SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/scatterlist.h>
  44
  45 #include <xen/xenbus.h>
  46 #include <xen/grant_table.h>
  47 #include <xen/events.h>
  48 #include <xen/page.h>
  49
  50 #include <xen/interface/grant_table.h>
  51 #include <xen/interface/io/blkif.h>
  52 #include <xen/interface/io/protocols.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         struct gendisk *gd;
  81         int vdevice;
  82         blkif_vdev_t handle;
  83         enum blkif_state connected;
  84         int ring_ref;
  85         struct blkif_front_ring ring;
  86         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  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         int is_ready;
  95
  96         /**
  97          * The number of people holding this device open.  We won't allow a
  98          * hot-unplug unless this is 0.
  99          */
 100         int users;
 101 };
 102
 103 static DEFINE_SPINLOCK(blkif_io_lock);
 104
 105 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 106         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 107 #define GRANT_INVALID_REF       0
 108
 109 #define PARTS_PER_DISK          16
 110 #define PARTS_PER_EXT_DISK      256
 111
 112 #define BLKIF_MAJOR(dev) ((dev)>>8)
 113 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 114
 115 #define EXT_SHIFT 28
 116 #define EXTENDED (1<<EXT_SHIFT)
 117 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 118 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 119
 120 #define DEV_NAME        "xvd"   /* name in /dev */
 121
 122 static int get_id_from_freelist(struct blkfront_info *info)
 123 {
 124         unsigned long free = info->shadow_free;
 125         BUG_ON(free > BLK_RING_SIZE);
 126         info->shadow_free = info->shadow[free].req.id;
 127         info->shadow[free].req.id = 0x0fffffee; /* debug */
 128         return free;
 129 }
 130
 131 static void add_id_to_freelist(struct blkfront_info *info,
 132                                unsigned long id)
 133 {
 134         info->shadow[id].req.id  = info->shadow_free;
 135         info->shadow[id].request = 0;
 136         info->shadow_free = id;
 137 }
 138
 139 static void blkif_restart_queue_callback(void *arg)
 140 {
 141         struct blkfront_info *info = (struct blkfront_info *)arg;
 142         schedule_work(&info->work);
 143 }
 144
 145 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 146 {
 147         /* We don't have real geometry info, but let's at least return
 148            values consistent with the size of the device */
 149         sector_t nsect = get_capacity(bd->bd_disk);
 150         sector_t cylinders = nsect;
 151
 152         hg->heads = 0xff;
 153         hg->sectors = 0x3f;
 154         sector_div(cylinders, hg->heads * hg->sectors);
 155         hg->cylinders = cylinders;
 156         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 157                 hg->cylinders = 0xffff;
 158         return 0;
 159 }
 160
 161 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 162                        unsigned command, unsigned long argument)
 163 {
 164         struct blkfront_info *info = bdev->bd_disk->private_data;
 165         int i;
 166
 167         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 168                 command, (long)argument);
 169
 170         switch (command) {
 171         case CDROMMULTISESSION:
 172                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 173                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 174                         if (put_user(0, (char __user *)(argument + i)))
 175                                 return -EFAULT;
 176                 return 0;
 177
 178         case CDROM_GET_CAPABILITY: {
 179                 struct gendisk *gd = info->gd;
 180                 if (gd->flags & GENHD_FL_CD)
 181                         return 0;
 182                 return -EINVAL;
 183         }
 184
 185         default:
 186                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 187                   command);*/
 188                 return -EINVAL; /* same return as native Linux */
 189         }
 190
 191         return 0;
 192 }
 193
 194 /*
 195  * blkif_queue_request
 196  *
 197  * request block io
 198  *
 199  * id: for guest use only.
 200  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 201  * buffer: buffer to read/write into. this should be a
 202  *   virtual address in the guest os.
 203  */
 204 static int blkif_queue_request(struct request *req)
 205 {
 206         struct blkfront_info *info = req->rq_disk->private_data;
 207         unsigned long buffer_mfn;
 208         struct blkif_request *ring_req;
 209         unsigned long id;
 210         unsigned int fsect, lsect;
 211         int i, ref;
 212         grant_ref_t gref_head;
 213         struct scatterlist *sg;
 214
 215         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 216                 return 1;
 217
 218         if (gnttab_alloc_grant_references(
 219                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 220                 gnttab_request_free_callback(
 221                         &info->callback,
 222                         blkif_restart_queue_callback,
 223                         info,
 224                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 225                 return 1;
 226         }
 227
 228         /* Fill out a communications ring structure. */
 229         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 230         id = get_id_from_freelist(info);
 231         info->shadow[id].request = (unsigned long)req;
 232
 233         ring_req->id = id;
 234         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
 235         ring_req->handle = info->handle;
 236
 237         ring_req->operation = rq_data_dir(req) ?
 238                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 239         if (blk_barrier_rq(req))
 240                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 241
 242         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
 243         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 244
 245         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
 246                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 247                 fsect = sg->offset >> 9;
 248                 lsect = fsect + (sg->length >> 9) - 1;
 249                 /* install a grant reference. */
 250                 ref = gnttab_claim_grant_reference(&gref_head);
 251                 BUG_ON(ref == -ENOSPC);
 252
 253                 gnttab_grant_foreign_access_ref(
 254                                 ref,
 255                                 info->xbdev->otherend_id,
 256                                 buffer_mfn,
 257                                 rq_data_dir(req) );
 258
 259                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 260                 ring_req->seg[i] =
 261                                 (struct blkif_request_segment) {
 262                                         .gref       = ref,
 263                                         .first_sect = fsect,
 264                                         .last_sect  = lsect };
 265         }
 266
 267         info->ring.req_prod_pvt++;
 268
 269         /* Keep a private copy so we can reissue requests when recovering. */
 270         info->shadow[id].req = *ring_req;
 271
 272         gnttab_free_grant_references(gref_head);
 273
 274         return 0;
 275 }
 276
 277
 278 static inline void flush_requests(struct blkfront_info *info)
 279 {
 280         int notify;
 281
 282         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 283
 284         if (notify)
 285                 notify_remote_via_irq(info->irq);
 286 }
 287
 288 /*
 289  * do_blkif_request
 290  *  read a block; request is in a request queue
 291  */
 292 static void do_blkif_request(struct request_queue *rq)
 293 {
 294         struct blkfront_info *info = NULL;
 295         struct request *req;
 296         int queued;
 297
 298         pr_debug("Entered do_blkif_request\n");
 299
 300         queued = 0;
 301
 302         while ((req = elv_next_request(rq)) != NULL) {
 303                 info = req->rq_disk->private_data;
 304                 if (!blk_fs_request(req)) {
 305                         __blk_end_request_cur(req, -EIO);
 306                         continue;
 307                 }
 308
 309                 if (RING_FULL(&info->ring))
 310                         goto wait;
 311
 312                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 313                          "(%u/%u) buffer:%p [%s]\n",
 314                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 315                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 316                          req->buffer, rq_data_dir(req) ? "write" : "read");
 317
 318
 319                 blkdev_dequeue_request(req);
 320                 if (blkif_queue_request(req)) {
 321                         blk_requeue_request(rq, req);
 322 wait:
 323                         /* Avoid pointless unplugs. */
 324                         blk_stop_queue(rq);
 325                         break;
 326                 }
 327
 328                 queued++;
 329         }
 330
 331         if (queued != 0)
 332                 flush_requests(info);
 333 }
 334
 335 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 336 {
 337         struct request_queue *rq;
 338
 339         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 340         if (rq == NULL)
 341                 return -1;
 342
 343         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 344
 345         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 346         blk_queue_hardsect_size(rq, sector_size);
 347         blk_queue_max_sectors(rq, 512);
 348
 349         /* Each segment in a request is up to an aligned page in size. */
 350         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 351         blk_queue_max_segment_size(rq, PAGE_SIZE);
 352
 353         /* Ensure a merged request will fit in a single I/O ring slot. */
 354         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 355         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 356
 357         /* Make sure buffer addresses are sector-aligned. */
 358         blk_queue_dma_alignment(rq, 511);
 359
 360         /* Make sure we don't use bounce buffers. */
 361         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 362
 363         gd->queue = rq;
 364
 365         return 0;
 366 }
 367
 368
 369 static int xlvbd_barrier(struct blkfront_info *info)
 370 {
 371         int err;
 372
 373         err = blk_queue_ordered(info->rq,
 374                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 375                                 NULL);
 376
 377         if (err)
 378                 return err;
 379
 380         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 381                info->gd->disk_name,
 382                info->feature_barrier ? "enabled" : "disabled");
 383         return 0;
 384 }
 385
 386
 387 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 388                                struct blkfront_info *info,
 389                                u16 vdisk_info, u16 sector_size)
 390 {
 391         struct gendisk *gd;
 392         int nr_minors = 1;
 393         int err = -ENODEV;
 394         unsigned int offset;
 395         int minor;
 396         int nr_parts;
 397
 398         BUG_ON(info->gd != NULL);
 399         BUG_ON(info->rq != NULL);
 400
 401         if ((info->vdevice>>EXT_SHIFT) > 1) {
 402                 /* this is above the extended range; something is wrong */
 403                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 404                 return -ENODEV;
 405         }
 406
 407         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 408                 minor = BLKIF_MINOR(info->vdevice);
 409                 nr_parts = PARTS_PER_DISK;
 410         } else {
 411                 minor = BLKIF_MINOR_EXT(info->vdevice);
 412                 nr_parts = PARTS_PER_EXT_DISK;
 413         }
 414
 415         if ((minor % nr_parts) == 0)
 416                 nr_minors = nr_parts;
 417
 418         gd = alloc_disk(nr_minors);
 419         if (gd == NULL)
 420                 goto out;
 421
 422         offset = minor / nr_parts;
 423
 424         if (nr_minors > 1) {
 425                 if (offset < 26)
 426                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 427                 else
 428                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 429                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 430         } else {
 431                 if (offset < 26)
 432                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 433                                 'a' + offset,
 434                                 minor & (nr_parts - 1));
 435                 else
 436                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 437                                 'a' + ((offset / 26) - 1),
 438                                 'a' + (offset % 26),
 439                                 minor & (nr_parts - 1));
 440         }
 441
 442         gd->major = XENVBD_MAJOR;
 443         gd->first_minor = minor;
 444         gd->fops = &xlvbd_block_fops;
 445         gd->private_data = info;
 446         gd->driverfs_dev = &(info->xbdev->dev);
 447         set_capacity(gd, capacity);
 448
 449         if (xlvbd_init_blk_queue(gd, sector_size)) {
 450                 del_gendisk(gd);
 451                 goto out;
 452         }
 453
 454         info->rq = gd->queue;
 455         info->gd = gd;
 456
 457         if (info->feature_barrier)
 458                 xlvbd_barrier(info);
 459
 460         if (vdisk_info & VDISK_READONLY)
 461                 set_disk_ro(gd, 1);
 462
 463         if (vdisk_info & VDISK_REMOVABLE)
 464                 gd->flags |= GENHD_FL_REMOVABLE;
 465
 466         if (vdisk_info & VDISK_CDROM)
 467                 gd->flags |= GENHD_FL_CD;
 468
 469         return 0;
 470
 471  out:
 472         return err;
 473 }
 474
 475 static void kick_pending_request_queues(struct blkfront_info *info)
 476 {
 477         if (!RING_FULL(&info->ring)) {
 478                 /* Re-enable calldowns. */
 479                 blk_start_queue(info->rq);
 480                 /* Kick things off immediately. */
 481                 do_blkif_request(info->rq);
 482         }
 483 }
 484
 485 static void blkif_restart_queue(struct work_struct *work)
 486 {
 487         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 488
 489         spin_lock_irq(&blkif_io_lock);
 490         if (info->connected == BLKIF_STATE_CONNECTED)
 491                 kick_pending_request_queues(info);
 492         spin_unlock_irq(&blkif_io_lock);
 493 }
 494
 495 static void blkif_free(struct blkfront_info *info, int suspend)
 496 {
 497         /* Prevent new requests being issued until we fix things up. */
 498         spin_lock_irq(&blkif_io_lock);
 499         info->connected = suspend ?
 500                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 501         /* No more blkif_request(). */
 502         if (info->rq)
 503                 blk_stop_queue(info->rq);
 504         /* No more gnttab callback work. */
 505         gnttab_cancel_free_callback(&info->callback);
 506         spin_unlock_irq(&blkif_io_lock);
 507
 508         /* Flush gnttab callback work. Must be done with no locks held. */
 509         flush_scheduled_work();
 510
 511         /* Free resources associated with old device channel. */
 512         if (info->ring_ref != GRANT_INVALID_REF) {
 513                 gnttab_end_foreign_access(info->ring_ref, 0,
 514                                           (unsigned long)info->ring.sring);
 515                 info->ring_ref = GRANT_INVALID_REF;
 516                 info->ring.sring = NULL;
 517         }
 518         if (info->irq)
 519                 unbind_from_irqhandler(info->irq, info);
 520         info->evtchn = info->irq = 0;
 521
 522 }
 523
 524 static void blkif_completion(struct blk_shadow *s)
 525 {
 526         int i;
 527         for (i = 0; i < s->req.nr_segments; i++)
 528                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 529 }
 530
 531 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 532 {
 533         struct request *req;
 534         struct blkif_response *bret;
 535         RING_IDX i, rp;
 536         unsigned long flags;
 537         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 538         int error;
 539
 540         spin_lock_irqsave(&blkif_io_lock, flags);
 541
 542         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 543                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 544                 return IRQ_HANDLED;
 545         }
 546
 547  again:
 548         rp = info->ring.sring->rsp_prod;
 549         rmb(); /* Ensure we see queued responses up to 'rp'. */
 550
 551         for (i = info->ring.rsp_cons; i != rp; i++) {
 552                 unsigned long id;
 553
 554                 bret = RING_GET_RESPONSE(&info->ring, i);
 555                 id   = bret->id;
 556                 req  = (struct request *)info->shadow[id].request;
 557
 558                 blkif_completion(&info->shadow[id]);
 559
 560                 add_id_to_freelist(info, id);
 561
 562                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 563                 switch (bret->operation) {
 564                 case BLKIF_OP_WRITE_BARRIER:
 565                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 566                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 567                                        info->gd->disk_name);
 568                                 error = -EOPNOTSUPP;
 569                                 info->feature_barrier = 0;
 570                                 xlvbd_barrier(info);
 571                         }
 572                         /* fall through */
 573                 case BLKIF_OP_READ:
 574                 case BLKIF_OP_WRITE:
 575                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 576                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 577                                         "request: %x\n", bret->status);
 578
 579                         __blk_end_request_all(req, error);
 580                         break;
 581                 default:
 582                         BUG();
 583                 }
 584         }
 585
 586         info->ring.rsp_cons = i;
 587
 588         if (i != info->ring.req_prod_pvt) {
 589                 int more_to_do;
 590                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 591                 if (more_to_do)
 592                         goto again;
 593         } else
 594                 info->ring.sring->rsp_event = i + 1;
 595
 596         kick_pending_request_queues(info);
 597
 598         spin_unlock_irqrestore(&blkif_io_lock, flags);
 599
 600         return IRQ_HANDLED;
 601 }
 602
 603
 604 static int setup_blkring(struct xenbus_device *dev,
 605                          struct blkfront_info *info)
 606 {
 607         struct blkif_sring *sring;
 608         int err;
 609
 610         info->ring_ref = GRANT_INVALID_REF;
 611
 612         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 613         if (!sring) {
 614                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 615                 return -ENOMEM;
 616         }
 617         SHARED_RING_INIT(sring);
 618         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 619
 620         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 621
 622         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 623         if (err < 0) {
 624                 free_page((unsigned long)sring);
 625                 info->ring.sring = NULL;
 626                 goto fail;
 627         }
 628         info->ring_ref = err;
 629
 630         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 631         if (err)
 632                 goto fail;
 633
 634         err = bind_evtchn_to_irqhandler(info->evtchn,
 635                                         blkif_interrupt,
 636                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 637         if (err <= 0) {
 638                 xenbus_dev_fatal(dev, err,
 639                                  "bind_evtchn_to_irqhandler failed");
 640                 goto fail;
 641         }
 642         info->irq = err;
 643
 644         return 0;
 645 fail:
 646         blkif_free(info, 0);
 647         return err;
 648 }
 649
 650
 651 /* Common code used when first setting up, and when resuming. */
 652 static int talk_to_backend(struct xenbus_device *dev,
 653                            struct blkfront_info *info)
 654 {
 655         const char *message = NULL;
 656         struct xenbus_transaction xbt;
 657         int err;
 658
 659         /* Create shared ring, alloc event channel. */
 660         err = setup_blkring(dev, info);
 661         if (err)
 662                 goto out;
 663
 664 again:
 665         err = xenbus_transaction_start(&xbt);
 666         if (err) {
 667                 xenbus_dev_fatal(dev, err, "starting transaction");
 668                 goto destroy_blkring;
 669         }
 670
 671         err = xenbus_printf(xbt, dev->nodename,
 672                             "ring-ref", "%u", info->ring_ref);
 673         if (err) {
 674                 message = "writing ring-ref";
 675                 goto abort_transaction;
 676         }
 677         err = xenbus_printf(xbt, dev->nodename,
 678                             "event-channel", "%u", info->evtchn);
 679         if (err) {
 680                 message = "writing event-channel";
 681                 goto abort_transaction;
 682         }
 683         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 684                             XEN_IO_PROTO_ABI_NATIVE);
 685         if (err) {
 686                 message = "writing protocol";
 687                 goto abort_transaction;
 688         }
 689
 690         err = xenbus_transaction_end(xbt, 0);
 691         if (err) {
 692                 if (err == -EAGAIN)
 693                         goto again;
 694                 xenbus_dev_fatal(dev, err, "completing transaction");
 695                 goto destroy_blkring;
 696         }
 697
 698         xenbus_switch_state(dev, XenbusStateInitialised);
 699
 700         return 0;
 701
 702  abort_transaction:
 703         xenbus_transaction_end(xbt, 1);
 704         if (message)
 705                 xenbus_dev_fatal(dev, err, "%s", message);
 706  destroy_blkring:
 707         blkif_free(info, 0);
 708  out:
 709         return err;
 710 }
 711
 712
 713 /**
 714  * Entry point to this code when a new device is created.  Allocate the basic
 715  * structures and the ring buffer for communication with the backend, and
 716  * inform the backend of the appropriate details for those.  Switch to
 717  * Initialised state.
 718  */
 719 static int blkfront_probe(struct xenbus_device *dev,
 720                           const struct xenbus_device_id *id)
 721 {
 722         int err, vdevice, i;
 723         struct blkfront_info *info;
 724
 725         /* FIXME: Use dynamic device id if this is not set. */
 726         err = xenbus_scanf(XBT_NIL, dev->nodename,
 727                            "virtual-device", "%i", &vdevice);
 728         if (err != 1) {
 729                 /* go looking in the extended area instead */
 730                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 731                                    "%i", &vdevice);
 732                 if (err != 1) {
 733                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 734                         return err;
 735                 }
 736         }
 737
 738         info = kzalloc(sizeof(*info), GFP_KERNEL);
 739         if (!info) {
 740                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 741                 return -ENOMEM;
 742         }
 743
 744         info->xbdev = dev;
 745         info->vdevice = vdevice;
 746         info->connected = BLKIF_STATE_DISCONNECTED;
 747         INIT_WORK(&info->work, blkif_restart_queue);
 748
 749         for (i = 0; i < BLK_RING_SIZE; i++)
 750                 info->shadow[i].req.id = i+1;
 751         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 752
 753         /* Front end dir is a number, which is used as the id. */
 754         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 755         dev->dev.driver_data = info;
 756
 757         err = talk_to_backend(dev, info);
 758         if (err) {
 759                 kfree(info);
 760                 dev->dev.driver_data = NULL;
 761                 return err;
 762         }
 763
 764         return 0;
 765 }
 766
 767
 768 static int blkif_recover(struct blkfront_info *info)
 769 {
 770         int i;
 771         struct blkif_request *req;
 772         struct blk_shadow *copy;
 773         int j;
 774
 775         /* Stage 1: Make a safe copy of the shadow state. */
 776         copy = kmalloc(sizeof(info->shadow),
 777                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 778         if (!copy)
 779                 return -ENOMEM;
 780         memcpy(copy, info->shadow, sizeof(info->shadow));
 781
 782         /* Stage 2: Set up free list. */
 783         memset(&info->shadow, 0, sizeof(info->shadow));
 784         for (i = 0; i < BLK_RING_SIZE; i++)
 785                 info->shadow[i].req.id = i+1;
 786         info->shadow_free = info->ring.req_prod_pvt;
 787         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 788
 789         /* Stage 3: Find pending requests and requeue them. */
 790         for (i = 0; i < BLK_RING_SIZE; i++) {
 791                 /* Not in use? */
 792                 if (copy[i].request == 0)
 793                         continue;
 794
 795                 /* Grab a request slot and copy shadow state into it. */
 796                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 797                 *req = copy[i].req;
 798
 799                 /* We get a new request id, and must reset the shadow state. */
 800                 req->id = get_id_from_freelist(info);
 801                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 802
 803                 /* Rewrite any grant references invalidated by susp/resume. */
 804                 for (j = 0; j < req->nr_segments; j++)
 805                         gnttab_grant_foreign_access_ref(
 806                                 req->seg[j].gref,
 807                                 info->xbdev->otherend_id,
 808                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 809                                 rq_data_dir(
 810                                         (struct request *)
 811                                         info->shadow[req->id].request));
 812                 info->shadow[req->id].req = *req;
 813
 814                 info->ring.req_prod_pvt++;
 815         }
 816
 817         kfree(copy);
 818
 819         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 820
 821         spin_lock_irq(&blkif_io_lock);
 822
 823         /* Now safe for us to use the shared ring */
 824         info->connected = BLKIF_STATE_CONNECTED;
 825
 826         /* Send off requeued requests */
 827         flush_requests(info);
 828
 829         /* Kick any other new requests queued since we resumed */
 830         kick_pending_request_queues(info);
 831
 832         spin_unlock_irq(&blkif_io_lock);
 833
 834         return 0;
 835 }
 836
 837 /**
 838  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 839  * driver restart.  We tear down our blkif structure and recreate it, but
 840  * leave the device-layer structures intact so that this is transparent to the
 841  * rest of the kernel.
 842  */
 843 static int blkfront_resume(struct xenbus_device *dev)
 844 {
 845         struct blkfront_info *info = dev->dev.driver_data;
 846         int err;
 847
 848         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 849
 850         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 851
 852         err = talk_to_backend(dev, info);
 853         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 854                 err = blkif_recover(info);
 855
 856         return err;
 857 }
 858
 859
 860 /*
 861  * Invoked when the backend is finally 'ready' (and has told produced
 862  * the details about the physical device - #sectors, size, etc).
 863  */
 864 static void blkfront_connect(struct blkfront_info *info)
 865 {
 866         unsigned long long sectors;
 867         unsigned long sector_size;
 868         unsigned int binfo;
 869         int err;
 870
 871         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 872             (info->connected == BLKIF_STATE_SUSPENDED) )
 873                 return;
 874
 875         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 876                 __func__, info->xbdev->otherend);
 877
 878         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 879                             "sectors", "%llu", &sectors,
 880                             "info", "%u", &binfo,
 881                             "sector-size", "%lu", &sector_size,
 882                             NULL);
 883         if (err) {
 884                 xenbus_dev_fatal(info->xbdev, err,
 885                                  "reading backend fields at %s",
 886                                  info->xbdev->otherend);
 887                 return;
 888         }
 889
 890         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 891                             "feature-barrier", "%lu", &info->feature_barrier,
 892                             NULL);
 893         if (err)
 894                 info->feature_barrier = 0;
 895
 896         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
 897         if (err) {
 898                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 899                                  info->xbdev->otherend);
 900                 return;
 901         }
 902
 903         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 904
 905         /* Kick pending requests. */
 906         spin_lock_irq(&blkif_io_lock);
 907         info->connected = BLKIF_STATE_CONNECTED;
 908         kick_pending_request_queues(info);
 909         spin_unlock_irq(&blkif_io_lock);
 910
 911         add_disk(info->gd);
 912
 913         info->is_ready = 1;
 914 }
 915
 916 /**
 917  * Handle the change of state of the backend to Closing.  We must delete our
 918  * device-layer structures now, to ensure that writes are flushed through to
 919  * the backend.  Once is this done, we can switch to Closed in
 920  * acknowledgement.
 921  */
 922 static void blkfront_closing(struct xenbus_device *dev)
 923 {
 924         struct blkfront_info *info = dev->dev.driver_data;
 925         unsigned long flags;
 926
 927         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 928
 929         if (info->rq == NULL)
 930                 goto out;
 931
 932         spin_lock_irqsave(&blkif_io_lock, flags);
 933
 934         del_gendisk(info->gd);
 935
 936         /* No more blkif_request(). */
 937         blk_stop_queue(info->rq);
 938
 939         /* No more gnttab callback work. */
 940         gnttab_cancel_free_callback(&info->callback);
 941         spin_unlock_irqrestore(&blkif_io_lock, flags);
 942
 943         /* Flush gnttab callback work. Must be done with no locks held. */
 944         flush_scheduled_work();
 945
 946         blk_cleanup_queue(info->rq);
 947         info->rq = NULL;
 948
 949  out:
 950         xenbus_frontend_closed(dev);
 951 }
 952
 953 /**
 954  * Callback received when the backend's state changes.
 955  */
 956 static void backend_changed(struct xenbus_device *dev,
 957                             enum xenbus_state backend_state)
 958 {
 959         struct blkfront_info *info = dev->dev.driver_data;
 960         struct block_device *bd;
 961
 962         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 963
 964         switch (backend_state) {
 965         case XenbusStateInitialising:
 966         case XenbusStateInitWait:
 967         case XenbusStateInitialised:
 968         case XenbusStateUnknown:
 969         case XenbusStateClosed:
 970                 break;
 971
 972         case XenbusStateConnected:
 973                 blkfront_connect(info);
 974                 break;
 975
 976         case XenbusStateClosing:
 977                 if (info->gd == NULL)
 978                         xenbus_dev_fatal(dev, -ENODEV, "gd is NULL");
 979                 bd = bdget_disk(info->gd, 0);
 980                 if (bd == NULL)
 981                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 982
 983                 mutex_lock(&bd->bd_mutex);
 984                 if (info->users > 0)
 985                         xenbus_dev_error(dev, -EBUSY,
 986                                          "Device in use; refusing to close");
 987                 else
 988                         blkfront_closing(dev);
 989                 mutex_unlock(&bd->bd_mutex);
 990                 bdput(bd);
 991                 break;
 992         }
 993 }
 994
 995 static int blkfront_remove(struct xenbus_device *dev)
 996 {
 997         struct blkfront_info *info = dev->dev.driver_data;
 998
 999         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1000
1001         blkif_free(info, 0);
1002
1003         kfree(info);
1004
1005         return 0;
1006 }
1007
1008 static int blkfront_is_ready(struct xenbus_device *dev)
1009 {
1010         struct blkfront_info *info = dev->dev.driver_data;
1011
1012         return info->is_ready;
1013 }
1014
1015 static int blkif_open(struct block_device *bdev, fmode_t mode)
1016 {
1017         struct blkfront_info *info = bdev->bd_disk->private_data;
1018         info->users++;
1019         return 0;
1020 }
1021
1022 static int blkif_release(struct gendisk *disk, fmode_t mode)
1023 {
1024         struct blkfront_info *info = disk->private_data;
1025         info->users--;
1026         if (info->users == 0) {
1027                 /* Check whether we have been instructed to close.  We will
1028                    have ignored this request initially, as the device was
1029                    still mounted. */
1030                 struct xenbus_device *dev = info->xbdev;
1031                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1032
1033                 if (state == XenbusStateClosing && info->is_ready)
1034                         blkfront_closing(dev);
1035         }
1036         return 0;
1037 }
1038
1039 static struct block_device_operations xlvbd_block_fops =
1040 {
1041         .owner = THIS_MODULE,
1042         .open = blkif_open,
1043         .release = blkif_release,
1044         .getgeo = blkif_getgeo,
1045         .locked_ioctl = blkif_ioctl,
1046 };
1047
1048
1049 static struct xenbus_device_id blkfront_ids[] = {
1050         { "vbd" },
1051         { "" }
1052 };
1053
1054 static struct xenbus_driver blkfront = {
1055         .name = "vbd",
1056         .owner = THIS_MODULE,
1057         .ids = blkfront_ids,
1058         .probe = blkfront_probe,
1059         .remove = blkfront_remove,
1060         .resume = blkfront_resume,
1061         .otherend_changed = backend_changed,
1062         .is_ready = blkfront_is_ready,
1063 };
1064
1065 static int __init xlblk_init(void)
1066 {
1067         if (!xen_domain())
1068                 return -ENODEV;
1069
1070         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1071                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1072                        XENVBD_MAJOR, DEV_NAME);
1073                 return -ENODEV;
1074         }
1075
1076         return xenbus_register_frontend(&blkfront);
1077 }
1078 module_init(xlblk_init);
1079
1080
1081 static void __exit xlblk_exit(void)
1082 {
1083         return xenbus_unregister_driver(&blkfront);
1084 }
1085 module_exit(xlblk_exit);
1086
1087 MODULE_DESCRIPTION("Xen virtual block device frontend");
1088 MODULE_LICENSE("GPL");
1089 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1090 MODULE_ALIAS("xen:vbd");
1091 MODULE_ALIAS("xenblk");