drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/cdrom.h>
  42 #include <linux/module.h>
  43 #include <linux/slab.h>
  44 #include <linux/scatterlist.h>
  45
  46 #include <xen/xen.h>
  47 #include <xen/xenbus.h>
  48 #include <xen/grant_table.h>
  49 #include <xen/events.h>
  50 #include <xen/page.h>
  51 #include <xen/platform_pci.h>
  52
  53 #include <xen/interface/grant_table.h>
  54 #include <xen/interface/io/blkif.h>
  55 #include <xen/interface/io/protocols.h>
  56
  57 #include <asm/xen/hypervisor.h>
  58
  59 enum blkif_state {
  60         BLKIF_STATE_DISCONNECTED,
  61         BLKIF_STATE_CONNECTED,
  62         BLKIF_STATE_SUSPENDED,
  63 };
  64
  65 struct blk_shadow {
  66         struct blkif_request req;
  67         unsigned long request;
  68         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  69 };
  70
  71 static const struct block_device_operations xlvbd_block_fops;
  72
  73 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  74
  75 /*
  76  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  77  * hang in private_data off the gendisk structure. We may end up
  78  * putting all kinds of interesting stuff here :-)
  79  */
  80 struct blkfront_info
  81 {
  82         struct xenbus_device *xbdev;
  83         struct gendisk *gd;
  84         int vdevice;
  85         blkif_vdev_t handle;
  86         enum blkif_state connected;
  87         int ring_ref;
  88         struct blkif_front_ring ring;
  89         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  90         unsigned int evtchn, irq;
  91         struct request_queue *rq;
  92         struct work_struct work;
  93         struct gnttab_free_callback callback;
  94         struct blk_shadow shadow[BLK_RING_SIZE];
  95         unsigned long shadow_free;
  96         int feature_barrier;
  97         int is_ready;
  98
  99         /**
 100          * The number of people holding this device open.  We won't allow a
 101          * hot-unplug unless this is 0.
 102          */
 103         int users;
 104 };
 105
 106 static DEFINE_SPINLOCK(blkif_io_lock);
 107
 108 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 109         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 110 #define GRANT_INVALID_REF       0
 111
 112 #define PARTS_PER_DISK          16
 113 #define PARTS_PER_EXT_DISK      256
 114
 115 #define BLKIF_MAJOR(dev) ((dev)>>8)
 116 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 117
 118 #define EXT_SHIFT 28
 119 #define EXTENDED (1<<EXT_SHIFT)
 120 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 121 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 122
 123 #define DEV_NAME        "xvd"   /* name in /dev */
 124
 125 static int get_id_from_freelist(struct blkfront_info *info)
 126 {
 127         unsigned long free = info->shadow_free;
 128         BUG_ON(free >= BLK_RING_SIZE);
 129         info->shadow_free = info->shadow[free].req.id;
 130         info->shadow[free].req.id = 0x0fffffee; /* debug */
 131         return free;
 132 }
 133
 134 static void add_id_to_freelist(struct blkfront_info *info,
 135                                unsigned long id)
 136 {
 137         info->shadow[id].req.id  = info->shadow_free;
 138         info->shadow[id].request = 0;
 139         info->shadow_free = id;
 140 }
 141
 142 static void blkif_restart_queue_callback(void *arg)
 143 {
 144         struct blkfront_info *info = (struct blkfront_info *)arg;
 145         schedule_work(&info->work);
 146 }
 147
 148 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 149 {
 150         /* We don't have real geometry info, but let's at least return
 151            values consistent with the size of the device */
 152         sector_t nsect = get_capacity(bd->bd_disk);
 153         sector_t cylinders = nsect;
 154
 155         hg->heads = 0xff;
 156         hg->sectors = 0x3f;
 157         sector_div(cylinders, hg->heads * hg->sectors);
 158         hg->cylinders = cylinders;
 159         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 160                 hg->cylinders = 0xffff;
 161         return 0;
 162 }
 163
 164 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 165                        unsigned command, unsigned long argument)
 166 {
 167         struct blkfront_info *info = bdev->bd_disk->private_data;
 168         int i;
 169
 170         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 171                 command, (long)argument);
 172
 173         switch (command) {
 174         case CDROMMULTISESSION:
 175                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 176                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 177                         if (put_user(0, (char __user *)(argument + i)))
 178                                 return -EFAULT;
 179                 return 0;
 180
 181         case CDROM_GET_CAPABILITY: {
 182                 struct gendisk *gd = info->gd;
 183                 if (gd->flags & GENHD_FL_CD)
 184                         return 0;
 185                 return -EINVAL;
 186         }
 187
 188         default:
 189                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 190                   command);*/
 191                 return -EINVAL; /* same return as native Linux */
 192         }
 193
 194         return 0;
 195 }
 196
 197 /*
 198  * blkif_queue_request
 199  *
 200  * request block io
 201  *
 202  * id: for guest use only.
 203  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 204  * buffer: buffer to read/write into. this should be a
 205  *   virtual address in the guest os.
 206  */
 207 static int blkif_queue_request(struct request *req)
 208 {
 209         struct blkfront_info *info = req->rq_disk->private_data;
 210         unsigned long buffer_mfn;
 211         struct blkif_request *ring_req;
 212         unsigned long id;
 213         unsigned int fsect, lsect;
 214         int i, ref;
 215         grant_ref_t gref_head;
 216         struct scatterlist *sg;
 217
 218         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 219                 return 1;
 220
 221         if (gnttab_alloc_grant_references(
 222                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 223                 gnttab_request_free_callback(
 224                         &info->callback,
 225                         blkif_restart_queue_callback,
 226                         info,
 227                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 228                 return 1;
 229         }
 230
 231         /* Fill out a communications ring structure. */
 232         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 233         id = get_id_from_freelist(info);
 234         info->shadow[id].request = (unsigned long)req;
 235
 236         ring_req->id = id;
 237         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
 238         ring_req->handle = info->handle;
 239
 240         ring_req->operation = rq_data_dir(req) ?
 241                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 242         if (blk_barrier_rq(req))
 243                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 244
 245         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
 246         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 247
 248         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
 249                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 250                 fsect = sg->offset >> 9;
 251                 lsect = fsect + (sg->length >> 9) - 1;
 252                 /* install a grant reference. */
 253                 ref = gnttab_claim_grant_reference(&gref_head);
 254                 BUG_ON(ref == -ENOSPC);
 255
 256                 gnttab_grant_foreign_access_ref(
 257                                 ref,
 258                                 info->xbdev->otherend_id,
 259                                 buffer_mfn,
 260                                 rq_data_dir(req) );
 261
 262                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 263                 ring_req->seg[i] =
 264                                 (struct blkif_request_segment) {
 265                                         .gref       = ref,
 266                                         .first_sect = fsect,
 267                                         .last_sect  = lsect };
 268         }
 269
 270         info->ring.req_prod_pvt++;
 271
 272         /* Keep a private copy so we can reissue requests when recovering. */
 273         info->shadow[id].req = *ring_req;
 274
 275         gnttab_free_grant_references(gref_head);
 276
 277         return 0;
 278 }
 279
 280
 281 static inline void flush_requests(struct blkfront_info *info)
 282 {
 283         int notify;
 284
 285         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 286
 287         if (notify)
 288                 notify_remote_via_irq(info->irq);
 289 }
 290
 291 /*
 292  * do_blkif_request
 293  *  read a block; request is in a request queue
 294  */
 295 static void do_blkif_request(struct request_queue *rq)
 296 {
 297         struct blkfront_info *info = NULL;
 298         struct request *req;
 299         int queued;
 300
 301         pr_debug("Entered do_blkif_request\n");
 302
 303         queued = 0;
 304
 305         while ((req = blk_peek_request(rq)) != NULL) {
 306                 info = req->rq_disk->private_data;
 307
 308                 if (RING_FULL(&info->ring))
 309                         goto wait;
 310
 311                 blk_start_request(req);
 312
 313                 if (!blk_fs_request(req)) {
 314                         __blk_end_request_all(req, -EIO);
 315                         continue;
 316                 }
 317
 318                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 319                          "(%u/%u) buffer:%p [%s]\n",
 320                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 321                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 322                          req->buffer, rq_data_dir(req) ? "write" : "read");
 323
 324                 if (blkif_queue_request(req)) {
 325                         blk_requeue_request(rq, req);
 326 wait:
 327                         /* Avoid pointless unplugs. */
 328                         blk_stop_queue(rq);
 329                         break;
 330                 }
 331
 332                 queued++;
 333         }
 334
 335         if (queued != 0)
 336                 flush_requests(info);
 337 }
 338
 339 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 340 {
 341         struct request_queue *rq;
 342
 343         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 344         if (rq == NULL)
 345                 return -1;
 346
 347         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 348
 349         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 350         blk_queue_logical_block_size(rq, sector_size);
 351         blk_queue_max_hw_sectors(rq, 512);
 352
 353         /* Each segment in a request is up to an aligned page in size. */
 354         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 355         blk_queue_max_segment_size(rq, PAGE_SIZE);
 356
 357         /* Ensure a merged request will fit in a single I/O ring slot. */
 358         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 359
 360         /* Make sure buffer addresses are sector-aligned. */
 361         blk_queue_dma_alignment(rq, 511);
 362
 363         /* Make sure we don't use bounce buffers. */
 364         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 365
 366         gd->queue = rq;
 367
 368         return 0;
 369 }
 370
 371
 372 static int xlvbd_barrier(struct blkfront_info *info)
 373 {
 374         int err;
 375
 376         err = blk_queue_ordered(info->rq,
 377                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 378                                 NULL);
 379
 380         if (err)
 381                 return err;
 382
 383         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 384                info->gd->disk_name,
 385                info->feature_barrier ? "enabled" : "disabled");
 386         return 0;
 387 }
 388
 389
 390 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 391                                struct blkfront_info *info,
 392                                u16 vdisk_info, u16 sector_size)
 393 {
 394         struct gendisk *gd;
 395         int nr_minors = 1;
 396         int err = -ENODEV;
 397         unsigned int offset;
 398         int minor;
 399         int nr_parts;
 400
 401         BUG_ON(info->gd != NULL);
 402         BUG_ON(info->rq != NULL);
 403
 404         if ((info->vdevice>>EXT_SHIFT) > 1) {
 405                 /* this is above the extended range; something is wrong */
 406                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 407                 return -ENODEV;
 408         }
 409
 410         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 411                 minor = BLKIF_MINOR(info->vdevice);
 412                 nr_parts = PARTS_PER_DISK;
 413         } else {
 414                 minor = BLKIF_MINOR_EXT(info->vdevice);
 415                 nr_parts = PARTS_PER_EXT_DISK;
 416         }
 417
 418         if ((minor % nr_parts) == 0)
 419                 nr_minors = nr_parts;
 420
 421         gd = alloc_disk(nr_minors);
 422         if (gd == NULL)
 423                 goto out;
 424
 425         offset = minor / nr_parts;
 426
 427         if (nr_minors > 1) {
 428                 if (offset < 26)
 429                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 430                 else
 431                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 432                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 433         } else {
 434                 if (offset < 26)
 435                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 436                                 'a' + offset,
 437                                 minor & (nr_parts - 1));
 438                 else
 439                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 440                                 'a' + ((offset / 26) - 1),
 441                                 'a' + (offset % 26),
 442                                 minor & (nr_parts - 1));
 443         }
 444
 445         gd->major = XENVBD_MAJOR;
 446         gd->first_minor = minor;
 447         gd->fops = &xlvbd_block_fops;
 448         gd->private_data = info;
 449         gd->driverfs_dev = &(info->xbdev->dev);
 450         set_capacity(gd, capacity);
 451
 452         if (xlvbd_init_blk_queue(gd, sector_size)) {
 453                 del_gendisk(gd);
 454                 goto out;
 455         }
 456
 457         info->rq = gd->queue;
 458         info->gd = gd;
 459
 460         if (info->feature_barrier)
 461                 xlvbd_barrier(info);
 462
 463         if (vdisk_info & VDISK_READONLY)
 464                 set_disk_ro(gd, 1);
 465
 466         if (vdisk_info & VDISK_REMOVABLE)
 467                 gd->flags |= GENHD_FL_REMOVABLE;
 468
 469         if (vdisk_info & VDISK_CDROM)
 470                 gd->flags |= GENHD_FL_CD;
 471
 472         return 0;
 473
 474  out:
 475         return err;
 476 }
 477
 478 static void kick_pending_request_queues(struct blkfront_info *info)
 479 {
 480         if (!RING_FULL(&info->ring)) {
 481                 /* Re-enable calldowns. */
 482                 blk_start_queue(info->rq);
 483                 /* Kick things off immediately. */
 484                 do_blkif_request(info->rq);
 485         }
 486 }
 487
 488 static void blkif_restart_queue(struct work_struct *work)
 489 {
 490         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 491
 492         spin_lock_irq(&blkif_io_lock);
 493         if (info->connected == BLKIF_STATE_CONNECTED)
 494                 kick_pending_request_queues(info);
 495         spin_unlock_irq(&blkif_io_lock);
 496 }
 497
 498 static void blkif_free(struct blkfront_info *info, int suspend)
 499 {
 500         /* Prevent new requests being issued until we fix things up. */
 501         spin_lock_irq(&blkif_io_lock);
 502         info->connected = suspend ?
 503                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 504         /* No more blkif_request(). */
 505         if (info->rq)
 506                 blk_stop_queue(info->rq);
 507         /* No more gnttab callback work. */
 508         gnttab_cancel_free_callback(&info->callback);
 509         spin_unlock_irq(&blkif_io_lock);
 510
 511         /* Flush gnttab callback work. Must be done with no locks held. */
 512         flush_scheduled_work();
 513
 514         /* Free resources associated with old device channel. */
 515         if (info->ring_ref != GRANT_INVALID_REF) {
 516                 gnttab_end_foreign_access(info->ring_ref, 0,
 517                                           (unsigned long)info->ring.sring);
 518                 info->ring_ref = GRANT_INVALID_REF;
 519                 info->ring.sring = NULL;
 520         }
 521         if (info->irq)
 522                 unbind_from_irqhandler(info->irq, info);
 523         info->evtchn = info->irq = 0;
 524
 525 }
 526
 527 static void blkif_completion(struct blk_shadow *s)
 528 {
 529         int i;
 530         for (i = 0; i < s->req.nr_segments; i++)
 531                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 532 }
 533
 534 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 535 {
 536         struct request *req;
 537         struct blkif_response *bret;
 538         RING_IDX i, rp;
 539         unsigned long flags;
 540         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 541         int error;
 542
 543         spin_lock_irqsave(&blkif_io_lock, flags);
 544
 545         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 546                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 547                 return IRQ_HANDLED;
 548         }
 549
 550  again:
 551         rp = info->ring.sring->rsp_prod;
 552         rmb(); /* Ensure we see queued responses up to 'rp'. */
 553
 554         for (i = info->ring.rsp_cons; i != rp; i++) {
 555                 unsigned long id;
 556
 557                 bret = RING_GET_RESPONSE(&info->ring, i);
 558                 id   = bret->id;
 559                 req  = (struct request *)info->shadow[id].request;
 560
 561                 blkif_completion(&info->shadow[id]);
 562
 563                 add_id_to_freelist(info, id);
 564
 565                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 566                 switch (bret->operation) {
 567                 case BLKIF_OP_WRITE_BARRIER:
 568                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 569                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 570                                        info->gd->disk_name);
 571                                 error = -EOPNOTSUPP;
 572                                 info->feature_barrier = 0;
 573                                 xlvbd_barrier(info);
 574                         }
 575                         /* fall through */
 576                 case BLKIF_OP_READ:
 577                 case BLKIF_OP_WRITE:
 578                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 579                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 580                                         "request: %x\n", bret->status);
 581
 582                         __blk_end_request_all(req, error);
 583                         break;
 584                 default:
 585                         BUG();
 586                 }
 587         }
 588
 589         info->ring.rsp_cons = i;
 590
 591         if (i != info->ring.req_prod_pvt) {
 592                 int more_to_do;
 593                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 594                 if (more_to_do)
 595                         goto again;
 596         } else
 597                 info->ring.sring->rsp_event = i + 1;
 598
 599         kick_pending_request_queues(info);
 600
 601         spin_unlock_irqrestore(&blkif_io_lock, flags);
 602
 603         return IRQ_HANDLED;
 604 }
 605
 606
 607 static int setup_blkring(struct xenbus_device *dev,
 608                          struct blkfront_info *info)
 609 {
 610         struct blkif_sring *sring;
 611         int err;
 612
 613         info->ring_ref = GRANT_INVALID_REF;
 614
 615         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 616         if (!sring) {
 617                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 618                 return -ENOMEM;
 619         }
 620         SHARED_RING_INIT(sring);
 621         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 622
 623         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 624
 625         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 626         if (err < 0) {
 627                 free_page((unsigned long)sring);
 628                 info->ring.sring = NULL;
 629                 goto fail;
 630         }
 631         info->ring_ref = err;
 632
 633         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 634         if (err)
 635                 goto fail;
 636
 637         err = bind_evtchn_to_irqhandler(info->evtchn,
 638                                         blkif_interrupt,
 639                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 640         if (err <= 0) {
 641                 xenbus_dev_fatal(dev, err,
 642                                  "bind_evtchn_to_irqhandler failed");
 643                 goto fail;
 644         }
 645         info->irq = err;
 646
 647         return 0;
 648 fail:
 649         blkif_free(info, 0);
 650         return err;
 651 }
 652
 653
 654 /* Common code used when first setting up, and when resuming. */
 655 static int talk_to_backend(struct xenbus_device *dev,
 656                            struct blkfront_info *info)
 657 {
 658         const char *message = NULL;
 659         struct xenbus_transaction xbt;
 660         int err;
 661
 662         /* Create shared ring, alloc event channel. */
 663         err = setup_blkring(dev, info);
 664         if (err)
 665                 goto out;
 666
 667 again:
 668         err = xenbus_transaction_start(&xbt);
 669         if (err) {
 670                 xenbus_dev_fatal(dev, err, "starting transaction");
 671                 goto destroy_blkring;
 672         }
 673
 674         err = xenbus_printf(xbt, dev->nodename,
 675                             "ring-ref", "%u", info->ring_ref);
 676         if (err) {
 677                 message = "writing ring-ref";
 678                 goto abort_transaction;
 679         }
 680         err = xenbus_printf(xbt, dev->nodename,
 681                             "event-channel", "%u", info->evtchn);
 682         if (err) {
 683                 message = "writing event-channel";
 684                 goto abort_transaction;
 685         }
 686         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 687                             XEN_IO_PROTO_ABI_NATIVE);
 688         if (err) {
 689                 message = "writing protocol";
 690                 goto abort_transaction;
 691         }
 692
 693         err = xenbus_transaction_end(xbt, 0);
 694         if (err) {
 695                 if (err == -EAGAIN)
 696                         goto again;
 697                 xenbus_dev_fatal(dev, err, "completing transaction");
 698                 goto destroy_blkring;
 699         }
 700
 701         xenbus_switch_state(dev, XenbusStateInitialised);
 702
 703         return 0;
 704
 705  abort_transaction:
 706         xenbus_transaction_end(xbt, 1);
 707         if (message)
 708                 xenbus_dev_fatal(dev, err, "%s", message);
 709  destroy_blkring:
 710         blkif_free(info, 0);
 711  out:
 712         return err;
 713 }
 714
 715
 716 /**
 717  * Entry point to this code when a new device is created.  Allocate the basic
 718  * structures and the ring buffer for communication with the backend, and
 719  * inform the backend of the appropriate details for those.  Switch to
 720  * Initialised state.
 721  */
 722 static int blkfront_probe(struct xenbus_device *dev,
 723                           const struct xenbus_device_id *id)
 724 {
 725         int err, vdevice, i;
 726         struct blkfront_info *info;
 727
 728         /* FIXME: Use dynamic device id if this is not set. */
 729         err = xenbus_scanf(XBT_NIL, dev->nodename,
 730                            "virtual-device", "%i", &vdevice);
 731         if (err != 1) {
 732                 /* go looking in the extended area instead */
 733                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 734                                    "%i", &vdevice);
 735                 if (err != 1) {
 736                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 737                         return err;
 738                 }
 739         }
 740
 741         if (xen_hvm_domain()) {
 742                 char *type;
 743                 int len;
 744                 /* no unplug has been done: do not hook devices != xen vbds */
 745                 if (xen_platform_pci_unplug & XEN_UNPLUG_IGNORE) {
 746                         int major;
 747
 748                         if (!VDEV_IS_EXTENDED(vdevice))
 749                                 major = BLKIF_MAJOR(vdevice);
 750                         else
 751                                 major = XENVBD_MAJOR;
 752
 753                         if (major != XENVBD_MAJOR) {
 754                                 printk(KERN_INFO
 755                                                 "%s: HVM does not support vbd %d as xen block device\n",
 756                                                 __FUNCTION__, vdevice);
 757                                 return -ENODEV;
 758                         }
 759                 }
 760                 /* do not create a PV cdrom device if we are an HVM guest */
 761                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 762                 if (IS_ERR(type))
 763                         return -ENODEV;
 764                 if (strncmp(type, "cdrom", 5) == 0) {
 765                         kfree(type);
 766                         return -ENODEV;
 767                 }
 768                 kfree(type);
 769         }
 770         info = kzalloc(sizeof(*info), GFP_KERNEL);
 771         if (!info) {
 772                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 773                 return -ENOMEM;
 774         }
 775
 776         info->xbdev = dev;
 777         info->vdevice = vdevice;
 778         info->connected = BLKIF_STATE_DISCONNECTED;
 779         INIT_WORK(&info->work, blkif_restart_queue);
 780
 781         for (i = 0; i < BLK_RING_SIZE; i++)
 782                 info->shadow[i].req.id = i+1;
 783         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 784
 785         /* Front end dir is a number, which is used as the id. */
 786         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 787         dev_set_drvdata(&dev->dev, info);
 788
 789         err = talk_to_backend(dev, info);
 790         if (err) {
 791                 kfree(info);
 792                 dev_set_drvdata(&dev->dev, NULL);
 793                 return err;
 794         }
 795
 796         return 0;
 797 }
 798
 799
 800 static int blkif_recover(struct blkfront_info *info)
 801 {
 802         int i;
 803         struct blkif_request *req;
 804         struct blk_shadow *copy;
 805         int j;
 806
 807         /* Stage 1: Make a safe copy of the shadow state. */
 808         copy = kmalloc(sizeof(info->shadow),
 809                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 810         if (!copy)
 811                 return -ENOMEM;
 812         memcpy(copy, info->shadow, sizeof(info->shadow));
 813
 814         /* Stage 2: Set up free list. */
 815         memset(&info->shadow, 0, sizeof(info->shadow));
 816         for (i = 0; i < BLK_RING_SIZE; i++)
 817                 info->shadow[i].req.id = i+1;
 818         info->shadow_free = info->ring.req_prod_pvt;
 819         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 820
 821         /* Stage 3: Find pending requests and requeue them. */
 822         for (i = 0; i < BLK_RING_SIZE; i++) {
 823                 /* Not in use? */
 824                 if (copy[i].request == 0)
 825                         continue;
 826
 827                 /* Grab a request slot and copy shadow state into it. */
 828                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 829                 *req = copy[i].req;
 830
 831                 /* We get a new request id, and must reset the shadow state. */
 832                 req->id = get_id_from_freelist(info);
 833                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 834
 835                 /* Rewrite any grant references invalidated by susp/resume. */
 836                 for (j = 0; j < req->nr_segments; j++)
 837                         gnttab_grant_foreign_access_ref(
 838                                 req->seg[j].gref,
 839                                 info->xbdev->otherend_id,
 840                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 841                                 rq_data_dir(
 842                                         (struct request *)
 843                                         info->shadow[req->id].request));
 844                 info->shadow[req->id].req = *req;
 845
 846                 info->ring.req_prod_pvt++;
 847         }
 848
 849         kfree(copy);
 850
 851         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 852
 853         spin_lock_irq(&blkif_io_lock);
 854
 855         /* Now safe for us to use the shared ring */
 856         info->connected = BLKIF_STATE_CONNECTED;
 857
 858         /* Send off requeued requests */
 859         flush_requests(info);
 860
 861         /* Kick any other new requests queued since we resumed */
 862         kick_pending_request_queues(info);
 863
 864         spin_unlock_irq(&blkif_io_lock);
 865
 866         return 0;
 867 }
 868
 869 /**
 870  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 871  * driver restart.  We tear down our blkif structure and recreate it, but
 872  * leave the device-layer structures intact so that this is transparent to the
 873  * rest of the kernel.
 874  */
 875 static int blkfront_resume(struct xenbus_device *dev)
 876 {
 877         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 878         int err;
 879
 880         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 881
 882         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 883
 884         err = talk_to_backend(dev, info);
 885         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 886                 err = blkif_recover(info);
 887
 888         return err;
 889 }
 890
 891
 892 /*
 893  * Invoked when the backend is finally 'ready' (and has told produced
 894  * the details about the physical device - #sectors, size, etc).
 895  */
 896 static void blkfront_connect(struct blkfront_info *info)
 897 {
 898         unsigned long long sectors;
 899         unsigned long sector_size;
 900         unsigned int binfo;
 901         int err;
 902
 903         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 904             (info->connected == BLKIF_STATE_SUSPENDED) )
 905                 return;
 906
 907         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 908                 __func__, info->xbdev->otherend);
 909
 910         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 911                             "sectors", "%llu", &sectors,
 912                             "info", "%u", &binfo,
 913                             "sector-size", "%lu", &sector_size,
 914                             NULL);
 915         if (err) {
 916                 xenbus_dev_fatal(info->xbdev, err,
 917                                  "reading backend fields at %s",
 918                                  info->xbdev->otherend);
 919                 return;
 920         }
 921
 922         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 923                             "feature-barrier", "%lu", &info->feature_barrier,
 924                             NULL);
 925         if (err)
 926                 info->feature_barrier = 0;
 927
 928         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
 929         if (err) {
 930                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 931                                  info->xbdev->otherend);
 932                 return;
 933         }
 934
 935         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 936
 937         /* Kick pending requests. */
 938         spin_lock_irq(&blkif_io_lock);
 939         info->connected = BLKIF_STATE_CONNECTED;
 940         kick_pending_request_queues(info);
 941         spin_unlock_irq(&blkif_io_lock);
 942
 943         add_disk(info->gd);
 944
 945         info->is_ready = 1;
 946 }
 947
 948 /**
 949  * Handle the change of state of the backend to Closing.  We must delete our
 950  * device-layer structures now, to ensure that writes are flushed through to
 951  * the backend.  Once is this done, we can switch to Closed in
 952  * acknowledgement.
 953  */
 954 static void blkfront_closing(struct xenbus_device *dev)
 955 {
 956         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 957         unsigned long flags;
 958
 959         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 960
 961         if (info->rq == NULL)
 962                 goto out;
 963
 964         spin_lock_irqsave(&blkif_io_lock, flags);
 965
 966         /* No more blkif_request(). */
 967         blk_stop_queue(info->rq);
 968
 969         /* No more gnttab callback work. */
 970         gnttab_cancel_free_callback(&info->callback);
 971         spin_unlock_irqrestore(&blkif_io_lock, flags);
 972
 973         /* Flush gnttab callback work. Must be done with no locks held. */
 974         flush_scheduled_work();
 975
 976         blk_cleanup_queue(info->rq);
 977         info->rq = NULL;
 978
 979         del_gendisk(info->gd);
 980
 981  out:
 982         xenbus_frontend_closed(dev);
 983 }
 984
 985 /**
 986  * Callback received when the backend's state changes.
 987  */
 988 static void backend_changed(struct xenbus_device *dev,
 989                             enum xenbus_state backend_state)
 990 {
 991         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 992         struct block_device *bd;
 993
 994         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 995
 996         switch (backend_state) {
 997         case XenbusStateInitialising:
 998         case XenbusStateInitWait:
 999         case XenbusStateInitialised:
1000         case XenbusStateUnknown:
1001         case XenbusStateClosed:
1002                 break;
1003
1004         case XenbusStateConnected:
1005                 blkfront_connect(info);
1006                 break;
1007
1008         case XenbusStateClosing:
1009                 if (info->gd == NULL) {
1010                         xenbus_frontend_closed(dev);
1011                         break;
1012                 }
1013                 bd = bdget_disk(info->gd, 0);
1014                 if (bd == NULL)
1015                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
1016
1017                 mutex_lock(&bd->bd_mutex);
1018                 if (info->users > 0)
1019                         xenbus_dev_error(dev, -EBUSY,
1020                                          "Device in use; refusing to close");
1021                 else
1022                         blkfront_closing(dev);
1023                 mutex_unlock(&bd->bd_mutex);
1024                 bdput(bd);
1025                 break;
1026         }
1027 }
1028
1029 static int blkfront_remove(struct xenbus_device *dev)
1030 {
1031         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1032
1033         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1034
1035         blkif_free(info, 0);
1036
1037         kfree(info);
1038
1039         return 0;
1040 }
1041
1042 static int blkfront_is_ready(struct xenbus_device *dev)
1043 {
1044         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1045
1046         return info->is_ready;
1047 }
1048
1049 static int blkif_open(struct block_device *bdev, fmode_t mode)
1050 {
1051         struct blkfront_info *info = bdev->bd_disk->private_data;
1052         info->users++;
1053         return 0;
1054 }
1055
1056 static int blkif_release(struct gendisk *disk, fmode_t mode)
1057 {
1058         struct blkfront_info *info = disk->private_data;
1059         info->users--;
1060         if (info->users == 0) {
1061                 /* Check whether we have been instructed to close.  We will
1062                    have ignored this request initially, as the device was
1063                    still mounted. */
1064                 struct xenbus_device *dev = info->xbdev;
1065                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1066
1067                 if (state == XenbusStateClosing && info->is_ready)
1068                         blkfront_closing(dev);
1069         }
1070         return 0;
1071 }
1072
1073 static const struct block_device_operations xlvbd_block_fops =
1074 {
1075         .owner = THIS_MODULE,
1076         .open = blkif_open,
1077         .release = blkif_release,
1078         .getgeo = blkif_getgeo,
1079         .locked_ioctl = blkif_ioctl,
1080 };
1081
1082
1083 static const struct xenbus_device_id blkfront_ids[] = {
1084         { "vbd" },
1085         { "" }
1086 };
1087
1088 static struct xenbus_driver blkfront = {
1089         .name = "vbd",
1090         .owner = THIS_MODULE,
1091         .ids = blkfront_ids,
1092         .probe = blkfront_probe,
1093         .remove = blkfront_remove,
1094         .resume = blkfront_resume,
1095         .otherend_changed = backend_changed,
1096         .is_ready = blkfront_is_ready,
1097 };
1098
1099 static int __init xlblk_init(void)
1100 {
1101         if (!xen_domain())
1102                 return -ENODEV;
1103
1104         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1105                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1106                        XENVBD_MAJOR, DEV_NAME);
1107                 return -ENODEV;
1108         }
1109
1110         return xenbus_register_frontend(&blkfront);
1111 }
1112 module_init(xlblk_init);
1113
1114
1115 static void __exit xlblk_exit(void)
1116 {
1117         return xenbus_unregister_driver(&blkfront);
1118 }
1119 module_exit(xlblk_exit);
1120
1121 MODULE_DESCRIPTION("Xen virtual block device frontend");
1122 MODULE_LICENSE("GPL");
1123 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1124 MODULE_ALIAS("xen:vbd");
1125 MODULE_ALIAS("xenblk");