907                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 908
 909                 mutex_lock(&bd->bd_mutex);
 910                 if (info->users > 0)
 911                         xenbus_dev_error(dev, -EBUSY,
 912                                          "Device in use; refusing to close");
 913                 else
 914                         blkfront_closing(dev);
 915                 mutex_unlock(&bd->bd_mutex);
 916                 bdput(bd);
 917                 break;
 918         }
 919 }
 920
 921 static int blkfront_remove(struct xenbus_device *dev)
 922 {
 923         struct blkfront_info *info = dev->dev.driver_data;
 924
 925         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
 926
 927         blkif_free(info, 0);
 928
 929         kfree(info);
 930
 931         return 0;
 932 }
 933
 934 static int blkif_open(struct inode *inode, struct file *filep)
 935 {
 936         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 937         info->users++;
 938         return 0;
 939 }
 940
 941 static int blkif_release(struct inode *inode, struct file *filep)
 942 {
 943         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 944         info->users--;
 945         if (info->users == 0) {
 946                 /* Check whether we have been instructed to close.  We will
 947                    have ignored this request initially, as the device was
 948                    still mounted. */
 949                 struct xenbus_device *dev = info->xbdev;
 950                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
 951
 952                 if (state == XenbusStateClosing)
 953                         blkfront_closing(dev);
 954         }
 955         return 0;
 956 }
 957
 958 static struct block_device_operations xlvbd_block_fops =
 959 {
 960         .owner = THIS_MODULE,
 961         .open = blkif_open,
 962         .release = blkif_release,
 963 <<<<<<< HEAD:drivers/block/xen-blkfront.c
 964 =======
 965         .getgeo = blkif_getgeo,
 966 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:drivers/block/xen-blkfront.c
 967 };
 968
 969
 970 static struct xenbus_device_id blkfront_ids[] = {
 971         { "vbd" },
 972         { "" }
 973 };
 974
 975 static struct xenbus_driver blkfront = {
 976         .name = "vbd",
 977         .owner = THIS_MODULE,
 978         .ids = blkfront_ids,
 979         .probe = blkfront_probe,
 980         .remove = blkfront_remove,
 981         .resume = blkfront_resume,
 982         .otherend_changed = backend_changed,
 983 };
 984
 985 static int __init xlblk_init(void)
 986 {
 987         if (!is_running_on_xen())
 988                 return -ENODEV;
 989
 990         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
 991                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
 992                        XENVBD_MAJOR, DEV_NAME);
 993                 return -ENODEV;
 994         }
 995
 996         return xenbus_register_frontend(&blkfront);
 997 }
 998 module_init(xlblk_init);
 999
1000
1001 static void xlblk_exit(void)
1002 {
1003         return xenbus_unregister_driver(&blkfront);
1004 }
1005 module_exit(xlblk_exit);
1006
1007 MODULE_DESCRIPTION("Xen virtual block device frontend");
1008 MODULE_LICENSE("GPL");
1009 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);