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/smp_lock.h>
  45 #include <linux/scatterlist.h>
  46
  47 #include <xen/xen.h>
  48 #include <xen/xenbus.h>
  49 #include <xen/grant_table.h>
  50 #include <xen/events.h>
  51 #include <xen/page.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 mutex mutex;
  83         struct xenbus_device *xbdev;
  84         struct gendisk *gd;
  85         int vdevice;
  86         blkif_vdev_t handle;
  87         enum blkif_state connected;
  88         int ring_ref;
  89         struct blkif_front_ring ring;
  90         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  91         unsigned int evtchn, irq;
  92         struct request_queue *rq;
  93         struct work_struct work;
  94         struct gnttab_free_callback callback;
  95         struct blk_shadow shadow[BLK_RING_SIZE];
  96         unsigned long shadow_free;
  97         int feature_barrier;
  98         int is_ready;
  99
 100         /**
 101          * The number of people holding this device open.  We won't allow a
 102          * hot-unplug unless this is 0.
 103          */
 104         int users;
 105 };
 106
 107 static DEFINE_SPINLOCK(blkif_io_lock);
 108
 109 static unsigned int nr_minors;
 110 static unsigned long *minors;
 111 static DEFINE_SPINLOCK(minor_lock);
 112
 113 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 114         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 115 #define GRANT_INVALID_REF       0
 116
 117 #define PARTS_PER_DISK          16
 118 #define PARTS_PER_EXT_DISK      256
 119
 120 #define BLKIF_MAJOR(dev) ((dev)>>8)
 121 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 122
 123 #define EXT_SHIFT 28
 124 #define EXTENDED (1<<EXT_SHIFT)
 125 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 126 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 127
 128 #define DEV_NAME        "xvd"   /* name in /dev */
 129
 130 static int get_id_from_freelist(struct blkfront_info *info)
 131 {
 132         unsigned long free = info->shadow_free;
 133         BUG_ON(free >= BLK_RING_SIZE);
 134         info->shadow_free = info->shadow[free].req.id;
 135         info->shadow[free].req.id = 0x0fffffee; /* debug */
 136         return free;
 137 }
 138
 139 static void add_id_to_freelist(struct blkfront_info *info,
 140                                unsigned long id)
 141 {
 142         info->shadow[id].req.id  = info->shadow_free;
 143         info->shadow[id].request = 0;
 144         info->shadow_free = id;
 145 }
 146
 147 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 148 {
 149         unsigned int end = minor + nr;
 150         int rc;
 151
 152         if (end > nr_minors) {
 153                 unsigned long *bitmap, *old;
 154
 155                 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
 156                                  GFP_KERNEL);
 157                 if (bitmap == NULL)
 158                         return -ENOMEM;
 159
 160                 spin_lock(&minor_lock);
 161                 if (end > nr_minors) {
 162                         old = minors;
 163                         memcpy(bitmap, minors,
 164                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 165                         minors = bitmap;
 166                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 167                 } else
 168                         old = bitmap;
 169                 spin_unlock(&minor_lock);
 170                 kfree(old);
 171         }
 172
 173         spin_lock(&minor_lock);
 174         if (find_next_bit(minors, end, minor) >= end) {
 175                 for (; minor < end; ++minor)
 176                         __set_bit(minor, minors);
 177                 rc = 0;
 178         } else
 179                 rc = -EBUSY;
 180         spin_unlock(&minor_lock);
 181
 182         return rc;
 183 }
 184
 185 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 186 {
 187         unsigned int end = minor + nr;
 188
 189         BUG_ON(end > nr_minors);
 190         spin_lock(&minor_lock);
 191         for (; minor < end; ++minor)
 192                 __clear_bit(minor, minors);
 193         spin_unlock(&minor_lock);
 194 }
 195
 196 static void blkif_restart_queue_callback(void *arg)
 197 {
 198         struct blkfront_info *info = (struct blkfront_info *)arg;
 199         schedule_work(&info->work);
 200 }
 201
 202 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 203 {
 204         /* We don't have real geometry info, but let's at least return
 205            values consistent with the size of the device */
 206         sector_t nsect = get_capacity(bd->bd_disk);
 207         sector_t cylinders = nsect;
 208
 209         hg->heads = 0xff;
 210         hg->sectors = 0x3f;
 211         sector_div(cylinders, hg->heads * hg->sectors);
 212         hg->cylinders = cylinders;
 213         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 214                 hg->cylinders = 0xffff;
 215         return 0;
 216 }
 217
 218 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 219                        unsigned command, unsigned long argument)
 220 {
 221         struct blkfront_info *info = bdev->bd_disk->private_data;
 222         int i;
 223
 224         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 225                 command, (long)argument);
 226
 227         switch (command) {
 228         case CDROMMULTISESSION:
 229                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 230                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 231                         if (put_user(0, (char __user *)(argument + i)))
 232                                 return -EFAULT;
 233                 return 0;
 234
 235         case CDROM_GET_CAPABILITY: {
 236                 struct gendisk *gd = info->gd;
 237                 if (gd->flags & GENHD_FL_CD)
 238                         return 0;
 239                 return -EINVAL;
 240         }
 241
 242         default:
 243                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 244                   command);*/
 245                 return -EINVAL; /* same return as native Linux */
 246         }
 247
 248         return 0;
 249 }
 250
 251 /*
 252  * blkif_queue_request
 253  *
 254  * request block io
 255  *
 256  * id: for guest use only.
 257  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 258  * buffer: buffer to read/write into. this should be a
 259  *   virtual address in the guest os.
 260  */
 261 static int blkif_queue_request(struct request *req)
 262 {
 263         struct blkfront_info *info = req->rq_disk->private_data;
 264         unsigned long buffer_mfn;
 265         struct blkif_request *ring_req;
 266         unsigned long id;
 267         unsigned int fsect, lsect;
 268         int i, ref;
 269         grant_ref_t gref_head;
 270         struct scatterlist *sg;
 271
 272         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 273                 return 1;
 274
 275         if (gnttab_alloc_grant_references(
 276                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 277                 gnttab_request_free_callback(
 278                         &info->callback,
 279                         blkif_restart_queue_callback,
 280                         info,
 281                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 282                 return 1;
 283         }
 284
 285         /* Fill out a communications ring structure. */
 286         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 287         id = get_id_from_freelist(info);
 288         info->shadow[id].request = (unsigned long)req;
 289
 290         ring_req->id = id;
 291         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
 292         ring_req->handle = info->handle;
 293
 294         ring_req->operation = rq_data_dir(req) ?
 295                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 296         if (req->cmd_flags & REQ_HARDBARRIER)
 297                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 298
 299         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
 300         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 301
 302         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
 303                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 304                 fsect = sg->offset >> 9;
 305                 lsect = fsect + (sg->length >> 9) - 1;
 306                 /* install a grant reference. */
 307                 ref = gnttab_claim_grant_reference(&gref_head);
 308                 BUG_ON(ref == -ENOSPC);
 309
 310                 gnttab_grant_foreign_access_ref(
 311                                 ref,
 312                                 info->xbdev->otherend_id,
 313                                 buffer_mfn,
 314                                 rq_data_dir(req) );
 315
 316                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 317                 ring_req->seg[i] =
 318                                 (struct blkif_request_segment) {
 319                                         .gref       = ref,
 320                                         .first_sect = fsect,
 321                                         .last_sect  = lsect };
 322         }
 323
 324         info->ring.req_prod_pvt++;
 325
 326         /* Keep a private copy so we can reissue requests when recovering. */
 327         info->shadow[id].req = *ring_req;
 328
 329         gnttab_free_grant_references(gref_head);
 330
 331         return 0;
 332 }
 333
 334
 335 static inline void flush_requests(struct blkfront_info *info)
 336 {
 337         int notify;
 338
 339         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 340
 341         if (notify)
 342                 notify_remote_via_irq(info->irq);
 343 }
 344
 345 /*
 346  * do_blkif_request
 347  *  read a block; request is in a request queue
 348  */
 349 static void do_blkif_request(struct request_queue *rq)
 350 {
 351         struct blkfront_info *info = NULL;
 352         struct request *req;
 353         int queued;
 354
 355         pr_debug("Entered do_blkif_request\n");
 356
 357         queued = 0;
 358
 359         while ((req = blk_peek_request(rq)) != NULL) {
 360                 info = req->rq_disk->private_data;
 361
 362                 if (RING_FULL(&info->ring))
 363                         goto wait;
 364
 365                 blk_start_request(req);
 366
 367                 if (req->cmd_type != REQ_TYPE_FS) {
 368                         __blk_end_request_all(req, -EIO);
 369                         continue;
 370                 }
 371
 372                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 373                          "(%u/%u) buffer:%p [%s]\n",
 374                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 375                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 376                          req->buffer, rq_data_dir(req) ? "write" : "read");
 377
 378                 if (blkif_queue_request(req)) {
 379                         blk_requeue_request(rq, req);
 380 wait:
 381                         /* Avoid pointless unplugs. */
 382                         blk_stop_queue(rq);
 383                         break;
 384                 }
 385
 386                 queued++;
 387         }
 388
 389         if (queued != 0)
 390                 flush_requests(info);
 391 }
 392
 393 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 394 {
 395         struct request_queue *rq;
 396
 397         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 398         if (rq == NULL)
 399                 return -1;
 400
 401         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 402
 403         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 404         blk_queue_logical_block_size(rq, sector_size);
 405         blk_queue_max_hw_sectors(rq, 512);
 406
 407         /* Each segment in a request is up to an aligned page in size. */
 408         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 409         blk_queue_max_segment_size(rq, PAGE_SIZE);
 410
 411         /* Ensure a merged request will fit in a single I/O ring slot. */
 412         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 413
 414         /* Make sure buffer addresses are sector-aligned. */
 415         blk_queue_dma_alignment(rq, 511);
 416
 417         /* Make sure we don't use bounce buffers. */
 418         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 419
 420         gd->queue = rq;
 421
 422         return 0;
 423 }
 424
 425
 426 static int xlvbd_barrier(struct blkfront_info *info)
 427 {
 428         int err;
 429
 430         err = blk_queue_ordered(info->rq,
 431                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE);
 432
 433         if (err)
 434                 return err;
 435
 436         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 437                info->gd->disk_name,
 438                info->feature_barrier ? "enabled" : "disabled");
 439         return 0;
 440 }
 441
 442
 443 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 444                                struct blkfront_info *info,
 445                                u16 vdisk_info, u16 sector_size)
 446 {
 447         struct gendisk *gd;
 448         int nr_minors = 1;
 449         int err = -ENODEV;
 450         unsigned int offset;
 451         int minor;
 452         int nr_parts;
 453
 454         BUG_ON(info->gd != NULL);
 455         BUG_ON(info->rq != NULL);
 456
 457         if ((info->vdevice>>EXT_SHIFT) > 1) {
 458                 /* this is above the extended range; something is wrong */
 459                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 460                 return -ENODEV;
 461         }
 462
 463         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 464                 minor = BLKIF_MINOR(info->vdevice);
 465                 nr_parts = PARTS_PER_DISK;
 466         } else {
 467                 minor = BLKIF_MINOR_EXT(info->vdevice);
 468                 nr_parts = PARTS_PER_EXT_DISK;
 469         }
 470
 471         if ((minor % nr_parts) == 0)
 472                 nr_minors = nr_parts;
 473
 474         err = xlbd_reserve_minors(minor, nr_minors);
 475         if (err)
 476                 goto out;
 477         err = -ENODEV;
 478
 479         gd = alloc_disk(nr_minors);
 480         if (gd == NULL)
 481                 goto release;
 482
 483         offset = minor / nr_parts;
 484
 485         if (nr_minors > 1) {
 486                 if (offset < 26)
 487                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 488                 else
 489                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 490                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 491         } else {
 492                 if (offset < 26)
 493                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 494                                 'a' + offset,
 495                                 minor & (nr_parts - 1));
 496                 else
 497                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 498                                 'a' + ((offset / 26) - 1),
 499                                 'a' + (offset % 26),
 500                                 minor & (nr_parts - 1));
 501         }
 502
 503         gd->major = XENVBD_MAJOR;
 504         gd->first_minor = minor;
 505         gd->fops = &xlvbd_block_fops;
 506         gd->private_data = info;
 507         gd->driverfs_dev = &(info->xbdev->dev);
 508         set_capacity(gd, capacity);
 509
 510         if (xlvbd_init_blk_queue(gd, sector_size)) {
 511                 del_gendisk(gd);
 512                 goto release;
 513         }
 514
 515         info->rq = gd->queue;
 516         info->gd = gd;
 517
 518         if (info->feature_barrier)
 519                 xlvbd_barrier(info);
 520
 521         if (vdisk_info & VDISK_READONLY)
 522                 set_disk_ro(gd, 1);
 523
 524         if (vdisk_info & VDISK_REMOVABLE)
 525                 gd->flags |= GENHD_FL_REMOVABLE;
 526
 527         if (vdisk_info & VDISK_CDROM)
 528                 gd->flags |= GENHD_FL_CD;
 529
 530         return 0;
 531
 532  release:
 533         xlbd_release_minors(minor, nr_minors);
 534  out:
 535         return err;
 536 }
 537
 538 static void xlvbd_release_gendisk(struct blkfront_info *info)
 539 {
 540         unsigned int minor, nr_minors;
 541         unsigned long flags;
 542
 543         if (info->rq == NULL)
 544                 return;
 545
 546         spin_lock_irqsave(&blkif_io_lock, flags);
 547
 548         /* No more blkif_request(). */
 549         blk_stop_queue(info->rq);
 550
 551         /* No more gnttab callback work. */
 552         gnttab_cancel_free_callback(&info->callback);
 553         spin_unlock_irqrestore(&blkif_io_lock, flags);
 554
 555         /* Flush gnttab callback work. Must be done with no locks held. */
 556         flush_scheduled_work();
 557
 558         del_gendisk(info->gd);
 559
 560         minor = info->gd->first_minor;
 561         nr_minors = info->gd->minors;
 562         xlbd_release_minors(minor, nr_minors);
 563
 564         blk_cleanup_queue(info->rq);
 565         info->rq = NULL;
 566
 567         put_disk(info->gd);
 568         info->gd = NULL;
 569 }
 570
 571 static void kick_pending_request_queues(struct blkfront_info *info)
 572 {
 573         if (!RING_FULL(&info->ring)) {
 574                 /* Re-enable calldowns. */
 575                 blk_start_queue(info->rq);
 576                 /* Kick things off immediately. */
 577                 do_blkif_request(info->rq);
 578         }
 579 }
 580
 581 static void blkif_restart_queue(struct work_struct *work)
 582 {
 583         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 584
 585         spin_lock_irq(&blkif_io_lock);
 586         if (info->connected == BLKIF_STATE_CONNECTED)
 587                 kick_pending_request_queues(info);
 588         spin_unlock_irq(&blkif_io_lock);
 589 }
 590
 591 static void blkif_free(struct blkfront_info *info, int suspend)
 592 {
 593         /* Prevent new requests being issued until we fix things up. */
 594         spin_lock_irq(&blkif_io_lock);
 595         info->connected = suspend ?
 596                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 597         /* No more blkif_request(). */
 598         if (info->rq)
 599                 blk_stop_queue(info->rq);
 600         /* No more gnttab callback work. */
 601         gnttab_cancel_free_callback(&info->callback);
 602         spin_unlock_irq(&blkif_io_lock);
 603
 604         /* Flush gnttab callback work. Must be done with no locks held. */
 605         flush_scheduled_work();
 606
 607         /* Free resources associated with old device channel. */
 608         if (info->ring_ref != GRANT_INVALID_REF) {
 609                 gnttab_end_foreign_access(info->ring_ref, 0,
 610                                           (unsigned long)info->ring.sring);
 611                 info->ring_ref = GRANT_INVALID_REF;
 612                 info->ring.sring = NULL;
 613         }
 614         if (info->irq)
 615                 unbind_from_irqhandler(info->irq, info);
 616         info->evtchn = info->irq = 0;
 617
 618 }
 619
 620 static void blkif_completion(struct blk_shadow *s)
 621 {
 622         int i;
 623         for (i = 0; i < s->req.nr_segments; i++)
 624                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 625 }
 626
 627 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 628 {
 629         struct request *req;
 630         struct blkif_response *bret;
 631         RING_IDX i, rp;
 632         unsigned long flags;
 633         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 634         int error;
 635
 636         spin_lock_irqsave(&blkif_io_lock, flags);
 637
 638         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 639                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 640                 return IRQ_HANDLED;
 641         }
 642
 643  again:
 644         rp = info->ring.sring->rsp_prod;
 645         rmb(); /* Ensure we see queued responses up to 'rp'. */
 646
 647         for (i = info->ring.rsp_cons; i != rp; i++) {
 648                 unsigned long id;
 649
 650                 bret = RING_GET_RESPONSE(&info->ring, i);
 651                 id   = bret->id;
 652                 req  = (struct request *)info->shadow[id].request;
 653
 654                 blkif_completion(&info->shadow[id]);
 655
 656                 add_id_to_freelist(info, id);
 657
 658                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 659                 switch (bret->operation) {
 660                 case BLKIF_OP_WRITE_BARRIER:
 661                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 662                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 663                                        info->gd->disk_name);
 664                                 error = -EOPNOTSUPP;
 665                                 info->feature_barrier = 0;
 666                                 xlvbd_barrier(info);
 667                         }
 668                         /* fall through */
 669                 case BLKIF_OP_READ:
 670                 case BLKIF_OP_WRITE:
 671                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 672                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 673                                         "request: %x\n", bret->status);
 674
 675                         __blk_end_request_all(req, error);
 676                         break;
 677                 default:
 678                         BUG();
 679                 }
 680         }
 681
 682         info->ring.rsp_cons = i;
 683
 684         if (i != info->ring.req_prod_pvt) {
 685                 int more_to_do;
 686                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 687                 if (more_to_do)
 688                         goto again;
 689         } else
 690                 info->ring.sring->rsp_event = i + 1;
 691
 692         kick_pending_request_queues(info);
 693
 694         spin_unlock_irqrestore(&blkif_io_lock, flags);
 695
 696         return IRQ_HANDLED;
 697 }
 698
 699
 700 static int setup_blkring(struct xenbus_device *dev,
 701                          struct blkfront_info *info)
 702 {
 703         struct blkif_sring *sring;
 704         int err;
 705
 706         info->ring_ref = GRANT_INVALID_REF;
 707
 708         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 709         if (!sring) {
 710                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 711                 return -ENOMEM;
 712         }
 713         SHARED_RING_INIT(sring);
 714         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 715
 716         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 717
 718         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 719         if (err < 0) {
 720                 free_page((unsigned long)sring);
 721                 info->ring.sring = NULL;
 722                 goto fail;
 723         }
 724         info->ring_ref = err;
 725
 726         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 727         if (err)
 728                 goto fail;
 729
 730         err = bind_evtchn_to_irqhandler(info->evtchn,
 731                                         blkif_interrupt,
 732                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 733         if (err <= 0) {
 734                 xenbus_dev_fatal(dev, err,
 735                                  "bind_evtchn_to_irqhandler failed");
 736                 goto fail;
 737         }
 738         info->irq = err;
 739
 740         return 0;
 741 fail:
 742         blkif_free(info, 0);
 743         return err;
 744 }
 745
 746
 747 /* Common code used when first setting up, and when resuming. */
 748 static int talk_to_blkback(struct xenbus_device *dev,
 749                            struct blkfront_info *info)
 750 {
 751         const char *message = NULL;
 752         struct xenbus_transaction xbt;
 753         int err;
 754
 755         /* Create shared ring, alloc event channel. */
 756         err = setup_blkring(dev, info);
 757         if (err)
 758                 goto out;
 759
 760 again:
 761         err = xenbus_transaction_start(&xbt);
 762         if (err) {
 763                 xenbus_dev_fatal(dev, err, "starting transaction");
 764                 goto destroy_blkring;
 765         }
 766
 767         err = xenbus_printf(xbt, dev->nodename,
 768                             "ring-ref", "%u", info->ring_ref);
 769         if (err) {
 770                 message = "writing ring-ref";
 771                 goto abort_transaction;
 772         }
 773         err = xenbus_printf(xbt, dev->nodename,
 774                             "event-channel", "%u", info->evtchn);
 775         if (err) {
 776                 message = "writing event-channel";
 777                 goto abort_transaction;
 778         }
 779         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 780                             XEN_IO_PROTO_ABI_NATIVE);
 781         if (err) {
 782                 message = "writing protocol";
 783                 goto abort_transaction;
 784         }
 785
 786         err = xenbus_transaction_end(xbt, 0);
 787         if (err) {
 788                 if (err == -EAGAIN)
 789                         goto again;
 790                 xenbus_dev_fatal(dev, err, "completing transaction");
 791                 goto destroy_blkring;
 792         }
 793
 794         xenbus_switch_state(dev, XenbusStateInitialised);
 795
 796         return 0;
 797
 798  abort_transaction:
 799         xenbus_transaction_end(xbt, 1);
 800         if (message)
 801                 xenbus_dev_fatal(dev, err, "%s", message);
 802  destroy_blkring:
 803         blkif_free(info, 0);
 804  out:
 805         return err;
 806 }
 807
 808 /**
 809  * Entry point to this code when a new device is created.  Allocate the basic
 810  * structures and the ring buffer for communication with the backend, and
 811  * inform the backend of the appropriate details for those.  Switch to
 812  * Initialised state.
 813  */
 814 static int blkfront_probe(struct xenbus_device *dev,
 815                           const struct xenbus_device_id *id)
 816 {
 817         int err, vdevice, i;
 818         struct blkfront_info *info;
 819
 820         /* FIXME: Use dynamic device id if this is not set. */
 821         err = xenbus_scanf(XBT_NIL, dev->nodename,
 822                            "virtual-device", "%i", &vdevice);
 823         if (err != 1) {
 824                 /* go looking in the extended area instead */
 825                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 826                                    "%i", &vdevice);
 827                 if (err != 1) {
 828                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 829                         return err;
 830                 }
 831         }
 832
 833         info = kzalloc(sizeof(*info), GFP_KERNEL);
 834         if (!info) {
 835                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 836                 return -ENOMEM;
 837         }
 838
 839         mutex_init(&info->mutex);
 840         info->xbdev = dev;
 841         info->vdevice = vdevice;
 842         info->connected = BLKIF_STATE_DISCONNECTED;
 843         INIT_WORK(&info->work, blkif_restart_queue);
 844
 845         for (i = 0; i < BLK_RING_SIZE; i++)
 846                 info->shadow[i].req.id = i+1;
 847         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 848
 849         /* Front end dir is a number, which is used as the id. */
 850         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 851         dev_set_drvdata(&dev->dev, info);
 852
 853         err = talk_to_blkback(dev, info);
 854         if (err) {
 855                 kfree(info);
 856                 dev_set_drvdata(&dev->dev, NULL);
 857                 return err;
 858         }
 859
 860         return 0;
 861 }
 862
 863
 864 static int blkif_recover(struct blkfront_info *info)
 865 {
 866         int i;
 867         struct blkif_request *req;
 868         struct blk_shadow *copy;
 869         int j;
 870
 871         /* Stage 1: Make a safe copy of the shadow state. */
 872         copy = kmalloc(sizeof(info->shadow),
 873                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 874         if (!copy)
 875                 return -ENOMEM;
 876         memcpy(copy, info->shadow, sizeof(info->shadow));
 877
 878         /* Stage 2: Set up free list. */
 879         memset(&info->shadow, 0, sizeof(info->shadow));
 880         for (i = 0; i < BLK_RING_SIZE; i++)
 881                 info->shadow[i].req.id = i+1;
 882         info->shadow_free = info->ring.req_prod_pvt;
 883         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 884
 885         /* Stage 3: Find pending requests and requeue them. */
 886         for (i = 0; i < BLK_RING_SIZE; i++) {
 887                 /* Not in use? */
 888                 if (copy[i].request == 0)
 889                         continue;
 890
 891                 /* Grab a request slot and copy shadow state into it. */
 892                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 893                 *req = copy[i].req;
 894
 895                 /* We get a new request id, and must reset the shadow state. */
 896                 req->id = get_id_from_freelist(info);
 897                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 898
 899                 /* Rewrite any grant references invalidated by susp/resume. */
 900                 for (j = 0; j < req->nr_segments; j++)
 901                         gnttab_grant_foreign_access_ref(
 902                                 req->seg[j].gref,
 903                                 info->xbdev->otherend_id,
 904                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 905                                 rq_data_dir(
 906                                         (struct request *)
 907                                         info->shadow[req->id].request));
 908                 info->shadow[req->id].req = *req;
 909
 910                 info->ring.req_prod_pvt++;
 911         }
 912
 913         kfree(copy);
 914
 915         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 916
 917         spin_lock_irq(&blkif_io_lock);
 918
 919         /* Now safe for us to use the shared ring */
 920         info->connected = BLKIF_STATE_CONNECTED;
 921
 922         /* Send off requeued requests */
 923         flush_requests(info);
 924
 925         /* Kick any other new requests queued since we resumed */
 926         kick_pending_request_queues(info);
 927
 928         spin_unlock_irq(&blkif_io_lock);
 929
 930         return 0;
 931 }
 932
 933 /**
 934  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 935  * driver restart.  We tear down our blkif structure and recreate it, but
 936  * leave the device-layer structures intact so that this is transparent to the
 937  * rest of the kernel.
 938  */
 939 static int blkfront_resume(struct xenbus_device *dev)
 940 {
 941         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 942         int err;
 943
 944         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 945
 946         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 947
 948         err = talk_to_blkback(dev, info);
 949         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 950                 err = blkif_recover(info);
 951
 952         return err;
 953 }
 954
 955 static void
 956 blkfront_closing(struct blkfront_info *info)
 957 {
 958         struct xenbus_device *xbdev = info->xbdev;
 959         struct block_device *bdev = NULL;
 960
 961         mutex_lock(&info->mutex);
 962
 963         if (xbdev->state == XenbusStateClosing) {
 964                 mutex_unlock(&info->mutex);
 965                 return;
 966         }
 967
 968         if (info->gd)
 969                 bdev = bdget_disk(info->gd, 0);
 970
 971         mutex_unlock(&info->mutex);
 972
 973         if (!bdev) {
 974                 xenbus_frontend_closed(xbdev);
 975                 return;
 976         }
 977
 978         mutex_lock(&bdev->bd_mutex);
 979
 980         if (info->users) {
 981                 xenbus_dev_error(xbdev, -EBUSY,
 982                                  "Device in use; refusing to close");
 983                 xenbus_switch_state(xbdev, XenbusStateClosing);
 984         } else {
 985                 xlvbd_release_gendisk(info);
 986                 xenbus_frontend_closed(xbdev);
 987         }
 988
 989         mutex_unlock(&bdev->bd_mutex);
 990         bdput(bdev);
 991 }
 992
 993 /*
 994  * Invoked when the backend is finally 'ready' (and has told produced
 995  * the details about the physical device - #sectors, size, etc).
 996  */
 997 static void blkfront_connect(struct blkfront_info *info)
 998 {
 999         unsigned long long sectors;
1000         unsigned long sector_size;
1001         unsigned int binfo;
1002         int err;
1003
1004         switch (info->connected) {
1005         case BLKIF_STATE_CONNECTED:
1006                 /*
1007                  * Potentially, the back-end may be signalling
1008                  * a capacity change; update the capacity.
1009                  */
1010                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1011                                    "sectors", "%Lu", &sectors);
1012                 if (XENBUS_EXIST_ERR(err))
1013                         return;
1014                 printk(KERN_INFO "Setting capacity to %Lu\n",
1015                        sectors);
1016                 set_capacity(info->gd, sectors);
1017                 revalidate_disk(info->gd);
1018
1019                 /* fall through */
1020         case BLKIF_STATE_SUSPENDED:
1021                 return;
1022
1023         default:
1024                 break;
1025         }
1026
1027         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1028                 __func__, info->xbdev->otherend);
1029
1030         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1031                             "sectors", "%llu", &sectors,
1032                             "info", "%u", &binfo,
1033                             "sector-size", "%lu", &sector_size,
1034                             NULL);
1035         if (err) {
1036                 xenbus_dev_fatal(info->xbdev, err,
1037                                  "reading backend fields at %s",
1038                                  info->xbdev->otherend);
1039                 return;
1040         }
1041
1042         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1043                             "feature-barrier", "%lu", &info->feature_barrier,
1044                             NULL);
1045         if (err)
1046                 info->feature_barrier = 0;
1047
1048         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1049         if (err) {
1050                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1051                                  info->xbdev->otherend);
1052                 return;
1053         }
1054
1055         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1056
1057         /* Kick pending requests. */
1058         spin_lock_irq(&blkif_io_lock);
1059         info->connected = BLKIF_STATE_CONNECTED;
1060         kick_pending_request_queues(info);
1061         spin_unlock_irq(&blkif_io_lock);
1062
1063         add_disk(info->gd);
1064
1065         info->is_ready = 1;
1066 }
1067
1068 /**
1069  * Callback received when the backend's state changes.
1070  */
1071 static void blkback_changed(struct xenbus_device *dev,
1072                             enum xenbus_state backend_state)
1073 {
1074         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1075
1076         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1077
1078         switch (backend_state) {
1079         case XenbusStateInitialising:
1080         case XenbusStateInitWait:
1081         case XenbusStateInitialised:
1082         case XenbusStateUnknown:
1083         case XenbusStateClosed:
1084                 break;
1085
1086         case XenbusStateConnected:
1087                 blkfront_connect(info);
1088                 break;
1089
1090         case XenbusStateClosing:
1091                 blkfront_closing(info);
1092                 break;
1093         }
1094 }
1095
1096 static int blkfront_remove(struct xenbus_device *dev)
1097 {
1098         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1099
1100         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1101
1102         blkif_free(info, 0);
1103
1104         if(info->users == 0)
1105                 kfree(info);
1106         else
1107                 info->xbdev = NULL;
1108
1109         return 0;
1110 }
1111
1112 static int blkfront_is_ready(struct xenbus_device *dev)
1113 {
1114         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1115
1116         return info->is_ready && info->xbdev;
1117 }
1118
1119 static int blkif_open(struct block_device *bdev, fmode_t mode)
1120 {
1121         struct blkfront_info *info = bdev->bd_disk->private_data;
1122
1123         if (!info->xbdev)
1124                 return -ENODEV;
1125
1126         lock_kernel();
1127         info->users++;
1128         unlock_kernel();
1129
1130         return 0;
1131 }
1132
1133 static int blkif_release(struct gendisk *disk, fmode_t mode)
1134 {
1135         struct blkfront_info *info = disk->private_data;
1136         lock_kernel();
1137         info->users--;
1138         if (info->users == 0) {
1139                 /* Check whether we have been instructed to close.  We will
1140                    have ignored this request initially, as the device was
1141                    still mounted. */
1142                 struct xenbus_device *dev = info->xbdev;
1143
1144                 if (!dev) {
1145                         xlvbd_release_gendisk(info);
1146                         kfree(info);
1147                 } else if (xenbus_read_driver_state(dev->otherend)
1148                            == XenbusStateClosing && info->is_ready) {
1149                         xlvbd_release_gendisk(info);
1150                         xenbus_frontend_closed(dev);
1151                 }
1152         }
1153         unlock_kernel();
1154         return 0;
1155 }
1156
1157 static const struct block_device_operations xlvbd_block_fops =
1158 {
1159         .owner = THIS_MODULE,
1160         .open = blkif_open,
1161         .release = blkif_release,
1162         .getgeo = blkif_getgeo,
1163         .ioctl = blkif_ioctl,
1164 };
1165
1166
1167 static const struct xenbus_device_id blkfront_ids[] = {
1168         { "vbd" },
1169         { "" }
1170 };
1171
1172 static struct xenbus_driver blkfront = {
1173         .name = "vbd",
1174         .owner = THIS_MODULE,
1175         .ids = blkfront_ids,
1176         .probe = blkfront_probe,
1177         .remove = blkfront_remove,
1178         .resume = blkfront_resume,
1179         .otherend_changed = blkback_changed,
1180         .is_ready = blkfront_is_ready,
1181 };
1182
1183 static int __init xlblk_init(void)
1184 {
1185         if (!xen_domain())
1186                 return -ENODEV;
1187
1188         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1189                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1190                        XENVBD_MAJOR, DEV_NAME);
1191                 return -ENODEV;
1192         }
1193
1194         return xenbus_register_frontend(&blkfront);
1195 }
1196 module_init(xlblk_init);
1197
1198
1199 static void __exit xlblk_exit(void)
1200 {
1201         return xenbus_unregister_driver(&blkfront);
1202 }
1203 module_exit(xlblk_exit);
1204
1205 MODULE_DESCRIPTION("Xen virtual block device frontend");
1206 MODULE_LICENSE("GPL");
1207 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1208 MODULE_ALIAS("xen:vbd");
1209 MODULE_ALIAS("xenblk");