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)req->sector;
 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                         end_request(req, 0);
 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/%li) buffer:%p [%s]\n",
 314                          req, req->cmd, (unsigned long)req->sector,
 315                          req->current_nr_sectors,
 316                          req->nr_sectors, req->buffer,
 317                          rq_data_dir(req) ? "write" : "read");
 318
 319
 320                 blkdev_dequeue_request(req);
 321                 if (blkif_queue_request(req)) {
 322                         blk_requeue_request(rq, req);
 323 wait:
 324                         /* Avoid pointless unplugs. */
 325                         blk_stop_queue(rq);
 326                         break;
 327                 }
 328
 329                 queued++;
 330         }
 331
 332         if (queued != 0)
 333                 flush_requests(info);
 334 }
 335
 336 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 337 {
 338         struct request_queue *rq;
 339
 340         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 341         if (rq == NULL)
 342                 return -1;
 343
 344         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 345
 346         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 347         blk_queue_hardsect_size(rq, sector_size);
 348         blk_queue_max_sectors(rq, 512);
 349
 350         /* Each segment in a request is up to an aligned page in size. */
 351         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 352         blk_queue_max_segment_size(rq, PAGE_SIZE);
 353
 354         /* Ensure a merged request will fit in a single I/O ring slot. */
 355         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 356         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 357
 358         /* Make sure buffer addresses are sector-aligned. */
 359         blk_queue_dma_alignment(rq, 511);
 360
 361         /* Make sure we don't use bounce buffers. */
 362         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 363
 364         gd->queue = rq;
 365
 366         return 0;
 367 }
 368
 369
 370 static int xlvbd_barrier(struct blkfront_info *info)
 371 {
 372         int err;
 373
 374         err = blk_queue_ordered(info->rq,
 375                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 376                                 NULL);
 377
 378         if (err)
 379                 return err;
 380
 381         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 382                info->gd->disk_name,
 383                info->feature_barrier ? "enabled" : "disabled");
 384         return 0;
 385 }
 386
 387
 388 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 389                                struct blkfront_info *info,
 390                                u16 vdisk_info, u16 sector_size)
 391 {
 392         struct gendisk *gd;
 393         int nr_minors = 1;
 394         int err = -ENODEV;
 395         unsigned int offset;
 396         int minor;
 397         int nr_parts;
 398
 399         BUG_ON(info->gd != NULL);
 400         BUG_ON(info->rq != NULL);
 401
 402         if ((info->vdevice>>EXT_SHIFT) > 1) {
 403                 /* this is above the extended range; something is wrong */
 404                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 405                 return -ENODEV;
 406         }
 407
 408         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 409                 minor = BLKIF_MINOR(info->vdevice);
 410                 nr_parts = PARTS_PER_DISK;
 411         } else {
 412                 minor = BLKIF_MINOR_EXT(info->vdevice);
 413                 nr_parts = PARTS_PER_EXT_DISK;
 414         }
 415
 416         if ((minor % nr_parts) == 0)
 417                 nr_minors = nr_parts;
 418
 419         gd = alloc_disk(nr_minors);
 420         if (gd == NULL)
 421                 goto out;
 422
 423         offset = minor / nr_parts;
 424
 425         if (nr_minors > 1) {
 426                 if (offset < 26)
 427                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 428                 else
 429                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 430                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 431         } else {
 432                 if (offset < 26)
 433                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 434                                 'a' + offset,
 435                                 minor & (nr_parts - 1));
 436                 else
 437                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 438                                 'a' + ((offset / 26) - 1),
 439                                 'a' + (offset % 26),
 440                                 minor & (nr_parts - 1));
 441         }
 442
 443         gd->major = XENVBD_MAJOR;
 444         gd->first_minor = minor;
 445         gd->fops = &xlvbd_block_fops;
 446         gd->private_data = info;
 447         gd->driverfs_dev = &(info->xbdev->dev);
 448         set_capacity(gd, capacity);
 449
 450         if (xlvbd_init_blk_queue(gd, sector_size)) {
 451                 del_gendisk(gd);
 452                 goto out;
 453         }
 454
 455         info->rq = gd->queue;
 456         info->gd = gd;
 457
 458         if (info->feature_barrier)
 459                 xlvbd_barrier(info);
 460
 461         if (vdisk_info & VDISK_READONLY)
 462                 set_disk_ro(gd, 1);
 463
 464         if (vdisk_info & VDISK_REMOVABLE)
 465                 gd->flags |= GENHD_FL_REMOVABLE;
 466
 467         if (vdisk_info & VDISK_CDROM)
 468                 gd->flags |= GENHD_FL_CD;
 469
 470         return 0;
 471
 472  out:
 473         return err;
 474 }
 475
 476 static void kick_pending_request_queues(struct blkfront_info *info)
 477 {
 478         if (!RING_FULL(&info->ring)) {
 479                 /* Re-enable calldowns. */
 480                 blk_start_queue(info->rq);
 481                 /* Kick things off immediately. */
 482                 do_blkif_request(info->rq);
 483         }
 484 }
 485
 486 static void blkif_restart_queue(struct work_struct *work)
 487 {
 488         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 489
 490         spin_lock_irq(&blkif_io_lock);
 491         if (info->connected == BLKIF_STATE_CONNECTED)
 492                 kick_pending_request_queues(info);
 493         spin_unlock_irq(&blkif_io_lock);
 494 }
 495
 496 static void blkif_free(struct blkfront_info *info, int suspend)
 497 {
 498         /* Prevent new requests being issued until we fix things up. */
 499         spin_lock_irq(&blkif_io_lock);
 500         info->connected = suspend ?
 501                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 502         /* No more blkif_request(). */
 503         if (info->rq)
 504                 blk_stop_queue(info->rq);
 505         /* No more gnttab callback work. */
 506         gnttab_cancel_free_callback(&info->callback);
 507         spin_unlock_irq(&blkif_io_lock);
 508
 509         /* Flush gnttab callback work. Must be done with no locks held. */
 510         flush_scheduled_work();
 511
 512         /* Free resources associated with old device channel. */
 513         if (info->ring_ref != GRANT_INVALID_REF) {
 514                 gnttab_end_foreign_access(info->ring_ref, 0,
 515                                           (unsigned long)info->ring.sring);
 516                 info->ring_ref = GRANT_INVALID_REF;
 517                 info->ring.sring = NULL;
 518         }
 519         if (info->irq)
 520                 unbind_from_irqhandler(info->irq, info);
 521         info->evtchn = info->irq = 0;
 522
 523 }
 524
 525 static void blkif_completion(struct blk_shadow *s)
 526 {
 527         int i;
 528         for (i = 0; i < s->req.nr_segments; i++)
 529                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 530 }
 531
 532 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 533 {
 534         struct request *req;
 535         struct blkif_response *bret;
 536         RING_IDX i, rp;
 537         unsigned long flags;
 538         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 539         int error;
 540
 541         spin_lock_irqsave(&blkif_io_lock, flags);
 542
 543         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 544                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 545                 return IRQ_HANDLED;
 546         }
 547
 548  again:
 549         rp = info->ring.sring->rsp_prod;
 550         rmb(); /* Ensure we see queued responses up to 'rp'. */
 551
 552         for (i = info->ring.rsp_cons; i != rp; i++) {
 553                 unsigned long id;
 554                 int ret;
 555
 556                 bret = RING_GET_RESPONSE(&info->ring, i);
 557                 id   = bret->id;
 558                 req  = (struct request *)info->shadow[id].request;
 559
 560                 blkif_completion(&info->shadow[id]);
 561
 562                 add_id_to_freelist(info, id);
 563
 564                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 565                 switch (bret->operation) {
 566                 case BLKIF_OP_WRITE_BARRIER:
 567                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 568                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 569                                        info->gd->disk_name);
 570                                 error = -EOPNOTSUPP;
 571                                 info->feature_barrier = 0;
 572                                 xlvbd_barrier(info);
 573                         }
 574                         /* fall through */
 575                 case BLKIF_OP_READ:
 576                 case BLKIF_OP_WRITE:
 577                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 578                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 579                                         "request: %x\n", bret->status);
 580
 581                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
 582                         BUG_ON(ret);
 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         info = kzalloc(sizeof(*info), GFP_KERNEL);
 742         if (!info) {
 743                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 744                 return -ENOMEM;
 745         }
 746
 747         info->xbdev = dev;
 748         info->vdevice = vdevice;
 749         info->connected = BLKIF_STATE_DISCONNECTED;
 750         INIT_WORK(&info->work, blkif_restart_queue);
 751
 752         for (i = 0; i < BLK_RING_SIZE; i++)
 753                 info->shadow[i].req.id = i+1;
 754         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 755
 756         /* Front end dir is a number, which is used as the id. */
 757         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 758         dev->dev.driver_data = info;
 759
 760         err = talk_to_backend(dev, info);
 761         if (err) {
 762                 kfree(info);
 763                 dev->dev.driver_data = NULL;
 764                 return err;
 765         }
 766
 767         return 0;
 768 }
 769
 770
 771 static int blkif_recover(struct blkfront_info *info)
 772 {
 773         int i;
 774         struct blkif_request *req;
 775         struct blk_shadow *copy;
 776         int j;
 777
 778         /* Stage 1: Make a safe copy of the shadow state. */
 779         copy = kmalloc(sizeof(info->shadow),
 780                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 781         if (!copy)
 782                 return -ENOMEM;
 783         memcpy(copy, info->shadow, sizeof(info->shadow));
 784
 785         /* Stage 2: Set up free list. */
 786         memset(&info->shadow, 0, sizeof(info->shadow));
 787         for (i = 0; i < BLK_RING_SIZE; i++)
 788                 info->shadow[i].req.id = i+1;
 789         info->shadow_free = info->ring.req_prod_pvt;
 790         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 791
 792         /* Stage 3: Find pending requests and requeue them. */
 793         for (i = 0; i < BLK_RING_SIZE; i++) {
 794                 /* Not in use? */
 795                 if (copy[i].request == 0)
 796                         continue;
 797
 798                 /* Grab a request slot and copy shadow state into it. */
 799                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 800                 *req = copy[i].req;
 801
 802                 /* We get a new request id, and must reset the shadow state. */
 803                 req->id = get_id_from_freelist(info);
 804                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 805
 806                 /* Rewrite any grant references invalidated by susp/resume. */
 807                 for (j = 0; j < req->nr_segments; j++)
 808                         gnttab_grant_foreign_access_ref(
 809                                 req->seg[j].gref,
 810                                 info->xbdev->otherend_id,
 811                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 812                                 rq_data_dir(
 813                                         (struct request *)
 814                                         info->shadow[req->id].request));
 815                 info->shadow[req->id].req = *req;
 816
 817                 info->ring.req_prod_pvt++;
 818         }
 819
 820         kfree(copy);
 821
 822         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 823
 824         spin_lock_irq(&blkif_io_lock);
 825
 826         /* Now safe for us to use the shared ring */
 827         info->connected = BLKIF_STATE_CONNECTED;
 828
 829         /* Send off requeued requests */
 830         flush_requests(info);
 831
 832         /* Kick any other new requests queued since we resumed */
 833         kick_pending_request_queues(info);
 834
 835         spin_unlock_irq(&blkif_io_lock);
 836
 837         return 0;
 838 }
 839
 840 /**
 841  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 842  * driver restart.  We tear down our blkif structure and recreate it, but
 843  * leave the device-layer structures intact so that this is transparent to the
 844  * rest of the kernel.
 845  */
 846 static int blkfront_resume(struct xenbus_device *dev)
 847 {
 848         struct blkfront_info *info = dev->dev.driver_data;
 849         int err;
 850
 851         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 852
 853         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 854
 855         err = talk_to_backend(dev, info);
 856         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 857                 err = blkif_recover(info);
 858
 859         return err;
 860 }
 861
 862
 863 /*
 864  * Invoked when the backend is finally 'ready' (and has told produced
 865  * the details about the physical device - #sectors, size, etc).
 866  */
 867 static void blkfront_connect(struct blkfront_info *info)
 868 {
 869         unsigned long long sectors;
 870         unsigned long sector_size;
 871         unsigned int binfo;
 872         int err;
 873
 874         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 875             (info->connected == BLKIF_STATE_SUSPENDED) )
 876                 return;
 877
 878         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 879                 __func__, info->xbdev->otherend);
 880
 881         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 882                             "sectors", "%llu", &sectors,
 883                             "info", "%u", &binfo,
 884                             "sector-size", "%lu", &sector_size,
 885                             NULL);
 886         if (err) {
 887                 xenbus_dev_fatal(info->xbdev, err,
 888                                  "reading backend fields at %s",
 889                                  info->xbdev->otherend);
 890                 return;
 891         }
 892
 893         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 894                             "feature-barrier", "%lu", &info->feature_barrier,
 895                             NULL);
 896         if (err)
 897                 info->feature_barrier = 0;
 898
 899         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
 900         if (err) {
 901                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 902                                  info->xbdev->otherend);
 903                 return;
 904         }
 905
 906         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 907
 908         /* Kick pending requests. */
 909         spin_lock_irq(&blkif_io_lock);
 910         info->connected = BLKIF_STATE_CONNECTED;
 911         kick_pending_request_queues(info);
 912         spin_unlock_irq(&blkif_io_lock);
 913
 914         add_disk(info->gd);
 915
 916         info->is_ready = 1;
 917 }
 918
 919 /**
 920  * Handle the change of state of the backend to Closing.  We must delete our
 921  * device-layer structures now, to ensure that writes are flushed through to
 922  * the backend.  Once is this done, we can switch to Closed in
 923  * acknowledgement.
 924  */
 925 static void blkfront_closing(struct xenbus_device *dev)
 926 {
 927         struct blkfront_info *info = dev->dev.driver_data;
 928         unsigned long flags;
 929
 930         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 931
 932         if (info->rq == NULL)
 933                 goto out;
 934
 935         spin_lock_irqsave(&blkif_io_lock, flags);
 936
 937         /* No more blkif_request(). */
 938         blk_stop_queue(info->rq);
 939
 940         /* No more gnttab callback work. */
 941         gnttab_cancel_free_callback(&info->callback);
 942         spin_unlock_irqrestore(&blkif_io_lock, flags);
 943
 944         /* Flush gnttab callback work. Must be done with no locks held. */
 945         flush_scheduled_work();
 946
 947         blk_cleanup_queue(info->rq);
 948         info->rq = NULL;
 949
 950         del_gendisk(info->gd);
 951
 952  out:
 953         xenbus_frontend_closed(dev);
 954 }
 955
 956 /**
 957  * Callback received when the backend's state changes.
 958  */
 959 static void backend_changed(struct xenbus_device *dev,
 960                             enum xenbus_state backend_state)
 961 {
 962         struct blkfront_info *info = dev->dev.driver_data;
 963         struct block_device *bd;
 964
 965         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 966
 967         switch (backend_state) {
 968         case XenbusStateInitialising:
 969         case XenbusStateInitWait:
 970         case XenbusStateInitialised:
 971         case XenbusStateUnknown:
 972         case XenbusStateClosed:
 973                 break;
 974
 975         case XenbusStateConnected:
 976                 blkfront_connect(info);
 977                 break;
 978
 979         case XenbusStateClosing:
 980                 if (info->gd == NULL) {
 981                         xenbus_frontend_closed(dev);
 982                         break;
 983                 }
 984                 bd = bdget_disk(info->gd, 0);
 985                 if (bd == NULL)
 986                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 987
 988                 mutex_lock(&bd->bd_mutex);
 989                 if (info->users > 0)
 990                         xenbus_dev_error(dev, -EBUSY,
 991                                          "Device in use; refusing to close");
 992                 else
 993                         blkfront_closing(dev);
 994                 mutex_unlock(&bd->bd_mutex);
 995                 bdput(bd);
 996                 break;
 997         }
 998 }
 999
1000 static int blkfront_remove(struct xenbus_device *dev)
1001 {
1002         struct blkfront_info *info = dev->dev.driver_data;
1003
1004         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1005
1006         blkif_free(info, 0);
1007
1008         kfree(info);
1009
1010         return 0;
1011 }
1012
1013 static int blkfront_is_ready(struct xenbus_device *dev)
1014 {
1015         struct blkfront_info *info = dev->dev.driver_data;
1016
1017         return info->is_ready;
1018 }
1019
1020 static int blkif_open(struct block_device *bdev, fmode_t mode)
1021 {
1022         struct blkfront_info *info = bdev->bd_disk->private_data;
1023         info->users++;
1024         return 0;
1025 }
1026
1027 static int blkif_release(struct gendisk *disk, fmode_t mode)
1028 {
1029         struct blkfront_info *info = disk->private_data;
1030         info->users--;
1031         if (info->users == 0) {
1032                 /* Check whether we have been instructed to close.  We will
1033                    have ignored this request initially, as the device was
1034                    still mounted. */
1035                 struct xenbus_device *dev = info->xbdev;
1036                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1037
1038                 if (state == XenbusStateClosing && info->is_ready)
1039                         blkfront_closing(dev);
1040         }
1041         return 0;
1042 }
1043
1044 static struct block_device_operations xlvbd_block_fops =
1045 {
1046         .owner = THIS_MODULE,
1047         .open = blkif_open,
1048         .release = blkif_release,
1049         .getgeo = blkif_getgeo,
1050         .locked_ioctl = blkif_ioctl,
1051 };
1052
1053
1054 static struct xenbus_device_id blkfront_ids[] = {
1055         { "vbd" },
1056         { "" }
1057 };
1058
1059 static struct xenbus_driver blkfront = {
1060         .name = "vbd",
1061         .owner = THIS_MODULE,
1062         .ids = blkfront_ids,
1063         .probe = blkfront_probe,
1064         .remove = blkfront_remove,
1065         .resume = blkfront_resume,
1066         .otherend_changed = backend_changed,
1067         .is_ready = blkfront_is_ready,
1068 };
1069
1070 static int __init xlblk_init(void)
1071 {
1072         if (!xen_domain())
1073                 return -ENODEV;
1074
1075         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1076                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1077                        XENVBD_MAJOR, DEV_NAME);
1078                 return -ENODEV;
1079         }
1080
1081         return xenbus_register_frontend(&blkfront);
1082 }
1083 module_init(xlblk_init);
1084
1085
1086 static void __exit xlblk_exit(void)
1087 {
1088         return xenbus_unregister_driver(&blkfront);
1089 }
1090 module_exit(xlblk_exit);
1091
1092 MODULE_DESCRIPTION("Xen virtual block device frontend");
1093 MODULE_LICENSE("GPL");
1094 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1095 MODULE_ALIAS("xen:vbd");
1096 MODULE_ALIAS("xenblk");