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